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 <trace/events/block.h>
71 #include "md-cluster.h"
74 static void autostart_arrays(int part);
77 /* pers_list is a list of registered personalities protected
79 * pers_lock does extra service to protect accesses to
80 * mddev->thread when the mutex cannot be held.
82 static LIST_HEAD(pers_list);
83 static DEFINE_SPINLOCK(pers_lock);
85 struct md_cluster_operations *md_cluster_ops;
86 EXPORT_SYMBOL(md_cluster_ops);
87 struct module *md_cluster_mod;
88 EXPORT_SYMBOL(md_cluster_mod);
90 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
91 static struct workqueue_struct *md_wq;
92 static struct workqueue_struct *md_misc_wq;
94 static int remove_and_add_spares(struct mddev *mddev,
95 struct md_rdev *this);
96 static void mddev_detach(struct mddev *mddev);
99 * Default number of read corrections we'll attempt on an rdev
100 * before ejecting it from the array. We divide the read error
101 * count by 2 for every hour elapsed between read errors.
103 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
105 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
106 * is 1000 KB/sec, so the extra system load does not show up that much.
107 * Increase it if you want to have more _guaranteed_ speed. Note that
108 * the RAID driver will use the maximum available bandwidth if the IO
109 * subsystem is idle. There is also an 'absolute maximum' reconstruction
110 * speed limit - in case reconstruction slows down your system despite
113 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
114 * or /sys/block/mdX/md/sync_speed_{min,max}
117 static int sysctl_speed_limit_min = 1000;
118 static int sysctl_speed_limit_max = 200000;
119 static inline int speed_min(struct mddev *mddev)
121 return mddev->sync_speed_min ?
122 mddev->sync_speed_min : sysctl_speed_limit_min;
125 static inline int speed_max(struct mddev *mddev)
127 return mddev->sync_speed_max ?
128 mddev->sync_speed_max : sysctl_speed_limit_max;
131 static struct ctl_table_header *raid_table_header;
133 static struct ctl_table raid_table[] = {
135 .procname = "speed_limit_min",
136 .data = &sysctl_speed_limit_min,
137 .maxlen = sizeof(int),
138 .mode = S_IRUGO|S_IWUSR,
139 .proc_handler = proc_dointvec,
142 .procname = "speed_limit_max",
143 .data = &sysctl_speed_limit_max,
144 .maxlen = sizeof(int),
145 .mode = S_IRUGO|S_IWUSR,
146 .proc_handler = proc_dointvec,
151 static struct ctl_table raid_dir_table[] = {
155 .mode = S_IRUGO|S_IXUGO,
161 static struct ctl_table raid_root_table[] = {
166 .child = raid_dir_table,
171 static const struct block_device_operations md_fops;
173 static int start_readonly;
176 * like bio_clone, but with a local bio set
179 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
184 if (!mddev || !mddev->bio_set)
185 return bio_alloc(gfp_mask, nr_iovecs);
187 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
192 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
195 * We have a system wide 'event count' that is incremented
196 * on any 'interesting' event, and readers of /proc/mdstat
197 * can use 'poll' or 'select' to find out when the event
201 * start array, stop array, error, add device, remove device,
202 * start build, activate spare
204 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
205 static atomic_t md_event_count;
206 void md_new_event(struct mddev *mddev)
208 atomic_inc(&md_event_count);
209 wake_up(&md_event_waiters);
211 EXPORT_SYMBOL_GPL(md_new_event);
214 * Enables to iterate over all existing md arrays
215 * all_mddevs_lock protects this list.
217 static LIST_HEAD(all_mddevs);
218 static DEFINE_SPINLOCK(all_mddevs_lock);
221 * iterates through all used mddevs in the system.
222 * We take care to grab the all_mddevs_lock whenever navigating
223 * the list, and to always hold a refcount when unlocked.
224 * Any code which breaks out of this loop while own
225 * a reference to the current mddev and must mddev_put it.
227 #define for_each_mddev(_mddev,_tmp) \
229 for (({ spin_lock(&all_mddevs_lock); \
230 _tmp = all_mddevs.next; \
232 ({ if (_tmp != &all_mddevs) \
233 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
234 spin_unlock(&all_mddevs_lock); \
235 if (_mddev) mddev_put(_mddev); \
236 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
237 _tmp != &all_mddevs;}); \
238 ({ spin_lock(&all_mddevs_lock); \
239 _tmp = _tmp->next;}) \
242 /* Rather than calling directly into the personality make_request function,
243 * IO requests come here first so that we can check if the device is
244 * being suspended pending a reconfiguration.
245 * We hold a refcount over the call to ->make_request. By the time that
246 * call has finished, the bio has been linked into some internal structure
247 * and so is visible to ->quiesce(), so we don't need the refcount any more.
249 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
251 const int rw = bio_data_dir(bio);
252 struct mddev *mddev = q->queuedata;
253 unsigned int sectors;
256 blk_queue_split(q, &bio, q->bio_split);
258 if (mddev == NULL || mddev->pers == NULL) {
260 return BLK_QC_T_NONE;
262 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
263 if (bio_sectors(bio) != 0)
264 bio->bi_error = -EROFS;
266 return BLK_QC_T_NONE;
268 smp_rmb(); /* Ensure implications of 'active' are visible */
270 if (mddev->suspended) {
273 prepare_to_wait(&mddev->sb_wait, &__wait,
274 TASK_UNINTERRUPTIBLE);
275 if (!mddev->suspended)
281 finish_wait(&mddev->sb_wait, &__wait);
283 atomic_inc(&mddev->active_io);
287 * save the sectors now since our bio can
288 * go away inside make_request
290 sectors = bio_sectors(bio);
291 /* bio could be mergeable after passing to underlayer */
292 bio->bi_opf &= ~REQ_NOMERGE;
293 mddev->pers->make_request(mddev, bio);
295 cpu = part_stat_lock();
296 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
297 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
300 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
301 wake_up(&mddev->sb_wait);
303 return BLK_QC_T_NONE;
306 /* mddev_suspend makes sure no new requests are submitted
307 * to the device, and that any requests that have been submitted
308 * are completely handled.
309 * Once mddev_detach() is called and completes, the module will be
312 void mddev_suspend(struct mddev *mddev)
314 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
315 if (mddev->suspended++)
318 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
319 mddev->pers->quiesce(mddev, 1);
321 del_timer_sync(&mddev->safemode_timer);
323 EXPORT_SYMBOL_GPL(mddev_suspend);
325 void mddev_resume(struct mddev *mddev)
327 if (--mddev->suspended)
329 wake_up(&mddev->sb_wait);
330 mddev->pers->quiesce(mddev, 0);
332 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
333 md_wakeup_thread(mddev->thread);
334 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
336 EXPORT_SYMBOL_GPL(mddev_resume);
338 int mddev_congested(struct mddev *mddev, int bits)
340 struct md_personality *pers = mddev->pers;
344 if (mddev->suspended)
346 else if (pers && pers->congested)
347 ret = pers->congested(mddev, bits);
351 EXPORT_SYMBOL_GPL(mddev_congested);
352 static int md_congested(void *data, int bits)
354 struct mddev *mddev = data;
355 return mddev_congested(mddev, bits);
359 * Generic flush handling for md
362 static void md_end_flush(struct bio *bio)
364 struct md_rdev *rdev = bio->bi_private;
365 struct mddev *mddev = rdev->mddev;
367 rdev_dec_pending(rdev, mddev);
369 if (atomic_dec_and_test(&mddev->flush_pending)) {
370 /* The pre-request flush has finished */
371 queue_work(md_wq, &mddev->flush_work);
376 static void md_submit_flush_data(struct work_struct *ws);
378 static void submit_flushes(struct work_struct *ws)
380 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
381 struct md_rdev *rdev;
383 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
384 atomic_set(&mddev->flush_pending, 1);
386 rdev_for_each_rcu(rdev, mddev)
387 if (rdev->raid_disk >= 0 &&
388 !test_bit(Faulty, &rdev->flags)) {
389 /* Take two references, one is dropped
390 * when request finishes, one after
391 * we reclaim rcu_read_lock
394 atomic_inc(&rdev->nr_pending);
395 atomic_inc(&rdev->nr_pending);
397 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
398 bi->bi_end_io = md_end_flush;
399 bi->bi_private = rdev;
400 bi->bi_bdev = rdev->bdev;
401 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
402 atomic_inc(&mddev->flush_pending);
405 rdev_dec_pending(rdev, mddev);
408 if (atomic_dec_and_test(&mddev->flush_pending))
409 queue_work(md_wq, &mddev->flush_work);
412 static void md_submit_flush_data(struct work_struct *ws)
414 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
415 struct bio *bio = mddev->flush_bio;
417 if (bio->bi_iter.bi_size == 0)
418 /* an empty barrier - all done */
421 bio->bi_opf &= ~REQ_PREFLUSH;
422 mddev->pers->make_request(mddev, bio);
425 mddev->flush_bio = NULL;
426 wake_up(&mddev->sb_wait);
429 void md_flush_request(struct mddev *mddev, struct bio *bio)
431 spin_lock_irq(&mddev->lock);
432 wait_event_lock_irq(mddev->sb_wait,
435 mddev->flush_bio = bio;
436 spin_unlock_irq(&mddev->lock);
438 INIT_WORK(&mddev->flush_work, submit_flushes);
439 queue_work(md_wq, &mddev->flush_work);
441 EXPORT_SYMBOL(md_flush_request);
443 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
445 struct mddev *mddev = cb->data;
446 md_wakeup_thread(mddev->thread);
449 EXPORT_SYMBOL(md_unplug);
451 static inline struct mddev *mddev_get(struct mddev *mddev)
453 atomic_inc(&mddev->active);
457 static void mddev_delayed_delete(struct work_struct *ws);
459 static void mddev_put(struct mddev *mddev)
461 struct bio_set *bs = NULL;
463 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
465 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
466 mddev->ctime == 0 && !mddev->hold_active) {
467 /* Array is not configured at all, and not held active,
469 list_del_init(&mddev->all_mddevs);
471 mddev->bio_set = NULL;
472 if (mddev->gendisk) {
473 /* We did a probe so need to clean up. Call
474 * queue_work inside the spinlock so that
475 * flush_workqueue() after mddev_find will
476 * succeed in waiting for the work to be done.
478 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
479 queue_work(md_misc_wq, &mddev->del_work);
483 spin_unlock(&all_mddevs_lock);
488 static void md_safemode_timeout(unsigned long data);
490 void mddev_init(struct mddev *mddev)
492 mutex_init(&mddev->open_mutex);
493 mutex_init(&mddev->reconfig_mutex);
494 mutex_init(&mddev->bitmap_info.mutex);
495 INIT_LIST_HEAD(&mddev->disks);
496 INIT_LIST_HEAD(&mddev->all_mddevs);
497 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
498 (unsigned long) mddev);
499 atomic_set(&mddev->active, 1);
500 atomic_set(&mddev->openers, 0);
501 atomic_set(&mddev->active_io, 0);
502 spin_lock_init(&mddev->lock);
503 atomic_set(&mddev->flush_pending, 0);
504 init_waitqueue_head(&mddev->sb_wait);
505 init_waitqueue_head(&mddev->recovery_wait);
506 mddev->reshape_position = MaxSector;
507 mddev->reshape_backwards = 0;
508 mddev->last_sync_action = "none";
509 mddev->resync_min = 0;
510 mddev->resync_max = MaxSector;
511 mddev->level = LEVEL_NONE;
513 EXPORT_SYMBOL_GPL(mddev_init);
515 static struct mddev *mddev_find(dev_t unit)
517 struct mddev *mddev, *new = NULL;
519 if (unit && MAJOR(unit) != MD_MAJOR)
520 unit &= ~((1<<MdpMinorShift)-1);
523 spin_lock(&all_mddevs_lock);
526 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
527 if (mddev->unit == unit) {
529 spin_unlock(&all_mddevs_lock);
535 list_add(&new->all_mddevs, &all_mddevs);
536 spin_unlock(&all_mddevs_lock);
537 new->hold_active = UNTIL_IOCTL;
541 /* find an unused unit number */
542 static int next_minor = 512;
543 int start = next_minor;
547 dev = MKDEV(MD_MAJOR, next_minor);
549 if (next_minor > MINORMASK)
551 if (next_minor == start) {
552 /* Oh dear, all in use. */
553 spin_unlock(&all_mddevs_lock);
559 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
560 if (mddev->unit == dev) {
566 new->md_minor = MINOR(dev);
567 new->hold_active = UNTIL_STOP;
568 list_add(&new->all_mddevs, &all_mddevs);
569 spin_unlock(&all_mddevs_lock);
572 spin_unlock(&all_mddevs_lock);
574 new = kzalloc(sizeof(*new), GFP_KERNEL);
579 if (MAJOR(unit) == MD_MAJOR)
580 new->md_minor = MINOR(unit);
582 new->md_minor = MINOR(unit) >> MdpMinorShift;
589 static struct attribute_group md_redundancy_group;
591 void mddev_unlock(struct mddev *mddev)
593 if (mddev->to_remove) {
594 /* These cannot be removed under reconfig_mutex as
595 * an access to the files will try to take reconfig_mutex
596 * while holding the file unremovable, which leads to
598 * So hold set sysfs_active while the remove in happeing,
599 * and anything else which might set ->to_remove or my
600 * otherwise change the sysfs namespace will fail with
601 * -EBUSY if sysfs_active is still set.
602 * We set sysfs_active under reconfig_mutex and elsewhere
603 * test it under the same mutex to ensure its correct value
606 struct attribute_group *to_remove = mddev->to_remove;
607 mddev->to_remove = NULL;
608 mddev->sysfs_active = 1;
609 mutex_unlock(&mddev->reconfig_mutex);
611 if (mddev->kobj.sd) {
612 if (to_remove != &md_redundancy_group)
613 sysfs_remove_group(&mddev->kobj, to_remove);
614 if (mddev->pers == NULL ||
615 mddev->pers->sync_request == NULL) {
616 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
617 if (mddev->sysfs_action)
618 sysfs_put(mddev->sysfs_action);
619 mddev->sysfs_action = NULL;
622 mddev->sysfs_active = 0;
624 mutex_unlock(&mddev->reconfig_mutex);
626 /* As we've dropped the mutex we need a spinlock to
627 * make sure the thread doesn't disappear
629 spin_lock(&pers_lock);
630 md_wakeup_thread(mddev->thread);
631 spin_unlock(&pers_lock);
633 EXPORT_SYMBOL_GPL(mddev_unlock);
635 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
637 struct md_rdev *rdev;
639 rdev_for_each_rcu(rdev, mddev)
640 if (rdev->desc_nr == nr)
645 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
647 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
649 struct md_rdev *rdev;
651 rdev_for_each(rdev, mddev)
652 if (rdev->bdev->bd_dev == dev)
658 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
660 struct md_rdev *rdev;
662 rdev_for_each_rcu(rdev, mddev)
663 if (rdev->bdev->bd_dev == dev)
669 static struct md_personality *find_pers(int level, char *clevel)
671 struct md_personality *pers;
672 list_for_each_entry(pers, &pers_list, list) {
673 if (level != LEVEL_NONE && pers->level == level)
675 if (strcmp(pers->name, clevel)==0)
681 /* return the offset of the super block in 512byte sectors */
682 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
684 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
685 return MD_NEW_SIZE_SECTORS(num_sectors);
688 static int alloc_disk_sb(struct md_rdev *rdev)
690 rdev->sb_page = alloc_page(GFP_KERNEL);
696 void md_rdev_clear(struct md_rdev *rdev)
699 put_page(rdev->sb_page);
701 rdev->sb_page = NULL;
706 put_page(rdev->bb_page);
707 rdev->bb_page = NULL;
709 badblocks_exit(&rdev->badblocks);
711 EXPORT_SYMBOL_GPL(md_rdev_clear);
713 static void super_written(struct bio *bio)
715 struct md_rdev *rdev = bio->bi_private;
716 struct mddev *mddev = rdev->mddev;
719 pr_err("md: super_written gets error=%d\n", bio->bi_error);
720 md_error(mddev, rdev);
721 if (!test_bit(Faulty, &rdev->flags)
722 && (bio->bi_opf & MD_FAILFAST)) {
723 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
724 set_bit(LastDev, &rdev->flags);
727 clear_bit(LastDev, &rdev->flags);
729 if (atomic_dec_and_test(&mddev->pending_writes))
730 wake_up(&mddev->sb_wait);
731 rdev_dec_pending(rdev, mddev);
735 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
736 sector_t sector, int size, struct page *page)
738 /* write first size bytes of page to sector of rdev
739 * Increment mddev->pending_writes before returning
740 * and decrement it on completion, waking up sb_wait
741 * if zero is reached.
742 * If an error occurred, call md_error
747 if (test_bit(Faulty, &rdev->flags))
750 bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
752 atomic_inc(&rdev->nr_pending);
754 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
755 bio->bi_iter.bi_sector = sector;
756 bio_add_page(bio, page, size, 0);
757 bio->bi_private = rdev;
758 bio->bi_end_io = super_written;
760 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
761 test_bit(FailFast, &rdev->flags) &&
762 !test_bit(LastDev, &rdev->flags))
764 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_FUA | ff;
766 atomic_inc(&mddev->pending_writes);
770 int md_super_wait(struct mddev *mddev)
772 /* wait for all superblock writes that were scheduled to complete */
773 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
774 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
779 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
780 struct page *page, int op, int op_flags, bool metadata_op)
782 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
785 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
786 rdev->meta_bdev : rdev->bdev;
787 bio_set_op_attrs(bio, op, op_flags);
789 bio->bi_iter.bi_sector = sector + rdev->sb_start;
790 else if (rdev->mddev->reshape_position != MaxSector &&
791 (rdev->mddev->reshape_backwards ==
792 (sector >= rdev->mddev->reshape_position)))
793 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
795 bio->bi_iter.bi_sector = sector + rdev->data_offset;
796 bio_add_page(bio, page, size, 0);
798 submit_bio_wait(bio);
800 ret = !bio->bi_error;
804 EXPORT_SYMBOL_GPL(sync_page_io);
806 static int read_disk_sb(struct md_rdev *rdev, int size)
808 char b[BDEVNAME_SIZE];
813 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
819 pr_err("md: disabled device %s, could not read superblock.\n",
820 bdevname(rdev->bdev,b));
824 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
826 return sb1->set_uuid0 == sb2->set_uuid0 &&
827 sb1->set_uuid1 == sb2->set_uuid1 &&
828 sb1->set_uuid2 == sb2->set_uuid2 &&
829 sb1->set_uuid3 == sb2->set_uuid3;
832 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
835 mdp_super_t *tmp1, *tmp2;
837 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
838 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
840 if (!tmp1 || !tmp2) {
849 * nr_disks is not constant
854 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
861 static u32 md_csum_fold(u32 csum)
863 csum = (csum & 0xffff) + (csum >> 16);
864 return (csum & 0xffff) + (csum >> 16);
867 static unsigned int calc_sb_csum(mdp_super_t *sb)
870 u32 *sb32 = (u32*)sb;
872 unsigned int disk_csum, csum;
874 disk_csum = sb->sb_csum;
877 for (i = 0; i < MD_SB_BYTES/4 ; i++)
879 csum = (newcsum & 0xffffffff) + (newcsum>>32);
882 /* This used to use csum_partial, which was wrong for several
883 * reasons including that different results are returned on
884 * different architectures. It isn't critical that we get exactly
885 * the same return value as before (we always csum_fold before
886 * testing, and that removes any differences). However as we
887 * know that csum_partial always returned a 16bit value on
888 * alphas, do a fold to maximise conformity to previous behaviour.
890 sb->sb_csum = md_csum_fold(disk_csum);
892 sb->sb_csum = disk_csum;
898 * Handle superblock details.
899 * We want to be able to handle multiple superblock formats
900 * so we have a common interface to them all, and an array of
901 * different handlers.
902 * We rely on user-space to write the initial superblock, and support
903 * reading and updating of superblocks.
904 * Interface methods are:
905 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
906 * loads and validates a superblock on dev.
907 * if refdev != NULL, compare superblocks on both devices
909 * 0 - dev has a superblock that is compatible with refdev
910 * 1 - dev has a superblock that is compatible and newer than refdev
911 * so dev should be used as the refdev in future
912 * -EINVAL superblock incompatible or invalid
913 * -othererror e.g. -EIO
915 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
916 * Verify that dev is acceptable into mddev.
917 * The first time, mddev->raid_disks will be 0, and data from
918 * dev should be merged in. Subsequent calls check that dev
919 * is new enough. Return 0 or -EINVAL
921 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
922 * Update the superblock for rdev with data in mddev
923 * This does not write to disc.
929 struct module *owner;
930 int (*load_super)(struct md_rdev *rdev,
931 struct md_rdev *refdev,
933 int (*validate_super)(struct mddev *mddev,
934 struct md_rdev *rdev);
935 void (*sync_super)(struct mddev *mddev,
936 struct md_rdev *rdev);
937 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
938 sector_t num_sectors);
939 int (*allow_new_offset)(struct md_rdev *rdev,
940 unsigned long long new_offset);
944 * Check that the given mddev has no bitmap.
946 * This function is called from the run method of all personalities that do not
947 * support bitmaps. It prints an error message and returns non-zero if mddev
948 * has a bitmap. Otherwise, it returns 0.
951 int md_check_no_bitmap(struct mddev *mddev)
953 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
955 pr_warn("%s: bitmaps are not supported for %s\n",
956 mdname(mddev), mddev->pers->name);
959 EXPORT_SYMBOL(md_check_no_bitmap);
962 * load_super for 0.90.0
964 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
966 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
971 * Calculate the position of the superblock (512byte sectors),
972 * it's at the end of the disk.
974 * It also happens to be a multiple of 4Kb.
976 rdev->sb_start = calc_dev_sboffset(rdev);
978 ret = read_disk_sb(rdev, MD_SB_BYTES);
984 bdevname(rdev->bdev, b);
985 sb = page_address(rdev->sb_page);
987 if (sb->md_magic != MD_SB_MAGIC) {
988 pr_warn("md: invalid raid superblock magic on %s\n", b);
992 if (sb->major_version != 0 ||
993 sb->minor_version < 90 ||
994 sb->minor_version > 91) {
995 pr_warn("Bad version number %d.%d on %s\n",
996 sb->major_version, sb->minor_version, b);
1000 if (sb->raid_disks <= 0)
1003 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1004 pr_warn("md: invalid superblock checksum on %s\n", b);
1008 rdev->preferred_minor = sb->md_minor;
1009 rdev->data_offset = 0;
1010 rdev->new_data_offset = 0;
1011 rdev->sb_size = MD_SB_BYTES;
1012 rdev->badblocks.shift = -1;
1014 if (sb->level == LEVEL_MULTIPATH)
1017 rdev->desc_nr = sb->this_disk.number;
1023 mdp_super_t *refsb = page_address(refdev->sb_page);
1024 if (!uuid_equal(refsb, sb)) {
1025 pr_warn("md: %s has different UUID to %s\n",
1026 b, bdevname(refdev->bdev,b2));
1029 if (!sb_equal(refsb, sb)) {
1030 pr_warn("md: %s has same UUID but different superblock to %s\n",
1031 b, bdevname(refdev->bdev, b2));
1035 ev2 = md_event(refsb);
1041 rdev->sectors = rdev->sb_start;
1042 /* Limit to 4TB as metadata cannot record more than that.
1043 * (not needed for Linear and RAID0 as metadata doesn't
1046 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1048 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1050 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1051 /* "this cannot possibly happen" ... */
1059 * validate_super for 0.90.0
1061 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1064 mdp_super_t *sb = page_address(rdev->sb_page);
1065 __u64 ev1 = md_event(sb);
1067 rdev->raid_disk = -1;
1068 clear_bit(Faulty, &rdev->flags);
1069 clear_bit(In_sync, &rdev->flags);
1070 clear_bit(Bitmap_sync, &rdev->flags);
1071 clear_bit(WriteMostly, &rdev->flags);
1073 if (mddev->raid_disks == 0) {
1074 mddev->major_version = 0;
1075 mddev->minor_version = sb->minor_version;
1076 mddev->patch_version = sb->patch_version;
1077 mddev->external = 0;
1078 mddev->chunk_sectors = sb->chunk_size >> 9;
1079 mddev->ctime = sb->ctime;
1080 mddev->utime = sb->utime;
1081 mddev->level = sb->level;
1082 mddev->clevel[0] = 0;
1083 mddev->layout = sb->layout;
1084 mddev->raid_disks = sb->raid_disks;
1085 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1086 mddev->events = ev1;
1087 mddev->bitmap_info.offset = 0;
1088 mddev->bitmap_info.space = 0;
1089 /* bitmap can use 60 K after the 4K superblocks */
1090 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1091 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1092 mddev->reshape_backwards = 0;
1094 if (mddev->minor_version >= 91) {
1095 mddev->reshape_position = sb->reshape_position;
1096 mddev->delta_disks = sb->delta_disks;
1097 mddev->new_level = sb->new_level;
1098 mddev->new_layout = sb->new_layout;
1099 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1100 if (mddev->delta_disks < 0)
1101 mddev->reshape_backwards = 1;
1103 mddev->reshape_position = MaxSector;
1104 mddev->delta_disks = 0;
1105 mddev->new_level = mddev->level;
1106 mddev->new_layout = mddev->layout;
1107 mddev->new_chunk_sectors = mddev->chunk_sectors;
1110 if (sb->state & (1<<MD_SB_CLEAN))
1111 mddev->recovery_cp = MaxSector;
1113 if (sb->events_hi == sb->cp_events_hi &&
1114 sb->events_lo == sb->cp_events_lo) {
1115 mddev->recovery_cp = sb->recovery_cp;
1117 mddev->recovery_cp = 0;
1120 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1121 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1122 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1123 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1125 mddev->max_disks = MD_SB_DISKS;
1127 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1128 mddev->bitmap_info.file == NULL) {
1129 mddev->bitmap_info.offset =
1130 mddev->bitmap_info.default_offset;
1131 mddev->bitmap_info.space =
1132 mddev->bitmap_info.default_space;
1135 } else if (mddev->pers == NULL) {
1136 /* Insist on good event counter while assembling, except
1137 * for spares (which don't need an event count) */
1139 if (sb->disks[rdev->desc_nr].state & (
1140 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1141 if (ev1 < mddev->events)
1143 } else if (mddev->bitmap) {
1144 /* if adding to array with a bitmap, then we can accept an
1145 * older device ... but not too old.
1147 if (ev1 < mddev->bitmap->events_cleared)
1149 if (ev1 < mddev->events)
1150 set_bit(Bitmap_sync, &rdev->flags);
1152 if (ev1 < mddev->events)
1153 /* just a hot-add of a new device, leave raid_disk at -1 */
1157 if (mddev->level != LEVEL_MULTIPATH) {
1158 desc = sb->disks + rdev->desc_nr;
1160 if (desc->state & (1<<MD_DISK_FAULTY))
1161 set_bit(Faulty, &rdev->flags);
1162 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1163 desc->raid_disk < mddev->raid_disks */) {
1164 set_bit(In_sync, &rdev->flags);
1165 rdev->raid_disk = desc->raid_disk;
1166 rdev->saved_raid_disk = desc->raid_disk;
1167 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1168 /* active but not in sync implies recovery up to
1169 * reshape position. We don't know exactly where
1170 * that is, so set to zero for now */
1171 if (mddev->minor_version >= 91) {
1172 rdev->recovery_offset = 0;
1173 rdev->raid_disk = desc->raid_disk;
1176 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1177 set_bit(WriteMostly, &rdev->flags);
1178 if (desc->state & (1<<MD_DISK_FAILFAST))
1179 set_bit(FailFast, &rdev->flags);
1180 } else /* MULTIPATH are always insync */
1181 set_bit(In_sync, &rdev->flags);
1186 * sync_super for 0.90.0
1188 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1191 struct md_rdev *rdev2;
1192 int next_spare = mddev->raid_disks;
1194 /* make rdev->sb match mddev data..
1197 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1198 * 3/ any empty disks < next_spare become removed
1200 * disks[0] gets initialised to REMOVED because
1201 * we cannot be sure from other fields if it has
1202 * been initialised or not.
1205 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1207 rdev->sb_size = MD_SB_BYTES;
1209 sb = page_address(rdev->sb_page);
1211 memset(sb, 0, sizeof(*sb));
1213 sb->md_magic = MD_SB_MAGIC;
1214 sb->major_version = mddev->major_version;
1215 sb->patch_version = mddev->patch_version;
1216 sb->gvalid_words = 0; /* ignored */
1217 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1218 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1219 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1220 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1222 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1223 sb->level = mddev->level;
1224 sb->size = mddev->dev_sectors / 2;
1225 sb->raid_disks = mddev->raid_disks;
1226 sb->md_minor = mddev->md_minor;
1227 sb->not_persistent = 0;
1228 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1230 sb->events_hi = (mddev->events>>32);
1231 sb->events_lo = (u32)mddev->events;
1233 if (mddev->reshape_position == MaxSector)
1234 sb->minor_version = 90;
1236 sb->minor_version = 91;
1237 sb->reshape_position = mddev->reshape_position;
1238 sb->new_level = mddev->new_level;
1239 sb->delta_disks = mddev->delta_disks;
1240 sb->new_layout = mddev->new_layout;
1241 sb->new_chunk = mddev->new_chunk_sectors << 9;
1243 mddev->minor_version = sb->minor_version;
1246 sb->recovery_cp = mddev->recovery_cp;
1247 sb->cp_events_hi = (mddev->events>>32);
1248 sb->cp_events_lo = (u32)mddev->events;
1249 if (mddev->recovery_cp == MaxSector)
1250 sb->state = (1<< MD_SB_CLEAN);
1252 sb->recovery_cp = 0;
1254 sb->layout = mddev->layout;
1255 sb->chunk_size = mddev->chunk_sectors << 9;
1257 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1258 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1260 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1261 rdev_for_each(rdev2, mddev) {
1264 int is_active = test_bit(In_sync, &rdev2->flags);
1266 if (rdev2->raid_disk >= 0 &&
1267 sb->minor_version >= 91)
1268 /* we have nowhere to store the recovery_offset,
1269 * but if it is not below the reshape_position,
1270 * we can piggy-back on that.
1273 if (rdev2->raid_disk < 0 ||
1274 test_bit(Faulty, &rdev2->flags))
1277 desc_nr = rdev2->raid_disk;
1279 desc_nr = next_spare++;
1280 rdev2->desc_nr = desc_nr;
1281 d = &sb->disks[rdev2->desc_nr];
1283 d->number = rdev2->desc_nr;
1284 d->major = MAJOR(rdev2->bdev->bd_dev);
1285 d->minor = MINOR(rdev2->bdev->bd_dev);
1287 d->raid_disk = rdev2->raid_disk;
1289 d->raid_disk = rdev2->desc_nr; /* compatibility */
1290 if (test_bit(Faulty, &rdev2->flags))
1291 d->state = (1<<MD_DISK_FAULTY);
1292 else if (is_active) {
1293 d->state = (1<<MD_DISK_ACTIVE);
1294 if (test_bit(In_sync, &rdev2->flags))
1295 d->state |= (1<<MD_DISK_SYNC);
1303 if (test_bit(WriteMostly, &rdev2->flags))
1304 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1305 if (test_bit(FailFast, &rdev2->flags))
1306 d->state |= (1<<MD_DISK_FAILFAST);
1308 /* now set the "removed" and "faulty" bits on any missing devices */
1309 for (i=0 ; i < mddev->raid_disks ; i++) {
1310 mdp_disk_t *d = &sb->disks[i];
1311 if (d->state == 0 && d->number == 0) {
1314 d->state = (1<<MD_DISK_REMOVED);
1315 d->state |= (1<<MD_DISK_FAULTY);
1319 sb->nr_disks = nr_disks;
1320 sb->active_disks = active;
1321 sb->working_disks = working;
1322 sb->failed_disks = failed;
1323 sb->spare_disks = spare;
1325 sb->this_disk = sb->disks[rdev->desc_nr];
1326 sb->sb_csum = calc_sb_csum(sb);
1330 * rdev_size_change for 0.90.0
1332 static unsigned long long
1333 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1335 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1336 return 0; /* component must fit device */
1337 if (rdev->mddev->bitmap_info.offset)
1338 return 0; /* can't move bitmap */
1339 rdev->sb_start = calc_dev_sboffset(rdev);
1340 if (!num_sectors || num_sectors > rdev->sb_start)
1341 num_sectors = rdev->sb_start;
1342 /* Limit to 4TB as metadata cannot record more than that.
1343 * 4TB == 2^32 KB, or 2*2^32 sectors.
1345 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1346 rdev->mddev->level >= 1)
1347 num_sectors = (sector_t)(2ULL << 32) - 2;
1349 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1351 } while (md_super_wait(rdev->mddev) < 0);
1356 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1358 /* non-zero offset changes not possible with v0.90 */
1359 return new_offset == 0;
1363 * version 1 superblock
1366 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1370 unsigned long long newcsum;
1371 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1372 __le32 *isuper = (__le32*)sb;
1374 disk_csum = sb->sb_csum;
1377 for (; size >= 4; size -= 4)
1378 newcsum += le32_to_cpu(*isuper++);
1381 newcsum += le16_to_cpu(*(__le16*) isuper);
1383 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1384 sb->sb_csum = disk_csum;
1385 return cpu_to_le32(csum);
1388 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1390 struct mdp_superblock_1 *sb;
1394 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1398 * Calculate the position of the superblock in 512byte sectors.
1399 * It is always aligned to a 4K boundary and
1400 * depeding on minor_version, it can be:
1401 * 0: At least 8K, but less than 12K, from end of device
1402 * 1: At start of device
1403 * 2: 4K from start of device.
1405 switch(minor_version) {
1407 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1409 sb_start &= ~(sector_t)(4*2-1);
1420 rdev->sb_start = sb_start;
1422 /* superblock is rarely larger than 1K, but it can be larger,
1423 * and it is safe to read 4k, so we do that
1425 ret = read_disk_sb(rdev, 4096);
1426 if (ret) return ret;
1428 sb = page_address(rdev->sb_page);
1430 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1431 sb->major_version != cpu_to_le32(1) ||
1432 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1433 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1434 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1437 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1438 pr_warn("md: invalid superblock checksum on %s\n",
1439 bdevname(rdev->bdev,b));
1442 if (le64_to_cpu(sb->data_size) < 10) {
1443 pr_warn("md: data_size too small on %s\n",
1444 bdevname(rdev->bdev,b));
1449 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1450 /* Some padding is non-zero, might be a new feature */
1453 rdev->preferred_minor = 0xffff;
1454 rdev->data_offset = le64_to_cpu(sb->data_offset);
1455 rdev->new_data_offset = rdev->data_offset;
1456 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1457 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1458 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1459 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1461 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1462 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1463 if (rdev->sb_size & bmask)
1464 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1467 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1470 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1473 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1476 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1478 if (!rdev->bb_page) {
1479 rdev->bb_page = alloc_page(GFP_KERNEL);
1483 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1484 rdev->badblocks.count == 0) {
1485 /* need to load the bad block list.
1486 * Currently we limit it to one page.
1492 int sectors = le16_to_cpu(sb->bblog_size);
1493 if (sectors > (PAGE_SIZE / 512))
1495 offset = le32_to_cpu(sb->bblog_offset);
1498 bb_sector = (long long)offset;
1499 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1500 rdev->bb_page, REQ_OP_READ, 0, true))
1502 bbp = (u64 *)page_address(rdev->bb_page);
1503 rdev->badblocks.shift = sb->bblog_shift;
1504 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1505 u64 bb = le64_to_cpu(*bbp);
1506 int count = bb & (0x3ff);
1507 u64 sector = bb >> 10;
1508 sector <<= sb->bblog_shift;
1509 count <<= sb->bblog_shift;
1512 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1515 } else if (sb->bblog_offset != 0)
1516 rdev->badblocks.shift = 0;
1522 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1524 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1525 sb->level != refsb->level ||
1526 sb->layout != refsb->layout ||
1527 sb->chunksize != refsb->chunksize) {
1528 pr_warn("md: %s has strangely different superblock to %s\n",
1529 bdevname(rdev->bdev,b),
1530 bdevname(refdev->bdev,b2));
1533 ev1 = le64_to_cpu(sb->events);
1534 ev2 = le64_to_cpu(refsb->events);
1541 if (minor_version) {
1542 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1543 sectors -= rdev->data_offset;
1545 sectors = rdev->sb_start;
1546 if (sectors < le64_to_cpu(sb->data_size))
1548 rdev->sectors = le64_to_cpu(sb->data_size);
1552 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1554 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1555 __u64 ev1 = le64_to_cpu(sb->events);
1557 rdev->raid_disk = -1;
1558 clear_bit(Faulty, &rdev->flags);
1559 clear_bit(In_sync, &rdev->flags);
1560 clear_bit(Bitmap_sync, &rdev->flags);
1561 clear_bit(WriteMostly, &rdev->flags);
1563 if (mddev->raid_disks == 0) {
1564 mddev->major_version = 1;
1565 mddev->patch_version = 0;
1566 mddev->external = 0;
1567 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1568 mddev->ctime = le64_to_cpu(sb->ctime);
1569 mddev->utime = le64_to_cpu(sb->utime);
1570 mddev->level = le32_to_cpu(sb->level);
1571 mddev->clevel[0] = 0;
1572 mddev->layout = le32_to_cpu(sb->layout);
1573 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1574 mddev->dev_sectors = le64_to_cpu(sb->size);
1575 mddev->events = ev1;
1576 mddev->bitmap_info.offset = 0;
1577 mddev->bitmap_info.space = 0;
1578 /* Default location for bitmap is 1K after superblock
1579 * using 3K - total of 4K
1581 mddev->bitmap_info.default_offset = 1024 >> 9;
1582 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1583 mddev->reshape_backwards = 0;
1585 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1586 memcpy(mddev->uuid, sb->set_uuid, 16);
1588 mddev->max_disks = (4096-256)/2;
1590 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1591 mddev->bitmap_info.file == NULL) {
1592 mddev->bitmap_info.offset =
1593 (__s32)le32_to_cpu(sb->bitmap_offset);
1594 /* Metadata doesn't record how much space is available.
1595 * For 1.0, we assume we can use up to the superblock
1596 * if before, else to 4K beyond superblock.
1597 * For others, assume no change is possible.
1599 if (mddev->minor_version > 0)
1600 mddev->bitmap_info.space = 0;
1601 else if (mddev->bitmap_info.offset > 0)
1602 mddev->bitmap_info.space =
1603 8 - mddev->bitmap_info.offset;
1605 mddev->bitmap_info.space =
1606 -mddev->bitmap_info.offset;
1609 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1610 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1611 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1612 mddev->new_level = le32_to_cpu(sb->new_level);
1613 mddev->new_layout = le32_to_cpu(sb->new_layout);
1614 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1615 if (mddev->delta_disks < 0 ||
1616 (mddev->delta_disks == 0 &&
1617 (le32_to_cpu(sb->feature_map)
1618 & MD_FEATURE_RESHAPE_BACKWARDS)))
1619 mddev->reshape_backwards = 1;
1621 mddev->reshape_position = MaxSector;
1622 mddev->delta_disks = 0;
1623 mddev->new_level = mddev->level;
1624 mddev->new_layout = mddev->layout;
1625 mddev->new_chunk_sectors = mddev->chunk_sectors;
1628 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1629 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1630 } else if (mddev->pers == NULL) {
1631 /* Insist of good event counter while assembling, except for
1632 * spares (which don't need an event count) */
1634 if (rdev->desc_nr >= 0 &&
1635 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1636 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1637 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1638 if (ev1 < mddev->events)
1640 } else if (mddev->bitmap) {
1641 /* If adding to array with a bitmap, then we can accept an
1642 * older device, but not too old.
1644 if (ev1 < mddev->bitmap->events_cleared)
1646 if (ev1 < mddev->events)
1647 set_bit(Bitmap_sync, &rdev->flags);
1649 if (ev1 < mddev->events)
1650 /* just a hot-add of a new device, leave raid_disk at -1 */
1653 if (mddev->level != LEVEL_MULTIPATH) {
1655 if (rdev->desc_nr < 0 ||
1656 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1657 role = MD_DISK_ROLE_SPARE;
1660 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1662 case MD_DISK_ROLE_SPARE: /* spare */
1664 case MD_DISK_ROLE_FAULTY: /* faulty */
1665 set_bit(Faulty, &rdev->flags);
1667 case MD_DISK_ROLE_JOURNAL: /* journal device */
1668 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1669 /* journal device without journal feature */
1670 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1673 set_bit(Journal, &rdev->flags);
1674 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1675 rdev->raid_disk = 0;
1678 rdev->saved_raid_disk = role;
1679 if ((le32_to_cpu(sb->feature_map) &
1680 MD_FEATURE_RECOVERY_OFFSET)) {
1681 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1682 if (!(le32_to_cpu(sb->feature_map) &
1683 MD_FEATURE_RECOVERY_BITMAP))
1684 rdev->saved_raid_disk = -1;
1686 set_bit(In_sync, &rdev->flags);
1687 rdev->raid_disk = role;
1690 if (sb->devflags & WriteMostly1)
1691 set_bit(WriteMostly, &rdev->flags);
1692 if (sb->devflags & FailFast1)
1693 set_bit(FailFast, &rdev->flags);
1694 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1695 set_bit(Replacement, &rdev->flags);
1696 } else /* MULTIPATH are always insync */
1697 set_bit(In_sync, &rdev->flags);
1702 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1704 struct mdp_superblock_1 *sb;
1705 struct md_rdev *rdev2;
1707 /* make rdev->sb match mddev and rdev data. */
1709 sb = page_address(rdev->sb_page);
1711 sb->feature_map = 0;
1713 sb->recovery_offset = cpu_to_le64(0);
1714 memset(sb->pad3, 0, sizeof(sb->pad3));
1716 sb->utime = cpu_to_le64((__u64)mddev->utime);
1717 sb->events = cpu_to_le64(mddev->events);
1719 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1720 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1721 sb->resync_offset = cpu_to_le64(MaxSector);
1723 sb->resync_offset = cpu_to_le64(0);
1725 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1727 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1728 sb->size = cpu_to_le64(mddev->dev_sectors);
1729 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1730 sb->level = cpu_to_le32(mddev->level);
1731 sb->layout = cpu_to_le32(mddev->layout);
1732 if (test_bit(FailFast, &rdev->flags))
1733 sb->devflags |= FailFast1;
1735 sb->devflags &= ~FailFast1;
1737 if (test_bit(WriteMostly, &rdev->flags))
1738 sb->devflags |= WriteMostly1;
1740 sb->devflags &= ~WriteMostly1;
1741 sb->data_offset = cpu_to_le64(rdev->data_offset);
1742 sb->data_size = cpu_to_le64(rdev->sectors);
1744 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1745 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1746 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1749 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1750 !test_bit(In_sync, &rdev->flags)) {
1752 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1753 sb->recovery_offset =
1754 cpu_to_le64(rdev->recovery_offset);
1755 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1757 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1759 /* Note: recovery_offset and journal_tail share space */
1760 if (test_bit(Journal, &rdev->flags))
1761 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1762 if (test_bit(Replacement, &rdev->flags))
1764 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1766 if (mddev->reshape_position != MaxSector) {
1767 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1768 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1769 sb->new_layout = cpu_to_le32(mddev->new_layout);
1770 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1771 sb->new_level = cpu_to_le32(mddev->new_level);
1772 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1773 if (mddev->delta_disks == 0 &&
1774 mddev->reshape_backwards)
1776 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1777 if (rdev->new_data_offset != rdev->data_offset) {
1779 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1780 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1781 - rdev->data_offset));
1785 if (mddev_is_clustered(mddev))
1786 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1788 if (rdev->badblocks.count == 0)
1789 /* Nothing to do for bad blocks*/ ;
1790 else if (sb->bblog_offset == 0)
1791 /* Cannot record bad blocks on this device */
1792 md_error(mddev, rdev);
1794 struct badblocks *bb = &rdev->badblocks;
1795 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1797 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1802 seq = read_seqbegin(&bb->lock);
1804 memset(bbp, 0xff, PAGE_SIZE);
1806 for (i = 0 ; i < bb->count ; i++) {
1807 u64 internal_bb = p[i];
1808 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1809 | BB_LEN(internal_bb));
1810 bbp[i] = cpu_to_le64(store_bb);
1813 if (read_seqretry(&bb->lock, seq))
1816 bb->sector = (rdev->sb_start +
1817 (int)le32_to_cpu(sb->bblog_offset));
1818 bb->size = le16_to_cpu(sb->bblog_size);
1823 rdev_for_each(rdev2, mddev)
1824 if (rdev2->desc_nr+1 > max_dev)
1825 max_dev = rdev2->desc_nr+1;
1827 if (max_dev > le32_to_cpu(sb->max_dev)) {
1829 sb->max_dev = cpu_to_le32(max_dev);
1830 rdev->sb_size = max_dev * 2 + 256;
1831 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1832 if (rdev->sb_size & bmask)
1833 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1835 max_dev = le32_to_cpu(sb->max_dev);
1837 for (i=0; i<max_dev;i++)
1838 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1840 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1841 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1843 rdev_for_each(rdev2, mddev) {
1845 if (test_bit(Faulty, &rdev2->flags))
1846 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1847 else if (test_bit(In_sync, &rdev2->flags))
1848 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1849 else if (test_bit(Journal, &rdev2->flags))
1850 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1851 else if (rdev2->raid_disk >= 0)
1852 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1854 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1857 sb->sb_csum = calc_sb_1_csum(sb);
1860 static unsigned long long
1861 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1863 struct mdp_superblock_1 *sb;
1864 sector_t max_sectors;
1865 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1866 return 0; /* component must fit device */
1867 if (rdev->data_offset != rdev->new_data_offset)
1868 return 0; /* too confusing */
1869 if (rdev->sb_start < rdev->data_offset) {
1870 /* minor versions 1 and 2; superblock before data */
1871 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1872 max_sectors -= rdev->data_offset;
1873 if (!num_sectors || num_sectors > max_sectors)
1874 num_sectors = max_sectors;
1875 } else if (rdev->mddev->bitmap_info.offset) {
1876 /* minor version 0 with bitmap we can't move */
1879 /* minor version 0; superblock after data */
1881 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1882 sb_start &= ~(sector_t)(4*2 - 1);
1883 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1884 if (!num_sectors || num_sectors > max_sectors)
1885 num_sectors = max_sectors;
1886 rdev->sb_start = sb_start;
1888 sb = page_address(rdev->sb_page);
1889 sb->data_size = cpu_to_le64(num_sectors);
1890 sb->super_offset = rdev->sb_start;
1891 sb->sb_csum = calc_sb_1_csum(sb);
1893 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1895 } while (md_super_wait(rdev->mddev) < 0);
1901 super_1_allow_new_offset(struct md_rdev *rdev,
1902 unsigned long long new_offset)
1904 /* All necessary checks on new >= old have been done */
1905 struct bitmap *bitmap;
1906 if (new_offset >= rdev->data_offset)
1909 /* with 1.0 metadata, there is no metadata to tread on
1910 * so we can always move back */
1911 if (rdev->mddev->minor_version == 0)
1914 /* otherwise we must be sure not to step on
1915 * any metadata, so stay:
1916 * 36K beyond start of superblock
1917 * beyond end of badblocks
1918 * beyond write-intent bitmap
1920 if (rdev->sb_start + (32+4)*2 > new_offset)
1922 bitmap = rdev->mddev->bitmap;
1923 if (bitmap && !rdev->mddev->bitmap_info.file &&
1924 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1925 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1927 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1933 static struct super_type super_types[] = {
1936 .owner = THIS_MODULE,
1937 .load_super = super_90_load,
1938 .validate_super = super_90_validate,
1939 .sync_super = super_90_sync,
1940 .rdev_size_change = super_90_rdev_size_change,
1941 .allow_new_offset = super_90_allow_new_offset,
1945 .owner = THIS_MODULE,
1946 .load_super = super_1_load,
1947 .validate_super = super_1_validate,
1948 .sync_super = super_1_sync,
1949 .rdev_size_change = super_1_rdev_size_change,
1950 .allow_new_offset = super_1_allow_new_offset,
1954 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1956 if (mddev->sync_super) {
1957 mddev->sync_super(mddev, rdev);
1961 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1963 super_types[mddev->major_version].sync_super(mddev, rdev);
1966 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1968 struct md_rdev *rdev, *rdev2;
1971 rdev_for_each_rcu(rdev, mddev1) {
1972 if (test_bit(Faulty, &rdev->flags) ||
1973 test_bit(Journal, &rdev->flags) ||
1974 rdev->raid_disk == -1)
1976 rdev_for_each_rcu(rdev2, mddev2) {
1977 if (test_bit(Faulty, &rdev2->flags) ||
1978 test_bit(Journal, &rdev2->flags) ||
1979 rdev2->raid_disk == -1)
1981 if (rdev->bdev->bd_contains ==
1982 rdev2->bdev->bd_contains) {
1992 static LIST_HEAD(pending_raid_disks);
1995 * Try to register data integrity profile for an mddev
1997 * This is called when an array is started and after a disk has been kicked
1998 * from the array. It only succeeds if all working and active component devices
1999 * are integrity capable with matching profiles.
2001 int md_integrity_register(struct mddev *mddev)
2003 struct md_rdev *rdev, *reference = NULL;
2005 if (list_empty(&mddev->disks))
2006 return 0; /* nothing to do */
2007 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2008 return 0; /* shouldn't register, or already is */
2009 rdev_for_each(rdev, mddev) {
2010 /* skip spares and non-functional disks */
2011 if (test_bit(Faulty, &rdev->flags))
2013 if (rdev->raid_disk < 0)
2016 /* Use the first rdev as the reference */
2020 /* does this rdev's profile match the reference profile? */
2021 if (blk_integrity_compare(reference->bdev->bd_disk,
2022 rdev->bdev->bd_disk) < 0)
2025 if (!reference || !bdev_get_integrity(reference->bdev))
2028 * All component devices are integrity capable and have matching
2029 * profiles, register the common profile for the md device.
2031 blk_integrity_register(mddev->gendisk,
2032 bdev_get_integrity(reference->bdev));
2034 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2035 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2036 pr_err("md: failed to create integrity pool for %s\n",
2042 EXPORT_SYMBOL(md_integrity_register);
2045 * Attempt to add an rdev, but only if it is consistent with the current
2048 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2050 struct blk_integrity *bi_rdev;
2051 struct blk_integrity *bi_mddev;
2052 char name[BDEVNAME_SIZE];
2054 if (!mddev->gendisk)
2057 bi_rdev = bdev_get_integrity(rdev->bdev);
2058 bi_mddev = blk_get_integrity(mddev->gendisk);
2060 if (!bi_mddev) /* nothing to do */
2063 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2064 pr_err("%s: incompatible integrity profile for %s\n",
2065 mdname(mddev), bdevname(rdev->bdev, name));
2071 EXPORT_SYMBOL(md_integrity_add_rdev);
2073 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2075 char b[BDEVNAME_SIZE];
2079 /* prevent duplicates */
2080 if (find_rdev(mddev, rdev->bdev->bd_dev))
2083 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2084 if (!test_bit(Journal, &rdev->flags) &&
2086 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2088 /* Cannot change size, so fail
2089 * If mddev->level <= 0, then we don't care
2090 * about aligning sizes (e.g. linear)
2092 if (mddev->level > 0)
2095 mddev->dev_sectors = rdev->sectors;
2098 /* Verify rdev->desc_nr is unique.
2099 * If it is -1, assign a free number, else
2100 * check number is not in use
2103 if (rdev->desc_nr < 0) {
2106 choice = mddev->raid_disks;
2107 while (md_find_rdev_nr_rcu(mddev, choice))
2109 rdev->desc_nr = choice;
2111 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2117 if (!test_bit(Journal, &rdev->flags) &&
2118 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2119 pr_warn("md: %s: array is limited to %d devices\n",
2120 mdname(mddev), mddev->max_disks);
2123 bdevname(rdev->bdev,b);
2124 strreplace(b, '/', '!');
2126 rdev->mddev = mddev;
2127 pr_debug("md: bind<%s>\n", b);
2129 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2132 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2133 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2134 /* failure here is OK */;
2135 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2137 list_add_rcu(&rdev->same_set, &mddev->disks);
2138 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2140 /* May as well allow recovery to be retried once */
2141 mddev->recovery_disabled++;
2146 pr_warn("md: failed to register dev-%s for %s\n",
2151 static void md_delayed_delete(struct work_struct *ws)
2153 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2154 kobject_del(&rdev->kobj);
2155 kobject_put(&rdev->kobj);
2158 static void unbind_rdev_from_array(struct md_rdev *rdev)
2160 char b[BDEVNAME_SIZE];
2162 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2163 list_del_rcu(&rdev->same_set);
2164 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2166 sysfs_remove_link(&rdev->kobj, "block");
2167 sysfs_put(rdev->sysfs_state);
2168 rdev->sysfs_state = NULL;
2169 rdev->badblocks.count = 0;
2170 /* We need to delay this, otherwise we can deadlock when
2171 * writing to 'remove' to "dev/state". We also need
2172 * to delay it due to rcu usage.
2175 INIT_WORK(&rdev->del_work, md_delayed_delete);
2176 kobject_get(&rdev->kobj);
2177 queue_work(md_misc_wq, &rdev->del_work);
2181 * prevent the device from being mounted, repartitioned or
2182 * otherwise reused by a RAID array (or any other kernel
2183 * subsystem), by bd_claiming the device.
2185 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2188 struct block_device *bdev;
2189 char b[BDEVNAME_SIZE];
2191 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2192 shared ? (struct md_rdev *)lock_rdev : rdev);
2194 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2195 return PTR_ERR(bdev);
2201 static void unlock_rdev(struct md_rdev *rdev)
2203 struct block_device *bdev = rdev->bdev;
2205 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2208 void md_autodetect_dev(dev_t dev);
2210 static void export_rdev(struct md_rdev *rdev)
2212 char b[BDEVNAME_SIZE];
2214 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2215 md_rdev_clear(rdev);
2217 if (test_bit(AutoDetected, &rdev->flags))
2218 md_autodetect_dev(rdev->bdev->bd_dev);
2221 kobject_put(&rdev->kobj);
2224 void md_kick_rdev_from_array(struct md_rdev *rdev)
2226 unbind_rdev_from_array(rdev);
2229 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2231 static void export_array(struct mddev *mddev)
2233 struct md_rdev *rdev;
2235 while (!list_empty(&mddev->disks)) {
2236 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2238 md_kick_rdev_from_array(rdev);
2240 mddev->raid_disks = 0;
2241 mddev->major_version = 0;
2244 static void sync_sbs(struct mddev *mddev, int nospares)
2246 /* Update each superblock (in-memory image), but
2247 * if we are allowed to, skip spares which already
2248 * have the right event counter, or have one earlier
2249 * (which would mean they aren't being marked as dirty
2250 * with the rest of the array)
2252 struct md_rdev *rdev;
2253 rdev_for_each(rdev, mddev) {
2254 if (rdev->sb_events == mddev->events ||
2256 rdev->raid_disk < 0 &&
2257 rdev->sb_events+1 == mddev->events)) {
2258 /* Don't update this superblock */
2259 rdev->sb_loaded = 2;
2261 sync_super(mddev, rdev);
2262 rdev->sb_loaded = 1;
2267 static bool does_sb_need_changing(struct mddev *mddev)
2269 struct md_rdev *rdev;
2270 struct mdp_superblock_1 *sb;
2273 /* Find a good rdev */
2274 rdev_for_each(rdev, mddev)
2275 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2278 /* No good device found. */
2282 sb = page_address(rdev->sb_page);
2283 /* Check if a device has become faulty or a spare become active */
2284 rdev_for_each(rdev, mddev) {
2285 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2286 /* Device activated? */
2287 if (role == 0xffff && rdev->raid_disk >=0 &&
2288 !test_bit(Faulty, &rdev->flags))
2290 /* Device turned faulty? */
2291 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2295 /* Check if any mddev parameters have changed */
2296 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2297 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2298 (mddev->layout != le64_to_cpu(sb->layout)) ||
2299 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2300 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2306 void md_update_sb(struct mddev *mddev, int force_change)
2308 struct md_rdev *rdev;
2311 int any_badblocks_changed = 0;
2316 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2321 if (mddev_is_clustered(mddev)) {
2322 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2324 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2326 ret = md_cluster_ops->metadata_update_start(mddev);
2327 /* Has someone else has updated the sb */
2328 if (!does_sb_need_changing(mddev)) {
2330 md_cluster_ops->metadata_update_cancel(mddev);
2331 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2332 BIT(MD_SB_CHANGE_DEVS) |
2333 BIT(MD_SB_CHANGE_CLEAN));
2338 /* First make sure individual recovery_offsets are correct */
2339 rdev_for_each(rdev, mddev) {
2340 if (rdev->raid_disk >= 0 &&
2341 mddev->delta_disks >= 0 &&
2342 !test_bit(Journal, &rdev->flags) &&
2343 !test_bit(In_sync, &rdev->flags) &&
2344 mddev->curr_resync_completed > rdev->recovery_offset)
2345 rdev->recovery_offset = mddev->curr_resync_completed;
2348 if (!mddev->persistent) {
2349 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2350 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2351 if (!mddev->external) {
2352 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2353 rdev_for_each(rdev, mddev) {
2354 if (rdev->badblocks.changed) {
2355 rdev->badblocks.changed = 0;
2356 ack_all_badblocks(&rdev->badblocks);
2357 md_error(mddev, rdev);
2359 clear_bit(Blocked, &rdev->flags);
2360 clear_bit(BlockedBadBlocks, &rdev->flags);
2361 wake_up(&rdev->blocked_wait);
2364 wake_up(&mddev->sb_wait);
2368 spin_lock(&mddev->lock);
2370 mddev->utime = ktime_get_real_seconds();
2372 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2374 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2375 /* just a clean<-> dirty transition, possibly leave spares alone,
2376 * though if events isn't the right even/odd, we will have to do
2382 if (mddev->degraded)
2383 /* If the array is degraded, then skipping spares is both
2384 * dangerous and fairly pointless.
2385 * Dangerous because a device that was removed from the array
2386 * might have a event_count that still looks up-to-date,
2387 * so it can be re-added without a resync.
2388 * Pointless because if there are any spares to skip,
2389 * then a recovery will happen and soon that array won't
2390 * be degraded any more and the spare can go back to sleep then.
2394 sync_req = mddev->in_sync;
2396 /* If this is just a dirty<->clean transition, and the array is clean
2397 * and 'events' is odd, we can roll back to the previous clean state */
2399 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2400 && mddev->can_decrease_events
2401 && mddev->events != 1) {
2403 mddev->can_decrease_events = 0;
2405 /* otherwise we have to go forward and ... */
2407 mddev->can_decrease_events = nospares;
2411 * This 64-bit counter should never wrap.
2412 * Either we are in around ~1 trillion A.C., assuming
2413 * 1 reboot per second, or we have a bug...
2415 WARN_ON(mddev->events == 0);
2417 rdev_for_each(rdev, mddev) {
2418 if (rdev->badblocks.changed)
2419 any_badblocks_changed++;
2420 if (test_bit(Faulty, &rdev->flags))
2421 set_bit(FaultRecorded, &rdev->flags);
2424 sync_sbs(mddev, nospares);
2425 spin_unlock(&mddev->lock);
2427 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2428 mdname(mddev), mddev->in_sync);
2431 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2433 bitmap_update_sb(mddev->bitmap);
2434 rdev_for_each(rdev, mddev) {
2435 char b[BDEVNAME_SIZE];
2437 if (rdev->sb_loaded != 1)
2438 continue; /* no noise on spare devices */
2440 if (!test_bit(Faulty, &rdev->flags)) {
2441 md_super_write(mddev,rdev,
2442 rdev->sb_start, rdev->sb_size,
2444 pr_debug("md: (write) %s's sb offset: %llu\n",
2445 bdevname(rdev->bdev, b),
2446 (unsigned long long)rdev->sb_start);
2447 rdev->sb_events = mddev->events;
2448 if (rdev->badblocks.size) {
2449 md_super_write(mddev, rdev,
2450 rdev->badblocks.sector,
2451 rdev->badblocks.size << 9,
2453 rdev->badblocks.size = 0;
2457 pr_debug("md: %s (skipping faulty)\n",
2458 bdevname(rdev->bdev, b));
2460 if (mddev->level == LEVEL_MULTIPATH)
2461 /* only need to write one superblock... */
2464 if (md_super_wait(mddev) < 0)
2466 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2468 if (mddev_is_clustered(mddev) && ret == 0)
2469 md_cluster_ops->metadata_update_finish(mddev);
2471 if (mddev->in_sync != sync_req ||
2472 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2473 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2474 /* have to write it out again */
2476 wake_up(&mddev->sb_wait);
2477 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2478 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2480 rdev_for_each(rdev, mddev) {
2481 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2482 clear_bit(Blocked, &rdev->flags);
2484 if (any_badblocks_changed)
2485 ack_all_badblocks(&rdev->badblocks);
2486 clear_bit(BlockedBadBlocks, &rdev->flags);
2487 wake_up(&rdev->blocked_wait);
2490 EXPORT_SYMBOL(md_update_sb);
2492 static int add_bound_rdev(struct md_rdev *rdev)
2494 struct mddev *mddev = rdev->mddev;
2496 bool add_journal = test_bit(Journal, &rdev->flags);
2498 if (!mddev->pers->hot_remove_disk || add_journal) {
2499 /* If there is hot_add_disk but no hot_remove_disk
2500 * then added disks for geometry changes,
2501 * and should be added immediately.
2503 super_types[mddev->major_version].
2504 validate_super(mddev, rdev);
2506 mddev_suspend(mddev);
2507 err = mddev->pers->hot_add_disk(mddev, rdev);
2509 mddev_resume(mddev);
2511 md_kick_rdev_from_array(rdev);
2515 sysfs_notify_dirent_safe(rdev->sysfs_state);
2517 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2518 if (mddev->degraded)
2519 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2520 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2521 md_new_event(mddev);
2522 md_wakeup_thread(mddev->thread);
2526 /* words written to sysfs files may, or may not, be \n terminated.
2527 * We want to accept with case. For this we use cmd_match.
2529 static int cmd_match(const char *cmd, const char *str)
2531 /* See if cmd, written into a sysfs file, matches
2532 * str. They must either be the same, or cmd can
2533 * have a trailing newline
2535 while (*cmd && *str && *cmd == *str) {
2546 struct rdev_sysfs_entry {
2547 struct attribute attr;
2548 ssize_t (*show)(struct md_rdev *, char *);
2549 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2553 state_show(struct md_rdev *rdev, char *page)
2557 unsigned long flags = ACCESS_ONCE(rdev->flags);
2559 if (test_bit(Faulty, &flags) ||
2560 (!test_bit(ExternalBbl, &flags) &&
2561 rdev->badblocks.unacked_exist))
2562 len += sprintf(page+len, "faulty%s", sep);
2563 if (test_bit(In_sync, &flags))
2564 len += sprintf(page+len, "in_sync%s", sep);
2565 if (test_bit(Journal, &flags))
2566 len += sprintf(page+len, "journal%s", sep);
2567 if (test_bit(WriteMostly, &flags))
2568 len += sprintf(page+len, "write_mostly%s", sep);
2569 if (test_bit(Blocked, &flags) ||
2570 (rdev->badblocks.unacked_exist
2571 && !test_bit(Faulty, &flags)))
2572 len += sprintf(page+len, "blocked%s", sep);
2573 if (!test_bit(Faulty, &flags) &&
2574 !test_bit(Journal, &flags) &&
2575 !test_bit(In_sync, &flags))
2576 len += sprintf(page+len, "spare%s", sep);
2577 if (test_bit(WriteErrorSeen, &flags))
2578 len += sprintf(page+len, "write_error%s", sep);
2579 if (test_bit(WantReplacement, &flags))
2580 len += sprintf(page+len, "want_replacement%s", sep);
2581 if (test_bit(Replacement, &flags))
2582 len += sprintf(page+len, "replacement%s", sep);
2583 if (test_bit(ExternalBbl, &flags))
2584 len += sprintf(page+len, "external_bbl%s", sep);
2585 if (test_bit(FailFast, &flags))
2586 len += sprintf(page+len, "failfast%s", sep);
2591 return len+sprintf(page+len, "\n");
2595 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2598 * faulty - simulates an error
2599 * remove - disconnects the device
2600 * writemostly - sets write_mostly
2601 * -writemostly - clears write_mostly
2602 * blocked - sets the Blocked flags
2603 * -blocked - clears the Blocked and possibly simulates an error
2604 * insync - sets Insync providing device isn't active
2605 * -insync - clear Insync for a device with a slot assigned,
2606 * so that it gets rebuilt based on bitmap
2607 * write_error - sets WriteErrorSeen
2608 * -write_error - clears WriteErrorSeen
2609 * {,-}failfast - set/clear FailFast
2612 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2613 md_error(rdev->mddev, rdev);
2614 if (test_bit(Faulty, &rdev->flags))
2618 } else if (cmd_match(buf, "remove")) {
2619 if (rdev->mddev->pers) {
2620 clear_bit(Blocked, &rdev->flags);
2621 remove_and_add_spares(rdev->mddev, rdev);
2623 if (rdev->raid_disk >= 0)
2626 struct mddev *mddev = rdev->mddev;
2628 if (mddev_is_clustered(mddev))
2629 err = md_cluster_ops->remove_disk(mddev, rdev);
2632 md_kick_rdev_from_array(rdev);
2634 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2635 md_wakeup_thread(mddev->thread);
2637 md_new_event(mddev);
2640 } else if (cmd_match(buf, "writemostly")) {
2641 set_bit(WriteMostly, &rdev->flags);
2643 } else if (cmd_match(buf, "-writemostly")) {
2644 clear_bit(WriteMostly, &rdev->flags);
2646 } else if (cmd_match(buf, "blocked")) {
2647 set_bit(Blocked, &rdev->flags);
2649 } else if (cmd_match(buf, "-blocked")) {
2650 if (!test_bit(Faulty, &rdev->flags) &&
2651 !test_bit(ExternalBbl, &rdev->flags) &&
2652 rdev->badblocks.unacked_exist) {
2653 /* metadata handler doesn't understand badblocks,
2654 * so we need to fail the device
2656 md_error(rdev->mddev, rdev);
2658 clear_bit(Blocked, &rdev->flags);
2659 clear_bit(BlockedBadBlocks, &rdev->flags);
2660 wake_up(&rdev->blocked_wait);
2661 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2662 md_wakeup_thread(rdev->mddev->thread);
2665 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2666 set_bit(In_sync, &rdev->flags);
2668 } else if (cmd_match(buf, "failfast")) {
2669 set_bit(FailFast, &rdev->flags);
2671 } else if (cmd_match(buf, "-failfast")) {
2672 clear_bit(FailFast, &rdev->flags);
2674 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2675 !test_bit(Journal, &rdev->flags)) {
2676 if (rdev->mddev->pers == NULL) {
2677 clear_bit(In_sync, &rdev->flags);
2678 rdev->saved_raid_disk = rdev->raid_disk;
2679 rdev->raid_disk = -1;
2682 } else if (cmd_match(buf, "write_error")) {
2683 set_bit(WriteErrorSeen, &rdev->flags);
2685 } else if (cmd_match(buf, "-write_error")) {
2686 clear_bit(WriteErrorSeen, &rdev->flags);
2688 } else if (cmd_match(buf, "want_replacement")) {
2689 /* Any non-spare device that is not a replacement can
2690 * become want_replacement at any time, but we then need to
2691 * check if recovery is needed.
2693 if (rdev->raid_disk >= 0 &&
2694 !test_bit(Journal, &rdev->flags) &&
2695 !test_bit(Replacement, &rdev->flags))
2696 set_bit(WantReplacement, &rdev->flags);
2697 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2698 md_wakeup_thread(rdev->mddev->thread);
2700 } else if (cmd_match(buf, "-want_replacement")) {
2701 /* Clearing 'want_replacement' is always allowed.
2702 * Once replacements starts it is too late though.
2705 clear_bit(WantReplacement, &rdev->flags);
2706 } else if (cmd_match(buf, "replacement")) {
2707 /* Can only set a device as a replacement when array has not
2708 * yet been started. Once running, replacement is automatic
2709 * from spares, or by assigning 'slot'.
2711 if (rdev->mddev->pers)
2714 set_bit(Replacement, &rdev->flags);
2717 } else if (cmd_match(buf, "-replacement")) {
2718 /* Similarly, can only clear Replacement before start */
2719 if (rdev->mddev->pers)
2722 clear_bit(Replacement, &rdev->flags);
2725 } else if (cmd_match(buf, "re-add")) {
2726 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2727 /* clear_bit is performed _after_ all the devices
2728 * have their local Faulty bit cleared. If any writes
2729 * happen in the meantime in the local node, they
2730 * will land in the local bitmap, which will be synced
2731 * by this node eventually
2733 if (!mddev_is_clustered(rdev->mddev) ||
2734 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2735 clear_bit(Faulty, &rdev->flags);
2736 err = add_bound_rdev(rdev);
2740 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2741 set_bit(ExternalBbl, &rdev->flags);
2742 rdev->badblocks.shift = 0;
2744 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2745 clear_bit(ExternalBbl, &rdev->flags);
2749 sysfs_notify_dirent_safe(rdev->sysfs_state);
2750 return err ? err : len;
2752 static struct rdev_sysfs_entry rdev_state =
2753 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2756 errors_show(struct md_rdev *rdev, char *page)
2758 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2762 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2767 rv = kstrtouint(buf, 10, &n);
2770 atomic_set(&rdev->corrected_errors, n);
2773 static struct rdev_sysfs_entry rdev_errors =
2774 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2777 slot_show(struct md_rdev *rdev, char *page)
2779 if (test_bit(Journal, &rdev->flags))
2780 return sprintf(page, "journal\n");
2781 else if (rdev->raid_disk < 0)
2782 return sprintf(page, "none\n");
2784 return sprintf(page, "%d\n", rdev->raid_disk);
2788 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2793 if (test_bit(Journal, &rdev->flags))
2795 if (strncmp(buf, "none", 4)==0)
2798 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2802 if (rdev->mddev->pers && slot == -1) {
2803 /* Setting 'slot' on an active array requires also
2804 * updating the 'rd%d' link, and communicating
2805 * with the personality with ->hot_*_disk.
2806 * For now we only support removing
2807 * failed/spare devices. This normally happens automatically,
2808 * but not when the metadata is externally managed.
2810 if (rdev->raid_disk == -1)
2812 /* personality does all needed checks */
2813 if (rdev->mddev->pers->hot_remove_disk == NULL)
2815 clear_bit(Blocked, &rdev->flags);
2816 remove_and_add_spares(rdev->mddev, rdev);
2817 if (rdev->raid_disk >= 0)
2819 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2820 md_wakeup_thread(rdev->mddev->thread);
2821 } else if (rdev->mddev->pers) {
2822 /* Activating a spare .. or possibly reactivating
2823 * if we ever get bitmaps working here.
2827 if (rdev->raid_disk != -1)
2830 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2833 if (rdev->mddev->pers->hot_add_disk == NULL)
2836 if (slot >= rdev->mddev->raid_disks &&
2837 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2840 rdev->raid_disk = slot;
2841 if (test_bit(In_sync, &rdev->flags))
2842 rdev->saved_raid_disk = slot;
2844 rdev->saved_raid_disk = -1;
2845 clear_bit(In_sync, &rdev->flags);
2846 clear_bit(Bitmap_sync, &rdev->flags);
2847 err = rdev->mddev->pers->
2848 hot_add_disk(rdev->mddev, rdev);
2850 rdev->raid_disk = -1;
2853 sysfs_notify_dirent_safe(rdev->sysfs_state);
2854 if (sysfs_link_rdev(rdev->mddev, rdev))
2855 /* failure here is OK */;
2856 /* don't wakeup anyone, leave that to userspace. */
2858 if (slot >= rdev->mddev->raid_disks &&
2859 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2861 rdev->raid_disk = slot;
2862 /* assume it is working */
2863 clear_bit(Faulty, &rdev->flags);
2864 clear_bit(WriteMostly, &rdev->flags);
2865 set_bit(In_sync, &rdev->flags);
2866 sysfs_notify_dirent_safe(rdev->sysfs_state);
2871 static struct rdev_sysfs_entry rdev_slot =
2872 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2875 offset_show(struct md_rdev *rdev, char *page)
2877 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2881 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2883 unsigned long long offset;
2884 if (kstrtoull(buf, 10, &offset) < 0)
2886 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2888 if (rdev->sectors && rdev->mddev->external)
2889 /* Must set offset before size, so overlap checks
2892 rdev->data_offset = offset;
2893 rdev->new_data_offset = offset;
2897 static struct rdev_sysfs_entry rdev_offset =
2898 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2900 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2902 return sprintf(page, "%llu\n",
2903 (unsigned long long)rdev->new_data_offset);
2906 static ssize_t new_offset_store(struct md_rdev *rdev,
2907 const char *buf, size_t len)
2909 unsigned long long new_offset;
2910 struct mddev *mddev = rdev->mddev;
2912 if (kstrtoull(buf, 10, &new_offset) < 0)
2915 if (mddev->sync_thread ||
2916 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2918 if (new_offset == rdev->data_offset)
2919 /* reset is always permitted */
2921 else if (new_offset > rdev->data_offset) {
2922 /* must not push array size beyond rdev_sectors */
2923 if (new_offset - rdev->data_offset
2924 + mddev->dev_sectors > rdev->sectors)
2927 /* Metadata worries about other space details. */
2929 /* decreasing the offset is inconsistent with a backwards
2932 if (new_offset < rdev->data_offset &&
2933 mddev->reshape_backwards)
2935 /* Increasing offset is inconsistent with forwards
2936 * reshape. reshape_direction should be set to
2937 * 'backwards' first.
2939 if (new_offset > rdev->data_offset &&
2940 !mddev->reshape_backwards)
2943 if (mddev->pers && mddev->persistent &&
2944 !super_types[mddev->major_version]
2945 .allow_new_offset(rdev, new_offset))
2947 rdev->new_data_offset = new_offset;
2948 if (new_offset > rdev->data_offset)
2949 mddev->reshape_backwards = 1;
2950 else if (new_offset < rdev->data_offset)
2951 mddev->reshape_backwards = 0;
2955 static struct rdev_sysfs_entry rdev_new_offset =
2956 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2959 rdev_size_show(struct md_rdev *rdev, char *page)
2961 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2964 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2966 /* check if two start/length pairs overlap */
2974 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2976 unsigned long long blocks;
2979 if (kstrtoull(buf, 10, &blocks) < 0)
2982 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2983 return -EINVAL; /* sector conversion overflow */
2986 if (new != blocks * 2)
2987 return -EINVAL; /* unsigned long long to sector_t overflow */
2994 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2996 struct mddev *my_mddev = rdev->mddev;
2997 sector_t oldsectors = rdev->sectors;
3000 if (test_bit(Journal, &rdev->flags))
3002 if (strict_blocks_to_sectors(buf, §ors) < 0)
3004 if (rdev->data_offset != rdev->new_data_offset)
3005 return -EINVAL; /* too confusing */
3006 if (my_mddev->pers && rdev->raid_disk >= 0) {
3007 if (my_mddev->persistent) {
3008 sectors = super_types[my_mddev->major_version].
3009 rdev_size_change(rdev, sectors);
3012 } else if (!sectors)
3013 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3015 if (!my_mddev->pers->resize)
3016 /* Cannot change size for RAID0 or Linear etc */
3019 if (sectors < my_mddev->dev_sectors)
3020 return -EINVAL; /* component must fit device */
3022 rdev->sectors = sectors;
3023 if (sectors > oldsectors && my_mddev->external) {
3024 /* Need to check that all other rdevs with the same
3025 * ->bdev do not overlap. 'rcu' is sufficient to walk
3026 * the rdev lists safely.
3027 * This check does not provide a hard guarantee, it
3028 * just helps avoid dangerous mistakes.
3030 struct mddev *mddev;
3032 struct list_head *tmp;
3035 for_each_mddev(mddev, tmp) {
3036 struct md_rdev *rdev2;
3038 rdev_for_each(rdev2, mddev)
3039 if (rdev->bdev == rdev2->bdev &&
3041 overlaps(rdev->data_offset, rdev->sectors,
3054 /* Someone else could have slipped in a size
3055 * change here, but doing so is just silly.
3056 * We put oldsectors back because we *know* it is
3057 * safe, and trust userspace not to race with
3060 rdev->sectors = oldsectors;
3067 static struct rdev_sysfs_entry rdev_size =
3068 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3070 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3072 unsigned long long recovery_start = rdev->recovery_offset;
3074 if (test_bit(In_sync, &rdev->flags) ||
3075 recovery_start == MaxSector)
3076 return sprintf(page, "none\n");
3078 return sprintf(page, "%llu\n", recovery_start);
3081 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3083 unsigned long long recovery_start;
3085 if (cmd_match(buf, "none"))
3086 recovery_start = MaxSector;
3087 else if (kstrtoull(buf, 10, &recovery_start))
3090 if (rdev->mddev->pers &&
3091 rdev->raid_disk >= 0)
3094 rdev->recovery_offset = recovery_start;
3095 if (recovery_start == MaxSector)
3096 set_bit(In_sync, &rdev->flags);
3098 clear_bit(In_sync, &rdev->flags);
3102 static struct rdev_sysfs_entry rdev_recovery_start =
3103 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3105 /* sysfs access to bad-blocks list.
3106 * We present two files.
3107 * 'bad-blocks' lists sector numbers and lengths of ranges that
3108 * are recorded as bad. The list is truncated to fit within
3109 * the one-page limit of sysfs.
3110 * Writing "sector length" to this file adds an acknowledged
3112 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3113 * been acknowledged. Writing to this file adds bad blocks
3114 * without acknowledging them. This is largely for testing.
3116 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3118 return badblocks_show(&rdev->badblocks, page, 0);
3120 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3122 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3123 /* Maybe that ack was all we needed */
3124 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3125 wake_up(&rdev->blocked_wait);
3128 static struct rdev_sysfs_entry rdev_bad_blocks =
3129 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3131 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3133 return badblocks_show(&rdev->badblocks, page, 1);
3135 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3137 return badblocks_store(&rdev->badblocks, page, len, 1);
3139 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3140 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3142 static struct attribute *rdev_default_attrs[] = {
3147 &rdev_new_offset.attr,
3149 &rdev_recovery_start.attr,
3150 &rdev_bad_blocks.attr,
3151 &rdev_unack_bad_blocks.attr,
3155 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3157 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3158 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3164 return entry->show(rdev, page);
3168 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3169 const char *page, size_t length)
3171 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3172 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3174 struct mddev *mddev = rdev->mddev;
3178 if (!capable(CAP_SYS_ADMIN))
3180 rv = mddev ? mddev_lock(mddev): -EBUSY;
3182 if (rdev->mddev == NULL)
3185 rv = entry->store(rdev, page, length);
3186 mddev_unlock(mddev);
3191 static void rdev_free(struct kobject *ko)
3193 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3196 static const struct sysfs_ops rdev_sysfs_ops = {
3197 .show = rdev_attr_show,
3198 .store = rdev_attr_store,
3200 static struct kobj_type rdev_ktype = {
3201 .release = rdev_free,
3202 .sysfs_ops = &rdev_sysfs_ops,
3203 .default_attrs = rdev_default_attrs,
3206 int md_rdev_init(struct md_rdev *rdev)
3209 rdev->saved_raid_disk = -1;
3210 rdev->raid_disk = -1;
3212 rdev->data_offset = 0;
3213 rdev->new_data_offset = 0;
3214 rdev->sb_events = 0;
3215 rdev->last_read_error = 0;
3216 rdev->sb_loaded = 0;
3217 rdev->bb_page = NULL;
3218 atomic_set(&rdev->nr_pending, 0);
3219 atomic_set(&rdev->read_errors, 0);
3220 atomic_set(&rdev->corrected_errors, 0);
3222 INIT_LIST_HEAD(&rdev->same_set);
3223 init_waitqueue_head(&rdev->blocked_wait);
3225 /* Add space to store bad block list.
3226 * This reserves the space even on arrays where it cannot
3227 * be used - I wonder if that matters
3229 return badblocks_init(&rdev->badblocks, 0);
3231 EXPORT_SYMBOL_GPL(md_rdev_init);
3233 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3235 * mark the device faulty if:
3237 * - the device is nonexistent (zero size)
3238 * - the device has no valid superblock
3240 * a faulty rdev _never_ has rdev->sb set.
3242 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3244 char b[BDEVNAME_SIZE];
3246 struct md_rdev *rdev;
3249 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3251 return ERR_PTR(-ENOMEM);
3253 err = md_rdev_init(rdev);
3256 err = alloc_disk_sb(rdev);
3260 err = lock_rdev(rdev, newdev, super_format == -2);
3264 kobject_init(&rdev->kobj, &rdev_ktype);
3266 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3268 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3269 bdevname(rdev->bdev,b));
3274 if (super_format >= 0) {
3275 err = super_types[super_format].
3276 load_super(rdev, NULL, super_minor);
3277 if (err == -EINVAL) {
3278 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3279 bdevname(rdev->bdev,b),
3280 super_format, super_minor);
3284 pr_warn("md: could not read %s's sb, not importing!\n",
3285 bdevname(rdev->bdev,b));
3295 md_rdev_clear(rdev);
3297 return ERR_PTR(err);
3301 * Check a full RAID array for plausibility
3304 static void analyze_sbs(struct mddev *mddev)
3307 struct md_rdev *rdev, *freshest, *tmp;
3308 char b[BDEVNAME_SIZE];
3311 rdev_for_each_safe(rdev, tmp, mddev)
3312 switch (super_types[mddev->major_version].
3313 load_super(rdev, freshest, mddev->minor_version)) {
3320 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3321 bdevname(rdev->bdev,b));
3322 md_kick_rdev_from_array(rdev);
3325 super_types[mddev->major_version].
3326 validate_super(mddev, freshest);
3329 rdev_for_each_safe(rdev, tmp, mddev) {
3330 if (mddev->max_disks &&
3331 (rdev->desc_nr >= mddev->max_disks ||
3332 i > mddev->max_disks)) {
3333 pr_warn("md: %s: %s: only %d devices permitted\n",
3334 mdname(mddev), bdevname(rdev->bdev, b),
3336 md_kick_rdev_from_array(rdev);
3339 if (rdev != freshest) {
3340 if (super_types[mddev->major_version].
3341 validate_super(mddev, rdev)) {
3342 pr_warn("md: kicking non-fresh %s from array!\n",
3343 bdevname(rdev->bdev,b));
3344 md_kick_rdev_from_array(rdev);
3348 if (mddev->level == LEVEL_MULTIPATH) {
3349 rdev->desc_nr = i++;
3350 rdev->raid_disk = rdev->desc_nr;
3351 set_bit(In_sync, &rdev->flags);
3352 } else if (rdev->raid_disk >=
3353 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3354 !test_bit(Journal, &rdev->flags)) {
3355 rdev->raid_disk = -1;
3356 clear_bit(In_sync, &rdev->flags);
3361 /* Read a fixed-point number.
3362 * Numbers in sysfs attributes should be in "standard" units where
3363 * possible, so time should be in seconds.
3364 * However we internally use a a much smaller unit such as
3365 * milliseconds or jiffies.
3366 * This function takes a decimal number with a possible fractional
3367 * component, and produces an integer which is the result of
3368 * multiplying that number by 10^'scale'.
3369 * all without any floating-point arithmetic.
3371 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3373 unsigned long result = 0;
3375 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3378 else if (decimals < scale) {
3381 result = result * 10 + value;
3393 while (decimals < scale) {
3402 safe_delay_show(struct mddev *mddev, char *page)
3404 int msec = (mddev->safemode_delay*1000)/HZ;
3405 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3408 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3412 if (mddev_is_clustered(mddev)) {
3413 pr_warn("md: Safemode is disabled for clustered mode\n");
3417 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3420 mddev->safemode_delay = 0;
3422 unsigned long old_delay = mddev->safemode_delay;
3423 unsigned long new_delay = (msec*HZ)/1000;
3427 mddev->safemode_delay = new_delay;
3428 if (new_delay < old_delay || old_delay == 0)
3429 mod_timer(&mddev->safemode_timer, jiffies+1);
3433 static struct md_sysfs_entry md_safe_delay =
3434 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3437 level_show(struct mddev *mddev, char *page)
3439 struct md_personality *p;
3441 spin_lock(&mddev->lock);
3444 ret = sprintf(page, "%s\n", p->name);
3445 else if (mddev->clevel[0])
3446 ret = sprintf(page, "%s\n", mddev->clevel);
3447 else if (mddev->level != LEVEL_NONE)
3448 ret = sprintf(page, "%d\n", mddev->level);
3451 spin_unlock(&mddev->lock);
3456 level_store(struct mddev *mddev, const char *buf, size_t len)
3461 struct md_personality *pers, *oldpers;
3463 void *priv, *oldpriv;
3464 struct md_rdev *rdev;
3466 if (slen == 0 || slen >= sizeof(clevel))
3469 rv = mddev_lock(mddev);
3473 if (mddev->pers == NULL) {
3474 strncpy(mddev->clevel, buf, slen);
3475 if (mddev->clevel[slen-1] == '\n')
3477 mddev->clevel[slen] = 0;
3478 mddev->level = LEVEL_NONE;
3486 /* request to change the personality. Need to ensure:
3487 * - array is not engaged in resync/recovery/reshape
3488 * - old personality can be suspended
3489 * - new personality will access other array.
3493 if (mddev->sync_thread ||
3494 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3495 mddev->reshape_position != MaxSector ||
3496 mddev->sysfs_active)
3500 if (!mddev->pers->quiesce) {
3501 pr_warn("md: %s: %s does not support online personality change\n",
3502 mdname(mddev), mddev->pers->name);
3506 /* Now find the new personality */
3507 strncpy(clevel, buf, slen);
3508 if (clevel[slen-1] == '\n')
3511 if (kstrtol(clevel, 10, &level))
3514 if (request_module("md-%s", clevel) != 0)
3515 request_module("md-level-%s", clevel);
3516 spin_lock(&pers_lock);
3517 pers = find_pers(level, clevel);
3518 if (!pers || !try_module_get(pers->owner)) {
3519 spin_unlock(&pers_lock);
3520 pr_warn("md: personality %s not loaded\n", clevel);
3524 spin_unlock(&pers_lock);
3526 if (pers == mddev->pers) {
3527 /* Nothing to do! */
3528 module_put(pers->owner);
3532 if (!pers->takeover) {
3533 module_put(pers->owner);
3534 pr_warn("md: %s: %s does not support personality takeover\n",
3535 mdname(mddev), clevel);
3540 rdev_for_each(rdev, mddev)
3541 rdev->new_raid_disk = rdev->raid_disk;
3543 /* ->takeover must set new_* and/or delta_disks
3544 * if it succeeds, and may set them when it fails.
3546 priv = pers->takeover(mddev);
3548 mddev->new_level = mddev->level;
3549 mddev->new_layout = mddev->layout;
3550 mddev->new_chunk_sectors = mddev->chunk_sectors;
3551 mddev->raid_disks -= mddev->delta_disks;
3552 mddev->delta_disks = 0;
3553 mddev->reshape_backwards = 0;
3554 module_put(pers->owner);
3555 pr_warn("md: %s: %s would not accept array\n",
3556 mdname(mddev), clevel);
3561 /* Looks like we have a winner */
3562 mddev_suspend(mddev);
3563 mddev_detach(mddev);
3565 spin_lock(&mddev->lock);
3566 oldpers = mddev->pers;
3567 oldpriv = mddev->private;
3569 mddev->private = priv;
3570 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3571 mddev->level = mddev->new_level;
3572 mddev->layout = mddev->new_layout;
3573 mddev->chunk_sectors = mddev->new_chunk_sectors;
3574 mddev->delta_disks = 0;
3575 mddev->reshape_backwards = 0;
3576 mddev->degraded = 0;
3577 spin_unlock(&mddev->lock);
3579 if (oldpers->sync_request == NULL &&
3581 /* We are converting from a no-redundancy array
3582 * to a redundancy array and metadata is managed
3583 * externally so we need to be sure that writes
3584 * won't block due to a need to transition
3586 * until external management is started.
3589 mddev->safemode_delay = 0;
3590 mddev->safemode = 0;
3593 oldpers->free(mddev, oldpriv);
3595 if (oldpers->sync_request == NULL &&
3596 pers->sync_request != NULL) {
3597 /* need to add the md_redundancy_group */
3598 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3599 pr_warn("md: cannot register extra attributes for %s\n",
3601 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3603 if (oldpers->sync_request != NULL &&
3604 pers->sync_request == NULL) {
3605 /* need to remove the md_redundancy_group */
3606 if (mddev->to_remove == NULL)
3607 mddev->to_remove = &md_redundancy_group;
3610 module_put(oldpers->owner);
3612 rdev_for_each(rdev, mddev) {
3613 if (rdev->raid_disk < 0)
3615 if (rdev->new_raid_disk >= mddev->raid_disks)
3616 rdev->new_raid_disk = -1;
3617 if (rdev->new_raid_disk == rdev->raid_disk)
3619 sysfs_unlink_rdev(mddev, rdev);
3621 rdev_for_each(rdev, mddev) {
3622 if (rdev->raid_disk < 0)
3624 if (rdev->new_raid_disk == rdev->raid_disk)
3626 rdev->raid_disk = rdev->new_raid_disk;
3627 if (rdev->raid_disk < 0)
3628 clear_bit(In_sync, &rdev->flags);
3630 if (sysfs_link_rdev(mddev, rdev))
3631 pr_warn("md: cannot register rd%d for %s after level change\n",
3632 rdev->raid_disk, mdname(mddev));
3636 if (pers->sync_request == NULL) {
3637 /* this is now an array without redundancy, so
3638 * it must always be in_sync
3641 del_timer_sync(&mddev->safemode_timer);
3643 blk_set_stacking_limits(&mddev->queue->limits);
3645 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3646 mddev_resume(mddev);
3648 md_update_sb(mddev, 1);
3649 sysfs_notify(&mddev->kobj, NULL, "level");
3650 md_new_event(mddev);
3653 mddev_unlock(mddev);
3657 static struct md_sysfs_entry md_level =
3658 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3661 layout_show(struct mddev *mddev, char *page)
3663 /* just a number, not meaningful for all levels */
3664 if (mddev->reshape_position != MaxSector &&
3665 mddev->layout != mddev->new_layout)
3666 return sprintf(page, "%d (%d)\n",
3667 mddev->new_layout, mddev->layout);
3668 return sprintf(page, "%d\n", mddev->layout);
3672 layout_store(struct mddev *mddev, const char *buf, size_t len)
3677 err = kstrtouint(buf, 10, &n);
3680 err = mddev_lock(mddev);
3685 if (mddev->pers->check_reshape == NULL)
3690 mddev->new_layout = n;
3691 err = mddev->pers->check_reshape(mddev);
3693 mddev->new_layout = mddev->layout;
3696 mddev->new_layout = n;
3697 if (mddev->reshape_position == MaxSector)
3700 mddev_unlock(mddev);
3703 static struct md_sysfs_entry md_layout =
3704 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3707 raid_disks_show(struct mddev *mddev, char *page)
3709 if (mddev->raid_disks == 0)
3711 if (mddev->reshape_position != MaxSector &&
3712 mddev->delta_disks != 0)
3713 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3714 mddev->raid_disks - mddev->delta_disks);
3715 return sprintf(page, "%d\n", mddev->raid_disks);
3718 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3721 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3726 err = kstrtouint(buf, 10, &n);
3730 err = mddev_lock(mddev);
3734 err = update_raid_disks(mddev, n);
3735 else if (mddev->reshape_position != MaxSector) {
3736 struct md_rdev *rdev;
3737 int olddisks = mddev->raid_disks - mddev->delta_disks;
3740 rdev_for_each(rdev, mddev) {
3742 rdev->data_offset < rdev->new_data_offset)
3745 rdev->data_offset > rdev->new_data_offset)
3749 mddev->delta_disks = n - olddisks;
3750 mddev->raid_disks = n;
3751 mddev->reshape_backwards = (mddev->delta_disks < 0);
3753 mddev->raid_disks = n;
3755 mddev_unlock(mddev);
3756 return err ? err : len;
3758 static struct md_sysfs_entry md_raid_disks =
3759 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3762 chunk_size_show(struct mddev *mddev, char *page)
3764 if (mddev->reshape_position != MaxSector &&
3765 mddev->chunk_sectors != mddev->new_chunk_sectors)
3766 return sprintf(page, "%d (%d)\n",
3767 mddev->new_chunk_sectors << 9,
3768 mddev->chunk_sectors << 9);
3769 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3773 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3778 err = kstrtoul(buf, 10, &n);
3782 err = mddev_lock(mddev);
3786 if (mddev->pers->check_reshape == NULL)
3791 mddev->new_chunk_sectors = n >> 9;
3792 err = mddev->pers->check_reshape(mddev);
3794 mddev->new_chunk_sectors = mddev->chunk_sectors;
3797 mddev->new_chunk_sectors = n >> 9;
3798 if (mddev->reshape_position == MaxSector)
3799 mddev->chunk_sectors = n >> 9;
3801 mddev_unlock(mddev);
3804 static struct md_sysfs_entry md_chunk_size =
3805 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3808 resync_start_show(struct mddev *mddev, char *page)
3810 if (mddev->recovery_cp == MaxSector)
3811 return sprintf(page, "none\n");
3812 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3816 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3818 unsigned long long n;
3821 if (cmd_match(buf, "none"))
3824 err = kstrtoull(buf, 10, &n);
3827 if (n != (sector_t)n)
3831 err = mddev_lock(mddev);
3834 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3838 mddev->recovery_cp = n;
3840 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3842 mddev_unlock(mddev);
3845 static struct md_sysfs_entry md_resync_start =
3846 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3847 resync_start_show, resync_start_store);
3850 * The array state can be:
3853 * No devices, no size, no level
3854 * Equivalent to STOP_ARRAY ioctl
3856 * May have some settings, but array is not active
3857 * all IO results in error
3858 * When written, doesn't tear down array, but just stops it
3859 * suspended (not supported yet)
3860 * All IO requests will block. The array can be reconfigured.
3861 * Writing this, if accepted, will block until array is quiescent
3863 * no resync can happen. no superblocks get written.
3864 * write requests fail
3866 * like readonly, but behaves like 'clean' on a write request.
3868 * clean - no pending writes, but otherwise active.
3869 * When written to inactive array, starts without resync
3870 * If a write request arrives then
3871 * if metadata is known, mark 'dirty' and switch to 'active'.
3872 * if not known, block and switch to write-pending
3873 * If written to an active array that has pending writes, then fails.
3875 * fully active: IO and resync can be happening.
3876 * When written to inactive array, starts with resync
3879 * clean, but writes are blocked waiting for 'active' to be written.
3882 * like active, but no writes have been seen for a while (100msec).
3885 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3886 write_pending, active_idle, bad_word};
3887 static char *array_states[] = {
3888 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3889 "write-pending", "active-idle", NULL };
3891 static int match_word(const char *word, char **list)
3894 for (n=0; list[n]; n++)
3895 if (cmd_match(word, list[n]))
3901 array_state_show(struct mddev *mddev, char *page)
3903 enum array_state st = inactive;
3914 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
3916 else if (mddev->in_sync)
3918 else if (mddev->safemode)
3924 if (list_empty(&mddev->disks) &&
3925 mddev->raid_disks == 0 &&
3926 mddev->dev_sectors == 0)
3931 return sprintf(page, "%s\n", array_states[st]);
3934 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3935 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3936 static int do_md_run(struct mddev *mddev);
3937 static int restart_array(struct mddev *mddev);
3940 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3943 enum array_state st = match_word(buf, array_states);
3945 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3946 /* don't take reconfig_mutex when toggling between
3949 spin_lock(&mddev->lock);
3951 restart_array(mddev);
3952 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
3953 md_wakeup_thread(mddev->thread);
3954 wake_up(&mddev->sb_wait);
3956 } else /* st == clean */ {
3957 restart_array(mddev);
3958 if (atomic_read(&mddev->writes_pending) == 0) {
3959 if (mddev->in_sync == 0) {
3961 if (mddev->safemode == 1)
3962 mddev->safemode = 0;
3963 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3970 sysfs_notify_dirent_safe(mddev->sysfs_state);
3971 spin_unlock(&mddev->lock);
3974 err = mddev_lock(mddev);
3982 /* stopping an active array */
3983 err = do_md_stop(mddev, 0, NULL);
3986 /* stopping an active array */
3988 err = do_md_stop(mddev, 2, NULL);
3990 err = 0; /* already inactive */
3993 break; /* not supported yet */
3996 err = md_set_readonly(mddev, NULL);
3999 set_disk_ro(mddev->gendisk, 1);
4000 err = do_md_run(mddev);
4006 err = md_set_readonly(mddev, NULL);
4007 else if (mddev->ro == 1)
4008 err = restart_array(mddev);
4011 set_disk_ro(mddev->gendisk, 0);
4015 err = do_md_run(mddev);
4020 err = restart_array(mddev);
4023 spin_lock(&mddev->lock);
4024 if (atomic_read(&mddev->writes_pending) == 0) {
4025 if (mddev->in_sync == 0) {
4027 if (mddev->safemode == 1)
4028 mddev->safemode = 0;
4029 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4034 spin_unlock(&mddev->lock);
4040 err = restart_array(mddev);
4043 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4044 wake_up(&mddev->sb_wait);
4048 set_disk_ro(mddev->gendisk, 0);
4049 err = do_md_run(mddev);
4054 /* these cannot be set */
4059 if (mddev->hold_active == UNTIL_IOCTL)
4060 mddev->hold_active = 0;
4061 sysfs_notify_dirent_safe(mddev->sysfs_state);
4063 mddev_unlock(mddev);
4066 static struct md_sysfs_entry md_array_state =
4067 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4070 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4071 return sprintf(page, "%d\n",
4072 atomic_read(&mddev->max_corr_read_errors));
4076 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4081 rv = kstrtouint(buf, 10, &n);
4084 atomic_set(&mddev->max_corr_read_errors, n);
4088 static struct md_sysfs_entry max_corr_read_errors =
4089 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4090 max_corrected_read_errors_store);
4093 null_show(struct mddev *mddev, char *page)
4099 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4101 /* buf must be %d:%d\n? giving major and minor numbers */
4102 /* The new device is added to the array.
4103 * If the array has a persistent superblock, we read the
4104 * superblock to initialise info and check validity.
4105 * Otherwise, only checking done is that in bind_rdev_to_array,
4106 * which mainly checks size.
4109 int major = simple_strtoul(buf, &e, 10);
4112 struct md_rdev *rdev;
4115 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4117 minor = simple_strtoul(e+1, &e, 10);
4118 if (*e && *e != '\n')
4120 dev = MKDEV(major, minor);
4121 if (major != MAJOR(dev) ||
4122 minor != MINOR(dev))
4125 flush_workqueue(md_misc_wq);
4127 err = mddev_lock(mddev);
4130 if (mddev->persistent) {
4131 rdev = md_import_device(dev, mddev->major_version,
4132 mddev->minor_version);
4133 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4134 struct md_rdev *rdev0
4135 = list_entry(mddev->disks.next,
4136 struct md_rdev, same_set);
4137 err = super_types[mddev->major_version]
4138 .load_super(rdev, rdev0, mddev->minor_version);
4142 } else if (mddev->external)
4143 rdev = md_import_device(dev, -2, -1);
4145 rdev = md_import_device(dev, -1, -1);
4148 mddev_unlock(mddev);
4149 return PTR_ERR(rdev);
4151 err = bind_rdev_to_array(rdev, mddev);
4155 mddev_unlock(mddev);
4156 return err ? err : len;
4159 static struct md_sysfs_entry md_new_device =
4160 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4163 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4166 unsigned long chunk, end_chunk;
4169 err = mddev_lock(mddev);
4174 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4176 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4177 if (buf == end) break;
4178 if (*end == '-') { /* range */
4180 end_chunk = simple_strtoul(buf, &end, 0);
4181 if (buf == end) break;
4183 if (*end && !isspace(*end)) break;
4184 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4185 buf = skip_spaces(end);
4187 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4189 mddev_unlock(mddev);
4193 static struct md_sysfs_entry md_bitmap =
4194 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4197 size_show(struct mddev *mddev, char *page)
4199 return sprintf(page, "%llu\n",
4200 (unsigned long long)mddev->dev_sectors / 2);
4203 static int update_size(struct mddev *mddev, sector_t num_sectors);
4206 size_store(struct mddev *mddev, const char *buf, size_t len)
4208 /* If array is inactive, we can reduce the component size, but
4209 * not increase it (except from 0).
4210 * If array is active, we can try an on-line resize
4213 int err = strict_blocks_to_sectors(buf, §ors);
4217 err = mddev_lock(mddev);
4221 err = update_size(mddev, sectors);
4223 md_update_sb(mddev, 1);
4225 if (mddev->dev_sectors == 0 ||
4226 mddev->dev_sectors > sectors)
4227 mddev->dev_sectors = sectors;
4231 mddev_unlock(mddev);
4232 return err ? err : len;
4235 static struct md_sysfs_entry md_size =
4236 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4238 /* Metadata version.
4240 * 'none' for arrays with no metadata (good luck...)
4241 * 'external' for arrays with externally managed metadata,
4242 * or N.M for internally known formats
4245 metadata_show(struct mddev *mddev, char *page)
4247 if (mddev->persistent)
4248 return sprintf(page, "%d.%d\n",
4249 mddev->major_version, mddev->minor_version);
4250 else if (mddev->external)
4251 return sprintf(page, "external:%s\n", mddev->metadata_type);
4253 return sprintf(page, "none\n");
4257 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4262 /* Changing the details of 'external' metadata is
4263 * always permitted. Otherwise there must be
4264 * no devices attached to the array.
4267 err = mddev_lock(mddev);
4271 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4273 else if (!list_empty(&mddev->disks))
4277 if (cmd_match(buf, "none")) {
4278 mddev->persistent = 0;
4279 mddev->external = 0;
4280 mddev->major_version = 0;
4281 mddev->minor_version = 90;
4284 if (strncmp(buf, "external:", 9) == 0) {
4285 size_t namelen = len-9;
4286 if (namelen >= sizeof(mddev->metadata_type))
4287 namelen = sizeof(mddev->metadata_type)-1;
4288 strncpy(mddev->metadata_type, buf+9, namelen);
4289 mddev->metadata_type[namelen] = 0;
4290 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4291 mddev->metadata_type[--namelen] = 0;
4292 mddev->persistent = 0;
4293 mddev->external = 1;
4294 mddev->major_version = 0;
4295 mddev->minor_version = 90;
4298 major = simple_strtoul(buf, &e, 10);
4300 if (e==buf || *e != '.')
4303 minor = simple_strtoul(buf, &e, 10);
4304 if (e==buf || (*e && *e != '\n') )
4307 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4309 mddev->major_version = major;
4310 mddev->minor_version = minor;
4311 mddev->persistent = 1;
4312 mddev->external = 0;
4315 mddev_unlock(mddev);
4319 static struct md_sysfs_entry md_metadata =
4320 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4323 action_show(struct mddev *mddev, char *page)
4325 char *type = "idle";
4326 unsigned long recovery = mddev->recovery;
4327 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4329 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4330 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4331 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4333 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4334 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4336 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4340 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4342 else if (mddev->reshape_position != MaxSector)
4345 return sprintf(page, "%s\n", type);
4349 action_store(struct mddev *mddev, const char *page, size_t len)
4351 if (!mddev->pers || !mddev->pers->sync_request)
4355 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4356 if (cmd_match(page, "frozen"))
4357 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4359 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4360 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4361 mddev_lock(mddev) == 0) {
4362 flush_workqueue(md_misc_wq);
4363 if (mddev->sync_thread) {
4364 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4365 md_reap_sync_thread(mddev);
4367 mddev_unlock(mddev);
4369 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4371 else if (cmd_match(page, "resync"))
4372 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4373 else if (cmd_match(page, "recover")) {
4374 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4375 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4376 } else if (cmd_match(page, "reshape")) {
4378 if (mddev->pers->start_reshape == NULL)
4380 err = mddev_lock(mddev);
4382 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4385 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4386 err = mddev->pers->start_reshape(mddev);
4388 mddev_unlock(mddev);
4392 sysfs_notify(&mddev->kobj, NULL, "degraded");
4394 if (cmd_match(page, "check"))
4395 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4396 else if (!cmd_match(page, "repair"))
4398 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4399 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4400 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4402 if (mddev->ro == 2) {
4403 /* A write to sync_action is enough to justify
4404 * canceling read-auto mode
4407 md_wakeup_thread(mddev->sync_thread);
4409 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4410 md_wakeup_thread(mddev->thread);
4411 sysfs_notify_dirent_safe(mddev->sysfs_action);
4415 static struct md_sysfs_entry md_scan_mode =
4416 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4419 last_sync_action_show(struct mddev *mddev, char *page)
4421 return sprintf(page, "%s\n", mddev->last_sync_action);
4424 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4427 mismatch_cnt_show(struct mddev *mddev, char *page)
4429 return sprintf(page, "%llu\n",
4430 (unsigned long long)
4431 atomic64_read(&mddev->resync_mismatches));
4434 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4437 sync_min_show(struct mddev *mddev, char *page)
4439 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4440 mddev->sync_speed_min ? "local": "system");
4444 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4449 if (strncmp(buf, "system", 6)==0) {
4452 rv = kstrtouint(buf, 10, &min);
4458 mddev->sync_speed_min = min;
4462 static struct md_sysfs_entry md_sync_min =
4463 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4466 sync_max_show(struct mddev *mddev, char *page)
4468 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4469 mddev->sync_speed_max ? "local": "system");
4473 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4478 if (strncmp(buf, "system", 6)==0) {
4481 rv = kstrtouint(buf, 10, &max);
4487 mddev->sync_speed_max = max;
4491 static struct md_sysfs_entry md_sync_max =
4492 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4495 degraded_show(struct mddev *mddev, char *page)
4497 return sprintf(page, "%d\n", mddev->degraded);
4499 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4502 sync_force_parallel_show(struct mddev *mddev, char *page)
4504 return sprintf(page, "%d\n", mddev->parallel_resync);
4508 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4512 if (kstrtol(buf, 10, &n))
4515 if (n != 0 && n != 1)
4518 mddev->parallel_resync = n;
4520 if (mddev->sync_thread)
4521 wake_up(&resync_wait);
4526 /* force parallel resync, even with shared block devices */
4527 static struct md_sysfs_entry md_sync_force_parallel =
4528 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4529 sync_force_parallel_show, sync_force_parallel_store);
4532 sync_speed_show(struct mddev *mddev, char *page)
4534 unsigned long resync, dt, db;
4535 if (mddev->curr_resync == 0)
4536 return sprintf(page, "none\n");
4537 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4538 dt = (jiffies - mddev->resync_mark) / HZ;
4540 db = resync - mddev->resync_mark_cnt;
4541 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4544 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4547 sync_completed_show(struct mddev *mddev, char *page)
4549 unsigned long long max_sectors, resync;
4551 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4552 return sprintf(page, "none\n");
4554 if (mddev->curr_resync == 1 ||
4555 mddev->curr_resync == 2)
4556 return sprintf(page, "delayed\n");
4558 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4559 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4560 max_sectors = mddev->resync_max_sectors;
4562 max_sectors = mddev->dev_sectors;
4564 resync = mddev->curr_resync_completed;
4565 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4568 static struct md_sysfs_entry md_sync_completed =
4569 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4572 min_sync_show(struct mddev *mddev, char *page)
4574 return sprintf(page, "%llu\n",
4575 (unsigned long long)mddev->resync_min);
4578 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4580 unsigned long long min;
4583 if (kstrtoull(buf, 10, &min))
4586 spin_lock(&mddev->lock);
4588 if (min > mddev->resync_max)
4592 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4595 /* Round down to multiple of 4K for safety */
4596 mddev->resync_min = round_down(min, 8);
4600 spin_unlock(&mddev->lock);
4604 static struct md_sysfs_entry md_min_sync =
4605 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4608 max_sync_show(struct mddev *mddev, char *page)
4610 if (mddev->resync_max == MaxSector)
4611 return sprintf(page, "max\n");
4613 return sprintf(page, "%llu\n",
4614 (unsigned long long)mddev->resync_max);
4617 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4620 spin_lock(&mddev->lock);
4621 if (strncmp(buf, "max", 3) == 0)
4622 mddev->resync_max = MaxSector;
4624 unsigned long long max;
4628 if (kstrtoull(buf, 10, &max))
4630 if (max < mddev->resync_min)
4634 if (max < mddev->resync_max &&
4636 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4639 /* Must be a multiple of chunk_size */
4640 chunk = mddev->chunk_sectors;
4642 sector_t temp = max;
4645 if (sector_div(temp, chunk))
4648 mddev->resync_max = max;
4650 wake_up(&mddev->recovery_wait);
4653 spin_unlock(&mddev->lock);
4657 static struct md_sysfs_entry md_max_sync =
4658 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4661 suspend_lo_show(struct mddev *mddev, char *page)
4663 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4667 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4669 unsigned long long old, new;
4672 err = kstrtoull(buf, 10, &new);
4675 if (new != (sector_t)new)
4678 err = mddev_lock(mddev);
4682 if (mddev->pers == NULL ||
4683 mddev->pers->quiesce == NULL)
4685 old = mddev->suspend_lo;
4686 mddev->suspend_lo = new;
4688 /* Shrinking suspended region */
4689 mddev->pers->quiesce(mddev, 2);
4691 /* Expanding suspended region - need to wait */
4692 mddev->pers->quiesce(mddev, 1);
4693 mddev->pers->quiesce(mddev, 0);
4697 mddev_unlock(mddev);
4700 static struct md_sysfs_entry md_suspend_lo =
4701 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4704 suspend_hi_show(struct mddev *mddev, char *page)
4706 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4710 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4712 unsigned long long old, new;
4715 err = kstrtoull(buf, 10, &new);
4718 if (new != (sector_t)new)
4721 err = mddev_lock(mddev);
4725 if (mddev->pers == NULL ||
4726 mddev->pers->quiesce == NULL)
4728 old = mddev->suspend_hi;
4729 mddev->suspend_hi = new;
4731 /* Shrinking suspended region */
4732 mddev->pers->quiesce(mddev, 2);
4734 /* Expanding suspended region - need to wait */
4735 mddev->pers->quiesce(mddev, 1);
4736 mddev->pers->quiesce(mddev, 0);
4740 mddev_unlock(mddev);
4743 static struct md_sysfs_entry md_suspend_hi =
4744 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4747 reshape_position_show(struct mddev *mddev, char *page)
4749 if (mddev->reshape_position != MaxSector)
4750 return sprintf(page, "%llu\n",
4751 (unsigned long long)mddev->reshape_position);
4752 strcpy(page, "none\n");
4757 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4759 struct md_rdev *rdev;
4760 unsigned long long new;
4763 err = kstrtoull(buf, 10, &new);
4766 if (new != (sector_t)new)
4768 err = mddev_lock(mddev);
4774 mddev->reshape_position = new;
4775 mddev->delta_disks = 0;
4776 mddev->reshape_backwards = 0;
4777 mddev->new_level = mddev->level;
4778 mddev->new_layout = mddev->layout;
4779 mddev->new_chunk_sectors = mddev->chunk_sectors;
4780 rdev_for_each(rdev, mddev)
4781 rdev->new_data_offset = rdev->data_offset;
4784 mddev_unlock(mddev);
4788 static struct md_sysfs_entry md_reshape_position =
4789 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4790 reshape_position_store);
4793 reshape_direction_show(struct mddev *mddev, char *page)
4795 return sprintf(page, "%s\n",
4796 mddev->reshape_backwards ? "backwards" : "forwards");
4800 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4805 if (cmd_match(buf, "forwards"))
4807 else if (cmd_match(buf, "backwards"))
4811 if (mddev->reshape_backwards == backwards)
4814 err = mddev_lock(mddev);
4817 /* check if we are allowed to change */
4818 if (mddev->delta_disks)
4820 else if (mddev->persistent &&
4821 mddev->major_version == 0)
4824 mddev->reshape_backwards = backwards;
4825 mddev_unlock(mddev);
4829 static struct md_sysfs_entry md_reshape_direction =
4830 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4831 reshape_direction_store);
4834 array_size_show(struct mddev *mddev, char *page)
4836 if (mddev->external_size)
4837 return sprintf(page, "%llu\n",
4838 (unsigned long long)mddev->array_sectors/2);
4840 return sprintf(page, "default\n");
4844 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4849 err = mddev_lock(mddev);
4853 /* cluster raid doesn't support change array_sectors */
4854 if (mddev_is_clustered(mddev))
4857 if (strncmp(buf, "default", 7) == 0) {
4859 sectors = mddev->pers->size(mddev, 0, 0);
4861 sectors = mddev->array_sectors;
4863 mddev->external_size = 0;
4865 if (strict_blocks_to_sectors(buf, §ors) < 0)
4867 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4870 mddev->external_size = 1;
4874 mddev->array_sectors = sectors;
4876 set_capacity(mddev->gendisk, mddev->array_sectors);
4877 revalidate_disk(mddev->gendisk);
4880 mddev_unlock(mddev);
4884 static struct md_sysfs_entry md_array_size =
4885 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4888 static struct attribute *md_default_attrs[] = {
4891 &md_raid_disks.attr,
4892 &md_chunk_size.attr,
4894 &md_resync_start.attr,
4896 &md_new_device.attr,
4897 &md_safe_delay.attr,
4898 &md_array_state.attr,
4899 &md_reshape_position.attr,
4900 &md_reshape_direction.attr,
4901 &md_array_size.attr,
4902 &max_corr_read_errors.attr,
4906 static struct attribute *md_redundancy_attrs[] = {
4908 &md_last_scan_mode.attr,
4909 &md_mismatches.attr,
4912 &md_sync_speed.attr,
4913 &md_sync_force_parallel.attr,
4914 &md_sync_completed.attr,
4917 &md_suspend_lo.attr,
4918 &md_suspend_hi.attr,
4923 static struct attribute_group md_redundancy_group = {
4925 .attrs = md_redundancy_attrs,
4929 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4931 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4932 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4937 spin_lock(&all_mddevs_lock);
4938 if (list_empty(&mddev->all_mddevs)) {
4939 spin_unlock(&all_mddevs_lock);
4943 spin_unlock(&all_mddevs_lock);
4945 rv = entry->show(mddev, page);
4951 md_attr_store(struct kobject *kobj, struct attribute *attr,
4952 const char *page, size_t length)
4954 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4955 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4960 if (!capable(CAP_SYS_ADMIN))
4962 spin_lock(&all_mddevs_lock);
4963 if (list_empty(&mddev->all_mddevs)) {
4964 spin_unlock(&all_mddevs_lock);
4968 spin_unlock(&all_mddevs_lock);
4969 rv = entry->store(mddev, page, length);
4974 static void md_free(struct kobject *ko)
4976 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4978 if (mddev->sysfs_state)
4979 sysfs_put(mddev->sysfs_state);
4982 blk_cleanup_queue(mddev->queue);
4983 if (mddev->gendisk) {
4984 del_gendisk(mddev->gendisk);
4985 put_disk(mddev->gendisk);
4991 static const struct sysfs_ops md_sysfs_ops = {
4992 .show = md_attr_show,
4993 .store = md_attr_store,
4995 static struct kobj_type md_ktype = {
4997 .sysfs_ops = &md_sysfs_ops,
4998 .default_attrs = md_default_attrs,
5003 static void mddev_delayed_delete(struct work_struct *ws)
5005 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5007 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5008 kobject_del(&mddev->kobj);
5009 kobject_put(&mddev->kobj);
5012 static int md_alloc(dev_t dev, char *name)
5014 static DEFINE_MUTEX(disks_mutex);
5015 struct mddev *mddev = mddev_find(dev);
5016 struct gendisk *disk;
5025 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5026 shift = partitioned ? MdpMinorShift : 0;
5027 unit = MINOR(mddev->unit) >> shift;
5029 /* wait for any previous instance of this device to be
5030 * completely removed (mddev_delayed_delete).
5032 flush_workqueue(md_misc_wq);
5034 mutex_lock(&disks_mutex);
5040 /* Need to ensure that 'name' is not a duplicate.
5042 struct mddev *mddev2;
5043 spin_lock(&all_mddevs_lock);
5045 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5046 if (mddev2->gendisk &&
5047 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5048 spin_unlock(&all_mddevs_lock);
5051 spin_unlock(&all_mddevs_lock);
5055 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5058 mddev->queue->queuedata = mddev;
5060 blk_queue_make_request(mddev->queue, md_make_request);
5061 blk_set_stacking_limits(&mddev->queue->limits);
5063 disk = alloc_disk(1 << shift);
5065 blk_cleanup_queue(mddev->queue);
5066 mddev->queue = NULL;
5069 disk->major = MAJOR(mddev->unit);
5070 disk->first_minor = unit << shift;
5072 strcpy(disk->disk_name, name);
5073 else if (partitioned)
5074 sprintf(disk->disk_name, "md_d%d", unit);
5076 sprintf(disk->disk_name, "md%d", unit);
5077 disk->fops = &md_fops;
5078 disk->private_data = mddev;
5079 disk->queue = mddev->queue;
5080 blk_queue_write_cache(mddev->queue, true, true);
5081 /* Allow extended partitions. This makes the
5082 * 'mdp' device redundant, but we can't really
5085 disk->flags |= GENHD_FL_EXT_DEVT;
5086 mddev->gendisk = disk;
5087 /* As soon as we call add_disk(), another thread could get
5088 * through to md_open, so make sure it doesn't get too far
5090 mutex_lock(&mddev->open_mutex);
5093 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5094 &disk_to_dev(disk)->kobj, "%s", "md");
5096 /* This isn't possible, but as kobject_init_and_add is marked
5097 * __must_check, we must do something with the result
5099 pr_debug("md: cannot register %s/md - name in use\n",
5103 if (mddev->kobj.sd &&
5104 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5105 pr_debug("pointless warning\n");
5106 mutex_unlock(&mddev->open_mutex);
5108 mutex_unlock(&disks_mutex);
5109 if (!error && mddev->kobj.sd) {
5110 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5111 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5117 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5119 md_alloc(dev, NULL);
5123 static int add_named_array(const char *val, struct kernel_param *kp)
5125 /* val must be "md_*" where * is not all digits.
5126 * We allocate an array with a large free minor number, and
5127 * set the name to val. val must not already be an active name.
5129 int len = strlen(val);
5130 char buf[DISK_NAME_LEN];
5132 while (len && val[len-1] == '\n')
5134 if (len >= DISK_NAME_LEN)
5136 strlcpy(buf, val, len+1);
5137 if (strncmp(buf, "md_", 3) != 0)
5139 return md_alloc(0, buf);
5142 static void md_safemode_timeout(unsigned long data)
5144 struct mddev *mddev = (struct mddev *) data;
5146 if (!atomic_read(&mddev->writes_pending)) {
5147 mddev->safemode = 1;
5148 if (mddev->external)
5149 sysfs_notify_dirent_safe(mddev->sysfs_state);
5151 md_wakeup_thread(mddev->thread);
5154 static int start_dirty_degraded;
5156 int md_run(struct mddev *mddev)
5159 struct md_rdev *rdev;
5160 struct md_personality *pers;
5162 if (list_empty(&mddev->disks))
5163 /* cannot run an array with no devices.. */
5168 /* Cannot run until previous stop completes properly */
5169 if (mddev->sysfs_active)
5173 * Analyze all RAID superblock(s)
5175 if (!mddev->raid_disks) {
5176 if (!mddev->persistent)
5181 if (mddev->level != LEVEL_NONE)
5182 request_module("md-level-%d", mddev->level);
5183 else if (mddev->clevel[0])
5184 request_module("md-%s", mddev->clevel);
5187 * Drop all container device buffers, from now on
5188 * the only valid external interface is through the md
5191 rdev_for_each(rdev, mddev) {
5192 if (test_bit(Faulty, &rdev->flags))
5194 sync_blockdev(rdev->bdev);
5195 invalidate_bdev(rdev->bdev);
5197 /* perform some consistency tests on the device.
5198 * We don't want the data to overlap the metadata,
5199 * Internal Bitmap issues have been handled elsewhere.
5201 if (rdev->meta_bdev) {
5202 /* Nothing to check */;
5203 } else if (rdev->data_offset < rdev->sb_start) {
5204 if (mddev->dev_sectors &&
5205 rdev->data_offset + mddev->dev_sectors
5207 pr_warn("md: %s: data overlaps metadata\n",
5212 if (rdev->sb_start + rdev->sb_size/512
5213 > rdev->data_offset) {
5214 pr_warn("md: %s: metadata overlaps data\n",
5219 sysfs_notify_dirent_safe(rdev->sysfs_state);
5222 if (mddev->bio_set == NULL) {
5223 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5224 if (!mddev->bio_set)
5228 spin_lock(&pers_lock);
5229 pers = find_pers(mddev->level, mddev->clevel);
5230 if (!pers || !try_module_get(pers->owner)) {
5231 spin_unlock(&pers_lock);
5232 if (mddev->level != LEVEL_NONE)
5233 pr_warn("md: personality for level %d is not loaded!\n",
5236 pr_warn("md: personality for level %s is not loaded!\n",
5240 spin_unlock(&pers_lock);
5241 if (mddev->level != pers->level) {
5242 mddev->level = pers->level;
5243 mddev->new_level = pers->level;
5245 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5247 if (mddev->reshape_position != MaxSector &&
5248 pers->start_reshape == NULL) {
5249 /* This personality cannot handle reshaping... */
5250 module_put(pers->owner);
5254 if (pers->sync_request) {
5255 /* Warn if this is a potentially silly
5258 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5259 struct md_rdev *rdev2;
5262 rdev_for_each(rdev, mddev)
5263 rdev_for_each(rdev2, mddev) {
5265 rdev->bdev->bd_contains ==
5266 rdev2->bdev->bd_contains) {
5267 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5269 bdevname(rdev->bdev,b),
5270 bdevname(rdev2->bdev,b2));
5276 pr_warn("True protection against single-disk failure might be compromised.\n");
5279 mddev->recovery = 0;
5280 /* may be over-ridden by personality */
5281 mddev->resync_max_sectors = mddev->dev_sectors;
5283 mddev->ok_start_degraded = start_dirty_degraded;
5285 if (start_readonly && mddev->ro == 0)
5286 mddev->ro = 2; /* read-only, but switch on first write */
5289 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
5290 * up mddev->thread. It is important to initialize critical
5291 * resources for mddev->thread BEFORE calling pers->run().
5293 err = pers->run(mddev);
5295 pr_warn("md: pers->run() failed ...\n");
5296 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5297 WARN_ONCE(!mddev->external_size,
5298 "%s: default size too small, but 'external_size' not in effect?\n",
5300 pr_warn("md: invalid array_size %llu > default size %llu\n",
5301 (unsigned long long)mddev->array_sectors / 2,
5302 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5305 if (err == 0 && pers->sync_request &&
5306 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5307 struct bitmap *bitmap;
5309 bitmap = bitmap_create(mddev, -1);
5310 if (IS_ERR(bitmap)) {
5311 err = PTR_ERR(bitmap);
5312 pr_warn("%s: failed to create bitmap (%d)\n",
5313 mdname(mddev), err);
5315 mddev->bitmap = bitmap;
5319 mddev_detach(mddev);
5321 pers->free(mddev, mddev->private);
5322 mddev->private = NULL;
5323 module_put(pers->owner);
5324 bitmap_destroy(mddev);
5330 rdev_for_each(rdev, mddev) {
5331 if (rdev->raid_disk >= 0 &&
5332 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5337 if (mddev->degraded)
5340 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5342 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5343 mddev->queue->backing_dev_info->congested_data = mddev;
5344 mddev->queue->backing_dev_info->congested_fn = md_congested;
5346 if (pers->sync_request) {
5347 if (mddev->kobj.sd &&
5348 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5349 pr_warn("md: cannot register extra attributes for %s\n",
5351 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5352 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5355 atomic_set(&mddev->writes_pending,0);
5356 atomic_set(&mddev->max_corr_read_errors,
5357 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5358 mddev->safemode = 0;
5359 if (mddev_is_clustered(mddev))
5360 mddev->safemode_delay = 0;
5362 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5365 spin_lock(&mddev->lock);
5367 spin_unlock(&mddev->lock);
5368 rdev_for_each(rdev, mddev)
5369 if (rdev->raid_disk >= 0)
5370 if (sysfs_link_rdev(mddev, rdev))
5371 /* failure here is OK */;
5373 if (mddev->degraded && !mddev->ro)
5374 /* This ensures that recovering status is reported immediately
5375 * via sysfs - until a lack of spares is confirmed.
5377 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5378 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5380 if (mddev->sb_flags)
5381 md_update_sb(mddev, 0);
5383 md_new_event(mddev);
5384 sysfs_notify_dirent_safe(mddev->sysfs_state);
5385 sysfs_notify_dirent_safe(mddev->sysfs_action);
5386 sysfs_notify(&mddev->kobj, NULL, "degraded");
5389 EXPORT_SYMBOL_GPL(md_run);
5391 static int do_md_run(struct mddev *mddev)
5395 err = md_run(mddev);
5398 err = bitmap_load(mddev);
5400 bitmap_destroy(mddev);
5404 if (mddev_is_clustered(mddev))
5405 md_allow_write(mddev);
5407 md_wakeup_thread(mddev->thread);
5408 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5410 set_capacity(mddev->gendisk, mddev->array_sectors);
5411 revalidate_disk(mddev->gendisk);
5413 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5418 static int restart_array(struct mddev *mddev)
5420 struct gendisk *disk = mddev->gendisk;
5422 /* Complain if it has no devices */
5423 if (list_empty(&mddev->disks))
5429 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5430 struct md_rdev *rdev;
5431 bool has_journal = false;
5434 rdev_for_each_rcu(rdev, mddev) {
5435 if (test_bit(Journal, &rdev->flags) &&
5436 !test_bit(Faulty, &rdev->flags)) {
5443 /* Don't restart rw with journal missing/faulty */
5448 mddev->safemode = 0;
5450 set_disk_ro(disk, 0);
5451 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5452 /* Kick recovery or resync if necessary */
5453 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5454 md_wakeup_thread(mddev->thread);
5455 md_wakeup_thread(mddev->sync_thread);
5456 sysfs_notify_dirent_safe(mddev->sysfs_state);
5460 static void md_clean(struct mddev *mddev)
5462 mddev->array_sectors = 0;
5463 mddev->external_size = 0;
5464 mddev->dev_sectors = 0;
5465 mddev->raid_disks = 0;
5466 mddev->recovery_cp = 0;
5467 mddev->resync_min = 0;
5468 mddev->resync_max = MaxSector;
5469 mddev->reshape_position = MaxSector;
5470 mddev->external = 0;
5471 mddev->persistent = 0;
5472 mddev->level = LEVEL_NONE;
5473 mddev->clevel[0] = 0;
5475 mddev->sb_flags = 0;
5477 mddev->metadata_type[0] = 0;
5478 mddev->chunk_sectors = 0;
5479 mddev->ctime = mddev->utime = 0;
5481 mddev->max_disks = 0;
5483 mddev->can_decrease_events = 0;
5484 mddev->delta_disks = 0;
5485 mddev->reshape_backwards = 0;
5486 mddev->new_level = LEVEL_NONE;
5487 mddev->new_layout = 0;
5488 mddev->new_chunk_sectors = 0;
5489 mddev->curr_resync = 0;
5490 atomic64_set(&mddev->resync_mismatches, 0);
5491 mddev->suspend_lo = mddev->suspend_hi = 0;
5492 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5493 mddev->recovery = 0;
5496 mddev->degraded = 0;
5497 mddev->safemode = 0;
5498 mddev->private = NULL;
5499 mddev->cluster_info = NULL;
5500 mddev->bitmap_info.offset = 0;
5501 mddev->bitmap_info.default_offset = 0;
5502 mddev->bitmap_info.default_space = 0;
5503 mddev->bitmap_info.chunksize = 0;
5504 mddev->bitmap_info.daemon_sleep = 0;
5505 mddev->bitmap_info.max_write_behind = 0;
5506 mddev->bitmap_info.nodes = 0;
5509 static void __md_stop_writes(struct mddev *mddev)
5511 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5512 flush_workqueue(md_misc_wq);
5513 if (mddev->sync_thread) {
5514 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5515 md_reap_sync_thread(mddev);
5518 del_timer_sync(&mddev->safemode_timer);
5520 if (mddev->pers && mddev->pers->quiesce) {
5521 mddev->pers->quiesce(mddev, 1);
5522 mddev->pers->quiesce(mddev, 0);
5524 bitmap_flush(mddev);
5526 if (mddev->ro == 0 &&
5527 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5529 /* mark array as shutdown cleanly */
5530 if (!mddev_is_clustered(mddev))
5532 md_update_sb(mddev, 1);
5536 void md_stop_writes(struct mddev *mddev)
5538 mddev_lock_nointr(mddev);
5539 __md_stop_writes(mddev);
5540 mddev_unlock(mddev);
5542 EXPORT_SYMBOL_GPL(md_stop_writes);
5544 static void mddev_detach(struct mddev *mddev)
5546 struct bitmap *bitmap = mddev->bitmap;
5547 /* wait for behind writes to complete */
5548 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5549 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
5551 /* need to kick something here to make sure I/O goes? */
5552 wait_event(bitmap->behind_wait,
5553 atomic_read(&bitmap->behind_writes) == 0);
5555 if (mddev->pers && mddev->pers->quiesce) {
5556 mddev->pers->quiesce(mddev, 1);
5557 mddev->pers->quiesce(mddev, 0);
5559 md_unregister_thread(&mddev->thread);
5561 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5564 static void __md_stop(struct mddev *mddev)
5566 struct md_personality *pers = mddev->pers;
5567 mddev_detach(mddev);
5568 /* Ensure ->event_work is done */
5569 flush_workqueue(md_misc_wq);
5570 spin_lock(&mddev->lock);
5572 spin_unlock(&mddev->lock);
5573 pers->free(mddev, mddev->private);
5574 mddev->private = NULL;
5575 if (pers->sync_request && mddev->to_remove == NULL)
5576 mddev->to_remove = &md_redundancy_group;
5577 module_put(pers->owner);
5578 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5581 void md_stop(struct mddev *mddev)
5583 /* stop the array and free an attached data structures.
5584 * This is called from dm-raid
5587 bitmap_destroy(mddev);
5589 bioset_free(mddev->bio_set);
5592 EXPORT_SYMBOL_GPL(md_stop);
5594 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5599 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5601 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5602 md_wakeup_thread(mddev->thread);
5604 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5605 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5606 if (mddev->sync_thread)
5607 /* Thread might be blocked waiting for metadata update
5608 * which will now never happen */
5609 wake_up_process(mddev->sync_thread->tsk);
5611 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5613 mddev_unlock(mddev);
5614 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5616 wait_event(mddev->sb_wait,
5617 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5618 mddev_lock_nointr(mddev);
5620 mutex_lock(&mddev->open_mutex);
5621 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5622 mddev->sync_thread ||
5623 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5624 pr_warn("md: %s still in use.\n",mdname(mddev));
5626 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5627 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5628 md_wakeup_thread(mddev->thread);
5634 __md_stop_writes(mddev);
5640 set_disk_ro(mddev->gendisk, 1);
5641 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5642 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5643 md_wakeup_thread(mddev->thread);
5644 sysfs_notify_dirent_safe(mddev->sysfs_state);
5648 mutex_unlock(&mddev->open_mutex);
5653 * 0 - completely stop and dis-assemble array
5654 * 2 - stop but do not disassemble array
5656 static int do_md_stop(struct mddev *mddev, int mode,
5657 struct block_device *bdev)
5659 struct gendisk *disk = mddev->gendisk;
5660 struct md_rdev *rdev;
5663 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5665 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5666 md_wakeup_thread(mddev->thread);
5668 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5669 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5670 if (mddev->sync_thread)
5671 /* Thread might be blocked waiting for metadata update
5672 * which will now never happen */
5673 wake_up_process(mddev->sync_thread->tsk);
5675 mddev_unlock(mddev);
5676 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5677 !test_bit(MD_RECOVERY_RUNNING,
5678 &mddev->recovery)));
5679 mddev_lock_nointr(mddev);
5681 mutex_lock(&mddev->open_mutex);
5682 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5683 mddev->sysfs_active ||
5684 mddev->sync_thread ||
5685 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5686 pr_warn("md: %s still in use.\n",mdname(mddev));
5687 mutex_unlock(&mddev->open_mutex);
5689 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5690 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5691 md_wakeup_thread(mddev->thread);
5697 set_disk_ro(disk, 0);
5699 __md_stop_writes(mddev);
5701 mddev->queue->backing_dev_info->congested_fn = NULL;
5703 /* tell userspace to handle 'inactive' */
5704 sysfs_notify_dirent_safe(mddev->sysfs_state);
5706 rdev_for_each(rdev, mddev)
5707 if (rdev->raid_disk >= 0)
5708 sysfs_unlink_rdev(mddev, rdev);
5710 set_capacity(disk, 0);
5711 mutex_unlock(&mddev->open_mutex);
5713 revalidate_disk(disk);
5718 mutex_unlock(&mddev->open_mutex);
5720 * Free resources if final stop
5723 pr_info("md: %s stopped.\n", mdname(mddev));
5725 bitmap_destroy(mddev);
5726 if (mddev->bitmap_info.file) {
5727 struct file *f = mddev->bitmap_info.file;
5728 spin_lock(&mddev->lock);
5729 mddev->bitmap_info.file = NULL;
5730 spin_unlock(&mddev->lock);
5733 mddev->bitmap_info.offset = 0;
5735 export_array(mddev);
5738 if (mddev->hold_active == UNTIL_STOP)
5739 mddev->hold_active = 0;
5741 md_new_event(mddev);
5742 sysfs_notify_dirent_safe(mddev->sysfs_state);
5747 static void autorun_array(struct mddev *mddev)
5749 struct md_rdev *rdev;
5752 if (list_empty(&mddev->disks))
5755 pr_info("md: running: ");
5757 rdev_for_each(rdev, mddev) {
5758 char b[BDEVNAME_SIZE];
5759 pr_cont("<%s>", bdevname(rdev->bdev,b));
5763 err = do_md_run(mddev);
5765 pr_warn("md: do_md_run() returned %d\n", err);
5766 do_md_stop(mddev, 0, NULL);
5771 * lets try to run arrays based on all disks that have arrived
5772 * until now. (those are in pending_raid_disks)
5774 * the method: pick the first pending disk, collect all disks with
5775 * the same UUID, remove all from the pending list and put them into
5776 * the 'same_array' list. Then order this list based on superblock
5777 * update time (freshest comes first), kick out 'old' disks and
5778 * compare superblocks. If everything's fine then run it.
5780 * If "unit" is allocated, then bump its reference count
5782 static void autorun_devices(int part)
5784 struct md_rdev *rdev0, *rdev, *tmp;
5785 struct mddev *mddev;
5786 char b[BDEVNAME_SIZE];
5788 pr_info("md: autorun ...\n");
5789 while (!list_empty(&pending_raid_disks)) {
5792 LIST_HEAD(candidates);
5793 rdev0 = list_entry(pending_raid_disks.next,
5794 struct md_rdev, same_set);
5796 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
5797 INIT_LIST_HEAD(&candidates);
5798 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5799 if (super_90_load(rdev, rdev0, 0) >= 0) {
5800 pr_debug("md: adding %s ...\n",
5801 bdevname(rdev->bdev,b));
5802 list_move(&rdev->same_set, &candidates);
5805 * now we have a set of devices, with all of them having
5806 * mostly sane superblocks. It's time to allocate the
5810 dev = MKDEV(mdp_major,
5811 rdev0->preferred_minor << MdpMinorShift);
5812 unit = MINOR(dev) >> MdpMinorShift;
5814 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5817 if (rdev0->preferred_minor != unit) {
5818 pr_warn("md: unit number in %s is bad: %d\n",
5819 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5823 md_probe(dev, NULL, NULL);
5824 mddev = mddev_find(dev);
5825 if (!mddev || !mddev->gendisk) {
5830 if (mddev_lock(mddev))
5831 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
5832 else if (mddev->raid_disks || mddev->major_version
5833 || !list_empty(&mddev->disks)) {
5834 pr_warn("md: %s already running, cannot run %s\n",
5835 mdname(mddev), bdevname(rdev0->bdev,b));
5836 mddev_unlock(mddev);
5838 pr_debug("md: created %s\n", mdname(mddev));
5839 mddev->persistent = 1;
5840 rdev_for_each_list(rdev, tmp, &candidates) {
5841 list_del_init(&rdev->same_set);
5842 if (bind_rdev_to_array(rdev, mddev))
5845 autorun_array(mddev);
5846 mddev_unlock(mddev);
5848 /* on success, candidates will be empty, on error
5851 rdev_for_each_list(rdev, tmp, &candidates) {
5852 list_del_init(&rdev->same_set);
5857 pr_info("md: ... autorun DONE.\n");
5859 #endif /* !MODULE */
5861 static int get_version(void __user *arg)
5865 ver.major = MD_MAJOR_VERSION;
5866 ver.minor = MD_MINOR_VERSION;
5867 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5869 if (copy_to_user(arg, &ver, sizeof(ver)))
5875 static int get_array_info(struct mddev *mddev, void __user *arg)
5877 mdu_array_info_t info;
5878 int nr,working,insync,failed,spare;
5879 struct md_rdev *rdev;
5881 nr = working = insync = failed = spare = 0;
5883 rdev_for_each_rcu(rdev, mddev) {
5885 if (test_bit(Faulty, &rdev->flags))
5889 if (test_bit(In_sync, &rdev->flags))
5891 else if (test_bit(Journal, &rdev->flags))
5892 /* TODO: add journal count to md_u.h */
5900 info.major_version = mddev->major_version;
5901 info.minor_version = mddev->minor_version;
5902 info.patch_version = MD_PATCHLEVEL_VERSION;
5903 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5904 info.level = mddev->level;
5905 info.size = mddev->dev_sectors / 2;
5906 if (info.size != mddev->dev_sectors / 2) /* overflow */
5909 info.raid_disks = mddev->raid_disks;
5910 info.md_minor = mddev->md_minor;
5911 info.not_persistent= !mddev->persistent;
5913 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5916 info.state = (1<<MD_SB_CLEAN);
5917 if (mddev->bitmap && mddev->bitmap_info.offset)
5918 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5919 if (mddev_is_clustered(mddev))
5920 info.state |= (1<<MD_SB_CLUSTERED);
5921 info.active_disks = insync;
5922 info.working_disks = working;
5923 info.failed_disks = failed;
5924 info.spare_disks = spare;
5926 info.layout = mddev->layout;
5927 info.chunk_size = mddev->chunk_sectors << 9;
5929 if (copy_to_user(arg, &info, sizeof(info)))
5935 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5937 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5941 file = kzalloc(sizeof(*file), GFP_NOIO);
5946 spin_lock(&mddev->lock);
5947 /* bitmap enabled */
5948 if (mddev->bitmap_info.file) {
5949 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5950 sizeof(file->pathname));
5954 memmove(file->pathname, ptr,
5955 sizeof(file->pathname)-(ptr-file->pathname));
5957 spin_unlock(&mddev->lock);
5960 copy_to_user(arg, file, sizeof(*file)))
5967 static int get_disk_info(struct mddev *mddev, void __user * arg)
5969 mdu_disk_info_t info;
5970 struct md_rdev *rdev;
5972 if (copy_from_user(&info, arg, sizeof(info)))
5976 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5978 info.major = MAJOR(rdev->bdev->bd_dev);
5979 info.minor = MINOR(rdev->bdev->bd_dev);
5980 info.raid_disk = rdev->raid_disk;
5982 if (test_bit(Faulty, &rdev->flags))
5983 info.state |= (1<<MD_DISK_FAULTY);
5984 else if (test_bit(In_sync, &rdev->flags)) {
5985 info.state |= (1<<MD_DISK_ACTIVE);
5986 info.state |= (1<<MD_DISK_SYNC);
5988 if (test_bit(Journal, &rdev->flags))
5989 info.state |= (1<<MD_DISK_JOURNAL);
5990 if (test_bit(WriteMostly, &rdev->flags))
5991 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5992 if (test_bit(FailFast, &rdev->flags))
5993 info.state |= (1<<MD_DISK_FAILFAST);
5995 info.major = info.minor = 0;
5996 info.raid_disk = -1;
5997 info.state = (1<<MD_DISK_REMOVED);
6001 if (copy_to_user(arg, &info, sizeof(info)))
6007 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6009 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6010 struct md_rdev *rdev;
6011 dev_t dev = MKDEV(info->major,info->minor);
6013 if (mddev_is_clustered(mddev) &&
6014 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6015 pr_warn("%s: Cannot add to clustered mddev.\n",
6020 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6023 if (!mddev->raid_disks) {
6025 /* expecting a device which has a superblock */
6026 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6028 pr_warn("md: md_import_device returned %ld\n",
6030 return PTR_ERR(rdev);
6032 if (!list_empty(&mddev->disks)) {
6033 struct md_rdev *rdev0
6034 = list_entry(mddev->disks.next,
6035 struct md_rdev, same_set);
6036 err = super_types[mddev->major_version]
6037 .load_super(rdev, rdev0, mddev->minor_version);
6039 pr_warn("md: %s has different UUID to %s\n",
6040 bdevname(rdev->bdev,b),
6041 bdevname(rdev0->bdev,b2));
6046 err = bind_rdev_to_array(rdev, mddev);
6053 * add_new_disk can be used once the array is assembled
6054 * to add "hot spares". They must already have a superblock
6059 if (!mddev->pers->hot_add_disk) {
6060 pr_warn("%s: personality does not support diskops!\n",
6064 if (mddev->persistent)
6065 rdev = md_import_device(dev, mddev->major_version,
6066 mddev->minor_version);
6068 rdev = md_import_device(dev, -1, -1);
6070 pr_warn("md: md_import_device returned %ld\n",
6072 return PTR_ERR(rdev);
6074 /* set saved_raid_disk if appropriate */
6075 if (!mddev->persistent) {
6076 if (info->state & (1<<MD_DISK_SYNC) &&
6077 info->raid_disk < mddev->raid_disks) {
6078 rdev->raid_disk = info->raid_disk;
6079 set_bit(In_sync, &rdev->flags);
6080 clear_bit(Bitmap_sync, &rdev->flags);
6082 rdev->raid_disk = -1;
6083 rdev->saved_raid_disk = rdev->raid_disk;
6085 super_types[mddev->major_version].
6086 validate_super(mddev, rdev);
6087 if ((info->state & (1<<MD_DISK_SYNC)) &&
6088 rdev->raid_disk != info->raid_disk) {
6089 /* This was a hot-add request, but events doesn't
6090 * match, so reject it.
6096 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6097 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6098 set_bit(WriteMostly, &rdev->flags);
6100 clear_bit(WriteMostly, &rdev->flags);
6101 if (info->state & (1<<MD_DISK_FAILFAST))
6102 set_bit(FailFast, &rdev->flags);
6104 clear_bit(FailFast, &rdev->flags);
6106 if (info->state & (1<<MD_DISK_JOURNAL)) {
6107 struct md_rdev *rdev2;
6108 bool has_journal = false;
6110 /* make sure no existing journal disk */
6111 rdev_for_each(rdev2, mddev) {
6112 if (test_bit(Journal, &rdev2->flags)) {
6121 set_bit(Journal, &rdev->flags);
6124 * check whether the device shows up in other nodes
6126 if (mddev_is_clustered(mddev)) {
6127 if (info->state & (1 << MD_DISK_CANDIDATE))
6128 set_bit(Candidate, &rdev->flags);
6129 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6130 /* --add initiated by this node */
6131 err = md_cluster_ops->add_new_disk(mddev, rdev);
6139 rdev->raid_disk = -1;
6140 err = bind_rdev_to_array(rdev, mddev);
6145 if (mddev_is_clustered(mddev)) {
6146 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6148 err = md_cluster_ops->new_disk_ack(mddev,
6151 md_kick_rdev_from_array(rdev);
6155 md_cluster_ops->add_new_disk_cancel(mddev);
6157 err = add_bound_rdev(rdev);
6161 err = add_bound_rdev(rdev);
6166 /* otherwise, add_new_disk is only allowed
6167 * for major_version==0 superblocks
6169 if (mddev->major_version != 0) {
6170 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6174 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6176 rdev = md_import_device(dev, -1, 0);
6178 pr_warn("md: error, md_import_device() returned %ld\n",
6180 return PTR_ERR(rdev);
6182 rdev->desc_nr = info->number;
6183 if (info->raid_disk < mddev->raid_disks)
6184 rdev->raid_disk = info->raid_disk;
6186 rdev->raid_disk = -1;
6188 if (rdev->raid_disk < mddev->raid_disks)
6189 if (info->state & (1<<MD_DISK_SYNC))
6190 set_bit(In_sync, &rdev->flags);
6192 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6193 set_bit(WriteMostly, &rdev->flags);
6194 if (info->state & (1<<MD_DISK_FAILFAST))
6195 set_bit(FailFast, &rdev->flags);
6197 if (!mddev->persistent) {
6198 pr_debug("md: nonpersistent superblock ...\n");
6199 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6201 rdev->sb_start = calc_dev_sboffset(rdev);
6202 rdev->sectors = rdev->sb_start;
6204 err = bind_rdev_to_array(rdev, mddev);
6214 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6216 char b[BDEVNAME_SIZE];
6217 struct md_rdev *rdev;
6219 rdev = find_rdev(mddev, dev);
6223 if (rdev->raid_disk < 0)
6226 clear_bit(Blocked, &rdev->flags);
6227 remove_and_add_spares(mddev, rdev);
6229 if (rdev->raid_disk >= 0)
6233 if (mddev_is_clustered(mddev))
6234 md_cluster_ops->remove_disk(mddev, rdev);
6236 md_kick_rdev_from_array(rdev);
6237 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6239 md_wakeup_thread(mddev->thread);
6241 md_update_sb(mddev, 1);
6242 md_new_event(mddev);
6246 pr_debug("md: cannot remove active disk %s from %s ...\n",
6247 bdevname(rdev->bdev,b), mdname(mddev));
6251 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6253 char b[BDEVNAME_SIZE];
6255 struct md_rdev *rdev;
6260 if (mddev->major_version != 0) {
6261 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6265 if (!mddev->pers->hot_add_disk) {
6266 pr_warn("%s: personality does not support diskops!\n",
6271 rdev = md_import_device(dev, -1, 0);
6273 pr_warn("md: error, md_import_device() returned %ld\n",
6278 if (mddev->persistent)
6279 rdev->sb_start = calc_dev_sboffset(rdev);
6281 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6283 rdev->sectors = rdev->sb_start;
6285 if (test_bit(Faulty, &rdev->flags)) {
6286 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6287 bdevname(rdev->bdev,b), mdname(mddev));
6292 clear_bit(In_sync, &rdev->flags);
6294 rdev->saved_raid_disk = -1;
6295 err = bind_rdev_to_array(rdev, mddev);
6300 * The rest should better be atomic, we can have disk failures
6301 * noticed in interrupt contexts ...
6304 rdev->raid_disk = -1;
6306 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6308 md_update_sb(mddev, 1);
6310 * Kick recovery, maybe this spare has to be added to the
6311 * array immediately.
6313 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6314 md_wakeup_thread(mddev->thread);
6315 md_new_event(mddev);
6323 static int set_bitmap_file(struct mddev *mddev, int fd)
6328 if (!mddev->pers->quiesce || !mddev->thread)
6330 if (mddev->recovery || mddev->sync_thread)
6332 /* we should be able to change the bitmap.. */
6336 struct inode *inode;
6339 if (mddev->bitmap || mddev->bitmap_info.file)
6340 return -EEXIST; /* cannot add when bitmap is present */
6344 pr_warn("%s: error: failed to get bitmap file\n",
6349 inode = f->f_mapping->host;
6350 if (!S_ISREG(inode->i_mode)) {
6351 pr_warn("%s: error: bitmap file must be a regular file\n",
6354 } else if (!(f->f_mode & FMODE_WRITE)) {
6355 pr_warn("%s: error: bitmap file must open for write\n",
6358 } else if (atomic_read(&inode->i_writecount) != 1) {
6359 pr_warn("%s: error: bitmap file is already in use\n",
6367 mddev->bitmap_info.file = f;
6368 mddev->bitmap_info.offset = 0; /* file overrides offset */
6369 } else if (mddev->bitmap == NULL)
6370 return -ENOENT; /* cannot remove what isn't there */
6373 mddev->pers->quiesce(mddev, 1);
6375 struct bitmap *bitmap;
6377 bitmap = bitmap_create(mddev, -1);
6378 if (!IS_ERR(bitmap)) {
6379 mddev->bitmap = bitmap;
6380 err = bitmap_load(mddev);
6382 err = PTR_ERR(bitmap);
6384 if (fd < 0 || err) {
6385 bitmap_destroy(mddev);
6386 fd = -1; /* make sure to put the file */
6388 mddev->pers->quiesce(mddev, 0);
6391 struct file *f = mddev->bitmap_info.file;
6393 spin_lock(&mddev->lock);
6394 mddev->bitmap_info.file = NULL;
6395 spin_unlock(&mddev->lock);
6404 * set_array_info is used two different ways
6405 * The original usage is when creating a new array.
6406 * In this usage, raid_disks is > 0 and it together with
6407 * level, size, not_persistent,layout,chunksize determine the
6408 * shape of the array.
6409 * This will always create an array with a type-0.90.0 superblock.
6410 * The newer usage is when assembling an array.
6411 * In this case raid_disks will be 0, and the major_version field is
6412 * use to determine which style super-blocks are to be found on the devices.
6413 * The minor and patch _version numbers are also kept incase the
6414 * super_block handler wishes to interpret them.
6416 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6419 if (info->raid_disks == 0) {
6420 /* just setting version number for superblock loading */
6421 if (info->major_version < 0 ||
6422 info->major_version >= ARRAY_SIZE(super_types) ||
6423 super_types[info->major_version].name == NULL) {
6424 /* maybe try to auto-load a module? */
6425 pr_warn("md: superblock version %d not known\n",
6426 info->major_version);
6429 mddev->major_version = info->major_version;
6430 mddev->minor_version = info->minor_version;
6431 mddev->patch_version = info->patch_version;
6432 mddev->persistent = !info->not_persistent;
6433 /* ensure mddev_put doesn't delete this now that there
6434 * is some minimal configuration.
6436 mddev->ctime = ktime_get_real_seconds();
6439 mddev->major_version = MD_MAJOR_VERSION;
6440 mddev->minor_version = MD_MINOR_VERSION;
6441 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6442 mddev->ctime = ktime_get_real_seconds();
6444 mddev->level = info->level;
6445 mddev->clevel[0] = 0;
6446 mddev->dev_sectors = 2 * (sector_t)info->size;
6447 mddev->raid_disks = info->raid_disks;
6448 /* don't set md_minor, it is determined by which /dev/md* was
6451 if (info->state & (1<<MD_SB_CLEAN))
6452 mddev->recovery_cp = MaxSector;
6454 mddev->recovery_cp = 0;
6455 mddev->persistent = ! info->not_persistent;
6456 mddev->external = 0;
6458 mddev->layout = info->layout;
6459 mddev->chunk_sectors = info->chunk_size >> 9;
6461 mddev->max_disks = MD_SB_DISKS;
6463 if (mddev->persistent) {
6465 mddev->sb_flags = 0;
6467 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6469 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6470 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6471 mddev->bitmap_info.offset = 0;
6473 mddev->reshape_position = MaxSector;
6476 * Generate a 128 bit UUID
6478 get_random_bytes(mddev->uuid, 16);
6480 mddev->new_level = mddev->level;
6481 mddev->new_chunk_sectors = mddev->chunk_sectors;
6482 mddev->new_layout = mddev->layout;
6483 mddev->delta_disks = 0;
6484 mddev->reshape_backwards = 0;
6489 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6491 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6493 if (mddev->external_size)
6496 mddev->array_sectors = array_sectors;
6498 EXPORT_SYMBOL(md_set_array_sectors);
6500 static int update_size(struct mddev *mddev, sector_t num_sectors)
6502 struct md_rdev *rdev;
6504 int fit = (num_sectors == 0);
6506 /* cluster raid doesn't support update size */
6507 if (mddev_is_clustered(mddev))
6510 if (mddev->pers->resize == NULL)
6512 /* The "num_sectors" is the number of sectors of each device that
6513 * is used. This can only make sense for arrays with redundancy.
6514 * linear and raid0 always use whatever space is available. We can only
6515 * consider changing this number if no resync or reconstruction is
6516 * happening, and if the new size is acceptable. It must fit before the
6517 * sb_start or, if that is <data_offset, it must fit before the size
6518 * of each device. If num_sectors is zero, we find the largest size
6521 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6527 rdev_for_each(rdev, mddev) {
6528 sector_t avail = rdev->sectors;
6530 if (fit && (num_sectors == 0 || num_sectors > avail))
6531 num_sectors = avail;
6532 if (avail < num_sectors)
6535 rv = mddev->pers->resize(mddev, num_sectors);
6537 revalidate_disk(mddev->gendisk);
6541 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6544 struct md_rdev *rdev;
6545 /* change the number of raid disks */
6546 if (mddev->pers->check_reshape == NULL)
6550 if (raid_disks <= 0 ||
6551 (mddev->max_disks && raid_disks >= mddev->max_disks))
6553 if (mddev->sync_thread ||
6554 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6555 mddev->reshape_position != MaxSector)
6558 rdev_for_each(rdev, mddev) {
6559 if (mddev->raid_disks < raid_disks &&
6560 rdev->data_offset < rdev->new_data_offset)
6562 if (mddev->raid_disks > raid_disks &&
6563 rdev->data_offset > rdev->new_data_offset)
6567 mddev->delta_disks = raid_disks - mddev->raid_disks;
6568 if (mddev->delta_disks < 0)
6569 mddev->reshape_backwards = 1;
6570 else if (mddev->delta_disks > 0)
6571 mddev->reshape_backwards = 0;
6573 rv = mddev->pers->check_reshape(mddev);
6575 mddev->delta_disks = 0;
6576 mddev->reshape_backwards = 0;
6582 * update_array_info is used to change the configuration of an
6584 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6585 * fields in the info are checked against the array.
6586 * Any differences that cannot be handled will cause an error.
6587 * Normally, only one change can be managed at a time.
6589 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6595 /* calculate expected state,ignoring low bits */
6596 if (mddev->bitmap && mddev->bitmap_info.offset)
6597 state |= (1 << MD_SB_BITMAP_PRESENT);
6599 if (mddev->major_version != info->major_version ||
6600 mddev->minor_version != info->minor_version ||
6601 /* mddev->patch_version != info->patch_version || */
6602 mddev->ctime != info->ctime ||
6603 mddev->level != info->level ||
6604 /* mddev->layout != info->layout || */
6605 mddev->persistent != !info->not_persistent ||
6606 mddev->chunk_sectors != info->chunk_size >> 9 ||
6607 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6608 ((state^info->state) & 0xfffffe00)
6611 /* Check there is only one change */
6612 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6614 if (mddev->raid_disks != info->raid_disks)
6616 if (mddev->layout != info->layout)
6618 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6625 if (mddev->layout != info->layout) {
6627 * we don't need to do anything at the md level, the
6628 * personality will take care of it all.
6630 if (mddev->pers->check_reshape == NULL)
6633 mddev->new_layout = info->layout;
6634 rv = mddev->pers->check_reshape(mddev);
6636 mddev->new_layout = mddev->layout;
6640 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6641 rv = update_size(mddev, (sector_t)info->size * 2);
6643 if (mddev->raid_disks != info->raid_disks)
6644 rv = update_raid_disks(mddev, info->raid_disks);
6646 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6647 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6651 if (mddev->recovery || mddev->sync_thread) {
6655 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6656 struct bitmap *bitmap;
6657 /* add the bitmap */
6658 if (mddev->bitmap) {
6662 if (mddev->bitmap_info.default_offset == 0) {
6666 mddev->bitmap_info.offset =
6667 mddev->bitmap_info.default_offset;
6668 mddev->bitmap_info.space =
6669 mddev->bitmap_info.default_space;
6670 mddev->pers->quiesce(mddev, 1);
6671 bitmap = bitmap_create(mddev, -1);
6672 if (!IS_ERR(bitmap)) {
6673 mddev->bitmap = bitmap;
6674 rv = bitmap_load(mddev);
6676 rv = PTR_ERR(bitmap);
6678 bitmap_destroy(mddev);
6679 mddev->pers->quiesce(mddev, 0);
6681 /* remove the bitmap */
6682 if (!mddev->bitmap) {
6686 if (mddev->bitmap->storage.file) {
6690 if (mddev->bitmap_info.nodes) {
6691 /* hold PW on all the bitmap lock */
6692 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6693 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6695 md_cluster_ops->unlock_all_bitmaps(mddev);
6699 mddev->bitmap_info.nodes = 0;
6700 md_cluster_ops->leave(mddev);
6702 mddev->pers->quiesce(mddev, 1);
6703 bitmap_destroy(mddev);
6704 mddev->pers->quiesce(mddev, 0);
6705 mddev->bitmap_info.offset = 0;
6708 md_update_sb(mddev, 1);
6714 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6716 struct md_rdev *rdev;
6719 if (mddev->pers == NULL)
6723 rdev = find_rdev_rcu(mddev, dev);
6727 md_error(mddev, rdev);
6728 if (!test_bit(Faulty, &rdev->flags))
6736 * We have a problem here : there is no easy way to give a CHS
6737 * virtual geometry. We currently pretend that we have a 2 heads
6738 * 4 sectors (with a BIG number of cylinders...). This drives
6739 * dosfs just mad... ;-)
6741 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6743 struct mddev *mddev = bdev->bd_disk->private_data;
6747 geo->cylinders = mddev->array_sectors / 8;
6751 static inline bool md_ioctl_valid(unsigned int cmd)
6756 case GET_ARRAY_INFO:
6757 case GET_BITMAP_FILE:
6760 case HOT_REMOVE_DISK:
6763 case RESTART_ARRAY_RW:
6765 case SET_ARRAY_INFO:
6766 case SET_BITMAP_FILE:
6767 case SET_DISK_FAULTY:
6770 case CLUSTERED_DISK_NACK:
6777 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6778 unsigned int cmd, unsigned long arg)
6781 void __user *argp = (void __user *)arg;
6782 struct mddev *mddev = NULL;
6785 if (!md_ioctl_valid(cmd))
6790 case GET_ARRAY_INFO:
6794 if (!capable(CAP_SYS_ADMIN))
6799 * Commands dealing with the RAID driver but not any
6804 err = get_version(argp);
6810 autostart_arrays(arg);
6817 * Commands creating/starting a new array:
6820 mddev = bdev->bd_disk->private_data;
6827 /* Some actions do not requires the mutex */
6829 case GET_ARRAY_INFO:
6830 if (!mddev->raid_disks && !mddev->external)
6833 err = get_array_info(mddev, argp);
6837 if (!mddev->raid_disks && !mddev->external)
6840 err = get_disk_info(mddev, argp);
6843 case SET_DISK_FAULTY:
6844 err = set_disk_faulty(mddev, new_decode_dev(arg));
6847 case GET_BITMAP_FILE:
6848 err = get_bitmap_file(mddev, argp);
6853 if (cmd == ADD_NEW_DISK)
6854 /* need to ensure md_delayed_delete() has completed */
6855 flush_workqueue(md_misc_wq);
6857 if (cmd == HOT_REMOVE_DISK)
6858 /* need to ensure recovery thread has run */
6859 wait_event_interruptible_timeout(mddev->sb_wait,
6860 !test_bit(MD_RECOVERY_NEEDED,
6862 msecs_to_jiffies(5000));
6863 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6864 /* Need to flush page cache, and ensure no-one else opens
6867 mutex_lock(&mddev->open_mutex);
6868 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6869 mutex_unlock(&mddev->open_mutex);
6873 set_bit(MD_CLOSING, &mddev->flags);
6874 mutex_unlock(&mddev->open_mutex);
6875 sync_blockdev(bdev);
6877 err = mddev_lock(mddev);
6879 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
6884 if (cmd == SET_ARRAY_INFO) {
6885 mdu_array_info_t info;
6887 memset(&info, 0, sizeof(info));
6888 else if (copy_from_user(&info, argp, sizeof(info))) {
6893 err = update_array_info(mddev, &info);
6895 pr_warn("md: couldn't update array info. %d\n", err);
6900 if (!list_empty(&mddev->disks)) {
6901 pr_warn("md: array %s already has disks!\n", mdname(mddev));
6905 if (mddev->raid_disks) {
6906 pr_warn("md: array %s already initialised!\n", mdname(mddev));
6910 err = set_array_info(mddev, &info);
6912 pr_warn("md: couldn't set array info. %d\n", err);
6919 * Commands querying/configuring an existing array:
6921 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6922 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6923 if ((!mddev->raid_disks && !mddev->external)
6924 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6925 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6926 && cmd != GET_BITMAP_FILE) {
6932 * Commands even a read-only array can execute:
6935 case RESTART_ARRAY_RW:
6936 err = restart_array(mddev);
6940 err = do_md_stop(mddev, 0, bdev);
6944 err = md_set_readonly(mddev, bdev);
6947 case HOT_REMOVE_DISK:
6948 err = hot_remove_disk(mddev, new_decode_dev(arg));
6952 /* We can support ADD_NEW_DISK on read-only arrays
6953 * only if we are re-adding a preexisting device.
6954 * So require mddev->pers and MD_DISK_SYNC.
6957 mdu_disk_info_t info;
6958 if (copy_from_user(&info, argp, sizeof(info)))
6960 else if (!(info.state & (1<<MD_DISK_SYNC)))
6961 /* Need to clear read-only for this */
6964 err = add_new_disk(mddev, &info);
6970 if (get_user(ro, (int __user *)(arg))) {
6976 /* if the bdev is going readonly the value of mddev->ro
6977 * does not matter, no writes are coming
6982 /* are we are already prepared for writes? */
6986 /* transitioning to readauto need only happen for
6987 * arrays that call md_write_start
6990 err = restart_array(mddev);
6993 set_disk_ro(mddev->gendisk, 0);
7000 * The remaining ioctls are changing the state of the
7001 * superblock, so we do not allow them on read-only arrays.
7003 if (mddev->ro && mddev->pers) {
7004 if (mddev->ro == 2) {
7006 sysfs_notify_dirent_safe(mddev->sysfs_state);
7007 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7008 /* mddev_unlock will wake thread */
7009 /* If a device failed while we were read-only, we
7010 * need to make sure the metadata is updated now.
7012 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7013 mddev_unlock(mddev);
7014 wait_event(mddev->sb_wait,
7015 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7016 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7017 mddev_lock_nointr(mddev);
7028 mdu_disk_info_t info;
7029 if (copy_from_user(&info, argp, sizeof(info)))
7032 err = add_new_disk(mddev, &info);
7036 case CLUSTERED_DISK_NACK:
7037 if (mddev_is_clustered(mddev))
7038 md_cluster_ops->new_disk_ack(mddev, false);
7044 err = hot_add_disk(mddev, new_decode_dev(arg));
7048 err = do_md_run(mddev);
7051 case SET_BITMAP_FILE:
7052 err = set_bitmap_file(mddev, (int)arg);
7061 if (mddev->hold_active == UNTIL_IOCTL &&
7063 mddev->hold_active = 0;
7064 mddev_unlock(mddev);
7068 #ifdef CONFIG_COMPAT
7069 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7070 unsigned int cmd, unsigned long arg)
7073 case HOT_REMOVE_DISK:
7075 case SET_DISK_FAULTY:
7076 case SET_BITMAP_FILE:
7077 /* These take in integer arg, do not convert */
7080 arg = (unsigned long)compat_ptr(arg);
7084 return md_ioctl(bdev, mode, cmd, arg);
7086 #endif /* CONFIG_COMPAT */
7088 static int md_open(struct block_device *bdev, fmode_t mode)
7091 * Succeed if we can lock the mddev, which confirms that
7092 * it isn't being stopped right now.
7094 struct mddev *mddev = mddev_find(bdev->bd_dev);
7100 if (mddev->gendisk != bdev->bd_disk) {
7101 /* we are racing with mddev_put which is discarding this
7105 /* Wait until bdev->bd_disk is definitely gone */
7106 flush_workqueue(md_misc_wq);
7107 /* Then retry the open from the top */
7108 return -ERESTARTSYS;
7110 BUG_ON(mddev != bdev->bd_disk->private_data);
7112 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7115 if (test_bit(MD_CLOSING, &mddev->flags)) {
7116 mutex_unlock(&mddev->open_mutex);
7122 atomic_inc(&mddev->openers);
7123 mutex_unlock(&mddev->open_mutex);
7125 check_disk_change(bdev);
7132 static void md_release(struct gendisk *disk, fmode_t mode)
7134 struct mddev *mddev = disk->private_data;
7137 atomic_dec(&mddev->openers);
7141 static int md_media_changed(struct gendisk *disk)
7143 struct mddev *mddev = disk->private_data;
7145 return mddev->changed;
7148 static int md_revalidate(struct gendisk *disk)
7150 struct mddev *mddev = disk->private_data;
7155 static const struct block_device_operations md_fops =
7157 .owner = THIS_MODULE,
7159 .release = md_release,
7161 #ifdef CONFIG_COMPAT
7162 .compat_ioctl = md_compat_ioctl,
7164 .getgeo = md_getgeo,
7165 .media_changed = md_media_changed,
7166 .revalidate_disk= md_revalidate,
7169 static int md_thread(void *arg)
7171 struct md_thread *thread = arg;
7174 * md_thread is a 'system-thread', it's priority should be very
7175 * high. We avoid resource deadlocks individually in each
7176 * raid personality. (RAID5 does preallocation) We also use RR and
7177 * the very same RT priority as kswapd, thus we will never get
7178 * into a priority inversion deadlock.
7180 * we definitely have to have equal or higher priority than
7181 * bdflush, otherwise bdflush will deadlock if there are too
7182 * many dirty RAID5 blocks.
7185 allow_signal(SIGKILL);
7186 while (!kthread_should_stop()) {
7188 /* We need to wait INTERRUPTIBLE so that
7189 * we don't add to the load-average.
7190 * That means we need to be sure no signals are
7193 if (signal_pending(current))
7194 flush_signals(current);
7196 wait_event_interruptible_timeout
7198 test_bit(THREAD_WAKEUP, &thread->flags)
7199 || kthread_should_stop() || kthread_should_park(),
7202 clear_bit(THREAD_WAKEUP, &thread->flags);
7203 if (kthread_should_park())
7205 if (!kthread_should_stop())
7206 thread->run(thread);
7212 void md_wakeup_thread(struct md_thread *thread)
7215 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7216 set_bit(THREAD_WAKEUP, &thread->flags);
7217 wake_up(&thread->wqueue);
7220 EXPORT_SYMBOL(md_wakeup_thread);
7222 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7223 struct mddev *mddev, const char *name)
7225 struct md_thread *thread;
7227 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7231 init_waitqueue_head(&thread->wqueue);
7234 thread->mddev = mddev;
7235 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7236 thread->tsk = kthread_run(md_thread, thread,
7238 mdname(thread->mddev),
7240 if (IS_ERR(thread->tsk)) {
7246 EXPORT_SYMBOL(md_register_thread);
7248 void md_unregister_thread(struct md_thread **threadp)
7250 struct md_thread *thread = *threadp;
7253 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7254 /* Locking ensures that mddev_unlock does not wake_up a
7255 * non-existent thread
7257 spin_lock(&pers_lock);
7259 spin_unlock(&pers_lock);
7261 kthread_stop(thread->tsk);
7264 EXPORT_SYMBOL(md_unregister_thread);
7266 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7268 if (!rdev || test_bit(Faulty, &rdev->flags))
7271 if (!mddev->pers || !mddev->pers->error_handler)
7273 mddev->pers->error_handler(mddev,rdev);
7274 if (mddev->degraded)
7275 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7276 sysfs_notify_dirent_safe(rdev->sysfs_state);
7277 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7278 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7279 md_wakeup_thread(mddev->thread);
7280 if (mddev->event_work.func)
7281 queue_work(md_misc_wq, &mddev->event_work);
7282 md_new_event(mddev);
7284 EXPORT_SYMBOL(md_error);
7286 /* seq_file implementation /proc/mdstat */
7288 static void status_unused(struct seq_file *seq)
7291 struct md_rdev *rdev;
7293 seq_printf(seq, "unused devices: ");
7295 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7296 char b[BDEVNAME_SIZE];
7298 seq_printf(seq, "%s ",
7299 bdevname(rdev->bdev,b));
7302 seq_printf(seq, "<none>");
7304 seq_printf(seq, "\n");
7307 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7309 sector_t max_sectors, resync, res;
7310 unsigned long dt, db;
7313 unsigned int per_milli;
7315 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7316 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7317 max_sectors = mddev->resync_max_sectors;
7319 max_sectors = mddev->dev_sectors;
7321 resync = mddev->curr_resync;
7323 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7324 /* Still cleaning up */
7325 resync = max_sectors;
7327 resync -= atomic_read(&mddev->recovery_active);
7330 if (mddev->recovery_cp < MaxSector) {
7331 seq_printf(seq, "\tresync=PENDING");
7337 seq_printf(seq, "\tresync=DELAYED");
7341 WARN_ON(max_sectors == 0);
7342 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7343 * in a sector_t, and (max_sectors>>scale) will fit in a
7344 * u32, as those are the requirements for sector_div.
7345 * Thus 'scale' must be at least 10
7348 if (sizeof(sector_t) > sizeof(unsigned long)) {
7349 while ( max_sectors/2 > (1ULL<<(scale+32)))
7352 res = (resync>>scale)*1000;
7353 sector_div(res, (u32)((max_sectors>>scale)+1));
7357 int i, x = per_milli/50, y = 20-x;
7358 seq_printf(seq, "[");
7359 for (i = 0; i < x; i++)
7360 seq_printf(seq, "=");
7361 seq_printf(seq, ">");
7362 for (i = 0; i < y; i++)
7363 seq_printf(seq, ".");
7364 seq_printf(seq, "] ");
7366 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7367 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7369 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7371 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7372 "resync" : "recovery"))),
7373 per_milli/10, per_milli % 10,
7374 (unsigned long long) resync/2,
7375 (unsigned long long) max_sectors/2);
7378 * dt: time from mark until now
7379 * db: blocks written from mark until now
7380 * rt: remaining time
7382 * rt is a sector_t, so could be 32bit or 64bit.
7383 * So we divide before multiply in case it is 32bit and close
7385 * We scale the divisor (db) by 32 to avoid losing precision
7386 * near the end of resync when the number of remaining sectors
7388 * We then divide rt by 32 after multiplying by db to compensate.
7389 * The '+1' avoids division by zero if db is very small.
7391 dt = ((jiffies - mddev->resync_mark) / HZ);
7393 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7394 - mddev->resync_mark_cnt;
7396 rt = max_sectors - resync; /* number of remaining sectors */
7397 sector_div(rt, db/32+1);
7401 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7402 ((unsigned long)rt % 60)/6);
7404 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7408 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7410 struct list_head *tmp;
7412 struct mddev *mddev;
7420 spin_lock(&all_mddevs_lock);
7421 list_for_each(tmp,&all_mddevs)
7423 mddev = list_entry(tmp, struct mddev, all_mddevs);
7425 spin_unlock(&all_mddevs_lock);
7428 spin_unlock(&all_mddevs_lock);
7430 return (void*)2;/* tail */
7434 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7436 struct list_head *tmp;
7437 struct mddev *next_mddev, *mddev = v;
7443 spin_lock(&all_mddevs_lock);
7445 tmp = all_mddevs.next;
7447 tmp = mddev->all_mddevs.next;
7448 if (tmp != &all_mddevs)
7449 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7451 next_mddev = (void*)2;
7454 spin_unlock(&all_mddevs_lock);
7462 static void md_seq_stop(struct seq_file *seq, void *v)
7464 struct mddev *mddev = v;
7466 if (mddev && v != (void*)1 && v != (void*)2)
7470 static int md_seq_show(struct seq_file *seq, void *v)
7472 struct mddev *mddev = v;
7474 struct md_rdev *rdev;
7476 if (v == (void*)1) {
7477 struct md_personality *pers;
7478 seq_printf(seq, "Personalities : ");
7479 spin_lock(&pers_lock);
7480 list_for_each_entry(pers, &pers_list, list)
7481 seq_printf(seq, "[%s] ", pers->name);
7483 spin_unlock(&pers_lock);
7484 seq_printf(seq, "\n");
7485 seq->poll_event = atomic_read(&md_event_count);
7488 if (v == (void*)2) {
7493 spin_lock(&mddev->lock);
7494 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7495 seq_printf(seq, "%s : %sactive", mdname(mddev),
7496 mddev->pers ? "" : "in");
7499 seq_printf(seq, " (read-only)");
7501 seq_printf(seq, " (auto-read-only)");
7502 seq_printf(seq, " %s", mddev->pers->name);
7507 rdev_for_each_rcu(rdev, mddev) {
7508 char b[BDEVNAME_SIZE];
7509 seq_printf(seq, " %s[%d]",
7510 bdevname(rdev->bdev,b), rdev->desc_nr);
7511 if (test_bit(WriteMostly, &rdev->flags))
7512 seq_printf(seq, "(W)");
7513 if (test_bit(Journal, &rdev->flags))
7514 seq_printf(seq, "(J)");
7515 if (test_bit(Faulty, &rdev->flags)) {
7516 seq_printf(seq, "(F)");
7519 if (rdev->raid_disk < 0)
7520 seq_printf(seq, "(S)"); /* spare */
7521 if (test_bit(Replacement, &rdev->flags))
7522 seq_printf(seq, "(R)");
7523 sectors += rdev->sectors;
7527 if (!list_empty(&mddev->disks)) {
7529 seq_printf(seq, "\n %llu blocks",
7530 (unsigned long long)
7531 mddev->array_sectors / 2);
7533 seq_printf(seq, "\n %llu blocks",
7534 (unsigned long long)sectors / 2);
7536 if (mddev->persistent) {
7537 if (mddev->major_version != 0 ||
7538 mddev->minor_version != 90) {
7539 seq_printf(seq," super %d.%d",
7540 mddev->major_version,
7541 mddev->minor_version);
7543 } else if (mddev->external)
7544 seq_printf(seq, " super external:%s",
7545 mddev->metadata_type);
7547 seq_printf(seq, " super non-persistent");
7550 mddev->pers->status(seq, mddev);
7551 seq_printf(seq, "\n ");
7552 if (mddev->pers->sync_request) {
7553 if (status_resync(seq, mddev))
7554 seq_printf(seq, "\n ");
7557 seq_printf(seq, "\n ");
7559 bitmap_status(seq, mddev->bitmap);
7561 seq_printf(seq, "\n");
7563 spin_unlock(&mddev->lock);
7568 static const struct seq_operations md_seq_ops = {
7569 .start = md_seq_start,
7570 .next = md_seq_next,
7571 .stop = md_seq_stop,
7572 .show = md_seq_show,
7575 static int md_seq_open(struct inode *inode, struct file *file)
7577 struct seq_file *seq;
7580 error = seq_open(file, &md_seq_ops);
7584 seq = file->private_data;
7585 seq->poll_event = atomic_read(&md_event_count);
7589 static int md_unloading;
7590 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7592 struct seq_file *seq = filp->private_data;
7596 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7597 poll_wait(filp, &md_event_waiters, wait);
7599 /* always allow read */
7600 mask = POLLIN | POLLRDNORM;
7602 if (seq->poll_event != atomic_read(&md_event_count))
7603 mask |= POLLERR | POLLPRI;
7607 static const struct file_operations md_seq_fops = {
7608 .owner = THIS_MODULE,
7609 .open = md_seq_open,
7611 .llseek = seq_lseek,
7612 .release = seq_release_private,
7613 .poll = mdstat_poll,
7616 int register_md_personality(struct md_personality *p)
7618 pr_debug("md: %s personality registered for level %d\n",
7620 spin_lock(&pers_lock);
7621 list_add_tail(&p->list, &pers_list);
7622 spin_unlock(&pers_lock);
7625 EXPORT_SYMBOL(register_md_personality);
7627 int unregister_md_personality(struct md_personality *p)
7629 pr_debug("md: %s personality unregistered\n", p->name);
7630 spin_lock(&pers_lock);
7631 list_del_init(&p->list);
7632 spin_unlock(&pers_lock);
7635 EXPORT_SYMBOL(unregister_md_personality);
7637 int register_md_cluster_operations(struct md_cluster_operations *ops,
7638 struct module *module)
7641 spin_lock(&pers_lock);
7642 if (md_cluster_ops != NULL)
7645 md_cluster_ops = ops;
7646 md_cluster_mod = module;
7648 spin_unlock(&pers_lock);
7651 EXPORT_SYMBOL(register_md_cluster_operations);
7653 int unregister_md_cluster_operations(void)
7655 spin_lock(&pers_lock);
7656 md_cluster_ops = NULL;
7657 spin_unlock(&pers_lock);
7660 EXPORT_SYMBOL(unregister_md_cluster_operations);
7662 int md_setup_cluster(struct mddev *mddev, int nodes)
7664 if (!md_cluster_ops)
7665 request_module("md-cluster");
7666 spin_lock(&pers_lock);
7667 /* ensure module won't be unloaded */
7668 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7669 pr_warn("can't find md-cluster module or get it's reference.\n");
7670 spin_unlock(&pers_lock);
7673 spin_unlock(&pers_lock);
7675 return md_cluster_ops->join(mddev, nodes);
7678 void md_cluster_stop(struct mddev *mddev)
7680 if (!md_cluster_ops)
7682 md_cluster_ops->leave(mddev);
7683 module_put(md_cluster_mod);
7686 static int is_mddev_idle(struct mddev *mddev, int init)
7688 struct md_rdev *rdev;
7694 rdev_for_each_rcu(rdev, mddev) {
7695 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7696 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7697 (int)part_stat_read(&disk->part0, sectors[1]) -
7698 atomic_read(&disk->sync_io);
7699 /* sync IO will cause sync_io to increase before the disk_stats
7700 * as sync_io is counted when a request starts, and
7701 * disk_stats is counted when it completes.
7702 * So resync activity will cause curr_events to be smaller than
7703 * when there was no such activity.
7704 * non-sync IO will cause disk_stat to increase without
7705 * increasing sync_io so curr_events will (eventually)
7706 * be larger than it was before. Once it becomes
7707 * substantially larger, the test below will cause
7708 * the array to appear non-idle, and resync will slow
7710 * If there is a lot of outstanding resync activity when
7711 * we set last_event to curr_events, then all that activity
7712 * completing might cause the array to appear non-idle
7713 * and resync will be slowed down even though there might
7714 * not have been non-resync activity. This will only
7715 * happen once though. 'last_events' will soon reflect
7716 * the state where there is little or no outstanding
7717 * resync requests, and further resync activity will
7718 * always make curr_events less than last_events.
7721 if (init || curr_events - rdev->last_events > 64) {
7722 rdev->last_events = curr_events;
7730 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7732 /* another "blocks" (512byte) blocks have been synced */
7733 atomic_sub(blocks, &mddev->recovery_active);
7734 wake_up(&mddev->recovery_wait);
7736 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7737 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7738 md_wakeup_thread(mddev->thread);
7739 // stop recovery, signal do_sync ....
7742 EXPORT_SYMBOL(md_done_sync);
7744 /* md_write_start(mddev, bi)
7745 * If we need to update some array metadata (e.g. 'active' flag
7746 * in superblock) before writing, schedule a superblock update
7747 * and wait for it to complete.
7749 void md_write_start(struct mddev *mddev, struct bio *bi)
7752 if (bio_data_dir(bi) != WRITE)
7755 BUG_ON(mddev->ro == 1);
7756 if (mddev->ro == 2) {
7757 /* need to switch to read/write */
7759 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7760 md_wakeup_thread(mddev->thread);
7761 md_wakeup_thread(mddev->sync_thread);
7764 atomic_inc(&mddev->writes_pending);
7765 if (mddev->safemode == 1)
7766 mddev->safemode = 0;
7767 if (mddev->in_sync) {
7768 spin_lock(&mddev->lock);
7769 if (mddev->in_sync) {
7771 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7772 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7773 md_wakeup_thread(mddev->thread);
7776 spin_unlock(&mddev->lock);
7779 sysfs_notify_dirent_safe(mddev->sysfs_state);
7780 wait_event(mddev->sb_wait,
7781 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7783 EXPORT_SYMBOL(md_write_start);
7785 void md_write_end(struct mddev *mddev)
7787 if (atomic_dec_and_test(&mddev->writes_pending)) {
7788 if (mddev->safemode == 2)
7789 md_wakeup_thread(mddev->thread);
7790 else if (mddev->safemode_delay)
7791 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7794 EXPORT_SYMBOL(md_write_end);
7796 /* md_allow_write(mddev)
7797 * Calling this ensures that the array is marked 'active' so that writes
7798 * may proceed without blocking. It is important to call this before
7799 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7800 * Must be called with mddev_lock held.
7802 * In the ->external case MD_SB_CHANGE_PENDING can not be cleared until mddev->lock
7803 * is dropped, so return -EAGAIN after notifying userspace.
7805 int md_allow_write(struct mddev *mddev)
7811 if (!mddev->pers->sync_request)
7814 spin_lock(&mddev->lock);
7815 if (mddev->in_sync) {
7817 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7818 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7819 if (mddev->safemode_delay &&
7820 mddev->safemode == 0)
7821 mddev->safemode = 1;
7822 spin_unlock(&mddev->lock);
7823 md_update_sb(mddev, 0);
7824 sysfs_notify_dirent_safe(mddev->sysfs_state);
7826 spin_unlock(&mddev->lock);
7828 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
7833 EXPORT_SYMBOL_GPL(md_allow_write);
7835 #define SYNC_MARKS 10
7836 #define SYNC_MARK_STEP (3*HZ)
7837 #define UPDATE_FREQUENCY (5*60*HZ)
7838 void md_do_sync(struct md_thread *thread)
7840 struct mddev *mddev = thread->mddev;
7841 struct mddev *mddev2;
7842 unsigned int currspeed = 0,
7844 sector_t max_sectors,j, io_sectors, recovery_done;
7845 unsigned long mark[SYNC_MARKS];
7846 unsigned long update_time;
7847 sector_t mark_cnt[SYNC_MARKS];
7849 struct list_head *tmp;
7850 sector_t last_check;
7852 struct md_rdev *rdev;
7853 char *desc, *action = NULL;
7854 struct blk_plug plug;
7857 /* just incase thread restarts... */
7858 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7860 if (mddev->ro) {/* never try to sync a read-only array */
7861 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7865 if (mddev_is_clustered(mddev)) {
7866 ret = md_cluster_ops->resync_start(mddev);
7870 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7871 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7872 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7873 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7874 && ((unsigned long long)mddev->curr_resync_completed
7875 < (unsigned long long)mddev->resync_max_sectors))
7879 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7880 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7881 desc = "data-check";
7883 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7884 desc = "requested-resync";
7888 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7893 mddev->last_sync_action = action ?: desc;
7895 /* we overload curr_resync somewhat here.
7896 * 0 == not engaged in resync at all
7897 * 2 == checking that there is no conflict with another sync
7898 * 1 == like 2, but have yielded to allow conflicting resync to
7900 * other == active in resync - this many blocks
7902 * Before starting a resync we must have set curr_resync to
7903 * 2, and then checked that every "conflicting" array has curr_resync
7904 * less than ours. When we find one that is the same or higher
7905 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7906 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7907 * This will mean we have to start checking from the beginning again.
7912 int mddev2_minor = -1;
7913 mddev->curr_resync = 2;
7916 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7918 for_each_mddev(mddev2, tmp) {
7919 if (mddev2 == mddev)
7921 if (!mddev->parallel_resync
7922 && mddev2->curr_resync
7923 && match_mddev_units(mddev, mddev2)) {
7925 if (mddev < mddev2 && mddev->curr_resync == 2) {
7926 /* arbitrarily yield */
7927 mddev->curr_resync = 1;
7928 wake_up(&resync_wait);
7930 if (mddev > mddev2 && mddev->curr_resync == 1)
7931 /* no need to wait here, we can wait the next
7932 * time 'round when curr_resync == 2
7935 /* We need to wait 'interruptible' so as not to
7936 * contribute to the load average, and not to
7937 * be caught by 'softlockup'
7939 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7940 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7941 mddev2->curr_resync >= mddev->curr_resync) {
7942 if (mddev2_minor != mddev2->md_minor) {
7943 mddev2_minor = mddev2->md_minor;
7944 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
7945 desc, mdname(mddev),
7949 if (signal_pending(current))
7950 flush_signals(current);
7952 finish_wait(&resync_wait, &wq);
7955 finish_wait(&resync_wait, &wq);
7958 } while (mddev->curr_resync < 2);
7961 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7962 /* resync follows the size requested by the personality,
7963 * which defaults to physical size, but can be virtual size
7965 max_sectors = mddev->resync_max_sectors;
7966 atomic64_set(&mddev->resync_mismatches, 0);
7967 /* we don't use the checkpoint if there's a bitmap */
7968 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7969 j = mddev->resync_min;
7970 else if (!mddev->bitmap)
7971 j = mddev->recovery_cp;
7973 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7974 max_sectors = mddev->resync_max_sectors;
7976 /* recovery follows the physical size of devices */
7977 max_sectors = mddev->dev_sectors;
7980 rdev_for_each_rcu(rdev, mddev)
7981 if (rdev->raid_disk >= 0 &&
7982 !test_bit(Journal, &rdev->flags) &&
7983 !test_bit(Faulty, &rdev->flags) &&
7984 !test_bit(In_sync, &rdev->flags) &&
7985 rdev->recovery_offset < j)
7986 j = rdev->recovery_offset;
7989 /* If there is a bitmap, we need to make sure all
7990 * writes that started before we added a spare
7991 * complete before we start doing a recovery.
7992 * Otherwise the write might complete and (via
7993 * bitmap_endwrite) set a bit in the bitmap after the
7994 * recovery has checked that bit and skipped that
7997 if (mddev->bitmap) {
7998 mddev->pers->quiesce(mddev, 1);
7999 mddev->pers->quiesce(mddev, 0);
8003 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8004 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8005 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8006 speed_max(mddev), desc);
8008 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8011 for (m = 0; m < SYNC_MARKS; m++) {
8013 mark_cnt[m] = io_sectors;
8016 mddev->resync_mark = mark[last_mark];
8017 mddev->resync_mark_cnt = mark_cnt[last_mark];
8020 * Tune reconstruction:
8022 window = 32*(PAGE_SIZE/512);
8023 pr_debug("md: using %dk window, over a total of %lluk.\n",
8024 window/2, (unsigned long long)max_sectors/2);
8026 atomic_set(&mddev->recovery_active, 0);
8030 pr_debug("md: resuming %s of %s from checkpoint.\n",
8031 desc, mdname(mddev));
8032 mddev->curr_resync = j;
8034 mddev->curr_resync = 3; /* no longer delayed */
8035 mddev->curr_resync_completed = j;
8036 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8037 md_new_event(mddev);
8038 update_time = jiffies;
8040 blk_start_plug(&plug);
8041 while (j < max_sectors) {
8046 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8047 ((mddev->curr_resync > mddev->curr_resync_completed &&
8048 (mddev->curr_resync - mddev->curr_resync_completed)
8049 > (max_sectors >> 4)) ||
8050 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8051 (j - mddev->curr_resync_completed)*2
8052 >= mddev->resync_max - mddev->curr_resync_completed ||
8053 mddev->curr_resync_completed > mddev->resync_max
8055 /* time to update curr_resync_completed */
8056 wait_event(mddev->recovery_wait,
8057 atomic_read(&mddev->recovery_active) == 0);
8058 mddev->curr_resync_completed = j;
8059 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8060 j > mddev->recovery_cp)
8061 mddev->recovery_cp = j;
8062 update_time = jiffies;
8063 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8064 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8067 while (j >= mddev->resync_max &&
8068 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8069 /* As this condition is controlled by user-space,
8070 * we can block indefinitely, so use '_interruptible'
8071 * to avoid triggering warnings.
8073 flush_signals(current); /* just in case */
8074 wait_event_interruptible(mddev->recovery_wait,
8075 mddev->resync_max > j
8076 || test_bit(MD_RECOVERY_INTR,
8080 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8083 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8085 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8089 if (!skipped) { /* actual IO requested */
8090 io_sectors += sectors;
8091 atomic_add(sectors, &mddev->recovery_active);
8094 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8098 if (j > max_sectors)
8099 /* when skipping, extra large numbers can be returned. */
8102 mddev->curr_resync = j;
8103 mddev->curr_mark_cnt = io_sectors;
8104 if (last_check == 0)
8105 /* this is the earliest that rebuild will be
8106 * visible in /proc/mdstat
8108 md_new_event(mddev);
8110 if (last_check + window > io_sectors || j == max_sectors)
8113 last_check = io_sectors;
8115 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8117 int next = (last_mark+1) % SYNC_MARKS;
8119 mddev->resync_mark = mark[next];
8120 mddev->resync_mark_cnt = mark_cnt[next];
8121 mark[next] = jiffies;
8122 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8126 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8130 * this loop exits only if either when we are slower than
8131 * the 'hard' speed limit, or the system was IO-idle for
8133 * the system might be non-idle CPU-wise, but we only care
8134 * about not overloading the IO subsystem. (things like an
8135 * e2fsck being done on the RAID array should execute fast)
8139 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8140 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8141 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8143 if (currspeed > speed_min(mddev)) {
8144 if (currspeed > speed_max(mddev)) {
8148 if (!is_mddev_idle(mddev, 0)) {
8150 * Give other IO more of a chance.
8151 * The faster the devices, the less we wait.
8153 wait_event(mddev->recovery_wait,
8154 !atomic_read(&mddev->recovery_active));
8158 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8159 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8160 ? "interrupted" : "done");
8162 * this also signals 'finished resyncing' to md_stop
8164 blk_finish_plug(&plug);
8165 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8167 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8168 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8169 mddev->curr_resync > 3) {
8170 mddev->curr_resync_completed = mddev->curr_resync;
8171 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8173 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8175 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8176 mddev->curr_resync > 3) {
8177 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8178 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8179 if (mddev->curr_resync >= mddev->recovery_cp) {
8180 pr_debug("md: checkpointing %s of %s.\n",
8181 desc, mdname(mddev));
8182 if (test_bit(MD_RECOVERY_ERROR,
8184 mddev->recovery_cp =
8185 mddev->curr_resync_completed;
8187 mddev->recovery_cp =
8191 mddev->recovery_cp = MaxSector;
8193 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8194 mddev->curr_resync = MaxSector;
8196 rdev_for_each_rcu(rdev, mddev)
8197 if (rdev->raid_disk >= 0 &&
8198 mddev->delta_disks >= 0 &&
8199 !test_bit(Journal, &rdev->flags) &&
8200 !test_bit(Faulty, &rdev->flags) &&
8201 !test_bit(In_sync, &rdev->flags) &&
8202 rdev->recovery_offset < mddev->curr_resync)
8203 rdev->recovery_offset = mddev->curr_resync;
8208 /* set CHANGE_PENDING here since maybe another update is needed,
8209 * so other nodes are informed. It should be harmless for normal
8211 set_mask_bits(&mddev->sb_flags, 0,
8212 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8214 spin_lock(&mddev->lock);
8215 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8216 /* We completed so min/max setting can be forgotten if used. */
8217 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8218 mddev->resync_min = 0;
8219 mddev->resync_max = MaxSector;
8220 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8221 mddev->resync_min = mddev->curr_resync_completed;
8222 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8223 mddev->curr_resync = 0;
8224 spin_unlock(&mddev->lock);
8226 wake_up(&resync_wait);
8227 md_wakeup_thread(mddev->thread);
8230 EXPORT_SYMBOL_GPL(md_do_sync);
8232 static int remove_and_add_spares(struct mddev *mddev,
8233 struct md_rdev *this)
8235 struct md_rdev *rdev;
8238 bool remove_some = false;
8240 rdev_for_each(rdev, mddev) {
8241 if ((this == NULL || rdev == this) &&
8242 rdev->raid_disk >= 0 &&
8243 !test_bit(Blocked, &rdev->flags) &&
8244 test_bit(Faulty, &rdev->flags) &&
8245 atomic_read(&rdev->nr_pending)==0) {
8246 /* Faulty non-Blocked devices with nr_pending == 0
8247 * never get nr_pending incremented,
8248 * never get Faulty cleared, and never get Blocked set.
8249 * So we can synchronize_rcu now rather than once per device
8252 set_bit(RemoveSynchronized, &rdev->flags);
8258 rdev_for_each(rdev, mddev) {
8259 if ((this == NULL || rdev == this) &&
8260 rdev->raid_disk >= 0 &&
8261 !test_bit(Blocked, &rdev->flags) &&
8262 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8263 (!test_bit(In_sync, &rdev->flags) &&
8264 !test_bit(Journal, &rdev->flags))) &&
8265 atomic_read(&rdev->nr_pending)==0)) {
8266 if (mddev->pers->hot_remove_disk(
8267 mddev, rdev) == 0) {
8268 sysfs_unlink_rdev(mddev, rdev);
8269 rdev->raid_disk = -1;
8273 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8274 clear_bit(RemoveSynchronized, &rdev->flags);
8277 if (removed && mddev->kobj.sd)
8278 sysfs_notify(&mddev->kobj, NULL, "degraded");
8280 if (this && removed)
8283 rdev_for_each(rdev, mddev) {
8284 if (this && this != rdev)
8286 if (test_bit(Candidate, &rdev->flags))
8288 if (rdev->raid_disk >= 0 &&
8289 !test_bit(In_sync, &rdev->flags) &&
8290 !test_bit(Journal, &rdev->flags) &&
8291 !test_bit(Faulty, &rdev->flags))
8293 if (rdev->raid_disk >= 0)
8295 if (test_bit(Faulty, &rdev->flags))
8297 if (!test_bit(Journal, &rdev->flags)) {
8299 ! (rdev->saved_raid_disk >= 0 &&
8300 !test_bit(Bitmap_sync, &rdev->flags)))
8303 rdev->recovery_offset = 0;
8306 hot_add_disk(mddev, rdev) == 0) {
8307 if (sysfs_link_rdev(mddev, rdev))
8308 /* failure here is OK */;
8309 if (!test_bit(Journal, &rdev->flags))
8311 md_new_event(mddev);
8312 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8317 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8321 static void md_start_sync(struct work_struct *ws)
8323 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8325 mddev->sync_thread = md_register_thread(md_do_sync,
8328 if (!mddev->sync_thread) {
8329 pr_warn("%s: could not start resync thread...\n",
8331 /* leave the spares where they are, it shouldn't hurt */
8332 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8333 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8334 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8335 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8336 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8337 wake_up(&resync_wait);
8338 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8340 if (mddev->sysfs_action)
8341 sysfs_notify_dirent_safe(mddev->sysfs_action);
8343 md_wakeup_thread(mddev->sync_thread);
8344 sysfs_notify_dirent_safe(mddev->sysfs_action);
8345 md_new_event(mddev);
8349 * This routine is regularly called by all per-raid-array threads to
8350 * deal with generic issues like resync and super-block update.
8351 * Raid personalities that don't have a thread (linear/raid0) do not
8352 * need this as they never do any recovery or update the superblock.
8354 * It does not do any resync itself, but rather "forks" off other threads
8355 * to do that as needed.
8356 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8357 * "->recovery" and create a thread at ->sync_thread.
8358 * When the thread finishes it sets MD_RECOVERY_DONE
8359 * and wakeups up this thread which will reap the thread and finish up.
8360 * This thread also removes any faulty devices (with nr_pending == 0).
8362 * The overall approach is:
8363 * 1/ if the superblock needs updating, update it.
8364 * 2/ If a recovery thread is running, don't do anything else.
8365 * 3/ If recovery has finished, clean up, possibly marking spares active.
8366 * 4/ If there are any faulty devices, remove them.
8367 * 5/ If array is degraded, try to add spares devices
8368 * 6/ If array has spares or is not in-sync, start a resync thread.
8370 void md_check_recovery(struct mddev *mddev)
8372 if (mddev->suspended)
8376 bitmap_daemon_work(mddev);
8378 if (signal_pending(current)) {
8379 if (mddev->pers->sync_request && !mddev->external) {
8380 pr_debug("md: %s in immediate safe mode\n",
8382 mddev->safemode = 2;
8384 flush_signals(current);
8387 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8390 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8391 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8392 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8393 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8394 (mddev->external == 0 && mddev->safemode == 1) ||
8395 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8396 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8400 if (mddev_trylock(mddev)) {
8404 struct md_rdev *rdev;
8405 if (!mddev->external && mddev->in_sync)
8406 /* 'Blocked' flag not needed as failed devices
8407 * will be recorded if array switched to read/write.
8408 * Leaving it set will prevent the device
8409 * from being removed.
8411 rdev_for_each(rdev, mddev)
8412 clear_bit(Blocked, &rdev->flags);
8413 /* On a read-only array we can:
8414 * - remove failed devices
8415 * - add already-in_sync devices if the array itself
8417 * As we only add devices that are already in-sync,
8418 * we can activate the spares immediately.
8420 remove_and_add_spares(mddev, NULL);
8421 /* There is no thread, but we need to call
8422 * ->spare_active and clear saved_raid_disk
8424 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8425 md_reap_sync_thread(mddev);
8426 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8427 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8428 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8432 if (mddev_is_clustered(mddev)) {
8433 struct md_rdev *rdev;
8434 /* kick the device if another node issued a
8437 rdev_for_each(rdev, mddev) {
8438 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8439 rdev->raid_disk < 0)
8440 md_kick_rdev_from_array(rdev);
8443 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8444 md_reload_sb(mddev, mddev->good_device_nr);
8447 if (!mddev->external) {
8449 spin_lock(&mddev->lock);
8450 if (mddev->safemode &&
8451 !atomic_read(&mddev->writes_pending) &&
8453 mddev->recovery_cp == MaxSector) {
8456 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8458 if (mddev->safemode == 1)
8459 mddev->safemode = 0;
8460 spin_unlock(&mddev->lock);
8462 sysfs_notify_dirent_safe(mddev->sysfs_state);
8465 if (mddev->sb_flags)
8466 md_update_sb(mddev, 0);
8468 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8469 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8470 /* resync/recovery still happening */
8471 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8474 if (mddev->sync_thread) {
8475 md_reap_sync_thread(mddev);
8478 /* Set RUNNING before clearing NEEDED to avoid
8479 * any transients in the value of "sync_action".
8481 mddev->curr_resync_completed = 0;
8482 spin_lock(&mddev->lock);
8483 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8484 spin_unlock(&mddev->lock);
8485 /* Clear some bits that don't mean anything, but
8488 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8489 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8491 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8492 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8494 /* no recovery is running.
8495 * remove any failed drives, then
8496 * add spares if possible.
8497 * Spares are also removed and re-added, to allow
8498 * the personality to fail the re-add.
8501 if (mddev->reshape_position != MaxSector) {
8502 if (mddev->pers->check_reshape == NULL ||
8503 mddev->pers->check_reshape(mddev) != 0)
8504 /* Cannot proceed */
8506 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8507 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8508 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8509 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8510 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8511 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8512 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8513 } else if (mddev->recovery_cp < MaxSector) {
8514 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8515 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8516 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8517 /* nothing to be done ... */
8520 if (mddev->pers->sync_request) {
8522 /* We are adding a device or devices to an array
8523 * which has the bitmap stored on all devices.
8524 * So make sure all bitmap pages get written
8526 bitmap_write_all(mddev->bitmap);
8528 INIT_WORK(&mddev->del_work, md_start_sync);
8529 queue_work(md_misc_wq, &mddev->del_work);
8533 if (!mddev->sync_thread) {
8534 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8535 wake_up(&resync_wait);
8536 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8538 if (mddev->sysfs_action)
8539 sysfs_notify_dirent_safe(mddev->sysfs_action);
8542 wake_up(&mddev->sb_wait);
8543 mddev_unlock(mddev);
8546 EXPORT_SYMBOL(md_check_recovery);
8548 void md_reap_sync_thread(struct mddev *mddev)
8550 struct md_rdev *rdev;
8552 /* resync has finished, collect result */
8553 md_unregister_thread(&mddev->sync_thread);
8554 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8555 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8557 /* activate any spares */
8558 if (mddev->pers->spare_active(mddev)) {
8559 sysfs_notify(&mddev->kobj, NULL,
8561 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8564 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8565 mddev->pers->finish_reshape)
8566 mddev->pers->finish_reshape(mddev);
8568 /* If array is no-longer degraded, then any saved_raid_disk
8569 * information must be scrapped.
8571 if (!mddev->degraded)
8572 rdev_for_each(rdev, mddev)
8573 rdev->saved_raid_disk = -1;
8575 md_update_sb(mddev, 1);
8576 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8577 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8579 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8580 md_cluster_ops->resync_finish(mddev);
8581 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8582 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8583 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8584 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8585 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8586 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8587 wake_up(&resync_wait);
8588 /* flag recovery needed just to double check */
8589 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8590 sysfs_notify_dirent_safe(mddev->sysfs_action);
8591 md_new_event(mddev);
8592 if (mddev->event_work.func)
8593 queue_work(md_misc_wq, &mddev->event_work);
8595 EXPORT_SYMBOL(md_reap_sync_thread);
8597 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8599 sysfs_notify_dirent_safe(rdev->sysfs_state);
8600 wait_event_timeout(rdev->blocked_wait,
8601 !test_bit(Blocked, &rdev->flags) &&
8602 !test_bit(BlockedBadBlocks, &rdev->flags),
8603 msecs_to_jiffies(5000));
8604 rdev_dec_pending(rdev, mddev);
8606 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8608 void md_finish_reshape(struct mddev *mddev)
8610 /* called be personality module when reshape completes. */
8611 struct md_rdev *rdev;
8613 rdev_for_each(rdev, mddev) {
8614 if (rdev->data_offset > rdev->new_data_offset)
8615 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8617 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8618 rdev->data_offset = rdev->new_data_offset;
8621 EXPORT_SYMBOL(md_finish_reshape);
8623 /* Bad block management */
8625 /* Returns 1 on success, 0 on failure */
8626 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8629 struct mddev *mddev = rdev->mddev;
8632 s += rdev->new_data_offset;
8634 s += rdev->data_offset;
8635 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8637 /* Make sure they get written out promptly */
8638 if (test_bit(ExternalBbl, &rdev->flags))
8639 sysfs_notify(&rdev->kobj, NULL,
8640 "unacknowledged_bad_blocks");
8641 sysfs_notify_dirent_safe(rdev->sysfs_state);
8642 set_mask_bits(&mddev->sb_flags, 0,
8643 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
8644 md_wakeup_thread(rdev->mddev->thread);
8649 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8651 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8656 s += rdev->new_data_offset;
8658 s += rdev->data_offset;
8659 rv = badblocks_clear(&rdev->badblocks, s, sectors);
8660 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
8661 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
8664 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8666 static int md_notify_reboot(struct notifier_block *this,
8667 unsigned long code, void *x)
8669 struct list_head *tmp;
8670 struct mddev *mddev;
8673 for_each_mddev(mddev, tmp) {
8674 if (mddev_trylock(mddev)) {
8676 __md_stop_writes(mddev);
8677 if (mddev->persistent)
8678 mddev->safemode = 2;
8679 mddev_unlock(mddev);
8684 * certain more exotic SCSI devices are known to be
8685 * volatile wrt too early system reboots. While the
8686 * right place to handle this issue is the given
8687 * driver, we do want to have a safe RAID driver ...
8695 static struct notifier_block md_notifier = {
8696 .notifier_call = md_notify_reboot,
8698 .priority = INT_MAX, /* before any real devices */
8701 static void md_geninit(void)
8703 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8705 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8708 static int __init md_init(void)
8712 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8716 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8720 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8723 if ((ret = register_blkdev(0, "mdp")) < 0)
8727 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8728 md_probe, NULL, NULL);
8729 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8730 md_probe, NULL, NULL);
8732 register_reboot_notifier(&md_notifier);
8733 raid_table_header = register_sysctl_table(raid_root_table);
8739 unregister_blkdev(MD_MAJOR, "md");
8741 destroy_workqueue(md_misc_wq);
8743 destroy_workqueue(md_wq);
8748 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8750 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8751 struct md_rdev *rdev2;
8753 char b[BDEVNAME_SIZE];
8755 /* Check for change of roles in the active devices */
8756 rdev_for_each(rdev2, mddev) {
8757 if (test_bit(Faulty, &rdev2->flags))
8760 /* Check if the roles changed */
8761 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8763 if (test_bit(Candidate, &rdev2->flags)) {
8764 if (role == 0xfffe) {
8765 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8766 md_kick_rdev_from_array(rdev2);
8770 clear_bit(Candidate, &rdev2->flags);
8773 if (role != rdev2->raid_disk) {
8775 if (rdev2->raid_disk == -1 && role != 0xffff) {
8776 rdev2->saved_raid_disk = role;
8777 ret = remove_and_add_spares(mddev, rdev2);
8778 pr_info("Activated spare: %s\n",
8779 bdevname(rdev2->bdev,b));
8780 /* wakeup mddev->thread here, so array could
8781 * perform resync with the new activated disk */
8782 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8783 md_wakeup_thread(mddev->thread);
8787 * We just want to do the minimum to mark the disk
8788 * as faulty. The recovery is performed by the
8789 * one who initiated the error.
8791 if ((role == 0xfffe) || (role == 0xfffd)) {
8792 md_error(mddev, rdev2);
8793 clear_bit(Blocked, &rdev2->flags);
8798 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8799 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8801 /* Finally set the event to be up to date */
8802 mddev->events = le64_to_cpu(sb->events);
8805 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8808 struct page *swapout = rdev->sb_page;
8809 struct mdp_superblock_1 *sb;
8811 /* Store the sb page of the rdev in the swapout temporary
8812 * variable in case we err in the future
8814 rdev->sb_page = NULL;
8815 err = alloc_disk_sb(rdev);
8817 ClearPageUptodate(rdev->sb_page);
8818 rdev->sb_loaded = 0;
8819 err = super_types[mddev->major_version].
8820 load_super(rdev, NULL, mddev->minor_version);
8823 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8824 __func__, __LINE__, rdev->desc_nr, err);
8826 put_page(rdev->sb_page);
8827 rdev->sb_page = swapout;
8828 rdev->sb_loaded = 1;
8832 sb = page_address(rdev->sb_page);
8833 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8837 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8838 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8840 /* The other node finished recovery, call spare_active to set
8841 * device In_sync and mddev->degraded
8843 if (rdev->recovery_offset == MaxSector &&
8844 !test_bit(In_sync, &rdev->flags) &&
8845 mddev->pers->spare_active(mddev))
8846 sysfs_notify(&mddev->kobj, NULL, "degraded");
8852 void md_reload_sb(struct mddev *mddev, int nr)
8854 struct md_rdev *rdev;
8858 rdev_for_each_rcu(rdev, mddev) {
8859 if (rdev->desc_nr == nr)
8863 if (!rdev || rdev->desc_nr != nr) {
8864 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8868 err = read_rdev(mddev, rdev);
8872 check_sb_changes(mddev, rdev);
8874 /* Read all rdev's to update recovery_offset */
8875 rdev_for_each_rcu(rdev, mddev)
8876 read_rdev(mddev, rdev);
8878 EXPORT_SYMBOL(md_reload_sb);
8883 * Searches all registered partitions for autorun RAID arrays
8887 static DEFINE_MUTEX(detected_devices_mutex);
8888 static LIST_HEAD(all_detected_devices);
8889 struct detected_devices_node {
8890 struct list_head list;
8894 void md_autodetect_dev(dev_t dev)
8896 struct detected_devices_node *node_detected_dev;
8898 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8899 if (node_detected_dev) {
8900 node_detected_dev->dev = dev;
8901 mutex_lock(&detected_devices_mutex);
8902 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8903 mutex_unlock(&detected_devices_mutex);
8907 static void autostart_arrays(int part)
8909 struct md_rdev *rdev;
8910 struct detected_devices_node *node_detected_dev;
8912 int i_scanned, i_passed;
8917 pr_info("md: Autodetecting RAID arrays.\n");
8919 mutex_lock(&detected_devices_mutex);
8920 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8922 node_detected_dev = list_entry(all_detected_devices.next,
8923 struct detected_devices_node, list);
8924 list_del(&node_detected_dev->list);
8925 dev = node_detected_dev->dev;
8926 kfree(node_detected_dev);
8927 mutex_unlock(&detected_devices_mutex);
8928 rdev = md_import_device(dev,0, 90);
8929 mutex_lock(&detected_devices_mutex);
8933 if (test_bit(Faulty, &rdev->flags))
8936 set_bit(AutoDetected, &rdev->flags);
8937 list_add(&rdev->same_set, &pending_raid_disks);
8940 mutex_unlock(&detected_devices_mutex);
8942 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
8944 autorun_devices(part);
8947 #endif /* !MODULE */
8949 static __exit void md_exit(void)
8951 struct mddev *mddev;
8952 struct list_head *tmp;
8955 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8956 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8958 unregister_blkdev(MD_MAJOR,"md");
8959 unregister_blkdev(mdp_major, "mdp");
8960 unregister_reboot_notifier(&md_notifier);
8961 unregister_sysctl_table(raid_table_header);
8963 /* We cannot unload the modules while some process is
8964 * waiting for us in select() or poll() - wake them up
8967 while (waitqueue_active(&md_event_waiters)) {
8968 /* not safe to leave yet */
8969 wake_up(&md_event_waiters);
8973 remove_proc_entry("mdstat", NULL);
8975 for_each_mddev(mddev, tmp) {
8976 export_array(mddev);
8978 mddev->hold_active = 0;
8980 * for_each_mddev() will call mddev_put() at the end of each
8981 * iteration. As the mddev is now fully clear, this will
8982 * schedule the mddev for destruction by a workqueue, and the
8983 * destroy_workqueue() below will wait for that to complete.
8986 destroy_workqueue(md_misc_wq);
8987 destroy_workqueue(md_wq);
8990 subsys_initcall(md_init);
8991 module_exit(md_exit)
8993 static int get_ro(char *buffer, struct kernel_param *kp)
8995 return sprintf(buffer, "%d", start_readonly);
8997 static int set_ro(const char *val, struct kernel_param *kp)
8999 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9002 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9003 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9004 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9006 MODULE_LICENSE("GPL");
9007 MODULE_DESCRIPTION("MD RAID framework");
9009 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);