2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 Errors, Warnings, etc.
36 pr_crit() for error conditions that risk data loss
37 pr_err() for error conditions that are unexpected, like an IO error
38 or internal inconsistency
39 pr_warn() for error conditions that could have been predicated, like
40 adding a device to an array when it has incompatible metadata
41 pr_info() for every interesting, very rare events, like an array starting
42 or stopping, or resync starting or stopping
43 pr_debug() for everything else.
47 #include <linux/sched/signal.h>
48 #include <linux/kthread.h>
49 #include <linux/blkdev.h>
50 #include <linux/badblocks.h>
51 #include <linux/sysctl.h>
52 #include <linux/seq_file.h>
54 #include <linux/poll.h>
55 #include <linux/ctype.h>
56 #include <linux/string.h>
57 #include <linux/hdreg.h>
58 #include <linux/proc_fs.h>
59 #include <linux/random.h>
60 #include <linux/module.h>
61 #include <linux/reboot.h>
62 #include <linux/file.h>
63 #include <linux/compat.h>
64 #include <linux/delay.h>
65 #include <linux/raid/md_p.h>
66 #include <linux/raid/md_u.h>
67 #include <linux/slab.h>
68 #include <linux/percpu-refcount.h>
70 #include <trace/events/block.h>
72 #include "md-bitmap.h"
73 #include "md-cluster.h"
76 static void autostart_arrays(int part);
79 /* pers_list is a list of registered personalities protected
81 * pers_lock does extra service to protect accesses to
82 * mddev->thread when the mutex cannot be held.
84 static LIST_HEAD(pers_list);
85 static DEFINE_SPINLOCK(pers_lock);
87 struct md_cluster_operations *md_cluster_ops;
88 EXPORT_SYMBOL(md_cluster_ops);
89 struct module *md_cluster_mod;
90 EXPORT_SYMBOL(md_cluster_mod);
92 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
93 static struct workqueue_struct *md_wq;
94 static struct workqueue_struct *md_misc_wq;
96 static int remove_and_add_spares(struct mddev *mddev,
97 struct md_rdev *this);
98 static void mddev_detach(struct mddev *mddev);
101 * Default number of read corrections we'll attempt on an rdev
102 * before ejecting it from the array. We divide the read error
103 * count by 2 for every hour elapsed between read errors.
105 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
107 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
108 * is 1000 KB/sec, so the extra system load does not show up that much.
109 * Increase it if you want to have more _guaranteed_ speed. Note that
110 * the RAID driver will use the maximum available bandwidth if the IO
111 * subsystem is idle. There is also an 'absolute maximum' reconstruction
112 * speed limit - in case reconstruction slows down your system despite
115 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
116 * or /sys/block/mdX/md/sync_speed_{min,max}
119 static int sysctl_speed_limit_min = 1000;
120 static int sysctl_speed_limit_max = 200000;
121 static inline int speed_min(struct mddev *mddev)
123 return mddev->sync_speed_min ?
124 mddev->sync_speed_min : sysctl_speed_limit_min;
127 static inline int speed_max(struct mddev *mddev)
129 return mddev->sync_speed_max ?
130 mddev->sync_speed_max : sysctl_speed_limit_max;
133 static struct ctl_table_header *raid_table_header;
135 static struct ctl_table raid_table[] = {
137 .procname = "speed_limit_min",
138 .data = &sysctl_speed_limit_min,
139 .maxlen = sizeof(int),
140 .mode = S_IRUGO|S_IWUSR,
141 .proc_handler = proc_dointvec,
144 .procname = "speed_limit_max",
145 .data = &sysctl_speed_limit_max,
146 .maxlen = sizeof(int),
147 .mode = S_IRUGO|S_IWUSR,
148 .proc_handler = proc_dointvec,
153 static struct ctl_table raid_dir_table[] = {
157 .mode = S_IRUGO|S_IXUGO,
163 static struct ctl_table raid_root_table[] = {
168 .child = raid_dir_table,
173 static const struct block_device_operations md_fops;
175 static int start_readonly;
178 * The original mechanism for creating an md device is to create
179 * a device node in /dev and to open it. This causes races with device-close.
180 * The preferred method is to write to the "new_array" module parameter.
181 * This can avoid races.
182 * Setting create_on_open to false disables the original mechanism
183 * so all the races disappear.
185 static bool create_on_open = true;
188 * like bio_clone_bioset, but with a local bio set
191 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
196 if (!mddev || !mddev->bio_set)
197 return bio_alloc(gfp_mask, nr_iovecs);
199 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
204 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
206 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
208 if (!mddev || !mddev->sync_set)
209 return bio_alloc(GFP_NOIO, 1);
211 return bio_alloc_bioset(GFP_NOIO, 1, mddev->sync_set);
215 * We have a system wide 'event count' that is incremented
216 * on any 'interesting' event, and readers of /proc/mdstat
217 * can use 'poll' or 'select' to find out when the event
221 * start array, stop array, error, add device, remove device,
222 * start build, activate spare
224 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
225 static atomic_t md_event_count;
226 void md_new_event(struct mddev *mddev)
228 atomic_inc(&md_event_count);
229 wake_up(&md_event_waiters);
231 EXPORT_SYMBOL_GPL(md_new_event);
234 * Enables to iterate over all existing md arrays
235 * all_mddevs_lock protects this list.
237 static LIST_HEAD(all_mddevs);
238 static DEFINE_SPINLOCK(all_mddevs_lock);
241 * iterates through all used mddevs in the system.
242 * We take care to grab the all_mddevs_lock whenever navigating
243 * the list, and to always hold a refcount when unlocked.
244 * Any code which breaks out of this loop while own
245 * a reference to the current mddev and must mddev_put it.
247 #define for_each_mddev(_mddev,_tmp) \
249 for (({ spin_lock(&all_mddevs_lock); \
250 _tmp = all_mddevs.next; \
252 ({ if (_tmp != &all_mddevs) \
253 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
254 spin_unlock(&all_mddevs_lock); \
255 if (_mddev) mddev_put(_mddev); \
256 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
257 _tmp != &all_mddevs;}); \
258 ({ spin_lock(&all_mddevs_lock); \
259 _tmp = _tmp->next;}) \
262 /* Rather than calling directly into the personality make_request function,
263 * IO requests come here first so that we can check if the device is
264 * being suspended pending a reconfiguration.
265 * We hold a refcount over the call to ->make_request. By the time that
266 * call has finished, the bio has been linked into some internal structure
267 * and so is visible to ->quiesce(), so we don't need the refcount any more.
269 static bool is_suspended(struct mddev *mddev, struct bio *bio)
271 if (mddev->suspended)
273 if (bio_data_dir(bio) != WRITE)
275 if (mddev->suspend_lo >= mddev->suspend_hi)
277 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
279 if (bio_end_sector(bio) < mddev->suspend_lo)
284 void md_handle_request(struct mddev *mddev, struct bio *bio)
288 if (is_suspended(mddev, bio)) {
291 prepare_to_wait(&mddev->sb_wait, &__wait,
292 TASK_UNINTERRUPTIBLE);
293 if (!is_suspended(mddev, bio))
299 finish_wait(&mddev->sb_wait, &__wait);
301 atomic_inc(&mddev->active_io);
304 if (!mddev->pers->make_request(mddev, bio)) {
305 atomic_dec(&mddev->active_io);
306 wake_up(&mddev->sb_wait);
307 goto check_suspended;
310 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
311 wake_up(&mddev->sb_wait);
313 EXPORT_SYMBOL(md_handle_request);
315 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
317 const int rw = bio_data_dir(bio);
318 struct mddev *mddev = q->queuedata;
319 unsigned int sectors;
322 blk_queue_split(q, &bio);
324 if (mddev == NULL || mddev->pers == NULL) {
326 return BLK_QC_T_NONE;
328 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
329 if (bio_sectors(bio) != 0)
330 bio->bi_status = BLK_STS_IOERR;
332 return BLK_QC_T_NONE;
336 * save the sectors now since our bio can
337 * go away inside make_request
339 sectors = bio_sectors(bio);
340 /* bio could be mergeable after passing to underlayer */
341 bio->bi_opf &= ~REQ_NOMERGE;
343 md_handle_request(mddev, bio);
345 cpu = part_stat_lock();
346 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
347 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
350 return BLK_QC_T_NONE;
353 /* mddev_suspend makes sure no new requests are submitted
354 * to the device, and that any requests that have been submitted
355 * are completely handled.
356 * Once mddev_detach() is called and completes, the module will be
359 void mddev_suspend(struct mddev *mddev)
361 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
362 lockdep_assert_held(&mddev->reconfig_mutex);
363 if (mddev->suspended++)
366 wake_up(&mddev->sb_wait);
367 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
368 smp_mb__after_atomic();
369 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
370 mddev->pers->quiesce(mddev, 1);
371 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
372 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
374 del_timer_sync(&mddev->safemode_timer);
376 EXPORT_SYMBOL_GPL(mddev_suspend);
378 void mddev_resume(struct mddev *mddev)
380 lockdep_assert_held(&mddev->reconfig_mutex);
381 if (--mddev->suspended)
383 wake_up(&mddev->sb_wait);
384 mddev->pers->quiesce(mddev, 0);
386 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
387 md_wakeup_thread(mddev->thread);
388 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
390 EXPORT_SYMBOL_GPL(mddev_resume);
392 int mddev_congested(struct mddev *mddev, int bits)
394 struct md_personality *pers = mddev->pers;
398 if (mddev->suspended)
400 else if (pers && pers->congested)
401 ret = pers->congested(mddev, bits);
405 EXPORT_SYMBOL_GPL(mddev_congested);
406 static int md_congested(void *data, int bits)
408 struct mddev *mddev = data;
409 return mddev_congested(mddev, bits);
413 * Generic flush handling for md
416 static void md_end_flush(struct bio *bio)
418 struct md_rdev *rdev = bio->bi_private;
419 struct mddev *mddev = rdev->mddev;
421 rdev_dec_pending(rdev, mddev);
423 if (atomic_dec_and_test(&mddev->flush_pending)) {
424 /* The pre-request flush has finished */
425 queue_work(md_wq, &mddev->flush_work);
430 static void md_submit_flush_data(struct work_struct *ws);
432 static void submit_flushes(struct work_struct *ws)
434 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
435 struct md_rdev *rdev;
437 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
438 atomic_set(&mddev->flush_pending, 1);
440 rdev_for_each_rcu(rdev, mddev)
441 if (rdev->raid_disk >= 0 &&
442 !test_bit(Faulty, &rdev->flags)) {
443 /* Take two references, one is dropped
444 * when request finishes, one after
445 * we reclaim rcu_read_lock
448 atomic_inc(&rdev->nr_pending);
449 atomic_inc(&rdev->nr_pending);
451 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
452 bi->bi_end_io = md_end_flush;
453 bi->bi_private = rdev;
454 bio_set_dev(bi, rdev->bdev);
455 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
456 atomic_inc(&mddev->flush_pending);
459 rdev_dec_pending(rdev, mddev);
462 if (atomic_dec_and_test(&mddev->flush_pending))
463 queue_work(md_wq, &mddev->flush_work);
466 static void md_submit_flush_data(struct work_struct *ws)
468 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
469 struct bio *bio = mddev->flush_bio;
472 * must reset flush_bio before calling into md_handle_request to avoid a
473 * deadlock, because other bios passed md_handle_request suspend check
474 * could wait for this and below md_handle_request could wait for those
475 * bios because of suspend check
477 mddev->flush_bio = NULL;
478 wake_up(&mddev->sb_wait);
480 if (bio->bi_iter.bi_size == 0)
481 /* an empty barrier - all done */
484 bio->bi_opf &= ~REQ_PREFLUSH;
485 md_handle_request(mddev, bio);
489 void md_flush_request(struct mddev *mddev, struct bio *bio)
491 spin_lock_irq(&mddev->lock);
492 wait_event_lock_irq(mddev->sb_wait,
495 mddev->flush_bio = bio;
496 spin_unlock_irq(&mddev->lock);
498 INIT_WORK(&mddev->flush_work, submit_flushes);
499 queue_work(md_wq, &mddev->flush_work);
501 EXPORT_SYMBOL(md_flush_request);
503 static inline struct mddev *mddev_get(struct mddev *mddev)
505 atomic_inc(&mddev->active);
509 static void mddev_delayed_delete(struct work_struct *ws);
511 static void mddev_put(struct mddev *mddev)
513 struct bio_set *bs = NULL, *sync_bs = NULL;
515 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
517 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
518 mddev->ctime == 0 && !mddev->hold_active) {
519 /* Array is not configured at all, and not held active,
521 list_del_init(&mddev->all_mddevs);
523 sync_bs = mddev->sync_set;
524 mddev->bio_set = NULL;
525 mddev->sync_set = NULL;
526 if (mddev->gendisk) {
527 /* We did a probe so need to clean up. Call
528 * queue_work inside the spinlock so that
529 * flush_workqueue() after mddev_find will
530 * succeed in waiting for the work to be done.
532 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
533 queue_work(md_misc_wq, &mddev->del_work);
537 spin_unlock(&all_mddevs_lock);
541 bioset_free(sync_bs);
544 static void md_safemode_timeout(struct timer_list *t);
546 void mddev_init(struct mddev *mddev)
548 mutex_init(&mddev->open_mutex);
549 mutex_init(&mddev->reconfig_mutex);
550 mutex_init(&mddev->bitmap_info.mutex);
551 INIT_LIST_HEAD(&mddev->disks);
552 INIT_LIST_HEAD(&mddev->all_mddevs);
553 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
554 atomic_set(&mddev->active, 1);
555 atomic_set(&mddev->openers, 0);
556 atomic_set(&mddev->active_io, 0);
557 spin_lock_init(&mddev->lock);
558 atomic_set(&mddev->flush_pending, 0);
559 init_waitqueue_head(&mddev->sb_wait);
560 init_waitqueue_head(&mddev->recovery_wait);
561 mddev->reshape_position = MaxSector;
562 mddev->reshape_backwards = 0;
563 mddev->last_sync_action = "none";
564 mddev->resync_min = 0;
565 mddev->resync_max = MaxSector;
566 mddev->level = LEVEL_NONE;
568 EXPORT_SYMBOL_GPL(mddev_init);
570 static struct mddev *mddev_find(dev_t unit)
572 struct mddev *mddev, *new = NULL;
574 if (unit && MAJOR(unit) != MD_MAJOR)
575 unit &= ~((1<<MdpMinorShift)-1);
578 spin_lock(&all_mddevs_lock);
581 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
582 if (mddev->unit == unit) {
584 spin_unlock(&all_mddevs_lock);
590 list_add(&new->all_mddevs, &all_mddevs);
591 spin_unlock(&all_mddevs_lock);
592 new->hold_active = UNTIL_IOCTL;
596 /* find an unused unit number */
597 static int next_minor = 512;
598 int start = next_minor;
602 dev = MKDEV(MD_MAJOR, next_minor);
604 if (next_minor > MINORMASK)
606 if (next_minor == start) {
607 /* Oh dear, all in use. */
608 spin_unlock(&all_mddevs_lock);
614 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
615 if (mddev->unit == dev) {
621 new->md_minor = MINOR(dev);
622 new->hold_active = UNTIL_STOP;
623 list_add(&new->all_mddevs, &all_mddevs);
624 spin_unlock(&all_mddevs_lock);
627 spin_unlock(&all_mddevs_lock);
629 new = kzalloc(sizeof(*new), GFP_KERNEL);
634 if (MAJOR(unit) == MD_MAJOR)
635 new->md_minor = MINOR(unit);
637 new->md_minor = MINOR(unit) >> MdpMinorShift;
644 static struct attribute_group md_redundancy_group;
646 void mddev_unlock(struct mddev *mddev)
648 if (mddev->to_remove) {
649 /* These cannot be removed under reconfig_mutex as
650 * an access to the files will try to take reconfig_mutex
651 * while holding the file unremovable, which leads to
653 * So hold set sysfs_active while the remove in happeing,
654 * and anything else which might set ->to_remove or my
655 * otherwise change the sysfs namespace will fail with
656 * -EBUSY if sysfs_active is still set.
657 * We set sysfs_active under reconfig_mutex and elsewhere
658 * test it under the same mutex to ensure its correct value
661 struct attribute_group *to_remove = mddev->to_remove;
662 mddev->to_remove = NULL;
663 mddev->sysfs_active = 1;
664 mutex_unlock(&mddev->reconfig_mutex);
666 if (mddev->kobj.sd) {
667 if (to_remove != &md_redundancy_group)
668 sysfs_remove_group(&mddev->kobj, to_remove);
669 if (mddev->pers == NULL ||
670 mddev->pers->sync_request == NULL) {
671 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
672 if (mddev->sysfs_action)
673 sysfs_put(mddev->sysfs_action);
674 mddev->sysfs_action = NULL;
677 mddev->sysfs_active = 0;
679 mutex_unlock(&mddev->reconfig_mutex);
681 /* As we've dropped the mutex we need a spinlock to
682 * make sure the thread doesn't disappear
684 spin_lock(&pers_lock);
685 md_wakeup_thread(mddev->thread);
686 wake_up(&mddev->sb_wait);
687 spin_unlock(&pers_lock);
689 EXPORT_SYMBOL_GPL(mddev_unlock);
691 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
693 struct md_rdev *rdev;
695 rdev_for_each_rcu(rdev, mddev)
696 if (rdev->desc_nr == nr)
701 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
703 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
705 struct md_rdev *rdev;
707 rdev_for_each(rdev, mddev)
708 if (rdev->bdev->bd_dev == dev)
714 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
716 struct md_rdev *rdev;
718 rdev_for_each_rcu(rdev, mddev)
719 if (rdev->bdev->bd_dev == dev)
725 static struct md_personality *find_pers(int level, char *clevel)
727 struct md_personality *pers;
728 list_for_each_entry(pers, &pers_list, list) {
729 if (level != LEVEL_NONE && pers->level == level)
731 if (strcmp(pers->name, clevel)==0)
737 /* return the offset of the super block in 512byte sectors */
738 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
740 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
741 return MD_NEW_SIZE_SECTORS(num_sectors);
744 static int alloc_disk_sb(struct md_rdev *rdev)
746 rdev->sb_page = alloc_page(GFP_KERNEL);
752 void md_rdev_clear(struct md_rdev *rdev)
755 put_page(rdev->sb_page);
757 rdev->sb_page = NULL;
762 put_page(rdev->bb_page);
763 rdev->bb_page = NULL;
765 badblocks_exit(&rdev->badblocks);
767 EXPORT_SYMBOL_GPL(md_rdev_clear);
769 static void super_written(struct bio *bio)
771 struct md_rdev *rdev = bio->bi_private;
772 struct mddev *mddev = rdev->mddev;
774 if (bio->bi_status) {
775 pr_err("md: super_written gets error=%d\n", bio->bi_status);
776 md_error(mddev, rdev);
777 if (!test_bit(Faulty, &rdev->flags)
778 && (bio->bi_opf & MD_FAILFAST)) {
779 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
780 set_bit(LastDev, &rdev->flags);
783 clear_bit(LastDev, &rdev->flags);
785 if (atomic_dec_and_test(&mddev->pending_writes))
786 wake_up(&mddev->sb_wait);
787 rdev_dec_pending(rdev, mddev);
791 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
792 sector_t sector, int size, struct page *page)
794 /* write first size bytes of page to sector of rdev
795 * Increment mddev->pending_writes before returning
796 * and decrement it on completion, waking up sb_wait
797 * if zero is reached.
798 * If an error occurred, call md_error
803 if (test_bit(Faulty, &rdev->flags))
806 bio = md_bio_alloc_sync(mddev);
808 atomic_inc(&rdev->nr_pending);
810 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
811 bio->bi_iter.bi_sector = sector;
812 bio_add_page(bio, page, size, 0);
813 bio->bi_private = rdev;
814 bio->bi_end_io = super_written;
816 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
817 test_bit(FailFast, &rdev->flags) &&
818 !test_bit(LastDev, &rdev->flags))
820 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
822 atomic_inc(&mddev->pending_writes);
826 int md_super_wait(struct mddev *mddev)
828 /* wait for all superblock writes that were scheduled to complete */
829 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
830 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
835 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
836 struct page *page, int op, int op_flags, bool metadata_op)
838 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
841 if (metadata_op && rdev->meta_bdev)
842 bio_set_dev(bio, rdev->meta_bdev);
844 bio_set_dev(bio, rdev->bdev);
845 bio_set_op_attrs(bio, op, op_flags);
847 bio->bi_iter.bi_sector = sector + rdev->sb_start;
848 else if (rdev->mddev->reshape_position != MaxSector &&
849 (rdev->mddev->reshape_backwards ==
850 (sector >= rdev->mddev->reshape_position)))
851 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
853 bio->bi_iter.bi_sector = sector + rdev->data_offset;
854 bio_add_page(bio, page, size, 0);
856 submit_bio_wait(bio);
858 ret = !bio->bi_status;
862 EXPORT_SYMBOL_GPL(sync_page_io);
864 static int read_disk_sb(struct md_rdev *rdev, int size)
866 char b[BDEVNAME_SIZE];
871 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
877 pr_err("md: disabled device %s, could not read superblock.\n",
878 bdevname(rdev->bdev,b));
882 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
884 return sb1->set_uuid0 == sb2->set_uuid0 &&
885 sb1->set_uuid1 == sb2->set_uuid1 &&
886 sb1->set_uuid2 == sb2->set_uuid2 &&
887 sb1->set_uuid3 == sb2->set_uuid3;
890 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
893 mdp_super_t *tmp1, *tmp2;
895 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
896 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
898 if (!tmp1 || !tmp2) {
907 * nr_disks is not constant
912 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
919 static u32 md_csum_fold(u32 csum)
921 csum = (csum & 0xffff) + (csum >> 16);
922 return (csum & 0xffff) + (csum >> 16);
925 static unsigned int calc_sb_csum(mdp_super_t *sb)
928 u32 *sb32 = (u32*)sb;
930 unsigned int disk_csum, csum;
932 disk_csum = sb->sb_csum;
935 for (i = 0; i < MD_SB_BYTES/4 ; i++)
937 csum = (newcsum & 0xffffffff) + (newcsum>>32);
940 /* This used to use csum_partial, which was wrong for several
941 * reasons including that different results are returned on
942 * different architectures. It isn't critical that we get exactly
943 * the same return value as before (we always csum_fold before
944 * testing, and that removes any differences). However as we
945 * know that csum_partial always returned a 16bit value on
946 * alphas, do a fold to maximise conformity to previous behaviour.
948 sb->sb_csum = md_csum_fold(disk_csum);
950 sb->sb_csum = disk_csum;
956 * Handle superblock details.
957 * We want to be able to handle multiple superblock formats
958 * so we have a common interface to them all, and an array of
959 * different handlers.
960 * We rely on user-space to write the initial superblock, and support
961 * reading and updating of superblocks.
962 * Interface methods are:
963 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
964 * loads and validates a superblock on dev.
965 * if refdev != NULL, compare superblocks on both devices
967 * 0 - dev has a superblock that is compatible with refdev
968 * 1 - dev has a superblock that is compatible and newer than refdev
969 * so dev should be used as the refdev in future
970 * -EINVAL superblock incompatible or invalid
971 * -othererror e.g. -EIO
973 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
974 * Verify that dev is acceptable into mddev.
975 * The first time, mddev->raid_disks will be 0, and data from
976 * dev should be merged in. Subsequent calls check that dev
977 * is new enough. Return 0 or -EINVAL
979 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
980 * Update the superblock for rdev with data in mddev
981 * This does not write to disc.
987 struct module *owner;
988 int (*load_super)(struct md_rdev *rdev,
989 struct md_rdev *refdev,
991 int (*validate_super)(struct mddev *mddev,
992 struct md_rdev *rdev);
993 void (*sync_super)(struct mddev *mddev,
994 struct md_rdev *rdev);
995 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
996 sector_t num_sectors);
997 int (*allow_new_offset)(struct md_rdev *rdev,
998 unsigned long long new_offset);
1002 * Check that the given mddev has no bitmap.
1004 * This function is called from the run method of all personalities that do not
1005 * support bitmaps. It prints an error message and returns non-zero if mddev
1006 * has a bitmap. Otherwise, it returns 0.
1009 int md_check_no_bitmap(struct mddev *mddev)
1011 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1013 pr_warn("%s: bitmaps are not supported for %s\n",
1014 mdname(mddev), mddev->pers->name);
1017 EXPORT_SYMBOL(md_check_no_bitmap);
1020 * load_super for 0.90.0
1022 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1024 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1029 * Calculate the position of the superblock (512byte sectors),
1030 * it's at the end of the disk.
1032 * It also happens to be a multiple of 4Kb.
1034 rdev->sb_start = calc_dev_sboffset(rdev);
1036 ret = read_disk_sb(rdev, MD_SB_BYTES);
1042 bdevname(rdev->bdev, b);
1043 sb = page_address(rdev->sb_page);
1045 if (sb->md_magic != MD_SB_MAGIC) {
1046 pr_warn("md: invalid raid superblock magic on %s\n", b);
1050 if (sb->major_version != 0 ||
1051 sb->minor_version < 90 ||
1052 sb->minor_version > 91) {
1053 pr_warn("Bad version number %d.%d on %s\n",
1054 sb->major_version, sb->minor_version, b);
1058 if (sb->raid_disks <= 0)
1061 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1062 pr_warn("md: invalid superblock checksum on %s\n", b);
1066 rdev->preferred_minor = sb->md_minor;
1067 rdev->data_offset = 0;
1068 rdev->new_data_offset = 0;
1069 rdev->sb_size = MD_SB_BYTES;
1070 rdev->badblocks.shift = -1;
1072 if (sb->level == LEVEL_MULTIPATH)
1075 rdev->desc_nr = sb->this_disk.number;
1081 mdp_super_t *refsb = page_address(refdev->sb_page);
1082 if (!md_uuid_equal(refsb, sb)) {
1083 pr_warn("md: %s has different UUID to %s\n",
1084 b, bdevname(refdev->bdev,b2));
1087 if (!md_sb_equal(refsb, sb)) {
1088 pr_warn("md: %s has same UUID but different superblock to %s\n",
1089 b, bdevname(refdev->bdev, b2));
1093 ev2 = md_event(refsb);
1099 rdev->sectors = rdev->sb_start;
1100 /* Limit to 4TB as metadata cannot record more than that.
1101 * (not needed for Linear and RAID0 as metadata doesn't
1104 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1106 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1108 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1109 /* "this cannot possibly happen" ... */
1117 * validate_super for 0.90.0
1119 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1122 mdp_super_t *sb = page_address(rdev->sb_page);
1123 __u64 ev1 = md_event(sb);
1125 rdev->raid_disk = -1;
1126 clear_bit(Faulty, &rdev->flags);
1127 clear_bit(In_sync, &rdev->flags);
1128 clear_bit(Bitmap_sync, &rdev->flags);
1129 clear_bit(WriteMostly, &rdev->flags);
1131 if (mddev->raid_disks == 0) {
1132 mddev->major_version = 0;
1133 mddev->minor_version = sb->minor_version;
1134 mddev->patch_version = sb->patch_version;
1135 mddev->external = 0;
1136 mddev->chunk_sectors = sb->chunk_size >> 9;
1137 mddev->ctime = sb->ctime;
1138 mddev->utime = sb->utime;
1139 mddev->level = sb->level;
1140 mddev->clevel[0] = 0;
1141 mddev->layout = sb->layout;
1142 mddev->raid_disks = sb->raid_disks;
1143 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1144 mddev->events = ev1;
1145 mddev->bitmap_info.offset = 0;
1146 mddev->bitmap_info.space = 0;
1147 /* bitmap can use 60 K after the 4K superblocks */
1148 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1149 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1150 mddev->reshape_backwards = 0;
1152 if (mddev->minor_version >= 91) {
1153 mddev->reshape_position = sb->reshape_position;
1154 mddev->delta_disks = sb->delta_disks;
1155 mddev->new_level = sb->new_level;
1156 mddev->new_layout = sb->new_layout;
1157 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1158 if (mddev->delta_disks < 0)
1159 mddev->reshape_backwards = 1;
1161 mddev->reshape_position = MaxSector;
1162 mddev->delta_disks = 0;
1163 mddev->new_level = mddev->level;
1164 mddev->new_layout = mddev->layout;
1165 mddev->new_chunk_sectors = mddev->chunk_sectors;
1168 if (sb->state & (1<<MD_SB_CLEAN))
1169 mddev->recovery_cp = MaxSector;
1171 if (sb->events_hi == sb->cp_events_hi &&
1172 sb->events_lo == sb->cp_events_lo) {
1173 mddev->recovery_cp = sb->recovery_cp;
1175 mddev->recovery_cp = 0;
1178 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1179 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1180 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1181 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1183 mddev->max_disks = MD_SB_DISKS;
1185 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1186 mddev->bitmap_info.file == NULL) {
1187 mddev->bitmap_info.offset =
1188 mddev->bitmap_info.default_offset;
1189 mddev->bitmap_info.space =
1190 mddev->bitmap_info.default_space;
1193 } else if (mddev->pers == NULL) {
1194 /* Insist on good event counter while assembling, except
1195 * for spares (which don't need an event count) */
1197 if (sb->disks[rdev->desc_nr].state & (
1198 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1199 if (ev1 < mddev->events)
1201 } else if (mddev->bitmap) {
1202 /* if adding to array with a bitmap, then we can accept an
1203 * older device ... but not too old.
1205 if (ev1 < mddev->bitmap->events_cleared)
1207 if (ev1 < mddev->events)
1208 set_bit(Bitmap_sync, &rdev->flags);
1210 if (ev1 < mddev->events)
1211 /* just a hot-add of a new device, leave raid_disk at -1 */
1215 if (mddev->level != LEVEL_MULTIPATH) {
1216 desc = sb->disks + rdev->desc_nr;
1218 if (desc->state & (1<<MD_DISK_FAULTY))
1219 set_bit(Faulty, &rdev->flags);
1220 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1221 desc->raid_disk < mddev->raid_disks */) {
1222 set_bit(In_sync, &rdev->flags);
1223 rdev->raid_disk = desc->raid_disk;
1224 rdev->saved_raid_disk = desc->raid_disk;
1225 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1226 /* active but not in sync implies recovery up to
1227 * reshape position. We don't know exactly where
1228 * that is, so set to zero for now */
1229 if (mddev->minor_version >= 91) {
1230 rdev->recovery_offset = 0;
1231 rdev->raid_disk = desc->raid_disk;
1234 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1235 set_bit(WriteMostly, &rdev->flags);
1236 if (desc->state & (1<<MD_DISK_FAILFAST))
1237 set_bit(FailFast, &rdev->flags);
1238 } else /* MULTIPATH are always insync */
1239 set_bit(In_sync, &rdev->flags);
1244 * sync_super for 0.90.0
1246 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1249 struct md_rdev *rdev2;
1250 int next_spare = mddev->raid_disks;
1252 /* make rdev->sb match mddev data..
1255 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1256 * 3/ any empty disks < next_spare become removed
1258 * disks[0] gets initialised to REMOVED because
1259 * we cannot be sure from other fields if it has
1260 * been initialised or not.
1263 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1265 rdev->sb_size = MD_SB_BYTES;
1267 sb = page_address(rdev->sb_page);
1269 memset(sb, 0, sizeof(*sb));
1271 sb->md_magic = MD_SB_MAGIC;
1272 sb->major_version = mddev->major_version;
1273 sb->patch_version = mddev->patch_version;
1274 sb->gvalid_words = 0; /* ignored */
1275 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1276 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1277 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1278 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1280 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1281 sb->level = mddev->level;
1282 sb->size = mddev->dev_sectors / 2;
1283 sb->raid_disks = mddev->raid_disks;
1284 sb->md_minor = mddev->md_minor;
1285 sb->not_persistent = 0;
1286 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1288 sb->events_hi = (mddev->events>>32);
1289 sb->events_lo = (u32)mddev->events;
1291 if (mddev->reshape_position == MaxSector)
1292 sb->minor_version = 90;
1294 sb->minor_version = 91;
1295 sb->reshape_position = mddev->reshape_position;
1296 sb->new_level = mddev->new_level;
1297 sb->delta_disks = mddev->delta_disks;
1298 sb->new_layout = mddev->new_layout;
1299 sb->new_chunk = mddev->new_chunk_sectors << 9;
1301 mddev->minor_version = sb->minor_version;
1304 sb->recovery_cp = mddev->recovery_cp;
1305 sb->cp_events_hi = (mddev->events>>32);
1306 sb->cp_events_lo = (u32)mddev->events;
1307 if (mddev->recovery_cp == MaxSector)
1308 sb->state = (1<< MD_SB_CLEAN);
1310 sb->recovery_cp = 0;
1312 sb->layout = mddev->layout;
1313 sb->chunk_size = mddev->chunk_sectors << 9;
1315 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1316 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1318 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1319 rdev_for_each(rdev2, mddev) {
1322 int is_active = test_bit(In_sync, &rdev2->flags);
1324 if (rdev2->raid_disk >= 0 &&
1325 sb->minor_version >= 91)
1326 /* we have nowhere to store the recovery_offset,
1327 * but if it is not below the reshape_position,
1328 * we can piggy-back on that.
1331 if (rdev2->raid_disk < 0 ||
1332 test_bit(Faulty, &rdev2->flags))
1335 desc_nr = rdev2->raid_disk;
1337 desc_nr = next_spare++;
1338 rdev2->desc_nr = desc_nr;
1339 d = &sb->disks[rdev2->desc_nr];
1341 d->number = rdev2->desc_nr;
1342 d->major = MAJOR(rdev2->bdev->bd_dev);
1343 d->minor = MINOR(rdev2->bdev->bd_dev);
1345 d->raid_disk = rdev2->raid_disk;
1347 d->raid_disk = rdev2->desc_nr; /* compatibility */
1348 if (test_bit(Faulty, &rdev2->flags))
1349 d->state = (1<<MD_DISK_FAULTY);
1350 else if (is_active) {
1351 d->state = (1<<MD_DISK_ACTIVE);
1352 if (test_bit(In_sync, &rdev2->flags))
1353 d->state |= (1<<MD_DISK_SYNC);
1361 if (test_bit(WriteMostly, &rdev2->flags))
1362 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1363 if (test_bit(FailFast, &rdev2->flags))
1364 d->state |= (1<<MD_DISK_FAILFAST);
1366 /* now set the "removed" and "faulty" bits on any missing devices */
1367 for (i=0 ; i < mddev->raid_disks ; i++) {
1368 mdp_disk_t *d = &sb->disks[i];
1369 if (d->state == 0 && d->number == 0) {
1372 d->state = (1<<MD_DISK_REMOVED);
1373 d->state |= (1<<MD_DISK_FAULTY);
1377 sb->nr_disks = nr_disks;
1378 sb->active_disks = active;
1379 sb->working_disks = working;
1380 sb->failed_disks = failed;
1381 sb->spare_disks = spare;
1383 sb->this_disk = sb->disks[rdev->desc_nr];
1384 sb->sb_csum = calc_sb_csum(sb);
1388 * rdev_size_change for 0.90.0
1390 static unsigned long long
1391 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1393 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1394 return 0; /* component must fit device */
1395 if (rdev->mddev->bitmap_info.offset)
1396 return 0; /* can't move bitmap */
1397 rdev->sb_start = calc_dev_sboffset(rdev);
1398 if (!num_sectors || num_sectors > rdev->sb_start)
1399 num_sectors = rdev->sb_start;
1400 /* Limit to 4TB as metadata cannot record more than that.
1401 * 4TB == 2^32 KB, or 2*2^32 sectors.
1403 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1404 rdev->mddev->level >= 1)
1405 num_sectors = (sector_t)(2ULL << 32) - 2;
1407 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1409 } while (md_super_wait(rdev->mddev) < 0);
1414 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1416 /* non-zero offset changes not possible with v0.90 */
1417 return new_offset == 0;
1421 * version 1 superblock
1424 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1428 unsigned long long newcsum;
1429 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1430 __le32 *isuper = (__le32*)sb;
1432 disk_csum = sb->sb_csum;
1435 for (; size >= 4; size -= 4)
1436 newcsum += le32_to_cpu(*isuper++);
1439 newcsum += le16_to_cpu(*(__le16*) isuper);
1441 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1442 sb->sb_csum = disk_csum;
1443 return cpu_to_le32(csum);
1446 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1448 struct mdp_superblock_1 *sb;
1452 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1456 * Calculate the position of the superblock in 512byte sectors.
1457 * It is always aligned to a 4K boundary and
1458 * depeding on minor_version, it can be:
1459 * 0: At least 8K, but less than 12K, from end of device
1460 * 1: At start of device
1461 * 2: 4K from start of device.
1463 switch(minor_version) {
1465 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1467 sb_start &= ~(sector_t)(4*2-1);
1478 rdev->sb_start = sb_start;
1480 /* superblock is rarely larger than 1K, but it can be larger,
1481 * and it is safe to read 4k, so we do that
1483 ret = read_disk_sb(rdev, 4096);
1484 if (ret) return ret;
1486 sb = page_address(rdev->sb_page);
1488 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1489 sb->major_version != cpu_to_le32(1) ||
1490 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1491 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1492 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1495 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1496 pr_warn("md: invalid superblock checksum on %s\n",
1497 bdevname(rdev->bdev,b));
1500 if (le64_to_cpu(sb->data_size) < 10) {
1501 pr_warn("md: data_size too small on %s\n",
1502 bdevname(rdev->bdev,b));
1507 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1508 /* Some padding is non-zero, might be a new feature */
1511 rdev->preferred_minor = 0xffff;
1512 rdev->data_offset = le64_to_cpu(sb->data_offset);
1513 rdev->new_data_offset = rdev->data_offset;
1514 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1515 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1516 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1517 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1519 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1520 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1521 if (rdev->sb_size & bmask)
1522 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1525 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1528 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1531 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1534 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1536 if (!rdev->bb_page) {
1537 rdev->bb_page = alloc_page(GFP_KERNEL);
1541 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1542 rdev->badblocks.count == 0) {
1543 /* need to load the bad block list.
1544 * Currently we limit it to one page.
1550 int sectors = le16_to_cpu(sb->bblog_size);
1551 if (sectors > (PAGE_SIZE / 512))
1553 offset = le32_to_cpu(sb->bblog_offset);
1556 bb_sector = (long long)offset;
1557 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1558 rdev->bb_page, REQ_OP_READ, 0, true))
1560 bbp = (u64 *)page_address(rdev->bb_page);
1561 rdev->badblocks.shift = sb->bblog_shift;
1562 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1563 u64 bb = le64_to_cpu(*bbp);
1564 int count = bb & (0x3ff);
1565 u64 sector = bb >> 10;
1566 sector <<= sb->bblog_shift;
1567 count <<= sb->bblog_shift;
1570 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1573 } else if (sb->bblog_offset != 0)
1574 rdev->badblocks.shift = 0;
1576 if ((le32_to_cpu(sb->feature_map) &
1577 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1578 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1579 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1580 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1587 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1589 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1590 sb->level != refsb->level ||
1591 sb->layout != refsb->layout ||
1592 sb->chunksize != refsb->chunksize) {
1593 pr_warn("md: %s has strangely different superblock to %s\n",
1594 bdevname(rdev->bdev,b),
1595 bdevname(refdev->bdev,b2));
1598 ev1 = le64_to_cpu(sb->events);
1599 ev2 = le64_to_cpu(refsb->events);
1606 if (minor_version) {
1607 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1608 sectors -= rdev->data_offset;
1610 sectors = rdev->sb_start;
1611 if (sectors < le64_to_cpu(sb->data_size))
1613 rdev->sectors = le64_to_cpu(sb->data_size);
1617 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1619 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1620 __u64 ev1 = le64_to_cpu(sb->events);
1622 rdev->raid_disk = -1;
1623 clear_bit(Faulty, &rdev->flags);
1624 clear_bit(In_sync, &rdev->flags);
1625 clear_bit(Bitmap_sync, &rdev->flags);
1626 clear_bit(WriteMostly, &rdev->flags);
1628 if (mddev->raid_disks == 0) {
1629 mddev->major_version = 1;
1630 mddev->patch_version = 0;
1631 mddev->external = 0;
1632 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1633 mddev->ctime = le64_to_cpu(sb->ctime);
1634 mddev->utime = le64_to_cpu(sb->utime);
1635 mddev->level = le32_to_cpu(sb->level);
1636 mddev->clevel[0] = 0;
1637 mddev->layout = le32_to_cpu(sb->layout);
1638 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1639 mddev->dev_sectors = le64_to_cpu(sb->size);
1640 mddev->events = ev1;
1641 mddev->bitmap_info.offset = 0;
1642 mddev->bitmap_info.space = 0;
1643 /* Default location for bitmap is 1K after superblock
1644 * using 3K - total of 4K
1646 mddev->bitmap_info.default_offset = 1024 >> 9;
1647 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1648 mddev->reshape_backwards = 0;
1650 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1651 memcpy(mddev->uuid, sb->set_uuid, 16);
1653 mddev->max_disks = (4096-256)/2;
1655 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1656 mddev->bitmap_info.file == NULL) {
1657 mddev->bitmap_info.offset =
1658 (__s32)le32_to_cpu(sb->bitmap_offset);
1659 /* Metadata doesn't record how much space is available.
1660 * For 1.0, we assume we can use up to the superblock
1661 * if before, else to 4K beyond superblock.
1662 * For others, assume no change is possible.
1664 if (mddev->minor_version > 0)
1665 mddev->bitmap_info.space = 0;
1666 else if (mddev->bitmap_info.offset > 0)
1667 mddev->bitmap_info.space =
1668 8 - mddev->bitmap_info.offset;
1670 mddev->bitmap_info.space =
1671 -mddev->bitmap_info.offset;
1674 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1675 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1676 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1677 mddev->new_level = le32_to_cpu(sb->new_level);
1678 mddev->new_layout = le32_to_cpu(sb->new_layout);
1679 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1680 if (mddev->delta_disks < 0 ||
1681 (mddev->delta_disks == 0 &&
1682 (le32_to_cpu(sb->feature_map)
1683 & MD_FEATURE_RESHAPE_BACKWARDS)))
1684 mddev->reshape_backwards = 1;
1686 mddev->reshape_position = MaxSector;
1687 mddev->delta_disks = 0;
1688 mddev->new_level = mddev->level;
1689 mddev->new_layout = mddev->layout;
1690 mddev->new_chunk_sectors = mddev->chunk_sectors;
1693 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1694 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1696 if (le32_to_cpu(sb->feature_map) &
1697 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1698 if (le32_to_cpu(sb->feature_map) &
1699 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1701 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1702 (le32_to_cpu(sb->feature_map) &
1703 MD_FEATURE_MULTIPLE_PPLS))
1705 set_bit(MD_HAS_PPL, &mddev->flags);
1707 } else if (mddev->pers == NULL) {
1708 /* Insist of good event counter while assembling, except for
1709 * spares (which don't need an event count) */
1711 if (rdev->desc_nr >= 0 &&
1712 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1713 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1714 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1715 if (ev1 < mddev->events)
1717 } else if (mddev->bitmap) {
1718 /* If adding to array with a bitmap, then we can accept an
1719 * older device, but not too old.
1721 if (ev1 < mddev->bitmap->events_cleared)
1723 if (ev1 < mddev->events)
1724 set_bit(Bitmap_sync, &rdev->flags);
1726 if (ev1 < mddev->events)
1727 /* just a hot-add of a new device, leave raid_disk at -1 */
1730 if (mddev->level != LEVEL_MULTIPATH) {
1732 if (rdev->desc_nr < 0 ||
1733 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1734 role = MD_DISK_ROLE_SPARE;
1737 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1739 case MD_DISK_ROLE_SPARE: /* spare */
1741 case MD_DISK_ROLE_FAULTY: /* faulty */
1742 set_bit(Faulty, &rdev->flags);
1744 case MD_DISK_ROLE_JOURNAL: /* journal device */
1745 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1746 /* journal device without journal feature */
1747 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1750 set_bit(Journal, &rdev->flags);
1751 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1752 rdev->raid_disk = 0;
1755 rdev->saved_raid_disk = role;
1756 if ((le32_to_cpu(sb->feature_map) &
1757 MD_FEATURE_RECOVERY_OFFSET)) {
1758 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1759 if (!(le32_to_cpu(sb->feature_map) &
1760 MD_FEATURE_RECOVERY_BITMAP))
1761 rdev->saved_raid_disk = -1;
1763 set_bit(In_sync, &rdev->flags);
1764 rdev->raid_disk = role;
1767 if (sb->devflags & WriteMostly1)
1768 set_bit(WriteMostly, &rdev->flags);
1769 if (sb->devflags & FailFast1)
1770 set_bit(FailFast, &rdev->flags);
1771 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1772 set_bit(Replacement, &rdev->flags);
1773 } else /* MULTIPATH are always insync */
1774 set_bit(In_sync, &rdev->flags);
1779 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1781 struct mdp_superblock_1 *sb;
1782 struct md_rdev *rdev2;
1784 /* make rdev->sb match mddev and rdev data. */
1786 sb = page_address(rdev->sb_page);
1788 sb->feature_map = 0;
1790 sb->recovery_offset = cpu_to_le64(0);
1791 memset(sb->pad3, 0, sizeof(sb->pad3));
1793 sb->utime = cpu_to_le64((__u64)mddev->utime);
1794 sb->events = cpu_to_le64(mddev->events);
1796 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1797 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1798 sb->resync_offset = cpu_to_le64(MaxSector);
1800 sb->resync_offset = cpu_to_le64(0);
1802 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1804 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1805 sb->size = cpu_to_le64(mddev->dev_sectors);
1806 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1807 sb->level = cpu_to_le32(mddev->level);
1808 sb->layout = cpu_to_le32(mddev->layout);
1809 if (test_bit(FailFast, &rdev->flags))
1810 sb->devflags |= FailFast1;
1812 sb->devflags &= ~FailFast1;
1814 if (test_bit(WriteMostly, &rdev->flags))
1815 sb->devflags |= WriteMostly1;
1817 sb->devflags &= ~WriteMostly1;
1818 sb->data_offset = cpu_to_le64(rdev->data_offset);
1819 sb->data_size = cpu_to_le64(rdev->sectors);
1821 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1822 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1823 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1826 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1827 !test_bit(In_sync, &rdev->flags)) {
1829 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1830 sb->recovery_offset =
1831 cpu_to_le64(rdev->recovery_offset);
1832 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1834 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1836 /* Note: recovery_offset and journal_tail share space */
1837 if (test_bit(Journal, &rdev->flags))
1838 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1839 if (test_bit(Replacement, &rdev->flags))
1841 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1843 if (mddev->reshape_position != MaxSector) {
1844 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1845 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1846 sb->new_layout = cpu_to_le32(mddev->new_layout);
1847 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1848 sb->new_level = cpu_to_le32(mddev->new_level);
1849 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1850 if (mddev->delta_disks == 0 &&
1851 mddev->reshape_backwards)
1853 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1854 if (rdev->new_data_offset != rdev->data_offset) {
1856 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1857 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1858 - rdev->data_offset));
1862 if (mddev_is_clustered(mddev))
1863 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1865 if (rdev->badblocks.count == 0)
1866 /* Nothing to do for bad blocks*/ ;
1867 else if (sb->bblog_offset == 0)
1868 /* Cannot record bad blocks on this device */
1869 md_error(mddev, rdev);
1871 struct badblocks *bb = &rdev->badblocks;
1872 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1874 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1879 seq = read_seqbegin(&bb->lock);
1881 memset(bbp, 0xff, PAGE_SIZE);
1883 for (i = 0 ; i < bb->count ; i++) {
1884 u64 internal_bb = p[i];
1885 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1886 | BB_LEN(internal_bb));
1887 bbp[i] = cpu_to_le64(store_bb);
1890 if (read_seqretry(&bb->lock, seq))
1893 bb->sector = (rdev->sb_start +
1894 (int)le32_to_cpu(sb->bblog_offset));
1895 bb->size = le16_to_cpu(sb->bblog_size);
1900 rdev_for_each(rdev2, mddev)
1901 if (rdev2->desc_nr+1 > max_dev)
1902 max_dev = rdev2->desc_nr+1;
1904 if (max_dev > le32_to_cpu(sb->max_dev)) {
1906 sb->max_dev = cpu_to_le32(max_dev);
1907 rdev->sb_size = max_dev * 2 + 256;
1908 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1909 if (rdev->sb_size & bmask)
1910 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1912 max_dev = le32_to_cpu(sb->max_dev);
1914 for (i=0; i<max_dev;i++)
1915 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1917 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1918 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1920 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
1921 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
1923 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
1925 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
1926 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
1927 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
1930 rdev_for_each(rdev2, mddev) {
1932 if (test_bit(Faulty, &rdev2->flags))
1933 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1934 else if (test_bit(In_sync, &rdev2->flags))
1935 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1936 else if (test_bit(Journal, &rdev2->flags))
1937 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1938 else if (rdev2->raid_disk >= 0)
1939 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1941 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1944 sb->sb_csum = calc_sb_1_csum(sb);
1947 static unsigned long long
1948 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1950 struct mdp_superblock_1 *sb;
1951 sector_t max_sectors;
1952 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1953 return 0; /* component must fit device */
1954 if (rdev->data_offset != rdev->new_data_offset)
1955 return 0; /* too confusing */
1956 if (rdev->sb_start < rdev->data_offset) {
1957 /* minor versions 1 and 2; superblock before data */
1958 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1959 max_sectors -= rdev->data_offset;
1960 if (!num_sectors || num_sectors > max_sectors)
1961 num_sectors = max_sectors;
1962 } else if (rdev->mddev->bitmap_info.offset) {
1963 /* minor version 0 with bitmap we can't move */
1966 /* minor version 0; superblock after data */
1968 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1969 sb_start &= ~(sector_t)(4*2 - 1);
1970 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1971 if (!num_sectors || num_sectors > max_sectors)
1972 num_sectors = max_sectors;
1973 rdev->sb_start = sb_start;
1975 sb = page_address(rdev->sb_page);
1976 sb->data_size = cpu_to_le64(num_sectors);
1977 sb->super_offset = cpu_to_le64(rdev->sb_start);
1978 sb->sb_csum = calc_sb_1_csum(sb);
1980 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1982 } while (md_super_wait(rdev->mddev) < 0);
1988 super_1_allow_new_offset(struct md_rdev *rdev,
1989 unsigned long long new_offset)
1991 /* All necessary checks on new >= old have been done */
1992 struct bitmap *bitmap;
1993 if (new_offset >= rdev->data_offset)
1996 /* with 1.0 metadata, there is no metadata to tread on
1997 * so we can always move back */
1998 if (rdev->mddev->minor_version == 0)
2001 /* otherwise we must be sure not to step on
2002 * any metadata, so stay:
2003 * 36K beyond start of superblock
2004 * beyond end of badblocks
2005 * beyond write-intent bitmap
2007 if (rdev->sb_start + (32+4)*2 > new_offset)
2009 bitmap = rdev->mddev->bitmap;
2010 if (bitmap && !rdev->mddev->bitmap_info.file &&
2011 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2012 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2014 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2020 static struct super_type super_types[] = {
2023 .owner = THIS_MODULE,
2024 .load_super = super_90_load,
2025 .validate_super = super_90_validate,
2026 .sync_super = super_90_sync,
2027 .rdev_size_change = super_90_rdev_size_change,
2028 .allow_new_offset = super_90_allow_new_offset,
2032 .owner = THIS_MODULE,
2033 .load_super = super_1_load,
2034 .validate_super = super_1_validate,
2035 .sync_super = super_1_sync,
2036 .rdev_size_change = super_1_rdev_size_change,
2037 .allow_new_offset = super_1_allow_new_offset,
2041 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2043 if (mddev->sync_super) {
2044 mddev->sync_super(mddev, rdev);
2048 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2050 super_types[mddev->major_version].sync_super(mddev, rdev);
2053 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2055 struct md_rdev *rdev, *rdev2;
2058 rdev_for_each_rcu(rdev, mddev1) {
2059 if (test_bit(Faulty, &rdev->flags) ||
2060 test_bit(Journal, &rdev->flags) ||
2061 rdev->raid_disk == -1)
2063 rdev_for_each_rcu(rdev2, mddev2) {
2064 if (test_bit(Faulty, &rdev2->flags) ||
2065 test_bit(Journal, &rdev2->flags) ||
2066 rdev2->raid_disk == -1)
2068 if (rdev->bdev->bd_contains ==
2069 rdev2->bdev->bd_contains) {
2079 static LIST_HEAD(pending_raid_disks);
2082 * Try to register data integrity profile for an mddev
2084 * This is called when an array is started and after a disk has been kicked
2085 * from the array. It only succeeds if all working and active component devices
2086 * are integrity capable with matching profiles.
2088 int md_integrity_register(struct mddev *mddev)
2090 struct md_rdev *rdev, *reference = NULL;
2092 if (list_empty(&mddev->disks))
2093 return 0; /* nothing to do */
2094 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2095 return 0; /* shouldn't register, or already is */
2096 rdev_for_each(rdev, mddev) {
2097 /* skip spares and non-functional disks */
2098 if (test_bit(Faulty, &rdev->flags))
2100 if (rdev->raid_disk < 0)
2103 /* Use the first rdev as the reference */
2107 /* does this rdev's profile match the reference profile? */
2108 if (blk_integrity_compare(reference->bdev->bd_disk,
2109 rdev->bdev->bd_disk) < 0)
2112 if (!reference || !bdev_get_integrity(reference->bdev))
2115 * All component devices are integrity capable and have matching
2116 * profiles, register the common profile for the md device.
2118 blk_integrity_register(mddev->gendisk,
2119 bdev_get_integrity(reference->bdev));
2121 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2122 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2123 pr_err("md: failed to create integrity pool for %s\n",
2129 EXPORT_SYMBOL(md_integrity_register);
2132 * Attempt to add an rdev, but only if it is consistent with the current
2135 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2137 struct blk_integrity *bi_rdev;
2138 struct blk_integrity *bi_mddev;
2139 char name[BDEVNAME_SIZE];
2141 if (!mddev->gendisk)
2144 bi_rdev = bdev_get_integrity(rdev->bdev);
2145 bi_mddev = blk_get_integrity(mddev->gendisk);
2147 if (!bi_mddev) /* nothing to do */
2150 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2151 pr_err("%s: incompatible integrity profile for %s\n",
2152 mdname(mddev), bdevname(rdev->bdev, name));
2158 EXPORT_SYMBOL(md_integrity_add_rdev);
2160 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2162 char b[BDEVNAME_SIZE];
2166 /* prevent duplicates */
2167 if (find_rdev(mddev, rdev->bdev->bd_dev))
2170 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2174 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2175 if (!test_bit(Journal, &rdev->flags) &&
2177 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2179 /* Cannot change size, so fail
2180 * If mddev->level <= 0, then we don't care
2181 * about aligning sizes (e.g. linear)
2183 if (mddev->level > 0)
2186 mddev->dev_sectors = rdev->sectors;
2189 /* Verify rdev->desc_nr is unique.
2190 * If it is -1, assign a free number, else
2191 * check number is not in use
2194 if (rdev->desc_nr < 0) {
2197 choice = mddev->raid_disks;
2198 while (md_find_rdev_nr_rcu(mddev, choice))
2200 rdev->desc_nr = choice;
2202 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2208 if (!test_bit(Journal, &rdev->flags) &&
2209 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2210 pr_warn("md: %s: array is limited to %d devices\n",
2211 mdname(mddev), mddev->max_disks);
2214 bdevname(rdev->bdev,b);
2215 strreplace(b, '/', '!');
2217 rdev->mddev = mddev;
2218 pr_debug("md: bind<%s>\n", b);
2220 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2223 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2224 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2225 /* failure here is OK */;
2226 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2228 list_add_rcu(&rdev->same_set, &mddev->disks);
2229 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2231 /* May as well allow recovery to be retried once */
2232 mddev->recovery_disabled++;
2237 pr_warn("md: failed to register dev-%s for %s\n",
2242 static void md_delayed_delete(struct work_struct *ws)
2244 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2245 kobject_del(&rdev->kobj);
2246 kobject_put(&rdev->kobj);
2249 static void unbind_rdev_from_array(struct md_rdev *rdev)
2251 char b[BDEVNAME_SIZE];
2253 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2254 list_del_rcu(&rdev->same_set);
2255 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2257 sysfs_remove_link(&rdev->kobj, "block");
2258 sysfs_put(rdev->sysfs_state);
2259 rdev->sysfs_state = NULL;
2260 rdev->badblocks.count = 0;
2261 /* We need to delay this, otherwise we can deadlock when
2262 * writing to 'remove' to "dev/state". We also need
2263 * to delay it due to rcu usage.
2266 INIT_WORK(&rdev->del_work, md_delayed_delete);
2267 kobject_get(&rdev->kobj);
2268 queue_work(md_misc_wq, &rdev->del_work);
2272 * prevent the device from being mounted, repartitioned or
2273 * otherwise reused by a RAID array (or any other kernel
2274 * subsystem), by bd_claiming the device.
2276 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2279 struct block_device *bdev;
2280 char b[BDEVNAME_SIZE];
2282 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2283 shared ? (struct md_rdev *)lock_rdev : rdev);
2285 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2286 return PTR_ERR(bdev);
2292 static void unlock_rdev(struct md_rdev *rdev)
2294 struct block_device *bdev = rdev->bdev;
2296 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2299 void md_autodetect_dev(dev_t dev);
2301 static void export_rdev(struct md_rdev *rdev)
2303 char b[BDEVNAME_SIZE];
2305 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2306 md_rdev_clear(rdev);
2308 if (test_bit(AutoDetected, &rdev->flags))
2309 md_autodetect_dev(rdev->bdev->bd_dev);
2312 kobject_put(&rdev->kobj);
2315 void md_kick_rdev_from_array(struct md_rdev *rdev)
2317 unbind_rdev_from_array(rdev);
2320 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2322 static void export_array(struct mddev *mddev)
2324 struct md_rdev *rdev;
2326 while (!list_empty(&mddev->disks)) {
2327 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2329 md_kick_rdev_from_array(rdev);
2331 mddev->raid_disks = 0;
2332 mddev->major_version = 0;
2335 static bool set_in_sync(struct mddev *mddev)
2337 lockdep_assert_held(&mddev->lock);
2338 if (!mddev->in_sync) {
2339 mddev->sync_checkers++;
2340 spin_unlock(&mddev->lock);
2341 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2342 spin_lock(&mddev->lock);
2343 if (!mddev->in_sync &&
2344 percpu_ref_is_zero(&mddev->writes_pending)) {
2347 * Ensure ->in_sync is visible before we clear
2351 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2352 sysfs_notify_dirent_safe(mddev->sysfs_state);
2354 if (--mddev->sync_checkers == 0)
2355 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2357 if (mddev->safemode == 1)
2358 mddev->safemode = 0;
2359 return mddev->in_sync;
2362 static void sync_sbs(struct mddev *mddev, int nospares)
2364 /* Update each superblock (in-memory image), but
2365 * if we are allowed to, skip spares which already
2366 * have the right event counter, or have one earlier
2367 * (which would mean they aren't being marked as dirty
2368 * with the rest of the array)
2370 struct md_rdev *rdev;
2371 rdev_for_each(rdev, mddev) {
2372 if (rdev->sb_events == mddev->events ||
2374 rdev->raid_disk < 0 &&
2375 rdev->sb_events+1 == mddev->events)) {
2376 /* Don't update this superblock */
2377 rdev->sb_loaded = 2;
2379 sync_super(mddev, rdev);
2380 rdev->sb_loaded = 1;
2385 static bool does_sb_need_changing(struct mddev *mddev)
2387 struct md_rdev *rdev;
2388 struct mdp_superblock_1 *sb;
2391 /* Find a good rdev */
2392 rdev_for_each(rdev, mddev)
2393 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2396 /* No good device found. */
2400 sb = page_address(rdev->sb_page);
2401 /* Check if a device has become faulty or a spare become active */
2402 rdev_for_each(rdev, mddev) {
2403 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2404 /* Device activated? */
2405 if (role == 0xffff && rdev->raid_disk >=0 &&
2406 !test_bit(Faulty, &rdev->flags))
2408 /* Device turned faulty? */
2409 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2413 /* Check if any mddev parameters have changed */
2414 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2415 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2416 (mddev->layout != le32_to_cpu(sb->layout)) ||
2417 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2418 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2424 void md_update_sb(struct mddev *mddev, int force_change)
2426 struct md_rdev *rdev;
2429 int any_badblocks_changed = 0;
2434 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2439 if (mddev_is_clustered(mddev)) {
2440 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2442 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2444 ret = md_cluster_ops->metadata_update_start(mddev);
2445 /* Has someone else has updated the sb */
2446 if (!does_sb_need_changing(mddev)) {
2448 md_cluster_ops->metadata_update_cancel(mddev);
2449 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2450 BIT(MD_SB_CHANGE_DEVS) |
2451 BIT(MD_SB_CHANGE_CLEAN));
2457 * First make sure individual recovery_offsets are correct
2458 * curr_resync_completed can only be used during recovery.
2459 * During reshape/resync it might use array-addresses rather
2460 * that device addresses.
2462 rdev_for_each(rdev, mddev) {
2463 if (rdev->raid_disk >= 0 &&
2464 mddev->delta_disks >= 0 &&
2465 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2466 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2467 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2468 !test_bit(Journal, &rdev->flags) &&
2469 !test_bit(In_sync, &rdev->flags) &&
2470 mddev->curr_resync_completed > rdev->recovery_offset)
2471 rdev->recovery_offset = mddev->curr_resync_completed;
2474 if (!mddev->persistent) {
2475 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2476 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2477 if (!mddev->external) {
2478 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2479 rdev_for_each(rdev, mddev) {
2480 if (rdev->badblocks.changed) {
2481 rdev->badblocks.changed = 0;
2482 ack_all_badblocks(&rdev->badblocks);
2483 md_error(mddev, rdev);
2485 clear_bit(Blocked, &rdev->flags);
2486 clear_bit(BlockedBadBlocks, &rdev->flags);
2487 wake_up(&rdev->blocked_wait);
2490 wake_up(&mddev->sb_wait);
2494 spin_lock(&mddev->lock);
2496 mddev->utime = ktime_get_real_seconds();
2498 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2500 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2501 /* just a clean<-> dirty transition, possibly leave spares alone,
2502 * though if events isn't the right even/odd, we will have to do
2508 if (mddev->degraded)
2509 /* If the array is degraded, then skipping spares is both
2510 * dangerous and fairly pointless.
2511 * Dangerous because a device that was removed from the array
2512 * might have a event_count that still looks up-to-date,
2513 * so it can be re-added without a resync.
2514 * Pointless because if there are any spares to skip,
2515 * then a recovery will happen and soon that array won't
2516 * be degraded any more and the spare can go back to sleep then.
2520 sync_req = mddev->in_sync;
2522 /* If this is just a dirty<->clean transition, and the array is clean
2523 * and 'events' is odd, we can roll back to the previous clean state */
2525 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2526 && mddev->can_decrease_events
2527 && mddev->events != 1) {
2529 mddev->can_decrease_events = 0;
2531 /* otherwise we have to go forward and ... */
2533 mddev->can_decrease_events = nospares;
2537 * This 64-bit counter should never wrap.
2538 * Either we are in around ~1 trillion A.C., assuming
2539 * 1 reboot per second, or we have a bug...
2541 WARN_ON(mddev->events == 0);
2543 rdev_for_each(rdev, mddev) {
2544 if (rdev->badblocks.changed)
2545 any_badblocks_changed++;
2546 if (test_bit(Faulty, &rdev->flags))
2547 set_bit(FaultRecorded, &rdev->flags);
2550 sync_sbs(mddev, nospares);
2551 spin_unlock(&mddev->lock);
2553 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2554 mdname(mddev), mddev->in_sync);
2557 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2559 bitmap_update_sb(mddev->bitmap);
2560 rdev_for_each(rdev, mddev) {
2561 char b[BDEVNAME_SIZE];
2563 if (rdev->sb_loaded != 1)
2564 continue; /* no noise on spare devices */
2566 if (!test_bit(Faulty, &rdev->flags)) {
2567 md_super_write(mddev,rdev,
2568 rdev->sb_start, rdev->sb_size,
2570 pr_debug("md: (write) %s's sb offset: %llu\n",
2571 bdevname(rdev->bdev, b),
2572 (unsigned long long)rdev->sb_start);
2573 rdev->sb_events = mddev->events;
2574 if (rdev->badblocks.size) {
2575 md_super_write(mddev, rdev,
2576 rdev->badblocks.sector,
2577 rdev->badblocks.size << 9,
2579 rdev->badblocks.size = 0;
2583 pr_debug("md: %s (skipping faulty)\n",
2584 bdevname(rdev->bdev, b));
2586 if (mddev->level == LEVEL_MULTIPATH)
2587 /* only need to write one superblock... */
2590 if (md_super_wait(mddev) < 0)
2592 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2594 if (mddev_is_clustered(mddev) && ret == 0)
2595 md_cluster_ops->metadata_update_finish(mddev);
2597 if (mddev->in_sync != sync_req ||
2598 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2599 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2600 /* have to write it out again */
2602 wake_up(&mddev->sb_wait);
2603 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2604 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2606 rdev_for_each(rdev, mddev) {
2607 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2608 clear_bit(Blocked, &rdev->flags);
2610 if (any_badblocks_changed)
2611 ack_all_badblocks(&rdev->badblocks);
2612 clear_bit(BlockedBadBlocks, &rdev->flags);
2613 wake_up(&rdev->blocked_wait);
2616 EXPORT_SYMBOL(md_update_sb);
2618 static int add_bound_rdev(struct md_rdev *rdev)
2620 struct mddev *mddev = rdev->mddev;
2622 bool add_journal = test_bit(Journal, &rdev->flags);
2624 if (!mddev->pers->hot_remove_disk || add_journal) {
2625 /* If there is hot_add_disk but no hot_remove_disk
2626 * then added disks for geometry changes,
2627 * and should be added immediately.
2629 super_types[mddev->major_version].
2630 validate_super(mddev, rdev);
2632 mddev_suspend(mddev);
2633 err = mddev->pers->hot_add_disk(mddev, rdev);
2635 mddev_resume(mddev);
2637 md_kick_rdev_from_array(rdev);
2641 sysfs_notify_dirent_safe(rdev->sysfs_state);
2643 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2644 if (mddev->degraded)
2645 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2646 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2647 md_new_event(mddev);
2648 md_wakeup_thread(mddev->thread);
2652 /* words written to sysfs files may, or may not, be \n terminated.
2653 * We want to accept with case. For this we use cmd_match.
2655 static int cmd_match(const char *cmd, const char *str)
2657 /* See if cmd, written into a sysfs file, matches
2658 * str. They must either be the same, or cmd can
2659 * have a trailing newline
2661 while (*cmd && *str && *cmd == *str) {
2672 struct rdev_sysfs_entry {
2673 struct attribute attr;
2674 ssize_t (*show)(struct md_rdev *, char *);
2675 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2679 state_show(struct md_rdev *rdev, char *page)
2683 unsigned long flags = READ_ONCE(rdev->flags);
2685 if (test_bit(Faulty, &flags) ||
2686 (!test_bit(ExternalBbl, &flags) &&
2687 rdev->badblocks.unacked_exist))
2688 len += sprintf(page+len, "faulty%s", sep);
2689 if (test_bit(In_sync, &flags))
2690 len += sprintf(page+len, "in_sync%s", sep);
2691 if (test_bit(Journal, &flags))
2692 len += sprintf(page+len, "journal%s", sep);
2693 if (test_bit(WriteMostly, &flags))
2694 len += sprintf(page+len, "write_mostly%s", sep);
2695 if (test_bit(Blocked, &flags) ||
2696 (rdev->badblocks.unacked_exist
2697 && !test_bit(Faulty, &flags)))
2698 len += sprintf(page+len, "blocked%s", sep);
2699 if (!test_bit(Faulty, &flags) &&
2700 !test_bit(Journal, &flags) &&
2701 !test_bit(In_sync, &flags))
2702 len += sprintf(page+len, "spare%s", sep);
2703 if (test_bit(WriteErrorSeen, &flags))
2704 len += sprintf(page+len, "write_error%s", sep);
2705 if (test_bit(WantReplacement, &flags))
2706 len += sprintf(page+len, "want_replacement%s", sep);
2707 if (test_bit(Replacement, &flags))
2708 len += sprintf(page+len, "replacement%s", sep);
2709 if (test_bit(ExternalBbl, &flags))
2710 len += sprintf(page+len, "external_bbl%s", sep);
2711 if (test_bit(FailFast, &flags))
2712 len += sprintf(page+len, "failfast%s", sep);
2717 return len+sprintf(page+len, "\n");
2721 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2724 * faulty - simulates an error
2725 * remove - disconnects the device
2726 * writemostly - sets write_mostly
2727 * -writemostly - clears write_mostly
2728 * blocked - sets the Blocked flags
2729 * -blocked - clears the Blocked and possibly simulates an error
2730 * insync - sets Insync providing device isn't active
2731 * -insync - clear Insync for a device with a slot assigned,
2732 * so that it gets rebuilt based on bitmap
2733 * write_error - sets WriteErrorSeen
2734 * -write_error - clears WriteErrorSeen
2735 * {,-}failfast - set/clear FailFast
2738 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2739 md_error(rdev->mddev, rdev);
2740 if (test_bit(Faulty, &rdev->flags))
2744 } else if (cmd_match(buf, "remove")) {
2745 if (rdev->mddev->pers) {
2746 clear_bit(Blocked, &rdev->flags);
2747 remove_and_add_spares(rdev->mddev, rdev);
2749 if (rdev->raid_disk >= 0)
2752 struct mddev *mddev = rdev->mddev;
2754 if (mddev_is_clustered(mddev))
2755 err = md_cluster_ops->remove_disk(mddev, rdev);
2758 md_kick_rdev_from_array(rdev);
2760 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2761 md_wakeup_thread(mddev->thread);
2763 md_new_event(mddev);
2766 } else if (cmd_match(buf, "writemostly")) {
2767 set_bit(WriteMostly, &rdev->flags);
2769 } else if (cmd_match(buf, "-writemostly")) {
2770 clear_bit(WriteMostly, &rdev->flags);
2772 } else if (cmd_match(buf, "blocked")) {
2773 set_bit(Blocked, &rdev->flags);
2775 } else if (cmd_match(buf, "-blocked")) {
2776 if (!test_bit(Faulty, &rdev->flags) &&
2777 !test_bit(ExternalBbl, &rdev->flags) &&
2778 rdev->badblocks.unacked_exist) {
2779 /* metadata handler doesn't understand badblocks,
2780 * so we need to fail the device
2782 md_error(rdev->mddev, rdev);
2784 clear_bit(Blocked, &rdev->flags);
2785 clear_bit(BlockedBadBlocks, &rdev->flags);
2786 wake_up(&rdev->blocked_wait);
2787 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2788 md_wakeup_thread(rdev->mddev->thread);
2791 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2792 set_bit(In_sync, &rdev->flags);
2794 } else if (cmd_match(buf, "failfast")) {
2795 set_bit(FailFast, &rdev->flags);
2797 } else if (cmd_match(buf, "-failfast")) {
2798 clear_bit(FailFast, &rdev->flags);
2800 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2801 !test_bit(Journal, &rdev->flags)) {
2802 if (rdev->mddev->pers == NULL) {
2803 clear_bit(In_sync, &rdev->flags);
2804 rdev->saved_raid_disk = rdev->raid_disk;
2805 rdev->raid_disk = -1;
2808 } else if (cmd_match(buf, "write_error")) {
2809 set_bit(WriteErrorSeen, &rdev->flags);
2811 } else if (cmd_match(buf, "-write_error")) {
2812 clear_bit(WriteErrorSeen, &rdev->flags);
2814 } else if (cmd_match(buf, "want_replacement")) {
2815 /* Any non-spare device that is not a replacement can
2816 * become want_replacement at any time, but we then need to
2817 * check if recovery is needed.
2819 if (rdev->raid_disk >= 0 &&
2820 !test_bit(Journal, &rdev->flags) &&
2821 !test_bit(Replacement, &rdev->flags))
2822 set_bit(WantReplacement, &rdev->flags);
2823 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2824 md_wakeup_thread(rdev->mddev->thread);
2826 } else if (cmd_match(buf, "-want_replacement")) {
2827 /* Clearing 'want_replacement' is always allowed.
2828 * Once replacements starts it is too late though.
2831 clear_bit(WantReplacement, &rdev->flags);
2832 } else if (cmd_match(buf, "replacement")) {
2833 /* Can only set a device as a replacement when array has not
2834 * yet been started. Once running, replacement is automatic
2835 * from spares, or by assigning 'slot'.
2837 if (rdev->mddev->pers)
2840 set_bit(Replacement, &rdev->flags);
2843 } else if (cmd_match(buf, "-replacement")) {
2844 /* Similarly, can only clear Replacement before start */
2845 if (rdev->mddev->pers)
2848 clear_bit(Replacement, &rdev->flags);
2851 } else if (cmd_match(buf, "re-add")) {
2852 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2853 /* clear_bit is performed _after_ all the devices
2854 * have their local Faulty bit cleared. If any writes
2855 * happen in the meantime in the local node, they
2856 * will land in the local bitmap, which will be synced
2857 * by this node eventually
2859 if (!mddev_is_clustered(rdev->mddev) ||
2860 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2861 clear_bit(Faulty, &rdev->flags);
2862 err = add_bound_rdev(rdev);
2866 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2867 set_bit(ExternalBbl, &rdev->flags);
2868 rdev->badblocks.shift = 0;
2870 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2871 clear_bit(ExternalBbl, &rdev->flags);
2875 sysfs_notify_dirent_safe(rdev->sysfs_state);
2876 return err ? err : len;
2878 static struct rdev_sysfs_entry rdev_state =
2879 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2882 errors_show(struct md_rdev *rdev, char *page)
2884 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2888 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2893 rv = kstrtouint(buf, 10, &n);
2896 atomic_set(&rdev->corrected_errors, n);
2899 static struct rdev_sysfs_entry rdev_errors =
2900 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2903 slot_show(struct md_rdev *rdev, char *page)
2905 if (test_bit(Journal, &rdev->flags))
2906 return sprintf(page, "journal\n");
2907 else if (rdev->raid_disk < 0)
2908 return sprintf(page, "none\n");
2910 return sprintf(page, "%d\n", rdev->raid_disk);
2914 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2919 if (test_bit(Journal, &rdev->flags))
2921 if (strncmp(buf, "none", 4)==0)
2924 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2928 if (rdev->mddev->pers && slot == -1) {
2929 /* Setting 'slot' on an active array requires also
2930 * updating the 'rd%d' link, and communicating
2931 * with the personality with ->hot_*_disk.
2932 * For now we only support removing
2933 * failed/spare devices. This normally happens automatically,
2934 * but not when the metadata is externally managed.
2936 if (rdev->raid_disk == -1)
2938 /* personality does all needed checks */
2939 if (rdev->mddev->pers->hot_remove_disk == NULL)
2941 clear_bit(Blocked, &rdev->flags);
2942 remove_and_add_spares(rdev->mddev, rdev);
2943 if (rdev->raid_disk >= 0)
2945 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2946 md_wakeup_thread(rdev->mddev->thread);
2947 } else if (rdev->mddev->pers) {
2948 /* Activating a spare .. or possibly reactivating
2949 * if we ever get bitmaps working here.
2953 if (rdev->raid_disk != -1)
2956 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2959 if (rdev->mddev->pers->hot_add_disk == NULL)
2962 if (slot >= rdev->mddev->raid_disks &&
2963 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2966 rdev->raid_disk = slot;
2967 if (test_bit(In_sync, &rdev->flags))
2968 rdev->saved_raid_disk = slot;
2970 rdev->saved_raid_disk = -1;
2971 clear_bit(In_sync, &rdev->flags);
2972 clear_bit(Bitmap_sync, &rdev->flags);
2973 err = rdev->mddev->pers->
2974 hot_add_disk(rdev->mddev, rdev);
2976 rdev->raid_disk = -1;
2979 sysfs_notify_dirent_safe(rdev->sysfs_state);
2980 if (sysfs_link_rdev(rdev->mddev, rdev))
2981 /* failure here is OK */;
2982 /* don't wakeup anyone, leave that to userspace. */
2984 if (slot >= rdev->mddev->raid_disks &&
2985 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2987 rdev->raid_disk = slot;
2988 /* assume it is working */
2989 clear_bit(Faulty, &rdev->flags);
2990 clear_bit(WriteMostly, &rdev->flags);
2991 set_bit(In_sync, &rdev->flags);
2992 sysfs_notify_dirent_safe(rdev->sysfs_state);
2997 static struct rdev_sysfs_entry rdev_slot =
2998 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3001 offset_show(struct md_rdev *rdev, char *page)
3003 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3007 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3009 unsigned long long offset;
3010 if (kstrtoull(buf, 10, &offset) < 0)
3012 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3014 if (rdev->sectors && rdev->mddev->external)
3015 /* Must set offset before size, so overlap checks
3018 rdev->data_offset = offset;
3019 rdev->new_data_offset = offset;
3023 static struct rdev_sysfs_entry rdev_offset =
3024 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3026 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3028 return sprintf(page, "%llu\n",
3029 (unsigned long long)rdev->new_data_offset);
3032 static ssize_t new_offset_store(struct md_rdev *rdev,
3033 const char *buf, size_t len)
3035 unsigned long long new_offset;
3036 struct mddev *mddev = rdev->mddev;
3038 if (kstrtoull(buf, 10, &new_offset) < 0)
3041 if (mddev->sync_thread ||
3042 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3044 if (new_offset == rdev->data_offset)
3045 /* reset is always permitted */
3047 else if (new_offset > rdev->data_offset) {
3048 /* must not push array size beyond rdev_sectors */
3049 if (new_offset - rdev->data_offset
3050 + mddev->dev_sectors > rdev->sectors)
3053 /* Metadata worries about other space details. */
3055 /* decreasing the offset is inconsistent with a backwards
3058 if (new_offset < rdev->data_offset &&
3059 mddev->reshape_backwards)
3061 /* Increasing offset is inconsistent with forwards
3062 * reshape. reshape_direction should be set to
3063 * 'backwards' first.
3065 if (new_offset > rdev->data_offset &&
3066 !mddev->reshape_backwards)
3069 if (mddev->pers && mddev->persistent &&
3070 !super_types[mddev->major_version]
3071 .allow_new_offset(rdev, new_offset))
3073 rdev->new_data_offset = new_offset;
3074 if (new_offset > rdev->data_offset)
3075 mddev->reshape_backwards = 1;
3076 else if (new_offset < rdev->data_offset)
3077 mddev->reshape_backwards = 0;
3081 static struct rdev_sysfs_entry rdev_new_offset =
3082 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3085 rdev_size_show(struct md_rdev *rdev, char *page)
3087 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3090 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3092 /* check if two start/length pairs overlap */
3100 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3102 unsigned long long blocks;
3105 if (kstrtoull(buf, 10, &blocks) < 0)
3108 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3109 return -EINVAL; /* sector conversion overflow */
3112 if (new != blocks * 2)
3113 return -EINVAL; /* unsigned long long to sector_t overflow */
3120 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3122 struct mddev *my_mddev = rdev->mddev;
3123 sector_t oldsectors = rdev->sectors;
3126 if (test_bit(Journal, &rdev->flags))
3128 if (strict_blocks_to_sectors(buf, §ors) < 0)
3130 if (rdev->data_offset != rdev->new_data_offset)
3131 return -EINVAL; /* too confusing */
3132 if (my_mddev->pers && rdev->raid_disk >= 0) {
3133 if (my_mddev->persistent) {
3134 sectors = super_types[my_mddev->major_version].
3135 rdev_size_change(rdev, sectors);
3138 } else if (!sectors)
3139 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3141 if (!my_mddev->pers->resize)
3142 /* Cannot change size for RAID0 or Linear etc */
3145 if (sectors < my_mddev->dev_sectors)
3146 return -EINVAL; /* component must fit device */
3148 rdev->sectors = sectors;
3149 if (sectors > oldsectors && my_mddev->external) {
3150 /* Need to check that all other rdevs with the same
3151 * ->bdev do not overlap. 'rcu' is sufficient to walk
3152 * the rdev lists safely.
3153 * This check does not provide a hard guarantee, it
3154 * just helps avoid dangerous mistakes.
3156 struct mddev *mddev;
3158 struct list_head *tmp;
3161 for_each_mddev(mddev, tmp) {
3162 struct md_rdev *rdev2;
3164 rdev_for_each(rdev2, mddev)
3165 if (rdev->bdev == rdev2->bdev &&
3167 overlaps(rdev->data_offset, rdev->sectors,
3180 /* Someone else could have slipped in a size
3181 * change here, but doing so is just silly.
3182 * We put oldsectors back because we *know* it is
3183 * safe, and trust userspace not to race with
3186 rdev->sectors = oldsectors;
3193 static struct rdev_sysfs_entry rdev_size =
3194 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3196 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3198 unsigned long long recovery_start = rdev->recovery_offset;
3200 if (test_bit(In_sync, &rdev->flags) ||
3201 recovery_start == MaxSector)
3202 return sprintf(page, "none\n");
3204 return sprintf(page, "%llu\n", recovery_start);
3207 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3209 unsigned long long recovery_start;
3211 if (cmd_match(buf, "none"))
3212 recovery_start = MaxSector;
3213 else if (kstrtoull(buf, 10, &recovery_start))
3216 if (rdev->mddev->pers &&
3217 rdev->raid_disk >= 0)
3220 rdev->recovery_offset = recovery_start;
3221 if (recovery_start == MaxSector)
3222 set_bit(In_sync, &rdev->flags);
3224 clear_bit(In_sync, &rdev->flags);
3228 static struct rdev_sysfs_entry rdev_recovery_start =
3229 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3231 /* sysfs access to bad-blocks list.
3232 * We present two files.
3233 * 'bad-blocks' lists sector numbers and lengths of ranges that
3234 * are recorded as bad. The list is truncated to fit within
3235 * the one-page limit of sysfs.
3236 * Writing "sector length" to this file adds an acknowledged
3238 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3239 * been acknowledged. Writing to this file adds bad blocks
3240 * without acknowledging them. This is largely for testing.
3242 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3244 return badblocks_show(&rdev->badblocks, page, 0);
3246 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3248 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3249 /* Maybe that ack was all we needed */
3250 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3251 wake_up(&rdev->blocked_wait);
3254 static struct rdev_sysfs_entry rdev_bad_blocks =
3255 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3257 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3259 return badblocks_show(&rdev->badblocks, page, 1);
3261 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3263 return badblocks_store(&rdev->badblocks, page, len, 1);
3265 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3266 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3269 ppl_sector_show(struct md_rdev *rdev, char *page)
3271 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3275 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3277 unsigned long long sector;
3279 if (kstrtoull(buf, 10, §or) < 0)
3281 if (sector != (sector_t)sector)
3284 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3285 rdev->raid_disk >= 0)
3288 if (rdev->mddev->persistent) {
3289 if (rdev->mddev->major_version == 0)
3291 if ((sector > rdev->sb_start &&
3292 sector - rdev->sb_start > S16_MAX) ||
3293 (sector < rdev->sb_start &&
3294 rdev->sb_start - sector > -S16_MIN))
3296 rdev->ppl.offset = sector - rdev->sb_start;
3297 } else if (!rdev->mddev->external) {
3300 rdev->ppl.sector = sector;
3304 static struct rdev_sysfs_entry rdev_ppl_sector =
3305 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3308 ppl_size_show(struct md_rdev *rdev, char *page)
3310 return sprintf(page, "%u\n", rdev->ppl.size);
3314 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3318 if (kstrtouint(buf, 10, &size) < 0)
3321 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3322 rdev->raid_disk >= 0)
3325 if (rdev->mddev->persistent) {
3326 if (rdev->mddev->major_version == 0)
3330 } else if (!rdev->mddev->external) {
3333 rdev->ppl.size = size;
3337 static struct rdev_sysfs_entry rdev_ppl_size =
3338 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3340 static struct attribute *rdev_default_attrs[] = {
3345 &rdev_new_offset.attr,
3347 &rdev_recovery_start.attr,
3348 &rdev_bad_blocks.attr,
3349 &rdev_unack_bad_blocks.attr,
3350 &rdev_ppl_sector.attr,
3351 &rdev_ppl_size.attr,
3355 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3357 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3358 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3364 return entry->show(rdev, page);
3368 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3369 const char *page, size_t length)
3371 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3372 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3374 struct mddev *mddev = rdev->mddev;
3378 if (!capable(CAP_SYS_ADMIN))
3380 rv = mddev ? mddev_lock(mddev): -EBUSY;
3382 if (rdev->mddev == NULL)
3385 rv = entry->store(rdev, page, length);
3386 mddev_unlock(mddev);
3391 static void rdev_free(struct kobject *ko)
3393 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3396 static const struct sysfs_ops rdev_sysfs_ops = {
3397 .show = rdev_attr_show,
3398 .store = rdev_attr_store,
3400 static struct kobj_type rdev_ktype = {
3401 .release = rdev_free,
3402 .sysfs_ops = &rdev_sysfs_ops,
3403 .default_attrs = rdev_default_attrs,
3406 int md_rdev_init(struct md_rdev *rdev)
3409 rdev->saved_raid_disk = -1;
3410 rdev->raid_disk = -1;
3412 rdev->data_offset = 0;
3413 rdev->new_data_offset = 0;
3414 rdev->sb_events = 0;
3415 rdev->last_read_error = 0;
3416 rdev->sb_loaded = 0;
3417 rdev->bb_page = NULL;
3418 atomic_set(&rdev->nr_pending, 0);
3419 atomic_set(&rdev->read_errors, 0);
3420 atomic_set(&rdev->corrected_errors, 0);
3422 INIT_LIST_HEAD(&rdev->same_set);
3423 init_waitqueue_head(&rdev->blocked_wait);
3425 /* Add space to store bad block list.
3426 * This reserves the space even on arrays where it cannot
3427 * be used - I wonder if that matters
3429 return badblocks_init(&rdev->badblocks, 0);
3431 EXPORT_SYMBOL_GPL(md_rdev_init);
3433 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3435 * mark the device faulty if:
3437 * - the device is nonexistent (zero size)
3438 * - the device has no valid superblock
3440 * a faulty rdev _never_ has rdev->sb set.
3442 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3444 char b[BDEVNAME_SIZE];
3446 struct md_rdev *rdev;
3449 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3451 return ERR_PTR(-ENOMEM);
3453 err = md_rdev_init(rdev);
3456 err = alloc_disk_sb(rdev);
3460 err = lock_rdev(rdev, newdev, super_format == -2);
3464 kobject_init(&rdev->kobj, &rdev_ktype);
3466 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3468 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3469 bdevname(rdev->bdev,b));
3474 if (super_format >= 0) {
3475 err = super_types[super_format].
3476 load_super(rdev, NULL, super_minor);
3477 if (err == -EINVAL) {
3478 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3479 bdevname(rdev->bdev,b),
3480 super_format, super_minor);
3484 pr_warn("md: could not read %s's sb, not importing!\n",
3485 bdevname(rdev->bdev,b));
3495 md_rdev_clear(rdev);
3497 return ERR_PTR(err);
3501 * Check a full RAID array for plausibility
3504 static void analyze_sbs(struct mddev *mddev)
3507 struct md_rdev *rdev, *freshest, *tmp;
3508 char b[BDEVNAME_SIZE];
3511 rdev_for_each_safe(rdev, tmp, mddev)
3512 switch (super_types[mddev->major_version].
3513 load_super(rdev, freshest, mddev->minor_version)) {
3520 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3521 bdevname(rdev->bdev,b));
3522 md_kick_rdev_from_array(rdev);
3525 super_types[mddev->major_version].
3526 validate_super(mddev, freshest);
3529 rdev_for_each_safe(rdev, tmp, mddev) {
3530 if (mddev->max_disks &&
3531 (rdev->desc_nr >= mddev->max_disks ||
3532 i > mddev->max_disks)) {
3533 pr_warn("md: %s: %s: only %d devices permitted\n",
3534 mdname(mddev), bdevname(rdev->bdev, b),
3536 md_kick_rdev_from_array(rdev);
3539 if (rdev != freshest) {
3540 if (super_types[mddev->major_version].
3541 validate_super(mddev, rdev)) {
3542 pr_warn("md: kicking non-fresh %s from array!\n",
3543 bdevname(rdev->bdev,b));
3544 md_kick_rdev_from_array(rdev);
3548 if (mddev->level == LEVEL_MULTIPATH) {
3549 rdev->desc_nr = i++;
3550 rdev->raid_disk = rdev->desc_nr;
3551 set_bit(In_sync, &rdev->flags);
3552 } else if (rdev->raid_disk >=
3553 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3554 !test_bit(Journal, &rdev->flags)) {
3555 rdev->raid_disk = -1;
3556 clear_bit(In_sync, &rdev->flags);
3561 /* Read a fixed-point number.
3562 * Numbers in sysfs attributes should be in "standard" units where
3563 * possible, so time should be in seconds.
3564 * However we internally use a a much smaller unit such as
3565 * milliseconds or jiffies.
3566 * This function takes a decimal number with a possible fractional
3567 * component, and produces an integer which is the result of
3568 * multiplying that number by 10^'scale'.
3569 * all without any floating-point arithmetic.
3571 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3573 unsigned long result = 0;
3575 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3578 else if (decimals < scale) {
3581 result = result * 10 + value;
3593 while (decimals < scale) {
3602 safe_delay_show(struct mddev *mddev, char *page)
3604 int msec = (mddev->safemode_delay*1000)/HZ;
3605 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3608 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3612 if (mddev_is_clustered(mddev)) {
3613 pr_warn("md: Safemode is disabled for clustered mode\n");
3617 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3620 mddev->safemode_delay = 0;
3622 unsigned long old_delay = mddev->safemode_delay;
3623 unsigned long new_delay = (msec*HZ)/1000;
3627 mddev->safemode_delay = new_delay;
3628 if (new_delay < old_delay || old_delay == 0)
3629 mod_timer(&mddev->safemode_timer, jiffies+1);
3633 static struct md_sysfs_entry md_safe_delay =
3634 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3637 level_show(struct mddev *mddev, char *page)
3639 struct md_personality *p;
3641 spin_lock(&mddev->lock);
3644 ret = sprintf(page, "%s\n", p->name);
3645 else if (mddev->clevel[0])
3646 ret = sprintf(page, "%s\n", mddev->clevel);
3647 else if (mddev->level != LEVEL_NONE)
3648 ret = sprintf(page, "%d\n", mddev->level);
3651 spin_unlock(&mddev->lock);
3656 level_store(struct mddev *mddev, const char *buf, size_t len)
3661 struct md_personality *pers, *oldpers;
3663 void *priv, *oldpriv;
3664 struct md_rdev *rdev;
3666 if (slen == 0 || slen >= sizeof(clevel))
3669 rv = mddev_lock(mddev);
3673 if (mddev->pers == NULL) {
3674 strncpy(mddev->clevel, buf, slen);
3675 if (mddev->clevel[slen-1] == '\n')
3677 mddev->clevel[slen] = 0;
3678 mddev->level = LEVEL_NONE;
3686 /* request to change the personality. Need to ensure:
3687 * - array is not engaged in resync/recovery/reshape
3688 * - old personality can be suspended
3689 * - new personality will access other array.
3693 if (mddev->sync_thread ||
3694 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3695 mddev->reshape_position != MaxSector ||
3696 mddev->sysfs_active)
3700 if (!mddev->pers->quiesce) {
3701 pr_warn("md: %s: %s does not support online personality change\n",
3702 mdname(mddev), mddev->pers->name);
3706 /* Now find the new personality */
3707 strncpy(clevel, buf, slen);
3708 if (clevel[slen-1] == '\n')
3711 if (kstrtol(clevel, 10, &level))
3714 if (request_module("md-%s", clevel) != 0)
3715 request_module("md-level-%s", clevel);
3716 spin_lock(&pers_lock);
3717 pers = find_pers(level, clevel);
3718 if (!pers || !try_module_get(pers->owner)) {
3719 spin_unlock(&pers_lock);
3720 pr_warn("md: personality %s not loaded\n", clevel);
3724 spin_unlock(&pers_lock);
3726 if (pers == mddev->pers) {
3727 /* Nothing to do! */
3728 module_put(pers->owner);
3732 if (!pers->takeover) {
3733 module_put(pers->owner);
3734 pr_warn("md: %s: %s does not support personality takeover\n",
3735 mdname(mddev), clevel);
3740 rdev_for_each(rdev, mddev)
3741 rdev->new_raid_disk = rdev->raid_disk;
3743 /* ->takeover must set new_* and/or delta_disks
3744 * if it succeeds, and may set them when it fails.
3746 priv = pers->takeover(mddev);
3748 mddev->new_level = mddev->level;
3749 mddev->new_layout = mddev->layout;
3750 mddev->new_chunk_sectors = mddev->chunk_sectors;
3751 mddev->raid_disks -= mddev->delta_disks;
3752 mddev->delta_disks = 0;
3753 mddev->reshape_backwards = 0;
3754 module_put(pers->owner);
3755 pr_warn("md: %s: %s would not accept array\n",
3756 mdname(mddev), clevel);
3761 /* Looks like we have a winner */
3762 mddev_suspend(mddev);
3763 mddev_detach(mddev);
3765 spin_lock(&mddev->lock);
3766 oldpers = mddev->pers;
3767 oldpriv = mddev->private;
3769 mddev->private = priv;
3770 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3771 mddev->level = mddev->new_level;
3772 mddev->layout = mddev->new_layout;
3773 mddev->chunk_sectors = mddev->new_chunk_sectors;
3774 mddev->delta_disks = 0;
3775 mddev->reshape_backwards = 0;
3776 mddev->degraded = 0;
3777 spin_unlock(&mddev->lock);
3779 if (oldpers->sync_request == NULL &&
3781 /* We are converting from a no-redundancy array
3782 * to a redundancy array and metadata is managed
3783 * externally so we need to be sure that writes
3784 * won't block due to a need to transition
3786 * until external management is started.
3789 mddev->safemode_delay = 0;
3790 mddev->safemode = 0;
3793 oldpers->free(mddev, oldpriv);
3795 if (oldpers->sync_request == NULL &&
3796 pers->sync_request != NULL) {
3797 /* need to add the md_redundancy_group */
3798 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3799 pr_warn("md: cannot register extra attributes for %s\n",
3801 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3803 if (oldpers->sync_request != NULL &&
3804 pers->sync_request == NULL) {
3805 /* need to remove the md_redundancy_group */
3806 if (mddev->to_remove == NULL)
3807 mddev->to_remove = &md_redundancy_group;
3810 module_put(oldpers->owner);
3812 rdev_for_each(rdev, mddev) {
3813 if (rdev->raid_disk < 0)
3815 if (rdev->new_raid_disk >= mddev->raid_disks)
3816 rdev->new_raid_disk = -1;
3817 if (rdev->new_raid_disk == rdev->raid_disk)
3819 sysfs_unlink_rdev(mddev, rdev);
3821 rdev_for_each(rdev, mddev) {
3822 if (rdev->raid_disk < 0)
3824 if (rdev->new_raid_disk == rdev->raid_disk)
3826 rdev->raid_disk = rdev->new_raid_disk;
3827 if (rdev->raid_disk < 0)
3828 clear_bit(In_sync, &rdev->flags);
3830 if (sysfs_link_rdev(mddev, rdev))
3831 pr_warn("md: cannot register rd%d for %s after level change\n",
3832 rdev->raid_disk, mdname(mddev));
3836 if (pers->sync_request == NULL) {
3837 /* this is now an array without redundancy, so
3838 * it must always be in_sync
3841 del_timer_sync(&mddev->safemode_timer);
3843 blk_set_stacking_limits(&mddev->queue->limits);
3845 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3846 mddev_resume(mddev);
3848 md_update_sb(mddev, 1);
3849 sysfs_notify(&mddev->kobj, NULL, "level");
3850 md_new_event(mddev);
3853 mddev_unlock(mddev);
3857 static struct md_sysfs_entry md_level =
3858 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3861 layout_show(struct mddev *mddev, char *page)
3863 /* just a number, not meaningful for all levels */
3864 if (mddev->reshape_position != MaxSector &&
3865 mddev->layout != mddev->new_layout)
3866 return sprintf(page, "%d (%d)\n",
3867 mddev->new_layout, mddev->layout);
3868 return sprintf(page, "%d\n", mddev->layout);
3872 layout_store(struct mddev *mddev, const char *buf, size_t len)
3877 err = kstrtouint(buf, 10, &n);
3880 err = mddev_lock(mddev);
3885 if (mddev->pers->check_reshape == NULL)
3890 mddev->new_layout = n;
3891 err = mddev->pers->check_reshape(mddev);
3893 mddev->new_layout = mddev->layout;
3896 mddev->new_layout = n;
3897 if (mddev->reshape_position == MaxSector)
3900 mddev_unlock(mddev);
3903 static struct md_sysfs_entry md_layout =
3904 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3907 raid_disks_show(struct mddev *mddev, char *page)
3909 if (mddev->raid_disks == 0)
3911 if (mddev->reshape_position != MaxSector &&
3912 mddev->delta_disks != 0)
3913 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3914 mddev->raid_disks - mddev->delta_disks);
3915 return sprintf(page, "%d\n", mddev->raid_disks);
3918 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3921 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3926 err = kstrtouint(buf, 10, &n);
3930 err = mddev_lock(mddev);
3934 err = update_raid_disks(mddev, n);
3935 else if (mddev->reshape_position != MaxSector) {
3936 struct md_rdev *rdev;
3937 int olddisks = mddev->raid_disks - mddev->delta_disks;
3940 rdev_for_each(rdev, mddev) {
3942 rdev->data_offset < rdev->new_data_offset)
3945 rdev->data_offset > rdev->new_data_offset)
3949 mddev->delta_disks = n - olddisks;
3950 mddev->raid_disks = n;
3951 mddev->reshape_backwards = (mddev->delta_disks < 0);
3953 mddev->raid_disks = n;
3955 mddev_unlock(mddev);
3956 return err ? err : len;
3958 static struct md_sysfs_entry md_raid_disks =
3959 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3962 chunk_size_show(struct mddev *mddev, char *page)
3964 if (mddev->reshape_position != MaxSector &&
3965 mddev->chunk_sectors != mddev->new_chunk_sectors)
3966 return sprintf(page, "%d (%d)\n",
3967 mddev->new_chunk_sectors << 9,
3968 mddev->chunk_sectors << 9);
3969 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3973 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3978 err = kstrtoul(buf, 10, &n);
3982 err = mddev_lock(mddev);
3986 if (mddev->pers->check_reshape == NULL)
3991 mddev->new_chunk_sectors = n >> 9;
3992 err = mddev->pers->check_reshape(mddev);
3994 mddev->new_chunk_sectors = mddev->chunk_sectors;
3997 mddev->new_chunk_sectors = n >> 9;
3998 if (mddev->reshape_position == MaxSector)
3999 mddev->chunk_sectors = n >> 9;
4001 mddev_unlock(mddev);
4004 static struct md_sysfs_entry md_chunk_size =
4005 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4008 resync_start_show(struct mddev *mddev, char *page)
4010 if (mddev->recovery_cp == MaxSector)
4011 return sprintf(page, "none\n");
4012 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4016 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4018 unsigned long long n;
4021 if (cmd_match(buf, "none"))
4024 err = kstrtoull(buf, 10, &n);
4027 if (n != (sector_t)n)
4031 err = mddev_lock(mddev);
4034 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4038 mddev->recovery_cp = n;
4040 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4042 mddev_unlock(mddev);
4045 static struct md_sysfs_entry md_resync_start =
4046 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4047 resync_start_show, resync_start_store);
4050 * The array state can be:
4053 * No devices, no size, no level
4054 * Equivalent to STOP_ARRAY ioctl
4056 * May have some settings, but array is not active
4057 * all IO results in error
4058 * When written, doesn't tear down array, but just stops it
4059 * suspended (not supported yet)
4060 * All IO requests will block. The array can be reconfigured.
4061 * Writing this, if accepted, will block until array is quiescent
4063 * no resync can happen. no superblocks get written.
4064 * write requests fail
4066 * like readonly, but behaves like 'clean' on a write request.
4068 * clean - no pending writes, but otherwise active.
4069 * When written to inactive array, starts without resync
4070 * If a write request arrives then
4071 * if metadata is known, mark 'dirty' and switch to 'active'.
4072 * if not known, block and switch to write-pending
4073 * If written to an active array that has pending writes, then fails.
4075 * fully active: IO and resync can be happening.
4076 * When written to inactive array, starts with resync
4079 * clean, but writes are blocked waiting for 'active' to be written.
4082 * like active, but no writes have been seen for a while (100msec).
4085 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4086 write_pending, active_idle, bad_word};
4087 static char *array_states[] = {
4088 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4089 "write-pending", "active-idle", NULL };
4091 static int match_word(const char *word, char **list)
4094 for (n=0; list[n]; n++)
4095 if (cmd_match(word, list[n]))
4101 array_state_show(struct mddev *mddev, char *page)
4103 enum array_state st = inactive;
4114 spin_lock(&mddev->lock);
4115 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4117 else if (mddev->in_sync)
4119 else if (mddev->safemode)
4123 spin_unlock(&mddev->lock);
4126 if (list_empty(&mddev->disks) &&
4127 mddev->raid_disks == 0 &&
4128 mddev->dev_sectors == 0)
4133 return sprintf(page, "%s\n", array_states[st]);
4136 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4137 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4138 static int do_md_run(struct mddev *mddev);
4139 static int restart_array(struct mddev *mddev);
4142 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4145 enum array_state st = match_word(buf, array_states);
4147 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4148 /* don't take reconfig_mutex when toggling between
4151 spin_lock(&mddev->lock);
4153 restart_array(mddev);
4154 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4155 md_wakeup_thread(mddev->thread);
4156 wake_up(&mddev->sb_wait);
4157 } else /* st == clean */ {
4158 restart_array(mddev);
4159 if (!set_in_sync(mddev))
4163 sysfs_notify_dirent_safe(mddev->sysfs_state);
4164 spin_unlock(&mddev->lock);
4167 err = mddev_lock(mddev);
4175 /* stopping an active array */
4176 err = do_md_stop(mddev, 0, NULL);
4179 /* stopping an active array */
4181 err = do_md_stop(mddev, 2, NULL);
4183 err = 0; /* already inactive */
4186 break; /* not supported yet */
4189 err = md_set_readonly(mddev, NULL);
4192 set_disk_ro(mddev->gendisk, 1);
4193 err = do_md_run(mddev);
4199 err = md_set_readonly(mddev, NULL);
4200 else if (mddev->ro == 1)
4201 err = restart_array(mddev);
4204 set_disk_ro(mddev->gendisk, 0);
4208 err = do_md_run(mddev);
4213 err = restart_array(mddev);
4216 spin_lock(&mddev->lock);
4217 if (!set_in_sync(mddev))
4219 spin_unlock(&mddev->lock);
4225 err = restart_array(mddev);
4228 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4229 wake_up(&mddev->sb_wait);
4233 set_disk_ro(mddev->gendisk, 0);
4234 err = do_md_run(mddev);
4239 /* these cannot be set */
4244 if (mddev->hold_active == UNTIL_IOCTL)
4245 mddev->hold_active = 0;
4246 sysfs_notify_dirent_safe(mddev->sysfs_state);
4248 mddev_unlock(mddev);
4251 static struct md_sysfs_entry md_array_state =
4252 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4255 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4256 return sprintf(page, "%d\n",
4257 atomic_read(&mddev->max_corr_read_errors));
4261 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4266 rv = kstrtouint(buf, 10, &n);
4269 atomic_set(&mddev->max_corr_read_errors, n);
4273 static struct md_sysfs_entry max_corr_read_errors =
4274 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4275 max_corrected_read_errors_store);
4278 null_show(struct mddev *mddev, char *page)
4284 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4286 /* buf must be %d:%d\n? giving major and minor numbers */
4287 /* The new device is added to the array.
4288 * If the array has a persistent superblock, we read the
4289 * superblock to initialise info and check validity.
4290 * Otherwise, only checking done is that in bind_rdev_to_array,
4291 * which mainly checks size.
4294 int major = simple_strtoul(buf, &e, 10);
4297 struct md_rdev *rdev;
4300 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4302 minor = simple_strtoul(e+1, &e, 10);
4303 if (*e && *e != '\n')
4305 dev = MKDEV(major, minor);
4306 if (major != MAJOR(dev) ||
4307 minor != MINOR(dev))
4310 flush_workqueue(md_misc_wq);
4312 err = mddev_lock(mddev);
4315 if (mddev->persistent) {
4316 rdev = md_import_device(dev, mddev->major_version,
4317 mddev->minor_version);
4318 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4319 struct md_rdev *rdev0
4320 = list_entry(mddev->disks.next,
4321 struct md_rdev, same_set);
4322 err = super_types[mddev->major_version]
4323 .load_super(rdev, rdev0, mddev->minor_version);
4327 } else if (mddev->external)
4328 rdev = md_import_device(dev, -2, -1);
4330 rdev = md_import_device(dev, -1, -1);
4333 mddev_unlock(mddev);
4334 return PTR_ERR(rdev);
4336 err = bind_rdev_to_array(rdev, mddev);
4340 mddev_unlock(mddev);
4342 md_new_event(mddev);
4343 return err ? err : len;
4346 static struct md_sysfs_entry md_new_device =
4347 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4350 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4353 unsigned long chunk, end_chunk;
4356 err = mddev_lock(mddev);
4361 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4363 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4364 if (buf == end) break;
4365 if (*end == '-') { /* range */
4367 end_chunk = simple_strtoul(buf, &end, 0);
4368 if (buf == end) break;
4370 if (*end && !isspace(*end)) break;
4371 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4372 buf = skip_spaces(end);
4374 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4376 mddev_unlock(mddev);
4380 static struct md_sysfs_entry md_bitmap =
4381 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4384 size_show(struct mddev *mddev, char *page)
4386 return sprintf(page, "%llu\n",
4387 (unsigned long long)mddev->dev_sectors / 2);
4390 static int update_size(struct mddev *mddev, sector_t num_sectors);
4393 size_store(struct mddev *mddev, const char *buf, size_t len)
4395 /* If array is inactive, we can reduce the component size, but
4396 * not increase it (except from 0).
4397 * If array is active, we can try an on-line resize
4400 int err = strict_blocks_to_sectors(buf, §ors);
4404 err = mddev_lock(mddev);
4408 err = update_size(mddev, sectors);
4410 md_update_sb(mddev, 1);
4412 if (mddev->dev_sectors == 0 ||
4413 mddev->dev_sectors > sectors)
4414 mddev->dev_sectors = sectors;
4418 mddev_unlock(mddev);
4419 return err ? err : len;
4422 static struct md_sysfs_entry md_size =
4423 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4425 /* Metadata version.
4427 * 'none' for arrays with no metadata (good luck...)
4428 * 'external' for arrays with externally managed metadata,
4429 * or N.M for internally known formats
4432 metadata_show(struct mddev *mddev, char *page)
4434 if (mddev->persistent)
4435 return sprintf(page, "%d.%d\n",
4436 mddev->major_version, mddev->minor_version);
4437 else if (mddev->external)
4438 return sprintf(page, "external:%s\n", mddev->metadata_type);
4440 return sprintf(page, "none\n");
4444 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4449 /* Changing the details of 'external' metadata is
4450 * always permitted. Otherwise there must be
4451 * no devices attached to the array.
4454 err = mddev_lock(mddev);
4458 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4460 else if (!list_empty(&mddev->disks))
4464 if (cmd_match(buf, "none")) {
4465 mddev->persistent = 0;
4466 mddev->external = 0;
4467 mddev->major_version = 0;
4468 mddev->minor_version = 90;
4471 if (strncmp(buf, "external:", 9) == 0) {
4472 size_t namelen = len-9;
4473 if (namelen >= sizeof(mddev->metadata_type))
4474 namelen = sizeof(mddev->metadata_type)-1;
4475 strncpy(mddev->metadata_type, buf+9, namelen);
4476 mddev->metadata_type[namelen] = 0;
4477 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4478 mddev->metadata_type[--namelen] = 0;
4479 mddev->persistent = 0;
4480 mddev->external = 1;
4481 mddev->major_version = 0;
4482 mddev->minor_version = 90;
4485 major = simple_strtoul(buf, &e, 10);
4487 if (e==buf || *e != '.')
4490 minor = simple_strtoul(buf, &e, 10);
4491 if (e==buf || (*e && *e != '\n') )
4494 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4496 mddev->major_version = major;
4497 mddev->minor_version = minor;
4498 mddev->persistent = 1;
4499 mddev->external = 0;
4502 mddev_unlock(mddev);
4506 static struct md_sysfs_entry md_metadata =
4507 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4510 action_show(struct mddev *mddev, char *page)
4512 char *type = "idle";
4513 unsigned long recovery = mddev->recovery;
4514 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4516 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4517 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4518 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4520 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4521 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4523 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4527 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4529 else if (mddev->reshape_position != MaxSector)
4532 return sprintf(page, "%s\n", type);
4536 action_store(struct mddev *mddev, const char *page, size_t len)
4538 if (!mddev->pers || !mddev->pers->sync_request)
4542 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4543 if (cmd_match(page, "frozen"))
4544 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4546 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4547 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4548 mddev_lock(mddev) == 0) {
4549 flush_workqueue(md_misc_wq);
4550 if (mddev->sync_thread) {
4551 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4552 md_reap_sync_thread(mddev);
4554 mddev_unlock(mddev);
4556 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4558 else if (cmd_match(page, "resync"))
4559 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4560 else if (cmd_match(page, "recover")) {
4561 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4562 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4563 } else if (cmd_match(page, "reshape")) {
4565 if (mddev->pers->start_reshape == NULL)
4567 err = mddev_lock(mddev);
4569 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4572 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4573 err = mddev->pers->start_reshape(mddev);
4575 mddev_unlock(mddev);
4579 sysfs_notify(&mddev->kobj, NULL, "degraded");
4581 if (cmd_match(page, "check"))
4582 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4583 else if (!cmd_match(page, "repair"))
4585 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4586 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4587 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4589 if (mddev->ro == 2) {
4590 /* A write to sync_action is enough to justify
4591 * canceling read-auto mode
4594 md_wakeup_thread(mddev->sync_thread);
4596 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4597 md_wakeup_thread(mddev->thread);
4598 sysfs_notify_dirent_safe(mddev->sysfs_action);
4602 static struct md_sysfs_entry md_scan_mode =
4603 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4606 last_sync_action_show(struct mddev *mddev, char *page)
4608 return sprintf(page, "%s\n", mddev->last_sync_action);
4611 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4614 mismatch_cnt_show(struct mddev *mddev, char *page)
4616 return sprintf(page, "%llu\n",
4617 (unsigned long long)
4618 atomic64_read(&mddev->resync_mismatches));
4621 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4624 sync_min_show(struct mddev *mddev, char *page)
4626 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4627 mddev->sync_speed_min ? "local": "system");
4631 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4636 if (strncmp(buf, "system", 6)==0) {
4639 rv = kstrtouint(buf, 10, &min);
4645 mddev->sync_speed_min = min;
4649 static struct md_sysfs_entry md_sync_min =
4650 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4653 sync_max_show(struct mddev *mddev, char *page)
4655 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4656 mddev->sync_speed_max ? "local": "system");
4660 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4665 if (strncmp(buf, "system", 6)==0) {
4668 rv = kstrtouint(buf, 10, &max);
4674 mddev->sync_speed_max = max;
4678 static struct md_sysfs_entry md_sync_max =
4679 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4682 degraded_show(struct mddev *mddev, char *page)
4684 return sprintf(page, "%d\n", mddev->degraded);
4686 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4689 sync_force_parallel_show(struct mddev *mddev, char *page)
4691 return sprintf(page, "%d\n", mddev->parallel_resync);
4695 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4699 if (kstrtol(buf, 10, &n))
4702 if (n != 0 && n != 1)
4705 mddev->parallel_resync = n;
4707 if (mddev->sync_thread)
4708 wake_up(&resync_wait);
4713 /* force parallel resync, even with shared block devices */
4714 static struct md_sysfs_entry md_sync_force_parallel =
4715 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4716 sync_force_parallel_show, sync_force_parallel_store);
4719 sync_speed_show(struct mddev *mddev, char *page)
4721 unsigned long resync, dt, db;
4722 if (mddev->curr_resync == 0)
4723 return sprintf(page, "none\n");
4724 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4725 dt = (jiffies - mddev->resync_mark) / HZ;
4727 db = resync - mddev->resync_mark_cnt;
4728 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4731 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4734 sync_completed_show(struct mddev *mddev, char *page)
4736 unsigned long long max_sectors, resync;
4738 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4739 return sprintf(page, "none\n");
4741 if (mddev->curr_resync == 1 ||
4742 mddev->curr_resync == 2)
4743 return sprintf(page, "delayed\n");
4745 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4746 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4747 max_sectors = mddev->resync_max_sectors;
4749 max_sectors = mddev->dev_sectors;
4751 resync = mddev->curr_resync_completed;
4752 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4755 static struct md_sysfs_entry md_sync_completed =
4756 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4759 min_sync_show(struct mddev *mddev, char *page)
4761 return sprintf(page, "%llu\n",
4762 (unsigned long long)mddev->resync_min);
4765 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4767 unsigned long long min;
4770 if (kstrtoull(buf, 10, &min))
4773 spin_lock(&mddev->lock);
4775 if (min > mddev->resync_max)
4779 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4782 /* Round down to multiple of 4K for safety */
4783 mddev->resync_min = round_down(min, 8);
4787 spin_unlock(&mddev->lock);
4791 static struct md_sysfs_entry md_min_sync =
4792 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4795 max_sync_show(struct mddev *mddev, char *page)
4797 if (mddev->resync_max == MaxSector)
4798 return sprintf(page, "max\n");
4800 return sprintf(page, "%llu\n",
4801 (unsigned long long)mddev->resync_max);
4804 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4807 spin_lock(&mddev->lock);
4808 if (strncmp(buf, "max", 3) == 0)
4809 mddev->resync_max = MaxSector;
4811 unsigned long long max;
4815 if (kstrtoull(buf, 10, &max))
4817 if (max < mddev->resync_min)
4821 if (max < mddev->resync_max &&
4823 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4826 /* Must be a multiple of chunk_size */
4827 chunk = mddev->chunk_sectors;
4829 sector_t temp = max;
4832 if (sector_div(temp, chunk))
4835 mddev->resync_max = max;
4837 wake_up(&mddev->recovery_wait);
4840 spin_unlock(&mddev->lock);
4844 static struct md_sysfs_entry md_max_sync =
4845 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4848 suspend_lo_show(struct mddev *mddev, char *page)
4850 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4854 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4856 unsigned long long new;
4859 err = kstrtoull(buf, 10, &new);
4862 if (new != (sector_t)new)
4865 err = mddev_lock(mddev);
4869 if (mddev->pers == NULL ||
4870 mddev->pers->quiesce == NULL)
4872 mddev_suspend(mddev);
4873 mddev->suspend_lo = new;
4874 mddev_resume(mddev);
4878 mddev_unlock(mddev);
4881 static struct md_sysfs_entry md_suspend_lo =
4882 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4885 suspend_hi_show(struct mddev *mddev, char *page)
4887 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4891 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4893 unsigned long long new;
4896 err = kstrtoull(buf, 10, &new);
4899 if (new != (sector_t)new)
4902 err = mddev_lock(mddev);
4906 if (mddev->pers == NULL)
4909 mddev_suspend(mddev);
4910 mddev->suspend_hi = new;
4911 mddev_resume(mddev);
4915 mddev_unlock(mddev);
4918 static struct md_sysfs_entry md_suspend_hi =
4919 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4922 reshape_position_show(struct mddev *mddev, char *page)
4924 if (mddev->reshape_position != MaxSector)
4925 return sprintf(page, "%llu\n",
4926 (unsigned long long)mddev->reshape_position);
4927 strcpy(page, "none\n");
4932 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4934 struct md_rdev *rdev;
4935 unsigned long long new;
4938 err = kstrtoull(buf, 10, &new);
4941 if (new != (sector_t)new)
4943 err = mddev_lock(mddev);
4949 mddev->reshape_position = new;
4950 mddev->delta_disks = 0;
4951 mddev->reshape_backwards = 0;
4952 mddev->new_level = mddev->level;
4953 mddev->new_layout = mddev->layout;
4954 mddev->new_chunk_sectors = mddev->chunk_sectors;
4955 rdev_for_each(rdev, mddev)
4956 rdev->new_data_offset = rdev->data_offset;
4959 mddev_unlock(mddev);
4963 static struct md_sysfs_entry md_reshape_position =
4964 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4965 reshape_position_store);
4968 reshape_direction_show(struct mddev *mddev, char *page)
4970 return sprintf(page, "%s\n",
4971 mddev->reshape_backwards ? "backwards" : "forwards");
4975 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4980 if (cmd_match(buf, "forwards"))
4982 else if (cmd_match(buf, "backwards"))
4986 if (mddev->reshape_backwards == backwards)
4989 err = mddev_lock(mddev);
4992 /* check if we are allowed to change */
4993 if (mddev->delta_disks)
4995 else if (mddev->persistent &&
4996 mddev->major_version == 0)
4999 mddev->reshape_backwards = backwards;
5000 mddev_unlock(mddev);
5004 static struct md_sysfs_entry md_reshape_direction =
5005 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5006 reshape_direction_store);
5009 array_size_show(struct mddev *mddev, char *page)
5011 if (mddev->external_size)
5012 return sprintf(page, "%llu\n",
5013 (unsigned long long)mddev->array_sectors/2);
5015 return sprintf(page, "default\n");
5019 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5024 err = mddev_lock(mddev);
5028 /* cluster raid doesn't support change array_sectors */
5029 if (mddev_is_clustered(mddev)) {
5030 mddev_unlock(mddev);
5034 if (strncmp(buf, "default", 7) == 0) {
5036 sectors = mddev->pers->size(mddev, 0, 0);
5038 sectors = mddev->array_sectors;
5040 mddev->external_size = 0;
5042 if (strict_blocks_to_sectors(buf, §ors) < 0)
5044 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5047 mddev->external_size = 1;
5051 mddev->array_sectors = sectors;
5053 set_capacity(mddev->gendisk, mddev->array_sectors);
5054 revalidate_disk(mddev->gendisk);
5057 mddev_unlock(mddev);
5061 static struct md_sysfs_entry md_array_size =
5062 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5066 consistency_policy_show(struct mddev *mddev, char *page)
5070 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5071 ret = sprintf(page, "journal\n");
5072 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5073 ret = sprintf(page, "ppl\n");
5074 } else if (mddev->bitmap) {
5075 ret = sprintf(page, "bitmap\n");
5076 } else if (mddev->pers) {
5077 if (mddev->pers->sync_request)
5078 ret = sprintf(page, "resync\n");
5080 ret = sprintf(page, "none\n");
5082 ret = sprintf(page, "unknown\n");
5089 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5094 if (mddev->pers->change_consistency_policy)
5095 err = mddev->pers->change_consistency_policy(mddev, buf);
5098 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5099 set_bit(MD_HAS_PPL, &mddev->flags);
5104 return err ? err : len;
5107 static struct md_sysfs_entry md_consistency_policy =
5108 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5109 consistency_policy_store);
5111 static struct attribute *md_default_attrs[] = {
5114 &md_raid_disks.attr,
5115 &md_chunk_size.attr,
5117 &md_resync_start.attr,
5119 &md_new_device.attr,
5120 &md_safe_delay.attr,
5121 &md_array_state.attr,
5122 &md_reshape_position.attr,
5123 &md_reshape_direction.attr,
5124 &md_array_size.attr,
5125 &max_corr_read_errors.attr,
5126 &md_consistency_policy.attr,
5130 static struct attribute *md_redundancy_attrs[] = {
5132 &md_last_scan_mode.attr,
5133 &md_mismatches.attr,
5136 &md_sync_speed.attr,
5137 &md_sync_force_parallel.attr,
5138 &md_sync_completed.attr,
5141 &md_suspend_lo.attr,
5142 &md_suspend_hi.attr,
5147 static struct attribute_group md_redundancy_group = {
5149 .attrs = md_redundancy_attrs,
5153 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5155 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5156 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5161 spin_lock(&all_mddevs_lock);
5162 if (list_empty(&mddev->all_mddevs)) {
5163 spin_unlock(&all_mddevs_lock);
5167 spin_unlock(&all_mddevs_lock);
5169 rv = entry->show(mddev, page);
5175 md_attr_store(struct kobject *kobj, struct attribute *attr,
5176 const char *page, size_t length)
5178 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5179 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5184 if (!capable(CAP_SYS_ADMIN))
5186 spin_lock(&all_mddevs_lock);
5187 if (list_empty(&mddev->all_mddevs)) {
5188 spin_unlock(&all_mddevs_lock);
5192 spin_unlock(&all_mddevs_lock);
5193 rv = entry->store(mddev, page, length);
5198 static void md_free(struct kobject *ko)
5200 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5202 if (mddev->sysfs_state)
5203 sysfs_put(mddev->sysfs_state);
5206 blk_cleanup_queue(mddev->queue);
5207 if (mddev->gendisk) {
5208 del_gendisk(mddev->gendisk);
5209 put_disk(mddev->gendisk);
5211 percpu_ref_exit(&mddev->writes_pending);
5216 static const struct sysfs_ops md_sysfs_ops = {
5217 .show = md_attr_show,
5218 .store = md_attr_store,
5220 static struct kobj_type md_ktype = {
5222 .sysfs_ops = &md_sysfs_ops,
5223 .default_attrs = md_default_attrs,
5228 static void mddev_delayed_delete(struct work_struct *ws)
5230 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5232 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5233 kobject_del(&mddev->kobj);
5234 kobject_put(&mddev->kobj);
5237 static void no_op(struct percpu_ref *r) {}
5239 int mddev_init_writes_pending(struct mddev *mddev)
5241 if (mddev->writes_pending.percpu_count_ptr)
5243 if (percpu_ref_init(&mddev->writes_pending, no_op, 0, GFP_KERNEL) < 0)
5245 /* We want to start with the refcount at zero */
5246 percpu_ref_put(&mddev->writes_pending);
5249 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5251 static int md_alloc(dev_t dev, char *name)
5254 * If dev is zero, name is the name of a device to allocate with
5255 * an arbitrary minor number. It will be "md_???"
5256 * If dev is non-zero it must be a device number with a MAJOR of
5257 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5258 * the device is being created by opening a node in /dev.
5259 * If "name" is not NULL, the device is being created by
5260 * writing to /sys/module/md_mod/parameters/new_array.
5262 static DEFINE_MUTEX(disks_mutex);
5263 struct mddev *mddev = mddev_find(dev);
5264 struct gendisk *disk;
5273 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5274 shift = partitioned ? MdpMinorShift : 0;
5275 unit = MINOR(mddev->unit) >> shift;
5277 /* wait for any previous instance of this device to be
5278 * completely removed (mddev_delayed_delete).
5280 flush_workqueue(md_misc_wq);
5282 mutex_lock(&disks_mutex);
5288 /* Need to ensure that 'name' is not a duplicate.
5290 struct mddev *mddev2;
5291 spin_lock(&all_mddevs_lock);
5293 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5294 if (mddev2->gendisk &&
5295 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5296 spin_unlock(&all_mddevs_lock);
5299 spin_unlock(&all_mddevs_lock);
5303 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5305 mddev->hold_active = UNTIL_STOP;
5308 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5311 mddev->queue->queuedata = mddev;
5313 blk_queue_make_request(mddev->queue, md_make_request);
5314 blk_set_stacking_limits(&mddev->queue->limits);
5316 disk = alloc_disk(1 << shift);
5318 blk_cleanup_queue(mddev->queue);
5319 mddev->queue = NULL;
5322 disk->major = MAJOR(mddev->unit);
5323 disk->first_minor = unit << shift;
5325 strcpy(disk->disk_name, name);
5326 else if (partitioned)
5327 sprintf(disk->disk_name, "md_d%d", unit);
5329 sprintf(disk->disk_name, "md%d", unit);
5330 disk->fops = &md_fops;
5331 disk->private_data = mddev;
5332 disk->queue = mddev->queue;
5333 blk_queue_write_cache(mddev->queue, true, true);
5334 /* Allow extended partitions. This makes the
5335 * 'mdp' device redundant, but we can't really
5338 disk->flags |= GENHD_FL_EXT_DEVT;
5339 mddev->gendisk = disk;
5340 /* As soon as we call add_disk(), another thread could get
5341 * through to md_open, so make sure it doesn't get too far
5343 mutex_lock(&mddev->open_mutex);
5346 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5347 &disk_to_dev(disk)->kobj, "%s", "md");
5349 /* This isn't possible, but as kobject_init_and_add is marked
5350 * __must_check, we must do something with the result
5352 pr_debug("md: cannot register %s/md - name in use\n",
5356 if (mddev->kobj.sd &&
5357 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5358 pr_debug("pointless warning\n");
5359 mutex_unlock(&mddev->open_mutex);
5361 mutex_unlock(&disks_mutex);
5362 if (!error && mddev->kobj.sd) {
5363 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5364 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5370 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5373 md_alloc(dev, NULL);
5377 static int add_named_array(const char *val, const struct kernel_param *kp)
5380 * val must be "md_*" or "mdNNN".
5381 * For "md_*" we allocate an array with a large free minor number, and
5382 * set the name to val. val must not already be an active name.
5383 * For "mdNNN" we allocate an array with the minor number NNN
5384 * which must not already be in use.
5386 int len = strlen(val);
5387 char buf[DISK_NAME_LEN];
5388 unsigned long devnum;
5390 while (len && val[len-1] == '\n')
5392 if (len >= DISK_NAME_LEN)
5394 strlcpy(buf, val, len+1);
5395 if (strncmp(buf, "md_", 3) == 0)
5396 return md_alloc(0, buf);
5397 if (strncmp(buf, "md", 2) == 0 &&
5399 kstrtoul(buf+2, 10, &devnum) == 0 &&
5400 devnum <= MINORMASK)
5401 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5406 static void md_safemode_timeout(struct timer_list *t)
5408 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5410 mddev->safemode = 1;
5411 if (mddev->external)
5412 sysfs_notify_dirent_safe(mddev->sysfs_state);
5414 md_wakeup_thread(mddev->thread);
5417 static int start_dirty_degraded;
5419 int md_run(struct mddev *mddev)
5422 struct md_rdev *rdev;
5423 struct md_personality *pers;
5425 if (list_empty(&mddev->disks))
5426 /* cannot run an array with no devices.. */
5431 /* Cannot run until previous stop completes properly */
5432 if (mddev->sysfs_active)
5436 * Analyze all RAID superblock(s)
5438 if (!mddev->raid_disks) {
5439 if (!mddev->persistent)
5444 if (mddev->level != LEVEL_NONE)
5445 request_module("md-level-%d", mddev->level);
5446 else if (mddev->clevel[0])
5447 request_module("md-%s", mddev->clevel);
5450 * Drop all container device buffers, from now on
5451 * the only valid external interface is through the md
5454 rdev_for_each(rdev, mddev) {
5455 if (test_bit(Faulty, &rdev->flags))
5457 sync_blockdev(rdev->bdev);
5458 invalidate_bdev(rdev->bdev);
5459 if (mddev->ro != 1 &&
5460 (bdev_read_only(rdev->bdev) ||
5461 bdev_read_only(rdev->meta_bdev))) {
5464 set_disk_ro(mddev->gendisk, 1);
5467 /* perform some consistency tests on the device.
5468 * We don't want the data to overlap the metadata,
5469 * Internal Bitmap issues have been handled elsewhere.
5471 if (rdev->meta_bdev) {
5472 /* Nothing to check */;
5473 } else if (rdev->data_offset < rdev->sb_start) {
5474 if (mddev->dev_sectors &&
5475 rdev->data_offset + mddev->dev_sectors
5477 pr_warn("md: %s: data overlaps metadata\n",
5482 if (rdev->sb_start + rdev->sb_size/512
5483 > rdev->data_offset) {
5484 pr_warn("md: %s: metadata overlaps data\n",
5489 sysfs_notify_dirent_safe(rdev->sysfs_state);
5492 if (mddev->bio_set == NULL) {
5493 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5494 if (!mddev->bio_set)
5497 if (mddev->sync_set == NULL) {
5498 mddev->sync_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5499 if (!mddev->sync_set)
5503 spin_lock(&pers_lock);
5504 pers = find_pers(mddev->level, mddev->clevel);
5505 if (!pers || !try_module_get(pers->owner)) {
5506 spin_unlock(&pers_lock);
5507 if (mddev->level != LEVEL_NONE)
5508 pr_warn("md: personality for level %d is not loaded!\n",
5511 pr_warn("md: personality for level %s is not loaded!\n",
5515 spin_unlock(&pers_lock);
5516 if (mddev->level != pers->level) {
5517 mddev->level = pers->level;
5518 mddev->new_level = pers->level;
5520 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5522 if (mddev->reshape_position != MaxSector &&
5523 pers->start_reshape == NULL) {
5524 /* This personality cannot handle reshaping... */
5525 module_put(pers->owner);
5529 if (pers->sync_request) {
5530 /* Warn if this is a potentially silly
5533 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5534 struct md_rdev *rdev2;
5537 rdev_for_each(rdev, mddev)
5538 rdev_for_each(rdev2, mddev) {
5540 rdev->bdev->bd_contains ==
5541 rdev2->bdev->bd_contains) {
5542 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5544 bdevname(rdev->bdev,b),
5545 bdevname(rdev2->bdev,b2));
5551 pr_warn("True protection against single-disk failure might be compromised.\n");
5554 mddev->recovery = 0;
5555 /* may be over-ridden by personality */
5556 mddev->resync_max_sectors = mddev->dev_sectors;
5558 mddev->ok_start_degraded = start_dirty_degraded;
5560 if (start_readonly && mddev->ro == 0)
5561 mddev->ro = 2; /* read-only, but switch on first write */
5564 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
5565 * up mddev->thread. It is important to initialize critical
5566 * resources for mddev->thread BEFORE calling pers->run().
5568 err = pers->run(mddev);
5570 pr_warn("md: pers->run() failed ...\n");
5571 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5572 WARN_ONCE(!mddev->external_size,
5573 "%s: default size too small, but 'external_size' not in effect?\n",
5575 pr_warn("md: invalid array_size %llu > default size %llu\n",
5576 (unsigned long long)mddev->array_sectors / 2,
5577 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5580 if (err == 0 && pers->sync_request &&
5581 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5582 struct bitmap *bitmap;
5584 bitmap = bitmap_create(mddev, -1);
5585 if (IS_ERR(bitmap)) {
5586 err = PTR_ERR(bitmap);
5587 pr_warn("%s: failed to create bitmap (%d)\n",
5588 mdname(mddev), err);
5590 mddev->bitmap = bitmap;
5594 mddev_detach(mddev);
5596 pers->free(mddev, mddev->private);
5597 mddev->private = NULL;
5598 module_put(pers->owner);
5599 bitmap_destroy(mddev);
5605 rdev_for_each(rdev, mddev) {
5606 if (rdev->raid_disk >= 0 &&
5607 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5612 if (mddev->degraded)
5615 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5617 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, mddev->queue);
5618 mddev->queue->backing_dev_info->congested_data = mddev;
5619 mddev->queue->backing_dev_info->congested_fn = md_congested;
5621 if (pers->sync_request) {
5622 if (mddev->kobj.sd &&
5623 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5624 pr_warn("md: cannot register extra attributes for %s\n",
5626 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5627 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5630 atomic_set(&mddev->max_corr_read_errors,
5631 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5632 mddev->safemode = 0;
5633 if (mddev_is_clustered(mddev))
5634 mddev->safemode_delay = 0;
5636 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5639 spin_lock(&mddev->lock);
5641 spin_unlock(&mddev->lock);
5642 rdev_for_each(rdev, mddev)
5643 if (rdev->raid_disk >= 0)
5644 if (sysfs_link_rdev(mddev, rdev))
5645 /* failure here is OK */;
5647 if (mddev->degraded && !mddev->ro)
5648 /* This ensures that recovering status is reported immediately
5649 * via sysfs - until a lack of spares is confirmed.
5651 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5652 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5654 if (mddev->sb_flags)
5655 md_update_sb(mddev, 0);
5657 md_new_event(mddev);
5658 sysfs_notify_dirent_safe(mddev->sysfs_state);
5659 sysfs_notify_dirent_safe(mddev->sysfs_action);
5660 sysfs_notify(&mddev->kobj, NULL, "degraded");
5663 EXPORT_SYMBOL_GPL(md_run);
5665 static int do_md_run(struct mddev *mddev)
5669 err = md_run(mddev);
5672 err = bitmap_load(mddev);
5674 bitmap_destroy(mddev);
5678 if (mddev_is_clustered(mddev))
5679 md_allow_write(mddev);
5681 md_wakeup_thread(mddev->thread);
5682 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5684 set_capacity(mddev->gendisk, mddev->array_sectors);
5685 revalidate_disk(mddev->gendisk);
5687 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5692 static int restart_array(struct mddev *mddev)
5694 struct gendisk *disk = mddev->gendisk;
5695 struct md_rdev *rdev;
5696 bool has_journal = false;
5697 bool has_readonly = false;
5699 /* Complain if it has no devices */
5700 if (list_empty(&mddev->disks))
5708 rdev_for_each_rcu(rdev, mddev) {
5709 if (test_bit(Journal, &rdev->flags) &&
5710 !test_bit(Faulty, &rdev->flags))
5712 if (bdev_read_only(rdev->bdev))
5713 has_readonly = true;
5716 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5717 /* Don't restart rw with journal missing/faulty */
5722 mddev->safemode = 0;
5724 set_disk_ro(disk, 0);
5725 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5726 /* Kick recovery or resync if necessary */
5727 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5728 md_wakeup_thread(mddev->thread);
5729 md_wakeup_thread(mddev->sync_thread);
5730 sysfs_notify_dirent_safe(mddev->sysfs_state);
5734 static void md_clean(struct mddev *mddev)
5736 mddev->array_sectors = 0;
5737 mddev->external_size = 0;
5738 mddev->dev_sectors = 0;
5739 mddev->raid_disks = 0;
5740 mddev->recovery_cp = 0;
5741 mddev->resync_min = 0;
5742 mddev->resync_max = MaxSector;
5743 mddev->reshape_position = MaxSector;
5744 mddev->external = 0;
5745 mddev->persistent = 0;
5746 mddev->level = LEVEL_NONE;
5747 mddev->clevel[0] = 0;
5749 mddev->sb_flags = 0;
5751 mddev->metadata_type[0] = 0;
5752 mddev->chunk_sectors = 0;
5753 mddev->ctime = mddev->utime = 0;
5755 mddev->max_disks = 0;
5757 mddev->can_decrease_events = 0;
5758 mddev->delta_disks = 0;
5759 mddev->reshape_backwards = 0;
5760 mddev->new_level = LEVEL_NONE;
5761 mddev->new_layout = 0;
5762 mddev->new_chunk_sectors = 0;
5763 mddev->curr_resync = 0;
5764 atomic64_set(&mddev->resync_mismatches, 0);
5765 mddev->suspend_lo = mddev->suspend_hi = 0;
5766 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5767 mddev->recovery = 0;
5770 mddev->degraded = 0;
5771 mddev->safemode = 0;
5772 mddev->private = NULL;
5773 mddev->cluster_info = NULL;
5774 mddev->bitmap_info.offset = 0;
5775 mddev->bitmap_info.default_offset = 0;
5776 mddev->bitmap_info.default_space = 0;
5777 mddev->bitmap_info.chunksize = 0;
5778 mddev->bitmap_info.daemon_sleep = 0;
5779 mddev->bitmap_info.max_write_behind = 0;
5780 mddev->bitmap_info.nodes = 0;
5783 static void __md_stop_writes(struct mddev *mddev)
5785 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5786 flush_workqueue(md_misc_wq);
5787 if (mddev->sync_thread) {
5788 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5789 md_reap_sync_thread(mddev);
5792 del_timer_sync(&mddev->safemode_timer);
5794 if (mddev->pers && mddev->pers->quiesce) {
5795 mddev->pers->quiesce(mddev, 1);
5796 mddev->pers->quiesce(mddev, 0);
5798 bitmap_flush(mddev);
5800 if (mddev->ro == 0 &&
5801 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5803 /* mark array as shutdown cleanly */
5804 if (!mddev_is_clustered(mddev))
5806 md_update_sb(mddev, 1);
5810 void md_stop_writes(struct mddev *mddev)
5812 mddev_lock_nointr(mddev);
5813 __md_stop_writes(mddev);
5814 mddev_unlock(mddev);
5816 EXPORT_SYMBOL_GPL(md_stop_writes);
5818 static void mddev_detach(struct mddev *mddev)
5820 bitmap_wait_behind_writes(mddev);
5821 if (mddev->pers && mddev->pers->quiesce) {
5822 mddev->pers->quiesce(mddev, 1);
5823 mddev->pers->quiesce(mddev, 0);
5825 md_unregister_thread(&mddev->thread);
5827 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5830 static void __md_stop(struct mddev *mddev)
5832 struct md_personality *pers = mddev->pers;
5833 bitmap_destroy(mddev);
5834 mddev_detach(mddev);
5835 /* Ensure ->event_work is done */
5836 flush_workqueue(md_misc_wq);
5837 spin_lock(&mddev->lock);
5839 spin_unlock(&mddev->lock);
5840 pers->free(mddev, mddev->private);
5841 mddev->private = NULL;
5842 if (pers->sync_request && mddev->to_remove == NULL)
5843 mddev->to_remove = &md_redundancy_group;
5844 module_put(pers->owner);
5845 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5848 void md_stop(struct mddev *mddev)
5850 /* stop the array and free an attached data structures.
5851 * This is called from dm-raid
5854 if (mddev->bio_set) {
5855 bioset_free(mddev->bio_set);
5856 mddev->bio_set = NULL;
5858 if (mddev->sync_set) {
5859 bioset_free(mddev->sync_set);
5860 mddev->sync_set = NULL;
5864 EXPORT_SYMBOL_GPL(md_stop);
5866 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5871 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5873 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5874 md_wakeup_thread(mddev->thread);
5876 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5877 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5878 if (mddev->sync_thread)
5879 /* Thread might be blocked waiting for metadata update
5880 * which will now never happen */
5881 wake_up_process(mddev->sync_thread->tsk);
5883 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
5885 mddev_unlock(mddev);
5886 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5888 wait_event(mddev->sb_wait,
5889 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
5890 mddev_lock_nointr(mddev);
5892 mutex_lock(&mddev->open_mutex);
5893 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5894 mddev->sync_thread ||
5895 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5896 pr_warn("md: %s still in use.\n",mdname(mddev));
5898 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5899 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5900 md_wakeup_thread(mddev->thread);
5906 __md_stop_writes(mddev);
5912 set_disk_ro(mddev->gendisk, 1);
5913 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5914 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5915 md_wakeup_thread(mddev->thread);
5916 sysfs_notify_dirent_safe(mddev->sysfs_state);
5920 mutex_unlock(&mddev->open_mutex);
5925 * 0 - completely stop and dis-assemble array
5926 * 2 - stop but do not disassemble array
5928 static int do_md_stop(struct mddev *mddev, int mode,
5929 struct block_device *bdev)
5931 struct gendisk *disk = mddev->gendisk;
5932 struct md_rdev *rdev;
5935 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5937 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5938 md_wakeup_thread(mddev->thread);
5940 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5941 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5942 if (mddev->sync_thread)
5943 /* Thread might be blocked waiting for metadata update
5944 * which will now never happen */
5945 wake_up_process(mddev->sync_thread->tsk);
5947 mddev_unlock(mddev);
5948 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5949 !test_bit(MD_RECOVERY_RUNNING,
5950 &mddev->recovery)));
5951 mddev_lock_nointr(mddev);
5953 mutex_lock(&mddev->open_mutex);
5954 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5955 mddev->sysfs_active ||
5956 mddev->sync_thread ||
5957 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
5958 pr_warn("md: %s still in use.\n",mdname(mddev));
5959 mutex_unlock(&mddev->open_mutex);
5961 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5962 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5963 md_wakeup_thread(mddev->thread);
5969 set_disk_ro(disk, 0);
5971 __md_stop_writes(mddev);
5973 mddev->queue->backing_dev_info->congested_fn = NULL;
5975 /* tell userspace to handle 'inactive' */
5976 sysfs_notify_dirent_safe(mddev->sysfs_state);
5978 rdev_for_each(rdev, mddev)
5979 if (rdev->raid_disk >= 0)
5980 sysfs_unlink_rdev(mddev, rdev);
5982 set_capacity(disk, 0);
5983 mutex_unlock(&mddev->open_mutex);
5985 revalidate_disk(disk);
5990 mutex_unlock(&mddev->open_mutex);
5992 * Free resources if final stop
5995 pr_info("md: %s stopped.\n", mdname(mddev));
5997 if (mddev->bitmap_info.file) {
5998 struct file *f = mddev->bitmap_info.file;
5999 spin_lock(&mddev->lock);
6000 mddev->bitmap_info.file = NULL;
6001 spin_unlock(&mddev->lock);
6004 mddev->bitmap_info.offset = 0;
6006 export_array(mddev);
6009 if (mddev->hold_active == UNTIL_STOP)
6010 mddev->hold_active = 0;
6012 md_new_event(mddev);
6013 sysfs_notify_dirent_safe(mddev->sysfs_state);
6018 static void autorun_array(struct mddev *mddev)
6020 struct md_rdev *rdev;
6023 if (list_empty(&mddev->disks))
6026 pr_info("md: running: ");
6028 rdev_for_each(rdev, mddev) {
6029 char b[BDEVNAME_SIZE];
6030 pr_cont("<%s>", bdevname(rdev->bdev,b));
6034 err = do_md_run(mddev);
6036 pr_warn("md: do_md_run() returned %d\n", err);
6037 do_md_stop(mddev, 0, NULL);
6042 * lets try to run arrays based on all disks that have arrived
6043 * until now. (those are in pending_raid_disks)
6045 * the method: pick the first pending disk, collect all disks with
6046 * the same UUID, remove all from the pending list and put them into
6047 * the 'same_array' list. Then order this list based on superblock
6048 * update time (freshest comes first), kick out 'old' disks and
6049 * compare superblocks. If everything's fine then run it.
6051 * If "unit" is allocated, then bump its reference count
6053 static void autorun_devices(int part)
6055 struct md_rdev *rdev0, *rdev, *tmp;
6056 struct mddev *mddev;
6057 char b[BDEVNAME_SIZE];
6059 pr_info("md: autorun ...\n");
6060 while (!list_empty(&pending_raid_disks)) {
6063 LIST_HEAD(candidates);
6064 rdev0 = list_entry(pending_raid_disks.next,
6065 struct md_rdev, same_set);
6067 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6068 INIT_LIST_HEAD(&candidates);
6069 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6070 if (super_90_load(rdev, rdev0, 0) >= 0) {
6071 pr_debug("md: adding %s ...\n",
6072 bdevname(rdev->bdev,b));
6073 list_move(&rdev->same_set, &candidates);
6076 * now we have a set of devices, with all of them having
6077 * mostly sane superblocks. It's time to allocate the
6081 dev = MKDEV(mdp_major,
6082 rdev0->preferred_minor << MdpMinorShift);
6083 unit = MINOR(dev) >> MdpMinorShift;
6085 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6088 if (rdev0->preferred_minor != unit) {
6089 pr_warn("md: unit number in %s is bad: %d\n",
6090 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6094 md_probe(dev, NULL, NULL);
6095 mddev = mddev_find(dev);
6096 if (!mddev || !mddev->gendisk) {
6101 if (mddev_lock(mddev))
6102 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6103 else if (mddev->raid_disks || mddev->major_version
6104 || !list_empty(&mddev->disks)) {
6105 pr_warn("md: %s already running, cannot run %s\n",
6106 mdname(mddev), bdevname(rdev0->bdev,b));
6107 mddev_unlock(mddev);
6109 pr_debug("md: created %s\n", mdname(mddev));
6110 mddev->persistent = 1;
6111 rdev_for_each_list(rdev, tmp, &candidates) {
6112 list_del_init(&rdev->same_set);
6113 if (bind_rdev_to_array(rdev, mddev))
6116 autorun_array(mddev);
6117 mddev_unlock(mddev);
6119 /* on success, candidates will be empty, on error
6122 rdev_for_each_list(rdev, tmp, &candidates) {
6123 list_del_init(&rdev->same_set);
6128 pr_info("md: ... autorun DONE.\n");
6130 #endif /* !MODULE */
6132 static int get_version(void __user *arg)
6136 ver.major = MD_MAJOR_VERSION;
6137 ver.minor = MD_MINOR_VERSION;
6138 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6140 if (copy_to_user(arg, &ver, sizeof(ver)))
6146 static int get_array_info(struct mddev *mddev, void __user *arg)
6148 mdu_array_info_t info;
6149 int nr,working,insync,failed,spare;
6150 struct md_rdev *rdev;
6152 nr = working = insync = failed = spare = 0;
6154 rdev_for_each_rcu(rdev, mddev) {
6156 if (test_bit(Faulty, &rdev->flags))
6160 if (test_bit(In_sync, &rdev->flags))
6162 else if (test_bit(Journal, &rdev->flags))
6163 /* TODO: add journal count to md_u.h */
6171 info.major_version = mddev->major_version;
6172 info.minor_version = mddev->minor_version;
6173 info.patch_version = MD_PATCHLEVEL_VERSION;
6174 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6175 info.level = mddev->level;
6176 info.size = mddev->dev_sectors / 2;
6177 if (info.size != mddev->dev_sectors / 2) /* overflow */
6180 info.raid_disks = mddev->raid_disks;
6181 info.md_minor = mddev->md_minor;
6182 info.not_persistent= !mddev->persistent;
6184 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6187 info.state = (1<<MD_SB_CLEAN);
6188 if (mddev->bitmap && mddev->bitmap_info.offset)
6189 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6190 if (mddev_is_clustered(mddev))
6191 info.state |= (1<<MD_SB_CLUSTERED);
6192 info.active_disks = insync;
6193 info.working_disks = working;
6194 info.failed_disks = failed;
6195 info.spare_disks = spare;
6197 info.layout = mddev->layout;
6198 info.chunk_size = mddev->chunk_sectors << 9;
6200 if (copy_to_user(arg, &info, sizeof(info)))
6206 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6208 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6212 file = kzalloc(sizeof(*file), GFP_NOIO);
6217 spin_lock(&mddev->lock);
6218 /* bitmap enabled */
6219 if (mddev->bitmap_info.file) {
6220 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6221 sizeof(file->pathname));
6225 memmove(file->pathname, ptr,
6226 sizeof(file->pathname)-(ptr-file->pathname));
6228 spin_unlock(&mddev->lock);
6231 copy_to_user(arg, file, sizeof(*file)))
6238 static int get_disk_info(struct mddev *mddev, void __user * arg)
6240 mdu_disk_info_t info;
6241 struct md_rdev *rdev;
6243 if (copy_from_user(&info, arg, sizeof(info)))
6247 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6249 info.major = MAJOR(rdev->bdev->bd_dev);
6250 info.minor = MINOR(rdev->bdev->bd_dev);
6251 info.raid_disk = rdev->raid_disk;
6253 if (test_bit(Faulty, &rdev->flags))
6254 info.state |= (1<<MD_DISK_FAULTY);
6255 else if (test_bit(In_sync, &rdev->flags)) {
6256 info.state |= (1<<MD_DISK_ACTIVE);
6257 info.state |= (1<<MD_DISK_SYNC);
6259 if (test_bit(Journal, &rdev->flags))
6260 info.state |= (1<<MD_DISK_JOURNAL);
6261 if (test_bit(WriteMostly, &rdev->flags))
6262 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6263 if (test_bit(FailFast, &rdev->flags))
6264 info.state |= (1<<MD_DISK_FAILFAST);
6266 info.major = info.minor = 0;
6267 info.raid_disk = -1;
6268 info.state = (1<<MD_DISK_REMOVED);
6272 if (copy_to_user(arg, &info, sizeof(info)))
6278 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6280 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6281 struct md_rdev *rdev;
6282 dev_t dev = MKDEV(info->major,info->minor);
6284 if (mddev_is_clustered(mddev) &&
6285 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6286 pr_warn("%s: Cannot add to clustered mddev.\n",
6291 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6294 if (!mddev->raid_disks) {
6296 /* expecting a device which has a superblock */
6297 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6299 pr_warn("md: md_import_device returned %ld\n",
6301 return PTR_ERR(rdev);
6303 if (!list_empty(&mddev->disks)) {
6304 struct md_rdev *rdev0
6305 = list_entry(mddev->disks.next,
6306 struct md_rdev, same_set);
6307 err = super_types[mddev->major_version]
6308 .load_super(rdev, rdev0, mddev->minor_version);
6310 pr_warn("md: %s has different UUID to %s\n",
6311 bdevname(rdev->bdev,b),
6312 bdevname(rdev0->bdev,b2));
6317 err = bind_rdev_to_array(rdev, mddev);
6324 * add_new_disk can be used once the array is assembled
6325 * to add "hot spares". They must already have a superblock
6330 if (!mddev->pers->hot_add_disk) {
6331 pr_warn("%s: personality does not support diskops!\n",
6335 if (mddev->persistent)
6336 rdev = md_import_device(dev, mddev->major_version,
6337 mddev->minor_version);
6339 rdev = md_import_device(dev, -1, -1);
6341 pr_warn("md: md_import_device returned %ld\n",
6343 return PTR_ERR(rdev);
6345 /* set saved_raid_disk if appropriate */
6346 if (!mddev->persistent) {
6347 if (info->state & (1<<MD_DISK_SYNC) &&
6348 info->raid_disk < mddev->raid_disks) {
6349 rdev->raid_disk = info->raid_disk;
6350 set_bit(In_sync, &rdev->flags);
6351 clear_bit(Bitmap_sync, &rdev->flags);
6353 rdev->raid_disk = -1;
6354 rdev->saved_raid_disk = rdev->raid_disk;
6356 super_types[mddev->major_version].
6357 validate_super(mddev, rdev);
6358 if ((info->state & (1<<MD_DISK_SYNC)) &&
6359 rdev->raid_disk != info->raid_disk) {
6360 /* This was a hot-add request, but events doesn't
6361 * match, so reject it.
6367 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6368 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6369 set_bit(WriteMostly, &rdev->flags);
6371 clear_bit(WriteMostly, &rdev->flags);
6372 if (info->state & (1<<MD_DISK_FAILFAST))
6373 set_bit(FailFast, &rdev->flags);
6375 clear_bit(FailFast, &rdev->flags);
6377 if (info->state & (1<<MD_DISK_JOURNAL)) {
6378 struct md_rdev *rdev2;
6379 bool has_journal = false;
6381 /* make sure no existing journal disk */
6382 rdev_for_each(rdev2, mddev) {
6383 if (test_bit(Journal, &rdev2->flags)) {
6388 if (has_journal || mddev->bitmap) {
6392 set_bit(Journal, &rdev->flags);
6395 * check whether the device shows up in other nodes
6397 if (mddev_is_clustered(mddev)) {
6398 if (info->state & (1 << MD_DISK_CANDIDATE))
6399 set_bit(Candidate, &rdev->flags);
6400 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6401 /* --add initiated by this node */
6402 err = md_cluster_ops->add_new_disk(mddev, rdev);
6410 rdev->raid_disk = -1;
6411 err = bind_rdev_to_array(rdev, mddev);
6416 if (mddev_is_clustered(mddev)) {
6417 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6419 err = md_cluster_ops->new_disk_ack(mddev,
6422 md_kick_rdev_from_array(rdev);
6426 md_cluster_ops->add_new_disk_cancel(mddev);
6428 err = add_bound_rdev(rdev);
6432 err = add_bound_rdev(rdev);
6437 /* otherwise, add_new_disk is only allowed
6438 * for major_version==0 superblocks
6440 if (mddev->major_version != 0) {
6441 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6445 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6447 rdev = md_import_device(dev, -1, 0);
6449 pr_warn("md: error, md_import_device() returned %ld\n",
6451 return PTR_ERR(rdev);
6453 rdev->desc_nr = info->number;
6454 if (info->raid_disk < mddev->raid_disks)
6455 rdev->raid_disk = info->raid_disk;
6457 rdev->raid_disk = -1;
6459 if (rdev->raid_disk < mddev->raid_disks)
6460 if (info->state & (1<<MD_DISK_SYNC))
6461 set_bit(In_sync, &rdev->flags);
6463 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6464 set_bit(WriteMostly, &rdev->flags);
6465 if (info->state & (1<<MD_DISK_FAILFAST))
6466 set_bit(FailFast, &rdev->flags);
6468 if (!mddev->persistent) {
6469 pr_debug("md: nonpersistent superblock ...\n");
6470 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6472 rdev->sb_start = calc_dev_sboffset(rdev);
6473 rdev->sectors = rdev->sb_start;
6475 err = bind_rdev_to_array(rdev, mddev);
6485 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6487 char b[BDEVNAME_SIZE];
6488 struct md_rdev *rdev;
6490 rdev = find_rdev(mddev, dev);
6494 if (rdev->raid_disk < 0)
6497 clear_bit(Blocked, &rdev->flags);
6498 remove_and_add_spares(mddev, rdev);
6500 if (rdev->raid_disk >= 0)
6504 if (mddev_is_clustered(mddev))
6505 md_cluster_ops->remove_disk(mddev, rdev);
6507 md_kick_rdev_from_array(rdev);
6508 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6510 md_wakeup_thread(mddev->thread);
6512 md_update_sb(mddev, 1);
6513 md_new_event(mddev);
6517 pr_debug("md: cannot remove active disk %s from %s ...\n",
6518 bdevname(rdev->bdev,b), mdname(mddev));
6522 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6524 char b[BDEVNAME_SIZE];
6526 struct md_rdev *rdev;
6531 if (mddev->major_version != 0) {
6532 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6536 if (!mddev->pers->hot_add_disk) {
6537 pr_warn("%s: personality does not support diskops!\n",
6542 rdev = md_import_device(dev, -1, 0);
6544 pr_warn("md: error, md_import_device() returned %ld\n",
6549 if (mddev->persistent)
6550 rdev->sb_start = calc_dev_sboffset(rdev);
6552 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6554 rdev->sectors = rdev->sb_start;
6556 if (test_bit(Faulty, &rdev->flags)) {
6557 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6558 bdevname(rdev->bdev,b), mdname(mddev));
6563 clear_bit(In_sync, &rdev->flags);
6565 rdev->saved_raid_disk = -1;
6566 err = bind_rdev_to_array(rdev, mddev);
6571 * The rest should better be atomic, we can have disk failures
6572 * noticed in interrupt contexts ...
6575 rdev->raid_disk = -1;
6577 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6579 md_update_sb(mddev, 1);
6581 * Kick recovery, maybe this spare has to be added to the
6582 * array immediately.
6584 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6585 md_wakeup_thread(mddev->thread);
6586 md_new_event(mddev);
6594 static int set_bitmap_file(struct mddev *mddev, int fd)
6599 if (!mddev->pers->quiesce || !mddev->thread)
6601 if (mddev->recovery || mddev->sync_thread)
6603 /* we should be able to change the bitmap.. */
6607 struct inode *inode;
6610 if (mddev->bitmap || mddev->bitmap_info.file)
6611 return -EEXIST; /* cannot add when bitmap is present */
6615 pr_warn("%s: error: failed to get bitmap file\n",
6620 inode = f->f_mapping->host;
6621 if (!S_ISREG(inode->i_mode)) {
6622 pr_warn("%s: error: bitmap file must be a regular file\n",
6625 } else if (!(f->f_mode & FMODE_WRITE)) {
6626 pr_warn("%s: error: bitmap file must open for write\n",
6629 } else if (atomic_read(&inode->i_writecount) != 1) {
6630 pr_warn("%s: error: bitmap file is already in use\n",
6638 mddev->bitmap_info.file = f;
6639 mddev->bitmap_info.offset = 0; /* file overrides offset */
6640 } else if (mddev->bitmap == NULL)
6641 return -ENOENT; /* cannot remove what isn't there */
6645 struct bitmap *bitmap;
6647 bitmap = bitmap_create(mddev, -1);
6648 mddev_suspend(mddev);
6649 if (!IS_ERR(bitmap)) {
6650 mddev->bitmap = bitmap;
6651 err = bitmap_load(mddev);
6653 err = PTR_ERR(bitmap);
6655 bitmap_destroy(mddev);
6658 mddev_resume(mddev);
6659 } else if (fd < 0) {
6660 mddev_suspend(mddev);
6661 bitmap_destroy(mddev);
6662 mddev_resume(mddev);
6666 struct file *f = mddev->bitmap_info.file;
6668 spin_lock(&mddev->lock);
6669 mddev->bitmap_info.file = NULL;
6670 spin_unlock(&mddev->lock);
6679 * set_array_info is used two different ways
6680 * The original usage is when creating a new array.
6681 * In this usage, raid_disks is > 0 and it together with
6682 * level, size, not_persistent,layout,chunksize determine the
6683 * shape of the array.
6684 * This will always create an array with a type-0.90.0 superblock.
6685 * The newer usage is when assembling an array.
6686 * In this case raid_disks will be 0, and the major_version field is
6687 * use to determine which style super-blocks are to be found on the devices.
6688 * The minor and patch _version numbers are also kept incase the
6689 * super_block handler wishes to interpret them.
6691 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6694 if (info->raid_disks == 0) {
6695 /* just setting version number for superblock loading */
6696 if (info->major_version < 0 ||
6697 info->major_version >= ARRAY_SIZE(super_types) ||
6698 super_types[info->major_version].name == NULL) {
6699 /* maybe try to auto-load a module? */
6700 pr_warn("md: superblock version %d not known\n",
6701 info->major_version);
6704 mddev->major_version = info->major_version;
6705 mddev->minor_version = info->minor_version;
6706 mddev->patch_version = info->patch_version;
6707 mddev->persistent = !info->not_persistent;
6708 /* ensure mddev_put doesn't delete this now that there
6709 * is some minimal configuration.
6711 mddev->ctime = ktime_get_real_seconds();
6714 mddev->major_version = MD_MAJOR_VERSION;
6715 mddev->minor_version = MD_MINOR_VERSION;
6716 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6717 mddev->ctime = ktime_get_real_seconds();
6719 mddev->level = info->level;
6720 mddev->clevel[0] = 0;
6721 mddev->dev_sectors = 2 * (sector_t)info->size;
6722 mddev->raid_disks = info->raid_disks;
6723 /* don't set md_minor, it is determined by which /dev/md* was
6726 if (info->state & (1<<MD_SB_CLEAN))
6727 mddev->recovery_cp = MaxSector;
6729 mddev->recovery_cp = 0;
6730 mddev->persistent = ! info->not_persistent;
6731 mddev->external = 0;
6733 mddev->layout = info->layout;
6734 mddev->chunk_sectors = info->chunk_size >> 9;
6736 if (mddev->persistent) {
6737 mddev->max_disks = MD_SB_DISKS;
6739 mddev->sb_flags = 0;
6741 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6743 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6744 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6745 mddev->bitmap_info.offset = 0;
6747 mddev->reshape_position = MaxSector;
6750 * Generate a 128 bit UUID
6752 get_random_bytes(mddev->uuid, 16);
6754 mddev->new_level = mddev->level;
6755 mddev->new_chunk_sectors = mddev->chunk_sectors;
6756 mddev->new_layout = mddev->layout;
6757 mddev->delta_disks = 0;
6758 mddev->reshape_backwards = 0;
6763 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6765 lockdep_assert_held(&mddev->reconfig_mutex);
6767 if (mddev->external_size)
6770 mddev->array_sectors = array_sectors;
6772 EXPORT_SYMBOL(md_set_array_sectors);
6774 static int update_size(struct mddev *mddev, sector_t num_sectors)
6776 struct md_rdev *rdev;
6778 int fit = (num_sectors == 0);
6779 sector_t old_dev_sectors = mddev->dev_sectors;
6781 if (mddev->pers->resize == NULL)
6783 /* The "num_sectors" is the number of sectors of each device that
6784 * is used. This can only make sense for arrays with redundancy.
6785 * linear and raid0 always use whatever space is available. We can only
6786 * consider changing this number if no resync or reconstruction is
6787 * happening, and if the new size is acceptable. It must fit before the
6788 * sb_start or, if that is <data_offset, it must fit before the size
6789 * of each device. If num_sectors is zero, we find the largest size
6792 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6798 rdev_for_each(rdev, mddev) {
6799 sector_t avail = rdev->sectors;
6801 if (fit && (num_sectors == 0 || num_sectors > avail))
6802 num_sectors = avail;
6803 if (avail < num_sectors)
6806 rv = mddev->pers->resize(mddev, num_sectors);
6808 if (mddev_is_clustered(mddev))
6809 md_cluster_ops->update_size(mddev, old_dev_sectors);
6810 else if (mddev->queue) {
6811 set_capacity(mddev->gendisk, mddev->array_sectors);
6812 revalidate_disk(mddev->gendisk);
6818 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6821 struct md_rdev *rdev;
6822 /* change the number of raid disks */
6823 if (mddev->pers->check_reshape == NULL)
6827 if (raid_disks <= 0 ||
6828 (mddev->max_disks && raid_disks >= mddev->max_disks))
6830 if (mddev->sync_thread ||
6831 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6832 mddev->reshape_position != MaxSector)
6835 rdev_for_each(rdev, mddev) {
6836 if (mddev->raid_disks < raid_disks &&
6837 rdev->data_offset < rdev->new_data_offset)
6839 if (mddev->raid_disks > raid_disks &&
6840 rdev->data_offset > rdev->new_data_offset)
6844 mddev->delta_disks = raid_disks - mddev->raid_disks;
6845 if (mddev->delta_disks < 0)
6846 mddev->reshape_backwards = 1;
6847 else if (mddev->delta_disks > 0)
6848 mddev->reshape_backwards = 0;
6850 rv = mddev->pers->check_reshape(mddev);
6852 mddev->delta_disks = 0;
6853 mddev->reshape_backwards = 0;
6859 * update_array_info is used to change the configuration of an
6861 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6862 * fields in the info are checked against the array.
6863 * Any differences that cannot be handled will cause an error.
6864 * Normally, only one change can be managed at a time.
6866 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6872 /* calculate expected state,ignoring low bits */
6873 if (mddev->bitmap && mddev->bitmap_info.offset)
6874 state |= (1 << MD_SB_BITMAP_PRESENT);
6876 if (mddev->major_version != info->major_version ||
6877 mddev->minor_version != info->minor_version ||
6878 /* mddev->patch_version != info->patch_version || */
6879 mddev->ctime != info->ctime ||
6880 mddev->level != info->level ||
6881 /* mddev->layout != info->layout || */
6882 mddev->persistent != !info->not_persistent ||
6883 mddev->chunk_sectors != info->chunk_size >> 9 ||
6884 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6885 ((state^info->state) & 0xfffffe00)
6888 /* Check there is only one change */
6889 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6891 if (mddev->raid_disks != info->raid_disks)
6893 if (mddev->layout != info->layout)
6895 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6902 if (mddev->layout != info->layout) {
6904 * we don't need to do anything at the md level, the
6905 * personality will take care of it all.
6907 if (mddev->pers->check_reshape == NULL)
6910 mddev->new_layout = info->layout;
6911 rv = mddev->pers->check_reshape(mddev);
6913 mddev->new_layout = mddev->layout;
6917 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6918 rv = update_size(mddev, (sector_t)info->size * 2);
6920 if (mddev->raid_disks != info->raid_disks)
6921 rv = update_raid_disks(mddev, info->raid_disks);
6923 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6924 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6928 if (mddev->recovery || mddev->sync_thread) {
6932 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6933 struct bitmap *bitmap;
6934 /* add the bitmap */
6935 if (mddev->bitmap) {
6939 if (mddev->bitmap_info.default_offset == 0) {
6943 mddev->bitmap_info.offset =
6944 mddev->bitmap_info.default_offset;
6945 mddev->bitmap_info.space =
6946 mddev->bitmap_info.default_space;
6947 bitmap = bitmap_create(mddev, -1);
6948 mddev_suspend(mddev);
6949 if (!IS_ERR(bitmap)) {
6950 mddev->bitmap = bitmap;
6951 rv = bitmap_load(mddev);
6953 rv = PTR_ERR(bitmap);
6955 bitmap_destroy(mddev);
6956 mddev_resume(mddev);
6958 /* remove the bitmap */
6959 if (!mddev->bitmap) {
6963 if (mddev->bitmap->storage.file) {
6967 if (mddev->bitmap_info.nodes) {
6968 /* hold PW on all the bitmap lock */
6969 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6970 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
6972 md_cluster_ops->unlock_all_bitmaps(mddev);
6976 mddev->bitmap_info.nodes = 0;
6977 md_cluster_ops->leave(mddev);
6979 mddev_suspend(mddev);
6980 bitmap_destroy(mddev);
6981 mddev_resume(mddev);
6982 mddev->bitmap_info.offset = 0;
6985 md_update_sb(mddev, 1);
6991 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6993 struct md_rdev *rdev;
6996 if (mddev->pers == NULL)
7000 rdev = find_rdev_rcu(mddev, dev);
7004 md_error(mddev, rdev);
7005 if (!test_bit(Faulty, &rdev->flags))
7013 * We have a problem here : there is no easy way to give a CHS
7014 * virtual geometry. We currently pretend that we have a 2 heads
7015 * 4 sectors (with a BIG number of cylinders...). This drives
7016 * dosfs just mad... ;-)
7018 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7020 struct mddev *mddev = bdev->bd_disk->private_data;
7024 geo->cylinders = mddev->array_sectors / 8;
7028 static inline bool md_ioctl_valid(unsigned int cmd)
7033 case GET_ARRAY_INFO:
7034 case GET_BITMAP_FILE:
7037 case HOT_REMOVE_DISK:
7040 case RESTART_ARRAY_RW:
7042 case SET_ARRAY_INFO:
7043 case SET_BITMAP_FILE:
7044 case SET_DISK_FAULTY:
7047 case CLUSTERED_DISK_NACK:
7054 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7055 unsigned int cmd, unsigned long arg)
7058 void __user *argp = (void __user *)arg;
7059 struct mddev *mddev = NULL;
7061 bool did_set_md_closing = false;
7063 if (!md_ioctl_valid(cmd))
7068 case GET_ARRAY_INFO:
7072 if (!capable(CAP_SYS_ADMIN))
7077 * Commands dealing with the RAID driver but not any
7082 err = get_version(argp);
7088 autostart_arrays(arg);
7095 * Commands creating/starting a new array:
7098 mddev = bdev->bd_disk->private_data;
7105 /* Some actions do not requires the mutex */
7107 case GET_ARRAY_INFO:
7108 if (!mddev->raid_disks && !mddev->external)
7111 err = get_array_info(mddev, argp);
7115 if (!mddev->raid_disks && !mddev->external)
7118 err = get_disk_info(mddev, argp);
7121 case SET_DISK_FAULTY:
7122 err = set_disk_faulty(mddev, new_decode_dev(arg));
7125 case GET_BITMAP_FILE:
7126 err = get_bitmap_file(mddev, argp);
7131 if (cmd == ADD_NEW_DISK)
7132 /* need to ensure md_delayed_delete() has completed */
7133 flush_workqueue(md_misc_wq);
7135 if (cmd == HOT_REMOVE_DISK)
7136 /* need to ensure recovery thread has run */
7137 wait_event_interruptible_timeout(mddev->sb_wait,
7138 !test_bit(MD_RECOVERY_NEEDED,
7140 msecs_to_jiffies(5000));
7141 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7142 /* Need to flush page cache, and ensure no-one else opens
7145 mutex_lock(&mddev->open_mutex);
7146 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7147 mutex_unlock(&mddev->open_mutex);
7151 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7152 set_bit(MD_CLOSING, &mddev->flags);
7153 did_set_md_closing = true;
7154 mutex_unlock(&mddev->open_mutex);
7155 sync_blockdev(bdev);
7157 err = mddev_lock(mddev);
7159 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7164 if (cmd == SET_ARRAY_INFO) {
7165 mdu_array_info_t info;
7167 memset(&info, 0, sizeof(info));
7168 else if (copy_from_user(&info, argp, sizeof(info))) {
7173 err = update_array_info(mddev, &info);
7175 pr_warn("md: couldn't update array info. %d\n", err);
7180 if (!list_empty(&mddev->disks)) {
7181 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7185 if (mddev->raid_disks) {
7186 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7190 err = set_array_info(mddev, &info);
7192 pr_warn("md: couldn't set array info. %d\n", err);
7199 * Commands querying/configuring an existing array:
7201 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7202 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7203 if ((!mddev->raid_disks && !mddev->external)
7204 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7205 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7206 && cmd != GET_BITMAP_FILE) {
7212 * Commands even a read-only array can execute:
7215 case RESTART_ARRAY_RW:
7216 err = restart_array(mddev);
7220 err = do_md_stop(mddev, 0, bdev);
7224 err = md_set_readonly(mddev, bdev);
7227 case HOT_REMOVE_DISK:
7228 err = hot_remove_disk(mddev, new_decode_dev(arg));
7232 /* We can support ADD_NEW_DISK on read-only arrays
7233 * only if we are re-adding a preexisting device.
7234 * So require mddev->pers and MD_DISK_SYNC.
7237 mdu_disk_info_t info;
7238 if (copy_from_user(&info, argp, sizeof(info)))
7240 else if (!(info.state & (1<<MD_DISK_SYNC)))
7241 /* Need to clear read-only for this */
7244 err = add_new_disk(mddev, &info);
7250 if (get_user(ro, (int __user *)(arg))) {
7256 /* if the bdev is going readonly the value of mddev->ro
7257 * does not matter, no writes are coming
7262 /* are we are already prepared for writes? */
7266 /* transitioning to readauto need only happen for
7267 * arrays that call md_write_start
7270 err = restart_array(mddev);
7273 set_disk_ro(mddev->gendisk, 0);
7280 * The remaining ioctls are changing the state of the
7281 * superblock, so we do not allow them on read-only arrays.
7283 if (mddev->ro && mddev->pers) {
7284 if (mddev->ro == 2) {
7286 sysfs_notify_dirent_safe(mddev->sysfs_state);
7287 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7288 /* mddev_unlock will wake thread */
7289 /* If a device failed while we were read-only, we
7290 * need to make sure the metadata is updated now.
7292 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7293 mddev_unlock(mddev);
7294 wait_event(mddev->sb_wait,
7295 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7296 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7297 mddev_lock_nointr(mddev);
7308 mdu_disk_info_t info;
7309 if (copy_from_user(&info, argp, sizeof(info)))
7312 err = add_new_disk(mddev, &info);
7316 case CLUSTERED_DISK_NACK:
7317 if (mddev_is_clustered(mddev))
7318 md_cluster_ops->new_disk_ack(mddev, false);
7324 err = hot_add_disk(mddev, new_decode_dev(arg));
7328 err = do_md_run(mddev);
7331 case SET_BITMAP_FILE:
7332 err = set_bitmap_file(mddev, (int)arg);
7341 if (mddev->hold_active == UNTIL_IOCTL &&
7343 mddev->hold_active = 0;
7344 mddev_unlock(mddev);
7346 if(did_set_md_closing)
7347 clear_bit(MD_CLOSING, &mddev->flags);
7350 #ifdef CONFIG_COMPAT
7351 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7352 unsigned int cmd, unsigned long arg)
7355 case HOT_REMOVE_DISK:
7357 case SET_DISK_FAULTY:
7358 case SET_BITMAP_FILE:
7359 /* These take in integer arg, do not convert */
7362 arg = (unsigned long)compat_ptr(arg);
7366 return md_ioctl(bdev, mode, cmd, arg);
7368 #endif /* CONFIG_COMPAT */
7370 static int md_open(struct block_device *bdev, fmode_t mode)
7373 * Succeed if we can lock the mddev, which confirms that
7374 * it isn't being stopped right now.
7376 struct mddev *mddev = mddev_find(bdev->bd_dev);
7382 if (mddev->gendisk != bdev->bd_disk) {
7383 /* we are racing with mddev_put which is discarding this
7387 /* Wait until bdev->bd_disk is definitely gone */
7388 flush_workqueue(md_misc_wq);
7389 /* Then retry the open from the top */
7390 return -ERESTARTSYS;
7392 BUG_ON(mddev != bdev->bd_disk->private_data);
7394 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7397 if (test_bit(MD_CLOSING, &mddev->flags)) {
7398 mutex_unlock(&mddev->open_mutex);
7404 atomic_inc(&mddev->openers);
7405 mutex_unlock(&mddev->open_mutex);
7407 check_disk_change(bdev);
7414 static void md_release(struct gendisk *disk, fmode_t mode)
7416 struct mddev *mddev = disk->private_data;
7419 atomic_dec(&mddev->openers);
7423 static int md_media_changed(struct gendisk *disk)
7425 struct mddev *mddev = disk->private_data;
7427 return mddev->changed;
7430 static int md_revalidate(struct gendisk *disk)
7432 struct mddev *mddev = disk->private_data;
7437 static const struct block_device_operations md_fops =
7439 .owner = THIS_MODULE,
7441 .release = md_release,
7443 #ifdef CONFIG_COMPAT
7444 .compat_ioctl = md_compat_ioctl,
7446 .getgeo = md_getgeo,
7447 .media_changed = md_media_changed,
7448 .revalidate_disk= md_revalidate,
7451 static int md_thread(void *arg)
7453 struct md_thread *thread = arg;
7456 * md_thread is a 'system-thread', it's priority should be very
7457 * high. We avoid resource deadlocks individually in each
7458 * raid personality. (RAID5 does preallocation) We also use RR and
7459 * the very same RT priority as kswapd, thus we will never get
7460 * into a priority inversion deadlock.
7462 * we definitely have to have equal or higher priority than
7463 * bdflush, otherwise bdflush will deadlock if there are too
7464 * many dirty RAID5 blocks.
7467 allow_signal(SIGKILL);
7468 while (!kthread_should_stop()) {
7470 /* We need to wait INTERRUPTIBLE so that
7471 * we don't add to the load-average.
7472 * That means we need to be sure no signals are
7475 if (signal_pending(current))
7476 flush_signals(current);
7478 wait_event_interruptible_timeout
7480 test_bit(THREAD_WAKEUP, &thread->flags)
7481 || kthread_should_stop() || kthread_should_park(),
7484 clear_bit(THREAD_WAKEUP, &thread->flags);
7485 if (kthread_should_park())
7487 if (!kthread_should_stop())
7488 thread->run(thread);
7494 void md_wakeup_thread(struct md_thread *thread)
7497 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7498 set_bit(THREAD_WAKEUP, &thread->flags);
7499 wake_up(&thread->wqueue);
7502 EXPORT_SYMBOL(md_wakeup_thread);
7504 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7505 struct mddev *mddev, const char *name)
7507 struct md_thread *thread;
7509 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7513 init_waitqueue_head(&thread->wqueue);
7516 thread->mddev = mddev;
7517 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7518 thread->tsk = kthread_run(md_thread, thread,
7520 mdname(thread->mddev),
7522 if (IS_ERR(thread->tsk)) {
7528 EXPORT_SYMBOL(md_register_thread);
7530 void md_unregister_thread(struct md_thread **threadp)
7532 struct md_thread *thread = *threadp;
7535 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7536 /* Locking ensures that mddev_unlock does not wake_up a
7537 * non-existent thread
7539 spin_lock(&pers_lock);
7541 spin_unlock(&pers_lock);
7543 kthread_stop(thread->tsk);
7546 EXPORT_SYMBOL(md_unregister_thread);
7548 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7550 if (!rdev || test_bit(Faulty, &rdev->flags))
7553 if (!mddev->pers || !mddev->pers->error_handler)
7555 mddev->pers->error_handler(mddev,rdev);
7556 if (mddev->degraded)
7557 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7558 sysfs_notify_dirent_safe(rdev->sysfs_state);
7559 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7560 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7561 md_wakeup_thread(mddev->thread);
7562 if (mddev->event_work.func)
7563 queue_work(md_misc_wq, &mddev->event_work);
7564 md_new_event(mddev);
7566 EXPORT_SYMBOL(md_error);
7568 /* seq_file implementation /proc/mdstat */
7570 static void status_unused(struct seq_file *seq)
7573 struct md_rdev *rdev;
7575 seq_printf(seq, "unused devices: ");
7577 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7578 char b[BDEVNAME_SIZE];
7580 seq_printf(seq, "%s ",
7581 bdevname(rdev->bdev,b));
7584 seq_printf(seq, "<none>");
7586 seq_printf(seq, "\n");
7589 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7591 sector_t max_sectors, resync, res;
7592 unsigned long dt, db;
7595 unsigned int per_milli;
7597 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7598 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7599 max_sectors = mddev->resync_max_sectors;
7601 max_sectors = mddev->dev_sectors;
7603 resync = mddev->curr_resync;
7605 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7606 /* Still cleaning up */
7607 resync = max_sectors;
7608 } else if (resync > max_sectors)
7609 resync = max_sectors;
7611 resync -= atomic_read(&mddev->recovery_active);
7614 if (mddev->recovery_cp < MaxSector) {
7615 seq_printf(seq, "\tresync=PENDING");
7621 seq_printf(seq, "\tresync=DELAYED");
7625 WARN_ON(max_sectors == 0);
7626 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7627 * in a sector_t, and (max_sectors>>scale) will fit in a
7628 * u32, as those are the requirements for sector_div.
7629 * Thus 'scale' must be at least 10
7632 if (sizeof(sector_t) > sizeof(unsigned long)) {
7633 while ( max_sectors/2 > (1ULL<<(scale+32)))
7636 res = (resync>>scale)*1000;
7637 sector_div(res, (u32)((max_sectors>>scale)+1));
7641 int i, x = per_milli/50, y = 20-x;
7642 seq_printf(seq, "[");
7643 for (i = 0; i < x; i++)
7644 seq_printf(seq, "=");
7645 seq_printf(seq, ">");
7646 for (i = 0; i < y; i++)
7647 seq_printf(seq, ".");
7648 seq_printf(seq, "] ");
7650 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7651 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7653 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7655 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7656 "resync" : "recovery"))),
7657 per_milli/10, per_milli % 10,
7658 (unsigned long long) resync/2,
7659 (unsigned long long) max_sectors/2);
7662 * dt: time from mark until now
7663 * db: blocks written from mark until now
7664 * rt: remaining time
7666 * rt is a sector_t, so could be 32bit or 64bit.
7667 * So we divide before multiply in case it is 32bit and close
7669 * We scale the divisor (db) by 32 to avoid losing precision
7670 * near the end of resync when the number of remaining sectors
7672 * We then divide rt by 32 after multiplying by db to compensate.
7673 * The '+1' avoids division by zero if db is very small.
7675 dt = ((jiffies - mddev->resync_mark) / HZ);
7677 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7678 - mddev->resync_mark_cnt;
7680 rt = max_sectors - resync; /* number of remaining sectors */
7681 sector_div(rt, db/32+1);
7685 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7686 ((unsigned long)rt % 60)/6);
7688 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7692 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7694 struct list_head *tmp;
7696 struct mddev *mddev;
7704 spin_lock(&all_mddevs_lock);
7705 list_for_each(tmp,&all_mddevs)
7707 mddev = list_entry(tmp, struct mddev, all_mddevs);
7709 spin_unlock(&all_mddevs_lock);
7712 spin_unlock(&all_mddevs_lock);
7714 return (void*)2;/* tail */
7718 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7720 struct list_head *tmp;
7721 struct mddev *next_mddev, *mddev = v;
7727 spin_lock(&all_mddevs_lock);
7729 tmp = all_mddevs.next;
7731 tmp = mddev->all_mddevs.next;
7732 if (tmp != &all_mddevs)
7733 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7735 next_mddev = (void*)2;
7738 spin_unlock(&all_mddevs_lock);
7746 static void md_seq_stop(struct seq_file *seq, void *v)
7748 struct mddev *mddev = v;
7750 if (mddev && v != (void*)1 && v != (void*)2)
7754 static int md_seq_show(struct seq_file *seq, void *v)
7756 struct mddev *mddev = v;
7758 struct md_rdev *rdev;
7760 if (v == (void*)1) {
7761 struct md_personality *pers;
7762 seq_printf(seq, "Personalities : ");
7763 spin_lock(&pers_lock);
7764 list_for_each_entry(pers, &pers_list, list)
7765 seq_printf(seq, "[%s] ", pers->name);
7767 spin_unlock(&pers_lock);
7768 seq_printf(seq, "\n");
7769 seq->poll_event = atomic_read(&md_event_count);
7772 if (v == (void*)2) {
7777 spin_lock(&mddev->lock);
7778 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7779 seq_printf(seq, "%s : %sactive", mdname(mddev),
7780 mddev->pers ? "" : "in");
7783 seq_printf(seq, " (read-only)");
7785 seq_printf(seq, " (auto-read-only)");
7786 seq_printf(seq, " %s", mddev->pers->name);
7791 rdev_for_each_rcu(rdev, mddev) {
7792 char b[BDEVNAME_SIZE];
7793 seq_printf(seq, " %s[%d]",
7794 bdevname(rdev->bdev,b), rdev->desc_nr);
7795 if (test_bit(WriteMostly, &rdev->flags))
7796 seq_printf(seq, "(W)");
7797 if (test_bit(Journal, &rdev->flags))
7798 seq_printf(seq, "(J)");
7799 if (test_bit(Faulty, &rdev->flags)) {
7800 seq_printf(seq, "(F)");
7803 if (rdev->raid_disk < 0)
7804 seq_printf(seq, "(S)"); /* spare */
7805 if (test_bit(Replacement, &rdev->flags))
7806 seq_printf(seq, "(R)");
7807 sectors += rdev->sectors;
7811 if (!list_empty(&mddev->disks)) {
7813 seq_printf(seq, "\n %llu blocks",
7814 (unsigned long long)
7815 mddev->array_sectors / 2);
7817 seq_printf(seq, "\n %llu blocks",
7818 (unsigned long long)sectors / 2);
7820 if (mddev->persistent) {
7821 if (mddev->major_version != 0 ||
7822 mddev->minor_version != 90) {
7823 seq_printf(seq," super %d.%d",
7824 mddev->major_version,
7825 mddev->minor_version);
7827 } else if (mddev->external)
7828 seq_printf(seq, " super external:%s",
7829 mddev->metadata_type);
7831 seq_printf(seq, " super non-persistent");
7834 mddev->pers->status(seq, mddev);
7835 seq_printf(seq, "\n ");
7836 if (mddev->pers->sync_request) {
7837 if (status_resync(seq, mddev))
7838 seq_printf(seq, "\n ");
7841 seq_printf(seq, "\n ");
7843 bitmap_status(seq, mddev->bitmap);
7845 seq_printf(seq, "\n");
7847 spin_unlock(&mddev->lock);
7852 static const struct seq_operations md_seq_ops = {
7853 .start = md_seq_start,
7854 .next = md_seq_next,
7855 .stop = md_seq_stop,
7856 .show = md_seq_show,
7859 static int md_seq_open(struct inode *inode, struct file *file)
7861 struct seq_file *seq;
7864 error = seq_open(file, &md_seq_ops);
7868 seq = file->private_data;
7869 seq->poll_event = atomic_read(&md_event_count);
7873 static int md_unloading;
7874 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7876 struct seq_file *seq = filp->private_data;
7880 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7881 poll_wait(filp, &md_event_waiters, wait);
7883 /* always allow read */
7884 mask = POLLIN | POLLRDNORM;
7886 if (seq->poll_event != atomic_read(&md_event_count))
7887 mask |= POLLERR | POLLPRI;
7891 static const struct file_operations md_seq_fops = {
7892 .owner = THIS_MODULE,
7893 .open = md_seq_open,
7895 .llseek = seq_lseek,
7896 .release = seq_release,
7897 .poll = mdstat_poll,
7900 int register_md_personality(struct md_personality *p)
7902 pr_debug("md: %s personality registered for level %d\n",
7904 spin_lock(&pers_lock);
7905 list_add_tail(&p->list, &pers_list);
7906 spin_unlock(&pers_lock);
7909 EXPORT_SYMBOL(register_md_personality);
7911 int unregister_md_personality(struct md_personality *p)
7913 pr_debug("md: %s personality unregistered\n", p->name);
7914 spin_lock(&pers_lock);
7915 list_del_init(&p->list);
7916 spin_unlock(&pers_lock);
7919 EXPORT_SYMBOL(unregister_md_personality);
7921 int register_md_cluster_operations(struct md_cluster_operations *ops,
7922 struct module *module)
7925 spin_lock(&pers_lock);
7926 if (md_cluster_ops != NULL)
7929 md_cluster_ops = ops;
7930 md_cluster_mod = module;
7932 spin_unlock(&pers_lock);
7935 EXPORT_SYMBOL(register_md_cluster_operations);
7937 int unregister_md_cluster_operations(void)
7939 spin_lock(&pers_lock);
7940 md_cluster_ops = NULL;
7941 spin_unlock(&pers_lock);
7944 EXPORT_SYMBOL(unregister_md_cluster_operations);
7946 int md_setup_cluster(struct mddev *mddev, int nodes)
7948 if (!md_cluster_ops)
7949 request_module("md-cluster");
7950 spin_lock(&pers_lock);
7951 /* ensure module won't be unloaded */
7952 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7953 pr_warn("can't find md-cluster module or get it's reference.\n");
7954 spin_unlock(&pers_lock);
7957 spin_unlock(&pers_lock);
7959 return md_cluster_ops->join(mddev, nodes);
7962 void md_cluster_stop(struct mddev *mddev)
7964 if (!md_cluster_ops)
7966 md_cluster_ops->leave(mddev);
7967 module_put(md_cluster_mod);
7970 static int is_mddev_idle(struct mddev *mddev, int init)
7972 struct md_rdev *rdev;
7978 rdev_for_each_rcu(rdev, mddev) {
7979 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7980 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7981 (int)part_stat_read(&disk->part0, sectors[1]) -
7982 atomic_read(&disk->sync_io);
7983 /* sync IO will cause sync_io to increase before the disk_stats
7984 * as sync_io is counted when a request starts, and
7985 * disk_stats is counted when it completes.
7986 * So resync activity will cause curr_events to be smaller than
7987 * when there was no such activity.
7988 * non-sync IO will cause disk_stat to increase without
7989 * increasing sync_io so curr_events will (eventually)
7990 * be larger than it was before. Once it becomes
7991 * substantially larger, the test below will cause
7992 * the array to appear non-idle, and resync will slow
7994 * If there is a lot of outstanding resync activity when
7995 * we set last_event to curr_events, then all that activity
7996 * completing might cause the array to appear non-idle
7997 * and resync will be slowed down even though there might
7998 * not have been non-resync activity. This will only
7999 * happen once though. 'last_events' will soon reflect
8000 * the state where there is little or no outstanding
8001 * resync requests, and further resync activity will
8002 * always make curr_events less than last_events.
8005 if (init || curr_events - rdev->last_events > 64) {
8006 rdev->last_events = curr_events;
8014 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8016 /* another "blocks" (512byte) blocks have been synced */
8017 atomic_sub(blocks, &mddev->recovery_active);
8018 wake_up(&mddev->recovery_wait);
8020 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8021 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8022 md_wakeup_thread(mddev->thread);
8023 // stop recovery, signal do_sync ....
8026 EXPORT_SYMBOL(md_done_sync);
8028 /* md_write_start(mddev, bi)
8029 * If we need to update some array metadata (e.g. 'active' flag
8030 * in superblock) before writing, schedule a superblock update
8031 * and wait for it to complete.
8032 * A return value of 'false' means that the write wasn't recorded
8033 * and cannot proceed as the array is being suspend.
8035 bool md_write_start(struct mddev *mddev, struct bio *bi)
8038 if (bio_data_dir(bi) != WRITE)
8041 BUG_ON(mddev->ro == 1);
8042 if (mddev->ro == 2) {
8043 /* need to switch to read/write */
8045 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8046 md_wakeup_thread(mddev->thread);
8047 md_wakeup_thread(mddev->sync_thread);
8051 percpu_ref_get(&mddev->writes_pending);
8052 smp_mb(); /* Match smp_mb in set_in_sync() */
8053 if (mddev->safemode == 1)
8054 mddev->safemode = 0;
8055 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8056 if (mddev->in_sync || mddev->sync_checkers) {
8057 spin_lock(&mddev->lock);
8058 if (mddev->in_sync) {
8060 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8061 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8062 md_wakeup_thread(mddev->thread);
8065 spin_unlock(&mddev->lock);
8069 sysfs_notify_dirent_safe(mddev->sysfs_state);
8070 wait_event(mddev->sb_wait,
8071 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8073 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8074 percpu_ref_put(&mddev->writes_pending);
8079 EXPORT_SYMBOL(md_write_start);
8081 /* md_write_inc can only be called when md_write_start() has
8082 * already been called at least once of the current request.
8083 * It increments the counter and is useful when a single request
8084 * is split into several parts. Each part causes an increment and
8085 * so needs a matching md_write_end().
8086 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8087 * a spinlocked region.
8089 void md_write_inc(struct mddev *mddev, struct bio *bi)
8091 if (bio_data_dir(bi) != WRITE)
8093 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8094 percpu_ref_get(&mddev->writes_pending);
8096 EXPORT_SYMBOL(md_write_inc);
8098 void md_write_end(struct mddev *mddev)
8100 percpu_ref_put(&mddev->writes_pending);
8102 if (mddev->safemode == 2)
8103 md_wakeup_thread(mddev->thread);
8104 else if (mddev->safemode_delay)
8105 /* The roundup() ensures this only performs locking once
8106 * every ->safemode_delay jiffies
8108 mod_timer(&mddev->safemode_timer,
8109 roundup(jiffies, mddev->safemode_delay) +
8110 mddev->safemode_delay);
8113 EXPORT_SYMBOL(md_write_end);
8115 /* md_allow_write(mddev)
8116 * Calling this ensures that the array is marked 'active' so that writes
8117 * may proceed without blocking. It is important to call this before
8118 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8119 * Must be called with mddev_lock held.
8121 void md_allow_write(struct mddev *mddev)
8127 if (!mddev->pers->sync_request)
8130 spin_lock(&mddev->lock);
8131 if (mddev->in_sync) {
8133 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8134 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8135 if (mddev->safemode_delay &&
8136 mddev->safemode == 0)
8137 mddev->safemode = 1;
8138 spin_unlock(&mddev->lock);
8139 md_update_sb(mddev, 0);
8140 sysfs_notify_dirent_safe(mddev->sysfs_state);
8141 /* wait for the dirty state to be recorded in the metadata */
8142 wait_event(mddev->sb_wait,
8143 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8145 spin_unlock(&mddev->lock);
8147 EXPORT_SYMBOL_GPL(md_allow_write);
8149 #define SYNC_MARKS 10
8150 #define SYNC_MARK_STEP (3*HZ)
8151 #define UPDATE_FREQUENCY (5*60*HZ)
8152 void md_do_sync(struct md_thread *thread)
8154 struct mddev *mddev = thread->mddev;
8155 struct mddev *mddev2;
8156 unsigned int currspeed = 0,
8158 sector_t max_sectors,j, io_sectors, recovery_done;
8159 unsigned long mark[SYNC_MARKS];
8160 unsigned long update_time;
8161 sector_t mark_cnt[SYNC_MARKS];
8163 struct list_head *tmp;
8164 sector_t last_check;
8166 struct md_rdev *rdev;
8167 char *desc, *action = NULL;
8168 struct blk_plug plug;
8171 /* just incase thread restarts... */
8172 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
8174 if (mddev->ro) {/* never try to sync a read-only array */
8175 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8179 if (mddev_is_clustered(mddev)) {
8180 ret = md_cluster_ops->resync_start(mddev);
8184 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8185 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8186 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8187 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8188 && ((unsigned long long)mddev->curr_resync_completed
8189 < (unsigned long long)mddev->resync_max_sectors))
8193 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8194 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8195 desc = "data-check";
8197 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8198 desc = "requested-resync";
8202 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8207 mddev->last_sync_action = action ?: desc;
8209 /* we overload curr_resync somewhat here.
8210 * 0 == not engaged in resync at all
8211 * 2 == checking that there is no conflict with another sync
8212 * 1 == like 2, but have yielded to allow conflicting resync to
8214 * other == active in resync - this many blocks
8216 * Before starting a resync we must have set curr_resync to
8217 * 2, and then checked that every "conflicting" array has curr_resync
8218 * less than ours. When we find one that is the same or higher
8219 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8220 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8221 * This will mean we have to start checking from the beginning again.
8226 int mddev2_minor = -1;
8227 mddev->curr_resync = 2;
8230 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8232 for_each_mddev(mddev2, tmp) {
8233 if (mddev2 == mddev)
8235 if (!mddev->parallel_resync
8236 && mddev2->curr_resync
8237 && match_mddev_units(mddev, mddev2)) {
8239 if (mddev < mddev2 && mddev->curr_resync == 2) {
8240 /* arbitrarily yield */
8241 mddev->curr_resync = 1;
8242 wake_up(&resync_wait);
8244 if (mddev > mddev2 && mddev->curr_resync == 1)
8245 /* no need to wait here, we can wait the next
8246 * time 'round when curr_resync == 2
8249 /* We need to wait 'interruptible' so as not to
8250 * contribute to the load average, and not to
8251 * be caught by 'softlockup'
8253 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8254 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8255 mddev2->curr_resync >= mddev->curr_resync) {
8256 if (mddev2_minor != mddev2->md_minor) {
8257 mddev2_minor = mddev2->md_minor;
8258 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8259 desc, mdname(mddev),
8263 if (signal_pending(current))
8264 flush_signals(current);
8266 finish_wait(&resync_wait, &wq);
8269 finish_wait(&resync_wait, &wq);
8272 } while (mddev->curr_resync < 2);
8275 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8276 /* resync follows the size requested by the personality,
8277 * which defaults to physical size, but can be virtual size
8279 max_sectors = mddev->resync_max_sectors;
8280 atomic64_set(&mddev->resync_mismatches, 0);
8281 /* we don't use the checkpoint if there's a bitmap */
8282 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8283 j = mddev->resync_min;
8284 else if (!mddev->bitmap)
8285 j = mddev->recovery_cp;
8287 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8288 max_sectors = mddev->resync_max_sectors;
8290 /* recovery follows the physical size of devices */
8291 max_sectors = mddev->dev_sectors;
8294 rdev_for_each_rcu(rdev, mddev)
8295 if (rdev->raid_disk >= 0 &&
8296 !test_bit(Journal, &rdev->flags) &&
8297 !test_bit(Faulty, &rdev->flags) &&
8298 !test_bit(In_sync, &rdev->flags) &&
8299 rdev->recovery_offset < j)
8300 j = rdev->recovery_offset;
8303 /* If there is a bitmap, we need to make sure all
8304 * writes that started before we added a spare
8305 * complete before we start doing a recovery.
8306 * Otherwise the write might complete and (via
8307 * bitmap_endwrite) set a bit in the bitmap after the
8308 * recovery has checked that bit and skipped that
8311 if (mddev->bitmap) {
8312 mddev->pers->quiesce(mddev, 1);
8313 mddev->pers->quiesce(mddev, 0);
8317 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8318 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8319 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8320 speed_max(mddev), desc);
8322 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8325 for (m = 0; m < SYNC_MARKS; m++) {
8327 mark_cnt[m] = io_sectors;
8330 mddev->resync_mark = mark[last_mark];
8331 mddev->resync_mark_cnt = mark_cnt[last_mark];
8334 * Tune reconstruction:
8336 window = 32*(PAGE_SIZE/512);
8337 pr_debug("md: using %dk window, over a total of %lluk.\n",
8338 window/2, (unsigned long long)max_sectors/2);
8340 atomic_set(&mddev->recovery_active, 0);
8344 pr_debug("md: resuming %s of %s from checkpoint.\n",
8345 desc, mdname(mddev));
8346 mddev->curr_resync = j;
8348 mddev->curr_resync = 3; /* no longer delayed */
8349 mddev->curr_resync_completed = j;
8350 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8351 md_new_event(mddev);
8352 update_time = jiffies;
8354 blk_start_plug(&plug);
8355 while (j < max_sectors) {
8360 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8361 ((mddev->curr_resync > mddev->curr_resync_completed &&
8362 (mddev->curr_resync - mddev->curr_resync_completed)
8363 > (max_sectors >> 4)) ||
8364 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8365 (j - mddev->curr_resync_completed)*2
8366 >= mddev->resync_max - mddev->curr_resync_completed ||
8367 mddev->curr_resync_completed > mddev->resync_max
8369 /* time to update curr_resync_completed */
8370 wait_event(mddev->recovery_wait,
8371 atomic_read(&mddev->recovery_active) == 0);
8372 mddev->curr_resync_completed = j;
8373 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8374 j > mddev->recovery_cp)
8375 mddev->recovery_cp = j;
8376 update_time = jiffies;
8377 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8378 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8381 while (j >= mddev->resync_max &&
8382 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8383 /* As this condition is controlled by user-space,
8384 * we can block indefinitely, so use '_interruptible'
8385 * to avoid triggering warnings.
8387 flush_signals(current); /* just in case */
8388 wait_event_interruptible(mddev->recovery_wait,
8389 mddev->resync_max > j
8390 || test_bit(MD_RECOVERY_INTR,
8394 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8397 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8399 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8403 if (!skipped) { /* actual IO requested */
8404 io_sectors += sectors;
8405 atomic_add(sectors, &mddev->recovery_active);
8408 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8412 if (j > max_sectors)
8413 /* when skipping, extra large numbers can be returned. */
8416 mddev->curr_resync = j;
8417 mddev->curr_mark_cnt = io_sectors;
8418 if (last_check == 0)
8419 /* this is the earliest that rebuild will be
8420 * visible in /proc/mdstat
8422 md_new_event(mddev);
8424 if (last_check + window > io_sectors || j == max_sectors)
8427 last_check = io_sectors;
8429 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8431 int next = (last_mark+1) % SYNC_MARKS;
8433 mddev->resync_mark = mark[next];
8434 mddev->resync_mark_cnt = mark_cnt[next];
8435 mark[next] = jiffies;
8436 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8440 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8444 * this loop exits only if either when we are slower than
8445 * the 'hard' speed limit, or the system was IO-idle for
8447 * the system might be non-idle CPU-wise, but we only care
8448 * about not overloading the IO subsystem. (things like an
8449 * e2fsck being done on the RAID array should execute fast)
8453 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8454 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8455 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8457 if (currspeed > speed_min(mddev)) {
8458 if (currspeed > speed_max(mddev)) {
8462 if (!is_mddev_idle(mddev, 0)) {
8464 * Give other IO more of a chance.
8465 * The faster the devices, the less we wait.
8467 wait_event(mddev->recovery_wait,
8468 !atomic_read(&mddev->recovery_active));
8472 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8473 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8474 ? "interrupted" : "done");
8476 * this also signals 'finished resyncing' to md_stop
8478 blk_finish_plug(&plug);
8479 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8481 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8482 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8483 mddev->curr_resync > 3) {
8484 mddev->curr_resync_completed = mddev->curr_resync;
8485 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8487 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8489 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8490 mddev->curr_resync > 3) {
8491 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8492 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8493 if (mddev->curr_resync >= mddev->recovery_cp) {
8494 pr_debug("md: checkpointing %s of %s.\n",
8495 desc, mdname(mddev));
8496 if (test_bit(MD_RECOVERY_ERROR,
8498 mddev->recovery_cp =
8499 mddev->curr_resync_completed;
8501 mddev->recovery_cp =
8505 mddev->recovery_cp = MaxSector;
8507 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8508 mddev->curr_resync = MaxSector;
8509 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8510 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8512 rdev_for_each_rcu(rdev, mddev)
8513 if (rdev->raid_disk >= 0 &&
8514 mddev->delta_disks >= 0 &&
8515 !test_bit(Journal, &rdev->flags) &&
8516 !test_bit(Faulty, &rdev->flags) &&
8517 !test_bit(In_sync, &rdev->flags) &&
8518 rdev->recovery_offset < mddev->curr_resync)
8519 rdev->recovery_offset = mddev->curr_resync;
8525 /* set CHANGE_PENDING here since maybe another update is needed,
8526 * so other nodes are informed. It should be harmless for normal
8528 set_mask_bits(&mddev->sb_flags, 0,
8529 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8531 spin_lock(&mddev->lock);
8532 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8533 /* We completed so min/max setting can be forgotten if used. */
8534 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8535 mddev->resync_min = 0;
8536 mddev->resync_max = MaxSector;
8537 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8538 mddev->resync_min = mddev->curr_resync_completed;
8539 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8540 mddev->curr_resync = 0;
8541 spin_unlock(&mddev->lock);
8543 wake_up(&resync_wait);
8544 md_wakeup_thread(mddev->thread);
8547 EXPORT_SYMBOL_GPL(md_do_sync);
8549 static int remove_and_add_spares(struct mddev *mddev,
8550 struct md_rdev *this)
8552 struct md_rdev *rdev;
8555 bool remove_some = false;
8557 rdev_for_each(rdev, mddev) {
8558 if ((this == NULL || rdev == this) &&
8559 rdev->raid_disk >= 0 &&
8560 !test_bit(Blocked, &rdev->flags) &&
8561 test_bit(Faulty, &rdev->flags) &&
8562 atomic_read(&rdev->nr_pending)==0) {
8563 /* Faulty non-Blocked devices with nr_pending == 0
8564 * never get nr_pending incremented,
8565 * never get Faulty cleared, and never get Blocked set.
8566 * So we can synchronize_rcu now rather than once per device
8569 set_bit(RemoveSynchronized, &rdev->flags);
8575 rdev_for_each(rdev, mddev) {
8576 if ((this == NULL || rdev == this) &&
8577 rdev->raid_disk >= 0 &&
8578 !test_bit(Blocked, &rdev->flags) &&
8579 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8580 (!test_bit(In_sync, &rdev->flags) &&
8581 !test_bit(Journal, &rdev->flags))) &&
8582 atomic_read(&rdev->nr_pending)==0)) {
8583 if (mddev->pers->hot_remove_disk(
8584 mddev, rdev) == 0) {
8585 sysfs_unlink_rdev(mddev, rdev);
8586 rdev->raid_disk = -1;
8590 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8591 clear_bit(RemoveSynchronized, &rdev->flags);
8594 if (removed && mddev->kobj.sd)
8595 sysfs_notify(&mddev->kobj, NULL, "degraded");
8597 if (this && removed)
8600 rdev_for_each(rdev, mddev) {
8601 if (this && this != rdev)
8603 if (test_bit(Candidate, &rdev->flags))
8605 if (rdev->raid_disk >= 0 &&
8606 !test_bit(In_sync, &rdev->flags) &&
8607 !test_bit(Journal, &rdev->flags) &&
8608 !test_bit(Faulty, &rdev->flags))
8610 if (rdev->raid_disk >= 0)
8612 if (test_bit(Faulty, &rdev->flags))
8614 if (!test_bit(Journal, &rdev->flags)) {
8616 ! (rdev->saved_raid_disk >= 0 &&
8617 !test_bit(Bitmap_sync, &rdev->flags)))
8620 rdev->recovery_offset = 0;
8623 hot_add_disk(mddev, rdev) == 0) {
8624 if (sysfs_link_rdev(mddev, rdev))
8625 /* failure here is OK */;
8626 if (!test_bit(Journal, &rdev->flags))
8628 md_new_event(mddev);
8629 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8634 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8638 static void md_start_sync(struct work_struct *ws)
8640 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8642 mddev->sync_thread = md_register_thread(md_do_sync,
8645 if (!mddev->sync_thread) {
8646 pr_warn("%s: could not start resync thread...\n",
8648 /* leave the spares where they are, it shouldn't hurt */
8649 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8650 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8651 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8652 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8653 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8654 wake_up(&resync_wait);
8655 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8657 if (mddev->sysfs_action)
8658 sysfs_notify_dirent_safe(mddev->sysfs_action);
8660 md_wakeup_thread(mddev->sync_thread);
8661 sysfs_notify_dirent_safe(mddev->sysfs_action);
8662 md_new_event(mddev);
8666 * This routine is regularly called by all per-raid-array threads to
8667 * deal with generic issues like resync and super-block update.
8668 * Raid personalities that don't have a thread (linear/raid0) do not
8669 * need this as they never do any recovery or update the superblock.
8671 * It does not do any resync itself, but rather "forks" off other threads
8672 * to do that as needed.
8673 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8674 * "->recovery" and create a thread at ->sync_thread.
8675 * When the thread finishes it sets MD_RECOVERY_DONE
8676 * and wakeups up this thread which will reap the thread and finish up.
8677 * This thread also removes any faulty devices (with nr_pending == 0).
8679 * The overall approach is:
8680 * 1/ if the superblock needs updating, update it.
8681 * 2/ If a recovery thread is running, don't do anything else.
8682 * 3/ If recovery has finished, clean up, possibly marking spares active.
8683 * 4/ If there are any faulty devices, remove them.
8684 * 5/ If array is degraded, try to add spares devices
8685 * 6/ If array has spares or is not in-sync, start a resync thread.
8687 void md_check_recovery(struct mddev *mddev)
8689 if (mddev->suspended)
8693 bitmap_daemon_work(mddev);
8695 if (signal_pending(current)) {
8696 if (mddev->pers->sync_request && !mddev->external) {
8697 pr_debug("md: %s in immediate safe mode\n",
8699 mddev->safemode = 2;
8701 flush_signals(current);
8704 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8707 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8708 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8709 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8710 (mddev->external == 0 && mddev->safemode == 1) ||
8711 (mddev->safemode == 2
8712 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8716 if (mddev_trylock(mddev)) {
8719 if (!mddev->external && mddev->safemode == 1)
8720 mddev->safemode = 0;
8723 struct md_rdev *rdev;
8724 if (!mddev->external && mddev->in_sync)
8725 /* 'Blocked' flag not needed as failed devices
8726 * will be recorded if array switched to read/write.
8727 * Leaving it set will prevent the device
8728 * from being removed.
8730 rdev_for_each(rdev, mddev)
8731 clear_bit(Blocked, &rdev->flags);
8732 /* On a read-only array we can:
8733 * - remove failed devices
8734 * - add already-in_sync devices if the array itself
8736 * As we only add devices that are already in-sync,
8737 * we can activate the spares immediately.
8739 remove_and_add_spares(mddev, NULL);
8740 /* There is no thread, but we need to call
8741 * ->spare_active and clear saved_raid_disk
8743 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8744 md_reap_sync_thread(mddev);
8745 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8746 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8747 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8751 if (mddev_is_clustered(mddev)) {
8752 struct md_rdev *rdev;
8753 /* kick the device if another node issued a
8756 rdev_for_each(rdev, mddev) {
8757 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8758 rdev->raid_disk < 0)
8759 md_kick_rdev_from_array(rdev);
8763 if (!mddev->external && !mddev->in_sync) {
8764 spin_lock(&mddev->lock);
8766 spin_unlock(&mddev->lock);
8769 if (mddev->sb_flags)
8770 md_update_sb(mddev, 0);
8772 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8773 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8774 /* resync/recovery still happening */
8775 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8778 if (mddev->sync_thread) {
8779 md_reap_sync_thread(mddev);
8782 /* Set RUNNING before clearing NEEDED to avoid
8783 * any transients in the value of "sync_action".
8785 mddev->curr_resync_completed = 0;
8786 spin_lock(&mddev->lock);
8787 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8788 spin_unlock(&mddev->lock);
8789 /* Clear some bits that don't mean anything, but
8792 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8793 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8795 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8796 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8798 /* no recovery is running.
8799 * remove any failed drives, then
8800 * add spares if possible.
8801 * Spares are also removed and re-added, to allow
8802 * the personality to fail the re-add.
8805 if (mddev->reshape_position != MaxSector) {
8806 if (mddev->pers->check_reshape == NULL ||
8807 mddev->pers->check_reshape(mddev) != 0)
8808 /* Cannot proceed */
8810 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8811 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8812 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8813 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8814 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8815 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8816 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8817 } else if (mddev->recovery_cp < MaxSector) {
8818 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8819 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8820 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8821 /* nothing to be done ... */
8824 if (mddev->pers->sync_request) {
8826 /* We are adding a device or devices to an array
8827 * which has the bitmap stored on all devices.
8828 * So make sure all bitmap pages get written
8830 bitmap_write_all(mddev->bitmap);
8832 INIT_WORK(&mddev->del_work, md_start_sync);
8833 queue_work(md_misc_wq, &mddev->del_work);
8837 if (!mddev->sync_thread) {
8838 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8839 wake_up(&resync_wait);
8840 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8842 if (mddev->sysfs_action)
8843 sysfs_notify_dirent_safe(mddev->sysfs_action);
8846 wake_up(&mddev->sb_wait);
8847 mddev_unlock(mddev);
8848 } else if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8849 /* Write superblock - thread that called mddev_suspend()
8850 * holds reconfig_mutex for us.
8852 set_bit(MD_UPDATING_SB, &mddev->flags);
8853 smp_mb__after_atomic();
8854 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8855 md_update_sb(mddev, 0);
8856 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8857 wake_up(&mddev->sb_wait);
8860 EXPORT_SYMBOL(md_check_recovery);
8862 void md_reap_sync_thread(struct mddev *mddev)
8864 struct md_rdev *rdev;
8866 /* resync has finished, collect result */
8867 md_unregister_thread(&mddev->sync_thread);
8868 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8869 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8871 /* activate any spares */
8872 if (mddev->pers->spare_active(mddev)) {
8873 sysfs_notify(&mddev->kobj, NULL,
8875 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8878 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8879 mddev->pers->finish_reshape)
8880 mddev->pers->finish_reshape(mddev);
8882 /* If array is no-longer degraded, then any saved_raid_disk
8883 * information must be scrapped.
8885 if (!mddev->degraded)
8886 rdev_for_each(rdev, mddev)
8887 rdev->saved_raid_disk = -1;
8889 md_update_sb(mddev, 1);
8890 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
8891 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8893 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8894 md_cluster_ops->resync_finish(mddev);
8895 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8896 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8897 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8898 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8899 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8900 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8901 wake_up(&resync_wait);
8902 /* flag recovery needed just to double check */
8903 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8904 sysfs_notify_dirent_safe(mddev->sysfs_action);
8905 md_new_event(mddev);
8906 if (mddev->event_work.func)
8907 queue_work(md_misc_wq, &mddev->event_work);
8909 EXPORT_SYMBOL(md_reap_sync_thread);
8911 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8913 sysfs_notify_dirent_safe(rdev->sysfs_state);
8914 wait_event_timeout(rdev->blocked_wait,
8915 !test_bit(Blocked, &rdev->flags) &&
8916 !test_bit(BlockedBadBlocks, &rdev->flags),
8917 msecs_to_jiffies(5000));
8918 rdev_dec_pending(rdev, mddev);
8920 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8922 void md_finish_reshape(struct mddev *mddev)
8924 /* called be personality module when reshape completes. */
8925 struct md_rdev *rdev;
8927 rdev_for_each(rdev, mddev) {
8928 if (rdev->data_offset > rdev->new_data_offset)
8929 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8931 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8932 rdev->data_offset = rdev->new_data_offset;
8935 EXPORT_SYMBOL(md_finish_reshape);
8937 /* Bad block management */
8939 /* Returns 1 on success, 0 on failure */
8940 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8943 struct mddev *mddev = rdev->mddev;
8946 s += rdev->new_data_offset;
8948 s += rdev->data_offset;
8949 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8951 /* Make sure they get written out promptly */
8952 if (test_bit(ExternalBbl, &rdev->flags))
8953 sysfs_notify(&rdev->kobj, NULL,
8954 "unacknowledged_bad_blocks");
8955 sysfs_notify_dirent_safe(rdev->sysfs_state);
8956 set_mask_bits(&mddev->sb_flags, 0,
8957 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
8958 md_wakeup_thread(rdev->mddev->thread);
8963 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8965 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8970 s += rdev->new_data_offset;
8972 s += rdev->data_offset;
8973 rv = badblocks_clear(&rdev->badblocks, s, sectors);
8974 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
8975 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
8978 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8980 static int md_notify_reboot(struct notifier_block *this,
8981 unsigned long code, void *x)
8983 struct list_head *tmp;
8984 struct mddev *mddev;
8987 for_each_mddev(mddev, tmp) {
8988 if (mddev_trylock(mddev)) {
8990 __md_stop_writes(mddev);
8991 if (mddev->persistent)
8992 mddev->safemode = 2;
8993 mddev_unlock(mddev);
8998 * certain more exotic SCSI devices are known to be
8999 * volatile wrt too early system reboots. While the
9000 * right place to handle this issue is the given
9001 * driver, we do want to have a safe RAID driver ...
9009 static struct notifier_block md_notifier = {
9010 .notifier_call = md_notify_reboot,
9012 .priority = INT_MAX, /* before any real devices */
9015 static void md_geninit(void)
9017 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9019 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9022 static int __init md_init(void)
9026 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9030 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9034 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9037 if ((ret = register_blkdev(0, "mdp")) < 0)
9041 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9042 md_probe, NULL, NULL);
9043 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9044 md_probe, NULL, NULL);
9046 register_reboot_notifier(&md_notifier);
9047 raid_table_header = register_sysctl_table(raid_root_table);
9053 unregister_blkdev(MD_MAJOR, "md");
9055 destroy_workqueue(md_misc_wq);
9057 destroy_workqueue(md_wq);
9062 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9064 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9065 struct md_rdev *rdev2;
9067 char b[BDEVNAME_SIZE];
9070 * If size is changed in another node then we need to
9071 * do resize as well.
9073 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9074 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9076 pr_info("md-cluster: resize failed\n");
9078 bitmap_update_sb(mddev->bitmap);
9081 /* Check for change of roles in the active devices */
9082 rdev_for_each(rdev2, mddev) {
9083 if (test_bit(Faulty, &rdev2->flags))
9086 /* Check if the roles changed */
9087 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9089 if (test_bit(Candidate, &rdev2->flags)) {
9090 if (role == 0xfffe) {
9091 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9092 md_kick_rdev_from_array(rdev2);
9096 clear_bit(Candidate, &rdev2->flags);
9099 if (role != rdev2->raid_disk) {
9101 if (rdev2->raid_disk == -1 && role != 0xffff) {
9102 rdev2->saved_raid_disk = role;
9103 ret = remove_and_add_spares(mddev, rdev2);
9104 pr_info("Activated spare: %s\n",
9105 bdevname(rdev2->bdev,b));
9106 /* wakeup mddev->thread here, so array could
9107 * perform resync with the new activated disk */
9108 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9109 md_wakeup_thread(mddev->thread);
9113 * We just want to do the minimum to mark the disk
9114 * as faulty. The recovery is performed by the
9115 * one who initiated the error.
9117 if ((role == 0xfffe) || (role == 0xfffd)) {
9118 md_error(mddev, rdev2);
9119 clear_bit(Blocked, &rdev2->flags);
9124 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9125 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9127 /* Finally set the event to be up to date */
9128 mddev->events = le64_to_cpu(sb->events);
9131 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9134 struct page *swapout = rdev->sb_page;
9135 struct mdp_superblock_1 *sb;
9137 /* Store the sb page of the rdev in the swapout temporary
9138 * variable in case we err in the future
9140 rdev->sb_page = NULL;
9141 err = alloc_disk_sb(rdev);
9143 ClearPageUptodate(rdev->sb_page);
9144 rdev->sb_loaded = 0;
9145 err = super_types[mddev->major_version].
9146 load_super(rdev, NULL, mddev->minor_version);
9149 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9150 __func__, __LINE__, rdev->desc_nr, err);
9152 put_page(rdev->sb_page);
9153 rdev->sb_page = swapout;
9154 rdev->sb_loaded = 1;
9158 sb = page_address(rdev->sb_page);
9159 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9163 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9164 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9166 /* The other node finished recovery, call spare_active to set
9167 * device In_sync and mddev->degraded
9169 if (rdev->recovery_offset == MaxSector &&
9170 !test_bit(In_sync, &rdev->flags) &&
9171 mddev->pers->spare_active(mddev))
9172 sysfs_notify(&mddev->kobj, NULL, "degraded");
9178 void md_reload_sb(struct mddev *mddev, int nr)
9180 struct md_rdev *rdev;
9184 rdev_for_each_rcu(rdev, mddev) {
9185 if (rdev->desc_nr == nr)
9189 if (!rdev || rdev->desc_nr != nr) {
9190 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9194 err = read_rdev(mddev, rdev);
9198 check_sb_changes(mddev, rdev);
9200 /* Read all rdev's to update recovery_offset */
9201 rdev_for_each_rcu(rdev, mddev)
9202 read_rdev(mddev, rdev);
9204 EXPORT_SYMBOL(md_reload_sb);
9209 * Searches all registered partitions for autorun RAID arrays
9213 static DEFINE_MUTEX(detected_devices_mutex);
9214 static LIST_HEAD(all_detected_devices);
9215 struct detected_devices_node {
9216 struct list_head list;
9220 void md_autodetect_dev(dev_t dev)
9222 struct detected_devices_node *node_detected_dev;
9224 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9225 if (node_detected_dev) {
9226 node_detected_dev->dev = dev;
9227 mutex_lock(&detected_devices_mutex);
9228 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9229 mutex_unlock(&detected_devices_mutex);
9233 static void autostart_arrays(int part)
9235 struct md_rdev *rdev;
9236 struct detected_devices_node *node_detected_dev;
9238 int i_scanned, i_passed;
9243 pr_info("md: Autodetecting RAID arrays.\n");
9245 mutex_lock(&detected_devices_mutex);
9246 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9248 node_detected_dev = list_entry(all_detected_devices.next,
9249 struct detected_devices_node, list);
9250 list_del(&node_detected_dev->list);
9251 dev = node_detected_dev->dev;
9252 kfree(node_detected_dev);
9253 mutex_unlock(&detected_devices_mutex);
9254 rdev = md_import_device(dev,0, 90);
9255 mutex_lock(&detected_devices_mutex);
9259 if (test_bit(Faulty, &rdev->flags))
9262 set_bit(AutoDetected, &rdev->flags);
9263 list_add(&rdev->same_set, &pending_raid_disks);
9266 mutex_unlock(&detected_devices_mutex);
9268 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9270 autorun_devices(part);
9273 #endif /* !MODULE */
9275 static __exit void md_exit(void)
9277 struct mddev *mddev;
9278 struct list_head *tmp;
9281 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9282 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9284 unregister_blkdev(MD_MAJOR,"md");
9285 unregister_blkdev(mdp_major, "mdp");
9286 unregister_reboot_notifier(&md_notifier);
9287 unregister_sysctl_table(raid_table_header);
9289 /* We cannot unload the modules while some process is
9290 * waiting for us in select() or poll() - wake them up
9293 while (waitqueue_active(&md_event_waiters)) {
9294 /* not safe to leave yet */
9295 wake_up(&md_event_waiters);
9299 remove_proc_entry("mdstat", NULL);
9301 for_each_mddev(mddev, tmp) {
9302 export_array(mddev);
9304 mddev->hold_active = 0;
9306 * for_each_mddev() will call mddev_put() at the end of each
9307 * iteration. As the mddev is now fully clear, this will
9308 * schedule the mddev for destruction by a workqueue, and the
9309 * destroy_workqueue() below will wait for that to complete.
9312 destroy_workqueue(md_misc_wq);
9313 destroy_workqueue(md_wq);
9316 subsys_initcall(md_init);
9317 module_exit(md_exit)
9319 static int get_ro(char *buffer, const struct kernel_param *kp)
9321 return sprintf(buffer, "%d", start_readonly);
9323 static int set_ro(const char *val, const struct kernel_param *kp)
9325 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9328 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9329 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9330 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9331 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9333 MODULE_LICENSE("GPL");
9334 MODULE_DESCRIPTION("MD RAID framework");
9336 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);