1 // SPDX-License-Identifier: GPL-2.0-or-later
3 md.c : Multiple Devices driver for Linux
4 Copyright (C) 1998, 1999, 2000 Ingo Molnar
6 completely rewritten, based on the MD driver code from Marc Zyngier
10 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
11 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
12 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
13 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
14 - kmod support by: Cyrus Durgin
15 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
16 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
18 - lots of fixes and improvements to the RAID1/RAID5 and generic
19 RAID code (such as request based resynchronization):
21 Neil Brown <neilb@cse.unsw.edu.au>.
23 - persistent bitmap code
24 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
27 Errors, Warnings, etc.
29 pr_crit() for error conditions that risk data loss
30 pr_err() for error conditions that are unexpected, like an IO error
31 or internal inconsistency
32 pr_warn() for error conditions that could have been predicated, like
33 adding a device to an array when it has incompatible metadata
34 pr_info() for every interesting, very rare events, like an array starting
35 or stopping, or resync starting or stopping
36 pr_debug() for everything else.
40 #include <linux/sched/mm.h>
41 #include <linux/sched/signal.h>
42 #include <linux/kthread.h>
43 #include <linux/blkdev.h>
44 #include <linux/badblocks.h>
45 #include <linux/sysctl.h>
46 #include <linux/seq_file.h>
48 #include <linux/poll.h>
49 #include <linux/ctype.h>
50 #include <linux/string.h>
51 #include <linux/hdreg.h>
52 #include <linux/proc_fs.h>
53 #include <linux/random.h>
54 #include <linux/module.h>
55 #include <linux/reboot.h>
56 #include <linux/file.h>
57 #include <linux/compat.h>
58 #include <linux/delay.h>
59 #include <linux/raid/md_p.h>
60 #include <linux/raid/md_u.h>
61 #include <linux/slab.h>
62 #include <linux/percpu-refcount.h>
64 #include <trace/events/block.h>
66 #include "md-bitmap.h"
67 #include "md-cluster.h"
70 static void autostart_arrays(int part);
73 /* pers_list is a list of registered personalities protected
75 * pers_lock does extra service to protect accesses to
76 * mddev->thread when the mutex cannot be held.
78 static LIST_HEAD(pers_list);
79 static DEFINE_SPINLOCK(pers_lock);
81 static struct kobj_type md_ktype;
83 struct md_cluster_operations *md_cluster_ops;
84 EXPORT_SYMBOL(md_cluster_ops);
85 static struct module *md_cluster_mod;
87 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
88 static struct workqueue_struct *md_wq;
89 static struct workqueue_struct *md_misc_wq;
91 static int remove_and_add_spares(struct mddev *mddev,
92 struct md_rdev *this);
93 static void mddev_detach(struct mddev *mddev);
96 * Default number of read corrections we'll attempt on an rdev
97 * before ejecting it from the array. We divide the read error
98 * count by 2 for every hour elapsed between read errors.
100 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
102 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
103 * is 1000 KB/sec, so the extra system load does not show up that much.
104 * Increase it if you want to have more _guaranteed_ speed. Note that
105 * the RAID driver will use the maximum available bandwidth if the IO
106 * subsystem is idle. There is also an 'absolute maximum' reconstruction
107 * speed limit - in case reconstruction slows down your system despite
110 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
111 * or /sys/block/mdX/md/sync_speed_{min,max}
114 static int sysctl_speed_limit_min = 1000;
115 static int sysctl_speed_limit_max = 200000;
116 static inline int speed_min(struct mddev *mddev)
118 return mddev->sync_speed_min ?
119 mddev->sync_speed_min : sysctl_speed_limit_min;
122 static inline int speed_max(struct mddev *mddev)
124 return mddev->sync_speed_max ?
125 mddev->sync_speed_max : sysctl_speed_limit_max;
128 static int rdev_init_wb(struct md_rdev *rdev)
130 if (rdev->bdev->bd_queue->nr_hw_queues == 1)
133 spin_lock_init(&rdev->wb_list_lock);
134 INIT_LIST_HEAD(&rdev->wb_list);
135 init_waitqueue_head(&rdev->wb_io_wait);
136 set_bit(WBCollisionCheck, &rdev->flags);
142 * Create wb_info_pool if rdev is the first multi-queue device flaged
143 * with writemostly, also write-behind mode is enabled.
145 void mddev_create_wb_pool(struct mddev *mddev, struct md_rdev *rdev,
148 if (mddev->bitmap_info.max_write_behind == 0)
151 if (!test_bit(WriteMostly, &rdev->flags) || !rdev_init_wb(rdev))
154 if (mddev->wb_info_pool == NULL) {
155 unsigned int noio_flag;
158 mddev_suspend(mddev);
159 noio_flag = memalloc_noio_save();
160 mddev->wb_info_pool = mempool_create_kmalloc_pool(NR_WB_INFOS,
161 sizeof(struct wb_info));
162 memalloc_noio_restore(noio_flag);
163 if (!mddev->wb_info_pool)
164 pr_err("can't alloc memory pool for writemostly\n");
169 EXPORT_SYMBOL_GPL(mddev_create_wb_pool);
172 * destroy wb_info_pool if rdev is the last device flaged with WBCollisionCheck.
174 static void mddev_destroy_wb_pool(struct mddev *mddev, struct md_rdev *rdev)
176 if (!test_and_clear_bit(WBCollisionCheck, &rdev->flags))
179 if (mddev->wb_info_pool) {
180 struct md_rdev *temp;
184 * Check if other rdevs need wb_info_pool.
186 rdev_for_each(temp, mddev)
188 test_bit(WBCollisionCheck, &temp->flags))
191 mddev_suspend(rdev->mddev);
192 mempool_destroy(mddev->wb_info_pool);
193 mddev->wb_info_pool = NULL;
194 mddev_resume(rdev->mddev);
199 static struct ctl_table_header *raid_table_header;
201 static struct ctl_table raid_table[] = {
203 .procname = "speed_limit_min",
204 .data = &sysctl_speed_limit_min,
205 .maxlen = sizeof(int),
206 .mode = S_IRUGO|S_IWUSR,
207 .proc_handler = proc_dointvec,
210 .procname = "speed_limit_max",
211 .data = &sysctl_speed_limit_max,
212 .maxlen = sizeof(int),
213 .mode = S_IRUGO|S_IWUSR,
214 .proc_handler = proc_dointvec,
219 static struct ctl_table raid_dir_table[] = {
223 .mode = S_IRUGO|S_IXUGO,
229 static struct ctl_table raid_root_table[] = {
234 .child = raid_dir_table,
239 static const struct block_device_operations md_fops;
241 static int start_readonly;
244 * The original mechanism for creating an md device is to create
245 * a device node in /dev and to open it. This causes races with device-close.
246 * The preferred method is to write to the "new_array" module parameter.
247 * This can avoid races.
248 * Setting create_on_open to false disables the original mechanism
249 * so all the races disappear.
251 static bool create_on_open = true;
253 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
256 if (!mddev || !bioset_initialized(&mddev->bio_set))
257 return bio_alloc(gfp_mask, nr_iovecs);
259 return bio_alloc_bioset(gfp_mask, nr_iovecs, &mddev->bio_set);
261 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
263 static struct bio *md_bio_alloc_sync(struct mddev *mddev)
265 if (!mddev || !bioset_initialized(&mddev->sync_set))
266 return bio_alloc(GFP_NOIO, 1);
268 return bio_alloc_bioset(GFP_NOIO, 1, &mddev->sync_set);
272 * We have a system wide 'event count' that is incremented
273 * on any 'interesting' event, and readers of /proc/mdstat
274 * can use 'poll' or 'select' to find out when the event
278 * start array, stop array, error, add device, remove device,
279 * start build, activate spare
281 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
282 static atomic_t md_event_count;
283 void md_new_event(struct mddev *mddev)
285 atomic_inc(&md_event_count);
286 wake_up(&md_event_waiters);
288 EXPORT_SYMBOL_GPL(md_new_event);
291 * Enables to iterate over all existing md arrays
292 * all_mddevs_lock protects this list.
294 static LIST_HEAD(all_mddevs);
295 static DEFINE_SPINLOCK(all_mddevs_lock);
298 * iterates through all used mddevs in the system.
299 * We take care to grab the all_mddevs_lock whenever navigating
300 * the list, and to always hold a refcount when unlocked.
301 * Any code which breaks out of this loop while own
302 * a reference to the current mddev and must mddev_put it.
304 #define for_each_mddev(_mddev,_tmp) \
306 for (({ spin_lock(&all_mddevs_lock); \
307 _tmp = all_mddevs.next; \
309 ({ if (_tmp != &all_mddevs) \
310 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
311 spin_unlock(&all_mddevs_lock); \
312 if (_mddev) mddev_put(_mddev); \
313 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
314 _tmp != &all_mddevs;}); \
315 ({ spin_lock(&all_mddevs_lock); \
316 _tmp = _tmp->next;}) \
319 /* Rather than calling directly into the personality make_request function,
320 * IO requests come here first so that we can check if the device is
321 * being suspended pending a reconfiguration.
322 * We hold a refcount over the call to ->make_request. By the time that
323 * call has finished, the bio has been linked into some internal structure
324 * and so is visible to ->quiesce(), so we don't need the refcount any more.
326 static bool is_suspended(struct mddev *mddev, struct bio *bio)
328 if (mddev->suspended)
330 if (bio_data_dir(bio) != WRITE)
332 if (mddev->suspend_lo >= mddev->suspend_hi)
334 if (bio->bi_iter.bi_sector >= mddev->suspend_hi)
336 if (bio_end_sector(bio) < mddev->suspend_lo)
341 void md_handle_request(struct mddev *mddev, struct bio *bio)
345 if (is_suspended(mddev, bio)) {
348 prepare_to_wait(&mddev->sb_wait, &__wait,
349 TASK_UNINTERRUPTIBLE);
350 if (!is_suspended(mddev, bio))
356 finish_wait(&mddev->sb_wait, &__wait);
358 atomic_inc(&mddev->active_io);
361 if (!mddev->pers->make_request(mddev, bio)) {
362 atomic_dec(&mddev->active_io);
363 wake_up(&mddev->sb_wait);
364 goto check_suspended;
367 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
368 wake_up(&mddev->sb_wait);
370 EXPORT_SYMBOL(md_handle_request);
372 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
374 const int rw = bio_data_dir(bio);
375 const int sgrp = op_stat_group(bio_op(bio));
376 struct mddev *mddev = q->queuedata;
377 unsigned int sectors;
379 if (unlikely(test_bit(MD_BROKEN, &mddev->flags)) && (rw == WRITE)) {
381 return BLK_QC_T_NONE;
384 blk_queue_split(q, &bio);
386 if (mddev == NULL || mddev->pers == NULL) {
388 return BLK_QC_T_NONE;
390 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
391 if (bio_sectors(bio) != 0)
392 bio->bi_status = BLK_STS_IOERR;
394 return BLK_QC_T_NONE;
398 * save the sectors now since our bio can
399 * go away inside make_request
401 sectors = bio_sectors(bio);
402 /* bio could be mergeable after passing to underlayer */
403 bio->bi_opf &= ~REQ_NOMERGE;
405 md_handle_request(mddev, bio);
408 part_stat_inc(&mddev->gendisk->part0, ios[sgrp]);
409 part_stat_add(&mddev->gendisk->part0, sectors[sgrp], sectors);
412 return BLK_QC_T_NONE;
415 /* mddev_suspend makes sure no new requests are submitted
416 * to the device, and that any requests that have been submitted
417 * are completely handled.
418 * Once mddev_detach() is called and completes, the module will be
421 void mddev_suspend(struct mddev *mddev)
423 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
424 lockdep_assert_held(&mddev->reconfig_mutex);
425 if (mddev->suspended++)
428 wake_up(&mddev->sb_wait);
429 set_bit(MD_ALLOW_SB_UPDATE, &mddev->flags);
430 smp_mb__after_atomic();
431 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
432 mddev->pers->quiesce(mddev, 1);
433 clear_bit_unlock(MD_ALLOW_SB_UPDATE, &mddev->flags);
434 wait_event(mddev->sb_wait, !test_bit(MD_UPDATING_SB, &mddev->flags));
436 del_timer_sync(&mddev->safemode_timer);
438 EXPORT_SYMBOL_GPL(mddev_suspend);
440 void mddev_resume(struct mddev *mddev)
442 lockdep_assert_held(&mddev->reconfig_mutex);
443 if (--mddev->suspended)
445 wake_up(&mddev->sb_wait);
446 mddev->pers->quiesce(mddev, 0);
448 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
449 md_wakeup_thread(mddev->thread);
450 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
452 EXPORT_SYMBOL_GPL(mddev_resume);
454 int mddev_congested(struct mddev *mddev, int bits)
456 struct md_personality *pers = mddev->pers;
460 if (mddev->suspended)
462 else if (pers && pers->congested)
463 ret = pers->congested(mddev, bits);
467 EXPORT_SYMBOL_GPL(mddev_congested);
468 static int md_congested(void *data, int bits)
470 struct mddev *mddev = data;
471 return mddev_congested(mddev, bits);
475 * Generic flush handling for md
478 static void md_end_flush(struct bio *bio)
480 struct md_rdev *rdev = bio->bi_private;
481 struct mddev *mddev = rdev->mddev;
483 rdev_dec_pending(rdev, mddev);
485 if (atomic_dec_and_test(&mddev->flush_pending)) {
486 /* The pre-request flush has finished */
487 queue_work(md_wq, &mddev->flush_work);
492 static void md_submit_flush_data(struct work_struct *ws);
494 static void submit_flushes(struct work_struct *ws)
496 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
497 struct md_rdev *rdev;
499 mddev->start_flush = ktime_get_boottime();
500 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
501 atomic_set(&mddev->flush_pending, 1);
503 rdev_for_each_rcu(rdev, mddev)
504 if (rdev->raid_disk >= 0 &&
505 !test_bit(Faulty, &rdev->flags)) {
506 /* Take two references, one is dropped
507 * when request finishes, one after
508 * we reclaim rcu_read_lock
511 atomic_inc(&rdev->nr_pending);
512 atomic_inc(&rdev->nr_pending);
514 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
515 bi->bi_end_io = md_end_flush;
516 bi->bi_private = rdev;
517 bio_set_dev(bi, rdev->bdev);
518 bi->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
519 atomic_inc(&mddev->flush_pending);
522 rdev_dec_pending(rdev, mddev);
525 if (atomic_dec_and_test(&mddev->flush_pending))
526 queue_work(md_wq, &mddev->flush_work);
529 static void md_submit_flush_data(struct work_struct *ws)
531 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
532 struct bio *bio = mddev->flush_bio;
535 * must reset flush_bio before calling into md_handle_request to avoid a
536 * deadlock, because other bios passed md_handle_request suspend check
537 * could wait for this and below md_handle_request could wait for those
538 * bios because of suspend check
540 mddev->last_flush = mddev->start_flush;
541 mddev->flush_bio = NULL;
542 wake_up(&mddev->sb_wait);
544 if (bio->bi_iter.bi_size == 0) {
545 /* an empty barrier - all done */
548 bio->bi_opf &= ~REQ_PREFLUSH;
549 md_handle_request(mddev, bio);
554 * Manages consolidation of flushes and submitting any flushes needed for
555 * a bio with REQ_PREFLUSH. Returns true if the bio is finished or is
556 * being finished in another context. Returns false if the flushing is
557 * complete but still needs the I/O portion of the bio to be processed.
559 bool md_flush_request(struct mddev *mddev, struct bio *bio)
561 ktime_t start = ktime_get_boottime();
562 spin_lock_irq(&mddev->lock);
563 wait_event_lock_irq(mddev->sb_wait,
565 ktime_after(mddev->last_flush, start),
567 if (!ktime_after(mddev->last_flush, start)) {
568 WARN_ON(mddev->flush_bio);
569 mddev->flush_bio = bio;
572 spin_unlock_irq(&mddev->lock);
575 INIT_WORK(&mddev->flush_work, submit_flushes);
576 queue_work(md_wq, &mddev->flush_work);
578 /* flush was performed for some other bio while we waited. */
579 if (bio->bi_iter.bi_size == 0)
580 /* an empty barrier - all done */
583 bio->bi_opf &= ~REQ_PREFLUSH;
589 EXPORT_SYMBOL(md_flush_request);
591 static inline struct mddev *mddev_get(struct mddev *mddev)
593 atomic_inc(&mddev->active);
597 static void mddev_delayed_delete(struct work_struct *ws);
599 static void mddev_put(struct mddev *mddev)
601 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
603 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
604 mddev->ctime == 0 && !mddev->hold_active) {
605 /* Array is not configured at all, and not held active,
607 list_del_init(&mddev->all_mddevs);
610 * Call queue_work inside the spinlock so that
611 * flush_workqueue() after mddev_find will succeed in waiting
612 * for the work to be done.
614 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
615 queue_work(md_misc_wq, &mddev->del_work);
617 spin_unlock(&all_mddevs_lock);
620 static void md_safemode_timeout(struct timer_list *t);
622 void mddev_init(struct mddev *mddev)
624 kobject_init(&mddev->kobj, &md_ktype);
625 mutex_init(&mddev->open_mutex);
626 mutex_init(&mddev->reconfig_mutex);
627 mutex_init(&mddev->bitmap_info.mutex);
628 INIT_LIST_HEAD(&mddev->disks);
629 INIT_LIST_HEAD(&mddev->all_mddevs);
630 timer_setup(&mddev->safemode_timer, md_safemode_timeout, 0);
631 atomic_set(&mddev->active, 1);
632 atomic_set(&mddev->openers, 0);
633 atomic_set(&mddev->active_io, 0);
634 spin_lock_init(&mddev->lock);
635 atomic_set(&mddev->flush_pending, 0);
636 init_waitqueue_head(&mddev->sb_wait);
637 init_waitqueue_head(&mddev->recovery_wait);
638 mddev->reshape_position = MaxSector;
639 mddev->reshape_backwards = 0;
640 mddev->last_sync_action = "none";
641 mddev->resync_min = 0;
642 mddev->resync_max = MaxSector;
643 mddev->level = LEVEL_NONE;
645 EXPORT_SYMBOL_GPL(mddev_init);
647 static struct mddev *mddev_find(dev_t unit)
649 struct mddev *mddev, *new = NULL;
651 if (unit && MAJOR(unit) != MD_MAJOR)
652 unit &= ~((1<<MdpMinorShift)-1);
655 spin_lock(&all_mddevs_lock);
658 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
659 if (mddev->unit == unit) {
661 spin_unlock(&all_mddevs_lock);
667 list_add(&new->all_mddevs, &all_mddevs);
668 spin_unlock(&all_mddevs_lock);
669 new->hold_active = UNTIL_IOCTL;
673 /* find an unused unit number */
674 static int next_minor = 512;
675 int start = next_minor;
679 dev = MKDEV(MD_MAJOR, next_minor);
681 if (next_minor > MINORMASK)
683 if (next_minor == start) {
684 /* Oh dear, all in use. */
685 spin_unlock(&all_mddevs_lock);
691 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
692 if (mddev->unit == dev) {
698 new->md_minor = MINOR(dev);
699 new->hold_active = UNTIL_STOP;
700 list_add(&new->all_mddevs, &all_mddevs);
701 spin_unlock(&all_mddevs_lock);
704 spin_unlock(&all_mddevs_lock);
706 new = kzalloc(sizeof(*new), GFP_KERNEL);
711 if (MAJOR(unit) == MD_MAJOR)
712 new->md_minor = MINOR(unit);
714 new->md_minor = MINOR(unit) >> MdpMinorShift;
721 static struct attribute_group md_redundancy_group;
723 void mddev_unlock(struct mddev *mddev)
725 if (mddev->to_remove) {
726 /* These cannot be removed under reconfig_mutex as
727 * an access to the files will try to take reconfig_mutex
728 * while holding the file unremovable, which leads to
730 * So hold set sysfs_active while the remove in happeing,
731 * and anything else which might set ->to_remove or my
732 * otherwise change the sysfs namespace will fail with
733 * -EBUSY if sysfs_active is still set.
734 * We set sysfs_active under reconfig_mutex and elsewhere
735 * test it under the same mutex to ensure its correct value
738 struct attribute_group *to_remove = mddev->to_remove;
739 mddev->to_remove = NULL;
740 mddev->sysfs_active = 1;
741 mutex_unlock(&mddev->reconfig_mutex);
743 if (mddev->kobj.sd) {
744 if (to_remove != &md_redundancy_group)
745 sysfs_remove_group(&mddev->kobj, to_remove);
746 if (mddev->pers == NULL ||
747 mddev->pers->sync_request == NULL) {
748 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
749 if (mddev->sysfs_action)
750 sysfs_put(mddev->sysfs_action);
751 mddev->sysfs_action = NULL;
754 mddev->sysfs_active = 0;
756 mutex_unlock(&mddev->reconfig_mutex);
758 /* As we've dropped the mutex we need a spinlock to
759 * make sure the thread doesn't disappear
761 spin_lock(&pers_lock);
762 md_wakeup_thread(mddev->thread);
763 wake_up(&mddev->sb_wait);
764 spin_unlock(&pers_lock);
766 EXPORT_SYMBOL_GPL(mddev_unlock);
768 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
770 struct md_rdev *rdev;
772 rdev_for_each_rcu(rdev, mddev)
773 if (rdev->desc_nr == nr)
778 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
780 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
782 struct md_rdev *rdev;
784 rdev_for_each(rdev, mddev)
785 if (rdev->bdev->bd_dev == dev)
791 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
793 struct md_rdev *rdev;
795 rdev_for_each_rcu(rdev, mddev)
796 if (rdev->bdev->bd_dev == dev)
801 EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
803 static struct md_personality *find_pers(int level, char *clevel)
805 struct md_personality *pers;
806 list_for_each_entry(pers, &pers_list, list) {
807 if (level != LEVEL_NONE && pers->level == level)
809 if (strcmp(pers->name, clevel)==0)
815 /* return the offset of the super block in 512byte sectors */
816 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
818 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
819 return MD_NEW_SIZE_SECTORS(num_sectors);
822 static int alloc_disk_sb(struct md_rdev *rdev)
824 rdev->sb_page = alloc_page(GFP_KERNEL);
830 void md_rdev_clear(struct md_rdev *rdev)
833 put_page(rdev->sb_page);
835 rdev->sb_page = NULL;
840 put_page(rdev->bb_page);
841 rdev->bb_page = NULL;
843 badblocks_exit(&rdev->badblocks);
845 EXPORT_SYMBOL_GPL(md_rdev_clear);
847 static void super_written(struct bio *bio)
849 struct md_rdev *rdev = bio->bi_private;
850 struct mddev *mddev = rdev->mddev;
852 if (bio->bi_status) {
853 pr_err("md: super_written gets error=%d\n", bio->bi_status);
854 md_error(mddev, rdev);
855 if (!test_bit(Faulty, &rdev->flags)
856 && (bio->bi_opf & MD_FAILFAST)) {
857 set_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags);
858 set_bit(LastDev, &rdev->flags);
861 clear_bit(LastDev, &rdev->flags);
863 if (atomic_dec_and_test(&mddev->pending_writes))
864 wake_up(&mddev->sb_wait);
865 rdev_dec_pending(rdev, mddev);
869 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
870 sector_t sector, int size, struct page *page)
872 /* write first size bytes of page to sector of rdev
873 * Increment mddev->pending_writes before returning
874 * and decrement it on completion, waking up sb_wait
875 * if zero is reached.
876 * If an error occurred, call md_error
884 if (test_bit(Faulty, &rdev->flags))
887 bio = md_bio_alloc_sync(mddev);
889 atomic_inc(&rdev->nr_pending);
891 bio_set_dev(bio, rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev);
892 bio->bi_iter.bi_sector = sector;
893 bio_add_page(bio, page, size, 0);
894 bio->bi_private = rdev;
895 bio->bi_end_io = super_written;
897 if (test_bit(MD_FAILFAST_SUPPORTED, &mddev->flags) &&
898 test_bit(FailFast, &rdev->flags) &&
899 !test_bit(LastDev, &rdev->flags))
901 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC | REQ_PREFLUSH | REQ_FUA | ff;
903 atomic_inc(&mddev->pending_writes);
907 int md_super_wait(struct mddev *mddev)
909 /* wait for all superblock writes that were scheduled to complete */
910 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
911 if (test_and_clear_bit(MD_SB_NEED_REWRITE, &mddev->sb_flags))
916 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
917 struct page *page, int op, int op_flags, bool metadata_op)
919 struct bio *bio = md_bio_alloc_sync(rdev->mddev);
922 if (metadata_op && rdev->meta_bdev)
923 bio_set_dev(bio, rdev->meta_bdev);
925 bio_set_dev(bio, rdev->bdev);
926 bio_set_op_attrs(bio, op, op_flags);
928 bio->bi_iter.bi_sector = sector + rdev->sb_start;
929 else if (rdev->mddev->reshape_position != MaxSector &&
930 (rdev->mddev->reshape_backwards ==
931 (sector >= rdev->mddev->reshape_position)))
932 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
934 bio->bi_iter.bi_sector = sector + rdev->data_offset;
935 bio_add_page(bio, page, size, 0);
937 submit_bio_wait(bio);
939 ret = !bio->bi_status;
943 EXPORT_SYMBOL_GPL(sync_page_io);
945 static int read_disk_sb(struct md_rdev *rdev, int size)
947 char b[BDEVNAME_SIZE];
952 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
958 pr_err("md: disabled device %s, could not read superblock.\n",
959 bdevname(rdev->bdev,b));
963 static int md_uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
965 return sb1->set_uuid0 == sb2->set_uuid0 &&
966 sb1->set_uuid1 == sb2->set_uuid1 &&
967 sb1->set_uuid2 == sb2->set_uuid2 &&
968 sb1->set_uuid3 == sb2->set_uuid3;
971 static int md_sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
974 mdp_super_t *tmp1, *tmp2;
976 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
977 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
979 if (!tmp1 || !tmp2) {
988 * nr_disks is not constant
993 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
1000 static u32 md_csum_fold(u32 csum)
1002 csum = (csum & 0xffff) + (csum >> 16);
1003 return (csum & 0xffff) + (csum >> 16);
1006 static unsigned int calc_sb_csum(mdp_super_t *sb)
1009 u32 *sb32 = (u32*)sb;
1011 unsigned int disk_csum, csum;
1013 disk_csum = sb->sb_csum;
1016 for (i = 0; i < MD_SB_BYTES/4 ; i++)
1018 csum = (newcsum & 0xffffffff) + (newcsum>>32);
1021 /* This used to use csum_partial, which was wrong for several
1022 * reasons including that different results are returned on
1023 * different architectures. It isn't critical that we get exactly
1024 * the same return value as before (we always csum_fold before
1025 * testing, and that removes any differences). However as we
1026 * know that csum_partial always returned a 16bit value on
1027 * alphas, do a fold to maximise conformity to previous behaviour.
1029 sb->sb_csum = md_csum_fold(disk_csum);
1031 sb->sb_csum = disk_csum;
1037 * Handle superblock details.
1038 * We want to be able to handle multiple superblock formats
1039 * so we have a common interface to them all, and an array of
1040 * different handlers.
1041 * We rely on user-space to write the initial superblock, and support
1042 * reading and updating of superblocks.
1043 * Interface methods are:
1044 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1045 * loads and validates a superblock on dev.
1046 * if refdev != NULL, compare superblocks on both devices
1048 * 0 - dev has a superblock that is compatible with refdev
1049 * 1 - dev has a superblock that is compatible and newer than refdev
1050 * so dev should be used as the refdev in future
1051 * -EINVAL superblock incompatible or invalid
1052 * -othererror e.g. -EIO
1054 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
1055 * Verify that dev is acceptable into mddev.
1056 * The first time, mddev->raid_disks will be 0, and data from
1057 * dev should be merged in. Subsequent calls check that dev
1058 * is new enough. Return 0 or -EINVAL
1060 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
1061 * Update the superblock for rdev with data in mddev
1062 * This does not write to disc.
1068 struct module *owner;
1069 int (*load_super)(struct md_rdev *rdev,
1070 struct md_rdev *refdev,
1072 int (*validate_super)(struct mddev *mddev,
1073 struct md_rdev *rdev);
1074 void (*sync_super)(struct mddev *mddev,
1075 struct md_rdev *rdev);
1076 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
1077 sector_t num_sectors);
1078 int (*allow_new_offset)(struct md_rdev *rdev,
1079 unsigned long long new_offset);
1083 * Check that the given mddev has no bitmap.
1085 * This function is called from the run method of all personalities that do not
1086 * support bitmaps. It prints an error message and returns non-zero if mddev
1087 * has a bitmap. Otherwise, it returns 0.
1090 int md_check_no_bitmap(struct mddev *mddev)
1092 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1094 pr_warn("%s: bitmaps are not supported for %s\n",
1095 mdname(mddev), mddev->pers->name);
1098 EXPORT_SYMBOL(md_check_no_bitmap);
1101 * load_super for 0.90.0
1103 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1105 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1108 bool spare_disk = true;
1111 * Calculate the position of the superblock (512byte sectors),
1112 * it's at the end of the disk.
1114 * It also happens to be a multiple of 4Kb.
1116 rdev->sb_start = calc_dev_sboffset(rdev);
1118 ret = read_disk_sb(rdev, MD_SB_BYTES);
1124 bdevname(rdev->bdev, b);
1125 sb = page_address(rdev->sb_page);
1127 if (sb->md_magic != MD_SB_MAGIC) {
1128 pr_warn("md: invalid raid superblock magic on %s\n", b);
1132 if (sb->major_version != 0 ||
1133 sb->minor_version < 90 ||
1134 sb->minor_version > 91) {
1135 pr_warn("Bad version number %d.%d on %s\n",
1136 sb->major_version, sb->minor_version, b);
1140 if (sb->raid_disks <= 0)
1143 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1144 pr_warn("md: invalid superblock checksum on %s\n", b);
1148 rdev->preferred_minor = sb->md_minor;
1149 rdev->data_offset = 0;
1150 rdev->new_data_offset = 0;
1151 rdev->sb_size = MD_SB_BYTES;
1152 rdev->badblocks.shift = -1;
1154 if (sb->level == LEVEL_MULTIPATH)
1157 rdev->desc_nr = sb->this_disk.number;
1159 /* not spare disk, or LEVEL_MULTIPATH */
1160 if (sb->level == LEVEL_MULTIPATH ||
1161 (rdev->desc_nr >= 0 &&
1162 rdev->desc_nr < MD_SB_DISKS &&
1163 sb->disks[rdev->desc_nr].state &
1164 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1174 mdp_super_t *refsb = page_address(refdev->sb_page);
1175 if (!md_uuid_equal(refsb, sb)) {
1176 pr_warn("md: %s has different UUID to %s\n",
1177 b, bdevname(refdev->bdev,b2));
1180 if (!md_sb_equal(refsb, sb)) {
1181 pr_warn("md: %s has same UUID but different superblock to %s\n",
1182 b, bdevname(refdev->bdev, b2));
1186 ev2 = md_event(refsb);
1188 if (!spare_disk && ev1 > ev2)
1193 rdev->sectors = rdev->sb_start;
1194 /* Limit to 4TB as metadata cannot record more than that.
1195 * (not needed for Linear and RAID0 as metadata doesn't
1198 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1199 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1201 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1202 /* "this cannot possibly happen" ... */
1210 * validate_super for 0.90.0
1212 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1215 mdp_super_t *sb = page_address(rdev->sb_page);
1216 __u64 ev1 = md_event(sb);
1218 rdev->raid_disk = -1;
1219 clear_bit(Faulty, &rdev->flags);
1220 clear_bit(In_sync, &rdev->flags);
1221 clear_bit(Bitmap_sync, &rdev->flags);
1222 clear_bit(WriteMostly, &rdev->flags);
1224 if (mddev->raid_disks == 0) {
1225 mddev->major_version = 0;
1226 mddev->minor_version = sb->minor_version;
1227 mddev->patch_version = sb->patch_version;
1228 mddev->external = 0;
1229 mddev->chunk_sectors = sb->chunk_size >> 9;
1230 mddev->ctime = sb->ctime;
1231 mddev->utime = sb->utime;
1232 mddev->level = sb->level;
1233 mddev->clevel[0] = 0;
1234 mddev->layout = sb->layout;
1235 mddev->raid_disks = sb->raid_disks;
1236 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1237 mddev->events = ev1;
1238 mddev->bitmap_info.offset = 0;
1239 mddev->bitmap_info.space = 0;
1240 /* bitmap can use 60 K after the 4K superblocks */
1241 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1242 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1243 mddev->reshape_backwards = 0;
1245 if (mddev->minor_version >= 91) {
1246 mddev->reshape_position = sb->reshape_position;
1247 mddev->delta_disks = sb->delta_disks;
1248 mddev->new_level = sb->new_level;
1249 mddev->new_layout = sb->new_layout;
1250 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1251 if (mddev->delta_disks < 0)
1252 mddev->reshape_backwards = 1;
1254 mddev->reshape_position = MaxSector;
1255 mddev->delta_disks = 0;
1256 mddev->new_level = mddev->level;
1257 mddev->new_layout = mddev->layout;
1258 mddev->new_chunk_sectors = mddev->chunk_sectors;
1260 if (mddev->level == 0)
1263 if (sb->state & (1<<MD_SB_CLEAN))
1264 mddev->recovery_cp = MaxSector;
1266 if (sb->events_hi == sb->cp_events_hi &&
1267 sb->events_lo == sb->cp_events_lo) {
1268 mddev->recovery_cp = sb->recovery_cp;
1270 mddev->recovery_cp = 0;
1273 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1274 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1275 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1276 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1278 mddev->max_disks = MD_SB_DISKS;
1280 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1281 mddev->bitmap_info.file == NULL) {
1282 mddev->bitmap_info.offset =
1283 mddev->bitmap_info.default_offset;
1284 mddev->bitmap_info.space =
1285 mddev->bitmap_info.default_space;
1288 } else if (mddev->pers == NULL) {
1289 /* Insist on good event counter while assembling, except
1290 * for spares (which don't need an event count) */
1292 if (sb->disks[rdev->desc_nr].state & (
1293 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1294 if (ev1 < mddev->events)
1296 } else if (mddev->bitmap) {
1297 /* if adding to array with a bitmap, then we can accept an
1298 * older device ... but not too old.
1300 if (ev1 < mddev->bitmap->events_cleared)
1302 if (ev1 < mddev->events)
1303 set_bit(Bitmap_sync, &rdev->flags);
1305 if (ev1 < mddev->events)
1306 /* just a hot-add of a new device, leave raid_disk at -1 */
1310 if (mddev->level != LEVEL_MULTIPATH) {
1311 desc = sb->disks + rdev->desc_nr;
1313 if (desc->state & (1<<MD_DISK_FAULTY))
1314 set_bit(Faulty, &rdev->flags);
1315 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1316 desc->raid_disk < mddev->raid_disks */) {
1317 set_bit(In_sync, &rdev->flags);
1318 rdev->raid_disk = desc->raid_disk;
1319 rdev->saved_raid_disk = desc->raid_disk;
1320 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1321 /* active but not in sync implies recovery up to
1322 * reshape position. We don't know exactly where
1323 * that is, so set to zero for now */
1324 if (mddev->minor_version >= 91) {
1325 rdev->recovery_offset = 0;
1326 rdev->raid_disk = desc->raid_disk;
1329 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1330 set_bit(WriteMostly, &rdev->flags);
1331 if (desc->state & (1<<MD_DISK_FAILFAST))
1332 set_bit(FailFast, &rdev->flags);
1333 } else /* MULTIPATH are always insync */
1334 set_bit(In_sync, &rdev->flags);
1339 * sync_super for 0.90.0
1341 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1344 struct md_rdev *rdev2;
1345 int next_spare = mddev->raid_disks;
1347 /* make rdev->sb match mddev data..
1350 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1351 * 3/ any empty disks < next_spare become removed
1353 * disks[0] gets initialised to REMOVED because
1354 * we cannot be sure from other fields if it has
1355 * been initialised or not.
1358 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1360 rdev->sb_size = MD_SB_BYTES;
1362 sb = page_address(rdev->sb_page);
1364 memset(sb, 0, sizeof(*sb));
1366 sb->md_magic = MD_SB_MAGIC;
1367 sb->major_version = mddev->major_version;
1368 sb->patch_version = mddev->patch_version;
1369 sb->gvalid_words = 0; /* ignored */
1370 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1371 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1372 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1373 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1375 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1376 sb->level = mddev->level;
1377 sb->size = mddev->dev_sectors / 2;
1378 sb->raid_disks = mddev->raid_disks;
1379 sb->md_minor = mddev->md_minor;
1380 sb->not_persistent = 0;
1381 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1383 sb->events_hi = (mddev->events>>32);
1384 sb->events_lo = (u32)mddev->events;
1386 if (mddev->reshape_position == MaxSector)
1387 sb->minor_version = 90;
1389 sb->minor_version = 91;
1390 sb->reshape_position = mddev->reshape_position;
1391 sb->new_level = mddev->new_level;
1392 sb->delta_disks = mddev->delta_disks;
1393 sb->new_layout = mddev->new_layout;
1394 sb->new_chunk = mddev->new_chunk_sectors << 9;
1396 mddev->minor_version = sb->minor_version;
1399 sb->recovery_cp = mddev->recovery_cp;
1400 sb->cp_events_hi = (mddev->events>>32);
1401 sb->cp_events_lo = (u32)mddev->events;
1402 if (mddev->recovery_cp == MaxSector)
1403 sb->state = (1<< MD_SB_CLEAN);
1405 sb->recovery_cp = 0;
1407 sb->layout = mddev->layout;
1408 sb->chunk_size = mddev->chunk_sectors << 9;
1410 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1411 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1413 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1414 rdev_for_each(rdev2, mddev) {
1417 int is_active = test_bit(In_sync, &rdev2->flags);
1419 if (rdev2->raid_disk >= 0 &&
1420 sb->minor_version >= 91)
1421 /* we have nowhere to store the recovery_offset,
1422 * but if it is not below the reshape_position,
1423 * we can piggy-back on that.
1426 if (rdev2->raid_disk < 0 ||
1427 test_bit(Faulty, &rdev2->flags))
1430 desc_nr = rdev2->raid_disk;
1432 desc_nr = next_spare++;
1433 rdev2->desc_nr = desc_nr;
1434 d = &sb->disks[rdev2->desc_nr];
1436 d->number = rdev2->desc_nr;
1437 d->major = MAJOR(rdev2->bdev->bd_dev);
1438 d->minor = MINOR(rdev2->bdev->bd_dev);
1440 d->raid_disk = rdev2->raid_disk;
1442 d->raid_disk = rdev2->desc_nr; /* compatibility */
1443 if (test_bit(Faulty, &rdev2->flags))
1444 d->state = (1<<MD_DISK_FAULTY);
1445 else if (is_active) {
1446 d->state = (1<<MD_DISK_ACTIVE);
1447 if (test_bit(In_sync, &rdev2->flags))
1448 d->state |= (1<<MD_DISK_SYNC);
1456 if (test_bit(WriteMostly, &rdev2->flags))
1457 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1458 if (test_bit(FailFast, &rdev2->flags))
1459 d->state |= (1<<MD_DISK_FAILFAST);
1461 /* now set the "removed" and "faulty" bits on any missing devices */
1462 for (i=0 ; i < mddev->raid_disks ; i++) {
1463 mdp_disk_t *d = &sb->disks[i];
1464 if (d->state == 0 && d->number == 0) {
1467 d->state = (1<<MD_DISK_REMOVED);
1468 d->state |= (1<<MD_DISK_FAULTY);
1472 sb->nr_disks = nr_disks;
1473 sb->active_disks = active;
1474 sb->working_disks = working;
1475 sb->failed_disks = failed;
1476 sb->spare_disks = spare;
1478 sb->this_disk = sb->disks[rdev->desc_nr];
1479 sb->sb_csum = calc_sb_csum(sb);
1483 * rdev_size_change for 0.90.0
1485 static unsigned long long
1486 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1488 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1489 return 0; /* component must fit device */
1490 if (rdev->mddev->bitmap_info.offset)
1491 return 0; /* can't move bitmap */
1492 rdev->sb_start = calc_dev_sboffset(rdev);
1493 if (!num_sectors || num_sectors > rdev->sb_start)
1494 num_sectors = rdev->sb_start;
1495 /* Limit to 4TB as metadata cannot record more than that.
1496 * 4TB == 2^32 KB, or 2*2^32 sectors.
1498 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1499 num_sectors = (sector_t)(2ULL << 32) - 2;
1501 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1503 } while (md_super_wait(rdev->mddev) < 0);
1508 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1510 /* non-zero offset changes not possible with v0.90 */
1511 return new_offset == 0;
1515 * version 1 superblock
1518 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1522 unsigned long long newcsum;
1523 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1524 __le32 *isuper = (__le32*)sb;
1526 disk_csum = sb->sb_csum;
1529 for (; size >= 4; size -= 4)
1530 newcsum += le32_to_cpu(*isuper++);
1533 newcsum += le16_to_cpu(*(__le16*) isuper);
1535 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1536 sb->sb_csum = disk_csum;
1537 return cpu_to_le32(csum);
1540 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1542 struct mdp_superblock_1 *sb;
1546 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1548 bool spare_disk = true;
1551 * Calculate the position of the superblock in 512byte sectors.
1552 * It is always aligned to a 4K boundary and
1553 * depeding on minor_version, it can be:
1554 * 0: At least 8K, but less than 12K, from end of device
1555 * 1: At start of device
1556 * 2: 4K from start of device.
1558 switch(minor_version) {
1560 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1562 sb_start &= ~(sector_t)(4*2-1);
1573 rdev->sb_start = sb_start;
1575 /* superblock is rarely larger than 1K, but it can be larger,
1576 * and it is safe to read 4k, so we do that
1578 ret = read_disk_sb(rdev, 4096);
1579 if (ret) return ret;
1581 sb = page_address(rdev->sb_page);
1583 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1584 sb->major_version != cpu_to_le32(1) ||
1585 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1586 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1587 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1590 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1591 pr_warn("md: invalid superblock checksum on %s\n",
1592 bdevname(rdev->bdev,b));
1595 if (le64_to_cpu(sb->data_size) < 10) {
1596 pr_warn("md: data_size too small on %s\n",
1597 bdevname(rdev->bdev,b));
1602 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1603 /* Some padding is non-zero, might be a new feature */
1606 rdev->preferred_minor = 0xffff;
1607 rdev->data_offset = le64_to_cpu(sb->data_offset);
1608 rdev->new_data_offset = rdev->data_offset;
1609 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1610 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1611 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1612 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1614 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1615 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1616 if (rdev->sb_size & bmask)
1617 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1620 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1623 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1626 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1629 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1631 if (!rdev->bb_page) {
1632 rdev->bb_page = alloc_page(GFP_KERNEL);
1636 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1637 rdev->badblocks.count == 0) {
1638 /* need to load the bad block list.
1639 * Currently we limit it to one page.
1645 int sectors = le16_to_cpu(sb->bblog_size);
1646 if (sectors > (PAGE_SIZE / 512))
1648 offset = le32_to_cpu(sb->bblog_offset);
1651 bb_sector = (long long)offset;
1652 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1653 rdev->bb_page, REQ_OP_READ, 0, true))
1655 bbp = (__le64 *)page_address(rdev->bb_page);
1656 rdev->badblocks.shift = sb->bblog_shift;
1657 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1658 u64 bb = le64_to_cpu(*bbp);
1659 int count = bb & (0x3ff);
1660 u64 sector = bb >> 10;
1661 sector <<= sb->bblog_shift;
1662 count <<= sb->bblog_shift;
1665 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1668 } else if (sb->bblog_offset != 0)
1669 rdev->badblocks.shift = 0;
1671 if ((le32_to_cpu(sb->feature_map) &
1672 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1673 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1674 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1675 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1678 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1682 /* not spare disk, or LEVEL_MULTIPATH */
1683 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1684 (rdev->desc_nr >= 0 &&
1685 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1686 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1687 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1697 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1699 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1700 sb->level != refsb->level ||
1701 sb->layout != refsb->layout ||
1702 sb->chunksize != refsb->chunksize) {
1703 pr_warn("md: %s has strangely different superblock to %s\n",
1704 bdevname(rdev->bdev,b),
1705 bdevname(refdev->bdev,b2));
1708 ev1 = le64_to_cpu(sb->events);
1709 ev2 = le64_to_cpu(refsb->events);
1711 if (!spare_disk && ev1 > ev2)
1716 if (minor_version) {
1717 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1718 sectors -= rdev->data_offset;
1720 sectors = rdev->sb_start;
1721 if (sectors < le64_to_cpu(sb->data_size))
1723 rdev->sectors = le64_to_cpu(sb->data_size);
1727 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1729 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1730 __u64 ev1 = le64_to_cpu(sb->events);
1732 rdev->raid_disk = -1;
1733 clear_bit(Faulty, &rdev->flags);
1734 clear_bit(In_sync, &rdev->flags);
1735 clear_bit(Bitmap_sync, &rdev->flags);
1736 clear_bit(WriteMostly, &rdev->flags);
1738 if (mddev->raid_disks == 0) {
1739 mddev->major_version = 1;
1740 mddev->patch_version = 0;
1741 mddev->external = 0;
1742 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1743 mddev->ctime = le64_to_cpu(sb->ctime);
1744 mddev->utime = le64_to_cpu(sb->utime);
1745 mddev->level = le32_to_cpu(sb->level);
1746 mddev->clevel[0] = 0;
1747 mddev->layout = le32_to_cpu(sb->layout);
1748 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1749 mddev->dev_sectors = le64_to_cpu(sb->size);
1750 mddev->events = ev1;
1751 mddev->bitmap_info.offset = 0;
1752 mddev->bitmap_info.space = 0;
1753 /* Default location for bitmap is 1K after superblock
1754 * using 3K - total of 4K
1756 mddev->bitmap_info.default_offset = 1024 >> 9;
1757 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1758 mddev->reshape_backwards = 0;
1760 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1761 memcpy(mddev->uuid, sb->set_uuid, 16);
1763 mddev->max_disks = (4096-256)/2;
1765 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1766 mddev->bitmap_info.file == NULL) {
1767 mddev->bitmap_info.offset =
1768 (__s32)le32_to_cpu(sb->bitmap_offset);
1769 /* Metadata doesn't record how much space is available.
1770 * For 1.0, we assume we can use up to the superblock
1771 * if before, else to 4K beyond superblock.
1772 * For others, assume no change is possible.
1774 if (mddev->minor_version > 0)
1775 mddev->bitmap_info.space = 0;
1776 else if (mddev->bitmap_info.offset > 0)
1777 mddev->bitmap_info.space =
1778 8 - mddev->bitmap_info.offset;
1780 mddev->bitmap_info.space =
1781 -mddev->bitmap_info.offset;
1784 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1785 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1786 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1787 mddev->new_level = le32_to_cpu(sb->new_level);
1788 mddev->new_layout = le32_to_cpu(sb->new_layout);
1789 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1790 if (mddev->delta_disks < 0 ||
1791 (mddev->delta_disks == 0 &&
1792 (le32_to_cpu(sb->feature_map)
1793 & MD_FEATURE_RESHAPE_BACKWARDS)))
1794 mddev->reshape_backwards = 1;
1796 mddev->reshape_position = MaxSector;
1797 mddev->delta_disks = 0;
1798 mddev->new_level = mddev->level;
1799 mddev->new_layout = mddev->layout;
1800 mddev->new_chunk_sectors = mddev->chunk_sectors;
1803 if (mddev->level == 0 &&
1804 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1807 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1808 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1810 if (le32_to_cpu(sb->feature_map) &
1811 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1812 if (le32_to_cpu(sb->feature_map) &
1813 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1815 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1816 (le32_to_cpu(sb->feature_map) &
1817 MD_FEATURE_MULTIPLE_PPLS))
1819 set_bit(MD_HAS_PPL, &mddev->flags);
1821 } else if (mddev->pers == NULL) {
1822 /* Insist of good event counter while assembling, except for
1823 * spares (which don't need an event count) */
1825 if (rdev->desc_nr >= 0 &&
1826 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1827 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1828 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1829 if (ev1 < mddev->events)
1831 } else if (mddev->bitmap) {
1832 /* If adding to array with a bitmap, then we can accept an
1833 * older device, but not too old.
1835 if (ev1 < mddev->bitmap->events_cleared)
1837 if (ev1 < mddev->events)
1838 set_bit(Bitmap_sync, &rdev->flags);
1840 if (ev1 < mddev->events)
1841 /* just a hot-add of a new device, leave raid_disk at -1 */
1844 if (mddev->level != LEVEL_MULTIPATH) {
1846 if (rdev->desc_nr < 0 ||
1847 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1848 role = MD_DISK_ROLE_SPARE;
1851 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1853 case MD_DISK_ROLE_SPARE: /* spare */
1855 case MD_DISK_ROLE_FAULTY: /* faulty */
1856 set_bit(Faulty, &rdev->flags);
1858 case MD_DISK_ROLE_JOURNAL: /* journal device */
1859 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1860 /* journal device without journal feature */
1861 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1864 set_bit(Journal, &rdev->flags);
1865 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1866 rdev->raid_disk = 0;
1869 rdev->saved_raid_disk = role;
1870 if ((le32_to_cpu(sb->feature_map) &
1871 MD_FEATURE_RECOVERY_OFFSET)) {
1872 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1873 if (!(le32_to_cpu(sb->feature_map) &
1874 MD_FEATURE_RECOVERY_BITMAP))
1875 rdev->saved_raid_disk = -1;
1878 * If the array is FROZEN, then the device can't
1879 * be in_sync with rest of array.
1881 if (!test_bit(MD_RECOVERY_FROZEN,
1883 set_bit(In_sync, &rdev->flags);
1885 rdev->raid_disk = role;
1888 if (sb->devflags & WriteMostly1)
1889 set_bit(WriteMostly, &rdev->flags);
1890 if (sb->devflags & FailFast1)
1891 set_bit(FailFast, &rdev->flags);
1892 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1893 set_bit(Replacement, &rdev->flags);
1894 } else /* MULTIPATH are always insync */
1895 set_bit(In_sync, &rdev->flags);
1900 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1902 struct mdp_superblock_1 *sb;
1903 struct md_rdev *rdev2;
1905 /* make rdev->sb match mddev and rdev data. */
1907 sb = page_address(rdev->sb_page);
1909 sb->feature_map = 0;
1911 sb->recovery_offset = cpu_to_le64(0);
1912 memset(sb->pad3, 0, sizeof(sb->pad3));
1914 sb->utime = cpu_to_le64((__u64)mddev->utime);
1915 sb->events = cpu_to_le64(mddev->events);
1917 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1918 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1919 sb->resync_offset = cpu_to_le64(MaxSector);
1921 sb->resync_offset = cpu_to_le64(0);
1923 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1925 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1926 sb->size = cpu_to_le64(mddev->dev_sectors);
1927 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1928 sb->level = cpu_to_le32(mddev->level);
1929 sb->layout = cpu_to_le32(mddev->layout);
1930 if (test_bit(FailFast, &rdev->flags))
1931 sb->devflags |= FailFast1;
1933 sb->devflags &= ~FailFast1;
1935 if (test_bit(WriteMostly, &rdev->flags))
1936 sb->devflags |= WriteMostly1;
1938 sb->devflags &= ~WriteMostly1;
1939 sb->data_offset = cpu_to_le64(rdev->data_offset);
1940 sb->data_size = cpu_to_le64(rdev->sectors);
1942 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1943 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1944 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1947 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1948 !test_bit(In_sync, &rdev->flags)) {
1950 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1951 sb->recovery_offset =
1952 cpu_to_le64(rdev->recovery_offset);
1953 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1955 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1957 /* Note: recovery_offset and journal_tail share space */
1958 if (test_bit(Journal, &rdev->flags))
1959 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1960 if (test_bit(Replacement, &rdev->flags))
1962 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1964 if (mddev->reshape_position != MaxSector) {
1965 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1966 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1967 sb->new_layout = cpu_to_le32(mddev->new_layout);
1968 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1969 sb->new_level = cpu_to_le32(mddev->new_level);
1970 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1971 if (mddev->delta_disks == 0 &&
1972 mddev->reshape_backwards)
1974 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1975 if (rdev->new_data_offset != rdev->data_offset) {
1977 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1978 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1979 - rdev->data_offset));
1983 if (mddev_is_clustered(mddev))
1984 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1986 if (rdev->badblocks.count == 0)
1987 /* Nothing to do for bad blocks*/ ;
1988 else if (sb->bblog_offset == 0)
1989 /* Cannot record bad blocks on this device */
1990 md_error(mddev, rdev);
1992 struct badblocks *bb = &rdev->badblocks;
1993 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
1995 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
2000 seq = read_seqbegin(&bb->lock);
2002 memset(bbp, 0xff, PAGE_SIZE);
2004 for (i = 0 ; i < bb->count ; i++) {
2005 u64 internal_bb = p[i];
2006 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2007 | BB_LEN(internal_bb));
2008 bbp[i] = cpu_to_le64(store_bb);
2011 if (read_seqretry(&bb->lock, seq))
2014 bb->sector = (rdev->sb_start +
2015 (int)le32_to_cpu(sb->bblog_offset));
2016 bb->size = le16_to_cpu(sb->bblog_size);
2021 rdev_for_each(rdev2, mddev)
2022 if (rdev2->desc_nr+1 > max_dev)
2023 max_dev = rdev2->desc_nr+1;
2025 if (max_dev > le32_to_cpu(sb->max_dev)) {
2027 sb->max_dev = cpu_to_le32(max_dev);
2028 rdev->sb_size = max_dev * 2 + 256;
2029 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2030 if (rdev->sb_size & bmask)
2031 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2033 max_dev = le32_to_cpu(sb->max_dev);
2035 for (i=0; i<max_dev;i++)
2036 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2038 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2039 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2041 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2042 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2044 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2046 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2047 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2048 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2051 rdev_for_each(rdev2, mddev) {
2053 if (test_bit(Faulty, &rdev2->flags))
2054 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2055 else if (test_bit(In_sync, &rdev2->flags))
2056 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2057 else if (test_bit(Journal, &rdev2->flags))
2058 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2059 else if (rdev2->raid_disk >= 0)
2060 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2062 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2065 sb->sb_csum = calc_sb_1_csum(sb);
2068 static unsigned long long
2069 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2071 struct mdp_superblock_1 *sb;
2072 sector_t max_sectors;
2073 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2074 return 0; /* component must fit device */
2075 if (rdev->data_offset != rdev->new_data_offset)
2076 return 0; /* too confusing */
2077 if (rdev->sb_start < rdev->data_offset) {
2078 /* minor versions 1 and 2; superblock before data */
2079 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2080 max_sectors -= rdev->data_offset;
2081 if (!num_sectors || num_sectors > max_sectors)
2082 num_sectors = max_sectors;
2083 } else if (rdev->mddev->bitmap_info.offset) {
2084 /* minor version 0 with bitmap we can't move */
2087 /* minor version 0; superblock after data */
2089 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
2090 sb_start &= ~(sector_t)(4*2 - 1);
2091 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
2092 if (!num_sectors || num_sectors > max_sectors)
2093 num_sectors = max_sectors;
2094 rdev->sb_start = sb_start;
2096 sb = page_address(rdev->sb_page);
2097 sb->data_size = cpu_to_le64(num_sectors);
2098 sb->super_offset = cpu_to_le64(rdev->sb_start);
2099 sb->sb_csum = calc_sb_1_csum(sb);
2101 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2103 } while (md_super_wait(rdev->mddev) < 0);
2109 super_1_allow_new_offset(struct md_rdev *rdev,
2110 unsigned long long new_offset)
2112 /* All necessary checks on new >= old have been done */
2113 struct bitmap *bitmap;
2114 if (new_offset >= rdev->data_offset)
2117 /* with 1.0 metadata, there is no metadata to tread on
2118 * so we can always move back */
2119 if (rdev->mddev->minor_version == 0)
2122 /* otherwise we must be sure not to step on
2123 * any metadata, so stay:
2124 * 36K beyond start of superblock
2125 * beyond end of badblocks
2126 * beyond write-intent bitmap
2128 if (rdev->sb_start + (32+4)*2 > new_offset)
2130 bitmap = rdev->mddev->bitmap;
2131 if (bitmap && !rdev->mddev->bitmap_info.file &&
2132 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2133 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2135 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2141 static struct super_type super_types[] = {
2144 .owner = THIS_MODULE,
2145 .load_super = super_90_load,
2146 .validate_super = super_90_validate,
2147 .sync_super = super_90_sync,
2148 .rdev_size_change = super_90_rdev_size_change,
2149 .allow_new_offset = super_90_allow_new_offset,
2153 .owner = THIS_MODULE,
2154 .load_super = super_1_load,
2155 .validate_super = super_1_validate,
2156 .sync_super = super_1_sync,
2157 .rdev_size_change = super_1_rdev_size_change,
2158 .allow_new_offset = super_1_allow_new_offset,
2162 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2164 if (mddev->sync_super) {
2165 mddev->sync_super(mddev, rdev);
2169 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2171 super_types[mddev->major_version].sync_super(mddev, rdev);
2174 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2176 struct md_rdev *rdev, *rdev2;
2179 rdev_for_each_rcu(rdev, mddev1) {
2180 if (test_bit(Faulty, &rdev->flags) ||
2181 test_bit(Journal, &rdev->flags) ||
2182 rdev->raid_disk == -1)
2184 rdev_for_each_rcu(rdev2, mddev2) {
2185 if (test_bit(Faulty, &rdev2->flags) ||
2186 test_bit(Journal, &rdev2->flags) ||
2187 rdev2->raid_disk == -1)
2189 if (rdev->bdev->bd_contains ==
2190 rdev2->bdev->bd_contains) {
2200 static LIST_HEAD(pending_raid_disks);
2203 * Try to register data integrity profile for an mddev
2205 * This is called when an array is started and after a disk has been kicked
2206 * from the array. It only succeeds if all working and active component devices
2207 * are integrity capable with matching profiles.
2209 int md_integrity_register(struct mddev *mddev)
2211 struct md_rdev *rdev, *reference = NULL;
2213 if (list_empty(&mddev->disks))
2214 return 0; /* nothing to do */
2215 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2216 return 0; /* shouldn't register, or already is */
2217 rdev_for_each(rdev, mddev) {
2218 /* skip spares and non-functional disks */
2219 if (test_bit(Faulty, &rdev->flags))
2221 if (rdev->raid_disk < 0)
2224 /* Use the first rdev as the reference */
2228 /* does this rdev's profile match the reference profile? */
2229 if (blk_integrity_compare(reference->bdev->bd_disk,
2230 rdev->bdev->bd_disk) < 0)
2233 if (!reference || !bdev_get_integrity(reference->bdev))
2236 * All component devices are integrity capable and have matching
2237 * profiles, register the common profile for the md device.
2239 blk_integrity_register(mddev->gendisk,
2240 bdev_get_integrity(reference->bdev));
2242 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2243 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2244 pr_err("md: failed to create integrity pool for %s\n",
2250 EXPORT_SYMBOL(md_integrity_register);
2253 * Attempt to add an rdev, but only if it is consistent with the current
2256 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2258 struct blk_integrity *bi_mddev;
2259 char name[BDEVNAME_SIZE];
2261 if (!mddev->gendisk)
2264 bi_mddev = blk_get_integrity(mddev->gendisk);
2266 if (!bi_mddev) /* nothing to do */
2269 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2270 pr_err("%s: incompatible integrity profile for %s\n",
2271 mdname(mddev), bdevname(rdev->bdev, name));
2277 EXPORT_SYMBOL(md_integrity_add_rdev);
2279 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2281 char b[BDEVNAME_SIZE];
2285 /* prevent duplicates */
2286 if (find_rdev(mddev, rdev->bdev->bd_dev))
2289 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2293 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2294 if (!test_bit(Journal, &rdev->flags) &&
2296 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2298 /* Cannot change size, so fail
2299 * If mddev->level <= 0, then we don't care
2300 * about aligning sizes (e.g. linear)
2302 if (mddev->level > 0)
2305 mddev->dev_sectors = rdev->sectors;
2308 /* Verify rdev->desc_nr is unique.
2309 * If it is -1, assign a free number, else
2310 * check number is not in use
2313 if (rdev->desc_nr < 0) {
2316 choice = mddev->raid_disks;
2317 while (md_find_rdev_nr_rcu(mddev, choice))
2319 rdev->desc_nr = choice;
2321 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2327 if (!test_bit(Journal, &rdev->flags) &&
2328 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2329 pr_warn("md: %s: array is limited to %d devices\n",
2330 mdname(mddev), mddev->max_disks);
2333 bdevname(rdev->bdev,b);
2334 strreplace(b, '/', '!');
2336 rdev->mddev = mddev;
2337 pr_debug("md: bind<%s>\n", b);
2339 if (mddev->raid_disks)
2340 mddev_create_wb_pool(mddev, rdev, false);
2342 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2345 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2346 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2347 /* failure here is OK */;
2348 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2350 list_add_rcu(&rdev->same_set, &mddev->disks);
2351 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2353 /* May as well allow recovery to be retried once */
2354 mddev->recovery_disabled++;
2359 pr_warn("md: failed to register dev-%s for %s\n",
2364 static void md_delayed_delete(struct work_struct *ws)
2366 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2367 kobject_del(&rdev->kobj);
2368 kobject_put(&rdev->kobj);
2371 static void unbind_rdev_from_array(struct md_rdev *rdev)
2373 char b[BDEVNAME_SIZE];
2375 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2376 list_del_rcu(&rdev->same_set);
2377 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2378 mddev_destroy_wb_pool(rdev->mddev, rdev);
2380 sysfs_remove_link(&rdev->kobj, "block");
2381 sysfs_put(rdev->sysfs_state);
2382 rdev->sysfs_state = NULL;
2383 rdev->badblocks.count = 0;
2384 /* We need to delay this, otherwise we can deadlock when
2385 * writing to 'remove' to "dev/state". We also need
2386 * to delay it due to rcu usage.
2389 INIT_WORK(&rdev->del_work, md_delayed_delete);
2390 kobject_get(&rdev->kobj);
2391 queue_work(md_misc_wq, &rdev->del_work);
2395 * prevent the device from being mounted, repartitioned or
2396 * otherwise reused by a RAID array (or any other kernel
2397 * subsystem), by bd_claiming the device.
2399 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2402 struct block_device *bdev;
2403 char b[BDEVNAME_SIZE];
2405 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2406 shared ? (struct md_rdev *)lock_rdev : rdev);
2408 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2409 return PTR_ERR(bdev);
2415 static void unlock_rdev(struct md_rdev *rdev)
2417 struct block_device *bdev = rdev->bdev;
2419 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2422 void md_autodetect_dev(dev_t dev);
2424 static void export_rdev(struct md_rdev *rdev)
2426 char b[BDEVNAME_SIZE];
2428 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2429 md_rdev_clear(rdev);
2431 if (test_bit(AutoDetected, &rdev->flags))
2432 md_autodetect_dev(rdev->bdev->bd_dev);
2435 kobject_put(&rdev->kobj);
2438 void md_kick_rdev_from_array(struct md_rdev *rdev)
2440 unbind_rdev_from_array(rdev);
2443 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2445 static void export_array(struct mddev *mddev)
2447 struct md_rdev *rdev;
2449 while (!list_empty(&mddev->disks)) {
2450 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2452 md_kick_rdev_from_array(rdev);
2454 mddev->raid_disks = 0;
2455 mddev->major_version = 0;
2458 static bool set_in_sync(struct mddev *mddev)
2460 lockdep_assert_held(&mddev->lock);
2461 if (!mddev->in_sync) {
2462 mddev->sync_checkers++;
2463 spin_unlock(&mddev->lock);
2464 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2465 spin_lock(&mddev->lock);
2466 if (!mddev->in_sync &&
2467 percpu_ref_is_zero(&mddev->writes_pending)) {
2470 * Ensure ->in_sync is visible before we clear
2474 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2475 sysfs_notify_dirent_safe(mddev->sysfs_state);
2477 if (--mddev->sync_checkers == 0)
2478 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2480 if (mddev->safemode == 1)
2481 mddev->safemode = 0;
2482 return mddev->in_sync;
2485 static void sync_sbs(struct mddev *mddev, int nospares)
2487 /* Update each superblock (in-memory image), but
2488 * if we are allowed to, skip spares which already
2489 * have the right event counter, or have one earlier
2490 * (which would mean they aren't being marked as dirty
2491 * with the rest of the array)
2493 struct md_rdev *rdev;
2494 rdev_for_each(rdev, mddev) {
2495 if (rdev->sb_events == mddev->events ||
2497 rdev->raid_disk < 0 &&
2498 rdev->sb_events+1 == mddev->events)) {
2499 /* Don't update this superblock */
2500 rdev->sb_loaded = 2;
2502 sync_super(mddev, rdev);
2503 rdev->sb_loaded = 1;
2508 static bool does_sb_need_changing(struct mddev *mddev)
2510 struct md_rdev *rdev;
2511 struct mdp_superblock_1 *sb;
2514 /* Find a good rdev */
2515 rdev_for_each(rdev, mddev)
2516 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2519 /* No good device found. */
2523 sb = page_address(rdev->sb_page);
2524 /* Check if a device has become faulty or a spare become active */
2525 rdev_for_each(rdev, mddev) {
2526 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2527 /* Device activated? */
2528 if (role == 0xffff && rdev->raid_disk >=0 &&
2529 !test_bit(Faulty, &rdev->flags))
2531 /* Device turned faulty? */
2532 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2536 /* Check if any mddev parameters have changed */
2537 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2538 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2539 (mddev->layout != le32_to_cpu(sb->layout)) ||
2540 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2541 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2547 void md_update_sb(struct mddev *mddev, int force_change)
2549 struct md_rdev *rdev;
2552 int any_badblocks_changed = 0;
2557 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2562 if (mddev_is_clustered(mddev)) {
2563 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2565 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2567 ret = md_cluster_ops->metadata_update_start(mddev);
2568 /* Has someone else has updated the sb */
2569 if (!does_sb_need_changing(mddev)) {
2571 md_cluster_ops->metadata_update_cancel(mddev);
2572 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2573 BIT(MD_SB_CHANGE_DEVS) |
2574 BIT(MD_SB_CHANGE_CLEAN));
2580 * First make sure individual recovery_offsets are correct
2581 * curr_resync_completed can only be used during recovery.
2582 * During reshape/resync it might use array-addresses rather
2583 * that device addresses.
2585 rdev_for_each(rdev, mddev) {
2586 if (rdev->raid_disk >= 0 &&
2587 mddev->delta_disks >= 0 &&
2588 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2589 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2590 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2591 !test_bit(Journal, &rdev->flags) &&
2592 !test_bit(In_sync, &rdev->flags) &&
2593 mddev->curr_resync_completed > rdev->recovery_offset)
2594 rdev->recovery_offset = mddev->curr_resync_completed;
2597 if (!mddev->persistent) {
2598 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2599 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2600 if (!mddev->external) {
2601 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2602 rdev_for_each(rdev, mddev) {
2603 if (rdev->badblocks.changed) {
2604 rdev->badblocks.changed = 0;
2605 ack_all_badblocks(&rdev->badblocks);
2606 md_error(mddev, rdev);
2608 clear_bit(Blocked, &rdev->flags);
2609 clear_bit(BlockedBadBlocks, &rdev->flags);
2610 wake_up(&rdev->blocked_wait);
2613 wake_up(&mddev->sb_wait);
2617 spin_lock(&mddev->lock);
2619 mddev->utime = ktime_get_real_seconds();
2621 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2623 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2624 /* just a clean<-> dirty transition, possibly leave spares alone,
2625 * though if events isn't the right even/odd, we will have to do
2631 if (mddev->degraded)
2632 /* If the array is degraded, then skipping spares is both
2633 * dangerous and fairly pointless.
2634 * Dangerous because a device that was removed from the array
2635 * might have a event_count that still looks up-to-date,
2636 * so it can be re-added without a resync.
2637 * Pointless because if there are any spares to skip,
2638 * then a recovery will happen and soon that array won't
2639 * be degraded any more and the spare can go back to sleep then.
2643 sync_req = mddev->in_sync;
2645 /* If this is just a dirty<->clean transition, and the array is clean
2646 * and 'events' is odd, we can roll back to the previous clean state */
2648 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2649 && mddev->can_decrease_events
2650 && mddev->events != 1) {
2652 mddev->can_decrease_events = 0;
2654 /* otherwise we have to go forward and ... */
2656 mddev->can_decrease_events = nospares;
2660 * This 64-bit counter should never wrap.
2661 * Either we are in around ~1 trillion A.C., assuming
2662 * 1 reboot per second, or we have a bug...
2664 WARN_ON(mddev->events == 0);
2666 rdev_for_each(rdev, mddev) {
2667 if (rdev->badblocks.changed)
2668 any_badblocks_changed++;
2669 if (test_bit(Faulty, &rdev->flags))
2670 set_bit(FaultRecorded, &rdev->flags);
2673 sync_sbs(mddev, nospares);
2674 spin_unlock(&mddev->lock);
2676 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2677 mdname(mddev), mddev->in_sync);
2680 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2682 md_bitmap_update_sb(mddev->bitmap);
2683 rdev_for_each(rdev, mddev) {
2684 char b[BDEVNAME_SIZE];
2686 if (rdev->sb_loaded != 1)
2687 continue; /* no noise on spare devices */
2689 if (!test_bit(Faulty, &rdev->flags)) {
2690 md_super_write(mddev,rdev,
2691 rdev->sb_start, rdev->sb_size,
2693 pr_debug("md: (write) %s's sb offset: %llu\n",
2694 bdevname(rdev->bdev, b),
2695 (unsigned long long)rdev->sb_start);
2696 rdev->sb_events = mddev->events;
2697 if (rdev->badblocks.size) {
2698 md_super_write(mddev, rdev,
2699 rdev->badblocks.sector,
2700 rdev->badblocks.size << 9,
2702 rdev->badblocks.size = 0;
2706 pr_debug("md: %s (skipping faulty)\n",
2707 bdevname(rdev->bdev, b));
2709 if (mddev->level == LEVEL_MULTIPATH)
2710 /* only need to write one superblock... */
2713 if (md_super_wait(mddev) < 0)
2715 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2717 if (mddev_is_clustered(mddev) && ret == 0)
2718 md_cluster_ops->metadata_update_finish(mddev);
2720 if (mddev->in_sync != sync_req ||
2721 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2722 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2723 /* have to write it out again */
2725 wake_up(&mddev->sb_wait);
2726 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2727 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2729 rdev_for_each(rdev, mddev) {
2730 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2731 clear_bit(Blocked, &rdev->flags);
2733 if (any_badblocks_changed)
2734 ack_all_badblocks(&rdev->badblocks);
2735 clear_bit(BlockedBadBlocks, &rdev->flags);
2736 wake_up(&rdev->blocked_wait);
2739 EXPORT_SYMBOL(md_update_sb);
2741 static int add_bound_rdev(struct md_rdev *rdev)
2743 struct mddev *mddev = rdev->mddev;
2745 bool add_journal = test_bit(Journal, &rdev->flags);
2747 if (!mddev->pers->hot_remove_disk || add_journal) {
2748 /* If there is hot_add_disk but no hot_remove_disk
2749 * then added disks for geometry changes,
2750 * and should be added immediately.
2752 super_types[mddev->major_version].
2753 validate_super(mddev, rdev);
2755 mddev_suspend(mddev);
2756 err = mddev->pers->hot_add_disk(mddev, rdev);
2758 mddev_resume(mddev);
2760 md_kick_rdev_from_array(rdev);
2764 sysfs_notify_dirent_safe(rdev->sysfs_state);
2766 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2767 if (mddev->degraded)
2768 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2769 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2770 md_new_event(mddev);
2771 md_wakeup_thread(mddev->thread);
2775 /* words written to sysfs files may, or may not, be \n terminated.
2776 * We want to accept with case. For this we use cmd_match.
2778 static int cmd_match(const char *cmd, const char *str)
2780 /* See if cmd, written into a sysfs file, matches
2781 * str. They must either be the same, or cmd can
2782 * have a trailing newline
2784 while (*cmd && *str && *cmd == *str) {
2795 struct rdev_sysfs_entry {
2796 struct attribute attr;
2797 ssize_t (*show)(struct md_rdev *, char *);
2798 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2802 state_show(struct md_rdev *rdev, char *page)
2806 unsigned long flags = READ_ONCE(rdev->flags);
2808 if (test_bit(Faulty, &flags) ||
2809 (!test_bit(ExternalBbl, &flags) &&
2810 rdev->badblocks.unacked_exist))
2811 len += sprintf(page+len, "faulty%s", sep);
2812 if (test_bit(In_sync, &flags))
2813 len += sprintf(page+len, "in_sync%s", sep);
2814 if (test_bit(Journal, &flags))
2815 len += sprintf(page+len, "journal%s", sep);
2816 if (test_bit(WriteMostly, &flags))
2817 len += sprintf(page+len, "write_mostly%s", sep);
2818 if (test_bit(Blocked, &flags) ||
2819 (rdev->badblocks.unacked_exist
2820 && !test_bit(Faulty, &flags)))
2821 len += sprintf(page+len, "blocked%s", sep);
2822 if (!test_bit(Faulty, &flags) &&
2823 !test_bit(Journal, &flags) &&
2824 !test_bit(In_sync, &flags))
2825 len += sprintf(page+len, "spare%s", sep);
2826 if (test_bit(WriteErrorSeen, &flags))
2827 len += sprintf(page+len, "write_error%s", sep);
2828 if (test_bit(WantReplacement, &flags))
2829 len += sprintf(page+len, "want_replacement%s", sep);
2830 if (test_bit(Replacement, &flags))
2831 len += sprintf(page+len, "replacement%s", sep);
2832 if (test_bit(ExternalBbl, &flags))
2833 len += sprintf(page+len, "external_bbl%s", sep);
2834 if (test_bit(FailFast, &flags))
2835 len += sprintf(page+len, "failfast%s", sep);
2840 return len+sprintf(page+len, "\n");
2844 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2847 * faulty - simulates an error
2848 * remove - disconnects the device
2849 * writemostly - sets write_mostly
2850 * -writemostly - clears write_mostly
2851 * blocked - sets the Blocked flags
2852 * -blocked - clears the Blocked and possibly simulates an error
2853 * insync - sets Insync providing device isn't active
2854 * -insync - clear Insync for a device with a slot assigned,
2855 * so that it gets rebuilt based on bitmap
2856 * write_error - sets WriteErrorSeen
2857 * -write_error - clears WriteErrorSeen
2858 * {,-}failfast - set/clear FailFast
2861 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2862 md_error(rdev->mddev, rdev);
2863 if (test_bit(Faulty, &rdev->flags))
2867 } else if (cmd_match(buf, "remove")) {
2868 if (rdev->mddev->pers) {
2869 clear_bit(Blocked, &rdev->flags);
2870 remove_and_add_spares(rdev->mddev, rdev);
2872 if (rdev->raid_disk >= 0)
2875 struct mddev *mddev = rdev->mddev;
2877 if (mddev_is_clustered(mddev))
2878 err = md_cluster_ops->remove_disk(mddev, rdev);
2881 md_kick_rdev_from_array(rdev);
2883 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2884 md_wakeup_thread(mddev->thread);
2886 md_new_event(mddev);
2889 } else if (cmd_match(buf, "writemostly")) {
2890 set_bit(WriteMostly, &rdev->flags);
2891 mddev_create_wb_pool(rdev->mddev, rdev, false);
2893 } else if (cmd_match(buf, "-writemostly")) {
2894 mddev_destroy_wb_pool(rdev->mddev, rdev);
2895 clear_bit(WriteMostly, &rdev->flags);
2897 } else if (cmd_match(buf, "blocked")) {
2898 set_bit(Blocked, &rdev->flags);
2900 } else if (cmd_match(buf, "-blocked")) {
2901 if (!test_bit(Faulty, &rdev->flags) &&
2902 !test_bit(ExternalBbl, &rdev->flags) &&
2903 rdev->badblocks.unacked_exist) {
2904 /* metadata handler doesn't understand badblocks,
2905 * so we need to fail the device
2907 md_error(rdev->mddev, rdev);
2909 clear_bit(Blocked, &rdev->flags);
2910 clear_bit(BlockedBadBlocks, &rdev->flags);
2911 wake_up(&rdev->blocked_wait);
2912 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2913 md_wakeup_thread(rdev->mddev->thread);
2916 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2917 set_bit(In_sync, &rdev->flags);
2919 } else if (cmd_match(buf, "failfast")) {
2920 set_bit(FailFast, &rdev->flags);
2922 } else if (cmd_match(buf, "-failfast")) {
2923 clear_bit(FailFast, &rdev->flags);
2925 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2926 !test_bit(Journal, &rdev->flags)) {
2927 if (rdev->mddev->pers == NULL) {
2928 clear_bit(In_sync, &rdev->flags);
2929 rdev->saved_raid_disk = rdev->raid_disk;
2930 rdev->raid_disk = -1;
2933 } else if (cmd_match(buf, "write_error")) {
2934 set_bit(WriteErrorSeen, &rdev->flags);
2936 } else if (cmd_match(buf, "-write_error")) {
2937 clear_bit(WriteErrorSeen, &rdev->flags);
2939 } else if (cmd_match(buf, "want_replacement")) {
2940 /* Any non-spare device that is not a replacement can
2941 * become want_replacement at any time, but we then need to
2942 * check if recovery is needed.
2944 if (rdev->raid_disk >= 0 &&
2945 !test_bit(Journal, &rdev->flags) &&
2946 !test_bit(Replacement, &rdev->flags))
2947 set_bit(WantReplacement, &rdev->flags);
2948 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2949 md_wakeup_thread(rdev->mddev->thread);
2951 } else if (cmd_match(buf, "-want_replacement")) {
2952 /* Clearing 'want_replacement' is always allowed.
2953 * Once replacements starts it is too late though.
2956 clear_bit(WantReplacement, &rdev->flags);
2957 } else if (cmd_match(buf, "replacement")) {
2958 /* Can only set a device as a replacement when array has not
2959 * yet been started. Once running, replacement is automatic
2960 * from spares, or by assigning 'slot'.
2962 if (rdev->mddev->pers)
2965 set_bit(Replacement, &rdev->flags);
2968 } else if (cmd_match(buf, "-replacement")) {
2969 /* Similarly, can only clear Replacement before start */
2970 if (rdev->mddev->pers)
2973 clear_bit(Replacement, &rdev->flags);
2976 } else if (cmd_match(buf, "re-add")) {
2977 if (!rdev->mddev->pers)
2979 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2980 rdev->saved_raid_disk >= 0) {
2981 /* clear_bit is performed _after_ all the devices
2982 * have their local Faulty bit cleared. If any writes
2983 * happen in the meantime in the local node, they
2984 * will land in the local bitmap, which will be synced
2985 * by this node eventually
2987 if (!mddev_is_clustered(rdev->mddev) ||
2988 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2989 clear_bit(Faulty, &rdev->flags);
2990 err = add_bound_rdev(rdev);
2994 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2995 set_bit(ExternalBbl, &rdev->flags);
2996 rdev->badblocks.shift = 0;
2998 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2999 clear_bit(ExternalBbl, &rdev->flags);
3003 sysfs_notify_dirent_safe(rdev->sysfs_state);
3004 return err ? err : len;
3006 static struct rdev_sysfs_entry rdev_state =
3007 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3010 errors_show(struct md_rdev *rdev, char *page)
3012 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3016 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3021 rv = kstrtouint(buf, 10, &n);
3024 atomic_set(&rdev->corrected_errors, n);
3027 static struct rdev_sysfs_entry rdev_errors =
3028 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3031 slot_show(struct md_rdev *rdev, char *page)
3033 if (test_bit(Journal, &rdev->flags))
3034 return sprintf(page, "journal\n");
3035 else if (rdev->raid_disk < 0)
3036 return sprintf(page, "none\n");
3038 return sprintf(page, "%d\n", rdev->raid_disk);
3042 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3047 if (test_bit(Journal, &rdev->flags))
3049 if (strncmp(buf, "none", 4)==0)
3052 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3056 if (rdev->mddev->pers && slot == -1) {
3057 /* Setting 'slot' on an active array requires also
3058 * updating the 'rd%d' link, and communicating
3059 * with the personality with ->hot_*_disk.
3060 * For now we only support removing
3061 * failed/spare devices. This normally happens automatically,
3062 * but not when the metadata is externally managed.
3064 if (rdev->raid_disk == -1)
3066 /* personality does all needed checks */
3067 if (rdev->mddev->pers->hot_remove_disk == NULL)
3069 clear_bit(Blocked, &rdev->flags);
3070 remove_and_add_spares(rdev->mddev, rdev);
3071 if (rdev->raid_disk >= 0)
3073 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3074 md_wakeup_thread(rdev->mddev->thread);
3075 } else if (rdev->mddev->pers) {
3076 /* Activating a spare .. or possibly reactivating
3077 * if we ever get bitmaps working here.
3081 if (rdev->raid_disk != -1)
3084 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3087 if (rdev->mddev->pers->hot_add_disk == NULL)
3090 if (slot >= rdev->mddev->raid_disks &&
3091 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3094 rdev->raid_disk = slot;
3095 if (test_bit(In_sync, &rdev->flags))
3096 rdev->saved_raid_disk = slot;
3098 rdev->saved_raid_disk = -1;
3099 clear_bit(In_sync, &rdev->flags);
3100 clear_bit(Bitmap_sync, &rdev->flags);
3101 err = rdev->mddev->pers->
3102 hot_add_disk(rdev->mddev, rdev);
3104 rdev->raid_disk = -1;
3107 sysfs_notify_dirent_safe(rdev->sysfs_state);
3108 if (sysfs_link_rdev(rdev->mddev, rdev))
3109 /* failure here is OK */;
3110 /* don't wakeup anyone, leave that to userspace. */
3112 if (slot >= rdev->mddev->raid_disks &&
3113 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3115 rdev->raid_disk = slot;
3116 /* assume it is working */
3117 clear_bit(Faulty, &rdev->flags);
3118 clear_bit(WriteMostly, &rdev->flags);
3119 set_bit(In_sync, &rdev->flags);
3120 sysfs_notify_dirent_safe(rdev->sysfs_state);
3125 static struct rdev_sysfs_entry rdev_slot =
3126 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3129 offset_show(struct md_rdev *rdev, char *page)
3131 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3135 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3137 unsigned long long offset;
3138 if (kstrtoull(buf, 10, &offset) < 0)
3140 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3142 if (rdev->sectors && rdev->mddev->external)
3143 /* Must set offset before size, so overlap checks
3146 rdev->data_offset = offset;
3147 rdev->new_data_offset = offset;
3151 static struct rdev_sysfs_entry rdev_offset =
3152 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3154 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3156 return sprintf(page, "%llu\n",
3157 (unsigned long long)rdev->new_data_offset);
3160 static ssize_t new_offset_store(struct md_rdev *rdev,
3161 const char *buf, size_t len)
3163 unsigned long long new_offset;
3164 struct mddev *mddev = rdev->mddev;
3166 if (kstrtoull(buf, 10, &new_offset) < 0)
3169 if (mddev->sync_thread ||
3170 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3172 if (new_offset == rdev->data_offset)
3173 /* reset is always permitted */
3175 else if (new_offset > rdev->data_offset) {
3176 /* must not push array size beyond rdev_sectors */
3177 if (new_offset - rdev->data_offset
3178 + mddev->dev_sectors > rdev->sectors)
3181 /* Metadata worries about other space details. */
3183 /* decreasing the offset is inconsistent with a backwards
3186 if (new_offset < rdev->data_offset &&
3187 mddev->reshape_backwards)
3189 /* Increasing offset is inconsistent with forwards
3190 * reshape. reshape_direction should be set to
3191 * 'backwards' first.
3193 if (new_offset > rdev->data_offset &&
3194 !mddev->reshape_backwards)
3197 if (mddev->pers && mddev->persistent &&
3198 !super_types[mddev->major_version]
3199 .allow_new_offset(rdev, new_offset))
3201 rdev->new_data_offset = new_offset;
3202 if (new_offset > rdev->data_offset)
3203 mddev->reshape_backwards = 1;
3204 else if (new_offset < rdev->data_offset)
3205 mddev->reshape_backwards = 0;
3209 static struct rdev_sysfs_entry rdev_new_offset =
3210 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3213 rdev_size_show(struct md_rdev *rdev, char *page)
3215 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3218 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3220 /* check if two start/length pairs overlap */
3228 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3230 unsigned long long blocks;
3233 if (kstrtoull(buf, 10, &blocks) < 0)
3236 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3237 return -EINVAL; /* sector conversion overflow */
3240 if (new != blocks * 2)
3241 return -EINVAL; /* unsigned long long to sector_t overflow */
3248 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3250 struct mddev *my_mddev = rdev->mddev;
3251 sector_t oldsectors = rdev->sectors;
3254 if (test_bit(Journal, &rdev->flags))
3256 if (strict_blocks_to_sectors(buf, §ors) < 0)
3258 if (rdev->data_offset != rdev->new_data_offset)
3259 return -EINVAL; /* too confusing */
3260 if (my_mddev->pers && rdev->raid_disk >= 0) {
3261 if (my_mddev->persistent) {
3262 sectors = super_types[my_mddev->major_version].
3263 rdev_size_change(rdev, sectors);
3266 } else if (!sectors)
3267 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3269 if (!my_mddev->pers->resize)
3270 /* Cannot change size for RAID0 or Linear etc */
3273 if (sectors < my_mddev->dev_sectors)
3274 return -EINVAL; /* component must fit device */
3276 rdev->sectors = sectors;
3277 if (sectors > oldsectors && my_mddev->external) {
3278 /* Need to check that all other rdevs with the same
3279 * ->bdev do not overlap. 'rcu' is sufficient to walk
3280 * the rdev lists safely.
3281 * This check does not provide a hard guarantee, it
3282 * just helps avoid dangerous mistakes.
3284 struct mddev *mddev;
3286 struct list_head *tmp;
3289 for_each_mddev(mddev, tmp) {
3290 struct md_rdev *rdev2;
3292 rdev_for_each(rdev2, mddev)
3293 if (rdev->bdev == rdev2->bdev &&
3295 overlaps(rdev->data_offset, rdev->sectors,
3308 /* Someone else could have slipped in a size
3309 * change here, but doing so is just silly.
3310 * We put oldsectors back because we *know* it is
3311 * safe, and trust userspace not to race with
3314 rdev->sectors = oldsectors;
3321 static struct rdev_sysfs_entry rdev_size =
3322 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3324 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3326 unsigned long long recovery_start = rdev->recovery_offset;
3328 if (test_bit(In_sync, &rdev->flags) ||
3329 recovery_start == MaxSector)
3330 return sprintf(page, "none\n");
3332 return sprintf(page, "%llu\n", recovery_start);
3335 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3337 unsigned long long recovery_start;
3339 if (cmd_match(buf, "none"))
3340 recovery_start = MaxSector;
3341 else if (kstrtoull(buf, 10, &recovery_start))
3344 if (rdev->mddev->pers &&
3345 rdev->raid_disk >= 0)
3348 rdev->recovery_offset = recovery_start;
3349 if (recovery_start == MaxSector)
3350 set_bit(In_sync, &rdev->flags);
3352 clear_bit(In_sync, &rdev->flags);
3356 static struct rdev_sysfs_entry rdev_recovery_start =
3357 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3359 /* sysfs access to bad-blocks list.
3360 * We present two files.
3361 * 'bad-blocks' lists sector numbers and lengths of ranges that
3362 * are recorded as bad. The list is truncated to fit within
3363 * the one-page limit of sysfs.
3364 * Writing "sector length" to this file adds an acknowledged
3366 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3367 * been acknowledged. Writing to this file adds bad blocks
3368 * without acknowledging them. This is largely for testing.
3370 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3372 return badblocks_show(&rdev->badblocks, page, 0);
3374 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3376 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3377 /* Maybe that ack was all we needed */
3378 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3379 wake_up(&rdev->blocked_wait);
3382 static struct rdev_sysfs_entry rdev_bad_blocks =
3383 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3385 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3387 return badblocks_show(&rdev->badblocks, page, 1);
3389 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3391 return badblocks_store(&rdev->badblocks, page, len, 1);
3393 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3394 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3397 ppl_sector_show(struct md_rdev *rdev, char *page)
3399 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3403 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3405 unsigned long long sector;
3407 if (kstrtoull(buf, 10, §or) < 0)
3409 if (sector != (sector_t)sector)
3412 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3413 rdev->raid_disk >= 0)
3416 if (rdev->mddev->persistent) {
3417 if (rdev->mddev->major_version == 0)
3419 if ((sector > rdev->sb_start &&
3420 sector - rdev->sb_start > S16_MAX) ||
3421 (sector < rdev->sb_start &&
3422 rdev->sb_start - sector > -S16_MIN))
3424 rdev->ppl.offset = sector - rdev->sb_start;
3425 } else if (!rdev->mddev->external) {
3428 rdev->ppl.sector = sector;
3432 static struct rdev_sysfs_entry rdev_ppl_sector =
3433 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3436 ppl_size_show(struct md_rdev *rdev, char *page)
3438 return sprintf(page, "%u\n", rdev->ppl.size);
3442 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3446 if (kstrtouint(buf, 10, &size) < 0)
3449 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3450 rdev->raid_disk >= 0)
3453 if (rdev->mddev->persistent) {
3454 if (rdev->mddev->major_version == 0)
3458 } else if (!rdev->mddev->external) {
3461 rdev->ppl.size = size;
3465 static struct rdev_sysfs_entry rdev_ppl_size =
3466 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3468 static struct attribute *rdev_default_attrs[] = {
3473 &rdev_new_offset.attr,
3475 &rdev_recovery_start.attr,
3476 &rdev_bad_blocks.attr,
3477 &rdev_unack_bad_blocks.attr,
3478 &rdev_ppl_sector.attr,
3479 &rdev_ppl_size.attr,
3483 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3485 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3486 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3492 return entry->show(rdev, page);
3496 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3497 const char *page, size_t length)
3499 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3500 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3502 struct mddev *mddev = rdev->mddev;
3506 if (!capable(CAP_SYS_ADMIN))
3508 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3510 if (rdev->mddev == NULL)
3513 rv = entry->store(rdev, page, length);
3514 mddev_unlock(mddev);
3519 static void rdev_free(struct kobject *ko)
3521 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3524 static const struct sysfs_ops rdev_sysfs_ops = {
3525 .show = rdev_attr_show,
3526 .store = rdev_attr_store,
3528 static struct kobj_type rdev_ktype = {
3529 .release = rdev_free,
3530 .sysfs_ops = &rdev_sysfs_ops,
3531 .default_attrs = rdev_default_attrs,
3534 int md_rdev_init(struct md_rdev *rdev)
3537 rdev->saved_raid_disk = -1;
3538 rdev->raid_disk = -1;
3540 rdev->data_offset = 0;
3541 rdev->new_data_offset = 0;
3542 rdev->sb_events = 0;
3543 rdev->last_read_error = 0;
3544 rdev->sb_loaded = 0;
3545 rdev->bb_page = NULL;
3546 atomic_set(&rdev->nr_pending, 0);
3547 atomic_set(&rdev->read_errors, 0);
3548 atomic_set(&rdev->corrected_errors, 0);
3550 INIT_LIST_HEAD(&rdev->same_set);
3551 init_waitqueue_head(&rdev->blocked_wait);
3553 /* Add space to store bad block list.
3554 * This reserves the space even on arrays where it cannot
3555 * be used - I wonder if that matters
3557 return badblocks_init(&rdev->badblocks, 0);
3559 EXPORT_SYMBOL_GPL(md_rdev_init);
3561 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3563 * mark the device faulty if:
3565 * - the device is nonexistent (zero size)
3566 * - the device has no valid superblock
3568 * a faulty rdev _never_ has rdev->sb set.
3570 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3572 char b[BDEVNAME_SIZE];
3574 struct md_rdev *rdev;
3577 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3579 return ERR_PTR(-ENOMEM);
3581 err = md_rdev_init(rdev);
3584 err = alloc_disk_sb(rdev);
3588 err = lock_rdev(rdev, newdev, super_format == -2);
3592 kobject_init(&rdev->kobj, &rdev_ktype);
3594 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3596 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3597 bdevname(rdev->bdev,b));
3602 if (super_format >= 0) {
3603 err = super_types[super_format].
3604 load_super(rdev, NULL, super_minor);
3605 if (err == -EINVAL) {
3606 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3607 bdevname(rdev->bdev,b),
3608 super_format, super_minor);
3612 pr_warn("md: could not read %s's sb, not importing!\n",
3613 bdevname(rdev->bdev,b));
3623 md_rdev_clear(rdev);
3625 return ERR_PTR(err);
3629 * Check a full RAID array for plausibility
3632 static int analyze_sbs(struct mddev *mddev)
3635 struct md_rdev *rdev, *freshest, *tmp;
3636 char b[BDEVNAME_SIZE];
3639 rdev_for_each_safe(rdev, tmp, mddev)
3640 switch (super_types[mddev->major_version].
3641 load_super(rdev, freshest, mddev->minor_version)) {
3648 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3649 bdevname(rdev->bdev,b));
3650 md_kick_rdev_from_array(rdev);
3653 /* Cannot find a valid fresh disk */
3655 pr_warn("md: cannot find a valid disk\n");
3659 super_types[mddev->major_version].
3660 validate_super(mddev, freshest);
3663 rdev_for_each_safe(rdev, tmp, mddev) {
3664 if (mddev->max_disks &&
3665 (rdev->desc_nr >= mddev->max_disks ||
3666 i > mddev->max_disks)) {
3667 pr_warn("md: %s: %s: only %d devices permitted\n",
3668 mdname(mddev), bdevname(rdev->bdev, b),
3670 md_kick_rdev_from_array(rdev);
3673 if (rdev != freshest) {
3674 if (super_types[mddev->major_version].
3675 validate_super(mddev, rdev)) {
3676 pr_warn("md: kicking non-fresh %s from array!\n",
3677 bdevname(rdev->bdev,b));
3678 md_kick_rdev_from_array(rdev);
3682 if (mddev->level == LEVEL_MULTIPATH) {
3683 rdev->desc_nr = i++;
3684 rdev->raid_disk = rdev->desc_nr;
3685 set_bit(In_sync, &rdev->flags);
3686 } else if (rdev->raid_disk >=
3687 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3688 !test_bit(Journal, &rdev->flags)) {
3689 rdev->raid_disk = -1;
3690 clear_bit(In_sync, &rdev->flags);
3697 /* Read a fixed-point number.
3698 * Numbers in sysfs attributes should be in "standard" units where
3699 * possible, so time should be in seconds.
3700 * However we internally use a a much smaller unit such as
3701 * milliseconds or jiffies.
3702 * This function takes a decimal number with a possible fractional
3703 * component, and produces an integer which is the result of
3704 * multiplying that number by 10^'scale'.
3705 * all without any floating-point arithmetic.
3707 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3709 unsigned long result = 0;
3711 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3714 else if (decimals < scale) {
3717 result = result * 10 + value;
3729 *res = result * int_pow(10, scale - decimals);
3734 safe_delay_show(struct mddev *mddev, char *page)
3736 int msec = (mddev->safemode_delay*1000)/HZ;
3737 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3740 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3744 if (mddev_is_clustered(mddev)) {
3745 pr_warn("md: Safemode is disabled for clustered mode\n");
3749 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3752 mddev->safemode_delay = 0;
3754 unsigned long old_delay = mddev->safemode_delay;
3755 unsigned long new_delay = (msec*HZ)/1000;
3759 mddev->safemode_delay = new_delay;
3760 if (new_delay < old_delay || old_delay == 0)
3761 mod_timer(&mddev->safemode_timer, jiffies+1);
3765 static struct md_sysfs_entry md_safe_delay =
3766 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3769 level_show(struct mddev *mddev, char *page)
3771 struct md_personality *p;
3773 spin_lock(&mddev->lock);
3776 ret = sprintf(page, "%s\n", p->name);
3777 else if (mddev->clevel[0])
3778 ret = sprintf(page, "%s\n", mddev->clevel);
3779 else if (mddev->level != LEVEL_NONE)
3780 ret = sprintf(page, "%d\n", mddev->level);
3783 spin_unlock(&mddev->lock);
3788 level_store(struct mddev *mddev, const char *buf, size_t len)
3793 struct md_personality *pers, *oldpers;
3795 void *priv, *oldpriv;
3796 struct md_rdev *rdev;
3798 if (slen == 0 || slen >= sizeof(clevel))
3801 rv = mddev_lock(mddev);
3805 if (mddev->pers == NULL) {
3806 strncpy(mddev->clevel, buf, slen);
3807 if (mddev->clevel[slen-1] == '\n')
3809 mddev->clevel[slen] = 0;
3810 mddev->level = LEVEL_NONE;
3818 /* request to change the personality. Need to ensure:
3819 * - array is not engaged in resync/recovery/reshape
3820 * - old personality can be suspended
3821 * - new personality will access other array.
3825 if (mddev->sync_thread ||
3826 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3827 mddev->reshape_position != MaxSector ||
3828 mddev->sysfs_active)
3832 if (!mddev->pers->quiesce) {
3833 pr_warn("md: %s: %s does not support online personality change\n",
3834 mdname(mddev), mddev->pers->name);
3838 /* Now find the new personality */
3839 strncpy(clevel, buf, slen);
3840 if (clevel[slen-1] == '\n')
3843 if (kstrtol(clevel, 10, &level))
3846 if (request_module("md-%s", clevel) != 0)
3847 request_module("md-level-%s", clevel);
3848 spin_lock(&pers_lock);
3849 pers = find_pers(level, clevel);
3850 if (!pers || !try_module_get(pers->owner)) {
3851 spin_unlock(&pers_lock);
3852 pr_warn("md: personality %s not loaded\n", clevel);
3856 spin_unlock(&pers_lock);
3858 if (pers == mddev->pers) {
3859 /* Nothing to do! */
3860 module_put(pers->owner);
3864 if (!pers->takeover) {
3865 module_put(pers->owner);
3866 pr_warn("md: %s: %s does not support personality takeover\n",
3867 mdname(mddev), clevel);
3872 rdev_for_each(rdev, mddev)
3873 rdev->new_raid_disk = rdev->raid_disk;
3875 /* ->takeover must set new_* and/or delta_disks
3876 * if it succeeds, and may set them when it fails.
3878 priv = pers->takeover(mddev);
3880 mddev->new_level = mddev->level;
3881 mddev->new_layout = mddev->layout;
3882 mddev->new_chunk_sectors = mddev->chunk_sectors;
3883 mddev->raid_disks -= mddev->delta_disks;
3884 mddev->delta_disks = 0;
3885 mddev->reshape_backwards = 0;
3886 module_put(pers->owner);
3887 pr_warn("md: %s: %s would not accept array\n",
3888 mdname(mddev), clevel);
3893 /* Looks like we have a winner */
3894 mddev_suspend(mddev);
3895 mddev_detach(mddev);
3897 spin_lock(&mddev->lock);
3898 oldpers = mddev->pers;
3899 oldpriv = mddev->private;
3901 mddev->private = priv;
3902 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3903 mddev->level = mddev->new_level;
3904 mddev->layout = mddev->new_layout;
3905 mddev->chunk_sectors = mddev->new_chunk_sectors;
3906 mddev->delta_disks = 0;
3907 mddev->reshape_backwards = 0;
3908 mddev->degraded = 0;
3909 spin_unlock(&mddev->lock);
3911 if (oldpers->sync_request == NULL &&
3913 /* We are converting from a no-redundancy array
3914 * to a redundancy array and metadata is managed
3915 * externally so we need to be sure that writes
3916 * won't block due to a need to transition
3918 * until external management is started.
3921 mddev->safemode_delay = 0;
3922 mddev->safemode = 0;
3925 oldpers->free(mddev, oldpriv);
3927 if (oldpers->sync_request == NULL &&
3928 pers->sync_request != NULL) {
3929 /* need to add the md_redundancy_group */
3930 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3931 pr_warn("md: cannot register extra attributes for %s\n",
3933 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3935 if (oldpers->sync_request != NULL &&
3936 pers->sync_request == NULL) {
3937 /* need to remove the md_redundancy_group */
3938 if (mddev->to_remove == NULL)
3939 mddev->to_remove = &md_redundancy_group;
3942 module_put(oldpers->owner);
3944 rdev_for_each(rdev, mddev) {
3945 if (rdev->raid_disk < 0)
3947 if (rdev->new_raid_disk >= mddev->raid_disks)
3948 rdev->new_raid_disk = -1;
3949 if (rdev->new_raid_disk == rdev->raid_disk)
3951 sysfs_unlink_rdev(mddev, rdev);
3953 rdev_for_each(rdev, mddev) {
3954 if (rdev->raid_disk < 0)
3956 if (rdev->new_raid_disk == rdev->raid_disk)
3958 rdev->raid_disk = rdev->new_raid_disk;
3959 if (rdev->raid_disk < 0)
3960 clear_bit(In_sync, &rdev->flags);
3962 if (sysfs_link_rdev(mddev, rdev))
3963 pr_warn("md: cannot register rd%d for %s after level change\n",
3964 rdev->raid_disk, mdname(mddev));
3968 if (pers->sync_request == NULL) {
3969 /* this is now an array without redundancy, so
3970 * it must always be in_sync
3973 del_timer_sync(&mddev->safemode_timer);
3975 blk_set_stacking_limits(&mddev->queue->limits);
3977 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3978 mddev_resume(mddev);
3980 md_update_sb(mddev, 1);
3981 sysfs_notify(&mddev->kobj, NULL, "level");
3982 md_new_event(mddev);
3985 mddev_unlock(mddev);
3989 static struct md_sysfs_entry md_level =
3990 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3993 layout_show(struct mddev *mddev, char *page)
3995 /* just a number, not meaningful for all levels */
3996 if (mddev->reshape_position != MaxSector &&
3997 mddev->layout != mddev->new_layout)
3998 return sprintf(page, "%d (%d)\n",
3999 mddev->new_layout, mddev->layout);
4000 return sprintf(page, "%d\n", mddev->layout);
4004 layout_store(struct mddev *mddev, const char *buf, size_t len)
4009 err = kstrtouint(buf, 10, &n);
4012 err = mddev_lock(mddev);
4017 if (mddev->pers->check_reshape == NULL)
4022 mddev->new_layout = n;
4023 err = mddev->pers->check_reshape(mddev);
4025 mddev->new_layout = mddev->layout;
4028 mddev->new_layout = n;
4029 if (mddev->reshape_position == MaxSector)
4032 mddev_unlock(mddev);
4035 static struct md_sysfs_entry md_layout =
4036 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4039 raid_disks_show(struct mddev *mddev, char *page)
4041 if (mddev->raid_disks == 0)
4043 if (mddev->reshape_position != MaxSector &&
4044 mddev->delta_disks != 0)
4045 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4046 mddev->raid_disks - mddev->delta_disks);
4047 return sprintf(page, "%d\n", mddev->raid_disks);
4050 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4053 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4058 err = kstrtouint(buf, 10, &n);
4062 err = mddev_lock(mddev);
4066 err = update_raid_disks(mddev, n);
4067 else if (mddev->reshape_position != MaxSector) {
4068 struct md_rdev *rdev;
4069 int olddisks = mddev->raid_disks - mddev->delta_disks;
4072 rdev_for_each(rdev, mddev) {
4074 rdev->data_offset < rdev->new_data_offset)
4077 rdev->data_offset > rdev->new_data_offset)
4081 mddev->delta_disks = n - olddisks;
4082 mddev->raid_disks = n;
4083 mddev->reshape_backwards = (mddev->delta_disks < 0);
4085 mddev->raid_disks = n;
4087 mddev_unlock(mddev);
4088 return err ? err : len;
4090 static struct md_sysfs_entry md_raid_disks =
4091 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4094 chunk_size_show(struct mddev *mddev, char *page)
4096 if (mddev->reshape_position != MaxSector &&
4097 mddev->chunk_sectors != mddev->new_chunk_sectors)
4098 return sprintf(page, "%d (%d)\n",
4099 mddev->new_chunk_sectors << 9,
4100 mddev->chunk_sectors << 9);
4101 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4105 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4110 err = kstrtoul(buf, 10, &n);
4114 err = mddev_lock(mddev);
4118 if (mddev->pers->check_reshape == NULL)
4123 mddev->new_chunk_sectors = n >> 9;
4124 err = mddev->pers->check_reshape(mddev);
4126 mddev->new_chunk_sectors = mddev->chunk_sectors;
4129 mddev->new_chunk_sectors = n >> 9;
4130 if (mddev->reshape_position == MaxSector)
4131 mddev->chunk_sectors = n >> 9;
4133 mddev_unlock(mddev);
4136 static struct md_sysfs_entry md_chunk_size =
4137 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4140 resync_start_show(struct mddev *mddev, char *page)
4142 if (mddev->recovery_cp == MaxSector)
4143 return sprintf(page, "none\n");
4144 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4148 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4150 unsigned long long n;
4153 if (cmd_match(buf, "none"))
4156 err = kstrtoull(buf, 10, &n);
4159 if (n != (sector_t)n)
4163 err = mddev_lock(mddev);
4166 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4170 mddev->recovery_cp = n;
4172 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4174 mddev_unlock(mddev);
4177 static struct md_sysfs_entry md_resync_start =
4178 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4179 resync_start_show, resync_start_store);
4182 * The array state can be:
4185 * No devices, no size, no level
4186 * Equivalent to STOP_ARRAY ioctl
4188 * May have some settings, but array is not active
4189 * all IO results in error
4190 * When written, doesn't tear down array, but just stops it
4191 * suspended (not supported yet)
4192 * All IO requests will block. The array can be reconfigured.
4193 * Writing this, if accepted, will block until array is quiescent
4195 * no resync can happen. no superblocks get written.
4196 * write requests fail
4198 * like readonly, but behaves like 'clean' on a write request.
4200 * clean - no pending writes, but otherwise active.
4201 * When written to inactive array, starts without resync
4202 * If a write request arrives then
4203 * if metadata is known, mark 'dirty' and switch to 'active'.
4204 * if not known, block and switch to write-pending
4205 * If written to an active array that has pending writes, then fails.
4207 * fully active: IO and resync can be happening.
4208 * When written to inactive array, starts with resync
4211 * clean, but writes are blocked waiting for 'active' to be written.
4214 * like active, but no writes have been seen for a while (100msec).
4217 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4218 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4219 * when a member is gone, so this state will at least alert the
4220 * user that something is wrong.
4222 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4223 write_pending, active_idle, broken, bad_word};
4224 static char *array_states[] = {
4225 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4226 "write-pending", "active-idle", "broken", NULL };
4228 static int match_word(const char *word, char **list)
4231 for (n=0; list[n]; n++)
4232 if (cmd_match(word, list[n]))
4238 array_state_show(struct mddev *mddev, char *page)
4240 enum array_state st = inactive;
4242 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4251 spin_lock(&mddev->lock);
4252 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4254 else if (mddev->in_sync)
4256 else if (mddev->safemode)
4260 spin_unlock(&mddev->lock);
4263 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4266 if (list_empty(&mddev->disks) &&
4267 mddev->raid_disks == 0 &&
4268 mddev->dev_sectors == 0)
4273 return sprintf(page, "%s\n", array_states[st]);
4276 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4277 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4278 static int do_md_run(struct mddev *mddev);
4279 static int restart_array(struct mddev *mddev);
4282 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4285 enum array_state st = match_word(buf, array_states);
4287 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4288 /* don't take reconfig_mutex when toggling between
4291 spin_lock(&mddev->lock);
4293 restart_array(mddev);
4294 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4295 md_wakeup_thread(mddev->thread);
4296 wake_up(&mddev->sb_wait);
4297 } else /* st == clean */ {
4298 restart_array(mddev);
4299 if (!set_in_sync(mddev))
4303 sysfs_notify_dirent_safe(mddev->sysfs_state);
4304 spin_unlock(&mddev->lock);
4307 err = mddev_lock(mddev);
4315 /* stopping an active array */
4316 err = do_md_stop(mddev, 0, NULL);
4319 /* stopping an active array */
4321 err = do_md_stop(mddev, 2, NULL);
4323 err = 0; /* already inactive */
4326 break; /* not supported yet */
4329 err = md_set_readonly(mddev, NULL);
4332 set_disk_ro(mddev->gendisk, 1);
4333 err = do_md_run(mddev);
4339 err = md_set_readonly(mddev, NULL);
4340 else if (mddev->ro == 1)
4341 err = restart_array(mddev);
4344 set_disk_ro(mddev->gendisk, 0);
4348 err = do_md_run(mddev);
4353 err = restart_array(mddev);
4356 spin_lock(&mddev->lock);
4357 if (!set_in_sync(mddev))
4359 spin_unlock(&mddev->lock);
4365 err = restart_array(mddev);
4368 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4369 wake_up(&mddev->sb_wait);
4373 set_disk_ro(mddev->gendisk, 0);
4374 err = do_md_run(mddev);
4380 /* these cannot be set */
4385 if (mddev->hold_active == UNTIL_IOCTL)
4386 mddev->hold_active = 0;
4387 sysfs_notify_dirent_safe(mddev->sysfs_state);
4389 mddev_unlock(mddev);
4392 static struct md_sysfs_entry md_array_state =
4393 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4396 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4397 return sprintf(page, "%d\n",
4398 atomic_read(&mddev->max_corr_read_errors));
4402 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4407 rv = kstrtouint(buf, 10, &n);
4410 atomic_set(&mddev->max_corr_read_errors, n);
4414 static struct md_sysfs_entry max_corr_read_errors =
4415 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4416 max_corrected_read_errors_store);
4419 null_show(struct mddev *mddev, char *page)
4425 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4427 /* buf must be %d:%d\n? giving major and minor numbers */
4428 /* The new device is added to the array.
4429 * If the array has a persistent superblock, we read the
4430 * superblock to initialise info and check validity.
4431 * Otherwise, only checking done is that in bind_rdev_to_array,
4432 * which mainly checks size.
4435 int major = simple_strtoul(buf, &e, 10);
4438 struct md_rdev *rdev;
4441 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4443 minor = simple_strtoul(e+1, &e, 10);
4444 if (*e && *e != '\n')
4446 dev = MKDEV(major, minor);
4447 if (major != MAJOR(dev) ||
4448 minor != MINOR(dev))
4451 flush_workqueue(md_misc_wq);
4453 err = mddev_lock(mddev);
4456 if (mddev->persistent) {
4457 rdev = md_import_device(dev, mddev->major_version,
4458 mddev->minor_version);
4459 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4460 struct md_rdev *rdev0
4461 = list_entry(mddev->disks.next,
4462 struct md_rdev, same_set);
4463 err = super_types[mddev->major_version]
4464 .load_super(rdev, rdev0, mddev->minor_version);
4468 } else if (mddev->external)
4469 rdev = md_import_device(dev, -2, -1);
4471 rdev = md_import_device(dev, -1, -1);
4474 mddev_unlock(mddev);
4475 return PTR_ERR(rdev);
4477 err = bind_rdev_to_array(rdev, mddev);
4481 mddev_unlock(mddev);
4483 md_new_event(mddev);
4484 return err ? err : len;
4487 static struct md_sysfs_entry md_new_device =
4488 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4491 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4494 unsigned long chunk, end_chunk;
4497 err = mddev_lock(mddev);
4502 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4504 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4505 if (buf == end) break;
4506 if (*end == '-') { /* range */
4508 end_chunk = simple_strtoul(buf, &end, 0);
4509 if (buf == end) break;
4511 if (*end && !isspace(*end)) break;
4512 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4513 buf = skip_spaces(end);
4515 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4517 mddev_unlock(mddev);
4521 static struct md_sysfs_entry md_bitmap =
4522 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4525 size_show(struct mddev *mddev, char *page)
4527 return sprintf(page, "%llu\n",
4528 (unsigned long long)mddev->dev_sectors / 2);
4531 static int update_size(struct mddev *mddev, sector_t num_sectors);
4534 size_store(struct mddev *mddev, const char *buf, size_t len)
4536 /* If array is inactive, we can reduce the component size, but
4537 * not increase it (except from 0).
4538 * If array is active, we can try an on-line resize
4541 int err = strict_blocks_to_sectors(buf, §ors);
4545 err = mddev_lock(mddev);
4549 err = update_size(mddev, sectors);
4551 md_update_sb(mddev, 1);
4553 if (mddev->dev_sectors == 0 ||
4554 mddev->dev_sectors > sectors)
4555 mddev->dev_sectors = sectors;
4559 mddev_unlock(mddev);
4560 return err ? err : len;
4563 static struct md_sysfs_entry md_size =
4564 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4566 /* Metadata version.
4568 * 'none' for arrays with no metadata (good luck...)
4569 * 'external' for arrays with externally managed metadata,
4570 * or N.M for internally known formats
4573 metadata_show(struct mddev *mddev, char *page)
4575 if (mddev->persistent)
4576 return sprintf(page, "%d.%d\n",
4577 mddev->major_version, mddev->minor_version);
4578 else if (mddev->external)
4579 return sprintf(page, "external:%s\n", mddev->metadata_type);
4581 return sprintf(page, "none\n");
4585 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4590 /* Changing the details of 'external' metadata is
4591 * always permitted. Otherwise there must be
4592 * no devices attached to the array.
4595 err = mddev_lock(mddev);
4599 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4601 else if (!list_empty(&mddev->disks))
4605 if (cmd_match(buf, "none")) {
4606 mddev->persistent = 0;
4607 mddev->external = 0;
4608 mddev->major_version = 0;
4609 mddev->minor_version = 90;
4612 if (strncmp(buf, "external:", 9) == 0) {
4613 size_t namelen = len-9;
4614 if (namelen >= sizeof(mddev->metadata_type))
4615 namelen = sizeof(mddev->metadata_type)-1;
4616 strncpy(mddev->metadata_type, buf+9, namelen);
4617 mddev->metadata_type[namelen] = 0;
4618 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4619 mddev->metadata_type[--namelen] = 0;
4620 mddev->persistent = 0;
4621 mddev->external = 1;
4622 mddev->major_version = 0;
4623 mddev->minor_version = 90;
4626 major = simple_strtoul(buf, &e, 10);
4628 if (e==buf || *e != '.')
4631 minor = simple_strtoul(buf, &e, 10);
4632 if (e==buf || (*e && *e != '\n') )
4635 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4637 mddev->major_version = major;
4638 mddev->minor_version = minor;
4639 mddev->persistent = 1;
4640 mddev->external = 0;
4643 mddev_unlock(mddev);
4647 static struct md_sysfs_entry md_metadata =
4648 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4651 action_show(struct mddev *mddev, char *page)
4653 char *type = "idle";
4654 unsigned long recovery = mddev->recovery;
4655 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4657 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4658 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4659 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4661 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4662 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4664 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4668 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4670 else if (mddev->reshape_position != MaxSector)
4673 return sprintf(page, "%s\n", type);
4677 action_store(struct mddev *mddev, const char *page, size_t len)
4679 if (!mddev->pers || !mddev->pers->sync_request)
4683 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4684 if (cmd_match(page, "frozen"))
4685 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4687 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4688 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4689 mddev_lock(mddev) == 0) {
4690 flush_workqueue(md_misc_wq);
4691 if (mddev->sync_thread) {
4692 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4693 md_reap_sync_thread(mddev);
4695 mddev_unlock(mddev);
4697 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4699 else if (cmd_match(page, "resync"))
4700 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4701 else if (cmd_match(page, "recover")) {
4702 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4703 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4704 } else if (cmd_match(page, "reshape")) {
4706 if (mddev->pers->start_reshape == NULL)
4708 err = mddev_lock(mddev);
4710 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4713 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4714 err = mddev->pers->start_reshape(mddev);
4716 mddev_unlock(mddev);
4720 sysfs_notify(&mddev->kobj, NULL, "degraded");
4722 if (cmd_match(page, "check"))
4723 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4724 else if (!cmd_match(page, "repair"))
4726 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4727 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4728 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4730 if (mddev->ro == 2) {
4731 /* A write to sync_action is enough to justify
4732 * canceling read-auto mode
4735 md_wakeup_thread(mddev->sync_thread);
4737 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4738 md_wakeup_thread(mddev->thread);
4739 sysfs_notify_dirent_safe(mddev->sysfs_action);
4743 static struct md_sysfs_entry md_scan_mode =
4744 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4747 last_sync_action_show(struct mddev *mddev, char *page)
4749 return sprintf(page, "%s\n", mddev->last_sync_action);
4752 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4755 mismatch_cnt_show(struct mddev *mddev, char *page)
4757 return sprintf(page, "%llu\n",
4758 (unsigned long long)
4759 atomic64_read(&mddev->resync_mismatches));
4762 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4765 sync_min_show(struct mddev *mddev, char *page)
4767 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4768 mddev->sync_speed_min ? "local": "system");
4772 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4777 if (strncmp(buf, "system", 6)==0) {
4780 rv = kstrtouint(buf, 10, &min);
4786 mddev->sync_speed_min = min;
4790 static struct md_sysfs_entry md_sync_min =
4791 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4794 sync_max_show(struct mddev *mddev, char *page)
4796 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4797 mddev->sync_speed_max ? "local": "system");
4801 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4806 if (strncmp(buf, "system", 6)==0) {
4809 rv = kstrtouint(buf, 10, &max);
4815 mddev->sync_speed_max = max;
4819 static struct md_sysfs_entry md_sync_max =
4820 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4823 degraded_show(struct mddev *mddev, char *page)
4825 return sprintf(page, "%d\n", mddev->degraded);
4827 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4830 sync_force_parallel_show(struct mddev *mddev, char *page)
4832 return sprintf(page, "%d\n", mddev->parallel_resync);
4836 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4840 if (kstrtol(buf, 10, &n))
4843 if (n != 0 && n != 1)
4846 mddev->parallel_resync = n;
4848 if (mddev->sync_thread)
4849 wake_up(&resync_wait);
4854 /* force parallel resync, even with shared block devices */
4855 static struct md_sysfs_entry md_sync_force_parallel =
4856 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4857 sync_force_parallel_show, sync_force_parallel_store);
4860 sync_speed_show(struct mddev *mddev, char *page)
4862 unsigned long resync, dt, db;
4863 if (mddev->curr_resync == 0)
4864 return sprintf(page, "none\n");
4865 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4866 dt = (jiffies - mddev->resync_mark) / HZ;
4868 db = resync - mddev->resync_mark_cnt;
4869 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4872 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4875 sync_completed_show(struct mddev *mddev, char *page)
4877 unsigned long long max_sectors, resync;
4879 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4880 return sprintf(page, "none\n");
4882 if (mddev->curr_resync == 1 ||
4883 mddev->curr_resync == 2)
4884 return sprintf(page, "delayed\n");
4886 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4887 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4888 max_sectors = mddev->resync_max_sectors;
4890 max_sectors = mddev->dev_sectors;
4892 resync = mddev->curr_resync_completed;
4893 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4896 static struct md_sysfs_entry md_sync_completed =
4897 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4900 min_sync_show(struct mddev *mddev, char *page)
4902 return sprintf(page, "%llu\n",
4903 (unsigned long long)mddev->resync_min);
4906 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4908 unsigned long long min;
4911 if (kstrtoull(buf, 10, &min))
4914 spin_lock(&mddev->lock);
4916 if (min > mddev->resync_max)
4920 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4923 /* Round down to multiple of 4K for safety */
4924 mddev->resync_min = round_down(min, 8);
4928 spin_unlock(&mddev->lock);
4932 static struct md_sysfs_entry md_min_sync =
4933 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4936 max_sync_show(struct mddev *mddev, char *page)
4938 if (mddev->resync_max == MaxSector)
4939 return sprintf(page, "max\n");
4941 return sprintf(page, "%llu\n",
4942 (unsigned long long)mddev->resync_max);
4945 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4948 spin_lock(&mddev->lock);
4949 if (strncmp(buf, "max", 3) == 0)
4950 mddev->resync_max = MaxSector;
4952 unsigned long long max;
4956 if (kstrtoull(buf, 10, &max))
4958 if (max < mddev->resync_min)
4962 if (max < mddev->resync_max &&
4964 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4967 /* Must be a multiple of chunk_size */
4968 chunk = mddev->chunk_sectors;
4970 sector_t temp = max;
4973 if (sector_div(temp, chunk))
4976 mddev->resync_max = max;
4978 wake_up(&mddev->recovery_wait);
4981 spin_unlock(&mddev->lock);
4985 static struct md_sysfs_entry md_max_sync =
4986 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4989 suspend_lo_show(struct mddev *mddev, char *page)
4991 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4995 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4997 unsigned long long new;
5000 err = kstrtoull(buf, 10, &new);
5003 if (new != (sector_t)new)
5006 err = mddev_lock(mddev);
5010 if (mddev->pers == NULL ||
5011 mddev->pers->quiesce == NULL)
5013 mddev_suspend(mddev);
5014 mddev->suspend_lo = new;
5015 mddev_resume(mddev);
5019 mddev_unlock(mddev);
5022 static struct md_sysfs_entry md_suspend_lo =
5023 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5026 suspend_hi_show(struct mddev *mddev, char *page)
5028 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5032 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5034 unsigned long long new;
5037 err = kstrtoull(buf, 10, &new);
5040 if (new != (sector_t)new)
5043 err = mddev_lock(mddev);
5047 if (mddev->pers == NULL)
5050 mddev_suspend(mddev);
5051 mddev->suspend_hi = new;
5052 mddev_resume(mddev);
5056 mddev_unlock(mddev);
5059 static struct md_sysfs_entry md_suspend_hi =
5060 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5063 reshape_position_show(struct mddev *mddev, char *page)
5065 if (mddev->reshape_position != MaxSector)
5066 return sprintf(page, "%llu\n",
5067 (unsigned long long)mddev->reshape_position);
5068 strcpy(page, "none\n");
5073 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5075 struct md_rdev *rdev;
5076 unsigned long long new;
5079 err = kstrtoull(buf, 10, &new);
5082 if (new != (sector_t)new)
5084 err = mddev_lock(mddev);
5090 mddev->reshape_position = new;
5091 mddev->delta_disks = 0;
5092 mddev->reshape_backwards = 0;
5093 mddev->new_level = mddev->level;
5094 mddev->new_layout = mddev->layout;
5095 mddev->new_chunk_sectors = mddev->chunk_sectors;
5096 rdev_for_each(rdev, mddev)
5097 rdev->new_data_offset = rdev->data_offset;
5100 mddev_unlock(mddev);
5104 static struct md_sysfs_entry md_reshape_position =
5105 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5106 reshape_position_store);
5109 reshape_direction_show(struct mddev *mddev, char *page)
5111 return sprintf(page, "%s\n",
5112 mddev->reshape_backwards ? "backwards" : "forwards");
5116 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5121 if (cmd_match(buf, "forwards"))
5123 else if (cmd_match(buf, "backwards"))
5127 if (mddev->reshape_backwards == backwards)
5130 err = mddev_lock(mddev);
5133 /* check if we are allowed to change */
5134 if (mddev->delta_disks)
5136 else if (mddev->persistent &&
5137 mddev->major_version == 0)
5140 mddev->reshape_backwards = backwards;
5141 mddev_unlock(mddev);
5145 static struct md_sysfs_entry md_reshape_direction =
5146 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5147 reshape_direction_store);
5150 array_size_show(struct mddev *mddev, char *page)
5152 if (mddev->external_size)
5153 return sprintf(page, "%llu\n",
5154 (unsigned long long)mddev->array_sectors/2);
5156 return sprintf(page, "default\n");
5160 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5165 err = mddev_lock(mddev);
5169 /* cluster raid doesn't support change array_sectors */
5170 if (mddev_is_clustered(mddev)) {
5171 mddev_unlock(mddev);
5175 if (strncmp(buf, "default", 7) == 0) {
5177 sectors = mddev->pers->size(mddev, 0, 0);
5179 sectors = mddev->array_sectors;
5181 mddev->external_size = 0;
5183 if (strict_blocks_to_sectors(buf, §ors) < 0)
5185 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5188 mddev->external_size = 1;
5192 mddev->array_sectors = sectors;
5194 set_capacity(mddev->gendisk, mddev->array_sectors);
5195 revalidate_disk(mddev->gendisk);
5198 mddev_unlock(mddev);
5202 static struct md_sysfs_entry md_array_size =
5203 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5207 consistency_policy_show(struct mddev *mddev, char *page)
5211 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5212 ret = sprintf(page, "journal\n");
5213 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5214 ret = sprintf(page, "ppl\n");
5215 } else if (mddev->bitmap) {
5216 ret = sprintf(page, "bitmap\n");
5217 } else if (mddev->pers) {
5218 if (mddev->pers->sync_request)
5219 ret = sprintf(page, "resync\n");
5221 ret = sprintf(page, "none\n");
5223 ret = sprintf(page, "unknown\n");
5230 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5235 if (mddev->pers->change_consistency_policy)
5236 err = mddev->pers->change_consistency_policy(mddev, buf);
5239 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5240 set_bit(MD_HAS_PPL, &mddev->flags);
5245 return err ? err : len;
5248 static struct md_sysfs_entry md_consistency_policy =
5249 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5250 consistency_policy_store);
5252 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5254 return sprintf(page, "%d\n", mddev->fail_last_dev);
5258 * Setting fail_last_dev to true to allow last device to be forcibly removed
5259 * from RAID1/RAID10.
5262 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5267 ret = kstrtobool(buf, &value);
5271 if (value != mddev->fail_last_dev)
5272 mddev->fail_last_dev = value;
5276 static struct md_sysfs_entry md_fail_last_dev =
5277 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5278 fail_last_dev_store);
5280 static struct attribute *md_default_attrs[] = {
5283 &md_raid_disks.attr,
5284 &md_chunk_size.attr,
5286 &md_resync_start.attr,
5288 &md_new_device.attr,
5289 &md_safe_delay.attr,
5290 &md_array_state.attr,
5291 &md_reshape_position.attr,
5292 &md_reshape_direction.attr,
5293 &md_array_size.attr,
5294 &max_corr_read_errors.attr,
5295 &md_consistency_policy.attr,
5296 &md_fail_last_dev.attr,
5300 static struct attribute *md_redundancy_attrs[] = {
5302 &md_last_scan_mode.attr,
5303 &md_mismatches.attr,
5306 &md_sync_speed.attr,
5307 &md_sync_force_parallel.attr,
5308 &md_sync_completed.attr,
5311 &md_suspend_lo.attr,
5312 &md_suspend_hi.attr,
5317 static struct attribute_group md_redundancy_group = {
5319 .attrs = md_redundancy_attrs,
5323 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5325 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5326 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5331 spin_lock(&all_mddevs_lock);
5332 if (list_empty(&mddev->all_mddevs)) {
5333 spin_unlock(&all_mddevs_lock);
5337 spin_unlock(&all_mddevs_lock);
5339 rv = entry->show(mddev, page);
5345 md_attr_store(struct kobject *kobj, struct attribute *attr,
5346 const char *page, size_t length)
5348 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5349 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5354 if (!capable(CAP_SYS_ADMIN))
5356 spin_lock(&all_mddevs_lock);
5357 if (list_empty(&mddev->all_mddevs)) {
5358 spin_unlock(&all_mddevs_lock);
5362 spin_unlock(&all_mddevs_lock);
5363 rv = entry->store(mddev, page, length);
5368 static void md_free(struct kobject *ko)
5370 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5372 if (mddev->sysfs_state)
5373 sysfs_put(mddev->sysfs_state);
5376 del_gendisk(mddev->gendisk);
5378 blk_cleanup_queue(mddev->queue);
5380 put_disk(mddev->gendisk);
5381 percpu_ref_exit(&mddev->writes_pending);
5383 bioset_exit(&mddev->bio_set);
5384 bioset_exit(&mddev->sync_set);
5388 static const struct sysfs_ops md_sysfs_ops = {
5389 .show = md_attr_show,
5390 .store = md_attr_store,
5392 static struct kobj_type md_ktype = {
5394 .sysfs_ops = &md_sysfs_ops,
5395 .default_attrs = md_default_attrs,
5400 static void mddev_delayed_delete(struct work_struct *ws)
5402 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5404 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5405 kobject_del(&mddev->kobj);
5406 kobject_put(&mddev->kobj);
5409 static void no_op(struct percpu_ref *r) {}
5411 int mddev_init_writes_pending(struct mddev *mddev)
5413 if (mddev->writes_pending.percpu_count_ptr)
5415 if (percpu_ref_init(&mddev->writes_pending, no_op,
5416 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5418 /* We want to start with the refcount at zero */
5419 percpu_ref_put(&mddev->writes_pending);
5422 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5424 static int md_alloc(dev_t dev, char *name)
5427 * If dev is zero, name is the name of a device to allocate with
5428 * an arbitrary minor number. It will be "md_???"
5429 * If dev is non-zero it must be a device number with a MAJOR of
5430 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5431 * the device is being created by opening a node in /dev.
5432 * If "name" is not NULL, the device is being created by
5433 * writing to /sys/module/md_mod/parameters/new_array.
5435 static DEFINE_MUTEX(disks_mutex);
5436 struct mddev *mddev = mddev_find(dev);
5437 struct gendisk *disk;
5446 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5447 shift = partitioned ? MdpMinorShift : 0;
5448 unit = MINOR(mddev->unit) >> shift;
5450 /* wait for any previous instance of this device to be
5451 * completely removed (mddev_delayed_delete).
5453 flush_workqueue(md_misc_wq);
5455 mutex_lock(&disks_mutex);
5461 /* Need to ensure that 'name' is not a duplicate.
5463 struct mddev *mddev2;
5464 spin_lock(&all_mddevs_lock);
5466 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5467 if (mddev2->gendisk &&
5468 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5469 spin_unlock(&all_mddevs_lock);
5472 spin_unlock(&all_mddevs_lock);
5476 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5478 mddev->hold_active = UNTIL_STOP;
5481 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5484 mddev->queue->queuedata = mddev;
5486 blk_queue_make_request(mddev->queue, md_make_request);
5487 blk_set_stacking_limits(&mddev->queue->limits);
5489 disk = alloc_disk(1 << shift);
5491 blk_cleanup_queue(mddev->queue);
5492 mddev->queue = NULL;
5495 disk->major = MAJOR(mddev->unit);
5496 disk->first_minor = unit << shift;
5498 strcpy(disk->disk_name, name);
5499 else if (partitioned)
5500 sprintf(disk->disk_name, "md_d%d", unit);
5502 sprintf(disk->disk_name, "md%d", unit);
5503 disk->fops = &md_fops;
5504 disk->private_data = mddev;
5505 disk->queue = mddev->queue;
5506 blk_queue_write_cache(mddev->queue, true, true);
5507 /* Allow extended partitions. This makes the
5508 * 'mdp' device redundant, but we can't really
5511 disk->flags |= GENHD_FL_EXT_DEVT;
5512 mddev->gendisk = disk;
5513 /* As soon as we call add_disk(), another thread could get
5514 * through to md_open, so make sure it doesn't get too far
5516 mutex_lock(&mddev->open_mutex);
5519 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5521 /* This isn't possible, but as kobject_init_and_add is marked
5522 * __must_check, we must do something with the result
5524 pr_debug("md: cannot register %s/md - name in use\n",
5528 if (mddev->kobj.sd &&
5529 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5530 pr_debug("pointless warning\n");
5531 mutex_unlock(&mddev->open_mutex);
5533 mutex_unlock(&disks_mutex);
5534 if (!error && mddev->kobj.sd) {
5535 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5536 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5542 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5545 md_alloc(dev, NULL);
5549 static int add_named_array(const char *val, const struct kernel_param *kp)
5552 * val must be "md_*" or "mdNNN".
5553 * For "md_*" we allocate an array with a large free minor number, and
5554 * set the name to val. val must not already be an active name.
5555 * For "mdNNN" we allocate an array with the minor number NNN
5556 * which must not already be in use.
5558 int len = strlen(val);
5559 char buf[DISK_NAME_LEN];
5560 unsigned long devnum;
5562 while (len && val[len-1] == '\n')
5564 if (len >= DISK_NAME_LEN)
5566 strlcpy(buf, val, len+1);
5567 if (strncmp(buf, "md_", 3) == 0)
5568 return md_alloc(0, buf);
5569 if (strncmp(buf, "md", 2) == 0 &&
5571 kstrtoul(buf+2, 10, &devnum) == 0 &&
5572 devnum <= MINORMASK)
5573 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5578 static void md_safemode_timeout(struct timer_list *t)
5580 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5582 mddev->safemode = 1;
5583 if (mddev->external)
5584 sysfs_notify_dirent_safe(mddev->sysfs_state);
5586 md_wakeup_thread(mddev->thread);
5589 static int start_dirty_degraded;
5591 int md_run(struct mddev *mddev)
5594 struct md_rdev *rdev;
5595 struct md_personality *pers;
5597 if (list_empty(&mddev->disks))
5598 /* cannot run an array with no devices.. */
5603 /* Cannot run until previous stop completes properly */
5604 if (mddev->sysfs_active)
5608 * Analyze all RAID superblock(s)
5610 if (!mddev->raid_disks) {
5611 if (!mddev->persistent)
5613 err = analyze_sbs(mddev);
5618 if (mddev->level != LEVEL_NONE)
5619 request_module("md-level-%d", mddev->level);
5620 else if (mddev->clevel[0])
5621 request_module("md-%s", mddev->clevel);
5624 * Drop all container device buffers, from now on
5625 * the only valid external interface is through the md
5628 mddev->has_superblocks = false;
5629 rdev_for_each(rdev, mddev) {
5630 if (test_bit(Faulty, &rdev->flags))
5632 sync_blockdev(rdev->bdev);
5633 invalidate_bdev(rdev->bdev);
5634 if (mddev->ro != 1 &&
5635 (bdev_read_only(rdev->bdev) ||
5636 bdev_read_only(rdev->meta_bdev))) {
5639 set_disk_ro(mddev->gendisk, 1);
5643 mddev->has_superblocks = true;
5645 /* perform some consistency tests on the device.
5646 * We don't want the data to overlap the metadata,
5647 * Internal Bitmap issues have been handled elsewhere.
5649 if (rdev->meta_bdev) {
5650 /* Nothing to check */;
5651 } else if (rdev->data_offset < rdev->sb_start) {
5652 if (mddev->dev_sectors &&
5653 rdev->data_offset + mddev->dev_sectors
5655 pr_warn("md: %s: data overlaps metadata\n",
5660 if (rdev->sb_start + rdev->sb_size/512
5661 > rdev->data_offset) {
5662 pr_warn("md: %s: metadata overlaps data\n",
5667 sysfs_notify_dirent_safe(rdev->sysfs_state);
5670 if (!bioset_initialized(&mddev->bio_set)) {
5671 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5675 if (!bioset_initialized(&mddev->sync_set)) {
5676 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5681 spin_lock(&pers_lock);
5682 pers = find_pers(mddev->level, mddev->clevel);
5683 if (!pers || !try_module_get(pers->owner)) {
5684 spin_unlock(&pers_lock);
5685 if (mddev->level != LEVEL_NONE)
5686 pr_warn("md: personality for level %d is not loaded!\n",
5689 pr_warn("md: personality for level %s is not loaded!\n",
5694 spin_unlock(&pers_lock);
5695 if (mddev->level != pers->level) {
5696 mddev->level = pers->level;
5697 mddev->new_level = pers->level;
5699 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5701 if (mddev->reshape_position != MaxSector &&
5702 pers->start_reshape == NULL) {
5703 /* This personality cannot handle reshaping... */
5704 module_put(pers->owner);
5709 if (pers->sync_request) {
5710 /* Warn if this is a potentially silly
5713 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5714 struct md_rdev *rdev2;
5717 rdev_for_each(rdev, mddev)
5718 rdev_for_each(rdev2, mddev) {
5720 rdev->bdev->bd_contains ==
5721 rdev2->bdev->bd_contains) {
5722 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5724 bdevname(rdev->bdev,b),
5725 bdevname(rdev2->bdev,b2));
5731 pr_warn("True protection against single-disk failure might be compromised.\n");
5734 mddev->recovery = 0;
5735 /* may be over-ridden by personality */
5736 mddev->resync_max_sectors = mddev->dev_sectors;
5738 mddev->ok_start_degraded = start_dirty_degraded;
5740 if (start_readonly && mddev->ro == 0)
5741 mddev->ro = 2; /* read-only, but switch on first write */
5743 err = pers->run(mddev);
5745 pr_warn("md: pers->run() failed ...\n");
5746 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5747 WARN_ONCE(!mddev->external_size,
5748 "%s: default size too small, but 'external_size' not in effect?\n",
5750 pr_warn("md: invalid array_size %llu > default size %llu\n",
5751 (unsigned long long)mddev->array_sectors / 2,
5752 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5755 if (err == 0 && pers->sync_request &&
5756 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5757 struct bitmap *bitmap;
5759 bitmap = md_bitmap_create(mddev, -1);
5760 if (IS_ERR(bitmap)) {
5761 err = PTR_ERR(bitmap);
5762 pr_warn("%s: failed to create bitmap (%d)\n",
5763 mdname(mddev), err);
5765 mddev->bitmap = bitmap;
5771 if (mddev->bitmap_info.max_write_behind > 0) {
5772 bool creat_pool = false;
5774 rdev_for_each(rdev, mddev) {
5775 if (test_bit(WriteMostly, &rdev->flags) &&
5779 if (creat_pool && mddev->wb_info_pool == NULL) {
5780 mddev->wb_info_pool =
5781 mempool_create_kmalloc_pool(NR_WB_INFOS,
5782 sizeof(struct wb_info));
5783 if (!mddev->wb_info_pool) {
5793 rdev_for_each(rdev, mddev) {
5794 if (rdev->raid_disk >= 0 &&
5795 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5800 if (mddev->degraded)
5803 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5805 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5806 mddev->queue->backing_dev_info->congested_data = mddev;
5807 mddev->queue->backing_dev_info->congested_fn = md_congested;
5809 if (pers->sync_request) {
5810 if (mddev->kobj.sd &&
5811 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5812 pr_warn("md: cannot register extra attributes for %s\n",
5814 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5815 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5818 atomic_set(&mddev->max_corr_read_errors,
5819 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5820 mddev->safemode = 0;
5821 if (mddev_is_clustered(mddev))
5822 mddev->safemode_delay = 0;
5824 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5827 spin_lock(&mddev->lock);
5829 spin_unlock(&mddev->lock);
5830 rdev_for_each(rdev, mddev)
5831 if (rdev->raid_disk >= 0)
5832 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
5834 if (mddev->degraded && !mddev->ro)
5835 /* This ensures that recovering status is reported immediately
5836 * via sysfs - until a lack of spares is confirmed.
5838 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5839 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5841 if (mddev->sb_flags)
5842 md_update_sb(mddev, 0);
5844 md_new_event(mddev);
5848 mddev_detach(mddev);
5850 pers->free(mddev, mddev->private);
5851 mddev->private = NULL;
5852 module_put(pers->owner);
5853 md_bitmap_destroy(mddev);
5855 bioset_exit(&mddev->bio_set);
5856 bioset_exit(&mddev->sync_set);
5859 EXPORT_SYMBOL_GPL(md_run);
5861 static int do_md_run(struct mddev *mddev)
5865 set_bit(MD_NOT_READY, &mddev->flags);
5866 err = md_run(mddev);
5869 err = md_bitmap_load(mddev);
5871 md_bitmap_destroy(mddev);
5875 if (mddev_is_clustered(mddev))
5876 md_allow_write(mddev);
5878 /* run start up tasks that require md_thread */
5881 md_wakeup_thread(mddev->thread);
5882 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5884 set_capacity(mddev->gendisk, mddev->array_sectors);
5885 revalidate_disk(mddev->gendisk);
5886 clear_bit(MD_NOT_READY, &mddev->flags);
5888 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5889 sysfs_notify_dirent_safe(mddev->sysfs_state);
5890 sysfs_notify_dirent_safe(mddev->sysfs_action);
5891 sysfs_notify(&mddev->kobj, NULL, "degraded");
5893 clear_bit(MD_NOT_READY, &mddev->flags);
5897 int md_start(struct mddev *mddev)
5901 if (mddev->pers->start) {
5902 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5903 md_wakeup_thread(mddev->thread);
5904 ret = mddev->pers->start(mddev);
5905 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5906 md_wakeup_thread(mddev->sync_thread);
5910 EXPORT_SYMBOL_GPL(md_start);
5912 static int restart_array(struct mddev *mddev)
5914 struct gendisk *disk = mddev->gendisk;
5915 struct md_rdev *rdev;
5916 bool has_journal = false;
5917 bool has_readonly = false;
5919 /* Complain if it has no devices */
5920 if (list_empty(&mddev->disks))
5928 rdev_for_each_rcu(rdev, mddev) {
5929 if (test_bit(Journal, &rdev->flags) &&
5930 !test_bit(Faulty, &rdev->flags))
5932 if (bdev_read_only(rdev->bdev))
5933 has_readonly = true;
5936 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5937 /* Don't restart rw with journal missing/faulty */
5942 mddev->safemode = 0;
5944 set_disk_ro(disk, 0);
5945 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5946 /* Kick recovery or resync if necessary */
5947 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5948 md_wakeup_thread(mddev->thread);
5949 md_wakeup_thread(mddev->sync_thread);
5950 sysfs_notify_dirent_safe(mddev->sysfs_state);
5954 static void md_clean(struct mddev *mddev)
5956 mddev->array_sectors = 0;
5957 mddev->external_size = 0;
5958 mddev->dev_sectors = 0;
5959 mddev->raid_disks = 0;
5960 mddev->recovery_cp = 0;
5961 mddev->resync_min = 0;
5962 mddev->resync_max = MaxSector;
5963 mddev->reshape_position = MaxSector;
5964 mddev->external = 0;
5965 mddev->persistent = 0;
5966 mddev->level = LEVEL_NONE;
5967 mddev->clevel[0] = 0;
5969 mddev->sb_flags = 0;
5971 mddev->metadata_type[0] = 0;
5972 mddev->chunk_sectors = 0;
5973 mddev->ctime = mddev->utime = 0;
5975 mddev->max_disks = 0;
5977 mddev->can_decrease_events = 0;
5978 mddev->delta_disks = 0;
5979 mddev->reshape_backwards = 0;
5980 mddev->new_level = LEVEL_NONE;
5981 mddev->new_layout = 0;
5982 mddev->new_chunk_sectors = 0;
5983 mddev->curr_resync = 0;
5984 atomic64_set(&mddev->resync_mismatches, 0);
5985 mddev->suspend_lo = mddev->suspend_hi = 0;
5986 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5987 mddev->recovery = 0;
5990 mddev->degraded = 0;
5991 mddev->safemode = 0;
5992 mddev->private = NULL;
5993 mddev->cluster_info = NULL;
5994 mddev->bitmap_info.offset = 0;
5995 mddev->bitmap_info.default_offset = 0;
5996 mddev->bitmap_info.default_space = 0;
5997 mddev->bitmap_info.chunksize = 0;
5998 mddev->bitmap_info.daemon_sleep = 0;
5999 mddev->bitmap_info.max_write_behind = 0;
6000 mddev->bitmap_info.nodes = 0;
6003 static void __md_stop_writes(struct mddev *mddev)
6005 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6006 flush_workqueue(md_misc_wq);
6007 if (mddev->sync_thread) {
6008 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6009 md_reap_sync_thread(mddev);
6012 del_timer_sync(&mddev->safemode_timer);
6014 if (mddev->pers && mddev->pers->quiesce) {
6015 mddev->pers->quiesce(mddev, 1);
6016 mddev->pers->quiesce(mddev, 0);
6018 md_bitmap_flush(mddev);
6020 if (mddev->ro == 0 &&
6021 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6023 /* mark array as shutdown cleanly */
6024 if (!mddev_is_clustered(mddev))
6026 md_update_sb(mddev, 1);
6028 mempool_destroy(mddev->wb_info_pool);
6029 mddev->wb_info_pool = NULL;
6032 void md_stop_writes(struct mddev *mddev)
6034 mddev_lock_nointr(mddev);
6035 __md_stop_writes(mddev);
6036 mddev_unlock(mddev);
6038 EXPORT_SYMBOL_GPL(md_stop_writes);
6040 static void mddev_detach(struct mddev *mddev)
6042 md_bitmap_wait_behind_writes(mddev);
6043 if (mddev->pers && mddev->pers->quiesce) {
6044 mddev->pers->quiesce(mddev, 1);
6045 mddev->pers->quiesce(mddev, 0);
6047 md_unregister_thread(&mddev->thread);
6049 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6052 static void __md_stop(struct mddev *mddev)
6054 struct md_personality *pers = mddev->pers;
6055 md_bitmap_destroy(mddev);
6056 mddev_detach(mddev);
6057 /* Ensure ->event_work is done */
6058 flush_workqueue(md_misc_wq);
6059 spin_lock(&mddev->lock);
6061 spin_unlock(&mddev->lock);
6062 pers->free(mddev, mddev->private);
6063 mddev->private = NULL;
6064 if (pers->sync_request && mddev->to_remove == NULL)
6065 mddev->to_remove = &md_redundancy_group;
6066 module_put(pers->owner);
6067 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6070 void md_stop(struct mddev *mddev)
6072 /* stop the array and free an attached data structures.
6073 * This is called from dm-raid
6076 bioset_exit(&mddev->bio_set);
6077 bioset_exit(&mddev->sync_set);
6080 EXPORT_SYMBOL_GPL(md_stop);
6082 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6087 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6089 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6090 md_wakeup_thread(mddev->thread);
6092 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6093 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6094 if (mddev->sync_thread)
6095 /* Thread might be blocked waiting for metadata update
6096 * which will now never happen */
6097 wake_up_process(mddev->sync_thread->tsk);
6099 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6101 mddev_unlock(mddev);
6102 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6104 wait_event(mddev->sb_wait,
6105 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6106 mddev_lock_nointr(mddev);
6108 mutex_lock(&mddev->open_mutex);
6109 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6110 mddev->sync_thread ||
6111 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6112 pr_warn("md: %s still in use.\n",mdname(mddev));
6114 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6115 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6116 md_wakeup_thread(mddev->thread);
6122 __md_stop_writes(mddev);
6128 set_disk_ro(mddev->gendisk, 1);
6129 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6130 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6131 md_wakeup_thread(mddev->thread);
6132 sysfs_notify_dirent_safe(mddev->sysfs_state);
6136 mutex_unlock(&mddev->open_mutex);
6141 * 0 - completely stop and dis-assemble array
6142 * 2 - stop but do not disassemble array
6144 static int do_md_stop(struct mddev *mddev, int mode,
6145 struct block_device *bdev)
6147 struct gendisk *disk = mddev->gendisk;
6148 struct md_rdev *rdev;
6151 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6153 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6154 md_wakeup_thread(mddev->thread);
6156 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6157 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6158 if (mddev->sync_thread)
6159 /* Thread might be blocked waiting for metadata update
6160 * which will now never happen */
6161 wake_up_process(mddev->sync_thread->tsk);
6163 mddev_unlock(mddev);
6164 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6165 !test_bit(MD_RECOVERY_RUNNING,
6166 &mddev->recovery)));
6167 mddev_lock_nointr(mddev);
6169 mutex_lock(&mddev->open_mutex);
6170 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6171 mddev->sysfs_active ||
6172 mddev->sync_thread ||
6173 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6174 pr_warn("md: %s still in use.\n",mdname(mddev));
6175 mutex_unlock(&mddev->open_mutex);
6177 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6178 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6179 md_wakeup_thread(mddev->thread);
6185 set_disk_ro(disk, 0);
6187 __md_stop_writes(mddev);
6189 mddev->queue->backing_dev_info->congested_fn = NULL;
6191 /* tell userspace to handle 'inactive' */
6192 sysfs_notify_dirent_safe(mddev->sysfs_state);
6194 rdev_for_each(rdev, mddev)
6195 if (rdev->raid_disk >= 0)
6196 sysfs_unlink_rdev(mddev, rdev);
6198 set_capacity(disk, 0);
6199 mutex_unlock(&mddev->open_mutex);
6201 revalidate_disk(disk);
6206 mutex_unlock(&mddev->open_mutex);
6208 * Free resources if final stop
6211 pr_info("md: %s stopped.\n", mdname(mddev));
6213 if (mddev->bitmap_info.file) {
6214 struct file *f = mddev->bitmap_info.file;
6215 spin_lock(&mddev->lock);
6216 mddev->bitmap_info.file = NULL;
6217 spin_unlock(&mddev->lock);
6220 mddev->bitmap_info.offset = 0;
6222 export_array(mddev);
6225 if (mddev->hold_active == UNTIL_STOP)
6226 mddev->hold_active = 0;
6228 md_new_event(mddev);
6229 sysfs_notify_dirent_safe(mddev->sysfs_state);
6234 static void autorun_array(struct mddev *mddev)
6236 struct md_rdev *rdev;
6239 if (list_empty(&mddev->disks))
6242 pr_info("md: running: ");
6244 rdev_for_each(rdev, mddev) {
6245 char b[BDEVNAME_SIZE];
6246 pr_cont("<%s>", bdevname(rdev->bdev,b));
6250 err = do_md_run(mddev);
6252 pr_warn("md: do_md_run() returned %d\n", err);
6253 do_md_stop(mddev, 0, NULL);
6258 * lets try to run arrays based on all disks that have arrived
6259 * until now. (those are in pending_raid_disks)
6261 * the method: pick the first pending disk, collect all disks with
6262 * the same UUID, remove all from the pending list and put them into
6263 * the 'same_array' list. Then order this list based on superblock
6264 * update time (freshest comes first), kick out 'old' disks and
6265 * compare superblocks. If everything's fine then run it.
6267 * If "unit" is allocated, then bump its reference count
6269 static void autorun_devices(int part)
6271 struct md_rdev *rdev0, *rdev, *tmp;
6272 struct mddev *mddev;
6273 char b[BDEVNAME_SIZE];
6275 pr_info("md: autorun ...\n");
6276 while (!list_empty(&pending_raid_disks)) {
6279 LIST_HEAD(candidates);
6280 rdev0 = list_entry(pending_raid_disks.next,
6281 struct md_rdev, same_set);
6283 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6284 INIT_LIST_HEAD(&candidates);
6285 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6286 if (super_90_load(rdev, rdev0, 0) >= 0) {
6287 pr_debug("md: adding %s ...\n",
6288 bdevname(rdev->bdev,b));
6289 list_move(&rdev->same_set, &candidates);
6292 * now we have a set of devices, with all of them having
6293 * mostly sane superblocks. It's time to allocate the
6297 dev = MKDEV(mdp_major,
6298 rdev0->preferred_minor << MdpMinorShift);
6299 unit = MINOR(dev) >> MdpMinorShift;
6301 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6304 if (rdev0->preferred_minor != unit) {
6305 pr_warn("md: unit number in %s is bad: %d\n",
6306 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6310 md_probe(dev, NULL, NULL);
6311 mddev = mddev_find(dev);
6312 if (!mddev || !mddev->gendisk) {
6317 if (mddev_lock(mddev))
6318 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6319 else if (mddev->raid_disks || mddev->major_version
6320 || !list_empty(&mddev->disks)) {
6321 pr_warn("md: %s already running, cannot run %s\n",
6322 mdname(mddev), bdevname(rdev0->bdev,b));
6323 mddev_unlock(mddev);
6325 pr_debug("md: created %s\n", mdname(mddev));
6326 mddev->persistent = 1;
6327 rdev_for_each_list(rdev, tmp, &candidates) {
6328 list_del_init(&rdev->same_set);
6329 if (bind_rdev_to_array(rdev, mddev))
6332 autorun_array(mddev);
6333 mddev_unlock(mddev);
6335 /* on success, candidates will be empty, on error
6338 rdev_for_each_list(rdev, tmp, &candidates) {
6339 list_del_init(&rdev->same_set);
6344 pr_info("md: ... autorun DONE.\n");
6346 #endif /* !MODULE */
6348 static int get_version(void __user *arg)
6352 ver.major = MD_MAJOR_VERSION;
6353 ver.minor = MD_MINOR_VERSION;
6354 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6356 if (copy_to_user(arg, &ver, sizeof(ver)))
6362 static int get_array_info(struct mddev *mddev, void __user *arg)
6364 mdu_array_info_t info;
6365 int nr,working,insync,failed,spare;
6366 struct md_rdev *rdev;
6368 nr = working = insync = failed = spare = 0;
6370 rdev_for_each_rcu(rdev, mddev) {
6372 if (test_bit(Faulty, &rdev->flags))
6376 if (test_bit(In_sync, &rdev->flags))
6378 else if (test_bit(Journal, &rdev->flags))
6379 /* TODO: add journal count to md_u.h */
6387 info.major_version = mddev->major_version;
6388 info.minor_version = mddev->minor_version;
6389 info.patch_version = MD_PATCHLEVEL_VERSION;
6390 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6391 info.level = mddev->level;
6392 info.size = mddev->dev_sectors / 2;
6393 if (info.size != mddev->dev_sectors / 2) /* overflow */
6396 info.raid_disks = mddev->raid_disks;
6397 info.md_minor = mddev->md_minor;
6398 info.not_persistent= !mddev->persistent;
6400 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6403 info.state = (1<<MD_SB_CLEAN);
6404 if (mddev->bitmap && mddev->bitmap_info.offset)
6405 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6406 if (mddev_is_clustered(mddev))
6407 info.state |= (1<<MD_SB_CLUSTERED);
6408 info.active_disks = insync;
6409 info.working_disks = working;
6410 info.failed_disks = failed;
6411 info.spare_disks = spare;
6413 info.layout = mddev->layout;
6414 info.chunk_size = mddev->chunk_sectors << 9;
6416 if (copy_to_user(arg, &info, sizeof(info)))
6422 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6424 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6428 file = kzalloc(sizeof(*file), GFP_NOIO);
6433 spin_lock(&mddev->lock);
6434 /* bitmap enabled */
6435 if (mddev->bitmap_info.file) {
6436 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6437 sizeof(file->pathname));
6441 memmove(file->pathname, ptr,
6442 sizeof(file->pathname)-(ptr-file->pathname));
6444 spin_unlock(&mddev->lock);
6447 copy_to_user(arg, file, sizeof(*file)))
6454 static int get_disk_info(struct mddev *mddev, void __user * arg)
6456 mdu_disk_info_t info;
6457 struct md_rdev *rdev;
6459 if (copy_from_user(&info, arg, sizeof(info)))
6463 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6465 info.major = MAJOR(rdev->bdev->bd_dev);
6466 info.minor = MINOR(rdev->bdev->bd_dev);
6467 info.raid_disk = rdev->raid_disk;
6469 if (test_bit(Faulty, &rdev->flags))
6470 info.state |= (1<<MD_DISK_FAULTY);
6471 else if (test_bit(In_sync, &rdev->flags)) {
6472 info.state |= (1<<MD_DISK_ACTIVE);
6473 info.state |= (1<<MD_DISK_SYNC);
6475 if (test_bit(Journal, &rdev->flags))
6476 info.state |= (1<<MD_DISK_JOURNAL);
6477 if (test_bit(WriteMostly, &rdev->flags))
6478 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6479 if (test_bit(FailFast, &rdev->flags))
6480 info.state |= (1<<MD_DISK_FAILFAST);
6482 info.major = info.minor = 0;
6483 info.raid_disk = -1;
6484 info.state = (1<<MD_DISK_REMOVED);
6488 if (copy_to_user(arg, &info, sizeof(info)))
6494 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6496 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6497 struct md_rdev *rdev;
6498 dev_t dev = MKDEV(info->major,info->minor);
6500 if (mddev_is_clustered(mddev) &&
6501 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6502 pr_warn("%s: Cannot add to clustered mddev.\n",
6507 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6510 if (!mddev->raid_disks) {
6512 /* expecting a device which has a superblock */
6513 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6515 pr_warn("md: md_import_device returned %ld\n",
6517 return PTR_ERR(rdev);
6519 if (!list_empty(&mddev->disks)) {
6520 struct md_rdev *rdev0
6521 = list_entry(mddev->disks.next,
6522 struct md_rdev, same_set);
6523 err = super_types[mddev->major_version]
6524 .load_super(rdev, rdev0, mddev->minor_version);
6526 pr_warn("md: %s has different UUID to %s\n",
6527 bdevname(rdev->bdev,b),
6528 bdevname(rdev0->bdev,b2));
6533 err = bind_rdev_to_array(rdev, mddev);
6540 * add_new_disk can be used once the array is assembled
6541 * to add "hot spares". They must already have a superblock
6546 if (!mddev->pers->hot_add_disk) {
6547 pr_warn("%s: personality does not support diskops!\n",
6551 if (mddev->persistent)
6552 rdev = md_import_device(dev, mddev->major_version,
6553 mddev->minor_version);
6555 rdev = md_import_device(dev, -1, -1);
6557 pr_warn("md: md_import_device returned %ld\n",
6559 return PTR_ERR(rdev);
6561 /* set saved_raid_disk if appropriate */
6562 if (!mddev->persistent) {
6563 if (info->state & (1<<MD_DISK_SYNC) &&
6564 info->raid_disk < mddev->raid_disks) {
6565 rdev->raid_disk = info->raid_disk;
6566 set_bit(In_sync, &rdev->flags);
6567 clear_bit(Bitmap_sync, &rdev->flags);
6569 rdev->raid_disk = -1;
6570 rdev->saved_raid_disk = rdev->raid_disk;
6572 super_types[mddev->major_version].
6573 validate_super(mddev, rdev);
6574 if ((info->state & (1<<MD_DISK_SYNC)) &&
6575 rdev->raid_disk != info->raid_disk) {
6576 /* This was a hot-add request, but events doesn't
6577 * match, so reject it.
6583 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6584 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6585 set_bit(WriteMostly, &rdev->flags);
6587 clear_bit(WriteMostly, &rdev->flags);
6588 if (info->state & (1<<MD_DISK_FAILFAST))
6589 set_bit(FailFast, &rdev->flags);
6591 clear_bit(FailFast, &rdev->flags);
6593 if (info->state & (1<<MD_DISK_JOURNAL)) {
6594 struct md_rdev *rdev2;
6595 bool has_journal = false;
6597 /* make sure no existing journal disk */
6598 rdev_for_each(rdev2, mddev) {
6599 if (test_bit(Journal, &rdev2->flags)) {
6604 if (has_journal || mddev->bitmap) {
6608 set_bit(Journal, &rdev->flags);
6611 * check whether the device shows up in other nodes
6613 if (mddev_is_clustered(mddev)) {
6614 if (info->state & (1 << MD_DISK_CANDIDATE))
6615 set_bit(Candidate, &rdev->flags);
6616 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6617 /* --add initiated by this node */
6618 err = md_cluster_ops->add_new_disk(mddev, rdev);
6626 rdev->raid_disk = -1;
6627 err = bind_rdev_to_array(rdev, mddev);
6632 if (mddev_is_clustered(mddev)) {
6633 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6635 err = md_cluster_ops->new_disk_ack(mddev,
6638 md_kick_rdev_from_array(rdev);
6642 md_cluster_ops->add_new_disk_cancel(mddev);
6644 err = add_bound_rdev(rdev);
6648 err = add_bound_rdev(rdev);
6653 /* otherwise, add_new_disk is only allowed
6654 * for major_version==0 superblocks
6656 if (mddev->major_version != 0) {
6657 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6661 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6663 rdev = md_import_device(dev, -1, 0);
6665 pr_warn("md: error, md_import_device() returned %ld\n",
6667 return PTR_ERR(rdev);
6669 rdev->desc_nr = info->number;
6670 if (info->raid_disk < mddev->raid_disks)
6671 rdev->raid_disk = info->raid_disk;
6673 rdev->raid_disk = -1;
6675 if (rdev->raid_disk < mddev->raid_disks)
6676 if (info->state & (1<<MD_DISK_SYNC))
6677 set_bit(In_sync, &rdev->flags);
6679 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6680 set_bit(WriteMostly, &rdev->flags);
6681 if (info->state & (1<<MD_DISK_FAILFAST))
6682 set_bit(FailFast, &rdev->flags);
6684 if (!mddev->persistent) {
6685 pr_debug("md: nonpersistent superblock ...\n");
6686 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6688 rdev->sb_start = calc_dev_sboffset(rdev);
6689 rdev->sectors = rdev->sb_start;
6691 err = bind_rdev_to_array(rdev, mddev);
6701 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6703 char b[BDEVNAME_SIZE];
6704 struct md_rdev *rdev;
6709 rdev = find_rdev(mddev, dev);
6713 if (rdev->raid_disk < 0)
6716 clear_bit(Blocked, &rdev->flags);
6717 remove_and_add_spares(mddev, rdev);
6719 if (rdev->raid_disk >= 0)
6723 if (mddev_is_clustered(mddev))
6724 md_cluster_ops->remove_disk(mddev, rdev);
6726 md_kick_rdev_from_array(rdev);
6727 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6729 md_wakeup_thread(mddev->thread);
6731 md_update_sb(mddev, 1);
6732 md_new_event(mddev);
6736 pr_debug("md: cannot remove active disk %s from %s ...\n",
6737 bdevname(rdev->bdev,b), mdname(mddev));
6741 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6743 char b[BDEVNAME_SIZE];
6745 struct md_rdev *rdev;
6750 if (mddev->major_version != 0) {
6751 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6755 if (!mddev->pers->hot_add_disk) {
6756 pr_warn("%s: personality does not support diskops!\n",
6761 rdev = md_import_device(dev, -1, 0);
6763 pr_warn("md: error, md_import_device() returned %ld\n",
6768 if (mddev->persistent)
6769 rdev->sb_start = calc_dev_sboffset(rdev);
6771 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6773 rdev->sectors = rdev->sb_start;
6775 if (test_bit(Faulty, &rdev->flags)) {
6776 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6777 bdevname(rdev->bdev,b), mdname(mddev));
6782 clear_bit(In_sync, &rdev->flags);
6784 rdev->saved_raid_disk = -1;
6785 err = bind_rdev_to_array(rdev, mddev);
6790 * The rest should better be atomic, we can have disk failures
6791 * noticed in interrupt contexts ...
6794 rdev->raid_disk = -1;
6796 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6798 md_update_sb(mddev, 1);
6800 * Kick recovery, maybe this spare has to be added to the
6801 * array immediately.
6803 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6804 md_wakeup_thread(mddev->thread);
6805 md_new_event(mddev);
6813 static int set_bitmap_file(struct mddev *mddev, int fd)
6818 if (!mddev->pers->quiesce || !mddev->thread)
6820 if (mddev->recovery || mddev->sync_thread)
6822 /* we should be able to change the bitmap.. */
6826 struct inode *inode;
6829 if (mddev->bitmap || mddev->bitmap_info.file)
6830 return -EEXIST; /* cannot add when bitmap is present */
6834 pr_warn("%s: error: failed to get bitmap file\n",
6839 inode = f->f_mapping->host;
6840 if (!S_ISREG(inode->i_mode)) {
6841 pr_warn("%s: error: bitmap file must be a regular file\n",
6844 } else if (!(f->f_mode & FMODE_WRITE)) {
6845 pr_warn("%s: error: bitmap file must open for write\n",
6848 } else if (atomic_read(&inode->i_writecount) != 1) {
6849 pr_warn("%s: error: bitmap file is already in use\n",
6857 mddev->bitmap_info.file = f;
6858 mddev->bitmap_info.offset = 0; /* file overrides offset */
6859 } else if (mddev->bitmap == NULL)
6860 return -ENOENT; /* cannot remove what isn't there */
6864 struct bitmap *bitmap;
6866 bitmap = md_bitmap_create(mddev, -1);
6867 mddev_suspend(mddev);
6868 if (!IS_ERR(bitmap)) {
6869 mddev->bitmap = bitmap;
6870 err = md_bitmap_load(mddev);
6872 err = PTR_ERR(bitmap);
6874 md_bitmap_destroy(mddev);
6877 mddev_resume(mddev);
6878 } else if (fd < 0) {
6879 mddev_suspend(mddev);
6880 md_bitmap_destroy(mddev);
6881 mddev_resume(mddev);
6885 struct file *f = mddev->bitmap_info.file;
6887 spin_lock(&mddev->lock);
6888 mddev->bitmap_info.file = NULL;
6889 spin_unlock(&mddev->lock);
6898 * set_array_info is used two different ways
6899 * The original usage is when creating a new array.
6900 * In this usage, raid_disks is > 0 and it together with
6901 * level, size, not_persistent,layout,chunksize determine the
6902 * shape of the array.
6903 * This will always create an array with a type-0.90.0 superblock.
6904 * The newer usage is when assembling an array.
6905 * In this case raid_disks will be 0, and the major_version field is
6906 * use to determine which style super-blocks are to be found on the devices.
6907 * The minor and patch _version numbers are also kept incase the
6908 * super_block handler wishes to interpret them.
6910 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6913 if (info->raid_disks == 0) {
6914 /* just setting version number for superblock loading */
6915 if (info->major_version < 0 ||
6916 info->major_version >= ARRAY_SIZE(super_types) ||
6917 super_types[info->major_version].name == NULL) {
6918 /* maybe try to auto-load a module? */
6919 pr_warn("md: superblock version %d not known\n",
6920 info->major_version);
6923 mddev->major_version = info->major_version;
6924 mddev->minor_version = info->minor_version;
6925 mddev->patch_version = info->patch_version;
6926 mddev->persistent = !info->not_persistent;
6927 /* ensure mddev_put doesn't delete this now that there
6928 * is some minimal configuration.
6930 mddev->ctime = ktime_get_real_seconds();
6933 mddev->major_version = MD_MAJOR_VERSION;
6934 mddev->minor_version = MD_MINOR_VERSION;
6935 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6936 mddev->ctime = ktime_get_real_seconds();
6938 mddev->level = info->level;
6939 mddev->clevel[0] = 0;
6940 mddev->dev_sectors = 2 * (sector_t)info->size;
6941 mddev->raid_disks = info->raid_disks;
6942 /* don't set md_minor, it is determined by which /dev/md* was
6945 if (info->state & (1<<MD_SB_CLEAN))
6946 mddev->recovery_cp = MaxSector;
6948 mddev->recovery_cp = 0;
6949 mddev->persistent = ! info->not_persistent;
6950 mddev->external = 0;
6952 mddev->layout = info->layout;
6953 if (mddev->level == 0)
6954 /* Cannot trust RAID0 layout info here */
6956 mddev->chunk_sectors = info->chunk_size >> 9;
6958 if (mddev->persistent) {
6959 mddev->max_disks = MD_SB_DISKS;
6961 mddev->sb_flags = 0;
6963 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6965 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6966 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6967 mddev->bitmap_info.offset = 0;
6969 mddev->reshape_position = MaxSector;
6972 * Generate a 128 bit UUID
6974 get_random_bytes(mddev->uuid, 16);
6976 mddev->new_level = mddev->level;
6977 mddev->new_chunk_sectors = mddev->chunk_sectors;
6978 mddev->new_layout = mddev->layout;
6979 mddev->delta_disks = 0;
6980 mddev->reshape_backwards = 0;
6985 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6987 lockdep_assert_held(&mddev->reconfig_mutex);
6989 if (mddev->external_size)
6992 mddev->array_sectors = array_sectors;
6994 EXPORT_SYMBOL(md_set_array_sectors);
6996 static int update_size(struct mddev *mddev, sector_t num_sectors)
6998 struct md_rdev *rdev;
7000 int fit = (num_sectors == 0);
7001 sector_t old_dev_sectors = mddev->dev_sectors;
7003 if (mddev->pers->resize == NULL)
7005 /* The "num_sectors" is the number of sectors of each device that
7006 * is used. This can only make sense for arrays with redundancy.
7007 * linear and raid0 always use whatever space is available. We can only
7008 * consider changing this number if no resync or reconstruction is
7009 * happening, and if the new size is acceptable. It must fit before the
7010 * sb_start or, if that is <data_offset, it must fit before the size
7011 * of each device. If num_sectors is zero, we find the largest size
7014 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7020 rdev_for_each(rdev, mddev) {
7021 sector_t avail = rdev->sectors;
7023 if (fit && (num_sectors == 0 || num_sectors > avail))
7024 num_sectors = avail;
7025 if (avail < num_sectors)
7028 rv = mddev->pers->resize(mddev, num_sectors);
7030 if (mddev_is_clustered(mddev))
7031 md_cluster_ops->update_size(mddev, old_dev_sectors);
7032 else if (mddev->queue) {
7033 set_capacity(mddev->gendisk, mddev->array_sectors);
7034 revalidate_disk(mddev->gendisk);
7040 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7043 struct md_rdev *rdev;
7044 /* change the number of raid disks */
7045 if (mddev->pers->check_reshape == NULL)
7049 if (raid_disks <= 0 ||
7050 (mddev->max_disks && raid_disks >= mddev->max_disks))
7052 if (mddev->sync_thread ||
7053 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7054 mddev->reshape_position != MaxSector)
7057 rdev_for_each(rdev, mddev) {
7058 if (mddev->raid_disks < raid_disks &&
7059 rdev->data_offset < rdev->new_data_offset)
7061 if (mddev->raid_disks > raid_disks &&
7062 rdev->data_offset > rdev->new_data_offset)
7066 mddev->delta_disks = raid_disks - mddev->raid_disks;
7067 if (mddev->delta_disks < 0)
7068 mddev->reshape_backwards = 1;
7069 else if (mddev->delta_disks > 0)
7070 mddev->reshape_backwards = 0;
7072 rv = mddev->pers->check_reshape(mddev);
7074 mddev->delta_disks = 0;
7075 mddev->reshape_backwards = 0;
7081 * update_array_info is used to change the configuration of an
7083 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7084 * fields in the info are checked against the array.
7085 * Any differences that cannot be handled will cause an error.
7086 * Normally, only one change can be managed at a time.
7088 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7094 /* calculate expected state,ignoring low bits */
7095 if (mddev->bitmap && mddev->bitmap_info.offset)
7096 state |= (1 << MD_SB_BITMAP_PRESENT);
7098 if (mddev->major_version != info->major_version ||
7099 mddev->minor_version != info->minor_version ||
7100 /* mddev->patch_version != info->patch_version || */
7101 mddev->ctime != info->ctime ||
7102 mddev->level != info->level ||
7103 /* mddev->layout != info->layout || */
7104 mddev->persistent != !info->not_persistent ||
7105 mddev->chunk_sectors != info->chunk_size >> 9 ||
7106 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7107 ((state^info->state) & 0xfffffe00)
7110 /* Check there is only one change */
7111 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7113 if (mddev->raid_disks != info->raid_disks)
7115 if (mddev->layout != info->layout)
7117 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7124 if (mddev->layout != info->layout) {
7126 * we don't need to do anything at the md level, the
7127 * personality will take care of it all.
7129 if (mddev->pers->check_reshape == NULL)
7132 mddev->new_layout = info->layout;
7133 rv = mddev->pers->check_reshape(mddev);
7135 mddev->new_layout = mddev->layout;
7139 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7140 rv = update_size(mddev, (sector_t)info->size * 2);
7142 if (mddev->raid_disks != info->raid_disks)
7143 rv = update_raid_disks(mddev, info->raid_disks);
7145 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7146 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7150 if (mddev->recovery || mddev->sync_thread) {
7154 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7155 struct bitmap *bitmap;
7156 /* add the bitmap */
7157 if (mddev->bitmap) {
7161 if (mddev->bitmap_info.default_offset == 0) {
7165 mddev->bitmap_info.offset =
7166 mddev->bitmap_info.default_offset;
7167 mddev->bitmap_info.space =
7168 mddev->bitmap_info.default_space;
7169 bitmap = md_bitmap_create(mddev, -1);
7170 mddev_suspend(mddev);
7171 if (!IS_ERR(bitmap)) {
7172 mddev->bitmap = bitmap;
7173 rv = md_bitmap_load(mddev);
7175 rv = PTR_ERR(bitmap);
7177 md_bitmap_destroy(mddev);
7178 mddev_resume(mddev);
7180 /* remove the bitmap */
7181 if (!mddev->bitmap) {
7185 if (mddev->bitmap->storage.file) {
7189 if (mddev->bitmap_info.nodes) {
7190 /* hold PW on all the bitmap lock */
7191 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7192 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7194 md_cluster_ops->unlock_all_bitmaps(mddev);
7198 mddev->bitmap_info.nodes = 0;
7199 md_cluster_ops->leave(mddev);
7201 mddev_suspend(mddev);
7202 md_bitmap_destroy(mddev);
7203 mddev_resume(mddev);
7204 mddev->bitmap_info.offset = 0;
7207 md_update_sb(mddev, 1);
7213 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7215 struct md_rdev *rdev;
7218 if (mddev->pers == NULL)
7222 rdev = md_find_rdev_rcu(mddev, dev);
7226 md_error(mddev, rdev);
7227 if (!test_bit(Faulty, &rdev->flags))
7235 * We have a problem here : there is no easy way to give a CHS
7236 * virtual geometry. We currently pretend that we have a 2 heads
7237 * 4 sectors (with a BIG number of cylinders...). This drives
7238 * dosfs just mad... ;-)
7240 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7242 struct mddev *mddev = bdev->bd_disk->private_data;
7246 geo->cylinders = mddev->array_sectors / 8;
7250 static inline bool md_ioctl_valid(unsigned int cmd)
7255 case GET_ARRAY_INFO:
7256 case GET_BITMAP_FILE:
7259 case HOT_REMOVE_DISK:
7262 case RESTART_ARRAY_RW:
7264 case SET_ARRAY_INFO:
7265 case SET_BITMAP_FILE:
7266 case SET_DISK_FAULTY:
7269 case CLUSTERED_DISK_NACK:
7276 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7277 unsigned int cmd, unsigned long arg)
7280 void __user *argp = (void __user *)arg;
7281 struct mddev *mddev = NULL;
7283 bool did_set_md_closing = false;
7285 if (!md_ioctl_valid(cmd))
7290 case GET_ARRAY_INFO:
7294 if (!capable(CAP_SYS_ADMIN))
7299 * Commands dealing with the RAID driver but not any
7304 err = get_version(argp);
7310 autostart_arrays(arg);
7317 * Commands creating/starting a new array:
7320 mddev = bdev->bd_disk->private_data;
7327 /* Some actions do not requires the mutex */
7329 case GET_ARRAY_INFO:
7330 if (!mddev->raid_disks && !mddev->external)
7333 err = get_array_info(mddev, argp);
7337 if (!mddev->raid_disks && !mddev->external)
7340 err = get_disk_info(mddev, argp);
7343 case SET_DISK_FAULTY:
7344 err = set_disk_faulty(mddev, new_decode_dev(arg));
7347 case GET_BITMAP_FILE:
7348 err = get_bitmap_file(mddev, argp);
7353 if (cmd == ADD_NEW_DISK)
7354 /* need to ensure md_delayed_delete() has completed */
7355 flush_workqueue(md_misc_wq);
7357 if (cmd == HOT_REMOVE_DISK)
7358 /* need to ensure recovery thread has run */
7359 wait_event_interruptible_timeout(mddev->sb_wait,
7360 !test_bit(MD_RECOVERY_NEEDED,
7362 msecs_to_jiffies(5000));
7363 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7364 /* Need to flush page cache, and ensure no-one else opens
7367 mutex_lock(&mddev->open_mutex);
7368 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7369 mutex_unlock(&mddev->open_mutex);
7373 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7374 set_bit(MD_CLOSING, &mddev->flags);
7375 did_set_md_closing = true;
7376 mutex_unlock(&mddev->open_mutex);
7377 sync_blockdev(bdev);
7379 err = mddev_lock(mddev);
7381 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7386 if (cmd == SET_ARRAY_INFO) {
7387 mdu_array_info_t info;
7389 memset(&info, 0, sizeof(info));
7390 else if (copy_from_user(&info, argp, sizeof(info))) {
7395 err = update_array_info(mddev, &info);
7397 pr_warn("md: couldn't update array info. %d\n", err);
7402 if (!list_empty(&mddev->disks)) {
7403 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7407 if (mddev->raid_disks) {
7408 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7412 err = set_array_info(mddev, &info);
7414 pr_warn("md: couldn't set array info. %d\n", err);
7421 * Commands querying/configuring an existing array:
7423 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7424 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7425 if ((!mddev->raid_disks && !mddev->external)
7426 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7427 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7428 && cmd != GET_BITMAP_FILE) {
7434 * Commands even a read-only array can execute:
7437 case RESTART_ARRAY_RW:
7438 err = restart_array(mddev);
7442 err = do_md_stop(mddev, 0, bdev);
7446 err = md_set_readonly(mddev, bdev);
7449 case HOT_REMOVE_DISK:
7450 err = hot_remove_disk(mddev, new_decode_dev(arg));
7454 /* We can support ADD_NEW_DISK on read-only arrays
7455 * only if we are re-adding a preexisting device.
7456 * So require mddev->pers and MD_DISK_SYNC.
7459 mdu_disk_info_t info;
7460 if (copy_from_user(&info, argp, sizeof(info)))
7462 else if (!(info.state & (1<<MD_DISK_SYNC)))
7463 /* Need to clear read-only for this */
7466 err = add_new_disk(mddev, &info);
7472 if (get_user(ro, (int __user *)(arg))) {
7478 /* if the bdev is going readonly the value of mddev->ro
7479 * does not matter, no writes are coming
7484 /* are we are already prepared for writes? */
7488 /* transitioning to readauto need only happen for
7489 * arrays that call md_write_start
7492 err = restart_array(mddev);
7495 set_disk_ro(mddev->gendisk, 0);
7502 * The remaining ioctls are changing the state of the
7503 * superblock, so we do not allow them on read-only arrays.
7505 if (mddev->ro && mddev->pers) {
7506 if (mddev->ro == 2) {
7508 sysfs_notify_dirent_safe(mddev->sysfs_state);
7509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7510 /* mddev_unlock will wake thread */
7511 /* If a device failed while we were read-only, we
7512 * need to make sure the metadata is updated now.
7514 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7515 mddev_unlock(mddev);
7516 wait_event(mddev->sb_wait,
7517 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7518 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7519 mddev_lock_nointr(mddev);
7530 mdu_disk_info_t info;
7531 if (copy_from_user(&info, argp, sizeof(info)))
7534 err = add_new_disk(mddev, &info);
7538 case CLUSTERED_DISK_NACK:
7539 if (mddev_is_clustered(mddev))
7540 md_cluster_ops->new_disk_ack(mddev, false);
7546 err = hot_add_disk(mddev, new_decode_dev(arg));
7550 err = do_md_run(mddev);
7553 case SET_BITMAP_FILE:
7554 err = set_bitmap_file(mddev, (int)arg);
7563 if (mddev->hold_active == UNTIL_IOCTL &&
7565 mddev->hold_active = 0;
7566 mddev_unlock(mddev);
7568 if(did_set_md_closing)
7569 clear_bit(MD_CLOSING, &mddev->flags);
7572 #ifdef CONFIG_COMPAT
7573 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7574 unsigned int cmd, unsigned long arg)
7577 case HOT_REMOVE_DISK:
7579 case SET_DISK_FAULTY:
7580 case SET_BITMAP_FILE:
7581 /* These take in integer arg, do not convert */
7584 arg = (unsigned long)compat_ptr(arg);
7588 return md_ioctl(bdev, mode, cmd, arg);
7590 #endif /* CONFIG_COMPAT */
7592 static int md_open(struct block_device *bdev, fmode_t mode)
7595 * Succeed if we can lock the mddev, which confirms that
7596 * it isn't being stopped right now.
7598 struct mddev *mddev = mddev_find(bdev->bd_dev);
7604 if (mddev->gendisk != bdev->bd_disk) {
7605 /* we are racing with mddev_put which is discarding this
7609 /* Wait until bdev->bd_disk is definitely gone */
7610 flush_workqueue(md_misc_wq);
7611 /* Then retry the open from the top */
7612 return -ERESTARTSYS;
7614 BUG_ON(mddev != bdev->bd_disk->private_data);
7616 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7619 if (test_bit(MD_CLOSING, &mddev->flags)) {
7620 mutex_unlock(&mddev->open_mutex);
7626 atomic_inc(&mddev->openers);
7627 mutex_unlock(&mddev->open_mutex);
7629 check_disk_change(bdev);
7636 static void md_release(struct gendisk *disk, fmode_t mode)
7638 struct mddev *mddev = disk->private_data;
7641 atomic_dec(&mddev->openers);
7645 static int md_media_changed(struct gendisk *disk)
7647 struct mddev *mddev = disk->private_data;
7649 return mddev->changed;
7652 static int md_revalidate(struct gendisk *disk)
7654 struct mddev *mddev = disk->private_data;
7659 static const struct block_device_operations md_fops =
7661 .owner = THIS_MODULE,
7663 .release = md_release,
7665 #ifdef CONFIG_COMPAT
7666 .compat_ioctl = md_compat_ioctl,
7668 .getgeo = md_getgeo,
7669 .media_changed = md_media_changed,
7670 .revalidate_disk= md_revalidate,
7673 static int md_thread(void *arg)
7675 struct md_thread *thread = arg;
7678 * md_thread is a 'system-thread', it's priority should be very
7679 * high. We avoid resource deadlocks individually in each
7680 * raid personality. (RAID5 does preallocation) We also use RR and
7681 * the very same RT priority as kswapd, thus we will never get
7682 * into a priority inversion deadlock.
7684 * we definitely have to have equal or higher priority than
7685 * bdflush, otherwise bdflush will deadlock if there are too
7686 * many dirty RAID5 blocks.
7689 allow_signal(SIGKILL);
7690 while (!kthread_should_stop()) {
7692 /* We need to wait INTERRUPTIBLE so that
7693 * we don't add to the load-average.
7694 * That means we need to be sure no signals are
7697 if (signal_pending(current))
7698 flush_signals(current);
7700 wait_event_interruptible_timeout
7702 test_bit(THREAD_WAKEUP, &thread->flags)
7703 || kthread_should_stop() || kthread_should_park(),
7706 clear_bit(THREAD_WAKEUP, &thread->flags);
7707 if (kthread_should_park())
7709 if (!kthread_should_stop())
7710 thread->run(thread);
7716 void md_wakeup_thread(struct md_thread *thread)
7719 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7720 set_bit(THREAD_WAKEUP, &thread->flags);
7721 wake_up(&thread->wqueue);
7724 EXPORT_SYMBOL(md_wakeup_thread);
7726 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7727 struct mddev *mddev, const char *name)
7729 struct md_thread *thread;
7731 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7735 init_waitqueue_head(&thread->wqueue);
7738 thread->mddev = mddev;
7739 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7740 thread->tsk = kthread_run(md_thread, thread,
7742 mdname(thread->mddev),
7744 if (IS_ERR(thread->tsk)) {
7750 EXPORT_SYMBOL(md_register_thread);
7752 void md_unregister_thread(struct md_thread **threadp)
7754 struct md_thread *thread = *threadp;
7757 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7758 /* Locking ensures that mddev_unlock does not wake_up a
7759 * non-existent thread
7761 spin_lock(&pers_lock);
7763 spin_unlock(&pers_lock);
7765 kthread_stop(thread->tsk);
7768 EXPORT_SYMBOL(md_unregister_thread);
7770 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7772 if (!rdev || test_bit(Faulty, &rdev->flags))
7775 if (!mddev->pers || !mddev->pers->error_handler)
7777 mddev->pers->error_handler(mddev,rdev);
7778 if (mddev->degraded)
7779 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7780 sysfs_notify_dirent_safe(rdev->sysfs_state);
7781 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7782 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7783 md_wakeup_thread(mddev->thread);
7784 if (mddev->event_work.func)
7785 queue_work(md_misc_wq, &mddev->event_work);
7786 md_new_event(mddev);
7788 EXPORT_SYMBOL(md_error);
7790 /* seq_file implementation /proc/mdstat */
7792 static void status_unused(struct seq_file *seq)
7795 struct md_rdev *rdev;
7797 seq_printf(seq, "unused devices: ");
7799 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7800 char b[BDEVNAME_SIZE];
7802 seq_printf(seq, "%s ",
7803 bdevname(rdev->bdev,b));
7806 seq_printf(seq, "<none>");
7808 seq_printf(seq, "\n");
7811 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7813 sector_t max_sectors, resync, res;
7814 unsigned long dt, db = 0;
7815 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7816 int scale, recovery_active;
7817 unsigned int per_milli;
7819 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7820 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7821 max_sectors = mddev->resync_max_sectors;
7823 max_sectors = mddev->dev_sectors;
7825 resync = mddev->curr_resync;
7827 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7828 /* Still cleaning up */
7829 resync = max_sectors;
7830 } else if (resync > max_sectors)
7831 resync = max_sectors;
7833 resync -= atomic_read(&mddev->recovery_active);
7836 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
7837 struct md_rdev *rdev;
7839 rdev_for_each(rdev, mddev)
7840 if (rdev->raid_disk >= 0 &&
7841 !test_bit(Faulty, &rdev->flags) &&
7842 rdev->recovery_offset != MaxSector &&
7843 rdev->recovery_offset) {
7844 seq_printf(seq, "\trecover=REMOTE");
7847 if (mddev->reshape_position != MaxSector)
7848 seq_printf(seq, "\treshape=REMOTE");
7850 seq_printf(seq, "\tresync=REMOTE");
7853 if (mddev->recovery_cp < MaxSector) {
7854 seq_printf(seq, "\tresync=PENDING");
7860 seq_printf(seq, "\tresync=DELAYED");
7864 WARN_ON(max_sectors == 0);
7865 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7866 * in a sector_t, and (max_sectors>>scale) will fit in a
7867 * u32, as those are the requirements for sector_div.
7868 * Thus 'scale' must be at least 10
7871 if (sizeof(sector_t) > sizeof(unsigned long)) {
7872 while ( max_sectors/2 > (1ULL<<(scale+32)))
7875 res = (resync>>scale)*1000;
7876 sector_div(res, (u32)((max_sectors>>scale)+1));
7880 int i, x = per_milli/50, y = 20-x;
7881 seq_printf(seq, "[");
7882 for (i = 0; i < x; i++)
7883 seq_printf(seq, "=");
7884 seq_printf(seq, ">");
7885 for (i = 0; i < y; i++)
7886 seq_printf(seq, ".");
7887 seq_printf(seq, "] ");
7889 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7890 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7892 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7894 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7895 "resync" : "recovery"))),
7896 per_milli/10, per_milli % 10,
7897 (unsigned long long) resync/2,
7898 (unsigned long long) max_sectors/2);
7901 * dt: time from mark until now
7902 * db: blocks written from mark until now
7903 * rt: remaining time
7905 * rt is a sector_t, which is always 64bit now. We are keeping
7906 * the original algorithm, but it is not really necessary.
7908 * Original algorithm:
7909 * So we divide before multiply in case it is 32bit and close
7911 * We scale the divisor (db) by 32 to avoid losing precision
7912 * near the end of resync when the number of remaining sectors
7914 * We then divide rt by 32 after multiplying by db to compensate.
7915 * The '+1' avoids division by zero if db is very small.
7917 dt = ((jiffies - mddev->resync_mark) / HZ);
7920 curr_mark_cnt = mddev->curr_mark_cnt;
7921 recovery_active = atomic_read(&mddev->recovery_active);
7922 resync_mark_cnt = mddev->resync_mark_cnt;
7924 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
7925 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
7927 rt = max_sectors - resync; /* number of remaining sectors */
7928 rt = div64_u64(rt, db/32+1);
7932 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7933 ((unsigned long)rt % 60)/6);
7935 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7939 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7941 struct list_head *tmp;
7943 struct mddev *mddev;
7951 spin_lock(&all_mddevs_lock);
7952 list_for_each(tmp,&all_mddevs)
7954 mddev = list_entry(tmp, struct mddev, all_mddevs);
7956 spin_unlock(&all_mddevs_lock);
7959 spin_unlock(&all_mddevs_lock);
7961 return (void*)2;/* tail */
7965 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7967 struct list_head *tmp;
7968 struct mddev *next_mddev, *mddev = v;
7974 spin_lock(&all_mddevs_lock);
7976 tmp = all_mddevs.next;
7978 tmp = mddev->all_mddevs.next;
7979 if (tmp != &all_mddevs)
7980 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7982 next_mddev = (void*)2;
7985 spin_unlock(&all_mddevs_lock);
7993 static void md_seq_stop(struct seq_file *seq, void *v)
7995 struct mddev *mddev = v;
7997 if (mddev && v != (void*)1 && v != (void*)2)
8001 static int md_seq_show(struct seq_file *seq, void *v)
8003 struct mddev *mddev = v;
8005 struct md_rdev *rdev;
8007 if (v == (void*)1) {
8008 struct md_personality *pers;
8009 seq_printf(seq, "Personalities : ");
8010 spin_lock(&pers_lock);
8011 list_for_each_entry(pers, &pers_list, list)
8012 seq_printf(seq, "[%s] ", pers->name);
8014 spin_unlock(&pers_lock);
8015 seq_printf(seq, "\n");
8016 seq->poll_event = atomic_read(&md_event_count);
8019 if (v == (void*)2) {
8024 spin_lock(&mddev->lock);
8025 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8026 seq_printf(seq, "%s : %sactive", mdname(mddev),
8027 mddev->pers ? "" : "in");
8030 seq_printf(seq, " (read-only)");
8032 seq_printf(seq, " (auto-read-only)");
8033 seq_printf(seq, " %s", mddev->pers->name);
8038 rdev_for_each_rcu(rdev, mddev) {
8039 char b[BDEVNAME_SIZE];
8040 seq_printf(seq, " %s[%d]",
8041 bdevname(rdev->bdev,b), rdev->desc_nr);
8042 if (test_bit(WriteMostly, &rdev->flags))
8043 seq_printf(seq, "(W)");
8044 if (test_bit(Journal, &rdev->flags))
8045 seq_printf(seq, "(J)");
8046 if (test_bit(Faulty, &rdev->flags)) {
8047 seq_printf(seq, "(F)");
8050 if (rdev->raid_disk < 0)
8051 seq_printf(seq, "(S)"); /* spare */
8052 if (test_bit(Replacement, &rdev->flags))
8053 seq_printf(seq, "(R)");
8054 sectors += rdev->sectors;
8058 if (!list_empty(&mddev->disks)) {
8060 seq_printf(seq, "\n %llu blocks",
8061 (unsigned long long)
8062 mddev->array_sectors / 2);
8064 seq_printf(seq, "\n %llu blocks",
8065 (unsigned long long)sectors / 2);
8067 if (mddev->persistent) {
8068 if (mddev->major_version != 0 ||
8069 mddev->minor_version != 90) {
8070 seq_printf(seq," super %d.%d",
8071 mddev->major_version,
8072 mddev->minor_version);
8074 } else if (mddev->external)
8075 seq_printf(seq, " super external:%s",
8076 mddev->metadata_type);
8078 seq_printf(seq, " super non-persistent");
8081 mddev->pers->status(seq, mddev);
8082 seq_printf(seq, "\n ");
8083 if (mddev->pers->sync_request) {
8084 if (status_resync(seq, mddev))
8085 seq_printf(seq, "\n ");
8088 seq_printf(seq, "\n ");
8090 md_bitmap_status(seq, mddev->bitmap);
8092 seq_printf(seq, "\n");
8094 spin_unlock(&mddev->lock);
8099 static const struct seq_operations md_seq_ops = {
8100 .start = md_seq_start,
8101 .next = md_seq_next,
8102 .stop = md_seq_stop,
8103 .show = md_seq_show,
8106 static int md_seq_open(struct inode *inode, struct file *file)
8108 struct seq_file *seq;
8111 error = seq_open(file, &md_seq_ops);
8115 seq = file->private_data;
8116 seq->poll_event = atomic_read(&md_event_count);
8120 static int md_unloading;
8121 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8123 struct seq_file *seq = filp->private_data;
8127 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8128 poll_wait(filp, &md_event_waiters, wait);
8130 /* always allow read */
8131 mask = EPOLLIN | EPOLLRDNORM;
8133 if (seq->poll_event != atomic_read(&md_event_count))
8134 mask |= EPOLLERR | EPOLLPRI;
8138 static const struct file_operations md_seq_fops = {
8139 .owner = THIS_MODULE,
8140 .open = md_seq_open,
8142 .llseek = seq_lseek,
8143 .release = seq_release,
8144 .poll = mdstat_poll,
8147 int register_md_personality(struct md_personality *p)
8149 pr_debug("md: %s personality registered for level %d\n",
8151 spin_lock(&pers_lock);
8152 list_add_tail(&p->list, &pers_list);
8153 spin_unlock(&pers_lock);
8156 EXPORT_SYMBOL(register_md_personality);
8158 int unregister_md_personality(struct md_personality *p)
8160 pr_debug("md: %s personality unregistered\n", p->name);
8161 spin_lock(&pers_lock);
8162 list_del_init(&p->list);
8163 spin_unlock(&pers_lock);
8166 EXPORT_SYMBOL(unregister_md_personality);
8168 int register_md_cluster_operations(struct md_cluster_operations *ops,
8169 struct module *module)
8172 spin_lock(&pers_lock);
8173 if (md_cluster_ops != NULL)
8176 md_cluster_ops = ops;
8177 md_cluster_mod = module;
8179 spin_unlock(&pers_lock);
8182 EXPORT_SYMBOL(register_md_cluster_operations);
8184 int unregister_md_cluster_operations(void)
8186 spin_lock(&pers_lock);
8187 md_cluster_ops = NULL;
8188 spin_unlock(&pers_lock);
8191 EXPORT_SYMBOL(unregister_md_cluster_operations);
8193 int md_setup_cluster(struct mddev *mddev, int nodes)
8195 if (!md_cluster_ops)
8196 request_module("md-cluster");
8197 spin_lock(&pers_lock);
8198 /* ensure module won't be unloaded */
8199 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8200 pr_warn("can't find md-cluster module or get it's reference.\n");
8201 spin_unlock(&pers_lock);
8204 spin_unlock(&pers_lock);
8206 return md_cluster_ops->join(mddev, nodes);
8209 void md_cluster_stop(struct mddev *mddev)
8211 if (!md_cluster_ops)
8213 md_cluster_ops->leave(mddev);
8214 module_put(md_cluster_mod);
8217 static int is_mddev_idle(struct mddev *mddev, int init)
8219 struct md_rdev *rdev;
8225 rdev_for_each_rcu(rdev, mddev) {
8226 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
8227 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8228 atomic_read(&disk->sync_io);
8229 /* sync IO will cause sync_io to increase before the disk_stats
8230 * as sync_io is counted when a request starts, and
8231 * disk_stats is counted when it completes.
8232 * So resync activity will cause curr_events to be smaller than
8233 * when there was no such activity.
8234 * non-sync IO will cause disk_stat to increase without
8235 * increasing sync_io so curr_events will (eventually)
8236 * be larger than it was before. Once it becomes
8237 * substantially larger, the test below will cause
8238 * the array to appear non-idle, and resync will slow
8240 * If there is a lot of outstanding resync activity when
8241 * we set last_event to curr_events, then all that activity
8242 * completing might cause the array to appear non-idle
8243 * and resync will be slowed down even though there might
8244 * not have been non-resync activity. This will only
8245 * happen once though. 'last_events' will soon reflect
8246 * the state where there is little or no outstanding
8247 * resync requests, and further resync activity will
8248 * always make curr_events less than last_events.
8251 if (init || curr_events - rdev->last_events > 64) {
8252 rdev->last_events = curr_events;
8260 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8262 /* another "blocks" (512byte) blocks have been synced */
8263 atomic_sub(blocks, &mddev->recovery_active);
8264 wake_up(&mddev->recovery_wait);
8266 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8267 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8268 md_wakeup_thread(mddev->thread);
8269 // stop recovery, signal do_sync ....
8272 EXPORT_SYMBOL(md_done_sync);
8274 /* md_write_start(mddev, bi)
8275 * If we need to update some array metadata (e.g. 'active' flag
8276 * in superblock) before writing, schedule a superblock update
8277 * and wait for it to complete.
8278 * A return value of 'false' means that the write wasn't recorded
8279 * and cannot proceed as the array is being suspend.
8281 bool md_write_start(struct mddev *mddev, struct bio *bi)
8285 if (bio_data_dir(bi) != WRITE)
8288 BUG_ON(mddev->ro == 1);
8289 if (mddev->ro == 2) {
8290 /* need to switch to read/write */
8292 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8293 md_wakeup_thread(mddev->thread);
8294 md_wakeup_thread(mddev->sync_thread);
8298 percpu_ref_get(&mddev->writes_pending);
8299 smp_mb(); /* Match smp_mb in set_in_sync() */
8300 if (mddev->safemode == 1)
8301 mddev->safemode = 0;
8302 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8303 if (mddev->in_sync || mddev->sync_checkers) {
8304 spin_lock(&mddev->lock);
8305 if (mddev->in_sync) {
8307 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8308 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8309 md_wakeup_thread(mddev->thread);
8312 spin_unlock(&mddev->lock);
8316 sysfs_notify_dirent_safe(mddev->sysfs_state);
8317 if (!mddev->has_superblocks)
8319 wait_event(mddev->sb_wait,
8320 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8322 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8323 percpu_ref_put(&mddev->writes_pending);
8328 EXPORT_SYMBOL(md_write_start);
8330 /* md_write_inc can only be called when md_write_start() has
8331 * already been called at least once of the current request.
8332 * It increments the counter and is useful when a single request
8333 * is split into several parts. Each part causes an increment and
8334 * so needs a matching md_write_end().
8335 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8336 * a spinlocked region.
8338 void md_write_inc(struct mddev *mddev, struct bio *bi)
8340 if (bio_data_dir(bi) != WRITE)
8342 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8343 percpu_ref_get(&mddev->writes_pending);
8345 EXPORT_SYMBOL(md_write_inc);
8347 void md_write_end(struct mddev *mddev)
8349 percpu_ref_put(&mddev->writes_pending);
8351 if (mddev->safemode == 2)
8352 md_wakeup_thread(mddev->thread);
8353 else if (mddev->safemode_delay)
8354 /* The roundup() ensures this only performs locking once
8355 * every ->safemode_delay jiffies
8357 mod_timer(&mddev->safemode_timer,
8358 roundup(jiffies, mddev->safemode_delay) +
8359 mddev->safemode_delay);
8362 EXPORT_SYMBOL(md_write_end);
8364 /* md_allow_write(mddev)
8365 * Calling this ensures that the array is marked 'active' so that writes
8366 * may proceed without blocking. It is important to call this before
8367 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8368 * Must be called with mddev_lock held.
8370 void md_allow_write(struct mddev *mddev)
8376 if (!mddev->pers->sync_request)
8379 spin_lock(&mddev->lock);
8380 if (mddev->in_sync) {
8382 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8383 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8384 if (mddev->safemode_delay &&
8385 mddev->safemode == 0)
8386 mddev->safemode = 1;
8387 spin_unlock(&mddev->lock);
8388 md_update_sb(mddev, 0);
8389 sysfs_notify_dirent_safe(mddev->sysfs_state);
8390 /* wait for the dirty state to be recorded in the metadata */
8391 wait_event(mddev->sb_wait,
8392 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8394 spin_unlock(&mddev->lock);
8396 EXPORT_SYMBOL_GPL(md_allow_write);
8398 #define SYNC_MARKS 10
8399 #define SYNC_MARK_STEP (3*HZ)
8400 #define UPDATE_FREQUENCY (5*60*HZ)
8401 void md_do_sync(struct md_thread *thread)
8403 struct mddev *mddev = thread->mddev;
8404 struct mddev *mddev2;
8405 unsigned int currspeed = 0, window;
8406 sector_t max_sectors,j, io_sectors, recovery_done;
8407 unsigned long mark[SYNC_MARKS];
8408 unsigned long update_time;
8409 sector_t mark_cnt[SYNC_MARKS];
8411 struct list_head *tmp;
8412 sector_t last_check;
8414 struct md_rdev *rdev;
8415 char *desc, *action = NULL;
8416 struct blk_plug plug;
8419 /* just incase thread restarts... */
8420 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8421 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8423 if (mddev->ro) {/* never try to sync a read-only array */
8424 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8428 if (mddev_is_clustered(mddev)) {
8429 ret = md_cluster_ops->resync_start(mddev);
8433 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8434 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8435 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8436 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8437 && ((unsigned long long)mddev->curr_resync_completed
8438 < (unsigned long long)mddev->resync_max_sectors))
8442 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8443 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8444 desc = "data-check";
8446 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8447 desc = "requested-resync";
8451 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8456 mddev->last_sync_action = action ?: desc;
8458 /* we overload curr_resync somewhat here.
8459 * 0 == not engaged in resync at all
8460 * 2 == checking that there is no conflict with another sync
8461 * 1 == like 2, but have yielded to allow conflicting resync to
8463 * other == active in resync - this many blocks
8465 * Before starting a resync we must have set curr_resync to
8466 * 2, and then checked that every "conflicting" array has curr_resync
8467 * less than ours. When we find one that is the same or higher
8468 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8469 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8470 * This will mean we have to start checking from the beginning again.
8475 int mddev2_minor = -1;
8476 mddev->curr_resync = 2;
8479 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8481 for_each_mddev(mddev2, tmp) {
8482 if (mddev2 == mddev)
8484 if (!mddev->parallel_resync
8485 && mddev2->curr_resync
8486 && match_mddev_units(mddev, mddev2)) {
8488 if (mddev < mddev2 && mddev->curr_resync == 2) {
8489 /* arbitrarily yield */
8490 mddev->curr_resync = 1;
8491 wake_up(&resync_wait);
8493 if (mddev > mddev2 && mddev->curr_resync == 1)
8494 /* no need to wait here, we can wait the next
8495 * time 'round when curr_resync == 2
8498 /* We need to wait 'interruptible' so as not to
8499 * contribute to the load average, and not to
8500 * be caught by 'softlockup'
8502 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8503 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8504 mddev2->curr_resync >= mddev->curr_resync) {
8505 if (mddev2_minor != mddev2->md_minor) {
8506 mddev2_minor = mddev2->md_minor;
8507 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8508 desc, mdname(mddev),
8512 if (signal_pending(current))
8513 flush_signals(current);
8515 finish_wait(&resync_wait, &wq);
8518 finish_wait(&resync_wait, &wq);
8521 } while (mddev->curr_resync < 2);
8524 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8525 /* resync follows the size requested by the personality,
8526 * which defaults to physical size, but can be virtual size
8528 max_sectors = mddev->resync_max_sectors;
8529 atomic64_set(&mddev->resync_mismatches, 0);
8530 /* we don't use the checkpoint if there's a bitmap */
8531 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8532 j = mddev->resync_min;
8533 else if (!mddev->bitmap)
8534 j = mddev->recovery_cp;
8536 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8537 max_sectors = mddev->resync_max_sectors;
8539 * If the original node aborts reshaping then we continue the
8540 * reshaping, so set j again to avoid restart reshape from the
8543 if (mddev_is_clustered(mddev) &&
8544 mddev->reshape_position != MaxSector)
8545 j = mddev->reshape_position;
8547 /* recovery follows the physical size of devices */
8548 max_sectors = mddev->dev_sectors;
8551 rdev_for_each_rcu(rdev, mddev)
8552 if (rdev->raid_disk >= 0 &&
8553 !test_bit(Journal, &rdev->flags) &&
8554 !test_bit(Faulty, &rdev->flags) &&
8555 !test_bit(In_sync, &rdev->flags) &&
8556 rdev->recovery_offset < j)
8557 j = rdev->recovery_offset;
8560 /* If there is a bitmap, we need to make sure all
8561 * writes that started before we added a spare
8562 * complete before we start doing a recovery.
8563 * Otherwise the write might complete and (via
8564 * bitmap_endwrite) set a bit in the bitmap after the
8565 * recovery has checked that bit and skipped that
8568 if (mddev->bitmap) {
8569 mddev->pers->quiesce(mddev, 1);
8570 mddev->pers->quiesce(mddev, 0);
8574 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8575 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8576 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8577 speed_max(mddev), desc);
8579 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8582 for (m = 0; m < SYNC_MARKS; m++) {
8584 mark_cnt[m] = io_sectors;
8587 mddev->resync_mark = mark[last_mark];
8588 mddev->resync_mark_cnt = mark_cnt[last_mark];
8591 * Tune reconstruction:
8593 window = 32 * (PAGE_SIZE / 512);
8594 pr_debug("md: using %dk window, over a total of %lluk.\n",
8595 window/2, (unsigned long long)max_sectors/2);
8597 atomic_set(&mddev->recovery_active, 0);
8601 pr_debug("md: resuming %s of %s from checkpoint.\n",
8602 desc, mdname(mddev));
8603 mddev->curr_resync = j;
8605 mddev->curr_resync = 3; /* no longer delayed */
8606 mddev->curr_resync_completed = j;
8607 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8608 md_new_event(mddev);
8609 update_time = jiffies;
8611 blk_start_plug(&plug);
8612 while (j < max_sectors) {
8617 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8618 ((mddev->curr_resync > mddev->curr_resync_completed &&
8619 (mddev->curr_resync - mddev->curr_resync_completed)
8620 > (max_sectors >> 4)) ||
8621 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8622 (j - mddev->curr_resync_completed)*2
8623 >= mddev->resync_max - mddev->curr_resync_completed ||
8624 mddev->curr_resync_completed > mddev->resync_max
8626 /* time to update curr_resync_completed */
8627 wait_event(mddev->recovery_wait,
8628 atomic_read(&mddev->recovery_active) == 0);
8629 mddev->curr_resync_completed = j;
8630 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8631 j > mddev->recovery_cp)
8632 mddev->recovery_cp = j;
8633 update_time = jiffies;
8634 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8635 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8638 while (j >= mddev->resync_max &&
8639 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8640 /* As this condition is controlled by user-space,
8641 * we can block indefinitely, so use '_interruptible'
8642 * to avoid triggering warnings.
8644 flush_signals(current); /* just in case */
8645 wait_event_interruptible(mddev->recovery_wait,
8646 mddev->resync_max > j
8647 || test_bit(MD_RECOVERY_INTR,
8651 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8654 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8656 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8660 if (!skipped) { /* actual IO requested */
8661 io_sectors += sectors;
8662 atomic_add(sectors, &mddev->recovery_active);
8665 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8669 if (j > max_sectors)
8670 /* when skipping, extra large numbers can be returned. */
8673 mddev->curr_resync = j;
8674 mddev->curr_mark_cnt = io_sectors;
8675 if (last_check == 0)
8676 /* this is the earliest that rebuild will be
8677 * visible in /proc/mdstat
8679 md_new_event(mddev);
8681 if (last_check + window > io_sectors || j == max_sectors)
8684 last_check = io_sectors;
8686 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8688 int next = (last_mark+1) % SYNC_MARKS;
8690 mddev->resync_mark = mark[next];
8691 mddev->resync_mark_cnt = mark_cnt[next];
8692 mark[next] = jiffies;
8693 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8697 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8701 * this loop exits only if either when we are slower than
8702 * the 'hard' speed limit, or the system was IO-idle for
8704 * the system might be non-idle CPU-wise, but we only care
8705 * about not overloading the IO subsystem. (things like an
8706 * e2fsck being done on the RAID array should execute fast)
8710 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8711 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8712 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8714 if (currspeed > speed_min(mddev)) {
8715 if (currspeed > speed_max(mddev)) {
8719 if (!is_mddev_idle(mddev, 0)) {
8721 * Give other IO more of a chance.
8722 * The faster the devices, the less we wait.
8724 wait_event(mddev->recovery_wait,
8725 !atomic_read(&mddev->recovery_active));
8729 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8730 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8731 ? "interrupted" : "done");
8733 * this also signals 'finished resyncing' to md_stop
8735 blk_finish_plug(&plug);
8736 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8738 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8739 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8740 mddev->curr_resync > 3) {
8741 mddev->curr_resync_completed = mddev->curr_resync;
8742 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8744 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8746 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8747 mddev->curr_resync > 3) {
8748 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8749 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8750 if (mddev->curr_resync >= mddev->recovery_cp) {
8751 pr_debug("md: checkpointing %s of %s.\n",
8752 desc, mdname(mddev));
8753 if (test_bit(MD_RECOVERY_ERROR,
8755 mddev->recovery_cp =
8756 mddev->curr_resync_completed;
8758 mddev->recovery_cp =
8762 mddev->recovery_cp = MaxSector;
8764 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8765 mddev->curr_resync = MaxSector;
8766 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8767 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8769 rdev_for_each_rcu(rdev, mddev)
8770 if (rdev->raid_disk >= 0 &&
8771 mddev->delta_disks >= 0 &&
8772 !test_bit(Journal, &rdev->flags) &&
8773 !test_bit(Faulty, &rdev->flags) &&
8774 !test_bit(In_sync, &rdev->flags) &&
8775 rdev->recovery_offset < mddev->curr_resync)
8776 rdev->recovery_offset = mddev->curr_resync;
8782 /* set CHANGE_PENDING here since maybe another update is needed,
8783 * so other nodes are informed. It should be harmless for normal
8785 set_mask_bits(&mddev->sb_flags, 0,
8786 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8788 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8789 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8790 mddev->delta_disks > 0 &&
8791 mddev->pers->finish_reshape &&
8792 mddev->pers->size &&
8794 mddev_lock_nointr(mddev);
8795 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8796 mddev_unlock(mddev);
8797 if (!mddev_is_clustered(mddev)) {
8798 set_capacity(mddev->gendisk, mddev->array_sectors);
8799 revalidate_disk(mddev->gendisk);
8803 spin_lock(&mddev->lock);
8804 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8805 /* We completed so min/max setting can be forgotten if used. */
8806 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8807 mddev->resync_min = 0;
8808 mddev->resync_max = MaxSector;
8809 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8810 mddev->resync_min = mddev->curr_resync_completed;
8811 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8812 mddev->curr_resync = 0;
8813 spin_unlock(&mddev->lock);
8815 wake_up(&resync_wait);
8816 md_wakeup_thread(mddev->thread);
8819 EXPORT_SYMBOL_GPL(md_do_sync);
8821 static int remove_and_add_spares(struct mddev *mddev,
8822 struct md_rdev *this)
8824 struct md_rdev *rdev;
8827 bool remove_some = false;
8829 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8830 /* Mustn't remove devices when resync thread is running */
8833 rdev_for_each(rdev, mddev) {
8834 if ((this == NULL || rdev == this) &&
8835 rdev->raid_disk >= 0 &&
8836 !test_bit(Blocked, &rdev->flags) &&
8837 test_bit(Faulty, &rdev->flags) &&
8838 atomic_read(&rdev->nr_pending)==0) {
8839 /* Faulty non-Blocked devices with nr_pending == 0
8840 * never get nr_pending incremented,
8841 * never get Faulty cleared, and never get Blocked set.
8842 * So we can synchronize_rcu now rather than once per device
8845 set_bit(RemoveSynchronized, &rdev->flags);
8851 rdev_for_each(rdev, mddev) {
8852 if ((this == NULL || rdev == this) &&
8853 rdev->raid_disk >= 0 &&
8854 !test_bit(Blocked, &rdev->flags) &&
8855 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8856 (!test_bit(In_sync, &rdev->flags) &&
8857 !test_bit(Journal, &rdev->flags))) &&
8858 atomic_read(&rdev->nr_pending)==0)) {
8859 if (mddev->pers->hot_remove_disk(
8860 mddev, rdev) == 0) {
8861 sysfs_unlink_rdev(mddev, rdev);
8862 rdev->saved_raid_disk = rdev->raid_disk;
8863 rdev->raid_disk = -1;
8867 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8868 clear_bit(RemoveSynchronized, &rdev->flags);
8871 if (removed && mddev->kobj.sd)
8872 sysfs_notify(&mddev->kobj, NULL, "degraded");
8874 if (this && removed)
8877 rdev_for_each(rdev, mddev) {
8878 if (this && this != rdev)
8880 if (test_bit(Candidate, &rdev->flags))
8882 if (rdev->raid_disk >= 0 &&
8883 !test_bit(In_sync, &rdev->flags) &&
8884 !test_bit(Journal, &rdev->flags) &&
8885 !test_bit(Faulty, &rdev->flags))
8887 if (rdev->raid_disk >= 0)
8889 if (test_bit(Faulty, &rdev->flags))
8891 if (!test_bit(Journal, &rdev->flags)) {
8893 ! (rdev->saved_raid_disk >= 0 &&
8894 !test_bit(Bitmap_sync, &rdev->flags)))
8897 rdev->recovery_offset = 0;
8900 hot_add_disk(mddev, rdev) == 0) {
8901 if (sysfs_link_rdev(mddev, rdev))
8902 /* failure here is OK */;
8903 if (!test_bit(Journal, &rdev->flags))
8905 md_new_event(mddev);
8906 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8911 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8915 static void md_start_sync(struct work_struct *ws)
8917 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8919 mddev->sync_thread = md_register_thread(md_do_sync,
8922 if (!mddev->sync_thread) {
8923 pr_warn("%s: could not start resync thread...\n",
8925 /* leave the spares where they are, it shouldn't hurt */
8926 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8927 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8928 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8929 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8930 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8931 wake_up(&resync_wait);
8932 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8934 if (mddev->sysfs_action)
8935 sysfs_notify_dirent_safe(mddev->sysfs_action);
8937 md_wakeup_thread(mddev->sync_thread);
8938 sysfs_notify_dirent_safe(mddev->sysfs_action);
8939 md_new_event(mddev);
8943 * This routine is regularly called by all per-raid-array threads to
8944 * deal with generic issues like resync and super-block update.
8945 * Raid personalities that don't have a thread (linear/raid0) do not
8946 * need this as they never do any recovery or update the superblock.
8948 * It does not do any resync itself, but rather "forks" off other threads
8949 * to do that as needed.
8950 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8951 * "->recovery" and create a thread at ->sync_thread.
8952 * When the thread finishes it sets MD_RECOVERY_DONE
8953 * and wakeups up this thread which will reap the thread and finish up.
8954 * This thread also removes any faulty devices (with nr_pending == 0).
8956 * The overall approach is:
8957 * 1/ if the superblock needs updating, update it.
8958 * 2/ If a recovery thread is running, don't do anything else.
8959 * 3/ If recovery has finished, clean up, possibly marking spares active.
8960 * 4/ If there are any faulty devices, remove them.
8961 * 5/ If array is degraded, try to add spares devices
8962 * 6/ If array has spares or is not in-sync, start a resync thread.
8964 void md_check_recovery(struct mddev *mddev)
8966 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8967 /* Write superblock - thread that called mddev_suspend()
8968 * holds reconfig_mutex for us.
8970 set_bit(MD_UPDATING_SB, &mddev->flags);
8971 smp_mb__after_atomic();
8972 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8973 md_update_sb(mddev, 0);
8974 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8975 wake_up(&mddev->sb_wait);
8978 if (mddev->suspended)
8982 md_bitmap_daemon_work(mddev);
8984 if (signal_pending(current)) {
8985 if (mddev->pers->sync_request && !mddev->external) {
8986 pr_debug("md: %s in immediate safe mode\n",
8988 mddev->safemode = 2;
8990 flush_signals(current);
8993 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8996 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8997 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8998 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8999 (mddev->external == 0 && mddev->safemode == 1) ||
9000 (mddev->safemode == 2
9001 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9005 if (mddev_trylock(mddev)) {
9007 bool try_set_sync = mddev->safemode != 0;
9009 if (!mddev->external && mddev->safemode == 1)
9010 mddev->safemode = 0;
9013 struct md_rdev *rdev;
9014 if (!mddev->external && mddev->in_sync)
9015 /* 'Blocked' flag not needed as failed devices
9016 * will be recorded if array switched to read/write.
9017 * Leaving it set will prevent the device
9018 * from being removed.
9020 rdev_for_each(rdev, mddev)
9021 clear_bit(Blocked, &rdev->flags);
9022 /* On a read-only array we can:
9023 * - remove failed devices
9024 * - add already-in_sync devices if the array itself
9026 * As we only add devices that are already in-sync,
9027 * we can activate the spares immediately.
9029 remove_and_add_spares(mddev, NULL);
9030 /* There is no thread, but we need to call
9031 * ->spare_active and clear saved_raid_disk
9033 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9034 md_reap_sync_thread(mddev);
9035 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9036 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9037 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9041 if (mddev_is_clustered(mddev)) {
9042 struct md_rdev *rdev;
9043 /* kick the device if another node issued a
9046 rdev_for_each(rdev, mddev) {
9047 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9048 rdev->raid_disk < 0)
9049 md_kick_rdev_from_array(rdev);
9053 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9054 spin_lock(&mddev->lock);
9056 spin_unlock(&mddev->lock);
9059 if (mddev->sb_flags)
9060 md_update_sb(mddev, 0);
9062 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9063 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9064 /* resync/recovery still happening */
9065 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9068 if (mddev->sync_thread) {
9069 md_reap_sync_thread(mddev);
9072 /* Set RUNNING before clearing NEEDED to avoid
9073 * any transients in the value of "sync_action".
9075 mddev->curr_resync_completed = 0;
9076 spin_lock(&mddev->lock);
9077 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9078 spin_unlock(&mddev->lock);
9079 /* Clear some bits that don't mean anything, but
9082 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9083 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9085 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9086 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9088 /* no recovery is running.
9089 * remove any failed drives, then
9090 * add spares if possible.
9091 * Spares are also removed and re-added, to allow
9092 * the personality to fail the re-add.
9095 if (mddev->reshape_position != MaxSector) {
9096 if (mddev->pers->check_reshape == NULL ||
9097 mddev->pers->check_reshape(mddev) != 0)
9098 /* Cannot proceed */
9100 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9101 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9102 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9103 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9104 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9105 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9106 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9107 } else if (mddev->recovery_cp < MaxSector) {
9108 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9109 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9110 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9111 /* nothing to be done ... */
9114 if (mddev->pers->sync_request) {
9116 /* We are adding a device or devices to an array
9117 * which has the bitmap stored on all devices.
9118 * So make sure all bitmap pages get written
9120 md_bitmap_write_all(mddev->bitmap);
9122 INIT_WORK(&mddev->del_work, md_start_sync);
9123 queue_work(md_misc_wq, &mddev->del_work);
9127 if (!mddev->sync_thread) {
9128 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9129 wake_up(&resync_wait);
9130 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9132 if (mddev->sysfs_action)
9133 sysfs_notify_dirent_safe(mddev->sysfs_action);
9136 wake_up(&mddev->sb_wait);
9137 mddev_unlock(mddev);
9140 EXPORT_SYMBOL(md_check_recovery);
9142 void md_reap_sync_thread(struct mddev *mddev)
9144 struct md_rdev *rdev;
9145 sector_t old_dev_sectors = mddev->dev_sectors;
9146 bool is_reshaped = false;
9148 /* resync has finished, collect result */
9149 md_unregister_thread(&mddev->sync_thread);
9150 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9151 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9152 mddev->degraded != mddev->raid_disks) {
9154 /* activate any spares */
9155 if (mddev->pers->spare_active(mddev)) {
9156 sysfs_notify(&mddev->kobj, NULL,
9158 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9161 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9162 mddev->pers->finish_reshape) {
9163 mddev->pers->finish_reshape(mddev);
9164 if (mddev_is_clustered(mddev))
9168 /* If array is no-longer degraded, then any saved_raid_disk
9169 * information must be scrapped.
9171 if (!mddev->degraded)
9172 rdev_for_each(rdev, mddev)
9173 rdev->saved_raid_disk = -1;
9175 md_update_sb(mddev, 1);
9176 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9177 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9179 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9180 md_cluster_ops->resync_finish(mddev);
9181 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9182 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9183 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9184 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9185 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9186 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9188 * We call md_cluster_ops->update_size here because sync_size could
9189 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9190 * so it is time to update size across cluster.
9192 if (mddev_is_clustered(mddev) && is_reshaped
9193 && !test_bit(MD_CLOSING, &mddev->flags))
9194 md_cluster_ops->update_size(mddev, old_dev_sectors);
9195 wake_up(&resync_wait);
9196 /* flag recovery needed just to double check */
9197 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9198 sysfs_notify_dirent_safe(mddev->sysfs_action);
9199 md_new_event(mddev);
9200 if (mddev->event_work.func)
9201 queue_work(md_misc_wq, &mddev->event_work);
9203 EXPORT_SYMBOL(md_reap_sync_thread);
9205 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9207 sysfs_notify_dirent_safe(rdev->sysfs_state);
9208 wait_event_timeout(rdev->blocked_wait,
9209 !test_bit(Blocked, &rdev->flags) &&
9210 !test_bit(BlockedBadBlocks, &rdev->flags),
9211 msecs_to_jiffies(5000));
9212 rdev_dec_pending(rdev, mddev);
9214 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9216 void md_finish_reshape(struct mddev *mddev)
9218 /* called be personality module when reshape completes. */
9219 struct md_rdev *rdev;
9221 rdev_for_each(rdev, mddev) {
9222 if (rdev->data_offset > rdev->new_data_offset)
9223 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9225 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9226 rdev->data_offset = rdev->new_data_offset;
9229 EXPORT_SYMBOL(md_finish_reshape);
9231 /* Bad block management */
9233 /* Returns 1 on success, 0 on failure */
9234 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9237 struct mddev *mddev = rdev->mddev;
9240 s += rdev->new_data_offset;
9242 s += rdev->data_offset;
9243 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9245 /* Make sure they get written out promptly */
9246 if (test_bit(ExternalBbl, &rdev->flags))
9247 sysfs_notify(&rdev->kobj, NULL,
9248 "unacknowledged_bad_blocks");
9249 sysfs_notify_dirent_safe(rdev->sysfs_state);
9250 set_mask_bits(&mddev->sb_flags, 0,
9251 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9252 md_wakeup_thread(rdev->mddev->thread);
9257 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9259 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9264 s += rdev->new_data_offset;
9266 s += rdev->data_offset;
9267 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9268 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9269 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9272 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9274 static int md_notify_reboot(struct notifier_block *this,
9275 unsigned long code, void *x)
9277 struct list_head *tmp;
9278 struct mddev *mddev;
9281 for_each_mddev(mddev, tmp) {
9282 if (mddev_trylock(mddev)) {
9284 __md_stop_writes(mddev);
9285 if (mddev->persistent)
9286 mddev->safemode = 2;
9287 mddev_unlock(mddev);
9292 * certain more exotic SCSI devices are known to be
9293 * volatile wrt too early system reboots. While the
9294 * right place to handle this issue is the given
9295 * driver, we do want to have a safe RAID driver ...
9303 static struct notifier_block md_notifier = {
9304 .notifier_call = md_notify_reboot,
9306 .priority = INT_MAX, /* before any real devices */
9309 static void md_geninit(void)
9311 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9313 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9316 static int __init md_init(void)
9320 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9324 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9328 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9331 if ((ret = register_blkdev(0, "mdp")) < 0)
9335 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9336 md_probe, NULL, NULL);
9337 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9338 md_probe, NULL, NULL);
9340 register_reboot_notifier(&md_notifier);
9341 raid_table_header = register_sysctl_table(raid_root_table);
9347 unregister_blkdev(MD_MAJOR, "md");
9349 destroy_workqueue(md_misc_wq);
9351 destroy_workqueue(md_wq);
9356 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9358 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9359 struct md_rdev *rdev2;
9361 char b[BDEVNAME_SIZE];
9364 * If size is changed in another node then we need to
9365 * do resize as well.
9367 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9368 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9370 pr_info("md-cluster: resize failed\n");
9372 md_bitmap_update_sb(mddev->bitmap);
9375 /* Check for change of roles in the active devices */
9376 rdev_for_each(rdev2, mddev) {
9377 if (test_bit(Faulty, &rdev2->flags))
9380 /* Check if the roles changed */
9381 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9383 if (test_bit(Candidate, &rdev2->flags)) {
9384 if (role == 0xfffe) {
9385 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9386 md_kick_rdev_from_array(rdev2);
9390 clear_bit(Candidate, &rdev2->flags);
9393 if (role != rdev2->raid_disk) {
9395 * got activated except reshape is happening.
9397 if (rdev2->raid_disk == -1 && role != 0xffff &&
9398 !(le32_to_cpu(sb->feature_map) &
9399 MD_FEATURE_RESHAPE_ACTIVE)) {
9400 rdev2->saved_raid_disk = role;
9401 ret = remove_and_add_spares(mddev, rdev2);
9402 pr_info("Activated spare: %s\n",
9403 bdevname(rdev2->bdev,b));
9404 /* wakeup mddev->thread here, so array could
9405 * perform resync with the new activated disk */
9406 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9407 md_wakeup_thread(mddev->thread);
9410 * We just want to do the minimum to mark the disk
9411 * as faulty. The recovery is performed by the
9412 * one who initiated the error.
9414 if ((role == 0xfffe) || (role == 0xfffd)) {
9415 md_error(mddev, rdev2);
9416 clear_bit(Blocked, &rdev2->flags);
9421 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9422 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9425 * Since mddev->delta_disks has already updated in update_raid_disks,
9426 * so it is time to check reshape.
9428 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9429 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9431 * reshape is happening in the remote node, we need to
9432 * update reshape_position and call start_reshape.
9434 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9435 if (mddev->pers->update_reshape_pos)
9436 mddev->pers->update_reshape_pos(mddev);
9437 if (mddev->pers->start_reshape)
9438 mddev->pers->start_reshape(mddev);
9439 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9440 mddev->reshape_position != MaxSector &&
9441 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9442 /* reshape is just done in another node. */
9443 mddev->reshape_position = MaxSector;
9444 if (mddev->pers->update_reshape_pos)
9445 mddev->pers->update_reshape_pos(mddev);
9448 /* Finally set the event to be up to date */
9449 mddev->events = le64_to_cpu(sb->events);
9452 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9455 struct page *swapout = rdev->sb_page;
9456 struct mdp_superblock_1 *sb;
9458 /* Store the sb page of the rdev in the swapout temporary
9459 * variable in case we err in the future
9461 rdev->sb_page = NULL;
9462 err = alloc_disk_sb(rdev);
9464 ClearPageUptodate(rdev->sb_page);
9465 rdev->sb_loaded = 0;
9466 err = super_types[mddev->major_version].
9467 load_super(rdev, NULL, mddev->minor_version);
9470 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9471 __func__, __LINE__, rdev->desc_nr, err);
9473 put_page(rdev->sb_page);
9474 rdev->sb_page = swapout;
9475 rdev->sb_loaded = 1;
9479 sb = page_address(rdev->sb_page);
9480 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9484 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9485 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9487 /* The other node finished recovery, call spare_active to set
9488 * device In_sync and mddev->degraded
9490 if (rdev->recovery_offset == MaxSector &&
9491 !test_bit(In_sync, &rdev->flags) &&
9492 mddev->pers->spare_active(mddev))
9493 sysfs_notify(&mddev->kobj, NULL, "degraded");
9499 void md_reload_sb(struct mddev *mddev, int nr)
9501 struct md_rdev *rdev;
9505 rdev_for_each_rcu(rdev, mddev) {
9506 if (rdev->desc_nr == nr)
9510 if (!rdev || rdev->desc_nr != nr) {
9511 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9515 err = read_rdev(mddev, rdev);
9519 check_sb_changes(mddev, rdev);
9521 /* Read all rdev's to update recovery_offset */
9522 rdev_for_each_rcu(rdev, mddev) {
9523 if (!test_bit(Faulty, &rdev->flags))
9524 read_rdev(mddev, rdev);
9527 EXPORT_SYMBOL(md_reload_sb);
9532 * Searches all registered partitions for autorun RAID arrays
9536 static DEFINE_MUTEX(detected_devices_mutex);
9537 static LIST_HEAD(all_detected_devices);
9538 struct detected_devices_node {
9539 struct list_head list;
9543 void md_autodetect_dev(dev_t dev)
9545 struct detected_devices_node *node_detected_dev;
9547 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9548 if (node_detected_dev) {
9549 node_detected_dev->dev = dev;
9550 mutex_lock(&detected_devices_mutex);
9551 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9552 mutex_unlock(&detected_devices_mutex);
9556 static void autostart_arrays(int part)
9558 struct md_rdev *rdev;
9559 struct detected_devices_node *node_detected_dev;
9561 int i_scanned, i_passed;
9566 pr_info("md: Autodetecting RAID arrays.\n");
9568 mutex_lock(&detected_devices_mutex);
9569 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9571 node_detected_dev = list_entry(all_detected_devices.next,
9572 struct detected_devices_node, list);
9573 list_del(&node_detected_dev->list);
9574 dev = node_detected_dev->dev;
9575 kfree(node_detected_dev);
9576 mutex_unlock(&detected_devices_mutex);
9577 rdev = md_import_device(dev,0, 90);
9578 mutex_lock(&detected_devices_mutex);
9582 if (test_bit(Faulty, &rdev->flags))
9585 set_bit(AutoDetected, &rdev->flags);
9586 list_add(&rdev->same_set, &pending_raid_disks);
9589 mutex_unlock(&detected_devices_mutex);
9591 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9593 autorun_devices(part);
9596 #endif /* !MODULE */
9598 static __exit void md_exit(void)
9600 struct mddev *mddev;
9601 struct list_head *tmp;
9604 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9605 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9607 unregister_blkdev(MD_MAJOR,"md");
9608 unregister_blkdev(mdp_major, "mdp");
9609 unregister_reboot_notifier(&md_notifier);
9610 unregister_sysctl_table(raid_table_header);
9612 /* We cannot unload the modules while some process is
9613 * waiting for us in select() or poll() - wake them up
9616 while (waitqueue_active(&md_event_waiters)) {
9617 /* not safe to leave yet */
9618 wake_up(&md_event_waiters);
9622 remove_proc_entry("mdstat", NULL);
9624 for_each_mddev(mddev, tmp) {
9625 export_array(mddev);
9627 mddev->hold_active = 0;
9629 * for_each_mddev() will call mddev_put() at the end of each
9630 * iteration. As the mddev is now fully clear, this will
9631 * schedule the mddev for destruction by a workqueue, and the
9632 * destroy_workqueue() below will wait for that to complete.
9635 destroy_workqueue(md_misc_wq);
9636 destroy_workqueue(md_wq);
9639 subsys_initcall(md_init);
9640 module_exit(md_exit)
9642 static int get_ro(char *buffer, const struct kernel_param *kp)
9644 return sprintf(buffer, "%d", start_readonly);
9646 static int set_ro(const char *val, const struct kernel_param *kp)
9648 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9651 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9652 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9653 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9654 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9656 MODULE_LICENSE("GPL");
9657 MODULE_DESCRIPTION("MD RAID framework");
9659 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);