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 struct md_cluster_operations *md_cluster_ops;
88 EXPORT_SYMBOL(md_cluster_ops);
89 struct module *md_cluster_mod;
90 EXPORT_SYMBOL(md_cluster_mod);
92 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
93 static struct workqueue_struct *md_wq;
94 static struct workqueue_struct *md_misc_wq;
96 static int remove_and_add_spares(struct mddev *mddev,
97 struct md_rdev *this);
98 static void mddev_detach(struct mddev *mddev);
101 * Default number of read corrections we'll attempt on an rdev
102 * before ejecting it from the array. We divide the read error
103 * count by 2 for every hour elapsed between read errors.
105 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
107 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
108 * is 1000 KB/sec, so the extra system load does not show up that much.
109 * Increase it if you want to have more _guaranteed_ speed. Note that
110 * the RAID driver will use the maximum available bandwidth if the IO
111 * subsystem is idle. There is also an 'absolute maximum' reconstruction
112 * speed limit - in case reconstruction slows down your system despite
115 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
116 * or /sys/block/mdX/md/sync_speed_{min,max}
119 static int sysctl_speed_limit_min = 1000;
120 static int sysctl_speed_limit_max = 200000;
121 static inline int speed_min(struct mddev *mddev)
123 return mddev->sync_speed_min ?
124 mddev->sync_speed_min : sysctl_speed_limit_min;
127 static inline int speed_max(struct mddev *mddev)
129 return mddev->sync_speed_max ?
130 mddev->sync_speed_max : sysctl_speed_limit_max;
133 static struct ctl_table_header *raid_table_header;
135 static struct ctl_table raid_table[] = {
137 .procname = "speed_limit_min",
138 .data = &sysctl_speed_limit_min,
139 .maxlen = sizeof(int),
140 .mode = S_IRUGO|S_IWUSR,
141 .proc_handler = proc_dointvec,
144 .procname = "speed_limit_max",
145 .data = &sysctl_speed_limit_max,
146 .maxlen = sizeof(int),
147 .mode = S_IRUGO|S_IWUSR,
148 .proc_handler = proc_dointvec,
153 static struct ctl_table raid_dir_table[] = {
157 .mode = S_IRUGO|S_IXUGO,
163 static struct ctl_table raid_root_table[] = {
168 .child = raid_dir_table,
173 static const struct block_device_operations md_fops;
175 static int start_readonly;
178 * The original mechanism for creating an md device is to create
179 * a device node in /dev and to open it. This causes races with device-close.
180 * The preferred method is to write to the "new_array" module parameter.
181 * This can avoid races.
182 * Setting create_on_open to false disables the original mechanism
183 * so all the races disappear.
185 static bool create_on_open = true;
188 * like bio_clone_bioset, but with a local bio set
191 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
196 if (!mddev || !bioset_initialized(&mddev->bio_set))
197 return bio_alloc(gfp_mask, nr_iovecs);
199 b = bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
204 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
206 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
208 if (!mddev || !bioset_initialized(&mddev->sync_set))
209 return bio_alloc(GFP_NOIO, 1);
211 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
215 * We have a system wide 'event count' that is incremented
216 * on any 'interesting' event, and readers of /proc/mdstat
217 * can use 'poll' or 'select' to find out when the event
221 * start array, stop array, error, add device, remove device,
222 * start build, activate spare
224 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
225 static atomic_t md_event_count;
226 void md_new_event(struct mddev *mddev)
228 atomic_inc(&md_event_count);
229 wake_up(&md_event_waiters);
231 EXPORT_SYMBOL_GPL(md_new_event);
234 * Enables to iterate over all existing md arrays
235 * all_mddevs_lock protects this list.
237 static LIST_HEAD(all_mddevs);
238 static DEFINE_SPINLOCK(all_mddevs_lock);
241 * iterates through all used mddevs in the system.
242 * We take care to grab the all_mddevs_lock whenever navigating
243 * the list, and to always hold a refcount when unlocked.
244 * Any code which breaks out of this loop while own
245 * a reference to the current mddev and must mddev_put it.
247 #define for_each_mddev(_mddev,_tmp) \
249 for (({ spin_lock(&all_mddevs_lock); \
250 _tmp = all_mddevs.next; \
252 ({ if (_tmp != &all_mddevs) \
253 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
254 spin_unlock(&all_mddevs_lock); \
255 if (_mddev) mddev_put(_mddev); \
256 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
257 _tmp != &all_mddevs;}); \
258 ({ spin_lock(&all_mddevs_lock); \
259 _tmp = _tmp->next;}) \
262 /* Rather than calling directly into the personality make_request function,
263 * IO requests come here first so that we can check if the device is
264 * being suspended pending a reconfiguration.
265 * We hold a refcount over the call to ->make_request. By the time that
266 * call has finished, the bio has been linked into some internal structure
267 * and so is visible to ->quiesce(), so we don't need the refcount any more.
269 static bool is_suspended(struct mddev *mddev, struct bio *bio)
271 if (mddev->suspended)
273 if (bio_data_dir(bio) != WRITE)
275 if (mddev->suspend_lo >= mddev->suspend_hi)
277 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
279 if (bio_end_sector(bio) < mddev->suspend_lo)
284 void md_handle_request(struct mddev *mddev, struct bio *bio)
288 if (is_suspended(mddev, bio)) {
291 prepare_to_wait(&mddev->sb_wait, &__wait,
292 TASK_UNINTERRUPTIBLE);
293 if (!is_suspended(mddev, bio))
299 finish_wait(&mddev->sb_wait, &__wait);
301 atomic_inc(&mddev->active_io);
304 if (!mddev->pers->make_request(mddev, bio)) {
305 atomic_dec(&mddev->active_io);
306 wake_up(&mddev->sb_wait);
307 goto check_suspended;
310 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
311 wake_up(&mddev->sb_wait);
313 EXPORT_SYMBOL(md_handle_request);
315 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
317 const int rw = bio_data_dir(bio);
318 struct mddev *mddev = q->queuedata;
319 unsigned int sectors;
322 blk_queue_split(q, &bio);
324 if (mddev == NULL || mddev->pers == NULL) {
326 return BLK_QC_T_NONE;
328 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
329 if (bio_sectors(bio) != 0)
330 bio->bi_status = BLK_STS_IOERR;
332 return BLK_QC_T_NONE;
336 * save the sectors now since our bio can
337 * go away inside make_request
339 sectors = bio_sectors(bio);
340 /* bio could be mergeable after passing to underlayer */
341 bio->bi_opf &= ~REQ_NOMERGE;
343 md_handle_request(mddev, bio);
345 cpu = part_stat_lock();
346 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
347 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
350 return BLK_QC_T_NONE;
353 /* mddev_suspend makes sure no new requests are submitted
354 * to the device, and that any requests that have been submitted
355 * are completely handled.
356 * Once mddev_detach() is called and completes, the module will be
359 void mddev_suspend(struct mddev *mddev)
361 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
362 lockdep_assert_held(&mddev->reconfig_mutex);
363 if (mddev->suspended++)
366 wake_up(&mddev->sb_wait);
367 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
368 smp_mb__after_atomic();
369 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
370 mddev->pers->quiesce(mddev, 1);
371 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
372 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
374 del_timer_sync(&mddev->safemode_timer);
376 EXPORT_SYMBOL_GPL(mddev_suspend);
378 void mddev_resume(struct mddev *mddev)
380 lockdep_assert_held(&mddev->reconfig_mutex);
381 if (--mddev->suspended)
383 wake_up(&mddev->sb_wait);
384 mddev->pers->quiesce(mddev, 0);
386 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
387 md_wakeup_thread(mddev->thread);
388 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
390 EXPORT_SYMBOL_GPL(mddev_resume);
392 int mddev_congested(struct mddev *mddev, int bits)
394 struct md_personality *pers = mddev->pers;
398 if (mddev->suspended)
400 else if (pers && pers->congested)
401 ret = pers->congested(mddev, bits);
405 EXPORT_SYMBOL_GPL(mddev_congested);
406 static int md_congested(void *data, int bits)
408 struct mddev *mddev = data;
409 return mddev_congested(mddev, bits);
413 * Generic flush handling for md
416 static void md_end_flush(struct bio *bio)
418 struct md_rdev *rdev = bio->bi_private;
419 struct mddev *mddev = rdev->mddev;
421 rdev_dec_pending(rdev, mddev);
423 if (atomic_dec_and_test(&mddev->flush_pending)) {
424 /* The pre-request flush has finished */
425 queue_work(md_wq, &mddev->flush_work);
430 static void md_submit_flush_data(struct work_struct *ws);
432 static void submit_flushes(struct work_struct *ws)
434 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
435 struct md_rdev *rdev;
437 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
438 atomic_set(&mddev->flush_pending, 1);
440 rdev_for_each_rcu(rdev, mddev)
441 if (rdev->raid_disk >= 0 &&
442 !test_bit(Faulty, &rdev->flags)) {
443 /* Take two references, one is dropped
444 * when request finishes, one after
445 * we reclaim rcu_read_lock
448 atomic_inc(&rdev->nr_pending);
449 atomic_inc(&rdev->nr_pending);
451 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
452 bi->bi_end_io = md_end_flush;
453 bi->bi_private = rdev;
454 bio_set_dev(bi, rdev->bdev);
455 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
456 atomic_inc(&mddev->flush_pending);
459 rdev_dec_pending(rdev, mddev);
462 if (atomic_dec_and_test(&mddev->flush_pending))
463 queue_work(md_wq, &mddev->flush_work);
466 static void md_submit_flush_data(struct work_struct *ws)
468 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
469 struct bio *bio = mddev->flush_bio;
472 * must reset flush_bio before calling into md_handle_request to avoid a
473 * deadlock, because other bios passed md_handle_request suspend check
474 * could wait for this and below md_handle_request could wait for those
475 * bios because of suspend check
477 mddev->flush_bio = NULL;
478 wake_up(&mddev->sb_wait);
480 if (bio->bi_iter.bi_size == 0)
481 /* an empty barrier - all done */
484 bio->bi_opf &= ~REQ_PREFLUSH;
485 md_handle_request(mddev, bio);
489 void md_flush_request(struct mddev *mddev, struct bio *bio)
491 spin_lock_irq(&mddev->lock);
492 wait_event_lock_irq(mddev->sb_wait,
495 mddev->flush_bio = bio;
496 spin_unlock_irq(&mddev->lock);
498 INIT_WORK(&mddev->flush_work, submit_flushes);
499 queue_work(md_wq, &mddev->flush_work);
501 EXPORT_SYMBOL(md_flush_request);
503 static inline struct mddev *mddev_get(struct mddev *mddev)
505 atomic_inc(&mddev->active);
509 static void mddev_delayed_delete(struct work_struct *ws);
511 static void mddev_put(struct mddev *mddev)
513 struct bio_set bs, sync_bs;
515 memset(&bs, 0, sizeof(bs));
516 memset(&sync_bs, 0, sizeof(sync_bs));
518 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
520 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
521 mddev->ctime == 0 && !mddev->hold_active) {
522 /* Array is not configured at all, and not held active,
524 list_del_init(&mddev->all_mddevs);
526 sync_bs = mddev->sync_set;
527 memset(&mddev->bio_set, 0, sizeof(mddev->bio_set));
528 memset(&mddev->sync_set, 0, sizeof(mddev->sync_set));
529 if (mddev->gendisk) {
530 /* We did a probe so need to clean up. Call
531 * queue_work inside the spinlock so that
532 * flush_workqueue() after mddev_find will
533 * succeed in waiting for the work to be done.
535 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
536 queue_work(md_misc_wq, &mddev->del_work);
540 spin_unlock(&all_mddevs_lock);
542 bioset_exit(&sync_bs);
545 static void md_safemode_timeout(struct timer_list *t);
547 void mddev_init(struct mddev *mddev)
549 mutex_init(&mddev->open_mutex);
550 mutex_init(&mddev->reconfig_mutex);
551 mutex_init(&mddev->bitmap_info.mutex);
552 INIT_LIST_HEAD(&mddev->disks);
553 INIT_LIST_HEAD(&mddev->all_mddevs);
554 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
555 atomic_set(&mddev->active, 1);
556 atomic_set(&mddev->openers, 0);
557 atomic_set(&mddev->active_io, 0);
558 spin_lock_init(&mddev->lock);
559 atomic_set(&mddev->flush_pending, 0);
560 init_waitqueue_head(&mddev->sb_wait);
561 init_waitqueue_head(&mddev->recovery_wait);
562 mddev->reshape_position = MaxSector;
563 mddev->reshape_backwards = 0;
564 mddev->last_sync_action = "none";
565 mddev->resync_min = 0;
566 mddev->resync_max = MaxSector;
567 mddev->level = LEVEL_NONE;
569 EXPORT_SYMBOL_GPL(mddev_init);
571 static struct mddev *mddev_find(dev_t unit)
573 struct mddev *mddev, *new = NULL;
575 if (unit && MAJOR(unit) != MD_MAJOR)
576 unit &= ~((1<<MdpMinorShift)-1);
579 spin_lock(&all_mddevs_lock);
582 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
583 if (mddev->unit == unit) {
585 spin_unlock(&all_mddevs_lock);
591 list_add(&new->all_mddevs, &all_mddevs);
592 spin_unlock(&all_mddevs_lock);
593 new->hold_active = UNTIL_IOCTL;
597 /* find an unused unit number */
598 static int next_minor = 512;
599 int start = next_minor;
603 dev = MKDEV(MD_MAJOR, next_minor);
605 if (next_minor > MINORMASK)
607 if (next_minor == start) {
608 /* Oh dear, all in use. */
609 spin_unlock(&all_mddevs_lock);
615 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
616 if (mddev->unit == dev) {
622 new->md_minor = MINOR(dev);
623 new->hold_active = UNTIL_STOP;
624 list_add(&new->all_mddevs, &all_mddevs);
625 spin_unlock(&all_mddevs_lock);
628 spin_unlock(&all_mddevs_lock);
630 new = kzalloc(sizeof(*new), GFP_KERNEL);
635 if (MAJOR(unit) == MD_MAJOR)
636 new->md_minor = MINOR(unit);
638 new->md_minor = MINOR(unit) >> MdpMinorShift;
645 static struct attribute_group md_redundancy_group;
647 void mddev_unlock(struct mddev *mddev)
649 if (mddev->to_remove) {
650 /* These cannot be removed under reconfig_mutex as
651 * an access to the files will try to take reconfig_mutex
652 * while holding the file unremovable, which leads to
654 * So hold set sysfs_active while the remove in happeing,
655 * and anything else which might set ->to_remove or my
656 * otherwise change the sysfs namespace will fail with
657 * -EBUSY if sysfs_active is still set.
658 * We set sysfs_active under reconfig_mutex and elsewhere
659 * test it under the same mutex to ensure its correct value
662 struct attribute_group *to_remove = mddev->to_remove;
663 mddev->to_remove = NULL;
664 mddev->sysfs_active = 1;
665 mutex_unlock(&mddev->reconfig_mutex);
667 if (mddev->kobj.sd) {
668 if (to_remove != &md_redundancy_group)
669 sysfs_remove_group(&mddev->kobj, to_remove);
670 if (mddev->pers == NULL ||
671 mddev->pers->sync_request == NULL) {
672 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
673 if (mddev->sysfs_action)
674 sysfs_put(mddev->sysfs_action);
675 mddev->sysfs_action = NULL;
678 mddev->sysfs_active = 0;
680 mutex_unlock(&mddev->reconfig_mutex);
682 /* As we've dropped the mutex we need a spinlock to
683 * make sure the thread doesn't disappear
685 spin_lock(&pers_lock);
686 md_wakeup_thread(mddev->thread);
687 wake_up(&mddev->sb_wait);
688 spin_unlock(&pers_lock);
690 EXPORT_SYMBOL_GPL(mddev_unlock);
692 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
694 struct md_rdev *rdev;
696 rdev_for_each_rcu(rdev, mddev)
697 if (rdev->desc_nr == nr)
702 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
704 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
706 struct md_rdev *rdev;
708 rdev_for_each(rdev, mddev)
709 if (rdev->bdev->bd_dev == dev)
715 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
717 struct md_rdev *rdev;
719 rdev_for_each_rcu(rdev, mddev)
720 if (rdev->bdev->bd_dev == dev)
725 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
727 static struct md_personality *find_pers(int level, char *clevel)
729 struct md_personality *pers;
730 list_for_each_entry(pers, &pers_list, list) {
731 if (level != LEVEL_NONE && pers->level == level)
733 if (strcmp(pers->name, clevel)==0)
739 /* return the offset of the super block in 512byte sectors */
740 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
742 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
743 return MD_NEW_SIZE_SECTORS(num_sectors);
746 static int alloc_disk_sb(struct md_rdev *rdev)
748 rdev->sb_page = alloc_page(GFP_KERNEL);
754 void md_rdev_clear(struct md_rdev *rdev)
757 put_page(rdev->sb_page);
759 rdev->sb_page = NULL;
764 put_page(rdev->bb_page);
765 rdev->bb_page = NULL;
767 badblocks_exit(&rdev->badblocks);
769 EXPORT_SYMBOL_GPL(md_rdev_clear);
771 static void super_written(struct bio *bio)
773 struct md_rdev *rdev = bio->bi_private;
774 struct mddev *mddev = rdev->mddev;
776 if (bio->bi_status) {
777 pr_err("md: super_written gets error=%d\n", bio->bi_status);
778 md_error(mddev, rdev);
779 if (!test_bit(Faulty, &rdev->flags)
780 && (bio->bi_opf & MD_FAILFAST)) {
781 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
782 set_bit(LastDev, &rdev->flags);
785 clear_bit(LastDev, &rdev->flags);
787 if (atomic_dec_and_test(&mddev->pending_writes))
788 wake_up(&mddev->sb_wait);
789 rdev_dec_pending(rdev, mddev);
793 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
794 sector_t sector, int size, struct page *page)
796 /* write first size bytes of page to sector of rdev
797 * Increment mddev->pending_writes before returning
798 * and decrement it on completion, waking up sb_wait
799 * if zero is reached.
800 * If an error occurred, call md_error
808 if (test_bit(Faulty, &rdev->flags))
811 bio = md_bio_alloc_sync(mddev);
813 atomic_inc(&rdev->nr_pending);
815 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
816 bio->bi_iter.bi_sector = sector;
817 bio_add_page(bio, page, size, 0);
818 bio->bi_private = rdev;
819 bio->bi_end_io = super_written;
821 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
822 test_bit(FailFast, &rdev->flags) &&
823 !test_bit(LastDev, &rdev->flags))
825 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
827 atomic_inc(&mddev->pending_writes);
831 int md_super_wait(struct mddev *mddev)
833 /* wait for all superblock writes that were scheduled to complete */
834 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
835 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
840 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
841 struct page *page, int op, int op_flags, bool metadata_op)
843 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
846 if (metadata_op && rdev->meta_bdev)
847 bio_set_dev(bio, rdev->meta_bdev);
849 bio_set_dev(bio, rdev->bdev);
850 bio_set_op_attrs(bio, op, op_flags);
852 bio->bi_iter.bi_sector = sector + rdev->sb_start;
853 else if (rdev->mddev->reshape_position != MaxSector &&
854 (rdev->mddev->reshape_backwards ==
855 (sector >= rdev->mddev->reshape_position)))
856 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
858 bio->bi_iter.bi_sector = sector + rdev->data_offset;
859 bio_add_page(bio, page, size, 0);
861 submit_bio_wait(bio);
863 ret = !bio->bi_status;
867 EXPORT_SYMBOL_GPL(sync_page_io);
869 static int read_disk_sb(struct md_rdev *rdev, int size)
871 char b[BDEVNAME_SIZE];
876 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
882 pr_err("md: disabled device %s, could not read superblock.\n",
883 bdevname(rdev->bdev,b));
887 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
889 return sb1->set_uuid0 == sb2->set_uuid0 &&
890 sb1->set_uuid1 == sb2->set_uuid1 &&
891 sb1->set_uuid2 == sb2->set_uuid2 &&
892 sb1->set_uuid3 == sb2->set_uuid3;
895 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
898 mdp_super_t *tmp1, *tmp2;
900 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
901 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
903 if (!tmp1 || !tmp2) {
912 * nr_disks is not constant
917 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
924 static u32 md_csum_fold(u32 csum)
926 csum = (csum & 0xffff) + (csum >> 16);
927 return (csum & 0xffff) + (csum >> 16);
930 static unsigned int calc_sb_csum(mdp_super_t *sb)
933 u32 *sb32 = (u32*)sb;
935 unsigned int disk_csum, csum;
937 disk_csum = sb->sb_csum;
940 for (i = 0; i < MD_SB_BYTES/4 ; i++)
942 csum = (newcsum & 0xffffffff) + (newcsum>>32);
945 /* This used to use csum_partial, which was wrong for several
946 * reasons including that different results are returned on
947 * different architectures. It isn't critical that we get exactly
948 * the same return value as before (we always csum_fold before
949 * testing, and that removes any differences). However as we
950 * know that csum_partial always returned a 16bit value on
951 * alphas, do a fold to maximise conformity to previous behaviour.
953 sb->sb_csum = md_csum_fold(disk_csum);
955 sb->sb_csum = disk_csum;
961 * Handle superblock details.
962 * We want to be able to handle multiple superblock formats
963 * so we have a common interface to them all, and an array of
964 * different handlers.
965 * We rely on user-space to write the initial superblock, and support
966 * reading and updating of superblocks.
967 * Interface methods are:
968 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
969 * loads and validates a superblock on dev.
970 * if refdev != NULL, compare superblocks on both devices
972 * 0 - dev has a superblock that is compatible with refdev
973 * 1 - dev has a superblock that is compatible and newer than refdev
974 * so dev should be used as the refdev in future
975 * -EINVAL superblock incompatible or invalid
976 * -othererror e.g. -EIO
978 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
979 * Verify that dev is acceptable into mddev.
980 * The first time, mddev->raid_disks will be 0, and data from
981 * dev should be merged in. Subsequent calls check that dev
982 * is new enough. Return 0 or -EINVAL
984 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
985 * Update the superblock for rdev with data in mddev
986 * This does not write to disc.
992 struct module *owner;
993 int (*load_super)(struct md_rdev *rdev,
994 struct md_rdev *refdev,
996 int (*validate_super)(struct mddev *mddev,
997 struct md_rdev *rdev);
998 void (*sync_super)(struct mddev *mddev,
999 struct md_rdev *rdev);
1000 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1001 sector_t num_sectors);
1002 int (*allow_new_offset)(struct md_rdev *rdev,
1003 unsigned long long new_offset);
1007 * Check that the given mddev has no bitmap.
1009 * This function is called from the run method of all personalities that do not
1010 * support bitmaps. It prints an error message and returns non-zero if mddev
1011 * has a bitmap. Otherwise, it returns 0.
1014 int md_check_no_bitmap(struct mddev *mddev)
1016 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1018 pr_warn("%s: bitmaps are not supported for %s\n",
1019 mdname(mddev), mddev->pers->name);
1022 EXPORT_SYMBOL(md_check_no_bitmap);
1025 * load_super for 0.90.0
1027 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1029 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1034 * Calculate the position of the superblock (512byte sectors),
1035 * it's at the end of the disk.
1037 * It also happens to be a multiple of 4Kb.
1039 rdev->sb_start = calc_dev_sboffset(rdev);
1041 ret = read_disk_sb(rdev, MD_SB_BYTES);
1047 bdevname(rdev->bdev, b);
1048 sb = page_address(rdev->sb_page);
1050 if (sb->md_magic != MD_SB_MAGIC) {
1051 pr_warn("md: invalid raid superblock magic on %s\n", b);
1055 if (sb->major_version != 0 ||
1056 sb->minor_version < 90 ||
1057 sb->minor_version > 91) {
1058 pr_warn("Bad version number %d.%d on %s\n",
1059 sb->major_version, sb->minor_version, b);
1063 if (sb->raid_disks <= 0)
1066 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1067 pr_warn("md: invalid superblock checksum on %s\n", b);
1071 rdev->preferred_minor = sb->md_minor;
1072 rdev->data_offset = 0;
1073 rdev->new_data_offset = 0;
1074 rdev->sb_size = MD_SB_BYTES;
1075 rdev->badblocks.shift = -1;
1077 if (sb->level == LEVEL_MULTIPATH)
1080 rdev->desc_nr = sb->this_disk.number;
1086 mdp_super_t *refsb = page_address(refdev->sb_page);
1087 if (!md_uuid_equal(refsb, sb)) {
1088 pr_warn("md: %s has different UUID to %s\n",
1089 b, bdevname(refdev->bdev,b2));
1092 if (!md_sb_equal(refsb, sb)) {
1093 pr_warn("md: %s has same UUID but different superblock to %s\n",
1094 b, bdevname(refdev->bdev, b2));
1098 ev2 = md_event(refsb);
1104 rdev->sectors = rdev->sb_start;
1105 /* Limit to 4TB as metadata cannot record more than that.
1106 * (not needed for Linear and RAID0 as metadata doesn't
1109 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1111 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1113 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1114 /* "this cannot possibly happen" ... */
1122 * validate_super for 0.90.0
1124 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1127 mdp_super_t *sb = page_address(rdev->sb_page);
1128 __u64 ev1 = md_event(sb);
1130 rdev->raid_disk = -1;
1131 clear_bit(Faulty, &rdev->flags);
1132 clear_bit(In_sync, &rdev->flags);
1133 clear_bit(Bitmap_sync, &rdev->flags);
1134 clear_bit(WriteMostly, &rdev->flags);
1136 if (mddev->raid_disks == 0) {
1137 mddev->major_version = 0;
1138 mddev->minor_version = sb->minor_version;
1139 mddev->patch_version = sb->patch_version;
1140 mddev->external = 0;
1141 mddev->chunk_sectors = sb->chunk_size >> 9;
1142 mddev->ctime = sb->ctime;
1143 mddev->utime = sb->utime;
1144 mddev->level = sb->level;
1145 mddev->clevel[0] = 0;
1146 mddev->layout = sb->layout;
1147 mddev->raid_disks = sb->raid_disks;
1148 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1149 mddev->events = ev1;
1150 mddev->bitmap_info.offset = 0;
1151 mddev->bitmap_info.space = 0;
1152 /* bitmap can use 60 K after the 4K superblocks */
1153 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1154 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1155 mddev->reshape_backwards = 0;
1157 if (mddev->minor_version >= 91) {
1158 mddev->reshape_position = sb->reshape_position;
1159 mddev->delta_disks = sb->delta_disks;
1160 mddev->new_level = sb->new_level;
1161 mddev->new_layout = sb->new_layout;
1162 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1163 if (mddev->delta_disks < 0)
1164 mddev->reshape_backwards = 1;
1166 mddev->reshape_position = MaxSector;
1167 mddev->delta_disks = 0;
1168 mddev->new_level = mddev->level;
1169 mddev->new_layout = mddev->layout;
1170 mddev->new_chunk_sectors = mddev->chunk_sectors;
1173 if (sb->state & (1<<MD_SB_CLEAN))
1174 mddev->recovery_cp = MaxSector;
1176 if (sb->events_hi == sb->cp_events_hi &&
1177 sb->events_lo == sb->cp_events_lo) {
1178 mddev->recovery_cp = sb->recovery_cp;
1180 mddev->recovery_cp = 0;
1183 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1184 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1185 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1186 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1188 mddev->max_disks = MD_SB_DISKS;
1190 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1191 mddev->bitmap_info.file == NULL) {
1192 mddev->bitmap_info.offset =
1193 mddev->bitmap_info.default_offset;
1194 mddev->bitmap_info.space =
1195 mddev->bitmap_info.default_space;
1198 } else if (mddev->pers == NULL) {
1199 /* Insist on good event counter while assembling, except
1200 * for spares (which don't need an event count) */
1202 if (sb->disks[rdev->desc_nr].state & (
1203 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1204 if (ev1 < mddev->events)
1206 } else if (mddev->bitmap) {
1207 /* if adding to array with a bitmap, then we can accept an
1208 * older device ... but not too old.
1210 if (ev1 < mddev->bitmap->events_cleared)
1212 if (ev1 < mddev->events)
1213 set_bit(Bitmap_sync, &rdev->flags);
1215 if (ev1 < mddev->events)
1216 /* just a hot-add of a new device, leave raid_disk at -1 */
1220 if (mddev->level != LEVEL_MULTIPATH) {
1221 desc = sb->disks + rdev->desc_nr;
1223 if (desc->state & (1<<MD_DISK_FAULTY))
1224 set_bit(Faulty, &rdev->flags);
1225 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1226 desc->raid_disk < mddev->raid_disks */) {
1227 set_bit(In_sync, &rdev->flags);
1228 rdev->raid_disk = desc->raid_disk;
1229 rdev->saved_raid_disk = desc->raid_disk;
1230 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1231 /* active but not in sync implies recovery up to
1232 * reshape position. We don't know exactly where
1233 * that is, so set to zero for now */
1234 if (mddev->minor_version >= 91) {
1235 rdev->recovery_offset = 0;
1236 rdev->raid_disk = desc->raid_disk;
1239 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1240 set_bit(WriteMostly, &rdev->flags);
1241 if (desc->state & (1<<MD_DISK_FAILFAST))
1242 set_bit(FailFast, &rdev->flags);
1243 } else /* MULTIPATH are always insync */
1244 set_bit(In_sync, &rdev->flags);
1249 * sync_super for 0.90.0
1251 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1254 struct md_rdev *rdev2;
1255 int next_spare = mddev->raid_disks;
1257 /* make rdev->sb match mddev data..
1260 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1261 * 3/ any empty disks < next_spare become removed
1263 * disks[0] gets initialised to REMOVED because
1264 * we cannot be sure from other fields if it has
1265 * been initialised or not.
1268 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1270 rdev->sb_size = MD_SB_BYTES;
1272 sb = page_address(rdev->sb_page);
1274 memset(sb, 0, sizeof(*sb));
1276 sb->md_magic = MD_SB_MAGIC;
1277 sb->major_version = mddev->major_version;
1278 sb->patch_version = mddev->patch_version;
1279 sb->gvalid_words = 0; /* ignored */
1280 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1281 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1282 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1283 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1285 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1286 sb->level = mddev->level;
1287 sb->size = mddev->dev_sectors / 2;
1288 sb->raid_disks = mddev->raid_disks;
1289 sb->md_minor = mddev->md_minor;
1290 sb->not_persistent = 0;
1291 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1293 sb->events_hi = (mddev->events>>32);
1294 sb->events_lo = (u32)mddev->events;
1296 if (mddev->reshape_position == MaxSector)
1297 sb->minor_version = 90;
1299 sb->minor_version = 91;
1300 sb->reshape_position = mddev->reshape_position;
1301 sb->new_level = mddev->new_level;
1302 sb->delta_disks = mddev->delta_disks;
1303 sb->new_layout = mddev->new_layout;
1304 sb->new_chunk = mddev->new_chunk_sectors << 9;
1306 mddev->minor_version = sb->minor_version;
1309 sb->recovery_cp = mddev->recovery_cp;
1310 sb->cp_events_hi = (mddev->events>>32);
1311 sb->cp_events_lo = (u32)mddev->events;
1312 if (mddev->recovery_cp == MaxSector)
1313 sb->state = (1<< MD_SB_CLEAN);
1315 sb->recovery_cp = 0;
1317 sb->layout = mddev->layout;
1318 sb->chunk_size = mddev->chunk_sectors << 9;
1320 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1321 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1323 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1324 rdev_for_each(rdev2, mddev) {
1327 int is_active = test_bit(In_sync, &rdev2->flags);
1329 if (rdev2->raid_disk >= 0 &&
1330 sb->minor_version >= 91)
1331 /* we have nowhere to store the recovery_offset,
1332 * but if it is not below the reshape_position,
1333 * we can piggy-back on that.
1336 if (rdev2->raid_disk < 0 ||
1337 test_bit(Faulty, &rdev2->flags))
1340 desc_nr = rdev2->raid_disk;
1342 desc_nr = next_spare++;
1343 rdev2->desc_nr = desc_nr;
1344 d = &sb->disks[rdev2->desc_nr];
1346 d->number = rdev2->desc_nr;
1347 d->major = MAJOR(rdev2->bdev->bd_dev);
1348 d->minor = MINOR(rdev2->bdev->bd_dev);
1350 d->raid_disk = rdev2->raid_disk;
1352 d->raid_disk = rdev2->desc_nr; /* compatibility */
1353 if (test_bit(Faulty, &rdev2->flags))
1354 d->state = (1<<MD_DISK_FAULTY);
1355 else if (is_active) {
1356 d->state = (1<<MD_DISK_ACTIVE);
1357 if (test_bit(In_sync, &rdev2->flags))
1358 d->state |= (1<<MD_DISK_SYNC);
1366 if (test_bit(WriteMostly, &rdev2->flags))
1367 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1368 if (test_bit(FailFast, &rdev2->flags))
1369 d->state |= (1<<MD_DISK_FAILFAST);
1371 /* now set the "removed" and "faulty" bits on any missing devices */
1372 for (i=0 ; i < mddev->raid_disks ; i++) {
1373 mdp_disk_t *d = &sb->disks[i];
1374 if (d->state == 0 && d->number == 0) {
1377 d->state = (1<<MD_DISK_REMOVED);
1378 d->state |= (1<<MD_DISK_FAULTY);
1382 sb->nr_disks = nr_disks;
1383 sb->active_disks = active;
1384 sb->working_disks = working;
1385 sb->failed_disks = failed;
1386 sb->spare_disks = spare;
1388 sb->this_disk = sb->disks[rdev->desc_nr];
1389 sb->sb_csum = calc_sb_csum(sb);
1393 * rdev_size_change for 0.90.0
1395 static unsigned long long
1396 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1398 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1399 return 0; /* component must fit device */
1400 if (rdev->mddev->bitmap_info.offset)
1401 return 0; /* can't move bitmap */
1402 rdev->sb_start = calc_dev_sboffset(rdev);
1403 if (!num_sectors || num_sectors > rdev->sb_start)
1404 num_sectors = rdev->sb_start;
1405 /* Limit to 4TB as metadata cannot record more than that.
1406 * 4TB == 2^32 KB, or 2*2^32 sectors.
1408 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1409 rdev->mddev->level >= 1)
1410 num_sectors = (sector_t)(2ULL << 32) - 2;
1412 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1414 } while (md_super_wait(rdev->mddev) < 0);
1419 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1421 /* non-zero offset changes not possible with v0.90 */
1422 return new_offset == 0;
1426 * version 1 superblock
1429 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1433 unsigned long long newcsum;
1434 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1435 __le32 *isuper = (__le32*)sb;
1437 disk_csum = sb->sb_csum;
1440 for (; size >= 4; size -= 4)
1441 newcsum += le32_to_cpu(*isuper++);
1444 newcsum += le16_to_cpu(*(__le16*) isuper);
1446 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1447 sb->sb_csum = disk_csum;
1448 return cpu_to_le32(csum);
1451 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1453 struct mdp_superblock_1 *sb;
1457 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1461 * Calculate the position of the superblock in 512byte sectors.
1462 * It is always aligned to a 4K boundary and
1463 * depeding on minor_version, it can be:
1464 * 0: At least 8K, but less than 12K, from end of device
1465 * 1: At start of device
1466 * 2: 4K from start of device.
1468 switch(minor_version) {
1470 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1472 sb_start &= ~(sector_t)(4*2-1);
1483 rdev->sb_start = sb_start;
1485 /* superblock is rarely larger than 1K, but it can be larger,
1486 * and it is safe to read 4k, so we do that
1488 ret = read_disk_sb(rdev, 4096);
1489 if (ret) return ret;
1491 sb = page_address(rdev->sb_page);
1493 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1494 sb->major_version != cpu_to_le32(1) ||
1495 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1496 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1497 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1500 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1501 pr_warn("md: invalid superblock checksum on %s\n",
1502 bdevname(rdev->bdev,b));
1505 if (le64_to_cpu(sb->data_size) < 10) {
1506 pr_warn("md: data_size too small on %s\n",
1507 bdevname(rdev->bdev,b));
1512 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1513 /* Some padding is non-zero, might be a new feature */
1516 rdev->preferred_minor = 0xffff;
1517 rdev->data_offset = le64_to_cpu(sb->data_offset);
1518 rdev->new_data_offset = rdev->data_offset;
1519 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1520 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1521 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1522 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1524 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1525 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1526 if (rdev->sb_size & bmask)
1527 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1530 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1533 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1536 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1539 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1541 if (!rdev->bb_page) {
1542 rdev->bb_page = alloc_page(GFP_KERNEL);
1546 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1547 rdev->badblocks.count == 0) {
1548 /* need to load the bad block list.
1549 * Currently we limit it to one page.
1555 int sectors = le16_to_cpu(sb->bblog_size);
1556 if (sectors > (PAGE_SIZE / 512))
1558 offset = le32_to_cpu(sb->bblog_offset);
1561 bb_sector = (long long)offset;
1562 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1563 rdev->bb_page, REQ_OP_READ, 0, true))
1565 bbp = (u64 *)page_address(rdev->bb_page);
1566 rdev->badblocks.shift = sb->bblog_shift;
1567 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1568 u64 bb = le64_to_cpu(*bbp);
1569 int count = bb & (0x3ff);
1570 u64 sector = bb >> 10;
1571 sector <<= sb->bblog_shift;
1572 count <<= sb->bblog_shift;
1575 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1578 } else if (sb->bblog_offset != 0)
1579 rdev->badblocks.shift = 0;
1581 if ((le32_to_cpu(sb->feature_map) &
1582 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1583 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1584 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1585 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1592 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1594 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1595 sb->level != refsb->level ||
1596 sb->layout != refsb->layout ||
1597 sb->chunksize != refsb->chunksize) {
1598 pr_warn("md: %s has strangely different superblock to %s\n",
1599 bdevname(rdev->bdev,b),
1600 bdevname(refdev->bdev,b2));
1603 ev1 = le64_to_cpu(sb->events);
1604 ev2 = le64_to_cpu(refsb->events);
1611 if (minor_version) {
1612 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1613 sectors -= rdev->data_offset;
1615 sectors = rdev->sb_start;
1616 if (sectors < le64_to_cpu(sb->data_size))
1618 rdev->sectors = le64_to_cpu(sb->data_size);
1622 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1624 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1625 __u64 ev1 = le64_to_cpu(sb->events);
1627 rdev->raid_disk = -1;
1628 clear_bit(Faulty, &rdev->flags);
1629 clear_bit(In_sync, &rdev->flags);
1630 clear_bit(Bitmap_sync, &rdev->flags);
1631 clear_bit(WriteMostly, &rdev->flags);
1633 if (mddev->raid_disks == 0) {
1634 mddev->major_version = 1;
1635 mddev->patch_version = 0;
1636 mddev->external = 0;
1637 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1638 mddev->ctime = le64_to_cpu(sb->ctime);
1639 mddev->utime = le64_to_cpu(sb->utime);
1640 mddev->level = le32_to_cpu(sb->level);
1641 mddev->clevel[0] = 0;
1642 mddev->layout = le32_to_cpu(sb->layout);
1643 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1644 mddev->dev_sectors = le64_to_cpu(sb->size);
1645 mddev->events = ev1;
1646 mddev->bitmap_info.offset = 0;
1647 mddev->bitmap_info.space = 0;
1648 /* Default location for bitmap is 1K after superblock
1649 * using 3K - total of 4K
1651 mddev->bitmap_info.default_offset = 1024 >> 9;
1652 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1653 mddev->reshape_backwards = 0;
1655 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1656 memcpy(mddev->uuid, sb->set_uuid, 16);
1658 mddev->max_disks = (4096-256)/2;
1660 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1661 mddev->bitmap_info.file == NULL) {
1662 mddev->bitmap_info.offset =
1663 (__s32)le32_to_cpu(sb->bitmap_offset);
1664 /* Metadata doesn't record how much space is available.
1665 * For 1.0, we assume we can use up to the superblock
1666 * if before, else to 4K beyond superblock.
1667 * For others, assume no change is possible.
1669 if (mddev->minor_version > 0)
1670 mddev->bitmap_info.space = 0;
1671 else if (mddev->bitmap_info.offset > 0)
1672 mddev->bitmap_info.space =
1673 8 - mddev->bitmap_info.offset;
1675 mddev->bitmap_info.space =
1676 -mddev->bitmap_info.offset;
1679 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1680 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1681 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1682 mddev->new_level = le32_to_cpu(sb->new_level);
1683 mddev->new_layout = le32_to_cpu(sb->new_layout);
1684 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1685 if (mddev->delta_disks < 0 ||
1686 (mddev->delta_disks == 0 &&
1687 (le32_to_cpu(sb->feature_map)
1688 & MD_FEATURE_RESHAPE_BACKWARDS)))
1689 mddev->reshape_backwards = 1;
1691 mddev->reshape_position = MaxSector;
1692 mddev->delta_disks = 0;
1693 mddev->new_level = mddev->level;
1694 mddev->new_layout = mddev->layout;
1695 mddev->new_chunk_sectors = mddev->chunk_sectors;
1698 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1699 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1701 if (le32_to_cpu(sb->feature_map) &
1702 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1703 if (le32_to_cpu(sb->feature_map) &
1704 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1706 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1707 (le32_to_cpu(sb->feature_map) &
1708 MD_FEATURE_MULTIPLE_PPLS))
1710 set_bit(MD_HAS_PPL, &mddev->flags);
1712 } else if (mddev->pers == NULL) {
1713 /* Insist of good event counter while assembling, except for
1714 * spares (which don't need an event count) */
1716 if (rdev->desc_nr >= 0 &&
1717 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1718 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1719 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1720 if (ev1 < mddev->events)
1722 } else if (mddev->bitmap) {
1723 /* If adding to array with a bitmap, then we can accept an
1724 * older device, but not too old.
1726 if (ev1 < mddev->bitmap->events_cleared)
1728 if (ev1 < mddev->events)
1729 set_bit(Bitmap_sync, &rdev->flags);
1731 if (ev1 < mddev->events)
1732 /* just a hot-add of a new device, leave raid_disk at -1 */
1735 if (mddev->level != LEVEL_MULTIPATH) {
1737 if (rdev->desc_nr < 0 ||
1738 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1739 role = MD_DISK_ROLE_SPARE;
1742 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1744 case MD_DISK_ROLE_SPARE: /* spare */
1746 case MD_DISK_ROLE_FAULTY: /* faulty */
1747 set_bit(Faulty, &rdev->flags);
1749 case MD_DISK_ROLE_JOURNAL: /* journal device */
1750 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1751 /* journal device without journal feature */
1752 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1755 set_bit(Journal, &rdev->flags);
1756 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1757 rdev->raid_disk = 0;
1760 rdev->saved_raid_disk = role;
1761 if ((le32_to_cpu(sb->feature_map) &
1762 MD_FEATURE_RECOVERY_OFFSET)) {
1763 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1764 if (!(le32_to_cpu(sb->feature_map) &
1765 MD_FEATURE_RECOVERY_BITMAP))
1766 rdev->saved_raid_disk = -1;
1768 set_bit(In_sync, &rdev->flags);
1769 rdev->raid_disk = role;
1772 if (sb->devflags & WriteMostly1)
1773 set_bit(WriteMostly, &rdev->flags);
1774 if (sb->devflags & FailFast1)
1775 set_bit(FailFast, &rdev->flags);
1776 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1777 set_bit(Replacement, &rdev->flags);
1778 } else /* MULTIPATH are always insync */
1779 set_bit(In_sync, &rdev->flags);
1784 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1786 struct mdp_superblock_1 *sb;
1787 struct md_rdev *rdev2;
1789 /* make rdev->sb match mddev and rdev data. */
1791 sb = page_address(rdev->sb_page);
1793 sb->feature_map = 0;
1795 sb->recovery_offset = cpu_to_le64(0);
1796 memset(sb->pad3, 0, sizeof(sb->pad3));
1798 sb->utime = cpu_to_le64((__u64)mddev->utime);
1799 sb->events = cpu_to_le64(mddev->events);
1801 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1802 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1803 sb->resync_offset = cpu_to_le64(MaxSector);
1805 sb->resync_offset = cpu_to_le64(0);
1807 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1809 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1810 sb->size = cpu_to_le64(mddev->dev_sectors);
1811 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1812 sb->level = cpu_to_le32(mddev->level);
1813 sb->layout = cpu_to_le32(mddev->layout);
1814 if (test_bit(FailFast, &rdev->flags))
1815 sb->devflags |= FailFast1;
1817 sb->devflags &= ~FailFast1;
1819 if (test_bit(WriteMostly, &rdev->flags))
1820 sb->devflags |= WriteMostly1;
1822 sb->devflags &= ~WriteMostly1;
1823 sb->data_offset = cpu_to_le64(rdev->data_offset);
1824 sb->data_size = cpu_to_le64(rdev->sectors);
1826 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1827 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1828 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1831 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1832 !test_bit(In_sync, &rdev->flags)) {
1834 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1835 sb->recovery_offset =
1836 cpu_to_le64(rdev->recovery_offset);
1837 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1839 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1841 /* Note: recovery_offset and journal_tail share space */
1842 if (test_bit(Journal, &rdev->flags))
1843 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1844 if (test_bit(Replacement, &rdev->flags))
1846 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1848 if (mddev->reshape_position != MaxSector) {
1849 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1850 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1851 sb->new_layout = cpu_to_le32(mddev->new_layout);
1852 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1853 sb->new_level = cpu_to_le32(mddev->new_level);
1854 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1855 if (mddev->delta_disks == 0 &&
1856 mddev->reshape_backwards)
1858 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1859 if (rdev->new_data_offset != rdev->data_offset) {
1861 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1862 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1863 - rdev->data_offset));
1867 if (mddev_is_clustered(mddev))
1868 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1870 if (rdev->badblocks.count == 0)
1871 /* Nothing to do for bad blocks*/ ;
1872 else if (sb->bblog_offset == 0)
1873 /* Cannot record bad blocks on this device */
1874 md_error(mddev, rdev);
1876 struct badblocks *bb = &rdev->badblocks;
1877 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1879 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1884 seq = read_seqbegin(&bb->lock);
1886 memset(bbp, 0xff, PAGE_SIZE);
1888 for (i = 0 ; i < bb->count ; i++) {
1889 u64 internal_bb = p[i];
1890 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1891 | BB_LEN(internal_bb));
1892 bbp[i] = cpu_to_le64(store_bb);
1895 if (read_seqretry(&bb->lock, seq))
1898 bb->sector = (rdev->sb_start +
1899 (int)le32_to_cpu(sb->bblog_offset));
1900 bb->size = le16_to_cpu(sb->bblog_size);
1905 rdev_for_each(rdev2, mddev)
1906 if (rdev2->desc_nr+1 > max_dev)
1907 max_dev = rdev2->desc_nr+1;
1909 if (max_dev > le32_to_cpu(sb->max_dev)) {
1911 sb->max_dev = cpu_to_le32(max_dev);
1912 rdev->sb_size = max_dev * 2 + 256;
1913 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1914 if (rdev->sb_size & bmask)
1915 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1917 max_dev = le32_to_cpu(sb->max_dev);
1919 for (i=0; i<max_dev;i++)
1920 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1922 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1923 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1925 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
1926 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
1928 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
1930 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
1931 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
1932 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
1935 rdev_for_each(rdev2, mddev) {
1937 if (test_bit(Faulty, &rdev2->flags))
1938 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1939 else if (test_bit(In_sync, &rdev2->flags))
1940 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1941 else if (test_bit(Journal, &rdev2->flags))
1942 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1943 else if (rdev2->raid_disk >= 0)
1944 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1946 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1949 sb->sb_csum = calc_sb_1_csum(sb);
1952 static unsigned long long
1953 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1955 struct mdp_superblock_1 *sb;
1956 sector_t max_sectors;
1957 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1958 return 0; /* component must fit device */
1959 if (rdev->data_offset != rdev->new_data_offset)
1960 return 0; /* too confusing */
1961 if (rdev->sb_start < rdev->data_offset) {
1962 /* minor versions 1 and 2; superblock before data */
1963 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1964 max_sectors -= rdev->data_offset;
1965 if (!num_sectors || num_sectors > max_sectors)
1966 num_sectors = max_sectors;
1967 } else if (rdev->mddev->bitmap_info.offset) {
1968 /* minor version 0 with bitmap we can't move */
1971 /* minor version 0; superblock after data */
1973 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1974 sb_start &= ~(sector_t)(4*2 - 1);
1975 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1976 if (!num_sectors || num_sectors > max_sectors)
1977 num_sectors = max_sectors;
1978 rdev->sb_start = sb_start;
1980 sb = page_address(rdev->sb_page);
1981 sb->data_size = cpu_to_le64(num_sectors);
1982 sb->super_offset = cpu_to_le64(rdev->sb_start);
1983 sb->sb_csum = calc_sb_1_csum(sb);
1985 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1987 } while (md_super_wait(rdev->mddev) < 0);
1993 super_1_allow_new_offset(struct md_rdev *rdev,
1994 unsigned long long new_offset)
1996 /* All necessary checks on new >= old have been done */
1997 struct bitmap *bitmap;
1998 if (new_offset >= rdev->data_offset)
2001 /* with 1.0 metadata, there is no metadata to tread on
2002 * so we can always move back */
2003 if (rdev->mddev->minor_version == 0)
2006 /* otherwise we must be sure not to step on
2007 * any metadata, so stay:
2008 * 36K beyond start of superblock
2009 * beyond end of badblocks
2010 * beyond write-intent bitmap
2012 if (rdev->sb_start + (32+4)*2 > new_offset)
2014 bitmap = rdev->mddev->bitmap;
2015 if (bitmap && !rdev->mddev->bitmap_info.file &&
2016 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2017 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2019 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2025 static struct super_type super_types[] = {
2028 .owner = THIS_MODULE,
2029 .load_super = super_90_load,
2030 .validate_super = super_90_validate,
2031 .sync_super = super_90_sync,
2032 .rdev_size_change = super_90_rdev_size_change,
2033 .allow_new_offset = super_90_allow_new_offset,
2037 .owner = THIS_MODULE,
2038 .load_super = super_1_load,
2039 .validate_super = super_1_validate,
2040 .sync_super = super_1_sync,
2041 .rdev_size_change = super_1_rdev_size_change,
2042 .allow_new_offset = super_1_allow_new_offset,
2046 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2048 if (mddev->sync_super) {
2049 mddev->sync_super(mddev, rdev);
2053 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2055 super_types[mddev->major_version].sync_super(mddev, rdev);
2058 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2060 struct md_rdev *rdev, *rdev2;
2063 rdev_for_each_rcu(rdev, mddev1) {
2064 if (test_bit(Faulty, &rdev->flags) ||
2065 test_bit(Journal, &rdev->flags) ||
2066 rdev->raid_disk == -1)
2068 rdev_for_each_rcu(rdev2, mddev2) {
2069 if (test_bit(Faulty, &rdev2->flags) ||
2070 test_bit(Journal, &rdev2->flags) ||
2071 rdev2->raid_disk == -1)
2073 if (rdev->bdev->bd_contains ==
2074 rdev2->bdev->bd_contains) {
2084 static LIST_HEAD(pending_raid_disks);
2087 * Try to register data integrity profile for an mddev
2089 * This is called when an array is started and after a disk has been kicked
2090 * from the array. It only succeeds if all working and active component devices
2091 * are integrity capable with matching profiles.
2093 int md_integrity_register(struct mddev *mddev)
2095 struct md_rdev *rdev, *reference = NULL;
2097 if (list_empty(&mddev->disks))
2098 return 0; /* nothing to do */
2099 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2100 return 0; /* shouldn't register, or already is */
2101 rdev_for_each(rdev, mddev) {
2102 /* skip spares and non-functional disks */
2103 if (test_bit(Faulty, &rdev->flags))
2105 if (rdev->raid_disk < 0)
2108 /* Use the first rdev as the reference */
2112 /* does this rdev's profile match the reference profile? */
2113 if (blk_integrity_compare(reference->bdev->bd_disk,
2114 rdev->bdev->bd_disk) < 0)
2117 if (!reference || !bdev_get_integrity(reference->bdev))
2120 * All component devices are integrity capable and have matching
2121 * profiles, register the common profile for the md device.
2123 blk_integrity_register(mddev->gendisk,
2124 bdev_get_integrity(reference->bdev));
2126 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2127 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2128 pr_err("md: failed to create integrity pool for %s\n",
2134 EXPORT_SYMBOL(md_integrity_register);
2137 * Attempt to add an rdev, but only if it is consistent with the current
2140 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2142 struct blk_integrity *bi_rdev;
2143 struct blk_integrity *bi_mddev;
2144 char name[BDEVNAME_SIZE];
2146 if (!mddev->gendisk)
2149 bi_rdev = bdev_get_integrity(rdev->bdev);
2150 bi_mddev = blk_get_integrity(mddev->gendisk);
2152 if (!bi_mddev) /* nothing to do */
2155 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2156 pr_err("%s: incompatible integrity profile for %s\n",
2157 mdname(mddev), bdevname(rdev->bdev, name));
2163 EXPORT_SYMBOL(md_integrity_add_rdev);
2165 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2167 char b[BDEVNAME_SIZE];
2171 /* prevent duplicates */
2172 if (find_rdev(mddev, rdev->bdev->bd_dev))
2175 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2179 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2180 if (!test_bit(Journal, &rdev->flags) &&
2182 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2184 /* Cannot change size, so fail
2185 * If mddev->level <= 0, then we don't care
2186 * about aligning sizes (e.g. linear)
2188 if (mddev->level > 0)
2191 mddev->dev_sectors = rdev->sectors;
2194 /* Verify rdev->desc_nr is unique.
2195 * If it is -1, assign a free number, else
2196 * check number is not in use
2199 if (rdev->desc_nr < 0) {
2202 choice = mddev->raid_disks;
2203 while (md_find_rdev_nr_rcu(mddev, choice))
2205 rdev->desc_nr = choice;
2207 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2213 if (!test_bit(Journal, &rdev->flags) &&
2214 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2215 pr_warn("md: %s: array is limited to %d devices\n",
2216 mdname(mddev), mddev->max_disks);
2219 bdevname(rdev->bdev,b);
2220 strreplace(b, '/', '!');
2222 rdev->mddev = mddev;
2223 pr_debug("md: bind<%s>\n", b);
2225 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2228 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2229 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2230 /* failure here is OK */;
2231 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2233 list_add_rcu(&rdev->same_set, &mddev->disks);
2234 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2236 /* May as well allow recovery to be retried once */
2237 mddev->recovery_disabled++;
2242 pr_warn("md: failed to register dev-%s for %s\n",
2247 static void md_delayed_delete(struct work_struct *ws)
2249 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2250 kobject_del(&rdev->kobj);
2251 kobject_put(&rdev->kobj);
2254 static void unbind_rdev_from_array(struct md_rdev *rdev)
2256 char b[BDEVNAME_SIZE];
2258 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2259 list_del_rcu(&rdev->same_set);
2260 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2262 sysfs_remove_link(&rdev->kobj, "block");
2263 sysfs_put(rdev->sysfs_state);
2264 rdev->sysfs_state = NULL;
2265 rdev->badblocks.count = 0;
2266 /* We need to delay this, otherwise we can deadlock when
2267 * writing to 'remove' to "dev/state". We also need
2268 * to delay it due to rcu usage.
2271 INIT_WORK(&rdev->del_work, md_delayed_delete);
2272 kobject_get(&rdev->kobj);
2273 queue_work(md_misc_wq, &rdev->del_work);
2277 * prevent the device from being mounted, repartitioned or
2278 * otherwise reused by a RAID array (or any other kernel
2279 * subsystem), by bd_claiming the device.
2281 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2284 struct block_device *bdev;
2285 char b[BDEVNAME_SIZE];
2287 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2288 shared ? (struct md_rdev *)lock_rdev : rdev);
2290 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2291 return PTR_ERR(bdev);
2297 static void unlock_rdev(struct md_rdev *rdev)
2299 struct block_device *bdev = rdev->bdev;
2301 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2304 void md_autodetect_dev(dev_t dev);
2306 static void export_rdev(struct md_rdev *rdev)
2308 char b[BDEVNAME_SIZE];
2310 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2311 md_rdev_clear(rdev);
2313 if (test_bit(AutoDetected, &rdev->flags))
2314 md_autodetect_dev(rdev->bdev->bd_dev);
2317 kobject_put(&rdev->kobj);
2320 void md_kick_rdev_from_array(struct md_rdev *rdev)
2322 unbind_rdev_from_array(rdev);
2325 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2327 static void export_array(struct mddev *mddev)
2329 struct md_rdev *rdev;
2331 while (!list_empty(&mddev->disks)) {
2332 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2334 md_kick_rdev_from_array(rdev);
2336 mddev->raid_disks = 0;
2337 mddev->major_version = 0;
2340 static bool set_in_sync(struct mddev *mddev)
2342 lockdep_assert_held(&mddev->lock);
2343 if (!mddev->in_sync) {
2344 mddev->sync_checkers++;
2345 spin_unlock(&mddev->lock);
2346 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2347 spin_lock(&mddev->lock);
2348 if (!mddev->in_sync &&
2349 percpu_ref_is_zero(&mddev->writes_pending)) {
2352 * Ensure ->in_sync is visible before we clear
2356 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2357 sysfs_notify_dirent_safe(mddev->sysfs_state);
2359 if (--mddev->sync_checkers == 0)
2360 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2362 if (mddev->safemode == 1)
2363 mddev->safemode = 0;
2364 return mddev->in_sync;
2367 static void sync_sbs(struct mddev *mddev, int nospares)
2369 /* Update each superblock (in-memory image), but
2370 * if we are allowed to, skip spares which already
2371 * have the right event counter, or have one earlier
2372 * (which would mean they aren't being marked as dirty
2373 * with the rest of the array)
2375 struct md_rdev *rdev;
2376 rdev_for_each(rdev, mddev) {
2377 if (rdev->sb_events == mddev->events ||
2379 rdev->raid_disk < 0 &&
2380 rdev->sb_events+1 == mddev->events)) {
2381 /* Don't update this superblock */
2382 rdev->sb_loaded = 2;
2384 sync_super(mddev, rdev);
2385 rdev->sb_loaded = 1;
2390 static bool does_sb_need_changing(struct mddev *mddev)
2392 struct md_rdev *rdev;
2393 struct mdp_superblock_1 *sb;
2396 /* Find a good rdev */
2397 rdev_for_each(rdev, mddev)
2398 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2401 /* No good device found. */
2405 sb = page_address(rdev->sb_page);
2406 /* Check if a device has become faulty or a spare become active */
2407 rdev_for_each(rdev, mddev) {
2408 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2409 /* Device activated? */
2410 if (role == 0xffff && rdev->raid_disk >=0 &&
2411 !test_bit(Faulty, &rdev->flags))
2413 /* Device turned faulty? */
2414 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2418 /* Check if any mddev parameters have changed */
2419 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2420 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2421 (mddev->layout != le32_to_cpu(sb->layout)) ||
2422 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2423 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2429 void md_update_sb(struct mddev *mddev, int force_change)
2431 struct md_rdev *rdev;
2434 int any_badblocks_changed = 0;
2439 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2444 if (mddev_is_clustered(mddev)) {
2445 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2447 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2449 ret = md_cluster_ops->metadata_update_start(mddev);
2450 /* Has someone else has updated the sb */
2451 if (!does_sb_need_changing(mddev)) {
2453 md_cluster_ops->metadata_update_cancel(mddev);
2454 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2455 BIT(MD_SB_CHANGE_DEVS) |
2456 BIT(MD_SB_CHANGE_CLEAN));
2462 * First make sure individual recovery_offsets are correct
2463 * curr_resync_completed can only be used during recovery.
2464 * During reshape/resync it might use array-addresses rather
2465 * that device addresses.
2467 rdev_for_each(rdev, mddev) {
2468 if (rdev->raid_disk >= 0 &&
2469 mddev->delta_disks >= 0 &&
2470 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2471 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2472 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2473 !test_bit(Journal, &rdev->flags) &&
2474 !test_bit(In_sync, &rdev->flags) &&
2475 mddev->curr_resync_completed > rdev->recovery_offset)
2476 rdev->recovery_offset = mddev->curr_resync_completed;
2479 if (!mddev->persistent) {
2480 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2481 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2482 if (!mddev->external) {
2483 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2484 rdev_for_each(rdev, mddev) {
2485 if (rdev->badblocks.changed) {
2486 rdev->badblocks.changed = 0;
2487 ack_all_badblocks(&rdev->badblocks);
2488 md_error(mddev, rdev);
2490 clear_bit(Blocked, &rdev->flags);
2491 clear_bit(BlockedBadBlocks, &rdev->flags);
2492 wake_up(&rdev->blocked_wait);
2495 wake_up(&mddev->sb_wait);
2499 spin_lock(&mddev->lock);
2501 mddev->utime = ktime_get_real_seconds();
2503 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2505 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2506 /* just a clean<-> dirty transition, possibly leave spares alone,
2507 * though if events isn't the right even/odd, we will have to do
2513 if (mddev->degraded)
2514 /* If the array is degraded, then skipping spares is both
2515 * dangerous and fairly pointless.
2516 * Dangerous because a device that was removed from the array
2517 * might have a event_count that still looks up-to-date,
2518 * so it can be re-added without a resync.
2519 * Pointless because if there are any spares to skip,
2520 * then a recovery will happen and soon that array won't
2521 * be degraded any more and the spare can go back to sleep then.
2525 sync_req = mddev->in_sync;
2527 /* If this is just a dirty<->clean transition, and the array is clean
2528 * and 'events' is odd, we can roll back to the previous clean state */
2530 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2531 && mddev->can_decrease_events
2532 && mddev->events != 1) {
2534 mddev->can_decrease_events = 0;
2536 /* otherwise we have to go forward and ... */
2538 mddev->can_decrease_events = nospares;
2542 * This 64-bit counter should never wrap.
2543 * Either we are in around ~1 trillion A.C., assuming
2544 * 1 reboot per second, or we have a bug...
2546 WARN_ON(mddev->events == 0);
2548 rdev_for_each(rdev, mddev) {
2549 if (rdev->badblocks.changed)
2550 any_badblocks_changed++;
2551 if (test_bit(Faulty, &rdev->flags))
2552 set_bit(FaultRecorded, &rdev->flags);
2555 sync_sbs(mddev, nospares);
2556 spin_unlock(&mddev->lock);
2558 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2559 mdname(mddev), mddev->in_sync);
2562 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2564 bitmap_update_sb(mddev->bitmap);
2565 rdev_for_each(rdev, mddev) {
2566 char b[BDEVNAME_SIZE];
2568 if (rdev->sb_loaded != 1)
2569 continue; /* no noise on spare devices */
2571 if (!test_bit(Faulty, &rdev->flags)) {
2572 md_super_write(mddev,rdev,
2573 rdev->sb_start, rdev->sb_size,
2575 pr_debug("md: (write) %s's sb offset: %llu\n",
2576 bdevname(rdev->bdev, b),
2577 (unsigned long long)rdev->sb_start);
2578 rdev->sb_events = mddev->events;
2579 if (rdev->badblocks.size) {
2580 md_super_write(mddev, rdev,
2581 rdev->badblocks.sector,
2582 rdev->badblocks.size << 9,
2584 rdev->badblocks.size = 0;
2588 pr_debug("md: %s (skipping faulty)\n",
2589 bdevname(rdev->bdev, b));
2591 if (mddev->level == LEVEL_MULTIPATH)
2592 /* only need to write one superblock... */
2595 if (md_super_wait(mddev) < 0)
2597 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2599 if (mddev_is_clustered(mddev) && ret == 0)
2600 md_cluster_ops->metadata_update_finish(mddev);
2602 if (mddev->in_sync != sync_req ||
2603 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2604 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2605 /* have to write it out again */
2607 wake_up(&mddev->sb_wait);
2608 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2609 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2611 rdev_for_each(rdev, mddev) {
2612 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2613 clear_bit(Blocked, &rdev->flags);
2615 if (any_badblocks_changed)
2616 ack_all_badblocks(&rdev->badblocks);
2617 clear_bit(BlockedBadBlocks, &rdev->flags);
2618 wake_up(&rdev->blocked_wait);
2621 EXPORT_SYMBOL(md_update_sb);
2623 static int add_bound_rdev(struct md_rdev *rdev)
2625 struct mddev *mddev = rdev->mddev;
2627 bool add_journal = test_bit(Journal, &rdev->flags);
2629 if (!mddev->pers->hot_remove_disk || add_journal) {
2630 /* If there is hot_add_disk but no hot_remove_disk
2631 * then added disks for geometry changes,
2632 * and should be added immediately.
2634 super_types[mddev->major_version].
2635 validate_super(mddev, rdev);
2637 mddev_suspend(mddev);
2638 err = mddev->pers->hot_add_disk(mddev, rdev);
2640 mddev_resume(mddev);
2642 md_kick_rdev_from_array(rdev);
2646 sysfs_notify_dirent_safe(rdev->sysfs_state);
2648 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2649 if (mddev->degraded)
2650 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2651 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2652 md_new_event(mddev);
2653 md_wakeup_thread(mddev->thread);
2657 /* words written to sysfs files may, or may not, be \n terminated.
2658 * We want to accept with case. For this we use cmd_match.
2660 static int cmd_match(const char *cmd, const char *str)
2662 /* See if cmd, written into a sysfs file, matches
2663 * str. They must either be the same, or cmd can
2664 * have a trailing newline
2666 while (*cmd && *str && *cmd == *str) {
2677 struct rdev_sysfs_entry {
2678 struct attribute attr;
2679 ssize_t (*show)(struct md_rdev *, char *);
2680 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2684 state_show(struct md_rdev *rdev, char *page)
2688 unsigned long flags = READ_ONCE(rdev->flags);
2690 if (test_bit(Faulty, &flags) ||
2691 (!test_bit(ExternalBbl, &flags) &&
2692 rdev->badblocks.unacked_exist))
2693 len += sprintf(page+len, "faulty%s", sep);
2694 if (test_bit(In_sync, &flags))
2695 len += sprintf(page+len, "in_sync%s", sep);
2696 if (test_bit(Journal, &flags))
2697 len += sprintf(page+len, "journal%s", sep);
2698 if (test_bit(WriteMostly, &flags))
2699 len += sprintf(page+len, "write_mostly%s", sep);
2700 if (test_bit(Blocked, &flags) ||
2701 (rdev->badblocks.unacked_exist
2702 && !test_bit(Faulty, &flags)))
2703 len += sprintf(page+len, "blocked%s", sep);
2704 if (!test_bit(Faulty, &flags) &&
2705 !test_bit(Journal, &flags) &&
2706 !test_bit(In_sync, &flags))
2707 len += sprintf(page+len, "spare%s", sep);
2708 if (test_bit(WriteErrorSeen, &flags))
2709 len += sprintf(page+len, "write_error%s", sep);
2710 if (test_bit(WantReplacement, &flags))
2711 len += sprintf(page+len, "want_replacement%s", sep);
2712 if (test_bit(Replacement, &flags))
2713 len += sprintf(page+len, "replacement%s", sep);
2714 if (test_bit(ExternalBbl, &flags))
2715 len += sprintf(page+len, "external_bbl%s", sep);
2716 if (test_bit(FailFast, &flags))
2717 len += sprintf(page+len, "failfast%s", sep);
2722 return len+sprintf(page+len, "\n");
2726 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2729 * faulty - simulates an error
2730 * remove - disconnects the device
2731 * writemostly - sets write_mostly
2732 * -writemostly - clears write_mostly
2733 * blocked - sets the Blocked flags
2734 * -blocked - clears the Blocked and possibly simulates an error
2735 * insync - sets Insync providing device isn't active
2736 * -insync - clear Insync for a device with a slot assigned,
2737 * so that it gets rebuilt based on bitmap
2738 * write_error - sets WriteErrorSeen
2739 * -write_error - clears WriteErrorSeen
2740 * {,-}failfast - set/clear FailFast
2743 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2744 md_error(rdev->mddev, rdev);
2745 if (test_bit(Faulty, &rdev->flags))
2749 } else if (cmd_match(buf, "remove")) {
2750 if (rdev->mddev->pers) {
2751 clear_bit(Blocked, &rdev->flags);
2752 remove_and_add_spares(rdev->mddev, rdev);
2754 if (rdev->raid_disk >= 0)
2757 struct mddev *mddev = rdev->mddev;
2759 if (mddev_is_clustered(mddev))
2760 err = md_cluster_ops->remove_disk(mddev, rdev);
2763 md_kick_rdev_from_array(rdev);
2765 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2766 md_wakeup_thread(mddev->thread);
2768 md_new_event(mddev);
2771 } else if (cmd_match(buf, "writemostly")) {
2772 set_bit(WriteMostly, &rdev->flags);
2774 } else if (cmd_match(buf, "-writemostly")) {
2775 clear_bit(WriteMostly, &rdev->flags);
2777 } else if (cmd_match(buf, "blocked")) {
2778 set_bit(Blocked, &rdev->flags);
2780 } else if (cmd_match(buf, "-blocked")) {
2781 if (!test_bit(Faulty, &rdev->flags) &&
2782 !test_bit(ExternalBbl, &rdev->flags) &&
2783 rdev->badblocks.unacked_exist) {
2784 /* metadata handler doesn't understand badblocks,
2785 * so we need to fail the device
2787 md_error(rdev->mddev, rdev);
2789 clear_bit(Blocked, &rdev->flags);
2790 clear_bit(BlockedBadBlocks, &rdev->flags);
2791 wake_up(&rdev->blocked_wait);
2792 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2793 md_wakeup_thread(rdev->mddev->thread);
2796 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2797 set_bit(In_sync, &rdev->flags);
2799 } else if (cmd_match(buf, "failfast")) {
2800 set_bit(FailFast, &rdev->flags);
2802 } else if (cmd_match(buf, "-failfast")) {
2803 clear_bit(FailFast, &rdev->flags);
2805 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2806 !test_bit(Journal, &rdev->flags)) {
2807 if (rdev->mddev->pers == NULL) {
2808 clear_bit(In_sync, &rdev->flags);
2809 rdev->saved_raid_disk = rdev->raid_disk;
2810 rdev->raid_disk = -1;
2813 } else if (cmd_match(buf, "write_error")) {
2814 set_bit(WriteErrorSeen, &rdev->flags);
2816 } else if (cmd_match(buf, "-write_error")) {
2817 clear_bit(WriteErrorSeen, &rdev->flags);
2819 } else if (cmd_match(buf, "want_replacement")) {
2820 /* Any non-spare device that is not a replacement can
2821 * become want_replacement at any time, but we then need to
2822 * check if recovery is needed.
2824 if (rdev->raid_disk >= 0 &&
2825 !test_bit(Journal, &rdev->flags) &&
2826 !test_bit(Replacement, &rdev->flags))
2827 set_bit(WantReplacement, &rdev->flags);
2828 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2829 md_wakeup_thread(rdev->mddev->thread);
2831 } else if (cmd_match(buf, "-want_replacement")) {
2832 /* Clearing 'want_replacement' is always allowed.
2833 * Once replacements starts it is too late though.
2836 clear_bit(WantReplacement, &rdev->flags);
2837 } else if (cmd_match(buf, "replacement")) {
2838 /* Can only set a device as a replacement when array has not
2839 * yet been started. Once running, replacement is automatic
2840 * from spares, or by assigning 'slot'.
2842 if (rdev->mddev->pers)
2845 set_bit(Replacement, &rdev->flags);
2848 } else if (cmd_match(buf, "-replacement")) {
2849 /* Similarly, can only clear Replacement before start */
2850 if (rdev->mddev->pers)
2853 clear_bit(Replacement, &rdev->flags);
2856 } else if (cmd_match(buf, "re-add")) {
2857 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2858 /* clear_bit is performed _after_ all the devices
2859 * have their local Faulty bit cleared. If any writes
2860 * happen in the meantime in the local node, they
2861 * will land in the local bitmap, which will be synced
2862 * by this node eventually
2864 if (!mddev_is_clustered(rdev->mddev) ||
2865 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2866 clear_bit(Faulty, &rdev->flags);
2867 err = add_bound_rdev(rdev);
2871 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2872 set_bit(ExternalBbl, &rdev->flags);
2873 rdev->badblocks.shift = 0;
2875 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2876 clear_bit(ExternalBbl, &rdev->flags);
2880 sysfs_notify_dirent_safe(rdev->sysfs_state);
2881 return err ? err : len;
2883 static struct rdev_sysfs_entry rdev_state =
2884 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2887 errors_show(struct md_rdev *rdev, char *page)
2889 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2893 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2898 rv = kstrtouint(buf, 10, &n);
2901 atomic_set(&rdev->corrected_errors, n);
2904 static struct rdev_sysfs_entry rdev_errors =
2905 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2908 slot_show(struct md_rdev *rdev, char *page)
2910 if (test_bit(Journal, &rdev->flags))
2911 return sprintf(page, "journal\n");
2912 else if (rdev->raid_disk < 0)
2913 return sprintf(page, "none\n");
2915 return sprintf(page, "%d\n", rdev->raid_disk);
2919 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2924 if (test_bit(Journal, &rdev->flags))
2926 if (strncmp(buf, "none", 4)==0)
2929 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2933 if (rdev->mddev->pers && slot == -1) {
2934 /* Setting 'slot' on an active array requires also
2935 * updating the 'rd%d' link, and communicating
2936 * with the personality with ->hot_*_disk.
2937 * For now we only support removing
2938 * failed/spare devices. This normally happens automatically,
2939 * but not when the metadata is externally managed.
2941 if (rdev->raid_disk == -1)
2943 /* personality does all needed checks */
2944 if (rdev->mddev->pers->hot_remove_disk == NULL)
2946 clear_bit(Blocked, &rdev->flags);
2947 remove_and_add_spares(rdev->mddev, rdev);
2948 if (rdev->raid_disk >= 0)
2950 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2951 md_wakeup_thread(rdev->mddev->thread);
2952 } else if (rdev->mddev->pers) {
2953 /* Activating a spare .. or possibly reactivating
2954 * if we ever get bitmaps working here.
2958 if (rdev->raid_disk != -1)
2961 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2964 if (rdev->mddev->pers->hot_add_disk == NULL)
2967 if (slot >= rdev->mddev->raid_disks &&
2968 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2971 rdev->raid_disk = slot;
2972 if (test_bit(In_sync, &rdev->flags))
2973 rdev->saved_raid_disk = slot;
2975 rdev->saved_raid_disk = -1;
2976 clear_bit(In_sync, &rdev->flags);
2977 clear_bit(Bitmap_sync, &rdev->flags);
2978 err = rdev->mddev->pers->
2979 hot_add_disk(rdev->mddev, rdev);
2981 rdev->raid_disk = -1;
2984 sysfs_notify_dirent_safe(rdev->sysfs_state);
2985 if (sysfs_link_rdev(rdev->mddev, rdev))
2986 /* failure here is OK */;
2987 /* don't wakeup anyone, leave that to userspace. */
2989 if (slot >= rdev->mddev->raid_disks &&
2990 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2992 rdev->raid_disk = slot;
2993 /* assume it is working */
2994 clear_bit(Faulty, &rdev->flags);
2995 clear_bit(WriteMostly, &rdev->flags);
2996 set_bit(In_sync, &rdev->flags);
2997 sysfs_notify_dirent_safe(rdev->sysfs_state);
3002 static struct rdev_sysfs_entry rdev_slot =
3003 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3006 offset_show(struct md_rdev *rdev, char *page)
3008 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3012 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3014 unsigned long long offset;
3015 if (kstrtoull(buf, 10, &offset) < 0)
3017 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3019 if (rdev->sectors && rdev->mddev->external)
3020 /* Must set offset before size, so overlap checks
3023 rdev->data_offset = offset;
3024 rdev->new_data_offset = offset;
3028 static struct rdev_sysfs_entry rdev_offset =
3029 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3031 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3033 return sprintf(page, "%llu\n",
3034 (unsigned long long)rdev->new_data_offset);
3037 static ssize_t new_offset_store(struct md_rdev *rdev,
3038 const char *buf, size_t len)
3040 unsigned long long new_offset;
3041 struct mddev *mddev = rdev->mddev;
3043 if (kstrtoull(buf, 10, &new_offset) < 0)
3046 if (mddev->sync_thread ||
3047 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3049 if (new_offset == rdev->data_offset)
3050 /* reset is always permitted */
3052 else if (new_offset > rdev->data_offset) {
3053 /* must not push array size beyond rdev_sectors */
3054 if (new_offset - rdev->data_offset
3055 + mddev->dev_sectors > rdev->sectors)
3058 /* Metadata worries about other space details. */
3060 /* decreasing the offset is inconsistent with a backwards
3063 if (new_offset < rdev->data_offset &&
3064 mddev->reshape_backwards)
3066 /* Increasing offset is inconsistent with forwards
3067 * reshape. reshape_direction should be set to
3068 * 'backwards' first.
3070 if (new_offset > rdev->data_offset &&
3071 !mddev->reshape_backwards)
3074 if (mddev->pers && mddev->persistent &&
3075 !super_types[mddev->major_version]
3076 .allow_new_offset(rdev, new_offset))
3078 rdev->new_data_offset = new_offset;
3079 if (new_offset > rdev->data_offset)
3080 mddev->reshape_backwards = 1;
3081 else if (new_offset < rdev->data_offset)
3082 mddev->reshape_backwards = 0;
3086 static struct rdev_sysfs_entry rdev_new_offset =
3087 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3090 rdev_size_show(struct md_rdev *rdev, char *page)
3092 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3095 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3097 /* check if two start/length pairs overlap */
3105 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3107 unsigned long long blocks;
3110 if (kstrtoull(buf, 10, &blocks) < 0)
3113 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3114 return -EINVAL; /* sector conversion overflow */
3117 if (new != blocks * 2)
3118 return -EINVAL; /* unsigned long long to sector_t overflow */
3125 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3127 struct mddev *my_mddev = rdev->mddev;
3128 sector_t oldsectors = rdev->sectors;
3131 if (test_bit(Journal, &rdev->flags))
3133 if (strict_blocks_to_sectors(buf, §ors) < 0)
3135 if (rdev->data_offset != rdev->new_data_offset)
3136 return -EINVAL; /* too confusing */
3137 if (my_mddev->pers && rdev->raid_disk >= 0) {
3138 if (my_mddev->persistent) {
3139 sectors = super_types[my_mddev->major_version].
3140 rdev_size_change(rdev, sectors);
3143 } else if (!sectors)
3144 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3146 if (!my_mddev->pers->resize)
3147 /* Cannot change size for RAID0 or Linear etc */
3150 if (sectors < my_mddev->dev_sectors)
3151 return -EINVAL; /* component must fit device */
3153 rdev->sectors = sectors;
3154 if (sectors > oldsectors && my_mddev->external) {
3155 /* Need to check that all other rdevs with the same
3156 * ->bdev do not overlap. 'rcu' is sufficient to walk
3157 * the rdev lists safely.
3158 * This check does not provide a hard guarantee, it
3159 * just helps avoid dangerous mistakes.
3161 struct mddev *mddev;
3163 struct list_head *tmp;
3166 for_each_mddev(mddev, tmp) {
3167 struct md_rdev *rdev2;
3169 rdev_for_each(rdev2, mddev)
3170 if (rdev->bdev == rdev2->bdev &&
3172 overlaps(rdev->data_offset, rdev->sectors,
3185 /* Someone else could have slipped in a size
3186 * change here, but doing so is just silly.
3187 * We put oldsectors back because we *know* it is
3188 * safe, and trust userspace not to race with
3191 rdev->sectors = oldsectors;
3198 static struct rdev_sysfs_entry rdev_size =
3199 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3201 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3203 unsigned long long recovery_start = rdev->recovery_offset;
3205 if (test_bit(In_sync, &rdev->flags) ||
3206 recovery_start == MaxSector)
3207 return sprintf(page, "none\n");
3209 return sprintf(page, "%llu\n", recovery_start);
3212 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3214 unsigned long long recovery_start;
3216 if (cmd_match(buf, "none"))
3217 recovery_start = MaxSector;
3218 else if (kstrtoull(buf, 10, &recovery_start))
3221 if (rdev->mddev->pers &&
3222 rdev->raid_disk >= 0)
3225 rdev->recovery_offset = recovery_start;
3226 if (recovery_start == MaxSector)
3227 set_bit(In_sync, &rdev->flags);
3229 clear_bit(In_sync, &rdev->flags);
3233 static struct rdev_sysfs_entry rdev_recovery_start =
3234 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3236 /* sysfs access to bad-blocks list.
3237 * We present two files.
3238 * 'bad-blocks' lists sector numbers and lengths of ranges that
3239 * are recorded as bad. The list is truncated to fit within
3240 * the one-page limit of sysfs.
3241 * Writing "sector length" to this file adds an acknowledged
3243 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3244 * been acknowledged. Writing to this file adds bad blocks
3245 * without acknowledging them. This is largely for testing.
3247 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3249 return badblocks_show(&rdev->badblocks, page, 0);
3251 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3253 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3254 /* Maybe that ack was all we needed */
3255 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3256 wake_up(&rdev->blocked_wait);
3259 static struct rdev_sysfs_entry rdev_bad_blocks =
3260 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3262 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3264 return badblocks_show(&rdev->badblocks, page, 1);
3266 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3268 return badblocks_store(&rdev->badblocks, page, len, 1);
3270 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3271 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3274 ppl_sector_show(struct md_rdev *rdev, char *page)
3276 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3280 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3282 unsigned long long sector;
3284 if (kstrtoull(buf, 10, §or) < 0)
3286 if (sector != (sector_t)sector)
3289 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3290 rdev->raid_disk >= 0)
3293 if (rdev->mddev->persistent) {
3294 if (rdev->mddev->major_version == 0)
3296 if ((sector > rdev->sb_start &&
3297 sector - rdev->sb_start > S16_MAX) ||
3298 (sector < rdev->sb_start &&
3299 rdev->sb_start - sector > -S16_MIN))
3301 rdev->ppl.offset = sector - rdev->sb_start;
3302 } else if (!rdev->mddev->external) {
3305 rdev->ppl.sector = sector;
3309 static struct rdev_sysfs_entry rdev_ppl_sector =
3310 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3313 ppl_size_show(struct md_rdev *rdev, char *page)
3315 return sprintf(page, "%u\n", rdev->ppl.size);
3319 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3323 if (kstrtouint(buf, 10, &size) < 0)
3326 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3327 rdev->raid_disk >= 0)
3330 if (rdev->mddev->persistent) {
3331 if (rdev->mddev->major_version == 0)
3335 } else if (!rdev->mddev->external) {
3338 rdev->ppl.size = size;
3342 static struct rdev_sysfs_entry rdev_ppl_size =
3343 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3345 static struct attribute *rdev_default_attrs[] = {
3350 &rdev_new_offset.attr,
3352 &rdev_recovery_start.attr,
3353 &rdev_bad_blocks.attr,
3354 &rdev_unack_bad_blocks.attr,
3355 &rdev_ppl_sector.attr,
3356 &rdev_ppl_size.attr,
3360 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3362 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3363 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3369 return entry->show(rdev, page);
3373 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3374 const char *page, size_t length)
3376 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3377 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3379 struct mddev *mddev = rdev->mddev;
3383 if (!capable(CAP_SYS_ADMIN))
3385 rv = mddev ? mddev_lock(mddev): -EBUSY;
3387 if (rdev->mddev == NULL)
3390 rv = entry->store(rdev, page, length);
3391 mddev_unlock(mddev);
3396 static void rdev_free(struct kobject *ko)
3398 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3401 static const struct sysfs_ops rdev_sysfs_ops = {
3402 .show = rdev_attr_show,
3403 .store = rdev_attr_store,
3405 static struct kobj_type rdev_ktype = {
3406 .release = rdev_free,
3407 .sysfs_ops = &rdev_sysfs_ops,
3408 .default_attrs = rdev_default_attrs,
3411 int md_rdev_init(struct md_rdev *rdev)
3414 rdev->saved_raid_disk = -1;
3415 rdev->raid_disk = -1;
3417 rdev->data_offset = 0;
3418 rdev->new_data_offset = 0;
3419 rdev->sb_events = 0;
3420 rdev->last_read_error = 0;
3421 rdev->sb_loaded = 0;
3422 rdev->bb_page = NULL;
3423 atomic_set(&rdev->nr_pending, 0);
3424 atomic_set(&rdev->read_errors, 0);
3425 atomic_set(&rdev->corrected_errors, 0);
3427 INIT_LIST_HEAD(&rdev->same_set);
3428 init_waitqueue_head(&rdev->blocked_wait);
3430 /* Add space to store bad block list.
3431 * This reserves the space even on arrays where it cannot
3432 * be used - I wonder if that matters
3434 return badblocks_init(&rdev->badblocks, 0);
3436 EXPORT_SYMBOL_GPL(md_rdev_init);
3438 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3440 * mark the device faulty if:
3442 * - the device is nonexistent (zero size)
3443 * - the device has no valid superblock
3445 * a faulty rdev _never_ has rdev->sb set.
3447 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3449 char b[BDEVNAME_SIZE];
3451 struct md_rdev *rdev;
3454 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3456 return ERR_PTR(-ENOMEM);
3458 err = md_rdev_init(rdev);
3461 err = alloc_disk_sb(rdev);
3465 err = lock_rdev(rdev, newdev, super_format == -2);
3469 kobject_init(&rdev->kobj, &rdev_ktype);
3471 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3473 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3474 bdevname(rdev->bdev,b));
3479 if (super_format >= 0) {
3480 err = super_types[super_format].
3481 load_super(rdev, NULL, super_minor);
3482 if (err == -EINVAL) {
3483 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3484 bdevname(rdev->bdev,b),
3485 super_format, super_minor);
3489 pr_warn("md: could not read %s's sb, not importing!\n",
3490 bdevname(rdev->bdev,b));
3500 md_rdev_clear(rdev);
3502 return ERR_PTR(err);
3506 * Check a full RAID array for plausibility
3509 static void analyze_sbs(struct mddev *mddev)
3512 struct md_rdev *rdev, *freshest, *tmp;
3513 char b[BDEVNAME_SIZE];
3516 rdev_for_each_safe(rdev, tmp, mddev)
3517 switch (super_types[mddev->major_version].
3518 load_super(rdev, freshest, mddev->minor_version)) {
3525 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3526 bdevname(rdev->bdev,b));
3527 md_kick_rdev_from_array(rdev);
3530 super_types[mddev->major_version].
3531 validate_super(mddev, freshest);
3534 rdev_for_each_safe(rdev, tmp, mddev) {
3535 if (mddev->max_disks &&
3536 (rdev->desc_nr >= mddev->max_disks ||
3537 i > mddev->max_disks)) {
3538 pr_warn("md: %s: %s: only %d devices permitted\n",
3539 mdname(mddev), bdevname(rdev->bdev, b),
3541 md_kick_rdev_from_array(rdev);
3544 if (rdev != freshest) {
3545 if (super_types[mddev->major_version].
3546 validate_super(mddev, rdev)) {
3547 pr_warn("md: kicking non-fresh %s from array!\n",
3548 bdevname(rdev->bdev,b));
3549 md_kick_rdev_from_array(rdev);
3553 if (mddev->level == LEVEL_MULTIPATH) {
3554 rdev->desc_nr = i++;
3555 rdev->raid_disk = rdev->desc_nr;
3556 set_bit(In_sync, &rdev->flags);
3557 } else if (rdev->raid_disk >=
3558 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3559 !test_bit(Journal, &rdev->flags)) {
3560 rdev->raid_disk = -1;
3561 clear_bit(In_sync, &rdev->flags);
3566 /* Read a fixed-point number.
3567 * Numbers in sysfs attributes should be in "standard" units where
3568 * possible, so time should be in seconds.
3569 * However we internally use a a much smaller unit such as
3570 * milliseconds or jiffies.
3571 * This function takes a decimal number with a possible fractional
3572 * component, and produces an integer which is the result of
3573 * multiplying that number by 10^'scale'.
3574 * all without any floating-point arithmetic.
3576 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3578 unsigned long result = 0;
3580 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3583 else if (decimals < scale) {
3586 result = result * 10 + value;
3598 while (decimals < scale) {
3607 safe_delay_show(struct mddev *mddev, char *page)
3609 int msec = (mddev->safemode_delay*1000)/HZ;
3610 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3613 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3617 if (mddev_is_clustered(mddev)) {
3618 pr_warn("md: Safemode is disabled for clustered mode\n");
3622 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3625 mddev->safemode_delay = 0;
3627 unsigned long old_delay = mddev->safemode_delay;
3628 unsigned long new_delay = (msec*HZ)/1000;
3632 mddev->safemode_delay = new_delay;
3633 if (new_delay < old_delay || old_delay == 0)
3634 mod_timer(&mddev->safemode_timer, jiffies+1);
3638 static struct md_sysfs_entry md_safe_delay =
3639 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3642 level_show(struct mddev *mddev, char *page)
3644 struct md_personality *p;
3646 spin_lock(&mddev->lock);
3649 ret = sprintf(page, "%s\n", p->name);
3650 else if (mddev->clevel[0])
3651 ret = sprintf(page, "%s\n", mddev->clevel);
3652 else if (mddev->level != LEVEL_NONE)
3653 ret = sprintf(page, "%d\n", mddev->level);
3656 spin_unlock(&mddev->lock);
3661 level_store(struct mddev *mddev, const char *buf, size_t len)
3666 struct md_personality *pers, *oldpers;
3668 void *priv, *oldpriv;
3669 struct md_rdev *rdev;
3671 if (slen == 0 || slen >= sizeof(clevel))
3674 rv = mddev_lock(mddev);
3678 if (mddev->pers == NULL) {
3679 strncpy(mddev->clevel, buf, slen);
3680 if (mddev->clevel[slen-1] == '\n')
3682 mddev->clevel[slen] = 0;
3683 mddev->level = LEVEL_NONE;
3691 /* request to change the personality. Need to ensure:
3692 * - array is not engaged in resync/recovery/reshape
3693 * - old personality can be suspended
3694 * - new personality will access other array.
3698 if (mddev->sync_thread ||
3699 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3700 mddev->reshape_position != MaxSector ||
3701 mddev->sysfs_active)
3705 if (!mddev->pers->quiesce) {
3706 pr_warn("md: %s: %s does not support online personality change\n",
3707 mdname(mddev), mddev->pers->name);
3711 /* Now find the new personality */
3712 strncpy(clevel, buf, slen);
3713 if (clevel[slen-1] == '\n')
3716 if (kstrtol(clevel, 10, &level))
3719 if (request_module("md-%s", clevel) != 0)
3720 request_module("md-level-%s", clevel);
3721 spin_lock(&pers_lock);
3722 pers = find_pers(level, clevel);
3723 if (!pers || !try_module_get(pers->owner)) {
3724 spin_unlock(&pers_lock);
3725 pr_warn("md: personality %s not loaded\n", clevel);
3729 spin_unlock(&pers_lock);
3731 if (pers == mddev->pers) {
3732 /* Nothing to do! */
3733 module_put(pers->owner);
3737 if (!pers->takeover) {
3738 module_put(pers->owner);
3739 pr_warn("md: %s: %s does not support personality takeover\n",
3740 mdname(mddev), clevel);
3745 rdev_for_each(rdev, mddev)
3746 rdev->new_raid_disk = rdev->raid_disk;
3748 /* ->takeover must set new_* and/or delta_disks
3749 * if it succeeds, and may set them when it fails.
3751 priv = pers->takeover(mddev);
3753 mddev->new_level = mddev->level;
3754 mddev->new_layout = mddev->layout;
3755 mddev->new_chunk_sectors = mddev->chunk_sectors;
3756 mddev->raid_disks -= mddev->delta_disks;
3757 mddev->delta_disks = 0;
3758 mddev->reshape_backwards = 0;
3759 module_put(pers->owner);
3760 pr_warn("md: %s: %s would not accept array\n",
3761 mdname(mddev), clevel);
3766 /* Looks like we have a winner */
3767 mddev_suspend(mddev);
3768 mddev_detach(mddev);
3770 spin_lock(&mddev->lock);
3771 oldpers = mddev->pers;
3772 oldpriv = mddev->private;
3774 mddev->private = priv;
3775 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3776 mddev->level = mddev->new_level;
3777 mddev->layout = mddev->new_layout;
3778 mddev->chunk_sectors = mddev->new_chunk_sectors;
3779 mddev->delta_disks = 0;
3780 mddev->reshape_backwards = 0;
3781 mddev->degraded = 0;
3782 spin_unlock(&mddev->lock);
3784 if (oldpers->sync_request == NULL &&
3786 /* We are converting from a no-redundancy array
3787 * to a redundancy array and metadata is managed
3788 * externally so we need to be sure that writes
3789 * won't block due to a need to transition
3791 * until external management is started.
3794 mddev->safemode_delay = 0;
3795 mddev->safemode = 0;
3798 oldpers->free(mddev, oldpriv);
3800 if (oldpers->sync_request == NULL &&
3801 pers->sync_request != NULL) {
3802 /* need to add the md_redundancy_group */
3803 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3804 pr_warn("md: cannot register extra attributes for %s\n",
3806 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3808 if (oldpers->sync_request != NULL &&
3809 pers->sync_request == NULL) {
3810 /* need to remove the md_redundancy_group */
3811 if (mddev->to_remove == NULL)
3812 mddev->to_remove = &md_redundancy_group;
3815 module_put(oldpers->owner);
3817 rdev_for_each(rdev, mddev) {
3818 if (rdev->raid_disk < 0)
3820 if (rdev->new_raid_disk >= mddev->raid_disks)
3821 rdev->new_raid_disk = -1;
3822 if (rdev->new_raid_disk == rdev->raid_disk)
3824 sysfs_unlink_rdev(mddev, rdev);
3826 rdev_for_each(rdev, mddev) {
3827 if (rdev->raid_disk < 0)
3829 if (rdev->new_raid_disk == rdev->raid_disk)
3831 rdev->raid_disk = rdev->new_raid_disk;
3832 if (rdev->raid_disk < 0)
3833 clear_bit(In_sync, &rdev->flags);
3835 if (sysfs_link_rdev(mddev, rdev))
3836 pr_warn("md: cannot register rd%d for %s after level change\n",
3837 rdev->raid_disk, mdname(mddev));
3841 if (pers->sync_request == NULL) {
3842 /* this is now an array without redundancy, so
3843 * it must always be in_sync
3846 del_timer_sync(&mddev->safemode_timer);
3848 blk_set_stacking_limits(&mddev->queue->limits);
3850 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3851 mddev_resume(mddev);
3853 md_update_sb(mddev, 1);
3854 sysfs_notify(&mddev->kobj, NULL, "level");
3855 md_new_event(mddev);
3858 mddev_unlock(mddev);
3862 static struct md_sysfs_entry md_level =
3863 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3866 layout_show(struct mddev *mddev, char *page)
3868 /* just a number, not meaningful for all levels */
3869 if (mddev->reshape_position != MaxSector &&
3870 mddev->layout != mddev->new_layout)
3871 return sprintf(page, "%d (%d)\n",
3872 mddev->new_layout, mddev->layout);
3873 return sprintf(page, "%d\n", mddev->layout);
3877 layout_store(struct mddev *mddev, const char *buf, size_t len)
3882 err = kstrtouint(buf, 10, &n);
3885 err = mddev_lock(mddev);
3890 if (mddev->pers->check_reshape == NULL)
3895 mddev->new_layout = n;
3896 err = mddev->pers->check_reshape(mddev);
3898 mddev->new_layout = mddev->layout;
3901 mddev->new_layout = n;
3902 if (mddev->reshape_position == MaxSector)
3905 mddev_unlock(mddev);
3908 static struct md_sysfs_entry md_layout =
3909 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3912 raid_disks_show(struct mddev *mddev, char *page)
3914 if (mddev->raid_disks == 0)
3916 if (mddev->reshape_position != MaxSector &&
3917 mddev->delta_disks != 0)
3918 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3919 mddev->raid_disks - mddev->delta_disks);
3920 return sprintf(page, "%d\n", mddev->raid_disks);
3923 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3926 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3931 err = kstrtouint(buf, 10, &n);
3935 err = mddev_lock(mddev);
3939 err = update_raid_disks(mddev, n);
3940 else if (mddev->reshape_position != MaxSector) {
3941 struct md_rdev *rdev;
3942 int olddisks = mddev->raid_disks - mddev->delta_disks;
3945 rdev_for_each(rdev, mddev) {
3947 rdev->data_offset < rdev->new_data_offset)
3950 rdev->data_offset > rdev->new_data_offset)
3954 mddev->delta_disks = n - olddisks;
3955 mddev->raid_disks = n;
3956 mddev->reshape_backwards = (mddev->delta_disks < 0);
3958 mddev->raid_disks = n;
3960 mddev_unlock(mddev);
3961 return err ? err : len;
3963 static struct md_sysfs_entry md_raid_disks =
3964 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3967 chunk_size_show(struct mddev *mddev, char *page)
3969 if (mddev->reshape_position != MaxSector &&
3970 mddev->chunk_sectors != mddev->new_chunk_sectors)
3971 return sprintf(page, "%d (%d)\n",
3972 mddev->new_chunk_sectors << 9,
3973 mddev->chunk_sectors << 9);
3974 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3978 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3983 err = kstrtoul(buf, 10, &n);
3987 err = mddev_lock(mddev);
3991 if (mddev->pers->check_reshape == NULL)
3996 mddev->new_chunk_sectors = n >> 9;
3997 err = mddev->pers->check_reshape(mddev);
3999 mddev->new_chunk_sectors = mddev->chunk_sectors;
4002 mddev->new_chunk_sectors = n >> 9;
4003 if (mddev->reshape_position == MaxSector)
4004 mddev->chunk_sectors = n >> 9;
4006 mddev_unlock(mddev);
4009 static struct md_sysfs_entry md_chunk_size =
4010 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4013 resync_start_show(struct mddev *mddev, char *page)
4015 if (mddev->recovery_cp == MaxSector)
4016 return sprintf(page, "none\n");
4017 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4021 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4023 unsigned long long n;
4026 if (cmd_match(buf, "none"))
4029 err = kstrtoull(buf, 10, &n);
4032 if (n != (sector_t)n)
4036 err = mddev_lock(mddev);
4039 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4043 mddev->recovery_cp = n;
4045 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4047 mddev_unlock(mddev);
4050 static struct md_sysfs_entry md_resync_start =
4051 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4052 resync_start_show, resync_start_store);
4055 * The array state can be:
4058 * No devices, no size, no level
4059 * Equivalent to STOP_ARRAY ioctl
4061 * May have some settings, but array is not active
4062 * all IO results in error
4063 * When written, doesn't tear down array, but just stops it
4064 * suspended (not supported yet)
4065 * All IO requests will block. The array can be reconfigured.
4066 * Writing this, if accepted, will block until array is quiescent
4068 * no resync can happen. no superblocks get written.
4069 * write requests fail
4071 * like readonly, but behaves like 'clean' on a write request.
4073 * clean - no pending writes, but otherwise active.
4074 * When written to inactive array, starts without resync
4075 * If a write request arrives then
4076 * if metadata is known, mark 'dirty' and switch to 'active'.
4077 * if not known, block and switch to write-pending
4078 * If written to an active array that has pending writes, then fails.
4080 * fully active: IO and resync can be happening.
4081 * When written to inactive array, starts with resync
4084 * clean, but writes are blocked waiting for 'active' to be written.
4087 * like active, but no writes have been seen for a while (100msec).
4090 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4091 write_pending, active_idle, bad_word};
4092 static char *array_states[] = {
4093 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4094 "write-pending", "active-idle", NULL };
4096 static int match_word(const char *word, char **list)
4099 for (n=0; list[n]; n++)
4100 if (cmd_match(word, list[n]))
4106 array_state_show(struct mddev *mddev, char *page)
4108 enum array_state st = inactive;
4119 spin_lock(&mddev->lock);
4120 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4122 else if (mddev->in_sync)
4124 else if (mddev->safemode)
4128 spin_unlock(&mddev->lock);
4131 if (list_empty(&mddev->disks) &&
4132 mddev->raid_disks == 0 &&
4133 mddev->dev_sectors == 0)
4138 return sprintf(page, "%s\n", array_states[st]);
4141 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4142 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4143 static int do_md_run(struct mddev *mddev);
4144 static int restart_array(struct mddev *mddev);
4147 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4150 enum array_state st = match_word(buf, array_states);
4152 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4153 /* don't take reconfig_mutex when toggling between
4156 spin_lock(&mddev->lock);
4158 restart_array(mddev);
4159 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4160 md_wakeup_thread(mddev->thread);
4161 wake_up(&mddev->sb_wait);
4162 } else /* st == clean */ {
4163 restart_array(mddev);
4164 if (!set_in_sync(mddev))
4168 sysfs_notify_dirent_safe(mddev->sysfs_state);
4169 spin_unlock(&mddev->lock);
4172 err = mddev_lock(mddev);
4180 /* stopping an active array */
4181 err = do_md_stop(mddev, 0, NULL);
4184 /* stopping an active array */
4186 err = do_md_stop(mddev, 2, NULL);
4188 err = 0; /* already inactive */
4191 break; /* not supported yet */
4194 err = md_set_readonly(mddev, NULL);
4197 set_disk_ro(mddev->gendisk, 1);
4198 err = do_md_run(mddev);
4204 err = md_set_readonly(mddev, NULL);
4205 else if (mddev->ro == 1)
4206 err = restart_array(mddev);
4209 set_disk_ro(mddev->gendisk, 0);
4213 err = do_md_run(mddev);
4218 err = restart_array(mddev);
4221 spin_lock(&mddev->lock);
4222 if (!set_in_sync(mddev))
4224 spin_unlock(&mddev->lock);
4230 err = restart_array(mddev);
4233 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4234 wake_up(&mddev->sb_wait);
4238 set_disk_ro(mddev->gendisk, 0);
4239 err = do_md_run(mddev);
4244 /* these cannot be set */
4249 if (mddev->hold_active == UNTIL_IOCTL)
4250 mddev->hold_active = 0;
4251 sysfs_notify_dirent_safe(mddev->sysfs_state);
4253 mddev_unlock(mddev);
4256 static struct md_sysfs_entry md_array_state =
4257 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4260 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4261 return sprintf(page, "%d\n",
4262 atomic_read(&mddev->max_corr_read_errors));
4266 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4271 rv = kstrtouint(buf, 10, &n);
4274 atomic_set(&mddev->max_corr_read_errors, n);
4278 static struct md_sysfs_entry max_corr_read_errors =
4279 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4280 max_corrected_read_errors_store);
4283 null_show(struct mddev *mddev, char *page)
4289 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4291 /* buf must be %d:%d\n? giving major and minor numbers */
4292 /* The new device is added to the array.
4293 * If the array has a persistent superblock, we read the
4294 * superblock to initialise info and check validity.
4295 * Otherwise, only checking done is that in bind_rdev_to_array,
4296 * which mainly checks size.
4299 int major = simple_strtoul(buf, &e, 10);
4302 struct md_rdev *rdev;
4305 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4307 minor = simple_strtoul(e+1, &e, 10);
4308 if (*e && *e != '\n')
4310 dev = MKDEV(major, minor);
4311 if (major != MAJOR(dev) ||
4312 minor != MINOR(dev))
4315 flush_workqueue(md_misc_wq);
4317 err = mddev_lock(mddev);
4320 if (mddev->persistent) {
4321 rdev = md_import_device(dev, mddev->major_version,
4322 mddev->minor_version);
4323 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4324 struct md_rdev *rdev0
4325 = list_entry(mddev->disks.next,
4326 struct md_rdev, same_set);
4327 err = super_types[mddev->major_version]
4328 .load_super(rdev, rdev0, mddev->minor_version);
4332 } else if (mddev->external)
4333 rdev = md_import_device(dev, -2, -1);
4335 rdev = md_import_device(dev, -1, -1);
4338 mddev_unlock(mddev);
4339 return PTR_ERR(rdev);
4341 err = bind_rdev_to_array(rdev, mddev);
4345 mddev_unlock(mddev);
4347 md_new_event(mddev);
4348 return err ? err : len;
4351 static struct md_sysfs_entry md_new_device =
4352 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4355 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4358 unsigned long chunk, end_chunk;
4361 err = mddev_lock(mddev);
4366 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4368 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4369 if (buf == end) break;
4370 if (*end == '-') { /* range */
4372 end_chunk = simple_strtoul(buf, &end, 0);
4373 if (buf == end) break;
4375 if (*end && !isspace(*end)) break;
4376 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4377 buf = skip_spaces(end);
4379 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4381 mddev_unlock(mddev);
4385 static struct md_sysfs_entry md_bitmap =
4386 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4389 size_show(struct mddev *mddev, char *page)
4391 return sprintf(page, "%llu\n",
4392 (unsigned long long)mddev->dev_sectors / 2);
4395 static int update_size(struct mddev *mddev, sector_t num_sectors);
4398 size_store(struct mddev *mddev, const char *buf, size_t len)
4400 /* If array is inactive, we can reduce the component size, but
4401 * not increase it (except from 0).
4402 * If array is active, we can try an on-line resize
4405 int err = strict_blocks_to_sectors(buf, §ors);
4409 err = mddev_lock(mddev);
4413 err = update_size(mddev, sectors);
4415 md_update_sb(mddev, 1);
4417 if (mddev->dev_sectors == 0 ||
4418 mddev->dev_sectors > sectors)
4419 mddev->dev_sectors = sectors;
4423 mddev_unlock(mddev);
4424 return err ? err : len;
4427 static struct md_sysfs_entry md_size =
4428 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4430 /* Metadata version.
4432 * 'none' for arrays with no metadata (good luck...)
4433 * 'external' for arrays with externally managed metadata,
4434 * or N.M for internally known formats
4437 metadata_show(struct mddev *mddev, char *page)
4439 if (mddev->persistent)
4440 return sprintf(page, "%d.%d\n",
4441 mddev->major_version, mddev->minor_version);
4442 else if (mddev->external)
4443 return sprintf(page, "external:%s\n", mddev->metadata_type);
4445 return sprintf(page, "none\n");
4449 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4454 /* Changing the details of 'external' metadata is
4455 * always permitted. Otherwise there must be
4456 * no devices attached to the array.
4459 err = mddev_lock(mddev);
4463 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4465 else if (!list_empty(&mddev->disks))
4469 if (cmd_match(buf, "none")) {
4470 mddev->persistent = 0;
4471 mddev->external = 0;
4472 mddev->major_version = 0;
4473 mddev->minor_version = 90;
4476 if (strncmp(buf, "external:", 9) == 0) {
4477 size_t namelen = len-9;
4478 if (namelen >= sizeof(mddev->metadata_type))
4479 namelen = sizeof(mddev->metadata_type)-1;
4480 strncpy(mddev->metadata_type, buf+9, namelen);
4481 mddev->metadata_type[namelen] = 0;
4482 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4483 mddev->metadata_type[--namelen] = 0;
4484 mddev->persistent = 0;
4485 mddev->external = 1;
4486 mddev->major_version = 0;
4487 mddev->minor_version = 90;
4490 major = simple_strtoul(buf, &e, 10);
4492 if (e==buf || *e != '.')
4495 minor = simple_strtoul(buf, &e, 10);
4496 if (e==buf || (*e && *e != '\n') )
4499 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4501 mddev->major_version = major;
4502 mddev->minor_version = minor;
4503 mddev->persistent = 1;
4504 mddev->external = 0;
4507 mddev_unlock(mddev);
4511 static struct md_sysfs_entry md_metadata =
4512 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4515 action_show(struct mddev *mddev, char *page)
4517 char *type = "idle";
4518 unsigned long recovery = mddev->recovery;
4519 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4521 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4522 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4523 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4525 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4526 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4528 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4532 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4534 else if (mddev->reshape_position != MaxSector)
4537 return sprintf(page, "%s\n", type);
4541 action_store(struct mddev *mddev, const char *page, size_t len)
4543 if (!mddev->pers || !mddev->pers->sync_request)
4547 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4548 if (cmd_match(page, "frozen"))
4549 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4551 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4552 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4553 mddev_lock(mddev) == 0) {
4554 flush_workqueue(md_misc_wq);
4555 if (mddev->sync_thread) {
4556 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4557 md_reap_sync_thread(mddev);
4559 mddev_unlock(mddev);
4561 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4563 else if (cmd_match(page, "resync"))
4564 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4565 else if (cmd_match(page, "recover")) {
4566 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4567 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4568 } else if (cmd_match(page, "reshape")) {
4570 if (mddev->pers->start_reshape == NULL)
4572 err = mddev_lock(mddev);
4574 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4577 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4578 err = mddev->pers->start_reshape(mddev);
4580 mddev_unlock(mddev);
4584 sysfs_notify(&mddev->kobj, NULL, "degraded");
4586 if (cmd_match(page, "check"))
4587 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4588 else if (!cmd_match(page, "repair"))
4590 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4591 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4592 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4594 if (mddev->ro == 2) {
4595 /* A write to sync_action is enough to justify
4596 * canceling read-auto mode
4599 md_wakeup_thread(mddev->sync_thread);
4601 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4602 md_wakeup_thread(mddev->thread);
4603 sysfs_notify_dirent_safe(mddev->sysfs_action);
4607 static struct md_sysfs_entry md_scan_mode =
4608 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4611 last_sync_action_show(struct mddev *mddev, char *page)
4613 return sprintf(page, "%s\n", mddev->last_sync_action);
4616 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4619 mismatch_cnt_show(struct mddev *mddev, char *page)
4621 return sprintf(page, "%llu\n",
4622 (unsigned long long)
4623 atomic64_read(&mddev->resync_mismatches));
4626 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4629 sync_min_show(struct mddev *mddev, char *page)
4631 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4632 mddev->sync_speed_min ? "local": "system");
4636 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4641 if (strncmp(buf, "system", 6)==0) {
4644 rv = kstrtouint(buf, 10, &min);
4650 mddev->sync_speed_min = min;
4654 static struct md_sysfs_entry md_sync_min =
4655 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4658 sync_max_show(struct mddev *mddev, char *page)
4660 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4661 mddev->sync_speed_max ? "local": "system");
4665 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4670 if (strncmp(buf, "system", 6)==0) {
4673 rv = kstrtouint(buf, 10, &max);
4679 mddev->sync_speed_max = max;
4683 static struct md_sysfs_entry md_sync_max =
4684 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4687 degraded_show(struct mddev *mddev, char *page)
4689 return sprintf(page, "%d\n", mddev->degraded);
4691 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4694 sync_force_parallel_show(struct mddev *mddev, char *page)
4696 return sprintf(page, "%d\n", mddev->parallel_resync);
4700 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4704 if (kstrtol(buf, 10, &n))
4707 if (n != 0 && n != 1)
4710 mddev->parallel_resync = n;
4712 if (mddev->sync_thread)
4713 wake_up(&resync_wait);
4718 /* force parallel resync, even with shared block devices */
4719 static struct md_sysfs_entry md_sync_force_parallel =
4720 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4721 sync_force_parallel_show, sync_force_parallel_store);
4724 sync_speed_show(struct mddev *mddev, char *page)
4726 unsigned long resync, dt, db;
4727 if (mddev->curr_resync == 0)
4728 return sprintf(page, "none\n");
4729 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4730 dt = (jiffies - mddev->resync_mark) / HZ;
4732 db = resync - mddev->resync_mark_cnt;
4733 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4736 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4739 sync_completed_show(struct mddev *mddev, char *page)
4741 unsigned long long max_sectors, resync;
4743 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4744 return sprintf(page, "none\n");
4746 if (mddev->curr_resync == 1 ||
4747 mddev->curr_resync == 2)
4748 return sprintf(page, "delayed\n");
4750 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4751 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4752 max_sectors = mddev->resync_max_sectors;
4754 max_sectors = mddev->dev_sectors;
4756 resync = mddev->curr_resync_completed;
4757 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4760 static struct md_sysfs_entry md_sync_completed =
4761 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4764 min_sync_show(struct mddev *mddev, char *page)
4766 return sprintf(page, "%llu\n",
4767 (unsigned long long)mddev->resync_min);
4770 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4772 unsigned long long min;
4775 if (kstrtoull(buf, 10, &min))
4778 spin_lock(&mddev->lock);
4780 if (min > mddev->resync_max)
4784 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4787 /* Round down to multiple of 4K for safety */
4788 mddev->resync_min = round_down(min, 8);
4792 spin_unlock(&mddev->lock);
4796 static struct md_sysfs_entry md_min_sync =
4797 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4800 max_sync_show(struct mddev *mddev, char *page)
4802 if (mddev->resync_max == MaxSector)
4803 return sprintf(page, "max\n");
4805 return sprintf(page, "%llu\n",
4806 (unsigned long long)mddev->resync_max);
4809 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4812 spin_lock(&mddev->lock);
4813 if (strncmp(buf, "max", 3) == 0)
4814 mddev->resync_max = MaxSector;
4816 unsigned long long max;
4820 if (kstrtoull(buf, 10, &max))
4822 if (max < mddev->resync_min)
4826 if (max < mddev->resync_max &&
4828 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4831 /* Must be a multiple of chunk_size */
4832 chunk = mddev->chunk_sectors;
4834 sector_t temp = max;
4837 if (sector_div(temp, chunk))
4840 mddev->resync_max = max;
4842 wake_up(&mddev->recovery_wait);
4845 spin_unlock(&mddev->lock);
4849 static struct md_sysfs_entry md_max_sync =
4850 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4853 suspend_lo_show(struct mddev *mddev, char *page)
4855 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4859 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4861 unsigned long long new;
4864 err = kstrtoull(buf, 10, &new);
4867 if (new != (sector_t)new)
4870 err = mddev_lock(mddev);
4874 if (mddev->pers == NULL ||
4875 mddev->pers->quiesce == NULL)
4877 mddev_suspend(mddev);
4878 mddev->suspend_lo = new;
4879 mddev_resume(mddev);
4883 mddev_unlock(mddev);
4886 static struct md_sysfs_entry md_suspend_lo =
4887 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4890 suspend_hi_show(struct mddev *mddev, char *page)
4892 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4896 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4898 unsigned long long new;
4901 err = kstrtoull(buf, 10, &new);
4904 if (new != (sector_t)new)
4907 err = mddev_lock(mddev);
4911 if (mddev->pers == NULL)
4914 mddev_suspend(mddev);
4915 mddev->suspend_hi = new;
4916 mddev_resume(mddev);
4920 mddev_unlock(mddev);
4923 static struct md_sysfs_entry md_suspend_hi =
4924 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4927 reshape_position_show(struct mddev *mddev, char *page)
4929 if (mddev->reshape_position != MaxSector)
4930 return sprintf(page, "%llu\n",
4931 (unsigned long long)mddev->reshape_position);
4932 strcpy(page, "none\n");
4937 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4939 struct md_rdev *rdev;
4940 unsigned long long new;
4943 err = kstrtoull(buf, 10, &new);
4946 if (new != (sector_t)new)
4948 err = mddev_lock(mddev);
4954 mddev->reshape_position = new;
4955 mddev->delta_disks = 0;
4956 mddev->reshape_backwards = 0;
4957 mddev->new_level = mddev->level;
4958 mddev->new_layout = mddev->layout;
4959 mddev->new_chunk_sectors = mddev->chunk_sectors;
4960 rdev_for_each(rdev, mddev)
4961 rdev->new_data_offset = rdev->data_offset;
4964 mddev_unlock(mddev);
4968 static struct md_sysfs_entry md_reshape_position =
4969 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4970 reshape_position_store);
4973 reshape_direction_show(struct mddev *mddev, char *page)
4975 return sprintf(page, "%s\n",
4976 mddev->reshape_backwards ? "backwards" : "forwards");
4980 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4985 if (cmd_match(buf, "forwards"))
4987 else if (cmd_match(buf, "backwards"))
4991 if (mddev->reshape_backwards == backwards)
4994 err = mddev_lock(mddev);
4997 /* check if we are allowed to change */
4998 if (mddev->delta_disks)
5000 else if (mddev->persistent &&
5001 mddev->major_version == 0)
5004 mddev->reshape_backwards = backwards;
5005 mddev_unlock(mddev);
5009 static struct md_sysfs_entry md_reshape_direction =
5010 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5011 reshape_direction_store);
5014 array_size_show(struct mddev *mddev, char *page)
5016 if (mddev->external_size)
5017 return sprintf(page, "%llu\n",
5018 (unsigned long long)mddev->array_sectors/2);
5020 return sprintf(page, "default\n");
5024 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5029 err = mddev_lock(mddev);
5033 /* cluster raid doesn't support change array_sectors */
5034 if (mddev_is_clustered(mddev)) {
5035 mddev_unlock(mddev);
5039 if (strncmp(buf, "default", 7) == 0) {
5041 sectors = mddev->pers->size(mddev, 0, 0);
5043 sectors = mddev->array_sectors;
5045 mddev->external_size = 0;
5047 if (strict_blocks_to_sectors(buf, §ors) < 0)
5049 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5052 mddev->external_size = 1;
5056 mddev->array_sectors = sectors;
5058 set_capacity(mddev->gendisk, mddev->array_sectors);
5059 revalidate_disk(mddev->gendisk);
5062 mddev_unlock(mddev);
5066 static struct md_sysfs_entry md_array_size =
5067 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5071 consistency_policy_show(struct mddev *mddev, char *page)
5075 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5076 ret = sprintf(page, "journal\n");
5077 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5078 ret = sprintf(page, "ppl\n");
5079 } else if (mddev->bitmap) {
5080 ret = sprintf(page, "bitmap\n");
5081 } else if (mddev->pers) {
5082 if (mddev->pers->sync_request)
5083 ret = sprintf(page, "resync\n");
5085 ret = sprintf(page, "none\n");
5087 ret = sprintf(page, "unknown\n");
5094 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5099 if (mddev->pers->change_consistency_policy)
5100 err = mddev->pers->change_consistency_policy(mddev, buf);
5103 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5104 set_bit(MD_HAS_PPL, &mddev->flags);
5109 return err ? err : len;
5112 static struct md_sysfs_entry md_consistency_policy =
5113 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5114 consistency_policy_store);
5116 static struct attribute *md_default_attrs[] = {
5119 &md_raid_disks.attr,
5120 &md_chunk_size.attr,
5122 &md_resync_start.attr,
5124 &md_new_device.attr,
5125 &md_safe_delay.attr,
5126 &md_array_state.attr,
5127 &md_reshape_position.attr,
5128 &md_reshape_direction.attr,
5129 &md_array_size.attr,
5130 &max_corr_read_errors.attr,
5131 &md_consistency_policy.attr,
5135 static struct attribute *md_redundancy_attrs[] = {
5137 &md_last_scan_mode.attr,
5138 &md_mismatches.attr,
5141 &md_sync_speed.attr,
5142 &md_sync_force_parallel.attr,
5143 &md_sync_completed.attr,
5146 &md_suspend_lo.attr,
5147 &md_suspend_hi.attr,
5152 static struct attribute_group md_redundancy_group = {
5154 .attrs = md_redundancy_attrs,
5158 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5160 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5161 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5166 spin_lock(&all_mddevs_lock);
5167 if (list_empty(&mddev->all_mddevs)) {
5168 spin_unlock(&all_mddevs_lock);
5172 spin_unlock(&all_mddevs_lock);
5174 rv = entry->show(mddev, page);
5180 md_attr_store(struct kobject *kobj, struct attribute *attr,
5181 const char *page, size_t length)
5183 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5184 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5189 if (!capable(CAP_SYS_ADMIN))
5191 spin_lock(&all_mddevs_lock);
5192 if (list_empty(&mddev->all_mddevs)) {
5193 spin_unlock(&all_mddevs_lock);
5197 spin_unlock(&all_mddevs_lock);
5198 rv = entry->store(mddev, page, length);
5203 static void md_free(struct kobject *ko)
5205 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5207 if (mddev->sysfs_state)
5208 sysfs_put(mddev->sysfs_state);
5211 del_gendisk(mddev->gendisk);
5213 blk_cleanup_queue(mddev->queue);
5215 put_disk(mddev->gendisk);
5216 percpu_ref_exit(&mddev->writes_pending);
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_init_and_add(&mddev->kobj, &md_ktype,
5352 &disk_to_dev(disk)->kobj, "%s", "md");
5354 /* This isn't possible, but as kobject_init_and_add is marked
5355 * __must_check, we must do something with the result
5357 pr_debug("md: cannot register %s/md - name in use\n",
5361 if (mddev->kobj.sd &&
5362 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5363 pr_debug("pointless warning\n");
5364 mutex_unlock(&mddev->open_mutex);
5366 mutex_unlock(&disks_mutex);
5367 if (!error && mddev->kobj.sd) {
5368 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5369 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5375 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5378 md_alloc(dev, NULL);
5382 static int add_named_array(const char *val, const struct kernel_param *kp)
5385 * val must be "md_*" or "mdNNN".
5386 * For "md_*" we allocate an array with a large free minor number, and
5387 * set the name to val. val must not already be an active name.
5388 * For "mdNNN" we allocate an array with the minor number NNN
5389 * which must not already be in use.
5391 int len = strlen(val);
5392 char buf[DISK_NAME_LEN];
5393 unsigned long devnum;
5395 while (len && val[len-1] == '\n')
5397 if (len >= DISK_NAME_LEN)
5399 strlcpy(buf, val, len+1);
5400 if (strncmp(buf, "md_", 3) == 0)
5401 return md_alloc(0, buf);
5402 if (strncmp(buf, "md", 2) == 0 &&
5404 kstrtoul(buf+2, 10, &devnum) == 0 &&
5405 devnum <= MINORMASK)
5406 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5411 static void md_safemode_timeout(struct timer_list *t)
5413 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5415 mddev->safemode = 1;
5416 if (mddev->external)
5417 sysfs_notify_dirent_safe(mddev->sysfs_state);
5419 md_wakeup_thread(mddev->thread);
5422 static int start_dirty_degraded;
5424 int md_run(struct mddev *mddev)
5427 struct md_rdev *rdev;
5428 struct md_personality *pers;
5430 if (list_empty(&mddev->disks))
5431 /* cannot run an array with no devices.. */
5436 /* Cannot run until previous stop completes properly */
5437 if (mddev->sysfs_active)
5441 * Analyze all RAID superblock(s)
5443 if (!mddev->raid_disks) {
5444 if (!mddev->persistent)
5449 if (mddev->level != LEVEL_NONE)
5450 request_module("md-level-%d", mddev->level);
5451 else if (mddev->clevel[0])
5452 request_module("md-%s", mddev->clevel);
5455 * Drop all container device buffers, from now on
5456 * the only valid external interface is through the md
5459 mddev->has_superblocks = false;
5460 rdev_for_each(rdev, mddev) {
5461 if (test_bit(Faulty, &rdev->flags))
5463 sync_blockdev(rdev->bdev);
5464 invalidate_bdev(rdev->bdev);
5465 if (mddev->ro != 1 &&
5466 (bdev_read_only(rdev->bdev) ||
5467 bdev_read_only(rdev->meta_bdev))) {
5470 set_disk_ro(mddev->gendisk, 1);
5474 mddev->has_superblocks = true;
5476 /* perform some consistency tests on the device.
5477 * We don't want the data to overlap the metadata,
5478 * Internal Bitmap issues have been handled elsewhere.
5480 if (rdev->meta_bdev) {
5481 /* Nothing to check */;
5482 } else if (rdev->data_offset < rdev->sb_start) {
5483 if (mddev->dev_sectors &&
5484 rdev->data_offset + mddev->dev_sectors
5486 pr_warn("md: %s: data overlaps metadata\n",
5491 if (rdev->sb_start + rdev->sb_size/512
5492 > rdev->data_offset) {
5493 pr_warn("md: %s: metadata overlaps data\n",
5498 sysfs_notify_dirent_safe(rdev->sysfs_state);
5501 if (!bioset_initialized(&mddev->bio_set)) {
5502 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5506 if (!bioset_initialized(&mddev->sync_set)) {
5507 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5512 spin_lock(&pers_lock);
5513 pers = find_pers(mddev->level, mddev->clevel);
5514 if (!pers || !try_module_get(pers->owner)) {
5515 spin_unlock(&pers_lock);
5516 if (mddev->level != LEVEL_NONE)
5517 pr_warn("md: personality for level %d is not loaded!\n",
5520 pr_warn("md: personality for level %s is not loaded!\n",
5525 spin_unlock(&pers_lock);
5526 if (mddev->level != pers->level) {
5527 mddev->level = pers->level;
5528 mddev->new_level = pers->level;
5530 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5532 if (mddev->reshape_position != MaxSector &&
5533 pers->start_reshape == NULL) {
5534 /* This personality cannot handle reshaping... */
5535 module_put(pers->owner);
5540 if (pers->sync_request) {
5541 /* Warn if this is a potentially silly
5544 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5545 struct md_rdev *rdev2;
5548 rdev_for_each(rdev, mddev)
5549 rdev_for_each(rdev2, mddev) {
5551 rdev->bdev->bd_contains ==
5552 rdev2->bdev->bd_contains) {
5553 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5555 bdevname(rdev->bdev,b),
5556 bdevname(rdev2->bdev,b2));
5562 pr_warn("True protection against single-disk failure might be compromised.\n");
5565 mddev->recovery = 0;
5566 /* may be over-ridden by personality */
5567 mddev->resync_max_sectors = mddev->dev_sectors;
5569 mddev->ok_start_degraded = start_dirty_degraded;
5571 if (start_readonly && mddev->ro == 0)
5572 mddev->ro = 2; /* read-only, but switch on first write */
5574 err = pers->run(mddev);
5576 pr_warn("md: pers->run() failed ...\n");
5577 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5578 WARN_ONCE(!mddev->external_size,
5579 "%s: default size too small, but 'external_size' not in effect?\n",
5581 pr_warn("md: invalid array_size %llu > default size %llu\n",
5582 (unsigned long long)mddev->array_sectors / 2,
5583 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5586 if (err == 0 && pers->sync_request &&
5587 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5588 struct bitmap *bitmap;
5590 bitmap = bitmap_create(mddev, -1);
5591 if (IS_ERR(bitmap)) {
5592 err = PTR_ERR(bitmap);
5593 pr_warn("%s: failed to create bitmap (%d)\n",
5594 mdname(mddev), err);
5596 mddev->bitmap = bitmap;
5600 mddev_detach(mddev);
5602 pers->free(mddev, mddev->private);
5603 mddev->private = NULL;
5604 module_put(pers->owner);
5605 bitmap_destroy(mddev);
5611 rdev_for_each(rdev, mddev) {
5612 if (rdev->raid_disk >= 0 &&
5613 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5618 if (mddev->degraded)
5621 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5623 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5624 mddev->queue->backing_dev_info->congested_data = mddev;
5625 mddev->queue->backing_dev_info->congested_fn = md_congested;
5627 if (pers->sync_request) {
5628 if (mddev->kobj.sd &&
5629 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5630 pr_warn("md: cannot register extra attributes for %s\n",
5632 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5633 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5636 atomic_set(&mddev->max_corr_read_errors,
5637 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5638 mddev->safemode = 0;
5639 if (mddev_is_clustered(mddev))
5640 mddev->safemode_delay = 0;
5642 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5645 spin_lock(&mddev->lock);
5647 spin_unlock(&mddev->lock);
5648 rdev_for_each(rdev, mddev)
5649 if (rdev->raid_disk >= 0)
5650 if (sysfs_link_rdev(mddev, rdev))
5651 /* failure here is OK */;
5653 if (mddev->degraded && !mddev->ro)
5654 /* This ensures that recovering status is reported immediately
5655 * via sysfs - until a lack of spares is confirmed.
5657 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5658 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5660 if (mddev->sb_flags)
5661 md_update_sb(mddev, 0);
5663 md_new_event(mddev);
5664 sysfs_notify_dirent_safe(mddev->sysfs_state);
5665 sysfs_notify_dirent_safe(mddev->sysfs_action);
5666 sysfs_notify(&mddev->kobj, NULL, "degraded");
5670 bioset_exit(&mddev->bio_set);
5671 bioset_exit(&mddev->sync_set);
5675 EXPORT_SYMBOL_GPL(md_run);
5677 static int do_md_run(struct mddev *mddev)
5681 err = md_run(mddev);
5684 err = bitmap_load(mddev);
5686 bitmap_destroy(mddev);
5690 if (mddev_is_clustered(mddev))
5691 md_allow_write(mddev);
5693 /* run start up tasks that require md_thread */
5696 md_wakeup_thread(mddev->thread);
5697 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5699 set_capacity(mddev->gendisk, mddev->array_sectors);
5700 revalidate_disk(mddev->gendisk);
5702 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5707 int md_start(struct mddev *mddev)
5711 if (mddev->pers->start) {
5712 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5713 md_wakeup_thread(mddev->thread);
5714 ret = mddev->pers->start(mddev);
5715 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5716 md_wakeup_thread(mddev->sync_thread);
5720 EXPORT_SYMBOL_GPL(md_start);
5722 static int restart_array(struct mddev *mddev)
5724 struct gendisk *disk = mddev->gendisk;
5725 struct md_rdev *rdev;
5726 bool has_journal = false;
5727 bool has_readonly = false;
5729 /* Complain if it has no devices */
5730 if (list_empty(&mddev->disks))
5738 rdev_for_each_rcu(rdev, mddev) {
5739 if (test_bit(Journal, &rdev->flags) &&
5740 !test_bit(Faulty, &rdev->flags))
5742 if (bdev_read_only(rdev->bdev))
5743 has_readonly = true;
5746 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5747 /* Don't restart rw with journal missing/faulty */
5752 mddev->safemode = 0;
5754 set_disk_ro(disk, 0);
5755 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5756 /* Kick recovery or resync if necessary */
5757 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5758 md_wakeup_thread(mddev->thread);
5759 md_wakeup_thread(mddev->sync_thread);
5760 sysfs_notify_dirent_safe(mddev->sysfs_state);
5764 static void md_clean(struct mddev *mddev)
5766 mddev->array_sectors = 0;
5767 mddev->external_size = 0;
5768 mddev->dev_sectors = 0;
5769 mddev->raid_disks = 0;
5770 mddev->recovery_cp = 0;
5771 mddev->resync_min = 0;
5772 mddev->resync_max = MaxSector;
5773 mddev->reshape_position = MaxSector;
5774 mddev->external = 0;
5775 mddev->persistent = 0;
5776 mddev->level = LEVEL_NONE;
5777 mddev->clevel[0] = 0;
5779 mddev->sb_flags = 0;
5781 mddev->metadata_type[0] = 0;
5782 mddev->chunk_sectors = 0;
5783 mddev->ctime = mddev->utime = 0;
5785 mddev->max_disks = 0;
5787 mddev->can_decrease_events = 0;
5788 mddev->delta_disks = 0;
5789 mddev->reshape_backwards = 0;
5790 mddev->new_level = LEVEL_NONE;
5791 mddev->new_layout = 0;
5792 mddev->new_chunk_sectors = 0;
5793 mddev->curr_resync = 0;
5794 atomic64_set(&mddev->resync_mismatches, 0);
5795 mddev->suspend_lo = mddev->suspend_hi = 0;
5796 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5797 mddev->recovery = 0;
5800 mddev->degraded = 0;
5801 mddev->safemode = 0;
5802 mddev->private = NULL;
5803 mddev->cluster_info = NULL;
5804 mddev->bitmap_info.offset = 0;
5805 mddev->bitmap_info.default_offset = 0;
5806 mddev->bitmap_info.default_space = 0;
5807 mddev->bitmap_info.chunksize = 0;
5808 mddev->bitmap_info.daemon_sleep = 0;
5809 mddev->bitmap_info.max_write_behind = 0;
5810 mddev->bitmap_info.nodes = 0;
5813 static void __md_stop_writes(struct mddev *mddev)
5815 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5816 flush_workqueue(md_misc_wq);
5817 if (mddev->sync_thread) {
5818 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5819 md_reap_sync_thread(mddev);
5822 del_timer_sync(&mddev->safemode_timer);
5824 if (mddev->pers && mddev->pers->quiesce) {
5825 mddev->pers->quiesce(mddev, 1);
5826 mddev->pers->quiesce(mddev, 0);
5828 bitmap_flush(mddev);
5830 if (mddev->ro == 0 &&
5831 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5833 /* mark array as shutdown cleanly */
5834 if (!mddev_is_clustered(mddev))
5836 md_update_sb(mddev, 1);
5840 void md_stop_writes(struct mddev *mddev)
5842 mddev_lock_nointr(mddev);
5843 __md_stop_writes(mddev);
5844 mddev_unlock(mddev);
5846 EXPORT_SYMBOL_GPL(md_stop_writes);
5848 static void mddev_detach(struct mddev *mddev)
5850 bitmap_wait_behind_writes(mddev);
5851 if (mddev->pers && mddev->pers->quiesce) {
5852 mddev->pers->quiesce(mddev, 1);
5853 mddev->pers->quiesce(mddev, 0);
5855 md_unregister_thread(&mddev->thread);
5857 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5860 static void __md_stop(struct mddev *mddev)
5862 struct md_personality *pers = mddev->pers;
5863 bitmap_destroy(mddev);
5864 mddev_detach(mddev);
5865 /* Ensure ->event_work is done */
5866 flush_workqueue(md_misc_wq);
5867 spin_lock(&mddev->lock);
5869 spin_unlock(&mddev->lock);
5870 pers->free(mddev, mddev->private);
5871 mddev->private = NULL;
5872 if (pers->sync_request && mddev->to_remove == NULL)
5873 mddev->to_remove = &md_redundancy_group;
5874 module_put(pers->owner);
5875 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5878 void md_stop(struct mddev *mddev)
5880 /* stop the array and free an attached data structures.
5881 * This is called from dm-raid
5884 bioset_exit(&mddev->bio_set);
5885 bioset_exit(&mddev->sync_set);
5888 EXPORT_SYMBOL_GPL(md_stop);
5890 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5895 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5897 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5898 md_wakeup_thread(mddev->thread);
5900 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5901 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5902 if (mddev->sync_thread)
5903 /* Thread might be blocked waiting for metadata update
5904 * which will now never happen */
5905 wake_up_process(mddev->sync_thread->tsk);
5907 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5909 mddev_unlock(mddev);
5910 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5912 wait_event(mddev->sb_wait,
5913 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5914 mddev_lock_nointr(mddev);
5916 mutex_lock(&mddev->open_mutex);
5917 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5918 mddev->sync_thread ||
5919 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5920 pr_warn("md: %s still in use.\n",mdname(mddev));
5922 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5923 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5924 md_wakeup_thread(mddev->thread);
5930 __md_stop_writes(mddev);
5936 set_disk_ro(mddev->gendisk, 1);
5937 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5938 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5939 md_wakeup_thread(mddev->thread);
5940 sysfs_notify_dirent_safe(mddev->sysfs_state);
5944 mutex_unlock(&mddev->open_mutex);
5949 * 0 - completely stop and dis-assemble array
5950 * 2 - stop but do not disassemble array
5952 static int do_md_stop(struct mddev *mddev, int mode,
5953 struct block_device *bdev)
5955 struct gendisk *disk = mddev->gendisk;
5956 struct md_rdev *rdev;
5959 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5961 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5962 md_wakeup_thread(mddev->thread);
5964 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5965 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5966 if (mddev->sync_thread)
5967 /* Thread might be blocked waiting for metadata update
5968 * which will now never happen */
5969 wake_up_process(mddev->sync_thread->tsk);
5971 mddev_unlock(mddev);
5972 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5973 !test_bit(MD_RECOVERY_RUNNING,
5974 &mddev->recovery)));
5975 mddev_lock_nointr(mddev);
5977 mutex_lock(&mddev->open_mutex);
5978 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5979 mddev->sysfs_active ||
5980 mddev->sync_thread ||
5981 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5982 pr_warn("md: %s still in use.\n",mdname(mddev));
5983 mutex_unlock(&mddev->open_mutex);
5985 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5986 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5987 md_wakeup_thread(mddev->thread);
5993 set_disk_ro(disk, 0);
5995 __md_stop_writes(mddev);
5997 mddev->queue->backing_dev_info->congested_fn = NULL;
5999 /* tell userspace to handle 'inactive' */
6000 sysfs_notify_dirent_safe(mddev->sysfs_state);
6002 rdev_for_each(rdev, mddev)
6003 if (rdev->raid_disk >= 0)
6004 sysfs_unlink_rdev(mddev, rdev);
6006 set_capacity(disk, 0);
6007 mutex_unlock(&mddev->open_mutex);
6009 revalidate_disk(disk);
6014 mutex_unlock(&mddev->open_mutex);
6016 * Free resources if final stop
6019 pr_info("md: %s stopped.\n", mdname(mddev));
6021 if (mddev->bitmap_info.file) {
6022 struct file *f = mddev->bitmap_info.file;
6023 spin_lock(&mddev->lock);
6024 mddev->bitmap_info.file = NULL;
6025 spin_unlock(&mddev->lock);
6028 mddev->bitmap_info.offset = 0;
6030 export_array(mddev);
6033 if (mddev->hold_active == UNTIL_STOP)
6034 mddev->hold_active = 0;
6036 md_new_event(mddev);
6037 sysfs_notify_dirent_safe(mddev->sysfs_state);
6042 static void autorun_array(struct mddev *mddev)
6044 struct md_rdev *rdev;
6047 if (list_empty(&mddev->disks))
6050 pr_info("md: running: ");
6052 rdev_for_each(rdev, mddev) {
6053 char b[BDEVNAME_SIZE];
6054 pr_cont("<%s>", bdevname(rdev->bdev,b));
6058 err = do_md_run(mddev);
6060 pr_warn("md: do_md_run() returned %d\n", err);
6061 do_md_stop(mddev, 0, NULL);
6066 * lets try to run arrays based on all disks that have arrived
6067 * until now. (those are in pending_raid_disks)
6069 * the method: pick the first pending disk, collect all disks with
6070 * the same UUID, remove all from the pending list and put them into
6071 * the 'same_array' list. Then order this list based on superblock
6072 * update time (freshest comes first), kick out 'old' disks and
6073 * compare superblocks. If everything's fine then run it.
6075 * If "unit" is allocated, then bump its reference count
6077 static void autorun_devices(int part)
6079 struct md_rdev *rdev0, *rdev, *tmp;
6080 struct mddev *mddev;
6081 char b[BDEVNAME_SIZE];
6083 pr_info("md: autorun ...\n");
6084 while (!list_empty(&pending_raid_disks)) {
6087 LIST_HEAD(candidates);
6088 rdev0 = list_entry(pending_raid_disks.next,
6089 struct md_rdev, same_set);
6091 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6092 INIT_LIST_HEAD(&candidates);
6093 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6094 if (super_90_load(rdev, rdev0, 0) >= 0) {
6095 pr_debug("md: adding %s ...\n",
6096 bdevname(rdev->bdev,b));
6097 list_move(&rdev->same_set, &candidates);
6100 * now we have a set of devices, with all of them having
6101 * mostly sane superblocks. It's time to allocate the
6105 dev = MKDEV(mdp_major,
6106 rdev0->preferred_minor << MdpMinorShift);
6107 unit = MINOR(dev) >> MdpMinorShift;
6109 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6112 if (rdev0->preferred_minor != unit) {
6113 pr_warn("md: unit number in %s is bad: %d\n",
6114 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6118 md_probe(dev, NULL, NULL);
6119 mddev = mddev_find(dev);
6120 if (!mddev || !mddev->gendisk) {
6125 if (mddev_lock(mddev))
6126 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6127 else if (mddev->raid_disks || mddev->major_version
6128 || !list_empty(&mddev->disks)) {
6129 pr_warn("md: %s already running, cannot run %s\n",
6130 mdname(mddev), bdevname(rdev0->bdev,b));
6131 mddev_unlock(mddev);
6133 pr_debug("md: created %s\n", mdname(mddev));
6134 mddev->persistent = 1;
6135 rdev_for_each_list(rdev, tmp, &candidates) {
6136 list_del_init(&rdev->same_set);
6137 if (bind_rdev_to_array(rdev, mddev))
6140 autorun_array(mddev);
6141 mddev_unlock(mddev);
6143 /* on success, candidates will be empty, on error
6146 rdev_for_each_list(rdev, tmp, &candidates) {
6147 list_del_init(&rdev->same_set);
6152 pr_info("md: ... autorun DONE.\n");
6154 #endif /* !MODULE */
6156 static int get_version(void __user *arg)
6160 ver.major = MD_MAJOR_VERSION;
6161 ver.minor = MD_MINOR_VERSION;
6162 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6164 if (copy_to_user(arg, &ver, sizeof(ver)))
6170 static int get_array_info(struct mddev *mddev, void __user *arg)
6172 mdu_array_info_t info;
6173 int nr,working,insync,failed,spare;
6174 struct md_rdev *rdev;
6176 nr = working = insync = failed = spare = 0;
6178 rdev_for_each_rcu(rdev, mddev) {
6180 if (test_bit(Faulty, &rdev->flags))
6184 if (test_bit(In_sync, &rdev->flags))
6186 else if (test_bit(Journal, &rdev->flags))
6187 /* TODO: add journal count to md_u.h */
6195 info.major_version = mddev->major_version;
6196 info.minor_version = mddev->minor_version;
6197 info.patch_version = MD_PATCHLEVEL_VERSION;
6198 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6199 info.level = mddev->level;
6200 info.size = mddev->dev_sectors / 2;
6201 if (info.size != mddev->dev_sectors / 2) /* overflow */
6204 info.raid_disks = mddev->raid_disks;
6205 info.md_minor = mddev->md_minor;
6206 info.not_persistent= !mddev->persistent;
6208 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6211 info.state = (1<<MD_SB_CLEAN);
6212 if (mddev->bitmap && mddev->bitmap_info.offset)
6213 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6214 if (mddev_is_clustered(mddev))
6215 info.state |= (1<<MD_SB_CLUSTERED);
6216 info.active_disks = insync;
6217 info.working_disks = working;
6218 info.failed_disks = failed;
6219 info.spare_disks = spare;
6221 info.layout = mddev->layout;
6222 info.chunk_size = mddev->chunk_sectors << 9;
6224 if (copy_to_user(arg, &info, sizeof(info)))
6230 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6232 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6236 file = kzalloc(sizeof(*file), GFP_NOIO);
6241 spin_lock(&mddev->lock);
6242 /* bitmap enabled */
6243 if (mddev->bitmap_info.file) {
6244 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6245 sizeof(file->pathname));
6249 memmove(file->pathname, ptr,
6250 sizeof(file->pathname)-(ptr-file->pathname));
6252 spin_unlock(&mddev->lock);
6255 copy_to_user(arg, file, sizeof(*file)))
6262 static int get_disk_info(struct mddev *mddev, void __user * arg)
6264 mdu_disk_info_t info;
6265 struct md_rdev *rdev;
6267 if (copy_from_user(&info, arg, sizeof(info)))
6271 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6273 info.major = MAJOR(rdev->bdev->bd_dev);
6274 info.minor = MINOR(rdev->bdev->bd_dev);
6275 info.raid_disk = rdev->raid_disk;
6277 if (test_bit(Faulty, &rdev->flags))
6278 info.state |= (1<<MD_DISK_FAULTY);
6279 else if (test_bit(In_sync, &rdev->flags)) {
6280 info.state |= (1<<MD_DISK_ACTIVE);
6281 info.state |= (1<<MD_DISK_SYNC);
6283 if (test_bit(Journal, &rdev->flags))
6284 info.state |= (1<<MD_DISK_JOURNAL);
6285 if (test_bit(WriteMostly, &rdev->flags))
6286 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6287 if (test_bit(FailFast, &rdev->flags))
6288 info.state |= (1<<MD_DISK_FAILFAST);
6290 info.major = info.minor = 0;
6291 info.raid_disk = -1;
6292 info.state = (1<<MD_DISK_REMOVED);
6296 if (copy_to_user(arg, &info, sizeof(info)))
6302 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6304 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6305 struct md_rdev *rdev;
6306 dev_t dev = MKDEV(info->major,info->minor);
6308 if (mddev_is_clustered(mddev) &&
6309 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6310 pr_warn("%s: Cannot add to clustered mddev.\n",
6315 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6318 if (!mddev->raid_disks) {
6320 /* expecting a device which has a superblock */
6321 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6323 pr_warn("md: md_import_device returned %ld\n",
6325 return PTR_ERR(rdev);
6327 if (!list_empty(&mddev->disks)) {
6328 struct md_rdev *rdev0
6329 = list_entry(mddev->disks.next,
6330 struct md_rdev, same_set);
6331 err = super_types[mddev->major_version]
6332 .load_super(rdev, rdev0, mddev->minor_version);
6334 pr_warn("md: %s has different UUID to %s\n",
6335 bdevname(rdev->bdev,b),
6336 bdevname(rdev0->bdev,b2));
6341 err = bind_rdev_to_array(rdev, mddev);
6348 * add_new_disk can be used once the array is assembled
6349 * to add "hot spares". They must already have a superblock
6354 if (!mddev->pers->hot_add_disk) {
6355 pr_warn("%s: personality does not support diskops!\n",
6359 if (mddev->persistent)
6360 rdev = md_import_device(dev, mddev->major_version,
6361 mddev->minor_version);
6363 rdev = md_import_device(dev, -1, -1);
6365 pr_warn("md: md_import_device returned %ld\n",
6367 return PTR_ERR(rdev);
6369 /* set saved_raid_disk if appropriate */
6370 if (!mddev->persistent) {
6371 if (info->state & (1<<MD_DISK_SYNC) &&
6372 info->raid_disk < mddev->raid_disks) {
6373 rdev->raid_disk = info->raid_disk;
6374 set_bit(In_sync, &rdev->flags);
6375 clear_bit(Bitmap_sync, &rdev->flags);
6377 rdev->raid_disk = -1;
6378 rdev->saved_raid_disk = rdev->raid_disk;
6380 super_types[mddev->major_version].
6381 validate_super(mddev, rdev);
6382 if ((info->state & (1<<MD_DISK_SYNC)) &&
6383 rdev->raid_disk != info->raid_disk) {
6384 /* This was a hot-add request, but events doesn't
6385 * match, so reject it.
6391 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6392 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6393 set_bit(WriteMostly, &rdev->flags);
6395 clear_bit(WriteMostly, &rdev->flags);
6396 if (info->state & (1<<MD_DISK_FAILFAST))
6397 set_bit(FailFast, &rdev->flags);
6399 clear_bit(FailFast, &rdev->flags);
6401 if (info->state & (1<<MD_DISK_JOURNAL)) {
6402 struct md_rdev *rdev2;
6403 bool has_journal = false;
6405 /* make sure no existing journal disk */
6406 rdev_for_each(rdev2, mddev) {
6407 if (test_bit(Journal, &rdev2->flags)) {
6412 if (has_journal || mddev->bitmap) {
6416 set_bit(Journal, &rdev->flags);
6419 * check whether the device shows up in other nodes
6421 if (mddev_is_clustered(mddev)) {
6422 if (info->state & (1 << MD_DISK_CANDIDATE))
6423 set_bit(Candidate, &rdev->flags);
6424 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6425 /* --add initiated by this node */
6426 err = md_cluster_ops->add_new_disk(mddev, rdev);
6434 rdev->raid_disk = -1;
6435 err = bind_rdev_to_array(rdev, mddev);
6440 if (mddev_is_clustered(mddev)) {
6441 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6443 err = md_cluster_ops->new_disk_ack(mddev,
6446 md_kick_rdev_from_array(rdev);
6450 md_cluster_ops->add_new_disk_cancel(mddev);
6452 err = add_bound_rdev(rdev);
6456 err = add_bound_rdev(rdev);
6461 /* otherwise, add_new_disk is only allowed
6462 * for major_version==0 superblocks
6464 if (mddev->major_version != 0) {
6465 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6469 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6471 rdev = md_import_device(dev, -1, 0);
6473 pr_warn("md: error, md_import_device() returned %ld\n",
6475 return PTR_ERR(rdev);
6477 rdev->desc_nr = info->number;
6478 if (info->raid_disk < mddev->raid_disks)
6479 rdev->raid_disk = info->raid_disk;
6481 rdev->raid_disk = -1;
6483 if (rdev->raid_disk < mddev->raid_disks)
6484 if (info->state & (1<<MD_DISK_SYNC))
6485 set_bit(In_sync, &rdev->flags);
6487 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6488 set_bit(WriteMostly, &rdev->flags);
6489 if (info->state & (1<<MD_DISK_FAILFAST))
6490 set_bit(FailFast, &rdev->flags);
6492 if (!mddev->persistent) {
6493 pr_debug("md: nonpersistent superblock ...\n");
6494 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6496 rdev->sb_start = calc_dev_sboffset(rdev);
6497 rdev->sectors = rdev->sb_start;
6499 err = bind_rdev_to_array(rdev, mddev);
6509 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6511 char b[BDEVNAME_SIZE];
6512 struct md_rdev *rdev;
6514 rdev = find_rdev(mddev, dev);
6518 if (rdev->raid_disk < 0)
6521 clear_bit(Blocked, &rdev->flags);
6522 remove_and_add_spares(mddev, rdev);
6524 if (rdev->raid_disk >= 0)
6528 if (mddev_is_clustered(mddev))
6529 md_cluster_ops->remove_disk(mddev, rdev);
6531 md_kick_rdev_from_array(rdev);
6532 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6534 md_wakeup_thread(mddev->thread);
6536 md_update_sb(mddev, 1);
6537 md_new_event(mddev);
6541 pr_debug("md: cannot remove active disk %s from %s ...\n",
6542 bdevname(rdev->bdev,b), mdname(mddev));
6546 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6548 char b[BDEVNAME_SIZE];
6550 struct md_rdev *rdev;
6555 if (mddev->major_version != 0) {
6556 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6560 if (!mddev->pers->hot_add_disk) {
6561 pr_warn("%s: personality does not support diskops!\n",
6566 rdev = md_import_device(dev, -1, 0);
6568 pr_warn("md: error, md_import_device() returned %ld\n",
6573 if (mddev->persistent)
6574 rdev->sb_start = calc_dev_sboffset(rdev);
6576 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6578 rdev->sectors = rdev->sb_start;
6580 if (test_bit(Faulty, &rdev->flags)) {
6581 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6582 bdevname(rdev->bdev,b), mdname(mddev));
6587 clear_bit(In_sync, &rdev->flags);
6589 rdev->saved_raid_disk = -1;
6590 err = bind_rdev_to_array(rdev, mddev);
6595 * The rest should better be atomic, we can have disk failures
6596 * noticed in interrupt contexts ...
6599 rdev->raid_disk = -1;
6601 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6603 md_update_sb(mddev, 1);
6605 * Kick recovery, maybe this spare has to be added to the
6606 * array immediately.
6608 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6609 md_wakeup_thread(mddev->thread);
6610 md_new_event(mddev);
6618 static int set_bitmap_file(struct mddev *mddev, int fd)
6623 if (!mddev->pers->quiesce || !mddev->thread)
6625 if (mddev->recovery || mddev->sync_thread)
6627 /* we should be able to change the bitmap.. */
6631 struct inode *inode;
6634 if (mddev->bitmap || mddev->bitmap_info.file)
6635 return -EEXIST; /* cannot add when bitmap is present */
6639 pr_warn("%s: error: failed to get bitmap file\n",
6644 inode = f->f_mapping->host;
6645 if (!S_ISREG(inode->i_mode)) {
6646 pr_warn("%s: error: bitmap file must be a regular file\n",
6649 } else if (!(f->f_mode & FMODE_WRITE)) {
6650 pr_warn("%s: error: bitmap file must open for write\n",
6653 } else if (atomic_read(&inode->i_writecount) != 1) {
6654 pr_warn("%s: error: bitmap file is already in use\n",
6662 mddev->bitmap_info.file = f;
6663 mddev->bitmap_info.offset = 0; /* file overrides offset */
6664 } else if (mddev->bitmap == NULL)
6665 return -ENOENT; /* cannot remove what isn't there */
6669 struct bitmap *bitmap;
6671 bitmap = bitmap_create(mddev, -1);
6672 mddev_suspend(mddev);
6673 if (!IS_ERR(bitmap)) {
6674 mddev->bitmap = bitmap;
6675 err = bitmap_load(mddev);
6677 err = PTR_ERR(bitmap);
6679 bitmap_destroy(mddev);
6682 mddev_resume(mddev);
6683 } else if (fd < 0) {
6684 mddev_suspend(mddev);
6685 bitmap_destroy(mddev);
6686 mddev_resume(mddev);
6690 struct file *f = mddev->bitmap_info.file;
6692 spin_lock(&mddev->lock);
6693 mddev->bitmap_info.file = NULL;
6694 spin_unlock(&mddev->lock);
6703 * set_array_info is used two different ways
6704 * The original usage is when creating a new array.
6705 * In this usage, raid_disks is > 0 and it together with
6706 * level, size, not_persistent,layout,chunksize determine the
6707 * shape of the array.
6708 * This will always create an array with a type-0.90.0 superblock.
6709 * The newer usage is when assembling an array.
6710 * In this case raid_disks will be 0, and the major_version field is
6711 * use to determine which style super-blocks are to be found on the devices.
6712 * The minor and patch _version numbers are also kept incase the
6713 * super_block handler wishes to interpret them.
6715 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6718 if (info->raid_disks == 0) {
6719 /* just setting version number for superblock loading */
6720 if (info->major_version < 0 ||
6721 info->major_version >= ARRAY_SIZE(super_types) ||
6722 super_types[info->major_version].name == NULL) {
6723 /* maybe try to auto-load a module? */
6724 pr_warn("md: superblock version %d not known\n",
6725 info->major_version);
6728 mddev->major_version = info->major_version;
6729 mddev->minor_version = info->minor_version;
6730 mddev->patch_version = info->patch_version;
6731 mddev->persistent = !info->not_persistent;
6732 /* ensure mddev_put doesn't delete this now that there
6733 * is some minimal configuration.
6735 mddev->ctime = ktime_get_real_seconds();
6738 mddev->major_version = MD_MAJOR_VERSION;
6739 mddev->minor_version = MD_MINOR_VERSION;
6740 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6741 mddev->ctime = ktime_get_real_seconds();
6743 mddev->level = info->level;
6744 mddev->clevel[0] = 0;
6745 mddev->dev_sectors = 2 * (sector_t)info->size;
6746 mddev->raid_disks = info->raid_disks;
6747 /* don't set md_minor, it is determined by which /dev/md* was
6750 if (info->state & (1<<MD_SB_CLEAN))
6751 mddev->recovery_cp = MaxSector;
6753 mddev->recovery_cp = 0;
6754 mddev->persistent = ! info->not_persistent;
6755 mddev->external = 0;
6757 mddev->layout = info->layout;
6758 mddev->chunk_sectors = info->chunk_size >> 9;
6760 if (mddev->persistent) {
6761 mddev->max_disks = MD_SB_DISKS;
6763 mddev->sb_flags = 0;
6765 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6767 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6768 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6769 mddev->bitmap_info.offset = 0;
6771 mddev->reshape_position = MaxSector;
6774 * Generate a 128 bit UUID
6776 get_random_bytes(mddev->uuid, 16);
6778 mddev->new_level = mddev->level;
6779 mddev->new_chunk_sectors = mddev->chunk_sectors;
6780 mddev->new_layout = mddev->layout;
6781 mddev->delta_disks = 0;
6782 mddev->reshape_backwards = 0;
6787 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6789 lockdep_assert_held(&mddev->reconfig_mutex);
6791 if (mddev->external_size)
6794 mddev->array_sectors = array_sectors;
6796 EXPORT_SYMBOL(md_set_array_sectors);
6798 static int update_size(struct mddev *mddev, sector_t num_sectors)
6800 struct md_rdev *rdev;
6802 int fit = (num_sectors == 0);
6803 sector_t old_dev_sectors = mddev->dev_sectors;
6805 if (mddev->pers->resize == NULL)
6807 /* The "num_sectors" is the number of sectors of each device that
6808 * is used. This can only make sense for arrays with redundancy.
6809 * linear and raid0 always use whatever space is available. We can only
6810 * consider changing this number if no resync or reconstruction is
6811 * happening, and if the new size is acceptable. It must fit before the
6812 * sb_start or, if that is <data_offset, it must fit before the size
6813 * of each device. If num_sectors is zero, we find the largest size
6816 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6822 rdev_for_each(rdev, mddev) {
6823 sector_t avail = rdev->sectors;
6825 if (fit && (num_sectors == 0 || num_sectors > avail))
6826 num_sectors = avail;
6827 if (avail < num_sectors)
6830 rv = mddev->pers->resize(mddev, num_sectors);
6832 if (mddev_is_clustered(mddev))
6833 md_cluster_ops->update_size(mddev, old_dev_sectors);
6834 else if (mddev->queue) {
6835 set_capacity(mddev->gendisk, mddev->array_sectors);
6836 revalidate_disk(mddev->gendisk);
6842 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6845 struct md_rdev *rdev;
6846 /* change the number of raid disks */
6847 if (mddev->pers->check_reshape == NULL)
6851 if (raid_disks <= 0 ||
6852 (mddev->max_disks && raid_disks >= mddev->max_disks))
6854 if (mddev->sync_thread ||
6855 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6856 mddev->reshape_position != MaxSector)
6859 rdev_for_each(rdev, mddev) {
6860 if (mddev->raid_disks < raid_disks &&
6861 rdev->data_offset < rdev->new_data_offset)
6863 if (mddev->raid_disks > raid_disks &&
6864 rdev->data_offset > rdev->new_data_offset)
6868 mddev->delta_disks = raid_disks - mddev->raid_disks;
6869 if (mddev->delta_disks < 0)
6870 mddev->reshape_backwards = 1;
6871 else if (mddev->delta_disks > 0)
6872 mddev->reshape_backwards = 0;
6874 rv = mddev->pers->check_reshape(mddev);
6876 mddev->delta_disks = 0;
6877 mddev->reshape_backwards = 0;
6883 * update_array_info is used to change the configuration of an
6885 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6886 * fields in the info are checked against the array.
6887 * Any differences that cannot be handled will cause an error.
6888 * Normally, only one change can be managed at a time.
6890 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6896 /* calculate expected state,ignoring low bits */
6897 if (mddev->bitmap && mddev->bitmap_info.offset)
6898 state |= (1 << MD_SB_BITMAP_PRESENT);
6900 if (mddev->major_version != info->major_version ||
6901 mddev->minor_version != info->minor_version ||
6902 /* mddev->patch_version != info->patch_version || */
6903 mddev->ctime != info->ctime ||
6904 mddev->level != info->level ||
6905 /* mddev->layout != info->layout || */
6906 mddev->persistent != !info->not_persistent ||
6907 mddev->chunk_sectors != info->chunk_size >> 9 ||
6908 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6909 ((state^info->state) & 0xfffffe00)
6912 /* Check there is only one change */
6913 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6915 if (mddev->raid_disks != info->raid_disks)
6917 if (mddev->layout != info->layout)
6919 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6926 if (mddev->layout != info->layout) {
6928 * we don't need to do anything at the md level, the
6929 * personality will take care of it all.
6931 if (mddev->pers->check_reshape == NULL)
6934 mddev->new_layout = info->layout;
6935 rv = mddev->pers->check_reshape(mddev);
6937 mddev->new_layout = mddev->layout;
6941 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6942 rv = update_size(mddev, (sector_t)info->size * 2);
6944 if (mddev->raid_disks != info->raid_disks)
6945 rv = update_raid_disks(mddev, info->raid_disks);
6947 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6948 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6952 if (mddev->recovery || mddev->sync_thread) {
6956 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6957 struct bitmap *bitmap;
6958 /* add the bitmap */
6959 if (mddev->bitmap) {
6963 if (mddev->bitmap_info.default_offset == 0) {
6967 mddev->bitmap_info.offset =
6968 mddev->bitmap_info.default_offset;
6969 mddev->bitmap_info.space =
6970 mddev->bitmap_info.default_space;
6971 bitmap = bitmap_create(mddev, -1);
6972 mddev_suspend(mddev);
6973 if (!IS_ERR(bitmap)) {
6974 mddev->bitmap = bitmap;
6975 rv = bitmap_load(mddev);
6977 rv = PTR_ERR(bitmap);
6979 bitmap_destroy(mddev);
6980 mddev_resume(mddev);
6982 /* remove the bitmap */
6983 if (!mddev->bitmap) {
6987 if (mddev->bitmap->storage.file) {
6991 if (mddev->bitmap_info.nodes) {
6992 /* hold PW on all the bitmap lock */
6993 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6994 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6996 md_cluster_ops->unlock_all_bitmaps(mddev);
7000 mddev->bitmap_info.nodes = 0;
7001 md_cluster_ops->leave(mddev);
7003 mddev_suspend(mddev);
7004 bitmap_destroy(mddev);
7005 mddev_resume(mddev);
7006 mddev->bitmap_info.offset = 0;
7009 md_update_sb(mddev, 1);
7015 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7017 struct md_rdev *rdev;
7020 if (mddev->pers == NULL)
7024 rdev = md_find_rdev_rcu(mddev, dev);
7028 md_error(mddev, rdev);
7029 if (!test_bit(Faulty, &rdev->flags))
7037 * We have a problem here : there is no easy way to give a CHS
7038 * virtual geometry. We currently pretend that we have a 2 heads
7039 * 4 sectors (with a BIG number of cylinders...). This drives
7040 * dosfs just mad... ;-)
7042 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7044 struct mddev *mddev = bdev->bd_disk->private_data;
7048 geo->cylinders = mddev->array_sectors / 8;
7052 static inline bool md_ioctl_valid(unsigned int cmd)
7057 case GET_ARRAY_INFO:
7058 case GET_BITMAP_FILE:
7061 case HOT_REMOVE_DISK:
7064 case RESTART_ARRAY_RW:
7066 case SET_ARRAY_INFO:
7067 case SET_BITMAP_FILE:
7068 case SET_DISK_FAULTY:
7071 case CLUSTERED_DISK_NACK:
7078 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7079 unsigned int cmd, unsigned long arg)
7082 void __user *argp = (void __user *)arg;
7083 struct mddev *mddev = NULL;
7085 bool did_set_md_closing = false;
7087 if (!md_ioctl_valid(cmd))
7092 case GET_ARRAY_INFO:
7096 if (!capable(CAP_SYS_ADMIN))
7101 * Commands dealing with the RAID driver but not any
7106 err = get_version(argp);
7112 autostart_arrays(arg);
7119 * Commands creating/starting a new array:
7122 mddev = bdev->bd_disk->private_data;
7129 /* Some actions do not requires the mutex */
7131 case GET_ARRAY_INFO:
7132 if (!mddev->raid_disks && !mddev->external)
7135 err = get_array_info(mddev, argp);
7139 if (!mddev->raid_disks && !mddev->external)
7142 err = get_disk_info(mddev, argp);
7145 case SET_DISK_FAULTY:
7146 err = set_disk_faulty(mddev, new_decode_dev(arg));
7149 case GET_BITMAP_FILE:
7150 err = get_bitmap_file(mddev, argp);
7155 if (cmd == ADD_NEW_DISK)
7156 /* need to ensure md_delayed_delete() has completed */
7157 flush_workqueue(md_misc_wq);
7159 if (cmd == HOT_REMOVE_DISK)
7160 /* need to ensure recovery thread has run */
7161 wait_event_interruptible_timeout(mddev->sb_wait,
7162 !test_bit(MD_RECOVERY_NEEDED,
7164 msecs_to_jiffies(5000));
7165 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7166 /* Need to flush page cache, and ensure no-one else opens
7169 mutex_lock(&mddev->open_mutex);
7170 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7171 mutex_unlock(&mddev->open_mutex);
7175 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7176 set_bit(MD_CLOSING, &mddev->flags);
7177 did_set_md_closing = true;
7178 mutex_unlock(&mddev->open_mutex);
7179 sync_blockdev(bdev);
7181 err = mddev_lock(mddev);
7183 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7188 if (cmd == SET_ARRAY_INFO) {
7189 mdu_array_info_t info;
7191 memset(&info, 0, sizeof(info));
7192 else if (copy_from_user(&info, argp, sizeof(info))) {
7197 err = update_array_info(mddev, &info);
7199 pr_warn("md: couldn't update array info. %d\n", err);
7204 if (!list_empty(&mddev->disks)) {
7205 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7209 if (mddev->raid_disks) {
7210 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7214 err = set_array_info(mddev, &info);
7216 pr_warn("md: couldn't set array info. %d\n", err);
7223 * Commands querying/configuring an existing array:
7225 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7226 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7227 if ((!mddev->raid_disks && !mddev->external)
7228 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7229 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7230 && cmd != GET_BITMAP_FILE) {
7236 * Commands even a read-only array can execute:
7239 case RESTART_ARRAY_RW:
7240 err = restart_array(mddev);
7244 err = do_md_stop(mddev, 0, bdev);
7248 err = md_set_readonly(mddev, bdev);
7251 case HOT_REMOVE_DISK:
7252 err = hot_remove_disk(mddev, new_decode_dev(arg));
7256 /* We can support ADD_NEW_DISK on read-only arrays
7257 * only if we are re-adding a preexisting device.
7258 * So require mddev->pers and MD_DISK_SYNC.
7261 mdu_disk_info_t info;
7262 if (copy_from_user(&info, argp, sizeof(info)))
7264 else if (!(info.state & (1<<MD_DISK_SYNC)))
7265 /* Need to clear read-only for this */
7268 err = add_new_disk(mddev, &info);
7274 if (get_user(ro, (int __user *)(arg))) {
7280 /* if the bdev is going readonly the value of mddev->ro
7281 * does not matter, no writes are coming
7286 /* are we are already prepared for writes? */
7290 /* transitioning to readauto need only happen for
7291 * arrays that call md_write_start
7294 err = restart_array(mddev);
7297 set_disk_ro(mddev->gendisk, 0);
7304 * The remaining ioctls are changing the state of the
7305 * superblock, so we do not allow them on read-only arrays.
7307 if (mddev->ro && mddev->pers) {
7308 if (mddev->ro == 2) {
7310 sysfs_notify_dirent_safe(mddev->sysfs_state);
7311 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7312 /* mddev_unlock will wake thread */
7313 /* If a device failed while we were read-only, we
7314 * need to make sure the metadata is updated now.
7316 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7317 mddev_unlock(mddev);
7318 wait_event(mddev->sb_wait,
7319 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7320 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7321 mddev_lock_nointr(mddev);
7332 mdu_disk_info_t info;
7333 if (copy_from_user(&info, argp, sizeof(info)))
7336 err = add_new_disk(mddev, &info);
7340 case CLUSTERED_DISK_NACK:
7341 if (mddev_is_clustered(mddev))
7342 md_cluster_ops->new_disk_ack(mddev, false);
7348 err = hot_add_disk(mddev, new_decode_dev(arg));
7352 err = do_md_run(mddev);
7355 case SET_BITMAP_FILE:
7356 err = set_bitmap_file(mddev, (int)arg);
7365 if (mddev->hold_active == UNTIL_IOCTL &&
7367 mddev->hold_active = 0;
7368 mddev_unlock(mddev);
7370 if(did_set_md_closing)
7371 clear_bit(MD_CLOSING, &mddev->flags);
7374 #ifdef CONFIG_COMPAT
7375 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7376 unsigned int cmd, unsigned long arg)
7379 case HOT_REMOVE_DISK:
7381 case SET_DISK_FAULTY:
7382 case SET_BITMAP_FILE:
7383 /* These take in integer arg, do not convert */
7386 arg = (unsigned long)compat_ptr(arg);
7390 return md_ioctl(bdev, mode, cmd, arg);
7392 #endif /* CONFIG_COMPAT */
7394 static int md_open(struct block_device *bdev, fmode_t mode)
7397 * Succeed if we can lock the mddev, which confirms that
7398 * it isn't being stopped right now.
7400 struct mddev *mddev = mddev_find(bdev->bd_dev);
7406 if (mddev->gendisk != bdev->bd_disk) {
7407 /* we are racing with mddev_put which is discarding this
7411 /* Wait until bdev->bd_disk is definitely gone */
7412 flush_workqueue(md_misc_wq);
7413 /* Then retry the open from the top */
7414 return -ERESTARTSYS;
7416 BUG_ON(mddev != bdev->bd_disk->private_data);
7418 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7421 if (test_bit(MD_CLOSING, &mddev->flags)) {
7422 mutex_unlock(&mddev->open_mutex);
7428 atomic_inc(&mddev->openers);
7429 mutex_unlock(&mddev->open_mutex);
7431 check_disk_change(bdev);
7438 static void md_release(struct gendisk *disk, fmode_t mode)
7440 struct mddev *mddev = disk->private_data;
7443 atomic_dec(&mddev->openers);
7447 static int md_media_changed(struct gendisk *disk)
7449 struct mddev *mddev = disk->private_data;
7451 return mddev->changed;
7454 static int md_revalidate(struct gendisk *disk)
7456 struct mddev *mddev = disk->private_data;
7461 static const struct block_device_operations md_fops =
7463 .owner = THIS_MODULE,
7465 .release = md_release,
7467 #ifdef CONFIG_COMPAT
7468 .compat_ioctl = md_compat_ioctl,
7470 .getgeo = md_getgeo,
7471 .media_changed = md_media_changed,
7472 .revalidate_disk= md_revalidate,
7475 static int md_thread(void *arg)
7477 struct md_thread *thread = arg;
7480 * md_thread is a 'system-thread', it's priority should be very
7481 * high. We avoid resource deadlocks individually in each
7482 * raid personality. (RAID5 does preallocation) We also use RR and
7483 * the very same RT priority as kswapd, thus we will never get
7484 * into a priority inversion deadlock.
7486 * we definitely have to have equal or higher priority than
7487 * bdflush, otherwise bdflush will deadlock if there are too
7488 * many dirty RAID5 blocks.
7491 allow_signal(SIGKILL);
7492 while (!kthread_should_stop()) {
7494 /* We need to wait INTERRUPTIBLE so that
7495 * we don't add to the load-average.
7496 * That means we need to be sure no signals are
7499 if (signal_pending(current))
7500 flush_signals(current);
7502 wait_event_interruptible_timeout
7504 test_bit(THREAD_WAKEUP, &thread->flags)
7505 || kthread_should_stop() || kthread_should_park(),
7508 clear_bit(THREAD_WAKEUP, &thread->flags);
7509 if (kthread_should_park())
7511 if (!kthread_should_stop())
7512 thread->run(thread);
7518 void md_wakeup_thread(struct md_thread *thread)
7521 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7522 set_bit(THREAD_WAKEUP, &thread->flags);
7523 wake_up(&thread->wqueue);
7526 EXPORT_SYMBOL(md_wakeup_thread);
7528 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7529 struct mddev *mddev, const char *name)
7531 struct md_thread *thread;
7533 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7537 init_waitqueue_head(&thread->wqueue);
7540 thread->mddev = mddev;
7541 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7542 thread->tsk = kthread_run(md_thread, thread,
7544 mdname(thread->mddev),
7546 if (IS_ERR(thread->tsk)) {
7552 EXPORT_SYMBOL(md_register_thread);
7554 void md_unregister_thread(struct md_thread **threadp)
7556 struct md_thread *thread = *threadp;
7559 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7560 /* Locking ensures that mddev_unlock does not wake_up a
7561 * non-existent thread
7563 spin_lock(&pers_lock);
7565 spin_unlock(&pers_lock);
7567 kthread_stop(thread->tsk);
7570 EXPORT_SYMBOL(md_unregister_thread);
7572 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7574 if (!rdev || test_bit(Faulty, &rdev->flags))
7577 if (!mddev->pers || !mddev->pers->error_handler)
7579 mddev->pers->error_handler(mddev,rdev);
7580 if (mddev->degraded)
7581 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7582 sysfs_notify_dirent_safe(rdev->sysfs_state);
7583 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7584 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7585 md_wakeup_thread(mddev->thread);
7586 if (mddev->event_work.func)
7587 queue_work(md_misc_wq, &mddev->event_work);
7588 md_new_event(mddev);
7590 EXPORT_SYMBOL(md_error);
7592 /* seq_file implementation /proc/mdstat */
7594 static void status_unused(struct seq_file *seq)
7597 struct md_rdev *rdev;
7599 seq_printf(seq, "unused devices: ");
7601 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7602 char b[BDEVNAME_SIZE];
7604 seq_printf(seq, "%s ",
7605 bdevname(rdev->bdev,b));
7608 seq_printf(seq, "<none>");
7610 seq_printf(seq, "\n");
7613 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7615 sector_t max_sectors, resync, res;
7616 unsigned long dt, db;
7619 unsigned int per_milli;
7621 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7622 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7623 max_sectors = mddev->resync_max_sectors;
7625 max_sectors = mddev->dev_sectors;
7627 resync = mddev->curr_resync;
7629 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7630 /* Still cleaning up */
7631 resync = max_sectors;
7632 } else if (resync > max_sectors)
7633 resync = max_sectors;
7635 resync -= atomic_read(&mddev->recovery_active);
7638 if (mddev->recovery_cp < MaxSector) {
7639 seq_printf(seq, "\tresync=PENDING");
7645 seq_printf(seq, "\tresync=DELAYED");
7649 WARN_ON(max_sectors == 0);
7650 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7651 * in a sector_t, and (max_sectors>>scale) will fit in a
7652 * u32, as those are the requirements for sector_div.
7653 * Thus 'scale' must be at least 10
7656 if (sizeof(sector_t) > sizeof(unsigned long)) {
7657 while ( max_sectors/2 > (1ULL<<(scale+32)))
7660 res = (resync>>scale)*1000;
7661 sector_div(res, (u32)((max_sectors>>scale)+1));
7665 int i, x = per_milli/50, y = 20-x;
7666 seq_printf(seq, "[");
7667 for (i = 0; i < x; i++)
7668 seq_printf(seq, "=");
7669 seq_printf(seq, ">");
7670 for (i = 0; i < y; i++)
7671 seq_printf(seq, ".");
7672 seq_printf(seq, "] ");
7674 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7675 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7677 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7679 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7680 "resync" : "recovery"))),
7681 per_milli/10, per_milli % 10,
7682 (unsigned long long) resync/2,
7683 (unsigned long long) max_sectors/2);
7686 * dt: time from mark until now
7687 * db: blocks written from mark until now
7688 * rt: remaining time
7690 * rt is a sector_t, so could be 32bit or 64bit.
7691 * So we divide before multiply in case it is 32bit and close
7693 * We scale the divisor (db) by 32 to avoid losing precision
7694 * near the end of resync when the number of remaining sectors
7696 * We then divide rt by 32 after multiplying by db to compensate.
7697 * The '+1' avoids division by zero if db is very small.
7699 dt = ((jiffies - mddev->resync_mark) / HZ);
7701 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7702 - mddev->resync_mark_cnt;
7704 rt = max_sectors - resync; /* number of remaining sectors */
7705 sector_div(rt, db/32+1);
7709 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7710 ((unsigned long)rt % 60)/6);
7712 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7716 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7718 struct list_head *tmp;
7720 struct mddev *mddev;
7728 spin_lock(&all_mddevs_lock);
7729 list_for_each(tmp,&all_mddevs)
7731 mddev = list_entry(tmp, struct mddev, all_mddevs);
7733 spin_unlock(&all_mddevs_lock);
7736 spin_unlock(&all_mddevs_lock);
7738 return (void*)2;/* tail */
7742 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7744 struct list_head *tmp;
7745 struct mddev *next_mddev, *mddev = v;
7751 spin_lock(&all_mddevs_lock);
7753 tmp = all_mddevs.next;
7755 tmp = mddev->all_mddevs.next;
7756 if (tmp != &all_mddevs)
7757 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7759 next_mddev = (void*)2;
7762 spin_unlock(&all_mddevs_lock);
7770 static void md_seq_stop(struct seq_file *seq, void *v)
7772 struct mddev *mddev = v;
7774 if (mddev && v != (void*)1 && v != (void*)2)
7778 static int md_seq_show(struct seq_file *seq, void *v)
7780 struct mddev *mddev = v;
7782 struct md_rdev *rdev;
7784 if (v == (void*)1) {
7785 struct md_personality *pers;
7786 seq_printf(seq, "Personalities : ");
7787 spin_lock(&pers_lock);
7788 list_for_each_entry(pers, &pers_list, list)
7789 seq_printf(seq, "[%s] ", pers->name);
7791 spin_unlock(&pers_lock);
7792 seq_printf(seq, "\n");
7793 seq->poll_event = atomic_read(&md_event_count);
7796 if (v == (void*)2) {
7801 spin_lock(&mddev->lock);
7802 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7803 seq_printf(seq, "%s : %sactive", mdname(mddev),
7804 mddev->pers ? "" : "in");
7807 seq_printf(seq, " (read-only)");
7809 seq_printf(seq, " (auto-read-only)");
7810 seq_printf(seq, " %s", mddev->pers->name);
7815 rdev_for_each_rcu(rdev, mddev) {
7816 char b[BDEVNAME_SIZE];
7817 seq_printf(seq, " %s[%d]",
7818 bdevname(rdev->bdev,b), rdev->desc_nr);
7819 if (test_bit(WriteMostly, &rdev->flags))
7820 seq_printf(seq, "(W)");
7821 if (test_bit(Journal, &rdev->flags))
7822 seq_printf(seq, "(J)");
7823 if (test_bit(Faulty, &rdev->flags)) {
7824 seq_printf(seq, "(F)");
7827 if (rdev->raid_disk < 0)
7828 seq_printf(seq, "(S)"); /* spare */
7829 if (test_bit(Replacement, &rdev->flags))
7830 seq_printf(seq, "(R)");
7831 sectors += rdev->sectors;
7835 if (!list_empty(&mddev->disks)) {
7837 seq_printf(seq, "\n %llu blocks",
7838 (unsigned long long)
7839 mddev->array_sectors / 2);
7841 seq_printf(seq, "\n %llu blocks",
7842 (unsigned long long)sectors / 2);
7844 if (mddev->persistent) {
7845 if (mddev->major_version != 0 ||
7846 mddev->minor_version != 90) {
7847 seq_printf(seq," super %d.%d",
7848 mddev->major_version,
7849 mddev->minor_version);
7851 } else if (mddev->external)
7852 seq_printf(seq, " super external:%s",
7853 mddev->metadata_type);
7855 seq_printf(seq, " super non-persistent");
7858 mddev->pers->status(seq, mddev);
7859 seq_printf(seq, "\n ");
7860 if (mddev->pers->sync_request) {
7861 if (status_resync(seq, mddev))
7862 seq_printf(seq, "\n ");
7865 seq_printf(seq, "\n ");
7867 bitmap_status(seq, mddev->bitmap);
7869 seq_printf(seq, "\n");
7871 spin_unlock(&mddev->lock);
7876 static const struct seq_operations md_seq_ops = {
7877 .start = md_seq_start,
7878 .next = md_seq_next,
7879 .stop = md_seq_stop,
7880 .show = md_seq_show,
7883 static int md_seq_open(struct inode *inode, struct file *file)
7885 struct seq_file *seq;
7888 error = seq_open(file, &md_seq_ops);
7892 seq = file->private_data;
7893 seq->poll_event = atomic_read(&md_event_count);
7897 static int md_unloading;
7898 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
7900 struct seq_file *seq = filp->private_data;
7904 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
7905 poll_wait(filp, &md_event_waiters, wait);
7907 /* always allow read */
7908 mask = EPOLLIN | EPOLLRDNORM;
7910 if (seq->poll_event != atomic_read(&md_event_count))
7911 mask |= EPOLLERR | EPOLLPRI;
7915 static const struct file_operations md_seq_fops = {
7916 .owner = THIS_MODULE,
7917 .open = md_seq_open,
7919 .llseek = seq_lseek,
7920 .release = seq_release,
7921 .poll = mdstat_poll,
7924 int register_md_personality(struct md_personality *p)
7926 pr_debug("md: %s personality registered for level %d\n",
7928 spin_lock(&pers_lock);
7929 list_add_tail(&p->list, &pers_list);
7930 spin_unlock(&pers_lock);
7933 EXPORT_SYMBOL(register_md_personality);
7935 int unregister_md_personality(struct md_personality *p)
7937 pr_debug("md: %s personality unregistered\n", p->name);
7938 spin_lock(&pers_lock);
7939 list_del_init(&p->list);
7940 spin_unlock(&pers_lock);
7943 EXPORT_SYMBOL(unregister_md_personality);
7945 int register_md_cluster_operations(struct md_cluster_operations *ops,
7946 struct module *module)
7949 spin_lock(&pers_lock);
7950 if (md_cluster_ops != NULL)
7953 md_cluster_ops = ops;
7954 md_cluster_mod = module;
7956 spin_unlock(&pers_lock);
7959 EXPORT_SYMBOL(register_md_cluster_operations);
7961 int unregister_md_cluster_operations(void)
7963 spin_lock(&pers_lock);
7964 md_cluster_ops = NULL;
7965 spin_unlock(&pers_lock);
7968 EXPORT_SYMBOL(unregister_md_cluster_operations);
7970 int md_setup_cluster(struct mddev *mddev, int nodes)
7972 if (!md_cluster_ops)
7973 request_module("md-cluster");
7974 spin_lock(&pers_lock);
7975 /* ensure module won't be unloaded */
7976 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7977 pr_warn("can't find md-cluster module or get it's reference.\n");
7978 spin_unlock(&pers_lock);
7981 spin_unlock(&pers_lock);
7983 return md_cluster_ops->join(mddev, nodes);
7986 void md_cluster_stop(struct mddev *mddev)
7988 if (!md_cluster_ops)
7990 md_cluster_ops->leave(mddev);
7991 module_put(md_cluster_mod);
7994 static int is_mddev_idle(struct mddev *mddev, int init)
7996 struct md_rdev *rdev;
8002 rdev_for_each_rcu(rdev, mddev) {
8003 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
8004 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
8005 (int)part_stat_read(&disk->part0, sectors[1]) -
8006 atomic_read(&disk->sync_io);
8007 /* sync IO will cause sync_io to increase before the disk_stats
8008 * as sync_io is counted when a request starts, and
8009 * disk_stats is counted when it completes.
8010 * So resync activity will cause curr_events to be smaller than
8011 * when there was no such activity.
8012 * non-sync IO will cause disk_stat to increase without
8013 * increasing sync_io so curr_events will (eventually)
8014 * be larger than it was before. Once it becomes
8015 * substantially larger, the test below will cause
8016 * the array to appear non-idle, and resync will slow
8018 * If there is a lot of outstanding resync activity when
8019 * we set last_event to curr_events, then all that activity
8020 * completing might cause the array to appear non-idle
8021 * and resync will be slowed down even though there might
8022 * not have been non-resync activity. This will only
8023 * happen once though. 'last_events' will soon reflect
8024 * the state where there is little or no outstanding
8025 * resync requests, and further resync activity will
8026 * always make curr_events less than last_events.
8029 if (init || curr_events - rdev->last_events > 64) {
8030 rdev->last_events = curr_events;
8038 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8040 /* another "blocks" (512byte) blocks have been synced */
8041 atomic_sub(blocks, &mddev->recovery_active);
8042 wake_up(&mddev->recovery_wait);
8044 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8045 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8046 md_wakeup_thread(mddev->thread);
8047 // stop recovery, signal do_sync ....
8050 EXPORT_SYMBOL(md_done_sync);
8052 /* md_write_start(mddev, bi)
8053 * If we need to update some array metadata (e.g. 'active' flag
8054 * in superblock) before writing, schedule a superblock update
8055 * and wait for it to complete.
8056 * A return value of 'false' means that the write wasn't recorded
8057 * and cannot proceed as the array is being suspend.
8059 bool md_write_start(struct mddev *mddev, struct bio *bi)
8063 if (bio_data_dir(bi) != WRITE)
8066 BUG_ON(mddev->ro == 1);
8067 if (mddev->ro == 2) {
8068 /* need to switch to read/write */
8070 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8071 md_wakeup_thread(mddev->thread);
8072 md_wakeup_thread(mddev->sync_thread);
8076 percpu_ref_get(&mddev->writes_pending);
8077 smp_mb(); /* Match smp_mb in set_in_sync() */
8078 if (mddev->safemode == 1)
8079 mddev->safemode = 0;
8080 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8081 if (mddev->in_sync || mddev->sync_checkers) {
8082 spin_lock(&mddev->lock);
8083 if (mddev->in_sync) {
8085 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8086 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8087 md_wakeup_thread(mddev->thread);
8090 spin_unlock(&mddev->lock);
8094 sysfs_notify_dirent_safe(mddev->sysfs_state);
8095 if (!mddev->has_superblocks)
8097 wait_event(mddev->sb_wait,
8098 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8100 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8101 percpu_ref_put(&mddev->writes_pending);
8106 EXPORT_SYMBOL(md_write_start);
8108 /* md_write_inc can only be called when md_write_start() has
8109 * already been called at least once of the current request.
8110 * It increments the counter and is useful when a single request
8111 * is split into several parts. Each part causes an increment and
8112 * so needs a matching md_write_end().
8113 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8114 * a spinlocked region.
8116 void md_write_inc(struct mddev *mddev, struct bio *bi)
8118 if (bio_data_dir(bi) != WRITE)
8120 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8121 percpu_ref_get(&mddev->writes_pending);
8123 EXPORT_SYMBOL(md_write_inc);
8125 void md_write_end(struct mddev *mddev)
8127 percpu_ref_put(&mddev->writes_pending);
8129 if (mddev->safemode == 2)
8130 md_wakeup_thread(mddev->thread);
8131 else if (mddev->safemode_delay)
8132 /* The roundup() ensures this only performs locking once
8133 * every ->safemode_delay jiffies
8135 mod_timer(&mddev->safemode_timer,
8136 roundup(jiffies, mddev->safemode_delay) +
8137 mddev->safemode_delay);
8140 EXPORT_SYMBOL(md_write_end);
8142 /* md_allow_write(mddev)
8143 * Calling this ensures that the array is marked 'active' so that writes
8144 * may proceed without blocking. It is important to call this before
8145 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8146 * Must be called with mddev_lock held.
8148 void md_allow_write(struct mddev *mddev)
8154 if (!mddev->pers->sync_request)
8157 spin_lock(&mddev->lock);
8158 if (mddev->in_sync) {
8160 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8161 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8162 if (mddev->safemode_delay &&
8163 mddev->safemode == 0)
8164 mddev->safemode = 1;
8165 spin_unlock(&mddev->lock);
8166 md_update_sb(mddev, 0);
8167 sysfs_notify_dirent_safe(mddev->sysfs_state);
8168 /* wait for the dirty state to be recorded in the metadata */
8169 wait_event(mddev->sb_wait,
8170 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8172 spin_unlock(&mddev->lock);
8174 EXPORT_SYMBOL_GPL(md_allow_write);
8176 #define SYNC_MARKS 10
8177 #define SYNC_MARK_STEP (3*HZ)
8178 #define UPDATE_FREQUENCY (5*60*HZ)
8179 void md_do_sync(struct md_thread *thread)
8181 struct mddev *mddev = thread->mddev;
8182 struct mddev *mddev2;
8183 unsigned int currspeed = 0,
8185 sector_t max_sectors,j, io_sectors, recovery_done;
8186 unsigned long mark[SYNC_MARKS];
8187 unsigned long update_time;
8188 sector_t mark_cnt[SYNC_MARKS];
8190 struct list_head *tmp;
8191 sector_t last_check;
8193 struct md_rdev *rdev;
8194 char *desc, *action = NULL;
8195 struct blk_plug plug;
8198 /* just incase thread restarts... */
8199 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8200 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8202 if (mddev->ro) {/* never try to sync a read-only array */
8203 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8207 if (mddev_is_clustered(mddev)) {
8208 ret = md_cluster_ops->resync_start(mddev);
8212 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8213 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8214 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8215 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8216 && ((unsigned long long)mddev->curr_resync_completed
8217 < (unsigned long long)mddev->resync_max_sectors))
8221 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8222 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8223 desc = "data-check";
8225 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8226 desc = "requested-resync";
8230 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8235 mddev->last_sync_action = action ?: desc;
8237 /* we overload curr_resync somewhat here.
8238 * 0 == not engaged in resync at all
8239 * 2 == checking that there is no conflict with another sync
8240 * 1 == like 2, but have yielded to allow conflicting resync to
8242 * other == active in resync - this many blocks
8244 * Before starting a resync we must have set curr_resync to
8245 * 2, and then checked that every "conflicting" array has curr_resync
8246 * less than ours. When we find one that is the same or higher
8247 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8248 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8249 * This will mean we have to start checking from the beginning again.
8254 int mddev2_minor = -1;
8255 mddev->curr_resync = 2;
8258 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8260 for_each_mddev(mddev2, tmp) {
8261 if (mddev2 == mddev)
8263 if (!mddev->parallel_resync
8264 && mddev2->curr_resync
8265 && match_mddev_units(mddev, mddev2)) {
8267 if (mddev < mddev2 && mddev->curr_resync == 2) {
8268 /* arbitrarily yield */
8269 mddev->curr_resync = 1;
8270 wake_up(&resync_wait);
8272 if (mddev > mddev2 && mddev->curr_resync == 1)
8273 /* no need to wait here, we can wait the next
8274 * time 'round when curr_resync == 2
8277 /* We need to wait 'interruptible' so as not to
8278 * contribute to the load average, and not to
8279 * be caught by 'softlockup'
8281 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8282 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8283 mddev2->curr_resync >= mddev->curr_resync) {
8284 if (mddev2_minor != mddev2->md_minor) {
8285 mddev2_minor = mddev2->md_minor;
8286 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8287 desc, mdname(mddev),
8291 if (signal_pending(current))
8292 flush_signals(current);
8294 finish_wait(&resync_wait, &wq);
8297 finish_wait(&resync_wait, &wq);
8300 } while (mddev->curr_resync < 2);
8303 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8304 /* resync follows the size requested by the personality,
8305 * which defaults to physical size, but can be virtual size
8307 max_sectors = mddev->resync_max_sectors;
8308 atomic64_set(&mddev->resync_mismatches, 0);
8309 /* we don't use the checkpoint if there's a bitmap */
8310 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8311 j = mddev->resync_min;
8312 else if (!mddev->bitmap)
8313 j = mddev->recovery_cp;
8315 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8316 max_sectors = mddev->resync_max_sectors;
8318 /* recovery follows the physical size of devices */
8319 max_sectors = mddev->dev_sectors;
8322 rdev_for_each_rcu(rdev, mddev)
8323 if (rdev->raid_disk >= 0 &&
8324 !test_bit(Journal, &rdev->flags) &&
8325 !test_bit(Faulty, &rdev->flags) &&
8326 !test_bit(In_sync, &rdev->flags) &&
8327 rdev->recovery_offset < j)
8328 j = rdev->recovery_offset;
8331 /* If there is a bitmap, we need to make sure all
8332 * writes that started before we added a spare
8333 * complete before we start doing a recovery.
8334 * Otherwise the write might complete and (via
8335 * bitmap_endwrite) set a bit in the bitmap after the
8336 * recovery has checked that bit and skipped that
8339 if (mddev->bitmap) {
8340 mddev->pers->quiesce(mddev, 1);
8341 mddev->pers->quiesce(mddev, 0);
8345 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8346 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8347 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8348 speed_max(mddev), desc);
8350 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8353 for (m = 0; m < SYNC_MARKS; m++) {
8355 mark_cnt[m] = io_sectors;
8358 mddev->resync_mark = mark[last_mark];
8359 mddev->resync_mark_cnt = mark_cnt[last_mark];
8362 * Tune reconstruction:
8364 window = 32*(PAGE_SIZE/512);
8365 pr_debug("md: using %dk window, over a total of %lluk.\n",
8366 window/2, (unsigned long long)max_sectors/2);
8368 atomic_set(&mddev->recovery_active, 0);
8372 pr_debug("md: resuming %s of %s from checkpoint.\n",
8373 desc, mdname(mddev));
8374 mddev->curr_resync = j;
8376 mddev->curr_resync = 3; /* no longer delayed */
8377 mddev->curr_resync_completed = j;
8378 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8379 md_new_event(mddev);
8380 update_time = jiffies;
8382 blk_start_plug(&plug);
8383 while (j < max_sectors) {
8388 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8389 ((mddev->curr_resync > mddev->curr_resync_completed &&
8390 (mddev->curr_resync - mddev->curr_resync_completed)
8391 > (max_sectors >> 4)) ||
8392 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8393 (j - mddev->curr_resync_completed)*2
8394 >= mddev->resync_max - mddev->curr_resync_completed ||
8395 mddev->curr_resync_completed > mddev->resync_max
8397 /* time to update curr_resync_completed */
8398 wait_event(mddev->recovery_wait,
8399 atomic_read(&mddev->recovery_active) == 0);
8400 mddev->curr_resync_completed = j;
8401 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8402 j > mddev->recovery_cp)
8403 mddev->recovery_cp = j;
8404 update_time = jiffies;
8405 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8406 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8409 while (j >= mddev->resync_max &&
8410 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8411 /* As this condition is controlled by user-space,
8412 * we can block indefinitely, so use '_interruptible'
8413 * to avoid triggering warnings.
8415 flush_signals(current); /* just in case */
8416 wait_event_interruptible(mddev->recovery_wait,
8417 mddev->resync_max > j
8418 || test_bit(MD_RECOVERY_INTR,
8422 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8425 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8427 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8431 if (!skipped) { /* actual IO requested */
8432 io_sectors += sectors;
8433 atomic_add(sectors, &mddev->recovery_active);
8436 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8440 if (j > max_sectors)
8441 /* when skipping, extra large numbers can be returned. */
8444 mddev->curr_resync = j;
8445 mddev->curr_mark_cnt = io_sectors;
8446 if (last_check == 0)
8447 /* this is the earliest that rebuild will be
8448 * visible in /proc/mdstat
8450 md_new_event(mddev);
8452 if (last_check + window > io_sectors || j == max_sectors)
8455 last_check = io_sectors;
8457 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8459 int next = (last_mark+1) % SYNC_MARKS;
8461 mddev->resync_mark = mark[next];
8462 mddev->resync_mark_cnt = mark_cnt[next];
8463 mark[next] = jiffies;
8464 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8468 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8472 * this loop exits only if either when we are slower than
8473 * the 'hard' speed limit, or the system was IO-idle for
8475 * the system might be non-idle CPU-wise, but we only care
8476 * about not overloading the IO subsystem. (things like an
8477 * e2fsck being done on the RAID array should execute fast)
8481 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8482 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8483 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8485 if (currspeed > speed_min(mddev)) {
8486 if (currspeed > speed_max(mddev)) {
8490 if (!is_mddev_idle(mddev, 0)) {
8492 * Give other IO more of a chance.
8493 * The faster the devices, the less we wait.
8495 wait_event(mddev->recovery_wait,
8496 !atomic_read(&mddev->recovery_active));
8500 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8501 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8502 ? "interrupted" : "done");
8504 * this also signals 'finished resyncing' to md_stop
8506 blk_finish_plug(&plug);
8507 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8509 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8510 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8511 mddev->curr_resync > 3) {
8512 mddev->curr_resync_completed = mddev->curr_resync;
8513 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8515 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8517 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8518 mddev->curr_resync > 3) {
8519 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8520 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8521 if (mddev->curr_resync >= mddev->recovery_cp) {
8522 pr_debug("md: checkpointing %s of %s.\n",
8523 desc, mdname(mddev));
8524 if (test_bit(MD_RECOVERY_ERROR,
8526 mddev->recovery_cp =
8527 mddev->curr_resync_completed;
8529 mddev->recovery_cp =
8533 mddev->recovery_cp = MaxSector;
8535 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8536 mddev->curr_resync = MaxSector;
8537 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8538 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8540 rdev_for_each_rcu(rdev, mddev)
8541 if (rdev->raid_disk >= 0 &&
8542 mddev->delta_disks >= 0 &&
8543 !test_bit(Journal, &rdev->flags) &&
8544 !test_bit(Faulty, &rdev->flags) &&
8545 !test_bit(In_sync, &rdev->flags) &&
8546 rdev->recovery_offset < mddev->curr_resync)
8547 rdev->recovery_offset = mddev->curr_resync;
8553 /* set CHANGE_PENDING here since maybe another update is needed,
8554 * so other nodes are informed. It should be harmless for normal
8556 set_mask_bits(&mddev->sb_flags, 0,
8557 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8559 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8560 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8561 mddev->delta_disks > 0 &&
8562 mddev->pers->finish_reshape &&
8563 mddev->pers->size &&
8565 mddev_lock_nointr(mddev);
8566 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8567 mddev_unlock(mddev);
8568 set_capacity(mddev->gendisk, mddev->array_sectors);
8569 revalidate_disk(mddev->gendisk);
8572 spin_lock(&mddev->lock);
8573 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8574 /* We completed so min/max setting can be forgotten if used. */
8575 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8576 mddev->resync_min = 0;
8577 mddev->resync_max = MaxSector;
8578 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8579 mddev->resync_min = mddev->curr_resync_completed;
8580 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8581 mddev->curr_resync = 0;
8582 spin_unlock(&mddev->lock);
8584 wake_up(&resync_wait);
8585 md_wakeup_thread(mddev->thread);
8588 EXPORT_SYMBOL_GPL(md_do_sync);
8590 static int remove_and_add_spares(struct mddev *mddev,
8591 struct md_rdev *this)
8593 struct md_rdev *rdev;
8596 bool remove_some = false;
8598 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8599 /* Mustn't remove devices when resync thread is running */
8602 rdev_for_each(rdev, mddev) {
8603 if ((this == NULL || rdev == this) &&
8604 rdev->raid_disk >= 0 &&
8605 !test_bit(Blocked, &rdev->flags) &&
8606 test_bit(Faulty, &rdev->flags) &&
8607 atomic_read(&rdev->nr_pending)==0) {
8608 /* Faulty non-Blocked devices with nr_pending == 0
8609 * never get nr_pending incremented,
8610 * never get Faulty cleared, and never get Blocked set.
8611 * So we can synchronize_rcu now rather than once per device
8614 set_bit(RemoveSynchronized, &rdev->flags);
8620 rdev_for_each(rdev, mddev) {
8621 if ((this == NULL || rdev == this) &&
8622 rdev->raid_disk >= 0 &&
8623 !test_bit(Blocked, &rdev->flags) &&
8624 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8625 (!test_bit(In_sync, &rdev->flags) &&
8626 !test_bit(Journal, &rdev->flags))) &&
8627 atomic_read(&rdev->nr_pending)==0)) {
8628 if (mddev->pers->hot_remove_disk(
8629 mddev, rdev) == 0) {
8630 sysfs_unlink_rdev(mddev, rdev);
8631 rdev->raid_disk = -1;
8635 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8636 clear_bit(RemoveSynchronized, &rdev->flags);
8639 if (removed && mddev->kobj.sd)
8640 sysfs_notify(&mddev->kobj, NULL, "degraded");
8642 if (this && removed)
8645 rdev_for_each(rdev, mddev) {
8646 if (this && this != rdev)
8648 if (test_bit(Candidate, &rdev->flags))
8650 if (rdev->raid_disk >= 0 &&
8651 !test_bit(In_sync, &rdev->flags) &&
8652 !test_bit(Journal, &rdev->flags) &&
8653 !test_bit(Faulty, &rdev->flags))
8655 if (rdev->raid_disk >= 0)
8657 if (test_bit(Faulty, &rdev->flags))
8659 if (!test_bit(Journal, &rdev->flags)) {
8661 ! (rdev->saved_raid_disk >= 0 &&
8662 !test_bit(Bitmap_sync, &rdev->flags)))
8665 rdev->recovery_offset = 0;
8668 hot_add_disk(mddev, rdev) == 0) {
8669 if (sysfs_link_rdev(mddev, rdev))
8670 /* failure here is OK */;
8671 if (!test_bit(Journal, &rdev->flags))
8673 md_new_event(mddev);
8674 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8679 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8683 static void md_start_sync(struct work_struct *ws)
8685 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8687 mddev->sync_thread = md_register_thread(md_do_sync,
8690 if (!mddev->sync_thread) {
8691 pr_warn("%s: could not start resync thread...\n",
8693 /* leave the spares where they are, it shouldn't hurt */
8694 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8695 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8696 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8697 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8698 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8699 wake_up(&resync_wait);
8700 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8702 if (mddev->sysfs_action)
8703 sysfs_notify_dirent_safe(mddev->sysfs_action);
8705 md_wakeup_thread(mddev->sync_thread);
8706 sysfs_notify_dirent_safe(mddev->sysfs_action);
8707 md_new_event(mddev);
8711 * This routine is regularly called by all per-raid-array threads to
8712 * deal with generic issues like resync and super-block update.
8713 * Raid personalities that don't have a thread (linear/raid0) do not
8714 * need this as they never do any recovery or update the superblock.
8716 * It does not do any resync itself, but rather "forks" off other threads
8717 * to do that as needed.
8718 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8719 * "->recovery" and create a thread at ->sync_thread.
8720 * When the thread finishes it sets MD_RECOVERY_DONE
8721 * and wakeups up this thread which will reap the thread and finish up.
8722 * This thread also removes any faulty devices (with nr_pending == 0).
8724 * The overall approach is:
8725 * 1/ if the superblock needs updating, update it.
8726 * 2/ If a recovery thread is running, don't do anything else.
8727 * 3/ If recovery has finished, clean up, possibly marking spares active.
8728 * 4/ If there are any faulty devices, remove them.
8729 * 5/ If array is degraded, try to add spares devices
8730 * 6/ If array has spares or is not in-sync, start a resync thread.
8732 void md_check_recovery(struct mddev *mddev)
8734 if (mddev->suspended)
8738 bitmap_daemon_work(mddev);
8740 if (signal_pending(current)) {
8741 if (mddev->pers->sync_request && !mddev->external) {
8742 pr_debug("md: %s in immediate safe mode\n",
8744 mddev->safemode = 2;
8746 flush_signals(current);
8749 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8752 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8753 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8754 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8755 (mddev->external == 0 && mddev->safemode == 1) ||
8756 (mddev->safemode == 2
8757 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8761 if (mddev_trylock(mddev)) {
8764 if (!mddev->external && mddev->safemode == 1)
8765 mddev->safemode = 0;
8768 struct md_rdev *rdev;
8769 if (!mddev->external && mddev->in_sync)
8770 /* 'Blocked' flag not needed as failed devices
8771 * will be recorded if array switched to read/write.
8772 * Leaving it set will prevent the device
8773 * from being removed.
8775 rdev_for_each(rdev, mddev)
8776 clear_bit(Blocked, &rdev->flags);
8777 /* On a read-only array we can:
8778 * - remove failed devices
8779 * - add already-in_sync devices if the array itself
8781 * As we only add devices that are already in-sync,
8782 * we can activate the spares immediately.
8784 remove_and_add_spares(mddev, NULL);
8785 /* There is no thread, but we need to call
8786 * ->spare_active and clear saved_raid_disk
8788 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8789 md_reap_sync_thread(mddev);
8790 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8791 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8792 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8796 if (mddev_is_clustered(mddev)) {
8797 struct md_rdev *rdev;
8798 /* kick the device if another node issued a
8801 rdev_for_each(rdev, mddev) {
8802 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8803 rdev->raid_disk < 0)
8804 md_kick_rdev_from_array(rdev);
8808 if (!mddev->external && !mddev->in_sync) {
8809 spin_lock(&mddev->lock);
8811 spin_unlock(&mddev->lock);
8814 if (mddev->sb_flags)
8815 md_update_sb(mddev, 0);
8817 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8818 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8819 /* resync/recovery still happening */
8820 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8823 if (mddev->sync_thread) {
8824 md_reap_sync_thread(mddev);
8827 /* Set RUNNING before clearing NEEDED to avoid
8828 * any transients in the value of "sync_action".
8830 mddev->curr_resync_completed = 0;
8831 spin_lock(&mddev->lock);
8832 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8833 spin_unlock(&mddev->lock);
8834 /* Clear some bits that don't mean anything, but
8837 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8838 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8840 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8841 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8843 /* no recovery is running.
8844 * remove any failed drives, then
8845 * add spares if possible.
8846 * Spares are also removed and re-added, to allow
8847 * the personality to fail the re-add.
8850 if (mddev->reshape_position != MaxSector) {
8851 if (mddev->pers->check_reshape == NULL ||
8852 mddev->pers->check_reshape(mddev) != 0)
8853 /* Cannot proceed */
8855 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8856 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8857 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8858 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8859 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8860 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8861 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8862 } else if (mddev->recovery_cp < MaxSector) {
8863 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8864 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8865 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8866 /* nothing to be done ... */
8869 if (mddev->pers->sync_request) {
8871 /* We are adding a device or devices to an array
8872 * which has the bitmap stored on all devices.
8873 * So make sure all bitmap pages get written
8875 bitmap_write_all(mddev->bitmap);
8877 INIT_WORK(&mddev->del_work, md_start_sync);
8878 queue_work(md_misc_wq, &mddev->del_work);
8882 if (!mddev->sync_thread) {
8883 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8884 wake_up(&resync_wait);
8885 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8887 if (mddev->sysfs_action)
8888 sysfs_notify_dirent_safe(mddev->sysfs_action);
8891 wake_up(&mddev->sb_wait);
8892 mddev_unlock(mddev);
8893 } else if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8894 /* Write superblock - thread that called mddev_suspend()
8895 * holds reconfig_mutex for us.
8897 set_bit(MD_UPDATING_SB, &mddev->flags);
8898 smp_mb__after_atomic();
8899 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8900 md_update_sb(mddev, 0);
8901 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8902 wake_up(&mddev->sb_wait);
8905 EXPORT_SYMBOL(md_check_recovery);
8907 void md_reap_sync_thread(struct mddev *mddev)
8909 struct md_rdev *rdev;
8911 /* resync has finished, collect result */
8912 md_unregister_thread(&mddev->sync_thread);
8913 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8914 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8916 /* activate any spares */
8917 if (mddev->pers->spare_active(mddev)) {
8918 sysfs_notify(&mddev->kobj, NULL,
8920 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8923 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8924 mddev->pers->finish_reshape)
8925 mddev->pers->finish_reshape(mddev);
8927 /* If array is no-longer degraded, then any saved_raid_disk
8928 * information must be scrapped.
8930 if (!mddev->degraded)
8931 rdev_for_each(rdev, mddev)
8932 rdev->saved_raid_disk = -1;
8934 md_update_sb(mddev, 1);
8935 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8936 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8938 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8939 md_cluster_ops->resync_finish(mddev);
8940 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8941 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8942 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8943 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8944 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8945 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8946 wake_up(&resync_wait);
8947 /* flag recovery needed just to double check */
8948 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8949 sysfs_notify_dirent_safe(mddev->sysfs_action);
8950 md_new_event(mddev);
8951 if (mddev->event_work.func)
8952 queue_work(md_misc_wq, &mddev->event_work);
8954 EXPORT_SYMBOL(md_reap_sync_thread);
8956 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8958 sysfs_notify_dirent_safe(rdev->sysfs_state);
8959 wait_event_timeout(rdev->blocked_wait,
8960 !test_bit(Blocked, &rdev->flags) &&
8961 !test_bit(BlockedBadBlocks, &rdev->flags),
8962 msecs_to_jiffies(5000));
8963 rdev_dec_pending(rdev, mddev);
8965 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8967 void md_finish_reshape(struct mddev *mddev)
8969 /* called be personality module when reshape completes. */
8970 struct md_rdev *rdev;
8972 rdev_for_each(rdev, mddev) {
8973 if (rdev->data_offset > rdev->new_data_offset)
8974 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8976 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8977 rdev->data_offset = rdev->new_data_offset;
8980 EXPORT_SYMBOL(md_finish_reshape);
8982 /* Bad block management */
8984 /* Returns 1 on success, 0 on failure */
8985 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8988 struct mddev *mddev = rdev->mddev;
8991 s += rdev->new_data_offset;
8993 s += rdev->data_offset;
8994 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8996 /* Make sure they get written out promptly */
8997 if (test_bit(ExternalBbl, &rdev->flags))
8998 sysfs_notify(&rdev->kobj, NULL,
8999 "unacknowledged_bad_blocks");
9000 sysfs_notify_dirent_safe(rdev->sysfs_state);
9001 set_mask_bits(&mddev->sb_flags, 0,
9002 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9003 md_wakeup_thread(rdev->mddev->thread);
9008 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9010 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9015 s += rdev->new_data_offset;
9017 s += rdev->data_offset;
9018 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9019 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9020 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9023 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9025 static int md_notify_reboot(struct notifier_block *this,
9026 unsigned long code, void *x)
9028 struct list_head *tmp;
9029 struct mddev *mddev;
9032 for_each_mddev(mddev, tmp) {
9033 if (mddev_trylock(mddev)) {
9035 __md_stop_writes(mddev);
9036 if (mddev->persistent)
9037 mddev->safemode = 2;
9038 mddev_unlock(mddev);
9043 * certain more exotic SCSI devices are known to be
9044 * volatile wrt too early system reboots. While the
9045 * right place to handle this issue is the given
9046 * driver, we do want to have a safe RAID driver ...
9054 static struct notifier_block md_notifier = {
9055 .notifier_call = md_notify_reboot,
9057 .priority = INT_MAX, /* before any real devices */
9060 static void md_geninit(void)
9062 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9064 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9067 static int __init md_init(void)
9071 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9075 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9079 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9082 if ((ret = register_blkdev(0, "mdp")) < 0)
9086 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9087 md_probe, NULL, NULL);
9088 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9089 md_probe, NULL, NULL);
9091 register_reboot_notifier(&md_notifier);
9092 raid_table_header = register_sysctl_table(raid_root_table);
9098 unregister_blkdev(MD_MAJOR, "md");
9100 destroy_workqueue(md_misc_wq);
9102 destroy_workqueue(md_wq);
9107 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9109 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9110 struct md_rdev *rdev2;
9112 char b[BDEVNAME_SIZE];
9115 * If size is changed in another node then we need to
9116 * do resize as well.
9118 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9119 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9121 pr_info("md-cluster: resize failed\n");
9123 bitmap_update_sb(mddev->bitmap);
9126 /* Check for change of roles in the active devices */
9127 rdev_for_each(rdev2, mddev) {
9128 if (test_bit(Faulty, &rdev2->flags))
9131 /* Check if the roles changed */
9132 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9134 if (test_bit(Candidate, &rdev2->flags)) {
9135 if (role == 0xfffe) {
9136 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9137 md_kick_rdev_from_array(rdev2);
9141 clear_bit(Candidate, &rdev2->flags);
9144 if (role != rdev2->raid_disk) {
9146 if (rdev2->raid_disk == -1 && role != 0xffff) {
9147 rdev2->saved_raid_disk = role;
9148 ret = remove_and_add_spares(mddev, rdev2);
9149 pr_info("Activated spare: %s\n",
9150 bdevname(rdev2->bdev,b));
9151 /* wakeup mddev->thread here, so array could
9152 * perform resync with the new activated disk */
9153 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9154 md_wakeup_thread(mddev->thread);
9158 * We just want to do the minimum to mark the disk
9159 * as faulty. The recovery is performed by the
9160 * one who initiated the error.
9162 if ((role == 0xfffe) || (role == 0xfffd)) {
9163 md_error(mddev, rdev2);
9164 clear_bit(Blocked, &rdev2->flags);
9169 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9170 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9172 /* Finally set the event to be up to date */
9173 mddev->events = le64_to_cpu(sb->events);
9176 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9179 struct page *swapout = rdev->sb_page;
9180 struct mdp_superblock_1 *sb;
9182 /* Store the sb page of the rdev in the swapout temporary
9183 * variable in case we err in the future
9185 rdev->sb_page = NULL;
9186 err = alloc_disk_sb(rdev);
9188 ClearPageUptodate(rdev->sb_page);
9189 rdev->sb_loaded = 0;
9190 err = super_types[mddev->major_version].
9191 load_super(rdev, NULL, mddev->minor_version);
9194 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9195 __func__, __LINE__, rdev->desc_nr, err);
9197 put_page(rdev->sb_page);
9198 rdev->sb_page = swapout;
9199 rdev->sb_loaded = 1;
9203 sb = page_address(rdev->sb_page);
9204 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9208 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9209 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9211 /* The other node finished recovery, call spare_active to set
9212 * device In_sync and mddev->degraded
9214 if (rdev->recovery_offset == MaxSector &&
9215 !test_bit(In_sync, &rdev->flags) &&
9216 mddev->pers->spare_active(mddev))
9217 sysfs_notify(&mddev->kobj, NULL, "degraded");
9223 void md_reload_sb(struct mddev *mddev, int nr)
9225 struct md_rdev *rdev;
9229 rdev_for_each_rcu(rdev, mddev) {
9230 if (rdev->desc_nr == nr)
9234 if (!rdev || rdev->desc_nr != nr) {
9235 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9239 err = read_rdev(mddev, rdev);
9243 check_sb_changes(mddev, rdev);
9245 /* Read all rdev's to update recovery_offset */
9246 rdev_for_each_rcu(rdev, mddev) {
9247 if (!test_bit(Faulty, &rdev->flags))
9248 read_rdev(mddev, rdev);
9251 EXPORT_SYMBOL(md_reload_sb);
9256 * Searches all registered partitions for autorun RAID arrays
9260 static DEFINE_MUTEX(detected_devices_mutex);
9261 static LIST_HEAD(all_detected_devices);
9262 struct detected_devices_node {
9263 struct list_head list;
9267 void md_autodetect_dev(dev_t dev)
9269 struct detected_devices_node *node_detected_dev;
9271 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9272 if (node_detected_dev) {
9273 node_detected_dev->dev = dev;
9274 mutex_lock(&detected_devices_mutex);
9275 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9276 mutex_unlock(&detected_devices_mutex);
9280 static void autostart_arrays(int part)
9282 struct md_rdev *rdev;
9283 struct detected_devices_node *node_detected_dev;
9285 int i_scanned, i_passed;
9290 pr_info("md: Autodetecting RAID arrays.\n");
9292 mutex_lock(&detected_devices_mutex);
9293 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9295 node_detected_dev = list_entry(all_detected_devices.next,
9296 struct detected_devices_node, list);
9297 list_del(&node_detected_dev->list);
9298 dev = node_detected_dev->dev;
9299 kfree(node_detected_dev);
9300 mutex_unlock(&detected_devices_mutex);
9301 rdev = md_import_device(dev,0, 90);
9302 mutex_lock(&detected_devices_mutex);
9306 if (test_bit(Faulty, &rdev->flags))
9309 set_bit(AutoDetected, &rdev->flags);
9310 list_add(&rdev->same_set, &pending_raid_disks);
9313 mutex_unlock(&detected_devices_mutex);
9315 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9317 autorun_devices(part);
9320 #endif /* !MODULE */
9322 static __exit void md_exit(void)
9324 struct mddev *mddev;
9325 struct list_head *tmp;
9328 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9329 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9331 unregister_blkdev(MD_MAJOR,"md");
9332 unregister_blkdev(mdp_major, "mdp");
9333 unregister_reboot_notifier(&md_notifier);
9334 unregister_sysctl_table(raid_table_header);
9336 /* We cannot unload the modules while some process is
9337 * waiting for us in select() or poll() - wake them up
9340 while (waitqueue_active(&md_event_waiters)) {
9341 /* not safe to leave yet */
9342 wake_up(&md_event_waiters);
9346 remove_proc_entry("mdstat", NULL);
9348 for_each_mddev(mddev, tmp) {
9349 export_array(mddev);
9351 mddev->hold_active = 0;
9353 * for_each_mddev() will call mddev_put() at the end of each
9354 * iteration. As the mddev is now fully clear, this will
9355 * schedule the mddev for destruction by a workqueue, and the
9356 * destroy_workqueue() below will wait for that to complete.
9359 destroy_workqueue(md_misc_wq);
9360 destroy_workqueue(md_wq);
9363 subsys_initcall(md_init);
9364 module_exit(md_exit)
9366 static int get_ro(char *buffer, const struct kernel_param *kp)
9368 return sprintf(buffer, "%d", start_readonly);
9370 static int set_ro(const char *val, const struct kernel_param *kp)
9372 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9375 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9376 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9377 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9378 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9380 MODULE_LICENSE("GPL");
9381 MODULE_DESCRIPTION("MD RAID framework");
9383 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);