2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 Errors, Warnings, etc.
36 pr_crit() for error conditions that risk data loss
37 pr_err() for error conditions that are unexpected, like an IO error
38 or internal inconsistency
39 pr_warn() for error conditions that could have been predicated, like
40 adding a device to an array when it has incompatible metadata
41 pr_info() for every interesting, very rare events, like an array starting
42 or stopping, or resync starting or stopping
43 pr_debug() for everything else.
47 #include <linux/sched/signal.h>
48 #include <linux/kthread.h>
49 #include <linux/blkdev.h>
50 #include <linux/badblocks.h>
51 #include <linux/sysctl.h>
52 #include <linux/seq_file.h>
54 #include <linux/poll.h>
55 #include <linux/ctype.h>
56 #include <linux/string.h>
57 #include <linux/hdreg.h>
58 #include <linux/proc_fs.h>
59 #include <linux/random.h>
60 #include <linux/module.h>
61 #include <linux/reboot.h>
62 #include <linux/file.h>
63 #include <linux/compat.h>
64 #include <linux/delay.h>
65 #include <linux/raid/md_p.h>
66 #include <linux/raid/md_u.h>
67 #include <linux/slab.h>
68 #include <trace/events/block.h>
71 #include "md-cluster.h"
74 static void autostart_arrays(int part);
77 /* pers_list is a list of registered personalities protected
79 * pers_lock does extra service to protect accesses to
80 * mddev->thread when the mutex cannot be held.
82 static LIST_HEAD(pers_list);
83 static DEFINE_SPINLOCK(pers_lock);
85 struct md_cluster_operations *md_cluster_ops;
86 EXPORT_SYMBOL(md_cluster_ops);
87 struct module *md_cluster_mod;
88 EXPORT_SYMBOL(md_cluster_mod);
90 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
91 static struct workqueue_struct *md_wq;
92 static struct workqueue_struct *md_misc_wq;
94 static int remove_and_add_spares(struct mddev *mddev,
95 struct md_rdev *this);
96 static void mddev_detach(struct mddev *mddev);
99 * Default number of read corrections we'll attempt on an rdev
100 * before ejecting it from the array. We divide the read error
101 * count by 2 for every hour elapsed between read errors.
103 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
105 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
106 * is 1000 KB/sec, so the extra system load does not show up that much.
107 * Increase it if you want to have more _guaranteed_ speed. Note that
108 * the RAID driver will use the maximum available bandwidth if the IO
109 * subsystem is idle. There is also an 'absolute maximum' reconstruction
110 * speed limit - in case reconstruction slows down your system despite
113 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
114 * or /sys/block/mdX/md/sync_speed_{min,max}
117 static int sysctl_speed_limit_min = 1000;
118 static int sysctl_speed_limit_max = 200000;
119 static inline int speed_min(struct mddev *mddev)
121 return mddev->sync_speed_min ?
122 mddev->sync_speed_min : sysctl_speed_limit_min;
125 static inline int speed_max(struct mddev *mddev)
127 return mddev->sync_speed_max ?
128 mddev->sync_speed_max : sysctl_speed_limit_max;
131 static struct ctl_table_header *raid_table_header;
133 static struct ctl_table raid_table[] = {
135 .procname = "speed_limit_min",
136 .data = &sysctl_speed_limit_min,
137 .maxlen = sizeof(int),
138 .mode = S_IRUGO|S_IWUSR,
139 .proc_handler = proc_dointvec,
142 .procname = "speed_limit_max",
143 .data = &sysctl_speed_limit_max,
144 .maxlen = sizeof(int),
145 .mode = S_IRUGO|S_IWUSR,
146 .proc_handler = proc_dointvec,
151 static struct ctl_table raid_dir_table[] = {
155 .mode = S_IRUGO|S_IXUGO,
161 static struct ctl_table raid_root_table[] = {
166 .child = raid_dir_table,
171 static const struct block_device_operations md_fops;
173 static int start_readonly;
176 * like bio_clone, but with a local bio set
179 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
184 if (!mddev || !mddev->bio_set)
185 return bio_alloc(gfp_mask, nr_iovecs);
187 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
192 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
195 * We have a system wide 'event count' that is incremented
196 * on any 'interesting' event, and readers of /proc/mdstat
197 * can use 'poll' or 'select' to find out when the event
201 * start array, stop array, error, add device, remove device,
202 * start build, activate spare
204 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
205 static atomic_t md_event_count;
206 void md_new_event(struct mddev *mddev)
208 atomic_inc(&md_event_count);
209 wake_up(&md_event_waiters);
211 EXPORT_SYMBOL_GPL(md_new_event);
214 * Enables to iterate over all existing md arrays
215 * all_mddevs_lock protects this list.
217 static LIST_HEAD(all_mddevs);
218 static DEFINE_SPINLOCK(all_mddevs_lock);
221 * iterates through all used mddevs in the system.
222 * We take care to grab the all_mddevs_lock whenever navigating
223 * the list, and to always hold a refcount when unlocked.
224 * Any code which breaks out of this loop while own
225 * a reference to the current mddev and must mddev_put it.
227 #define for_each_mddev(_mddev,_tmp) \
229 for (({ spin_lock(&all_mddevs_lock); \
230 _tmp = all_mddevs.next; \
232 ({ if (_tmp != &all_mddevs) \
233 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
234 spin_unlock(&all_mddevs_lock); \
235 if (_mddev) mddev_put(_mddev); \
236 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
237 _tmp != &all_mddevs;}); \
238 ({ spin_lock(&all_mddevs_lock); \
239 _tmp = _tmp->next;}) \
242 /* Rather than calling directly into the personality make_request function,
243 * IO requests come here first so that we can check if the device is
244 * being suspended pending a reconfiguration.
245 * We hold a refcount over the call to ->make_request. By the time that
246 * call has finished, the bio has been linked into some internal structure
247 * and so is visible to ->quiesce(), so we don't need the refcount any more.
249 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
251 const int rw = bio_data_dir(bio);
252 struct mddev *mddev = q->queuedata;
253 unsigned int sectors;
256 blk_queue_split(q, &bio, q->bio_split);
258 if (mddev == NULL || mddev->pers == NULL) {
260 return BLK_QC_T_NONE;
262 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
263 if (bio_sectors(bio) != 0)
264 bio->bi_error = -EROFS;
266 return BLK_QC_T_NONE;
268 smp_rmb(); /* Ensure implications of 'active' are visible */
270 if (mddev->suspended) {
273 prepare_to_wait(&mddev->sb_wait, &__wait,
274 TASK_UNINTERRUPTIBLE);
275 if (!mddev->suspended)
281 finish_wait(&mddev->sb_wait, &__wait);
283 atomic_inc(&mddev->active_io);
287 * save the sectors now since our bio can
288 * go away inside make_request
290 sectors = bio_sectors(bio);
291 /* bio could be mergeable after passing to underlayer */
292 bio->bi_opf &= ~REQ_NOMERGE;
293 mddev->pers->make_request(mddev, bio);
295 cpu = part_stat_lock();
296 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
297 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
300 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
301 wake_up(&mddev->sb_wait);
303 return BLK_QC_T_NONE;
306 /* mddev_suspend makes sure no new requests are submitted
307 * to the device, and that any requests that have been submitted
308 * are completely handled.
309 * Once mddev_detach() is called and completes, the module will be
312 void mddev_suspend(struct mddev *mddev)
314 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
315 if (mddev->suspended++)
318 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
319 mddev->pers->quiesce(mddev, 1);
321 del_timer_sync(&mddev->safemode_timer);
323 EXPORT_SYMBOL_GPL(mddev_suspend);
325 void mddev_resume(struct mddev *mddev)
327 if (--mddev->suspended)
329 wake_up(&mddev->sb_wait);
330 mddev->pers->quiesce(mddev, 0);
332 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
333 md_wakeup_thread(mddev->thread);
334 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
336 EXPORT_SYMBOL_GPL(mddev_resume);
338 int mddev_congested(struct mddev *mddev, int bits)
340 struct md_personality *pers = mddev->pers;
344 if (mddev->suspended)
346 else if (pers && pers->congested)
347 ret = pers->congested(mddev, bits);
351 EXPORT_SYMBOL_GPL(mddev_congested);
352 static int md_congested(void *data, int bits)
354 struct mddev *mddev = data;
355 return mddev_congested(mddev, bits);
359 * Generic flush handling for md
362 static void md_end_flush(struct bio *bio)
364 struct md_rdev *rdev = bio->bi_private;
365 struct mddev *mddev = rdev->mddev;
367 rdev_dec_pending(rdev, mddev);
369 if (atomic_dec_and_test(&mddev->flush_pending)) {
370 /* The pre-request flush has finished */
371 queue_work(md_wq, &mddev->flush_work);
376 static void md_submit_flush_data(struct work_struct *ws);
378 static void submit_flushes(struct work_struct *ws)
380 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
381 struct md_rdev *rdev;
383 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
384 atomic_set(&mddev->flush_pending, 1);
386 rdev_for_each_rcu(rdev, mddev)
387 if (rdev->raid_disk >= 0 &&
388 !test_bit(Faulty, &rdev->flags)) {
389 /* Take two references, one is dropped
390 * when request finishes, one after
391 * we reclaim rcu_read_lock
394 atomic_inc(&rdev->nr_pending);
395 atomic_inc(&rdev->nr_pending);
397 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
398 bi->bi_end_io = md_end_flush;
399 bi->bi_private = rdev;
400 bi->bi_bdev = rdev->bdev;
401 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
402 atomic_inc(&mddev->flush_pending);
405 rdev_dec_pending(rdev, mddev);
408 if (atomic_dec_and_test(&mddev->flush_pending))
409 queue_work(md_wq, &mddev->flush_work);
412 static void md_submit_flush_data(struct work_struct *ws)
414 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
415 struct bio *bio = mddev->flush_bio;
417 if (bio->bi_iter.bi_size == 0)
418 /* an empty barrier - all done */
421 bio->bi_opf &= ~REQ_PREFLUSH;
422 mddev->pers->make_request(mddev, bio);
425 mddev->flush_bio = NULL;
426 wake_up(&mddev->sb_wait);
429 void md_flush_request(struct mddev *mddev, struct bio *bio)
431 spin_lock_irq(&mddev->lock);
432 wait_event_lock_irq(mddev->sb_wait,
435 mddev->flush_bio = bio;
436 spin_unlock_irq(&mddev->lock);
438 INIT_WORK(&mddev->flush_work, submit_flushes);
439 queue_work(md_wq, &mddev->flush_work);
441 EXPORT_SYMBOL(md_flush_request);
443 static inline struct mddev *mddev_get(struct mddev *mddev)
445 atomic_inc(&mddev->active);
449 static void mddev_delayed_delete(struct work_struct *ws);
451 static void mddev_put(struct mddev *mddev)
453 struct bio_set *bs = NULL;
455 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
457 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
458 mddev->ctime == 0 && !mddev->hold_active) {
459 /* Array is not configured at all, and not held active,
461 list_del_init(&mddev->all_mddevs);
463 mddev->bio_set = NULL;
464 if (mddev->gendisk) {
465 /* We did a probe so need to clean up. Call
466 * queue_work inside the spinlock so that
467 * flush_workqueue() after mddev_find will
468 * succeed in waiting for the work to be done.
470 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
471 queue_work(md_misc_wq, &mddev->del_work);
475 spin_unlock(&all_mddevs_lock);
480 static void md_safemode_timeout(unsigned long data);
482 void mddev_init(struct mddev *mddev)
484 mutex_init(&mddev->open_mutex);
485 mutex_init(&mddev->reconfig_mutex);
486 mutex_init(&mddev->bitmap_info.mutex);
487 INIT_LIST_HEAD(&mddev->disks);
488 INIT_LIST_HEAD(&mddev->all_mddevs);
489 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
490 (unsigned long) mddev);
491 atomic_set(&mddev->active, 1);
492 atomic_set(&mddev->openers, 0);
493 atomic_set(&mddev->active_io, 0);
494 spin_lock_init(&mddev->lock);
495 atomic_set(&mddev->flush_pending, 0);
496 init_waitqueue_head(&mddev->sb_wait);
497 init_waitqueue_head(&mddev->recovery_wait);
498 mddev->reshape_position = MaxSector;
499 mddev->reshape_backwards = 0;
500 mddev->last_sync_action = "none";
501 mddev->resync_min = 0;
502 mddev->resync_max = MaxSector;
503 mddev->level = LEVEL_NONE;
505 EXPORT_SYMBOL_GPL(mddev_init);
507 static struct mddev *mddev_find(dev_t unit)
509 struct mddev *mddev, *new = NULL;
511 if (unit && MAJOR(unit) != MD_MAJOR)
512 unit &= ~((1<<MdpMinorShift)-1);
515 spin_lock(&all_mddevs_lock);
518 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
519 if (mddev->unit == unit) {
521 spin_unlock(&all_mddevs_lock);
527 list_add(&new->all_mddevs, &all_mddevs);
528 spin_unlock(&all_mddevs_lock);
529 new->hold_active = UNTIL_IOCTL;
533 /* find an unused unit number */
534 static int next_minor = 512;
535 int start = next_minor;
539 dev = MKDEV(MD_MAJOR, next_minor);
541 if (next_minor > MINORMASK)
543 if (next_minor == start) {
544 /* Oh dear, all in use. */
545 spin_unlock(&all_mddevs_lock);
551 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
552 if (mddev->unit == dev) {
558 new->md_minor = MINOR(dev);
559 new->hold_active = UNTIL_STOP;
560 list_add(&new->all_mddevs, &all_mddevs);
561 spin_unlock(&all_mddevs_lock);
564 spin_unlock(&all_mddevs_lock);
566 new = kzalloc(sizeof(*new), GFP_KERNEL);
571 if (MAJOR(unit) == MD_MAJOR)
572 new->md_minor = MINOR(unit);
574 new->md_minor = MINOR(unit) >> MdpMinorShift;
581 static struct attribute_group md_redundancy_group;
583 void mddev_unlock(struct mddev *mddev)
585 if (mddev->to_remove) {
586 /* These cannot be removed under reconfig_mutex as
587 * an access to the files will try to take reconfig_mutex
588 * while holding the file unremovable, which leads to
590 * So hold set sysfs_active while the remove in happeing,
591 * and anything else which might set ->to_remove or my
592 * otherwise change the sysfs namespace will fail with
593 * -EBUSY if sysfs_active is still set.
594 * We set sysfs_active under reconfig_mutex and elsewhere
595 * test it under the same mutex to ensure its correct value
598 struct attribute_group *to_remove = mddev->to_remove;
599 mddev->to_remove = NULL;
600 mddev->sysfs_active = 1;
601 mutex_unlock(&mddev->reconfig_mutex);
603 if (mddev->kobj.sd) {
604 if (to_remove != &md_redundancy_group)
605 sysfs_remove_group(&mddev->kobj, to_remove);
606 if (mddev->pers == NULL ||
607 mddev->pers->sync_request == NULL) {
608 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
609 if (mddev->sysfs_action)
610 sysfs_put(mddev->sysfs_action);
611 mddev->sysfs_action = NULL;
614 mddev->sysfs_active = 0;
616 mutex_unlock(&mddev->reconfig_mutex);
618 /* As we've dropped the mutex we need a spinlock to
619 * make sure the thread doesn't disappear
621 spin_lock(&pers_lock);
622 md_wakeup_thread(mddev->thread);
623 spin_unlock(&pers_lock);
625 EXPORT_SYMBOL_GPL(mddev_unlock);
627 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
629 struct md_rdev *rdev;
631 rdev_for_each_rcu(rdev, mddev)
632 if (rdev->desc_nr == nr)
637 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
639 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
641 struct md_rdev *rdev;
643 rdev_for_each(rdev, mddev)
644 if (rdev->bdev->bd_dev == dev)
650 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
652 struct md_rdev *rdev;
654 rdev_for_each_rcu(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
661 static struct md_personality *find_pers(int level, char *clevel)
663 struct md_personality *pers;
664 list_for_each_entry(pers, &pers_list, list) {
665 if (level != LEVEL_NONE && pers->level == level)
667 if (strcmp(pers->name, clevel)==0)
673 /* return the offset of the super block in 512byte sectors */
674 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
676 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
677 return MD_NEW_SIZE_SECTORS(num_sectors);
680 static int alloc_disk_sb(struct md_rdev *rdev)
682 rdev->sb_page = alloc_page(GFP_KERNEL);
688 void md_rdev_clear(struct md_rdev *rdev)
691 put_page(rdev->sb_page);
693 rdev->sb_page = NULL;
698 put_page(rdev->bb_page);
699 rdev->bb_page = NULL;
701 badblocks_exit(&rdev->badblocks);
703 EXPORT_SYMBOL_GPL(md_rdev_clear);
705 static void super_written(struct bio *bio)
707 struct md_rdev *rdev = bio->bi_private;
708 struct mddev *mddev = rdev->mddev;
711 pr_err("md: super_written gets error=%d\n", bio->bi_error);
712 md_error(mddev, rdev);
713 if (!test_bit(Faulty, &rdev->flags)
714 && (bio->bi_opf & MD_FAILFAST)) {
715 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
716 set_bit(LastDev, &rdev->flags);
719 clear_bit(LastDev, &rdev->flags);
721 if (atomic_dec_and_test(&mddev->pending_writes))
722 wake_up(&mddev->sb_wait);
723 rdev_dec_pending(rdev, mddev);
727 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
728 sector_t sector, int size, struct page *page)
730 /* write first size bytes of page to sector of rdev
731 * Increment mddev->pending_writes before returning
732 * and decrement it on completion, waking up sb_wait
733 * if zero is reached.
734 * If an error occurred, call md_error
739 if (test_bit(Faulty, &rdev->flags))
742 bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
744 atomic_inc(&rdev->nr_pending);
746 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
747 bio->bi_iter.bi_sector = sector;
748 bio_add_page(bio, page, size, 0);
749 bio->bi_private = rdev;
750 bio->bi_end_io = super_written;
752 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
753 test_bit(FailFast, &rdev->flags) &&
754 !test_bit(LastDev, &rdev->flags))
756 bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_FUA | ff;
758 atomic_inc(&mddev->pending_writes);
762 int md_super_wait(struct mddev *mddev)
764 /* wait for all superblock writes that were scheduled to complete */
765 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
766 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
771 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
772 struct page *page, int op, int op_flags, bool metadata_op)
774 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
777 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
778 rdev->meta_bdev : rdev->bdev;
779 bio_set_op_attrs(bio, op, op_flags);
781 bio->bi_iter.bi_sector = sector + rdev->sb_start;
782 else if (rdev->mddev->reshape_position != MaxSector &&
783 (rdev->mddev->reshape_backwards ==
784 (sector >= rdev->mddev->reshape_position)))
785 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
787 bio->bi_iter.bi_sector = sector + rdev->data_offset;
788 bio_add_page(bio, page, size, 0);
790 submit_bio_wait(bio);
792 ret = !bio->bi_error;
796 EXPORT_SYMBOL_GPL(sync_page_io);
798 static int read_disk_sb(struct md_rdev *rdev, int size)
800 char b[BDEVNAME_SIZE];
805 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
811 pr_err("md: disabled device %s, could not read superblock.\n",
812 bdevname(rdev->bdev,b));
816 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
818 return sb1->set_uuid0 == sb2->set_uuid0 &&
819 sb1->set_uuid1 == sb2->set_uuid1 &&
820 sb1->set_uuid2 == sb2->set_uuid2 &&
821 sb1->set_uuid3 == sb2->set_uuid3;
824 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
827 mdp_super_t *tmp1, *tmp2;
829 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
830 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
832 if (!tmp1 || !tmp2) {
841 * nr_disks is not constant
846 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
853 static u32 md_csum_fold(u32 csum)
855 csum = (csum & 0xffff) + (csum >> 16);
856 return (csum & 0xffff) + (csum >> 16);
859 static unsigned int calc_sb_csum(mdp_super_t *sb)
862 u32 *sb32 = (u32*)sb;
864 unsigned int disk_csum, csum;
866 disk_csum = sb->sb_csum;
869 for (i = 0; i < MD_SB_BYTES/4 ; i++)
871 csum = (newcsum & 0xffffffff) + (newcsum>>32);
874 /* This used to use csum_partial, which was wrong for several
875 * reasons including that different results are returned on
876 * different architectures. It isn't critical that we get exactly
877 * the same return value as before (we always csum_fold before
878 * testing, and that removes any differences). However as we
879 * know that csum_partial always returned a 16bit value on
880 * alphas, do a fold to maximise conformity to previous behaviour.
882 sb->sb_csum = md_csum_fold(disk_csum);
884 sb->sb_csum = disk_csum;
890 * Handle superblock details.
891 * We want to be able to handle multiple superblock formats
892 * so we have a common interface to them all, and an array of
893 * different handlers.
894 * We rely on user-space to write the initial superblock, and support
895 * reading and updating of superblocks.
896 * Interface methods are:
897 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
898 * loads and validates a superblock on dev.
899 * if refdev != NULL, compare superblocks on both devices
901 * 0 - dev has a superblock that is compatible with refdev
902 * 1 - dev has a superblock that is compatible and newer than refdev
903 * so dev should be used as the refdev in future
904 * -EINVAL superblock incompatible or invalid
905 * -othererror e.g. -EIO
907 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
908 * Verify that dev is acceptable into mddev.
909 * The first time, mddev->raid_disks will be 0, and data from
910 * dev should be merged in. Subsequent calls check that dev
911 * is new enough. Return 0 or -EINVAL
913 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
914 * Update the superblock for rdev with data in mddev
915 * This does not write to disc.
921 struct module *owner;
922 int (*load_super)(struct md_rdev *rdev,
923 struct md_rdev *refdev,
925 int (*validate_super)(struct mddev *mddev,
926 struct md_rdev *rdev);
927 void (*sync_super)(struct mddev *mddev,
928 struct md_rdev *rdev);
929 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
930 sector_t num_sectors);
931 int (*allow_new_offset)(struct md_rdev *rdev,
932 unsigned long long new_offset);
936 * Check that the given mddev has no bitmap.
938 * This function is called from the run method of all personalities that do not
939 * support bitmaps. It prints an error message and returns non-zero if mddev
940 * has a bitmap. Otherwise, it returns 0.
943 int md_check_no_bitmap(struct mddev *mddev)
945 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
947 pr_warn("%s: bitmaps are not supported for %s\n",
948 mdname(mddev), mddev->pers->name);
951 EXPORT_SYMBOL(md_check_no_bitmap);
954 * load_super for 0.90.0
956 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
958 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
963 * Calculate the position of the superblock (512byte sectors),
964 * it's at the end of the disk.
966 * It also happens to be a multiple of 4Kb.
968 rdev->sb_start = calc_dev_sboffset(rdev);
970 ret = read_disk_sb(rdev, MD_SB_BYTES);
976 bdevname(rdev->bdev, b);
977 sb = page_address(rdev->sb_page);
979 if (sb->md_magic != MD_SB_MAGIC) {
980 pr_warn("md: invalid raid superblock magic on %s\n", b);
984 if (sb->major_version != 0 ||
985 sb->minor_version < 90 ||
986 sb->minor_version > 91) {
987 pr_warn("Bad version number %d.%d on %s\n",
988 sb->major_version, sb->minor_version, b);
992 if (sb->raid_disks <= 0)
995 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
996 pr_warn("md: invalid superblock checksum on %s\n", b);
1000 rdev->preferred_minor = sb->md_minor;
1001 rdev->data_offset = 0;
1002 rdev->new_data_offset = 0;
1003 rdev->sb_size = MD_SB_BYTES;
1004 rdev->badblocks.shift = -1;
1006 if (sb->level == LEVEL_MULTIPATH)
1009 rdev->desc_nr = sb->this_disk.number;
1015 mdp_super_t *refsb = page_address(refdev->sb_page);
1016 if (!uuid_equal(refsb, sb)) {
1017 pr_warn("md: %s has different UUID to %s\n",
1018 b, bdevname(refdev->bdev,b2));
1021 if (!sb_equal(refsb, sb)) {
1022 pr_warn("md: %s has same UUID but different superblock to %s\n",
1023 b, bdevname(refdev->bdev, b2));
1027 ev2 = md_event(refsb);
1033 rdev->sectors = rdev->sb_start;
1034 /* Limit to 4TB as metadata cannot record more than that.
1035 * (not needed for Linear and RAID0 as metadata doesn't
1038 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1040 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1042 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1043 /* "this cannot possibly happen" ... */
1051 * validate_super for 0.90.0
1053 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1056 mdp_super_t *sb = page_address(rdev->sb_page);
1057 __u64 ev1 = md_event(sb);
1059 rdev->raid_disk = -1;
1060 clear_bit(Faulty, &rdev->flags);
1061 clear_bit(In_sync, &rdev->flags);
1062 clear_bit(Bitmap_sync, &rdev->flags);
1063 clear_bit(WriteMostly, &rdev->flags);
1065 if (mddev->raid_disks == 0) {
1066 mddev->major_version = 0;
1067 mddev->minor_version = sb->minor_version;
1068 mddev->patch_version = sb->patch_version;
1069 mddev->external = 0;
1070 mddev->chunk_sectors = sb->chunk_size >> 9;
1071 mddev->ctime = sb->ctime;
1072 mddev->utime = sb->utime;
1073 mddev->level = sb->level;
1074 mddev->clevel[0] = 0;
1075 mddev->layout = sb->layout;
1076 mddev->raid_disks = sb->raid_disks;
1077 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1078 mddev->events = ev1;
1079 mddev->bitmap_info.offset = 0;
1080 mddev->bitmap_info.space = 0;
1081 /* bitmap can use 60 K after the 4K superblocks */
1082 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1083 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1084 mddev->reshape_backwards = 0;
1086 if (mddev->minor_version >= 91) {
1087 mddev->reshape_position = sb->reshape_position;
1088 mddev->delta_disks = sb->delta_disks;
1089 mddev->new_level = sb->new_level;
1090 mddev->new_layout = sb->new_layout;
1091 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1092 if (mddev->delta_disks < 0)
1093 mddev->reshape_backwards = 1;
1095 mddev->reshape_position = MaxSector;
1096 mddev->delta_disks = 0;
1097 mddev->new_level = mddev->level;
1098 mddev->new_layout = mddev->layout;
1099 mddev->new_chunk_sectors = mddev->chunk_sectors;
1102 if (sb->state & (1<<MD_SB_CLEAN))
1103 mddev->recovery_cp = MaxSector;
1105 if (sb->events_hi == sb->cp_events_hi &&
1106 sb->events_lo == sb->cp_events_lo) {
1107 mddev->recovery_cp = sb->recovery_cp;
1109 mddev->recovery_cp = 0;
1112 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1113 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1114 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1115 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1117 mddev->max_disks = MD_SB_DISKS;
1119 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1120 mddev->bitmap_info.file == NULL) {
1121 mddev->bitmap_info.offset =
1122 mddev->bitmap_info.default_offset;
1123 mddev->bitmap_info.space =
1124 mddev->bitmap_info.default_space;
1127 } else if (mddev->pers == NULL) {
1128 /* Insist on good event counter while assembling, except
1129 * for spares (which don't need an event count) */
1131 if (sb->disks[rdev->desc_nr].state & (
1132 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1133 if (ev1 < mddev->events)
1135 } else if (mddev->bitmap) {
1136 /* if adding to array with a bitmap, then we can accept an
1137 * older device ... but not too old.
1139 if (ev1 < mddev->bitmap->events_cleared)
1141 if (ev1 < mddev->events)
1142 set_bit(Bitmap_sync, &rdev->flags);
1144 if (ev1 < mddev->events)
1145 /* just a hot-add of a new device, leave raid_disk at -1 */
1149 if (mddev->level != LEVEL_MULTIPATH) {
1150 desc = sb->disks + rdev->desc_nr;
1152 if (desc->state & (1<<MD_DISK_FAULTY))
1153 set_bit(Faulty, &rdev->flags);
1154 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1155 desc->raid_disk < mddev->raid_disks */) {
1156 set_bit(In_sync, &rdev->flags);
1157 rdev->raid_disk = desc->raid_disk;
1158 rdev->saved_raid_disk = desc->raid_disk;
1159 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1160 /* active but not in sync implies recovery up to
1161 * reshape position. We don't know exactly where
1162 * that is, so set to zero for now */
1163 if (mddev->minor_version >= 91) {
1164 rdev->recovery_offset = 0;
1165 rdev->raid_disk = desc->raid_disk;
1168 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1169 set_bit(WriteMostly, &rdev->flags);
1170 if (desc->state & (1<<MD_DISK_FAILFAST))
1171 set_bit(FailFast, &rdev->flags);
1172 } else /* MULTIPATH are always insync */
1173 set_bit(In_sync, &rdev->flags);
1178 * sync_super for 0.90.0
1180 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1183 struct md_rdev *rdev2;
1184 int next_spare = mddev->raid_disks;
1186 /* make rdev->sb match mddev data..
1189 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1190 * 3/ any empty disks < next_spare become removed
1192 * disks[0] gets initialised to REMOVED because
1193 * we cannot be sure from other fields if it has
1194 * been initialised or not.
1197 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1199 rdev->sb_size = MD_SB_BYTES;
1201 sb = page_address(rdev->sb_page);
1203 memset(sb, 0, sizeof(*sb));
1205 sb->md_magic = MD_SB_MAGIC;
1206 sb->major_version = mddev->major_version;
1207 sb->patch_version = mddev->patch_version;
1208 sb->gvalid_words = 0; /* ignored */
1209 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1210 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1211 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1212 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1214 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1215 sb->level = mddev->level;
1216 sb->size = mddev->dev_sectors / 2;
1217 sb->raid_disks = mddev->raid_disks;
1218 sb->md_minor = mddev->md_minor;
1219 sb->not_persistent = 0;
1220 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1222 sb->events_hi = (mddev->events>>32);
1223 sb->events_lo = (u32)mddev->events;
1225 if (mddev->reshape_position == MaxSector)
1226 sb->minor_version = 90;
1228 sb->minor_version = 91;
1229 sb->reshape_position = mddev->reshape_position;
1230 sb->new_level = mddev->new_level;
1231 sb->delta_disks = mddev->delta_disks;
1232 sb->new_layout = mddev->new_layout;
1233 sb->new_chunk = mddev->new_chunk_sectors << 9;
1235 mddev->minor_version = sb->minor_version;
1238 sb->recovery_cp = mddev->recovery_cp;
1239 sb->cp_events_hi = (mddev->events>>32);
1240 sb->cp_events_lo = (u32)mddev->events;
1241 if (mddev->recovery_cp == MaxSector)
1242 sb->state = (1<< MD_SB_CLEAN);
1244 sb->recovery_cp = 0;
1246 sb->layout = mddev->layout;
1247 sb->chunk_size = mddev->chunk_sectors << 9;
1249 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1250 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1252 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1253 rdev_for_each(rdev2, mddev) {
1256 int is_active = test_bit(In_sync, &rdev2->flags);
1258 if (rdev2->raid_disk >= 0 &&
1259 sb->minor_version >= 91)
1260 /* we have nowhere to store the recovery_offset,
1261 * but if it is not below the reshape_position,
1262 * we can piggy-back on that.
1265 if (rdev2->raid_disk < 0 ||
1266 test_bit(Faulty, &rdev2->flags))
1269 desc_nr = rdev2->raid_disk;
1271 desc_nr = next_spare++;
1272 rdev2->desc_nr = desc_nr;
1273 d = &sb->disks[rdev2->desc_nr];
1275 d->number = rdev2->desc_nr;
1276 d->major = MAJOR(rdev2->bdev->bd_dev);
1277 d->minor = MINOR(rdev2->bdev->bd_dev);
1279 d->raid_disk = rdev2->raid_disk;
1281 d->raid_disk = rdev2->desc_nr; /* compatibility */
1282 if (test_bit(Faulty, &rdev2->flags))
1283 d->state = (1<<MD_DISK_FAULTY);
1284 else if (is_active) {
1285 d->state = (1<<MD_DISK_ACTIVE);
1286 if (test_bit(In_sync, &rdev2->flags))
1287 d->state |= (1<<MD_DISK_SYNC);
1295 if (test_bit(WriteMostly, &rdev2->flags))
1296 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1297 if (test_bit(FailFast, &rdev2->flags))
1298 d->state |= (1<<MD_DISK_FAILFAST);
1300 /* now set the "removed" and "faulty" bits on any missing devices */
1301 for (i=0 ; i < mddev->raid_disks ; i++) {
1302 mdp_disk_t *d = &sb->disks[i];
1303 if (d->state == 0 && d->number == 0) {
1306 d->state = (1<<MD_DISK_REMOVED);
1307 d->state |= (1<<MD_DISK_FAULTY);
1311 sb->nr_disks = nr_disks;
1312 sb->active_disks = active;
1313 sb->working_disks = working;
1314 sb->failed_disks = failed;
1315 sb->spare_disks = spare;
1317 sb->this_disk = sb->disks[rdev->desc_nr];
1318 sb->sb_csum = calc_sb_csum(sb);
1322 * rdev_size_change for 0.90.0
1324 static unsigned long long
1325 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1327 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1328 return 0; /* component must fit device */
1329 if (rdev->mddev->bitmap_info.offset)
1330 return 0; /* can't move bitmap */
1331 rdev->sb_start = calc_dev_sboffset(rdev);
1332 if (!num_sectors || num_sectors > rdev->sb_start)
1333 num_sectors = rdev->sb_start;
1334 /* Limit to 4TB as metadata cannot record more than that.
1335 * 4TB == 2^32 KB, or 2*2^32 sectors.
1337 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1338 rdev->mddev->level >= 1)
1339 num_sectors = (sector_t)(2ULL << 32) - 2;
1341 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1343 } while (md_super_wait(rdev->mddev) < 0);
1348 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1350 /* non-zero offset changes not possible with v0.90 */
1351 return new_offset == 0;
1355 * version 1 superblock
1358 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1362 unsigned long long newcsum;
1363 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1364 __le32 *isuper = (__le32*)sb;
1366 disk_csum = sb->sb_csum;
1369 for (; size >= 4; size -= 4)
1370 newcsum += le32_to_cpu(*isuper++);
1373 newcsum += le16_to_cpu(*(__le16*) isuper);
1375 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1376 sb->sb_csum = disk_csum;
1377 return cpu_to_le32(csum);
1380 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1382 struct mdp_superblock_1 *sb;
1386 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1390 * Calculate the position of the superblock in 512byte sectors.
1391 * It is always aligned to a 4K boundary and
1392 * depeding on minor_version, it can be:
1393 * 0: At least 8K, but less than 12K, from end of device
1394 * 1: At start of device
1395 * 2: 4K from start of device.
1397 switch(minor_version) {
1399 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1401 sb_start &= ~(sector_t)(4*2-1);
1412 rdev->sb_start = sb_start;
1414 /* superblock is rarely larger than 1K, but it can be larger,
1415 * and it is safe to read 4k, so we do that
1417 ret = read_disk_sb(rdev, 4096);
1418 if (ret) return ret;
1420 sb = page_address(rdev->sb_page);
1422 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1423 sb->major_version != cpu_to_le32(1) ||
1424 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1425 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1426 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1429 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1430 pr_warn("md: invalid superblock checksum on %s\n",
1431 bdevname(rdev->bdev,b));
1434 if (le64_to_cpu(sb->data_size) < 10) {
1435 pr_warn("md: data_size too small on %s\n",
1436 bdevname(rdev->bdev,b));
1441 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1442 /* Some padding is non-zero, might be a new feature */
1445 rdev->preferred_minor = 0xffff;
1446 rdev->data_offset = le64_to_cpu(sb->data_offset);
1447 rdev->new_data_offset = rdev->data_offset;
1448 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1449 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1450 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1451 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1453 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1454 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1455 if (rdev->sb_size & bmask)
1456 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1459 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1462 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1465 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1468 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1470 if (!rdev->bb_page) {
1471 rdev->bb_page = alloc_page(GFP_KERNEL);
1475 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1476 rdev->badblocks.count == 0) {
1477 /* need to load the bad block list.
1478 * Currently we limit it to one page.
1484 int sectors = le16_to_cpu(sb->bblog_size);
1485 if (sectors > (PAGE_SIZE / 512))
1487 offset = le32_to_cpu(sb->bblog_offset);
1490 bb_sector = (long long)offset;
1491 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1492 rdev->bb_page, REQ_OP_READ, 0, true))
1494 bbp = (u64 *)page_address(rdev->bb_page);
1495 rdev->badblocks.shift = sb->bblog_shift;
1496 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1497 u64 bb = le64_to_cpu(*bbp);
1498 int count = bb & (0x3ff);
1499 u64 sector = bb >> 10;
1500 sector <<= sb->bblog_shift;
1501 count <<= sb->bblog_shift;
1504 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1507 } else if (sb->bblog_offset != 0)
1508 rdev->badblocks.shift = 0;
1510 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) {
1511 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1512 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1513 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1520 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1522 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1523 sb->level != refsb->level ||
1524 sb->layout != refsb->layout ||
1525 sb->chunksize != refsb->chunksize) {
1526 pr_warn("md: %s has strangely different superblock to %s\n",
1527 bdevname(rdev->bdev,b),
1528 bdevname(refdev->bdev,b2));
1531 ev1 = le64_to_cpu(sb->events);
1532 ev2 = le64_to_cpu(refsb->events);
1539 if (minor_version) {
1540 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1541 sectors -= rdev->data_offset;
1543 sectors = rdev->sb_start;
1544 if (sectors < le64_to_cpu(sb->data_size))
1546 rdev->sectors = le64_to_cpu(sb->data_size);
1550 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1552 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1553 __u64 ev1 = le64_to_cpu(sb->events);
1555 rdev->raid_disk = -1;
1556 clear_bit(Faulty, &rdev->flags);
1557 clear_bit(In_sync, &rdev->flags);
1558 clear_bit(Bitmap_sync, &rdev->flags);
1559 clear_bit(WriteMostly, &rdev->flags);
1561 if (mddev->raid_disks == 0) {
1562 mddev->major_version = 1;
1563 mddev->patch_version = 0;
1564 mddev->external = 0;
1565 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1566 mddev->ctime = le64_to_cpu(sb->ctime);
1567 mddev->utime = le64_to_cpu(sb->utime);
1568 mddev->level = le32_to_cpu(sb->level);
1569 mddev->clevel[0] = 0;
1570 mddev->layout = le32_to_cpu(sb->layout);
1571 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1572 mddev->dev_sectors = le64_to_cpu(sb->size);
1573 mddev->events = ev1;
1574 mddev->bitmap_info.offset = 0;
1575 mddev->bitmap_info.space = 0;
1576 /* Default location for bitmap is 1K after superblock
1577 * using 3K - total of 4K
1579 mddev->bitmap_info.default_offset = 1024 >> 9;
1580 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1581 mddev->reshape_backwards = 0;
1583 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1584 memcpy(mddev->uuid, sb->set_uuid, 16);
1586 mddev->max_disks = (4096-256)/2;
1588 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1589 mddev->bitmap_info.file == NULL) {
1590 mddev->bitmap_info.offset =
1591 (__s32)le32_to_cpu(sb->bitmap_offset);
1592 /* Metadata doesn't record how much space is available.
1593 * For 1.0, we assume we can use up to the superblock
1594 * if before, else to 4K beyond superblock.
1595 * For others, assume no change is possible.
1597 if (mddev->minor_version > 0)
1598 mddev->bitmap_info.space = 0;
1599 else if (mddev->bitmap_info.offset > 0)
1600 mddev->bitmap_info.space =
1601 8 - mddev->bitmap_info.offset;
1603 mddev->bitmap_info.space =
1604 -mddev->bitmap_info.offset;
1607 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1608 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1609 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1610 mddev->new_level = le32_to_cpu(sb->new_level);
1611 mddev->new_layout = le32_to_cpu(sb->new_layout);
1612 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1613 if (mddev->delta_disks < 0 ||
1614 (mddev->delta_disks == 0 &&
1615 (le32_to_cpu(sb->feature_map)
1616 & MD_FEATURE_RESHAPE_BACKWARDS)))
1617 mddev->reshape_backwards = 1;
1619 mddev->reshape_position = MaxSector;
1620 mddev->delta_disks = 0;
1621 mddev->new_level = mddev->level;
1622 mddev->new_layout = mddev->layout;
1623 mddev->new_chunk_sectors = mddev->chunk_sectors;
1626 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1627 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1629 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) {
1630 if (le32_to_cpu(sb->feature_map) &
1631 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1633 set_bit(MD_HAS_PPL, &mddev->flags);
1635 } else if (mddev->pers == NULL) {
1636 /* Insist of good event counter while assembling, except for
1637 * spares (which don't need an event count) */
1639 if (rdev->desc_nr >= 0 &&
1640 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1641 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1642 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1643 if (ev1 < mddev->events)
1645 } else if (mddev->bitmap) {
1646 /* If adding to array with a bitmap, then we can accept an
1647 * older device, but not too old.
1649 if (ev1 < mddev->bitmap->events_cleared)
1651 if (ev1 < mddev->events)
1652 set_bit(Bitmap_sync, &rdev->flags);
1654 if (ev1 < mddev->events)
1655 /* just a hot-add of a new device, leave raid_disk at -1 */
1658 if (mddev->level != LEVEL_MULTIPATH) {
1660 if (rdev->desc_nr < 0 ||
1661 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1662 role = MD_DISK_ROLE_SPARE;
1665 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1667 case MD_DISK_ROLE_SPARE: /* spare */
1669 case MD_DISK_ROLE_FAULTY: /* faulty */
1670 set_bit(Faulty, &rdev->flags);
1672 case MD_DISK_ROLE_JOURNAL: /* journal device */
1673 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1674 /* journal device without journal feature */
1675 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1678 set_bit(Journal, &rdev->flags);
1679 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1680 rdev->raid_disk = 0;
1683 rdev->saved_raid_disk = role;
1684 if ((le32_to_cpu(sb->feature_map) &
1685 MD_FEATURE_RECOVERY_OFFSET)) {
1686 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1687 if (!(le32_to_cpu(sb->feature_map) &
1688 MD_FEATURE_RECOVERY_BITMAP))
1689 rdev->saved_raid_disk = -1;
1691 set_bit(In_sync, &rdev->flags);
1692 rdev->raid_disk = role;
1695 if (sb->devflags & WriteMostly1)
1696 set_bit(WriteMostly, &rdev->flags);
1697 if (sb->devflags & FailFast1)
1698 set_bit(FailFast, &rdev->flags);
1699 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1700 set_bit(Replacement, &rdev->flags);
1701 } else /* MULTIPATH are always insync */
1702 set_bit(In_sync, &rdev->flags);
1707 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1709 struct mdp_superblock_1 *sb;
1710 struct md_rdev *rdev2;
1712 /* make rdev->sb match mddev and rdev data. */
1714 sb = page_address(rdev->sb_page);
1716 sb->feature_map = 0;
1718 sb->recovery_offset = cpu_to_le64(0);
1719 memset(sb->pad3, 0, sizeof(sb->pad3));
1721 sb->utime = cpu_to_le64((__u64)mddev->utime);
1722 sb->events = cpu_to_le64(mddev->events);
1724 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1725 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1726 sb->resync_offset = cpu_to_le64(MaxSector);
1728 sb->resync_offset = cpu_to_le64(0);
1730 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1732 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1733 sb->size = cpu_to_le64(mddev->dev_sectors);
1734 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1735 sb->level = cpu_to_le32(mddev->level);
1736 sb->layout = cpu_to_le32(mddev->layout);
1737 if (test_bit(FailFast, &rdev->flags))
1738 sb->devflags |= FailFast1;
1740 sb->devflags &= ~FailFast1;
1742 if (test_bit(WriteMostly, &rdev->flags))
1743 sb->devflags |= WriteMostly1;
1745 sb->devflags &= ~WriteMostly1;
1746 sb->data_offset = cpu_to_le64(rdev->data_offset);
1747 sb->data_size = cpu_to_le64(rdev->sectors);
1749 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1750 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1751 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1754 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1755 !test_bit(In_sync, &rdev->flags)) {
1757 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1758 sb->recovery_offset =
1759 cpu_to_le64(rdev->recovery_offset);
1760 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1762 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1764 /* Note: recovery_offset and journal_tail share space */
1765 if (test_bit(Journal, &rdev->flags))
1766 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1767 if (test_bit(Replacement, &rdev->flags))
1769 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1771 if (mddev->reshape_position != MaxSector) {
1772 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1773 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1774 sb->new_layout = cpu_to_le32(mddev->new_layout);
1775 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1776 sb->new_level = cpu_to_le32(mddev->new_level);
1777 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1778 if (mddev->delta_disks == 0 &&
1779 mddev->reshape_backwards)
1781 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1782 if (rdev->new_data_offset != rdev->data_offset) {
1784 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1785 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1786 - rdev->data_offset));
1790 if (mddev_is_clustered(mddev))
1791 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1793 if (rdev->badblocks.count == 0)
1794 /* Nothing to do for bad blocks*/ ;
1795 else if (sb->bblog_offset == 0)
1796 /* Cannot record bad blocks on this device */
1797 md_error(mddev, rdev);
1799 struct badblocks *bb = &rdev->badblocks;
1800 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1802 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1807 seq = read_seqbegin(&bb->lock);
1809 memset(bbp, 0xff, PAGE_SIZE);
1811 for (i = 0 ; i < bb->count ; i++) {
1812 u64 internal_bb = p[i];
1813 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1814 | BB_LEN(internal_bb));
1815 bbp[i] = cpu_to_le64(store_bb);
1818 if (read_seqretry(&bb->lock, seq))
1821 bb->sector = (rdev->sb_start +
1822 (int)le32_to_cpu(sb->bblog_offset));
1823 bb->size = le16_to_cpu(sb->bblog_size);
1828 rdev_for_each(rdev2, mddev)
1829 if (rdev2->desc_nr+1 > max_dev)
1830 max_dev = rdev2->desc_nr+1;
1832 if (max_dev > le32_to_cpu(sb->max_dev)) {
1834 sb->max_dev = cpu_to_le32(max_dev);
1835 rdev->sb_size = max_dev * 2 + 256;
1836 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1837 if (rdev->sb_size & bmask)
1838 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1840 max_dev = le32_to_cpu(sb->max_dev);
1842 for (i=0; i<max_dev;i++)
1843 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1845 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1846 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1848 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
1849 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
1850 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
1851 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
1854 rdev_for_each(rdev2, mddev) {
1856 if (test_bit(Faulty, &rdev2->flags))
1857 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1858 else if (test_bit(In_sync, &rdev2->flags))
1859 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1860 else if (test_bit(Journal, &rdev2->flags))
1861 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1862 else if (rdev2->raid_disk >= 0)
1863 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1865 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1868 sb->sb_csum = calc_sb_1_csum(sb);
1871 static unsigned long long
1872 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1874 struct mdp_superblock_1 *sb;
1875 sector_t max_sectors;
1876 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1877 return 0; /* component must fit device */
1878 if (rdev->data_offset != rdev->new_data_offset)
1879 return 0; /* too confusing */
1880 if (rdev->sb_start < rdev->data_offset) {
1881 /* minor versions 1 and 2; superblock before data */
1882 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1883 max_sectors -= rdev->data_offset;
1884 if (!num_sectors || num_sectors > max_sectors)
1885 num_sectors = max_sectors;
1886 } else if (rdev->mddev->bitmap_info.offset) {
1887 /* minor version 0 with bitmap we can't move */
1890 /* minor version 0; superblock after data */
1892 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1893 sb_start &= ~(sector_t)(4*2 - 1);
1894 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1895 if (!num_sectors || num_sectors > max_sectors)
1896 num_sectors = max_sectors;
1897 rdev->sb_start = sb_start;
1899 sb = page_address(rdev->sb_page);
1900 sb->data_size = cpu_to_le64(num_sectors);
1901 sb->super_offset = cpu_to_le64(rdev->sb_start);
1902 sb->sb_csum = calc_sb_1_csum(sb);
1904 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1906 } while (md_super_wait(rdev->mddev) < 0);
1912 super_1_allow_new_offset(struct md_rdev *rdev,
1913 unsigned long long new_offset)
1915 /* All necessary checks on new >= old have been done */
1916 struct bitmap *bitmap;
1917 if (new_offset >= rdev->data_offset)
1920 /* with 1.0 metadata, there is no metadata to tread on
1921 * so we can always move back */
1922 if (rdev->mddev->minor_version == 0)
1925 /* otherwise we must be sure not to step on
1926 * any metadata, so stay:
1927 * 36K beyond start of superblock
1928 * beyond end of badblocks
1929 * beyond write-intent bitmap
1931 if (rdev->sb_start + (32+4)*2 > new_offset)
1933 bitmap = rdev->mddev->bitmap;
1934 if (bitmap && !rdev->mddev->bitmap_info.file &&
1935 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1936 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1938 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1944 static struct super_type super_types[] = {
1947 .owner = THIS_MODULE,
1948 .load_super = super_90_load,
1949 .validate_super = super_90_validate,
1950 .sync_super = super_90_sync,
1951 .rdev_size_change = super_90_rdev_size_change,
1952 .allow_new_offset = super_90_allow_new_offset,
1956 .owner = THIS_MODULE,
1957 .load_super = super_1_load,
1958 .validate_super = super_1_validate,
1959 .sync_super = super_1_sync,
1960 .rdev_size_change = super_1_rdev_size_change,
1961 .allow_new_offset = super_1_allow_new_offset,
1965 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1967 if (mddev->sync_super) {
1968 mddev->sync_super(mddev, rdev);
1972 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1974 super_types[mddev->major_version].sync_super(mddev, rdev);
1977 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1979 struct md_rdev *rdev, *rdev2;
1982 rdev_for_each_rcu(rdev, mddev1) {
1983 if (test_bit(Faulty, &rdev->flags) ||
1984 test_bit(Journal, &rdev->flags) ||
1985 rdev->raid_disk == -1)
1987 rdev_for_each_rcu(rdev2, mddev2) {
1988 if (test_bit(Faulty, &rdev2->flags) ||
1989 test_bit(Journal, &rdev2->flags) ||
1990 rdev2->raid_disk == -1)
1992 if (rdev->bdev->bd_contains ==
1993 rdev2->bdev->bd_contains) {
2003 static LIST_HEAD(pending_raid_disks);
2006 * Try to register data integrity profile for an mddev
2008 * This is called when an array is started and after a disk has been kicked
2009 * from the array. It only succeeds if all working and active component devices
2010 * are integrity capable with matching profiles.
2012 int md_integrity_register(struct mddev *mddev)
2014 struct md_rdev *rdev, *reference = NULL;
2016 if (list_empty(&mddev->disks))
2017 return 0; /* nothing to do */
2018 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2019 return 0; /* shouldn't register, or already is */
2020 rdev_for_each(rdev, mddev) {
2021 /* skip spares and non-functional disks */
2022 if (test_bit(Faulty, &rdev->flags))
2024 if (rdev->raid_disk < 0)
2027 /* Use the first rdev as the reference */
2031 /* does this rdev's profile match the reference profile? */
2032 if (blk_integrity_compare(reference->bdev->bd_disk,
2033 rdev->bdev->bd_disk) < 0)
2036 if (!reference || !bdev_get_integrity(reference->bdev))
2039 * All component devices are integrity capable and have matching
2040 * profiles, register the common profile for the md device.
2042 blk_integrity_register(mddev->gendisk,
2043 bdev_get_integrity(reference->bdev));
2045 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2046 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2047 pr_err("md: failed to create integrity pool for %s\n",
2053 EXPORT_SYMBOL(md_integrity_register);
2056 * Attempt to add an rdev, but only if it is consistent with the current
2059 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2061 struct blk_integrity *bi_rdev;
2062 struct blk_integrity *bi_mddev;
2063 char name[BDEVNAME_SIZE];
2065 if (!mddev->gendisk)
2068 bi_rdev = bdev_get_integrity(rdev->bdev);
2069 bi_mddev = blk_get_integrity(mddev->gendisk);
2071 if (!bi_mddev) /* nothing to do */
2074 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2075 pr_err("%s: incompatible integrity profile for %s\n",
2076 mdname(mddev), bdevname(rdev->bdev, name));
2082 EXPORT_SYMBOL(md_integrity_add_rdev);
2084 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2086 char b[BDEVNAME_SIZE];
2090 /* prevent duplicates */
2091 if (find_rdev(mddev, rdev->bdev->bd_dev))
2094 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2095 if (!test_bit(Journal, &rdev->flags) &&
2097 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2099 /* Cannot change size, so fail
2100 * If mddev->level <= 0, then we don't care
2101 * about aligning sizes (e.g. linear)
2103 if (mddev->level > 0)
2106 mddev->dev_sectors = rdev->sectors;
2109 /* Verify rdev->desc_nr is unique.
2110 * If it is -1, assign a free number, else
2111 * check number is not in use
2114 if (rdev->desc_nr < 0) {
2117 choice = mddev->raid_disks;
2118 while (md_find_rdev_nr_rcu(mddev, choice))
2120 rdev->desc_nr = choice;
2122 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2128 if (!test_bit(Journal, &rdev->flags) &&
2129 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2130 pr_warn("md: %s: array is limited to %d devices\n",
2131 mdname(mddev), mddev->max_disks);
2134 bdevname(rdev->bdev,b);
2135 strreplace(b, '/', '!');
2137 rdev->mddev = mddev;
2138 pr_debug("md: bind<%s>\n", b);
2140 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2143 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2144 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2145 /* failure here is OK */;
2146 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2148 list_add_rcu(&rdev->same_set, &mddev->disks);
2149 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2151 /* May as well allow recovery to be retried once */
2152 mddev->recovery_disabled++;
2157 pr_warn("md: failed to register dev-%s for %s\n",
2162 static void md_delayed_delete(struct work_struct *ws)
2164 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2165 kobject_del(&rdev->kobj);
2166 kobject_put(&rdev->kobj);
2169 static void unbind_rdev_from_array(struct md_rdev *rdev)
2171 char b[BDEVNAME_SIZE];
2173 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2174 list_del_rcu(&rdev->same_set);
2175 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2177 sysfs_remove_link(&rdev->kobj, "block");
2178 sysfs_put(rdev->sysfs_state);
2179 rdev->sysfs_state = NULL;
2180 rdev->badblocks.count = 0;
2181 /* We need to delay this, otherwise we can deadlock when
2182 * writing to 'remove' to "dev/state". We also need
2183 * to delay it due to rcu usage.
2186 INIT_WORK(&rdev->del_work, md_delayed_delete);
2187 kobject_get(&rdev->kobj);
2188 queue_work(md_misc_wq, &rdev->del_work);
2192 * prevent the device from being mounted, repartitioned or
2193 * otherwise reused by a RAID array (or any other kernel
2194 * subsystem), by bd_claiming the device.
2196 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2199 struct block_device *bdev;
2200 char b[BDEVNAME_SIZE];
2202 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2203 shared ? (struct md_rdev *)lock_rdev : rdev);
2205 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2206 return PTR_ERR(bdev);
2212 static void unlock_rdev(struct md_rdev *rdev)
2214 struct block_device *bdev = rdev->bdev;
2216 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2219 void md_autodetect_dev(dev_t dev);
2221 static void export_rdev(struct md_rdev *rdev)
2223 char b[BDEVNAME_SIZE];
2225 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2226 md_rdev_clear(rdev);
2228 if (test_bit(AutoDetected, &rdev->flags))
2229 md_autodetect_dev(rdev->bdev->bd_dev);
2232 kobject_put(&rdev->kobj);
2235 void md_kick_rdev_from_array(struct md_rdev *rdev)
2237 unbind_rdev_from_array(rdev);
2240 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2242 static void export_array(struct mddev *mddev)
2244 struct md_rdev *rdev;
2246 while (!list_empty(&mddev->disks)) {
2247 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2249 md_kick_rdev_from_array(rdev);
2251 mddev->raid_disks = 0;
2252 mddev->major_version = 0;
2255 static void sync_sbs(struct mddev *mddev, int nospares)
2257 /* Update each superblock (in-memory image), but
2258 * if we are allowed to, skip spares which already
2259 * have the right event counter, or have one earlier
2260 * (which would mean they aren't being marked as dirty
2261 * with the rest of the array)
2263 struct md_rdev *rdev;
2264 rdev_for_each(rdev, mddev) {
2265 if (rdev->sb_events == mddev->events ||
2267 rdev->raid_disk < 0 &&
2268 rdev->sb_events+1 == mddev->events)) {
2269 /* Don't update this superblock */
2270 rdev->sb_loaded = 2;
2272 sync_super(mddev, rdev);
2273 rdev->sb_loaded = 1;
2278 static bool does_sb_need_changing(struct mddev *mddev)
2280 struct md_rdev *rdev;
2281 struct mdp_superblock_1 *sb;
2284 /* Find a good rdev */
2285 rdev_for_each(rdev, mddev)
2286 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2289 /* No good device found. */
2293 sb = page_address(rdev->sb_page);
2294 /* Check if a device has become faulty or a spare become active */
2295 rdev_for_each(rdev, mddev) {
2296 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2297 /* Device activated? */
2298 if (role == 0xffff && rdev->raid_disk >=0 &&
2299 !test_bit(Faulty, &rdev->flags))
2301 /* Device turned faulty? */
2302 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2306 /* Check if any mddev parameters have changed */
2307 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2308 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2309 (mddev->layout != le32_to_cpu(sb->layout)) ||
2310 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2311 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2317 void md_update_sb(struct mddev *mddev, int force_change)
2319 struct md_rdev *rdev;
2322 int any_badblocks_changed = 0;
2327 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2332 if (mddev_is_clustered(mddev)) {
2333 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2335 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2337 ret = md_cluster_ops->metadata_update_start(mddev);
2338 /* Has someone else has updated the sb */
2339 if (!does_sb_need_changing(mddev)) {
2341 md_cluster_ops->metadata_update_cancel(mddev);
2342 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2343 BIT(MD_SB_CHANGE_DEVS) |
2344 BIT(MD_SB_CHANGE_CLEAN));
2349 /* First make sure individual recovery_offsets are correct */
2350 rdev_for_each(rdev, mddev) {
2351 if (rdev->raid_disk >= 0 &&
2352 mddev->delta_disks >= 0 &&
2353 !test_bit(Journal, &rdev->flags) &&
2354 !test_bit(In_sync, &rdev->flags) &&
2355 mddev->curr_resync_completed > rdev->recovery_offset)
2356 rdev->recovery_offset = mddev->curr_resync_completed;
2359 if (!mddev->persistent) {
2360 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2361 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2362 if (!mddev->external) {
2363 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2364 rdev_for_each(rdev, mddev) {
2365 if (rdev->badblocks.changed) {
2366 rdev->badblocks.changed = 0;
2367 ack_all_badblocks(&rdev->badblocks);
2368 md_error(mddev, rdev);
2370 clear_bit(Blocked, &rdev->flags);
2371 clear_bit(BlockedBadBlocks, &rdev->flags);
2372 wake_up(&rdev->blocked_wait);
2375 wake_up(&mddev->sb_wait);
2379 spin_lock(&mddev->lock);
2381 mddev->utime = ktime_get_real_seconds();
2383 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2385 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2386 /* just a clean<-> dirty transition, possibly leave spares alone,
2387 * though if events isn't the right even/odd, we will have to do
2393 if (mddev->degraded)
2394 /* If the array is degraded, then skipping spares is both
2395 * dangerous and fairly pointless.
2396 * Dangerous because a device that was removed from the array
2397 * might have a event_count that still looks up-to-date,
2398 * so it can be re-added without a resync.
2399 * Pointless because if there are any spares to skip,
2400 * then a recovery will happen and soon that array won't
2401 * be degraded any more and the spare can go back to sleep then.
2405 sync_req = mddev->in_sync;
2407 /* If this is just a dirty<->clean transition, and the array is clean
2408 * and 'events' is odd, we can roll back to the previous clean state */
2410 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2411 && mddev->can_decrease_events
2412 && mddev->events != 1) {
2414 mddev->can_decrease_events = 0;
2416 /* otherwise we have to go forward and ... */
2418 mddev->can_decrease_events = nospares;
2422 * This 64-bit counter should never wrap.
2423 * Either we are in around ~1 trillion A.C., assuming
2424 * 1 reboot per second, or we have a bug...
2426 WARN_ON(mddev->events == 0);
2428 rdev_for_each(rdev, mddev) {
2429 if (rdev->badblocks.changed)
2430 any_badblocks_changed++;
2431 if (test_bit(Faulty, &rdev->flags))
2432 set_bit(FaultRecorded, &rdev->flags);
2435 sync_sbs(mddev, nospares);
2436 spin_unlock(&mddev->lock);
2438 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2439 mdname(mddev), mddev->in_sync);
2442 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2444 bitmap_update_sb(mddev->bitmap);
2445 rdev_for_each(rdev, mddev) {
2446 char b[BDEVNAME_SIZE];
2448 if (rdev->sb_loaded != 1)
2449 continue; /* no noise on spare devices */
2451 if (!test_bit(Faulty, &rdev->flags)) {
2452 md_super_write(mddev,rdev,
2453 rdev->sb_start, rdev->sb_size,
2455 pr_debug("md: (write) %s's sb offset: %llu\n",
2456 bdevname(rdev->bdev, b),
2457 (unsigned long long)rdev->sb_start);
2458 rdev->sb_events = mddev->events;
2459 if (rdev->badblocks.size) {
2460 md_super_write(mddev, rdev,
2461 rdev->badblocks.sector,
2462 rdev->badblocks.size << 9,
2464 rdev->badblocks.size = 0;
2468 pr_debug("md: %s (skipping faulty)\n",
2469 bdevname(rdev->bdev, b));
2471 if (mddev->level == LEVEL_MULTIPATH)
2472 /* only need to write one superblock... */
2475 if (md_super_wait(mddev) < 0)
2477 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2479 if (mddev_is_clustered(mddev) && ret == 0)
2480 md_cluster_ops->metadata_update_finish(mddev);
2482 if (mddev->in_sync != sync_req ||
2483 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2484 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2485 /* have to write it out again */
2487 wake_up(&mddev->sb_wait);
2488 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2489 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2491 rdev_for_each(rdev, mddev) {
2492 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2493 clear_bit(Blocked, &rdev->flags);
2495 if (any_badblocks_changed)
2496 ack_all_badblocks(&rdev->badblocks);
2497 clear_bit(BlockedBadBlocks, &rdev->flags);
2498 wake_up(&rdev->blocked_wait);
2501 EXPORT_SYMBOL(md_update_sb);
2503 static int add_bound_rdev(struct md_rdev *rdev)
2505 struct mddev *mddev = rdev->mddev;
2507 bool add_journal = test_bit(Journal, &rdev->flags);
2509 if (!mddev->pers->hot_remove_disk || add_journal) {
2510 /* If there is hot_add_disk but no hot_remove_disk
2511 * then added disks for geometry changes,
2512 * and should be added immediately.
2514 super_types[mddev->major_version].
2515 validate_super(mddev, rdev);
2517 mddev_suspend(mddev);
2518 err = mddev->pers->hot_add_disk(mddev, rdev);
2520 mddev_resume(mddev);
2522 md_kick_rdev_from_array(rdev);
2526 sysfs_notify_dirent_safe(rdev->sysfs_state);
2528 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2529 if (mddev->degraded)
2530 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2531 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2532 md_new_event(mddev);
2533 md_wakeup_thread(mddev->thread);
2537 /* words written to sysfs files may, or may not, be \n terminated.
2538 * We want to accept with case. For this we use cmd_match.
2540 static int cmd_match(const char *cmd, const char *str)
2542 /* See if cmd, written into a sysfs file, matches
2543 * str. They must either be the same, or cmd can
2544 * have a trailing newline
2546 while (*cmd && *str && *cmd == *str) {
2557 struct rdev_sysfs_entry {
2558 struct attribute attr;
2559 ssize_t (*show)(struct md_rdev *, char *);
2560 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2564 state_show(struct md_rdev *rdev, char *page)
2568 unsigned long flags = ACCESS_ONCE(rdev->flags);
2570 if (test_bit(Faulty, &flags) ||
2571 (!test_bit(ExternalBbl, &flags) &&
2572 rdev->badblocks.unacked_exist))
2573 len += sprintf(page+len, "faulty%s", sep);
2574 if (test_bit(In_sync, &flags))
2575 len += sprintf(page+len, "in_sync%s", sep);
2576 if (test_bit(Journal, &flags))
2577 len += sprintf(page+len, "journal%s", sep);
2578 if (test_bit(WriteMostly, &flags))
2579 len += sprintf(page+len, "write_mostly%s", sep);
2580 if (test_bit(Blocked, &flags) ||
2581 (rdev->badblocks.unacked_exist
2582 && !test_bit(Faulty, &flags)))
2583 len += sprintf(page+len, "blocked%s", sep);
2584 if (!test_bit(Faulty, &flags) &&
2585 !test_bit(Journal, &flags) &&
2586 !test_bit(In_sync, &flags))
2587 len += sprintf(page+len, "spare%s", sep);
2588 if (test_bit(WriteErrorSeen, &flags))
2589 len += sprintf(page+len, "write_error%s", sep);
2590 if (test_bit(WantReplacement, &flags))
2591 len += sprintf(page+len, "want_replacement%s", sep);
2592 if (test_bit(Replacement, &flags))
2593 len += sprintf(page+len, "replacement%s", sep);
2594 if (test_bit(ExternalBbl, &flags))
2595 len += sprintf(page+len, "external_bbl%s", sep);
2596 if (test_bit(FailFast, &flags))
2597 len += sprintf(page+len, "failfast%s", sep);
2602 return len+sprintf(page+len, "\n");
2606 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2609 * faulty - simulates an error
2610 * remove - disconnects the device
2611 * writemostly - sets write_mostly
2612 * -writemostly - clears write_mostly
2613 * blocked - sets the Blocked flags
2614 * -blocked - clears the Blocked and possibly simulates an error
2615 * insync - sets Insync providing device isn't active
2616 * -insync - clear Insync for a device with a slot assigned,
2617 * so that it gets rebuilt based on bitmap
2618 * write_error - sets WriteErrorSeen
2619 * -write_error - clears WriteErrorSeen
2620 * {,-}failfast - set/clear FailFast
2623 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2624 md_error(rdev->mddev, rdev);
2625 if (test_bit(Faulty, &rdev->flags))
2629 } else if (cmd_match(buf, "remove")) {
2630 if (rdev->mddev->pers) {
2631 clear_bit(Blocked, &rdev->flags);
2632 remove_and_add_spares(rdev->mddev, rdev);
2634 if (rdev->raid_disk >= 0)
2637 struct mddev *mddev = rdev->mddev;
2639 if (mddev_is_clustered(mddev))
2640 err = md_cluster_ops->remove_disk(mddev, rdev);
2643 md_kick_rdev_from_array(rdev);
2645 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2646 md_wakeup_thread(mddev->thread);
2648 md_new_event(mddev);
2651 } else if (cmd_match(buf, "writemostly")) {
2652 set_bit(WriteMostly, &rdev->flags);
2654 } else if (cmd_match(buf, "-writemostly")) {
2655 clear_bit(WriteMostly, &rdev->flags);
2657 } else if (cmd_match(buf, "blocked")) {
2658 set_bit(Blocked, &rdev->flags);
2660 } else if (cmd_match(buf, "-blocked")) {
2661 if (!test_bit(Faulty, &rdev->flags) &&
2662 !test_bit(ExternalBbl, &rdev->flags) &&
2663 rdev->badblocks.unacked_exist) {
2664 /* metadata handler doesn't understand badblocks,
2665 * so we need to fail the device
2667 md_error(rdev->mddev, rdev);
2669 clear_bit(Blocked, &rdev->flags);
2670 clear_bit(BlockedBadBlocks, &rdev->flags);
2671 wake_up(&rdev->blocked_wait);
2672 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2673 md_wakeup_thread(rdev->mddev->thread);
2676 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2677 set_bit(In_sync, &rdev->flags);
2679 } else if (cmd_match(buf, "failfast")) {
2680 set_bit(FailFast, &rdev->flags);
2682 } else if (cmd_match(buf, "-failfast")) {
2683 clear_bit(FailFast, &rdev->flags);
2685 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2686 !test_bit(Journal, &rdev->flags)) {
2687 if (rdev->mddev->pers == NULL) {
2688 clear_bit(In_sync, &rdev->flags);
2689 rdev->saved_raid_disk = rdev->raid_disk;
2690 rdev->raid_disk = -1;
2693 } else if (cmd_match(buf, "write_error")) {
2694 set_bit(WriteErrorSeen, &rdev->flags);
2696 } else if (cmd_match(buf, "-write_error")) {
2697 clear_bit(WriteErrorSeen, &rdev->flags);
2699 } else if (cmd_match(buf, "want_replacement")) {
2700 /* Any non-spare device that is not a replacement can
2701 * become want_replacement at any time, but we then need to
2702 * check if recovery is needed.
2704 if (rdev->raid_disk >= 0 &&
2705 !test_bit(Journal, &rdev->flags) &&
2706 !test_bit(Replacement, &rdev->flags))
2707 set_bit(WantReplacement, &rdev->flags);
2708 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2709 md_wakeup_thread(rdev->mddev->thread);
2711 } else if (cmd_match(buf, "-want_replacement")) {
2712 /* Clearing 'want_replacement' is always allowed.
2713 * Once replacements starts it is too late though.
2716 clear_bit(WantReplacement, &rdev->flags);
2717 } else if (cmd_match(buf, "replacement")) {
2718 /* Can only set a device as a replacement when array has not
2719 * yet been started. Once running, replacement is automatic
2720 * from spares, or by assigning 'slot'.
2722 if (rdev->mddev->pers)
2725 set_bit(Replacement, &rdev->flags);
2728 } else if (cmd_match(buf, "-replacement")) {
2729 /* Similarly, can only clear Replacement before start */
2730 if (rdev->mddev->pers)
2733 clear_bit(Replacement, &rdev->flags);
2736 } else if (cmd_match(buf, "re-add")) {
2737 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2738 /* clear_bit is performed _after_ all the devices
2739 * have their local Faulty bit cleared. If any writes
2740 * happen in the meantime in the local node, they
2741 * will land in the local bitmap, which will be synced
2742 * by this node eventually
2744 if (!mddev_is_clustered(rdev->mddev) ||
2745 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2746 clear_bit(Faulty, &rdev->flags);
2747 err = add_bound_rdev(rdev);
2751 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2752 set_bit(ExternalBbl, &rdev->flags);
2753 rdev->badblocks.shift = 0;
2755 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2756 clear_bit(ExternalBbl, &rdev->flags);
2760 sysfs_notify_dirent_safe(rdev->sysfs_state);
2761 return err ? err : len;
2763 static struct rdev_sysfs_entry rdev_state =
2764 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2767 errors_show(struct md_rdev *rdev, char *page)
2769 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2773 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2778 rv = kstrtouint(buf, 10, &n);
2781 atomic_set(&rdev->corrected_errors, n);
2784 static struct rdev_sysfs_entry rdev_errors =
2785 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2788 slot_show(struct md_rdev *rdev, char *page)
2790 if (test_bit(Journal, &rdev->flags))
2791 return sprintf(page, "journal\n");
2792 else if (rdev->raid_disk < 0)
2793 return sprintf(page, "none\n");
2795 return sprintf(page, "%d\n", rdev->raid_disk);
2799 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2804 if (test_bit(Journal, &rdev->flags))
2806 if (strncmp(buf, "none", 4)==0)
2809 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2813 if (rdev->mddev->pers && slot == -1) {
2814 /* Setting 'slot' on an active array requires also
2815 * updating the 'rd%d' link, and communicating
2816 * with the personality with ->hot_*_disk.
2817 * For now we only support removing
2818 * failed/spare devices. This normally happens automatically,
2819 * but not when the metadata is externally managed.
2821 if (rdev->raid_disk == -1)
2823 /* personality does all needed checks */
2824 if (rdev->mddev->pers->hot_remove_disk == NULL)
2826 clear_bit(Blocked, &rdev->flags);
2827 remove_and_add_spares(rdev->mddev, rdev);
2828 if (rdev->raid_disk >= 0)
2830 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2831 md_wakeup_thread(rdev->mddev->thread);
2832 } else if (rdev->mddev->pers) {
2833 /* Activating a spare .. or possibly reactivating
2834 * if we ever get bitmaps working here.
2838 if (rdev->raid_disk != -1)
2841 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2844 if (rdev->mddev->pers->hot_add_disk == NULL)
2847 if (slot >= rdev->mddev->raid_disks &&
2848 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2851 rdev->raid_disk = slot;
2852 if (test_bit(In_sync, &rdev->flags))
2853 rdev->saved_raid_disk = slot;
2855 rdev->saved_raid_disk = -1;
2856 clear_bit(In_sync, &rdev->flags);
2857 clear_bit(Bitmap_sync, &rdev->flags);
2858 err = rdev->mddev->pers->
2859 hot_add_disk(rdev->mddev, rdev);
2861 rdev->raid_disk = -1;
2864 sysfs_notify_dirent_safe(rdev->sysfs_state);
2865 if (sysfs_link_rdev(rdev->mddev, rdev))
2866 /* failure here is OK */;
2867 /* don't wakeup anyone, leave that to userspace. */
2869 if (slot >= rdev->mddev->raid_disks &&
2870 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2872 rdev->raid_disk = slot;
2873 /* assume it is working */
2874 clear_bit(Faulty, &rdev->flags);
2875 clear_bit(WriteMostly, &rdev->flags);
2876 set_bit(In_sync, &rdev->flags);
2877 sysfs_notify_dirent_safe(rdev->sysfs_state);
2882 static struct rdev_sysfs_entry rdev_slot =
2883 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2886 offset_show(struct md_rdev *rdev, char *page)
2888 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2892 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2894 unsigned long long offset;
2895 if (kstrtoull(buf, 10, &offset) < 0)
2897 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2899 if (rdev->sectors && rdev->mddev->external)
2900 /* Must set offset before size, so overlap checks
2903 rdev->data_offset = offset;
2904 rdev->new_data_offset = offset;
2908 static struct rdev_sysfs_entry rdev_offset =
2909 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2911 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2913 return sprintf(page, "%llu\n",
2914 (unsigned long long)rdev->new_data_offset);
2917 static ssize_t new_offset_store(struct md_rdev *rdev,
2918 const char *buf, size_t len)
2920 unsigned long long new_offset;
2921 struct mddev *mddev = rdev->mddev;
2923 if (kstrtoull(buf, 10, &new_offset) < 0)
2926 if (mddev->sync_thread ||
2927 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2929 if (new_offset == rdev->data_offset)
2930 /* reset is always permitted */
2932 else if (new_offset > rdev->data_offset) {
2933 /* must not push array size beyond rdev_sectors */
2934 if (new_offset - rdev->data_offset
2935 + mddev->dev_sectors > rdev->sectors)
2938 /* Metadata worries about other space details. */
2940 /* decreasing the offset is inconsistent with a backwards
2943 if (new_offset < rdev->data_offset &&
2944 mddev->reshape_backwards)
2946 /* Increasing offset is inconsistent with forwards
2947 * reshape. reshape_direction should be set to
2948 * 'backwards' first.
2950 if (new_offset > rdev->data_offset &&
2951 !mddev->reshape_backwards)
2954 if (mddev->pers && mddev->persistent &&
2955 !super_types[mddev->major_version]
2956 .allow_new_offset(rdev, new_offset))
2958 rdev->new_data_offset = new_offset;
2959 if (new_offset > rdev->data_offset)
2960 mddev->reshape_backwards = 1;
2961 else if (new_offset < rdev->data_offset)
2962 mddev->reshape_backwards = 0;
2966 static struct rdev_sysfs_entry rdev_new_offset =
2967 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2970 rdev_size_show(struct md_rdev *rdev, char *page)
2972 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2975 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2977 /* check if two start/length pairs overlap */
2985 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2987 unsigned long long blocks;
2990 if (kstrtoull(buf, 10, &blocks) < 0)
2993 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2994 return -EINVAL; /* sector conversion overflow */
2997 if (new != blocks * 2)
2998 return -EINVAL; /* unsigned long long to sector_t overflow */
3005 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3007 struct mddev *my_mddev = rdev->mddev;
3008 sector_t oldsectors = rdev->sectors;
3011 if (test_bit(Journal, &rdev->flags))
3013 if (strict_blocks_to_sectors(buf, §ors) < 0)
3015 if (rdev->data_offset != rdev->new_data_offset)
3016 return -EINVAL; /* too confusing */
3017 if (my_mddev->pers && rdev->raid_disk >= 0) {
3018 if (my_mddev->persistent) {
3019 sectors = super_types[my_mddev->major_version].
3020 rdev_size_change(rdev, sectors);
3023 } else if (!sectors)
3024 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3026 if (!my_mddev->pers->resize)
3027 /* Cannot change size for RAID0 or Linear etc */
3030 if (sectors < my_mddev->dev_sectors)
3031 return -EINVAL; /* component must fit device */
3033 rdev->sectors = sectors;
3034 if (sectors > oldsectors && my_mddev->external) {
3035 /* Need to check that all other rdevs with the same
3036 * ->bdev do not overlap. 'rcu' is sufficient to walk
3037 * the rdev lists safely.
3038 * This check does not provide a hard guarantee, it
3039 * just helps avoid dangerous mistakes.
3041 struct mddev *mddev;
3043 struct list_head *tmp;
3046 for_each_mddev(mddev, tmp) {
3047 struct md_rdev *rdev2;
3049 rdev_for_each(rdev2, mddev)
3050 if (rdev->bdev == rdev2->bdev &&
3052 overlaps(rdev->data_offset, rdev->sectors,
3065 /* Someone else could have slipped in a size
3066 * change here, but doing so is just silly.
3067 * We put oldsectors back because we *know* it is
3068 * safe, and trust userspace not to race with
3071 rdev->sectors = oldsectors;
3078 static struct rdev_sysfs_entry rdev_size =
3079 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3081 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3083 unsigned long long recovery_start = rdev->recovery_offset;
3085 if (test_bit(In_sync, &rdev->flags) ||
3086 recovery_start == MaxSector)
3087 return sprintf(page, "none\n");
3089 return sprintf(page, "%llu\n", recovery_start);
3092 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3094 unsigned long long recovery_start;
3096 if (cmd_match(buf, "none"))
3097 recovery_start = MaxSector;
3098 else if (kstrtoull(buf, 10, &recovery_start))
3101 if (rdev->mddev->pers &&
3102 rdev->raid_disk >= 0)
3105 rdev->recovery_offset = recovery_start;
3106 if (recovery_start == MaxSector)
3107 set_bit(In_sync, &rdev->flags);
3109 clear_bit(In_sync, &rdev->flags);
3113 static struct rdev_sysfs_entry rdev_recovery_start =
3114 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3116 /* sysfs access to bad-blocks list.
3117 * We present two files.
3118 * 'bad-blocks' lists sector numbers and lengths of ranges that
3119 * are recorded as bad. The list is truncated to fit within
3120 * the one-page limit of sysfs.
3121 * Writing "sector length" to this file adds an acknowledged
3123 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3124 * been acknowledged. Writing to this file adds bad blocks
3125 * without acknowledging them. This is largely for testing.
3127 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3129 return badblocks_show(&rdev->badblocks, page, 0);
3131 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3133 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3134 /* Maybe that ack was all we needed */
3135 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3136 wake_up(&rdev->blocked_wait);
3139 static struct rdev_sysfs_entry rdev_bad_blocks =
3140 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3142 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3144 return badblocks_show(&rdev->badblocks, page, 1);
3146 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3148 return badblocks_store(&rdev->badblocks, page, len, 1);
3150 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3151 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3154 ppl_sector_show(struct md_rdev *rdev, char *page)
3156 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3160 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3162 unsigned long long sector;
3164 if (kstrtoull(buf, 10, §or) < 0)
3166 if (sector != (sector_t)sector)
3169 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3170 rdev->raid_disk >= 0)
3173 if (rdev->mddev->persistent) {
3174 if (rdev->mddev->major_version == 0)
3176 if ((sector > rdev->sb_start &&
3177 sector - rdev->sb_start > S16_MAX) ||
3178 (sector < rdev->sb_start &&
3179 rdev->sb_start - sector > -S16_MIN))
3181 rdev->ppl.offset = sector - rdev->sb_start;
3182 } else if (!rdev->mddev->external) {
3185 rdev->ppl.sector = sector;
3189 static struct rdev_sysfs_entry rdev_ppl_sector =
3190 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3193 ppl_size_show(struct md_rdev *rdev, char *page)
3195 return sprintf(page, "%u\n", rdev->ppl.size);
3199 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3203 if (kstrtouint(buf, 10, &size) < 0)
3206 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3207 rdev->raid_disk >= 0)
3210 if (rdev->mddev->persistent) {
3211 if (rdev->mddev->major_version == 0)
3215 } else if (!rdev->mddev->external) {
3218 rdev->ppl.size = size;
3222 static struct rdev_sysfs_entry rdev_ppl_size =
3223 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3225 static struct attribute *rdev_default_attrs[] = {
3230 &rdev_new_offset.attr,
3232 &rdev_recovery_start.attr,
3233 &rdev_bad_blocks.attr,
3234 &rdev_unack_bad_blocks.attr,
3235 &rdev_ppl_sector.attr,
3236 &rdev_ppl_size.attr,
3240 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3242 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3243 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3249 return entry->show(rdev, page);
3253 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3254 const char *page, size_t length)
3256 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3257 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3259 struct mddev *mddev = rdev->mddev;
3263 if (!capable(CAP_SYS_ADMIN))
3265 rv = mddev ? mddev_lock(mddev): -EBUSY;
3267 if (rdev->mddev == NULL)
3270 rv = entry->store(rdev, page, length);
3271 mddev_unlock(mddev);
3276 static void rdev_free(struct kobject *ko)
3278 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3281 static const struct sysfs_ops rdev_sysfs_ops = {
3282 .show = rdev_attr_show,
3283 .store = rdev_attr_store,
3285 static struct kobj_type rdev_ktype = {
3286 .release = rdev_free,
3287 .sysfs_ops = &rdev_sysfs_ops,
3288 .default_attrs = rdev_default_attrs,
3291 int md_rdev_init(struct md_rdev *rdev)
3294 rdev->saved_raid_disk = -1;
3295 rdev->raid_disk = -1;
3297 rdev->data_offset = 0;
3298 rdev->new_data_offset = 0;
3299 rdev->sb_events = 0;
3300 rdev->last_read_error = 0;
3301 rdev->sb_loaded = 0;
3302 rdev->bb_page = NULL;
3303 atomic_set(&rdev->nr_pending, 0);
3304 atomic_set(&rdev->read_errors, 0);
3305 atomic_set(&rdev->corrected_errors, 0);
3307 INIT_LIST_HEAD(&rdev->same_set);
3308 init_waitqueue_head(&rdev->blocked_wait);
3310 /* Add space to store bad block list.
3311 * This reserves the space even on arrays where it cannot
3312 * be used - I wonder if that matters
3314 return badblocks_init(&rdev->badblocks, 0);
3316 EXPORT_SYMBOL_GPL(md_rdev_init);
3318 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3320 * mark the device faulty if:
3322 * - the device is nonexistent (zero size)
3323 * - the device has no valid superblock
3325 * a faulty rdev _never_ has rdev->sb set.
3327 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3329 char b[BDEVNAME_SIZE];
3331 struct md_rdev *rdev;
3334 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3336 return ERR_PTR(-ENOMEM);
3338 err = md_rdev_init(rdev);
3341 err = alloc_disk_sb(rdev);
3345 err = lock_rdev(rdev, newdev, super_format == -2);
3349 kobject_init(&rdev->kobj, &rdev_ktype);
3351 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3353 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3354 bdevname(rdev->bdev,b));
3359 if (super_format >= 0) {
3360 err = super_types[super_format].
3361 load_super(rdev, NULL, super_minor);
3362 if (err == -EINVAL) {
3363 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3364 bdevname(rdev->bdev,b),
3365 super_format, super_minor);
3369 pr_warn("md: could not read %s's sb, not importing!\n",
3370 bdevname(rdev->bdev,b));
3380 md_rdev_clear(rdev);
3382 return ERR_PTR(err);
3386 * Check a full RAID array for plausibility
3389 static void analyze_sbs(struct mddev *mddev)
3392 struct md_rdev *rdev, *freshest, *tmp;
3393 char b[BDEVNAME_SIZE];
3396 rdev_for_each_safe(rdev, tmp, mddev)
3397 switch (super_types[mddev->major_version].
3398 load_super(rdev, freshest, mddev->minor_version)) {
3405 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3406 bdevname(rdev->bdev,b));
3407 md_kick_rdev_from_array(rdev);
3410 super_types[mddev->major_version].
3411 validate_super(mddev, freshest);
3414 rdev_for_each_safe(rdev, tmp, mddev) {
3415 if (mddev->max_disks &&
3416 (rdev->desc_nr >= mddev->max_disks ||
3417 i > mddev->max_disks)) {
3418 pr_warn("md: %s: %s: only %d devices permitted\n",
3419 mdname(mddev), bdevname(rdev->bdev, b),
3421 md_kick_rdev_from_array(rdev);
3424 if (rdev != freshest) {
3425 if (super_types[mddev->major_version].
3426 validate_super(mddev, rdev)) {
3427 pr_warn("md: kicking non-fresh %s from array!\n",
3428 bdevname(rdev->bdev,b));
3429 md_kick_rdev_from_array(rdev);
3433 if (mddev->level == LEVEL_MULTIPATH) {
3434 rdev->desc_nr = i++;
3435 rdev->raid_disk = rdev->desc_nr;
3436 set_bit(In_sync, &rdev->flags);
3437 } else if (rdev->raid_disk >=
3438 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3439 !test_bit(Journal, &rdev->flags)) {
3440 rdev->raid_disk = -1;
3441 clear_bit(In_sync, &rdev->flags);
3446 /* Read a fixed-point number.
3447 * Numbers in sysfs attributes should be in "standard" units where
3448 * possible, so time should be in seconds.
3449 * However we internally use a a much smaller unit such as
3450 * milliseconds or jiffies.
3451 * This function takes a decimal number with a possible fractional
3452 * component, and produces an integer which is the result of
3453 * multiplying that number by 10^'scale'.
3454 * all without any floating-point arithmetic.
3456 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3458 unsigned long result = 0;
3460 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3463 else if (decimals < scale) {
3466 result = result * 10 + value;
3478 while (decimals < scale) {
3487 safe_delay_show(struct mddev *mddev, char *page)
3489 int msec = (mddev->safemode_delay*1000)/HZ;
3490 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3493 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3497 if (mddev_is_clustered(mddev)) {
3498 pr_warn("md: Safemode is disabled for clustered mode\n");
3502 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3505 mddev->safemode_delay = 0;
3507 unsigned long old_delay = mddev->safemode_delay;
3508 unsigned long new_delay = (msec*HZ)/1000;
3512 mddev->safemode_delay = new_delay;
3513 if (new_delay < old_delay || old_delay == 0)
3514 mod_timer(&mddev->safemode_timer, jiffies+1);
3518 static struct md_sysfs_entry md_safe_delay =
3519 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3522 level_show(struct mddev *mddev, char *page)
3524 struct md_personality *p;
3526 spin_lock(&mddev->lock);
3529 ret = sprintf(page, "%s\n", p->name);
3530 else if (mddev->clevel[0])
3531 ret = sprintf(page, "%s\n", mddev->clevel);
3532 else if (mddev->level != LEVEL_NONE)
3533 ret = sprintf(page, "%d\n", mddev->level);
3536 spin_unlock(&mddev->lock);
3541 level_store(struct mddev *mddev, const char *buf, size_t len)
3546 struct md_personality *pers, *oldpers;
3548 void *priv, *oldpriv;
3549 struct md_rdev *rdev;
3551 if (slen == 0 || slen >= sizeof(clevel))
3554 rv = mddev_lock(mddev);
3558 if (mddev->pers == NULL) {
3559 strncpy(mddev->clevel, buf, slen);
3560 if (mddev->clevel[slen-1] == '\n')
3562 mddev->clevel[slen] = 0;
3563 mddev->level = LEVEL_NONE;
3571 /* request to change the personality. Need to ensure:
3572 * - array is not engaged in resync/recovery/reshape
3573 * - old personality can be suspended
3574 * - new personality will access other array.
3578 if (mddev->sync_thread ||
3579 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3580 mddev->reshape_position != MaxSector ||
3581 mddev->sysfs_active)
3585 if (!mddev->pers->quiesce) {
3586 pr_warn("md: %s: %s does not support online personality change\n",
3587 mdname(mddev), mddev->pers->name);
3591 /* Now find the new personality */
3592 strncpy(clevel, buf, slen);
3593 if (clevel[slen-1] == '\n')
3596 if (kstrtol(clevel, 10, &level))
3599 if (request_module("md-%s", clevel) != 0)
3600 request_module("md-level-%s", clevel);
3601 spin_lock(&pers_lock);
3602 pers = find_pers(level, clevel);
3603 if (!pers || !try_module_get(pers->owner)) {
3604 spin_unlock(&pers_lock);
3605 pr_warn("md: personality %s not loaded\n", clevel);
3609 spin_unlock(&pers_lock);
3611 if (pers == mddev->pers) {
3612 /* Nothing to do! */
3613 module_put(pers->owner);
3617 if (!pers->takeover) {
3618 module_put(pers->owner);
3619 pr_warn("md: %s: %s does not support personality takeover\n",
3620 mdname(mddev), clevel);
3625 rdev_for_each(rdev, mddev)
3626 rdev->new_raid_disk = rdev->raid_disk;
3628 /* ->takeover must set new_* and/or delta_disks
3629 * if it succeeds, and may set them when it fails.
3631 priv = pers->takeover(mddev);
3633 mddev->new_level = mddev->level;
3634 mddev->new_layout = mddev->layout;
3635 mddev->new_chunk_sectors = mddev->chunk_sectors;
3636 mddev->raid_disks -= mddev->delta_disks;
3637 mddev->delta_disks = 0;
3638 mddev->reshape_backwards = 0;
3639 module_put(pers->owner);
3640 pr_warn("md: %s: %s would not accept array\n",
3641 mdname(mddev), clevel);
3646 /* Looks like we have a winner */
3647 mddev_suspend(mddev);
3648 mddev_detach(mddev);
3650 spin_lock(&mddev->lock);
3651 oldpers = mddev->pers;
3652 oldpriv = mddev->private;
3654 mddev->private = priv;
3655 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3656 mddev->level = mddev->new_level;
3657 mddev->layout = mddev->new_layout;
3658 mddev->chunk_sectors = mddev->new_chunk_sectors;
3659 mddev->delta_disks = 0;
3660 mddev->reshape_backwards = 0;
3661 mddev->degraded = 0;
3662 spin_unlock(&mddev->lock);
3664 if (oldpers->sync_request == NULL &&
3666 /* We are converting from a no-redundancy array
3667 * to a redundancy array and metadata is managed
3668 * externally so we need to be sure that writes
3669 * won't block due to a need to transition
3671 * until external management is started.
3674 mddev->safemode_delay = 0;
3675 mddev->safemode = 0;
3678 oldpers->free(mddev, oldpriv);
3680 if (oldpers->sync_request == NULL &&
3681 pers->sync_request != NULL) {
3682 /* need to add the md_redundancy_group */
3683 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3684 pr_warn("md: cannot register extra attributes for %s\n",
3686 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3688 if (oldpers->sync_request != NULL &&
3689 pers->sync_request == NULL) {
3690 /* need to remove the md_redundancy_group */
3691 if (mddev->to_remove == NULL)
3692 mddev->to_remove = &md_redundancy_group;
3695 module_put(oldpers->owner);
3697 rdev_for_each(rdev, mddev) {
3698 if (rdev->raid_disk < 0)
3700 if (rdev->new_raid_disk >= mddev->raid_disks)
3701 rdev->new_raid_disk = -1;
3702 if (rdev->new_raid_disk == rdev->raid_disk)
3704 sysfs_unlink_rdev(mddev, rdev);
3706 rdev_for_each(rdev, mddev) {
3707 if (rdev->raid_disk < 0)
3709 if (rdev->new_raid_disk == rdev->raid_disk)
3711 rdev->raid_disk = rdev->new_raid_disk;
3712 if (rdev->raid_disk < 0)
3713 clear_bit(In_sync, &rdev->flags);
3715 if (sysfs_link_rdev(mddev, rdev))
3716 pr_warn("md: cannot register rd%d for %s after level change\n",
3717 rdev->raid_disk, mdname(mddev));
3721 if (pers->sync_request == NULL) {
3722 /* this is now an array without redundancy, so
3723 * it must always be in_sync
3726 del_timer_sync(&mddev->safemode_timer);
3728 blk_set_stacking_limits(&mddev->queue->limits);
3730 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3731 mddev_resume(mddev);
3733 md_update_sb(mddev, 1);
3734 sysfs_notify(&mddev->kobj, NULL, "level");
3735 md_new_event(mddev);
3738 mddev_unlock(mddev);
3742 static struct md_sysfs_entry md_level =
3743 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3746 layout_show(struct mddev *mddev, char *page)
3748 /* just a number, not meaningful for all levels */
3749 if (mddev->reshape_position != MaxSector &&
3750 mddev->layout != mddev->new_layout)
3751 return sprintf(page, "%d (%d)\n",
3752 mddev->new_layout, mddev->layout);
3753 return sprintf(page, "%d\n", mddev->layout);
3757 layout_store(struct mddev *mddev, const char *buf, size_t len)
3762 err = kstrtouint(buf, 10, &n);
3765 err = mddev_lock(mddev);
3770 if (mddev->pers->check_reshape == NULL)
3775 mddev->new_layout = n;
3776 err = mddev->pers->check_reshape(mddev);
3778 mddev->new_layout = mddev->layout;
3781 mddev->new_layout = n;
3782 if (mddev->reshape_position == MaxSector)
3785 mddev_unlock(mddev);
3788 static struct md_sysfs_entry md_layout =
3789 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3792 raid_disks_show(struct mddev *mddev, char *page)
3794 if (mddev->raid_disks == 0)
3796 if (mddev->reshape_position != MaxSector &&
3797 mddev->delta_disks != 0)
3798 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3799 mddev->raid_disks - mddev->delta_disks);
3800 return sprintf(page, "%d\n", mddev->raid_disks);
3803 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3806 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3811 err = kstrtouint(buf, 10, &n);
3815 err = mddev_lock(mddev);
3819 err = update_raid_disks(mddev, n);
3820 else if (mddev->reshape_position != MaxSector) {
3821 struct md_rdev *rdev;
3822 int olddisks = mddev->raid_disks - mddev->delta_disks;
3825 rdev_for_each(rdev, mddev) {
3827 rdev->data_offset < rdev->new_data_offset)
3830 rdev->data_offset > rdev->new_data_offset)
3834 mddev->delta_disks = n - olddisks;
3835 mddev->raid_disks = n;
3836 mddev->reshape_backwards = (mddev->delta_disks < 0);
3838 mddev->raid_disks = n;
3840 mddev_unlock(mddev);
3841 return err ? err : len;
3843 static struct md_sysfs_entry md_raid_disks =
3844 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3847 chunk_size_show(struct mddev *mddev, char *page)
3849 if (mddev->reshape_position != MaxSector &&
3850 mddev->chunk_sectors != mddev->new_chunk_sectors)
3851 return sprintf(page, "%d (%d)\n",
3852 mddev->new_chunk_sectors << 9,
3853 mddev->chunk_sectors << 9);
3854 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3858 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3863 err = kstrtoul(buf, 10, &n);
3867 err = mddev_lock(mddev);
3871 if (mddev->pers->check_reshape == NULL)
3876 mddev->new_chunk_sectors = n >> 9;
3877 err = mddev->pers->check_reshape(mddev);
3879 mddev->new_chunk_sectors = mddev->chunk_sectors;
3882 mddev->new_chunk_sectors = n >> 9;
3883 if (mddev->reshape_position == MaxSector)
3884 mddev->chunk_sectors = n >> 9;
3886 mddev_unlock(mddev);
3889 static struct md_sysfs_entry md_chunk_size =
3890 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3893 resync_start_show(struct mddev *mddev, char *page)
3895 if (mddev->recovery_cp == MaxSector)
3896 return sprintf(page, "none\n");
3897 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3901 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3903 unsigned long long n;
3906 if (cmd_match(buf, "none"))
3909 err = kstrtoull(buf, 10, &n);
3912 if (n != (sector_t)n)
3916 err = mddev_lock(mddev);
3919 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3923 mddev->recovery_cp = n;
3925 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
3927 mddev_unlock(mddev);
3930 static struct md_sysfs_entry md_resync_start =
3931 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3932 resync_start_show, resync_start_store);
3935 * The array state can be:
3938 * No devices, no size, no level
3939 * Equivalent to STOP_ARRAY ioctl
3941 * May have some settings, but array is not active
3942 * all IO results in error
3943 * When written, doesn't tear down array, but just stops it
3944 * suspended (not supported yet)
3945 * All IO requests will block. The array can be reconfigured.
3946 * Writing this, if accepted, will block until array is quiescent
3948 * no resync can happen. no superblocks get written.
3949 * write requests fail
3951 * like readonly, but behaves like 'clean' on a write request.
3953 * clean - no pending writes, but otherwise active.
3954 * When written to inactive array, starts without resync
3955 * If a write request arrives then
3956 * if metadata is known, mark 'dirty' and switch to 'active'.
3957 * if not known, block and switch to write-pending
3958 * If written to an active array that has pending writes, then fails.
3960 * fully active: IO and resync can be happening.
3961 * When written to inactive array, starts with resync
3964 * clean, but writes are blocked waiting for 'active' to be written.
3967 * like active, but no writes have been seen for a while (100msec).
3970 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3971 write_pending, active_idle, bad_word};
3972 static char *array_states[] = {
3973 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3974 "write-pending", "active-idle", NULL };
3976 static int match_word(const char *word, char **list)
3979 for (n=0; list[n]; n++)
3980 if (cmd_match(word, list[n]))
3986 array_state_show(struct mddev *mddev, char *page)
3988 enum array_state st = inactive;
3999 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4001 else if (mddev->in_sync)
4003 else if (mddev->safemode)
4009 if (list_empty(&mddev->disks) &&
4010 mddev->raid_disks == 0 &&
4011 mddev->dev_sectors == 0)
4016 return sprintf(page, "%s\n", array_states[st]);
4019 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4020 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4021 static int do_md_run(struct mddev *mddev);
4022 static int restart_array(struct mddev *mddev);
4025 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4028 enum array_state st = match_word(buf, array_states);
4030 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4031 /* don't take reconfig_mutex when toggling between
4034 spin_lock(&mddev->lock);
4036 restart_array(mddev);
4037 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4038 md_wakeup_thread(mddev->thread);
4039 wake_up(&mddev->sb_wait);
4041 } else /* st == clean */ {
4042 restart_array(mddev);
4043 if (atomic_read(&mddev->writes_pending) == 0) {
4044 if (mddev->in_sync == 0) {
4046 if (mddev->safemode == 1)
4047 mddev->safemode = 0;
4048 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4055 sysfs_notify_dirent_safe(mddev->sysfs_state);
4056 spin_unlock(&mddev->lock);
4059 err = mddev_lock(mddev);
4067 /* stopping an active array */
4068 err = do_md_stop(mddev, 0, NULL);
4071 /* stopping an active array */
4073 err = do_md_stop(mddev, 2, NULL);
4075 err = 0; /* already inactive */
4078 break; /* not supported yet */
4081 err = md_set_readonly(mddev, NULL);
4084 set_disk_ro(mddev->gendisk, 1);
4085 err = do_md_run(mddev);
4091 err = md_set_readonly(mddev, NULL);
4092 else if (mddev->ro == 1)
4093 err = restart_array(mddev);
4096 set_disk_ro(mddev->gendisk, 0);
4100 err = do_md_run(mddev);
4105 err = restart_array(mddev);
4108 spin_lock(&mddev->lock);
4109 if (atomic_read(&mddev->writes_pending) == 0) {
4110 if (mddev->in_sync == 0) {
4112 if (mddev->safemode == 1)
4113 mddev->safemode = 0;
4114 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4119 spin_unlock(&mddev->lock);
4125 err = restart_array(mddev);
4128 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4129 wake_up(&mddev->sb_wait);
4133 set_disk_ro(mddev->gendisk, 0);
4134 err = do_md_run(mddev);
4139 /* these cannot be set */
4144 if (mddev->hold_active == UNTIL_IOCTL)
4145 mddev->hold_active = 0;
4146 sysfs_notify_dirent_safe(mddev->sysfs_state);
4148 mddev_unlock(mddev);
4151 static struct md_sysfs_entry md_array_state =
4152 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4155 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4156 return sprintf(page, "%d\n",
4157 atomic_read(&mddev->max_corr_read_errors));
4161 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4166 rv = kstrtouint(buf, 10, &n);
4169 atomic_set(&mddev->max_corr_read_errors, n);
4173 static struct md_sysfs_entry max_corr_read_errors =
4174 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4175 max_corrected_read_errors_store);
4178 null_show(struct mddev *mddev, char *page)
4184 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4186 /* buf must be %d:%d\n? giving major and minor numbers */
4187 /* The new device is added to the array.
4188 * If the array has a persistent superblock, we read the
4189 * superblock to initialise info and check validity.
4190 * Otherwise, only checking done is that in bind_rdev_to_array,
4191 * which mainly checks size.
4194 int major = simple_strtoul(buf, &e, 10);
4197 struct md_rdev *rdev;
4200 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4202 minor = simple_strtoul(e+1, &e, 10);
4203 if (*e && *e != '\n')
4205 dev = MKDEV(major, minor);
4206 if (major != MAJOR(dev) ||
4207 minor != MINOR(dev))
4210 flush_workqueue(md_misc_wq);
4212 err = mddev_lock(mddev);
4215 if (mddev->persistent) {
4216 rdev = md_import_device(dev, mddev->major_version,
4217 mddev->minor_version);
4218 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4219 struct md_rdev *rdev0
4220 = list_entry(mddev->disks.next,
4221 struct md_rdev, same_set);
4222 err = super_types[mddev->major_version]
4223 .load_super(rdev, rdev0, mddev->minor_version);
4227 } else if (mddev->external)
4228 rdev = md_import_device(dev, -2, -1);
4230 rdev = md_import_device(dev, -1, -1);
4233 mddev_unlock(mddev);
4234 return PTR_ERR(rdev);
4236 err = bind_rdev_to_array(rdev, mddev);
4240 mddev_unlock(mddev);
4241 return err ? err : len;
4244 static struct md_sysfs_entry md_new_device =
4245 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4248 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4251 unsigned long chunk, end_chunk;
4254 err = mddev_lock(mddev);
4259 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4261 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4262 if (buf == end) break;
4263 if (*end == '-') { /* range */
4265 end_chunk = simple_strtoul(buf, &end, 0);
4266 if (buf == end) break;
4268 if (*end && !isspace(*end)) break;
4269 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4270 buf = skip_spaces(end);
4272 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4274 mddev_unlock(mddev);
4278 static struct md_sysfs_entry md_bitmap =
4279 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4282 size_show(struct mddev *mddev, char *page)
4284 return sprintf(page, "%llu\n",
4285 (unsigned long long)mddev->dev_sectors / 2);
4288 static int update_size(struct mddev *mddev, sector_t num_sectors);
4291 size_store(struct mddev *mddev, const char *buf, size_t len)
4293 /* If array is inactive, we can reduce the component size, but
4294 * not increase it (except from 0).
4295 * If array is active, we can try an on-line resize
4298 int err = strict_blocks_to_sectors(buf, §ors);
4302 err = mddev_lock(mddev);
4306 err = update_size(mddev, sectors);
4308 md_update_sb(mddev, 1);
4310 if (mddev->dev_sectors == 0 ||
4311 mddev->dev_sectors > sectors)
4312 mddev->dev_sectors = sectors;
4316 mddev_unlock(mddev);
4317 return err ? err : len;
4320 static struct md_sysfs_entry md_size =
4321 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4323 /* Metadata version.
4325 * 'none' for arrays with no metadata (good luck...)
4326 * 'external' for arrays with externally managed metadata,
4327 * or N.M for internally known formats
4330 metadata_show(struct mddev *mddev, char *page)
4332 if (mddev->persistent)
4333 return sprintf(page, "%d.%d\n",
4334 mddev->major_version, mddev->minor_version);
4335 else if (mddev->external)
4336 return sprintf(page, "external:%s\n", mddev->metadata_type);
4338 return sprintf(page, "none\n");
4342 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4347 /* Changing the details of 'external' metadata is
4348 * always permitted. Otherwise there must be
4349 * no devices attached to the array.
4352 err = mddev_lock(mddev);
4356 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4358 else if (!list_empty(&mddev->disks))
4362 if (cmd_match(buf, "none")) {
4363 mddev->persistent = 0;
4364 mddev->external = 0;
4365 mddev->major_version = 0;
4366 mddev->minor_version = 90;
4369 if (strncmp(buf, "external:", 9) == 0) {
4370 size_t namelen = len-9;
4371 if (namelen >= sizeof(mddev->metadata_type))
4372 namelen = sizeof(mddev->metadata_type)-1;
4373 strncpy(mddev->metadata_type, buf+9, namelen);
4374 mddev->metadata_type[namelen] = 0;
4375 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4376 mddev->metadata_type[--namelen] = 0;
4377 mddev->persistent = 0;
4378 mddev->external = 1;
4379 mddev->major_version = 0;
4380 mddev->minor_version = 90;
4383 major = simple_strtoul(buf, &e, 10);
4385 if (e==buf || *e != '.')
4388 minor = simple_strtoul(buf, &e, 10);
4389 if (e==buf || (*e && *e != '\n') )
4392 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4394 mddev->major_version = major;
4395 mddev->minor_version = minor;
4396 mddev->persistent = 1;
4397 mddev->external = 0;
4400 mddev_unlock(mddev);
4404 static struct md_sysfs_entry md_metadata =
4405 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4408 action_show(struct mddev *mddev, char *page)
4410 char *type = "idle";
4411 unsigned long recovery = mddev->recovery;
4412 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4414 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4415 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4416 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4418 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4419 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4421 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4425 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4427 else if (mddev->reshape_position != MaxSector)
4430 return sprintf(page, "%s\n", type);
4434 action_store(struct mddev *mddev, const char *page, size_t len)
4436 if (!mddev->pers || !mddev->pers->sync_request)
4440 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4441 if (cmd_match(page, "frozen"))
4442 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4444 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4445 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4446 mddev_lock(mddev) == 0) {
4447 flush_workqueue(md_misc_wq);
4448 if (mddev->sync_thread) {
4449 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4450 md_reap_sync_thread(mddev);
4452 mddev_unlock(mddev);
4454 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4456 else if (cmd_match(page, "resync"))
4457 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4458 else if (cmd_match(page, "recover")) {
4459 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4460 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4461 } else if (cmd_match(page, "reshape")) {
4463 if (mddev->pers->start_reshape == NULL)
4465 err = mddev_lock(mddev);
4467 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4470 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4471 err = mddev->pers->start_reshape(mddev);
4473 mddev_unlock(mddev);
4477 sysfs_notify(&mddev->kobj, NULL, "degraded");
4479 if (cmd_match(page, "check"))
4480 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4481 else if (!cmd_match(page, "repair"))
4483 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4484 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4485 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4487 if (mddev->ro == 2) {
4488 /* A write to sync_action is enough to justify
4489 * canceling read-auto mode
4492 md_wakeup_thread(mddev->sync_thread);
4494 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4495 md_wakeup_thread(mddev->thread);
4496 sysfs_notify_dirent_safe(mddev->sysfs_action);
4500 static struct md_sysfs_entry md_scan_mode =
4501 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4504 last_sync_action_show(struct mddev *mddev, char *page)
4506 return sprintf(page, "%s\n", mddev->last_sync_action);
4509 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4512 mismatch_cnt_show(struct mddev *mddev, char *page)
4514 return sprintf(page, "%llu\n",
4515 (unsigned long long)
4516 atomic64_read(&mddev->resync_mismatches));
4519 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4522 sync_min_show(struct mddev *mddev, char *page)
4524 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4525 mddev->sync_speed_min ? "local": "system");
4529 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4534 if (strncmp(buf, "system", 6)==0) {
4537 rv = kstrtouint(buf, 10, &min);
4543 mddev->sync_speed_min = min;
4547 static struct md_sysfs_entry md_sync_min =
4548 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4551 sync_max_show(struct mddev *mddev, char *page)
4553 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4554 mddev->sync_speed_max ? "local": "system");
4558 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4563 if (strncmp(buf, "system", 6)==0) {
4566 rv = kstrtouint(buf, 10, &max);
4572 mddev->sync_speed_max = max;
4576 static struct md_sysfs_entry md_sync_max =
4577 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4580 degraded_show(struct mddev *mddev, char *page)
4582 return sprintf(page, "%d\n", mddev->degraded);
4584 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4587 sync_force_parallel_show(struct mddev *mddev, char *page)
4589 return sprintf(page, "%d\n", mddev->parallel_resync);
4593 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4597 if (kstrtol(buf, 10, &n))
4600 if (n != 0 && n != 1)
4603 mddev->parallel_resync = n;
4605 if (mddev->sync_thread)
4606 wake_up(&resync_wait);
4611 /* force parallel resync, even with shared block devices */
4612 static struct md_sysfs_entry md_sync_force_parallel =
4613 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4614 sync_force_parallel_show, sync_force_parallel_store);
4617 sync_speed_show(struct mddev *mddev, char *page)
4619 unsigned long resync, dt, db;
4620 if (mddev->curr_resync == 0)
4621 return sprintf(page, "none\n");
4622 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4623 dt = (jiffies - mddev->resync_mark) / HZ;
4625 db = resync - mddev->resync_mark_cnt;
4626 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4629 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4632 sync_completed_show(struct mddev *mddev, char *page)
4634 unsigned long long max_sectors, resync;
4636 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4637 return sprintf(page, "none\n");
4639 if (mddev->curr_resync == 1 ||
4640 mddev->curr_resync == 2)
4641 return sprintf(page, "delayed\n");
4643 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4644 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4645 max_sectors = mddev->resync_max_sectors;
4647 max_sectors = mddev->dev_sectors;
4649 resync = mddev->curr_resync_completed;
4650 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4653 static struct md_sysfs_entry md_sync_completed =
4654 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4657 min_sync_show(struct mddev *mddev, char *page)
4659 return sprintf(page, "%llu\n",
4660 (unsigned long long)mddev->resync_min);
4663 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4665 unsigned long long min;
4668 if (kstrtoull(buf, 10, &min))
4671 spin_lock(&mddev->lock);
4673 if (min > mddev->resync_max)
4677 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4680 /* Round down to multiple of 4K for safety */
4681 mddev->resync_min = round_down(min, 8);
4685 spin_unlock(&mddev->lock);
4689 static struct md_sysfs_entry md_min_sync =
4690 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4693 max_sync_show(struct mddev *mddev, char *page)
4695 if (mddev->resync_max == MaxSector)
4696 return sprintf(page, "max\n");
4698 return sprintf(page, "%llu\n",
4699 (unsigned long long)mddev->resync_max);
4702 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4705 spin_lock(&mddev->lock);
4706 if (strncmp(buf, "max", 3) == 0)
4707 mddev->resync_max = MaxSector;
4709 unsigned long long max;
4713 if (kstrtoull(buf, 10, &max))
4715 if (max < mddev->resync_min)
4719 if (max < mddev->resync_max &&
4721 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4724 /* Must be a multiple of chunk_size */
4725 chunk = mddev->chunk_sectors;
4727 sector_t temp = max;
4730 if (sector_div(temp, chunk))
4733 mddev->resync_max = max;
4735 wake_up(&mddev->recovery_wait);
4738 spin_unlock(&mddev->lock);
4742 static struct md_sysfs_entry md_max_sync =
4743 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4746 suspend_lo_show(struct mddev *mddev, char *page)
4748 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4752 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4754 unsigned long long old, new;
4757 err = kstrtoull(buf, 10, &new);
4760 if (new != (sector_t)new)
4763 err = mddev_lock(mddev);
4767 if (mddev->pers == NULL ||
4768 mddev->pers->quiesce == NULL)
4770 old = mddev->suspend_lo;
4771 mddev->suspend_lo = new;
4773 /* Shrinking suspended region */
4774 mddev->pers->quiesce(mddev, 2);
4776 /* Expanding suspended region - need to wait */
4777 mddev->pers->quiesce(mddev, 1);
4778 mddev->pers->quiesce(mddev, 0);
4782 mddev_unlock(mddev);
4785 static struct md_sysfs_entry md_suspend_lo =
4786 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4789 suspend_hi_show(struct mddev *mddev, char *page)
4791 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4795 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4797 unsigned long long old, new;
4800 err = kstrtoull(buf, 10, &new);
4803 if (new != (sector_t)new)
4806 err = mddev_lock(mddev);
4810 if (mddev->pers == NULL ||
4811 mddev->pers->quiesce == NULL)
4813 old = mddev->suspend_hi;
4814 mddev->suspend_hi = new;
4816 /* Shrinking suspended region */
4817 mddev->pers->quiesce(mddev, 2);
4819 /* Expanding suspended region - need to wait */
4820 mddev->pers->quiesce(mddev, 1);
4821 mddev->pers->quiesce(mddev, 0);
4825 mddev_unlock(mddev);
4828 static struct md_sysfs_entry md_suspend_hi =
4829 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4832 reshape_position_show(struct mddev *mddev, char *page)
4834 if (mddev->reshape_position != MaxSector)
4835 return sprintf(page, "%llu\n",
4836 (unsigned long long)mddev->reshape_position);
4837 strcpy(page, "none\n");
4842 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4844 struct md_rdev *rdev;
4845 unsigned long long new;
4848 err = kstrtoull(buf, 10, &new);
4851 if (new != (sector_t)new)
4853 err = mddev_lock(mddev);
4859 mddev->reshape_position = new;
4860 mddev->delta_disks = 0;
4861 mddev->reshape_backwards = 0;
4862 mddev->new_level = mddev->level;
4863 mddev->new_layout = mddev->layout;
4864 mddev->new_chunk_sectors = mddev->chunk_sectors;
4865 rdev_for_each(rdev, mddev)
4866 rdev->new_data_offset = rdev->data_offset;
4869 mddev_unlock(mddev);
4873 static struct md_sysfs_entry md_reshape_position =
4874 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4875 reshape_position_store);
4878 reshape_direction_show(struct mddev *mddev, char *page)
4880 return sprintf(page, "%s\n",
4881 mddev->reshape_backwards ? "backwards" : "forwards");
4885 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4890 if (cmd_match(buf, "forwards"))
4892 else if (cmd_match(buf, "backwards"))
4896 if (mddev->reshape_backwards == backwards)
4899 err = mddev_lock(mddev);
4902 /* check if we are allowed to change */
4903 if (mddev->delta_disks)
4905 else if (mddev->persistent &&
4906 mddev->major_version == 0)
4909 mddev->reshape_backwards = backwards;
4910 mddev_unlock(mddev);
4914 static struct md_sysfs_entry md_reshape_direction =
4915 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4916 reshape_direction_store);
4919 array_size_show(struct mddev *mddev, char *page)
4921 if (mddev->external_size)
4922 return sprintf(page, "%llu\n",
4923 (unsigned long long)mddev->array_sectors/2);
4925 return sprintf(page, "default\n");
4929 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4934 err = mddev_lock(mddev);
4938 /* cluster raid doesn't support change array_sectors */
4939 if (mddev_is_clustered(mddev))
4942 if (strncmp(buf, "default", 7) == 0) {
4944 sectors = mddev->pers->size(mddev, 0, 0);
4946 sectors = mddev->array_sectors;
4948 mddev->external_size = 0;
4950 if (strict_blocks_to_sectors(buf, §ors) < 0)
4952 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4955 mddev->external_size = 1;
4959 mddev->array_sectors = sectors;
4961 set_capacity(mddev->gendisk, mddev->array_sectors);
4962 revalidate_disk(mddev->gendisk);
4965 mddev_unlock(mddev);
4969 static struct md_sysfs_entry md_array_size =
4970 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4974 consistency_policy_show(struct mddev *mddev, char *page)
4978 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
4979 ret = sprintf(page, "journal\n");
4980 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
4981 ret = sprintf(page, "ppl\n");
4982 } else if (mddev->bitmap) {
4983 ret = sprintf(page, "bitmap\n");
4984 } else if (mddev->pers) {
4985 if (mddev->pers->sync_request)
4986 ret = sprintf(page, "resync\n");
4988 ret = sprintf(page, "none\n");
4990 ret = sprintf(page, "unknown\n");
4997 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5002 if (mddev->pers->change_consistency_policy)
5003 err = mddev->pers->change_consistency_policy(mddev, buf);
5006 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5007 set_bit(MD_HAS_PPL, &mddev->flags);
5012 return err ? err : len;
5015 static struct md_sysfs_entry md_consistency_policy =
5016 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5017 consistency_policy_store);
5019 static struct attribute *md_default_attrs[] = {
5022 &md_raid_disks.attr,
5023 &md_chunk_size.attr,
5025 &md_resync_start.attr,
5027 &md_new_device.attr,
5028 &md_safe_delay.attr,
5029 &md_array_state.attr,
5030 &md_reshape_position.attr,
5031 &md_reshape_direction.attr,
5032 &md_array_size.attr,
5033 &max_corr_read_errors.attr,
5034 &md_consistency_policy.attr,
5038 static struct attribute *md_redundancy_attrs[] = {
5040 &md_last_scan_mode.attr,
5041 &md_mismatches.attr,
5044 &md_sync_speed.attr,
5045 &md_sync_force_parallel.attr,
5046 &md_sync_completed.attr,
5049 &md_suspend_lo.attr,
5050 &md_suspend_hi.attr,
5055 static struct attribute_group md_redundancy_group = {
5057 .attrs = md_redundancy_attrs,
5061 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5063 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5064 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5069 spin_lock(&all_mddevs_lock);
5070 if (list_empty(&mddev->all_mddevs)) {
5071 spin_unlock(&all_mddevs_lock);
5075 spin_unlock(&all_mddevs_lock);
5077 rv = entry->show(mddev, page);
5083 md_attr_store(struct kobject *kobj, struct attribute *attr,
5084 const char *page, size_t length)
5086 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5087 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5092 if (!capable(CAP_SYS_ADMIN))
5094 spin_lock(&all_mddevs_lock);
5095 if (list_empty(&mddev->all_mddevs)) {
5096 spin_unlock(&all_mddevs_lock);
5100 spin_unlock(&all_mddevs_lock);
5101 rv = entry->store(mddev, page, length);
5106 static void md_free(struct kobject *ko)
5108 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5110 if (mddev->sysfs_state)
5111 sysfs_put(mddev->sysfs_state);
5114 blk_cleanup_queue(mddev->queue);
5115 if (mddev->gendisk) {
5116 del_gendisk(mddev->gendisk);
5117 put_disk(mddev->gendisk);
5123 static const struct sysfs_ops md_sysfs_ops = {
5124 .show = md_attr_show,
5125 .store = md_attr_store,
5127 static struct kobj_type md_ktype = {
5129 .sysfs_ops = &md_sysfs_ops,
5130 .default_attrs = md_default_attrs,
5135 static void mddev_delayed_delete(struct work_struct *ws)
5137 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5139 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5140 kobject_del(&mddev->kobj);
5141 kobject_put(&mddev->kobj);
5144 static int md_alloc(dev_t dev, char *name)
5146 static DEFINE_MUTEX(disks_mutex);
5147 struct mddev *mddev = mddev_find(dev);
5148 struct gendisk *disk;
5157 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5158 shift = partitioned ? MdpMinorShift : 0;
5159 unit = MINOR(mddev->unit) >> shift;
5161 /* wait for any previous instance of this device to be
5162 * completely removed (mddev_delayed_delete).
5164 flush_workqueue(md_misc_wq);
5166 mutex_lock(&disks_mutex);
5172 /* Need to ensure that 'name' is not a duplicate.
5174 struct mddev *mddev2;
5175 spin_lock(&all_mddevs_lock);
5177 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5178 if (mddev2->gendisk &&
5179 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5180 spin_unlock(&all_mddevs_lock);
5183 spin_unlock(&all_mddevs_lock);
5187 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5190 mddev->queue->queuedata = mddev;
5192 blk_queue_make_request(mddev->queue, md_make_request);
5193 blk_set_stacking_limits(&mddev->queue->limits);
5195 disk = alloc_disk(1 << shift);
5197 blk_cleanup_queue(mddev->queue);
5198 mddev->queue = NULL;
5201 disk->major = MAJOR(mddev->unit);
5202 disk->first_minor = unit << shift;
5204 strcpy(disk->disk_name, name);
5205 else if (partitioned)
5206 sprintf(disk->disk_name, "md_d%d", unit);
5208 sprintf(disk->disk_name, "md%d", unit);
5209 disk->fops = &md_fops;
5210 disk->private_data = mddev;
5211 disk->queue = mddev->queue;
5212 blk_queue_write_cache(mddev->queue, true, true);
5213 /* Allow extended partitions. This makes the
5214 * 'mdp' device redundant, but we can't really
5217 disk->flags |= GENHD_FL_EXT_DEVT;
5218 mddev->gendisk = disk;
5219 /* As soon as we call add_disk(), another thread could get
5220 * through to md_open, so make sure it doesn't get too far
5222 mutex_lock(&mddev->open_mutex);
5225 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5226 &disk_to_dev(disk)->kobj, "%s", "md");
5228 /* This isn't possible, but as kobject_init_and_add is marked
5229 * __must_check, we must do something with the result
5231 pr_debug("md: cannot register %s/md - name in use\n",
5235 if (mddev->kobj.sd &&
5236 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5237 pr_debug("pointless warning\n");
5238 mutex_unlock(&mddev->open_mutex);
5240 mutex_unlock(&disks_mutex);
5241 if (!error && mddev->kobj.sd) {
5242 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5243 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5249 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5251 md_alloc(dev, NULL);
5255 static int add_named_array(const char *val, struct kernel_param *kp)
5257 /* val must be "md_*" where * is not all digits.
5258 * We allocate an array with a large free minor number, and
5259 * set the name to val. val must not already be an active name.
5261 int len = strlen(val);
5262 char buf[DISK_NAME_LEN];
5264 while (len && val[len-1] == '\n')
5266 if (len >= DISK_NAME_LEN)
5268 strlcpy(buf, val, len+1);
5269 if (strncmp(buf, "md_", 3) != 0)
5271 return md_alloc(0, buf);
5274 static void md_safemode_timeout(unsigned long data)
5276 struct mddev *mddev = (struct mddev *) data;
5278 if (!atomic_read(&mddev->writes_pending)) {
5279 mddev->safemode = 1;
5280 if (mddev->external)
5281 sysfs_notify_dirent_safe(mddev->sysfs_state);
5283 md_wakeup_thread(mddev->thread);
5286 static int start_dirty_degraded;
5288 int md_run(struct mddev *mddev)
5291 struct md_rdev *rdev;
5292 struct md_personality *pers;
5294 if (list_empty(&mddev->disks))
5295 /* cannot run an array with no devices.. */
5300 /* Cannot run until previous stop completes properly */
5301 if (mddev->sysfs_active)
5305 * Analyze all RAID superblock(s)
5307 if (!mddev->raid_disks) {
5308 if (!mddev->persistent)
5313 if (mddev->level != LEVEL_NONE)
5314 request_module("md-level-%d", mddev->level);
5315 else if (mddev->clevel[0])
5316 request_module("md-%s", mddev->clevel);
5319 * Drop all container device buffers, from now on
5320 * the only valid external interface is through the md
5323 rdev_for_each(rdev, mddev) {
5324 if (test_bit(Faulty, &rdev->flags))
5326 sync_blockdev(rdev->bdev);
5327 invalidate_bdev(rdev->bdev);
5329 /* perform some consistency tests on the device.
5330 * We don't want the data to overlap the metadata,
5331 * Internal Bitmap issues have been handled elsewhere.
5333 if (rdev->meta_bdev) {
5334 /* Nothing to check */;
5335 } else if (rdev->data_offset < rdev->sb_start) {
5336 if (mddev->dev_sectors &&
5337 rdev->data_offset + mddev->dev_sectors
5339 pr_warn("md: %s: data overlaps metadata\n",
5344 if (rdev->sb_start + rdev->sb_size/512
5345 > rdev->data_offset) {
5346 pr_warn("md: %s: metadata overlaps data\n",
5351 sysfs_notify_dirent_safe(rdev->sysfs_state);
5354 if (mddev->bio_set == NULL) {
5355 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5356 if (!mddev->bio_set)
5360 spin_lock(&pers_lock);
5361 pers = find_pers(mddev->level, mddev->clevel);
5362 if (!pers || !try_module_get(pers->owner)) {
5363 spin_unlock(&pers_lock);
5364 if (mddev->level != LEVEL_NONE)
5365 pr_warn("md: personality for level %d is not loaded!\n",
5368 pr_warn("md: personality for level %s is not loaded!\n",
5372 spin_unlock(&pers_lock);
5373 if (mddev->level != pers->level) {
5374 mddev->level = pers->level;
5375 mddev->new_level = pers->level;
5377 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5379 if (mddev->reshape_position != MaxSector &&
5380 pers->start_reshape == NULL) {
5381 /* This personality cannot handle reshaping... */
5382 module_put(pers->owner);
5386 if (pers->sync_request) {
5387 /* Warn if this is a potentially silly
5390 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5391 struct md_rdev *rdev2;
5394 rdev_for_each(rdev, mddev)
5395 rdev_for_each(rdev2, mddev) {
5397 rdev->bdev->bd_contains ==
5398 rdev2->bdev->bd_contains) {
5399 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5401 bdevname(rdev->bdev,b),
5402 bdevname(rdev2->bdev,b2));
5408 pr_warn("True protection against single-disk failure might be compromised.\n");
5411 mddev->recovery = 0;
5412 /* may be over-ridden by personality */
5413 mddev->resync_max_sectors = mddev->dev_sectors;
5415 mddev->ok_start_degraded = start_dirty_degraded;
5417 if (start_readonly && mddev->ro == 0)
5418 mddev->ro = 2; /* read-only, but switch on first write */
5421 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
5422 * up mddev->thread. It is important to initialize critical
5423 * resources for mddev->thread BEFORE calling pers->run().
5425 err = pers->run(mddev);
5427 pr_warn("md: pers->run() failed ...\n");
5428 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5429 WARN_ONCE(!mddev->external_size,
5430 "%s: default size too small, but 'external_size' not in effect?\n",
5432 pr_warn("md: invalid array_size %llu > default size %llu\n",
5433 (unsigned long long)mddev->array_sectors / 2,
5434 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5437 if (err == 0 && pers->sync_request &&
5438 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5439 struct bitmap *bitmap;
5441 bitmap = bitmap_create(mddev, -1);
5442 if (IS_ERR(bitmap)) {
5443 err = PTR_ERR(bitmap);
5444 pr_warn("%s: failed to create bitmap (%d)\n",
5445 mdname(mddev), err);
5447 mddev->bitmap = bitmap;
5451 mddev_detach(mddev);
5453 pers->free(mddev, mddev->private);
5454 mddev->private = NULL;
5455 module_put(pers->owner);
5456 bitmap_destroy(mddev);
5462 rdev_for_each(rdev, mddev) {
5463 if (rdev->raid_disk >= 0 &&
5464 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5469 if (mddev->degraded)
5472 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5474 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5475 mddev->queue->backing_dev_info->congested_data = mddev;
5476 mddev->queue->backing_dev_info->congested_fn = md_congested;
5478 if (pers->sync_request) {
5479 if (mddev->kobj.sd &&
5480 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5481 pr_warn("md: cannot register extra attributes for %s\n",
5483 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5484 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5487 atomic_set(&mddev->writes_pending,0);
5488 atomic_set(&mddev->max_corr_read_errors,
5489 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5490 mddev->safemode = 0;
5491 if (mddev_is_clustered(mddev))
5492 mddev->safemode_delay = 0;
5494 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5497 spin_lock(&mddev->lock);
5499 spin_unlock(&mddev->lock);
5500 rdev_for_each(rdev, mddev)
5501 if (rdev->raid_disk >= 0)
5502 if (sysfs_link_rdev(mddev, rdev))
5503 /* failure here is OK */;
5505 if (mddev->degraded && !mddev->ro)
5506 /* This ensures that recovering status is reported immediately
5507 * via sysfs - until a lack of spares is confirmed.
5509 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5510 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5512 if (mddev->sb_flags)
5513 md_update_sb(mddev, 0);
5515 md_new_event(mddev);
5516 sysfs_notify_dirent_safe(mddev->sysfs_state);
5517 sysfs_notify_dirent_safe(mddev->sysfs_action);
5518 sysfs_notify(&mddev->kobj, NULL, "degraded");
5521 EXPORT_SYMBOL_GPL(md_run);
5523 static int do_md_run(struct mddev *mddev)
5527 err = md_run(mddev);
5530 err = bitmap_load(mddev);
5532 bitmap_destroy(mddev);
5536 if (mddev_is_clustered(mddev))
5537 md_allow_write(mddev);
5539 md_wakeup_thread(mddev->thread);
5540 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5542 set_capacity(mddev->gendisk, mddev->array_sectors);
5543 revalidate_disk(mddev->gendisk);
5545 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5550 static int restart_array(struct mddev *mddev)
5552 struct gendisk *disk = mddev->gendisk;
5554 /* Complain if it has no devices */
5555 if (list_empty(&mddev->disks))
5561 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5562 struct md_rdev *rdev;
5563 bool has_journal = false;
5566 rdev_for_each_rcu(rdev, mddev) {
5567 if (test_bit(Journal, &rdev->flags) &&
5568 !test_bit(Faulty, &rdev->flags)) {
5575 /* Don't restart rw with journal missing/faulty */
5580 mddev->safemode = 0;
5582 set_disk_ro(disk, 0);
5583 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5584 /* Kick recovery or resync if necessary */
5585 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5586 md_wakeup_thread(mddev->thread);
5587 md_wakeup_thread(mddev->sync_thread);
5588 sysfs_notify_dirent_safe(mddev->sysfs_state);
5592 static void md_clean(struct mddev *mddev)
5594 mddev->array_sectors = 0;
5595 mddev->external_size = 0;
5596 mddev->dev_sectors = 0;
5597 mddev->raid_disks = 0;
5598 mddev->recovery_cp = 0;
5599 mddev->resync_min = 0;
5600 mddev->resync_max = MaxSector;
5601 mddev->reshape_position = MaxSector;
5602 mddev->external = 0;
5603 mddev->persistent = 0;
5604 mddev->level = LEVEL_NONE;
5605 mddev->clevel[0] = 0;
5607 mddev->sb_flags = 0;
5609 mddev->metadata_type[0] = 0;
5610 mddev->chunk_sectors = 0;
5611 mddev->ctime = mddev->utime = 0;
5613 mddev->max_disks = 0;
5615 mddev->can_decrease_events = 0;
5616 mddev->delta_disks = 0;
5617 mddev->reshape_backwards = 0;
5618 mddev->new_level = LEVEL_NONE;
5619 mddev->new_layout = 0;
5620 mddev->new_chunk_sectors = 0;
5621 mddev->curr_resync = 0;
5622 atomic64_set(&mddev->resync_mismatches, 0);
5623 mddev->suspend_lo = mddev->suspend_hi = 0;
5624 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5625 mddev->recovery = 0;
5628 mddev->degraded = 0;
5629 mddev->safemode = 0;
5630 mddev->private = NULL;
5631 mddev->cluster_info = NULL;
5632 mddev->bitmap_info.offset = 0;
5633 mddev->bitmap_info.default_offset = 0;
5634 mddev->bitmap_info.default_space = 0;
5635 mddev->bitmap_info.chunksize = 0;
5636 mddev->bitmap_info.daemon_sleep = 0;
5637 mddev->bitmap_info.max_write_behind = 0;
5638 mddev->bitmap_info.nodes = 0;
5641 static void __md_stop_writes(struct mddev *mddev)
5643 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5644 flush_workqueue(md_misc_wq);
5645 if (mddev->sync_thread) {
5646 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5647 md_reap_sync_thread(mddev);
5650 del_timer_sync(&mddev->safemode_timer);
5652 if (mddev->pers && mddev->pers->quiesce) {
5653 mddev->pers->quiesce(mddev, 1);
5654 mddev->pers->quiesce(mddev, 0);
5656 bitmap_flush(mddev);
5658 if (mddev->ro == 0 &&
5659 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5661 /* mark array as shutdown cleanly */
5662 if (!mddev_is_clustered(mddev))
5664 md_update_sb(mddev, 1);
5668 void md_stop_writes(struct mddev *mddev)
5670 mddev_lock_nointr(mddev);
5671 __md_stop_writes(mddev);
5672 mddev_unlock(mddev);
5674 EXPORT_SYMBOL_GPL(md_stop_writes);
5676 static void mddev_detach(struct mddev *mddev)
5678 bitmap_wait_behind_writes(mddev);
5679 if (mddev->pers && mddev->pers->quiesce) {
5680 mddev->pers->quiesce(mddev, 1);
5681 mddev->pers->quiesce(mddev, 0);
5683 md_unregister_thread(&mddev->thread);
5685 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5688 static void __md_stop(struct mddev *mddev)
5690 struct md_personality *pers = mddev->pers;
5691 bitmap_destroy(mddev);
5692 mddev_detach(mddev);
5693 /* Ensure ->event_work is done */
5694 flush_workqueue(md_misc_wq);
5695 spin_lock(&mddev->lock);
5697 spin_unlock(&mddev->lock);
5698 pers->free(mddev, mddev->private);
5699 mddev->private = NULL;
5700 if (pers->sync_request && mddev->to_remove == NULL)
5701 mddev->to_remove = &md_redundancy_group;
5702 module_put(pers->owner);
5703 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5706 void md_stop(struct mddev *mddev)
5708 /* stop the array and free an attached data structures.
5709 * This is called from dm-raid
5713 bioset_free(mddev->bio_set);
5716 EXPORT_SYMBOL_GPL(md_stop);
5718 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5723 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5725 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5726 md_wakeup_thread(mddev->thread);
5728 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5729 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5730 if (mddev->sync_thread)
5731 /* Thread might be blocked waiting for metadata update
5732 * which will now never happen */
5733 wake_up_process(mddev->sync_thread->tsk);
5735 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5737 mddev_unlock(mddev);
5738 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5740 wait_event(mddev->sb_wait,
5741 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5742 mddev_lock_nointr(mddev);
5744 mutex_lock(&mddev->open_mutex);
5745 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5746 mddev->sync_thread ||
5747 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5748 pr_warn("md: %s still in use.\n",mdname(mddev));
5750 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5751 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5752 md_wakeup_thread(mddev->thread);
5758 __md_stop_writes(mddev);
5764 set_disk_ro(mddev->gendisk, 1);
5765 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5766 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5767 md_wakeup_thread(mddev->thread);
5768 sysfs_notify_dirent_safe(mddev->sysfs_state);
5772 mutex_unlock(&mddev->open_mutex);
5777 * 0 - completely stop and dis-assemble array
5778 * 2 - stop but do not disassemble array
5780 static int do_md_stop(struct mddev *mddev, int mode,
5781 struct block_device *bdev)
5783 struct gendisk *disk = mddev->gendisk;
5784 struct md_rdev *rdev;
5787 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5789 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5790 md_wakeup_thread(mddev->thread);
5792 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5793 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5794 if (mddev->sync_thread)
5795 /* Thread might be blocked waiting for metadata update
5796 * which will now never happen */
5797 wake_up_process(mddev->sync_thread->tsk);
5799 mddev_unlock(mddev);
5800 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5801 !test_bit(MD_RECOVERY_RUNNING,
5802 &mddev->recovery)));
5803 mddev_lock_nointr(mddev);
5805 mutex_lock(&mddev->open_mutex);
5806 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5807 mddev->sysfs_active ||
5808 mddev->sync_thread ||
5809 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5810 pr_warn("md: %s still in use.\n",mdname(mddev));
5811 mutex_unlock(&mddev->open_mutex);
5813 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5814 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5815 md_wakeup_thread(mddev->thread);
5821 set_disk_ro(disk, 0);
5823 __md_stop_writes(mddev);
5825 mddev->queue->backing_dev_info->congested_fn = NULL;
5827 /* tell userspace to handle 'inactive' */
5828 sysfs_notify_dirent_safe(mddev->sysfs_state);
5830 rdev_for_each(rdev, mddev)
5831 if (rdev->raid_disk >= 0)
5832 sysfs_unlink_rdev(mddev, rdev);
5834 set_capacity(disk, 0);
5835 mutex_unlock(&mddev->open_mutex);
5837 revalidate_disk(disk);
5842 mutex_unlock(&mddev->open_mutex);
5844 * Free resources if final stop
5847 pr_info("md: %s stopped.\n", mdname(mddev));
5849 if (mddev->bitmap_info.file) {
5850 struct file *f = mddev->bitmap_info.file;
5851 spin_lock(&mddev->lock);
5852 mddev->bitmap_info.file = NULL;
5853 spin_unlock(&mddev->lock);
5856 mddev->bitmap_info.offset = 0;
5858 export_array(mddev);
5861 if (mddev->hold_active == UNTIL_STOP)
5862 mddev->hold_active = 0;
5864 md_new_event(mddev);
5865 sysfs_notify_dirent_safe(mddev->sysfs_state);
5870 static void autorun_array(struct mddev *mddev)
5872 struct md_rdev *rdev;
5875 if (list_empty(&mddev->disks))
5878 pr_info("md: running: ");
5880 rdev_for_each(rdev, mddev) {
5881 char b[BDEVNAME_SIZE];
5882 pr_cont("<%s>", bdevname(rdev->bdev,b));
5886 err = do_md_run(mddev);
5888 pr_warn("md: do_md_run() returned %d\n", err);
5889 do_md_stop(mddev, 0, NULL);
5894 * lets try to run arrays based on all disks that have arrived
5895 * until now. (those are in pending_raid_disks)
5897 * the method: pick the first pending disk, collect all disks with
5898 * the same UUID, remove all from the pending list and put them into
5899 * the 'same_array' list. Then order this list based on superblock
5900 * update time (freshest comes first), kick out 'old' disks and
5901 * compare superblocks. If everything's fine then run it.
5903 * If "unit" is allocated, then bump its reference count
5905 static void autorun_devices(int part)
5907 struct md_rdev *rdev0, *rdev, *tmp;
5908 struct mddev *mddev;
5909 char b[BDEVNAME_SIZE];
5911 pr_info("md: autorun ...\n");
5912 while (!list_empty(&pending_raid_disks)) {
5915 LIST_HEAD(candidates);
5916 rdev0 = list_entry(pending_raid_disks.next,
5917 struct md_rdev, same_set);
5919 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
5920 INIT_LIST_HEAD(&candidates);
5921 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5922 if (super_90_load(rdev, rdev0, 0) >= 0) {
5923 pr_debug("md: adding %s ...\n",
5924 bdevname(rdev->bdev,b));
5925 list_move(&rdev->same_set, &candidates);
5928 * now we have a set of devices, with all of them having
5929 * mostly sane superblocks. It's time to allocate the
5933 dev = MKDEV(mdp_major,
5934 rdev0->preferred_minor << MdpMinorShift);
5935 unit = MINOR(dev) >> MdpMinorShift;
5937 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5940 if (rdev0->preferred_minor != unit) {
5941 pr_warn("md: unit number in %s is bad: %d\n",
5942 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5946 md_probe(dev, NULL, NULL);
5947 mddev = mddev_find(dev);
5948 if (!mddev || !mddev->gendisk) {
5953 if (mddev_lock(mddev))
5954 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
5955 else if (mddev->raid_disks || mddev->major_version
5956 || !list_empty(&mddev->disks)) {
5957 pr_warn("md: %s already running, cannot run %s\n",
5958 mdname(mddev), bdevname(rdev0->bdev,b));
5959 mddev_unlock(mddev);
5961 pr_debug("md: created %s\n", mdname(mddev));
5962 mddev->persistent = 1;
5963 rdev_for_each_list(rdev, tmp, &candidates) {
5964 list_del_init(&rdev->same_set);
5965 if (bind_rdev_to_array(rdev, mddev))
5968 autorun_array(mddev);
5969 mddev_unlock(mddev);
5971 /* on success, candidates will be empty, on error
5974 rdev_for_each_list(rdev, tmp, &candidates) {
5975 list_del_init(&rdev->same_set);
5980 pr_info("md: ... autorun DONE.\n");
5982 #endif /* !MODULE */
5984 static int get_version(void __user *arg)
5988 ver.major = MD_MAJOR_VERSION;
5989 ver.minor = MD_MINOR_VERSION;
5990 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5992 if (copy_to_user(arg, &ver, sizeof(ver)))
5998 static int get_array_info(struct mddev *mddev, void __user *arg)
6000 mdu_array_info_t info;
6001 int nr,working,insync,failed,spare;
6002 struct md_rdev *rdev;
6004 nr = working = insync = failed = spare = 0;
6006 rdev_for_each_rcu(rdev, mddev) {
6008 if (test_bit(Faulty, &rdev->flags))
6012 if (test_bit(In_sync, &rdev->flags))
6014 else if (test_bit(Journal, &rdev->flags))
6015 /* TODO: add journal count to md_u.h */
6023 info.major_version = mddev->major_version;
6024 info.minor_version = mddev->minor_version;
6025 info.patch_version = MD_PATCHLEVEL_VERSION;
6026 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6027 info.level = mddev->level;
6028 info.size = mddev->dev_sectors / 2;
6029 if (info.size != mddev->dev_sectors / 2) /* overflow */
6032 info.raid_disks = mddev->raid_disks;
6033 info.md_minor = mddev->md_minor;
6034 info.not_persistent= !mddev->persistent;
6036 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6039 info.state = (1<<MD_SB_CLEAN);
6040 if (mddev->bitmap && mddev->bitmap_info.offset)
6041 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6042 if (mddev_is_clustered(mddev))
6043 info.state |= (1<<MD_SB_CLUSTERED);
6044 info.active_disks = insync;
6045 info.working_disks = working;
6046 info.failed_disks = failed;
6047 info.spare_disks = spare;
6049 info.layout = mddev->layout;
6050 info.chunk_size = mddev->chunk_sectors << 9;
6052 if (copy_to_user(arg, &info, sizeof(info)))
6058 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6060 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6064 file = kzalloc(sizeof(*file), GFP_NOIO);
6069 spin_lock(&mddev->lock);
6070 /* bitmap enabled */
6071 if (mddev->bitmap_info.file) {
6072 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6073 sizeof(file->pathname));
6077 memmove(file->pathname, ptr,
6078 sizeof(file->pathname)-(ptr-file->pathname));
6080 spin_unlock(&mddev->lock);
6083 copy_to_user(arg, file, sizeof(*file)))
6090 static int get_disk_info(struct mddev *mddev, void __user * arg)
6092 mdu_disk_info_t info;
6093 struct md_rdev *rdev;
6095 if (copy_from_user(&info, arg, sizeof(info)))
6099 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6101 info.major = MAJOR(rdev->bdev->bd_dev);
6102 info.minor = MINOR(rdev->bdev->bd_dev);
6103 info.raid_disk = rdev->raid_disk;
6105 if (test_bit(Faulty, &rdev->flags))
6106 info.state |= (1<<MD_DISK_FAULTY);
6107 else if (test_bit(In_sync, &rdev->flags)) {
6108 info.state |= (1<<MD_DISK_ACTIVE);
6109 info.state |= (1<<MD_DISK_SYNC);
6111 if (test_bit(Journal, &rdev->flags))
6112 info.state |= (1<<MD_DISK_JOURNAL);
6113 if (test_bit(WriteMostly, &rdev->flags))
6114 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6115 if (test_bit(FailFast, &rdev->flags))
6116 info.state |= (1<<MD_DISK_FAILFAST);
6118 info.major = info.minor = 0;
6119 info.raid_disk = -1;
6120 info.state = (1<<MD_DISK_REMOVED);
6124 if (copy_to_user(arg, &info, sizeof(info)))
6130 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6132 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6133 struct md_rdev *rdev;
6134 dev_t dev = MKDEV(info->major,info->minor);
6136 if (mddev_is_clustered(mddev) &&
6137 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6138 pr_warn("%s: Cannot add to clustered mddev.\n",
6143 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6146 if (!mddev->raid_disks) {
6148 /* expecting a device which has a superblock */
6149 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6151 pr_warn("md: md_import_device returned %ld\n",
6153 return PTR_ERR(rdev);
6155 if (!list_empty(&mddev->disks)) {
6156 struct md_rdev *rdev0
6157 = list_entry(mddev->disks.next,
6158 struct md_rdev, same_set);
6159 err = super_types[mddev->major_version]
6160 .load_super(rdev, rdev0, mddev->minor_version);
6162 pr_warn("md: %s has different UUID to %s\n",
6163 bdevname(rdev->bdev,b),
6164 bdevname(rdev0->bdev,b2));
6169 err = bind_rdev_to_array(rdev, mddev);
6176 * add_new_disk can be used once the array is assembled
6177 * to add "hot spares". They must already have a superblock
6182 if (!mddev->pers->hot_add_disk) {
6183 pr_warn("%s: personality does not support diskops!\n",
6187 if (mddev->persistent)
6188 rdev = md_import_device(dev, mddev->major_version,
6189 mddev->minor_version);
6191 rdev = md_import_device(dev, -1, -1);
6193 pr_warn("md: md_import_device returned %ld\n",
6195 return PTR_ERR(rdev);
6197 /* set saved_raid_disk if appropriate */
6198 if (!mddev->persistent) {
6199 if (info->state & (1<<MD_DISK_SYNC) &&
6200 info->raid_disk < mddev->raid_disks) {
6201 rdev->raid_disk = info->raid_disk;
6202 set_bit(In_sync, &rdev->flags);
6203 clear_bit(Bitmap_sync, &rdev->flags);
6205 rdev->raid_disk = -1;
6206 rdev->saved_raid_disk = rdev->raid_disk;
6208 super_types[mddev->major_version].
6209 validate_super(mddev, rdev);
6210 if ((info->state & (1<<MD_DISK_SYNC)) &&
6211 rdev->raid_disk != info->raid_disk) {
6212 /* This was a hot-add request, but events doesn't
6213 * match, so reject it.
6219 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6220 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6221 set_bit(WriteMostly, &rdev->flags);
6223 clear_bit(WriteMostly, &rdev->flags);
6224 if (info->state & (1<<MD_DISK_FAILFAST))
6225 set_bit(FailFast, &rdev->flags);
6227 clear_bit(FailFast, &rdev->flags);
6229 if (info->state & (1<<MD_DISK_JOURNAL)) {
6230 struct md_rdev *rdev2;
6231 bool has_journal = false;
6233 /* make sure no existing journal disk */
6234 rdev_for_each(rdev2, mddev) {
6235 if (test_bit(Journal, &rdev2->flags)) {
6244 set_bit(Journal, &rdev->flags);
6247 * check whether the device shows up in other nodes
6249 if (mddev_is_clustered(mddev)) {
6250 if (info->state & (1 << MD_DISK_CANDIDATE))
6251 set_bit(Candidate, &rdev->flags);
6252 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6253 /* --add initiated by this node */
6254 err = md_cluster_ops->add_new_disk(mddev, rdev);
6262 rdev->raid_disk = -1;
6263 err = bind_rdev_to_array(rdev, mddev);
6268 if (mddev_is_clustered(mddev)) {
6269 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6271 err = md_cluster_ops->new_disk_ack(mddev,
6274 md_kick_rdev_from_array(rdev);
6278 md_cluster_ops->add_new_disk_cancel(mddev);
6280 err = add_bound_rdev(rdev);
6284 err = add_bound_rdev(rdev);
6289 /* otherwise, add_new_disk is only allowed
6290 * for major_version==0 superblocks
6292 if (mddev->major_version != 0) {
6293 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6297 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6299 rdev = md_import_device(dev, -1, 0);
6301 pr_warn("md: error, md_import_device() returned %ld\n",
6303 return PTR_ERR(rdev);
6305 rdev->desc_nr = info->number;
6306 if (info->raid_disk < mddev->raid_disks)
6307 rdev->raid_disk = info->raid_disk;
6309 rdev->raid_disk = -1;
6311 if (rdev->raid_disk < mddev->raid_disks)
6312 if (info->state & (1<<MD_DISK_SYNC))
6313 set_bit(In_sync, &rdev->flags);
6315 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6316 set_bit(WriteMostly, &rdev->flags);
6317 if (info->state & (1<<MD_DISK_FAILFAST))
6318 set_bit(FailFast, &rdev->flags);
6320 if (!mddev->persistent) {
6321 pr_debug("md: nonpersistent superblock ...\n");
6322 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6324 rdev->sb_start = calc_dev_sboffset(rdev);
6325 rdev->sectors = rdev->sb_start;
6327 err = bind_rdev_to_array(rdev, mddev);
6337 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6339 char b[BDEVNAME_SIZE];
6340 struct md_rdev *rdev;
6342 rdev = find_rdev(mddev, dev);
6346 if (rdev->raid_disk < 0)
6349 clear_bit(Blocked, &rdev->flags);
6350 remove_and_add_spares(mddev, rdev);
6352 if (rdev->raid_disk >= 0)
6356 if (mddev_is_clustered(mddev))
6357 md_cluster_ops->remove_disk(mddev, rdev);
6359 md_kick_rdev_from_array(rdev);
6360 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6362 md_wakeup_thread(mddev->thread);
6364 md_update_sb(mddev, 1);
6365 md_new_event(mddev);
6369 pr_debug("md: cannot remove active disk %s from %s ...\n",
6370 bdevname(rdev->bdev,b), mdname(mddev));
6374 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6376 char b[BDEVNAME_SIZE];
6378 struct md_rdev *rdev;
6383 if (mddev->major_version != 0) {
6384 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6388 if (!mddev->pers->hot_add_disk) {
6389 pr_warn("%s: personality does not support diskops!\n",
6394 rdev = md_import_device(dev, -1, 0);
6396 pr_warn("md: error, md_import_device() returned %ld\n",
6401 if (mddev->persistent)
6402 rdev->sb_start = calc_dev_sboffset(rdev);
6404 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6406 rdev->sectors = rdev->sb_start;
6408 if (test_bit(Faulty, &rdev->flags)) {
6409 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6410 bdevname(rdev->bdev,b), mdname(mddev));
6415 clear_bit(In_sync, &rdev->flags);
6417 rdev->saved_raid_disk = -1;
6418 err = bind_rdev_to_array(rdev, mddev);
6423 * The rest should better be atomic, we can have disk failures
6424 * noticed in interrupt contexts ...
6427 rdev->raid_disk = -1;
6429 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6431 md_update_sb(mddev, 1);
6433 * Kick recovery, maybe this spare has to be added to the
6434 * array immediately.
6436 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6437 md_wakeup_thread(mddev->thread);
6438 md_new_event(mddev);
6446 static int set_bitmap_file(struct mddev *mddev, int fd)
6451 if (!mddev->pers->quiesce || !mddev->thread)
6453 if (mddev->recovery || mddev->sync_thread)
6455 /* we should be able to change the bitmap.. */
6459 struct inode *inode;
6462 if (mddev->bitmap || mddev->bitmap_info.file)
6463 return -EEXIST; /* cannot add when bitmap is present */
6467 pr_warn("%s: error: failed to get bitmap file\n",
6472 inode = f->f_mapping->host;
6473 if (!S_ISREG(inode->i_mode)) {
6474 pr_warn("%s: error: bitmap file must be a regular file\n",
6477 } else if (!(f->f_mode & FMODE_WRITE)) {
6478 pr_warn("%s: error: bitmap file must open for write\n",
6481 } else if (atomic_read(&inode->i_writecount) != 1) {
6482 pr_warn("%s: error: bitmap file is already in use\n",
6490 mddev->bitmap_info.file = f;
6491 mddev->bitmap_info.offset = 0; /* file overrides offset */
6492 } else if (mddev->bitmap == NULL)
6493 return -ENOENT; /* cannot remove what isn't there */
6496 mddev->pers->quiesce(mddev, 1);
6498 struct bitmap *bitmap;
6500 bitmap = bitmap_create(mddev, -1);
6501 if (!IS_ERR(bitmap)) {
6502 mddev->bitmap = bitmap;
6503 err = bitmap_load(mddev);
6505 err = PTR_ERR(bitmap);
6507 if (fd < 0 || err) {
6508 bitmap_destroy(mddev);
6509 fd = -1; /* make sure to put the file */
6511 mddev->pers->quiesce(mddev, 0);
6514 struct file *f = mddev->bitmap_info.file;
6516 spin_lock(&mddev->lock);
6517 mddev->bitmap_info.file = NULL;
6518 spin_unlock(&mddev->lock);
6527 * set_array_info is used two different ways
6528 * The original usage is when creating a new array.
6529 * In this usage, raid_disks is > 0 and it together with
6530 * level, size, not_persistent,layout,chunksize determine the
6531 * shape of the array.
6532 * This will always create an array with a type-0.90.0 superblock.
6533 * The newer usage is when assembling an array.
6534 * In this case raid_disks will be 0, and the major_version field is
6535 * use to determine which style super-blocks are to be found on the devices.
6536 * The minor and patch _version numbers are also kept incase the
6537 * super_block handler wishes to interpret them.
6539 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6542 if (info->raid_disks == 0) {
6543 /* just setting version number for superblock loading */
6544 if (info->major_version < 0 ||
6545 info->major_version >= ARRAY_SIZE(super_types) ||
6546 super_types[info->major_version].name == NULL) {
6547 /* maybe try to auto-load a module? */
6548 pr_warn("md: superblock version %d not known\n",
6549 info->major_version);
6552 mddev->major_version = info->major_version;
6553 mddev->minor_version = info->minor_version;
6554 mddev->patch_version = info->patch_version;
6555 mddev->persistent = !info->not_persistent;
6556 /* ensure mddev_put doesn't delete this now that there
6557 * is some minimal configuration.
6559 mddev->ctime = ktime_get_real_seconds();
6562 mddev->major_version = MD_MAJOR_VERSION;
6563 mddev->minor_version = MD_MINOR_VERSION;
6564 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6565 mddev->ctime = ktime_get_real_seconds();
6567 mddev->level = info->level;
6568 mddev->clevel[0] = 0;
6569 mddev->dev_sectors = 2 * (sector_t)info->size;
6570 mddev->raid_disks = info->raid_disks;
6571 /* don't set md_minor, it is determined by which /dev/md* was
6574 if (info->state & (1<<MD_SB_CLEAN))
6575 mddev->recovery_cp = MaxSector;
6577 mddev->recovery_cp = 0;
6578 mddev->persistent = ! info->not_persistent;
6579 mddev->external = 0;
6581 mddev->layout = info->layout;
6582 mddev->chunk_sectors = info->chunk_size >> 9;
6584 if (mddev->persistent) {
6585 mddev->max_disks = MD_SB_DISKS;
6587 mddev->sb_flags = 0;
6589 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6591 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6592 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6593 mddev->bitmap_info.offset = 0;
6595 mddev->reshape_position = MaxSector;
6598 * Generate a 128 bit UUID
6600 get_random_bytes(mddev->uuid, 16);
6602 mddev->new_level = mddev->level;
6603 mddev->new_chunk_sectors = mddev->chunk_sectors;
6604 mddev->new_layout = mddev->layout;
6605 mddev->delta_disks = 0;
6606 mddev->reshape_backwards = 0;
6611 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6613 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6615 if (mddev->external_size)
6618 mddev->array_sectors = array_sectors;
6620 EXPORT_SYMBOL(md_set_array_sectors);
6622 static int update_size(struct mddev *mddev, sector_t num_sectors)
6624 struct md_rdev *rdev;
6626 int fit = (num_sectors == 0);
6627 sector_t old_dev_sectors = mddev->dev_sectors;
6629 if (mddev->pers->resize == NULL)
6631 /* The "num_sectors" is the number of sectors of each device that
6632 * is used. This can only make sense for arrays with redundancy.
6633 * linear and raid0 always use whatever space is available. We can only
6634 * consider changing this number if no resync or reconstruction is
6635 * happening, and if the new size is acceptable. It must fit before the
6636 * sb_start or, if that is <data_offset, it must fit before the size
6637 * of each device. If num_sectors is zero, we find the largest size
6640 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6646 rdev_for_each(rdev, mddev) {
6647 sector_t avail = rdev->sectors;
6649 if (fit && (num_sectors == 0 || num_sectors > avail))
6650 num_sectors = avail;
6651 if (avail < num_sectors)
6654 rv = mddev->pers->resize(mddev, num_sectors);
6656 if (mddev_is_clustered(mddev))
6657 md_cluster_ops->update_size(mddev, old_dev_sectors);
6658 else if (mddev->queue) {
6659 set_capacity(mddev->gendisk, mddev->array_sectors);
6660 revalidate_disk(mddev->gendisk);
6666 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6669 struct md_rdev *rdev;
6670 /* change the number of raid disks */
6671 if (mddev->pers->check_reshape == NULL)
6675 if (raid_disks <= 0 ||
6676 (mddev->max_disks && raid_disks >= mddev->max_disks))
6678 if (mddev->sync_thread ||
6679 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6680 mddev->reshape_position != MaxSector)
6683 rdev_for_each(rdev, mddev) {
6684 if (mddev->raid_disks < raid_disks &&
6685 rdev->data_offset < rdev->new_data_offset)
6687 if (mddev->raid_disks > raid_disks &&
6688 rdev->data_offset > rdev->new_data_offset)
6692 mddev->delta_disks = raid_disks - mddev->raid_disks;
6693 if (mddev->delta_disks < 0)
6694 mddev->reshape_backwards = 1;
6695 else if (mddev->delta_disks > 0)
6696 mddev->reshape_backwards = 0;
6698 rv = mddev->pers->check_reshape(mddev);
6700 mddev->delta_disks = 0;
6701 mddev->reshape_backwards = 0;
6707 * update_array_info is used to change the configuration of an
6709 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6710 * fields in the info are checked against the array.
6711 * Any differences that cannot be handled will cause an error.
6712 * Normally, only one change can be managed at a time.
6714 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6720 /* calculate expected state,ignoring low bits */
6721 if (mddev->bitmap && mddev->bitmap_info.offset)
6722 state |= (1 << MD_SB_BITMAP_PRESENT);
6724 if (mddev->major_version != info->major_version ||
6725 mddev->minor_version != info->minor_version ||
6726 /* mddev->patch_version != info->patch_version || */
6727 mddev->ctime != info->ctime ||
6728 mddev->level != info->level ||
6729 /* mddev->layout != info->layout || */
6730 mddev->persistent != !info->not_persistent ||
6731 mddev->chunk_sectors != info->chunk_size >> 9 ||
6732 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6733 ((state^info->state) & 0xfffffe00)
6736 /* Check there is only one change */
6737 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6739 if (mddev->raid_disks != info->raid_disks)
6741 if (mddev->layout != info->layout)
6743 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6750 if (mddev->layout != info->layout) {
6752 * we don't need to do anything at the md level, the
6753 * personality will take care of it all.
6755 if (mddev->pers->check_reshape == NULL)
6758 mddev->new_layout = info->layout;
6759 rv = mddev->pers->check_reshape(mddev);
6761 mddev->new_layout = mddev->layout;
6765 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6766 rv = update_size(mddev, (sector_t)info->size * 2);
6768 if (mddev->raid_disks != info->raid_disks)
6769 rv = update_raid_disks(mddev, info->raid_disks);
6771 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6772 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6776 if (mddev->recovery || mddev->sync_thread) {
6780 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6781 struct bitmap *bitmap;
6782 /* add the bitmap */
6783 if (mddev->bitmap) {
6787 if (mddev->bitmap_info.default_offset == 0) {
6791 mddev->bitmap_info.offset =
6792 mddev->bitmap_info.default_offset;
6793 mddev->bitmap_info.space =
6794 mddev->bitmap_info.default_space;
6795 mddev->pers->quiesce(mddev, 1);
6796 bitmap = bitmap_create(mddev, -1);
6797 if (!IS_ERR(bitmap)) {
6798 mddev->bitmap = bitmap;
6799 rv = bitmap_load(mddev);
6801 rv = PTR_ERR(bitmap);
6803 bitmap_destroy(mddev);
6804 mddev->pers->quiesce(mddev, 0);
6806 /* remove the bitmap */
6807 if (!mddev->bitmap) {
6811 if (mddev->bitmap->storage.file) {
6815 if (mddev->bitmap_info.nodes) {
6816 /* hold PW on all the bitmap lock */
6817 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6818 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6820 md_cluster_ops->unlock_all_bitmaps(mddev);
6824 mddev->bitmap_info.nodes = 0;
6825 md_cluster_ops->leave(mddev);
6827 mddev->pers->quiesce(mddev, 1);
6828 bitmap_destroy(mddev);
6829 mddev->pers->quiesce(mddev, 0);
6830 mddev->bitmap_info.offset = 0;
6833 md_update_sb(mddev, 1);
6839 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6841 struct md_rdev *rdev;
6844 if (mddev->pers == NULL)
6848 rdev = find_rdev_rcu(mddev, dev);
6852 md_error(mddev, rdev);
6853 if (!test_bit(Faulty, &rdev->flags))
6861 * We have a problem here : there is no easy way to give a CHS
6862 * virtual geometry. We currently pretend that we have a 2 heads
6863 * 4 sectors (with a BIG number of cylinders...). This drives
6864 * dosfs just mad... ;-)
6866 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6868 struct mddev *mddev = bdev->bd_disk->private_data;
6872 geo->cylinders = mddev->array_sectors / 8;
6876 static inline bool md_ioctl_valid(unsigned int cmd)
6881 case GET_ARRAY_INFO:
6882 case GET_BITMAP_FILE:
6885 case HOT_REMOVE_DISK:
6888 case RESTART_ARRAY_RW:
6890 case SET_ARRAY_INFO:
6891 case SET_BITMAP_FILE:
6892 case SET_DISK_FAULTY:
6895 case CLUSTERED_DISK_NACK:
6902 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6903 unsigned int cmd, unsigned long arg)
6906 void __user *argp = (void __user *)arg;
6907 struct mddev *mddev = NULL;
6910 if (!md_ioctl_valid(cmd))
6915 case GET_ARRAY_INFO:
6919 if (!capable(CAP_SYS_ADMIN))
6924 * Commands dealing with the RAID driver but not any
6929 err = get_version(argp);
6935 autostart_arrays(arg);
6942 * Commands creating/starting a new array:
6945 mddev = bdev->bd_disk->private_data;
6952 /* Some actions do not requires the mutex */
6954 case GET_ARRAY_INFO:
6955 if (!mddev->raid_disks && !mddev->external)
6958 err = get_array_info(mddev, argp);
6962 if (!mddev->raid_disks && !mddev->external)
6965 err = get_disk_info(mddev, argp);
6968 case SET_DISK_FAULTY:
6969 err = set_disk_faulty(mddev, new_decode_dev(arg));
6972 case GET_BITMAP_FILE:
6973 err = get_bitmap_file(mddev, argp);
6978 if (cmd == ADD_NEW_DISK)
6979 /* need to ensure md_delayed_delete() has completed */
6980 flush_workqueue(md_misc_wq);
6982 if (cmd == HOT_REMOVE_DISK)
6983 /* need to ensure recovery thread has run */
6984 wait_event_interruptible_timeout(mddev->sb_wait,
6985 !test_bit(MD_RECOVERY_NEEDED,
6987 msecs_to_jiffies(5000));
6988 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6989 /* Need to flush page cache, and ensure no-one else opens
6992 mutex_lock(&mddev->open_mutex);
6993 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6994 mutex_unlock(&mddev->open_mutex);
6998 set_bit(MD_CLOSING, &mddev->flags);
6999 mutex_unlock(&mddev->open_mutex);
7000 sync_blockdev(bdev);
7002 err = mddev_lock(mddev);
7004 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7009 if (cmd == SET_ARRAY_INFO) {
7010 mdu_array_info_t info;
7012 memset(&info, 0, sizeof(info));
7013 else if (copy_from_user(&info, argp, sizeof(info))) {
7018 err = update_array_info(mddev, &info);
7020 pr_warn("md: couldn't update array info. %d\n", err);
7025 if (!list_empty(&mddev->disks)) {
7026 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7030 if (mddev->raid_disks) {
7031 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7035 err = set_array_info(mddev, &info);
7037 pr_warn("md: couldn't set array info. %d\n", err);
7044 * Commands querying/configuring an existing array:
7046 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7047 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7048 if ((!mddev->raid_disks && !mddev->external)
7049 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7050 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7051 && cmd != GET_BITMAP_FILE) {
7057 * Commands even a read-only array can execute:
7060 case RESTART_ARRAY_RW:
7061 err = restart_array(mddev);
7065 err = do_md_stop(mddev, 0, bdev);
7069 err = md_set_readonly(mddev, bdev);
7072 case HOT_REMOVE_DISK:
7073 err = hot_remove_disk(mddev, new_decode_dev(arg));
7077 /* We can support ADD_NEW_DISK on read-only arrays
7078 * only if we are re-adding a preexisting device.
7079 * So require mddev->pers and MD_DISK_SYNC.
7082 mdu_disk_info_t info;
7083 if (copy_from_user(&info, argp, sizeof(info)))
7085 else if (!(info.state & (1<<MD_DISK_SYNC)))
7086 /* Need to clear read-only for this */
7089 err = add_new_disk(mddev, &info);
7095 if (get_user(ro, (int __user *)(arg))) {
7101 /* if the bdev is going readonly the value of mddev->ro
7102 * does not matter, no writes are coming
7107 /* are we are already prepared for writes? */
7111 /* transitioning to readauto need only happen for
7112 * arrays that call md_write_start
7115 err = restart_array(mddev);
7118 set_disk_ro(mddev->gendisk, 0);
7125 * The remaining ioctls are changing the state of the
7126 * superblock, so we do not allow them on read-only arrays.
7128 if (mddev->ro && mddev->pers) {
7129 if (mddev->ro == 2) {
7131 sysfs_notify_dirent_safe(mddev->sysfs_state);
7132 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7133 /* mddev_unlock will wake thread */
7134 /* If a device failed while we were read-only, we
7135 * need to make sure the metadata is updated now.
7137 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7138 mddev_unlock(mddev);
7139 wait_event(mddev->sb_wait,
7140 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7141 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7142 mddev_lock_nointr(mddev);
7153 mdu_disk_info_t info;
7154 if (copy_from_user(&info, argp, sizeof(info)))
7157 err = add_new_disk(mddev, &info);
7161 case CLUSTERED_DISK_NACK:
7162 if (mddev_is_clustered(mddev))
7163 md_cluster_ops->new_disk_ack(mddev, false);
7169 err = hot_add_disk(mddev, new_decode_dev(arg));
7173 err = do_md_run(mddev);
7176 case SET_BITMAP_FILE:
7177 err = set_bitmap_file(mddev, (int)arg);
7186 if (mddev->hold_active == UNTIL_IOCTL &&
7188 mddev->hold_active = 0;
7189 mddev_unlock(mddev);
7193 #ifdef CONFIG_COMPAT
7194 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7195 unsigned int cmd, unsigned long arg)
7198 case HOT_REMOVE_DISK:
7200 case SET_DISK_FAULTY:
7201 case SET_BITMAP_FILE:
7202 /* These take in integer arg, do not convert */
7205 arg = (unsigned long)compat_ptr(arg);
7209 return md_ioctl(bdev, mode, cmd, arg);
7211 #endif /* CONFIG_COMPAT */
7213 static int md_open(struct block_device *bdev, fmode_t mode)
7216 * Succeed if we can lock the mddev, which confirms that
7217 * it isn't being stopped right now.
7219 struct mddev *mddev = mddev_find(bdev->bd_dev);
7225 if (mddev->gendisk != bdev->bd_disk) {
7226 /* we are racing with mddev_put which is discarding this
7230 /* Wait until bdev->bd_disk is definitely gone */
7231 flush_workqueue(md_misc_wq);
7232 /* Then retry the open from the top */
7233 return -ERESTARTSYS;
7235 BUG_ON(mddev != bdev->bd_disk->private_data);
7237 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7240 if (test_bit(MD_CLOSING, &mddev->flags)) {
7241 mutex_unlock(&mddev->open_mutex);
7247 atomic_inc(&mddev->openers);
7248 mutex_unlock(&mddev->open_mutex);
7250 check_disk_change(bdev);
7257 static void md_release(struct gendisk *disk, fmode_t mode)
7259 struct mddev *mddev = disk->private_data;
7262 atomic_dec(&mddev->openers);
7266 static int md_media_changed(struct gendisk *disk)
7268 struct mddev *mddev = disk->private_data;
7270 return mddev->changed;
7273 static int md_revalidate(struct gendisk *disk)
7275 struct mddev *mddev = disk->private_data;
7280 static const struct block_device_operations md_fops =
7282 .owner = THIS_MODULE,
7284 .release = md_release,
7286 #ifdef CONFIG_COMPAT
7287 .compat_ioctl = md_compat_ioctl,
7289 .getgeo = md_getgeo,
7290 .media_changed = md_media_changed,
7291 .revalidate_disk= md_revalidate,
7294 static int md_thread(void *arg)
7296 struct md_thread *thread = arg;
7299 * md_thread is a 'system-thread', it's priority should be very
7300 * high. We avoid resource deadlocks individually in each
7301 * raid personality. (RAID5 does preallocation) We also use RR and
7302 * the very same RT priority as kswapd, thus we will never get
7303 * into a priority inversion deadlock.
7305 * we definitely have to have equal or higher priority than
7306 * bdflush, otherwise bdflush will deadlock if there are too
7307 * many dirty RAID5 blocks.
7310 allow_signal(SIGKILL);
7311 while (!kthread_should_stop()) {
7313 /* We need to wait INTERRUPTIBLE so that
7314 * we don't add to the load-average.
7315 * That means we need to be sure no signals are
7318 if (signal_pending(current))
7319 flush_signals(current);
7321 wait_event_interruptible_timeout
7323 test_bit(THREAD_WAKEUP, &thread->flags)
7324 || kthread_should_stop() || kthread_should_park(),
7327 clear_bit(THREAD_WAKEUP, &thread->flags);
7328 if (kthread_should_park())
7330 if (!kthread_should_stop())
7331 thread->run(thread);
7337 void md_wakeup_thread(struct md_thread *thread)
7340 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7341 set_bit(THREAD_WAKEUP, &thread->flags);
7342 wake_up(&thread->wqueue);
7345 EXPORT_SYMBOL(md_wakeup_thread);
7347 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7348 struct mddev *mddev, const char *name)
7350 struct md_thread *thread;
7352 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7356 init_waitqueue_head(&thread->wqueue);
7359 thread->mddev = mddev;
7360 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7361 thread->tsk = kthread_run(md_thread, thread,
7363 mdname(thread->mddev),
7365 if (IS_ERR(thread->tsk)) {
7371 EXPORT_SYMBOL(md_register_thread);
7373 void md_unregister_thread(struct md_thread **threadp)
7375 struct md_thread *thread = *threadp;
7378 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7379 /* Locking ensures that mddev_unlock does not wake_up a
7380 * non-existent thread
7382 spin_lock(&pers_lock);
7384 spin_unlock(&pers_lock);
7386 kthread_stop(thread->tsk);
7389 EXPORT_SYMBOL(md_unregister_thread);
7391 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7393 if (!rdev || test_bit(Faulty, &rdev->flags))
7396 if (!mddev->pers || !mddev->pers->error_handler)
7398 mddev->pers->error_handler(mddev,rdev);
7399 if (mddev->degraded)
7400 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7401 sysfs_notify_dirent_safe(rdev->sysfs_state);
7402 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7403 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7404 md_wakeup_thread(mddev->thread);
7405 if (mddev->event_work.func)
7406 queue_work(md_misc_wq, &mddev->event_work);
7407 md_new_event(mddev);
7409 EXPORT_SYMBOL(md_error);
7411 /* seq_file implementation /proc/mdstat */
7413 static void status_unused(struct seq_file *seq)
7416 struct md_rdev *rdev;
7418 seq_printf(seq, "unused devices: ");
7420 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7421 char b[BDEVNAME_SIZE];
7423 seq_printf(seq, "%s ",
7424 bdevname(rdev->bdev,b));
7427 seq_printf(seq, "<none>");
7429 seq_printf(seq, "\n");
7432 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7434 sector_t max_sectors, resync, res;
7435 unsigned long dt, db;
7438 unsigned int per_milli;
7440 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7441 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7442 max_sectors = mddev->resync_max_sectors;
7444 max_sectors = mddev->dev_sectors;
7446 resync = mddev->curr_resync;
7448 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7449 /* Still cleaning up */
7450 resync = max_sectors;
7452 resync -= atomic_read(&mddev->recovery_active);
7455 if (mddev->recovery_cp < MaxSector) {
7456 seq_printf(seq, "\tresync=PENDING");
7462 seq_printf(seq, "\tresync=DELAYED");
7466 WARN_ON(max_sectors == 0);
7467 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7468 * in a sector_t, and (max_sectors>>scale) will fit in a
7469 * u32, as those are the requirements for sector_div.
7470 * Thus 'scale' must be at least 10
7473 if (sizeof(sector_t) > sizeof(unsigned long)) {
7474 while ( max_sectors/2 > (1ULL<<(scale+32)))
7477 res = (resync>>scale)*1000;
7478 sector_div(res, (u32)((max_sectors>>scale)+1));
7482 int i, x = per_milli/50, y = 20-x;
7483 seq_printf(seq, "[");
7484 for (i = 0; i < x; i++)
7485 seq_printf(seq, "=");
7486 seq_printf(seq, ">");
7487 for (i = 0; i < y; i++)
7488 seq_printf(seq, ".");
7489 seq_printf(seq, "] ");
7491 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7492 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7494 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7496 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7497 "resync" : "recovery"))),
7498 per_milli/10, per_milli % 10,
7499 (unsigned long long) resync/2,
7500 (unsigned long long) max_sectors/2);
7503 * dt: time from mark until now
7504 * db: blocks written from mark until now
7505 * rt: remaining time
7507 * rt is a sector_t, so could be 32bit or 64bit.
7508 * So we divide before multiply in case it is 32bit and close
7510 * We scale the divisor (db) by 32 to avoid losing precision
7511 * near the end of resync when the number of remaining sectors
7513 * We then divide rt by 32 after multiplying by db to compensate.
7514 * The '+1' avoids division by zero if db is very small.
7516 dt = ((jiffies - mddev->resync_mark) / HZ);
7518 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7519 - mddev->resync_mark_cnt;
7521 rt = max_sectors - resync; /* number of remaining sectors */
7522 sector_div(rt, db/32+1);
7526 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7527 ((unsigned long)rt % 60)/6);
7529 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7533 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7535 struct list_head *tmp;
7537 struct mddev *mddev;
7545 spin_lock(&all_mddevs_lock);
7546 list_for_each(tmp,&all_mddevs)
7548 mddev = list_entry(tmp, struct mddev, all_mddevs);
7550 spin_unlock(&all_mddevs_lock);
7553 spin_unlock(&all_mddevs_lock);
7555 return (void*)2;/* tail */
7559 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7561 struct list_head *tmp;
7562 struct mddev *next_mddev, *mddev = v;
7568 spin_lock(&all_mddevs_lock);
7570 tmp = all_mddevs.next;
7572 tmp = mddev->all_mddevs.next;
7573 if (tmp != &all_mddevs)
7574 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7576 next_mddev = (void*)2;
7579 spin_unlock(&all_mddevs_lock);
7587 static void md_seq_stop(struct seq_file *seq, void *v)
7589 struct mddev *mddev = v;
7591 if (mddev && v != (void*)1 && v != (void*)2)
7595 static int md_seq_show(struct seq_file *seq, void *v)
7597 struct mddev *mddev = v;
7599 struct md_rdev *rdev;
7601 if (v == (void*)1) {
7602 struct md_personality *pers;
7603 seq_printf(seq, "Personalities : ");
7604 spin_lock(&pers_lock);
7605 list_for_each_entry(pers, &pers_list, list)
7606 seq_printf(seq, "[%s] ", pers->name);
7608 spin_unlock(&pers_lock);
7609 seq_printf(seq, "\n");
7610 seq->poll_event = atomic_read(&md_event_count);
7613 if (v == (void*)2) {
7618 spin_lock(&mddev->lock);
7619 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7620 seq_printf(seq, "%s : %sactive", mdname(mddev),
7621 mddev->pers ? "" : "in");
7624 seq_printf(seq, " (read-only)");
7626 seq_printf(seq, " (auto-read-only)");
7627 seq_printf(seq, " %s", mddev->pers->name);
7632 rdev_for_each_rcu(rdev, mddev) {
7633 char b[BDEVNAME_SIZE];
7634 seq_printf(seq, " %s[%d]",
7635 bdevname(rdev->bdev,b), rdev->desc_nr);
7636 if (test_bit(WriteMostly, &rdev->flags))
7637 seq_printf(seq, "(W)");
7638 if (test_bit(Journal, &rdev->flags))
7639 seq_printf(seq, "(J)");
7640 if (test_bit(Faulty, &rdev->flags)) {
7641 seq_printf(seq, "(F)");
7644 if (rdev->raid_disk < 0)
7645 seq_printf(seq, "(S)"); /* spare */
7646 if (test_bit(Replacement, &rdev->flags))
7647 seq_printf(seq, "(R)");
7648 sectors += rdev->sectors;
7652 if (!list_empty(&mddev->disks)) {
7654 seq_printf(seq, "\n %llu blocks",
7655 (unsigned long long)
7656 mddev->array_sectors / 2);
7658 seq_printf(seq, "\n %llu blocks",
7659 (unsigned long long)sectors / 2);
7661 if (mddev->persistent) {
7662 if (mddev->major_version != 0 ||
7663 mddev->minor_version != 90) {
7664 seq_printf(seq," super %d.%d",
7665 mddev->major_version,
7666 mddev->minor_version);
7668 } else if (mddev->external)
7669 seq_printf(seq, " super external:%s",
7670 mddev->metadata_type);
7672 seq_printf(seq, " super non-persistent");
7675 mddev->pers->status(seq, mddev);
7676 seq_printf(seq, "\n ");
7677 if (mddev->pers->sync_request) {
7678 if (status_resync(seq, mddev))
7679 seq_printf(seq, "\n ");
7682 seq_printf(seq, "\n ");
7684 bitmap_status(seq, mddev->bitmap);
7686 seq_printf(seq, "\n");
7688 spin_unlock(&mddev->lock);
7693 static const struct seq_operations md_seq_ops = {
7694 .start = md_seq_start,
7695 .next = md_seq_next,
7696 .stop = md_seq_stop,
7697 .show = md_seq_show,
7700 static int md_seq_open(struct inode *inode, struct file *file)
7702 struct seq_file *seq;
7705 error = seq_open(file, &md_seq_ops);
7709 seq = file->private_data;
7710 seq->poll_event = atomic_read(&md_event_count);
7714 static int md_unloading;
7715 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7717 struct seq_file *seq = filp->private_data;
7721 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7722 poll_wait(filp, &md_event_waiters, wait);
7724 /* always allow read */
7725 mask = POLLIN | POLLRDNORM;
7727 if (seq->poll_event != atomic_read(&md_event_count))
7728 mask |= POLLERR | POLLPRI;
7732 static const struct file_operations md_seq_fops = {
7733 .owner = THIS_MODULE,
7734 .open = md_seq_open,
7736 .llseek = seq_lseek,
7737 .release = seq_release_private,
7738 .poll = mdstat_poll,
7741 int register_md_personality(struct md_personality *p)
7743 pr_debug("md: %s personality registered for level %d\n",
7745 spin_lock(&pers_lock);
7746 list_add_tail(&p->list, &pers_list);
7747 spin_unlock(&pers_lock);
7750 EXPORT_SYMBOL(register_md_personality);
7752 int unregister_md_personality(struct md_personality *p)
7754 pr_debug("md: %s personality unregistered\n", p->name);
7755 spin_lock(&pers_lock);
7756 list_del_init(&p->list);
7757 spin_unlock(&pers_lock);
7760 EXPORT_SYMBOL(unregister_md_personality);
7762 int register_md_cluster_operations(struct md_cluster_operations *ops,
7763 struct module *module)
7766 spin_lock(&pers_lock);
7767 if (md_cluster_ops != NULL)
7770 md_cluster_ops = ops;
7771 md_cluster_mod = module;
7773 spin_unlock(&pers_lock);
7776 EXPORT_SYMBOL(register_md_cluster_operations);
7778 int unregister_md_cluster_operations(void)
7780 spin_lock(&pers_lock);
7781 md_cluster_ops = NULL;
7782 spin_unlock(&pers_lock);
7785 EXPORT_SYMBOL(unregister_md_cluster_operations);
7787 int md_setup_cluster(struct mddev *mddev, int nodes)
7789 if (!md_cluster_ops)
7790 request_module("md-cluster");
7791 spin_lock(&pers_lock);
7792 /* ensure module won't be unloaded */
7793 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7794 pr_warn("can't find md-cluster module or get it's reference.\n");
7795 spin_unlock(&pers_lock);
7798 spin_unlock(&pers_lock);
7800 return md_cluster_ops->join(mddev, nodes);
7803 void md_cluster_stop(struct mddev *mddev)
7805 if (!md_cluster_ops)
7807 md_cluster_ops->leave(mddev);
7808 module_put(md_cluster_mod);
7811 static int is_mddev_idle(struct mddev *mddev, int init)
7813 struct md_rdev *rdev;
7819 rdev_for_each_rcu(rdev, mddev) {
7820 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7821 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7822 (int)part_stat_read(&disk->part0, sectors[1]) -
7823 atomic_read(&disk->sync_io);
7824 /* sync IO will cause sync_io to increase before the disk_stats
7825 * as sync_io is counted when a request starts, and
7826 * disk_stats is counted when it completes.
7827 * So resync activity will cause curr_events to be smaller than
7828 * when there was no such activity.
7829 * non-sync IO will cause disk_stat to increase without
7830 * increasing sync_io so curr_events will (eventually)
7831 * be larger than it was before. Once it becomes
7832 * substantially larger, the test below will cause
7833 * the array to appear non-idle, and resync will slow
7835 * If there is a lot of outstanding resync activity when
7836 * we set last_event to curr_events, then all that activity
7837 * completing might cause the array to appear non-idle
7838 * and resync will be slowed down even though there might
7839 * not have been non-resync activity. This will only
7840 * happen once though. 'last_events' will soon reflect
7841 * the state where there is little or no outstanding
7842 * resync requests, and further resync activity will
7843 * always make curr_events less than last_events.
7846 if (init || curr_events - rdev->last_events > 64) {
7847 rdev->last_events = curr_events;
7855 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7857 /* another "blocks" (512byte) blocks have been synced */
7858 atomic_sub(blocks, &mddev->recovery_active);
7859 wake_up(&mddev->recovery_wait);
7861 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7862 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7863 md_wakeup_thread(mddev->thread);
7864 // stop recovery, signal do_sync ....
7867 EXPORT_SYMBOL(md_done_sync);
7869 /* md_write_start(mddev, bi)
7870 * If we need to update some array metadata (e.g. 'active' flag
7871 * in superblock) before writing, schedule a superblock update
7872 * and wait for it to complete.
7874 void md_write_start(struct mddev *mddev, struct bio *bi)
7877 if (bio_data_dir(bi) != WRITE)
7880 BUG_ON(mddev->ro == 1);
7881 if (mddev->ro == 2) {
7882 /* need to switch to read/write */
7884 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7885 md_wakeup_thread(mddev->thread);
7886 md_wakeup_thread(mddev->sync_thread);
7889 atomic_inc(&mddev->writes_pending);
7890 if (mddev->safemode == 1)
7891 mddev->safemode = 0;
7892 if (mddev->in_sync) {
7893 spin_lock(&mddev->lock);
7894 if (mddev->in_sync) {
7896 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7897 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7898 md_wakeup_thread(mddev->thread);
7901 spin_unlock(&mddev->lock);
7904 sysfs_notify_dirent_safe(mddev->sysfs_state);
7905 wait_event(mddev->sb_wait,
7906 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7908 EXPORT_SYMBOL(md_write_start);
7910 void md_write_end(struct mddev *mddev)
7912 if (atomic_dec_and_test(&mddev->writes_pending)) {
7913 if (mddev->safemode == 2)
7914 md_wakeup_thread(mddev->thread);
7915 else if (mddev->safemode_delay)
7916 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7919 EXPORT_SYMBOL(md_write_end);
7921 /* md_allow_write(mddev)
7922 * Calling this ensures that the array is marked 'active' so that writes
7923 * may proceed without blocking. It is important to call this before
7924 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7925 * Must be called with mddev_lock held.
7927 * In the ->external case MD_SB_CHANGE_PENDING can not be cleared until mddev->lock
7928 * is dropped, so return -EAGAIN after notifying userspace.
7930 int md_allow_write(struct mddev *mddev)
7936 if (!mddev->pers->sync_request)
7939 spin_lock(&mddev->lock);
7940 if (mddev->in_sync) {
7942 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
7943 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
7944 if (mddev->safemode_delay &&
7945 mddev->safemode == 0)
7946 mddev->safemode = 1;
7947 spin_unlock(&mddev->lock);
7948 md_update_sb(mddev, 0);
7949 sysfs_notify_dirent_safe(mddev->sysfs_state);
7951 spin_unlock(&mddev->lock);
7953 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
7958 EXPORT_SYMBOL_GPL(md_allow_write);
7960 #define SYNC_MARKS 10
7961 #define SYNC_MARK_STEP (3*HZ)
7962 #define UPDATE_FREQUENCY (5*60*HZ)
7963 void md_do_sync(struct md_thread *thread)
7965 struct mddev *mddev = thread->mddev;
7966 struct mddev *mddev2;
7967 unsigned int currspeed = 0,
7969 sector_t max_sectors,j, io_sectors, recovery_done;
7970 unsigned long mark[SYNC_MARKS];
7971 unsigned long update_time;
7972 sector_t mark_cnt[SYNC_MARKS];
7974 struct list_head *tmp;
7975 sector_t last_check;
7977 struct md_rdev *rdev;
7978 char *desc, *action = NULL;
7979 struct blk_plug plug;
7982 /* just incase thread restarts... */
7983 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7985 if (mddev->ro) {/* never try to sync a read-only array */
7986 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7990 if (mddev_is_clustered(mddev)) {
7991 ret = md_cluster_ops->resync_start(mddev);
7995 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7996 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7997 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7998 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7999 && ((unsigned long long)mddev->curr_resync_completed
8000 < (unsigned long long)mddev->resync_max_sectors))
8004 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8005 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8006 desc = "data-check";
8008 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8009 desc = "requested-resync";
8013 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8018 mddev->last_sync_action = action ?: desc;
8020 /* we overload curr_resync somewhat here.
8021 * 0 == not engaged in resync at all
8022 * 2 == checking that there is no conflict with another sync
8023 * 1 == like 2, but have yielded to allow conflicting resync to
8025 * other == active in resync - this many blocks
8027 * Before starting a resync we must have set curr_resync to
8028 * 2, and then checked that every "conflicting" array has curr_resync
8029 * less than ours. When we find one that is the same or higher
8030 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8031 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8032 * This will mean we have to start checking from the beginning again.
8037 int mddev2_minor = -1;
8038 mddev->curr_resync = 2;
8041 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8043 for_each_mddev(mddev2, tmp) {
8044 if (mddev2 == mddev)
8046 if (!mddev->parallel_resync
8047 && mddev2->curr_resync
8048 && match_mddev_units(mddev, mddev2)) {
8050 if (mddev < mddev2 && mddev->curr_resync == 2) {
8051 /* arbitrarily yield */
8052 mddev->curr_resync = 1;
8053 wake_up(&resync_wait);
8055 if (mddev > mddev2 && mddev->curr_resync == 1)
8056 /* no need to wait here, we can wait the next
8057 * time 'round when curr_resync == 2
8060 /* We need to wait 'interruptible' so as not to
8061 * contribute to the load average, and not to
8062 * be caught by 'softlockup'
8064 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8065 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8066 mddev2->curr_resync >= mddev->curr_resync) {
8067 if (mddev2_minor != mddev2->md_minor) {
8068 mddev2_minor = mddev2->md_minor;
8069 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8070 desc, mdname(mddev),
8074 if (signal_pending(current))
8075 flush_signals(current);
8077 finish_wait(&resync_wait, &wq);
8080 finish_wait(&resync_wait, &wq);
8083 } while (mddev->curr_resync < 2);
8086 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8087 /* resync follows the size requested by the personality,
8088 * which defaults to physical size, but can be virtual size
8090 max_sectors = mddev->resync_max_sectors;
8091 atomic64_set(&mddev->resync_mismatches, 0);
8092 /* we don't use the checkpoint if there's a bitmap */
8093 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8094 j = mddev->resync_min;
8095 else if (!mddev->bitmap)
8096 j = mddev->recovery_cp;
8098 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8099 max_sectors = mddev->resync_max_sectors;
8101 /* recovery follows the physical size of devices */
8102 max_sectors = mddev->dev_sectors;
8105 rdev_for_each_rcu(rdev, mddev)
8106 if (rdev->raid_disk >= 0 &&
8107 !test_bit(Journal, &rdev->flags) &&
8108 !test_bit(Faulty, &rdev->flags) &&
8109 !test_bit(In_sync, &rdev->flags) &&
8110 rdev->recovery_offset < j)
8111 j = rdev->recovery_offset;
8114 /* If there is a bitmap, we need to make sure all
8115 * writes that started before we added a spare
8116 * complete before we start doing a recovery.
8117 * Otherwise the write might complete and (via
8118 * bitmap_endwrite) set a bit in the bitmap after the
8119 * recovery has checked that bit and skipped that
8122 if (mddev->bitmap) {
8123 mddev->pers->quiesce(mddev, 1);
8124 mddev->pers->quiesce(mddev, 0);
8128 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8129 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8130 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8131 speed_max(mddev), desc);
8133 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8136 for (m = 0; m < SYNC_MARKS; m++) {
8138 mark_cnt[m] = io_sectors;
8141 mddev->resync_mark = mark[last_mark];
8142 mddev->resync_mark_cnt = mark_cnt[last_mark];
8145 * Tune reconstruction:
8147 window = 32*(PAGE_SIZE/512);
8148 pr_debug("md: using %dk window, over a total of %lluk.\n",
8149 window/2, (unsigned long long)max_sectors/2);
8151 atomic_set(&mddev->recovery_active, 0);
8155 pr_debug("md: resuming %s of %s from checkpoint.\n",
8156 desc, mdname(mddev));
8157 mddev->curr_resync = j;
8159 mddev->curr_resync = 3; /* no longer delayed */
8160 mddev->curr_resync_completed = j;
8161 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8162 md_new_event(mddev);
8163 update_time = jiffies;
8165 blk_start_plug(&plug);
8166 while (j < max_sectors) {
8171 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8172 ((mddev->curr_resync > mddev->curr_resync_completed &&
8173 (mddev->curr_resync - mddev->curr_resync_completed)
8174 > (max_sectors >> 4)) ||
8175 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8176 (j - mddev->curr_resync_completed)*2
8177 >= mddev->resync_max - mddev->curr_resync_completed ||
8178 mddev->curr_resync_completed > mddev->resync_max
8180 /* time to update curr_resync_completed */
8181 wait_event(mddev->recovery_wait,
8182 atomic_read(&mddev->recovery_active) == 0);
8183 mddev->curr_resync_completed = j;
8184 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8185 j > mddev->recovery_cp)
8186 mddev->recovery_cp = j;
8187 update_time = jiffies;
8188 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8189 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8192 while (j >= mddev->resync_max &&
8193 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8194 /* As this condition is controlled by user-space,
8195 * we can block indefinitely, so use '_interruptible'
8196 * to avoid triggering warnings.
8198 flush_signals(current); /* just in case */
8199 wait_event_interruptible(mddev->recovery_wait,
8200 mddev->resync_max > j
8201 || test_bit(MD_RECOVERY_INTR,
8205 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8208 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8210 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8214 if (!skipped) { /* actual IO requested */
8215 io_sectors += sectors;
8216 atomic_add(sectors, &mddev->recovery_active);
8219 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8223 if (j > max_sectors)
8224 /* when skipping, extra large numbers can be returned. */
8227 mddev->curr_resync = j;
8228 mddev->curr_mark_cnt = io_sectors;
8229 if (last_check == 0)
8230 /* this is the earliest that rebuild will be
8231 * visible in /proc/mdstat
8233 md_new_event(mddev);
8235 if (last_check + window > io_sectors || j == max_sectors)
8238 last_check = io_sectors;
8240 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8242 int next = (last_mark+1) % SYNC_MARKS;
8244 mddev->resync_mark = mark[next];
8245 mddev->resync_mark_cnt = mark_cnt[next];
8246 mark[next] = jiffies;
8247 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8251 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8255 * this loop exits only if either when we are slower than
8256 * the 'hard' speed limit, or the system was IO-idle for
8258 * the system might be non-idle CPU-wise, but we only care
8259 * about not overloading the IO subsystem. (things like an
8260 * e2fsck being done on the RAID array should execute fast)
8264 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8265 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8266 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8268 if (currspeed > speed_min(mddev)) {
8269 if (currspeed > speed_max(mddev)) {
8273 if (!is_mddev_idle(mddev, 0)) {
8275 * Give other IO more of a chance.
8276 * The faster the devices, the less we wait.
8278 wait_event(mddev->recovery_wait,
8279 !atomic_read(&mddev->recovery_active));
8283 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8284 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8285 ? "interrupted" : "done");
8287 * this also signals 'finished resyncing' to md_stop
8289 blk_finish_plug(&plug);
8290 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8292 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8293 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8294 mddev->curr_resync > 3) {
8295 mddev->curr_resync_completed = mddev->curr_resync;
8296 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8298 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8300 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8301 mddev->curr_resync > 3) {
8302 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8303 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8304 if (mddev->curr_resync >= mddev->recovery_cp) {
8305 pr_debug("md: checkpointing %s of %s.\n",
8306 desc, mdname(mddev));
8307 if (test_bit(MD_RECOVERY_ERROR,
8309 mddev->recovery_cp =
8310 mddev->curr_resync_completed;
8312 mddev->recovery_cp =
8316 mddev->recovery_cp = MaxSector;
8318 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8319 mddev->curr_resync = MaxSector;
8321 rdev_for_each_rcu(rdev, mddev)
8322 if (rdev->raid_disk >= 0 &&
8323 mddev->delta_disks >= 0 &&
8324 !test_bit(Journal, &rdev->flags) &&
8325 !test_bit(Faulty, &rdev->flags) &&
8326 !test_bit(In_sync, &rdev->flags) &&
8327 rdev->recovery_offset < mddev->curr_resync)
8328 rdev->recovery_offset = mddev->curr_resync;
8333 /* set CHANGE_PENDING here since maybe another update is needed,
8334 * so other nodes are informed. It should be harmless for normal
8336 set_mask_bits(&mddev->sb_flags, 0,
8337 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8339 spin_lock(&mddev->lock);
8340 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8341 /* We completed so min/max setting can be forgotten if used. */
8342 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8343 mddev->resync_min = 0;
8344 mddev->resync_max = MaxSector;
8345 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8346 mddev->resync_min = mddev->curr_resync_completed;
8347 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8348 mddev->curr_resync = 0;
8349 spin_unlock(&mddev->lock);
8351 wake_up(&resync_wait);
8352 md_wakeup_thread(mddev->thread);
8355 EXPORT_SYMBOL_GPL(md_do_sync);
8357 static int remove_and_add_spares(struct mddev *mddev,
8358 struct md_rdev *this)
8360 struct md_rdev *rdev;
8363 bool remove_some = false;
8365 rdev_for_each(rdev, mddev) {
8366 if ((this == NULL || rdev == this) &&
8367 rdev->raid_disk >= 0 &&
8368 !test_bit(Blocked, &rdev->flags) &&
8369 test_bit(Faulty, &rdev->flags) &&
8370 atomic_read(&rdev->nr_pending)==0) {
8371 /* Faulty non-Blocked devices with nr_pending == 0
8372 * never get nr_pending incremented,
8373 * never get Faulty cleared, and never get Blocked set.
8374 * So we can synchronize_rcu now rather than once per device
8377 set_bit(RemoveSynchronized, &rdev->flags);
8383 rdev_for_each(rdev, mddev) {
8384 if ((this == NULL || rdev == this) &&
8385 rdev->raid_disk >= 0 &&
8386 !test_bit(Blocked, &rdev->flags) &&
8387 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8388 (!test_bit(In_sync, &rdev->flags) &&
8389 !test_bit(Journal, &rdev->flags))) &&
8390 atomic_read(&rdev->nr_pending)==0)) {
8391 if (mddev->pers->hot_remove_disk(
8392 mddev, rdev) == 0) {
8393 sysfs_unlink_rdev(mddev, rdev);
8394 rdev->raid_disk = -1;
8398 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8399 clear_bit(RemoveSynchronized, &rdev->flags);
8402 if (removed && mddev->kobj.sd)
8403 sysfs_notify(&mddev->kobj, NULL, "degraded");
8405 if (this && removed)
8408 rdev_for_each(rdev, mddev) {
8409 if (this && this != rdev)
8411 if (test_bit(Candidate, &rdev->flags))
8413 if (rdev->raid_disk >= 0 &&
8414 !test_bit(In_sync, &rdev->flags) &&
8415 !test_bit(Journal, &rdev->flags) &&
8416 !test_bit(Faulty, &rdev->flags))
8418 if (rdev->raid_disk >= 0)
8420 if (test_bit(Faulty, &rdev->flags))
8422 if (!test_bit(Journal, &rdev->flags)) {
8424 ! (rdev->saved_raid_disk >= 0 &&
8425 !test_bit(Bitmap_sync, &rdev->flags)))
8428 rdev->recovery_offset = 0;
8431 hot_add_disk(mddev, rdev) == 0) {
8432 if (sysfs_link_rdev(mddev, rdev))
8433 /* failure here is OK */;
8434 if (!test_bit(Journal, &rdev->flags))
8436 md_new_event(mddev);
8437 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8442 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8446 static void md_start_sync(struct work_struct *ws)
8448 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8450 mddev->sync_thread = md_register_thread(md_do_sync,
8453 if (!mddev->sync_thread) {
8454 pr_warn("%s: could not start resync thread...\n",
8456 /* leave the spares where they are, it shouldn't hurt */
8457 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8458 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8459 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8460 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8461 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8462 wake_up(&resync_wait);
8463 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8465 if (mddev->sysfs_action)
8466 sysfs_notify_dirent_safe(mddev->sysfs_action);
8468 md_wakeup_thread(mddev->sync_thread);
8469 sysfs_notify_dirent_safe(mddev->sysfs_action);
8470 md_new_event(mddev);
8474 * This routine is regularly called by all per-raid-array threads to
8475 * deal with generic issues like resync and super-block update.
8476 * Raid personalities that don't have a thread (linear/raid0) do not
8477 * need this as they never do any recovery or update the superblock.
8479 * It does not do any resync itself, but rather "forks" off other threads
8480 * to do that as needed.
8481 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8482 * "->recovery" and create a thread at ->sync_thread.
8483 * When the thread finishes it sets MD_RECOVERY_DONE
8484 * and wakeups up this thread which will reap the thread and finish up.
8485 * This thread also removes any faulty devices (with nr_pending == 0).
8487 * The overall approach is:
8488 * 1/ if the superblock needs updating, update it.
8489 * 2/ If a recovery thread is running, don't do anything else.
8490 * 3/ If recovery has finished, clean up, possibly marking spares active.
8491 * 4/ If there are any faulty devices, remove them.
8492 * 5/ If array is degraded, try to add spares devices
8493 * 6/ If array has spares or is not in-sync, start a resync thread.
8495 void md_check_recovery(struct mddev *mddev)
8497 if (mddev->suspended)
8501 bitmap_daemon_work(mddev);
8503 if (signal_pending(current)) {
8504 if (mddev->pers->sync_request && !mddev->external) {
8505 pr_debug("md: %s in immediate safe mode\n",
8507 mddev->safemode = 2;
8509 flush_signals(current);
8512 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8515 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8516 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8517 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8518 (mddev->external == 0 && mddev->safemode == 1) ||
8519 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8520 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8524 if (mddev_trylock(mddev)) {
8528 struct md_rdev *rdev;
8529 if (!mddev->external && mddev->in_sync)
8530 /* 'Blocked' flag not needed as failed devices
8531 * will be recorded if array switched to read/write.
8532 * Leaving it set will prevent the device
8533 * from being removed.
8535 rdev_for_each(rdev, mddev)
8536 clear_bit(Blocked, &rdev->flags);
8537 /* On a read-only array we can:
8538 * - remove failed devices
8539 * - add already-in_sync devices if the array itself
8541 * As we only add devices that are already in-sync,
8542 * we can activate the spares immediately.
8544 remove_and_add_spares(mddev, NULL);
8545 /* There is no thread, but we need to call
8546 * ->spare_active and clear saved_raid_disk
8548 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8549 md_reap_sync_thread(mddev);
8550 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8551 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8552 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8556 if (mddev_is_clustered(mddev)) {
8557 struct md_rdev *rdev;
8558 /* kick the device if another node issued a
8561 rdev_for_each(rdev, mddev) {
8562 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8563 rdev->raid_disk < 0)
8564 md_kick_rdev_from_array(rdev);
8568 if (!mddev->external) {
8570 spin_lock(&mddev->lock);
8571 if (mddev->safemode &&
8572 !atomic_read(&mddev->writes_pending) &&
8574 mddev->recovery_cp == MaxSector) {
8577 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8579 if (mddev->safemode == 1)
8580 mddev->safemode = 0;
8581 spin_unlock(&mddev->lock);
8583 sysfs_notify_dirent_safe(mddev->sysfs_state);
8586 if (mddev->sb_flags)
8587 md_update_sb(mddev, 0);
8589 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8590 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8591 /* resync/recovery still happening */
8592 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8595 if (mddev->sync_thread) {
8596 md_reap_sync_thread(mddev);
8599 /* Set RUNNING before clearing NEEDED to avoid
8600 * any transients in the value of "sync_action".
8602 mddev->curr_resync_completed = 0;
8603 spin_lock(&mddev->lock);
8604 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8605 spin_unlock(&mddev->lock);
8606 /* Clear some bits that don't mean anything, but
8609 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8610 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8612 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8613 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8615 /* no recovery is running.
8616 * remove any failed drives, then
8617 * add spares if possible.
8618 * Spares are also removed and re-added, to allow
8619 * the personality to fail the re-add.
8622 if (mddev->reshape_position != MaxSector) {
8623 if (mddev->pers->check_reshape == NULL ||
8624 mddev->pers->check_reshape(mddev) != 0)
8625 /* Cannot proceed */
8627 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8628 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8629 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8630 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8631 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8632 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8633 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8634 } else if (mddev->recovery_cp < MaxSector) {
8635 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8636 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8637 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8638 /* nothing to be done ... */
8641 if (mddev->pers->sync_request) {
8643 /* We are adding a device or devices to an array
8644 * which has the bitmap stored on all devices.
8645 * So make sure all bitmap pages get written
8647 bitmap_write_all(mddev->bitmap);
8649 INIT_WORK(&mddev->del_work, md_start_sync);
8650 queue_work(md_misc_wq, &mddev->del_work);
8654 if (!mddev->sync_thread) {
8655 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8656 wake_up(&resync_wait);
8657 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8659 if (mddev->sysfs_action)
8660 sysfs_notify_dirent_safe(mddev->sysfs_action);
8663 wake_up(&mddev->sb_wait);
8664 mddev_unlock(mddev);
8667 EXPORT_SYMBOL(md_check_recovery);
8669 void md_reap_sync_thread(struct mddev *mddev)
8671 struct md_rdev *rdev;
8673 /* resync has finished, collect result */
8674 md_unregister_thread(&mddev->sync_thread);
8675 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8676 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8678 /* activate any spares */
8679 if (mddev->pers->spare_active(mddev)) {
8680 sysfs_notify(&mddev->kobj, NULL,
8682 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8685 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8686 mddev->pers->finish_reshape)
8687 mddev->pers->finish_reshape(mddev);
8689 /* If array is no-longer degraded, then any saved_raid_disk
8690 * information must be scrapped.
8692 if (!mddev->degraded)
8693 rdev_for_each(rdev, mddev)
8694 rdev->saved_raid_disk = -1;
8696 md_update_sb(mddev, 1);
8697 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8698 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8700 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8701 md_cluster_ops->resync_finish(mddev);
8702 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8703 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8704 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8705 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8706 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8707 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8708 wake_up(&resync_wait);
8709 /* flag recovery needed just to double check */
8710 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8711 sysfs_notify_dirent_safe(mddev->sysfs_action);
8712 md_new_event(mddev);
8713 if (mddev->event_work.func)
8714 queue_work(md_misc_wq, &mddev->event_work);
8716 EXPORT_SYMBOL(md_reap_sync_thread);
8718 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8720 sysfs_notify_dirent_safe(rdev->sysfs_state);
8721 wait_event_timeout(rdev->blocked_wait,
8722 !test_bit(Blocked, &rdev->flags) &&
8723 !test_bit(BlockedBadBlocks, &rdev->flags),
8724 msecs_to_jiffies(5000));
8725 rdev_dec_pending(rdev, mddev);
8727 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8729 void md_finish_reshape(struct mddev *mddev)
8731 /* called be personality module when reshape completes. */
8732 struct md_rdev *rdev;
8734 rdev_for_each(rdev, mddev) {
8735 if (rdev->data_offset > rdev->new_data_offset)
8736 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8738 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8739 rdev->data_offset = rdev->new_data_offset;
8742 EXPORT_SYMBOL(md_finish_reshape);
8744 /* Bad block management */
8746 /* Returns 1 on success, 0 on failure */
8747 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8750 struct mddev *mddev = rdev->mddev;
8753 s += rdev->new_data_offset;
8755 s += rdev->data_offset;
8756 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8758 /* Make sure they get written out promptly */
8759 if (test_bit(ExternalBbl, &rdev->flags))
8760 sysfs_notify(&rdev->kobj, NULL,
8761 "unacknowledged_bad_blocks");
8762 sysfs_notify_dirent_safe(rdev->sysfs_state);
8763 set_mask_bits(&mddev->sb_flags, 0,
8764 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
8765 md_wakeup_thread(rdev->mddev->thread);
8770 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8772 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8777 s += rdev->new_data_offset;
8779 s += rdev->data_offset;
8780 rv = badblocks_clear(&rdev->badblocks, s, sectors);
8781 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
8782 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
8785 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8787 static int md_notify_reboot(struct notifier_block *this,
8788 unsigned long code, void *x)
8790 struct list_head *tmp;
8791 struct mddev *mddev;
8794 for_each_mddev(mddev, tmp) {
8795 if (mddev_trylock(mddev)) {
8797 __md_stop_writes(mddev);
8798 if (mddev->persistent)
8799 mddev->safemode = 2;
8800 mddev_unlock(mddev);
8805 * certain more exotic SCSI devices are known to be
8806 * volatile wrt too early system reboots. While the
8807 * right place to handle this issue is the given
8808 * driver, we do want to have a safe RAID driver ...
8816 static struct notifier_block md_notifier = {
8817 .notifier_call = md_notify_reboot,
8819 .priority = INT_MAX, /* before any real devices */
8822 static void md_geninit(void)
8824 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8826 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8829 static int __init md_init(void)
8833 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8837 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8841 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8844 if ((ret = register_blkdev(0, "mdp")) < 0)
8848 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8849 md_probe, NULL, NULL);
8850 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8851 md_probe, NULL, NULL);
8853 register_reboot_notifier(&md_notifier);
8854 raid_table_header = register_sysctl_table(raid_root_table);
8860 unregister_blkdev(MD_MAJOR, "md");
8862 destroy_workqueue(md_misc_wq);
8864 destroy_workqueue(md_wq);
8869 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8871 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8872 struct md_rdev *rdev2;
8874 char b[BDEVNAME_SIZE];
8877 * If size is changed in another node then we need to
8878 * do resize as well.
8880 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
8881 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
8883 pr_info("md-cluster: resize failed\n");
8885 bitmap_update_sb(mddev->bitmap);
8888 /* Check for change of roles in the active devices */
8889 rdev_for_each(rdev2, mddev) {
8890 if (test_bit(Faulty, &rdev2->flags))
8893 /* Check if the roles changed */
8894 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8896 if (test_bit(Candidate, &rdev2->flags)) {
8897 if (role == 0xfffe) {
8898 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8899 md_kick_rdev_from_array(rdev2);
8903 clear_bit(Candidate, &rdev2->flags);
8906 if (role != rdev2->raid_disk) {
8908 if (rdev2->raid_disk == -1 && role != 0xffff) {
8909 rdev2->saved_raid_disk = role;
8910 ret = remove_and_add_spares(mddev, rdev2);
8911 pr_info("Activated spare: %s\n",
8912 bdevname(rdev2->bdev,b));
8913 /* wakeup mddev->thread here, so array could
8914 * perform resync with the new activated disk */
8915 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8916 md_wakeup_thread(mddev->thread);
8920 * We just want to do the minimum to mark the disk
8921 * as faulty. The recovery is performed by the
8922 * one who initiated the error.
8924 if ((role == 0xfffe) || (role == 0xfffd)) {
8925 md_error(mddev, rdev2);
8926 clear_bit(Blocked, &rdev2->flags);
8931 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8932 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8934 /* Finally set the event to be up to date */
8935 mddev->events = le64_to_cpu(sb->events);
8938 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8941 struct page *swapout = rdev->sb_page;
8942 struct mdp_superblock_1 *sb;
8944 /* Store the sb page of the rdev in the swapout temporary
8945 * variable in case we err in the future
8947 rdev->sb_page = NULL;
8948 err = alloc_disk_sb(rdev);
8950 ClearPageUptodate(rdev->sb_page);
8951 rdev->sb_loaded = 0;
8952 err = super_types[mddev->major_version].
8953 load_super(rdev, NULL, mddev->minor_version);
8956 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8957 __func__, __LINE__, rdev->desc_nr, err);
8959 put_page(rdev->sb_page);
8960 rdev->sb_page = swapout;
8961 rdev->sb_loaded = 1;
8965 sb = page_address(rdev->sb_page);
8966 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8970 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8971 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8973 /* The other node finished recovery, call spare_active to set
8974 * device In_sync and mddev->degraded
8976 if (rdev->recovery_offset == MaxSector &&
8977 !test_bit(In_sync, &rdev->flags) &&
8978 mddev->pers->spare_active(mddev))
8979 sysfs_notify(&mddev->kobj, NULL, "degraded");
8985 void md_reload_sb(struct mddev *mddev, int nr)
8987 struct md_rdev *rdev;
8991 rdev_for_each_rcu(rdev, mddev) {
8992 if (rdev->desc_nr == nr)
8996 if (!rdev || rdev->desc_nr != nr) {
8997 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9001 err = read_rdev(mddev, rdev);
9005 check_sb_changes(mddev, rdev);
9007 /* Read all rdev's to update recovery_offset */
9008 rdev_for_each_rcu(rdev, mddev)
9009 read_rdev(mddev, rdev);
9011 EXPORT_SYMBOL(md_reload_sb);
9016 * Searches all registered partitions for autorun RAID arrays
9020 static DEFINE_MUTEX(detected_devices_mutex);
9021 static LIST_HEAD(all_detected_devices);
9022 struct detected_devices_node {
9023 struct list_head list;
9027 void md_autodetect_dev(dev_t dev)
9029 struct detected_devices_node *node_detected_dev;
9031 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9032 if (node_detected_dev) {
9033 node_detected_dev->dev = dev;
9034 mutex_lock(&detected_devices_mutex);
9035 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9036 mutex_unlock(&detected_devices_mutex);
9040 static void autostart_arrays(int part)
9042 struct md_rdev *rdev;
9043 struct detected_devices_node *node_detected_dev;
9045 int i_scanned, i_passed;
9050 pr_info("md: Autodetecting RAID arrays.\n");
9052 mutex_lock(&detected_devices_mutex);
9053 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9055 node_detected_dev = list_entry(all_detected_devices.next,
9056 struct detected_devices_node, list);
9057 list_del(&node_detected_dev->list);
9058 dev = node_detected_dev->dev;
9059 kfree(node_detected_dev);
9060 mutex_unlock(&detected_devices_mutex);
9061 rdev = md_import_device(dev,0, 90);
9062 mutex_lock(&detected_devices_mutex);
9066 if (test_bit(Faulty, &rdev->flags))
9069 set_bit(AutoDetected, &rdev->flags);
9070 list_add(&rdev->same_set, &pending_raid_disks);
9073 mutex_unlock(&detected_devices_mutex);
9075 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9077 autorun_devices(part);
9080 #endif /* !MODULE */
9082 static __exit void md_exit(void)
9084 struct mddev *mddev;
9085 struct list_head *tmp;
9088 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9089 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9091 unregister_blkdev(MD_MAJOR,"md");
9092 unregister_blkdev(mdp_major, "mdp");
9093 unregister_reboot_notifier(&md_notifier);
9094 unregister_sysctl_table(raid_table_header);
9096 /* We cannot unload the modules while some process is
9097 * waiting for us in select() or poll() - wake them up
9100 while (waitqueue_active(&md_event_waiters)) {
9101 /* not safe to leave yet */
9102 wake_up(&md_event_waiters);
9106 remove_proc_entry("mdstat", NULL);
9108 for_each_mddev(mddev, tmp) {
9109 export_array(mddev);
9111 mddev->hold_active = 0;
9113 * for_each_mddev() will call mddev_put() at the end of each
9114 * iteration. As the mddev is now fully clear, this will
9115 * schedule the mddev for destruction by a workqueue, and the
9116 * destroy_workqueue() below will wait for that to complete.
9119 destroy_workqueue(md_misc_wq);
9120 destroy_workqueue(md_wq);
9123 subsys_initcall(md_init);
9124 module_exit(md_exit)
9126 static int get_ro(char *buffer, struct kernel_param *kp)
9128 return sprintf(buffer, "%d", start_readonly);
9130 static int set_ro(const char *val, struct kernel_param *kp)
9132 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9135 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9136 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9137 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9139 MODULE_LICENSE("GPL");
9140 MODULE_DESCRIPTION("MD RAID framework");
9142 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);