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/kthread.h>
48 #include <linux/blkdev.h>
49 #include <linux/badblocks.h>
50 #include <linux/sysctl.h>
51 #include <linux/seq_file.h>
53 #include <linux/poll.h>
54 #include <linux/ctype.h>
55 #include <linux/string.h>
56 #include <linux/hdreg.h>
57 #include <linux/proc_fs.h>
58 #include <linux/random.h>
59 #include <linux/module.h>
60 #include <linux/reboot.h>
61 #include <linux/file.h>
62 #include <linux/compat.h>
63 #include <linux/delay.h>
64 #include <linux/raid/md_p.h>
65 #include <linux/raid/md_u.h>
66 #include <linux/slab.h>
67 #include <trace/events/block.h>
70 #include "md-cluster.h"
73 static void autostart_arrays(int part);
76 /* pers_list is a list of registered personalities protected
78 * pers_lock does extra service to protect accesses to
79 * mddev->thread when the mutex cannot be held.
81 static LIST_HEAD(pers_list);
82 static DEFINE_SPINLOCK(pers_lock);
84 struct md_cluster_operations *md_cluster_ops;
85 EXPORT_SYMBOL(md_cluster_ops);
86 struct module *md_cluster_mod;
87 EXPORT_SYMBOL(md_cluster_mod);
89 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
90 static struct workqueue_struct *md_wq;
91 static struct workqueue_struct *md_misc_wq;
93 static int remove_and_add_spares(struct mddev *mddev,
94 struct md_rdev *this);
95 static void mddev_detach(struct mddev *mddev);
98 * Default number of read corrections we'll attempt on an rdev
99 * before ejecting it from the array. We divide the read error
100 * count by 2 for every hour elapsed between read errors.
102 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
104 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
105 * is 1000 KB/sec, so the extra system load does not show up that much.
106 * Increase it if you want to have more _guaranteed_ speed. Note that
107 * the RAID driver will use the maximum available bandwidth if the IO
108 * subsystem is idle. There is also an 'absolute maximum' reconstruction
109 * speed limit - in case reconstruction slows down your system despite
112 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
113 * or /sys/block/mdX/md/sync_speed_{min,max}
116 static int sysctl_speed_limit_min = 1000;
117 static int sysctl_speed_limit_max = 200000;
118 static inline int speed_min(struct mddev *mddev)
120 return mddev->sync_speed_min ?
121 mddev->sync_speed_min : sysctl_speed_limit_min;
124 static inline int speed_max(struct mddev *mddev)
126 return mddev->sync_speed_max ?
127 mddev->sync_speed_max : sysctl_speed_limit_max;
130 static struct ctl_table_header *raid_table_header;
132 static struct ctl_table raid_table[] = {
134 .procname = "speed_limit_min",
135 .data = &sysctl_speed_limit_min,
136 .maxlen = sizeof(int),
137 .mode = S_IRUGO|S_IWUSR,
138 .proc_handler = proc_dointvec,
141 .procname = "speed_limit_max",
142 .data = &sysctl_speed_limit_max,
143 .maxlen = sizeof(int),
144 .mode = S_IRUGO|S_IWUSR,
145 .proc_handler = proc_dointvec,
150 static struct ctl_table raid_dir_table[] = {
154 .mode = S_IRUGO|S_IXUGO,
160 static struct ctl_table raid_root_table[] = {
165 .child = raid_dir_table,
170 static const struct block_device_operations md_fops;
172 static int start_readonly;
175 * like bio_clone, but with a local bio set
178 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
183 if (!mddev || !mddev->bio_set)
184 return bio_alloc(gfp_mask, nr_iovecs);
186 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
191 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
193 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
196 if (!mddev || !mddev->bio_set)
197 return bio_clone(bio, gfp_mask);
199 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
201 EXPORT_SYMBOL_GPL(bio_clone_mddev);
204 * We have a system wide 'event count' that is incremented
205 * on any 'interesting' event, and readers of /proc/mdstat
206 * can use 'poll' or 'select' to find out when the event
210 * start array, stop array, error, add device, remove device,
211 * start build, activate spare
213 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
214 static atomic_t md_event_count;
215 void md_new_event(struct mddev *mddev)
217 atomic_inc(&md_event_count);
218 wake_up(&md_event_waiters);
220 EXPORT_SYMBOL_GPL(md_new_event);
223 * Enables to iterate over all existing md arrays
224 * all_mddevs_lock protects this list.
226 static LIST_HEAD(all_mddevs);
227 static DEFINE_SPINLOCK(all_mddevs_lock);
230 * iterates through all used mddevs in the system.
231 * We take care to grab the all_mddevs_lock whenever navigating
232 * the list, and to always hold a refcount when unlocked.
233 * Any code which breaks out of this loop while own
234 * a reference to the current mddev and must mddev_put it.
236 #define for_each_mddev(_mddev,_tmp) \
238 for (({ spin_lock(&all_mddevs_lock); \
239 _tmp = all_mddevs.next; \
241 ({ if (_tmp != &all_mddevs) \
242 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
243 spin_unlock(&all_mddevs_lock); \
244 if (_mddev) mddev_put(_mddev); \
245 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
246 _tmp != &all_mddevs;}); \
247 ({ spin_lock(&all_mddevs_lock); \
248 _tmp = _tmp->next;}) \
251 /* Rather than calling directly into the personality make_request function,
252 * IO requests come here first so that we can check if the device is
253 * being suspended pending a reconfiguration.
254 * We hold a refcount over the call to ->make_request. By the time that
255 * call has finished, the bio has been linked into some internal structure
256 * and so is visible to ->quiesce(), so we don't need the refcount any more.
258 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
260 const int rw = bio_data_dir(bio);
261 struct mddev *mddev = q->queuedata;
262 unsigned int sectors;
265 blk_queue_split(q, &bio, q->bio_split);
267 if (mddev == NULL || mddev->pers == NULL) {
269 return BLK_QC_T_NONE;
271 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
272 if (bio_sectors(bio) != 0)
273 bio->bi_error = -EROFS;
275 return BLK_QC_T_NONE;
277 smp_rmb(); /* Ensure implications of 'active' are visible */
279 if (mddev->suspended) {
282 prepare_to_wait(&mddev->sb_wait, &__wait,
283 TASK_UNINTERRUPTIBLE);
284 if (!mddev->suspended)
290 finish_wait(&mddev->sb_wait, &__wait);
292 atomic_inc(&mddev->active_io);
296 * save the sectors now since our bio can
297 * go away inside make_request
299 sectors = bio_sectors(bio);
300 /* bio could be mergeable after passing to underlayer */
301 bio->bi_opf &= ~REQ_NOMERGE;
302 mddev->pers->make_request(mddev, bio);
304 cpu = part_stat_lock();
305 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
306 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
309 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
310 wake_up(&mddev->sb_wait);
312 return BLK_QC_T_NONE;
315 /* mddev_suspend makes sure no new requests are submitted
316 * to the device, and that any requests that have been submitted
317 * are completely handled.
318 * Once mddev_detach() is called and completes, the module will be
321 void mddev_suspend(struct mddev *mddev)
323 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
324 if (mddev->suspended++)
327 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
328 mddev->pers->quiesce(mddev, 1);
330 del_timer_sync(&mddev->safemode_timer);
332 EXPORT_SYMBOL_GPL(mddev_suspend);
334 void mddev_resume(struct mddev *mddev)
336 if (--mddev->suspended)
338 wake_up(&mddev->sb_wait);
339 mddev->pers->quiesce(mddev, 0);
341 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
342 md_wakeup_thread(mddev->thread);
343 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
345 EXPORT_SYMBOL_GPL(mddev_resume);
347 int mddev_congested(struct mddev *mddev, int bits)
349 struct md_personality *pers = mddev->pers;
353 if (mddev->suspended)
355 else if (pers && pers->congested)
356 ret = pers->congested(mddev, bits);
360 EXPORT_SYMBOL_GPL(mddev_congested);
361 static int md_congested(void *data, int bits)
363 struct mddev *mddev = data;
364 return mddev_congested(mddev, bits);
368 * Generic flush handling for md
371 static void md_end_flush(struct bio *bio)
373 struct md_rdev *rdev = bio->bi_private;
374 struct mddev *mddev = rdev->mddev;
376 rdev_dec_pending(rdev, mddev);
378 if (atomic_dec_and_test(&mddev->flush_pending)) {
379 /* The pre-request flush has finished */
380 queue_work(md_wq, &mddev->flush_work);
385 static void md_submit_flush_data(struct work_struct *ws);
387 static void submit_flushes(struct work_struct *ws)
389 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
390 struct md_rdev *rdev;
392 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
393 atomic_set(&mddev->flush_pending, 1);
395 rdev_for_each_rcu(rdev, mddev)
396 if (rdev->raid_disk >= 0 &&
397 !test_bit(Faulty, &rdev->flags)) {
398 /* Take two references, one is dropped
399 * when request finishes, one after
400 * we reclaim rcu_read_lock
403 atomic_inc(&rdev->nr_pending);
404 atomic_inc(&rdev->nr_pending);
406 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
407 bi->bi_end_io = md_end_flush;
408 bi->bi_private = rdev;
409 bi->bi_bdev = rdev->bdev;
410 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
411 atomic_inc(&mddev->flush_pending);
414 rdev_dec_pending(rdev, mddev);
417 if (atomic_dec_and_test(&mddev->flush_pending))
418 queue_work(md_wq, &mddev->flush_work);
421 static void md_submit_flush_data(struct work_struct *ws)
423 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
424 struct bio *bio = mddev->flush_bio;
426 if (bio->bi_iter.bi_size == 0)
427 /* an empty barrier - all done */
430 bio->bi_opf &= ~REQ_PREFLUSH;
431 mddev->pers->make_request(mddev, bio);
434 mddev->flush_bio = NULL;
435 wake_up(&mddev->sb_wait);
438 void md_flush_request(struct mddev *mddev, struct bio *bio)
440 spin_lock_irq(&mddev->lock);
441 wait_event_lock_irq(mddev->sb_wait,
444 mddev->flush_bio = bio;
445 spin_unlock_irq(&mddev->lock);
447 INIT_WORK(&mddev->flush_work, submit_flushes);
448 queue_work(md_wq, &mddev->flush_work);
450 EXPORT_SYMBOL(md_flush_request);
452 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
454 struct mddev *mddev = cb->data;
455 md_wakeup_thread(mddev->thread);
458 EXPORT_SYMBOL(md_unplug);
460 static inline struct mddev *mddev_get(struct mddev *mddev)
462 atomic_inc(&mddev->active);
466 static void mddev_delayed_delete(struct work_struct *ws);
468 static void mddev_put(struct mddev *mddev)
470 struct bio_set *bs = NULL;
472 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
474 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
475 mddev->ctime == 0 && !mddev->hold_active) {
476 /* Array is not configured at all, and not held active,
478 list_del_init(&mddev->all_mddevs);
480 mddev->bio_set = NULL;
481 if (mddev->gendisk) {
482 /* We did a probe so need to clean up. Call
483 * queue_work inside the spinlock so that
484 * flush_workqueue() after mddev_find will
485 * succeed in waiting for the work to be done.
487 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
488 queue_work(md_misc_wq, &mddev->del_work);
492 spin_unlock(&all_mddevs_lock);
497 static void md_safemode_timeout(unsigned long data);
499 void mddev_init(struct mddev *mddev)
501 mutex_init(&mddev->open_mutex);
502 mutex_init(&mddev->reconfig_mutex);
503 mutex_init(&mddev->bitmap_info.mutex);
504 INIT_LIST_HEAD(&mddev->disks);
505 INIT_LIST_HEAD(&mddev->all_mddevs);
506 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
507 (unsigned long) mddev);
508 atomic_set(&mddev->active, 1);
509 atomic_set(&mddev->openers, 0);
510 atomic_set(&mddev->active_io, 0);
511 spin_lock_init(&mddev->lock);
512 atomic_set(&mddev->flush_pending, 0);
513 init_waitqueue_head(&mddev->sb_wait);
514 init_waitqueue_head(&mddev->recovery_wait);
515 mddev->reshape_position = MaxSector;
516 mddev->reshape_backwards = 0;
517 mddev->last_sync_action = "none";
518 mddev->resync_min = 0;
519 mddev->resync_max = MaxSector;
520 mddev->level = LEVEL_NONE;
522 EXPORT_SYMBOL_GPL(mddev_init);
524 static struct mddev *mddev_find(dev_t unit)
526 struct mddev *mddev, *new = NULL;
528 if (unit && MAJOR(unit) != MD_MAJOR)
529 unit &= ~((1<<MdpMinorShift)-1);
532 spin_lock(&all_mddevs_lock);
535 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
536 if (mddev->unit == unit) {
538 spin_unlock(&all_mddevs_lock);
544 list_add(&new->all_mddevs, &all_mddevs);
545 spin_unlock(&all_mddevs_lock);
546 new->hold_active = UNTIL_IOCTL;
550 /* find an unused unit number */
551 static int next_minor = 512;
552 int start = next_minor;
556 dev = MKDEV(MD_MAJOR, next_minor);
558 if (next_minor > MINORMASK)
560 if (next_minor == start) {
561 /* Oh dear, all in use. */
562 spin_unlock(&all_mddevs_lock);
568 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
569 if (mddev->unit == dev) {
575 new->md_minor = MINOR(dev);
576 new->hold_active = UNTIL_STOP;
577 list_add(&new->all_mddevs, &all_mddevs);
578 spin_unlock(&all_mddevs_lock);
581 spin_unlock(&all_mddevs_lock);
583 new = kzalloc(sizeof(*new), GFP_KERNEL);
588 if (MAJOR(unit) == MD_MAJOR)
589 new->md_minor = MINOR(unit);
591 new->md_minor = MINOR(unit) >> MdpMinorShift;
598 static struct attribute_group md_redundancy_group;
600 void mddev_unlock(struct mddev *mddev)
602 if (mddev->to_remove) {
603 /* These cannot be removed under reconfig_mutex as
604 * an access to the files will try to take reconfig_mutex
605 * while holding the file unremovable, which leads to
607 * So hold set sysfs_active while the remove in happeing,
608 * and anything else which might set ->to_remove or my
609 * otherwise change the sysfs namespace will fail with
610 * -EBUSY if sysfs_active is still set.
611 * We set sysfs_active under reconfig_mutex and elsewhere
612 * test it under the same mutex to ensure its correct value
615 struct attribute_group *to_remove = mddev->to_remove;
616 mddev->to_remove = NULL;
617 mddev->sysfs_active = 1;
618 mutex_unlock(&mddev->reconfig_mutex);
620 if (mddev->kobj.sd) {
621 if (to_remove != &md_redundancy_group)
622 sysfs_remove_group(&mddev->kobj, to_remove);
623 if (mddev->pers == NULL ||
624 mddev->pers->sync_request == NULL) {
625 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
626 if (mddev->sysfs_action)
627 sysfs_put(mddev->sysfs_action);
628 mddev->sysfs_action = NULL;
631 mddev->sysfs_active = 0;
633 mutex_unlock(&mddev->reconfig_mutex);
635 /* As we've dropped the mutex we need a spinlock to
636 * make sure the thread doesn't disappear
638 spin_lock(&pers_lock);
639 md_wakeup_thread(mddev->thread);
640 spin_unlock(&pers_lock);
642 EXPORT_SYMBOL_GPL(mddev_unlock);
644 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
646 struct md_rdev *rdev;
648 rdev_for_each_rcu(rdev, mddev)
649 if (rdev->desc_nr == nr)
654 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
656 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
658 struct md_rdev *rdev;
660 rdev_for_each(rdev, mddev)
661 if (rdev->bdev->bd_dev == dev)
667 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
669 struct md_rdev *rdev;
671 rdev_for_each_rcu(rdev, mddev)
672 if (rdev->bdev->bd_dev == dev)
678 static struct md_personality *find_pers(int level, char *clevel)
680 struct md_personality *pers;
681 list_for_each_entry(pers, &pers_list, list) {
682 if (level != LEVEL_NONE && pers->level == level)
684 if (strcmp(pers->name, clevel)==0)
690 /* return the offset of the super block in 512byte sectors */
691 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
693 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
694 return MD_NEW_SIZE_SECTORS(num_sectors);
697 static int alloc_disk_sb(struct md_rdev *rdev)
699 rdev->sb_page = alloc_page(GFP_KERNEL);
705 void md_rdev_clear(struct md_rdev *rdev)
708 put_page(rdev->sb_page);
710 rdev->sb_page = NULL;
715 put_page(rdev->bb_page);
716 rdev->bb_page = NULL;
718 badblocks_exit(&rdev->badblocks);
720 EXPORT_SYMBOL_GPL(md_rdev_clear);
722 static void super_written(struct bio *bio)
724 struct md_rdev *rdev = bio->bi_private;
725 struct mddev *mddev = rdev->mddev;
728 pr_err("md: super_written gets error=%d\n", bio->bi_error);
729 md_error(mddev, rdev);
730 if (!test_bit(Faulty, &rdev->flags)
731 && (bio->bi_opf & MD_FAILFAST)) {
732 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
733 set_bit(LastDev, &rdev->flags);
736 clear_bit(LastDev, &rdev->flags);
738 if (atomic_dec_and_test(&mddev->pending_writes))
739 wake_up(&mddev->sb_wait);
740 rdev_dec_pending(rdev, mddev);
744 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
745 sector_t sector, int size, struct page *page)
747 /* write first size bytes of page to sector of rdev
748 * Increment mddev->pending_writes before returning
749 * and decrement it on completion, waking up sb_wait
750 * if zero is reached.
751 * If an error occurred, call md_error
756 if (test_bit(Faulty, &rdev->flags))
759 bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
761 atomic_inc(&rdev->nr_pending);
763 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
764 bio->bi_iter.bi_sector = sector;
765 bio_add_page(bio, page, size, 0);
766 bio->bi_private = rdev;
767 bio->bi_end_io = super_written;
769 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
770 test_bit(FailFast, &rdev->flags) &&
771 !test_bit(LastDev, &rdev->flags))
773 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_FUA | ff;
775 atomic_inc(&mddev->pending_writes);
779 int md_super_wait(struct mddev *mddev)
781 /* wait for all superblock writes that were scheduled to complete */
782 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
783 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
788 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
789 struct page *page, int op, int op_flags, bool metadata_op)
791 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
794 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
795 rdev->meta_bdev : rdev->bdev;
796 bio_set_op_attrs(bio, op, op_flags);
798 bio->bi_iter.bi_sector = sector + rdev->sb_start;
799 else if (rdev->mddev->reshape_position != MaxSector &&
800 (rdev->mddev->reshape_backwards ==
801 (sector >= rdev->mddev->reshape_position)))
802 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
804 bio->bi_iter.bi_sector = sector + rdev->data_offset;
805 bio_add_page(bio, page, size, 0);
807 submit_bio_wait(bio);
809 ret = !bio->bi_error;
813 EXPORT_SYMBOL_GPL(sync_page_io);
815 static int read_disk_sb(struct md_rdev *rdev, int size)
817 char b[BDEVNAME_SIZE];
822 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
828 pr_err("md: disabled device %s, could not read superblock.\n",
829 bdevname(rdev->bdev,b));
833 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
835 return sb1->set_uuid0 == sb2->set_uuid0 &&
836 sb1->set_uuid1 == sb2->set_uuid1 &&
837 sb1->set_uuid2 == sb2->set_uuid2 &&
838 sb1->set_uuid3 == sb2->set_uuid3;
841 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
844 mdp_super_t *tmp1, *tmp2;
846 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
847 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
849 if (!tmp1 || !tmp2) {
858 * nr_disks is not constant
863 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
870 static u32 md_csum_fold(u32 csum)
872 csum = (csum & 0xffff) + (csum >> 16);
873 return (csum & 0xffff) + (csum >> 16);
876 static unsigned int calc_sb_csum(mdp_super_t *sb)
879 u32 *sb32 = (u32*)sb;
881 unsigned int disk_csum, csum;
883 disk_csum = sb->sb_csum;
886 for (i = 0; i < MD_SB_BYTES/4 ; i++)
888 csum = (newcsum & 0xffffffff) + (newcsum>>32);
891 /* This used to use csum_partial, which was wrong for several
892 * reasons including that different results are returned on
893 * different architectures. It isn't critical that we get exactly
894 * the same return value as before (we always csum_fold before
895 * testing, and that removes any differences). However as we
896 * know that csum_partial always returned a 16bit value on
897 * alphas, do a fold to maximise conformity to previous behaviour.
899 sb->sb_csum = md_csum_fold(disk_csum);
901 sb->sb_csum = disk_csum;
907 * Handle superblock details.
908 * We want to be able to handle multiple superblock formats
909 * so we have a common interface to them all, and an array of
910 * different handlers.
911 * We rely on user-space to write the initial superblock, and support
912 * reading and updating of superblocks.
913 * Interface methods are:
914 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
915 * loads and validates a superblock on dev.
916 * if refdev != NULL, compare superblocks on both devices
918 * 0 - dev has a superblock that is compatible with refdev
919 * 1 - dev has a superblock that is compatible and newer than refdev
920 * so dev should be used as the refdev in future
921 * -EINVAL superblock incompatible or invalid
922 * -othererror e.g. -EIO
924 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
925 * Verify that dev is acceptable into mddev.
926 * The first time, mddev->raid_disks will be 0, and data from
927 * dev should be merged in. Subsequent calls check that dev
928 * is new enough. Return 0 or -EINVAL
930 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
931 * Update the superblock for rdev with data in mddev
932 * This does not write to disc.
938 struct module *owner;
939 int (*load_super)(struct md_rdev *rdev,
940 struct md_rdev *refdev,
942 int (*validate_super)(struct mddev *mddev,
943 struct md_rdev *rdev);
944 void (*sync_super)(struct mddev *mddev,
945 struct md_rdev *rdev);
946 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
947 sector_t num_sectors);
948 int (*allow_new_offset)(struct md_rdev *rdev,
949 unsigned long long new_offset);
953 * Check that the given mddev has no bitmap.
955 * This function is called from the run method of all personalities that do not
956 * support bitmaps. It prints an error message and returns non-zero if mddev
957 * has a bitmap. Otherwise, it returns 0.
960 int md_check_no_bitmap(struct mddev *mddev)
962 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
964 pr_warn("%s: bitmaps are not supported for %s\n",
965 mdname(mddev), mddev->pers->name);
968 EXPORT_SYMBOL(md_check_no_bitmap);
971 * load_super for 0.90.0
973 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
975 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
980 * Calculate the position of the superblock (512byte sectors),
981 * it's at the end of the disk.
983 * It also happens to be a multiple of 4Kb.
985 rdev->sb_start = calc_dev_sboffset(rdev);
987 ret = read_disk_sb(rdev, MD_SB_BYTES);
993 bdevname(rdev->bdev, b);
994 sb = page_address(rdev->sb_page);
996 if (sb->md_magic != MD_SB_MAGIC) {
997 pr_warn("md: invalid raid superblock magic on %s\n", b);
1001 if (sb->major_version != 0 ||
1002 sb->minor_version < 90 ||
1003 sb->minor_version > 91) {
1004 pr_warn("Bad version number %d.%d on %s\n",
1005 sb->major_version, sb->minor_version, b);
1009 if (sb->raid_disks <= 0)
1012 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1013 pr_warn("md: invalid superblock checksum on %s\n", b);
1017 rdev->preferred_minor = sb->md_minor;
1018 rdev->data_offset = 0;
1019 rdev->new_data_offset = 0;
1020 rdev->sb_size = MD_SB_BYTES;
1021 rdev->badblocks.shift = -1;
1023 if (sb->level == LEVEL_MULTIPATH)
1026 rdev->desc_nr = sb->this_disk.number;
1032 mdp_super_t *refsb = page_address(refdev->sb_page);
1033 if (!uuid_equal(refsb, sb)) {
1034 pr_warn("md: %s has different UUID to %s\n",
1035 b, bdevname(refdev->bdev,b2));
1038 if (!sb_equal(refsb, sb)) {
1039 pr_warn("md: %s has same UUID but different superblock to %s\n",
1040 b, bdevname(refdev->bdev, b2));
1044 ev2 = md_event(refsb);
1050 rdev->sectors = rdev->sb_start;
1051 /* Limit to 4TB as metadata cannot record more than that.
1052 * (not needed for Linear and RAID0 as metadata doesn't
1055 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1057 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1059 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1060 /* "this cannot possibly happen" ... */
1068 * validate_super for 0.90.0
1070 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1073 mdp_super_t *sb = page_address(rdev->sb_page);
1074 __u64 ev1 = md_event(sb);
1076 rdev->raid_disk = -1;
1077 clear_bit(Faulty, &rdev->flags);
1078 clear_bit(In_sync, &rdev->flags);
1079 clear_bit(Bitmap_sync, &rdev->flags);
1080 clear_bit(WriteMostly, &rdev->flags);
1082 if (mddev->raid_disks == 0) {
1083 mddev->major_version = 0;
1084 mddev->minor_version = sb->minor_version;
1085 mddev->patch_version = sb->patch_version;
1086 mddev->external = 0;
1087 mddev->chunk_sectors = sb->chunk_size >> 9;
1088 mddev->ctime = sb->ctime;
1089 mddev->utime = sb->utime;
1090 mddev->level = sb->level;
1091 mddev->clevel[0] = 0;
1092 mddev->layout = sb->layout;
1093 mddev->raid_disks = sb->raid_disks;
1094 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1095 mddev->events = ev1;
1096 mddev->bitmap_info.offset = 0;
1097 mddev->bitmap_info.space = 0;
1098 /* bitmap can use 60 K after the 4K superblocks */
1099 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1100 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1101 mddev->reshape_backwards = 0;
1103 if (mddev->minor_version >= 91) {
1104 mddev->reshape_position = sb->reshape_position;
1105 mddev->delta_disks = sb->delta_disks;
1106 mddev->new_level = sb->new_level;
1107 mddev->new_layout = sb->new_layout;
1108 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1109 if (mddev->delta_disks < 0)
1110 mddev->reshape_backwards = 1;
1112 mddev->reshape_position = MaxSector;
1113 mddev->delta_disks = 0;
1114 mddev->new_level = mddev->level;
1115 mddev->new_layout = mddev->layout;
1116 mddev->new_chunk_sectors = mddev->chunk_sectors;
1119 if (sb->state & (1<<MD_SB_CLEAN))
1120 mddev->recovery_cp = MaxSector;
1122 if (sb->events_hi == sb->cp_events_hi &&
1123 sb->events_lo == sb->cp_events_lo) {
1124 mddev->recovery_cp = sb->recovery_cp;
1126 mddev->recovery_cp = 0;
1129 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1130 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1131 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1132 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1134 mddev->max_disks = MD_SB_DISKS;
1136 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1137 mddev->bitmap_info.file == NULL) {
1138 mddev->bitmap_info.offset =
1139 mddev->bitmap_info.default_offset;
1140 mddev->bitmap_info.space =
1141 mddev->bitmap_info.default_space;
1144 } else if (mddev->pers == NULL) {
1145 /* Insist on good event counter while assembling, except
1146 * for spares (which don't need an event count) */
1148 if (sb->disks[rdev->desc_nr].state & (
1149 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1150 if (ev1 < mddev->events)
1152 } else if (mddev->bitmap) {
1153 /* if adding to array with a bitmap, then we can accept an
1154 * older device ... but not too old.
1156 if (ev1 < mddev->bitmap->events_cleared)
1158 if (ev1 < mddev->events)
1159 set_bit(Bitmap_sync, &rdev->flags);
1161 if (ev1 < mddev->events)
1162 /* just a hot-add of a new device, leave raid_disk at -1 */
1166 if (mddev->level != LEVEL_MULTIPATH) {
1167 desc = sb->disks + rdev->desc_nr;
1169 if (desc->state & (1<<MD_DISK_FAULTY))
1170 set_bit(Faulty, &rdev->flags);
1171 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1172 desc->raid_disk < mddev->raid_disks */) {
1173 set_bit(In_sync, &rdev->flags);
1174 rdev->raid_disk = desc->raid_disk;
1175 rdev->saved_raid_disk = desc->raid_disk;
1176 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1177 /* active but not in sync implies recovery up to
1178 * reshape position. We don't know exactly where
1179 * that is, so set to zero for now */
1180 if (mddev->minor_version >= 91) {
1181 rdev->recovery_offset = 0;
1182 rdev->raid_disk = desc->raid_disk;
1185 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1186 set_bit(WriteMostly, &rdev->flags);
1187 if (desc->state & (1<<MD_DISK_FAILFAST))
1188 set_bit(FailFast, &rdev->flags);
1189 } else /* MULTIPATH are always insync */
1190 set_bit(In_sync, &rdev->flags);
1195 * sync_super for 0.90.0
1197 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1200 struct md_rdev *rdev2;
1201 int next_spare = mddev->raid_disks;
1203 /* make rdev->sb match mddev data..
1206 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1207 * 3/ any empty disks < next_spare become removed
1209 * disks[0] gets initialised to REMOVED because
1210 * we cannot be sure from other fields if it has
1211 * been initialised or not.
1214 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1216 rdev->sb_size = MD_SB_BYTES;
1218 sb = page_address(rdev->sb_page);
1220 memset(sb, 0, sizeof(*sb));
1222 sb->md_magic = MD_SB_MAGIC;
1223 sb->major_version = mddev->major_version;
1224 sb->patch_version = mddev->patch_version;
1225 sb->gvalid_words = 0; /* ignored */
1226 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1227 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1228 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1229 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1231 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1232 sb->level = mddev->level;
1233 sb->size = mddev->dev_sectors / 2;
1234 sb->raid_disks = mddev->raid_disks;
1235 sb->md_minor = mddev->md_minor;
1236 sb->not_persistent = 0;
1237 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1239 sb->events_hi = (mddev->events>>32);
1240 sb->events_lo = (u32)mddev->events;
1242 if (mddev->reshape_position == MaxSector)
1243 sb->minor_version = 90;
1245 sb->minor_version = 91;
1246 sb->reshape_position = mddev->reshape_position;
1247 sb->new_level = mddev->new_level;
1248 sb->delta_disks = mddev->delta_disks;
1249 sb->new_layout = mddev->new_layout;
1250 sb->new_chunk = mddev->new_chunk_sectors << 9;
1252 mddev->minor_version = sb->minor_version;
1255 sb->recovery_cp = mddev->recovery_cp;
1256 sb->cp_events_hi = (mddev->events>>32);
1257 sb->cp_events_lo = (u32)mddev->events;
1258 if (mddev->recovery_cp == MaxSector)
1259 sb->state = (1<< MD_SB_CLEAN);
1261 sb->recovery_cp = 0;
1263 sb->layout = mddev->layout;
1264 sb->chunk_size = mddev->chunk_sectors << 9;
1266 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1267 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1269 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1270 rdev_for_each(rdev2, mddev) {
1273 int is_active = test_bit(In_sync, &rdev2->flags);
1275 if (rdev2->raid_disk >= 0 &&
1276 sb->minor_version >= 91)
1277 /* we have nowhere to store the recovery_offset,
1278 * but if it is not below the reshape_position,
1279 * we can piggy-back on that.
1282 if (rdev2->raid_disk < 0 ||
1283 test_bit(Faulty, &rdev2->flags))
1286 desc_nr = rdev2->raid_disk;
1288 desc_nr = next_spare++;
1289 rdev2->desc_nr = desc_nr;
1290 d = &sb->disks[rdev2->desc_nr];
1292 d->number = rdev2->desc_nr;
1293 d->major = MAJOR(rdev2->bdev->bd_dev);
1294 d->minor = MINOR(rdev2->bdev->bd_dev);
1296 d->raid_disk = rdev2->raid_disk;
1298 d->raid_disk = rdev2->desc_nr; /* compatibility */
1299 if (test_bit(Faulty, &rdev2->flags))
1300 d->state = (1<<MD_DISK_FAULTY);
1301 else if (is_active) {
1302 d->state = (1<<MD_DISK_ACTIVE);
1303 if (test_bit(In_sync, &rdev2->flags))
1304 d->state |= (1<<MD_DISK_SYNC);
1312 if (test_bit(WriteMostly, &rdev2->flags))
1313 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1314 if (test_bit(FailFast, &rdev2->flags))
1315 d->state |= (1<<MD_DISK_FAILFAST);
1317 /* now set the "removed" and "faulty" bits on any missing devices */
1318 for (i=0 ; i < mddev->raid_disks ; i++) {
1319 mdp_disk_t *d = &sb->disks[i];
1320 if (d->state == 0 && d->number == 0) {
1323 d->state = (1<<MD_DISK_REMOVED);
1324 d->state |= (1<<MD_DISK_FAULTY);
1328 sb->nr_disks = nr_disks;
1329 sb->active_disks = active;
1330 sb->working_disks = working;
1331 sb->failed_disks = failed;
1332 sb->spare_disks = spare;
1334 sb->this_disk = sb->disks[rdev->desc_nr];
1335 sb->sb_csum = calc_sb_csum(sb);
1339 * rdev_size_change for 0.90.0
1341 static unsigned long long
1342 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1344 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1345 return 0; /* component must fit device */
1346 if (rdev->mddev->bitmap_info.offset)
1347 return 0; /* can't move bitmap */
1348 rdev->sb_start = calc_dev_sboffset(rdev);
1349 if (!num_sectors || num_sectors > rdev->sb_start)
1350 num_sectors = rdev->sb_start;
1351 /* Limit to 4TB as metadata cannot record more than that.
1352 * 4TB == 2^32 KB, or 2*2^32 sectors.
1354 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1355 rdev->mddev->level >= 1)
1356 num_sectors = (sector_t)(2ULL << 32) - 2;
1358 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1360 } while (md_super_wait(rdev->mddev) < 0);
1365 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1367 /* non-zero offset changes not possible with v0.90 */
1368 return new_offset == 0;
1372 * version 1 superblock
1375 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1379 unsigned long long newcsum;
1380 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1381 __le32 *isuper = (__le32*)sb;
1383 disk_csum = sb->sb_csum;
1386 for (; size >= 4; size -= 4)
1387 newcsum += le32_to_cpu(*isuper++);
1390 newcsum += le16_to_cpu(*(__le16*) isuper);
1392 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1393 sb->sb_csum = disk_csum;
1394 return cpu_to_le32(csum);
1397 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1399 struct mdp_superblock_1 *sb;
1403 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1407 * Calculate the position of the superblock in 512byte sectors.
1408 * It is always aligned to a 4K boundary and
1409 * depeding on minor_version, it can be:
1410 * 0: At least 8K, but less than 12K, from end of device
1411 * 1: At start of device
1412 * 2: 4K from start of device.
1414 switch(minor_version) {
1416 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1418 sb_start &= ~(sector_t)(4*2-1);
1429 rdev->sb_start = sb_start;
1431 /* superblock is rarely larger than 1K, but it can be larger,
1432 * and it is safe to read 4k, so we do that
1434 ret = read_disk_sb(rdev, 4096);
1435 if (ret) return ret;
1437 sb = page_address(rdev->sb_page);
1439 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1440 sb->major_version != cpu_to_le32(1) ||
1441 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1442 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1443 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1446 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1447 pr_warn("md: invalid superblock checksum on %s\n",
1448 bdevname(rdev->bdev,b));
1451 if (le64_to_cpu(sb->data_size) < 10) {
1452 pr_warn("md: data_size too small on %s\n",
1453 bdevname(rdev->bdev,b));
1458 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1459 /* Some padding is non-zero, might be a new feature */
1462 rdev->preferred_minor = 0xffff;
1463 rdev->data_offset = le64_to_cpu(sb->data_offset);
1464 rdev->new_data_offset = rdev->data_offset;
1465 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1466 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1467 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1468 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1470 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1471 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1472 if (rdev->sb_size & bmask)
1473 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1476 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1479 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1482 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1485 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1487 if (!rdev->bb_page) {
1488 rdev->bb_page = alloc_page(GFP_KERNEL);
1492 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1493 rdev->badblocks.count == 0) {
1494 /* need to load the bad block list.
1495 * Currently we limit it to one page.
1501 int sectors = le16_to_cpu(sb->bblog_size);
1502 if (sectors > (PAGE_SIZE / 512))
1504 offset = le32_to_cpu(sb->bblog_offset);
1507 bb_sector = (long long)offset;
1508 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1509 rdev->bb_page, REQ_OP_READ, 0, true))
1511 bbp = (u64 *)page_address(rdev->bb_page);
1512 rdev->badblocks.shift = sb->bblog_shift;
1513 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1514 u64 bb = le64_to_cpu(*bbp);
1515 int count = bb & (0x3ff);
1516 u64 sector = bb >> 10;
1517 sector <<= sb->bblog_shift;
1518 count <<= sb->bblog_shift;
1521 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1524 } else if (sb->bblog_offset != 0)
1525 rdev->badblocks.shift = 0;
1531 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1533 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1534 sb->level != refsb->level ||
1535 sb->layout != refsb->layout ||
1536 sb->chunksize != refsb->chunksize) {
1537 pr_warn("md: %s has strangely different superblock to %s\n",
1538 bdevname(rdev->bdev,b),
1539 bdevname(refdev->bdev,b2));
1542 ev1 = le64_to_cpu(sb->events);
1543 ev2 = le64_to_cpu(refsb->events);
1550 if (minor_version) {
1551 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1552 sectors -= rdev->data_offset;
1554 sectors = rdev->sb_start;
1555 if (sectors < le64_to_cpu(sb->data_size))
1557 rdev->sectors = le64_to_cpu(sb->data_size);
1561 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1563 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1564 __u64 ev1 = le64_to_cpu(sb->events);
1566 rdev->raid_disk = -1;
1567 clear_bit(Faulty, &rdev->flags);
1568 clear_bit(In_sync, &rdev->flags);
1569 clear_bit(Bitmap_sync, &rdev->flags);
1570 clear_bit(WriteMostly, &rdev->flags);
1572 if (mddev->raid_disks == 0) {
1573 mddev->major_version = 1;
1574 mddev->patch_version = 0;
1575 mddev->external = 0;
1576 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1577 mddev->ctime = le64_to_cpu(sb->ctime);
1578 mddev->utime = le64_to_cpu(sb->utime);
1579 mddev->level = le32_to_cpu(sb->level);
1580 mddev->clevel[0] = 0;
1581 mddev->layout = le32_to_cpu(sb->layout);
1582 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1583 mddev->dev_sectors = le64_to_cpu(sb->size);
1584 mddev->events = ev1;
1585 mddev->bitmap_info.offset = 0;
1586 mddev->bitmap_info.space = 0;
1587 /* Default location for bitmap is 1K after superblock
1588 * using 3K - total of 4K
1590 mddev->bitmap_info.default_offset = 1024 >> 9;
1591 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1592 mddev->reshape_backwards = 0;
1594 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1595 memcpy(mddev->uuid, sb->set_uuid, 16);
1597 mddev->max_disks = (4096-256)/2;
1599 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1600 mddev->bitmap_info.file == NULL) {
1601 mddev->bitmap_info.offset =
1602 (__s32)le32_to_cpu(sb->bitmap_offset);
1603 /* Metadata doesn't record how much space is available.
1604 * For 1.0, we assume we can use up to the superblock
1605 * if before, else to 4K beyond superblock.
1606 * For others, assume no change is possible.
1608 if (mddev->minor_version > 0)
1609 mddev->bitmap_info.space = 0;
1610 else if (mddev->bitmap_info.offset > 0)
1611 mddev->bitmap_info.space =
1612 8 - mddev->bitmap_info.offset;
1614 mddev->bitmap_info.space =
1615 -mddev->bitmap_info.offset;
1618 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1619 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1620 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1621 mddev->new_level = le32_to_cpu(sb->new_level);
1622 mddev->new_layout = le32_to_cpu(sb->new_layout);
1623 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1624 if (mddev->delta_disks < 0 ||
1625 (mddev->delta_disks == 0 &&
1626 (le32_to_cpu(sb->feature_map)
1627 & MD_FEATURE_RESHAPE_BACKWARDS)))
1628 mddev->reshape_backwards = 1;
1630 mddev->reshape_position = MaxSector;
1631 mddev->delta_disks = 0;
1632 mddev->new_level = mddev->level;
1633 mddev->new_layout = mddev->layout;
1634 mddev->new_chunk_sectors = mddev->chunk_sectors;
1637 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1638 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1639 } else if (mddev->pers == NULL) {
1640 /* Insist of good event counter while assembling, except for
1641 * spares (which don't need an event count) */
1643 if (rdev->desc_nr >= 0 &&
1644 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1645 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1646 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1647 if (ev1 < mddev->events)
1649 } else if (mddev->bitmap) {
1650 /* If adding to array with a bitmap, then we can accept an
1651 * older device, but not too old.
1653 if (ev1 < mddev->bitmap->events_cleared)
1655 if (ev1 < mddev->events)
1656 set_bit(Bitmap_sync, &rdev->flags);
1658 if (ev1 < mddev->events)
1659 /* just a hot-add of a new device, leave raid_disk at -1 */
1662 if (mddev->level != LEVEL_MULTIPATH) {
1664 if (rdev->desc_nr < 0 ||
1665 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1666 role = MD_DISK_ROLE_SPARE;
1669 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1671 case MD_DISK_ROLE_SPARE: /* spare */
1673 case MD_DISK_ROLE_FAULTY: /* faulty */
1674 set_bit(Faulty, &rdev->flags);
1676 case MD_DISK_ROLE_JOURNAL: /* journal device */
1677 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1678 /* journal device without journal feature */
1679 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1682 set_bit(Journal, &rdev->flags);
1683 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1684 rdev->raid_disk = 0;
1687 rdev->saved_raid_disk = role;
1688 if ((le32_to_cpu(sb->feature_map) &
1689 MD_FEATURE_RECOVERY_OFFSET)) {
1690 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1691 if (!(le32_to_cpu(sb->feature_map) &
1692 MD_FEATURE_RECOVERY_BITMAP))
1693 rdev->saved_raid_disk = -1;
1695 set_bit(In_sync, &rdev->flags);
1696 rdev->raid_disk = role;
1699 if (sb->devflags & WriteMostly1)
1700 set_bit(WriteMostly, &rdev->flags);
1701 if (sb->devflags & FailFast1)
1702 set_bit(FailFast, &rdev->flags);
1703 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1704 set_bit(Replacement, &rdev->flags);
1705 } else /* MULTIPATH are always insync */
1706 set_bit(In_sync, &rdev->flags);
1711 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1713 struct mdp_superblock_1 *sb;
1714 struct md_rdev *rdev2;
1716 /* make rdev->sb match mddev and rdev data. */
1718 sb = page_address(rdev->sb_page);
1720 sb->feature_map = 0;
1722 sb->recovery_offset = cpu_to_le64(0);
1723 memset(sb->pad3, 0, sizeof(sb->pad3));
1725 sb->utime = cpu_to_le64((__u64)mddev->utime);
1726 sb->events = cpu_to_le64(mddev->events);
1728 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1729 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1730 sb->resync_offset = cpu_to_le64(MaxSector);
1732 sb->resync_offset = cpu_to_le64(0);
1734 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1736 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1737 sb->size = cpu_to_le64(mddev->dev_sectors);
1738 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1739 sb->level = cpu_to_le32(mddev->level);
1740 sb->layout = cpu_to_le32(mddev->layout);
1741 if (test_bit(FailFast, &rdev->flags))
1742 sb->devflags |= FailFast1;
1744 sb->devflags &= ~FailFast1;
1746 if (test_bit(WriteMostly, &rdev->flags))
1747 sb->devflags |= WriteMostly1;
1749 sb->devflags &= ~WriteMostly1;
1750 sb->data_offset = cpu_to_le64(rdev->data_offset);
1751 sb->data_size = cpu_to_le64(rdev->sectors);
1753 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1754 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1755 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1758 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1759 !test_bit(In_sync, &rdev->flags)) {
1761 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1762 sb->recovery_offset =
1763 cpu_to_le64(rdev->recovery_offset);
1764 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1766 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1768 /* Note: recovery_offset and journal_tail share space */
1769 if (test_bit(Journal, &rdev->flags))
1770 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1771 if (test_bit(Replacement, &rdev->flags))
1773 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1775 if (mddev->reshape_position != MaxSector) {
1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1777 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1778 sb->new_layout = cpu_to_le32(mddev->new_layout);
1779 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1780 sb->new_level = cpu_to_le32(mddev->new_level);
1781 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1782 if (mddev->delta_disks == 0 &&
1783 mddev->reshape_backwards)
1785 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1786 if (rdev->new_data_offset != rdev->data_offset) {
1788 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1789 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1790 - rdev->data_offset));
1794 if (mddev_is_clustered(mddev))
1795 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1797 if (rdev->badblocks.count == 0)
1798 /* Nothing to do for bad blocks*/ ;
1799 else if (sb->bblog_offset == 0)
1800 /* Cannot record bad blocks on this device */
1801 md_error(mddev, rdev);
1803 struct badblocks *bb = &rdev->badblocks;
1804 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1806 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1811 seq = read_seqbegin(&bb->lock);
1813 memset(bbp, 0xff, PAGE_SIZE);
1815 for (i = 0 ; i < bb->count ; i++) {
1816 u64 internal_bb = p[i];
1817 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1818 | BB_LEN(internal_bb));
1819 bbp[i] = cpu_to_le64(store_bb);
1822 if (read_seqretry(&bb->lock, seq))
1825 bb->sector = (rdev->sb_start +
1826 (int)le32_to_cpu(sb->bblog_offset));
1827 bb->size = le16_to_cpu(sb->bblog_size);
1832 rdev_for_each(rdev2, mddev)
1833 if (rdev2->desc_nr+1 > max_dev)
1834 max_dev = rdev2->desc_nr+1;
1836 if (max_dev > le32_to_cpu(sb->max_dev)) {
1838 sb->max_dev = cpu_to_le32(max_dev);
1839 rdev->sb_size = max_dev * 2 + 256;
1840 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1841 if (rdev->sb_size & bmask)
1842 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1844 max_dev = le32_to_cpu(sb->max_dev);
1846 for (i=0; i<max_dev;i++)
1847 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1849 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1850 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1852 rdev_for_each(rdev2, mddev) {
1854 if (test_bit(Faulty, &rdev2->flags))
1855 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1856 else if (test_bit(In_sync, &rdev2->flags))
1857 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1858 else if (test_bit(Journal, &rdev2->flags))
1859 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1860 else if (rdev2->raid_disk >= 0)
1861 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1863 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1866 sb->sb_csum = calc_sb_1_csum(sb);
1869 static unsigned long long
1870 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1872 struct mdp_superblock_1 *sb;
1873 sector_t max_sectors;
1874 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1875 return 0; /* component must fit device */
1876 if (rdev->data_offset != rdev->new_data_offset)
1877 return 0; /* too confusing */
1878 if (rdev->sb_start < rdev->data_offset) {
1879 /* minor versions 1 and 2; superblock before data */
1880 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1881 max_sectors -= rdev->data_offset;
1882 if (!num_sectors || num_sectors > max_sectors)
1883 num_sectors = max_sectors;
1884 } else if (rdev->mddev->bitmap_info.offset) {
1885 /* minor version 0 with bitmap we can't move */
1888 /* minor version 0; superblock after data */
1890 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1891 sb_start &= ~(sector_t)(4*2 - 1);
1892 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1893 if (!num_sectors || num_sectors > max_sectors)
1894 num_sectors = max_sectors;
1895 rdev->sb_start = sb_start;
1897 sb = page_address(rdev->sb_page);
1898 sb->data_size = cpu_to_le64(num_sectors);
1899 sb->super_offset = rdev->sb_start;
1900 sb->sb_csum = calc_sb_1_csum(sb);
1902 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1904 } while (md_super_wait(rdev->mddev) < 0);
1910 super_1_allow_new_offset(struct md_rdev *rdev,
1911 unsigned long long new_offset)
1913 /* All necessary checks on new >= old have been done */
1914 struct bitmap *bitmap;
1915 if (new_offset >= rdev->data_offset)
1918 /* with 1.0 metadata, there is no metadata to tread on
1919 * so we can always move back */
1920 if (rdev->mddev->minor_version == 0)
1923 /* otherwise we must be sure not to step on
1924 * any metadata, so stay:
1925 * 36K beyond start of superblock
1926 * beyond end of badblocks
1927 * beyond write-intent bitmap
1929 if (rdev->sb_start + (32+4)*2 > new_offset)
1931 bitmap = rdev->mddev->bitmap;
1932 if (bitmap && !rdev->mddev->bitmap_info.file &&
1933 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1934 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1936 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1942 static struct super_type super_types[] = {
1945 .owner = THIS_MODULE,
1946 .load_super = super_90_load,
1947 .validate_super = super_90_validate,
1948 .sync_super = super_90_sync,
1949 .rdev_size_change = super_90_rdev_size_change,
1950 .allow_new_offset = super_90_allow_new_offset,
1954 .owner = THIS_MODULE,
1955 .load_super = super_1_load,
1956 .validate_super = super_1_validate,
1957 .sync_super = super_1_sync,
1958 .rdev_size_change = super_1_rdev_size_change,
1959 .allow_new_offset = super_1_allow_new_offset,
1963 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1965 if (mddev->sync_super) {
1966 mddev->sync_super(mddev, rdev);
1970 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1972 super_types[mddev->major_version].sync_super(mddev, rdev);
1975 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1977 struct md_rdev *rdev, *rdev2;
1980 rdev_for_each_rcu(rdev, mddev1) {
1981 if (test_bit(Faulty, &rdev->flags) ||
1982 test_bit(Journal, &rdev->flags) ||
1983 rdev->raid_disk == -1)
1985 rdev_for_each_rcu(rdev2, mddev2) {
1986 if (test_bit(Faulty, &rdev2->flags) ||
1987 test_bit(Journal, &rdev2->flags) ||
1988 rdev2->raid_disk == -1)
1990 if (rdev->bdev->bd_contains ==
1991 rdev2->bdev->bd_contains) {
2001 static LIST_HEAD(pending_raid_disks);
2004 * Try to register data integrity profile for an mddev
2006 * This is called when an array is started and after a disk has been kicked
2007 * from the array. It only succeeds if all working and active component devices
2008 * are integrity capable with matching profiles.
2010 int md_integrity_register(struct mddev *mddev)
2012 struct md_rdev *rdev, *reference = NULL;
2014 if (list_empty(&mddev->disks))
2015 return 0; /* nothing to do */
2016 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2017 return 0; /* shouldn't register, or already is */
2018 rdev_for_each(rdev, mddev) {
2019 /* skip spares and non-functional disks */
2020 if (test_bit(Faulty, &rdev->flags))
2022 if (rdev->raid_disk < 0)
2025 /* Use the first rdev as the reference */
2029 /* does this rdev's profile match the reference profile? */
2030 if (blk_integrity_compare(reference->bdev->bd_disk,
2031 rdev->bdev->bd_disk) < 0)
2034 if (!reference || !bdev_get_integrity(reference->bdev))
2037 * All component devices are integrity capable and have matching
2038 * profiles, register the common profile for the md device.
2040 blk_integrity_register(mddev->gendisk,
2041 bdev_get_integrity(reference->bdev));
2043 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2044 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2045 pr_err("md: failed to create integrity pool for %s\n",
2051 EXPORT_SYMBOL(md_integrity_register);
2054 * Attempt to add an rdev, but only if it is consistent with the current
2057 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2059 struct blk_integrity *bi_rdev;
2060 struct blk_integrity *bi_mddev;
2061 char name[BDEVNAME_SIZE];
2063 if (!mddev->gendisk)
2066 bi_rdev = bdev_get_integrity(rdev->bdev);
2067 bi_mddev = blk_get_integrity(mddev->gendisk);
2069 if (!bi_mddev) /* nothing to do */
2072 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2073 pr_err("%s: incompatible integrity profile for %s\n",
2074 mdname(mddev), bdevname(rdev->bdev, name));
2080 EXPORT_SYMBOL(md_integrity_add_rdev);
2082 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2084 char b[BDEVNAME_SIZE];
2088 /* prevent duplicates */
2089 if (find_rdev(mddev, rdev->bdev->bd_dev))
2092 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2093 if (!test_bit(Journal, &rdev->flags) &&
2095 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2097 /* Cannot change size, so fail
2098 * If mddev->level <= 0, then we don't care
2099 * about aligning sizes (e.g. linear)
2101 if (mddev->level > 0)
2104 mddev->dev_sectors = rdev->sectors;
2107 /* Verify rdev->desc_nr is unique.
2108 * If it is -1, assign a free number, else
2109 * check number is not in use
2112 if (rdev->desc_nr < 0) {
2115 choice = mddev->raid_disks;
2116 while (md_find_rdev_nr_rcu(mddev, choice))
2118 rdev->desc_nr = choice;
2120 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2126 if (!test_bit(Journal, &rdev->flags) &&
2127 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2128 pr_warn("md: %s: array is limited to %d devices\n",
2129 mdname(mddev), mddev->max_disks);
2132 bdevname(rdev->bdev,b);
2133 strreplace(b, '/', '!');
2135 rdev->mddev = mddev;
2136 pr_debug("md: bind<%s>\n", b);
2138 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2141 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2142 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2143 /* failure here is OK */;
2144 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2146 list_add_rcu(&rdev->same_set, &mddev->disks);
2147 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2149 /* May as well allow recovery to be retried once */
2150 mddev->recovery_disabled++;
2155 pr_warn("md: failed to register dev-%s for %s\n",
2160 static void md_delayed_delete(struct work_struct *ws)
2162 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2163 kobject_del(&rdev->kobj);
2164 kobject_put(&rdev->kobj);
2167 static void unbind_rdev_from_array(struct md_rdev *rdev)
2169 char b[BDEVNAME_SIZE];
2171 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2172 list_del_rcu(&rdev->same_set);
2173 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2175 sysfs_remove_link(&rdev->kobj, "block");
2176 sysfs_put(rdev->sysfs_state);
2177 rdev->sysfs_state = NULL;
2178 rdev->badblocks.count = 0;
2179 /* We need to delay this, otherwise we can deadlock when
2180 * writing to 'remove' to "dev/state". We also need
2181 * to delay it due to rcu usage.
2184 INIT_WORK(&rdev->del_work, md_delayed_delete);
2185 kobject_get(&rdev->kobj);
2186 queue_work(md_misc_wq, &rdev->del_work);
2190 * prevent the device from being mounted, repartitioned or
2191 * otherwise reused by a RAID array (or any other kernel
2192 * subsystem), by bd_claiming the device.
2194 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2197 struct block_device *bdev;
2198 char b[BDEVNAME_SIZE];
2200 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2201 shared ? (struct md_rdev *)lock_rdev : rdev);
2203 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2204 return PTR_ERR(bdev);
2210 static void unlock_rdev(struct md_rdev *rdev)
2212 struct block_device *bdev = rdev->bdev;
2214 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2217 void md_autodetect_dev(dev_t dev);
2219 static void export_rdev(struct md_rdev *rdev)
2221 char b[BDEVNAME_SIZE];
2223 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2224 md_rdev_clear(rdev);
2226 if (test_bit(AutoDetected, &rdev->flags))
2227 md_autodetect_dev(rdev->bdev->bd_dev);
2230 kobject_put(&rdev->kobj);
2233 void md_kick_rdev_from_array(struct md_rdev *rdev)
2235 unbind_rdev_from_array(rdev);
2238 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2240 static void export_array(struct mddev *mddev)
2242 struct md_rdev *rdev;
2244 while (!list_empty(&mddev->disks)) {
2245 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2247 md_kick_rdev_from_array(rdev);
2249 mddev->raid_disks = 0;
2250 mddev->major_version = 0;
2253 static void sync_sbs(struct mddev *mddev, int nospares)
2255 /* Update each superblock (in-memory image), but
2256 * if we are allowed to, skip spares which already
2257 * have the right event counter, or have one earlier
2258 * (which would mean they aren't being marked as dirty
2259 * with the rest of the array)
2261 struct md_rdev *rdev;
2262 rdev_for_each(rdev, mddev) {
2263 if (rdev->sb_events == mddev->events ||
2265 rdev->raid_disk < 0 &&
2266 rdev->sb_events+1 == mddev->events)) {
2267 /* Don't update this superblock */
2268 rdev->sb_loaded = 2;
2270 sync_super(mddev, rdev);
2271 rdev->sb_loaded = 1;
2276 static bool does_sb_need_changing(struct mddev *mddev)
2278 struct md_rdev *rdev;
2279 struct mdp_superblock_1 *sb;
2282 /* Find a good rdev */
2283 rdev_for_each(rdev, mddev)
2284 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2287 /* No good device found. */
2291 sb = page_address(rdev->sb_page);
2292 /* Check if a device has become faulty or a spare become active */
2293 rdev_for_each(rdev, mddev) {
2294 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2295 /* Device activated? */
2296 if (role == 0xffff && rdev->raid_disk >=0 &&
2297 !test_bit(Faulty, &rdev->flags))
2299 /* Device turned faulty? */
2300 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2304 /* Check if any mddev parameters have changed */
2305 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2306 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2307 (mddev->layout != le64_to_cpu(sb->layout)) ||
2308 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2309 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2315 void md_update_sb(struct mddev *mddev, int force_change)
2317 struct md_rdev *rdev;
2320 int any_badblocks_changed = 0;
2325 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2330 if (mddev_is_clustered(mddev)) {
2331 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2333 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2335 ret = md_cluster_ops->metadata_update_start(mddev);
2336 /* Has someone else has updated the sb */
2337 if (!does_sb_need_changing(mddev)) {
2339 md_cluster_ops->metadata_update_cancel(mddev);
2340 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2341 BIT(MD_SB_CHANGE_DEVS) |
2342 BIT(MD_SB_CHANGE_CLEAN));
2347 /* First make sure individual recovery_offsets are correct */
2348 rdev_for_each(rdev, mddev) {
2349 if (rdev->raid_disk >= 0 &&
2350 mddev->delta_disks >= 0 &&
2351 !test_bit(Journal, &rdev->flags) &&
2352 !test_bit(In_sync, &rdev->flags) &&
2353 mddev->curr_resync_completed > rdev->recovery_offset)
2354 rdev->recovery_offset = mddev->curr_resync_completed;
2357 if (!mddev->persistent) {
2358 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2359 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2360 if (!mddev->external) {
2361 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2362 rdev_for_each(rdev, mddev) {
2363 if (rdev->badblocks.changed) {
2364 rdev->badblocks.changed = 0;
2365 ack_all_badblocks(&rdev->badblocks);
2366 md_error(mddev, rdev);
2368 clear_bit(Blocked, &rdev->flags);
2369 clear_bit(BlockedBadBlocks, &rdev->flags);
2370 wake_up(&rdev->blocked_wait);
2373 wake_up(&mddev->sb_wait);
2377 spin_lock(&mddev->lock);
2379 mddev->utime = ktime_get_real_seconds();
2381 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2383 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2384 /* just a clean<-> dirty transition, possibly leave spares alone,
2385 * though if events isn't the right even/odd, we will have to do
2391 if (mddev->degraded)
2392 /* If the array is degraded, then skipping spares is both
2393 * dangerous and fairly pointless.
2394 * Dangerous because a device that was removed from the array
2395 * might have a event_count that still looks up-to-date,
2396 * so it can be re-added without a resync.
2397 * Pointless because if there are any spares to skip,
2398 * then a recovery will happen and soon that array won't
2399 * be degraded any more and the spare can go back to sleep then.
2403 sync_req = mddev->in_sync;
2405 /* If this is just a dirty<->clean transition, and the array is clean
2406 * and 'events' is odd, we can roll back to the previous clean state */
2408 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2409 && mddev->can_decrease_events
2410 && mddev->events != 1) {
2412 mddev->can_decrease_events = 0;
2414 /* otherwise we have to go forward and ... */
2416 mddev->can_decrease_events = nospares;
2420 * This 64-bit counter should never wrap.
2421 * Either we are in around ~1 trillion A.C., assuming
2422 * 1 reboot per second, or we have a bug...
2424 WARN_ON(mddev->events == 0);
2426 rdev_for_each(rdev, mddev) {
2427 if (rdev->badblocks.changed)
2428 any_badblocks_changed++;
2429 if (test_bit(Faulty, &rdev->flags))
2430 set_bit(FaultRecorded, &rdev->flags);
2433 sync_sbs(mddev, nospares);
2434 spin_unlock(&mddev->lock);
2436 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2437 mdname(mddev), mddev->in_sync);
2440 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2442 bitmap_update_sb(mddev->bitmap);
2443 rdev_for_each(rdev, mddev) {
2444 char b[BDEVNAME_SIZE];
2446 if (rdev->sb_loaded != 1)
2447 continue; /* no noise on spare devices */
2449 if (!test_bit(Faulty, &rdev->flags)) {
2450 md_super_write(mddev,rdev,
2451 rdev->sb_start, rdev->sb_size,
2453 pr_debug("md: (write) %s's sb offset: %llu\n",
2454 bdevname(rdev->bdev, b),
2455 (unsigned long long)rdev->sb_start);
2456 rdev->sb_events = mddev->events;
2457 if (rdev->badblocks.size) {
2458 md_super_write(mddev, rdev,
2459 rdev->badblocks.sector,
2460 rdev->badblocks.size << 9,
2462 rdev->badblocks.size = 0;
2466 pr_debug("md: %s (skipping faulty)\n",
2467 bdevname(rdev->bdev, b));
2469 if (mddev->level == LEVEL_MULTIPATH)
2470 /* only need to write one superblock... */
2473 if (md_super_wait(mddev) < 0)
2475 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2477 if (mddev_is_clustered(mddev) && ret == 0)
2478 md_cluster_ops->metadata_update_finish(mddev);
2480 if (mddev->in_sync != sync_req ||
2481 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2482 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2483 /* have to write it out again */
2485 wake_up(&mddev->sb_wait);
2486 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2487 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2489 rdev_for_each(rdev, mddev) {
2490 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2491 clear_bit(Blocked, &rdev->flags);
2493 if (any_badblocks_changed)
2494 ack_all_badblocks(&rdev->badblocks);
2495 clear_bit(BlockedBadBlocks, &rdev->flags);
2496 wake_up(&rdev->blocked_wait);
2499 EXPORT_SYMBOL(md_update_sb);
2501 static int add_bound_rdev(struct md_rdev *rdev)
2503 struct mddev *mddev = rdev->mddev;
2505 bool add_journal = test_bit(Journal, &rdev->flags);
2507 if (!mddev->pers->hot_remove_disk || add_journal) {
2508 /* If there is hot_add_disk but no hot_remove_disk
2509 * then added disks for geometry changes,
2510 * and should be added immediately.
2512 super_types[mddev->major_version].
2513 validate_super(mddev, rdev);
2515 mddev_suspend(mddev);
2516 err = mddev->pers->hot_add_disk(mddev, rdev);
2518 mddev_resume(mddev);
2520 md_kick_rdev_from_array(rdev);
2524 sysfs_notify_dirent_safe(rdev->sysfs_state);
2526 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2527 if (mddev->degraded)
2528 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2529 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2530 md_new_event(mddev);
2531 md_wakeup_thread(mddev->thread);
2535 /* words written to sysfs files may, or may not, be \n terminated.
2536 * We want to accept with case. For this we use cmd_match.
2538 static int cmd_match(const char *cmd, const char *str)
2540 /* See if cmd, written into a sysfs file, matches
2541 * str. They must either be the same, or cmd can
2542 * have a trailing newline
2544 while (*cmd && *str && *cmd == *str) {
2555 struct rdev_sysfs_entry {
2556 struct attribute attr;
2557 ssize_t (*show)(struct md_rdev *, char *);
2558 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2562 state_show(struct md_rdev *rdev, char *page)
2566 unsigned long flags = ACCESS_ONCE(rdev->flags);
2568 if (test_bit(Faulty, &flags) ||
2569 (!test_bit(ExternalBbl, &flags) &&
2570 rdev->badblocks.unacked_exist))
2571 len += sprintf(page+len, "faulty%s", sep);
2572 if (test_bit(In_sync, &flags))
2573 len += sprintf(page+len, "in_sync%s", sep);
2574 if (test_bit(Journal, &flags))
2575 len += sprintf(page+len, "journal%s", sep);
2576 if (test_bit(WriteMostly, &flags))
2577 len += sprintf(page+len, "write_mostly%s", sep);
2578 if (test_bit(Blocked, &flags) ||
2579 (rdev->badblocks.unacked_exist
2580 && !test_bit(Faulty, &flags)))
2581 len += sprintf(page+len, "blocked%s", sep);
2582 if (!test_bit(Faulty, &flags) &&
2583 !test_bit(Journal, &flags) &&
2584 !test_bit(In_sync, &flags))
2585 len += sprintf(page+len, "spare%s", sep);
2586 if (test_bit(WriteErrorSeen, &flags))
2587 len += sprintf(page+len, "write_error%s", sep);
2588 if (test_bit(WantReplacement, &flags))
2589 len += sprintf(page+len, "want_replacement%s", sep);
2590 if (test_bit(Replacement, &flags))
2591 len += sprintf(page+len, "replacement%s", sep);
2592 if (test_bit(ExternalBbl, &flags))
2593 len += sprintf(page+len, "external_bbl%s", sep);
2594 if (test_bit(FailFast, &flags))
2595 len += sprintf(page+len, "failfast%s", sep);
2600 return len+sprintf(page+len, "\n");
2604 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2607 * faulty - simulates an error
2608 * remove - disconnects the device
2609 * writemostly - sets write_mostly
2610 * -writemostly - clears write_mostly
2611 * blocked - sets the Blocked flags
2612 * -blocked - clears the Blocked and possibly simulates an error
2613 * insync - sets Insync providing device isn't active
2614 * -insync - clear Insync for a device with a slot assigned,
2615 * so that it gets rebuilt based on bitmap
2616 * write_error - sets WriteErrorSeen
2617 * -write_error - clears WriteErrorSeen
2618 * {,-}failfast - set/clear FailFast
2621 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2622 md_error(rdev->mddev, rdev);
2623 if (test_bit(Faulty, &rdev->flags))
2627 } else if (cmd_match(buf, "remove")) {
2628 if (rdev->mddev->pers) {
2629 clear_bit(Blocked, &rdev->flags);
2630 remove_and_add_spares(rdev->mddev, rdev);
2632 if (rdev->raid_disk >= 0)
2635 struct mddev *mddev = rdev->mddev;
2637 if (mddev_is_clustered(mddev))
2638 err = md_cluster_ops->remove_disk(mddev, rdev);
2641 md_kick_rdev_from_array(rdev);
2643 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2644 md_wakeup_thread(mddev->thread);
2646 md_new_event(mddev);
2649 } else if (cmd_match(buf, "writemostly")) {
2650 set_bit(WriteMostly, &rdev->flags);
2652 } else if (cmd_match(buf, "-writemostly")) {
2653 clear_bit(WriteMostly, &rdev->flags);
2655 } else if (cmd_match(buf, "blocked")) {
2656 set_bit(Blocked, &rdev->flags);
2658 } else if (cmd_match(buf, "-blocked")) {
2659 if (!test_bit(Faulty, &rdev->flags) &&
2660 !test_bit(ExternalBbl, &rdev->flags) &&
2661 rdev->badblocks.unacked_exist) {
2662 /* metadata handler doesn't understand badblocks,
2663 * so we need to fail the device
2665 md_error(rdev->mddev, rdev);
2667 clear_bit(Blocked, &rdev->flags);
2668 clear_bit(BlockedBadBlocks, &rdev->flags);
2669 wake_up(&rdev->blocked_wait);
2670 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2671 md_wakeup_thread(rdev->mddev->thread);
2674 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2675 set_bit(In_sync, &rdev->flags);
2677 } else if (cmd_match(buf, "failfast")) {
2678 set_bit(FailFast, &rdev->flags);
2680 } else if (cmd_match(buf, "-failfast")) {
2681 clear_bit(FailFast, &rdev->flags);
2683 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2684 !test_bit(Journal, &rdev->flags)) {
2685 if (rdev->mddev->pers == NULL) {
2686 clear_bit(In_sync, &rdev->flags);
2687 rdev->saved_raid_disk = rdev->raid_disk;
2688 rdev->raid_disk = -1;
2691 } else if (cmd_match(buf, "write_error")) {
2692 set_bit(WriteErrorSeen, &rdev->flags);
2694 } else if (cmd_match(buf, "-write_error")) {
2695 clear_bit(WriteErrorSeen, &rdev->flags);
2697 } else if (cmd_match(buf, "want_replacement")) {
2698 /* Any non-spare device that is not a replacement can
2699 * become want_replacement at any time, but we then need to
2700 * check if recovery is needed.
2702 if (rdev->raid_disk >= 0 &&
2703 !test_bit(Journal, &rdev->flags) &&
2704 !test_bit(Replacement, &rdev->flags))
2705 set_bit(WantReplacement, &rdev->flags);
2706 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2707 md_wakeup_thread(rdev->mddev->thread);
2709 } else if (cmd_match(buf, "-want_replacement")) {
2710 /* Clearing 'want_replacement' is always allowed.
2711 * Once replacements starts it is too late though.
2714 clear_bit(WantReplacement, &rdev->flags);
2715 } else if (cmd_match(buf, "replacement")) {
2716 /* Can only set a device as a replacement when array has not
2717 * yet been started. Once running, replacement is automatic
2718 * from spares, or by assigning 'slot'.
2720 if (rdev->mddev->pers)
2723 set_bit(Replacement, &rdev->flags);
2726 } else if (cmd_match(buf, "-replacement")) {
2727 /* Similarly, can only clear Replacement before start */
2728 if (rdev->mddev->pers)
2731 clear_bit(Replacement, &rdev->flags);
2734 } else if (cmd_match(buf, "re-add")) {
2735 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2736 /* clear_bit is performed _after_ all the devices
2737 * have their local Faulty bit cleared. If any writes
2738 * happen in the meantime in the local node, they
2739 * will land in the local bitmap, which will be synced
2740 * by this node eventually
2742 if (!mddev_is_clustered(rdev->mddev) ||
2743 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2744 clear_bit(Faulty, &rdev->flags);
2745 err = add_bound_rdev(rdev);
2749 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2750 set_bit(ExternalBbl, &rdev->flags);
2751 rdev->badblocks.shift = 0;
2753 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2754 clear_bit(ExternalBbl, &rdev->flags);
2758 sysfs_notify_dirent_safe(rdev->sysfs_state);
2759 return err ? err : len;
2761 static struct rdev_sysfs_entry rdev_state =
2762 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2765 errors_show(struct md_rdev *rdev, char *page)
2767 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2771 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2776 rv = kstrtouint(buf, 10, &n);
2779 atomic_set(&rdev->corrected_errors, n);
2782 static struct rdev_sysfs_entry rdev_errors =
2783 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2786 slot_show(struct md_rdev *rdev, char *page)
2788 if (test_bit(Journal, &rdev->flags))
2789 return sprintf(page, "journal\n");
2790 else if (rdev->raid_disk < 0)
2791 return sprintf(page, "none\n");
2793 return sprintf(page, "%d\n", rdev->raid_disk);
2797 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2802 if (test_bit(Journal, &rdev->flags))
2804 if (strncmp(buf, "none", 4)==0)
2807 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2811 if (rdev->mddev->pers && slot == -1) {
2812 /* Setting 'slot' on an active array requires also
2813 * updating the 'rd%d' link, and communicating
2814 * with the personality with ->hot_*_disk.
2815 * For now we only support removing
2816 * failed/spare devices. This normally happens automatically,
2817 * but not when the metadata is externally managed.
2819 if (rdev->raid_disk == -1)
2821 /* personality does all needed checks */
2822 if (rdev->mddev->pers->hot_remove_disk == NULL)
2824 clear_bit(Blocked, &rdev->flags);
2825 remove_and_add_spares(rdev->mddev, rdev);
2826 if (rdev->raid_disk >= 0)
2828 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2829 md_wakeup_thread(rdev->mddev->thread);
2830 } else if (rdev->mddev->pers) {
2831 /* Activating a spare .. or possibly reactivating
2832 * if we ever get bitmaps working here.
2836 if (rdev->raid_disk != -1)
2839 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2842 if (rdev->mddev->pers->hot_add_disk == NULL)
2845 if (slot >= rdev->mddev->raid_disks &&
2846 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2849 rdev->raid_disk = slot;
2850 if (test_bit(In_sync, &rdev->flags))
2851 rdev->saved_raid_disk = slot;
2853 rdev->saved_raid_disk = -1;
2854 clear_bit(In_sync, &rdev->flags);
2855 clear_bit(Bitmap_sync, &rdev->flags);
2856 err = rdev->mddev->pers->
2857 hot_add_disk(rdev->mddev, rdev);
2859 rdev->raid_disk = -1;
2862 sysfs_notify_dirent_safe(rdev->sysfs_state);
2863 if (sysfs_link_rdev(rdev->mddev, rdev))
2864 /* failure here is OK */;
2865 /* don't wakeup anyone, leave that to userspace. */
2867 if (slot >= rdev->mddev->raid_disks &&
2868 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2870 rdev->raid_disk = slot;
2871 /* assume it is working */
2872 clear_bit(Faulty, &rdev->flags);
2873 clear_bit(WriteMostly, &rdev->flags);
2874 set_bit(In_sync, &rdev->flags);
2875 sysfs_notify_dirent_safe(rdev->sysfs_state);
2880 static struct rdev_sysfs_entry rdev_slot =
2881 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2884 offset_show(struct md_rdev *rdev, char *page)
2886 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2890 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2892 unsigned long long offset;
2893 if (kstrtoull(buf, 10, &offset) < 0)
2895 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2897 if (rdev->sectors && rdev->mddev->external)
2898 /* Must set offset before size, so overlap checks
2901 rdev->data_offset = offset;
2902 rdev->new_data_offset = offset;
2906 static struct rdev_sysfs_entry rdev_offset =
2907 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2909 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2911 return sprintf(page, "%llu\n",
2912 (unsigned long long)rdev->new_data_offset);
2915 static ssize_t new_offset_store(struct md_rdev *rdev,
2916 const char *buf, size_t len)
2918 unsigned long long new_offset;
2919 struct mddev *mddev = rdev->mddev;
2921 if (kstrtoull(buf, 10, &new_offset) < 0)
2924 if (mddev->sync_thread ||
2925 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2927 if (new_offset == rdev->data_offset)
2928 /* reset is always permitted */
2930 else if (new_offset > rdev->data_offset) {
2931 /* must not push array size beyond rdev_sectors */
2932 if (new_offset - rdev->data_offset
2933 + mddev->dev_sectors > rdev->sectors)
2936 /* Metadata worries about other space details. */
2938 /* decreasing the offset is inconsistent with a backwards
2941 if (new_offset < rdev->data_offset &&
2942 mddev->reshape_backwards)
2944 /* Increasing offset is inconsistent with forwards
2945 * reshape. reshape_direction should be set to
2946 * 'backwards' first.
2948 if (new_offset > rdev->data_offset &&
2949 !mddev->reshape_backwards)
2952 if (mddev->pers && mddev->persistent &&
2953 !super_types[mddev->major_version]
2954 .allow_new_offset(rdev, new_offset))
2956 rdev->new_data_offset = new_offset;
2957 if (new_offset > rdev->data_offset)
2958 mddev->reshape_backwards = 1;
2959 else if (new_offset < rdev->data_offset)
2960 mddev->reshape_backwards = 0;
2964 static struct rdev_sysfs_entry rdev_new_offset =
2965 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2968 rdev_size_show(struct md_rdev *rdev, char *page)
2970 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2973 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2975 /* check if two start/length pairs overlap */
2983 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2985 unsigned long long blocks;
2988 if (kstrtoull(buf, 10, &blocks) < 0)
2991 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2992 return -EINVAL; /* sector conversion overflow */
2995 if (new != blocks * 2)
2996 return -EINVAL; /* unsigned long long to sector_t overflow */
3003 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3005 struct mddev *my_mddev = rdev->mddev;
3006 sector_t oldsectors = rdev->sectors;
3009 if (test_bit(Journal, &rdev->flags))
3011 if (strict_blocks_to_sectors(buf, §ors) < 0)
3013 if (rdev->data_offset != rdev->new_data_offset)
3014 return -EINVAL; /* too confusing */
3015 if (my_mddev->pers && rdev->raid_disk >= 0) {
3016 if (my_mddev->persistent) {
3017 sectors = super_types[my_mddev->major_version].
3018 rdev_size_change(rdev, sectors);
3021 } else if (!sectors)
3022 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3024 if (!my_mddev->pers->resize)
3025 /* Cannot change size for RAID0 or Linear etc */
3028 if (sectors < my_mddev->dev_sectors)
3029 return -EINVAL; /* component must fit device */
3031 rdev->sectors = sectors;
3032 if (sectors > oldsectors && my_mddev->external) {
3033 /* Need to check that all other rdevs with the same
3034 * ->bdev do not overlap. 'rcu' is sufficient to walk
3035 * the rdev lists safely.
3036 * This check does not provide a hard guarantee, it
3037 * just helps avoid dangerous mistakes.
3039 struct mddev *mddev;
3041 struct list_head *tmp;
3044 for_each_mddev(mddev, tmp) {
3045 struct md_rdev *rdev2;
3047 rdev_for_each(rdev2, mddev)
3048 if (rdev->bdev == rdev2->bdev &&
3050 overlaps(rdev->data_offset, rdev->sectors,
3063 /* Someone else could have slipped in a size
3064 * change here, but doing so is just silly.
3065 * We put oldsectors back because we *know* it is
3066 * safe, and trust userspace not to race with
3069 rdev->sectors = oldsectors;
3076 static struct rdev_sysfs_entry rdev_size =
3077 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3079 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3081 unsigned long long recovery_start = rdev->recovery_offset;
3083 if (test_bit(In_sync, &rdev->flags) ||
3084 recovery_start == MaxSector)
3085 return sprintf(page, "none\n");
3087 return sprintf(page, "%llu\n", recovery_start);
3090 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3092 unsigned long long recovery_start;
3094 if (cmd_match(buf, "none"))
3095 recovery_start = MaxSector;
3096 else if (kstrtoull(buf, 10, &recovery_start))
3099 if (rdev->mddev->pers &&
3100 rdev->raid_disk >= 0)
3103 rdev->recovery_offset = recovery_start;
3104 if (recovery_start == MaxSector)
3105 set_bit(In_sync, &rdev->flags);
3107 clear_bit(In_sync, &rdev->flags);
3111 static struct rdev_sysfs_entry rdev_recovery_start =
3112 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3114 /* sysfs access to bad-blocks list.
3115 * We present two files.
3116 * 'bad-blocks' lists sector numbers and lengths of ranges that
3117 * are recorded as bad. The list is truncated to fit within
3118 * the one-page limit of sysfs.
3119 * Writing "sector length" to this file adds an acknowledged
3121 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3122 * been acknowledged. Writing to this file adds bad blocks
3123 * without acknowledging them. This is largely for testing.
3125 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3127 return badblocks_show(&rdev->badblocks, page, 0);
3129 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3131 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3132 /* Maybe that ack was all we needed */
3133 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3134 wake_up(&rdev->blocked_wait);
3137 static struct rdev_sysfs_entry rdev_bad_blocks =
3138 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3140 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3142 return badblocks_show(&rdev->badblocks, page, 1);
3144 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3146 return badblocks_store(&rdev->badblocks, page, len, 1);
3148 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3149 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3151 static struct attribute *rdev_default_attrs[] = {
3156 &rdev_new_offset.attr,
3158 &rdev_recovery_start.attr,
3159 &rdev_bad_blocks.attr,
3160 &rdev_unack_bad_blocks.attr,
3164 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3166 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3167 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3173 return entry->show(rdev, page);
3177 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3178 const char *page, size_t length)
3180 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3181 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3183 struct mddev *mddev = rdev->mddev;
3187 if (!capable(CAP_SYS_ADMIN))
3189 rv = mddev ? mddev_lock(mddev): -EBUSY;
3191 if (rdev->mddev == NULL)
3194 rv = entry->store(rdev, page, length);
3195 mddev_unlock(mddev);
3200 static void rdev_free(struct kobject *ko)
3202 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3205 static const struct sysfs_ops rdev_sysfs_ops = {
3206 .show = rdev_attr_show,
3207 .store = rdev_attr_store,
3209 static struct kobj_type rdev_ktype = {
3210 .release = rdev_free,
3211 .sysfs_ops = &rdev_sysfs_ops,
3212 .default_attrs = rdev_default_attrs,
3215 int md_rdev_init(struct md_rdev *rdev)
3218 rdev->saved_raid_disk = -1;
3219 rdev->raid_disk = -1;
3221 rdev->data_offset = 0;
3222 rdev->new_data_offset = 0;
3223 rdev->sb_events = 0;
3224 rdev->last_read_error = 0;
3225 rdev->sb_loaded = 0;
3226 rdev->bb_page = NULL;
3227 atomic_set(&rdev->nr_pending, 0);
3228 atomic_set(&rdev->read_errors, 0);
3229 atomic_set(&rdev->corrected_errors, 0);
3231 INIT_LIST_HEAD(&rdev->same_set);
3232 init_waitqueue_head(&rdev->blocked_wait);
3234 /* Add space to store bad block list.
3235 * This reserves the space even on arrays where it cannot
3236 * be used - I wonder if that matters
3238 return badblocks_init(&rdev->badblocks, 0);
3240 EXPORT_SYMBOL_GPL(md_rdev_init);
3242 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3244 * mark the device faulty if:
3246 * - the device is nonexistent (zero size)
3247 * - the device has no valid superblock
3249 * a faulty rdev _never_ has rdev->sb set.
3251 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3253 char b[BDEVNAME_SIZE];
3255 struct md_rdev *rdev;
3258 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3260 return ERR_PTR(-ENOMEM);
3262 err = md_rdev_init(rdev);
3265 err = alloc_disk_sb(rdev);
3269 err = lock_rdev(rdev, newdev, super_format == -2);
3273 kobject_init(&rdev->kobj, &rdev_ktype);
3275 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3277 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3278 bdevname(rdev->bdev,b));
3283 if (super_format >= 0) {
3284 err = super_types[super_format].
3285 load_super(rdev, NULL, super_minor);
3286 if (err == -EINVAL) {
3287 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3288 bdevname(rdev->bdev,b),
3289 super_format, super_minor);
3293 pr_warn("md: could not read %s's sb, not importing!\n",
3294 bdevname(rdev->bdev,b));
3304 md_rdev_clear(rdev);
3306 return ERR_PTR(err);
3310 * Check a full RAID array for plausibility
3313 static void analyze_sbs(struct mddev *mddev)
3316 struct md_rdev *rdev, *freshest, *tmp;
3317 char b[BDEVNAME_SIZE];
3320 rdev_for_each_safe(rdev, tmp, mddev)
3321 switch (super_types[mddev->major_version].
3322 load_super(rdev, freshest, mddev->minor_version)) {
3329 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3330 bdevname(rdev->bdev,b));
3331 md_kick_rdev_from_array(rdev);
3334 super_types[mddev->major_version].
3335 validate_super(mddev, freshest);
3338 rdev_for_each_safe(rdev, tmp, mddev) {
3339 if (mddev->max_disks &&
3340 (rdev->desc_nr >= mddev->max_disks ||
3341 i > mddev->max_disks)) {
3342 pr_warn("md: %s: %s: only %d devices permitted\n",
3343 mdname(mddev), bdevname(rdev->bdev, b),
3345 md_kick_rdev_from_array(rdev);
3348 if (rdev != freshest) {
3349 if (super_types[mddev->major_version].
3350 validate_super(mddev, rdev)) {
3351 pr_warn("md: kicking non-fresh %s from array!\n",
3352 bdevname(rdev->bdev,b));
3353 md_kick_rdev_from_array(rdev);
3357 if (mddev->level == LEVEL_MULTIPATH) {
3358 rdev->desc_nr = i++;
3359 rdev->raid_disk = rdev->desc_nr;
3360 set_bit(In_sync, &rdev->flags);
3361 } else if (rdev->raid_disk >=
3362 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3363 !test_bit(Journal, &rdev->flags)) {
3364 rdev->raid_disk = -1;
3365 clear_bit(In_sync, &rdev->flags);
3370 /* Read a fixed-point number.
3371 * Numbers in sysfs attributes should be in "standard" units where
3372 * possible, so time should be in seconds.
3373 * However we internally use a a much smaller unit such as
3374 * milliseconds or jiffies.
3375 * This function takes a decimal number with a possible fractional
3376 * component, and produces an integer which is the result of
3377 * multiplying that number by 10^'scale'.
3378 * all without any floating-point arithmetic.
3380 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3382 unsigned long result = 0;
3384 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3387 else if (decimals < scale) {
3390 result = result * 10 + value;
3402 while (decimals < scale) {
3411 safe_delay_show(struct mddev *mddev, char *page)
3413 int msec = (mddev->safemode_delay*1000)/HZ;
3414 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3417 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3421 if (mddev_is_clustered(mddev)) {
3422 pr_warn("md: Safemode is disabled for clustered mode\n");
3426 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3429 mddev->safemode_delay = 0;
3431 unsigned long old_delay = mddev->safemode_delay;
3432 unsigned long new_delay = (msec*HZ)/1000;
3436 mddev->safemode_delay = new_delay;
3437 if (new_delay < old_delay || old_delay == 0)
3438 mod_timer(&mddev->safemode_timer, jiffies+1);
3442 static struct md_sysfs_entry md_safe_delay =
3443 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3446 level_show(struct mddev *mddev, char *page)
3448 struct md_personality *p;
3450 spin_lock(&mddev->lock);
3453 ret = sprintf(page, "%s\n", p->name);
3454 else if (mddev->clevel[0])
3455 ret = sprintf(page, "%s\n", mddev->clevel);
3456 else if (mddev->level != LEVEL_NONE)
3457 ret = sprintf(page, "%d\n", mddev->level);
3460 spin_unlock(&mddev->lock);
3465 level_store(struct mddev *mddev, const char *buf, size_t len)
3470 struct md_personality *pers, *oldpers;
3472 void *priv, *oldpriv;
3473 struct md_rdev *rdev;
3475 if (slen == 0 || slen >= sizeof(clevel))
3478 rv = mddev_lock(mddev);
3482 if (mddev->pers == NULL) {
3483 strncpy(mddev->clevel, buf, slen);
3484 if (mddev->clevel[slen-1] == '\n')
3486 mddev->clevel[slen] = 0;
3487 mddev->level = LEVEL_NONE;
3495 /* request to change the personality. Need to ensure:
3496 * - array is not engaged in resync/recovery/reshape
3497 * - old personality can be suspended
3498 * - new personality will access other array.
3502 if (mddev->sync_thread ||
3503 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3504 mddev->reshape_position != MaxSector ||
3505 mddev->sysfs_active)
3509 if (!mddev->pers->quiesce) {
3510 pr_warn("md: %s: %s does not support online personality change\n",
3511 mdname(mddev), mddev->pers->name);
3515 /* Now find the new personality */
3516 strncpy(clevel, buf, slen);
3517 if (clevel[slen-1] == '\n')
3520 if (kstrtol(clevel, 10, &level))
3523 if (request_module("md-%s", clevel) != 0)
3524 request_module("md-level-%s", clevel);
3525 spin_lock(&pers_lock);
3526 pers = find_pers(level, clevel);
3527 if (!pers || !try_module_get(pers->owner)) {
3528 spin_unlock(&pers_lock);
3529 pr_warn("md: personality %s not loaded\n", clevel);
3533 spin_unlock(&pers_lock);
3535 if (pers == mddev->pers) {
3536 /* Nothing to do! */
3537 module_put(pers->owner);
3541 if (!pers->takeover) {
3542 module_put(pers->owner);
3543 pr_warn("md: %s: %s does not support personality takeover\n",
3544 mdname(mddev), clevel);
3549 rdev_for_each(rdev, mddev)
3550 rdev->new_raid_disk = rdev->raid_disk;
3552 /* ->takeover must set new_* and/or delta_disks
3553 * if it succeeds, and may set them when it fails.
3555 priv = pers->takeover(mddev);
3557 mddev->new_level = mddev->level;
3558 mddev->new_layout = mddev->layout;
3559 mddev->new_chunk_sectors = mddev->chunk_sectors;
3560 mddev->raid_disks -= mddev->delta_disks;
3561 mddev->delta_disks = 0;
3562 mddev->reshape_backwards = 0;
3563 module_put(pers->owner);
3564 pr_warn("md: %s: %s would not accept array\n",
3565 mdname(mddev), clevel);
3570 /* Looks like we have a winner */
3571 mddev_suspend(mddev);
3572 mddev_detach(mddev);
3574 spin_lock(&mddev->lock);
3575 oldpers = mddev->pers;
3576 oldpriv = mddev->private;
3578 mddev->private = priv;
3579 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3580 mddev->level = mddev->new_level;
3581 mddev->layout = mddev->new_layout;
3582 mddev->chunk_sectors = mddev->new_chunk_sectors;
3583 mddev->delta_disks = 0;
3584 mddev->reshape_backwards = 0;
3585 mddev->degraded = 0;
3586 spin_unlock(&mddev->lock);
3588 if (oldpers->sync_request == NULL &&
3590 /* We are converting from a no-redundancy array
3591 * to a redundancy array and metadata is managed
3592 * externally so we need to be sure that writes
3593 * won't block due to a need to transition
3595 * until external management is started.
3598 mddev->safemode_delay = 0;
3599 mddev->safemode = 0;
3602 oldpers->free(mddev, oldpriv);
3604 if (oldpers->sync_request == NULL &&
3605 pers->sync_request != NULL) {
3606 /* need to add the md_redundancy_group */
3607 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3608 pr_warn("md: cannot register extra attributes for %s\n",
3610 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3612 if (oldpers->sync_request != NULL &&
3613 pers->sync_request == NULL) {
3614 /* need to remove the md_redundancy_group */
3615 if (mddev->to_remove == NULL)
3616 mddev->to_remove = &md_redundancy_group;
3619 module_put(oldpers->owner);
3621 rdev_for_each(rdev, mddev) {
3622 if (rdev->raid_disk < 0)
3624 if (rdev->new_raid_disk >= mddev->raid_disks)
3625 rdev->new_raid_disk = -1;
3626 if (rdev->new_raid_disk == rdev->raid_disk)
3628 sysfs_unlink_rdev(mddev, rdev);
3630 rdev_for_each(rdev, mddev) {
3631 if (rdev->raid_disk < 0)
3633 if (rdev->new_raid_disk == rdev->raid_disk)
3635 rdev->raid_disk = rdev->new_raid_disk;
3636 if (rdev->raid_disk < 0)
3637 clear_bit(In_sync, &rdev->flags);
3639 if (sysfs_link_rdev(mddev, rdev))
3640 pr_warn("md: cannot register rd%d for %s after level change\n",
3641 rdev->raid_disk, mdname(mddev));
3645 if (pers->sync_request == NULL) {
3646 /* this is now an array without redundancy, so
3647 * it must always be in_sync
3650 del_timer_sync(&mddev->safemode_timer);
3652 blk_set_stacking_limits(&mddev->queue->limits);
3654 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3655 mddev_resume(mddev);
3657 md_update_sb(mddev, 1);
3658 sysfs_notify(&mddev->kobj, NULL, "level");
3659 md_new_event(mddev);
3662 mddev_unlock(mddev);
3666 static struct md_sysfs_entry md_level =
3667 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3670 layout_show(struct mddev *mddev, char *page)
3672 /* just a number, not meaningful for all levels */
3673 if (mddev->reshape_position != MaxSector &&
3674 mddev->layout != mddev->new_layout)
3675 return sprintf(page, "%d (%d)\n",
3676 mddev->new_layout, mddev->layout);
3677 return sprintf(page, "%d\n", mddev->layout);
3681 layout_store(struct mddev *mddev, const char *buf, size_t len)
3686 err = kstrtouint(buf, 10, &n);
3689 err = mddev_lock(mddev);
3694 if (mddev->pers->check_reshape == NULL)
3699 mddev->new_layout = n;
3700 err = mddev->pers->check_reshape(mddev);
3702 mddev->new_layout = mddev->layout;
3705 mddev->new_layout = n;
3706 if (mddev->reshape_position == MaxSector)
3709 mddev_unlock(mddev);
3712 static struct md_sysfs_entry md_layout =
3713 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3716 raid_disks_show(struct mddev *mddev, char *page)
3718 if (mddev->raid_disks == 0)
3720 if (mddev->reshape_position != MaxSector &&
3721 mddev->delta_disks != 0)
3722 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3723 mddev->raid_disks - mddev->delta_disks);
3724 return sprintf(page, "%d\n", mddev->raid_disks);
3727 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3730 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3735 err = kstrtouint(buf, 10, &n);
3739 err = mddev_lock(mddev);
3743 err = update_raid_disks(mddev, n);
3744 else if (mddev->reshape_position != MaxSector) {
3745 struct md_rdev *rdev;
3746 int olddisks = mddev->raid_disks - mddev->delta_disks;
3749 rdev_for_each(rdev, mddev) {
3751 rdev->data_offset < rdev->new_data_offset)
3754 rdev->data_offset > rdev->new_data_offset)
3758 mddev->delta_disks = n - olddisks;
3759 mddev->raid_disks = n;
3760 mddev->reshape_backwards = (mddev->delta_disks < 0);
3762 mddev->raid_disks = n;
3764 mddev_unlock(mddev);
3765 return err ? err : len;
3767 static struct md_sysfs_entry md_raid_disks =
3768 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3771 chunk_size_show(struct mddev *mddev, char *page)
3773 if (mddev->reshape_position != MaxSector &&
3774 mddev->chunk_sectors != mddev->new_chunk_sectors)
3775 return sprintf(page, "%d (%d)\n",
3776 mddev->new_chunk_sectors << 9,
3777 mddev->chunk_sectors << 9);
3778 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3782 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3787 err = kstrtoul(buf, 10, &n);
3791 err = mddev_lock(mddev);
3795 if (mddev->pers->check_reshape == NULL)
3800 mddev->new_chunk_sectors = n >> 9;
3801 err = mddev->pers->check_reshape(mddev);
3803 mddev->new_chunk_sectors = mddev->chunk_sectors;
3806 mddev->new_chunk_sectors = n >> 9;
3807 if (mddev->reshape_position == MaxSector)
3808 mddev->chunk_sectors = n >> 9;
3810 mddev_unlock(mddev);
3813 static struct md_sysfs_entry md_chunk_size =
3814 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3817 resync_start_show(struct mddev *mddev, char *page)
3819 if (mddev->recovery_cp == MaxSector)
3820 return sprintf(page, "none\n");
3821 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3825 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3827 unsigned long long n;
3830 if (cmd_match(buf, "none"))
3833 err = kstrtoull(buf, 10, &n);
3836 if (n != (sector_t)n)
3840 err = mddev_lock(mddev);
3843 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3847 mddev->recovery_cp = n;
3849 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3851 mddev_unlock(mddev);
3854 static struct md_sysfs_entry md_resync_start =
3855 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3856 resync_start_show, resync_start_store);
3859 * The array state can be:
3862 * No devices, no size, no level
3863 * Equivalent to STOP_ARRAY ioctl
3865 * May have some settings, but array is not active
3866 * all IO results in error
3867 * When written, doesn't tear down array, but just stops it
3868 * suspended (not supported yet)
3869 * All IO requests will block. The array can be reconfigured.
3870 * Writing this, if accepted, will block until array is quiescent
3872 * no resync can happen. no superblocks get written.
3873 * write requests fail
3875 * like readonly, but behaves like 'clean' on a write request.
3877 * clean - no pending writes, but otherwise active.
3878 * When written to inactive array, starts without resync
3879 * If a write request arrives then
3880 * if metadata is known, mark 'dirty' and switch to 'active'.
3881 * if not known, block and switch to write-pending
3882 * If written to an active array that has pending writes, then fails.
3884 * fully active: IO and resync can be happening.
3885 * When written to inactive array, starts with resync
3888 * clean, but writes are blocked waiting for 'active' to be written.
3891 * like active, but no writes have been seen for a while (100msec).
3894 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3895 write_pending, active_idle, bad_word};
3896 static char *array_states[] = {
3897 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3898 "write-pending", "active-idle", NULL };
3900 static int match_word(const char *word, char **list)
3903 for (n=0; list[n]; n++)
3904 if (cmd_match(word, list[n]))
3910 array_state_show(struct mddev *mddev, char *page)
3912 enum array_state st = inactive;
3923 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
3925 else if (mddev->in_sync)
3927 else if (mddev->safemode)
3933 if (list_empty(&mddev->disks) &&
3934 mddev->raid_disks == 0 &&
3935 mddev->dev_sectors == 0)
3940 return sprintf(page, "%s\n", array_states[st]);
3943 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3944 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3945 static int do_md_run(struct mddev *mddev);
3946 static int restart_array(struct mddev *mddev);
3949 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3952 enum array_state st = match_word(buf, array_states);
3954 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3955 /* don't take reconfig_mutex when toggling between
3958 spin_lock(&mddev->lock);
3960 restart_array(mddev);
3961 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
3962 md_wakeup_thread(mddev->thread);
3963 wake_up(&mddev->sb_wait);
3965 } else /* st == clean */ {
3966 restart_array(mddev);
3967 if (atomic_read(&mddev->writes_pending) == 0) {
3968 if (mddev->in_sync == 0) {
3970 if (mddev->safemode == 1)
3971 mddev->safemode = 0;
3972 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3979 sysfs_notify_dirent_safe(mddev->sysfs_state);
3980 spin_unlock(&mddev->lock);
3983 err = mddev_lock(mddev);
3991 /* stopping an active array */
3992 err = do_md_stop(mddev, 0, NULL);
3995 /* stopping an active array */
3997 err = do_md_stop(mddev, 2, NULL);
3999 err = 0; /* already inactive */
4002 break; /* not supported yet */
4005 err = md_set_readonly(mddev, NULL);
4008 set_disk_ro(mddev->gendisk, 1);
4009 err = do_md_run(mddev);
4015 err = md_set_readonly(mddev, NULL);
4016 else if (mddev->ro == 1)
4017 err = restart_array(mddev);
4020 set_disk_ro(mddev->gendisk, 0);
4024 err = do_md_run(mddev);
4029 err = restart_array(mddev);
4032 spin_lock(&mddev->lock);
4033 if (atomic_read(&mddev->writes_pending) == 0) {
4034 if (mddev->in_sync == 0) {
4036 if (mddev->safemode == 1)
4037 mddev->safemode = 0;
4038 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4043 spin_unlock(&mddev->lock);
4049 err = restart_array(mddev);
4052 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4053 wake_up(&mddev->sb_wait);
4057 set_disk_ro(mddev->gendisk, 0);
4058 err = do_md_run(mddev);
4063 /* these cannot be set */
4068 if (mddev->hold_active == UNTIL_IOCTL)
4069 mddev->hold_active = 0;
4070 sysfs_notify_dirent_safe(mddev->sysfs_state);
4072 mddev_unlock(mddev);
4075 static struct md_sysfs_entry md_array_state =
4076 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4079 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4080 return sprintf(page, "%d\n",
4081 atomic_read(&mddev->max_corr_read_errors));
4085 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4090 rv = kstrtouint(buf, 10, &n);
4093 atomic_set(&mddev->max_corr_read_errors, n);
4097 static struct md_sysfs_entry max_corr_read_errors =
4098 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4099 max_corrected_read_errors_store);
4102 null_show(struct mddev *mddev, char *page)
4108 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4110 /* buf must be %d:%d\n? giving major and minor numbers */
4111 /* The new device is added to the array.
4112 * If the array has a persistent superblock, we read the
4113 * superblock to initialise info and check validity.
4114 * Otherwise, only checking done is that in bind_rdev_to_array,
4115 * which mainly checks size.
4118 int major = simple_strtoul(buf, &e, 10);
4121 struct md_rdev *rdev;
4124 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4126 minor = simple_strtoul(e+1, &e, 10);
4127 if (*e && *e != '\n')
4129 dev = MKDEV(major, minor);
4130 if (major != MAJOR(dev) ||
4131 minor != MINOR(dev))
4134 flush_workqueue(md_misc_wq);
4136 err = mddev_lock(mddev);
4139 if (mddev->persistent) {
4140 rdev = md_import_device(dev, mddev->major_version,
4141 mddev->minor_version);
4142 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4143 struct md_rdev *rdev0
4144 = list_entry(mddev->disks.next,
4145 struct md_rdev, same_set);
4146 err = super_types[mddev->major_version]
4147 .load_super(rdev, rdev0, mddev->minor_version);
4151 } else if (mddev->external)
4152 rdev = md_import_device(dev, -2, -1);
4154 rdev = md_import_device(dev, -1, -1);
4157 mddev_unlock(mddev);
4158 return PTR_ERR(rdev);
4160 err = bind_rdev_to_array(rdev, mddev);
4164 mddev_unlock(mddev);
4165 return err ? err : len;
4168 static struct md_sysfs_entry md_new_device =
4169 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4172 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4175 unsigned long chunk, end_chunk;
4178 err = mddev_lock(mddev);
4183 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4185 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4186 if (buf == end) break;
4187 if (*end == '-') { /* range */
4189 end_chunk = simple_strtoul(buf, &end, 0);
4190 if (buf == end) break;
4192 if (*end && !isspace(*end)) break;
4193 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4194 buf = skip_spaces(end);
4196 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4198 mddev_unlock(mddev);
4202 static struct md_sysfs_entry md_bitmap =
4203 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4206 size_show(struct mddev *mddev, char *page)
4208 return sprintf(page, "%llu\n",
4209 (unsigned long long)mddev->dev_sectors / 2);
4212 static int update_size(struct mddev *mddev, sector_t num_sectors);
4215 size_store(struct mddev *mddev, const char *buf, size_t len)
4217 /* If array is inactive, we can reduce the component size, but
4218 * not increase it (except from 0).
4219 * If array is active, we can try an on-line resize
4222 int err = strict_blocks_to_sectors(buf, §ors);
4226 err = mddev_lock(mddev);
4230 err = update_size(mddev, sectors);
4232 md_update_sb(mddev, 1);
4234 if (mddev->dev_sectors == 0 ||
4235 mddev->dev_sectors > sectors)
4236 mddev->dev_sectors = sectors;
4240 mddev_unlock(mddev);
4241 return err ? err : len;
4244 static struct md_sysfs_entry md_size =
4245 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4247 /* Metadata version.
4249 * 'none' for arrays with no metadata (good luck...)
4250 * 'external' for arrays with externally managed metadata,
4251 * or N.M for internally known formats
4254 metadata_show(struct mddev *mddev, char *page)
4256 if (mddev->persistent)
4257 return sprintf(page, "%d.%d\n",
4258 mddev->major_version, mddev->minor_version);
4259 else if (mddev->external)
4260 return sprintf(page, "external:%s\n", mddev->metadata_type);
4262 return sprintf(page, "none\n");
4266 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4271 /* Changing the details of 'external' metadata is
4272 * always permitted. Otherwise there must be
4273 * no devices attached to the array.
4276 err = mddev_lock(mddev);
4280 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4282 else if (!list_empty(&mddev->disks))
4286 if (cmd_match(buf, "none")) {
4287 mddev->persistent = 0;
4288 mddev->external = 0;
4289 mddev->major_version = 0;
4290 mddev->minor_version = 90;
4293 if (strncmp(buf, "external:", 9) == 0) {
4294 size_t namelen = len-9;
4295 if (namelen >= sizeof(mddev->metadata_type))
4296 namelen = sizeof(mddev->metadata_type)-1;
4297 strncpy(mddev->metadata_type, buf+9, namelen);
4298 mddev->metadata_type[namelen] = 0;
4299 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4300 mddev->metadata_type[--namelen] = 0;
4301 mddev->persistent = 0;
4302 mddev->external = 1;
4303 mddev->major_version = 0;
4304 mddev->minor_version = 90;
4307 major = simple_strtoul(buf, &e, 10);
4309 if (e==buf || *e != '.')
4312 minor = simple_strtoul(buf, &e, 10);
4313 if (e==buf || (*e && *e != '\n') )
4316 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4318 mddev->major_version = major;
4319 mddev->minor_version = minor;
4320 mddev->persistent = 1;
4321 mddev->external = 0;
4324 mddev_unlock(mddev);
4328 static struct md_sysfs_entry md_metadata =
4329 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4332 action_show(struct mddev *mddev, char *page)
4334 char *type = "idle";
4335 unsigned long recovery = mddev->recovery;
4336 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4338 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4339 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4340 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4342 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4343 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4345 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4349 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4351 else if (mddev->reshape_position != MaxSector)
4354 return sprintf(page, "%s\n", type);
4358 action_store(struct mddev *mddev, const char *page, size_t len)
4360 if (!mddev->pers || !mddev->pers->sync_request)
4364 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4365 if (cmd_match(page, "frozen"))
4366 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4368 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4369 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4370 mddev_lock(mddev) == 0) {
4371 flush_workqueue(md_misc_wq);
4372 if (mddev->sync_thread) {
4373 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4374 md_reap_sync_thread(mddev);
4376 mddev_unlock(mddev);
4378 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4380 else if (cmd_match(page, "resync"))
4381 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4382 else if (cmd_match(page, "recover")) {
4383 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4384 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4385 } else if (cmd_match(page, "reshape")) {
4387 if (mddev->pers->start_reshape == NULL)
4389 err = mddev_lock(mddev);
4391 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4394 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4395 err = mddev->pers->start_reshape(mddev);
4397 mddev_unlock(mddev);
4401 sysfs_notify(&mddev->kobj, NULL, "degraded");
4403 if (cmd_match(page, "check"))
4404 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4405 else if (!cmd_match(page, "repair"))
4407 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4408 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4409 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4411 if (mddev->ro == 2) {
4412 /* A write to sync_action is enough to justify
4413 * canceling read-auto mode
4416 md_wakeup_thread(mddev->sync_thread);
4418 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4419 md_wakeup_thread(mddev->thread);
4420 sysfs_notify_dirent_safe(mddev->sysfs_action);
4424 static struct md_sysfs_entry md_scan_mode =
4425 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4428 last_sync_action_show(struct mddev *mddev, char *page)
4430 return sprintf(page, "%s\n", mddev->last_sync_action);
4433 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4436 mismatch_cnt_show(struct mddev *mddev, char *page)
4438 return sprintf(page, "%llu\n",
4439 (unsigned long long)
4440 atomic64_read(&mddev->resync_mismatches));
4443 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4446 sync_min_show(struct mddev *mddev, char *page)
4448 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4449 mddev->sync_speed_min ? "local": "system");
4453 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4458 if (strncmp(buf, "system", 6)==0) {
4461 rv = kstrtouint(buf, 10, &min);
4467 mddev->sync_speed_min = min;
4471 static struct md_sysfs_entry md_sync_min =
4472 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4475 sync_max_show(struct mddev *mddev, char *page)
4477 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4478 mddev->sync_speed_max ? "local": "system");
4482 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4487 if (strncmp(buf, "system", 6)==0) {
4490 rv = kstrtouint(buf, 10, &max);
4496 mddev->sync_speed_max = max;
4500 static struct md_sysfs_entry md_sync_max =
4501 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4504 degraded_show(struct mddev *mddev, char *page)
4506 return sprintf(page, "%d\n", mddev->degraded);
4508 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4511 sync_force_parallel_show(struct mddev *mddev, char *page)
4513 return sprintf(page, "%d\n", mddev->parallel_resync);
4517 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4521 if (kstrtol(buf, 10, &n))
4524 if (n != 0 && n != 1)
4527 mddev->parallel_resync = n;
4529 if (mddev->sync_thread)
4530 wake_up(&resync_wait);
4535 /* force parallel resync, even with shared block devices */
4536 static struct md_sysfs_entry md_sync_force_parallel =
4537 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4538 sync_force_parallel_show, sync_force_parallel_store);
4541 sync_speed_show(struct mddev *mddev, char *page)
4543 unsigned long resync, dt, db;
4544 if (mddev->curr_resync == 0)
4545 return sprintf(page, "none\n");
4546 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4547 dt = (jiffies - mddev->resync_mark) / HZ;
4549 db = resync - mddev->resync_mark_cnt;
4550 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4553 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4556 sync_completed_show(struct mddev *mddev, char *page)
4558 unsigned long long max_sectors, resync;
4560 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4561 return sprintf(page, "none\n");
4563 if (mddev->curr_resync == 1 ||
4564 mddev->curr_resync == 2)
4565 return sprintf(page, "delayed\n");
4567 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4568 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4569 max_sectors = mddev->resync_max_sectors;
4571 max_sectors = mddev->dev_sectors;
4573 resync = mddev->curr_resync_completed;
4574 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4577 static struct md_sysfs_entry md_sync_completed =
4578 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4581 min_sync_show(struct mddev *mddev, char *page)
4583 return sprintf(page, "%llu\n",
4584 (unsigned long long)mddev->resync_min);
4587 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4589 unsigned long long min;
4592 if (kstrtoull(buf, 10, &min))
4595 spin_lock(&mddev->lock);
4597 if (min > mddev->resync_max)
4601 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4604 /* Round down to multiple of 4K for safety */
4605 mddev->resync_min = round_down(min, 8);
4609 spin_unlock(&mddev->lock);
4613 static struct md_sysfs_entry md_min_sync =
4614 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4617 max_sync_show(struct mddev *mddev, char *page)
4619 if (mddev->resync_max == MaxSector)
4620 return sprintf(page, "max\n");
4622 return sprintf(page, "%llu\n",
4623 (unsigned long long)mddev->resync_max);
4626 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4629 spin_lock(&mddev->lock);
4630 if (strncmp(buf, "max", 3) == 0)
4631 mddev->resync_max = MaxSector;
4633 unsigned long long max;
4637 if (kstrtoull(buf, 10, &max))
4639 if (max < mddev->resync_min)
4643 if (max < mddev->resync_max &&
4645 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4648 /* Must be a multiple of chunk_size */
4649 chunk = mddev->chunk_sectors;
4651 sector_t temp = max;
4654 if (sector_div(temp, chunk))
4657 mddev->resync_max = max;
4659 wake_up(&mddev->recovery_wait);
4662 spin_unlock(&mddev->lock);
4666 static struct md_sysfs_entry md_max_sync =
4667 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4670 suspend_lo_show(struct mddev *mddev, char *page)
4672 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4676 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4678 unsigned long long old, new;
4681 err = kstrtoull(buf, 10, &new);
4684 if (new != (sector_t)new)
4687 err = mddev_lock(mddev);
4691 if (mddev->pers == NULL ||
4692 mddev->pers->quiesce == NULL)
4694 old = mddev->suspend_lo;
4695 mddev->suspend_lo = new;
4697 /* Shrinking suspended region */
4698 mddev->pers->quiesce(mddev, 2);
4700 /* Expanding suspended region - need to wait */
4701 mddev->pers->quiesce(mddev, 1);
4702 mddev->pers->quiesce(mddev, 0);
4706 mddev_unlock(mddev);
4709 static struct md_sysfs_entry md_suspend_lo =
4710 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4713 suspend_hi_show(struct mddev *mddev, char *page)
4715 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4719 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4721 unsigned long long old, new;
4724 err = kstrtoull(buf, 10, &new);
4727 if (new != (sector_t)new)
4730 err = mddev_lock(mddev);
4734 if (mddev->pers == NULL ||
4735 mddev->pers->quiesce == NULL)
4737 old = mddev->suspend_hi;
4738 mddev->suspend_hi = new;
4740 /* Shrinking suspended region */
4741 mddev->pers->quiesce(mddev, 2);
4743 /* Expanding suspended region - need to wait */
4744 mddev->pers->quiesce(mddev, 1);
4745 mddev->pers->quiesce(mddev, 0);
4749 mddev_unlock(mddev);
4752 static struct md_sysfs_entry md_suspend_hi =
4753 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4756 reshape_position_show(struct mddev *mddev, char *page)
4758 if (mddev->reshape_position != MaxSector)
4759 return sprintf(page, "%llu\n",
4760 (unsigned long long)mddev->reshape_position);
4761 strcpy(page, "none\n");
4766 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4768 struct md_rdev *rdev;
4769 unsigned long long new;
4772 err = kstrtoull(buf, 10, &new);
4775 if (new != (sector_t)new)
4777 err = mddev_lock(mddev);
4783 mddev->reshape_position = new;
4784 mddev->delta_disks = 0;
4785 mddev->reshape_backwards = 0;
4786 mddev->new_level = mddev->level;
4787 mddev->new_layout = mddev->layout;
4788 mddev->new_chunk_sectors = mddev->chunk_sectors;
4789 rdev_for_each(rdev, mddev)
4790 rdev->new_data_offset = rdev->data_offset;
4793 mddev_unlock(mddev);
4797 static struct md_sysfs_entry md_reshape_position =
4798 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4799 reshape_position_store);
4802 reshape_direction_show(struct mddev *mddev, char *page)
4804 return sprintf(page, "%s\n",
4805 mddev->reshape_backwards ? "backwards" : "forwards");
4809 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4814 if (cmd_match(buf, "forwards"))
4816 else if (cmd_match(buf, "backwards"))
4820 if (mddev->reshape_backwards == backwards)
4823 err = mddev_lock(mddev);
4826 /* check if we are allowed to change */
4827 if (mddev->delta_disks)
4829 else if (mddev->persistent &&
4830 mddev->major_version == 0)
4833 mddev->reshape_backwards = backwards;
4834 mddev_unlock(mddev);
4838 static struct md_sysfs_entry md_reshape_direction =
4839 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4840 reshape_direction_store);
4843 array_size_show(struct mddev *mddev, char *page)
4845 if (mddev->external_size)
4846 return sprintf(page, "%llu\n",
4847 (unsigned long long)mddev->array_sectors/2);
4849 return sprintf(page, "default\n");
4853 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4858 err = mddev_lock(mddev);
4862 /* cluster raid doesn't support change array_sectors */
4863 if (mddev_is_clustered(mddev))
4866 if (strncmp(buf, "default", 7) == 0) {
4868 sectors = mddev->pers->size(mddev, 0, 0);
4870 sectors = mddev->array_sectors;
4872 mddev->external_size = 0;
4874 if (strict_blocks_to_sectors(buf, §ors) < 0)
4876 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4879 mddev->external_size = 1;
4883 mddev->array_sectors = sectors;
4885 set_capacity(mddev->gendisk, mddev->array_sectors);
4886 revalidate_disk(mddev->gendisk);
4889 mddev_unlock(mddev);
4893 static struct md_sysfs_entry md_array_size =
4894 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4897 static struct attribute *md_default_attrs[] = {
4900 &md_raid_disks.attr,
4901 &md_chunk_size.attr,
4903 &md_resync_start.attr,
4905 &md_new_device.attr,
4906 &md_safe_delay.attr,
4907 &md_array_state.attr,
4908 &md_reshape_position.attr,
4909 &md_reshape_direction.attr,
4910 &md_array_size.attr,
4911 &max_corr_read_errors.attr,
4915 static struct attribute *md_redundancy_attrs[] = {
4917 &md_last_scan_mode.attr,
4918 &md_mismatches.attr,
4921 &md_sync_speed.attr,
4922 &md_sync_force_parallel.attr,
4923 &md_sync_completed.attr,
4926 &md_suspend_lo.attr,
4927 &md_suspend_hi.attr,
4932 static struct attribute_group md_redundancy_group = {
4934 .attrs = md_redundancy_attrs,
4938 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4940 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4941 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4946 spin_lock(&all_mddevs_lock);
4947 if (list_empty(&mddev->all_mddevs)) {
4948 spin_unlock(&all_mddevs_lock);
4952 spin_unlock(&all_mddevs_lock);
4954 rv = entry->show(mddev, page);
4960 md_attr_store(struct kobject *kobj, struct attribute *attr,
4961 const char *page, size_t length)
4963 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4964 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4969 if (!capable(CAP_SYS_ADMIN))
4971 spin_lock(&all_mddevs_lock);
4972 if (list_empty(&mddev->all_mddevs)) {
4973 spin_unlock(&all_mddevs_lock);
4977 spin_unlock(&all_mddevs_lock);
4978 rv = entry->store(mddev, page, length);
4983 static void md_free(struct kobject *ko)
4985 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4987 if (mddev->sysfs_state)
4988 sysfs_put(mddev->sysfs_state);
4991 blk_cleanup_queue(mddev->queue);
4992 if (mddev->gendisk) {
4993 del_gendisk(mddev->gendisk);
4994 put_disk(mddev->gendisk);
5000 static const struct sysfs_ops md_sysfs_ops = {
5001 .show = md_attr_show,
5002 .store = md_attr_store,
5004 static struct kobj_type md_ktype = {
5006 .sysfs_ops = &md_sysfs_ops,
5007 .default_attrs = md_default_attrs,
5012 static void mddev_delayed_delete(struct work_struct *ws)
5014 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5016 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5017 kobject_del(&mddev->kobj);
5018 kobject_put(&mddev->kobj);
5021 static int md_alloc(dev_t dev, char *name)
5023 static DEFINE_MUTEX(disks_mutex);
5024 struct mddev *mddev = mddev_find(dev);
5025 struct gendisk *disk;
5034 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5035 shift = partitioned ? MdpMinorShift : 0;
5036 unit = MINOR(mddev->unit) >> shift;
5038 /* wait for any previous instance of this device to be
5039 * completely removed (mddev_delayed_delete).
5041 flush_workqueue(md_misc_wq);
5043 mutex_lock(&disks_mutex);
5049 /* Need to ensure that 'name' is not a duplicate.
5051 struct mddev *mddev2;
5052 spin_lock(&all_mddevs_lock);
5054 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5055 if (mddev2->gendisk &&
5056 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5057 spin_unlock(&all_mddevs_lock);
5060 spin_unlock(&all_mddevs_lock);
5064 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5067 mddev->queue->queuedata = mddev;
5069 blk_queue_make_request(mddev->queue, md_make_request);
5070 blk_set_stacking_limits(&mddev->queue->limits);
5072 disk = alloc_disk(1 << shift);
5074 blk_cleanup_queue(mddev->queue);
5075 mddev->queue = NULL;
5078 disk->major = MAJOR(mddev->unit);
5079 disk->first_minor = unit << shift;
5081 strcpy(disk->disk_name, name);
5082 else if (partitioned)
5083 sprintf(disk->disk_name, "md_d%d", unit);
5085 sprintf(disk->disk_name, "md%d", unit);
5086 disk->fops = &md_fops;
5087 disk->private_data = mddev;
5088 disk->queue = mddev->queue;
5089 blk_queue_write_cache(mddev->queue, true, true);
5090 /* Allow extended partitions. This makes the
5091 * 'mdp' device redundant, but we can't really
5094 disk->flags |= GENHD_FL_EXT_DEVT;
5095 mddev->gendisk = disk;
5096 /* As soon as we call add_disk(), another thread could get
5097 * through to md_open, so make sure it doesn't get too far
5099 mutex_lock(&mddev->open_mutex);
5102 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5103 &disk_to_dev(disk)->kobj, "%s", "md");
5105 /* This isn't possible, but as kobject_init_and_add is marked
5106 * __must_check, we must do something with the result
5108 pr_debug("md: cannot register %s/md - name in use\n",
5112 if (mddev->kobj.sd &&
5113 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5114 pr_debug("pointless warning\n");
5115 mutex_unlock(&mddev->open_mutex);
5117 mutex_unlock(&disks_mutex);
5118 if (!error && mddev->kobj.sd) {
5119 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5120 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5126 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5128 md_alloc(dev, NULL);
5132 static int add_named_array(const char *val, struct kernel_param *kp)
5134 /* val must be "md_*" where * is not all digits.
5135 * We allocate an array with a large free minor number, and
5136 * set the name to val. val must not already be an active name.
5138 int len = strlen(val);
5139 char buf[DISK_NAME_LEN];
5141 while (len && val[len-1] == '\n')
5143 if (len >= DISK_NAME_LEN)
5145 strlcpy(buf, val, len+1);
5146 if (strncmp(buf, "md_", 3) != 0)
5148 return md_alloc(0, buf);
5151 static void md_safemode_timeout(unsigned long data)
5153 struct mddev *mddev = (struct mddev *) data;
5155 if (!atomic_read(&mddev->writes_pending)) {
5156 mddev->safemode = 1;
5157 if (mddev->external)
5158 sysfs_notify_dirent_safe(mddev->sysfs_state);
5160 md_wakeup_thread(mddev->thread);
5163 static int start_dirty_degraded;
5165 int md_run(struct mddev *mddev)
5168 struct md_rdev *rdev;
5169 struct md_personality *pers;
5171 if (list_empty(&mddev->disks))
5172 /* cannot run an array with no devices.. */
5177 /* Cannot run until previous stop completes properly */
5178 if (mddev->sysfs_active)
5182 * Analyze all RAID superblock(s)
5184 if (!mddev->raid_disks) {
5185 if (!mddev->persistent)
5190 if (mddev->level != LEVEL_NONE)
5191 request_module("md-level-%d", mddev->level);
5192 else if (mddev->clevel[0])
5193 request_module("md-%s", mddev->clevel);
5196 * Drop all container device buffers, from now on
5197 * the only valid external interface is through the md
5200 rdev_for_each(rdev, mddev) {
5201 if (test_bit(Faulty, &rdev->flags))
5203 sync_blockdev(rdev->bdev);
5204 invalidate_bdev(rdev->bdev);
5206 /* perform some consistency tests on the device.
5207 * We don't want the data to overlap the metadata,
5208 * Internal Bitmap issues have been handled elsewhere.
5210 if (rdev->meta_bdev) {
5211 /* Nothing to check */;
5212 } else if (rdev->data_offset < rdev->sb_start) {
5213 if (mddev->dev_sectors &&
5214 rdev->data_offset + mddev->dev_sectors
5216 pr_warn("md: %s: data overlaps metadata\n",
5221 if (rdev->sb_start + rdev->sb_size/512
5222 > rdev->data_offset) {
5223 pr_warn("md: %s: metadata overlaps data\n",
5228 sysfs_notify_dirent_safe(rdev->sysfs_state);
5231 if (mddev->bio_set == NULL)
5232 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5234 spin_lock(&pers_lock);
5235 pers = find_pers(mddev->level, mddev->clevel);
5236 if (!pers || !try_module_get(pers->owner)) {
5237 spin_unlock(&pers_lock);
5238 if (mddev->level != LEVEL_NONE)
5239 pr_warn("md: personality for level %d is not loaded!\n",
5242 pr_warn("md: personality for level %s is not loaded!\n",
5246 spin_unlock(&pers_lock);
5247 if (mddev->level != pers->level) {
5248 mddev->level = pers->level;
5249 mddev->new_level = pers->level;
5251 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5253 if (mddev->reshape_position != MaxSector &&
5254 pers->start_reshape == NULL) {
5255 /* This personality cannot handle reshaping... */
5256 module_put(pers->owner);
5260 if (pers->sync_request) {
5261 /* Warn if this is a potentially silly
5264 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5265 struct md_rdev *rdev2;
5268 rdev_for_each(rdev, mddev)
5269 rdev_for_each(rdev2, mddev) {
5271 rdev->bdev->bd_contains ==
5272 rdev2->bdev->bd_contains) {
5273 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5275 bdevname(rdev->bdev,b),
5276 bdevname(rdev2->bdev,b2));
5282 pr_warn("True protection against single-disk failure might be compromised.\n");
5285 mddev->recovery = 0;
5286 /* may be over-ridden by personality */
5287 mddev->resync_max_sectors = mddev->dev_sectors;
5289 mddev->ok_start_degraded = start_dirty_degraded;
5291 if (start_readonly && mddev->ro == 0)
5292 mddev->ro = 2; /* read-only, but switch on first write */
5295 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
5296 * up mddev->thread. It is important to initialize critical
5297 * resources for mddev->thread BEFORE calling pers->run().
5299 err = pers->run(mddev);
5301 pr_warn("md: pers->run() failed ...\n");
5302 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5303 WARN_ONCE(!mddev->external_size,
5304 "%s: default size too small, but 'external_size' not in effect?\n",
5306 pr_warn("md: invalid array_size %llu > default size %llu\n",
5307 (unsigned long long)mddev->array_sectors / 2,
5308 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5311 if (err == 0 && pers->sync_request &&
5312 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5313 struct bitmap *bitmap;
5315 bitmap = bitmap_create(mddev, -1);
5316 if (IS_ERR(bitmap)) {
5317 err = PTR_ERR(bitmap);
5318 pr_warn("%s: failed to create bitmap (%d)\n",
5319 mdname(mddev), err);
5321 mddev->bitmap = bitmap;
5325 mddev_detach(mddev);
5327 pers->free(mddev, mddev->private);
5328 mddev->private = NULL;
5329 module_put(pers->owner);
5330 bitmap_destroy(mddev);
5336 rdev_for_each(rdev, mddev) {
5337 if (rdev->raid_disk >= 0 &&
5338 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5343 if (mddev->degraded)
5346 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5348 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5349 mddev->queue->backing_dev_info.congested_data = mddev;
5350 mddev->queue->backing_dev_info.congested_fn = md_congested;
5352 if (pers->sync_request) {
5353 if (mddev->kobj.sd &&
5354 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5355 pr_warn("md: cannot register extra attributes for %s\n",
5357 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5358 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5361 atomic_set(&mddev->writes_pending,0);
5362 atomic_set(&mddev->max_corr_read_errors,
5363 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5364 mddev->safemode = 0;
5365 if (mddev_is_clustered(mddev))
5366 mddev->safemode_delay = 0;
5368 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5371 spin_lock(&mddev->lock);
5373 spin_unlock(&mddev->lock);
5374 rdev_for_each(rdev, mddev)
5375 if (rdev->raid_disk >= 0)
5376 if (sysfs_link_rdev(mddev, rdev))
5377 /* failure here is OK */;
5379 if (mddev->degraded && !mddev->ro)
5380 /* This ensures that recovering status is reported immediately
5381 * via sysfs - until a lack of spares is confirmed.
5383 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5384 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5386 if (mddev->sb_flags)
5387 md_update_sb(mddev, 0);
5389 md_new_event(mddev);
5390 sysfs_notify_dirent_safe(mddev->sysfs_state);
5391 sysfs_notify_dirent_safe(mddev->sysfs_action);
5392 sysfs_notify(&mddev->kobj, NULL, "degraded");
5395 EXPORT_SYMBOL_GPL(md_run);
5397 static int do_md_run(struct mddev *mddev)
5401 err = md_run(mddev);
5404 err = bitmap_load(mddev);
5406 bitmap_destroy(mddev);
5410 if (mddev_is_clustered(mddev))
5411 md_allow_write(mddev);
5413 md_wakeup_thread(mddev->thread);
5414 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5416 set_capacity(mddev->gendisk, mddev->array_sectors);
5417 revalidate_disk(mddev->gendisk);
5419 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5424 static int restart_array(struct mddev *mddev)
5426 struct gendisk *disk = mddev->gendisk;
5428 /* Complain if it has no devices */
5429 if (list_empty(&mddev->disks))
5435 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5436 struct md_rdev *rdev;
5437 bool has_journal = false;
5440 rdev_for_each_rcu(rdev, mddev) {
5441 if (test_bit(Journal, &rdev->flags) &&
5442 !test_bit(Faulty, &rdev->flags)) {
5449 /* Don't restart rw with journal missing/faulty */
5454 mddev->safemode = 0;
5456 set_disk_ro(disk, 0);
5457 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5458 /* Kick recovery or resync if necessary */
5459 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5460 md_wakeup_thread(mddev->thread);
5461 md_wakeup_thread(mddev->sync_thread);
5462 sysfs_notify_dirent_safe(mddev->sysfs_state);
5466 static void md_clean(struct mddev *mddev)
5468 mddev->array_sectors = 0;
5469 mddev->external_size = 0;
5470 mddev->dev_sectors = 0;
5471 mddev->raid_disks = 0;
5472 mddev->recovery_cp = 0;
5473 mddev->resync_min = 0;
5474 mddev->resync_max = MaxSector;
5475 mddev->reshape_position = MaxSector;
5476 mddev->external = 0;
5477 mddev->persistent = 0;
5478 mddev->level = LEVEL_NONE;
5479 mddev->clevel[0] = 0;
5481 mddev->sb_flags = 0;
5483 mddev->metadata_type[0] = 0;
5484 mddev->chunk_sectors = 0;
5485 mddev->ctime = mddev->utime = 0;
5487 mddev->max_disks = 0;
5489 mddev->can_decrease_events = 0;
5490 mddev->delta_disks = 0;
5491 mddev->reshape_backwards = 0;
5492 mddev->new_level = LEVEL_NONE;
5493 mddev->new_layout = 0;
5494 mddev->new_chunk_sectors = 0;
5495 mddev->curr_resync = 0;
5496 atomic64_set(&mddev->resync_mismatches, 0);
5497 mddev->suspend_lo = mddev->suspend_hi = 0;
5498 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5499 mddev->recovery = 0;
5502 mddev->degraded = 0;
5503 mddev->safemode = 0;
5504 mddev->private = NULL;
5505 mddev->cluster_info = NULL;
5506 mddev->bitmap_info.offset = 0;
5507 mddev->bitmap_info.default_offset = 0;
5508 mddev->bitmap_info.default_space = 0;
5509 mddev->bitmap_info.chunksize = 0;
5510 mddev->bitmap_info.daemon_sleep = 0;
5511 mddev->bitmap_info.max_write_behind = 0;
5512 mddev->bitmap_info.nodes = 0;
5515 static void __md_stop_writes(struct mddev *mddev)
5517 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5518 flush_workqueue(md_misc_wq);
5519 if (mddev->sync_thread) {
5520 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5521 md_reap_sync_thread(mddev);
5524 del_timer_sync(&mddev->safemode_timer);
5526 if (mddev->pers && mddev->pers->quiesce) {
5527 mddev->pers->quiesce(mddev, 1);
5528 mddev->pers->quiesce(mddev, 0);
5530 bitmap_flush(mddev);
5532 if (mddev->ro == 0 &&
5533 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5535 /* mark array as shutdown cleanly */
5536 if (!mddev_is_clustered(mddev))
5538 md_update_sb(mddev, 1);
5542 void md_stop_writes(struct mddev *mddev)
5544 mddev_lock_nointr(mddev);
5545 __md_stop_writes(mddev);
5546 mddev_unlock(mddev);
5548 EXPORT_SYMBOL_GPL(md_stop_writes);
5550 static void mddev_detach(struct mddev *mddev)
5552 struct bitmap *bitmap = mddev->bitmap;
5553 /* wait for behind writes to complete */
5554 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5555 pr_debug("md:%s: behind writes in progress - waiting to stop.\n",
5557 /* need to kick something here to make sure I/O goes? */
5558 wait_event(bitmap->behind_wait,
5559 atomic_read(&bitmap->behind_writes) == 0);
5561 if (mddev->pers && mddev->pers->quiesce) {
5562 mddev->pers->quiesce(mddev, 1);
5563 mddev->pers->quiesce(mddev, 0);
5565 md_unregister_thread(&mddev->thread);
5567 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5570 static void __md_stop(struct mddev *mddev)
5572 struct md_personality *pers = mddev->pers;
5573 mddev_detach(mddev);
5574 /* Ensure ->event_work is done */
5575 flush_workqueue(md_misc_wq);
5576 spin_lock(&mddev->lock);
5578 spin_unlock(&mddev->lock);
5579 pers->free(mddev, mddev->private);
5580 mddev->private = NULL;
5581 if (pers->sync_request && mddev->to_remove == NULL)
5582 mddev->to_remove = &md_redundancy_group;
5583 module_put(pers->owner);
5584 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5587 void md_stop(struct mddev *mddev)
5589 /* stop the array and free an attached data structures.
5590 * This is called from dm-raid
5593 bitmap_destroy(mddev);
5595 bioset_free(mddev->bio_set);
5598 EXPORT_SYMBOL_GPL(md_stop);
5600 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5605 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5607 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5608 md_wakeup_thread(mddev->thread);
5610 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5611 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5612 if (mddev->sync_thread)
5613 /* Thread might be blocked waiting for metadata update
5614 * which will now never happen */
5615 wake_up_process(mddev->sync_thread->tsk);
5617 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5619 mddev_unlock(mddev);
5620 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5622 wait_event(mddev->sb_wait,
5623 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5624 mddev_lock_nointr(mddev);
5626 mutex_lock(&mddev->open_mutex);
5627 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5628 mddev->sync_thread ||
5629 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5630 pr_warn("md: %s still in use.\n",mdname(mddev));
5632 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5633 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5634 md_wakeup_thread(mddev->thread);
5640 __md_stop_writes(mddev);
5646 set_disk_ro(mddev->gendisk, 1);
5647 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5648 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5649 md_wakeup_thread(mddev->thread);
5650 sysfs_notify_dirent_safe(mddev->sysfs_state);
5654 mutex_unlock(&mddev->open_mutex);
5659 * 0 - completely stop and dis-assemble array
5660 * 2 - stop but do not disassemble array
5662 static int do_md_stop(struct mddev *mddev, int mode,
5663 struct block_device *bdev)
5665 struct gendisk *disk = mddev->gendisk;
5666 struct md_rdev *rdev;
5669 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5671 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5672 md_wakeup_thread(mddev->thread);
5674 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5675 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5676 if (mddev->sync_thread)
5677 /* Thread might be blocked waiting for metadata update
5678 * which will now never happen */
5679 wake_up_process(mddev->sync_thread->tsk);
5681 mddev_unlock(mddev);
5682 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5683 !test_bit(MD_RECOVERY_RUNNING,
5684 &mddev->recovery)));
5685 mddev_lock_nointr(mddev);
5687 mutex_lock(&mddev->open_mutex);
5688 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5689 mddev->sysfs_active ||
5690 mddev->sync_thread ||
5691 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5692 pr_warn("md: %s still in use.\n",mdname(mddev));
5693 mutex_unlock(&mddev->open_mutex);
5695 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5696 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5697 md_wakeup_thread(mddev->thread);
5703 set_disk_ro(disk, 0);
5705 __md_stop_writes(mddev);
5707 mddev->queue->backing_dev_info.congested_fn = NULL;
5709 /* tell userspace to handle 'inactive' */
5710 sysfs_notify_dirent_safe(mddev->sysfs_state);
5712 rdev_for_each(rdev, mddev)
5713 if (rdev->raid_disk >= 0)
5714 sysfs_unlink_rdev(mddev, rdev);
5716 set_capacity(disk, 0);
5717 mutex_unlock(&mddev->open_mutex);
5719 revalidate_disk(disk);
5724 mutex_unlock(&mddev->open_mutex);
5726 * Free resources if final stop
5729 pr_info("md: %s stopped.\n", mdname(mddev));
5731 bitmap_destroy(mddev);
5732 if (mddev->bitmap_info.file) {
5733 struct file *f = mddev->bitmap_info.file;
5734 spin_lock(&mddev->lock);
5735 mddev->bitmap_info.file = NULL;
5736 spin_unlock(&mddev->lock);
5739 mddev->bitmap_info.offset = 0;
5741 export_array(mddev);
5744 if (mddev->hold_active == UNTIL_STOP)
5745 mddev->hold_active = 0;
5747 md_new_event(mddev);
5748 sysfs_notify_dirent_safe(mddev->sysfs_state);
5753 static void autorun_array(struct mddev *mddev)
5755 struct md_rdev *rdev;
5758 if (list_empty(&mddev->disks))
5761 pr_info("md: running: ");
5763 rdev_for_each(rdev, mddev) {
5764 char b[BDEVNAME_SIZE];
5765 pr_cont("<%s>", bdevname(rdev->bdev,b));
5769 err = do_md_run(mddev);
5771 pr_warn("md: do_md_run() returned %d\n", err);
5772 do_md_stop(mddev, 0, NULL);
5777 * lets try to run arrays based on all disks that have arrived
5778 * until now. (those are in pending_raid_disks)
5780 * the method: pick the first pending disk, collect all disks with
5781 * the same UUID, remove all from the pending list and put them into
5782 * the 'same_array' list. Then order this list based on superblock
5783 * update time (freshest comes first), kick out 'old' disks and
5784 * compare superblocks. If everything's fine then run it.
5786 * If "unit" is allocated, then bump its reference count
5788 static void autorun_devices(int part)
5790 struct md_rdev *rdev0, *rdev, *tmp;
5791 struct mddev *mddev;
5792 char b[BDEVNAME_SIZE];
5794 pr_info("md: autorun ...\n");
5795 while (!list_empty(&pending_raid_disks)) {
5798 LIST_HEAD(candidates);
5799 rdev0 = list_entry(pending_raid_disks.next,
5800 struct md_rdev, same_set);
5802 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
5803 INIT_LIST_HEAD(&candidates);
5804 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5805 if (super_90_load(rdev, rdev0, 0) >= 0) {
5806 pr_debug("md: adding %s ...\n",
5807 bdevname(rdev->bdev,b));
5808 list_move(&rdev->same_set, &candidates);
5811 * now we have a set of devices, with all of them having
5812 * mostly sane superblocks. It's time to allocate the
5816 dev = MKDEV(mdp_major,
5817 rdev0->preferred_minor << MdpMinorShift);
5818 unit = MINOR(dev) >> MdpMinorShift;
5820 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5823 if (rdev0->preferred_minor != unit) {
5824 pr_warn("md: unit number in %s is bad: %d\n",
5825 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5829 md_probe(dev, NULL, NULL);
5830 mddev = mddev_find(dev);
5831 if (!mddev || !mddev->gendisk) {
5836 if (mddev_lock(mddev))
5837 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
5838 else if (mddev->raid_disks || mddev->major_version
5839 || !list_empty(&mddev->disks)) {
5840 pr_warn("md: %s already running, cannot run %s\n",
5841 mdname(mddev), bdevname(rdev0->bdev,b));
5842 mddev_unlock(mddev);
5844 pr_debug("md: created %s\n", mdname(mddev));
5845 mddev->persistent = 1;
5846 rdev_for_each_list(rdev, tmp, &candidates) {
5847 list_del_init(&rdev->same_set);
5848 if (bind_rdev_to_array(rdev, mddev))
5851 autorun_array(mddev);
5852 mddev_unlock(mddev);
5854 /* on success, candidates will be empty, on error
5857 rdev_for_each_list(rdev, tmp, &candidates) {
5858 list_del_init(&rdev->same_set);
5863 pr_info("md: ... autorun DONE.\n");
5865 #endif /* !MODULE */
5867 static int get_version(void __user *arg)
5871 ver.major = MD_MAJOR_VERSION;
5872 ver.minor = MD_MINOR_VERSION;
5873 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5875 if (copy_to_user(arg, &ver, sizeof(ver)))
5881 static int get_array_info(struct mddev *mddev, void __user *arg)
5883 mdu_array_info_t info;
5884 int nr,working,insync,failed,spare;
5885 struct md_rdev *rdev;
5887 nr = working = insync = failed = spare = 0;
5889 rdev_for_each_rcu(rdev, mddev) {
5891 if (test_bit(Faulty, &rdev->flags))
5895 if (test_bit(In_sync, &rdev->flags))
5897 else if (test_bit(Journal, &rdev->flags))
5898 /* TODO: add journal count to md_u.h */
5906 info.major_version = mddev->major_version;
5907 info.minor_version = mddev->minor_version;
5908 info.patch_version = MD_PATCHLEVEL_VERSION;
5909 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5910 info.level = mddev->level;
5911 info.size = mddev->dev_sectors / 2;
5912 if (info.size != mddev->dev_sectors / 2) /* overflow */
5915 info.raid_disks = mddev->raid_disks;
5916 info.md_minor = mddev->md_minor;
5917 info.not_persistent= !mddev->persistent;
5919 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5922 info.state = (1<<MD_SB_CLEAN);
5923 if (mddev->bitmap && mddev->bitmap_info.offset)
5924 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5925 if (mddev_is_clustered(mddev))
5926 info.state |= (1<<MD_SB_CLUSTERED);
5927 info.active_disks = insync;
5928 info.working_disks = working;
5929 info.failed_disks = failed;
5930 info.spare_disks = spare;
5932 info.layout = mddev->layout;
5933 info.chunk_size = mddev->chunk_sectors << 9;
5935 if (copy_to_user(arg, &info, sizeof(info)))
5941 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5943 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5947 file = kzalloc(sizeof(*file), GFP_NOIO);
5952 spin_lock(&mddev->lock);
5953 /* bitmap enabled */
5954 if (mddev->bitmap_info.file) {
5955 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5956 sizeof(file->pathname));
5960 memmove(file->pathname, ptr,
5961 sizeof(file->pathname)-(ptr-file->pathname));
5963 spin_unlock(&mddev->lock);
5966 copy_to_user(arg, file, sizeof(*file)))
5973 static int get_disk_info(struct mddev *mddev, void __user * arg)
5975 mdu_disk_info_t info;
5976 struct md_rdev *rdev;
5978 if (copy_from_user(&info, arg, sizeof(info)))
5982 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5984 info.major = MAJOR(rdev->bdev->bd_dev);
5985 info.minor = MINOR(rdev->bdev->bd_dev);
5986 info.raid_disk = rdev->raid_disk;
5988 if (test_bit(Faulty, &rdev->flags))
5989 info.state |= (1<<MD_DISK_FAULTY);
5990 else if (test_bit(In_sync, &rdev->flags)) {
5991 info.state |= (1<<MD_DISK_ACTIVE);
5992 info.state |= (1<<MD_DISK_SYNC);
5994 if (test_bit(Journal, &rdev->flags))
5995 info.state |= (1<<MD_DISK_JOURNAL);
5996 if (test_bit(WriteMostly, &rdev->flags))
5997 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5998 if (test_bit(FailFast, &rdev->flags))
5999 info.state |= (1<<MD_DISK_FAILFAST);
6001 info.major = info.minor = 0;
6002 info.raid_disk = -1;
6003 info.state = (1<<MD_DISK_REMOVED);
6007 if (copy_to_user(arg, &info, sizeof(info)))
6013 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6015 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6016 struct md_rdev *rdev;
6017 dev_t dev = MKDEV(info->major,info->minor);
6019 if (mddev_is_clustered(mddev) &&
6020 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6021 pr_warn("%s: Cannot add to clustered mddev.\n",
6026 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6029 if (!mddev->raid_disks) {
6031 /* expecting a device which has a superblock */
6032 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6034 pr_warn("md: md_import_device returned %ld\n",
6036 return PTR_ERR(rdev);
6038 if (!list_empty(&mddev->disks)) {
6039 struct md_rdev *rdev0
6040 = list_entry(mddev->disks.next,
6041 struct md_rdev, same_set);
6042 err = super_types[mddev->major_version]
6043 .load_super(rdev, rdev0, mddev->minor_version);
6045 pr_warn("md: %s has different UUID to %s\n",
6046 bdevname(rdev->bdev,b),
6047 bdevname(rdev0->bdev,b2));
6052 err = bind_rdev_to_array(rdev, mddev);
6059 * add_new_disk can be used once the array is assembled
6060 * to add "hot spares". They must already have a superblock
6065 if (!mddev->pers->hot_add_disk) {
6066 pr_warn("%s: personality does not support diskops!\n",
6070 if (mddev->persistent)
6071 rdev = md_import_device(dev, mddev->major_version,
6072 mddev->minor_version);
6074 rdev = md_import_device(dev, -1, -1);
6076 pr_warn("md: md_import_device returned %ld\n",
6078 return PTR_ERR(rdev);
6080 /* set saved_raid_disk if appropriate */
6081 if (!mddev->persistent) {
6082 if (info->state & (1<<MD_DISK_SYNC) &&
6083 info->raid_disk < mddev->raid_disks) {
6084 rdev->raid_disk = info->raid_disk;
6085 set_bit(In_sync, &rdev->flags);
6086 clear_bit(Bitmap_sync, &rdev->flags);
6088 rdev->raid_disk = -1;
6089 rdev->saved_raid_disk = rdev->raid_disk;
6091 super_types[mddev->major_version].
6092 validate_super(mddev, rdev);
6093 if ((info->state & (1<<MD_DISK_SYNC)) &&
6094 rdev->raid_disk != info->raid_disk) {
6095 /* This was a hot-add request, but events doesn't
6096 * match, so reject it.
6102 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6103 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6104 set_bit(WriteMostly, &rdev->flags);
6106 clear_bit(WriteMostly, &rdev->flags);
6107 if (info->state & (1<<MD_DISK_FAILFAST))
6108 set_bit(FailFast, &rdev->flags);
6110 clear_bit(FailFast, &rdev->flags);
6112 if (info->state & (1<<MD_DISK_JOURNAL)) {
6113 struct md_rdev *rdev2;
6114 bool has_journal = false;
6116 /* make sure no existing journal disk */
6117 rdev_for_each(rdev2, mddev) {
6118 if (test_bit(Journal, &rdev2->flags)) {
6127 set_bit(Journal, &rdev->flags);
6130 * check whether the device shows up in other nodes
6132 if (mddev_is_clustered(mddev)) {
6133 if (info->state & (1 << MD_DISK_CANDIDATE))
6134 set_bit(Candidate, &rdev->flags);
6135 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6136 /* --add initiated by this node */
6137 err = md_cluster_ops->add_new_disk(mddev, rdev);
6145 rdev->raid_disk = -1;
6146 err = bind_rdev_to_array(rdev, mddev);
6151 if (mddev_is_clustered(mddev)) {
6152 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6154 err = md_cluster_ops->new_disk_ack(mddev,
6157 md_kick_rdev_from_array(rdev);
6161 md_cluster_ops->add_new_disk_cancel(mddev);
6163 err = add_bound_rdev(rdev);
6167 err = add_bound_rdev(rdev);
6172 /* otherwise, add_new_disk is only allowed
6173 * for major_version==0 superblocks
6175 if (mddev->major_version != 0) {
6176 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6180 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6182 rdev = md_import_device(dev, -1, 0);
6184 pr_warn("md: error, md_import_device() returned %ld\n",
6186 return PTR_ERR(rdev);
6188 rdev->desc_nr = info->number;
6189 if (info->raid_disk < mddev->raid_disks)
6190 rdev->raid_disk = info->raid_disk;
6192 rdev->raid_disk = -1;
6194 if (rdev->raid_disk < mddev->raid_disks)
6195 if (info->state & (1<<MD_DISK_SYNC))
6196 set_bit(In_sync, &rdev->flags);
6198 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6199 set_bit(WriteMostly, &rdev->flags);
6200 if (info->state & (1<<MD_DISK_FAILFAST))
6201 set_bit(FailFast, &rdev->flags);
6203 if (!mddev->persistent) {
6204 pr_debug("md: nonpersistent superblock ...\n");
6205 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6207 rdev->sb_start = calc_dev_sboffset(rdev);
6208 rdev->sectors = rdev->sb_start;
6210 err = bind_rdev_to_array(rdev, mddev);
6220 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6222 char b[BDEVNAME_SIZE];
6223 struct md_rdev *rdev;
6225 rdev = find_rdev(mddev, dev);
6229 if (rdev->raid_disk < 0)
6232 clear_bit(Blocked, &rdev->flags);
6233 remove_and_add_spares(mddev, rdev);
6235 if (rdev->raid_disk >= 0)
6239 if (mddev_is_clustered(mddev))
6240 md_cluster_ops->remove_disk(mddev, rdev);
6242 md_kick_rdev_from_array(rdev);
6243 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6245 md_wakeup_thread(mddev->thread);
6247 md_update_sb(mddev, 1);
6248 md_new_event(mddev);
6252 pr_debug("md: cannot remove active disk %s from %s ...\n",
6253 bdevname(rdev->bdev,b), mdname(mddev));
6257 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6259 char b[BDEVNAME_SIZE];
6261 struct md_rdev *rdev;
6266 if (mddev->major_version != 0) {
6267 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6271 if (!mddev->pers->hot_add_disk) {
6272 pr_warn("%s: personality does not support diskops!\n",
6277 rdev = md_import_device(dev, -1, 0);
6279 pr_warn("md: error, md_import_device() returned %ld\n",
6284 if (mddev->persistent)
6285 rdev->sb_start = calc_dev_sboffset(rdev);
6287 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6289 rdev->sectors = rdev->sb_start;
6291 if (test_bit(Faulty, &rdev->flags)) {
6292 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6293 bdevname(rdev->bdev,b), mdname(mddev));
6298 clear_bit(In_sync, &rdev->flags);
6300 rdev->saved_raid_disk = -1;
6301 err = bind_rdev_to_array(rdev, mddev);
6306 * The rest should better be atomic, we can have disk failures
6307 * noticed in interrupt contexts ...
6310 rdev->raid_disk = -1;
6312 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6314 md_update_sb(mddev, 1);
6316 * Kick recovery, maybe this spare has to be added to the
6317 * array immediately.
6319 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6320 md_wakeup_thread(mddev->thread);
6321 md_new_event(mddev);
6329 static int set_bitmap_file(struct mddev *mddev, int fd)
6334 if (!mddev->pers->quiesce || !mddev->thread)
6336 if (mddev->recovery || mddev->sync_thread)
6338 /* we should be able to change the bitmap.. */
6342 struct inode *inode;
6345 if (mddev->bitmap || mddev->bitmap_info.file)
6346 return -EEXIST; /* cannot add when bitmap is present */
6350 pr_warn("%s: error: failed to get bitmap file\n",
6355 inode = f->f_mapping->host;
6356 if (!S_ISREG(inode->i_mode)) {
6357 pr_warn("%s: error: bitmap file must be a regular file\n",
6360 } else if (!(f->f_mode & FMODE_WRITE)) {
6361 pr_warn("%s: error: bitmap file must open for write\n",
6364 } else if (atomic_read(&inode->i_writecount) != 1) {
6365 pr_warn("%s: error: bitmap file is already in use\n",
6373 mddev->bitmap_info.file = f;
6374 mddev->bitmap_info.offset = 0; /* file overrides offset */
6375 } else if (mddev->bitmap == NULL)
6376 return -ENOENT; /* cannot remove what isn't there */
6379 mddev->pers->quiesce(mddev, 1);
6381 struct bitmap *bitmap;
6383 bitmap = bitmap_create(mddev, -1);
6384 if (!IS_ERR(bitmap)) {
6385 mddev->bitmap = bitmap;
6386 err = bitmap_load(mddev);
6388 err = PTR_ERR(bitmap);
6390 if (fd < 0 || err) {
6391 bitmap_destroy(mddev);
6392 fd = -1; /* make sure to put the file */
6394 mddev->pers->quiesce(mddev, 0);
6397 struct file *f = mddev->bitmap_info.file;
6399 spin_lock(&mddev->lock);
6400 mddev->bitmap_info.file = NULL;
6401 spin_unlock(&mddev->lock);
6410 * set_array_info is used two different ways
6411 * The original usage is when creating a new array.
6412 * In this usage, raid_disks is > 0 and it together with
6413 * level, size, not_persistent,layout,chunksize determine the
6414 * shape of the array.
6415 * This will always create an array with a type-0.90.0 superblock.
6416 * The newer usage is when assembling an array.
6417 * In this case raid_disks will be 0, and the major_version field is
6418 * use to determine which style super-blocks are to be found on the devices.
6419 * The minor and patch _version numbers are also kept incase the
6420 * super_block handler wishes to interpret them.
6422 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6425 if (info->raid_disks == 0) {
6426 /* just setting version number for superblock loading */
6427 if (info->major_version < 0 ||
6428 info->major_version >= ARRAY_SIZE(super_types) ||
6429 super_types[info->major_version].name == NULL) {
6430 /* maybe try to auto-load a module? */
6431 pr_warn("md: superblock version %d not known\n",
6432 info->major_version);
6435 mddev->major_version = info->major_version;
6436 mddev->minor_version = info->minor_version;
6437 mddev->patch_version = info->patch_version;
6438 mddev->persistent = !info->not_persistent;
6439 /* ensure mddev_put doesn't delete this now that there
6440 * is some minimal configuration.
6442 mddev->ctime = ktime_get_real_seconds();
6445 mddev->major_version = MD_MAJOR_VERSION;
6446 mddev->minor_version = MD_MINOR_VERSION;
6447 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6448 mddev->ctime = ktime_get_real_seconds();
6450 mddev->level = info->level;
6451 mddev->clevel[0] = 0;
6452 mddev->dev_sectors = 2 * (sector_t)info->size;
6453 mddev->raid_disks = info->raid_disks;
6454 /* don't set md_minor, it is determined by which /dev/md* was
6457 if (info->state & (1<<MD_SB_CLEAN))
6458 mddev->recovery_cp = MaxSector;
6460 mddev->recovery_cp = 0;
6461 mddev->persistent = ! info->not_persistent;
6462 mddev->external = 0;
6464 mddev->layout = info->layout;
6465 mddev->chunk_sectors = info->chunk_size >> 9;
6467 mddev->max_disks = MD_SB_DISKS;
6469 if (mddev->persistent) {
6471 mddev->sb_flags = 0;
6473 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6475 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6476 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6477 mddev->bitmap_info.offset = 0;
6479 mddev->reshape_position = MaxSector;
6482 * Generate a 128 bit UUID
6484 get_random_bytes(mddev->uuid, 16);
6486 mddev->new_level = mddev->level;
6487 mddev->new_chunk_sectors = mddev->chunk_sectors;
6488 mddev->new_layout = mddev->layout;
6489 mddev->delta_disks = 0;
6490 mddev->reshape_backwards = 0;
6495 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6497 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6499 if (mddev->external_size)
6502 mddev->array_sectors = array_sectors;
6504 EXPORT_SYMBOL(md_set_array_sectors);
6506 static int update_size(struct mddev *mddev, sector_t num_sectors)
6508 struct md_rdev *rdev;
6510 int fit = (num_sectors == 0);
6512 /* cluster raid doesn't support update size */
6513 if (mddev_is_clustered(mddev))
6516 if (mddev->pers->resize == NULL)
6518 /* The "num_sectors" is the number of sectors of each device that
6519 * is used. This can only make sense for arrays with redundancy.
6520 * linear and raid0 always use whatever space is available. We can only
6521 * consider changing this number if no resync or reconstruction is
6522 * happening, and if the new size is acceptable. It must fit before the
6523 * sb_start or, if that is <data_offset, it must fit before the size
6524 * of each device. If num_sectors is zero, we find the largest size
6527 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6533 rdev_for_each(rdev, mddev) {
6534 sector_t avail = rdev->sectors;
6536 if (fit && (num_sectors == 0 || num_sectors > avail))
6537 num_sectors = avail;
6538 if (avail < num_sectors)
6541 rv = mddev->pers->resize(mddev, num_sectors);
6543 revalidate_disk(mddev->gendisk);
6547 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6550 struct md_rdev *rdev;
6551 /* change the number of raid disks */
6552 if (mddev->pers->check_reshape == NULL)
6556 if (raid_disks <= 0 ||
6557 (mddev->max_disks && raid_disks >= mddev->max_disks))
6559 if (mddev->sync_thread ||
6560 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6561 mddev->reshape_position != MaxSector)
6564 rdev_for_each(rdev, mddev) {
6565 if (mddev->raid_disks < raid_disks &&
6566 rdev->data_offset < rdev->new_data_offset)
6568 if (mddev->raid_disks > raid_disks &&
6569 rdev->data_offset > rdev->new_data_offset)
6573 mddev->delta_disks = raid_disks - mddev->raid_disks;
6574 if (mddev->delta_disks < 0)
6575 mddev->reshape_backwards = 1;
6576 else if (mddev->delta_disks > 0)
6577 mddev->reshape_backwards = 0;
6579 rv = mddev->pers->check_reshape(mddev);
6581 mddev->delta_disks = 0;
6582 mddev->reshape_backwards = 0;
6588 * update_array_info is used to change the configuration of an
6590 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6591 * fields in the info are checked against the array.
6592 * Any differences that cannot be handled will cause an error.
6593 * Normally, only one change can be managed at a time.
6595 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6601 /* calculate expected state,ignoring low bits */
6602 if (mddev->bitmap && mddev->bitmap_info.offset)
6603 state |= (1 << MD_SB_BITMAP_PRESENT);
6605 if (mddev->major_version != info->major_version ||
6606 mddev->minor_version != info->minor_version ||
6607 /* mddev->patch_version != info->patch_version || */
6608 mddev->ctime != info->ctime ||
6609 mddev->level != info->level ||
6610 /* mddev->layout != info->layout || */
6611 mddev->persistent != !info->not_persistent ||
6612 mddev->chunk_sectors != info->chunk_size >> 9 ||
6613 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6614 ((state^info->state) & 0xfffffe00)
6617 /* Check there is only one change */
6618 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6620 if (mddev->raid_disks != info->raid_disks)
6622 if (mddev->layout != info->layout)
6624 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6631 if (mddev->layout != info->layout) {
6633 * we don't need to do anything at the md level, the
6634 * personality will take care of it all.
6636 if (mddev->pers->check_reshape == NULL)
6639 mddev->new_layout = info->layout;
6640 rv = mddev->pers->check_reshape(mddev);
6642 mddev->new_layout = mddev->layout;
6646 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6647 rv = update_size(mddev, (sector_t)info->size * 2);
6649 if (mddev->raid_disks != info->raid_disks)
6650 rv = update_raid_disks(mddev, info->raid_disks);
6652 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6653 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6657 if (mddev->recovery || mddev->sync_thread) {
6661 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6662 struct bitmap *bitmap;
6663 /* add the bitmap */
6664 if (mddev->bitmap) {
6668 if (mddev->bitmap_info.default_offset == 0) {
6672 mddev->bitmap_info.offset =
6673 mddev->bitmap_info.default_offset;
6674 mddev->bitmap_info.space =
6675 mddev->bitmap_info.default_space;
6676 mddev->pers->quiesce(mddev, 1);
6677 bitmap = bitmap_create(mddev, -1);
6678 if (!IS_ERR(bitmap)) {
6679 mddev->bitmap = bitmap;
6680 rv = bitmap_load(mddev);
6682 rv = PTR_ERR(bitmap);
6684 bitmap_destroy(mddev);
6685 mddev->pers->quiesce(mddev, 0);
6687 /* remove the bitmap */
6688 if (!mddev->bitmap) {
6692 if (mddev->bitmap->storage.file) {
6696 if (mddev->bitmap_info.nodes) {
6697 /* hold PW on all the bitmap lock */
6698 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6699 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6701 md_cluster_ops->unlock_all_bitmaps(mddev);
6705 mddev->bitmap_info.nodes = 0;
6706 md_cluster_ops->leave(mddev);
6708 mddev->pers->quiesce(mddev, 1);
6709 bitmap_destroy(mddev);
6710 mddev->pers->quiesce(mddev, 0);
6711 mddev->bitmap_info.offset = 0;
6714 md_update_sb(mddev, 1);
6720 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6722 struct md_rdev *rdev;
6725 if (mddev->pers == NULL)
6729 rdev = find_rdev_rcu(mddev, dev);
6733 md_error(mddev, rdev);
6734 if (!test_bit(Faulty, &rdev->flags))
6742 * We have a problem here : there is no easy way to give a CHS
6743 * virtual geometry. We currently pretend that we have a 2 heads
6744 * 4 sectors (with a BIG number of cylinders...). This drives
6745 * dosfs just mad... ;-)
6747 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6749 struct mddev *mddev = bdev->bd_disk->private_data;
6753 geo->cylinders = mddev->array_sectors / 8;
6757 static inline bool md_ioctl_valid(unsigned int cmd)
6762 case GET_ARRAY_INFO:
6763 case GET_BITMAP_FILE:
6766 case HOT_REMOVE_DISK:
6769 case RESTART_ARRAY_RW:
6771 case SET_ARRAY_INFO:
6772 case SET_BITMAP_FILE:
6773 case SET_DISK_FAULTY:
6776 case CLUSTERED_DISK_NACK:
6783 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6784 unsigned int cmd, unsigned long arg)
6787 void __user *argp = (void __user *)arg;
6788 struct mddev *mddev = NULL;
6791 if (!md_ioctl_valid(cmd))
6796 case GET_ARRAY_INFO:
6800 if (!capable(CAP_SYS_ADMIN))
6805 * Commands dealing with the RAID driver but not any
6810 err = get_version(argp);
6816 autostart_arrays(arg);
6823 * Commands creating/starting a new array:
6826 mddev = bdev->bd_disk->private_data;
6833 /* Some actions do not requires the mutex */
6835 case GET_ARRAY_INFO:
6836 if (!mddev->raid_disks && !mddev->external)
6839 err = get_array_info(mddev, argp);
6843 if (!mddev->raid_disks && !mddev->external)
6846 err = get_disk_info(mddev, argp);
6849 case SET_DISK_FAULTY:
6850 err = set_disk_faulty(mddev, new_decode_dev(arg));
6853 case GET_BITMAP_FILE:
6854 err = get_bitmap_file(mddev, argp);
6859 if (cmd == ADD_NEW_DISK)
6860 /* need to ensure md_delayed_delete() has completed */
6861 flush_workqueue(md_misc_wq);
6863 if (cmd == HOT_REMOVE_DISK)
6864 /* need to ensure recovery thread has run */
6865 wait_event_interruptible_timeout(mddev->sb_wait,
6866 !test_bit(MD_RECOVERY_NEEDED,
6868 msecs_to_jiffies(5000));
6869 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6870 /* Need to flush page cache, and ensure no-one else opens
6873 mutex_lock(&mddev->open_mutex);
6874 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6875 mutex_unlock(&mddev->open_mutex);
6879 set_bit(MD_CLOSING, &mddev->flags);
6880 mutex_unlock(&mddev->open_mutex);
6881 sync_blockdev(bdev);
6883 err = mddev_lock(mddev);
6885 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
6890 if (cmd == SET_ARRAY_INFO) {
6891 mdu_array_info_t info;
6893 memset(&info, 0, sizeof(info));
6894 else if (copy_from_user(&info, argp, sizeof(info))) {
6899 err = update_array_info(mddev, &info);
6901 pr_warn("md: couldn't update array info. %d\n", err);
6906 if (!list_empty(&mddev->disks)) {
6907 pr_warn("md: array %s already has disks!\n", mdname(mddev));
6911 if (mddev->raid_disks) {
6912 pr_warn("md: array %s already initialised!\n", mdname(mddev));
6916 err = set_array_info(mddev, &info);
6918 pr_warn("md: couldn't set array info. %d\n", err);
6925 * Commands querying/configuring an existing array:
6927 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6928 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6929 if ((!mddev->raid_disks && !mddev->external)
6930 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6931 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6932 && cmd != GET_BITMAP_FILE) {
6938 * Commands even a read-only array can execute:
6941 case RESTART_ARRAY_RW:
6942 err = restart_array(mddev);
6946 err = do_md_stop(mddev, 0, bdev);
6950 err = md_set_readonly(mddev, bdev);
6953 case HOT_REMOVE_DISK:
6954 err = hot_remove_disk(mddev, new_decode_dev(arg));
6958 /* We can support ADD_NEW_DISK on read-only arrays
6959 * only if we are re-adding a preexisting device.
6960 * So require mddev->pers and MD_DISK_SYNC.
6963 mdu_disk_info_t info;
6964 if (copy_from_user(&info, argp, sizeof(info)))
6966 else if (!(info.state & (1<<MD_DISK_SYNC)))
6967 /* Need to clear read-only for this */
6970 err = add_new_disk(mddev, &info);
6976 if (get_user(ro, (int __user *)(arg))) {
6982 /* if the bdev is going readonly the value of mddev->ro
6983 * does not matter, no writes are coming
6988 /* are we are already prepared for writes? */
6992 /* transitioning to readauto need only happen for
6993 * arrays that call md_write_start
6996 err = restart_array(mddev);
6999 set_disk_ro(mddev->gendisk, 0);
7006 * The remaining ioctls are changing the state of the
7007 * superblock, so we do not allow them on read-only arrays.
7009 if (mddev->ro && mddev->pers) {
7010 if (mddev->ro == 2) {
7012 sysfs_notify_dirent_safe(mddev->sysfs_state);
7013 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7014 /* mddev_unlock will wake thread */
7015 /* If a device failed while we were read-only, we
7016 * need to make sure the metadata is updated now.
7018 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7019 mddev_unlock(mddev);
7020 wait_event(mddev->sb_wait,
7021 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7022 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7023 mddev_lock_nointr(mddev);
7034 mdu_disk_info_t info;
7035 if (copy_from_user(&info, argp, sizeof(info)))
7038 err = add_new_disk(mddev, &info);
7042 case CLUSTERED_DISK_NACK:
7043 if (mddev_is_clustered(mddev))
7044 md_cluster_ops->new_disk_ack(mddev, false);
7050 err = hot_add_disk(mddev, new_decode_dev(arg));
7054 err = do_md_run(mddev);
7057 case SET_BITMAP_FILE:
7058 err = set_bitmap_file(mddev, (int)arg);
7067 if (mddev->hold_active == UNTIL_IOCTL &&
7069 mddev->hold_active = 0;
7070 mddev_unlock(mddev);
7074 #ifdef CONFIG_COMPAT
7075 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7076 unsigned int cmd, unsigned long arg)
7079 case HOT_REMOVE_DISK:
7081 case SET_DISK_FAULTY:
7082 case SET_BITMAP_FILE:
7083 /* These take in integer arg, do not convert */
7086 arg = (unsigned long)compat_ptr(arg);
7090 return md_ioctl(bdev, mode, cmd, arg);
7092 #endif /* CONFIG_COMPAT */
7094 static int md_open(struct block_device *bdev, fmode_t mode)
7097 * Succeed if we can lock the mddev, which confirms that
7098 * it isn't being stopped right now.
7100 struct mddev *mddev = mddev_find(bdev->bd_dev);
7106 if (mddev->gendisk != bdev->bd_disk) {
7107 /* we are racing with mddev_put which is discarding this
7111 /* Wait until bdev->bd_disk is definitely gone */
7112 flush_workqueue(md_misc_wq);
7113 /* Then retry the open from the top */
7114 return -ERESTARTSYS;
7116 BUG_ON(mddev != bdev->bd_disk->private_data);
7118 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7121 if (test_bit(MD_CLOSING, &mddev->flags)) {
7122 mutex_unlock(&mddev->open_mutex);
7128 atomic_inc(&mddev->openers);
7129 mutex_unlock(&mddev->open_mutex);
7131 check_disk_change(bdev);
7138 static void md_release(struct gendisk *disk, fmode_t mode)
7140 struct mddev *mddev = disk->private_data;
7143 atomic_dec(&mddev->openers);
7147 static int md_media_changed(struct gendisk *disk)
7149 struct mddev *mddev = disk->private_data;
7151 return mddev->changed;
7154 static int md_revalidate(struct gendisk *disk)
7156 struct mddev *mddev = disk->private_data;
7161 static const struct block_device_operations md_fops =
7163 .owner = THIS_MODULE,
7165 .release = md_release,
7167 #ifdef CONFIG_COMPAT
7168 .compat_ioctl = md_compat_ioctl,
7170 .getgeo = md_getgeo,
7171 .media_changed = md_media_changed,
7172 .revalidate_disk= md_revalidate,
7175 static int md_thread(void *arg)
7177 struct md_thread *thread = arg;
7180 * md_thread is a 'system-thread', it's priority should be very
7181 * high. We avoid resource deadlocks individually in each
7182 * raid personality. (RAID5 does preallocation) We also use RR and
7183 * the very same RT priority as kswapd, thus we will never get
7184 * into a priority inversion deadlock.
7186 * we definitely have to have equal or higher priority than
7187 * bdflush, otherwise bdflush will deadlock if there are too
7188 * many dirty RAID5 blocks.
7191 allow_signal(SIGKILL);
7192 while (!kthread_should_stop()) {
7194 /* We need to wait INTERRUPTIBLE so that
7195 * we don't add to the load-average.
7196 * That means we need to be sure no signals are
7199 if (signal_pending(current))
7200 flush_signals(current);
7202 wait_event_interruptible_timeout
7204 test_bit(THREAD_WAKEUP, &thread->flags)
7205 || kthread_should_stop() || kthread_should_park(),
7208 clear_bit(THREAD_WAKEUP, &thread->flags);
7209 if (kthread_should_park())
7211 if (!kthread_should_stop())
7212 thread->run(thread);
7218 void md_wakeup_thread(struct md_thread *thread)
7221 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7222 set_bit(THREAD_WAKEUP, &thread->flags);
7223 wake_up(&thread->wqueue);
7226 EXPORT_SYMBOL(md_wakeup_thread);
7228 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7229 struct mddev *mddev, const char *name)
7231 struct md_thread *thread;
7233 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7237 init_waitqueue_head(&thread->wqueue);
7240 thread->mddev = mddev;
7241 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7242 thread->tsk = kthread_run(md_thread, thread,
7244 mdname(thread->mddev),
7246 if (IS_ERR(thread->tsk)) {
7252 EXPORT_SYMBOL(md_register_thread);
7254 void md_unregister_thread(struct md_thread **threadp)
7256 struct md_thread *thread = *threadp;
7259 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7260 /* Locking ensures that mddev_unlock does not wake_up a
7261 * non-existent thread
7263 spin_lock(&pers_lock);
7265 spin_unlock(&pers_lock);
7267 kthread_stop(thread->tsk);
7270 EXPORT_SYMBOL(md_unregister_thread);
7272 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7274 if (!rdev || test_bit(Faulty, &rdev->flags))
7277 if (!mddev->pers || !mddev->pers->error_handler)
7279 mddev->pers->error_handler(mddev,rdev);
7280 if (mddev->degraded)
7281 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7282 sysfs_notify_dirent_safe(rdev->sysfs_state);
7283 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7284 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7285 md_wakeup_thread(mddev->thread);
7286 if (mddev->event_work.func)
7287 queue_work(md_misc_wq, &mddev->event_work);
7288 md_new_event(mddev);
7290 EXPORT_SYMBOL(md_error);
7292 /* seq_file implementation /proc/mdstat */
7294 static void status_unused(struct seq_file *seq)
7297 struct md_rdev *rdev;
7299 seq_printf(seq, "unused devices: ");
7301 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7302 char b[BDEVNAME_SIZE];
7304 seq_printf(seq, "%s ",
7305 bdevname(rdev->bdev,b));
7308 seq_printf(seq, "<none>");
7310 seq_printf(seq, "\n");
7313 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7315 sector_t max_sectors, resync, res;
7316 unsigned long dt, db;
7319 unsigned int per_milli;
7321 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7322 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7323 max_sectors = mddev->resync_max_sectors;
7325 max_sectors = mddev->dev_sectors;
7327 resync = mddev->curr_resync;
7329 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7330 /* Still cleaning up */
7331 resync = max_sectors;
7333 resync -= atomic_read(&mddev->recovery_active);
7336 if (mddev->recovery_cp < MaxSector) {
7337 seq_printf(seq, "\tresync=PENDING");
7343 seq_printf(seq, "\tresync=DELAYED");
7347 WARN_ON(max_sectors == 0);
7348 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7349 * in a sector_t, and (max_sectors>>scale) will fit in a
7350 * u32, as those are the requirements for sector_div.
7351 * Thus 'scale' must be at least 10
7354 if (sizeof(sector_t) > sizeof(unsigned long)) {
7355 while ( max_sectors/2 > (1ULL<<(scale+32)))
7358 res = (resync>>scale)*1000;
7359 sector_div(res, (u32)((max_sectors>>scale)+1));
7363 int i, x = per_milli/50, y = 20-x;
7364 seq_printf(seq, "[");
7365 for (i = 0; i < x; i++)
7366 seq_printf(seq, "=");
7367 seq_printf(seq, ">");
7368 for (i = 0; i < y; i++)
7369 seq_printf(seq, ".");
7370 seq_printf(seq, "] ");
7372 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7373 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7375 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7377 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7378 "resync" : "recovery"))),
7379 per_milli/10, per_milli % 10,
7380 (unsigned long long) resync/2,
7381 (unsigned long long) max_sectors/2);
7384 * dt: time from mark until now
7385 * db: blocks written from mark until now
7386 * rt: remaining time
7388 * rt is a sector_t, so could be 32bit or 64bit.
7389 * So we divide before multiply in case it is 32bit and close
7391 * We scale the divisor (db) by 32 to avoid losing precision
7392 * near the end of resync when the number of remaining sectors
7394 * We then divide rt by 32 after multiplying by db to compensate.
7395 * The '+1' avoids division by zero if db is very small.
7397 dt = ((jiffies - mddev->resync_mark) / HZ);
7399 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7400 - mddev->resync_mark_cnt;
7402 rt = max_sectors - resync; /* number of remaining sectors */
7403 sector_div(rt, db/32+1);
7407 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7408 ((unsigned long)rt % 60)/6);
7410 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7414 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7416 struct list_head *tmp;
7418 struct mddev *mddev;
7426 spin_lock(&all_mddevs_lock);
7427 list_for_each(tmp,&all_mddevs)
7429 mddev = list_entry(tmp, struct mddev, all_mddevs);
7431 spin_unlock(&all_mddevs_lock);
7434 spin_unlock(&all_mddevs_lock);
7436 return (void*)2;/* tail */
7440 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7442 struct list_head *tmp;
7443 struct mddev *next_mddev, *mddev = v;
7449 spin_lock(&all_mddevs_lock);
7451 tmp = all_mddevs.next;
7453 tmp = mddev->all_mddevs.next;
7454 if (tmp != &all_mddevs)
7455 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7457 next_mddev = (void*)2;
7460 spin_unlock(&all_mddevs_lock);
7468 static void md_seq_stop(struct seq_file *seq, void *v)
7470 struct mddev *mddev = v;
7472 if (mddev && v != (void*)1 && v != (void*)2)
7476 static int md_seq_show(struct seq_file *seq, void *v)
7478 struct mddev *mddev = v;
7480 struct md_rdev *rdev;
7482 if (v == (void*)1) {
7483 struct md_personality *pers;
7484 seq_printf(seq, "Personalities : ");
7485 spin_lock(&pers_lock);
7486 list_for_each_entry(pers, &pers_list, list)
7487 seq_printf(seq, "[%s] ", pers->name);
7489 spin_unlock(&pers_lock);
7490 seq_printf(seq, "\n");
7491 seq->poll_event = atomic_read(&md_event_count);
7494 if (v == (void*)2) {
7499 spin_lock(&mddev->lock);
7500 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7501 seq_printf(seq, "%s : %sactive", mdname(mddev),
7502 mddev->pers ? "" : "in");
7505 seq_printf(seq, " (read-only)");
7507 seq_printf(seq, " (auto-read-only)");
7508 seq_printf(seq, " %s", mddev->pers->name);
7513 rdev_for_each_rcu(rdev, mddev) {
7514 char b[BDEVNAME_SIZE];
7515 seq_printf(seq, " %s[%d]",
7516 bdevname(rdev->bdev,b), rdev->desc_nr);
7517 if (test_bit(WriteMostly, &rdev->flags))
7518 seq_printf(seq, "(W)");
7519 if (test_bit(Journal, &rdev->flags))
7520 seq_printf(seq, "(J)");
7521 if (test_bit(Faulty, &rdev->flags)) {
7522 seq_printf(seq, "(F)");
7525 if (rdev->raid_disk < 0)
7526 seq_printf(seq, "(S)"); /* spare */
7527 if (test_bit(Replacement, &rdev->flags))
7528 seq_printf(seq, "(R)");
7529 sectors += rdev->sectors;
7533 if (!list_empty(&mddev->disks)) {
7535 seq_printf(seq, "\n %llu blocks",
7536 (unsigned long long)
7537 mddev->array_sectors / 2);
7539 seq_printf(seq, "\n %llu blocks",
7540 (unsigned long long)sectors / 2);
7542 if (mddev->persistent) {
7543 if (mddev->major_version != 0 ||
7544 mddev->minor_version != 90) {
7545 seq_printf(seq," super %d.%d",
7546 mddev->major_version,
7547 mddev->minor_version);
7549 } else if (mddev->external)
7550 seq_printf(seq, " super external:%s",
7551 mddev->metadata_type);
7553 seq_printf(seq, " super non-persistent");
7556 mddev->pers->status(seq, mddev);
7557 seq_printf(seq, "\n ");
7558 if (mddev->pers->sync_request) {
7559 if (status_resync(seq, mddev))
7560 seq_printf(seq, "\n ");
7563 seq_printf(seq, "\n ");
7565 bitmap_status(seq, mddev->bitmap);
7567 seq_printf(seq, "\n");
7569 spin_unlock(&mddev->lock);
7574 static const struct seq_operations md_seq_ops = {
7575 .start = md_seq_start,
7576 .next = md_seq_next,
7577 .stop = md_seq_stop,
7578 .show = md_seq_show,
7581 static int md_seq_open(struct inode *inode, struct file *file)
7583 struct seq_file *seq;
7586 error = seq_open(file, &md_seq_ops);
7590 seq = file->private_data;
7591 seq->poll_event = atomic_read(&md_event_count);
7595 static int md_unloading;
7596 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7598 struct seq_file *seq = filp->private_data;
7602 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7603 poll_wait(filp, &md_event_waiters, wait);
7605 /* always allow read */
7606 mask = POLLIN | POLLRDNORM;
7608 if (seq->poll_event != atomic_read(&md_event_count))
7609 mask |= POLLERR | POLLPRI;
7613 static const struct file_operations md_seq_fops = {
7614 .owner = THIS_MODULE,
7615 .open = md_seq_open,
7617 .llseek = seq_lseek,
7618 .release = seq_release_private,
7619 .poll = mdstat_poll,
7622 int register_md_personality(struct md_personality *p)
7624 pr_debug("md: %s personality registered for level %d\n",
7626 spin_lock(&pers_lock);
7627 list_add_tail(&p->list, &pers_list);
7628 spin_unlock(&pers_lock);
7631 EXPORT_SYMBOL(register_md_personality);
7633 int unregister_md_personality(struct md_personality *p)
7635 pr_debug("md: %s personality unregistered\n", p->name);
7636 spin_lock(&pers_lock);
7637 list_del_init(&p->list);
7638 spin_unlock(&pers_lock);
7641 EXPORT_SYMBOL(unregister_md_personality);
7643 int register_md_cluster_operations(struct md_cluster_operations *ops,
7644 struct module *module)
7647 spin_lock(&pers_lock);
7648 if (md_cluster_ops != NULL)
7651 md_cluster_ops = ops;
7652 md_cluster_mod = module;
7654 spin_unlock(&pers_lock);
7657 EXPORT_SYMBOL(register_md_cluster_operations);
7659 int unregister_md_cluster_operations(void)
7661 spin_lock(&pers_lock);
7662 md_cluster_ops = NULL;
7663 spin_unlock(&pers_lock);
7666 EXPORT_SYMBOL(unregister_md_cluster_operations);
7668 int md_setup_cluster(struct mddev *mddev, int nodes)
7670 if (!md_cluster_ops)
7671 request_module("md-cluster");
7672 spin_lock(&pers_lock);
7673 /* ensure module won't be unloaded */
7674 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7675 pr_warn("can't find md-cluster module or get it's reference.\n");
7676 spin_unlock(&pers_lock);
7679 spin_unlock(&pers_lock);
7681 return md_cluster_ops->join(mddev, nodes);
7684 void md_cluster_stop(struct mddev *mddev)
7686 if (!md_cluster_ops)
7688 md_cluster_ops->leave(mddev);
7689 module_put(md_cluster_mod);
7692 static int is_mddev_idle(struct mddev *mddev, int init)
7694 struct md_rdev *rdev;
7700 rdev_for_each_rcu(rdev, mddev) {
7701 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7702 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7703 (int)part_stat_read(&disk->part0, sectors[1]) -
7704 atomic_read(&disk->sync_io);
7705 /* sync IO will cause sync_io to increase before the disk_stats
7706 * as sync_io is counted when a request starts, and
7707 * disk_stats is counted when it completes.
7708 * So resync activity will cause curr_events to be smaller than
7709 * when there was no such activity.
7710 * non-sync IO will cause disk_stat to increase without
7711 * increasing sync_io so curr_events will (eventually)
7712 * be larger than it was before. Once it becomes
7713 * substantially larger, the test below will cause
7714 * the array to appear non-idle, and resync will slow
7716 * If there is a lot of outstanding resync activity when
7717 * we set last_event to curr_events, then all that activity
7718 * completing might cause the array to appear non-idle
7719 * and resync will be slowed down even though there might
7720 * not have been non-resync activity. This will only
7721 * happen once though. 'last_events' will soon reflect
7722 * the state where there is little or no outstanding
7723 * resync requests, and further resync activity will
7724 * always make curr_events less than last_events.
7727 if (init || curr_events - rdev->last_events > 64) {
7728 rdev->last_events = curr_events;
7736 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7738 /* another "blocks" (512byte) blocks have been synced */
7739 atomic_sub(blocks, &mddev->recovery_active);
7740 wake_up(&mddev->recovery_wait);
7742 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7743 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7744 md_wakeup_thread(mddev->thread);
7745 // stop recovery, signal do_sync ....
7748 EXPORT_SYMBOL(md_done_sync);
7750 /* md_write_start(mddev, bi)
7751 * If we need to update some array metadata (e.g. 'active' flag
7752 * in superblock) before writing, schedule a superblock update
7753 * and wait for it to complete.
7755 void md_write_start(struct mddev *mddev, struct bio *bi)
7758 if (bio_data_dir(bi) != WRITE)
7761 BUG_ON(mddev->ro == 1);
7762 if (mddev->ro == 2) {
7763 /* need to switch to read/write */
7765 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7766 md_wakeup_thread(mddev->thread);
7767 md_wakeup_thread(mddev->sync_thread);
7770 atomic_inc(&mddev->writes_pending);
7771 if (mddev->safemode == 1)
7772 mddev->safemode = 0;
7773 if (mddev->in_sync) {
7774 spin_lock(&mddev->lock);
7775 if (mddev->in_sync) {
7777 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7778 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7779 md_wakeup_thread(mddev->thread);
7782 spin_unlock(&mddev->lock);
7785 sysfs_notify_dirent_safe(mddev->sysfs_state);
7786 wait_event(mddev->sb_wait,
7787 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7789 EXPORT_SYMBOL(md_write_start);
7791 void md_write_end(struct mddev *mddev)
7793 if (atomic_dec_and_test(&mddev->writes_pending)) {
7794 if (mddev->safemode == 2)
7795 md_wakeup_thread(mddev->thread);
7796 else if (mddev->safemode_delay)
7797 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7800 EXPORT_SYMBOL(md_write_end);
7802 /* md_allow_write(mddev)
7803 * Calling this ensures that the array is marked 'active' so that writes
7804 * may proceed without blocking. It is important to call this before
7805 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7806 * Must be called with mddev_lock held.
7808 * In the ->external case MD_SB_CHANGE_PENDING can not be cleared until mddev->lock
7809 * is dropped, so return -EAGAIN after notifying userspace.
7811 int md_allow_write(struct mddev *mddev)
7817 if (!mddev->pers->sync_request)
7820 spin_lock(&mddev->lock);
7821 if (mddev->in_sync) {
7823 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7824 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7825 if (mddev->safemode_delay &&
7826 mddev->safemode == 0)
7827 mddev->safemode = 1;
7828 spin_unlock(&mddev->lock);
7829 md_update_sb(mddev, 0);
7830 sysfs_notify_dirent_safe(mddev->sysfs_state);
7832 spin_unlock(&mddev->lock);
7834 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
7839 EXPORT_SYMBOL_GPL(md_allow_write);
7841 #define SYNC_MARKS 10
7842 #define SYNC_MARK_STEP (3*HZ)
7843 #define UPDATE_FREQUENCY (5*60*HZ)
7844 void md_do_sync(struct md_thread *thread)
7846 struct mddev *mddev = thread->mddev;
7847 struct mddev *mddev2;
7848 unsigned int currspeed = 0,
7850 sector_t max_sectors,j, io_sectors, recovery_done;
7851 unsigned long mark[SYNC_MARKS];
7852 unsigned long update_time;
7853 sector_t mark_cnt[SYNC_MARKS];
7855 struct list_head *tmp;
7856 sector_t last_check;
7858 struct md_rdev *rdev;
7859 char *desc, *action = NULL;
7860 struct blk_plug plug;
7863 /* just incase thread restarts... */
7864 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7866 if (mddev->ro) {/* never try to sync a read-only array */
7867 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7871 if (mddev_is_clustered(mddev)) {
7872 ret = md_cluster_ops->resync_start(mddev);
7876 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7877 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7878 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7879 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7880 && ((unsigned long long)mddev->curr_resync_completed
7881 < (unsigned long long)mddev->resync_max_sectors))
7885 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7886 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7887 desc = "data-check";
7889 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7890 desc = "requested-resync";
7894 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7899 mddev->last_sync_action = action ?: desc;
7901 /* we overload curr_resync somewhat here.
7902 * 0 == not engaged in resync at all
7903 * 2 == checking that there is no conflict with another sync
7904 * 1 == like 2, but have yielded to allow conflicting resync to
7906 * other == active in resync - this many blocks
7908 * Before starting a resync we must have set curr_resync to
7909 * 2, and then checked that every "conflicting" array has curr_resync
7910 * less than ours. When we find one that is the same or higher
7911 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7912 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7913 * This will mean we have to start checking from the beginning again.
7918 int mddev2_minor = -1;
7919 mddev->curr_resync = 2;
7922 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7924 for_each_mddev(mddev2, tmp) {
7925 if (mddev2 == mddev)
7927 if (!mddev->parallel_resync
7928 && mddev2->curr_resync
7929 && match_mddev_units(mddev, mddev2)) {
7931 if (mddev < mddev2 && mddev->curr_resync == 2) {
7932 /* arbitrarily yield */
7933 mddev->curr_resync = 1;
7934 wake_up(&resync_wait);
7936 if (mddev > mddev2 && mddev->curr_resync == 1)
7937 /* no need to wait here, we can wait the next
7938 * time 'round when curr_resync == 2
7941 /* We need to wait 'interruptible' so as not to
7942 * contribute to the load average, and not to
7943 * be caught by 'softlockup'
7945 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7946 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7947 mddev2->curr_resync >= mddev->curr_resync) {
7948 if (mddev2_minor != mddev2->md_minor) {
7949 mddev2_minor = mddev2->md_minor;
7950 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
7951 desc, mdname(mddev),
7955 if (signal_pending(current))
7956 flush_signals(current);
7958 finish_wait(&resync_wait, &wq);
7961 finish_wait(&resync_wait, &wq);
7964 } while (mddev->curr_resync < 2);
7967 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7968 /* resync follows the size requested by the personality,
7969 * which defaults to physical size, but can be virtual size
7971 max_sectors = mddev->resync_max_sectors;
7972 atomic64_set(&mddev->resync_mismatches, 0);
7973 /* we don't use the checkpoint if there's a bitmap */
7974 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7975 j = mddev->resync_min;
7976 else if (!mddev->bitmap)
7977 j = mddev->recovery_cp;
7979 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7980 max_sectors = mddev->resync_max_sectors;
7982 /* recovery follows the physical size of devices */
7983 max_sectors = mddev->dev_sectors;
7986 rdev_for_each_rcu(rdev, mddev)
7987 if (rdev->raid_disk >= 0 &&
7988 !test_bit(Journal, &rdev->flags) &&
7989 !test_bit(Faulty, &rdev->flags) &&
7990 !test_bit(In_sync, &rdev->flags) &&
7991 rdev->recovery_offset < j)
7992 j = rdev->recovery_offset;
7995 /* If there is a bitmap, we need to make sure all
7996 * writes that started before we added a spare
7997 * complete before we start doing a recovery.
7998 * Otherwise the write might complete and (via
7999 * bitmap_endwrite) set a bit in the bitmap after the
8000 * recovery has checked that bit and skipped that
8003 if (mddev->bitmap) {
8004 mddev->pers->quiesce(mddev, 1);
8005 mddev->pers->quiesce(mddev, 0);
8009 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8010 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8011 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8012 speed_max(mddev), desc);
8014 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8017 for (m = 0; m < SYNC_MARKS; m++) {
8019 mark_cnt[m] = io_sectors;
8022 mddev->resync_mark = mark[last_mark];
8023 mddev->resync_mark_cnt = mark_cnt[last_mark];
8026 * Tune reconstruction:
8028 window = 32*(PAGE_SIZE/512);
8029 pr_debug("md: using %dk window, over a total of %lluk.\n",
8030 window/2, (unsigned long long)max_sectors/2);
8032 atomic_set(&mddev->recovery_active, 0);
8036 pr_debug("md: resuming %s of %s from checkpoint.\n",
8037 desc, mdname(mddev));
8038 mddev->curr_resync = j;
8040 mddev->curr_resync = 3; /* no longer delayed */
8041 mddev->curr_resync_completed = j;
8042 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8043 md_new_event(mddev);
8044 update_time = jiffies;
8046 blk_start_plug(&plug);
8047 while (j < max_sectors) {
8052 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8053 ((mddev->curr_resync > mddev->curr_resync_completed &&
8054 (mddev->curr_resync - mddev->curr_resync_completed)
8055 > (max_sectors >> 4)) ||
8056 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8057 (j - mddev->curr_resync_completed)*2
8058 >= mddev->resync_max - mddev->curr_resync_completed ||
8059 mddev->curr_resync_completed > mddev->resync_max
8061 /* time to update curr_resync_completed */
8062 wait_event(mddev->recovery_wait,
8063 atomic_read(&mddev->recovery_active) == 0);
8064 mddev->curr_resync_completed = j;
8065 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8066 j > mddev->recovery_cp)
8067 mddev->recovery_cp = j;
8068 update_time = jiffies;
8069 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8070 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8073 while (j >= mddev->resync_max &&
8074 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8075 /* As this condition is controlled by user-space,
8076 * we can block indefinitely, so use '_interruptible'
8077 * to avoid triggering warnings.
8079 flush_signals(current); /* just in case */
8080 wait_event_interruptible(mddev->recovery_wait,
8081 mddev->resync_max > j
8082 || test_bit(MD_RECOVERY_INTR,
8086 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8089 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8091 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8095 if (!skipped) { /* actual IO requested */
8096 io_sectors += sectors;
8097 atomic_add(sectors, &mddev->recovery_active);
8100 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8104 if (j > max_sectors)
8105 /* when skipping, extra large numbers can be returned. */
8108 mddev->curr_resync = j;
8109 mddev->curr_mark_cnt = io_sectors;
8110 if (last_check == 0)
8111 /* this is the earliest that rebuild will be
8112 * visible in /proc/mdstat
8114 md_new_event(mddev);
8116 if (last_check + window > io_sectors || j == max_sectors)
8119 last_check = io_sectors;
8121 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8123 int next = (last_mark+1) % SYNC_MARKS;
8125 mddev->resync_mark = mark[next];
8126 mddev->resync_mark_cnt = mark_cnt[next];
8127 mark[next] = jiffies;
8128 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8132 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8136 * this loop exits only if either when we are slower than
8137 * the 'hard' speed limit, or the system was IO-idle for
8139 * the system might be non-idle CPU-wise, but we only care
8140 * about not overloading the IO subsystem. (things like an
8141 * e2fsck being done on the RAID array should execute fast)
8145 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8146 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8147 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8149 if (currspeed > speed_min(mddev)) {
8150 if (currspeed > speed_max(mddev)) {
8154 if (!is_mddev_idle(mddev, 0)) {
8156 * Give other IO more of a chance.
8157 * The faster the devices, the less we wait.
8159 wait_event(mddev->recovery_wait,
8160 !atomic_read(&mddev->recovery_active));
8164 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8165 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8166 ? "interrupted" : "done");
8168 * this also signals 'finished resyncing' to md_stop
8170 blk_finish_plug(&plug);
8171 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8173 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8174 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8175 mddev->curr_resync > 3) {
8176 mddev->curr_resync_completed = mddev->curr_resync;
8177 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8179 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8181 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8182 mddev->curr_resync > 3) {
8183 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8184 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8185 if (mddev->curr_resync >= mddev->recovery_cp) {
8186 pr_debug("md: checkpointing %s of %s.\n",
8187 desc, mdname(mddev));
8188 if (test_bit(MD_RECOVERY_ERROR,
8190 mddev->recovery_cp =
8191 mddev->curr_resync_completed;
8193 mddev->recovery_cp =
8197 mddev->recovery_cp = MaxSector;
8199 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8200 mddev->curr_resync = MaxSector;
8202 rdev_for_each_rcu(rdev, mddev)
8203 if (rdev->raid_disk >= 0 &&
8204 mddev->delta_disks >= 0 &&
8205 !test_bit(Journal, &rdev->flags) &&
8206 !test_bit(Faulty, &rdev->flags) &&
8207 !test_bit(In_sync, &rdev->flags) &&
8208 rdev->recovery_offset < mddev->curr_resync)
8209 rdev->recovery_offset = mddev->curr_resync;
8214 /* set CHANGE_PENDING here since maybe another update is needed,
8215 * so other nodes are informed. It should be harmless for normal
8217 set_mask_bits(&mddev->sb_flags, 0,
8218 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8220 spin_lock(&mddev->lock);
8221 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8222 /* We completed so min/max setting can be forgotten if used. */
8223 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8224 mddev->resync_min = 0;
8225 mddev->resync_max = MaxSector;
8226 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8227 mddev->resync_min = mddev->curr_resync_completed;
8228 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8229 mddev->curr_resync = 0;
8230 spin_unlock(&mddev->lock);
8232 wake_up(&resync_wait);
8233 md_wakeup_thread(mddev->thread);
8236 EXPORT_SYMBOL_GPL(md_do_sync);
8238 static int remove_and_add_spares(struct mddev *mddev,
8239 struct md_rdev *this)
8241 struct md_rdev *rdev;
8244 bool remove_some = false;
8246 rdev_for_each(rdev, mddev) {
8247 if ((this == NULL || rdev == this) &&
8248 rdev->raid_disk >= 0 &&
8249 !test_bit(Blocked, &rdev->flags) &&
8250 test_bit(Faulty, &rdev->flags) &&
8251 atomic_read(&rdev->nr_pending)==0) {
8252 /* Faulty non-Blocked devices with nr_pending == 0
8253 * never get nr_pending incremented,
8254 * never get Faulty cleared, and never get Blocked set.
8255 * So we can synchronize_rcu now rather than once per device
8258 set_bit(RemoveSynchronized, &rdev->flags);
8264 rdev_for_each(rdev, mddev) {
8265 if ((this == NULL || rdev == this) &&
8266 rdev->raid_disk >= 0 &&
8267 !test_bit(Blocked, &rdev->flags) &&
8268 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8269 (!test_bit(In_sync, &rdev->flags) &&
8270 !test_bit(Journal, &rdev->flags))) &&
8271 atomic_read(&rdev->nr_pending)==0)) {
8272 if (mddev->pers->hot_remove_disk(
8273 mddev, rdev) == 0) {
8274 sysfs_unlink_rdev(mddev, rdev);
8275 rdev->raid_disk = -1;
8279 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8280 clear_bit(RemoveSynchronized, &rdev->flags);
8283 if (removed && mddev->kobj.sd)
8284 sysfs_notify(&mddev->kobj, NULL, "degraded");
8286 if (this && removed)
8289 rdev_for_each(rdev, mddev) {
8290 if (this && this != rdev)
8292 if (test_bit(Candidate, &rdev->flags))
8294 if (rdev->raid_disk >= 0 &&
8295 !test_bit(In_sync, &rdev->flags) &&
8296 !test_bit(Journal, &rdev->flags) &&
8297 !test_bit(Faulty, &rdev->flags))
8299 if (rdev->raid_disk >= 0)
8301 if (test_bit(Faulty, &rdev->flags))
8303 if (!test_bit(Journal, &rdev->flags)) {
8305 ! (rdev->saved_raid_disk >= 0 &&
8306 !test_bit(Bitmap_sync, &rdev->flags)))
8309 rdev->recovery_offset = 0;
8312 hot_add_disk(mddev, rdev) == 0) {
8313 if (sysfs_link_rdev(mddev, rdev))
8314 /* failure here is OK */;
8315 if (!test_bit(Journal, &rdev->flags))
8317 md_new_event(mddev);
8318 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8323 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8327 static void md_start_sync(struct work_struct *ws)
8329 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8331 mddev->sync_thread = md_register_thread(md_do_sync,
8334 if (!mddev->sync_thread) {
8335 pr_warn("%s: could not start resync thread...\n",
8337 /* leave the spares where they are, it shouldn't hurt */
8338 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8339 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8340 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8341 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8342 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8343 wake_up(&resync_wait);
8344 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8346 if (mddev->sysfs_action)
8347 sysfs_notify_dirent_safe(mddev->sysfs_action);
8349 md_wakeup_thread(mddev->sync_thread);
8350 sysfs_notify_dirent_safe(mddev->sysfs_action);
8351 md_new_event(mddev);
8355 * This routine is regularly called by all per-raid-array threads to
8356 * deal with generic issues like resync and super-block update.
8357 * Raid personalities that don't have a thread (linear/raid0) do not
8358 * need this as they never do any recovery or update the superblock.
8360 * It does not do any resync itself, but rather "forks" off other threads
8361 * to do that as needed.
8362 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8363 * "->recovery" and create a thread at ->sync_thread.
8364 * When the thread finishes it sets MD_RECOVERY_DONE
8365 * and wakeups up this thread which will reap the thread and finish up.
8366 * This thread also removes any faulty devices (with nr_pending == 0).
8368 * The overall approach is:
8369 * 1/ if the superblock needs updating, update it.
8370 * 2/ If a recovery thread is running, don't do anything else.
8371 * 3/ If recovery has finished, clean up, possibly marking spares active.
8372 * 4/ If there are any faulty devices, remove them.
8373 * 5/ If array is degraded, try to add spares devices
8374 * 6/ If array has spares or is not in-sync, start a resync thread.
8376 void md_check_recovery(struct mddev *mddev)
8378 if (mddev->suspended)
8382 bitmap_daemon_work(mddev);
8384 if (signal_pending(current)) {
8385 if (mddev->pers->sync_request && !mddev->external) {
8386 pr_debug("md: %s in immediate safe mode\n",
8388 mddev->safemode = 2;
8390 flush_signals(current);
8393 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8396 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8397 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8398 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8399 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8400 (mddev->external == 0 && mddev->safemode == 1) ||
8401 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8402 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8406 if (mddev_trylock(mddev)) {
8410 struct md_rdev *rdev;
8411 if (!mddev->external && mddev->in_sync)
8412 /* 'Blocked' flag not needed as failed devices
8413 * will be recorded if array switched to read/write.
8414 * Leaving it set will prevent the device
8415 * from being removed.
8417 rdev_for_each(rdev, mddev)
8418 clear_bit(Blocked, &rdev->flags);
8419 /* On a read-only array we can:
8420 * - remove failed devices
8421 * - add already-in_sync devices if the array itself
8423 * As we only add devices that are already in-sync,
8424 * we can activate the spares immediately.
8426 remove_and_add_spares(mddev, NULL);
8427 /* There is no thread, but we need to call
8428 * ->spare_active and clear saved_raid_disk
8430 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8431 md_reap_sync_thread(mddev);
8432 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8433 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8434 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8438 if (mddev_is_clustered(mddev)) {
8439 struct md_rdev *rdev;
8440 /* kick the device if another node issued a
8443 rdev_for_each(rdev, mddev) {
8444 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8445 rdev->raid_disk < 0)
8446 md_kick_rdev_from_array(rdev);
8449 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8450 md_reload_sb(mddev, mddev->good_device_nr);
8453 if (!mddev->external) {
8455 spin_lock(&mddev->lock);
8456 if (mddev->safemode &&
8457 !atomic_read(&mddev->writes_pending) &&
8459 mddev->recovery_cp == MaxSector) {
8462 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8464 if (mddev->safemode == 1)
8465 mddev->safemode = 0;
8466 spin_unlock(&mddev->lock);
8468 sysfs_notify_dirent_safe(mddev->sysfs_state);
8471 if (mddev->sb_flags)
8472 md_update_sb(mddev, 0);
8474 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8475 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8476 /* resync/recovery still happening */
8477 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8480 if (mddev->sync_thread) {
8481 md_reap_sync_thread(mddev);
8484 /* Set RUNNING before clearing NEEDED to avoid
8485 * any transients in the value of "sync_action".
8487 mddev->curr_resync_completed = 0;
8488 spin_lock(&mddev->lock);
8489 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8490 spin_unlock(&mddev->lock);
8491 /* Clear some bits that don't mean anything, but
8494 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8495 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8497 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8498 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8500 /* no recovery is running.
8501 * remove any failed drives, then
8502 * add spares if possible.
8503 * Spares are also removed and re-added, to allow
8504 * the personality to fail the re-add.
8507 if (mddev->reshape_position != MaxSector) {
8508 if (mddev->pers->check_reshape == NULL ||
8509 mddev->pers->check_reshape(mddev) != 0)
8510 /* Cannot proceed */
8512 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8513 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8514 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8515 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8516 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8517 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8518 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8519 } else if (mddev->recovery_cp < MaxSector) {
8520 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8521 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8522 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8523 /* nothing to be done ... */
8526 if (mddev->pers->sync_request) {
8528 /* We are adding a device or devices to an array
8529 * which has the bitmap stored on all devices.
8530 * So make sure all bitmap pages get written
8532 bitmap_write_all(mddev->bitmap);
8534 INIT_WORK(&mddev->del_work, md_start_sync);
8535 queue_work(md_misc_wq, &mddev->del_work);
8539 if (!mddev->sync_thread) {
8540 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8541 wake_up(&resync_wait);
8542 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8544 if (mddev->sysfs_action)
8545 sysfs_notify_dirent_safe(mddev->sysfs_action);
8548 wake_up(&mddev->sb_wait);
8549 mddev_unlock(mddev);
8552 EXPORT_SYMBOL(md_check_recovery);
8554 void md_reap_sync_thread(struct mddev *mddev)
8556 struct md_rdev *rdev;
8558 /* resync has finished, collect result */
8559 md_unregister_thread(&mddev->sync_thread);
8560 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8561 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8563 /* activate any spares */
8564 if (mddev->pers->spare_active(mddev)) {
8565 sysfs_notify(&mddev->kobj, NULL,
8567 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8570 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8571 mddev->pers->finish_reshape)
8572 mddev->pers->finish_reshape(mddev);
8574 /* If array is no-longer degraded, then any saved_raid_disk
8575 * information must be scrapped.
8577 if (!mddev->degraded)
8578 rdev_for_each(rdev, mddev)
8579 rdev->saved_raid_disk = -1;
8581 md_update_sb(mddev, 1);
8582 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8583 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8585 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8586 md_cluster_ops->resync_finish(mddev);
8587 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8588 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8589 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8590 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8591 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8592 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8593 wake_up(&resync_wait);
8594 /* flag recovery needed just to double check */
8595 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8596 sysfs_notify_dirent_safe(mddev->sysfs_action);
8597 md_new_event(mddev);
8598 if (mddev->event_work.func)
8599 queue_work(md_misc_wq, &mddev->event_work);
8601 EXPORT_SYMBOL(md_reap_sync_thread);
8603 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8605 sysfs_notify_dirent_safe(rdev->sysfs_state);
8606 wait_event_timeout(rdev->blocked_wait,
8607 !test_bit(Blocked, &rdev->flags) &&
8608 !test_bit(BlockedBadBlocks, &rdev->flags),
8609 msecs_to_jiffies(5000));
8610 rdev_dec_pending(rdev, mddev);
8612 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8614 void md_finish_reshape(struct mddev *mddev)
8616 /* called be personality module when reshape completes. */
8617 struct md_rdev *rdev;
8619 rdev_for_each(rdev, mddev) {
8620 if (rdev->data_offset > rdev->new_data_offset)
8621 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8623 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8624 rdev->data_offset = rdev->new_data_offset;
8627 EXPORT_SYMBOL(md_finish_reshape);
8629 /* Bad block management */
8631 /* Returns 1 on success, 0 on failure */
8632 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8635 struct mddev *mddev = rdev->mddev;
8638 s += rdev->new_data_offset;
8640 s += rdev->data_offset;
8641 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8643 /* Make sure they get written out promptly */
8644 if (test_bit(ExternalBbl, &rdev->flags))
8645 sysfs_notify(&rdev->kobj, NULL,
8646 "unacknowledged_bad_blocks");
8647 sysfs_notify_dirent_safe(rdev->sysfs_state);
8648 set_mask_bits(&mddev->sb_flags, 0,
8649 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
8650 md_wakeup_thread(rdev->mddev->thread);
8655 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8657 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8662 s += rdev->new_data_offset;
8664 s += rdev->data_offset;
8665 rv = badblocks_clear(&rdev->badblocks, s, sectors);
8666 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
8667 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
8670 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8672 static int md_notify_reboot(struct notifier_block *this,
8673 unsigned long code, void *x)
8675 struct list_head *tmp;
8676 struct mddev *mddev;
8679 for_each_mddev(mddev, tmp) {
8680 if (mddev_trylock(mddev)) {
8682 __md_stop_writes(mddev);
8683 if (mddev->persistent)
8684 mddev->safemode = 2;
8685 mddev_unlock(mddev);
8690 * certain more exotic SCSI devices are known to be
8691 * volatile wrt too early system reboots. While the
8692 * right place to handle this issue is the given
8693 * driver, we do want to have a safe RAID driver ...
8701 static struct notifier_block md_notifier = {
8702 .notifier_call = md_notify_reboot,
8704 .priority = INT_MAX, /* before any real devices */
8707 static void md_geninit(void)
8709 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8711 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8714 static int __init md_init(void)
8718 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8722 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8726 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8729 if ((ret = register_blkdev(0, "mdp")) < 0)
8733 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8734 md_probe, NULL, NULL);
8735 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8736 md_probe, NULL, NULL);
8738 register_reboot_notifier(&md_notifier);
8739 raid_table_header = register_sysctl_table(raid_root_table);
8745 unregister_blkdev(MD_MAJOR, "md");
8747 destroy_workqueue(md_misc_wq);
8749 destroy_workqueue(md_wq);
8754 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8756 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8757 struct md_rdev *rdev2;
8759 char b[BDEVNAME_SIZE];
8761 /* Check for change of roles in the active devices */
8762 rdev_for_each(rdev2, mddev) {
8763 if (test_bit(Faulty, &rdev2->flags))
8766 /* Check if the roles changed */
8767 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8769 if (test_bit(Candidate, &rdev2->flags)) {
8770 if (role == 0xfffe) {
8771 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8772 md_kick_rdev_from_array(rdev2);
8776 clear_bit(Candidate, &rdev2->flags);
8779 if (role != rdev2->raid_disk) {
8781 if (rdev2->raid_disk == -1 && role != 0xffff) {
8782 rdev2->saved_raid_disk = role;
8783 ret = remove_and_add_spares(mddev, rdev2);
8784 pr_info("Activated spare: %s\n",
8785 bdevname(rdev2->bdev,b));
8786 /* wakeup mddev->thread here, so array could
8787 * perform resync with the new activated disk */
8788 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8789 md_wakeup_thread(mddev->thread);
8793 * We just want to do the minimum to mark the disk
8794 * as faulty. The recovery is performed by the
8795 * one who initiated the error.
8797 if ((role == 0xfffe) || (role == 0xfffd)) {
8798 md_error(mddev, rdev2);
8799 clear_bit(Blocked, &rdev2->flags);
8804 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8805 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8807 /* Finally set the event to be up to date */
8808 mddev->events = le64_to_cpu(sb->events);
8811 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8814 struct page *swapout = rdev->sb_page;
8815 struct mdp_superblock_1 *sb;
8817 /* Store the sb page of the rdev in the swapout temporary
8818 * variable in case we err in the future
8820 rdev->sb_page = NULL;
8821 err = alloc_disk_sb(rdev);
8823 ClearPageUptodate(rdev->sb_page);
8824 rdev->sb_loaded = 0;
8825 err = super_types[mddev->major_version].
8826 load_super(rdev, NULL, mddev->minor_version);
8829 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8830 __func__, __LINE__, rdev->desc_nr, err);
8832 put_page(rdev->sb_page);
8833 rdev->sb_page = swapout;
8834 rdev->sb_loaded = 1;
8838 sb = page_address(rdev->sb_page);
8839 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8843 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8844 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8846 /* The other node finished recovery, call spare_active to set
8847 * device In_sync and mddev->degraded
8849 if (rdev->recovery_offset == MaxSector &&
8850 !test_bit(In_sync, &rdev->flags) &&
8851 mddev->pers->spare_active(mddev))
8852 sysfs_notify(&mddev->kobj, NULL, "degraded");
8858 void md_reload_sb(struct mddev *mddev, int nr)
8860 struct md_rdev *rdev;
8864 rdev_for_each_rcu(rdev, mddev) {
8865 if (rdev->desc_nr == nr)
8869 if (!rdev || rdev->desc_nr != nr) {
8870 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8874 err = read_rdev(mddev, rdev);
8878 check_sb_changes(mddev, rdev);
8880 /* Read all rdev's to update recovery_offset */
8881 rdev_for_each_rcu(rdev, mddev)
8882 read_rdev(mddev, rdev);
8884 EXPORT_SYMBOL(md_reload_sb);
8889 * Searches all registered partitions for autorun RAID arrays
8893 static DEFINE_MUTEX(detected_devices_mutex);
8894 static LIST_HEAD(all_detected_devices);
8895 struct detected_devices_node {
8896 struct list_head list;
8900 void md_autodetect_dev(dev_t dev)
8902 struct detected_devices_node *node_detected_dev;
8904 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8905 if (node_detected_dev) {
8906 node_detected_dev->dev = dev;
8907 mutex_lock(&detected_devices_mutex);
8908 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8909 mutex_unlock(&detected_devices_mutex);
8913 static void autostart_arrays(int part)
8915 struct md_rdev *rdev;
8916 struct detected_devices_node *node_detected_dev;
8918 int i_scanned, i_passed;
8923 pr_info("md: Autodetecting RAID arrays.\n");
8925 mutex_lock(&detected_devices_mutex);
8926 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8928 node_detected_dev = list_entry(all_detected_devices.next,
8929 struct detected_devices_node, list);
8930 list_del(&node_detected_dev->list);
8931 dev = node_detected_dev->dev;
8932 kfree(node_detected_dev);
8933 mutex_unlock(&detected_devices_mutex);
8934 rdev = md_import_device(dev,0, 90);
8935 mutex_lock(&detected_devices_mutex);
8939 if (test_bit(Faulty, &rdev->flags))
8942 set_bit(AutoDetected, &rdev->flags);
8943 list_add(&rdev->same_set, &pending_raid_disks);
8946 mutex_unlock(&detected_devices_mutex);
8948 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
8950 autorun_devices(part);
8953 #endif /* !MODULE */
8955 static __exit void md_exit(void)
8957 struct mddev *mddev;
8958 struct list_head *tmp;
8961 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8962 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8964 unregister_blkdev(MD_MAJOR,"md");
8965 unregister_blkdev(mdp_major, "mdp");
8966 unregister_reboot_notifier(&md_notifier);
8967 unregister_sysctl_table(raid_table_header);
8969 /* We cannot unload the modules while some process is
8970 * waiting for us in select() or poll() - wake them up
8973 while (waitqueue_active(&md_event_waiters)) {
8974 /* not safe to leave yet */
8975 wake_up(&md_event_waiters);
8979 remove_proc_entry("mdstat", NULL);
8981 for_each_mddev(mddev, tmp) {
8982 export_array(mddev);
8983 mddev->hold_active = 0;
8985 destroy_workqueue(md_misc_wq);
8986 destroy_workqueue(md_wq);
8989 subsys_initcall(md_init);
8990 module_exit(md_exit)
8992 static int get_ro(char *buffer, struct kernel_param *kp)
8994 return sprintf(buffer, "%d", start_readonly);
8996 static int set_ro(const char *val, struct kernel_param *kp)
8998 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9001 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9002 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9003 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9005 MODULE_LICENSE("GPL");
9006 MODULE_DESCRIPTION("MD RAID framework");
9008 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);