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 sb->disks[rdev->desc_nr].state &
1163 ((1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE))))
1173 mdp_super_t *refsb = page_address(refdev->sb_page);
1174 if (!md_uuid_equal(refsb, sb)) {
1175 pr_warn("md: %s has different UUID to %s\n",
1176 b, bdevname(refdev->bdev,b2));
1179 if (!md_sb_equal(refsb, sb)) {
1180 pr_warn("md: %s has same UUID but different superblock to %s\n",
1181 b, bdevname(refdev->bdev, b2));
1185 ev2 = md_event(refsb);
1187 if (!spare_disk && ev1 > ev2)
1192 rdev->sectors = rdev->sb_start;
1193 /* Limit to 4TB as metadata cannot record more than that.
1194 * (not needed for Linear and RAID0 as metadata doesn't
1197 if ((u64)rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1198 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1200 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1201 /* "this cannot possibly happen" ... */
1209 * validate_super for 0.90.0
1211 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1214 mdp_super_t *sb = page_address(rdev->sb_page);
1215 __u64 ev1 = md_event(sb);
1217 rdev->raid_disk = -1;
1218 clear_bit(Faulty, &rdev->flags);
1219 clear_bit(In_sync, &rdev->flags);
1220 clear_bit(Bitmap_sync, &rdev->flags);
1221 clear_bit(WriteMostly, &rdev->flags);
1223 if (mddev->raid_disks == 0) {
1224 mddev->major_version = 0;
1225 mddev->minor_version = sb->minor_version;
1226 mddev->patch_version = sb->patch_version;
1227 mddev->external = 0;
1228 mddev->chunk_sectors = sb->chunk_size >> 9;
1229 mddev->ctime = sb->ctime;
1230 mddev->utime = sb->utime;
1231 mddev->level = sb->level;
1232 mddev->clevel[0] = 0;
1233 mddev->layout = sb->layout;
1234 mddev->raid_disks = sb->raid_disks;
1235 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1236 mddev->events = ev1;
1237 mddev->bitmap_info.offset = 0;
1238 mddev->bitmap_info.space = 0;
1239 /* bitmap can use 60 K after the 4K superblocks */
1240 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1241 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1242 mddev->reshape_backwards = 0;
1244 if (mddev->minor_version >= 91) {
1245 mddev->reshape_position = sb->reshape_position;
1246 mddev->delta_disks = sb->delta_disks;
1247 mddev->new_level = sb->new_level;
1248 mddev->new_layout = sb->new_layout;
1249 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1250 if (mddev->delta_disks < 0)
1251 mddev->reshape_backwards = 1;
1253 mddev->reshape_position = MaxSector;
1254 mddev->delta_disks = 0;
1255 mddev->new_level = mddev->level;
1256 mddev->new_layout = mddev->layout;
1257 mddev->new_chunk_sectors = mddev->chunk_sectors;
1259 if (mddev->level == 0)
1262 if (sb->state & (1<<MD_SB_CLEAN))
1263 mddev->recovery_cp = MaxSector;
1265 if (sb->events_hi == sb->cp_events_hi &&
1266 sb->events_lo == sb->cp_events_lo) {
1267 mddev->recovery_cp = sb->recovery_cp;
1269 mddev->recovery_cp = 0;
1272 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1273 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1274 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1275 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1277 mddev->max_disks = MD_SB_DISKS;
1279 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1280 mddev->bitmap_info.file == NULL) {
1281 mddev->bitmap_info.offset =
1282 mddev->bitmap_info.default_offset;
1283 mddev->bitmap_info.space =
1284 mddev->bitmap_info.default_space;
1287 } else if (mddev->pers == NULL) {
1288 /* Insist on good event counter while assembling, except
1289 * for spares (which don't need an event count) */
1291 if (sb->disks[rdev->desc_nr].state & (
1292 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1293 if (ev1 < mddev->events)
1295 } else if (mddev->bitmap) {
1296 /* if adding to array with a bitmap, then we can accept an
1297 * older device ... but not too old.
1299 if (ev1 < mddev->bitmap->events_cleared)
1301 if (ev1 < mddev->events)
1302 set_bit(Bitmap_sync, &rdev->flags);
1304 if (ev1 < mddev->events)
1305 /* just a hot-add of a new device, leave raid_disk at -1 */
1309 if (mddev->level != LEVEL_MULTIPATH) {
1310 desc = sb->disks + rdev->desc_nr;
1312 if (desc->state & (1<<MD_DISK_FAULTY))
1313 set_bit(Faulty, &rdev->flags);
1314 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1315 desc->raid_disk < mddev->raid_disks */) {
1316 set_bit(In_sync, &rdev->flags);
1317 rdev->raid_disk = desc->raid_disk;
1318 rdev->saved_raid_disk = desc->raid_disk;
1319 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1320 /* active but not in sync implies recovery up to
1321 * reshape position. We don't know exactly where
1322 * that is, so set to zero for now */
1323 if (mddev->minor_version >= 91) {
1324 rdev->recovery_offset = 0;
1325 rdev->raid_disk = desc->raid_disk;
1328 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1329 set_bit(WriteMostly, &rdev->flags);
1330 if (desc->state & (1<<MD_DISK_FAILFAST))
1331 set_bit(FailFast, &rdev->flags);
1332 } else /* MULTIPATH are always insync */
1333 set_bit(In_sync, &rdev->flags);
1338 * sync_super for 0.90.0
1340 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1343 struct md_rdev *rdev2;
1344 int next_spare = mddev->raid_disks;
1346 /* make rdev->sb match mddev data..
1349 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1350 * 3/ any empty disks < next_spare become removed
1352 * disks[0] gets initialised to REMOVED because
1353 * we cannot be sure from other fields if it has
1354 * been initialised or not.
1357 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1359 rdev->sb_size = MD_SB_BYTES;
1361 sb = page_address(rdev->sb_page);
1363 memset(sb, 0, sizeof(*sb));
1365 sb->md_magic = MD_SB_MAGIC;
1366 sb->major_version = mddev->major_version;
1367 sb->patch_version = mddev->patch_version;
1368 sb->gvalid_words = 0; /* ignored */
1369 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1370 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1371 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1372 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1374 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1375 sb->level = mddev->level;
1376 sb->size = mddev->dev_sectors / 2;
1377 sb->raid_disks = mddev->raid_disks;
1378 sb->md_minor = mddev->md_minor;
1379 sb->not_persistent = 0;
1380 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1382 sb->events_hi = (mddev->events>>32);
1383 sb->events_lo = (u32)mddev->events;
1385 if (mddev->reshape_position == MaxSector)
1386 sb->minor_version = 90;
1388 sb->minor_version = 91;
1389 sb->reshape_position = mddev->reshape_position;
1390 sb->new_level = mddev->new_level;
1391 sb->delta_disks = mddev->delta_disks;
1392 sb->new_layout = mddev->new_layout;
1393 sb->new_chunk = mddev->new_chunk_sectors << 9;
1395 mddev->minor_version = sb->minor_version;
1398 sb->recovery_cp = mddev->recovery_cp;
1399 sb->cp_events_hi = (mddev->events>>32);
1400 sb->cp_events_lo = (u32)mddev->events;
1401 if (mddev->recovery_cp == MaxSector)
1402 sb->state = (1<< MD_SB_CLEAN);
1404 sb->recovery_cp = 0;
1406 sb->layout = mddev->layout;
1407 sb->chunk_size = mddev->chunk_sectors << 9;
1409 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1410 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1412 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1413 rdev_for_each(rdev2, mddev) {
1416 int is_active = test_bit(In_sync, &rdev2->flags);
1418 if (rdev2->raid_disk >= 0 &&
1419 sb->minor_version >= 91)
1420 /* we have nowhere to store the recovery_offset,
1421 * but if it is not below the reshape_position,
1422 * we can piggy-back on that.
1425 if (rdev2->raid_disk < 0 ||
1426 test_bit(Faulty, &rdev2->flags))
1429 desc_nr = rdev2->raid_disk;
1431 desc_nr = next_spare++;
1432 rdev2->desc_nr = desc_nr;
1433 d = &sb->disks[rdev2->desc_nr];
1435 d->number = rdev2->desc_nr;
1436 d->major = MAJOR(rdev2->bdev->bd_dev);
1437 d->minor = MINOR(rdev2->bdev->bd_dev);
1439 d->raid_disk = rdev2->raid_disk;
1441 d->raid_disk = rdev2->desc_nr; /* compatibility */
1442 if (test_bit(Faulty, &rdev2->flags))
1443 d->state = (1<<MD_DISK_FAULTY);
1444 else if (is_active) {
1445 d->state = (1<<MD_DISK_ACTIVE);
1446 if (test_bit(In_sync, &rdev2->flags))
1447 d->state |= (1<<MD_DISK_SYNC);
1455 if (test_bit(WriteMostly, &rdev2->flags))
1456 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1457 if (test_bit(FailFast, &rdev2->flags))
1458 d->state |= (1<<MD_DISK_FAILFAST);
1460 /* now set the "removed" and "faulty" bits on any missing devices */
1461 for (i=0 ; i < mddev->raid_disks ; i++) {
1462 mdp_disk_t *d = &sb->disks[i];
1463 if (d->state == 0 && d->number == 0) {
1466 d->state = (1<<MD_DISK_REMOVED);
1467 d->state |= (1<<MD_DISK_FAULTY);
1471 sb->nr_disks = nr_disks;
1472 sb->active_disks = active;
1473 sb->working_disks = working;
1474 sb->failed_disks = failed;
1475 sb->spare_disks = spare;
1477 sb->this_disk = sb->disks[rdev->desc_nr];
1478 sb->sb_csum = calc_sb_csum(sb);
1482 * rdev_size_change for 0.90.0
1484 static unsigned long long
1485 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1487 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1488 return 0; /* component must fit device */
1489 if (rdev->mddev->bitmap_info.offset)
1490 return 0; /* can't move bitmap */
1491 rdev->sb_start = calc_dev_sboffset(rdev);
1492 if (!num_sectors || num_sectors > rdev->sb_start)
1493 num_sectors = rdev->sb_start;
1494 /* Limit to 4TB as metadata cannot record more than that.
1495 * 4TB == 2^32 KB, or 2*2^32 sectors.
1497 if ((u64)num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1498 num_sectors = (sector_t)(2ULL << 32) - 2;
1500 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1502 } while (md_super_wait(rdev->mddev) < 0);
1507 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1509 /* non-zero offset changes not possible with v0.90 */
1510 return new_offset == 0;
1514 * version 1 superblock
1517 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1521 unsigned long long newcsum;
1522 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1523 __le32 *isuper = (__le32*)sb;
1525 disk_csum = sb->sb_csum;
1528 for (; size >= 4; size -= 4)
1529 newcsum += le32_to_cpu(*isuper++);
1532 newcsum += le16_to_cpu(*(__le16*) isuper);
1534 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1535 sb->sb_csum = disk_csum;
1536 return cpu_to_le32(csum);
1539 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1541 struct mdp_superblock_1 *sb;
1545 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1547 bool spare_disk = true;
1550 * Calculate the position of the superblock in 512byte sectors.
1551 * It is always aligned to a 4K boundary and
1552 * depeding on minor_version, it can be:
1553 * 0: At least 8K, but less than 12K, from end of device
1554 * 1: At start of device
1555 * 2: 4K from start of device.
1557 switch(minor_version) {
1559 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1561 sb_start &= ~(sector_t)(4*2-1);
1572 rdev->sb_start = sb_start;
1574 /* superblock is rarely larger than 1K, but it can be larger,
1575 * and it is safe to read 4k, so we do that
1577 ret = read_disk_sb(rdev, 4096);
1578 if (ret) return ret;
1580 sb = page_address(rdev->sb_page);
1582 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1583 sb->major_version != cpu_to_le32(1) ||
1584 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1585 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1586 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1589 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1590 pr_warn("md: invalid superblock checksum on %s\n",
1591 bdevname(rdev->bdev,b));
1594 if (le64_to_cpu(sb->data_size) < 10) {
1595 pr_warn("md: data_size too small on %s\n",
1596 bdevname(rdev->bdev,b));
1601 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1602 /* Some padding is non-zero, might be a new feature */
1605 rdev->preferred_minor = 0xffff;
1606 rdev->data_offset = le64_to_cpu(sb->data_offset);
1607 rdev->new_data_offset = rdev->data_offset;
1608 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1609 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1610 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1611 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1613 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1614 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1615 if (rdev->sb_size & bmask)
1616 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1619 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1622 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1625 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1628 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1630 if (!rdev->bb_page) {
1631 rdev->bb_page = alloc_page(GFP_KERNEL);
1635 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1636 rdev->badblocks.count == 0) {
1637 /* need to load the bad block list.
1638 * Currently we limit it to one page.
1644 int sectors = le16_to_cpu(sb->bblog_size);
1645 if (sectors > (PAGE_SIZE / 512))
1647 offset = le32_to_cpu(sb->bblog_offset);
1650 bb_sector = (long long)offset;
1651 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1652 rdev->bb_page, REQ_OP_READ, 0, true))
1654 bbp = (__le64 *)page_address(rdev->bb_page);
1655 rdev->badblocks.shift = sb->bblog_shift;
1656 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1657 u64 bb = le64_to_cpu(*bbp);
1658 int count = bb & (0x3ff);
1659 u64 sector = bb >> 10;
1660 sector <<= sb->bblog_shift;
1661 count <<= sb->bblog_shift;
1664 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1667 } else if (sb->bblog_offset != 0)
1668 rdev->badblocks.shift = 0;
1670 if ((le32_to_cpu(sb->feature_map) &
1671 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS))) {
1672 rdev->ppl.offset = (__s16)le16_to_cpu(sb->ppl.offset);
1673 rdev->ppl.size = le16_to_cpu(sb->ppl.size);
1674 rdev->ppl.sector = rdev->sb_start + rdev->ppl.offset;
1677 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT) &&
1681 /* not spare disk, or LEVEL_MULTIPATH */
1682 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH) ||
1683 (rdev->desc_nr >= 0 &&
1684 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1685 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1686 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL)))
1696 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1698 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1699 sb->level != refsb->level ||
1700 sb->layout != refsb->layout ||
1701 sb->chunksize != refsb->chunksize) {
1702 pr_warn("md: %s has strangely different superblock to %s\n",
1703 bdevname(rdev->bdev,b),
1704 bdevname(refdev->bdev,b2));
1707 ev1 = le64_to_cpu(sb->events);
1708 ev2 = le64_to_cpu(refsb->events);
1710 if (!spare_disk && ev1 > ev2)
1715 if (minor_version) {
1716 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1717 sectors -= rdev->data_offset;
1719 sectors = rdev->sb_start;
1720 if (sectors < le64_to_cpu(sb->data_size))
1722 rdev->sectors = le64_to_cpu(sb->data_size);
1726 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1728 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1729 __u64 ev1 = le64_to_cpu(sb->events);
1731 rdev->raid_disk = -1;
1732 clear_bit(Faulty, &rdev->flags);
1733 clear_bit(In_sync, &rdev->flags);
1734 clear_bit(Bitmap_sync, &rdev->flags);
1735 clear_bit(WriteMostly, &rdev->flags);
1737 if (mddev->raid_disks == 0) {
1738 mddev->major_version = 1;
1739 mddev->patch_version = 0;
1740 mddev->external = 0;
1741 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1742 mddev->ctime = le64_to_cpu(sb->ctime);
1743 mddev->utime = le64_to_cpu(sb->utime);
1744 mddev->level = le32_to_cpu(sb->level);
1745 mddev->clevel[0] = 0;
1746 mddev->layout = le32_to_cpu(sb->layout);
1747 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1748 mddev->dev_sectors = le64_to_cpu(sb->size);
1749 mddev->events = ev1;
1750 mddev->bitmap_info.offset = 0;
1751 mddev->bitmap_info.space = 0;
1752 /* Default location for bitmap is 1K after superblock
1753 * using 3K - total of 4K
1755 mddev->bitmap_info.default_offset = 1024 >> 9;
1756 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1757 mddev->reshape_backwards = 0;
1759 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1760 memcpy(mddev->uuid, sb->set_uuid, 16);
1762 mddev->max_disks = (4096-256)/2;
1764 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1765 mddev->bitmap_info.file == NULL) {
1766 mddev->bitmap_info.offset =
1767 (__s32)le32_to_cpu(sb->bitmap_offset);
1768 /* Metadata doesn't record how much space is available.
1769 * For 1.0, we assume we can use up to the superblock
1770 * if before, else to 4K beyond superblock.
1771 * For others, assume no change is possible.
1773 if (mddev->minor_version > 0)
1774 mddev->bitmap_info.space = 0;
1775 else if (mddev->bitmap_info.offset > 0)
1776 mddev->bitmap_info.space =
1777 8 - mddev->bitmap_info.offset;
1779 mddev->bitmap_info.space =
1780 -mddev->bitmap_info.offset;
1783 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1784 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1785 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1786 mddev->new_level = le32_to_cpu(sb->new_level);
1787 mddev->new_layout = le32_to_cpu(sb->new_layout);
1788 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1789 if (mddev->delta_disks < 0 ||
1790 (mddev->delta_disks == 0 &&
1791 (le32_to_cpu(sb->feature_map)
1792 & MD_FEATURE_RESHAPE_BACKWARDS)))
1793 mddev->reshape_backwards = 1;
1795 mddev->reshape_position = MaxSector;
1796 mddev->delta_disks = 0;
1797 mddev->new_level = mddev->level;
1798 mddev->new_layout = mddev->layout;
1799 mddev->new_chunk_sectors = mddev->chunk_sectors;
1802 if (mddev->level == 0 &&
1803 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RAID0_LAYOUT))
1806 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1807 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1809 if (le32_to_cpu(sb->feature_map) &
1810 (MD_FEATURE_PPL | MD_FEATURE_MULTIPLE_PPLS)) {
1811 if (le32_to_cpu(sb->feature_map) &
1812 (MD_FEATURE_BITMAP_OFFSET | MD_FEATURE_JOURNAL))
1814 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_PPL) &&
1815 (le32_to_cpu(sb->feature_map) &
1816 MD_FEATURE_MULTIPLE_PPLS))
1818 set_bit(MD_HAS_PPL, &mddev->flags);
1820 } else if (mddev->pers == NULL) {
1821 /* Insist of good event counter while assembling, except for
1822 * spares (which don't need an event count) */
1824 if (rdev->desc_nr >= 0 &&
1825 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1826 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1827 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1828 if (ev1 < mddev->events)
1830 } else if (mddev->bitmap) {
1831 /* If adding to array with a bitmap, then we can accept an
1832 * older device, but not too old.
1834 if (ev1 < mddev->bitmap->events_cleared)
1836 if (ev1 < mddev->events)
1837 set_bit(Bitmap_sync, &rdev->flags);
1839 if (ev1 < mddev->events)
1840 /* just a hot-add of a new device, leave raid_disk at -1 */
1843 if (mddev->level != LEVEL_MULTIPATH) {
1845 if (rdev->desc_nr < 0 ||
1846 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1847 role = MD_DISK_ROLE_SPARE;
1850 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1852 case MD_DISK_ROLE_SPARE: /* spare */
1854 case MD_DISK_ROLE_FAULTY: /* faulty */
1855 set_bit(Faulty, &rdev->flags);
1857 case MD_DISK_ROLE_JOURNAL: /* journal device */
1858 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1859 /* journal device without journal feature */
1860 pr_warn("md: journal device provided without journal feature, ignoring the device\n");
1863 set_bit(Journal, &rdev->flags);
1864 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1865 rdev->raid_disk = 0;
1868 rdev->saved_raid_disk = role;
1869 if ((le32_to_cpu(sb->feature_map) &
1870 MD_FEATURE_RECOVERY_OFFSET)) {
1871 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1872 if (!(le32_to_cpu(sb->feature_map) &
1873 MD_FEATURE_RECOVERY_BITMAP))
1874 rdev->saved_raid_disk = -1;
1877 * If the array is FROZEN, then the device can't
1878 * be in_sync with rest of array.
1880 if (!test_bit(MD_RECOVERY_FROZEN,
1882 set_bit(In_sync, &rdev->flags);
1884 rdev->raid_disk = role;
1887 if (sb->devflags & WriteMostly1)
1888 set_bit(WriteMostly, &rdev->flags);
1889 if (sb->devflags & FailFast1)
1890 set_bit(FailFast, &rdev->flags);
1891 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1892 set_bit(Replacement, &rdev->flags);
1893 } else /* MULTIPATH are always insync */
1894 set_bit(In_sync, &rdev->flags);
1899 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1901 struct mdp_superblock_1 *sb;
1902 struct md_rdev *rdev2;
1904 /* make rdev->sb match mddev and rdev data. */
1906 sb = page_address(rdev->sb_page);
1908 sb->feature_map = 0;
1910 sb->recovery_offset = cpu_to_le64(0);
1911 memset(sb->pad3, 0, sizeof(sb->pad3));
1913 sb->utime = cpu_to_le64((__u64)mddev->utime);
1914 sb->events = cpu_to_le64(mddev->events);
1916 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1917 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1918 sb->resync_offset = cpu_to_le64(MaxSector);
1920 sb->resync_offset = cpu_to_le64(0);
1922 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1924 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1925 sb->size = cpu_to_le64(mddev->dev_sectors);
1926 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1927 sb->level = cpu_to_le32(mddev->level);
1928 sb->layout = cpu_to_le32(mddev->layout);
1929 if (test_bit(FailFast, &rdev->flags))
1930 sb->devflags |= FailFast1;
1932 sb->devflags &= ~FailFast1;
1934 if (test_bit(WriteMostly, &rdev->flags))
1935 sb->devflags |= WriteMostly1;
1937 sb->devflags &= ~WriteMostly1;
1938 sb->data_offset = cpu_to_le64(rdev->data_offset);
1939 sb->data_size = cpu_to_le64(rdev->sectors);
1941 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1942 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1943 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1946 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1947 !test_bit(In_sync, &rdev->flags)) {
1949 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1950 sb->recovery_offset =
1951 cpu_to_le64(rdev->recovery_offset);
1952 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1954 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1956 /* Note: recovery_offset and journal_tail share space */
1957 if (test_bit(Journal, &rdev->flags))
1958 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1959 if (test_bit(Replacement, &rdev->flags))
1961 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1963 if (mddev->reshape_position != MaxSector) {
1964 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1965 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1966 sb->new_layout = cpu_to_le32(mddev->new_layout);
1967 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1968 sb->new_level = cpu_to_le32(mddev->new_level);
1969 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1970 if (mddev->delta_disks == 0 &&
1971 mddev->reshape_backwards)
1973 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1974 if (rdev->new_data_offset != rdev->data_offset) {
1976 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1977 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1978 - rdev->data_offset));
1982 if (mddev_is_clustered(mddev))
1983 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1985 if (rdev->badblocks.count == 0)
1986 /* Nothing to do for bad blocks*/ ;
1987 else if (sb->bblog_offset == 0)
1988 /* Cannot record bad blocks on this device */
1989 md_error(mddev, rdev);
1991 struct badblocks *bb = &rdev->badblocks;
1992 __le64 *bbp = (__le64 *)page_address(rdev->bb_page);
1994 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1999 seq = read_seqbegin(&bb->lock);
2001 memset(bbp, 0xff, PAGE_SIZE);
2003 for (i = 0 ; i < bb->count ; i++) {
2004 u64 internal_bb = p[i];
2005 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
2006 | BB_LEN(internal_bb));
2007 bbp[i] = cpu_to_le64(store_bb);
2010 if (read_seqretry(&bb->lock, seq))
2013 bb->sector = (rdev->sb_start +
2014 (int)le32_to_cpu(sb->bblog_offset));
2015 bb->size = le16_to_cpu(sb->bblog_size);
2020 rdev_for_each(rdev2, mddev)
2021 if (rdev2->desc_nr+1 > max_dev)
2022 max_dev = rdev2->desc_nr+1;
2024 if (max_dev > le32_to_cpu(sb->max_dev)) {
2026 sb->max_dev = cpu_to_le32(max_dev);
2027 rdev->sb_size = max_dev * 2 + 256;
2028 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
2029 if (rdev->sb_size & bmask)
2030 rdev->sb_size = (rdev->sb_size | bmask) + 1;
2032 max_dev = le32_to_cpu(sb->max_dev);
2034 for (i=0; i<max_dev;i++)
2035 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2037 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
2038 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
2040 if (test_bit(MD_HAS_PPL, &mddev->flags)) {
2041 if (test_bit(MD_HAS_MULTIPLE_PPLS, &mddev->flags))
2043 cpu_to_le32(MD_FEATURE_MULTIPLE_PPLS);
2045 sb->feature_map |= cpu_to_le32(MD_FEATURE_PPL);
2046 sb->ppl.offset = cpu_to_le16(rdev->ppl.offset);
2047 sb->ppl.size = cpu_to_le16(rdev->ppl.size);
2050 rdev_for_each(rdev2, mddev) {
2052 if (test_bit(Faulty, &rdev2->flags))
2053 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
2054 else if (test_bit(In_sync, &rdev2->flags))
2055 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2056 else if (test_bit(Journal, &rdev2->flags))
2057 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
2058 else if (rdev2->raid_disk >= 0)
2059 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
2061 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
2064 sb->sb_csum = calc_sb_1_csum(sb);
2067 static unsigned long long
2068 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
2070 struct mdp_superblock_1 *sb;
2071 sector_t max_sectors;
2072 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
2073 return 0; /* component must fit device */
2074 if (rdev->data_offset != rdev->new_data_offset)
2075 return 0; /* too confusing */
2076 if (rdev->sb_start < rdev->data_offset) {
2077 /* minor versions 1 and 2; superblock before data */
2078 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
2079 max_sectors -= rdev->data_offset;
2080 if (!num_sectors || num_sectors > max_sectors)
2081 num_sectors = max_sectors;
2082 } else if (rdev->mddev->bitmap_info.offset) {
2083 /* minor version 0 with bitmap we can't move */
2086 /* minor version 0; superblock after data */
2088 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
2089 sb_start &= ~(sector_t)(4*2 - 1);
2090 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
2091 if (!num_sectors || num_sectors > max_sectors)
2092 num_sectors = max_sectors;
2093 rdev->sb_start = sb_start;
2095 sb = page_address(rdev->sb_page);
2096 sb->data_size = cpu_to_le64(num_sectors);
2097 sb->super_offset = cpu_to_le64(rdev->sb_start);
2098 sb->sb_csum = calc_sb_1_csum(sb);
2100 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
2102 } while (md_super_wait(rdev->mddev) < 0);
2108 super_1_allow_new_offset(struct md_rdev *rdev,
2109 unsigned long long new_offset)
2111 /* All necessary checks on new >= old have been done */
2112 struct bitmap *bitmap;
2113 if (new_offset >= rdev->data_offset)
2116 /* with 1.0 metadata, there is no metadata to tread on
2117 * so we can always move back */
2118 if (rdev->mddev->minor_version == 0)
2121 /* otherwise we must be sure not to step on
2122 * any metadata, so stay:
2123 * 36K beyond start of superblock
2124 * beyond end of badblocks
2125 * beyond write-intent bitmap
2127 if (rdev->sb_start + (32+4)*2 > new_offset)
2129 bitmap = rdev->mddev->bitmap;
2130 if (bitmap && !rdev->mddev->bitmap_info.file &&
2131 rdev->sb_start + rdev->mddev->bitmap_info.offset +
2132 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
2134 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
2140 static struct super_type super_types[] = {
2143 .owner = THIS_MODULE,
2144 .load_super = super_90_load,
2145 .validate_super = super_90_validate,
2146 .sync_super = super_90_sync,
2147 .rdev_size_change = super_90_rdev_size_change,
2148 .allow_new_offset = super_90_allow_new_offset,
2152 .owner = THIS_MODULE,
2153 .load_super = super_1_load,
2154 .validate_super = super_1_validate,
2155 .sync_super = super_1_sync,
2156 .rdev_size_change = super_1_rdev_size_change,
2157 .allow_new_offset = super_1_allow_new_offset,
2161 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
2163 if (mddev->sync_super) {
2164 mddev->sync_super(mddev, rdev);
2168 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
2170 super_types[mddev->major_version].sync_super(mddev, rdev);
2173 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
2175 struct md_rdev *rdev, *rdev2;
2178 rdev_for_each_rcu(rdev, mddev1) {
2179 if (test_bit(Faulty, &rdev->flags) ||
2180 test_bit(Journal, &rdev->flags) ||
2181 rdev->raid_disk == -1)
2183 rdev_for_each_rcu(rdev2, mddev2) {
2184 if (test_bit(Faulty, &rdev2->flags) ||
2185 test_bit(Journal, &rdev2->flags) ||
2186 rdev2->raid_disk == -1)
2188 if (rdev->bdev->bd_contains ==
2189 rdev2->bdev->bd_contains) {
2199 static LIST_HEAD(pending_raid_disks);
2202 * Try to register data integrity profile for an mddev
2204 * This is called when an array is started and after a disk has been kicked
2205 * from the array. It only succeeds if all working and active component devices
2206 * are integrity capable with matching profiles.
2208 int md_integrity_register(struct mddev *mddev)
2210 struct md_rdev *rdev, *reference = NULL;
2212 if (list_empty(&mddev->disks))
2213 return 0; /* nothing to do */
2214 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2215 return 0; /* shouldn't register, or already is */
2216 rdev_for_each(rdev, mddev) {
2217 /* skip spares and non-functional disks */
2218 if (test_bit(Faulty, &rdev->flags))
2220 if (rdev->raid_disk < 0)
2223 /* Use the first rdev as the reference */
2227 /* does this rdev's profile match the reference profile? */
2228 if (blk_integrity_compare(reference->bdev->bd_disk,
2229 rdev->bdev->bd_disk) < 0)
2232 if (!reference || !bdev_get_integrity(reference->bdev))
2235 * All component devices are integrity capable and have matching
2236 * profiles, register the common profile for the md device.
2238 blk_integrity_register(mddev->gendisk,
2239 bdev_get_integrity(reference->bdev));
2241 pr_debug("md: data integrity enabled on %s\n", mdname(mddev));
2242 if (bioset_integrity_create(&mddev->bio_set, BIO_POOL_SIZE)) {
2243 pr_err("md: failed to create integrity pool for %s\n",
2249 EXPORT_SYMBOL(md_integrity_register);
2252 * Attempt to add an rdev, but only if it is consistent with the current
2255 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2257 struct blk_integrity *bi_mddev;
2258 char name[BDEVNAME_SIZE];
2260 if (!mddev->gendisk)
2263 bi_mddev = blk_get_integrity(mddev->gendisk);
2265 if (!bi_mddev) /* nothing to do */
2268 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2269 pr_err("%s: incompatible integrity profile for %s\n",
2270 mdname(mddev), bdevname(rdev->bdev, name));
2276 EXPORT_SYMBOL(md_integrity_add_rdev);
2278 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2280 char b[BDEVNAME_SIZE];
2284 /* prevent duplicates */
2285 if (find_rdev(mddev, rdev->bdev->bd_dev))
2288 if ((bdev_read_only(rdev->bdev) || bdev_read_only(rdev->meta_bdev)) &&
2292 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2293 if (!test_bit(Journal, &rdev->flags) &&
2295 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2297 /* Cannot change size, so fail
2298 * If mddev->level <= 0, then we don't care
2299 * about aligning sizes (e.g. linear)
2301 if (mddev->level > 0)
2304 mddev->dev_sectors = rdev->sectors;
2307 /* Verify rdev->desc_nr is unique.
2308 * If it is -1, assign a free number, else
2309 * check number is not in use
2312 if (rdev->desc_nr < 0) {
2315 choice = mddev->raid_disks;
2316 while (md_find_rdev_nr_rcu(mddev, choice))
2318 rdev->desc_nr = choice;
2320 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2326 if (!test_bit(Journal, &rdev->flags) &&
2327 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2328 pr_warn("md: %s: array is limited to %d devices\n",
2329 mdname(mddev), mddev->max_disks);
2332 bdevname(rdev->bdev,b);
2333 strreplace(b, '/', '!');
2335 rdev->mddev = mddev;
2336 pr_debug("md: bind<%s>\n", b);
2338 if (mddev->raid_disks)
2339 mddev_create_wb_pool(mddev, rdev, false);
2341 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2344 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2345 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2346 /* failure here is OK */;
2347 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2349 list_add_rcu(&rdev->same_set, &mddev->disks);
2350 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2352 /* May as well allow recovery to be retried once */
2353 mddev->recovery_disabled++;
2358 pr_warn("md: failed to register dev-%s for %s\n",
2363 static void md_delayed_delete(struct work_struct *ws)
2365 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2366 kobject_del(&rdev->kobj);
2367 kobject_put(&rdev->kobj);
2370 static void unbind_rdev_from_array(struct md_rdev *rdev)
2372 char b[BDEVNAME_SIZE];
2374 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2375 list_del_rcu(&rdev->same_set);
2376 pr_debug("md: unbind<%s>\n", bdevname(rdev->bdev,b));
2377 mddev_destroy_wb_pool(rdev->mddev, rdev);
2379 sysfs_remove_link(&rdev->kobj, "block");
2380 sysfs_put(rdev->sysfs_state);
2381 rdev->sysfs_state = NULL;
2382 rdev->badblocks.count = 0;
2383 /* We need to delay this, otherwise we can deadlock when
2384 * writing to 'remove' to "dev/state". We also need
2385 * to delay it due to rcu usage.
2388 INIT_WORK(&rdev->del_work, md_delayed_delete);
2389 kobject_get(&rdev->kobj);
2390 queue_work(md_misc_wq, &rdev->del_work);
2394 * prevent the device from being mounted, repartitioned or
2395 * otherwise reused by a RAID array (or any other kernel
2396 * subsystem), by bd_claiming the device.
2398 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2401 struct block_device *bdev;
2402 char b[BDEVNAME_SIZE];
2404 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2405 shared ? (struct md_rdev *)lock_rdev : rdev);
2407 pr_warn("md: could not open %s.\n", __bdevname(dev, b));
2408 return PTR_ERR(bdev);
2414 static void unlock_rdev(struct md_rdev *rdev)
2416 struct block_device *bdev = rdev->bdev;
2418 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2421 void md_autodetect_dev(dev_t dev);
2423 static void export_rdev(struct md_rdev *rdev)
2425 char b[BDEVNAME_SIZE];
2427 pr_debug("md: export_rdev(%s)\n", bdevname(rdev->bdev,b));
2428 md_rdev_clear(rdev);
2430 if (test_bit(AutoDetected, &rdev->flags))
2431 md_autodetect_dev(rdev->bdev->bd_dev);
2434 kobject_put(&rdev->kobj);
2437 void md_kick_rdev_from_array(struct md_rdev *rdev)
2439 unbind_rdev_from_array(rdev);
2442 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2444 static void export_array(struct mddev *mddev)
2446 struct md_rdev *rdev;
2448 while (!list_empty(&mddev->disks)) {
2449 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2451 md_kick_rdev_from_array(rdev);
2453 mddev->raid_disks = 0;
2454 mddev->major_version = 0;
2457 static bool set_in_sync(struct mddev *mddev)
2459 lockdep_assert_held(&mddev->lock);
2460 if (!mddev->in_sync) {
2461 mddev->sync_checkers++;
2462 spin_unlock(&mddev->lock);
2463 percpu_ref_switch_to_atomic_sync(&mddev->writes_pending);
2464 spin_lock(&mddev->lock);
2465 if (!mddev->in_sync &&
2466 percpu_ref_is_zero(&mddev->writes_pending)) {
2469 * Ensure ->in_sync is visible before we clear
2473 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2474 sysfs_notify_dirent_safe(mddev->sysfs_state);
2476 if (--mddev->sync_checkers == 0)
2477 percpu_ref_switch_to_percpu(&mddev->writes_pending);
2479 if (mddev->safemode == 1)
2480 mddev->safemode = 0;
2481 return mddev->in_sync;
2484 static void sync_sbs(struct mddev *mddev, int nospares)
2486 /* Update each superblock (in-memory image), but
2487 * if we are allowed to, skip spares which already
2488 * have the right event counter, or have one earlier
2489 * (which would mean they aren't being marked as dirty
2490 * with the rest of the array)
2492 struct md_rdev *rdev;
2493 rdev_for_each(rdev, mddev) {
2494 if (rdev->sb_events == mddev->events ||
2496 rdev->raid_disk < 0 &&
2497 rdev->sb_events+1 == mddev->events)) {
2498 /* Don't update this superblock */
2499 rdev->sb_loaded = 2;
2501 sync_super(mddev, rdev);
2502 rdev->sb_loaded = 1;
2507 static bool does_sb_need_changing(struct mddev *mddev)
2509 struct md_rdev *rdev;
2510 struct mdp_superblock_1 *sb;
2513 /* Find a good rdev */
2514 rdev_for_each(rdev, mddev)
2515 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2518 /* No good device found. */
2522 sb = page_address(rdev->sb_page);
2523 /* Check if a device has become faulty or a spare become active */
2524 rdev_for_each(rdev, mddev) {
2525 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2526 /* Device activated? */
2527 if (role == 0xffff && rdev->raid_disk >=0 &&
2528 !test_bit(Faulty, &rdev->flags))
2530 /* Device turned faulty? */
2531 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2535 /* Check if any mddev parameters have changed */
2536 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2537 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2538 (mddev->layout != le32_to_cpu(sb->layout)) ||
2539 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2540 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2546 void md_update_sb(struct mddev *mddev, int force_change)
2548 struct md_rdev *rdev;
2551 int any_badblocks_changed = 0;
2556 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2561 if (mddev_is_clustered(mddev)) {
2562 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2564 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2566 ret = md_cluster_ops->metadata_update_start(mddev);
2567 /* Has someone else has updated the sb */
2568 if (!does_sb_need_changing(mddev)) {
2570 md_cluster_ops->metadata_update_cancel(mddev);
2571 bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2572 BIT(MD_SB_CHANGE_DEVS) |
2573 BIT(MD_SB_CHANGE_CLEAN));
2579 * First make sure individual recovery_offsets are correct
2580 * curr_resync_completed can only be used during recovery.
2581 * During reshape/resync it might use array-addresses rather
2582 * that device addresses.
2584 rdev_for_each(rdev, mddev) {
2585 if (rdev->raid_disk >= 0 &&
2586 mddev->delta_disks >= 0 &&
2587 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
2588 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery) &&
2589 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
2590 !test_bit(Journal, &rdev->flags) &&
2591 !test_bit(In_sync, &rdev->flags) &&
2592 mddev->curr_resync_completed > rdev->recovery_offset)
2593 rdev->recovery_offset = mddev->curr_resync_completed;
2596 if (!mddev->persistent) {
2597 clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
2598 clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2599 if (!mddev->external) {
2600 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
2601 rdev_for_each(rdev, mddev) {
2602 if (rdev->badblocks.changed) {
2603 rdev->badblocks.changed = 0;
2604 ack_all_badblocks(&rdev->badblocks);
2605 md_error(mddev, rdev);
2607 clear_bit(Blocked, &rdev->flags);
2608 clear_bit(BlockedBadBlocks, &rdev->flags);
2609 wake_up(&rdev->blocked_wait);
2612 wake_up(&mddev->sb_wait);
2616 spin_lock(&mddev->lock);
2618 mddev->utime = ktime_get_real_seconds();
2620 if (test_and_clear_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags))
2622 if (test_and_clear_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags))
2623 /* just a clean<-> dirty transition, possibly leave spares alone,
2624 * though if events isn't the right even/odd, we will have to do
2630 if (mddev->degraded)
2631 /* If the array is degraded, then skipping spares is both
2632 * dangerous and fairly pointless.
2633 * Dangerous because a device that was removed from the array
2634 * might have a event_count that still looks up-to-date,
2635 * so it can be re-added without a resync.
2636 * Pointless because if there are any spares to skip,
2637 * then a recovery will happen and soon that array won't
2638 * be degraded any more and the spare can go back to sleep then.
2642 sync_req = mddev->in_sync;
2644 /* If this is just a dirty<->clean transition, and the array is clean
2645 * and 'events' is odd, we can roll back to the previous clean state */
2647 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2648 && mddev->can_decrease_events
2649 && mddev->events != 1) {
2651 mddev->can_decrease_events = 0;
2653 /* otherwise we have to go forward and ... */
2655 mddev->can_decrease_events = nospares;
2659 * This 64-bit counter should never wrap.
2660 * Either we are in around ~1 trillion A.C., assuming
2661 * 1 reboot per second, or we have a bug...
2663 WARN_ON(mddev->events == 0);
2665 rdev_for_each(rdev, mddev) {
2666 if (rdev->badblocks.changed)
2667 any_badblocks_changed++;
2668 if (test_bit(Faulty, &rdev->flags))
2669 set_bit(FaultRecorded, &rdev->flags);
2672 sync_sbs(mddev, nospares);
2673 spin_unlock(&mddev->lock);
2675 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2676 mdname(mddev), mddev->in_sync);
2679 blk_add_trace_msg(mddev->queue, "md md_update_sb");
2681 md_bitmap_update_sb(mddev->bitmap);
2682 rdev_for_each(rdev, mddev) {
2683 char b[BDEVNAME_SIZE];
2685 if (rdev->sb_loaded != 1)
2686 continue; /* no noise on spare devices */
2688 if (!test_bit(Faulty, &rdev->flags)) {
2689 md_super_write(mddev,rdev,
2690 rdev->sb_start, rdev->sb_size,
2692 pr_debug("md: (write) %s's sb offset: %llu\n",
2693 bdevname(rdev->bdev, b),
2694 (unsigned long long)rdev->sb_start);
2695 rdev->sb_events = mddev->events;
2696 if (rdev->badblocks.size) {
2697 md_super_write(mddev, rdev,
2698 rdev->badblocks.sector,
2699 rdev->badblocks.size << 9,
2701 rdev->badblocks.size = 0;
2705 pr_debug("md: %s (skipping faulty)\n",
2706 bdevname(rdev->bdev, b));
2708 if (mddev->level == LEVEL_MULTIPATH)
2709 /* only need to write one superblock... */
2712 if (md_super_wait(mddev) < 0)
2714 /* if there was a failure, MD_SB_CHANGE_DEVS was set, and we re-write super */
2716 if (mddev_is_clustered(mddev) && ret == 0)
2717 md_cluster_ops->metadata_update_finish(mddev);
2719 if (mddev->in_sync != sync_req ||
2720 !bit_clear_unless(&mddev->sb_flags, BIT(MD_SB_CHANGE_PENDING),
2721 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_CLEAN)))
2722 /* have to write it out again */
2724 wake_up(&mddev->sb_wait);
2725 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2726 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2728 rdev_for_each(rdev, mddev) {
2729 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2730 clear_bit(Blocked, &rdev->flags);
2732 if (any_badblocks_changed)
2733 ack_all_badblocks(&rdev->badblocks);
2734 clear_bit(BlockedBadBlocks, &rdev->flags);
2735 wake_up(&rdev->blocked_wait);
2738 EXPORT_SYMBOL(md_update_sb);
2740 static int add_bound_rdev(struct md_rdev *rdev)
2742 struct mddev *mddev = rdev->mddev;
2744 bool add_journal = test_bit(Journal, &rdev->flags);
2746 if (!mddev->pers->hot_remove_disk || add_journal) {
2747 /* If there is hot_add_disk but no hot_remove_disk
2748 * then added disks for geometry changes,
2749 * and should be added immediately.
2751 super_types[mddev->major_version].
2752 validate_super(mddev, rdev);
2754 mddev_suspend(mddev);
2755 err = mddev->pers->hot_add_disk(mddev, rdev);
2757 mddev_resume(mddev);
2759 md_kick_rdev_from_array(rdev);
2763 sysfs_notify_dirent_safe(rdev->sysfs_state);
2765 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2766 if (mddev->degraded)
2767 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2768 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2769 md_new_event(mddev);
2770 md_wakeup_thread(mddev->thread);
2774 /* words written to sysfs files may, or may not, be \n terminated.
2775 * We want to accept with case. For this we use cmd_match.
2777 static int cmd_match(const char *cmd, const char *str)
2779 /* See if cmd, written into a sysfs file, matches
2780 * str. They must either be the same, or cmd can
2781 * have a trailing newline
2783 while (*cmd && *str && *cmd == *str) {
2794 struct rdev_sysfs_entry {
2795 struct attribute attr;
2796 ssize_t (*show)(struct md_rdev *, char *);
2797 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2801 state_show(struct md_rdev *rdev, char *page)
2805 unsigned long flags = READ_ONCE(rdev->flags);
2807 if (test_bit(Faulty, &flags) ||
2808 (!test_bit(ExternalBbl, &flags) &&
2809 rdev->badblocks.unacked_exist))
2810 len += sprintf(page+len, "faulty%s", sep);
2811 if (test_bit(In_sync, &flags))
2812 len += sprintf(page+len, "in_sync%s", sep);
2813 if (test_bit(Journal, &flags))
2814 len += sprintf(page+len, "journal%s", sep);
2815 if (test_bit(WriteMostly, &flags))
2816 len += sprintf(page+len, "write_mostly%s", sep);
2817 if (test_bit(Blocked, &flags) ||
2818 (rdev->badblocks.unacked_exist
2819 && !test_bit(Faulty, &flags)))
2820 len += sprintf(page+len, "blocked%s", sep);
2821 if (!test_bit(Faulty, &flags) &&
2822 !test_bit(Journal, &flags) &&
2823 !test_bit(In_sync, &flags))
2824 len += sprintf(page+len, "spare%s", sep);
2825 if (test_bit(WriteErrorSeen, &flags))
2826 len += sprintf(page+len, "write_error%s", sep);
2827 if (test_bit(WantReplacement, &flags))
2828 len += sprintf(page+len, "want_replacement%s", sep);
2829 if (test_bit(Replacement, &flags))
2830 len += sprintf(page+len, "replacement%s", sep);
2831 if (test_bit(ExternalBbl, &flags))
2832 len += sprintf(page+len, "external_bbl%s", sep);
2833 if (test_bit(FailFast, &flags))
2834 len += sprintf(page+len, "failfast%s", sep);
2839 return len+sprintf(page+len, "\n");
2843 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2846 * faulty - simulates an error
2847 * remove - disconnects the device
2848 * writemostly - sets write_mostly
2849 * -writemostly - clears write_mostly
2850 * blocked - sets the Blocked flags
2851 * -blocked - clears the Blocked and possibly simulates an error
2852 * insync - sets Insync providing device isn't active
2853 * -insync - clear Insync for a device with a slot assigned,
2854 * so that it gets rebuilt based on bitmap
2855 * write_error - sets WriteErrorSeen
2856 * -write_error - clears WriteErrorSeen
2857 * {,-}failfast - set/clear FailFast
2860 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2861 md_error(rdev->mddev, rdev);
2862 if (test_bit(Faulty, &rdev->flags))
2866 } else if (cmd_match(buf, "remove")) {
2867 if (rdev->mddev->pers) {
2868 clear_bit(Blocked, &rdev->flags);
2869 remove_and_add_spares(rdev->mddev, rdev);
2871 if (rdev->raid_disk >= 0)
2874 struct mddev *mddev = rdev->mddev;
2876 if (mddev_is_clustered(mddev))
2877 err = md_cluster_ops->remove_disk(mddev, rdev);
2880 md_kick_rdev_from_array(rdev);
2882 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2883 md_wakeup_thread(mddev->thread);
2885 md_new_event(mddev);
2888 } else if (cmd_match(buf, "writemostly")) {
2889 set_bit(WriteMostly, &rdev->flags);
2890 mddev_create_wb_pool(rdev->mddev, rdev, false);
2892 } else if (cmd_match(buf, "-writemostly")) {
2893 mddev_destroy_wb_pool(rdev->mddev, rdev);
2894 clear_bit(WriteMostly, &rdev->flags);
2896 } else if (cmd_match(buf, "blocked")) {
2897 set_bit(Blocked, &rdev->flags);
2899 } else if (cmd_match(buf, "-blocked")) {
2900 if (!test_bit(Faulty, &rdev->flags) &&
2901 !test_bit(ExternalBbl, &rdev->flags) &&
2902 rdev->badblocks.unacked_exist) {
2903 /* metadata handler doesn't understand badblocks,
2904 * so we need to fail the device
2906 md_error(rdev->mddev, rdev);
2908 clear_bit(Blocked, &rdev->flags);
2909 clear_bit(BlockedBadBlocks, &rdev->flags);
2910 wake_up(&rdev->blocked_wait);
2911 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2912 md_wakeup_thread(rdev->mddev->thread);
2915 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2916 set_bit(In_sync, &rdev->flags);
2918 } else if (cmd_match(buf, "failfast")) {
2919 set_bit(FailFast, &rdev->flags);
2921 } else if (cmd_match(buf, "-failfast")) {
2922 clear_bit(FailFast, &rdev->flags);
2924 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2925 !test_bit(Journal, &rdev->flags)) {
2926 if (rdev->mddev->pers == NULL) {
2927 clear_bit(In_sync, &rdev->flags);
2928 rdev->saved_raid_disk = rdev->raid_disk;
2929 rdev->raid_disk = -1;
2932 } else if (cmd_match(buf, "write_error")) {
2933 set_bit(WriteErrorSeen, &rdev->flags);
2935 } else if (cmd_match(buf, "-write_error")) {
2936 clear_bit(WriteErrorSeen, &rdev->flags);
2938 } else if (cmd_match(buf, "want_replacement")) {
2939 /* Any non-spare device that is not a replacement can
2940 * become want_replacement at any time, but we then need to
2941 * check if recovery is needed.
2943 if (rdev->raid_disk >= 0 &&
2944 !test_bit(Journal, &rdev->flags) &&
2945 !test_bit(Replacement, &rdev->flags))
2946 set_bit(WantReplacement, &rdev->flags);
2947 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2948 md_wakeup_thread(rdev->mddev->thread);
2950 } else if (cmd_match(buf, "-want_replacement")) {
2951 /* Clearing 'want_replacement' is always allowed.
2952 * Once replacements starts it is too late though.
2955 clear_bit(WantReplacement, &rdev->flags);
2956 } else if (cmd_match(buf, "replacement")) {
2957 /* Can only set a device as a replacement when array has not
2958 * yet been started. Once running, replacement is automatic
2959 * from spares, or by assigning 'slot'.
2961 if (rdev->mddev->pers)
2964 set_bit(Replacement, &rdev->flags);
2967 } else if (cmd_match(buf, "-replacement")) {
2968 /* Similarly, can only clear Replacement before start */
2969 if (rdev->mddev->pers)
2972 clear_bit(Replacement, &rdev->flags);
2975 } else if (cmd_match(buf, "re-add")) {
2976 if (!rdev->mddev->pers)
2978 else if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1) &&
2979 rdev->saved_raid_disk >= 0) {
2980 /* clear_bit is performed _after_ all the devices
2981 * have their local Faulty bit cleared. If any writes
2982 * happen in the meantime in the local node, they
2983 * will land in the local bitmap, which will be synced
2984 * by this node eventually
2986 if (!mddev_is_clustered(rdev->mddev) ||
2987 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2988 clear_bit(Faulty, &rdev->flags);
2989 err = add_bound_rdev(rdev);
2993 } else if (cmd_match(buf, "external_bbl") && (rdev->mddev->external)) {
2994 set_bit(ExternalBbl, &rdev->flags);
2995 rdev->badblocks.shift = 0;
2997 } else if (cmd_match(buf, "-external_bbl") && (rdev->mddev->external)) {
2998 clear_bit(ExternalBbl, &rdev->flags);
3002 sysfs_notify_dirent_safe(rdev->sysfs_state);
3003 return err ? err : len;
3005 static struct rdev_sysfs_entry rdev_state =
3006 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
3009 errors_show(struct md_rdev *rdev, char *page)
3011 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
3015 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
3020 rv = kstrtouint(buf, 10, &n);
3023 atomic_set(&rdev->corrected_errors, n);
3026 static struct rdev_sysfs_entry rdev_errors =
3027 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
3030 slot_show(struct md_rdev *rdev, char *page)
3032 if (test_bit(Journal, &rdev->flags))
3033 return sprintf(page, "journal\n");
3034 else if (rdev->raid_disk < 0)
3035 return sprintf(page, "none\n");
3037 return sprintf(page, "%d\n", rdev->raid_disk);
3041 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
3046 if (test_bit(Journal, &rdev->flags))
3048 if (strncmp(buf, "none", 4)==0)
3051 err = kstrtouint(buf, 10, (unsigned int *)&slot);
3055 if (rdev->mddev->pers && slot == -1) {
3056 /* Setting 'slot' on an active array requires also
3057 * updating the 'rd%d' link, and communicating
3058 * with the personality with ->hot_*_disk.
3059 * For now we only support removing
3060 * failed/spare devices. This normally happens automatically,
3061 * but not when the metadata is externally managed.
3063 if (rdev->raid_disk == -1)
3065 /* personality does all needed checks */
3066 if (rdev->mddev->pers->hot_remove_disk == NULL)
3068 clear_bit(Blocked, &rdev->flags);
3069 remove_and_add_spares(rdev->mddev, rdev);
3070 if (rdev->raid_disk >= 0)
3072 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
3073 md_wakeup_thread(rdev->mddev->thread);
3074 } else if (rdev->mddev->pers) {
3075 /* Activating a spare .. or possibly reactivating
3076 * if we ever get bitmaps working here.
3080 if (rdev->raid_disk != -1)
3083 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
3086 if (rdev->mddev->pers->hot_add_disk == NULL)
3089 if (slot >= rdev->mddev->raid_disks &&
3090 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3093 rdev->raid_disk = slot;
3094 if (test_bit(In_sync, &rdev->flags))
3095 rdev->saved_raid_disk = slot;
3097 rdev->saved_raid_disk = -1;
3098 clear_bit(In_sync, &rdev->flags);
3099 clear_bit(Bitmap_sync, &rdev->flags);
3100 err = rdev->mddev->pers->
3101 hot_add_disk(rdev->mddev, rdev);
3103 rdev->raid_disk = -1;
3106 sysfs_notify_dirent_safe(rdev->sysfs_state);
3107 if (sysfs_link_rdev(rdev->mddev, rdev))
3108 /* failure here is OK */;
3109 /* don't wakeup anyone, leave that to userspace. */
3111 if (slot >= rdev->mddev->raid_disks &&
3112 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
3114 rdev->raid_disk = slot;
3115 /* assume it is working */
3116 clear_bit(Faulty, &rdev->flags);
3117 clear_bit(WriteMostly, &rdev->flags);
3118 set_bit(In_sync, &rdev->flags);
3119 sysfs_notify_dirent_safe(rdev->sysfs_state);
3124 static struct rdev_sysfs_entry rdev_slot =
3125 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
3128 offset_show(struct md_rdev *rdev, char *page)
3130 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
3134 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
3136 unsigned long long offset;
3137 if (kstrtoull(buf, 10, &offset) < 0)
3139 if (rdev->mddev->pers && rdev->raid_disk >= 0)
3141 if (rdev->sectors && rdev->mddev->external)
3142 /* Must set offset before size, so overlap checks
3145 rdev->data_offset = offset;
3146 rdev->new_data_offset = offset;
3150 static struct rdev_sysfs_entry rdev_offset =
3151 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
3153 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
3155 return sprintf(page, "%llu\n",
3156 (unsigned long long)rdev->new_data_offset);
3159 static ssize_t new_offset_store(struct md_rdev *rdev,
3160 const char *buf, size_t len)
3162 unsigned long long new_offset;
3163 struct mddev *mddev = rdev->mddev;
3165 if (kstrtoull(buf, 10, &new_offset) < 0)
3168 if (mddev->sync_thread ||
3169 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
3171 if (new_offset == rdev->data_offset)
3172 /* reset is always permitted */
3174 else if (new_offset > rdev->data_offset) {
3175 /* must not push array size beyond rdev_sectors */
3176 if (new_offset - rdev->data_offset
3177 + mddev->dev_sectors > rdev->sectors)
3180 /* Metadata worries about other space details. */
3182 /* decreasing the offset is inconsistent with a backwards
3185 if (new_offset < rdev->data_offset &&
3186 mddev->reshape_backwards)
3188 /* Increasing offset is inconsistent with forwards
3189 * reshape. reshape_direction should be set to
3190 * 'backwards' first.
3192 if (new_offset > rdev->data_offset &&
3193 !mddev->reshape_backwards)
3196 if (mddev->pers && mddev->persistent &&
3197 !super_types[mddev->major_version]
3198 .allow_new_offset(rdev, new_offset))
3200 rdev->new_data_offset = new_offset;
3201 if (new_offset > rdev->data_offset)
3202 mddev->reshape_backwards = 1;
3203 else if (new_offset < rdev->data_offset)
3204 mddev->reshape_backwards = 0;
3208 static struct rdev_sysfs_entry rdev_new_offset =
3209 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
3212 rdev_size_show(struct md_rdev *rdev, char *page)
3214 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
3217 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
3219 /* check if two start/length pairs overlap */
3227 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
3229 unsigned long long blocks;
3232 if (kstrtoull(buf, 10, &blocks) < 0)
3235 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
3236 return -EINVAL; /* sector conversion overflow */
3239 if (new != blocks * 2)
3240 return -EINVAL; /* unsigned long long to sector_t overflow */
3247 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3249 struct mddev *my_mddev = rdev->mddev;
3250 sector_t oldsectors = rdev->sectors;
3253 if (test_bit(Journal, &rdev->flags))
3255 if (strict_blocks_to_sectors(buf, §ors) < 0)
3257 if (rdev->data_offset != rdev->new_data_offset)
3258 return -EINVAL; /* too confusing */
3259 if (my_mddev->pers && rdev->raid_disk >= 0) {
3260 if (my_mddev->persistent) {
3261 sectors = super_types[my_mddev->major_version].
3262 rdev_size_change(rdev, sectors);
3265 } else if (!sectors)
3266 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3268 if (!my_mddev->pers->resize)
3269 /* Cannot change size for RAID0 or Linear etc */
3272 if (sectors < my_mddev->dev_sectors)
3273 return -EINVAL; /* component must fit device */
3275 rdev->sectors = sectors;
3276 if (sectors > oldsectors && my_mddev->external) {
3277 /* Need to check that all other rdevs with the same
3278 * ->bdev do not overlap. 'rcu' is sufficient to walk
3279 * the rdev lists safely.
3280 * This check does not provide a hard guarantee, it
3281 * just helps avoid dangerous mistakes.
3283 struct mddev *mddev;
3285 struct list_head *tmp;
3288 for_each_mddev(mddev, tmp) {
3289 struct md_rdev *rdev2;
3291 rdev_for_each(rdev2, mddev)
3292 if (rdev->bdev == rdev2->bdev &&
3294 overlaps(rdev->data_offset, rdev->sectors,
3307 /* Someone else could have slipped in a size
3308 * change here, but doing so is just silly.
3309 * We put oldsectors back because we *know* it is
3310 * safe, and trust userspace not to race with
3313 rdev->sectors = oldsectors;
3320 static struct rdev_sysfs_entry rdev_size =
3321 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3323 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3325 unsigned long long recovery_start = rdev->recovery_offset;
3327 if (test_bit(In_sync, &rdev->flags) ||
3328 recovery_start == MaxSector)
3329 return sprintf(page, "none\n");
3331 return sprintf(page, "%llu\n", recovery_start);
3334 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3336 unsigned long long recovery_start;
3338 if (cmd_match(buf, "none"))
3339 recovery_start = MaxSector;
3340 else if (kstrtoull(buf, 10, &recovery_start))
3343 if (rdev->mddev->pers &&
3344 rdev->raid_disk >= 0)
3347 rdev->recovery_offset = recovery_start;
3348 if (recovery_start == MaxSector)
3349 set_bit(In_sync, &rdev->flags);
3351 clear_bit(In_sync, &rdev->flags);
3355 static struct rdev_sysfs_entry rdev_recovery_start =
3356 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3358 /* sysfs access to bad-blocks list.
3359 * We present two files.
3360 * 'bad-blocks' lists sector numbers and lengths of ranges that
3361 * are recorded as bad. The list is truncated to fit within
3362 * the one-page limit of sysfs.
3363 * Writing "sector length" to this file adds an acknowledged
3365 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3366 * been acknowledged. Writing to this file adds bad blocks
3367 * without acknowledging them. This is largely for testing.
3369 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3371 return badblocks_show(&rdev->badblocks, page, 0);
3373 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3375 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3376 /* Maybe that ack was all we needed */
3377 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3378 wake_up(&rdev->blocked_wait);
3381 static struct rdev_sysfs_entry rdev_bad_blocks =
3382 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3384 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3386 return badblocks_show(&rdev->badblocks, page, 1);
3388 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3390 return badblocks_store(&rdev->badblocks, page, len, 1);
3392 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3393 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3396 ppl_sector_show(struct md_rdev *rdev, char *page)
3398 return sprintf(page, "%llu\n", (unsigned long long)rdev->ppl.sector);
3402 ppl_sector_store(struct md_rdev *rdev, const char *buf, size_t len)
3404 unsigned long long sector;
3406 if (kstrtoull(buf, 10, §or) < 0)
3408 if (sector != (sector_t)sector)
3411 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3412 rdev->raid_disk >= 0)
3415 if (rdev->mddev->persistent) {
3416 if (rdev->mddev->major_version == 0)
3418 if ((sector > rdev->sb_start &&
3419 sector - rdev->sb_start > S16_MAX) ||
3420 (sector < rdev->sb_start &&
3421 rdev->sb_start - sector > -S16_MIN))
3423 rdev->ppl.offset = sector - rdev->sb_start;
3424 } else if (!rdev->mddev->external) {
3427 rdev->ppl.sector = sector;
3431 static struct rdev_sysfs_entry rdev_ppl_sector =
3432 __ATTR(ppl_sector, S_IRUGO|S_IWUSR, ppl_sector_show, ppl_sector_store);
3435 ppl_size_show(struct md_rdev *rdev, char *page)
3437 return sprintf(page, "%u\n", rdev->ppl.size);
3441 ppl_size_store(struct md_rdev *rdev, const char *buf, size_t len)
3445 if (kstrtouint(buf, 10, &size) < 0)
3448 if (rdev->mddev->pers && test_bit(MD_HAS_PPL, &rdev->mddev->flags) &&
3449 rdev->raid_disk >= 0)
3452 if (rdev->mddev->persistent) {
3453 if (rdev->mddev->major_version == 0)
3457 } else if (!rdev->mddev->external) {
3460 rdev->ppl.size = size;
3464 static struct rdev_sysfs_entry rdev_ppl_size =
3465 __ATTR(ppl_size, S_IRUGO|S_IWUSR, ppl_size_show, ppl_size_store);
3467 static struct attribute *rdev_default_attrs[] = {
3472 &rdev_new_offset.attr,
3474 &rdev_recovery_start.attr,
3475 &rdev_bad_blocks.attr,
3476 &rdev_unack_bad_blocks.attr,
3477 &rdev_ppl_sector.attr,
3478 &rdev_ppl_size.attr,
3482 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3484 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3485 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3491 return entry->show(rdev, page);
3495 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3496 const char *page, size_t length)
3498 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3499 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3501 struct mddev *mddev = rdev->mddev;
3505 if (!capable(CAP_SYS_ADMIN))
3507 rv = mddev ? mddev_lock(mddev) : -ENODEV;
3509 if (rdev->mddev == NULL)
3512 rv = entry->store(rdev, page, length);
3513 mddev_unlock(mddev);
3518 static void rdev_free(struct kobject *ko)
3520 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3523 static const struct sysfs_ops rdev_sysfs_ops = {
3524 .show = rdev_attr_show,
3525 .store = rdev_attr_store,
3527 static struct kobj_type rdev_ktype = {
3528 .release = rdev_free,
3529 .sysfs_ops = &rdev_sysfs_ops,
3530 .default_attrs = rdev_default_attrs,
3533 int md_rdev_init(struct md_rdev *rdev)
3536 rdev->saved_raid_disk = -1;
3537 rdev->raid_disk = -1;
3539 rdev->data_offset = 0;
3540 rdev->new_data_offset = 0;
3541 rdev->sb_events = 0;
3542 rdev->last_read_error = 0;
3543 rdev->sb_loaded = 0;
3544 rdev->bb_page = NULL;
3545 atomic_set(&rdev->nr_pending, 0);
3546 atomic_set(&rdev->read_errors, 0);
3547 atomic_set(&rdev->corrected_errors, 0);
3549 INIT_LIST_HEAD(&rdev->same_set);
3550 init_waitqueue_head(&rdev->blocked_wait);
3552 /* Add space to store bad block list.
3553 * This reserves the space even on arrays where it cannot
3554 * be used - I wonder if that matters
3556 return badblocks_init(&rdev->badblocks, 0);
3558 EXPORT_SYMBOL_GPL(md_rdev_init);
3560 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3562 * mark the device faulty if:
3564 * - the device is nonexistent (zero size)
3565 * - the device has no valid superblock
3567 * a faulty rdev _never_ has rdev->sb set.
3569 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3571 char b[BDEVNAME_SIZE];
3573 struct md_rdev *rdev;
3576 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3578 return ERR_PTR(-ENOMEM);
3580 err = md_rdev_init(rdev);
3583 err = alloc_disk_sb(rdev);
3587 err = lock_rdev(rdev, newdev, super_format == -2);
3591 kobject_init(&rdev->kobj, &rdev_ktype);
3593 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3595 pr_warn("md: %s has zero or unknown size, marking faulty!\n",
3596 bdevname(rdev->bdev,b));
3601 if (super_format >= 0) {
3602 err = super_types[super_format].
3603 load_super(rdev, NULL, super_minor);
3604 if (err == -EINVAL) {
3605 pr_warn("md: %s does not have a valid v%d.%d superblock, not importing!\n",
3606 bdevname(rdev->bdev,b),
3607 super_format, super_minor);
3611 pr_warn("md: could not read %s's sb, not importing!\n",
3612 bdevname(rdev->bdev,b));
3622 md_rdev_clear(rdev);
3624 return ERR_PTR(err);
3628 * Check a full RAID array for plausibility
3631 static int analyze_sbs(struct mddev *mddev)
3634 struct md_rdev *rdev, *freshest, *tmp;
3635 char b[BDEVNAME_SIZE];
3638 rdev_for_each_safe(rdev, tmp, mddev)
3639 switch (super_types[mddev->major_version].
3640 load_super(rdev, freshest, mddev->minor_version)) {
3647 pr_warn("md: fatal superblock inconsistency in %s -- removing from array\n",
3648 bdevname(rdev->bdev,b));
3649 md_kick_rdev_from_array(rdev);
3652 /* Cannot find a valid fresh disk */
3654 pr_warn("md: cannot find a valid disk\n");
3658 super_types[mddev->major_version].
3659 validate_super(mddev, freshest);
3662 rdev_for_each_safe(rdev, tmp, mddev) {
3663 if (mddev->max_disks &&
3664 (rdev->desc_nr >= mddev->max_disks ||
3665 i > mddev->max_disks)) {
3666 pr_warn("md: %s: %s: only %d devices permitted\n",
3667 mdname(mddev), bdevname(rdev->bdev, b),
3669 md_kick_rdev_from_array(rdev);
3672 if (rdev != freshest) {
3673 if (super_types[mddev->major_version].
3674 validate_super(mddev, rdev)) {
3675 pr_warn("md: kicking non-fresh %s from array!\n",
3676 bdevname(rdev->bdev,b));
3677 md_kick_rdev_from_array(rdev);
3681 if (mddev->level == LEVEL_MULTIPATH) {
3682 rdev->desc_nr = i++;
3683 rdev->raid_disk = rdev->desc_nr;
3684 set_bit(In_sync, &rdev->flags);
3685 } else if (rdev->raid_disk >=
3686 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3687 !test_bit(Journal, &rdev->flags)) {
3688 rdev->raid_disk = -1;
3689 clear_bit(In_sync, &rdev->flags);
3696 /* Read a fixed-point number.
3697 * Numbers in sysfs attributes should be in "standard" units where
3698 * possible, so time should be in seconds.
3699 * However we internally use a a much smaller unit such as
3700 * milliseconds or jiffies.
3701 * This function takes a decimal number with a possible fractional
3702 * component, and produces an integer which is the result of
3703 * multiplying that number by 10^'scale'.
3704 * all without any floating-point arithmetic.
3706 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3708 unsigned long result = 0;
3710 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3713 else if (decimals < scale) {
3716 result = result * 10 + value;
3728 *res = result * int_pow(10, scale - decimals);
3733 safe_delay_show(struct mddev *mddev, char *page)
3735 int msec = (mddev->safemode_delay*1000)/HZ;
3736 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3739 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3743 if (mddev_is_clustered(mddev)) {
3744 pr_warn("md: Safemode is disabled for clustered mode\n");
3748 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3751 mddev->safemode_delay = 0;
3753 unsigned long old_delay = mddev->safemode_delay;
3754 unsigned long new_delay = (msec*HZ)/1000;
3758 mddev->safemode_delay = new_delay;
3759 if (new_delay < old_delay || old_delay == 0)
3760 mod_timer(&mddev->safemode_timer, jiffies+1);
3764 static struct md_sysfs_entry md_safe_delay =
3765 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3768 level_show(struct mddev *mddev, char *page)
3770 struct md_personality *p;
3772 spin_lock(&mddev->lock);
3775 ret = sprintf(page, "%s\n", p->name);
3776 else if (mddev->clevel[0])
3777 ret = sprintf(page, "%s\n", mddev->clevel);
3778 else if (mddev->level != LEVEL_NONE)
3779 ret = sprintf(page, "%d\n", mddev->level);
3782 spin_unlock(&mddev->lock);
3787 level_store(struct mddev *mddev, const char *buf, size_t len)
3792 struct md_personality *pers, *oldpers;
3794 void *priv, *oldpriv;
3795 struct md_rdev *rdev;
3797 if (slen == 0 || slen >= sizeof(clevel))
3800 rv = mddev_lock(mddev);
3804 if (mddev->pers == NULL) {
3805 strncpy(mddev->clevel, buf, slen);
3806 if (mddev->clevel[slen-1] == '\n')
3808 mddev->clevel[slen] = 0;
3809 mddev->level = LEVEL_NONE;
3817 /* request to change the personality. Need to ensure:
3818 * - array is not engaged in resync/recovery/reshape
3819 * - old personality can be suspended
3820 * - new personality will access other array.
3824 if (mddev->sync_thread ||
3825 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3826 mddev->reshape_position != MaxSector ||
3827 mddev->sysfs_active)
3831 if (!mddev->pers->quiesce) {
3832 pr_warn("md: %s: %s does not support online personality change\n",
3833 mdname(mddev), mddev->pers->name);
3837 /* Now find the new personality */
3838 strncpy(clevel, buf, slen);
3839 if (clevel[slen-1] == '\n')
3842 if (kstrtol(clevel, 10, &level))
3845 if (request_module("md-%s", clevel) != 0)
3846 request_module("md-level-%s", clevel);
3847 spin_lock(&pers_lock);
3848 pers = find_pers(level, clevel);
3849 if (!pers || !try_module_get(pers->owner)) {
3850 spin_unlock(&pers_lock);
3851 pr_warn("md: personality %s not loaded\n", clevel);
3855 spin_unlock(&pers_lock);
3857 if (pers == mddev->pers) {
3858 /* Nothing to do! */
3859 module_put(pers->owner);
3863 if (!pers->takeover) {
3864 module_put(pers->owner);
3865 pr_warn("md: %s: %s does not support personality takeover\n",
3866 mdname(mddev), clevel);
3871 rdev_for_each(rdev, mddev)
3872 rdev->new_raid_disk = rdev->raid_disk;
3874 /* ->takeover must set new_* and/or delta_disks
3875 * if it succeeds, and may set them when it fails.
3877 priv = pers->takeover(mddev);
3879 mddev->new_level = mddev->level;
3880 mddev->new_layout = mddev->layout;
3881 mddev->new_chunk_sectors = mddev->chunk_sectors;
3882 mddev->raid_disks -= mddev->delta_disks;
3883 mddev->delta_disks = 0;
3884 mddev->reshape_backwards = 0;
3885 module_put(pers->owner);
3886 pr_warn("md: %s: %s would not accept array\n",
3887 mdname(mddev), clevel);
3892 /* Looks like we have a winner */
3893 mddev_suspend(mddev);
3894 mddev_detach(mddev);
3896 spin_lock(&mddev->lock);
3897 oldpers = mddev->pers;
3898 oldpriv = mddev->private;
3900 mddev->private = priv;
3901 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3902 mddev->level = mddev->new_level;
3903 mddev->layout = mddev->new_layout;
3904 mddev->chunk_sectors = mddev->new_chunk_sectors;
3905 mddev->delta_disks = 0;
3906 mddev->reshape_backwards = 0;
3907 mddev->degraded = 0;
3908 spin_unlock(&mddev->lock);
3910 if (oldpers->sync_request == NULL &&
3912 /* We are converting from a no-redundancy array
3913 * to a redundancy array and metadata is managed
3914 * externally so we need to be sure that writes
3915 * won't block due to a need to transition
3917 * until external management is started.
3920 mddev->safemode_delay = 0;
3921 mddev->safemode = 0;
3924 oldpers->free(mddev, oldpriv);
3926 if (oldpers->sync_request == NULL &&
3927 pers->sync_request != NULL) {
3928 /* need to add the md_redundancy_group */
3929 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3930 pr_warn("md: cannot register extra attributes for %s\n",
3932 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3934 if (oldpers->sync_request != NULL &&
3935 pers->sync_request == NULL) {
3936 /* need to remove the md_redundancy_group */
3937 if (mddev->to_remove == NULL)
3938 mddev->to_remove = &md_redundancy_group;
3941 module_put(oldpers->owner);
3943 rdev_for_each(rdev, mddev) {
3944 if (rdev->raid_disk < 0)
3946 if (rdev->new_raid_disk >= mddev->raid_disks)
3947 rdev->new_raid_disk = -1;
3948 if (rdev->new_raid_disk == rdev->raid_disk)
3950 sysfs_unlink_rdev(mddev, rdev);
3952 rdev_for_each(rdev, mddev) {
3953 if (rdev->raid_disk < 0)
3955 if (rdev->new_raid_disk == rdev->raid_disk)
3957 rdev->raid_disk = rdev->new_raid_disk;
3958 if (rdev->raid_disk < 0)
3959 clear_bit(In_sync, &rdev->flags);
3961 if (sysfs_link_rdev(mddev, rdev))
3962 pr_warn("md: cannot register rd%d for %s after level change\n",
3963 rdev->raid_disk, mdname(mddev));
3967 if (pers->sync_request == NULL) {
3968 /* this is now an array without redundancy, so
3969 * it must always be in_sync
3972 del_timer_sync(&mddev->safemode_timer);
3974 blk_set_stacking_limits(&mddev->queue->limits);
3976 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
3977 mddev_resume(mddev);
3979 md_update_sb(mddev, 1);
3980 sysfs_notify(&mddev->kobj, NULL, "level");
3981 md_new_event(mddev);
3984 mddev_unlock(mddev);
3988 static struct md_sysfs_entry md_level =
3989 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3992 layout_show(struct mddev *mddev, char *page)
3994 /* just a number, not meaningful for all levels */
3995 if (mddev->reshape_position != MaxSector &&
3996 mddev->layout != mddev->new_layout)
3997 return sprintf(page, "%d (%d)\n",
3998 mddev->new_layout, mddev->layout);
3999 return sprintf(page, "%d\n", mddev->layout);
4003 layout_store(struct mddev *mddev, const char *buf, size_t len)
4008 err = kstrtouint(buf, 10, &n);
4011 err = mddev_lock(mddev);
4016 if (mddev->pers->check_reshape == NULL)
4021 mddev->new_layout = n;
4022 err = mddev->pers->check_reshape(mddev);
4024 mddev->new_layout = mddev->layout;
4027 mddev->new_layout = n;
4028 if (mddev->reshape_position == MaxSector)
4031 mddev_unlock(mddev);
4034 static struct md_sysfs_entry md_layout =
4035 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
4038 raid_disks_show(struct mddev *mddev, char *page)
4040 if (mddev->raid_disks == 0)
4042 if (mddev->reshape_position != MaxSector &&
4043 mddev->delta_disks != 0)
4044 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
4045 mddev->raid_disks - mddev->delta_disks);
4046 return sprintf(page, "%d\n", mddev->raid_disks);
4049 static int update_raid_disks(struct mddev *mddev, int raid_disks);
4052 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
4057 err = kstrtouint(buf, 10, &n);
4061 err = mddev_lock(mddev);
4065 err = update_raid_disks(mddev, n);
4066 else if (mddev->reshape_position != MaxSector) {
4067 struct md_rdev *rdev;
4068 int olddisks = mddev->raid_disks - mddev->delta_disks;
4071 rdev_for_each(rdev, mddev) {
4073 rdev->data_offset < rdev->new_data_offset)
4076 rdev->data_offset > rdev->new_data_offset)
4080 mddev->delta_disks = n - olddisks;
4081 mddev->raid_disks = n;
4082 mddev->reshape_backwards = (mddev->delta_disks < 0);
4084 mddev->raid_disks = n;
4086 mddev_unlock(mddev);
4087 return err ? err : len;
4089 static struct md_sysfs_entry md_raid_disks =
4090 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
4093 chunk_size_show(struct mddev *mddev, char *page)
4095 if (mddev->reshape_position != MaxSector &&
4096 mddev->chunk_sectors != mddev->new_chunk_sectors)
4097 return sprintf(page, "%d (%d)\n",
4098 mddev->new_chunk_sectors << 9,
4099 mddev->chunk_sectors << 9);
4100 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
4104 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
4109 err = kstrtoul(buf, 10, &n);
4113 err = mddev_lock(mddev);
4117 if (mddev->pers->check_reshape == NULL)
4122 mddev->new_chunk_sectors = n >> 9;
4123 err = mddev->pers->check_reshape(mddev);
4125 mddev->new_chunk_sectors = mddev->chunk_sectors;
4128 mddev->new_chunk_sectors = n >> 9;
4129 if (mddev->reshape_position == MaxSector)
4130 mddev->chunk_sectors = n >> 9;
4132 mddev_unlock(mddev);
4135 static struct md_sysfs_entry md_chunk_size =
4136 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
4139 resync_start_show(struct mddev *mddev, char *page)
4141 if (mddev->recovery_cp == MaxSector)
4142 return sprintf(page, "none\n");
4143 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
4147 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
4149 unsigned long long n;
4152 if (cmd_match(buf, "none"))
4155 err = kstrtoull(buf, 10, &n);
4158 if (n != (sector_t)n)
4162 err = mddev_lock(mddev);
4165 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4169 mddev->recovery_cp = n;
4171 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
4173 mddev_unlock(mddev);
4176 static struct md_sysfs_entry md_resync_start =
4177 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
4178 resync_start_show, resync_start_store);
4181 * The array state can be:
4184 * No devices, no size, no level
4185 * Equivalent to STOP_ARRAY ioctl
4187 * May have some settings, but array is not active
4188 * all IO results in error
4189 * When written, doesn't tear down array, but just stops it
4190 * suspended (not supported yet)
4191 * All IO requests will block. The array can be reconfigured.
4192 * Writing this, if accepted, will block until array is quiescent
4194 * no resync can happen. no superblocks get written.
4195 * write requests fail
4197 * like readonly, but behaves like 'clean' on a write request.
4199 * clean - no pending writes, but otherwise active.
4200 * When written to inactive array, starts without resync
4201 * If a write request arrives then
4202 * if metadata is known, mark 'dirty' and switch to 'active'.
4203 * if not known, block and switch to write-pending
4204 * If written to an active array that has pending writes, then fails.
4206 * fully active: IO and resync can be happening.
4207 * When written to inactive array, starts with resync
4210 * clean, but writes are blocked waiting for 'active' to be written.
4213 * like active, but no writes have been seen for a while (100msec).
4216 * RAID0/LINEAR-only: same as clean, but array is missing a member.
4217 * It's useful because RAID0/LINEAR mounted-arrays aren't stopped
4218 * when a member is gone, so this state will at least alert the
4219 * user that something is wrong.
4221 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
4222 write_pending, active_idle, broken, bad_word};
4223 static char *array_states[] = {
4224 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
4225 "write-pending", "active-idle", "broken", NULL };
4227 static int match_word(const char *word, char **list)
4230 for (n=0; list[n]; n++)
4231 if (cmd_match(word, list[n]))
4237 array_state_show(struct mddev *mddev, char *page)
4239 enum array_state st = inactive;
4241 if (mddev->pers && !test_bit(MD_NOT_READY, &mddev->flags)) {
4250 spin_lock(&mddev->lock);
4251 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
4253 else if (mddev->in_sync)
4255 else if (mddev->safemode)
4259 spin_unlock(&mddev->lock);
4262 if (test_bit(MD_BROKEN, &mddev->flags) && st == clean)
4265 if (list_empty(&mddev->disks) &&
4266 mddev->raid_disks == 0 &&
4267 mddev->dev_sectors == 0)
4272 return sprintf(page, "%s\n", array_states[st]);
4275 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
4276 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
4277 static int do_md_run(struct mddev *mddev);
4278 static int restart_array(struct mddev *mddev);
4281 array_state_store(struct mddev *mddev, const char *buf, size_t len)
4284 enum array_state st = match_word(buf, array_states);
4286 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
4287 /* don't take reconfig_mutex when toggling between
4290 spin_lock(&mddev->lock);
4292 restart_array(mddev);
4293 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4294 md_wakeup_thread(mddev->thread);
4295 wake_up(&mddev->sb_wait);
4296 } else /* st == clean */ {
4297 restart_array(mddev);
4298 if (!set_in_sync(mddev))
4302 sysfs_notify_dirent_safe(mddev->sysfs_state);
4303 spin_unlock(&mddev->lock);
4306 err = mddev_lock(mddev);
4314 /* stopping an active array */
4315 err = do_md_stop(mddev, 0, NULL);
4318 /* stopping an active array */
4320 err = do_md_stop(mddev, 2, NULL);
4322 err = 0; /* already inactive */
4325 break; /* not supported yet */
4328 err = md_set_readonly(mddev, NULL);
4331 set_disk_ro(mddev->gendisk, 1);
4332 err = do_md_run(mddev);
4338 err = md_set_readonly(mddev, NULL);
4339 else if (mddev->ro == 1)
4340 err = restart_array(mddev);
4343 set_disk_ro(mddev->gendisk, 0);
4347 err = do_md_run(mddev);
4352 err = restart_array(mddev);
4355 spin_lock(&mddev->lock);
4356 if (!set_in_sync(mddev))
4358 spin_unlock(&mddev->lock);
4364 err = restart_array(mddev);
4367 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
4368 wake_up(&mddev->sb_wait);
4372 set_disk_ro(mddev->gendisk, 0);
4373 err = do_md_run(mddev);
4379 /* these cannot be set */
4384 if (mddev->hold_active == UNTIL_IOCTL)
4385 mddev->hold_active = 0;
4386 sysfs_notify_dirent_safe(mddev->sysfs_state);
4388 mddev_unlock(mddev);
4391 static struct md_sysfs_entry md_array_state =
4392 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4395 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4396 return sprintf(page, "%d\n",
4397 atomic_read(&mddev->max_corr_read_errors));
4401 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4406 rv = kstrtouint(buf, 10, &n);
4409 atomic_set(&mddev->max_corr_read_errors, n);
4413 static struct md_sysfs_entry max_corr_read_errors =
4414 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4415 max_corrected_read_errors_store);
4418 null_show(struct mddev *mddev, char *page)
4424 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4426 /* buf must be %d:%d\n? giving major and minor numbers */
4427 /* The new device is added to the array.
4428 * If the array has a persistent superblock, we read the
4429 * superblock to initialise info and check validity.
4430 * Otherwise, only checking done is that in bind_rdev_to_array,
4431 * which mainly checks size.
4434 int major = simple_strtoul(buf, &e, 10);
4437 struct md_rdev *rdev;
4440 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4442 minor = simple_strtoul(e+1, &e, 10);
4443 if (*e && *e != '\n')
4445 dev = MKDEV(major, minor);
4446 if (major != MAJOR(dev) ||
4447 minor != MINOR(dev))
4450 flush_workqueue(md_misc_wq);
4452 err = mddev_lock(mddev);
4455 if (mddev->persistent) {
4456 rdev = md_import_device(dev, mddev->major_version,
4457 mddev->minor_version);
4458 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4459 struct md_rdev *rdev0
4460 = list_entry(mddev->disks.next,
4461 struct md_rdev, same_set);
4462 err = super_types[mddev->major_version]
4463 .load_super(rdev, rdev0, mddev->minor_version);
4467 } else if (mddev->external)
4468 rdev = md_import_device(dev, -2, -1);
4470 rdev = md_import_device(dev, -1, -1);
4473 mddev_unlock(mddev);
4474 return PTR_ERR(rdev);
4476 err = bind_rdev_to_array(rdev, mddev);
4480 mddev_unlock(mddev);
4482 md_new_event(mddev);
4483 return err ? err : len;
4486 static struct md_sysfs_entry md_new_device =
4487 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4490 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4493 unsigned long chunk, end_chunk;
4496 err = mddev_lock(mddev);
4501 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4503 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4504 if (buf == end) break;
4505 if (*end == '-') { /* range */
4507 end_chunk = simple_strtoul(buf, &end, 0);
4508 if (buf == end) break;
4510 if (*end && !isspace(*end)) break;
4511 md_bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4512 buf = skip_spaces(end);
4514 md_bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4516 mddev_unlock(mddev);
4520 static struct md_sysfs_entry md_bitmap =
4521 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4524 size_show(struct mddev *mddev, char *page)
4526 return sprintf(page, "%llu\n",
4527 (unsigned long long)mddev->dev_sectors / 2);
4530 static int update_size(struct mddev *mddev, sector_t num_sectors);
4533 size_store(struct mddev *mddev, const char *buf, size_t len)
4535 /* If array is inactive, we can reduce the component size, but
4536 * not increase it (except from 0).
4537 * If array is active, we can try an on-line resize
4540 int err = strict_blocks_to_sectors(buf, §ors);
4544 err = mddev_lock(mddev);
4548 err = update_size(mddev, sectors);
4550 md_update_sb(mddev, 1);
4552 if (mddev->dev_sectors == 0 ||
4553 mddev->dev_sectors > sectors)
4554 mddev->dev_sectors = sectors;
4558 mddev_unlock(mddev);
4559 return err ? err : len;
4562 static struct md_sysfs_entry md_size =
4563 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4565 /* Metadata version.
4567 * 'none' for arrays with no metadata (good luck...)
4568 * 'external' for arrays with externally managed metadata,
4569 * or N.M for internally known formats
4572 metadata_show(struct mddev *mddev, char *page)
4574 if (mddev->persistent)
4575 return sprintf(page, "%d.%d\n",
4576 mddev->major_version, mddev->minor_version);
4577 else if (mddev->external)
4578 return sprintf(page, "external:%s\n", mddev->metadata_type);
4580 return sprintf(page, "none\n");
4584 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4589 /* Changing the details of 'external' metadata is
4590 * always permitted. Otherwise there must be
4591 * no devices attached to the array.
4594 err = mddev_lock(mddev);
4598 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4600 else if (!list_empty(&mddev->disks))
4604 if (cmd_match(buf, "none")) {
4605 mddev->persistent = 0;
4606 mddev->external = 0;
4607 mddev->major_version = 0;
4608 mddev->minor_version = 90;
4611 if (strncmp(buf, "external:", 9) == 0) {
4612 size_t namelen = len-9;
4613 if (namelen >= sizeof(mddev->metadata_type))
4614 namelen = sizeof(mddev->metadata_type)-1;
4615 strncpy(mddev->metadata_type, buf+9, namelen);
4616 mddev->metadata_type[namelen] = 0;
4617 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4618 mddev->metadata_type[--namelen] = 0;
4619 mddev->persistent = 0;
4620 mddev->external = 1;
4621 mddev->major_version = 0;
4622 mddev->minor_version = 90;
4625 major = simple_strtoul(buf, &e, 10);
4627 if (e==buf || *e != '.')
4630 minor = simple_strtoul(buf, &e, 10);
4631 if (e==buf || (*e && *e != '\n') )
4634 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4636 mddev->major_version = major;
4637 mddev->minor_version = minor;
4638 mddev->persistent = 1;
4639 mddev->external = 0;
4642 mddev_unlock(mddev);
4646 static struct md_sysfs_entry md_metadata =
4647 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4650 action_show(struct mddev *mddev, char *page)
4652 char *type = "idle";
4653 unsigned long recovery = mddev->recovery;
4654 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4656 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4657 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4658 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4660 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4661 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4663 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4667 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4669 else if (mddev->reshape_position != MaxSector)
4672 return sprintf(page, "%s\n", type);
4676 action_store(struct mddev *mddev, const char *page, size_t len)
4678 if (!mddev->pers || !mddev->pers->sync_request)
4682 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4683 if (cmd_match(page, "frozen"))
4684 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4686 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4687 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4688 mddev_lock(mddev) == 0) {
4689 flush_workqueue(md_misc_wq);
4690 if (mddev->sync_thread) {
4691 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4692 md_reap_sync_thread(mddev);
4694 mddev_unlock(mddev);
4696 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4698 else if (cmd_match(page, "resync"))
4699 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4700 else if (cmd_match(page, "recover")) {
4701 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4702 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4703 } else if (cmd_match(page, "reshape")) {
4705 if (mddev->pers->start_reshape == NULL)
4707 err = mddev_lock(mddev);
4709 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4712 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4713 err = mddev->pers->start_reshape(mddev);
4715 mddev_unlock(mddev);
4719 sysfs_notify(&mddev->kobj, NULL, "degraded");
4721 if (cmd_match(page, "check"))
4722 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4723 else if (!cmd_match(page, "repair"))
4725 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4726 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4727 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4729 if (mddev->ro == 2) {
4730 /* A write to sync_action is enough to justify
4731 * canceling read-auto mode
4734 md_wakeup_thread(mddev->sync_thread);
4736 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4737 md_wakeup_thread(mddev->thread);
4738 sysfs_notify_dirent_safe(mddev->sysfs_action);
4742 static struct md_sysfs_entry md_scan_mode =
4743 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4746 last_sync_action_show(struct mddev *mddev, char *page)
4748 return sprintf(page, "%s\n", mddev->last_sync_action);
4751 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4754 mismatch_cnt_show(struct mddev *mddev, char *page)
4756 return sprintf(page, "%llu\n",
4757 (unsigned long long)
4758 atomic64_read(&mddev->resync_mismatches));
4761 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4764 sync_min_show(struct mddev *mddev, char *page)
4766 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4767 mddev->sync_speed_min ? "local": "system");
4771 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4776 if (strncmp(buf, "system", 6)==0) {
4779 rv = kstrtouint(buf, 10, &min);
4785 mddev->sync_speed_min = min;
4789 static struct md_sysfs_entry md_sync_min =
4790 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4793 sync_max_show(struct mddev *mddev, char *page)
4795 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4796 mddev->sync_speed_max ? "local": "system");
4800 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4805 if (strncmp(buf, "system", 6)==0) {
4808 rv = kstrtouint(buf, 10, &max);
4814 mddev->sync_speed_max = max;
4818 static struct md_sysfs_entry md_sync_max =
4819 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4822 degraded_show(struct mddev *mddev, char *page)
4824 return sprintf(page, "%d\n", mddev->degraded);
4826 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4829 sync_force_parallel_show(struct mddev *mddev, char *page)
4831 return sprintf(page, "%d\n", mddev->parallel_resync);
4835 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4839 if (kstrtol(buf, 10, &n))
4842 if (n != 0 && n != 1)
4845 mddev->parallel_resync = n;
4847 if (mddev->sync_thread)
4848 wake_up(&resync_wait);
4853 /* force parallel resync, even with shared block devices */
4854 static struct md_sysfs_entry md_sync_force_parallel =
4855 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4856 sync_force_parallel_show, sync_force_parallel_store);
4859 sync_speed_show(struct mddev *mddev, char *page)
4861 unsigned long resync, dt, db;
4862 if (mddev->curr_resync == 0)
4863 return sprintf(page, "none\n");
4864 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4865 dt = (jiffies - mddev->resync_mark) / HZ;
4867 db = resync - mddev->resync_mark_cnt;
4868 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4871 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4874 sync_completed_show(struct mddev *mddev, char *page)
4876 unsigned long long max_sectors, resync;
4878 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4879 return sprintf(page, "none\n");
4881 if (mddev->curr_resync == 1 ||
4882 mddev->curr_resync == 2)
4883 return sprintf(page, "delayed\n");
4885 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4886 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4887 max_sectors = mddev->resync_max_sectors;
4889 max_sectors = mddev->dev_sectors;
4891 resync = mddev->curr_resync_completed;
4892 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4895 static struct md_sysfs_entry md_sync_completed =
4896 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4899 min_sync_show(struct mddev *mddev, char *page)
4901 return sprintf(page, "%llu\n",
4902 (unsigned long long)mddev->resync_min);
4905 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4907 unsigned long long min;
4910 if (kstrtoull(buf, 10, &min))
4913 spin_lock(&mddev->lock);
4915 if (min > mddev->resync_max)
4919 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4922 /* Round down to multiple of 4K for safety */
4923 mddev->resync_min = round_down(min, 8);
4927 spin_unlock(&mddev->lock);
4931 static struct md_sysfs_entry md_min_sync =
4932 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4935 max_sync_show(struct mddev *mddev, char *page)
4937 if (mddev->resync_max == MaxSector)
4938 return sprintf(page, "max\n");
4940 return sprintf(page, "%llu\n",
4941 (unsigned long long)mddev->resync_max);
4944 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4947 spin_lock(&mddev->lock);
4948 if (strncmp(buf, "max", 3) == 0)
4949 mddev->resync_max = MaxSector;
4951 unsigned long long max;
4955 if (kstrtoull(buf, 10, &max))
4957 if (max < mddev->resync_min)
4961 if (max < mddev->resync_max &&
4963 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4966 /* Must be a multiple of chunk_size */
4967 chunk = mddev->chunk_sectors;
4969 sector_t temp = max;
4972 if (sector_div(temp, chunk))
4975 mddev->resync_max = max;
4977 wake_up(&mddev->recovery_wait);
4980 spin_unlock(&mddev->lock);
4984 static struct md_sysfs_entry md_max_sync =
4985 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4988 suspend_lo_show(struct mddev *mddev, char *page)
4990 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4994 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4996 unsigned long long new;
4999 err = kstrtoull(buf, 10, &new);
5002 if (new != (sector_t)new)
5005 err = mddev_lock(mddev);
5009 if (mddev->pers == NULL ||
5010 mddev->pers->quiesce == NULL)
5012 mddev_suspend(mddev);
5013 mddev->suspend_lo = new;
5014 mddev_resume(mddev);
5018 mddev_unlock(mddev);
5021 static struct md_sysfs_entry md_suspend_lo =
5022 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
5025 suspend_hi_show(struct mddev *mddev, char *page)
5027 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
5031 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
5033 unsigned long long new;
5036 err = kstrtoull(buf, 10, &new);
5039 if (new != (sector_t)new)
5042 err = mddev_lock(mddev);
5046 if (mddev->pers == NULL)
5049 mddev_suspend(mddev);
5050 mddev->suspend_hi = new;
5051 mddev_resume(mddev);
5055 mddev_unlock(mddev);
5058 static struct md_sysfs_entry md_suspend_hi =
5059 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
5062 reshape_position_show(struct mddev *mddev, char *page)
5064 if (mddev->reshape_position != MaxSector)
5065 return sprintf(page, "%llu\n",
5066 (unsigned long long)mddev->reshape_position);
5067 strcpy(page, "none\n");
5072 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
5074 struct md_rdev *rdev;
5075 unsigned long long new;
5078 err = kstrtoull(buf, 10, &new);
5081 if (new != (sector_t)new)
5083 err = mddev_lock(mddev);
5089 mddev->reshape_position = new;
5090 mddev->delta_disks = 0;
5091 mddev->reshape_backwards = 0;
5092 mddev->new_level = mddev->level;
5093 mddev->new_layout = mddev->layout;
5094 mddev->new_chunk_sectors = mddev->chunk_sectors;
5095 rdev_for_each(rdev, mddev)
5096 rdev->new_data_offset = rdev->data_offset;
5099 mddev_unlock(mddev);
5103 static struct md_sysfs_entry md_reshape_position =
5104 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
5105 reshape_position_store);
5108 reshape_direction_show(struct mddev *mddev, char *page)
5110 return sprintf(page, "%s\n",
5111 mddev->reshape_backwards ? "backwards" : "forwards");
5115 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
5120 if (cmd_match(buf, "forwards"))
5122 else if (cmd_match(buf, "backwards"))
5126 if (mddev->reshape_backwards == backwards)
5129 err = mddev_lock(mddev);
5132 /* check if we are allowed to change */
5133 if (mddev->delta_disks)
5135 else if (mddev->persistent &&
5136 mddev->major_version == 0)
5139 mddev->reshape_backwards = backwards;
5140 mddev_unlock(mddev);
5144 static struct md_sysfs_entry md_reshape_direction =
5145 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
5146 reshape_direction_store);
5149 array_size_show(struct mddev *mddev, char *page)
5151 if (mddev->external_size)
5152 return sprintf(page, "%llu\n",
5153 (unsigned long long)mddev->array_sectors/2);
5155 return sprintf(page, "default\n");
5159 array_size_store(struct mddev *mddev, const char *buf, size_t len)
5164 err = mddev_lock(mddev);
5168 /* cluster raid doesn't support change array_sectors */
5169 if (mddev_is_clustered(mddev)) {
5170 mddev_unlock(mddev);
5174 if (strncmp(buf, "default", 7) == 0) {
5176 sectors = mddev->pers->size(mddev, 0, 0);
5178 sectors = mddev->array_sectors;
5180 mddev->external_size = 0;
5182 if (strict_blocks_to_sectors(buf, §ors) < 0)
5184 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
5187 mddev->external_size = 1;
5191 mddev->array_sectors = sectors;
5193 set_capacity(mddev->gendisk, mddev->array_sectors);
5194 revalidate_disk(mddev->gendisk);
5197 mddev_unlock(mddev);
5201 static struct md_sysfs_entry md_array_size =
5202 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
5206 consistency_policy_show(struct mddev *mddev, char *page)
5210 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5211 ret = sprintf(page, "journal\n");
5212 } else if (test_bit(MD_HAS_PPL, &mddev->flags)) {
5213 ret = sprintf(page, "ppl\n");
5214 } else if (mddev->bitmap) {
5215 ret = sprintf(page, "bitmap\n");
5216 } else if (mddev->pers) {
5217 if (mddev->pers->sync_request)
5218 ret = sprintf(page, "resync\n");
5220 ret = sprintf(page, "none\n");
5222 ret = sprintf(page, "unknown\n");
5229 consistency_policy_store(struct mddev *mddev, const char *buf, size_t len)
5234 if (mddev->pers->change_consistency_policy)
5235 err = mddev->pers->change_consistency_policy(mddev, buf);
5238 } else if (mddev->external && strncmp(buf, "ppl", 3) == 0) {
5239 set_bit(MD_HAS_PPL, &mddev->flags);
5244 return err ? err : len;
5247 static struct md_sysfs_entry md_consistency_policy =
5248 __ATTR(consistency_policy, S_IRUGO | S_IWUSR, consistency_policy_show,
5249 consistency_policy_store);
5251 static ssize_t fail_last_dev_show(struct mddev *mddev, char *page)
5253 return sprintf(page, "%d\n", mddev->fail_last_dev);
5257 * Setting fail_last_dev to true to allow last device to be forcibly removed
5258 * from RAID1/RAID10.
5261 fail_last_dev_store(struct mddev *mddev, const char *buf, size_t len)
5266 ret = kstrtobool(buf, &value);
5270 if (value != mddev->fail_last_dev)
5271 mddev->fail_last_dev = value;
5275 static struct md_sysfs_entry md_fail_last_dev =
5276 __ATTR(fail_last_dev, S_IRUGO | S_IWUSR, fail_last_dev_show,
5277 fail_last_dev_store);
5279 static struct attribute *md_default_attrs[] = {
5282 &md_raid_disks.attr,
5283 &md_chunk_size.attr,
5285 &md_resync_start.attr,
5287 &md_new_device.attr,
5288 &md_safe_delay.attr,
5289 &md_array_state.attr,
5290 &md_reshape_position.attr,
5291 &md_reshape_direction.attr,
5292 &md_array_size.attr,
5293 &max_corr_read_errors.attr,
5294 &md_consistency_policy.attr,
5295 &md_fail_last_dev.attr,
5299 static struct attribute *md_redundancy_attrs[] = {
5301 &md_last_scan_mode.attr,
5302 &md_mismatches.attr,
5305 &md_sync_speed.attr,
5306 &md_sync_force_parallel.attr,
5307 &md_sync_completed.attr,
5310 &md_suspend_lo.attr,
5311 &md_suspend_hi.attr,
5316 static struct attribute_group md_redundancy_group = {
5318 .attrs = md_redundancy_attrs,
5322 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
5324 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5325 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5330 spin_lock(&all_mddevs_lock);
5331 if (list_empty(&mddev->all_mddevs)) {
5332 spin_unlock(&all_mddevs_lock);
5336 spin_unlock(&all_mddevs_lock);
5338 rv = entry->show(mddev, page);
5344 md_attr_store(struct kobject *kobj, struct attribute *attr,
5345 const char *page, size_t length)
5347 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
5348 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
5353 if (!capable(CAP_SYS_ADMIN))
5355 spin_lock(&all_mddevs_lock);
5356 if (list_empty(&mddev->all_mddevs)) {
5357 spin_unlock(&all_mddevs_lock);
5361 spin_unlock(&all_mddevs_lock);
5362 rv = entry->store(mddev, page, length);
5367 static void md_free(struct kobject *ko)
5369 struct mddev *mddev = container_of(ko, struct mddev, kobj);
5371 if (mddev->sysfs_state)
5372 sysfs_put(mddev->sysfs_state);
5375 del_gendisk(mddev->gendisk);
5377 blk_cleanup_queue(mddev->queue);
5379 put_disk(mddev->gendisk);
5380 percpu_ref_exit(&mddev->writes_pending);
5382 bioset_exit(&mddev->bio_set);
5383 bioset_exit(&mddev->sync_set);
5387 static const struct sysfs_ops md_sysfs_ops = {
5388 .show = md_attr_show,
5389 .store = md_attr_store,
5391 static struct kobj_type md_ktype = {
5393 .sysfs_ops = &md_sysfs_ops,
5394 .default_attrs = md_default_attrs,
5399 static void mddev_delayed_delete(struct work_struct *ws)
5401 struct mddev *mddev = container_of(ws, struct mddev, del_work);
5403 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
5404 kobject_del(&mddev->kobj);
5405 kobject_put(&mddev->kobj);
5408 static void no_op(struct percpu_ref *r) {}
5410 int mddev_init_writes_pending(struct mddev *mddev)
5412 if (mddev->writes_pending.percpu_count_ptr)
5414 if (percpu_ref_init(&mddev->writes_pending, no_op,
5415 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL) < 0)
5417 /* We want to start with the refcount at zero */
5418 percpu_ref_put(&mddev->writes_pending);
5421 EXPORT_SYMBOL_GPL(mddev_init_writes_pending);
5423 static int md_alloc(dev_t dev, char *name)
5426 * If dev is zero, name is the name of a device to allocate with
5427 * an arbitrary minor number. It will be "md_???"
5428 * If dev is non-zero it must be a device number with a MAJOR of
5429 * MD_MAJOR or mdp_major. In this case, if "name" is NULL, then
5430 * the device is being created by opening a node in /dev.
5431 * If "name" is not NULL, the device is being created by
5432 * writing to /sys/module/md_mod/parameters/new_array.
5434 static DEFINE_MUTEX(disks_mutex);
5435 struct mddev *mddev = mddev_find(dev);
5436 struct gendisk *disk;
5445 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5446 shift = partitioned ? MdpMinorShift : 0;
5447 unit = MINOR(mddev->unit) >> shift;
5449 /* wait for any previous instance of this device to be
5450 * completely removed (mddev_delayed_delete).
5452 flush_workqueue(md_misc_wq);
5454 mutex_lock(&disks_mutex);
5460 /* Need to ensure that 'name' is not a duplicate.
5462 struct mddev *mddev2;
5463 spin_lock(&all_mddevs_lock);
5465 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5466 if (mddev2->gendisk &&
5467 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5468 spin_unlock(&all_mddevs_lock);
5471 spin_unlock(&all_mddevs_lock);
5475 * Creating /dev/mdNNN via "newarray", so adjust hold_active.
5477 mddev->hold_active = UNTIL_STOP;
5480 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5483 mddev->queue->queuedata = mddev;
5485 blk_queue_make_request(mddev->queue, md_make_request);
5486 blk_set_stacking_limits(&mddev->queue->limits);
5488 disk = alloc_disk(1 << shift);
5490 blk_cleanup_queue(mddev->queue);
5491 mddev->queue = NULL;
5494 disk->major = MAJOR(mddev->unit);
5495 disk->first_minor = unit << shift;
5497 strcpy(disk->disk_name, name);
5498 else if (partitioned)
5499 sprintf(disk->disk_name, "md_d%d", unit);
5501 sprintf(disk->disk_name, "md%d", unit);
5502 disk->fops = &md_fops;
5503 disk->private_data = mddev;
5504 disk->queue = mddev->queue;
5505 blk_queue_write_cache(mddev->queue, true, true);
5506 /* Allow extended partitions. This makes the
5507 * 'mdp' device redundant, but we can't really
5510 disk->flags |= GENHD_FL_EXT_DEVT;
5511 mddev->gendisk = disk;
5512 /* As soon as we call add_disk(), another thread could get
5513 * through to md_open, so make sure it doesn't get too far
5515 mutex_lock(&mddev->open_mutex);
5518 error = kobject_add(&mddev->kobj, &disk_to_dev(disk)->kobj, "%s", "md");
5520 /* This isn't possible, but as kobject_init_and_add is marked
5521 * __must_check, we must do something with the result
5523 pr_debug("md: cannot register %s/md - name in use\n",
5527 if (mddev->kobj.sd &&
5528 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5529 pr_debug("pointless warning\n");
5530 mutex_unlock(&mddev->open_mutex);
5532 mutex_unlock(&disks_mutex);
5533 if (!error && mddev->kobj.sd) {
5534 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5535 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5541 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5544 md_alloc(dev, NULL);
5548 static int add_named_array(const char *val, const struct kernel_param *kp)
5551 * val must be "md_*" or "mdNNN".
5552 * For "md_*" we allocate an array with a large free minor number, and
5553 * set the name to val. val must not already be an active name.
5554 * For "mdNNN" we allocate an array with the minor number NNN
5555 * which must not already be in use.
5557 int len = strlen(val);
5558 char buf[DISK_NAME_LEN];
5559 unsigned long devnum;
5561 while (len && val[len-1] == '\n')
5563 if (len >= DISK_NAME_LEN)
5565 strlcpy(buf, val, len+1);
5566 if (strncmp(buf, "md_", 3) == 0)
5567 return md_alloc(0, buf);
5568 if (strncmp(buf, "md", 2) == 0 &&
5570 kstrtoul(buf+2, 10, &devnum) == 0 &&
5571 devnum <= MINORMASK)
5572 return md_alloc(MKDEV(MD_MAJOR, devnum), NULL);
5577 static void md_safemode_timeout(struct timer_list *t)
5579 struct mddev *mddev = from_timer(mddev, t, safemode_timer);
5581 mddev->safemode = 1;
5582 if (mddev->external)
5583 sysfs_notify_dirent_safe(mddev->sysfs_state);
5585 md_wakeup_thread(mddev->thread);
5588 static int start_dirty_degraded;
5590 int md_run(struct mddev *mddev)
5593 struct md_rdev *rdev;
5594 struct md_personality *pers;
5596 if (list_empty(&mddev->disks))
5597 /* cannot run an array with no devices.. */
5602 /* Cannot run until previous stop completes properly */
5603 if (mddev->sysfs_active)
5607 * Analyze all RAID superblock(s)
5609 if (!mddev->raid_disks) {
5610 if (!mddev->persistent)
5612 err = analyze_sbs(mddev);
5617 if (mddev->level != LEVEL_NONE)
5618 request_module("md-level-%d", mddev->level);
5619 else if (mddev->clevel[0])
5620 request_module("md-%s", mddev->clevel);
5623 * Drop all container device buffers, from now on
5624 * the only valid external interface is through the md
5627 mddev->has_superblocks = false;
5628 rdev_for_each(rdev, mddev) {
5629 if (test_bit(Faulty, &rdev->flags))
5631 sync_blockdev(rdev->bdev);
5632 invalidate_bdev(rdev->bdev);
5633 if (mddev->ro != 1 &&
5634 (bdev_read_only(rdev->bdev) ||
5635 bdev_read_only(rdev->meta_bdev))) {
5638 set_disk_ro(mddev->gendisk, 1);
5642 mddev->has_superblocks = true;
5644 /* perform some consistency tests on the device.
5645 * We don't want the data to overlap the metadata,
5646 * Internal Bitmap issues have been handled elsewhere.
5648 if (rdev->meta_bdev) {
5649 /* Nothing to check */;
5650 } else if (rdev->data_offset < rdev->sb_start) {
5651 if (mddev->dev_sectors &&
5652 rdev->data_offset + mddev->dev_sectors
5654 pr_warn("md: %s: data overlaps metadata\n",
5659 if (rdev->sb_start + rdev->sb_size/512
5660 > rdev->data_offset) {
5661 pr_warn("md: %s: metadata overlaps data\n",
5666 sysfs_notify_dirent_safe(rdev->sysfs_state);
5669 if (!bioset_initialized(&mddev->bio_set)) {
5670 err = bioset_init(&mddev->bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5674 if (!bioset_initialized(&mddev->sync_set)) {
5675 err = bioset_init(&mddev->sync_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
5680 spin_lock(&pers_lock);
5681 pers = find_pers(mddev->level, mddev->clevel);
5682 if (!pers || !try_module_get(pers->owner)) {
5683 spin_unlock(&pers_lock);
5684 if (mddev->level != LEVEL_NONE)
5685 pr_warn("md: personality for level %d is not loaded!\n",
5688 pr_warn("md: personality for level %s is not loaded!\n",
5693 spin_unlock(&pers_lock);
5694 if (mddev->level != pers->level) {
5695 mddev->level = pers->level;
5696 mddev->new_level = pers->level;
5698 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5700 if (mddev->reshape_position != MaxSector &&
5701 pers->start_reshape == NULL) {
5702 /* This personality cannot handle reshaping... */
5703 module_put(pers->owner);
5708 if (pers->sync_request) {
5709 /* Warn if this is a potentially silly
5712 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5713 struct md_rdev *rdev2;
5716 rdev_for_each(rdev, mddev)
5717 rdev_for_each(rdev2, mddev) {
5719 rdev->bdev->bd_contains ==
5720 rdev2->bdev->bd_contains) {
5721 pr_warn("%s: WARNING: %s appears to be on the same physical disk as %s.\n",
5723 bdevname(rdev->bdev,b),
5724 bdevname(rdev2->bdev,b2));
5730 pr_warn("True protection against single-disk failure might be compromised.\n");
5733 mddev->recovery = 0;
5734 /* may be over-ridden by personality */
5735 mddev->resync_max_sectors = mddev->dev_sectors;
5737 mddev->ok_start_degraded = start_dirty_degraded;
5739 if (start_readonly && mddev->ro == 0)
5740 mddev->ro = 2; /* read-only, but switch on first write */
5742 err = pers->run(mddev);
5744 pr_warn("md: pers->run() failed ...\n");
5745 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5746 WARN_ONCE(!mddev->external_size,
5747 "%s: default size too small, but 'external_size' not in effect?\n",
5749 pr_warn("md: invalid array_size %llu > default size %llu\n",
5750 (unsigned long long)mddev->array_sectors / 2,
5751 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5754 if (err == 0 && pers->sync_request &&
5755 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5756 struct bitmap *bitmap;
5758 bitmap = md_bitmap_create(mddev, -1);
5759 if (IS_ERR(bitmap)) {
5760 err = PTR_ERR(bitmap);
5761 pr_warn("%s: failed to create bitmap (%d)\n",
5762 mdname(mddev), err);
5764 mddev->bitmap = bitmap;
5770 if (mddev->bitmap_info.max_write_behind > 0) {
5771 bool creat_pool = false;
5773 rdev_for_each(rdev, mddev) {
5774 if (test_bit(WriteMostly, &rdev->flags) &&
5778 if (creat_pool && mddev->wb_info_pool == NULL) {
5779 mddev->wb_info_pool =
5780 mempool_create_kmalloc_pool(NR_WB_INFOS,
5781 sizeof(struct wb_info));
5782 if (!mddev->wb_info_pool) {
5792 rdev_for_each(rdev, mddev) {
5793 if (rdev->raid_disk >= 0 &&
5794 !blk_queue_nonrot(bdev_get_queue(rdev->bdev))) {
5799 if (mddev->degraded)
5802 blk_queue_flag_set(QUEUE_FLAG_NONROT, mddev->queue);
5804 blk_queue_flag_clear(QUEUE_FLAG_NONROT, mddev->queue);
5805 mddev->queue->backing_dev_info->congested_data = mddev;
5806 mddev->queue->backing_dev_info->congested_fn = md_congested;
5808 if (pers->sync_request) {
5809 if (mddev->kobj.sd &&
5810 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5811 pr_warn("md: cannot register extra attributes for %s\n",
5813 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5814 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5817 atomic_set(&mddev->max_corr_read_errors,
5818 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5819 mddev->safemode = 0;
5820 if (mddev_is_clustered(mddev))
5821 mddev->safemode_delay = 0;
5823 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5826 spin_lock(&mddev->lock);
5828 spin_unlock(&mddev->lock);
5829 rdev_for_each(rdev, mddev)
5830 if (rdev->raid_disk >= 0)
5831 sysfs_link_rdev(mddev, rdev); /* failure here is OK */
5833 if (mddev->degraded && !mddev->ro)
5834 /* This ensures that recovering status is reported immediately
5835 * via sysfs - until a lack of spares is confirmed.
5837 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5838 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5840 if (mddev->sb_flags)
5841 md_update_sb(mddev, 0);
5843 md_new_event(mddev);
5847 mddev_detach(mddev);
5849 pers->free(mddev, mddev->private);
5850 mddev->private = NULL;
5851 module_put(pers->owner);
5852 md_bitmap_destroy(mddev);
5854 bioset_exit(&mddev->bio_set);
5855 bioset_exit(&mddev->sync_set);
5858 EXPORT_SYMBOL_GPL(md_run);
5860 static int do_md_run(struct mddev *mddev)
5864 set_bit(MD_NOT_READY, &mddev->flags);
5865 err = md_run(mddev);
5868 err = md_bitmap_load(mddev);
5870 md_bitmap_destroy(mddev);
5874 if (mddev_is_clustered(mddev))
5875 md_allow_write(mddev);
5877 /* run start up tasks that require md_thread */
5880 md_wakeup_thread(mddev->thread);
5881 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5883 set_capacity(mddev->gendisk, mddev->array_sectors);
5884 revalidate_disk(mddev->gendisk);
5885 clear_bit(MD_NOT_READY, &mddev->flags);
5887 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5888 sysfs_notify_dirent_safe(mddev->sysfs_state);
5889 sysfs_notify_dirent_safe(mddev->sysfs_action);
5890 sysfs_notify(&mddev->kobj, NULL, "degraded");
5892 clear_bit(MD_NOT_READY, &mddev->flags);
5896 int md_start(struct mddev *mddev)
5900 if (mddev->pers->start) {
5901 set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5902 md_wakeup_thread(mddev->thread);
5903 ret = mddev->pers->start(mddev);
5904 clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
5905 md_wakeup_thread(mddev->sync_thread);
5909 EXPORT_SYMBOL_GPL(md_start);
5911 static int restart_array(struct mddev *mddev)
5913 struct gendisk *disk = mddev->gendisk;
5914 struct md_rdev *rdev;
5915 bool has_journal = false;
5916 bool has_readonly = false;
5918 /* Complain if it has no devices */
5919 if (list_empty(&mddev->disks))
5927 rdev_for_each_rcu(rdev, mddev) {
5928 if (test_bit(Journal, &rdev->flags) &&
5929 !test_bit(Faulty, &rdev->flags))
5931 if (bdev_read_only(rdev->bdev))
5932 has_readonly = true;
5935 if (test_bit(MD_HAS_JOURNAL, &mddev->flags) && !has_journal)
5936 /* Don't restart rw with journal missing/faulty */
5941 mddev->safemode = 0;
5943 set_disk_ro(disk, 0);
5944 pr_debug("md: %s switched to read-write mode.\n", mdname(mddev));
5945 /* Kick recovery or resync if necessary */
5946 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5947 md_wakeup_thread(mddev->thread);
5948 md_wakeup_thread(mddev->sync_thread);
5949 sysfs_notify_dirent_safe(mddev->sysfs_state);
5953 static void md_clean(struct mddev *mddev)
5955 mddev->array_sectors = 0;
5956 mddev->external_size = 0;
5957 mddev->dev_sectors = 0;
5958 mddev->raid_disks = 0;
5959 mddev->recovery_cp = 0;
5960 mddev->resync_min = 0;
5961 mddev->resync_max = MaxSector;
5962 mddev->reshape_position = MaxSector;
5963 mddev->external = 0;
5964 mddev->persistent = 0;
5965 mddev->level = LEVEL_NONE;
5966 mddev->clevel[0] = 0;
5968 mddev->sb_flags = 0;
5970 mddev->metadata_type[0] = 0;
5971 mddev->chunk_sectors = 0;
5972 mddev->ctime = mddev->utime = 0;
5974 mddev->max_disks = 0;
5976 mddev->can_decrease_events = 0;
5977 mddev->delta_disks = 0;
5978 mddev->reshape_backwards = 0;
5979 mddev->new_level = LEVEL_NONE;
5980 mddev->new_layout = 0;
5981 mddev->new_chunk_sectors = 0;
5982 mddev->curr_resync = 0;
5983 atomic64_set(&mddev->resync_mismatches, 0);
5984 mddev->suspend_lo = mddev->suspend_hi = 0;
5985 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5986 mddev->recovery = 0;
5989 mddev->degraded = 0;
5990 mddev->safemode = 0;
5991 mddev->private = NULL;
5992 mddev->cluster_info = NULL;
5993 mddev->bitmap_info.offset = 0;
5994 mddev->bitmap_info.default_offset = 0;
5995 mddev->bitmap_info.default_space = 0;
5996 mddev->bitmap_info.chunksize = 0;
5997 mddev->bitmap_info.daemon_sleep = 0;
5998 mddev->bitmap_info.max_write_behind = 0;
5999 mddev->bitmap_info.nodes = 0;
6002 static void __md_stop_writes(struct mddev *mddev)
6004 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6005 flush_workqueue(md_misc_wq);
6006 if (mddev->sync_thread) {
6007 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6008 md_reap_sync_thread(mddev);
6011 del_timer_sync(&mddev->safemode_timer);
6013 if (mddev->pers && mddev->pers->quiesce) {
6014 mddev->pers->quiesce(mddev, 1);
6015 mddev->pers->quiesce(mddev, 0);
6017 md_bitmap_flush(mddev);
6019 if (mddev->ro == 0 &&
6020 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
6022 /* mark array as shutdown cleanly */
6023 if (!mddev_is_clustered(mddev))
6025 md_update_sb(mddev, 1);
6027 mempool_destroy(mddev->wb_info_pool);
6028 mddev->wb_info_pool = NULL;
6031 void md_stop_writes(struct mddev *mddev)
6033 mddev_lock_nointr(mddev);
6034 __md_stop_writes(mddev);
6035 mddev_unlock(mddev);
6037 EXPORT_SYMBOL_GPL(md_stop_writes);
6039 static void mddev_detach(struct mddev *mddev)
6041 md_bitmap_wait_behind_writes(mddev);
6042 if (mddev->pers && mddev->pers->quiesce) {
6043 mddev->pers->quiesce(mddev, 1);
6044 mddev->pers->quiesce(mddev, 0);
6046 md_unregister_thread(&mddev->thread);
6048 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
6051 static void __md_stop(struct mddev *mddev)
6053 struct md_personality *pers = mddev->pers;
6054 md_bitmap_destroy(mddev);
6055 mddev_detach(mddev);
6056 /* Ensure ->event_work is done */
6057 flush_workqueue(md_misc_wq);
6058 spin_lock(&mddev->lock);
6060 spin_unlock(&mddev->lock);
6061 pers->free(mddev, mddev->private);
6062 mddev->private = NULL;
6063 if (pers->sync_request && mddev->to_remove == NULL)
6064 mddev->to_remove = &md_redundancy_group;
6065 module_put(pers->owner);
6066 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6069 void md_stop(struct mddev *mddev)
6071 /* stop the array and free an attached data structures.
6072 * This is called from dm-raid
6075 bioset_exit(&mddev->bio_set);
6076 bioset_exit(&mddev->sync_set);
6079 EXPORT_SYMBOL_GPL(md_stop);
6081 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
6086 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6088 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6089 md_wakeup_thread(mddev->thread);
6091 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6092 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6093 if (mddev->sync_thread)
6094 /* Thread might be blocked waiting for metadata update
6095 * which will now never happen */
6096 wake_up_process(mddev->sync_thread->tsk);
6098 if (mddev->external && test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
6100 mddev_unlock(mddev);
6101 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
6103 wait_event(mddev->sb_wait,
6104 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
6105 mddev_lock_nointr(mddev);
6107 mutex_lock(&mddev->open_mutex);
6108 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6109 mddev->sync_thread ||
6110 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6111 pr_warn("md: %s still in use.\n",mdname(mddev));
6113 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6114 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6115 md_wakeup_thread(mddev->thread);
6121 __md_stop_writes(mddev);
6127 set_disk_ro(mddev->gendisk, 1);
6128 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6129 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6130 md_wakeup_thread(mddev->thread);
6131 sysfs_notify_dirent_safe(mddev->sysfs_state);
6135 mutex_unlock(&mddev->open_mutex);
6140 * 0 - completely stop and dis-assemble array
6141 * 2 - stop but do not disassemble array
6143 static int do_md_stop(struct mddev *mddev, int mode,
6144 struct block_device *bdev)
6146 struct gendisk *disk = mddev->gendisk;
6147 struct md_rdev *rdev;
6150 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
6152 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6153 md_wakeup_thread(mddev->thread);
6155 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
6156 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6157 if (mddev->sync_thread)
6158 /* Thread might be blocked waiting for metadata update
6159 * which will now never happen */
6160 wake_up_process(mddev->sync_thread->tsk);
6162 mddev_unlock(mddev);
6163 wait_event(resync_wait, (mddev->sync_thread == NULL &&
6164 !test_bit(MD_RECOVERY_RUNNING,
6165 &mddev->recovery)));
6166 mddev_lock_nointr(mddev);
6168 mutex_lock(&mddev->open_mutex);
6169 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
6170 mddev->sysfs_active ||
6171 mddev->sync_thread ||
6172 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) {
6173 pr_warn("md: %s still in use.\n",mdname(mddev));
6174 mutex_unlock(&mddev->open_mutex);
6176 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
6177 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6178 md_wakeup_thread(mddev->thread);
6184 set_disk_ro(disk, 0);
6186 __md_stop_writes(mddev);
6188 mddev->queue->backing_dev_info->congested_fn = NULL;
6190 /* tell userspace to handle 'inactive' */
6191 sysfs_notify_dirent_safe(mddev->sysfs_state);
6193 rdev_for_each(rdev, mddev)
6194 if (rdev->raid_disk >= 0)
6195 sysfs_unlink_rdev(mddev, rdev);
6197 set_capacity(disk, 0);
6198 mutex_unlock(&mddev->open_mutex);
6200 revalidate_disk(disk);
6205 mutex_unlock(&mddev->open_mutex);
6207 * Free resources if final stop
6210 pr_info("md: %s stopped.\n", mdname(mddev));
6212 if (mddev->bitmap_info.file) {
6213 struct file *f = mddev->bitmap_info.file;
6214 spin_lock(&mddev->lock);
6215 mddev->bitmap_info.file = NULL;
6216 spin_unlock(&mddev->lock);
6219 mddev->bitmap_info.offset = 0;
6221 export_array(mddev);
6224 if (mddev->hold_active == UNTIL_STOP)
6225 mddev->hold_active = 0;
6227 md_new_event(mddev);
6228 sysfs_notify_dirent_safe(mddev->sysfs_state);
6233 static void autorun_array(struct mddev *mddev)
6235 struct md_rdev *rdev;
6238 if (list_empty(&mddev->disks))
6241 pr_info("md: running: ");
6243 rdev_for_each(rdev, mddev) {
6244 char b[BDEVNAME_SIZE];
6245 pr_cont("<%s>", bdevname(rdev->bdev,b));
6249 err = do_md_run(mddev);
6251 pr_warn("md: do_md_run() returned %d\n", err);
6252 do_md_stop(mddev, 0, NULL);
6257 * lets try to run arrays based on all disks that have arrived
6258 * until now. (those are in pending_raid_disks)
6260 * the method: pick the first pending disk, collect all disks with
6261 * the same UUID, remove all from the pending list and put them into
6262 * the 'same_array' list. Then order this list based on superblock
6263 * update time (freshest comes first), kick out 'old' disks and
6264 * compare superblocks. If everything's fine then run it.
6266 * If "unit" is allocated, then bump its reference count
6268 static void autorun_devices(int part)
6270 struct md_rdev *rdev0, *rdev, *tmp;
6271 struct mddev *mddev;
6272 char b[BDEVNAME_SIZE];
6274 pr_info("md: autorun ...\n");
6275 while (!list_empty(&pending_raid_disks)) {
6278 LIST_HEAD(candidates);
6279 rdev0 = list_entry(pending_raid_disks.next,
6280 struct md_rdev, same_set);
6282 pr_debug("md: considering %s ...\n", bdevname(rdev0->bdev,b));
6283 INIT_LIST_HEAD(&candidates);
6284 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
6285 if (super_90_load(rdev, rdev0, 0) >= 0) {
6286 pr_debug("md: adding %s ...\n",
6287 bdevname(rdev->bdev,b));
6288 list_move(&rdev->same_set, &candidates);
6291 * now we have a set of devices, with all of them having
6292 * mostly sane superblocks. It's time to allocate the
6296 dev = MKDEV(mdp_major,
6297 rdev0->preferred_minor << MdpMinorShift);
6298 unit = MINOR(dev) >> MdpMinorShift;
6300 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
6303 if (rdev0->preferred_minor != unit) {
6304 pr_warn("md: unit number in %s is bad: %d\n",
6305 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
6309 md_probe(dev, NULL, NULL);
6310 mddev = mddev_find(dev);
6311 if (!mddev || !mddev->gendisk) {
6316 if (mddev_lock(mddev))
6317 pr_warn("md: %s locked, cannot run\n", mdname(mddev));
6318 else if (mddev->raid_disks || mddev->major_version
6319 || !list_empty(&mddev->disks)) {
6320 pr_warn("md: %s already running, cannot run %s\n",
6321 mdname(mddev), bdevname(rdev0->bdev,b));
6322 mddev_unlock(mddev);
6324 pr_debug("md: created %s\n", mdname(mddev));
6325 mddev->persistent = 1;
6326 rdev_for_each_list(rdev, tmp, &candidates) {
6327 list_del_init(&rdev->same_set);
6328 if (bind_rdev_to_array(rdev, mddev))
6331 autorun_array(mddev);
6332 mddev_unlock(mddev);
6334 /* on success, candidates will be empty, on error
6337 rdev_for_each_list(rdev, tmp, &candidates) {
6338 list_del_init(&rdev->same_set);
6343 pr_info("md: ... autorun DONE.\n");
6345 #endif /* !MODULE */
6347 static int get_version(void __user *arg)
6351 ver.major = MD_MAJOR_VERSION;
6352 ver.minor = MD_MINOR_VERSION;
6353 ver.patchlevel = MD_PATCHLEVEL_VERSION;
6355 if (copy_to_user(arg, &ver, sizeof(ver)))
6361 static int get_array_info(struct mddev *mddev, void __user *arg)
6363 mdu_array_info_t info;
6364 int nr,working,insync,failed,spare;
6365 struct md_rdev *rdev;
6367 nr = working = insync = failed = spare = 0;
6369 rdev_for_each_rcu(rdev, mddev) {
6371 if (test_bit(Faulty, &rdev->flags))
6375 if (test_bit(In_sync, &rdev->flags))
6377 else if (test_bit(Journal, &rdev->flags))
6378 /* TODO: add journal count to md_u.h */
6386 info.major_version = mddev->major_version;
6387 info.minor_version = mddev->minor_version;
6388 info.patch_version = MD_PATCHLEVEL_VERSION;
6389 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
6390 info.level = mddev->level;
6391 info.size = mddev->dev_sectors / 2;
6392 if (info.size != mddev->dev_sectors / 2) /* overflow */
6395 info.raid_disks = mddev->raid_disks;
6396 info.md_minor = mddev->md_minor;
6397 info.not_persistent= !mddev->persistent;
6399 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
6402 info.state = (1<<MD_SB_CLEAN);
6403 if (mddev->bitmap && mddev->bitmap_info.offset)
6404 info.state |= (1<<MD_SB_BITMAP_PRESENT);
6405 if (mddev_is_clustered(mddev))
6406 info.state |= (1<<MD_SB_CLUSTERED);
6407 info.active_disks = insync;
6408 info.working_disks = working;
6409 info.failed_disks = failed;
6410 info.spare_disks = spare;
6412 info.layout = mddev->layout;
6413 info.chunk_size = mddev->chunk_sectors << 9;
6415 if (copy_to_user(arg, &info, sizeof(info)))
6421 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
6423 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
6427 file = kzalloc(sizeof(*file), GFP_NOIO);
6432 spin_lock(&mddev->lock);
6433 /* bitmap enabled */
6434 if (mddev->bitmap_info.file) {
6435 ptr = file_path(mddev->bitmap_info.file, file->pathname,
6436 sizeof(file->pathname));
6440 memmove(file->pathname, ptr,
6441 sizeof(file->pathname)-(ptr-file->pathname));
6443 spin_unlock(&mddev->lock);
6446 copy_to_user(arg, file, sizeof(*file)))
6453 static int get_disk_info(struct mddev *mddev, void __user * arg)
6455 mdu_disk_info_t info;
6456 struct md_rdev *rdev;
6458 if (copy_from_user(&info, arg, sizeof(info)))
6462 rdev = md_find_rdev_nr_rcu(mddev, info.number);
6464 info.major = MAJOR(rdev->bdev->bd_dev);
6465 info.minor = MINOR(rdev->bdev->bd_dev);
6466 info.raid_disk = rdev->raid_disk;
6468 if (test_bit(Faulty, &rdev->flags))
6469 info.state |= (1<<MD_DISK_FAULTY);
6470 else if (test_bit(In_sync, &rdev->flags)) {
6471 info.state |= (1<<MD_DISK_ACTIVE);
6472 info.state |= (1<<MD_DISK_SYNC);
6474 if (test_bit(Journal, &rdev->flags))
6475 info.state |= (1<<MD_DISK_JOURNAL);
6476 if (test_bit(WriteMostly, &rdev->flags))
6477 info.state |= (1<<MD_DISK_WRITEMOSTLY);
6478 if (test_bit(FailFast, &rdev->flags))
6479 info.state |= (1<<MD_DISK_FAILFAST);
6481 info.major = info.minor = 0;
6482 info.raid_disk = -1;
6483 info.state = (1<<MD_DISK_REMOVED);
6487 if (copy_to_user(arg, &info, sizeof(info)))
6493 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
6495 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
6496 struct md_rdev *rdev;
6497 dev_t dev = MKDEV(info->major,info->minor);
6499 if (mddev_is_clustered(mddev) &&
6500 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
6501 pr_warn("%s: Cannot add to clustered mddev.\n",
6506 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
6509 if (!mddev->raid_disks) {
6511 /* expecting a device which has a superblock */
6512 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
6514 pr_warn("md: md_import_device returned %ld\n",
6516 return PTR_ERR(rdev);
6518 if (!list_empty(&mddev->disks)) {
6519 struct md_rdev *rdev0
6520 = list_entry(mddev->disks.next,
6521 struct md_rdev, same_set);
6522 err = super_types[mddev->major_version]
6523 .load_super(rdev, rdev0, mddev->minor_version);
6525 pr_warn("md: %s has different UUID to %s\n",
6526 bdevname(rdev->bdev,b),
6527 bdevname(rdev0->bdev,b2));
6532 err = bind_rdev_to_array(rdev, mddev);
6539 * add_new_disk can be used once the array is assembled
6540 * to add "hot spares". They must already have a superblock
6545 if (!mddev->pers->hot_add_disk) {
6546 pr_warn("%s: personality does not support diskops!\n",
6550 if (mddev->persistent)
6551 rdev = md_import_device(dev, mddev->major_version,
6552 mddev->minor_version);
6554 rdev = md_import_device(dev, -1, -1);
6556 pr_warn("md: md_import_device returned %ld\n",
6558 return PTR_ERR(rdev);
6560 /* set saved_raid_disk if appropriate */
6561 if (!mddev->persistent) {
6562 if (info->state & (1<<MD_DISK_SYNC) &&
6563 info->raid_disk < mddev->raid_disks) {
6564 rdev->raid_disk = info->raid_disk;
6565 set_bit(In_sync, &rdev->flags);
6566 clear_bit(Bitmap_sync, &rdev->flags);
6568 rdev->raid_disk = -1;
6569 rdev->saved_raid_disk = rdev->raid_disk;
6571 super_types[mddev->major_version].
6572 validate_super(mddev, rdev);
6573 if ((info->state & (1<<MD_DISK_SYNC)) &&
6574 rdev->raid_disk != info->raid_disk) {
6575 /* This was a hot-add request, but events doesn't
6576 * match, so reject it.
6582 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6583 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6584 set_bit(WriteMostly, &rdev->flags);
6586 clear_bit(WriteMostly, &rdev->flags);
6587 if (info->state & (1<<MD_DISK_FAILFAST))
6588 set_bit(FailFast, &rdev->flags);
6590 clear_bit(FailFast, &rdev->flags);
6592 if (info->state & (1<<MD_DISK_JOURNAL)) {
6593 struct md_rdev *rdev2;
6594 bool has_journal = false;
6596 /* make sure no existing journal disk */
6597 rdev_for_each(rdev2, mddev) {
6598 if (test_bit(Journal, &rdev2->flags)) {
6603 if (has_journal || mddev->bitmap) {
6607 set_bit(Journal, &rdev->flags);
6610 * check whether the device shows up in other nodes
6612 if (mddev_is_clustered(mddev)) {
6613 if (info->state & (1 << MD_DISK_CANDIDATE))
6614 set_bit(Candidate, &rdev->flags);
6615 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6616 /* --add initiated by this node */
6617 err = md_cluster_ops->add_new_disk(mddev, rdev);
6625 rdev->raid_disk = -1;
6626 err = bind_rdev_to_array(rdev, mddev);
6631 if (mddev_is_clustered(mddev)) {
6632 if (info->state & (1 << MD_DISK_CANDIDATE)) {
6634 err = md_cluster_ops->new_disk_ack(mddev,
6637 md_kick_rdev_from_array(rdev);
6641 md_cluster_ops->add_new_disk_cancel(mddev);
6643 err = add_bound_rdev(rdev);
6647 err = add_bound_rdev(rdev);
6652 /* otherwise, add_new_disk is only allowed
6653 * for major_version==0 superblocks
6655 if (mddev->major_version != 0) {
6656 pr_warn("%s: ADD_NEW_DISK not supported\n", mdname(mddev));
6660 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6662 rdev = md_import_device(dev, -1, 0);
6664 pr_warn("md: error, md_import_device() returned %ld\n",
6666 return PTR_ERR(rdev);
6668 rdev->desc_nr = info->number;
6669 if (info->raid_disk < mddev->raid_disks)
6670 rdev->raid_disk = info->raid_disk;
6672 rdev->raid_disk = -1;
6674 if (rdev->raid_disk < mddev->raid_disks)
6675 if (info->state & (1<<MD_DISK_SYNC))
6676 set_bit(In_sync, &rdev->flags);
6678 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6679 set_bit(WriteMostly, &rdev->flags);
6680 if (info->state & (1<<MD_DISK_FAILFAST))
6681 set_bit(FailFast, &rdev->flags);
6683 if (!mddev->persistent) {
6684 pr_debug("md: nonpersistent superblock ...\n");
6685 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6687 rdev->sb_start = calc_dev_sboffset(rdev);
6688 rdev->sectors = rdev->sb_start;
6690 err = bind_rdev_to_array(rdev, mddev);
6700 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6702 char b[BDEVNAME_SIZE];
6703 struct md_rdev *rdev;
6708 rdev = find_rdev(mddev, dev);
6712 if (rdev->raid_disk < 0)
6715 clear_bit(Blocked, &rdev->flags);
6716 remove_and_add_spares(mddev, rdev);
6718 if (rdev->raid_disk >= 0)
6722 if (mddev_is_clustered(mddev))
6723 md_cluster_ops->remove_disk(mddev, rdev);
6725 md_kick_rdev_from_array(rdev);
6726 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6728 md_wakeup_thread(mddev->thread);
6730 md_update_sb(mddev, 1);
6731 md_new_event(mddev);
6735 pr_debug("md: cannot remove active disk %s from %s ...\n",
6736 bdevname(rdev->bdev,b), mdname(mddev));
6740 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6742 char b[BDEVNAME_SIZE];
6744 struct md_rdev *rdev;
6749 if (mddev->major_version != 0) {
6750 pr_warn("%s: HOT_ADD may only be used with version-0 superblocks.\n",
6754 if (!mddev->pers->hot_add_disk) {
6755 pr_warn("%s: personality does not support diskops!\n",
6760 rdev = md_import_device(dev, -1, 0);
6762 pr_warn("md: error, md_import_device() returned %ld\n",
6767 if (mddev->persistent)
6768 rdev->sb_start = calc_dev_sboffset(rdev);
6770 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6772 rdev->sectors = rdev->sb_start;
6774 if (test_bit(Faulty, &rdev->flags)) {
6775 pr_warn("md: can not hot-add faulty %s disk to %s!\n",
6776 bdevname(rdev->bdev,b), mdname(mddev));
6781 clear_bit(In_sync, &rdev->flags);
6783 rdev->saved_raid_disk = -1;
6784 err = bind_rdev_to_array(rdev, mddev);
6789 * The rest should better be atomic, we can have disk failures
6790 * noticed in interrupt contexts ...
6793 rdev->raid_disk = -1;
6795 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6797 md_update_sb(mddev, 1);
6799 * Kick recovery, maybe this spare has to be added to the
6800 * array immediately.
6802 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6803 md_wakeup_thread(mddev->thread);
6804 md_new_event(mddev);
6812 static int set_bitmap_file(struct mddev *mddev, int fd)
6817 if (!mddev->pers->quiesce || !mddev->thread)
6819 if (mddev->recovery || mddev->sync_thread)
6821 /* we should be able to change the bitmap.. */
6825 struct inode *inode;
6828 if (mddev->bitmap || mddev->bitmap_info.file)
6829 return -EEXIST; /* cannot add when bitmap is present */
6833 pr_warn("%s: error: failed to get bitmap file\n",
6838 inode = f->f_mapping->host;
6839 if (!S_ISREG(inode->i_mode)) {
6840 pr_warn("%s: error: bitmap file must be a regular file\n",
6843 } else if (!(f->f_mode & FMODE_WRITE)) {
6844 pr_warn("%s: error: bitmap file must open for write\n",
6847 } else if (atomic_read(&inode->i_writecount) != 1) {
6848 pr_warn("%s: error: bitmap file is already in use\n",
6856 mddev->bitmap_info.file = f;
6857 mddev->bitmap_info.offset = 0; /* file overrides offset */
6858 } else if (mddev->bitmap == NULL)
6859 return -ENOENT; /* cannot remove what isn't there */
6863 struct bitmap *bitmap;
6865 bitmap = md_bitmap_create(mddev, -1);
6866 mddev_suspend(mddev);
6867 if (!IS_ERR(bitmap)) {
6868 mddev->bitmap = bitmap;
6869 err = md_bitmap_load(mddev);
6871 err = PTR_ERR(bitmap);
6873 md_bitmap_destroy(mddev);
6876 mddev_resume(mddev);
6877 } else if (fd < 0) {
6878 mddev_suspend(mddev);
6879 md_bitmap_destroy(mddev);
6880 mddev_resume(mddev);
6884 struct file *f = mddev->bitmap_info.file;
6886 spin_lock(&mddev->lock);
6887 mddev->bitmap_info.file = NULL;
6888 spin_unlock(&mddev->lock);
6897 * set_array_info is used two different ways
6898 * The original usage is when creating a new array.
6899 * In this usage, raid_disks is > 0 and it together with
6900 * level, size, not_persistent,layout,chunksize determine the
6901 * shape of the array.
6902 * This will always create an array with a type-0.90.0 superblock.
6903 * The newer usage is when assembling an array.
6904 * In this case raid_disks will be 0, and the major_version field is
6905 * use to determine which style super-blocks are to be found on the devices.
6906 * The minor and patch _version numbers are also kept incase the
6907 * super_block handler wishes to interpret them.
6909 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6912 if (info->raid_disks == 0) {
6913 /* just setting version number for superblock loading */
6914 if (info->major_version < 0 ||
6915 info->major_version >= ARRAY_SIZE(super_types) ||
6916 super_types[info->major_version].name == NULL) {
6917 /* maybe try to auto-load a module? */
6918 pr_warn("md: superblock version %d not known\n",
6919 info->major_version);
6922 mddev->major_version = info->major_version;
6923 mddev->minor_version = info->minor_version;
6924 mddev->patch_version = info->patch_version;
6925 mddev->persistent = !info->not_persistent;
6926 /* ensure mddev_put doesn't delete this now that there
6927 * is some minimal configuration.
6929 mddev->ctime = ktime_get_real_seconds();
6932 mddev->major_version = MD_MAJOR_VERSION;
6933 mddev->minor_version = MD_MINOR_VERSION;
6934 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6935 mddev->ctime = ktime_get_real_seconds();
6937 mddev->level = info->level;
6938 mddev->clevel[0] = 0;
6939 mddev->dev_sectors = 2 * (sector_t)info->size;
6940 mddev->raid_disks = info->raid_disks;
6941 /* don't set md_minor, it is determined by which /dev/md* was
6944 if (info->state & (1<<MD_SB_CLEAN))
6945 mddev->recovery_cp = MaxSector;
6947 mddev->recovery_cp = 0;
6948 mddev->persistent = ! info->not_persistent;
6949 mddev->external = 0;
6951 mddev->layout = info->layout;
6952 if (mddev->level == 0)
6953 /* Cannot trust RAID0 layout info here */
6955 mddev->chunk_sectors = info->chunk_size >> 9;
6957 if (mddev->persistent) {
6958 mddev->max_disks = MD_SB_DISKS;
6960 mddev->sb_flags = 0;
6962 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
6964 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6965 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6966 mddev->bitmap_info.offset = 0;
6968 mddev->reshape_position = MaxSector;
6971 * Generate a 128 bit UUID
6973 get_random_bytes(mddev->uuid, 16);
6975 mddev->new_level = mddev->level;
6976 mddev->new_chunk_sectors = mddev->chunk_sectors;
6977 mddev->new_layout = mddev->layout;
6978 mddev->delta_disks = 0;
6979 mddev->reshape_backwards = 0;
6984 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6986 lockdep_assert_held(&mddev->reconfig_mutex);
6988 if (mddev->external_size)
6991 mddev->array_sectors = array_sectors;
6993 EXPORT_SYMBOL(md_set_array_sectors);
6995 static int update_size(struct mddev *mddev, sector_t num_sectors)
6997 struct md_rdev *rdev;
6999 int fit = (num_sectors == 0);
7000 sector_t old_dev_sectors = mddev->dev_sectors;
7002 if (mddev->pers->resize == NULL)
7004 /* The "num_sectors" is the number of sectors of each device that
7005 * is used. This can only make sense for arrays with redundancy.
7006 * linear and raid0 always use whatever space is available. We can only
7007 * consider changing this number if no resync or reconstruction is
7008 * happening, and if the new size is acceptable. It must fit before the
7009 * sb_start or, if that is <data_offset, it must fit before the size
7010 * of each device. If num_sectors is zero, we find the largest size
7013 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7019 rdev_for_each(rdev, mddev) {
7020 sector_t avail = rdev->sectors;
7022 if (fit && (num_sectors == 0 || num_sectors > avail))
7023 num_sectors = avail;
7024 if (avail < num_sectors)
7027 rv = mddev->pers->resize(mddev, num_sectors);
7029 if (mddev_is_clustered(mddev))
7030 md_cluster_ops->update_size(mddev, old_dev_sectors);
7031 else if (mddev->queue) {
7032 set_capacity(mddev->gendisk, mddev->array_sectors);
7033 revalidate_disk(mddev->gendisk);
7039 static int update_raid_disks(struct mddev *mddev, int raid_disks)
7042 struct md_rdev *rdev;
7043 /* change the number of raid disks */
7044 if (mddev->pers->check_reshape == NULL)
7048 if (raid_disks <= 0 ||
7049 (mddev->max_disks && raid_disks >= mddev->max_disks))
7051 if (mddev->sync_thread ||
7052 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
7053 mddev->reshape_position != MaxSector)
7056 rdev_for_each(rdev, mddev) {
7057 if (mddev->raid_disks < raid_disks &&
7058 rdev->data_offset < rdev->new_data_offset)
7060 if (mddev->raid_disks > raid_disks &&
7061 rdev->data_offset > rdev->new_data_offset)
7065 mddev->delta_disks = raid_disks - mddev->raid_disks;
7066 if (mddev->delta_disks < 0)
7067 mddev->reshape_backwards = 1;
7068 else if (mddev->delta_disks > 0)
7069 mddev->reshape_backwards = 0;
7071 rv = mddev->pers->check_reshape(mddev);
7073 mddev->delta_disks = 0;
7074 mddev->reshape_backwards = 0;
7080 * update_array_info is used to change the configuration of an
7082 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
7083 * fields in the info are checked against the array.
7084 * Any differences that cannot be handled will cause an error.
7085 * Normally, only one change can be managed at a time.
7087 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
7093 /* calculate expected state,ignoring low bits */
7094 if (mddev->bitmap && mddev->bitmap_info.offset)
7095 state |= (1 << MD_SB_BITMAP_PRESENT);
7097 if (mddev->major_version != info->major_version ||
7098 mddev->minor_version != info->minor_version ||
7099 /* mddev->patch_version != info->patch_version || */
7100 mddev->ctime != info->ctime ||
7101 mddev->level != info->level ||
7102 /* mddev->layout != info->layout || */
7103 mddev->persistent != !info->not_persistent ||
7104 mddev->chunk_sectors != info->chunk_size >> 9 ||
7105 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
7106 ((state^info->state) & 0xfffffe00)
7109 /* Check there is only one change */
7110 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7112 if (mddev->raid_disks != info->raid_disks)
7114 if (mddev->layout != info->layout)
7116 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
7123 if (mddev->layout != info->layout) {
7125 * we don't need to do anything at the md level, the
7126 * personality will take care of it all.
7128 if (mddev->pers->check_reshape == NULL)
7131 mddev->new_layout = info->layout;
7132 rv = mddev->pers->check_reshape(mddev);
7134 mddev->new_layout = mddev->layout;
7138 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
7139 rv = update_size(mddev, (sector_t)info->size * 2);
7141 if (mddev->raid_disks != info->raid_disks)
7142 rv = update_raid_disks(mddev, info->raid_disks);
7144 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
7145 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
7149 if (mddev->recovery || mddev->sync_thread) {
7153 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
7154 struct bitmap *bitmap;
7155 /* add the bitmap */
7156 if (mddev->bitmap) {
7160 if (mddev->bitmap_info.default_offset == 0) {
7164 mddev->bitmap_info.offset =
7165 mddev->bitmap_info.default_offset;
7166 mddev->bitmap_info.space =
7167 mddev->bitmap_info.default_space;
7168 bitmap = md_bitmap_create(mddev, -1);
7169 mddev_suspend(mddev);
7170 if (!IS_ERR(bitmap)) {
7171 mddev->bitmap = bitmap;
7172 rv = md_bitmap_load(mddev);
7174 rv = PTR_ERR(bitmap);
7176 md_bitmap_destroy(mddev);
7177 mddev_resume(mddev);
7179 /* remove the bitmap */
7180 if (!mddev->bitmap) {
7184 if (mddev->bitmap->storage.file) {
7188 if (mddev->bitmap_info.nodes) {
7189 /* hold PW on all the bitmap lock */
7190 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
7191 pr_warn("md: can't change bitmap to none since the array is in use by more than one node\n");
7193 md_cluster_ops->unlock_all_bitmaps(mddev);
7197 mddev->bitmap_info.nodes = 0;
7198 md_cluster_ops->leave(mddev);
7200 mddev_suspend(mddev);
7201 md_bitmap_destroy(mddev);
7202 mddev_resume(mddev);
7203 mddev->bitmap_info.offset = 0;
7206 md_update_sb(mddev, 1);
7212 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
7214 struct md_rdev *rdev;
7217 if (mddev->pers == NULL)
7221 rdev = md_find_rdev_rcu(mddev, dev);
7225 md_error(mddev, rdev);
7226 if (!test_bit(Faulty, &rdev->flags))
7234 * We have a problem here : there is no easy way to give a CHS
7235 * virtual geometry. We currently pretend that we have a 2 heads
7236 * 4 sectors (with a BIG number of cylinders...). This drives
7237 * dosfs just mad... ;-)
7239 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
7241 struct mddev *mddev = bdev->bd_disk->private_data;
7245 geo->cylinders = mddev->array_sectors / 8;
7249 static inline bool md_ioctl_valid(unsigned int cmd)
7254 case GET_ARRAY_INFO:
7255 case GET_BITMAP_FILE:
7258 case HOT_REMOVE_DISK:
7261 case RESTART_ARRAY_RW:
7263 case SET_ARRAY_INFO:
7264 case SET_BITMAP_FILE:
7265 case SET_DISK_FAULTY:
7268 case CLUSTERED_DISK_NACK:
7275 static int md_ioctl(struct block_device *bdev, fmode_t mode,
7276 unsigned int cmd, unsigned long arg)
7279 void __user *argp = (void __user *)arg;
7280 struct mddev *mddev = NULL;
7282 bool did_set_md_closing = false;
7284 if (!md_ioctl_valid(cmd))
7289 case GET_ARRAY_INFO:
7293 if (!capable(CAP_SYS_ADMIN))
7298 * Commands dealing with the RAID driver but not any
7303 err = get_version(argp);
7309 autostart_arrays(arg);
7316 * Commands creating/starting a new array:
7319 mddev = bdev->bd_disk->private_data;
7326 /* Some actions do not requires the mutex */
7328 case GET_ARRAY_INFO:
7329 if (!mddev->raid_disks && !mddev->external)
7332 err = get_array_info(mddev, argp);
7336 if (!mddev->raid_disks && !mddev->external)
7339 err = get_disk_info(mddev, argp);
7342 case SET_DISK_FAULTY:
7343 err = set_disk_faulty(mddev, new_decode_dev(arg));
7346 case GET_BITMAP_FILE:
7347 err = get_bitmap_file(mddev, argp);
7352 if (cmd == ADD_NEW_DISK)
7353 /* need to ensure md_delayed_delete() has completed */
7354 flush_workqueue(md_misc_wq);
7356 if (cmd == HOT_REMOVE_DISK)
7357 /* need to ensure recovery thread has run */
7358 wait_event_interruptible_timeout(mddev->sb_wait,
7359 !test_bit(MD_RECOVERY_NEEDED,
7361 msecs_to_jiffies(5000));
7362 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
7363 /* Need to flush page cache, and ensure no-one else opens
7366 mutex_lock(&mddev->open_mutex);
7367 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
7368 mutex_unlock(&mddev->open_mutex);
7372 WARN_ON_ONCE(test_bit(MD_CLOSING, &mddev->flags));
7373 set_bit(MD_CLOSING, &mddev->flags);
7374 did_set_md_closing = true;
7375 mutex_unlock(&mddev->open_mutex);
7376 sync_blockdev(bdev);
7378 err = mddev_lock(mddev);
7380 pr_debug("md: ioctl lock interrupted, reason %d, cmd %d\n",
7385 if (cmd == SET_ARRAY_INFO) {
7386 mdu_array_info_t info;
7388 memset(&info, 0, sizeof(info));
7389 else if (copy_from_user(&info, argp, sizeof(info))) {
7394 err = update_array_info(mddev, &info);
7396 pr_warn("md: couldn't update array info. %d\n", err);
7401 if (!list_empty(&mddev->disks)) {
7402 pr_warn("md: array %s already has disks!\n", mdname(mddev));
7406 if (mddev->raid_disks) {
7407 pr_warn("md: array %s already initialised!\n", mdname(mddev));
7411 err = set_array_info(mddev, &info);
7413 pr_warn("md: couldn't set array info. %d\n", err);
7420 * Commands querying/configuring an existing array:
7422 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
7423 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
7424 if ((!mddev->raid_disks && !mddev->external)
7425 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
7426 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
7427 && cmd != GET_BITMAP_FILE) {
7433 * Commands even a read-only array can execute:
7436 case RESTART_ARRAY_RW:
7437 err = restart_array(mddev);
7441 err = do_md_stop(mddev, 0, bdev);
7445 err = md_set_readonly(mddev, bdev);
7448 case HOT_REMOVE_DISK:
7449 err = hot_remove_disk(mddev, new_decode_dev(arg));
7453 /* We can support ADD_NEW_DISK on read-only arrays
7454 * only if we are re-adding a preexisting device.
7455 * So require mddev->pers and MD_DISK_SYNC.
7458 mdu_disk_info_t info;
7459 if (copy_from_user(&info, argp, sizeof(info)))
7461 else if (!(info.state & (1<<MD_DISK_SYNC)))
7462 /* Need to clear read-only for this */
7465 err = add_new_disk(mddev, &info);
7471 if (get_user(ro, (int __user *)(arg))) {
7477 /* if the bdev is going readonly the value of mddev->ro
7478 * does not matter, no writes are coming
7483 /* are we are already prepared for writes? */
7487 /* transitioning to readauto need only happen for
7488 * arrays that call md_write_start
7491 err = restart_array(mddev);
7494 set_disk_ro(mddev->gendisk, 0);
7501 * The remaining ioctls are changing the state of the
7502 * superblock, so we do not allow them on read-only arrays.
7504 if (mddev->ro && mddev->pers) {
7505 if (mddev->ro == 2) {
7507 sysfs_notify_dirent_safe(mddev->sysfs_state);
7508 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7509 /* mddev_unlock will wake thread */
7510 /* If a device failed while we were read-only, we
7511 * need to make sure the metadata is updated now.
7513 if (test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags)) {
7514 mddev_unlock(mddev);
7515 wait_event(mddev->sb_wait,
7516 !test_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags) &&
7517 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
7518 mddev_lock_nointr(mddev);
7529 mdu_disk_info_t info;
7530 if (copy_from_user(&info, argp, sizeof(info)))
7533 err = add_new_disk(mddev, &info);
7537 case CLUSTERED_DISK_NACK:
7538 if (mddev_is_clustered(mddev))
7539 md_cluster_ops->new_disk_ack(mddev, false);
7545 err = hot_add_disk(mddev, new_decode_dev(arg));
7549 err = do_md_run(mddev);
7552 case SET_BITMAP_FILE:
7553 err = set_bitmap_file(mddev, (int)arg);
7562 if (mddev->hold_active == UNTIL_IOCTL &&
7564 mddev->hold_active = 0;
7565 mddev_unlock(mddev);
7567 if(did_set_md_closing)
7568 clear_bit(MD_CLOSING, &mddev->flags);
7571 #ifdef CONFIG_COMPAT
7572 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7573 unsigned int cmd, unsigned long arg)
7576 case HOT_REMOVE_DISK:
7578 case SET_DISK_FAULTY:
7579 case SET_BITMAP_FILE:
7580 /* These take in integer arg, do not convert */
7583 arg = (unsigned long)compat_ptr(arg);
7587 return md_ioctl(bdev, mode, cmd, arg);
7589 #endif /* CONFIG_COMPAT */
7591 static int md_open(struct block_device *bdev, fmode_t mode)
7594 * Succeed if we can lock the mddev, which confirms that
7595 * it isn't being stopped right now.
7597 struct mddev *mddev = mddev_find(bdev->bd_dev);
7603 if (mddev->gendisk != bdev->bd_disk) {
7604 /* we are racing with mddev_put which is discarding this
7608 /* Wait until bdev->bd_disk is definitely gone */
7609 flush_workqueue(md_misc_wq);
7610 /* Then retry the open from the top */
7611 return -ERESTARTSYS;
7613 BUG_ON(mddev != bdev->bd_disk->private_data);
7615 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7618 if (test_bit(MD_CLOSING, &mddev->flags)) {
7619 mutex_unlock(&mddev->open_mutex);
7625 atomic_inc(&mddev->openers);
7626 mutex_unlock(&mddev->open_mutex);
7628 check_disk_change(bdev);
7635 static void md_release(struct gendisk *disk, fmode_t mode)
7637 struct mddev *mddev = disk->private_data;
7640 atomic_dec(&mddev->openers);
7644 static int md_media_changed(struct gendisk *disk)
7646 struct mddev *mddev = disk->private_data;
7648 return mddev->changed;
7651 static int md_revalidate(struct gendisk *disk)
7653 struct mddev *mddev = disk->private_data;
7658 static const struct block_device_operations md_fops =
7660 .owner = THIS_MODULE,
7662 .release = md_release,
7664 #ifdef CONFIG_COMPAT
7665 .compat_ioctl = md_compat_ioctl,
7667 .getgeo = md_getgeo,
7668 .media_changed = md_media_changed,
7669 .revalidate_disk= md_revalidate,
7672 static int md_thread(void *arg)
7674 struct md_thread *thread = arg;
7677 * md_thread is a 'system-thread', it's priority should be very
7678 * high. We avoid resource deadlocks individually in each
7679 * raid personality. (RAID5 does preallocation) We also use RR and
7680 * the very same RT priority as kswapd, thus we will never get
7681 * into a priority inversion deadlock.
7683 * we definitely have to have equal or higher priority than
7684 * bdflush, otherwise bdflush will deadlock if there are too
7685 * many dirty RAID5 blocks.
7688 allow_signal(SIGKILL);
7689 while (!kthread_should_stop()) {
7691 /* We need to wait INTERRUPTIBLE so that
7692 * we don't add to the load-average.
7693 * That means we need to be sure no signals are
7696 if (signal_pending(current))
7697 flush_signals(current);
7699 wait_event_interruptible_timeout
7701 test_bit(THREAD_WAKEUP, &thread->flags)
7702 || kthread_should_stop() || kthread_should_park(),
7705 clear_bit(THREAD_WAKEUP, &thread->flags);
7706 if (kthread_should_park())
7708 if (!kthread_should_stop())
7709 thread->run(thread);
7715 void md_wakeup_thread(struct md_thread *thread)
7718 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7719 set_bit(THREAD_WAKEUP, &thread->flags);
7720 wake_up(&thread->wqueue);
7723 EXPORT_SYMBOL(md_wakeup_thread);
7725 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7726 struct mddev *mddev, const char *name)
7728 struct md_thread *thread;
7730 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7734 init_waitqueue_head(&thread->wqueue);
7737 thread->mddev = mddev;
7738 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7739 thread->tsk = kthread_run(md_thread, thread,
7741 mdname(thread->mddev),
7743 if (IS_ERR(thread->tsk)) {
7749 EXPORT_SYMBOL(md_register_thread);
7751 void md_unregister_thread(struct md_thread **threadp)
7753 struct md_thread *thread = *threadp;
7756 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7757 /* Locking ensures that mddev_unlock does not wake_up a
7758 * non-existent thread
7760 spin_lock(&pers_lock);
7762 spin_unlock(&pers_lock);
7764 kthread_stop(thread->tsk);
7767 EXPORT_SYMBOL(md_unregister_thread);
7769 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7771 if (!rdev || test_bit(Faulty, &rdev->flags))
7774 if (!mddev->pers || !mddev->pers->error_handler)
7776 mddev->pers->error_handler(mddev,rdev);
7777 if (mddev->degraded)
7778 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7779 sysfs_notify_dirent_safe(rdev->sysfs_state);
7780 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7781 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7782 md_wakeup_thread(mddev->thread);
7783 if (mddev->event_work.func)
7784 queue_work(md_misc_wq, &mddev->event_work);
7785 md_new_event(mddev);
7787 EXPORT_SYMBOL(md_error);
7789 /* seq_file implementation /proc/mdstat */
7791 static void status_unused(struct seq_file *seq)
7794 struct md_rdev *rdev;
7796 seq_printf(seq, "unused devices: ");
7798 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7799 char b[BDEVNAME_SIZE];
7801 seq_printf(seq, "%s ",
7802 bdevname(rdev->bdev,b));
7805 seq_printf(seq, "<none>");
7807 seq_printf(seq, "\n");
7810 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7812 sector_t max_sectors, resync, res;
7813 unsigned long dt, db = 0;
7814 sector_t rt, curr_mark_cnt, resync_mark_cnt;
7815 int scale, recovery_active;
7816 unsigned int per_milli;
7818 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7819 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7820 max_sectors = mddev->resync_max_sectors;
7822 max_sectors = mddev->dev_sectors;
7824 resync = mddev->curr_resync;
7826 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7827 /* Still cleaning up */
7828 resync = max_sectors;
7829 } else if (resync > max_sectors)
7830 resync = max_sectors;
7832 resync -= atomic_read(&mddev->recovery_active);
7835 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery)) {
7836 struct md_rdev *rdev;
7838 rdev_for_each(rdev, mddev)
7839 if (rdev->raid_disk >= 0 &&
7840 !test_bit(Faulty, &rdev->flags) &&
7841 rdev->recovery_offset != MaxSector &&
7842 rdev->recovery_offset) {
7843 seq_printf(seq, "\trecover=REMOTE");
7846 if (mddev->reshape_position != MaxSector)
7847 seq_printf(seq, "\treshape=REMOTE");
7849 seq_printf(seq, "\tresync=REMOTE");
7852 if (mddev->recovery_cp < MaxSector) {
7853 seq_printf(seq, "\tresync=PENDING");
7859 seq_printf(seq, "\tresync=DELAYED");
7863 WARN_ON(max_sectors == 0);
7864 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7865 * in a sector_t, and (max_sectors>>scale) will fit in a
7866 * u32, as those are the requirements for sector_div.
7867 * Thus 'scale' must be at least 10
7870 if (sizeof(sector_t) > sizeof(unsigned long)) {
7871 while ( max_sectors/2 > (1ULL<<(scale+32)))
7874 res = (resync>>scale)*1000;
7875 sector_div(res, (u32)((max_sectors>>scale)+1));
7879 int i, x = per_milli/50, y = 20-x;
7880 seq_printf(seq, "[");
7881 for (i = 0; i < x; i++)
7882 seq_printf(seq, "=");
7883 seq_printf(seq, ">");
7884 for (i = 0; i < y; i++)
7885 seq_printf(seq, ".");
7886 seq_printf(seq, "] ");
7888 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7889 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7891 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7893 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7894 "resync" : "recovery"))),
7895 per_milli/10, per_milli % 10,
7896 (unsigned long long) resync/2,
7897 (unsigned long long) max_sectors/2);
7900 * dt: time from mark until now
7901 * db: blocks written from mark until now
7902 * rt: remaining time
7904 * rt is a sector_t, which is always 64bit now. We are keeping
7905 * the original algorithm, but it is not really necessary.
7907 * Original algorithm:
7908 * So we divide before multiply in case it is 32bit and close
7910 * We scale the divisor (db) by 32 to avoid losing precision
7911 * near the end of resync when the number of remaining sectors
7913 * We then divide rt by 32 after multiplying by db to compensate.
7914 * The '+1' avoids division by zero if db is very small.
7916 dt = ((jiffies - mddev->resync_mark) / HZ);
7919 curr_mark_cnt = mddev->curr_mark_cnt;
7920 recovery_active = atomic_read(&mddev->recovery_active);
7921 resync_mark_cnt = mddev->resync_mark_cnt;
7923 if (curr_mark_cnt >= (recovery_active + resync_mark_cnt))
7924 db = curr_mark_cnt - (recovery_active + resync_mark_cnt);
7926 rt = max_sectors - resync; /* number of remaining sectors */
7927 rt = div64_u64(rt, db/32+1);
7931 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7932 ((unsigned long)rt % 60)/6);
7934 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7938 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7940 struct list_head *tmp;
7942 struct mddev *mddev;
7950 spin_lock(&all_mddevs_lock);
7951 list_for_each(tmp,&all_mddevs)
7953 mddev = list_entry(tmp, struct mddev, all_mddevs);
7955 spin_unlock(&all_mddevs_lock);
7958 spin_unlock(&all_mddevs_lock);
7960 return (void*)2;/* tail */
7964 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7966 struct list_head *tmp;
7967 struct mddev *next_mddev, *mddev = v;
7973 spin_lock(&all_mddevs_lock);
7975 tmp = all_mddevs.next;
7977 tmp = mddev->all_mddevs.next;
7978 if (tmp != &all_mddevs)
7979 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7981 next_mddev = (void*)2;
7984 spin_unlock(&all_mddevs_lock);
7992 static void md_seq_stop(struct seq_file *seq, void *v)
7994 struct mddev *mddev = v;
7996 if (mddev && v != (void*)1 && v != (void*)2)
8000 static int md_seq_show(struct seq_file *seq, void *v)
8002 struct mddev *mddev = v;
8004 struct md_rdev *rdev;
8006 if (v == (void*)1) {
8007 struct md_personality *pers;
8008 seq_printf(seq, "Personalities : ");
8009 spin_lock(&pers_lock);
8010 list_for_each_entry(pers, &pers_list, list)
8011 seq_printf(seq, "[%s] ", pers->name);
8013 spin_unlock(&pers_lock);
8014 seq_printf(seq, "\n");
8015 seq->poll_event = atomic_read(&md_event_count);
8018 if (v == (void*)2) {
8023 spin_lock(&mddev->lock);
8024 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
8025 seq_printf(seq, "%s : %sactive", mdname(mddev),
8026 mddev->pers ? "" : "in");
8029 seq_printf(seq, " (read-only)");
8031 seq_printf(seq, " (auto-read-only)");
8032 seq_printf(seq, " %s", mddev->pers->name);
8037 rdev_for_each_rcu(rdev, mddev) {
8038 char b[BDEVNAME_SIZE];
8039 seq_printf(seq, " %s[%d]",
8040 bdevname(rdev->bdev,b), rdev->desc_nr);
8041 if (test_bit(WriteMostly, &rdev->flags))
8042 seq_printf(seq, "(W)");
8043 if (test_bit(Journal, &rdev->flags))
8044 seq_printf(seq, "(J)");
8045 if (test_bit(Faulty, &rdev->flags)) {
8046 seq_printf(seq, "(F)");
8049 if (rdev->raid_disk < 0)
8050 seq_printf(seq, "(S)"); /* spare */
8051 if (test_bit(Replacement, &rdev->flags))
8052 seq_printf(seq, "(R)");
8053 sectors += rdev->sectors;
8057 if (!list_empty(&mddev->disks)) {
8059 seq_printf(seq, "\n %llu blocks",
8060 (unsigned long long)
8061 mddev->array_sectors / 2);
8063 seq_printf(seq, "\n %llu blocks",
8064 (unsigned long long)sectors / 2);
8066 if (mddev->persistent) {
8067 if (mddev->major_version != 0 ||
8068 mddev->minor_version != 90) {
8069 seq_printf(seq," super %d.%d",
8070 mddev->major_version,
8071 mddev->minor_version);
8073 } else if (mddev->external)
8074 seq_printf(seq, " super external:%s",
8075 mddev->metadata_type);
8077 seq_printf(seq, " super non-persistent");
8080 mddev->pers->status(seq, mddev);
8081 seq_printf(seq, "\n ");
8082 if (mddev->pers->sync_request) {
8083 if (status_resync(seq, mddev))
8084 seq_printf(seq, "\n ");
8087 seq_printf(seq, "\n ");
8089 md_bitmap_status(seq, mddev->bitmap);
8091 seq_printf(seq, "\n");
8093 spin_unlock(&mddev->lock);
8098 static const struct seq_operations md_seq_ops = {
8099 .start = md_seq_start,
8100 .next = md_seq_next,
8101 .stop = md_seq_stop,
8102 .show = md_seq_show,
8105 static int md_seq_open(struct inode *inode, struct file *file)
8107 struct seq_file *seq;
8110 error = seq_open(file, &md_seq_ops);
8114 seq = file->private_data;
8115 seq->poll_event = atomic_read(&md_event_count);
8119 static int md_unloading;
8120 static __poll_t mdstat_poll(struct file *filp, poll_table *wait)
8122 struct seq_file *seq = filp->private_data;
8126 return EPOLLIN|EPOLLRDNORM|EPOLLERR|EPOLLPRI;
8127 poll_wait(filp, &md_event_waiters, wait);
8129 /* always allow read */
8130 mask = EPOLLIN | EPOLLRDNORM;
8132 if (seq->poll_event != atomic_read(&md_event_count))
8133 mask |= EPOLLERR | EPOLLPRI;
8137 static const struct file_operations md_seq_fops = {
8138 .owner = THIS_MODULE,
8139 .open = md_seq_open,
8141 .llseek = seq_lseek,
8142 .release = seq_release,
8143 .poll = mdstat_poll,
8146 int register_md_personality(struct md_personality *p)
8148 pr_debug("md: %s personality registered for level %d\n",
8150 spin_lock(&pers_lock);
8151 list_add_tail(&p->list, &pers_list);
8152 spin_unlock(&pers_lock);
8155 EXPORT_SYMBOL(register_md_personality);
8157 int unregister_md_personality(struct md_personality *p)
8159 pr_debug("md: %s personality unregistered\n", p->name);
8160 spin_lock(&pers_lock);
8161 list_del_init(&p->list);
8162 spin_unlock(&pers_lock);
8165 EXPORT_SYMBOL(unregister_md_personality);
8167 int register_md_cluster_operations(struct md_cluster_operations *ops,
8168 struct module *module)
8171 spin_lock(&pers_lock);
8172 if (md_cluster_ops != NULL)
8175 md_cluster_ops = ops;
8176 md_cluster_mod = module;
8178 spin_unlock(&pers_lock);
8181 EXPORT_SYMBOL(register_md_cluster_operations);
8183 int unregister_md_cluster_operations(void)
8185 spin_lock(&pers_lock);
8186 md_cluster_ops = NULL;
8187 spin_unlock(&pers_lock);
8190 EXPORT_SYMBOL(unregister_md_cluster_operations);
8192 int md_setup_cluster(struct mddev *mddev, int nodes)
8194 if (!md_cluster_ops)
8195 request_module("md-cluster");
8196 spin_lock(&pers_lock);
8197 /* ensure module won't be unloaded */
8198 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
8199 pr_warn("can't find md-cluster module or get it's reference.\n");
8200 spin_unlock(&pers_lock);
8203 spin_unlock(&pers_lock);
8205 return md_cluster_ops->join(mddev, nodes);
8208 void md_cluster_stop(struct mddev *mddev)
8210 if (!md_cluster_ops)
8212 md_cluster_ops->leave(mddev);
8213 module_put(md_cluster_mod);
8216 static int is_mddev_idle(struct mddev *mddev, int init)
8218 struct md_rdev *rdev;
8224 rdev_for_each_rcu(rdev, mddev) {
8225 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
8226 curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
8227 atomic_read(&disk->sync_io);
8228 /* sync IO will cause sync_io to increase before the disk_stats
8229 * as sync_io is counted when a request starts, and
8230 * disk_stats is counted when it completes.
8231 * So resync activity will cause curr_events to be smaller than
8232 * when there was no such activity.
8233 * non-sync IO will cause disk_stat to increase without
8234 * increasing sync_io so curr_events will (eventually)
8235 * be larger than it was before. Once it becomes
8236 * substantially larger, the test below will cause
8237 * the array to appear non-idle, and resync will slow
8239 * If there is a lot of outstanding resync activity when
8240 * we set last_event to curr_events, then all that activity
8241 * completing might cause the array to appear non-idle
8242 * and resync will be slowed down even though there might
8243 * not have been non-resync activity. This will only
8244 * happen once though. 'last_events' will soon reflect
8245 * the state where there is little or no outstanding
8246 * resync requests, and further resync activity will
8247 * always make curr_events less than last_events.
8250 if (init || curr_events - rdev->last_events > 64) {
8251 rdev->last_events = curr_events;
8259 void md_done_sync(struct mddev *mddev, int blocks, int ok)
8261 /* another "blocks" (512byte) blocks have been synced */
8262 atomic_sub(blocks, &mddev->recovery_active);
8263 wake_up(&mddev->recovery_wait);
8265 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8266 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
8267 md_wakeup_thread(mddev->thread);
8268 // stop recovery, signal do_sync ....
8271 EXPORT_SYMBOL(md_done_sync);
8273 /* md_write_start(mddev, bi)
8274 * If we need to update some array metadata (e.g. 'active' flag
8275 * in superblock) before writing, schedule a superblock update
8276 * and wait for it to complete.
8277 * A return value of 'false' means that the write wasn't recorded
8278 * and cannot proceed as the array is being suspend.
8280 bool md_write_start(struct mddev *mddev, struct bio *bi)
8284 if (bio_data_dir(bi) != WRITE)
8287 BUG_ON(mddev->ro == 1);
8288 if (mddev->ro == 2) {
8289 /* need to switch to read/write */
8291 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8292 md_wakeup_thread(mddev->thread);
8293 md_wakeup_thread(mddev->sync_thread);
8297 percpu_ref_get(&mddev->writes_pending);
8298 smp_mb(); /* Match smp_mb in set_in_sync() */
8299 if (mddev->safemode == 1)
8300 mddev->safemode = 0;
8301 /* sync_checkers is always 0 when writes_pending is in per-cpu mode */
8302 if (mddev->in_sync || mddev->sync_checkers) {
8303 spin_lock(&mddev->lock);
8304 if (mddev->in_sync) {
8306 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8307 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8308 md_wakeup_thread(mddev->thread);
8311 spin_unlock(&mddev->lock);
8315 sysfs_notify_dirent_safe(mddev->sysfs_state);
8316 if (!mddev->has_superblocks)
8318 wait_event(mddev->sb_wait,
8319 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags) ||
8321 if (test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
8322 percpu_ref_put(&mddev->writes_pending);
8327 EXPORT_SYMBOL(md_write_start);
8329 /* md_write_inc can only be called when md_write_start() has
8330 * already been called at least once of the current request.
8331 * It increments the counter and is useful when a single request
8332 * is split into several parts. Each part causes an increment and
8333 * so needs a matching md_write_end().
8334 * Unlike md_write_start(), it is safe to call md_write_inc() inside
8335 * a spinlocked region.
8337 void md_write_inc(struct mddev *mddev, struct bio *bi)
8339 if (bio_data_dir(bi) != WRITE)
8341 WARN_ON_ONCE(mddev->in_sync || mddev->ro);
8342 percpu_ref_get(&mddev->writes_pending);
8344 EXPORT_SYMBOL(md_write_inc);
8346 void md_write_end(struct mddev *mddev)
8348 percpu_ref_put(&mddev->writes_pending);
8350 if (mddev->safemode == 2)
8351 md_wakeup_thread(mddev->thread);
8352 else if (mddev->safemode_delay)
8353 /* The roundup() ensures this only performs locking once
8354 * every ->safemode_delay jiffies
8356 mod_timer(&mddev->safemode_timer,
8357 roundup(jiffies, mddev->safemode_delay) +
8358 mddev->safemode_delay);
8361 EXPORT_SYMBOL(md_write_end);
8363 /* md_allow_write(mddev)
8364 * Calling this ensures that the array is marked 'active' so that writes
8365 * may proceed without blocking. It is important to call this before
8366 * attempting a GFP_KERNEL allocation while holding the mddev lock.
8367 * Must be called with mddev_lock held.
8369 void md_allow_write(struct mddev *mddev)
8375 if (!mddev->pers->sync_request)
8378 spin_lock(&mddev->lock);
8379 if (mddev->in_sync) {
8381 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8382 set_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
8383 if (mddev->safemode_delay &&
8384 mddev->safemode == 0)
8385 mddev->safemode = 1;
8386 spin_unlock(&mddev->lock);
8387 md_update_sb(mddev, 0);
8388 sysfs_notify_dirent_safe(mddev->sysfs_state);
8389 /* wait for the dirty state to be recorded in the metadata */
8390 wait_event(mddev->sb_wait,
8391 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags));
8393 spin_unlock(&mddev->lock);
8395 EXPORT_SYMBOL_GPL(md_allow_write);
8397 #define SYNC_MARKS 10
8398 #define SYNC_MARK_STEP (3*HZ)
8399 #define UPDATE_FREQUENCY (5*60*HZ)
8400 void md_do_sync(struct md_thread *thread)
8402 struct mddev *mddev = thread->mddev;
8403 struct mddev *mddev2;
8404 unsigned int currspeed = 0, window;
8405 sector_t max_sectors,j, io_sectors, recovery_done;
8406 unsigned long mark[SYNC_MARKS];
8407 unsigned long update_time;
8408 sector_t mark_cnt[SYNC_MARKS];
8410 struct list_head *tmp;
8411 sector_t last_check;
8413 struct md_rdev *rdev;
8414 char *desc, *action = NULL;
8415 struct blk_plug plug;
8418 /* just incase thread restarts... */
8419 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8420 test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
8422 if (mddev->ro) {/* never try to sync a read-only array */
8423 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8427 if (mddev_is_clustered(mddev)) {
8428 ret = md_cluster_ops->resync_start(mddev);
8432 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
8433 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
8434 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
8435 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
8436 && ((unsigned long long)mddev->curr_resync_completed
8437 < (unsigned long long)mddev->resync_max_sectors))
8441 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8442 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
8443 desc = "data-check";
8445 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8446 desc = "requested-resync";
8450 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
8455 mddev->last_sync_action = action ?: desc;
8457 /* we overload curr_resync somewhat here.
8458 * 0 == not engaged in resync at all
8459 * 2 == checking that there is no conflict with another sync
8460 * 1 == like 2, but have yielded to allow conflicting resync to
8462 * other == active in resync - this many blocks
8464 * Before starting a resync we must have set curr_resync to
8465 * 2, and then checked that every "conflicting" array has curr_resync
8466 * less than ours. When we find one that is the same or higher
8467 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
8468 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
8469 * This will mean we have to start checking from the beginning again.
8474 int mddev2_minor = -1;
8475 mddev->curr_resync = 2;
8478 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8480 for_each_mddev(mddev2, tmp) {
8481 if (mddev2 == mddev)
8483 if (!mddev->parallel_resync
8484 && mddev2->curr_resync
8485 && match_mddev_units(mddev, mddev2)) {
8487 if (mddev < mddev2 && mddev->curr_resync == 2) {
8488 /* arbitrarily yield */
8489 mddev->curr_resync = 1;
8490 wake_up(&resync_wait);
8492 if (mddev > mddev2 && mddev->curr_resync == 1)
8493 /* no need to wait here, we can wait the next
8494 * time 'round when curr_resync == 2
8497 /* We need to wait 'interruptible' so as not to
8498 * contribute to the load average, and not to
8499 * be caught by 'softlockup'
8501 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
8502 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8503 mddev2->curr_resync >= mddev->curr_resync) {
8504 if (mddev2_minor != mddev2->md_minor) {
8505 mddev2_minor = mddev2->md_minor;
8506 pr_info("md: delaying %s of %s until %s has finished (they share one or more physical units)\n",
8507 desc, mdname(mddev),
8511 if (signal_pending(current))
8512 flush_signals(current);
8514 finish_wait(&resync_wait, &wq);
8517 finish_wait(&resync_wait, &wq);
8520 } while (mddev->curr_resync < 2);
8523 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8524 /* resync follows the size requested by the personality,
8525 * which defaults to physical size, but can be virtual size
8527 max_sectors = mddev->resync_max_sectors;
8528 atomic64_set(&mddev->resync_mismatches, 0);
8529 /* we don't use the checkpoint if there's a bitmap */
8530 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8531 j = mddev->resync_min;
8532 else if (!mddev->bitmap)
8533 j = mddev->recovery_cp;
8535 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) {
8536 max_sectors = mddev->resync_max_sectors;
8538 * If the original node aborts reshaping then we continue the
8539 * reshaping, so set j again to avoid restart reshape from the
8542 if (mddev_is_clustered(mddev) &&
8543 mddev->reshape_position != MaxSector)
8544 j = mddev->reshape_position;
8546 /* recovery follows the physical size of devices */
8547 max_sectors = mddev->dev_sectors;
8550 rdev_for_each_rcu(rdev, mddev)
8551 if (rdev->raid_disk >= 0 &&
8552 !test_bit(Journal, &rdev->flags) &&
8553 !test_bit(Faulty, &rdev->flags) &&
8554 !test_bit(In_sync, &rdev->flags) &&
8555 rdev->recovery_offset < j)
8556 j = rdev->recovery_offset;
8559 /* If there is a bitmap, we need to make sure all
8560 * writes that started before we added a spare
8561 * complete before we start doing a recovery.
8562 * Otherwise the write might complete and (via
8563 * bitmap_endwrite) set a bit in the bitmap after the
8564 * recovery has checked that bit and skipped that
8567 if (mddev->bitmap) {
8568 mddev->pers->quiesce(mddev, 1);
8569 mddev->pers->quiesce(mddev, 0);
8573 pr_info("md: %s of RAID array %s\n", desc, mdname(mddev));
8574 pr_debug("md: minimum _guaranteed_ speed: %d KB/sec/disk.\n", speed_min(mddev));
8575 pr_debug("md: using maximum available idle IO bandwidth (but not more than %d KB/sec) for %s.\n",
8576 speed_max(mddev), desc);
8578 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
8581 for (m = 0; m < SYNC_MARKS; m++) {
8583 mark_cnt[m] = io_sectors;
8586 mddev->resync_mark = mark[last_mark];
8587 mddev->resync_mark_cnt = mark_cnt[last_mark];
8590 * Tune reconstruction:
8592 window = 32 * (PAGE_SIZE / 512);
8593 pr_debug("md: using %dk window, over a total of %lluk.\n",
8594 window/2, (unsigned long long)max_sectors/2);
8596 atomic_set(&mddev->recovery_active, 0);
8600 pr_debug("md: resuming %s of %s from checkpoint.\n",
8601 desc, mdname(mddev));
8602 mddev->curr_resync = j;
8604 mddev->curr_resync = 3; /* no longer delayed */
8605 mddev->curr_resync_completed = j;
8606 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8607 md_new_event(mddev);
8608 update_time = jiffies;
8610 blk_start_plug(&plug);
8611 while (j < max_sectors) {
8616 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8617 ((mddev->curr_resync > mddev->curr_resync_completed &&
8618 (mddev->curr_resync - mddev->curr_resync_completed)
8619 > (max_sectors >> 4)) ||
8620 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8621 (j - mddev->curr_resync_completed)*2
8622 >= mddev->resync_max - mddev->curr_resync_completed ||
8623 mddev->curr_resync_completed > mddev->resync_max
8625 /* time to update curr_resync_completed */
8626 wait_event(mddev->recovery_wait,
8627 atomic_read(&mddev->recovery_active) == 0);
8628 mddev->curr_resync_completed = j;
8629 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8630 j > mddev->recovery_cp)
8631 mddev->recovery_cp = j;
8632 update_time = jiffies;
8633 set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags);
8634 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8637 while (j >= mddev->resync_max &&
8638 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8639 /* As this condition is controlled by user-space,
8640 * we can block indefinitely, so use '_interruptible'
8641 * to avoid triggering warnings.
8643 flush_signals(current); /* just in case */
8644 wait_event_interruptible(mddev->recovery_wait,
8645 mddev->resync_max > j
8646 || test_bit(MD_RECOVERY_INTR,
8650 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8653 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8655 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8659 if (!skipped) { /* actual IO requested */
8660 io_sectors += sectors;
8661 atomic_add(sectors, &mddev->recovery_active);
8664 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8668 if (j > max_sectors)
8669 /* when skipping, extra large numbers can be returned. */
8672 mddev->curr_resync = j;
8673 mddev->curr_mark_cnt = io_sectors;
8674 if (last_check == 0)
8675 /* this is the earliest that rebuild will be
8676 * visible in /proc/mdstat
8678 md_new_event(mddev);
8680 if (last_check + window > io_sectors || j == max_sectors)
8683 last_check = io_sectors;
8685 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8687 int next = (last_mark+1) % SYNC_MARKS;
8689 mddev->resync_mark = mark[next];
8690 mddev->resync_mark_cnt = mark_cnt[next];
8691 mark[next] = jiffies;
8692 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8696 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8700 * this loop exits only if either when we are slower than
8701 * the 'hard' speed limit, or the system was IO-idle for
8703 * the system might be non-idle CPU-wise, but we only care
8704 * about not overloading the IO subsystem. (things like an
8705 * e2fsck being done on the RAID array should execute fast)
8709 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8710 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8711 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8713 if (currspeed > speed_min(mddev)) {
8714 if (currspeed > speed_max(mddev)) {
8718 if (!is_mddev_idle(mddev, 0)) {
8720 * Give other IO more of a chance.
8721 * The faster the devices, the less we wait.
8723 wait_event(mddev->recovery_wait,
8724 !atomic_read(&mddev->recovery_active));
8728 pr_info("md: %s: %s %s.\n",mdname(mddev), desc,
8729 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8730 ? "interrupted" : "done");
8732 * this also signals 'finished resyncing' to md_stop
8734 blk_finish_plug(&plug);
8735 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8737 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8738 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8739 mddev->curr_resync > 3) {
8740 mddev->curr_resync_completed = mddev->curr_resync;
8741 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8743 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8745 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8746 mddev->curr_resync > 3) {
8747 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8748 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8749 if (mddev->curr_resync >= mddev->recovery_cp) {
8750 pr_debug("md: checkpointing %s of %s.\n",
8751 desc, mdname(mddev));
8752 if (test_bit(MD_RECOVERY_ERROR,
8754 mddev->recovery_cp =
8755 mddev->curr_resync_completed;
8757 mddev->recovery_cp =
8761 mddev->recovery_cp = MaxSector;
8763 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8764 mddev->curr_resync = MaxSector;
8765 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8766 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery)) {
8768 rdev_for_each_rcu(rdev, mddev)
8769 if (rdev->raid_disk >= 0 &&
8770 mddev->delta_disks >= 0 &&
8771 !test_bit(Journal, &rdev->flags) &&
8772 !test_bit(Faulty, &rdev->flags) &&
8773 !test_bit(In_sync, &rdev->flags) &&
8774 rdev->recovery_offset < mddev->curr_resync)
8775 rdev->recovery_offset = mddev->curr_resync;
8781 /* set CHANGE_PENDING here since maybe another update is needed,
8782 * so other nodes are informed. It should be harmless for normal
8784 set_mask_bits(&mddev->sb_flags, 0,
8785 BIT(MD_SB_CHANGE_PENDING) | BIT(MD_SB_CHANGE_DEVS));
8787 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8788 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8789 mddev->delta_disks > 0 &&
8790 mddev->pers->finish_reshape &&
8791 mddev->pers->size &&
8793 mddev_lock_nointr(mddev);
8794 md_set_array_sectors(mddev, mddev->pers->size(mddev, 0, 0));
8795 mddev_unlock(mddev);
8796 if (!mddev_is_clustered(mddev)) {
8797 set_capacity(mddev->gendisk, mddev->array_sectors);
8798 revalidate_disk(mddev->gendisk);
8802 spin_lock(&mddev->lock);
8803 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8804 /* We completed so min/max setting can be forgotten if used. */
8805 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8806 mddev->resync_min = 0;
8807 mddev->resync_max = MaxSector;
8808 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8809 mddev->resync_min = mddev->curr_resync_completed;
8810 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8811 mddev->curr_resync = 0;
8812 spin_unlock(&mddev->lock);
8814 wake_up(&resync_wait);
8815 md_wakeup_thread(mddev->thread);
8818 EXPORT_SYMBOL_GPL(md_do_sync);
8820 static int remove_and_add_spares(struct mddev *mddev,
8821 struct md_rdev *this)
8823 struct md_rdev *rdev;
8826 bool remove_some = false;
8828 if (this && test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
8829 /* Mustn't remove devices when resync thread is running */
8832 rdev_for_each(rdev, mddev) {
8833 if ((this == NULL || rdev == this) &&
8834 rdev->raid_disk >= 0 &&
8835 !test_bit(Blocked, &rdev->flags) &&
8836 test_bit(Faulty, &rdev->flags) &&
8837 atomic_read(&rdev->nr_pending)==0) {
8838 /* Faulty non-Blocked devices with nr_pending == 0
8839 * never get nr_pending incremented,
8840 * never get Faulty cleared, and never get Blocked set.
8841 * So we can synchronize_rcu now rather than once per device
8844 set_bit(RemoveSynchronized, &rdev->flags);
8850 rdev_for_each(rdev, mddev) {
8851 if ((this == NULL || rdev == this) &&
8852 rdev->raid_disk >= 0 &&
8853 !test_bit(Blocked, &rdev->flags) &&
8854 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8855 (!test_bit(In_sync, &rdev->flags) &&
8856 !test_bit(Journal, &rdev->flags))) &&
8857 atomic_read(&rdev->nr_pending)==0)) {
8858 if (mddev->pers->hot_remove_disk(
8859 mddev, rdev) == 0) {
8860 sysfs_unlink_rdev(mddev, rdev);
8861 rdev->saved_raid_disk = rdev->raid_disk;
8862 rdev->raid_disk = -1;
8866 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8867 clear_bit(RemoveSynchronized, &rdev->flags);
8870 if (removed && mddev->kobj.sd)
8871 sysfs_notify(&mddev->kobj, NULL, "degraded");
8873 if (this && removed)
8876 rdev_for_each(rdev, mddev) {
8877 if (this && this != rdev)
8879 if (test_bit(Candidate, &rdev->flags))
8881 if (rdev->raid_disk >= 0 &&
8882 !test_bit(In_sync, &rdev->flags) &&
8883 !test_bit(Journal, &rdev->flags) &&
8884 !test_bit(Faulty, &rdev->flags))
8886 if (rdev->raid_disk >= 0)
8888 if (test_bit(Faulty, &rdev->flags))
8890 if (!test_bit(Journal, &rdev->flags)) {
8892 ! (rdev->saved_raid_disk >= 0 &&
8893 !test_bit(Bitmap_sync, &rdev->flags)))
8896 rdev->recovery_offset = 0;
8899 hot_add_disk(mddev, rdev) == 0) {
8900 if (sysfs_link_rdev(mddev, rdev))
8901 /* failure here is OK */;
8902 if (!test_bit(Journal, &rdev->flags))
8904 md_new_event(mddev);
8905 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8910 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
8914 static void md_start_sync(struct work_struct *ws)
8916 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8918 mddev->sync_thread = md_register_thread(md_do_sync,
8921 if (!mddev->sync_thread) {
8922 pr_warn("%s: could not start resync thread...\n",
8924 /* leave the spares where they are, it shouldn't hurt */
8925 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8926 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8927 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8928 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8929 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8930 wake_up(&resync_wait);
8931 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8933 if (mddev->sysfs_action)
8934 sysfs_notify_dirent_safe(mddev->sysfs_action);
8936 md_wakeup_thread(mddev->sync_thread);
8937 sysfs_notify_dirent_safe(mddev->sysfs_action);
8938 md_new_event(mddev);
8942 * This routine is regularly called by all per-raid-array threads to
8943 * deal with generic issues like resync and super-block update.
8944 * Raid personalities that don't have a thread (linear/raid0) do not
8945 * need this as they never do any recovery or update the superblock.
8947 * It does not do any resync itself, but rather "forks" off other threads
8948 * to do that as needed.
8949 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8950 * "->recovery" and create a thread at ->sync_thread.
8951 * When the thread finishes it sets MD_RECOVERY_DONE
8952 * and wakeups up this thread which will reap the thread and finish up.
8953 * This thread also removes any faulty devices (with nr_pending == 0).
8955 * The overall approach is:
8956 * 1/ if the superblock needs updating, update it.
8957 * 2/ If a recovery thread is running, don't do anything else.
8958 * 3/ If recovery has finished, clean up, possibly marking spares active.
8959 * 4/ If there are any faulty devices, remove them.
8960 * 5/ If array is degraded, try to add spares devices
8961 * 6/ If array has spares or is not in-sync, start a resync thread.
8963 void md_check_recovery(struct mddev *mddev)
8965 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags) && mddev->sb_flags) {
8966 /* Write superblock - thread that called mddev_suspend()
8967 * holds reconfig_mutex for us.
8969 set_bit(MD_UPDATING_SB, &mddev->flags);
8970 smp_mb__after_atomic();
8971 if (test_bit(MD_ALLOW_SB_UPDATE, &mddev->flags))
8972 md_update_sb(mddev, 0);
8973 clear_bit_unlock(MD_UPDATING_SB, &mddev->flags);
8974 wake_up(&mddev->sb_wait);
8977 if (mddev->suspended)
8981 md_bitmap_daemon_work(mddev);
8983 if (signal_pending(current)) {
8984 if (mddev->pers->sync_request && !mddev->external) {
8985 pr_debug("md: %s in immediate safe mode\n",
8987 mddev->safemode = 2;
8989 flush_signals(current);
8992 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8995 (mddev->sb_flags & ~ (1<<MD_SB_CHANGE_PENDING)) ||
8996 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8997 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8998 (mddev->external == 0 && mddev->safemode == 1) ||
8999 (mddev->safemode == 2
9000 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
9004 if (mddev_trylock(mddev)) {
9006 bool try_set_sync = mddev->safemode != 0;
9008 if (!mddev->external && mddev->safemode == 1)
9009 mddev->safemode = 0;
9012 struct md_rdev *rdev;
9013 if (!mddev->external && mddev->in_sync)
9014 /* 'Blocked' flag not needed as failed devices
9015 * will be recorded if array switched to read/write.
9016 * Leaving it set will prevent the device
9017 * from being removed.
9019 rdev_for_each(rdev, mddev)
9020 clear_bit(Blocked, &rdev->flags);
9021 /* On a read-only array we can:
9022 * - remove failed devices
9023 * - add already-in_sync devices if the array itself
9025 * As we only add devices that are already in-sync,
9026 * we can activate the spares immediately.
9028 remove_and_add_spares(mddev, NULL);
9029 /* There is no thread, but we need to call
9030 * ->spare_active and clear saved_raid_disk
9032 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
9033 md_reap_sync_thread(mddev);
9034 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9035 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9036 clear_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags);
9040 if (mddev_is_clustered(mddev)) {
9041 struct md_rdev *rdev;
9042 /* kick the device if another node issued a
9045 rdev_for_each(rdev, mddev) {
9046 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
9047 rdev->raid_disk < 0)
9048 md_kick_rdev_from_array(rdev);
9052 if (try_set_sync && !mddev->external && !mddev->in_sync) {
9053 spin_lock(&mddev->lock);
9055 spin_unlock(&mddev->lock);
9058 if (mddev->sb_flags)
9059 md_update_sb(mddev, 0);
9061 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
9062 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
9063 /* resync/recovery still happening */
9064 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9067 if (mddev->sync_thread) {
9068 md_reap_sync_thread(mddev);
9071 /* Set RUNNING before clearing NEEDED to avoid
9072 * any transients in the value of "sync_action".
9074 mddev->curr_resync_completed = 0;
9075 spin_lock(&mddev->lock);
9076 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9077 spin_unlock(&mddev->lock);
9078 /* Clear some bits that don't mean anything, but
9081 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
9082 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9084 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
9085 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
9087 /* no recovery is running.
9088 * remove any failed drives, then
9089 * add spares if possible.
9090 * Spares are also removed and re-added, to allow
9091 * the personality to fail the re-add.
9094 if (mddev->reshape_position != MaxSector) {
9095 if (mddev->pers->check_reshape == NULL ||
9096 mddev->pers->check_reshape(mddev) != 0)
9097 /* Cannot proceed */
9099 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9100 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9101 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
9102 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9103 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9104 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9105 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9106 } else if (mddev->recovery_cp < MaxSector) {
9107 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9108 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
9109 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
9110 /* nothing to be done ... */
9113 if (mddev->pers->sync_request) {
9115 /* We are adding a device or devices to an array
9116 * which has the bitmap stored on all devices.
9117 * So make sure all bitmap pages get written
9119 md_bitmap_write_all(mddev->bitmap);
9121 INIT_WORK(&mddev->del_work, md_start_sync);
9122 queue_work(md_misc_wq, &mddev->del_work);
9126 if (!mddev->sync_thread) {
9127 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9128 wake_up(&resync_wait);
9129 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
9131 if (mddev->sysfs_action)
9132 sysfs_notify_dirent_safe(mddev->sysfs_action);
9135 wake_up(&mddev->sb_wait);
9136 mddev_unlock(mddev);
9139 EXPORT_SYMBOL(md_check_recovery);
9141 void md_reap_sync_thread(struct mddev *mddev)
9143 struct md_rdev *rdev;
9144 sector_t old_dev_sectors = mddev->dev_sectors;
9145 bool is_reshaped = false;
9147 /* resync has finished, collect result */
9148 md_unregister_thread(&mddev->sync_thread);
9149 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
9150 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
9151 mddev->degraded != mddev->raid_disks) {
9153 /* activate any spares */
9154 if (mddev->pers->spare_active(mddev)) {
9155 sysfs_notify(&mddev->kobj, NULL,
9157 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
9160 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
9161 mddev->pers->finish_reshape) {
9162 mddev->pers->finish_reshape(mddev);
9163 if (mddev_is_clustered(mddev))
9167 /* If array is no-longer degraded, then any saved_raid_disk
9168 * information must be scrapped.
9170 if (!mddev->degraded)
9171 rdev_for_each(rdev, mddev)
9172 rdev->saved_raid_disk = -1;
9174 md_update_sb(mddev, 1);
9175 /* MD_SB_CHANGE_PENDING should be cleared by md_update_sb, so we can
9176 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
9178 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
9179 md_cluster_ops->resync_finish(mddev);
9180 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
9181 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
9182 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
9183 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
9184 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
9185 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
9187 * We call md_cluster_ops->update_size here because sync_size could
9188 * be changed by md_update_sb, and MD_RECOVERY_RESHAPE is cleared,
9189 * so it is time to update size across cluster.
9191 if (mddev_is_clustered(mddev) && is_reshaped
9192 && !test_bit(MD_CLOSING, &mddev->flags))
9193 md_cluster_ops->update_size(mddev, old_dev_sectors);
9194 wake_up(&resync_wait);
9195 /* flag recovery needed just to double check */
9196 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9197 sysfs_notify_dirent_safe(mddev->sysfs_action);
9198 md_new_event(mddev);
9199 if (mddev->event_work.func)
9200 queue_work(md_misc_wq, &mddev->event_work);
9202 EXPORT_SYMBOL(md_reap_sync_thread);
9204 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
9206 sysfs_notify_dirent_safe(rdev->sysfs_state);
9207 wait_event_timeout(rdev->blocked_wait,
9208 !test_bit(Blocked, &rdev->flags) &&
9209 !test_bit(BlockedBadBlocks, &rdev->flags),
9210 msecs_to_jiffies(5000));
9211 rdev_dec_pending(rdev, mddev);
9213 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
9215 void md_finish_reshape(struct mddev *mddev)
9217 /* called be personality module when reshape completes. */
9218 struct md_rdev *rdev;
9220 rdev_for_each(rdev, mddev) {
9221 if (rdev->data_offset > rdev->new_data_offset)
9222 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
9224 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
9225 rdev->data_offset = rdev->new_data_offset;
9228 EXPORT_SYMBOL(md_finish_reshape);
9230 /* Bad block management */
9232 /* Returns 1 on success, 0 on failure */
9233 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9236 struct mddev *mddev = rdev->mddev;
9239 s += rdev->new_data_offset;
9241 s += rdev->data_offset;
9242 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
9244 /* Make sure they get written out promptly */
9245 if (test_bit(ExternalBbl, &rdev->flags))
9246 sysfs_notify(&rdev->kobj, NULL,
9247 "unacknowledged_bad_blocks");
9248 sysfs_notify_dirent_safe(rdev->sysfs_state);
9249 set_mask_bits(&mddev->sb_flags, 0,
9250 BIT(MD_SB_CHANGE_CLEAN) | BIT(MD_SB_CHANGE_PENDING));
9251 md_wakeup_thread(rdev->mddev->thread);
9256 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
9258 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
9263 s += rdev->new_data_offset;
9265 s += rdev->data_offset;
9266 rv = badblocks_clear(&rdev->badblocks, s, sectors);
9267 if ((rv == 0) && test_bit(ExternalBbl, &rdev->flags))
9268 sysfs_notify(&rdev->kobj, NULL, "bad_blocks");
9271 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
9273 static int md_notify_reboot(struct notifier_block *this,
9274 unsigned long code, void *x)
9276 struct list_head *tmp;
9277 struct mddev *mddev;
9280 for_each_mddev(mddev, tmp) {
9281 if (mddev_trylock(mddev)) {
9283 __md_stop_writes(mddev);
9284 if (mddev->persistent)
9285 mddev->safemode = 2;
9286 mddev_unlock(mddev);
9291 * certain more exotic SCSI devices are known to be
9292 * volatile wrt too early system reboots. While the
9293 * right place to handle this issue is the given
9294 * driver, we do want to have a safe RAID driver ...
9302 static struct notifier_block md_notifier = {
9303 .notifier_call = md_notify_reboot,
9305 .priority = INT_MAX, /* before any real devices */
9308 static void md_geninit(void)
9310 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9312 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9315 static int __init md_init(void)
9319 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9323 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9327 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9330 if ((ret = register_blkdev(0, "mdp")) < 0)
9334 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9335 md_probe, NULL, NULL);
9336 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9337 md_probe, NULL, NULL);
9339 register_reboot_notifier(&md_notifier);
9340 raid_table_header = register_sysctl_table(raid_root_table);
9346 unregister_blkdev(MD_MAJOR, "md");
9348 destroy_workqueue(md_misc_wq);
9350 destroy_workqueue(md_wq);
9355 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9357 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9358 struct md_rdev *rdev2;
9360 char b[BDEVNAME_SIZE];
9363 * If size is changed in another node then we need to
9364 * do resize as well.
9366 if (mddev->dev_sectors != le64_to_cpu(sb->size)) {
9367 ret = mddev->pers->resize(mddev, le64_to_cpu(sb->size));
9369 pr_info("md-cluster: resize failed\n");
9371 md_bitmap_update_sb(mddev->bitmap);
9374 /* Check for change of roles in the active devices */
9375 rdev_for_each(rdev2, mddev) {
9376 if (test_bit(Faulty, &rdev2->flags))
9379 /* Check if the roles changed */
9380 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9382 if (test_bit(Candidate, &rdev2->flags)) {
9383 if (role == 0xfffe) {
9384 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9385 md_kick_rdev_from_array(rdev2);
9389 clear_bit(Candidate, &rdev2->flags);
9392 if (role != rdev2->raid_disk) {
9394 * got activated except reshape is happening.
9396 if (rdev2->raid_disk == -1 && role != 0xffff &&
9397 !(le32_to_cpu(sb->feature_map) &
9398 MD_FEATURE_RESHAPE_ACTIVE)) {
9399 rdev2->saved_raid_disk = role;
9400 ret = remove_and_add_spares(mddev, rdev2);
9401 pr_info("Activated spare: %s\n",
9402 bdevname(rdev2->bdev,b));
9403 /* wakeup mddev->thread here, so array could
9404 * perform resync with the new activated disk */
9405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
9406 md_wakeup_thread(mddev->thread);
9409 * We just want to do the minimum to mark the disk
9410 * as faulty. The recovery is performed by the
9411 * one who initiated the error.
9413 if ((role == 0xfffe) || (role == 0xfffd)) {
9414 md_error(mddev, rdev2);
9415 clear_bit(Blocked, &rdev2->flags);
9420 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9421 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9424 * Since mddev->delta_disks has already updated in update_raid_disks,
9425 * so it is time to check reshape.
9427 if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9428 (le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9430 * reshape is happening in the remote node, we need to
9431 * update reshape_position and call start_reshape.
9433 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
9434 if (mddev->pers->update_reshape_pos)
9435 mddev->pers->update_reshape_pos(mddev);
9436 if (mddev->pers->start_reshape)
9437 mddev->pers->start_reshape(mddev);
9438 } else if (test_bit(MD_RESYNCING_REMOTE, &mddev->recovery) &&
9439 mddev->reshape_position != MaxSector &&
9440 !(le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
9441 /* reshape is just done in another node. */
9442 mddev->reshape_position = MaxSector;
9443 if (mddev->pers->update_reshape_pos)
9444 mddev->pers->update_reshape_pos(mddev);
9447 /* Finally set the event to be up to date */
9448 mddev->events = le64_to_cpu(sb->events);
9451 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9454 struct page *swapout = rdev->sb_page;
9455 struct mdp_superblock_1 *sb;
9457 /* Store the sb page of the rdev in the swapout temporary
9458 * variable in case we err in the future
9460 rdev->sb_page = NULL;
9461 err = alloc_disk_sb(rdev);
9463 ClearPageUptodate(rdev->sb_page);
9464 rdev->sb_loaded = 0;
9465 err = super_types[mddev->major_version].
9466 load_super(rdev, NULL, mddev->minor_version);
9469 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9470 __func__, __LINE__, rdev->desc_nr, err);
9472 put_page(rdev->sb_page);
9473 rdev->sb_page = swapout;
9474 rdev->sb_loaded = 1;
9478 sb = page_address(rdev->sb_page);
9479 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9483 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9484 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9486 /* The other node finished recovery, call spare_active to set
9487 * device In_sync and mddev->degraded
9489 if (rdev->recovery_offset == MaxSector &&
9490 !test_bit(In_sync, &rdev->flags) &&
9491 mddev->pers->spare_active(mddev))
9492 sysfs_notify(&mddev->kobj, NULL, "degraded");
9498 void md_reload_sb(struct mddev *mddev, int nr)
9500 struct md_rdev *rdev;
9504 rdev_for_each_rcu(rdev, mddev) {
9505 if (rdev->desc_nr == nr)
9509 if (!rdev || rdev->desc_nr != nr) {
9510 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9514 err = read_rdev(mddev, rdev);
9518 check_sb_changes(mddev, rdev);
9520 /* Read all rdev's to update recovery_offset */
9521 rdev_for_each_rcu(rdev, mddev) {
9522 if (!test_bit(Faulty, &rdev->flags))
9523 read_rdev(mddev, rdev);
9526 EXPORT_SYMBOL(md_reload_sb);
9531 * Searches all registered partitions for autorun RAID arrays
9535 static DEFINE_MUTEX(detected_devices_mutex);
9536 static LIST_HEAD(all_detected_devices);
9537 struct detected_devices_node {
9538 struct list_head list;
9542 void md_autodetect_dev(dev_t dev)
9544 struct detected_devices_node *node_detected_dev;
9546 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9547 if (node_detected_dev) {
9548 node_detected_dev->dev = dev;
9549 mutex_lock(&detected_devices_mutex);
9550 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9551 mutex_unlock(&detected_devices_mutex);
9555 static void autostart_arrays(int part)
9557 struct md_rdev *rdev;
9558 struct detected_devices_node *node_detected_dev;
9560 int i_scanned, i_passed;
9565 pr_info("md: Autodetecting RAID arrays.\n");
9567 mutex_lock(&detected_devices_mutex);
9568 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9570 node_detected_dev = list_entry(all_detected_devices.next,
9571 struct detected_devices_node, list);
9572 list_del(&node_detected_dev->list);
9573 dev = node_detected_dev->dev;
9574 kfree(node_detected_dev);
9575 mutex_unlock(&detected_devices_mutex);
9576 rdev = md_import_device(dev,0, 90);
9577 mutex_lock(&detected_devices_mutex);
9581 if (test_bit(Faulty, &rdev->flags))
9584 set_bit(AutoDetected, &rdev->flags);
9585 list_add(&rdev->same_set, &pending_raid_disks);
9588 mutex_unlock(&detected_devices_mutex);
9590 pr_debug("md: Scanned %d and added %d devices.\n", i_scanned, i_passed);
9592 autorun_devices(part);
9595 #endif /* !MODULE */
9597 static __exit void md_exit(void)
9599 struct mddev *mddev;
9600 struct list_head *tmp;
9603 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9604 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9606 unregister_blkdev(MD_MAJOR,"md");
9607 unregister_blkdev(mdp_major, "mdp");
9608 unregister_reboot_notifier(&md_notifier);
9609 unregister_sysctl_table(raid_table_header);
9611 /* We cannot unload the modules while some process is
9612 * waiting for us in select() or poll() - wake them up
9615 while (waitqueue_active(&md_event_waiters)) {
9616 /* not safe to leave yet */
9617 wake_up(&md_event_waiters);
9621 remove_proc_entry("mdstat", NULL);
9623 for_each_mddev(mddev, tmp) {
9624 export_array(mddev);
9626 mddev->hold_active = 0;
9628 * for_each_mddev() will call mddev_put() at the end of each
9629 * iteration. As the mddev is now fully clear, this will
9630 * schedule the mddev for destruction by a workqueue, and the
9631 * destroy_workqueue() below will wait for that to complete.
9634 destroy_workqueue(md_misc_wq);
9635 destroy_workqueue(md_wq);
9638 subsys_initcall(md_init);
9639 module_exit(md_exit)
9641 static int get_ro(char *buffer, const struct kernel_param *kp)
9643 return sprintf(buffer, "%d", start_readonly);
9645 static int set_ro(const char *val, const struct kernel_param *kp)
9647 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9650 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9651 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9652 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9653 module_param(create_on_open, bool, S_IRUSR|S_IWUSR);
9655 MODULE_LICENSE("GPL");
9656 MODULE_DESCRIPTION("MD RAID framework");
9658 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);