2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/badblocks.h>
38 #include <linux/sysctl.h>
39 #include <linux/seq_file.h>
41 #include <linux/poll.h>
42 #include <linux/ctype.h>
43 #include <linux/string.h>
44 #include <linux/hdreg.h>
45 #include <linux/proc_fs.h>
46 #include <linux/random.h>
47 #include <linux/module.h>
48 #include <linux/reboot.h>
49 #include <linux/file.h>
50 #include <linux/compat.h>
51 #include <linux/delay.h>
52 #include <linux/raid/md_p.h>
53 #include <linux/raid/md_u.h>
54 #include <linux/slab.h>
57 #include "md-cluster.h"
60 static void autostart_arrays(int part);
63 /* pers_list is a list of registered personalities protected
65 * pers_lock does extra service to protect accesses to
66 * mddev->thread when the mutex cannot be held.
68 static LIST_HEAD(pers_list);
69 static DEFINE_SPINLOCK(pers_lock);
71 struct md_cluster_operations *md_cluster_ops;
72 EXPORT_SYMBOL(md_cluster_ops);
73 struct module *md_cluster_mod;
74 EXPORT_SYMBOL(md_cluster_mod);
76 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
77 static struct workqueue_struct *md_wq;
78 static struct workqueue_struct *md_misc_wq;
80 static int remove_and_add_spares(struct mddev *mddev,
81 struct md_rdev *this);
82 static void mddev_detach(struct mddev *mddev);
85 * Default number of read corrections we'll attempt on an rdev
86 * before ejecting it from the array. We divide the read error
87 * count by 2 for every hour elapsed between read errors.
89 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
91 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
92 * is 1000 KB/sec, so the extra system load does not show up that much.
93 * Increase it if you want to have more _guaranteed_ speed. Note that
94 * the RAID driver will use the maximum available bandwidth if the IO
95 * subsystem is idle. There is also an 'absolute maximum' reconstruction
96 * speed limit - in case reconstruction slows down your system despite
99 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
100 * or /sys/block/mdX/md/sync_speed_{min,max}
103 static int sysctl_speed_limit_min = 1000;
104 static int sysctl_speed_limit_max = 200000;
105 static inline int speed_min(struct mddev *mddev)
107 return mddev->sync_speed_min ?
108 mddev->sync_speed_min : sysctl_speed_limit_min;
111 static inline int speed_max(struct mddev *mddev)
113 return mddev->sync_speed_max ?
114 mddev->sync_speed_max : sysctl_speed_limit_max;
117 static struct ctl_table_header *raid_table_header;
119 static struct ctl_table raid_table[] = {
121 .procname = "speed_limit_min",
122 .data = &sysctl_speed_limit_min,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
128 .procname = "speed_limit_max",
129 .data = &sysctl_speed_limit_max,
130 .maxlen = sizeof(int),
131 .mode = S_IRUGO|S_IWUSR,
132 .proc_handler = proc_dointvec,
137 static struct ctl_table raid_dir_table[] = {
141 .mode = S_IRUGO|S_IXUGO,
147 static struct ctl_table raid_root_table[] = {
152 .child = raid_dir_table,
157 static const struct block_device_operations md_fops;
159 static int start_readonly;
162 * like bio_clone, but with a local bio set
165 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
170 if (!mddev || !mddev->bio_set)
171 return bio_alloc(gfp_mask, nr_iovecs);
173 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
178 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
180 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
183 if (!mddev || !mddev->bio_set)
184 return bio_clone(bio, gfp_mask);
186 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
188 EXPORT_SYMBOL_GPL(bio_clone_mddev);
191 * We have a system wide 'event count' that is incremented
192 * on any 'interesting' event, and readers of /proc/mdstat
193 * can use 'poll' or 'select' to find out when the event
197 * start array, stop array, error, add device, remove device,
198 * start build, activate spare
200 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
201 static atomic_t md_event_count;
202 void md_new_event(struct mddev *mddev)
204 atomic_inc(&md_event_count);
205 wake_up(&md_event_waiters);
207 EXPORT_SYMBOL_GPL(md_new_event);
210 * Enables to iterate over all existing md arrays
211 * all_mddevs_lock protects this list.
213 static LIST_HEAD(all_mddevs);
214 static DEFINE_SPINLOCK(all_mddevs_lock);
217 * iterates through all used mddevs in the system.
218 * We take care to grab the all_mddevs_lock whenever navigating
219 * the list, and to always hold a refcount when unlocked.
220 * Any code which breaks out of this loop while own
221 * a reference to the current mddev and must mddev_put it.
223 #define for_each_mddev(_mddev,_tmp) \
225 for (({ spin_lock(&all_mddevs_lock); \
226 _tmp = all_mddevs.next; \
228 ({ if (_tmp != &all_mddevs) \
229 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
230 spin_unlock(&all_mddevs_lock); \
231 if (_mddev) mddev_put(_mddev); \
232 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
233 _tmp != &all_mddevs;}); \
234 ({ spin_lock(&all_mddevs_lock); \
235 _tmp = _tmp->next;}) \
238 /* Rather than calling directly into the personality make_request function,
239 * IO requests come here first so that we can check if the device is
240 * being suspended pending a reconfiguration.
241 * We hold a refcount over the call to ->make_request. By the time that
242 * call has finished, the bio has been linked into some internal structure
243 * and so is visible to ->quiesce(), so we don't need the refcount any more.
245 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
247 const int rw = bio_data_dir(bio);
248 struct mddev *mddev = q->queuedata;
249 unsigned int sectors;
252 blk_queue_split(q, &bio, q->bio_split);
254 if (mddev == NULL || mddev->pers == NULL) {
256 return BLK_QC_T_NONE;
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 if (bio_sectors(bio) != 0)
260 bio->bi_error = -EROFS;
262 return BLK_QC_T_NONE;
264 smp_rmb(); /* Ensure implications of 'active' are visible */
266 if (mddev->suspended) {
269 prepare_to_wait(&mddev->sb_wait, &__wait,
270 TASK_UNINTERRUPTIBLE);
271 if (!mddev->suspended)
277 finish_wait(&mddev->sb_wait, &__wait);
279 atomic_inc(&mddev->active_io);
283 * save the sectors now since our bio can
284 * go away inside make_request
286 sectors = bio_sectors(bio);
287 /* bio could be mergeable after passing to underlayer */
288 bio->bi_rw &= ~REQ_NOMERGE;
289 mddev->pers->make_request(mddev, bio);
291 cpu = part_stat_lock();
292 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
293 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
296 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
297 wake_up(&mddev->sb_wait);
299 return BLK_QC_T_NONE;
302 /* mddev_suspend makes sure no new requests are submitted
303 * to the device, and that any requests that have been submitted
304 * are completely handled.
305 * Once mddev_detach() is called and completes, the module will be
308 void mddev_suspend(struct mddev *mddev)
310 WARN_ON_ONCE(mddev->thread && current == mddev->thread->tsk);
311 if (mddev->suspended++)
314 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
315 mddev->pers->quiesce(mddev, 1);
317 del_timer_sync(&mddev->safemode_timer);
319 EXPORT_SYMBOL_GPL(mddev_suspend);
321 void mddev_resume(struct mddev *mddev)
323 if (--mddev->suspended)
325 wake_up(&mddev->sb_wait);
326 mddev->pers->quiesce(mddev, 0);
328 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
329 md_wakeup_thread(mddev->thread);
330 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
332 EXPORT_SYMBOL_GPL(mddev_resume);
334 int mddev_congested(struct mddev *mddev, int bits)
336 struct md_personality *pers = mddev->pers;
340 if (mddev->suspended)
342 else if (pers && pers->congested)
343 ret = pers->congested(mddev, bits);
347 EXPORT_SYMBOL_GPL(mddev_congested);
348 static int md_congested(void *data, int bits)
350 struct mddev *mddev = data;
351 return mddev_congested(mddev, bits);
355 * Generic flush handling for md
358 static void md_end_flush(struct bio *bio)
360 struct md_rdev *rdev = bio->bi_private;
361 struct mddev *mddev = rdev->mddev;
363 rdev_dec_pending(rdev, mddev);
365 if (atomic_dec_and_test(&mddev->flush_pending)) {
366 /* The pre-request flush has finished */
367 queue_work(md_wq, &mddev->flush_work);
372 static void md_submit_flush_data(struct work_struct *ws);
374 static void submit_flushes(struct work_struct *ws)
376 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
377 struct md_rdev *rdev;
379 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
380 atomic_set(&mddev->flush_pending, 1);
382 rdev_for_each_rcu(rdev, mddev)
383 if (rdev->raid_disk >= 0 &&
384 !test_bit(Faulty, &rdev->flags)) {
385 /* Take two references, one is dropped
386 * when request finishes, one after
387 * we reclaim rcu_read_lock
390 atomic_inc(&rdev->nr_pending);
391 atomic_inc(&rdev->nr_pending);
393 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
394 bi->bi_end_io = md_end_flush;
395 bi->bi_private = rdev;
396 bi->bi_bdev = rdev->bdev;
397 bio_set_op_attrs(bi, REQ_OP_WRITE, WRITE_FLUSH);
398 atomic_inc(&mddev->flush_pending);
401 rdev_dec_pending(rdev, mddev);
404 if (atomic_dec_and_test(&mddev->flush_pending))
405 queue_work(md_wq, &mddev->flush_work);
408 static void md_submit_flush_data(struct work_struct *ws)
410 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
411 struct bio *bio = mddev->flush_bio;
413 if (bio->bi_iter.bi_size == 0)
414 /* an empty barrier - all done */
417 bio->bi_rw &= ~REQ_PREFLUSH;
418 mddev->pers->make_request(mddev, bio);
421 mddev->flush_bio = NULL;
422 wake_up(&mddev->sb_wait);
425 void md_flush_request(struct mddev *mddev, struct bio *bio)
427 spin_lock_irq(&mddev->lock);
428 wait_event_lock_irq(mddev->sb_wait,
431 mddev->flush_bio = bio;
432 spin_unlock_irq(&mddev->lock);
434 INIT_WORK(&mddev->flush_work, submit_flushes);
435 queue_work(md_wq, &mddev->flush_work);
437 EXPORT_SYMBOL(md_flush_request);
439 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
441 struct mddev *mddev = cb->data;
442 md_wakeup_thread(mddev->thread);
445 EXPORT_SYMBOL(md_unplug);
447 static inline struct mddev *mddev_get(struct mddev *mddev)
449 atomic_inc(&mddev->active);
453 static void mddev_delayed_delete(struct work_struct *ws);
455 static void mddev_put(struct mddev *mddev)
457 struct bio_set *bs = NULL;
459 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
461 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
462 mddev->ctime == 0 && !mddev->hold_active) {
463 /* Array is not configured at all, and not held active,
465 list_del_init(&mddev->all_mddevs);
467 mddev->bio_set = NULL;
468 if (mddev->gendisk) {
469 /* We did a probe so need to clean up. Call
470 * queue_work inside the spinlock so that
471 * flush_workqueue() after mddev_find will
472 * succeed in waiting for the work to be done.
474 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
475 queue_work(md_misc_wq, &mddev->del_work);
479 spin_unlock(&all_mddevs_lock);
484 static void md_safemode_timeout(unsigned long data);
486 void mddev_init(struct mddev *mddev)
488 mutex_init(&mddev->open_mutex);
489 mutex_init(&mddev->reconfig_mutex);
490 mutex_init(&mddev->bitmap_info.mutex);
491 INIT_LIST_HEAD(&mddev->disks);
492 INIT_LIST_HEAD(&mddev->all_mddevs);
493 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
494 (unsigned long) mddev);
495 atomic_set(&mddev->active, 1);
496 atomic_set(&mddev->openers, 0);
497 atomic_set(&mddev->active_io, 0);
498 spin_lock_init(&mddev->lock);
499 atomic_set(&mddev->flush_pending, 0);
500 init_waitqueue_head(&mddev->sb_wait);
501 init_waitqueue_head(&mddev->recovery_wait);
502 mddev->reshape_position = MaxSector;
503 mddev->reshape_backwards = 0;
504 mddev->last_sync_action = "none";
505 mddev->resync_min = 0;
506 mddev->resync_max = MaxSector;
507 mddev->level = LEVEL_NONE;
509 EXPORT_SYMBOL_GPL(mddev_init);
511 static struct mddev *mddev_find(dev_t unit)
513 struct mddev *mddev, *new = NULL;
515 if (unit && MAJOR(unit) != MD_MAJOR)
516 unit &= ~((1<<MdpMinorShift)-1);
519 spin_lock(&all_mddevs_lock);
522 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
523 if (mddev->unit == unit) {
525 spin_unlock(&all_mddevs_lock);
531 list_add(&new->all_mddevs, &all_mddevs);
532 spin_unlock(&all_mddevs_lock);
533 new->hold_active = UNTIL_IOCTL;
537 /* find an unused unit number */
538 static int next_minor = 512;
539 int start = next_minor;
543 dev = MKDEV(MD_MAJOR, next_minor);
545 if (next_minor > MINORMASK)
547 if (next_minor == start) {
548 /* Oh dear, all in use. */
549 spin_unlock(&all_mddevs_lock);
555 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
556 if (mddev->unit == dev) {
562 new->md_minor = MINOR(dev);
563 new->hold_active = UNTIL_STOP;
564 list_add(&new->all_mddevs, &all_mddevs);
565 spin_unlock(&all_mddevs_lock);
568 spin_unlock(&all_mddevs_lock);
570 new = kzalloc(sizeof(*new), GFP_KERNEL);
575 if (MAJOR(unit) == MD_MAJOR)
576 new->md_minor = MINOR(unit);
578 new->md_minor = MINOR(unit) >> MdpMinorShift;
585 static struct attribute_group md_redundancy_group;
587 void mddev_unlock(struct mddev *mddev)
589 if (mddev->to_remove) {
590 /* These cannot be removed under reconfig_mutex as
591 * an access to the files will try to take reconfig_mutex
592 * while holding the file unremovable, which leads to
594 * So hold set sysfs_active while the remove in happeing,
595 * and anything else which might set ->to_remove or my
596 * otherwise change the sysfs namespace will fail with
597 * -EBUSY if sysfs_active is still set.
598 * We set sysfs_active under reconfig_mutex and elsewhere
599 * test it under the same mutex to ensure its correct value
602 struct attribute_group *to_remove = mddev->to_remove;
603 mddev->to_remove = NULL;
604 mddev->sysfs_active = 1;
605 mutex_unlock(&mddev->reconfig_mutex);
607 if (mddev->kobj.sd) {
608 if (to_remove != &md_redundancy_group)
609 sysfs_remove_group(&mddev->kobj, to_remove);
610 if (mddev->pers == NULL ||
611 mddev->pers->sync_request == NULL) {
612 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
613 if (mddev->sysfs_action)
614 sysfs_put(mddev->sysfs_action);
615 mddev->sysfs_action = NULL;
618 mddev->sysfs_active = 0;
620 mutex_unlock(&mddev->reconfig_mutex);
622 /* As we've dropped the mutex we need a spinlock to
623 * make sure the thread doesn't disappear
625 spin_lock(&pers_lock);
626 md_wakeup_thread(mddev->thread);
627 spin_unlock(&pers_lock);
629 EXPORT_SYMBOL_GPL(mddev_unlock);
631 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
633 struct md_rdev *rdev;
635 rdev_for_each_rcu(rdev, mddev)
636 if (rdev->desc_nr == nr)
641 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
643 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
645 struct md_rdev *rdev;
647 rdev_for_each(rdev, mddev)
648 if (rdev->bdev->bd_dev == dev)
654 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
656 struct md_rdev *rdev;
658 rdev_for_each_rcu(rdev, mddev)
659 if (rdev->bdev->bd_dev == dev)
665 static struct md_personality *find_pers(int level, char *clevel)
667 struct md_personality *pers;
668 list_for_each_entry(pers, &pers_list, list) {
669 if (level != LEVEL_NONE && pers->level == level)
671 if (strcmp(pers->name, clevel)==0)
677 /* return the offset of the super block in 512byte sectors */
678 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
680 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
681 return MD_NEW_SIZE_SECTORS(num_sectors);
684 static int alloc_disk_sb(struct md_rdev *rdev)
686 rdev->sb_page = alloc_page(GFP_KERNEL);
687 if (!rdev->sb_page) {
688 printk(KERN_ALERT "md: out of memory.\n");
695 void md_rdev_clear(struct md_rdev *rdev)
698 put_page(rdev->sb_page);
700 rdev->sb_page = NULL;
705 put_page(rdev->bb_page);
706 rdev->bb_page = NULL;
708 badblocks_exit(&rdev->badblocks);
710 EXPORT_SYMBOL_GPL(md_rdev_clear);
712 static void super_written(struct bio *bio)
714 struct md_rdev *rdev = bio->bi_private;
715 struct mddev *mddev = rdev->mddev;
718 printk("md: super_written gets error=%d\n", bio->bi_error);
719 md_error(mddev, rdev);
722 if (atomic_dec_and_test(&mddev->pending_writes))
723 wake_up(&mddev->sb_wait);
724 rdev_dec_pending(rdev, mddev);
728 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
729 sector_t sector, int size, struct page *page)
731 /* write first size bytes of page to sector of rdev
732 * Increment mddev->pending_writes before returning
733 * and decrement it on completion, waking up sb_wait
734 * if zero is reached.
735 * If an error occurred, call md_error
737 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
739 atomic_inc(&rdev->nr_pending);
741 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
742 bio->bi_iter.bi_sector = sector;
743 bio_add_page(bio, page, size, 0);
744 bio->bi_private = rdev;
745 bio->bi_end_io = super_written;
746 bio_set_op_attrs(bio, REQ_OP_WRITE, WRITE_FLUSH_FUA);
748 atomic_inc(&mddev->pending_writes);
752 void md_super_wait(struct mddev *mddev)
754 /* wait for all superblock writes that were scheduled to complete */
755 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
758 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
759 struct page *page, int op, int op_flags, bool metadata_op)
761 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
764 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
765 rdev->meta_bdev : rdev->bdev;
766 bio_set_op_attrs(bio, op, op_flags);
768 bio->bi_iter.bi_sector = sector + rdev->sb_start;
769 else if (rdev->mddev->reshape_position != MaxSector &&
770 (rdev->mddev->reshape_backwards ==
771 (sector >= rdev->mddev->reshape_position)))
772 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
774 bio->bi_iter.bi_sector = sector + rdev->data_offset;
775 bio_add_page(bio, page, size, 0);
777 submit_bio_wait(bio);
779 ret = !bio->bi_error;
783 EXPORT_SYMBOL_GPL(sync_page_io);
785 static int read_disk_sb(struct md_rdev *rdev, int size)
787 char b[BDEVNAME_SIZE];
792 if (!sync_page_io(rdev, 0, size, rdev->sb_page, REQ_OP_READ, 0, true))
798 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
799 bdevname(rdev->bdev,b));
803 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
805 return sb1->set_uuid0 == sb2->set_uuid0 &&
806 sb1->set_uuid1 == sb2->set_uuid1 &&
807 sb1->set_uuid2 == sb2->set_uuid2 &&
808 sb1->set_uuid3 == sb2->set_uuid3;
811 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
814 mdp_super_t *tmp1, *tmp2;
816 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
817 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
819 if (!tmp1 || !tmp2) {
821 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
829 * nr_disks is not constant
834 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
841 static u32 md_csum_fold(u32 csum)
843 csum = (csum & 0xffff) + (csum >> 16);
844 return (csum & 0xffff) + (csum >> 16);
847 static unsigned int calc_sb_csum(mdp_super_t *sb)
850 u32 *sb32 = (u32*)sb;
852 unsigned int disk_csum, csum;
854 disk_csum = sb->sb_csum;
857 for (i = 0; i < MD_SB_BYTES/4 ; i++)
859 csum = (newcsum & 0xffffffff) + (newcsum>>32);
862 /* This used to use csum_partial, which was wrong for several
863 * reasons including that different results are returned on
864 * different architectures. It isn't critical that we get exactly
865 * the same return value as before (we always csum_fold before
866 * testing, and that removes any differences). However as we
867 * know that csum_partial always returned a 16bit value on
868 * alphas, do a fold to maximise conformity to previous behaviour.
870 sb->sb_csum = md_csum_fold(disk_csum);
872 sb->sb_csum = disk_csum;
878 * Handle superblock details.
879 * We want to be able to handle multiple superblock formats
880 * so we have a common interface to them all, and an array of
881 * different handlers.
882 * We rely on user-space to write the initial superblock, and support
883 * reading and updating of superblocks.
884 * Interface methods are:
885 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
886 * loads and validates a superblock on dev.
887 * if refdev != NULL, compare superblocks on both devices
889 * 0 - dev has a superblock that is compatible with refdev
890 * 1 - dev has a superblock that is compatible and newer than refdev
891 * so dev should be used as the refdev in future
892 * -EINVAL superblock incompatible or invalid
893 * -othererror e.g. -EIO
895 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
896 * Verify that dev is acceptable into mddev.
897 * The first time, mddev->raid_disks will be 0, and data from
898 * dev should be merged in. Subsequent calls check that dev
899 * is new enough. Return 0 or -EINVAL
901 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
902 * Update the superblock for rdev with data in mddev
903 * This does not write to disc.
909 struct module *owner;
910 int (*load_super)(struct md_rdev *rdev,
911 struct md_rdev *refdev,
913 int (*validate_super)(struct mddev *mddev,
914 struct md_rdev *rdev);
915 void (*sync_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
918 sector_t num_sectors);
919 int (*allow_new_offset)(struct md_rdev *rdev,
920 unsigned long long new_offset);
924 * Check that the given mddev has no bitmap.
926 * This function is called from the run method of all personalities that do not
927 * support bitmaps. It prints an error message and returns non-zero if mddev
928 * has a bitmap. Otherwise, it returns 0.
931 int md_check_no_bitmap(struct mddev *mddev)
933 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
935 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
936 mdname(mddev), mddev->pers->name);
939 EXPORT_SYMBOL(md_check_no_bitmap);
942 * load_super for 0.90.0
944 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
946 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
951 * Calculate the position of the superblock (512byte sectors),
952 * it's at the end of the disk.
954 * It also happens to be a multiple of 4Kb.
956 rdev->sb_start = calc_dev_sboffset(rdev);
958 ret = read_disk_sb(rdev, MD_SB_BYTES);
963 bdevname(rdev->bdev, b);
964 sb = page_address(rdev->sb_page);
966 if (sb->md_magic != MD_SB_MAGIC) {
967 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
972 if (sb->major_version != 0 ||
973 sb->minor_version < 90 ||
974 sb->minor_version > 91) {
975 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
976 sb->major_version, sb->minor_version,
981 if (sb->raid_disks <= 0)
984 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
985 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
990 rdev->preferred_minor = sb->md_minor;
991 rdev->data_offset = 0;
992 rdev->new_data_offset = 0;
993 rdev->sb_size = MD_SB_BYTES;
994 rdev->badblocks.shift = -1;
996 if (sb->level == LEVEL_MULTIPATH)
999 rdev->desc_nr = sb->this_disk.number;
1005 mdp_super_t *refsb = page_address(refdev->sb_page);
1006 if (!uuid_equal(refsb, sb)) {
1007 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1008 b, bdevname(refdev->bdev,b2));
1011 if (!sb_equal(refsb, sb)) {
1012 printk(KERN_WARNING "md: %s has same UUID"
1013 " but different superblock to %s\n",
1014 b, bdevname(refdev->bdev, b2));
1018 ev2 = md_event(refsb);
1024 rdev->sectors = rdev->sb_start;
1025 /* Limit to 4TB as metadata cannot record more than that.
1026 * (not needed for Linear and RAID0 as metadata doesn't
1029 if (IS_ENABLED(CONFIG_LBDAF) && (u64)rdev->sectors >= (2ULL << 32) &&
1031 rdev->sectors = (sector_t)(2ULL << 32) - 2;
1033 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1034 /* "this cannot possibly happen" ... */
1042 * validate_super for 0.90.0
1044 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1047 mdp_super_t *sb = page_address(rdev->sb_page);
1048 __u64 ev1 = md_event(sb);
1050 rdev->raid_disk = -1;
1051 clear_bit(Faulty, &rdev->flags);
1052 clear_bit(In_sync, &rdev->flags);
1053 clear_bit(Bitmap_sync, &rdev->flags);
1054 clear_bit(WriteMostly, &rdev->flags);
1056 if (mddev->raid_disks == 0) {
1057 mddev->major_version = 0;
1058 mddev->minor_version = sb->minor_version;
1059 mddev->patch_version = sb->patch_version;
1060 mddev->external = 0;
1061 mddev->chunk_sectors = sb->chunk_size >> 9;
1062 mddev->ctime = sb->ctime;
1063 mddev->utime = sb->utime;
1064 mddev->level = sb->level;
1065 mddev->clevel[0] = 0;
1066 mddev->layout = sb->layout;
1067 mddev->raid_disks = sb->raid_disks;
1068 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1069 mddev->events = ev1;
1070 mddev->bitmap_info.offset = 0;
1071 mddev->bitmap_info.space = 0;
1072 /* bitmap can use 60 K after the 4K superblocks */
1073 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1074 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1075 mddev->reshape_backwards = 0;
1077 if (mddev->minor_version >= 91) {
1078 mddev->reshape_position = sb->reshape_position;
1079 mddev->delta_disks = sb->delta_disks;
1080 mddev->new_level = sb->new_level;
1081 mddev->new_layout = sb->new_layout;
1082 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1083 if (mddev->delta_disks < 0)
1084 mddev->reshape_backwards = 1;
1086 mddev->reshape_position = MaxSector;
1087 mddev->delta_disks = 0;
1088 mddev->new_level = mddev->level;
1089 mddev->new_layout = mddev->layout;
1090 mddev->new_chunk_sectors = mddev->chunk_sectors;
1093 if (sb->state & (1<<MD_SB_CLEAN))
1094 mddev->recovery_cp = MaxSector;
1096 if (sb->events_hi == sb->cp_events_hi &&
1097 sb->events_lo == sb->cp_events_lo) {
1098 mddev->recovery_cp = sb->recovery_cp;
1100 mddev->recovery_cp = 0;
1103 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1104 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1105 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1106 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1108 mddev->max_disks = MD_SB_DISKS;
1110 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1111 mddev->bitmap_info.file == NULL) {
1112 mddev->bitmap_info.offset =
1113 mddev->bitmap_info.default_offset;
1114 mddev->bitmap_info.space =
1115 mddev->bitmap_info.default_space;
1118 } else if (mddev->pers == NULL) {
1119 /* Insist on good event counter while assembling, except
1120 * for spares (which don't need an event count) */
1122 if (sb->disks[rdev->desc_nr].state & (
1123 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1124 if (ev1 < mddev->events)
1126 } else if (mddev->bitmap) {
1127 /* if adding to array with a bitmap, then we can accept an
1128 * older device ... but not too old.
1130 if (ev1 < mddev->bitmap->events_cleared)
1132 if (ev1 < mddev->events)
1133 set_bit(Bitmap_sync, &rdev->flags);
1135 if (ev1 < mddev->events)
1136 /* just a hot-add of a new device, leave raid_disk at -1 */
1140 if (mddev->level != LEVEL_MULTIPATH) {
1141 desc = sb->disks + rdev->desc_nr;
1143 if (desc->state & (1<<MD_DISK_FAULTY))
1144 set_bit(Faulty, &rdev->flags);
1145 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1146 desc->raid_disk < mddev->raid_disks */) {
1147 set_bit(In_sync, &rdev->flags);
1148 rdev->raid_disk = desc->raid_disk;
1149 rdev->saved_raid_disk = desc->raid_disk;
1150 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1151 /* active but not in sync implies recovery up to
1152 * reshape position. We don't know exactly where
1153 * that is, so set to zero for now */
1154 if (mddev->minor_version >= 91) {
1155 rdev->recovery_offset = 0;
1156 rdev->raid_disk = desc->raid_disk;
1159 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1160 set_bit(WriteMostly, &rdev->flags);
1161 } else /* MULTIPATH are always insync */
1162 set_bit(In_sync, &rdev->flags);
1167 * sync_super for 0.90.0
1169 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1172 struct md_rdev *rdev2;
1173 int next_spare = mddev->raid_disks;
1175 /* make rdev->sb match mddev data..
1178 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1179 * 3/ any empty disks < next_spare become removed
1181 * disks[0] gets initialised to REMOVED because
1182 * we cannot be sure from other fields if it has
1183 * been initialised or not.
1186 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1188 rdev->sb_size = MD_SB_BYTES;
1190 sb = page_address(rdev->sb_page);
1192 memset(sb, 0, sizeof(*sb));
1194 sb->md_magic = MD_SB_MAGIC;
1195 sb->major_version = mddev->major_version;
1196 sb->patch_version = mddev->patch_version;
1197 sb->gvalid_words = 0; /* ignored */
1198 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1199 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1200 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1201 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1203 sb->ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
1204 sb->level = mddev->level;
1205 sb->size = mddev->dev_sectors / 2;
1206 sb->raid_disks = mddev->raid_disks;
1207 sb->md_minor = mddev->md_minor;
1208 sb->not_persistent = 0;
1209 sb->utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
1211 sb->events_hi = (mddev->events>>32);
1212 sb->events_lo = (u32)mddev->events;
1214 if (mddev->reshape_position == MaxSector)
1215 sb->minor_version = 90;
1217 sb->minor_version = 91;
1218 sb->reshape_position = mddev->reshape_position;
1219 sb->new_level = mddev->new_level;
1220 sb->delta_disks = mddev->delta_disks;
1221 sb->new_layout = mddev->new_layout;
1222 sb->new_chunk = mddev->new_chunk_sectors << 9;
1224 mddev->minor_version = sb->minor_version;
1227 sb->recovery_cp = mddev->recovery_cp;
1228 sb->cp_events_hi = (mddev->events>>32);
1229 sb->cp_events_lo = (u32)mddev->events;
1230 if (mddev->recovery_cp == MaxSector)
1231 sb->state = (1<< MD_SB_CLEAN);
1233 sb->recovery_cp = 0;
1235 sb->layout = mddev->layout;
1236 sb->chunk_size = mddev->chunk_sectors << 9;
1238 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1239 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1241 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1242 rdev_for_each(rdev2, mddev) {
1245 int is_active = test_bit(In_sync, &rdev2->flags);
1247 if (rdev2->raid_disk >= 0 &&
1248 sb->minor_version >= 91)
1249 /* we have nowhere to store the recovery_offset,
1250 * but if it is not below the reshape_position,
1251 * we can piggy-back on that.
1254 if (rdev2->raid_disk < 0 ||
1255 test_bit(Faulty, &rdev2->flags))
1258 desc_nr = rdev2->raid_disk;
1260 desc_nr = next_spare++;
1261 rdev2->desc_nr = desc_nr;
1262 d = &sb->disks[rdev2->desc_nr];
1264 d->number = rdev2->desc_nr;
1265 d->major = MAJOR(rdev2->bdev->bd_dev);
1266 d->minor = MINOR(rdev2->bdev->bd_dev);
1268 d->raid_disk = rdev2->raid_disk;
1270 d->raid_disk = rdev2->desc_nr; /* compatibility */
1271 if (test_bit(Faulty, &rdev2->flags))
1272 d->state = (1<<MD_DISK_FAULTY);
1273 else if (is_active) {
1274 d->state = (1<<MD_DISK_ACTIVE);
1275 if (test_bit(In_sync, &rdev2->flags))
1276 d->state |= (1<<MD_DISK_SYNC);
1284 if (test_bit(WriteMostly, &rdev2->flags))
1285 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1287 /* now set the "removed" and "faulty" bits on any missing devices */
1288 for (i=0 ; i < mddev->raid_disks ; i++) {
1289 mdp_disk_t *d = &sb->disks[i];
1290 if (d->state == 0 && d->number == 0) {
1293 d->state = (1<<MD_DISK_REMOVED);
1294 d->state |= (1<<MD_DISK_FAULTY);
1298 sb->nr_disks = nr_disks;
1299 sb->active_disks = active;
1300 sb->working_disks = working;
1301 sb->failed_disks = failed;
1302 sb->spare_disks = spare;
1304 sb->this_disk = sb->disks[rdev->desc_nr];
1305 sb->sb_csum = calc_sb_csum(sb);
1309 * rdev_size_change for 0.90.0
1311 static unsigned long long
1312 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1314 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1315 return 0; /* component must fit device */
1316 if (rdev->mddev->bitmap_info.offset)
1317 return 0; /* can't move bitmap */
1318 rdev->sb_start = calc_dev_sboffset(rdev);
1319 if (!num_sectors || num_sectors > rdev->sb_start)
1320 num_sectors = rdev->sb_start;
1321 /* Limit to 4TB as metadata cannot record more than that.
1322 * 4TB == 2^32 KB, or 2*2^32 sectors.
1324 if (IS_ENABLED(CONFIG_LBDAF) && (u64)num_sectors >= (2ULL << 32) &&
1325 rdev->mddev->level >= 1)
1326 num_sectors = (sector_t)(2ULL << 32) - 2;
1327 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1329 md_super_wait(rdev->mddev);
1334 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1336 /* non-zero offset changes not possible with v0.90 */
1337 return new_offset == 0;
1341 * version 1 superblock
1344 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1348 unsigned long long newcsum;
1349 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1350 __le32 *isuper = (__le32*)sb;
1352 disk_csum = sb->sb_csum;
1355 for (; size >= 4; size -= 4)
1356 newcsum += le32_to_cpu(*isuper++);
1359 newcsum += le16_to_cpu(*(__le16*) isuper);
1361 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1362 sb->sb_csum = disk_csum;
1363 return cpu_to_le32(csum);
1366 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1368 struct mdp_superblock_1 *sb;
1372 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1376 * Calculate the position of the superblock in 512byte sectors.
1377 * It is always aligned to a 4K boundary and
1378 * depeding on minor_version, it can be:
1379 * 0: At least 8K, but less than 12K, from end of device
1380 * 1: At start of device
1381 * 2: 4K from start of device.
1383 switch(minor_version) {
1385 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1387 sb_start &= ~(sector_t)(4*2-1);
1398 rdev->sb_start = sb_start;
1400 /* superblock is rarely larger than 1K, but it can be larger,
1401 * and it is safe to read 4k, so we do that
1403 ret = read_disk_sb(rdev, 4096);
1404 if (ret) return ret;
1406 sb = page_address(rdev->sb_page);
1408 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1409 sb->major_version != cpu_to_le32(1) ||
1410 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1411 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1412 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1415 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1416 printk("md: invalid superblock checksum on %s\n",
1417 bdevname(rdev->bdev,b));
1420 if (le64_to_cpu(sb->data_size) < 10) {
1421 printk("md: data_size too small on %s\n",
1422 bdevname(rdev->bdev,b));
1427 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1428 /* Some padding is non-zero, might be a new feature */
1431 rdev->preferred_minor = 0xffff;
1432 rdev->data_offset = le64_to_cpu(sb->data_offset);
1433 rdev->new_data_offset = rdev->data_offset;
1434 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1435 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1436 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1437 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1439 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1440 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1441 if (rdev->sb_size & bmask)
1442 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1445 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1448 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1451 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1454 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1456 if (!rdev->bb_page) {
1457 rdev->bb_page = alloc_page(GFP_KERNEL);
1461 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1462 rdev->badblocks.count == 0) {
1463 /* need to load the bad block list.
1464 * Currently we limit it to one page.
1470 int sectors = le16_to_cpu(sb->bblog_size);
1471 if (sectors > (PAGE_SIZE / 512))
1473 offset = le32_to_cpu(sb->bblog_offset);
1476 bb_sector = (long long)offset;
1477 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1478 rdev->bb_page, REQ_OP_READ, 0, true))
1480 bbp = (u64 *)page_address(rdev->bb_page);
1481 rdev->badblocks.shift = sb->bblog_shift;
1482 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1483 u64 bb = le64_to_cpu(*bbp);
1484 int count = bb & (0x3ff);
1485 u64 sector = bb >> 10;
1486 sector <<= sb->bblog_shift;
1487 count <<= sb->bblog_shift;
1490 if (badblocks_set(&rdev->badblocks, sector, count, 1))
1493 } else if (sb->bblog_offset != 0)
1494 rdev->badblocks.shift = 0;
1500 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1502 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1503 sb->level != refsb->level ||
1504 sb->layout != refsb->layout ||
1505 sb->chunksize != refsb->chunksize) {
1506 printk(KERN_WARNING "md: %s has strangely different"
1507 " superblock to %s\n",
1508 bdevname(rdev->bdev,b),
1509 bdevname(refdev->bdev,b2));
1512 ev1 = le64_to_cpu(sb->events);
1513 ev2 = le64_to_cpu(refsb->events);
1520 if (minor_version) {
1521 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1522 sectors -= rdev->data_offset;
1524 sectors = rdev->sb_start;
1525 if (sectors < le64_to_cpu(sb->data_size))
1527 rdev->sectors = le64_to_cpu(sb->data_size);
1531 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1533 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1534 __u64 ev1 = le64_to_cpu(sb->events);
1536 rdev->raid_disk = -1;
1537 clear_bit(Faulty, &rdev->flags);
1538 clear_bit(In_sync, &rdev->flags);
1539 clear_bit(Bitmap_sync, &rdev->flags);
1540 clear_bit(WriteMostly, &rdev->flags);
1542 if (mddev->raid_disks == 0) {
1543 mddev->major_version = 1;
1544 mddev->patch_version = 0;
1545 mddev->external = 0;
1546 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1547 mddev->ctime = le64_to_cpu(sb->ctime);
1548 mddev->utime = le64_to_cpu(sb->utime);
1549 mddev->level = le32_to_cpu(sb->level);
1550 mddev->clevel[0] = 0;
1551 mddev->layout = le32_to_cpu(sb->layout);
1552 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1553 mddev->dev_sectors = le64_to_cpu(sb->size);
1554 mddev->events = ev1;
1555 mddev->bitmap_info.offset = 0;
1556 mddev->bitmap_info.space = 0;
1557 /* Default location for bitmap is 1K after superblock
1558 * using 3K - total of 4K
1560 mddev->bitmap_info.default_offset = 1024 >> 9;
1561 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1562 mddev->reshape_backwards = 0;
1564 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1565 memcpy(mddev->uuid, sb->set_uuid, 16);
1567 mddev->max_disks = (4096-256)/2;
1569 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1570 mddev->bitmap_info.file == NULL) {
1571 mddev->bitmap_info.offset =
1572 (__s32)le32_to_cpu(sb->bitmap_offset);
1573 /* Metadata doesn't record how much space is available.
1574 * For 1.0, we assume we can use up to the superblock
1575 * if before, else to 4K beyond superblock.
1576 * For others, assume no change is possible.
1578 if (mddev->minor_version > 0)
1579 mddev->bitmap_info.space = 0;
1580 else if (mddev->bitmap_info.offset > 0)
1581 mddev->bitmap_info.space =
1582 8 - mddev->bitmap_info.offset;
1584 mddev->bitmap_info.space =
1585 -mddev->bitmap_info.offset;
1588 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1589 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1590 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1591 mddev->new_level = le32_to_cpu(sb->new_level);
1592 mddev->new_layout = le32_to_cpu(sb->new_layout);
1593 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1594 if (mddev->delta_disks < 0 ||
1595 (mddev->delta_disks == 0 &&
1596 (le32_to_cpu(sb->feature_map)
1597 & MD_FEATURE_RESHAPE_BACKWARDS)))
1598 mddev->reshape_backwards = 1;
1600 mddev->reshape_position = MaxSector;
1601 mddev->delta_disks = 0;
1602 mddev->new_level = mddev->level;
1603 mddev->new_layout = mddev->layout;
1604 mddev->new_chunk_sectors = mddev->chunk_sectors;
1607 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL) {
1608 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1609 if (mddev->recovery_cp == MaxSector)
1610 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1612 } else if (mddev->pers == NULL) {
1613 /* Insist of good event counter while assembling, except for
1614 * spares (which don't need an event count) */
1616 if (rdev->desc_nr >= 0 &&
1617 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1618 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1619 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1620 if (ev1 < mddev->events)
1622 } else if (mddev->bitmap) {
1623 /* If adding to array with a bitmap, then we can accept an
1624 * older device, but not too old.
1626 if (ev1 < mddev->bitmap->events_cleared)
1628 if (ev1 < mddev->events)
1629 set_bit(Bitmap_sync, &rdev->flags);
1631 if (ev1 < mddev->events)
1632 /* just a hot-add of a new device, leave raid_disk at -1 */
1635 if (mddev->level != LEVEL_MULTIPATH) {
1637 if (rdev->desc_nr < 0 ||
1638 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1639 role = MD_DISK_ROLE_SPARE;
1642 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1644 case MD_DISK_ROLE_SPARE: /* spare */
1646 case MD_DISK_ROLE_FAULTY: /* faulty */
1647 set_bit(Faulty, &rdev->flags);
1649 case MD_DISK_ROLE_JOURNAL: /* journal device */
1650 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1651 /* journal device without journal feature */
1653 "md: journal device provided without journal feature, ignoring the device\n");
1656 set_bit(Journal, &rdev->flags);
1657 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1658 rdev->raid_disk = 0;
1661 rdev->saved_raid_disk = role;
1662 if ((le32_to_cpu(sb->feature_map) &
1663 MD_FEATURE_RECOVERY_OFFSET)) {
1664 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1665 if (!(le32_to_cpu(sb->feature_map) &
1666 MD_FEATURE_RECOVERY_BITMAP))
1667 rdev->saved_raid_disk = -1;
1669 set_bit(In_sync, &rdev->flags);
1670 rdev->raid_disk = role;
1673 if (sb->devflags & WriteMostly1)
1674 set_bit(WriteMostly, &rdev->flags);
1675 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1676 set_bit(Replacement, &rdev->flags);
1677 } else /* MULTIPATH are always insync */
1678 set_bit(In_sync, &rdev->flags);
1683 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1685 struct mdp_superblock_1 *sb;
1686 struct md_rdev *rdev2;
1688 /* make rdev->sb match mddev and rdev data. */
1690 sb = page_address(rdev->sb_page);
1692 sb->feature_map = 0;
1694 sb->recovery_offset = cpu_to_le64(0);
1695 memset(sb->pad3, 0, sizeof(sb->pad3));
1697 sb->utime = cpu_to_le64((__u64)mddev->utime);
1698 sb->events = cpu_to_le64(mddev->events);
1700 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1701 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1702 sb->resync_offset = cpu_to_le64(MaxSector);
1704 sb->resync_offset = cpu_to_le64(0);
1706 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1708 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1709 sb->size = cpu_to_le64(mddev->dev_sectors);
1710 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1711 sb->level = cpu_to_le32(mddev->level);
1712 sb->layout = cpu_to_le32(mddev->layout);
1714 if (test_bit(WriteMostly, &rdev->flags))
1715 sb->devflags |= WriteMostly1;
1717 sb->devflags &= ~WriteMostly1;
1718 sb->data_offset = cpu_to_le64(rdev->data_offset);
1719 sb->data_size = cpu_to_le64(rdev->sectors);
1721 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1722 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1723 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1726 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1727 !test_bit(In_sync, &rdev->flags)) {
1729 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1730 sb->recovery_offset =
1731 cpu_to_le64(rdev->recovery_offset);
1732 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1734 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1736 /* Note: recovery_offset and journal_tail share space */
1737 if (test_bit(Journal, &rdev->flags))
1738 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1739 if (test_bit(Replacement, &rdev->flags))
1741 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1743 if (mddev->reshape_position != MaxSector) {
1744 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1745 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1746 sb->new_layout = cpu_to_le32(mddev->new_layout);
1747 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1748 sb->new_level = cpu_to_le32(mddev->new_level);
1749 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1750 if (mddev->delta_disks == 0 &&
1751 mddev->reshape_backwards)
1753 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1754 if (rdev->new_data_offset != rdev->data_offset) {
1756 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1757 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1758 - rdev->data_offset));
1762 if (mddev_is_clustered(mddev))
1763 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1765 if (rdev->badblocks.count == 0)
1766 /* Nothing to do for bad blocks*/ ;
1767 else if (sb->bblog_offset == 0)
1768 /* Cannot record bad blocks on this device */
1769 md_error(mddev, rdev);
1771 struct badblocks *bb = &rdev->badblocks;
1772 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1774 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1779 seq = read_seqbegin(&bb->lock);
1781 memset(bbp, 0xff, PAGE_SIZE);
1783 for (i = 0 ; i < bb->count ; i++) {
1784 u64 internal_bb = p[i];
1785 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1786 | BB_LEN(internal_bb));
1787 bbp[i] = cpu_to_le64(store_bb);
1790 if (read_seqretry(&bb->lock, seq))
1793 bb->sector = (rdev->sb_start +
1794 (int)le32_to_cpu(sb->bblog_offset));
1795 bb->size = le16_to_cpu(sb->bblog_size);
1800 rdev_for_each(rdev2, mddev)
1801 if (rdev2->desc_nr+1 > max_dev)
1802 max_dev = rdev2->desc_nr+1;
1804 if (max_dev > le32_to_cpu(sb->max_dev)) {
1806 sb->max_dev = cpu_to_le32(max_dev);
1807 rdev->sb_size = max_dev * 2 + 256;
1808 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1809 if (rdev->sb_size & bmask)
1810 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1812 max_dev = le32_to_cpu(sb->max_dev);
1814 for (i=0; i<max_dev;i++)
1815 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1817 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1818 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1820 rdev_for_each(rdev2, mddev) {
1822 if (test_bit(Faulty, &rdev2->flags))
1823 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1824 else if (test_bit(In_sync, &rdev2->flags))
1825 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1826 else if (test_bit(Journal, &rdev2->flags))
1827 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1828 else if (rdev2->raid_disk >= 0)
1829 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1831 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1834 sb->sb_csum = calc_sb_1_csum(sb);
1837 static unsigned long long
1838 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1840 struct mdp_superblock_1 *sb;
1841 sector_t max_sectors;
1842 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1843 return 0; /* component must fit device */
1844 if (rdev->data_offset != rdev->new_data_offset)
1845 return 0; /* too confusing */
1846 if (rdev->sb_start < rdev->data_offset) {
1847 /* minor versions 1 and 2; superblock before data */
1848 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1849 max_sectors -= rdev->data_offset;
1850 if (!num_sectors || num_sectors > max_sectors)
1851 num_sectors = max_sectors;
1852 } else if (rdev->mddev->bitmap_info.offset) {
1853 /* minor version 0 with bitmap we can't move */
1856 /* minor version 0; superblock after data */
1858 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1859 sb_start &= ~(sector_t)(4*2 - 1);
1860 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1861 if (!num_sectors || num_sectors > max_sectors)
1862 num_sectors = max_sectors;
1863 rdev->sb_start = sb_start;
1865 sb = page_address(rdev->sb_page);
1866 sb->data_size = cpu_to_le64(num_sectors);
1867 sb->super_offset = rdev->sb_start;
1868 sb->sb_csum = calc_sb_1_csum(sb);
1869 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1871 md_super_wait(rdev->mddev);
1877 super_1_allow_new_offset(struct md_rdev *rdev,
1878 unsigned long long new_offset)
1880 /* All necessary checks on new >= old have been done */
1881 struct bitmap *bitmap;
1882 if (new_offset >= rdev->data_offset)
1885 /* with 1.0 metadata, there is no metadata to tread on
1886 * so we can always move back */
1887 if (rdev->mddev->minor_version == 0)
1890 /* otherwise we must be sure not to step on
1891 * any metadata, so stay:
1892 * 36K beyond start of superblock
1893 * beyond end of badblocks
1894 * beyond write-intent bitmap
1896 if (rdev->sb_start + (32+4)*2 > new_offset)
1898 bitmap = rdev->mddev->bitmap;
1899 if (bitmap && !rdev->mddev->bitmap_info.file &&
1900 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1901 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1903 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1909 static struct super_type super_types[] = {
1912 .owner = THIS_MODULE,
1913 .load_super = super_90_load,
1914 .validate_super = super_90_validate,
1915 .sync_super = super_90_sync,
1916 .rdev_size_change = super_90_rdev_size_change,
1917 .allow_new_offset = super_90_allow_new_offset,
1921 .owner = THIS_MODULE,
1922 .load_super = super_1_load,
1923 .validate_super = super_1_validate,
1924 .sync_super = super_1_sync,
1925 .rdev_size_change = super_1_rdev_size_change,
1926 .allow_new_offset = super_1_allow_new_offset,
1930 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1932 if (mddev->sync_super) {
1933 mddev->sync_super(mddev, rdev);
1937 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1939 super_types[mddev->major_version].sync_super(mddev, rdev);
1942 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1944 struct md_rdev *rdev, *rdev2;
1947 rdev_for_each_rcu(rdev, mddev1) {
1948 if (test_bit(Faulty, &rdev->flags) ||
1949 test_bit(Journal, &rdev->flags) ||
1950 rdev->raid_disk == -1)
1952 rdev_for_each_rcu(rdev2, mddev2) {
1953 if (test_bit(Faulty, &rdev2->flags) ||
1954 test_bit(Journal, &rdev2->flags) ||
1955 rdev2->raid_disk == -1)
1957 if (rdev->bdev->bd_contains ==
1958 rdev2->bdev->bd_contains) {
1968 static LIST_HEAD(pending_raid_disks);
1971 * Try to register data integrity profile for an mddev
1973 * This is called when an array is started and after a disk has been kicked
1974 * from the array. It only succeeds if all working and active component devices
1975 * are integrity capable with matching profiles.
1977 int md_integrity_register(struct mddev *mddev)
1979 struct md_rdev *rdev, *reference = NULL;
1981 if (list_empty(&mddev->disks))
1982 return 0; /* nothing to do */
1983 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1984 return 0; /* shouldn't register, or already is */
1985 rdev_for_each(rdev, mddev) {
1986 /* skip spares and non-functional disks */
1987 if (test_bit(Faulty, &rdev->flags))
1989 if (rdev->raid_disk < 0)
1992 /* Use the first rdev as the reference */
1996 /* does this rdev's profile match the reference profile? */
1997 if (blk_integrity_compare(reference->bdev->bd_disk,
1998 rdev->bdev->bd_disk) < 0)
2001 if (!reference || !bdev_get_integrity(reference->bdev))
2004 * All component devices are integrity capable and have matching
2005 * profiles, register the common profile for the md device.
2007 blk_integrity_register(mddev->gendisk,
2008 bdev_get_integrity(reference->bdev));
2010 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2011 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2012 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2018 EXPORT_SYMBOL(md_integrity_register);
2021 * Attempt to add an rdev, but only if it is consistent with the current
2024 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2026 struct blk_integrity *bi_rdev;
2027 struct blk_integrity *bi_mddev;
2028 char name[BDEVNAME_SIZE];
2030 if (!mddev->gendisk)
2033 bi_rdev = bdev_get_integrity(rdev->bdev);
2034 bi_mddev = blk_get_integrity(mddev->gendisk);
2036 if (!bi_mddev) /* nothing to do */
2039 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2040 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2041 mdname(mddev), bdevname(rdev->bdev, name));
2047 EXPORT_SYMBOL(md_integrity_add_rdev);
2049 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2051 char b[BDEVNAME_SIZE];
2055 /* prevent duplicates */
2056 if (find_rdev(mddev, rdev->bdev->bd_dev))
2059 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2060 if (!test_bit(Journal, &rdev->flags) &&
2062 (mddev->dev_sectors == 0 || rdev->sectors < mddev->dev_sectors)) {
2064 /* Cannot change size, so fail
2065 * If mddev->level <= 0, then we don't care
2066 * about aligning sizes (e.g. linear)
2068 if (mddev->level > 0)
2071 mddev->dev_sectors = rdev->sectors;
2074 /* Verify rdev->desc_nr is unique.
2075 * If it is -1, assign a free number, else
2076 * check number is not in use
2079 if (rdev->desc_nr < 0) {
2082 choice = mddev->raid_disks;
2083 while (md_find_rdev_nr_rcu(mddev, choice))
2085 rdev->desc_nr = choice;
2087 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2093 if (!test_bit(Journal, &rdev->flags) &&
2094 mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2095 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2096 mdname(mddev), mddev->max_disks);
2099 bdevname(rdev->bdev,b);
2100 strreplace(b, '/', '!');
2102 rdev->mddev = mddev;
2103 printk(KERN_INFO "md: bind<%s>\n", b);
2105 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2108 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2109 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2110 /* failure here is OK */;
2111 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2113 list_add_rcu(&rdev->same_set, &mddev->disks);
2114 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2116 /* May as well allow recovery to be retried once */
2117 mddev->recovery_disabled++;
2122 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2127 static void md_delayed_delete(struct work_struct *ws)
2129 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2130 kobject_del(&rdev->kobj);
2131 kobject_put(&rdev->kobj);
2134 static void unbind_rdev_from_array(struct md_rdev *rdev)
2136 char b[BDEVNAME_SIZE];
2138 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2139 list_del_rcu(&rdev->same_set);
2140 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2142 sysfs_remove_link(&rdev->kobj, "block");
2143 sysfs_put(rdev->sysfs_state);
2144 rdev->sysfs_state = NULL;
2145 rdev->badblocks.count = 0;
2146 /* We need to delay this, otherwise we can deadlock when
2147 * writing to 'remove' to "dev/state". We also need
2148 * to delay it due to rcu usage.
2151 INIT_WORK(&rdev->del_work, md_delayed_delete);
2152 kobject_get(&rdev->kobj);
2153 queue_work(md_misc_wq, &rdev->del_work);
2157 * prevent the device from being mounted, repartitioned or
2158 * otherwise reused by a RAID array (or any other kernel
2159 * subsystem), by bd_claiming the device.
2161 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2164 struct block_device *bdev;
2165 char b[BDEVNAME_SIZE];
2167 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2168 shared ? (struct md_rdev *)lock_rdev : rdev);
2170 printk(KERN_ERR "md: could not open %s.\n",
2171 __bdevname(dev, b));
2172 return PTR_ERR(bdev);
2178 static void unlock_rdev(struct md_rdev *rdev)
2180 struct block_device *bdev = rdev->bdev;
2182 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2185 void md_autodetect_dev(dev_t dev);
2187 static void export_rdev(struct md_rdev *rdev)
2189 char b[BDEVNAME_SIZE];
2191 printk(KERN_INFO "md: export_rdev(%s)\n",
2192 bdevname(rdev->bdev,b));
2193 md_rdev_clear(rdev);
2195 if (test_bit(AutoDetected, &rdev->flags))
2196 md_autodetect_dev(rdev->bdev->bd_dev);
2199 kobject_put(&rdev->kobj);
2202 void md_kick_rdev_from_array(struct md_rdev *rdev)
2204 unbind_rdev_from_array(rdev);
2207 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2209 static void export_array(struct mddev *mddev)
2211 struct md_rdev *rdev;
2213 while (!list_empty(&mddev->disks)) {
2214 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2216 md_kick_rdev_from_array(rdev);
2218 mddev->raid_disks = 0;
2219 mddev->major_version = 0;
2222 static void sync_sbs(struct mddev *mddev, int nospares)
2224 /* Update each superblock (in-memory image), but
2225 * if we are allowed to, skip spares which already
2226 * have the right event counter, or have one earlier
2227 * (which would mean they aren't being marked as dirty
2228 * with the rest of the array)
2230 struct md_rdev *rdev;
2231 rdev_for_each(rdev, mddev) {
2232 if (rdev->sb_events == mddev->events ||
2234 rdev->raid_disk < 0 &&
2235 rdev->sb_events+1 == mddev->events)) {
2236 /* Don't update this superblock */
2237 rdev->sb_loaded = 2;
2239 sync_super(mddev, rdev);
2240 rdev->sb_loaded = 1;
2245 static bool does_sb_need_changing(struct mddev *mddev)
2247 struct md_rdev *rdev;
2248 struct mdp_superblock_1 *sb;
2251 /* Find a good rdev */
2252 rdev_for_each(rdev, mddev)
2253 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2256 /* No good device found. */
2260 sb = page_address(rdev->sb_page);
2261 /* Check if a device has become faulty or a spare become active */
2262 rdev_for_each(rdev, mddev) {
2263 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2264 /* Device activated? */
2265 if (role == 0xffff && rdev->raid_disk >=0 &&
2266 !test_bit(Faulty, &rdev->flags))
2268 /* Device turned faulty? */
2269 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2273 /* Check if any mddev parameters have changed */
2274 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2275 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2276 (mddev->layout != le64_to_cpu(sb->layout)) ||
2277 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2278 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2284 void md_update_sb(struct mddev *mddev, int force_change)
2286 struct md_rdev *rdev;
2289 int any_badblocks_changed = 0;
2294 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2299 if (mddev_is_clustered(mddev)) {
2300 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2302 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2304 ret = md_cluster_ops->metadata_update_start(mddev);
2305 /* Has someone else has updated the sb */
2306 if (!does_sb_need_changing(mddev)) {
2308 md_cluster_ops->metadata_update_cancel(mddev);
2309 bit_clear_unless(&mddev->flags, BIT(MD_CHANGE_PENDING),
2310 BIT(MD_CHANGE_DEVS) |
2311 BIT(MD_CHANGE_CLEAN));
2316 /* First make sure individual recovery_offsets are correct */
2317 rdev_for_each(rdev, mddev) {
2318 if (rdev->raid_disk >= 0 &&
2319 mddev->delta_disks >= 0 &&
2320 !test_bit(Journal, &rdev->flags) &&
2321 !test_bit(In_sync, &rdev->flags) &&
2322 mddev->curr_resync_completed > rdev->recovery_offset)
2323 rdev->recovery_offset = mddev->curr_resync_completed;
2326 if (!mddev->persistent) {
2327 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2328 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2329 if (!mddev->external) {
2330 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2331 rdev_for_each(rdev, mddev) {
2332 if (rdev->badblocks.changed) {
2333 rdev->badblocks.changed = 0;
2334 ack_all_badblocks(&rdev->badblocks);
2335 md_error(mddev, rdev);
2337 clear_bit(Blocked, &rdev->flags);
2338 clear_bit(BlockedBadBlocks, &rdev->flags);
2339 wake_up(&rdev->blocked_wait);
2342 wake_up(&mddev->sb_wait);
2346 spin_lock(&mddev->lock);
2348 mddev->utime = ktime_get_real_seconds();
2350 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2352 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2353 /* just a clean<-> dirty transition, possibly leave spares alone,
2354 * though if events isn't the right even/odd, we will have to do
2360 if (mddev->degraded)
2361 /* If the array is degraded, then skipping spares is both
2362 * dangerous and fairly pointless.
2363 * Dangerous because a device that was removed from the array
2364 * might have a event_count that still looks up-to-date,
2365 * so it can be re-added without a resync.
2366 * Pointless because if there are any spares to skip,
2367 * then a recovery will happen and soon that array won't
2368 * be degraded any more and the spare can go back to sleep then.
2372 sync_req = mddev->in_sync;
2374 /* If this is just a dirty<->clean transition, and the array is clean
2375 * and 'events' is odd, we can roll back to the previous clean state */
2377 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2378 && mddev->can_decrease_events
2379 && mddev->events != 1) {
2381 mddev->can_decrease_events = 0;
2383 /* otherwise we have to go forward and ... */
2385 mddev->can_decrease_events = nospares;
2389 * This 64-bit counter should never wrap.
2390 * Either we are in around ~1 trillion A.C., assuming
2391 * 1 reboot per second, or we have a bug...
2393 WARN_ON(mddev->events == 0);
2395 rdev_for_each(rdev, mddev) {
2396 if (rdev->badblocks.changed)
2397 any_badblocks_changed++;
2398 if (test_bit(Faulty, &rdev->flags))
2399 set_bit(FaultRecorded, &rdev->flags);
2402 sync_sbs(mddev, nospares);
2403 spin_unlock(&mddev->lock);
2405 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2406 mdname(mddev), mddev->in_sync);
2408 bitmap_update_sb(mddev->bitmap);
2409 rdev_for_each(rdev, mddev) {
2410 char b[BDEVNAME_SIZE];
2412 if (rdev->sb_loaded != 1)
2413 continue; /* no noise on spare devices */
2415 if (!test_bit(Faulty, &rdev->flags)) {
2416 md_super_write(mddev,rdev,
2417 rdev->sb_start, rdev->sb_size,
2419 pr_debug("md: (write) %s's sb offset: %llu\n",
2420 bdevname(rdev->bdev, b),
2421 (unsigned long long)rdev->sb_start);
2422 rdev->sb_events = mddev->events;
2423 if (rdev->badblocks.size) {
2424 md_super_write(mddev, rdev,
2425 rdev->badblocks.sector,
2426 rdev->badblocks.size << 9,
2428 rdev->badblocks.size = 0;
2432 pr_debug("md: %s (skipping faulty)\n",
2433 bdevname(rdev->bdev, b));
2435 if (mddev->level == LEVEL_MULTIPATH)
2436 /* only need to write one superblock... */
2439 md_super_wait(mddev);
2440 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2442 if (mddev_is_clustered(mddev) && ret == 0)
2443 md_cluster_ops->metadata_update_finish(mddev);
2445 if (mddev->in_sync != sync_req ||
2446 !bit_clear_unless(&mddev->flags, BIT(MD_CHANGE_PENDING),
2447 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_CLEAN)))
2448 /* have to write it out again */
2450 wake_up(&mddev->sb_wait);
2451 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2452 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2454 rdev_for_each(rdev, mddev) {
2455 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2456 clear_bit(Blocked, &rdev->flags);
2458 if (any_badblocks_changed)
2459 ack_all_badblocks(&rdev->badblocks);
2460 clear_bit(BlockedBadBlocks, &rdev->flags);
2461 wake_up(&rdev->blocked_wait);
2464 EXPORT_SYMBOL(md_update_sb);
2466 static int add_bound_rdev(struct md_rdev *rdev)
2468 struct mddev *mddev = rdev->mddev;
2470 bool add_journal = test_bit(Journal, &rdev->flags);
2472 if (!mddev->pers->hot_remove_disk || add_journal) {
2473 /* If there is hot_add_disk but no hot_remove_disk
2474 * then added disks for geometry changes,
2475 * and should be added immediately.
2477 super_types[mddev->major_version].
2478 validate_super(mddev, rdev);
2480 mddev_suspend(mddev);
2481 err = mddev->pers->hot_add_disk(mddev, rdev);
2483 mddev_resume(mddev);
2485 md_kick_rdev_from_array(rdev);
2489 sysfs_notify_dirent_safe(rdev->sysfs_state);
2491 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2492 if (mddev->degraded)
2493 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2494 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2495 md_new_event(mddev);
2496 md_wakeup_thread(mddev->thread);
2500 /* words written to sysfs files may, or may not, be \n terminated.
2501 * We want to accept with case. For this we use cmd_match.
2503 static int cmd_match(const char *cmd, const char *str)
2505 /* See if cmd, written into a sysfs file, matches
2506 * str. They must either be the same, or cmd can
2507 * have a trailing newline
2509 while (*cmd && *str && *cmd == *str) {
2520 struct rdev_sysfs_entry {
2521 struct attribute attr;
2522 ssize_t (*show)(struct md_rdev *, char *);
2523 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2527 state_show(struct md_rdev *rdev, char *page)
2531 unsigned long flags = ACCESS_ONCE(rdev->flags);
2533 if (test_bit(Faulty, &flags) ||
2534 rdev->badblocks.unacked_exist) {
2535 len+= sprintf(page+len, "%sfaulty",sep);
2538 if (test_bit(In_sync, &flags)) {
2539 len += sprintf(page+len, "%sin_sync",sep);
2542 if (test_bit(Journal, &flags)) {
2543 len += sprintf(page+len, "%sjournal",sep);
2546 if (test_bit(WriteMostly, &flags)) {
2547 len += sprintf(page+len, "%swrite_mostly",sep);
2550 if (test_bit(Blocked, &flags) ||
2551 (rdev->badblocks.unacked_exist
2552 && !test_bit(Faulty, &flags))) {
2553 len += sprintf(page+len, "%sblocked", sep);
2556 if (!test_bit(Faulty, &flags) &&
2557 !test_bit(Journal, &flags) &&
2558 !test_bit(In_sync, &flags)) {
2559 len += sprintf(page+len, "%sspare", sep);
2562 if (test_bit(WriteErrorSeen, &flags)) {
2563 len += sprintf(page+len, "%swrite_error", sep);
2566 if (test_bit(WantReplacement, &flags)) {
2567 len += sprintf(page+len, "%swant_replacement", sep);
2570 if (test_bit(Replacement, &flags)) {
2571 len += sprintf(page+len, "%sreplacement", sep);
2575 return len+sprintf(page+len, "\n");
2579 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2582 * faulty - simulates an error
2583 * remove - disconnects the device
2584 * writemostly - sets write_mostly
2585 * -writemostly - clears write_mostly
2586 * blocked - sets the Blocked flags
2587 * -blocked - clears the Blocked and possibly simulates an error
2588 * insync - sets Insync providing device isn't active
2589 * -insync - clear Insync for a device with a slot assigned,
2590 * so that it gets rebuilt based on bitmap
2591 * write_error - sets WriteErrorSeen
2592 * -write_error - clears WriteErrorSeen
2595 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2596 md_error(rdev->mddev, rdev);
2597 if (test_bit(Faulty, &rdev->flags))
2601 } else if (cmd_match(buf, "remove")) {
2602 if (rdev->mddev->pers) {
2603 clear_bit(Blocked, &rdev->flags);
2604 remove_and_add_spares(rdev->mddev, rdev);
2606 if (rdev->raid_disk >= 0)
2609 struct mddev *mddev = rdev->mddev;
2611 if (mddev_is_clustered(mddev))
2612 err = md_cluster_ops->remove_disk(mddev, rdev);
2615 md_kick_rdev_from_array(rdev);
2617 md_update_sb(mddev, 1);
2618 md_new_event(mddev);
2621 } else if (cmd_match(buf, "writemostly")) {
2622 set_bit(WriteMostly, &rdev->flags);
2624 } else if (cmd_match(buf, "-writemostly")) {
2625 clear_bit(WriteMostly, &rdev->flags);
2627 } else if (cmd_match(buf, "blocked")) {
2628 set_bit(Blocked, &rdev->flags);
2630 } else if (cmd_match(buf, "-blocked")) {
2631 if (!test_bit(Faulty, &rdev->flags) &&
2632 rdev->badblocks.unacked_exist) {
2633 /* metadata handler doesn't understand badblocks,
2634 * so we need to fail the device
2636 md_error(rdev->mddev, rdev);
2638 clear_bit(Blocked, &rdev->flags);
2639 clear_bit(BlockedBadBlocks, &rdev->flags);
2640 wake_up(&rdev->blocked_wait);
2641 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2642 md_wakeup_thread(rdev->mddev->thread);
2645 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2646 set_bit(In_sync, &rdev->flags);
2648 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2649 !test_bit(Journal, &rdev->flags)) {
2650 if (rdev->mddev->pers == NULL) {
2651 clear_bit(In_sync, &rdev->flags);
2652 rdev->saved_raid_disk = rdev->raid_disk;
2653 rdev->raid_disk = -1;
2656 } else if (cmd_match(buf, "write_error")) {
2657 set_bit(WriteErrorSeen, &rdev->flags);
2659 } else if (cmd_match(buf, "-write_error")) {
2660 clear_bit(WriteErrorSeen, &rdev->flags);
2662 } else if (cmd_match(buf, "want_replacement")) {
2663 /* Any non-spare device that is not a replacement can
2664 * become want_replacement at any time, but we then need to
2665 * check if recovery is needed.
2667 if (rdev->raid_disk >= 0 &&
2668 !test_bit(Journal, &rdev->flags) &&
2669 !test_bit(Replacement, &rdev->flags))
2670 set_bit(WantReplacement, &rdev->flags);
2671 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2672 md_wakeup_thread(rdev->mddev->thread);
2674 } else if (cmd_match(buf, "-want_replacement")) {
2675 /* Clearing 'want_replacement' is always allowed.
2676 * Once replacements starts it is too late though.
2679 clear_bit(WantReplacement, &rdev->flags);
2680 } else if (cmd_match(buf, "replacement")) {
2681 /* Can only set a device as a replacement when array has not
2682 * yet been started. Once running, replacement is automatic
2683 * from spares, or by assigning 'slot'.
2685 if (rdev->mddev->pers)
2688 set_bit(Replacement, &rdev->flags);
2691 } else if (cmd_match(buf, "-replacement")) {
2692 /* Similarly, can only clear Replacement before start */
2693 if (rdev->mddev->pers)
2696 clear_bit(Replacement, &rdev->flags);
2699 } else if (cmd_match(buf, "re-add")) {
2700 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2701 /* clear_bit is performed _after_ all the devices
2702 * have their local Faulty bit cleared. If any writes
2703 * happen in the meantime in the local node, they
2704 * will land in the local bitmap, which will be synced
2705 * by this node eventually
2707 if (!mddev_is_clustered(rdev->mddev) ||
2708 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2709 clear_bit(Faulty, &rdev->flags);
2710 err = add_bound_rdev(rdev);
2716 sysfs_notify_dirent_safe(rdev->sysfs_state);
2717 return err ? err : len;
2719 static struct rdev_sysfs_entry rdev_state =
2720 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2723 errors_show(struct md_rdev *rdev, char *page)
2725 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2729 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2734 rv = kstrtouint(buf, 10, &n);
2737 atomic_set(&rdev->corrected_errors, n);
2740 static struct rdev_sysfs_entry rdev_errors =
2741 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2744 slot_show(struct md_rdev *rdev, char *page)
2746 if (test_bit(Journal, &rdev->flags))
2747 return sprintf(page, "journal\n");
2748 else if (rdev->raid_disk < 0)
2749 return sprintf(page, "none\n");
2751 return sprintf(page, "%d\n", rdev->raid_disk);
2755 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2760 if (test_bit(Journal, &rdev->flags))
2762 if (strncmp(buf, "none", 4)==0)
2765 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2769 if (rdev->mddev->pers && slot == -1) {
2770 /* Setting 'slot' on an active array requires also
2771 * updating the 'rd%d' link, and communicating
2772 * with the personality with ->hot_*_disk.
2773 * For now we only support removing
2774 * failed/spare devices. This normally happens automatically,
2775 * but not when the metadata is externally managed.
2777 if (rdev->raid_disk == -1)
2779 /* personality does all needed checks */
2780 if (rdev->mddev->pers->hot_remove_disk == NULL)
2782 clear_bit(Blocked, &rdev->flags);
2783 remove_and_add_spares(rdev->mddev, rdev);
2784 if (rdev->raid_disk >= 0)
2786 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2787 md_wakeup_thread(rdev->mddev->thread);
2788 } else if (rdev->mddev->pers) {
2789 /* Activating a spare .. or possibly reactivating
2790 * if we ever get bitmaps working here.
2794 if (rdev->raid_disk != -1)
2797 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2800 if (rdev->mddev->pers->hot_add_disk == NULL)
2803 if (slot >= rdev->mddev->raid_disks &&
2804 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2807 rdev->raid_disk = slot;
2808 if (test_bit(In_sync, &rdev->flags))
2809 rdev->saved_raid_disk = slot;
2811 rdev->saved_raid_disk = -1;
2812 clear_bit(In_sync, &rdev->flags);
2813 clear_bit(Bitmap_sync, &rdev->flags);
2814 err = rdev->mddev->pers->
2815 hot_add_disk(rdev->mddev, rdev);
2817 rdev->raid_disk = -1;
2820 sysfs_notify_dirent_safe(rdev->sysfs_state);
2821 if (sysfs_link_rdev(rdev->mddev, rdev))
2822 /* failure here is OK */;
2823 /* don't wakeup anyone, leave that to userspace. */
2825 if (slot >= rdev->mddev->raid_disks &&
2826 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2828 rdev->raid_disk = slot;
2829 /* assume it is working */
2830 clear_bit(Faulty, &rdev->flags);
2831 clear_bit(WriteMostly, &rdev->flags);
2832 set_bit(In_sync, &rdev->flags);
2833 sysfs_notify_dirent_safe(rdev->sysfs_state);
2838 static struct rdev_sysfs_entry rdev_slot =
2839 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2842 offset_show(struct md_rdev *rdev, char *page)
2844 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2848 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2850 unsigned long long offset;
2851 if (kstrtoull(buf, 10, &offset) < 0)
2853 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2855 if (rdev->sectors && rdev->mddev->external)
2856 /* Must set offset before size, so overlap checks
2859 rdev->data_offset = offset;
2860 rdev->new_data_offset = offset;
2864 static struct rdev_sysfs_entry rdev_offset =
2865 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2867 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2869 return sprintf(page, "%llu\n",
2870 (unsigned long long)rdev->new_data_offset);
2873 static ssize_t new_offset_store(struct md_rdev *rdev,
2874 const char *buf, size_t len)
2876 unsigned long long new_offset;
2877 struct mddev *mddev = rdev->mddev;
2879 if (kstrtoull(buf, 10, &new_offset) < 0)
2882 if (mddev->sync_thread ||
2883 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2885 if (new_offset == rdev->data_offset)
2886 /* reset is always permitted */
2888 else if (new_offset > rdev->data_offset) {
2889 /* must not push array size beyond rdev_sectors */
2890 if (new_offset - rdev->data_offset
2891 + mddev->dev_sectors > rdev->sectors)
2894 /* Metadata worries about other space details. */
2896 /* decreasing the offset is inconsistent with a backwards
2899 if (new_offset < rdev->data_offset &&
2900 mddev->reshape_backwards)
2902 /* Increasing offset is inconsistent with forwards
2903 * reshape. reshape_direction should be set to
2904 * 'backwards' first.
2906 if (new_offset > rdev->data_offset &&
2907 !mddev->reshape_backwards)
2910 if (mddev->pers && mddev->persistent &&
2911 !super_types[mddev->major_version]
2912 .allow_new_offset(rdev, new_offset))
2914 rdev->new_data_offset = new_offset;
2915 if (new_offset > rdev->data_offset)
2916 mddev->reshape_backwards = 1;
2917 else if (new_offset < rdev->data_offset)
2918 mddev->reshape_backwards = 0;
2922 static struct rdev_sysfs_entry rdev_new_offset =
2923 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2926 rdev_size_show(struct md_rdev *rdev, char *page)
2928 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2931 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2933 /* check if two start/length pairs overlap */
2941 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2943 unsigned long long blocks;
2946 if (kstrtoull(buf, 10, &blocks) < 0)
2949 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2950 return -EINVAL; /* sector conversion overflow */
2953 if (new != blocks * 2)
2954 return -EINVAL; /* unsigned long long to sector_t overflow */
2961 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2963 struct mddev *my_mddev = rdev->mddev;
2964 sector_t oldsectors = rdev->sectors;
2967 if (test_bit(Journal, &rdev->flags))
2969 if (strict_blocks_to_sectors(buf, §ors) < 0)
2971 if (rdev->data_offset != rdev->new_data_offset)
2972 return -EINVAL; /* too confusing */
2973 if (my_mddev->pers && rdev->raid_disk >= 0) {
2974 if (my_mddev->persistent) {
2975 sectors = super_types[my_mddev->major_version].
2976 rdev_size_change(rdev, sectors);
2979 } else if (!sectors)
2980 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2982 if (!my_mddev->pers->resize)
2983 /* Cannot change size for RAID0 or Linear etc */
2986 if (sectors < my_mddev->dev_sectors)
2987 return -EINVAL; /* component must fit device */
2989 rdev->sectors = sectors;
2990 if (sectors > oldsectors && my_mddev->external) {
2991 /* Need to check that all other rdevs with the same
2992 * ->bdev do not overlap. 'rcu' is sufficient to walk
2993 * the rdev lists safely.
2994 * This check does not provide a hard guarantee, it
2995 * just helps avoid dangerous mistakes.
2997 struct mddev *mddev;
2999 struct list_head *tmp;
3002 for_each_mddev(mddev, tmp) {
3003 struct md_rdev *rdev2;
3005 rdev_for_each(rdev2, mddev)
3006 if (rdev->bdev == rdev2->bdev &&
3008 overlaps(rdev->data_offset, rdev->sectors,
3021 /* Someone else could have slipped in a size
3022 * change here, but doing so is just silly.
3023 * We put oldsectors back because we *know* it is
3024 * safe, and trust userspace not to race with
3027 rdev->sectors = oldsectors;
3034 static struct rdev_sysfs_entry rdev_size =
3035 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3037 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3039 unsigned long long recovery_start = rdev->recovery_offset;
3041 if (test_bit(In_sync, &rdev->flags) ||
3042 recovery_start == MaxSector)
3043 return sprintf(page, "none\n");
3045 return sprintf(page, "%llu\n", recovery_start);
3048 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3050 unsigned long long recovery_start;
3052 if (cmd_match(buf, "none"))
3053 recovery_start = MaxSector;
3054 else if (kstrtoull(buf, 10, &recovery_start))
3057 if (rdev->mddev->pers &&
3058 rdev->raid_disk >= 0)
3061 rdev->recovery_offset = recovery_start;
3062 if (recovery_start == MaxSector)
3063 set_bit(In_sync, &rdev->flags);
3065 clear_bit(In_sync, &rdev->flags);
3069 static struct rdev_sysfs_entry rdev_recovery_start =
3070 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3072 /* sysfs access to bad-blocks list.
3073 * We present two files.
3074 * 'bad-blocks' lists sector numbers and lengths of ranges that
3075 * are recorded as bad. The list is truncated to fit within
3076 * the one-page limit of sysfs.
3077 * Writing "sector length" to this file adds an acknowledged
3079 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
3080 * been acknowledged. Writing to this file adds bad blocks
3081 * without acknowledging them. This is largely for testing.
3083 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3085 return badblocks_show(&rdev->badblocks, page, 0);
3087 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3089 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3090 /* Maybe that ack was all we needed */
3091 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3092 wake_up(&rdev->blocked_wait);
3095 static struct rdev_sysfs_entry rdev_bad_blocks =
3096 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3098 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3100 return badblocks_show(&rdev->badblocks, page, 1);
3102 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3104 return badblocks_store(&rdev->badblocks, page, len, 1);
3106 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3107 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3109 static struct attribute *rdev_default_attrs[] = {
3114 &rdev_new_offset.attr,
3116 &rdev_recovery_start.attr,
3117 &rdev_bad_blocks.attr,
3118 &rdev_unack_bad_blocks.attr,
3122 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3124 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3125 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3131 return entry->show(rdev, page);
3135 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3136 const char *page, size_t length)
3138 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3139 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3141 struct mddev *mddev = rdev->mddev;
3145 if (!capable(CAP_SYS_ADMIN))
3147 rv = mddev ? mddev_lock(mddev): -EBUSY;
3149 if (rdev->mddev == NULL)
3152 rv = entry->store(rdev, page, length);
3153 mddev_unlock(mddev);
3158 static void rdev_free(struct kobject *ko)
3160 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3163 static const struct sysfs_ops rdev_sysfs_ops = {
3164 .show = rdev_attr_show,
3165 .store = rdev_attr_store,
3167 static struct kobj_type rdev_ktype = {
3168 .release = rdev_free,
3169 .sysfs_ops = &rdev_sysfs_ops,
3170 .default_attrs = rdev_default_attrs,
3173 int md_rdev_init(struct md_rdev *rdev)
3176 rdev->saved_raid_disk = -1;
3177 rdev->raid_disk = -1;
3179 rdev->data_offset = 0;
3180 rdev->new_data_offset = 0;
3181 rdev->sb_events = 0;
3182 rdev->last_read_error = 0;
3183 rdev->sb_loaded = 0;
3184 rdev->bb_page = NULL;
3185 atomic_set(&rdev->nr_pending, 0);
3186 atomic_set(&rdev->read_errors, 0);
3187 atomic_set(&rdev->corrected_errors, 0);
3189 INIT_LIST_HEAD(&rdev->same_set);
3190 init_waitqueue_head(&rdev->blocked_wait);
3192 /* Add space to store bad block list.
3193 * This reserves the space even on arrays where it cannot
3194 * be used - I wonder if that matters
3196 return badblocks_init(&rdev->badblocks, 0);
3198 EXPORT_SYMBOL_GPL(md_rdev_init);
3200 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3202 * mark the device faulty if:
3204 * - the device is nonexistent (zero size)
3205 * - the device has no valid superblock
3207 * a faulty rdev _never_ has rdev->sb set.
3209 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3211 char b[BDEVNAME_SIZE];
3213 struct md_rdev *rdev;
3216 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3218 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3219 return ERR_PTR(-ENOMEM);
3222 err = md_rdev_init(rdev);
3225 err = alloc_disk_sb(rdev);
3229 err = lock_rdev(rdev, newdev, super_format == -2);
3233 kobject_init(&rdev->kobj, &rdev_ktype);
3235 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3238 "md: %s has zero or unknown size, marking faulty!\n",
3239 bdevname(rdev->bdev,b));
3244 if (super_format >= 0) {
3245 err = super_types[super_format].
3246 load_super(rdev, NULL, super_minor);
3247 if (err == -EINVAL) {
3249 "md: %s does not have a valid v%d.%d "
3250 "superblock, not importing!\n",
3251 bdevname(rdev->bdev,b),
3252 super_format, super_minor);
3257 "md: could not read %s's sb, not importing!\n",
3258 bdevname(rdev->bdev,b));
3268 md_rdev_clear(rdev);
3270 return ERR_PTR(err);
3274 * Check a full RAID array for plausibility
3277 static void analyze_sbs(struct mddev *mddev)
3280 struct md_rdev *rdev, *freshest, *tmp;
3281 char b[BDEVNAME_SIZE];
3284 rdev_for_each_safe(rdev, tmp, mddev)
3285 switch (super_types[mddev->major_version].
3286 load_super(rdev, freshest, mddev->minor_version)) {
3294 "md: fatal superblock inconsistency in %s"
3295 " -- removing from array\n",
3296 bdevname(rdev->bdev,b));
3297 md_kick_rdev_from_array(rdev);
3300 super_types[mddev->major_version].
3301 validate_super(mddev, freshest);
3304 rdev_for_each_safe(rdev, tmp, mddev) {
3305 if (mddev->max_disks &&
3306 (rdev->desc_nr >= mddev->max_disks ||
3307 i > mddev->max_disks)) {
3309 "md: %s: %s: only %d devices permitted\n",
3310 mdname(mddev), bdevname(rdev->bdev, b),
3312 md_kick_rdev_from_array(rdev);
3315 if (rdev != freshest) {
3316 if (super_types[mddev->major_version].
3317 validate_super(mddev, rdev)) {
3318 printk(KERN_WARNING "md: kicking non-fresh %s"
3320 bdevname(rdev->bdev,b));
3321 md_kick_rdev_from_array(rdev);
3325 if (mddev->level == LEVEL_MULTIPATH) {
3326 rdev->desc_nr = i++;
3327 rdev->raid_disk = rdev->desc_nr;
3328 set_bit(In_sync, &rdev->flags);
3329 } else if (rdev->raid_disk >=
3330 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3331 !test_bit(Journal, &rdev->flags)) {
3332 rdev->raid_disk = -1;
3333 clear_bit(In_sync, &rdev->flags);
3338 /* Read a fixed-point number.
3339 * Numbers in sysfs attributes should be in "standard" units where
3340 * possible, so time should be in seconds.
3341 * However we internally use a a much smaller unit such as
3342 * milliseconds or jiffies.
3343 * This function takes a decimal number with a possible fractional
3344 * component, and produces an integer which is the result of
3345 * multiplying that number by 10^'scale'.
3346 * all without any floating-point arithmetic.
3348 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3350 unsigned long result = 0;
3352 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3355 else if (decimals < scale) {
3358 result = result * 10 + value;
3370 while (decimals < scale) {
3379 safe_delay_show(struct mddev *mddev, char *page)
3381 int msec = (mddev->safemode_delay*1000)/HZ;
3382 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3385 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3389 if (mddev_is_clustered(mddev)) {
3390 pr_info("md: Safemode is disabled for clustered mode\n");
3394 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3397 mddev->safemode_delay = 0;
3399 unsigned long old_delay = mddev->safemode_delay;
3400 unsigned long new_delay = (msec*HZ)/1000;
3404 mddev->safemode_delay = new_delay;
3405 if (new_delay < old_delay || old_delay == 0)
3406 mod_timer(&mddev->safemode_timer, jiffies+1);
3410 static struct md_sysfs_entry md_safe_delay =
3411 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3414 level_show(struct mddev *mddev, char *page)
3416 struct md_personality *p;
3418 spin_lock(&mddev->lock);
3421 ret = sprintf(page, "%s\n", p->name);
3422 else if (mddev->clevel[0])
3423 ret = sprintf(page, "%s\n", mddev->clevel);
3424 else if (mddev->level != LEVEL_NONE)
3425 ret = sprintf(page, "%d\n", mddev->level);
3428 spin_unlock(&mddev->lock);
3433 level_store(struct mddev *mddev, const char *buf, size_t len)
3438 struct md_personality *pers, *oldpers;
3440 void *priv, *oldpriv;
3441 struct md_rdev *rdev;
3443 if (slen == 0 || slen >= sizeof(clevel))
3446 rv = mddev_lock(mddev);
3450 if (mddev->pers == NULL) {
3451 strncpy(mddev->clevel, buf, slen);
3452 if (mddev->clevel[slen-1] == '\n')
3454 mddev->clevel[slen] = 0;
3455 mddev->level = LEVEL_NONE;
3463 /* request to change the personality. Need to ensure:
3464 * - array is not engaged in resync/recovery/reshape
3465 * - old personality can be suspended
3466 * - new personality will access other array.
3470 if (mddev->sync_thread ||
3471 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3472 mddev->reshape_position != MaxSector ||
3473 mddev->sysfs_active)
3477 if (!mddev->pers->quiesce) {
3478 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3479 mdname(mddev), mddev->pers->name);
3483 /* Now find the new personality */
3484 strncpy(clevel, buf, slen);
3485 if (clevel[slen-1] == '\n')
3488 if (kstrtol(clevel, 10, &level))
3491 if (request_module("md-%s", clevel) != 0)
3492 request_module("md-level-%s", clevel);
3493 spin_lock(&pers_lock);
3494 pers = find_pers(level, clevel);
3495 if (!pers || !try_module_get(pers->owner)) {
3496 spin_unlock(&pers_lock);
3497 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3501 spin_unlock(&pers_lock);
3503 if (pers == mddev->pers) {
3504 /* Nothing to do! */
3505 module_put(pers->owner);
3509 if (!pers->takeover) {
3510 module_put(pers->owner);
3511 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3512 mdname(mddev), clevel);
3517 rdev_for_each(rdev, mddev)
3518 rdev->new_raid_disk = rdev->raid_disk;
3520 /* ->takeover must set new_* and/or delta_disks
3521 * if it succeeds, and may set them when it fails.
3523 priv = pers->takeover(mddev);
3525 mddev->new_level = mddev->level;
3526 mddev->new_layout = mddev->layout;
3527 mddev->new_chunk_sectors = mddev->chunk_sectors;
3528 mddev->raid_disks -= mddev->delta_disks;
3529 mddev->delta_disks = 0;
3530 mddev->reshape_backwards = 0;
3531 module_put(pers->owner);
3532 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3533 mdname(mddev), clevel);
3538 /* Looks like we have a winner */
3539 mddev_suspend(mddev);
3540 mddev_detach(mddev);
3542 spin_lock(&mddev->lock);
3543 oldpers = mddev->pers;
3544 oldpriv = mddev->private;
3546 mddev->private = priv;
3547 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3548 mddev->level = mddev->new_level;
3549 mddev->layout = mddev->new_layout;
3550 mddev->chunk_sectors = mddev->new_chunk_sectors;
3551 mddev->delta_disks = 0;
3552 mddev->reshape_backwards = 0;
3553 mddev->degraded = 0;
3554 spin_unlock(&mddev->lock);
3556 if (oldpers->sync_request == NULL &&
3558 /* We are converting from a no-redundancy array
3559 * to a redundancy array and metadata is managed
3560 * externally so we need to be sure that writes
3561 * won't block due to a need to transition
3563 * until external management is started.
3566 mddev->safemode_delay = 0;
3567 mddev->safemode = 0;
3570 oldpers->free(mddev, oldpriv);
3572 if (oldpers->sync_request == NULL &&
3573 pers->sync_request != NULL) {
3574 /* need to add the md_redundancy_group */
3575 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3577 "md: cannot register extra attributes for %s\n",
3579 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3581 if (oldpers->sync_request != NULL &&
3582 pers->sync_request == NULL) {
3583 /* need to remove the md_redundancy_group */
3584 if (mddev->to_remove == NULL)
3585 mddev->to_remove = &md_redundancy_group;
3588 module_put(oldpers->owner);
3590 rdev_for_each(rdev, mddev) {
3591 if (rdev->raid_disk < 0)
3593 if (rdev->new_raid_disk >= mddev->raid_disks)
3594 rdev->new_raid_disk = -1;
3595 if (rdev->new_raid_disk == rdev->raid_disk)
3597 sysfs_unlink_rdev(mddev, rdev);
3599 rdev_for_each(rdev, mddev) {
3600 if (rdev->raid_disk < 0)
3602 if (rdev->new_raid_disk == rdev->raid_disk)
3604 rdev->raid_disk = rdev->new_raid_disk;
3605 if (rdev->raid_disk < 0)
3606 clear_bit(In_sync, &rdev->flags);
3608 if (sysfs_link_rdev(mddev, rdev))
3609 printk(KERN_WARNING "md: cannot register rd%d"
3610 " for %s after level change\n",
3611 rdev->raid_disk, mdname(mddev));
3615 if (pers->sync_request == NULL) {
3616 /* this is now an array without redundancy, so
3617 * it must always be in_sync
3620 del_timer_sync(&mddev->safemode_timer);
3622 blk_set_stacking_limits(&mddev->queue->limits);
3624 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3625 mddev_resume(mddev);
3627 md_update_sb(mddev, 1);
3628 sysfs_notify(&mddev->kobj, NULL, "level");
3629 md_new_event(mddev);
3632 mddev_unlock(mddev);
3636 static struct md_sysfs_entry md_level =
3637 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3640 layout_show(struct mddev *mddev, char *page)
3642 /* just a number, not meaningful for all levels */
3643 if (mddev->reshape_position != MaxSector &&
3644 mddev->layout != mddev->new_layout)
3645 return sprintf(page, "%d (%d)\n",
3646 mddev->new_layout, mddev->layout);
3647 return sprintf(page, "%d\n", mddev->layout);
3651 layout_store(struct mddev *mddev, const char *buf, size_t len)
3656 err = kstrtouint(buf, 10, &n);
3659 err = mddev_lock(mddev);
3664 if (mddev->pers->check_reshape == NULL)
3669 mddev->new_layout = n;
3670 err = mddev->pers->check_reshape(mddev);
3672 mddev->new_layout = mddev->layout;
3675 mddev->new_layout = n;
3676 if (mddev->reshape_position == MaxSector)
3679 mddev_unlock(mddev);
3682 static struct md_sysfs_entry md_layout =
3683 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3686 raid_disks_show(struct mddev *mddev, char *page)
3688 if (mddev->raid_disks == 0)
3690 if (mddev->reshape_position != MaxSector &&
3691 mddev->delta_disks != 0)
3692 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3693 mddev->raid_disks - mddev->delta_disks);
3694 return sprintf(page, "%d\n", mddev->raid_disks);
3697 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3700 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3705 err = kstrtouint(buf, 10, &n);
3709 err = mddev_lock(mddev);
3713 err = update_raid_disks(mddev, n);
3714 else if (mddev->reshape_position != MaxSector) {
3715 struct md_rdev *rdev;
3716 int olddisks = mddev->raid_disks - mddev->delta_disks;
3719 rdev_for_each(rdev, mddev) {
3721 rdev->data_offset < rdev->new_data_offset)
3724 rdev->data_offset > rdev->new_data_offset)
3728 mddev->delta_disks = n - olddisks;
3729 mddev->raid_disks = n;
3730 mddev->reshape_backwards = (mddev->delta_disks < 0);
3732 mddev->raid_disks = n;
3734 mddev_unlock(mddev);
3735 return err ? err : len;
3737 static struct md_sysfs_entry md_raid_disks =
3738 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3741 chunk_size_show(struct mddev *mddev, char *page)
3743 if (mddev->reshape_position != MaxSector &&
3744 mddev->chunk_sectors != mddev->new_chunk_sectors)
3745 return sprintf(page, "%d (%d)\n",
3746 mddev->new_chunk_sectors << 9,
3747 mddev->chunk_sectors << 9);
3748 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3752 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3757 err = kstrtoul(buf, 10, &n);
3761 err = mddev_lock(mddev);
3765 if (mddev->pers->check_reshape == NULL)
3770 mddev->new_chunk_sectors = n >> 9;
3771 err = mddev->pers->check_reshape(mddev);
3773 mddev->new_chunk_sectors = mddev->chunk_sectors;
3776 mddev->new_chunk_sectors = n >> 9;
3777 if (mddev->reshape_position == MaxSector)
3778 mddev->chunk_sectors = n >> 9;
3780 mddev_unlock(mddev);
3783 static struct md_sysfs_entry md_chunk_size =
3784 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3787 resync_start_show(struct mddev *mddev, char *page)
3789 if (mddev->recovery_cp == MaxSector)
3790 return sprintf(page, "none\n");
3791 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3795 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3797 unsigned long long n;
3800 if (cmd_match(buf, "none"))
3803 err = kstrtoull(buf, 10, &n);
3806 if (n != (sector_t)n)
3810 err = mddev_lock(mddev);
3813 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3817 mddev->recovery_cp = n;
3819 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3821 mddev_unlock(mddev);
3824 static struct md_sysfs_entry md_resync_start =
3825 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3826 resync_start_show, resync_start_store);
3829 * The array state can be:
3832 * No devices, no size, no level
3833 * Equivalent to STOP_ARRAY ioctl
3835 * May have some settings, but array is not active
3836 * all IO results in error
3837 * When written, doesn't tear down array, but just stops it
3838 * suspended (not supported yet)
3839 * All IO requests will block. The array can be reconfigured.
3840 * Writing this, if accepted, will block until array is quiescent
3842 * no resync can happen. no superblocks get written.
3843 * write requests fail
3845 * like readonly, but behaves like 'clean' on a write request.
3847 * clean - no pending writes, but otherwise active.
3848 * When written to inactive array, starts without resync
3849 * If a write request arrives then
3850 * if metadata is known, mark 'dirty' and switch to 'active'.
3851 * if not known, block and switch to write-pending
3852 * If written to an active array that has pending writes, then fails.
3854 * fully active: IO and resync can be happening.
3855 * When written to inactive array, starts with resync
3858 * clean, but writes are blocked waiting for 'active' to be written.
3861 * like active, but no writes have been seen for a while (100msec).
3864 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3865 write_pending, active_idle, bad_word};
3866 static char *array_states[] = {
3867 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3868 "write-pending", "active-idle", NULL };
3870 static int match_word(const char *word, char **list)
3873 for (n=0; list[n]; n++)
3874 if (cmd_match(word, list[n]))
3880 array_state_show(struct mddev *mddev, char *page)
3882 enum array_state st = inactive;
3895 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3897 else if (mddev->safemode)
3903 if (list_empty(&mddev->disks) &&
3904 mddev->raid_disks == 0 &&
3905 mddev->dev_sectors == 0)
3910 return sprintf(page, "%s\n", array_states[st]);
3913 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3914 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3915 static int do_md_run(struct mddev *mddev);
3916 static int restart_array(struct mddev *mddev);
3919 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3922 enum array_state st = match_word(buf, array_states);
3924 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3925 /* don't take reconfig_mutex when toggling between
3928 spin_lock(&mddev->lock);
3930 restart_array(mddev);
3931 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3932 wake_up(&mddev->sb_wait);
3934 } else /* st == clean */ {
3935 restart_array(mddev);
3936 if (atomic_read(&mddev->writes_pending) == 0) {
3937 if (mddev->in_sync == 0) {
3939 if (mddev->safemode == 1)
3940 mddev->safemode = 0;
3941 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3948 sysfs_notify_dirent_safe(mddev->sysfs_state);
3949 spin_unlock(&mddev->lock);
3952 err = mddev_lock(mddev);
3960 /* stopping an active array */
3961 err = do_md_stop(mddev, 0, NULL);
3964 /* stopping an active array */
3966 err = do_md_stop(mddev, 2, NULL);
3968 err = 0; /* already inactive */
3971 break; /* not supported yet */
3974 err = md_set_readonly(mddev, NULL);
3977 set_disk_ro(mddev->gendisk, 1);
3978 err = do_md_run(mddev);
3984 err = md_set_readonly(mddev, NULL);
3985 else if (mddev->ro == 1)
3986 err = restart_array(mddev);
3989 set_disk_ro(mddev->gendisk, 0);
3993 err = do_md_run(mddev);
3998 err = restart_array(mddev);
4001 spin_lock(&mddev->lock);
4002 if (atomic_read(&mddev->writes_pending) == 0) {
4003 if (mddev->in_sync == 0) {
4005 if (mddev->safemode == 1)
4006 mddev->safemode = 0;
4007 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
4012 spin_unlock(&mddev->lock);
4018 err = restart_array(mddev);
4021 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4022 wake_up(&mddev->sb_wait);
4026 set_disk_ro(mddev->gendisk, 0);
4027 err = do_md_run(mddev);
4032 /* these cannot be set */
4037 if (mddev->hold_active == UNTIL_IOCTL)
4038 mddev->hold_active = 0;
4039 sysfs_notify_dirent_safe(mddev->sysfs_state);
4041 mddev_unlock(mddev);
4044 static struct md_sysfs_entry md_array_state =
4045 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4048 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4049 return sprintf(page, "%d\n",
4050 atomic_read(&mddev->max_corr_read_errors));
4054 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4059 rv = kstrtouint(buf, 10, &n);
4062 atomic_set(&mddev->max_corr_read_errors, n);
4066 static struct md_sysfs_entry max_corr_read_errors =
4067 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4068 max_corrected_read_errors_store);
4071 null_show(struct mddev *mddev, char *page)
4077 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4079 /* buf must be %d:%d\n? giving major and minor numbers */
4080 /* The new device is added to the array.
4081 * If the array has a persistent superblock, we read the
4082 * superblock to initialise info and check validity.
4083 * Otherwise, only checking done is that in bind_rdev_to_array,
4084 * which mainly checks size.
4087 int major = simple_strtoul(buf, &e, 10);
4090 struct md_rdev *rdev;
4093 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4095 minor = simple_strtoul(e+1, &e, 10);
4096 if (*e && *e != '\n')
4098 dev = MKDEV(major, minor);
4099 if (major != MAJOR(dev) ||
4100 minor != MINOR(dev))
4103 flush_workqueue(md_misc_wq);
4105 err = mddev_lock(mddev);
4108 if (mddev->persistent) {
4109 rdev = md_import_device(dev, mddev->major_version,
4110 mddev->minor_version);
4111 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4112 struct md_rdev *rdev0
4113 = list_entry(mddev->disks.next,
4114 struct md_rdev, same_set);
4115 err = super_types[mddev->major_version]
4116 .load_super(rdev, rdev0, mddev->minor_version);
4120 } else if (mddev->external)
4121 rdev = md_import_device(dev, -2, -1);
4123 rdev = md_import_device(dev, -1, -1);
4126 mddev_unlock(mddev);
4127 return PTR_ERR(rdev);
4129 err = bind_rdev_to_array(rdev, mddev);
4133 mddev_unlock(mddev);
4134 return err ? err : len;
4137 static struct md_sysfs_entry md_new_device =
4138 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4141 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4144 unsigned long chunk, end_chunk;
4147 err = mddev_lock(mddev);
4152 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4154 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4155 if (buf == end) break;
4156 if (*end == '-') { /* range */
4158 end_chunk = simple_strtoul(buf, &end, 0);
4159 if (buf == end) break;
4161 if (*end && !isspace(*end)) break;
4162 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4163 buf = skip_spaces(end);
4165 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4167 mddev_unlock(mddev);
4171 static struct md_sysfs_entry md_bitmap =
4172 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4175 size_show(struct mddev *mddev, char *page)
4177 return sprintf(page, "%llu\n",
4178 (unsigned long long)mddev->dev_sectors / 2);
4181 static int update_size(struct mddev *mddev, sector_t num_sectors);
4184 size_store(struct mddev *mddev, const char *buf, size_t len)
4186 /* If array is inactive, we can reduce the component size, but
4187 * not increase it (except from 0).
4188 * If array is active, we can try an on-line resize
4191 int err = strict_blocks_to_sectors(buf, §ors);
4195 err = mddev_lock(mddev);
4199 err = update_size(mddev, sectors);
4201 md_update_sb(mddev, 1);
4203 if (mddev->dev_sectors == 0 ||
4204 mddev->dev_sectors > sectors)
4205 mddev->dev_sectors = sectors;
4209 mddev_unlock(mddev);
4210 return err ? err : len;
4213 static struct md_sysfs_entry md_size =
4214 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4216 /* Metadata version.
4218 * 'none' for arrays with no metadata (good luck...)
4219 * 'external' for arrays with externally managed metadata,
4220 * or N.M for internally known formats
4223 metadata_show(struct mddev *mddev, char *page)
4225 if (mddev->persistent)
4226 return sprintf(page, "%d.%d\n",
4227 mddev->major_version, mddev->minor_version);
4228 else if (mddev->external)
4229 return sprintf(page, "external:%s\n", mddev->metadata_type);
4231 return sprintf(page, "none\n");
4235 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4240 /* Changing the details of 'external' metadata is
4241 * always permitted. Otherwise there must be
4242 * no devices attached to the array.
4245 err = mddev_lock(mddev);
4249 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4251 else if (!list_empty(&mddev->disks))
4255 if (cmd_match(buf, "none")) {
4256 mddev->persistent = 0;
4257 mddev->external = 0;
4258 mddev->major_version = 0;
4259 mddev->minor_version = 90;
4262 if (strncmp(buf, "external:", 9) == 0) {
4263 size_t namelen = len-9;
4264 if (namelen >= sizeof(mddev->metadata_type))
4265 namelen = sizeof(mddev->metadata_type)-1;
4266 strncpy(mddev->metadata_type, buf+9, namelen);
4267 mddev->metadata_type[namelen] = 0;
4268 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4269 mddev->metadata_type[--namelen] = 0;
4270 mddev->persistent = 0;
4271 mddev->external = 1;
4272 mddev->major_version = 0;
4273 mddev->minor_version = 90;
4276 major = simple_strtoul(buf, &e, 10);
4278 if (e==buf || *e != '.')
4281 minor = simple_strtoul(buf, &e, 10);
4282 if (e==buf || (*e && *e != '\n') )
4285 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4287 mddev->major_version = major;
4288 mddev->minor_version = minor;
4289 mddev->persistent = 1;
4290 mddev->external = 0;
4293 mddev_unlock(mddev);
4297 static struct md_sysfs_entry md_metadata =
4298 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4301 action_show(struct mddev *mddev, char *page)
4303 char *type = "idle";
4304 unsigned long recovery = mddev->recovery;
4305 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4307 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4308 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4309 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4311 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4312 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4314 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4318 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4320 else if (mddev->reshape_position != MaxSector)
4323 return sprintf(page, "%s\n", type);
4327 action_store(struct mddev *mddev, const char *page, size_t len)
4329 if (!mddev->pers || !mddev->pers->sync_request)
4333 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4334 if (cmd_match(page, "frozen"))
4335 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4337 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4338 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4339 mddev_lock(mddev) == 0) {
4340 flush_workqueue(md_misc_wq);
4341 if (mddev->sync_thread) {
4342 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4343 md_reap_sync_thread(mddev);
4345 mddev_unlock(mddev);
4347 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4349 else if (cmd_match(page, "resync"))
4350 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4351 else if (cmd_match(page, "recover")) {
4352 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4353 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4354 } else if (cmd_match(page, "reshape")) {
4356 if (mddev->pers->start_reshape == NULL)
4358 err = mddev_lock(mddev);
4360 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4363 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4364 err = mddev->pers->start_reshape(mddev);
4366 mddev_unlock(mddev);
4370 sysfs_notify(&mddev->kobj, NULL, "degraded");
4372 if (cmd_match(page, "check"))
4373 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4374 else if (!cmd_match(page, "repair"))
4376 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4377 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4378 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4380 if (mddev->ro == 2) {
4381 /* A write to sync_action is enough to justify
4382 * canceling read-auto mode
4385 md_wakeup_thread(mddev->sync_thread);
4387 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4388 md_wakeup_thread(mddev->thread);
4389 sysfs_notify_dirent_safe(mddev->sysfs_action);
4393 static struct md_sysfs_entry md_scan_mode =
4394 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4397 last_sync_action_show(struct mddev *mddev, char *page)
4399 return sprintf(page, "%s\n", mddev->last_sync_action);
4402 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4405 mismatch_cnt_show(struct mddev *mddev, char *page)
4407 return sprintf(page, "%llu\n",
4408 (unsigned long long)
4409 atomic64_read(&mddev->resync_mismatches));
4412 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4415 sync_min_show(struct mddev *mddev, char *page)
4417 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4418 mddev->sync_speed_min ? "local": "system");
4422 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4427 if (strncmp(buf, "system", 6)==0) {
4430 rv = kstrtouint(buf, 10, &min);
4436 mddev->sync_speed_min = min;
4440 static struct md_sysfs_entry md_sync_min =
4441 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4444 sync_max_show(struct mddev *mddev, char *page)
4446 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4447 mddev->sync_speed_max ? "local": "system");
4451 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4456 if (strncmp(buf, "system", 6)==0) {
4459 rv = kstrtouint(buf, 10, &max);
4465 mddev->sync_speed_max = max;
4469 static struct md_sysfs_entry md_sync_max =
4470 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4473 degraded_show(struct mddev *mddev, char *page)
4475 return sprintf(page, "%d\n", mddev->degraded);
4477 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4480 sync_force_parallel_show(struct mddev *mddev, char *page)
4482 return sprintf(page, "%d\n", mddev->parallel_resync);
4486 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4490 if (kstrtol(buf, 10, &n))
4493 if (n != 0 && n != 1)
4496 mddev->parallel_resync = n;
4498 if (mddev->sync_thread)
4499 wake_up(&resync_wait);
4504 /* force parallel resync, even with shared block devices */
4505 static struct md_sysfs_entry md_sync_force_parallel =
4506 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4507 sync_force_parallel_show, sync_force_parallel_store);
4510 sync_speed_show(struct mddev *mddev, char *page)
4512 unsigned long resync, dt, db;
4513 if (mddev->curr_resync == 0)
4514 return sprintf(page, "none\n");
4515 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4516 dt = (jiffies - mddev->resync_mark) / HZ;
4518 db = resync - mddev->resync_mark_cnt;
4519 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4522 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4525 sync_completed_show(struct mddev *mddev, char *page)
4527 unsigned long long max_sectors, resync;
4529 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4530 return sprintf(page, "none\n");
4532 if (mddev->curr_resync == 1 ||
4533 mddev->curr_resync == 2)
4534 return sprintf(page, "delayed\n");
4536 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4537 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4538 max_sectors = mddev->resync_max_sectors;
4540 max_sectors = mddev->dev_sectors;
4542 resync = mddev->curr_resync_completed;
4543 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4546 static struct md_sysfs_entry md_sync_completed =
4547 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4550 min_sync_show(struct mddev *mddev, char *page)
4552 return sprintf(page, "%llu\n",
4553 (unsigned long long)mddev->resync_min);
4556 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4558 unsigned long long min;
4561 if (kstrtoull(buf, 10, &min))
4564 spin_lock(&mddev->lock);
4566 if (min > mddev->resync_max)
4570 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4573 /* Round down to multiple of 4K for safety */
4574 mddev->resync_min = round_down(min, 8);
4578 spin_unlock(&mddev->lock);
4582 static struct md_sysfs_entry md_min_sync =
4583 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4586 max_sync_show(struct mddev *mddev, char *page)
4588 if (mddev->resync_max == MaxSector)
4589 return sprintf(page, "max\n");
4591 return sprintf(page, "%llu\n",
4592 (unsigned long long)mddev->resync_max);
4595 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4598 spin_lock(&mddev->lock);
4599 if (strncmp(buf, "max", 3) == 0)
4600 mddev->resync_max = MaxSector;
4602 unsigned long long max;
4606 if (kstrtoull(buf, 10, &max))
4608 if (max < mddev->resync_min)
4612 if (max < mddev->resync_max &&
4614 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4617 /* Must be a multiple of chunk_size */
4618 chunk = mddev->chunk_sectors;
4620 sector_t temp = max;
4623 if (sector_div(temp, chunk))
4626 mddev->resync_max = max;
4628 wake_up(&mddev->recovery_wait);
4631 spin_unlock(&mddev->lock);
4635 static struct md_sysfs_entry md_max_sync =
4636 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4639 suspend_lo_show(struct mddev *mddev, char *page)
4641 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4645 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4647 unsigned long long old, new;
4650 err = kstrtoull(buf, 10, &new);
4653 if (new != (sector_t)new)
4656 err = mddev_lock(mddev);
4660 if (mddev->pers == NULL ||
4661 mddev->pers->quiesce == NULL)
4663 old = mddev->suspend_lo;
4664 mddev->suspend_lo = new;
4666 /* Shrinking suspended region */
4667 mddev->pers->quiesce(mddev, 2);
4669 /* Expanding suspended region - need to wait */
4670 mddev->pers->quiesce(mddev, 1);
4671 mddev->pers->quiesce(mddev, 0);
4675 mddev_unlock(mddev);
4678 static struct md_sysfs_entry md_suspend_lo =
4679 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4682 suspend_hi_show(struct mddev *mddev, char *page)
4684 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4688 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4690 unsigned long long old, new;
4693 err = kstrtoull(buf, 10, &new);
4696 if (new != (sector_t)new)
4699 err = mddev_lock(mddev);
4703 if (mddev->pers == NULL ||
4704 mddev->pers->quiesce == NULL)
4706 old = mddev->suspend_hi;
4707 mddev->suspend_hi = new;
4709 /* Shrinking suspended region */
4710 mddev->pers->quiesce(mddev, 2);
4712 /* Expanding suspended region - need to wait */
4713 mddev->pers->quiesce(mddev, 1);
4714 mddev->pers->quiesce(mddev, 0);
4718 mddev_unlock(mddev);
4721 static struct md_sysfs_entry md_suspend_hi =
4722 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4725 reshape_position_show(struct mddev *mddev, char *page)
4727 if (mddev->reshape_position != MaxSector)
4728 return sprintf(page, "%llu\n",
4729 (unsigned long long)mddev->reshape_position);
4730 strcpy(page, "none\n");
4735 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4737 struct md_rdev *rdev;
4738 unsigned long long new;
4741 err = kstrtoull(buf, 10, &new);
4744 if (new != (sector_t)new)
4746 err = mddev_lock(mddev);
4752 mddev->reshape_position = new;
4753 mddev->delta_disks = 0;
4754 mddev->reshape_backwards = 0;
4755 mddev->new_level = mddev->level;
4756 mddev->new_layout = mddev->layout;
4757 mddev->new_chunk_sectors = mddev->chunk_sectors;
4758 rdev_for_each(rdev, mddev)
4759 rdev->new_data_offset = rdev->data_offset;
4762 mddev_unlock(mddev);
4766 static struct md_sysfs_entry md_reshape_position =
4767 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4768 reshape_position_store);
4771 reshape_direction_show(struct mddev *mddev, char *page)
4773 return sprintf(page, "%s\n",
4774 mddev->reshape_backwards ? "backwards" : "forwards");
4778 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4783 if (cmd_match(buf, "forwards"))
4785 else if (cmd_match(buf, "backwards"))
4789 if (mddev->reshape_backwards == backwards)
4792 err = mddev_lock(mddev);
4795 /* check if we are allowed to change */
4796 if (mddev->delta_disks)
4798 else if (mddev->persistent &&
4799 mddev->major_version == 0)
4802 mddev->reshape_backwards = backwards;
4803 mddev_unlock(mddev);
4807 static struct md_sysfs_entry md_reshape_direction =
4808 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4809 reshape_direction_store);
4812 array_size_show(struct mddev *mddev, char *page)
4814 if (mddev->external_size)
4815 return sprintf(page, "%llu\n",
4816 (unsigned long long)mddev->array_sectors/2);
4818 return sprintf(page, "default\n");
4822 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4827 err = mddev_lock(mddev);
4831 /* cluster raid doesn't support change array_sectors */
4832 if (mddev_is_clustered(mddev))
4835 if (strncmp(buf, "default", 7) == 0) {
4837 sectors = mddev->pers->size(mddev, 0, 0);
4839 sectors = mddev->array_sectors;
4841 mddev->external_size = 0;
4843 if (strict_blocks_to_sectors(buf, §ors) < 0)
4845 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4848 mddev->external_size = 1;
4852 mddev->array_sectors = sectors;
4854 set_capacity(mddev->gendisk, mddev->array_sectors);
4855 revalidate_disk(mddev->gendisk);
4858 mddev_unlock(mddev);
4862 static struct md_sysfs_entry md_array_size =
4863 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4866 static struct attribute *md_default_attrs[] = {
4869 &md_raid_disks.attr,
4870 &md_chunk_size.attr,
4872 &md_resync_start.attr,
4874 &md_new_device.attr,
4875 &md_safe_delay.attr,
4876 &md_array_state.attr,
4877 &md_reshape_position.attr,
4878 &md_reshape_direction.attr,
4879 &md_array_size.attr,
4880 &max_corr_read_errors.attr,
4884 static struct attribute *md_redundancy_attrs[] = {
4886 &md_last_scan_mode.attr,
4887 &md_mismatches.attr,
4890 &md_sync_speed.attr,
4891 &md_sync_force_parallel.attr,
4892 &md_sync_completed.attr,
4895 &md_suspend_lo.attr,
4896 &md_suspend_hi.attr,
4901 static struct attribute_group md_redundancy_group = {
4903 .attrs = md_redundancy_attrs,
4907 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4909 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4910 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4915 spin_lock(&all_mddevs_lock);
4916 if (list_empty(&mddev->all_mddevs)) {
4917 spin_unlock(&all_mddevs_lock);
4921 spin_unlock(&all_mddevs_lock);
4923 rv = entry->show(mddev, page);
4929 md_attr_store(struct kobject *kobj, struct attribute *attr,
4930 const char *page, size_t length)
4932 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4933 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4938 if (!capable(CAP_SYS_ADMIN))
4940 spin_lock(&all_mddevs_lock);
4941 if (list_empty(&mddev->all_mddevs)) {
4942 spin_unlock(&all_mddevs_lock);
4946 spin_unlock(&all_mddevs_lock);
4947 rv = entry->store(mddev, page, length);
4952 static void md_free(struct kobject *ko)
4954 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4956 if (mddev->sysfs_state)
4957 sysfs_put(mddev->sysfs_state);
4960 blk_cleanup_queue(mddev->queue);
4961 if (mddev->gendisk) {
4962 del_gendisk(mddev->gendisk);
4963 put_disk(mddev->gendisk);
4969 static const struct sysfs_ops md_sysfs_ops = {
4970 .show = md_attr_show,
4971 .store = md_attr_store,
4973 static struct kobj_type md_ktype = {
4975 .sysfs_ops = &md_sysfs_ops,
4976 .default_attrs = md_default_attrs,
4981 static void mddev_delayed_delete(struct work_struct *ws)
4983 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4985 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4986 kobject_del(&mddev->kobj);
4987 kobject_put(&mddev->kobj);
4990 static int md_alloc(dev_t dev, char *name)
4992 static DEFINE_MUTEX(disks_mutex);
4993 struct mddev *mddev = mddev_find(dev);
4994 struct gendisk *disk;
5003 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
5004 shift = partitioned ? MdpMinorShift : 0;
5005 unit = MINOR(mddev->unit) >> shift;
5007 /* wait for any previous instance of this device to be
5008 * completely removed (mddev_delayed_delete).
5010 flush_workqueue(md_misc_wq);
5012 mutex_lock(&disks_mutex);
5018 /* Need to ensure that 'name' is not a duplicate.
5020 struct mddev *mddev2;
5021 spin_lock(&all_mddevs_lock);
5023 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5024 if (mddev2->gendisk &&
5025 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5026 spin_unlock(&all_mddevs_lock);
5029 spin_unlock(&all_mddevs_lock);
5033 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5036 mddev->queue->queuedata = mddev;
5038 blk_queue_make_request(mddev->queue, md_make_request);
5039 blk_set_stacking_limits(&mddev->queue->limits);
5041 disk = alloc_disk(1 << shift);
5043 blk_cleanup_queue(mddev->queue);
5044 mddev->queue = NULL;
5047 disk->major = MAJOR(mddev->unit);
5048 disk->first_minor = unit << shift;
5050 strcpy(disk->disk_name, name);
5051 else if (partitioned)
5052 sprintf(disk->disk_name, "md_d%d", unit);
5054 sprintf(disk->disk_name, "md%d", unit);
5055 disk->fops = &md_fops;
5056 disk->private_data = mddev;
5057 disk->queue = mddev->queue;
5058 blk_queue_write_cache(mddev->queue, true, true);
5059 /* Allow extended partitions. This makes the
5060 * 'mdp' device redundant, but we can't really
5063 disk->flags |= GENHD_FL_EXT_DEVT;
5064 mddev->gendisk = disk;
5065 /* As soon as we call add_disk(), another thread could get
5066 * through to md_open, so make sure it doesn't get too far
5068 mutex_lock(&mddev->open_mutex);
5071 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5072 &disk_to_dev(disk)->kobj, "%s", "md");
5074 /* This isn't possible, but as kobject_init_and_add is marked
5075 * __must_check, we must do something with the result
5077 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5081 if (mddev->kobj.sd &&
5082 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5083 printk(KERN_DEBUG "pointless warning\n");
5084 mutex_unlock(&mddev->open_mutex);
5086 mutex_unlock(&disks_mutex);
5087 if (!error && mddev->kobj.sd) {
5088 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5089 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5095 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5097 md_alloc(dev, NULL);
5101 static int add_named_array(const char *val, struct kernel_param *kp)
5103 /* val must be "md_*" where * is not all digits.
5104 * We allocate an array with a large free minor number, and
5105 * set the name to val. val must not already be an active name.
5107 int len = strlen(val);
5108 char buf[DISK_NAME_LEN];
5110 while (len && val[len-1] == '\n')
5112 if (len >= DISK_NAME_LEN)
5114 strlcpy(buf, val, len+1);
5115 if (strncmp(buf, "md_", 3) != 0)
5117 return md_alloc(0, buf);
5120 static void md_safemode_timeout(unsigned long data)
5122 struct mddev *mddev = (struct mddev *) data;
5124 if (!atomic_read(&mddev->writes_pending)) {
5125 mddev->safemode = 1;
5126 if (mddev->external)
5127 sysfs_notify_dirent_safe(mddev->sysfs_state);
5129 md_wakeup_thread(mddev->thread);
5132 static int start_dirty_degraded;
5134 int md_run(struct mddev *mddev)
5137 struct md_rdev *rdev;
5138 struct md_personality *pers;
5140 if (list_empty(&mddev->disks))
5141 /* cannot run an array with no devices.. */
5146 /* Cannot run until previous stop completes properly */
5147 if (mddev->sysfs_active)
5151 * Analyze all RAID superblock(s)
5153 if (!mddev->raid_disks) {
5154 if (!mddev->persistent)
5159 if (mddev->level != LEVEL_NONE)
5160 request_module("md-level-%d", mddev->level);
5161 else if (mddev->clevel[0])
5162 request_module("md-%s", mddev->clevel);
5165 * Drop all container device buffers, from now on
5166 * the only valid external interface is through the md
5169 rdev_for_each(rdev, mddev) {
5170 if (test_bit(Faulty, &rdev->flags))
5172 sync_blockdev(rdev->bdev);
5173 invalidate_bdev(rdev->bdev);
5175 /* perform some consistency tests on the device.
5176 * We don't want the data to overlap the metadata,
5177 * Internal Bitmap issues have been handled elsewhere.
5179 if (rdev->meta_bdev) {
5180 /* Nothing to check */;
5181 } else if (rdev->data_offset < rdev->sb_start) {
5182 if (mddev->dev_sectors &&
5183 rdev->data_offset + mddev->dev_sectors
5185 printk("md: %s: data overlaps metadata\n",
5190 if (rdev->sb_start + rdev->sb_size/512
5191 > rdev->data_offset) {
5192 printk("md: %s: metadata overlaps data\n",
5197 sysfs_notify_dirent_safe(rdev->sysfs_state);
5200 if (mddev->bio_set == NULL)
5201 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5203 spin_lock(&pers_lock);
5204 pers = find_pers(mddev->level, mddev->clevel);
5205 if (!pers || !try_module_get(pers->owner)) {
5206 spin_unlock(&pers_lock);
5207 if (mddev->level != LEVEL_NONE)
5208 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5211 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5215 spin_unlock(&pers_lock);
5216 if (mddev->level != pers->level) {
5217 mddev->level = pers->level;
5218 mddev->new_level = pers->level;
5220 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5222 if (mddev->reshape_position != MaxSector &&
5223 pers->start_reshape == NULL) {
5224 /* This personality cannot handle reshaping... */
5225 module_put(pers->owner);
5229 if (pers->sync_request) {
5230 /* Warn if this is a potentially silly
5233 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5234 struct md_rdev *rdev2;
5237 rdev_for_each(rdev, mddev)
5238 rdev_for_each(rdev2, mddev) {
5240 rdev->bdev->bd_contains ==
5241 rdev2->bdev->bd_contains) {
5243 "%s: WARNING: %s appears to be"
5244 " on the same physical disk as"
5247 bdevname(rdev->bdev,b),
5248 bdevname(rdev2->bdev,b2));
5255 "True protection against single-disk"
5256 " failure might be compromised.\n");
5259 mddev->recovery = 0;
5260 /* may be over-ridden by personality */
5261 mddev->resync_max_sectors = mddev->dev_sectors;
5263 mddev->ok_start_degraded = start_dirty_degraded;
5265 if (start_readonly && mddev->ro == 0)
5266 mddev->ro = 2; /* read-only, but switch on first write */
5268 err = pers->run(mddev);
5270 printk(KERN_ERR "md: pers->run() failed ...\n");
5271 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5272 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5273 " but 'external_size' not in effect?\n", __func__);
5275 "md: invalid array_size %llu > default size %llu\n",
5276 (unsigned long long)mddev->array_sectors / 2,
5277 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5280 if (err == 0 && pers->sync_request &&
5281 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5282 struct bitmap *bitmap;
5284 bitmap = bitmap_create(mddev, -1);
5285 if (IS_ERR(bitmap)) {
5286 err = PTR_ERR(bitmap);
5287 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5288 mdname(mddev), err);
5290 mddev->bitmap = bitmap;
5294 mddev_detach(mddev);
5296 pers->free(mddev, mddev->private);
5297 mddev->private = NULL;
5298 module_put(pers->owner);
5299 bitmap_destroy(mddev);
5303 mddev->queue->backing_dev_info.congested_data = mddev;
5304 mddev->queue->backing_dev_info.congested_fn = md_congested;
5306 if (pers->sync_request) {
5307 if (mddev->kobj.sd &&
5308 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5310 "md: cannot register extra attributes for %s\n",
5312 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5313 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5316 atomic_set(&mddev->writes_pending,0);
5317 atomic_set(&mddev->max_corr_read_errors,
5318 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5319 mddev->safemode = 0;
5320 if (mddev_is_clustered(mddev))
5321 mddev->safemode_delay = 0;
5323 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5326 spin_lock(&mddev->lock);
5328 spin_unlock(&mddev->lock);
5329 rdev_for_each(rdev, mddev)
5330 if (rdev->raid_disk >= 0)
5331 if (sysfs_link_rdev(mddev, rdev))
5332 /* failure here is OK */;
5334 if (mddev->degraded && !mddev->ro)
5335 /* This ensures that recovering status is reported immediately
5336 * via sysfs - until a lack of spares is confirmed.
5338 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5339 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5341 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5342 md_update_sb(mddev, 0);
5344 md_new_event(mddev);
5345 sysfs_notify_dirent_safe(mddev->sysfs_state);
5346 sysfs_notify_dirent_safe(mddev->sysfs_action);
5347 sysfs_notify(&mddev->kobj, NULL, "degraded");
5350 EXPORT_SYMBOL_GPL(md_run);
5352 static int do_md_run(struct mddev *mddev)
5356 err = md_run(mddev);
5359 err = bitmap_load(mddev);
5361 bitmap_destroy(mddev);
5365 if (mddev_is_clustered(mddev))
5366 md_allow_write(mddev);
5368 md_wakeup_thread(mddev->thread);
5369 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5371 set_capacity(mddev->gendisk, mddev->array_sectors);
5372 revalidate_disk(mddev->gendisk);
5374 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5379 static int restart_array(struct mddev *mddev)
5381 struct gendisk *disk = mddev->gendisk;
5383 /* Complain if it has no devices */
5384 if (list_empty(&mddev->disks))
5390 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5391 struct md_rdev *rdev;
5392 bool has_journal = false;
5395 rdev_for_each_rcu(rdev, mddev) {
5396 if (test_bit(Journal, &rdev->flags) &&
5397 !test_bit(Faulty, &rdev->flags)) {
5404 /* Don't restart rw with journal missing/faulty */
5409 mddev->safemode = 0;
5411 set_disk_ro(disk, 0);
5412 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5414 /* Kick recovery or resync if necessary */
5415 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5416 md_wakeup_thread(mddev->thread);
5417 md_wakeup_thread(mddev->sync_thread);
5418 sysfs_notify_dirent_safe(mddev->sysfs_state);
5422 static void md_clean(struct mddev *mddev)
5424 mddev->array_sectors = 0;
5425 mddev->external_size = 0;
5426 mddev->dev_sectors = 0;
5427 mddev->raid_disks = 0;
5428 mddev->recovery_cp = 0;
5429 mddev->resync_min = 0;
5430 mddev->resync_max = MaxSector;
5431 mddev->reshape_position = MaxSector;
5432 mddev->external = 0;
5433 mddev->persistent = 0;
5434 mddev->level = LEVEL_NONE;
5435 mddev->clevel[0] = 0;
5438 mddev->metadata_type[0] = 0;
5439 mddev->chunk_sectors = 0;
5440 mddev->ctime = mddev->utime = 0;
5442 mddev->max_disks = 0;
5444 mddev->can_decrease_events = 0;
5445 mddev->delta_disks = 0;
5446 mddev->reshape_backwards = 0;
5447 mddev->new_level = LEVEL_NONE;
5448 mddev->new_layout = 0;
5449 mddev->new_chunk_sectors = 0;
5450 mddev->curr_resync = 0;
5451 atomic64_set(&mddev->resync_mismatches, 0);
5452 mddev->suspend_lo = mddev->suspend_hi = 0;
5453 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5454 mddev->recovery = 0;
5457 mddev->degraded = 0;
5458 mddev->safemode = 0;
5459 mddev->private = NULL;
5460 mddev->bitmap_info.offset = 0;
5461 mddev->bitmap_info.default_offset = 0;
5462 mddev->bitmap_info.default_space = 0;
5463 mddev->bitmap_info.chunksize = 0;
5464 mddev->bitmap_info.daemon_sleep = 0;
5465 mddev->bitmap_info.max_write_behind = 0;
5468 static void __md_stop_writes(struct mddev *mddev)
5470 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5471 flush_workqueue(md_misc_wq);
5472 if (mddev->sync_thread) {
5473 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5474 md_reap_sync_thread(mddev);
5477 del_timer_sync(&mddev->safemode_timer);
5479 bitmap_flush(mddev);
5480 md_super_wait(mddev);
5482 if (mddev->ro == 0 &&
5483 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5484 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5485 /* mark array as shutdown cleanly */
5486 if (!mddev_is_clustered(mddev))
5488 md_update_sb(mddev, 1);
5492 void md_stop_writes(struct mddev *mddev)
5494 mddev_lock_nointr(mddev);
5495 __md_stop_writes(mddev);
5496 mddev_unlock(mddev);
5498 EXPORT_SYMBOL_GPL(md_stop_writes);
5500 static void mddev_detach(struct mddev *mddev)
5502 struct bitmap *bitmap = mddev->bitmap;
5503 /* wait for behind writes to complete */
5504 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5505 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5507 /* need to kick something here to make sure I/O goes? */
5508 wait_event(bitmap->behind_wait,
5509 atomic_read(&bitmap->behind_writes) == 0);
5511 if (mddev->pers && mddev->pers->quiesce) {
5512 mddev->pers->quiesce(mddev, 1);
5513 mddev->pers->quiesce(mddev, 0);
5515 md_unregister_thread(&mddev->thread);
5517 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5520 static void __md_stop(struct mddev *mddev)
5522 struct md_personality *pers = mddev->pers;
5523 mddev_detach(mddev);
5524 /* Ensure ->event_work is done */
5525 flush_workqueue(md_misc_wq);
5526 spin_lock(&mddev->lock);
5528 spin_unlock(&mddev->lock);
5529 pers->free(mddev, mddev->private);
5530 mddev->private = NULL;
5531 if (pers->sync_request && mddev->to_remove == NULL)
5532 mddev->to_remove = &md_redundancy_group;
5533 module_put(pers->owner);
5534 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5537 void md_stop(struct mddev *mddev)
5539 /* stop the array and free an attached data structures.
5540 * This is called from dm-raid
5543 bitmap_destroy(mddev);
5545 bioset_free(mddev->bio_set);
5548 EXPORT_SYMBOL_GPL(md_stop);
5550 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5555 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5557 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5558 md_wakeup_thread(mddev->thread);
5560 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5561 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5562 if (mddev->sync_thread)
5563 /* Thread might be blocked waiting for metadata update
5564 * which will now never happen */
5565 wake_up_process(mddev->sync_thread->tsk);
5567 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5569 mddev_unlock(mddev);
5570 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5572 wait_event(mddev->sb_wait,
5573 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5574 mddev_lock_nointr(mddev);
5576 mutex_lock(&mddev->open_mutex);
5577 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5578 mddev->sync_thread ||
5579 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5580 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5581 printk("md: %s still in use.\n",mdname(mddev));
5583 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5584 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5585 md_wakeup_thread(mddev->thread);
5591 __md_stop_writes(mddev);
5597 set_disk_ro(mddev->gendisk, 1);
5598 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5599 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5600 md_wakeup_thread(mddev->thread);
5601 sysfs_notify_dirent_safe(mddev->sysfs_state);
5605 mutex_unlock(&mddev->open_mutex);
5610 * 0 - completely stop and dis-assemble array
5611 * 2 - stop but do not disassemble array
5613 static int do_md_stop(struct mddev *mddev, int mode,
5614 struct block_device *bdev)
5616 struct gendisk *disk = mddev->gendisk;
5617 struct md_rdev *rdev;
5620 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5622 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5623 md_wakeup_thread(mddev->thread);
5625 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5626 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5627 if (mddev->sync_thread)
5628 /* Thread might be blocked waiting for metadata update
5629 * which will now never happen */
5630 wake_up_process(mddev->sync_thread->tsk);
5632 mddev_unlock(mddev);
5633 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5634 !test_bit(MD_RECOVERY_RUNNING,
5635 &mddev->recovery)));
5636 mddev_lock_nointr(mddev);
5638 mutex_lock(&mddev->open_mutex);
5639 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5640 mddev->sysfs_active ||
5641 mddev->sync_thread ||
5642 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5643 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5644 printk("md: %s still in use.\n",mdname(mddev));
5645 mutex_unlock(&mddev->open_mutex);
5647 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5648 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5649 md_wakeup_thread(mddev->thread);
5655 set_disk_ro(disk, 0);
5657 __md_stop_writes(mddev);
5659 mddev->queue->backing_dev_info.congested_fn = NULL;
5661 /* tell userspace to handle 'inactive' */
5662 sysfs_notify_dirent_safe(mddev->sysfs_state);
5664 rdev_for_each(rdev, mddev)
5665 if (rdev->raid_disk >= 0)
5666 sysfs_unlink_rdev(mddev, rdev);
5668 set_capacity(disk, 0);
5669 mutex_unlock(&mddev->open_mutex);
5671 revalidate_disk(disk);
5676 mutex_unlock(&mddev->open_mutex);
5678 * Free resources if final stop
5681 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5683 bitmap_destroy(mddev);
5684 if (mddev->bitmap_info.file) {
5685 struct file *f = mddev->bitmap_info.file;
5686 spin_lock(&mddev->lock);
5687 mddev->bitmap_info.file = NULL;
5688 spin_unlock(&mddev->lock);
5691 mddev->bitmap_info.offset = 0;
5693 export_array(mddev);
5696 if (mddev->hold_active == UNTIL_STOP)
5697 mddev->hold_active = 0;
5699 md_new_event(mddev);
5700 sysfs_notify_dirent_safe(mddev->sysfs_state);
5705 static void autorun_array(struct mddev *mddev)
5707 struct md_rdev *rdev;
5710 if (list_empty(&mddev->disks))
5713 printk(KERN_INFO "md: running: ");
5715 rdev_for_each(rdev, mddev) {
5716 char b[BDEVNAME_SIZE];
5717 printk("<%s>", bdevname(rdev->bdev,b));
5721 err = do_md_run(mddev);
5723 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5724 do_md_stop(mddev, 0, NULL);
5729 * lets try to run arrays based on all disks that have arrived
5730 * until now. (those are in pending_raid_disks)
5732 * the method: pick the first pending disk, collect all disks with
5733 * the same UUID, remove all from the pending list and put them into
5734 * the 'same_array' list. Then order this list based on superblock
5735 * update time (freshest comes first), kick out 'old' disks and
5736 * compare superblocks. If everything's fine then run it.
5738 * If "unit" is allocated, then bump its reference count
5740 static void autorun_devices(int part)
5742 struct md_rdev *rdev0, *rdev, *tmp;
5743 struct mddev *mddev;
5744 char b[BDEVNAME_SIZE];
5746 printk(KERN_INFO "md: autorun ...\n");
5747 while (!list_empty(&pending_raid_disks)) {
5750 LIST_HEAD(candidates);
5751 rdev0 = list_entry(pending_raid_disks.next,
5752 struct md_rdev, same_set);
5754 printk(KERN_INFO "md: considering %s ...\n",
5755 bdevname(rdev0->bdev,b));
5756 INIT_LIST_HEAD(&candidates);
5757 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5758 if (super_90_load(rdev, rdev0, 0) >= 0) {
5759 printk(KERN_INFO "md: adding %s ...\n",
5760 bdevname(rdev->bdev,b));
5761 list_move(&rdev->same_set, &candidates);
5764 * now we have a set of devices, with all of them having
5765 * mostly sane superblocks. It's time to allocate the
5769 dev = MKDEV(mdp_major,
5770 rdev0->preferred_minor << MdpMinorShift);
5771 unit = MINOR(dev) >> MdpMinorShift;
5773 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5776 if (rdev0->preferred_minor != unit) {
5777 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5778 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5782 md_probe(dev, NULL, NULL);
5783 mddev = mddev_find(dev);
5784 if (!mddev || !mddev->gendisk) {
5788 "md: cannot allocate memory for md drive.\n");
5791 if (mddev_lock(mddev))
5792 printk(KERN_WARNING "md: %s locked, cannot run\n",
5794 else if (mddev->raid_disks || mddev->major_version
5795 || !list_empty(&mddev->disks)) {
5797 "md: %s already running, cannot run %s\n",
5798 mdname(mddev), bdevname(rdev0->bdev,b));
5799 mddev_unlock(mddev);
5801 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5802 mddev->persistent = 1;
5803 rdev_for_each_list(rdev, tmp, &candidates) {
5804 list_del_init(&rdev->same_set);
5805 if (bind_rdev_to_array(rdev, mddev))
5808 autorun_array(mddev);
5809 mddev_unlock(mddev);
5811 /* on success, candidates will be empty, on error
5814 rdev_for_each_list(rdev, tmp, &candidates) {
5815 list_del_init(&rdev->same_set);
5820 printk(KERN_INFO "md: ... autorun DONE.\n");
5822 #endif /* !MODULE */
5824 static int get_version(void __user *arg)
5828 ver.major = MD_MAJOR_VERSION;
5829 ver.minor = MD_MINOR_VERSION;
5830 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5832 if (copy_to_user(arg, &ver, sizeof(ver)))
5838 static int get_array_info(struct mddev *mddev, void __user *arg)
5840 mdu_array_info_t info;
5841 int nr,working,insync,failed,spare;
5842 struct md_rdev *rdev;
5844 nr = working = insync = failed = spare = 0;
5846 rdev_for_each_rcu(rdev, mddev) {
5848 if (test_bit(Faulty, &rdev->flags))
5852 if (test_bit(In_sync, &rdev->flags))
5860 info.major_version = mddev->major_version;
5861 info.minor_version = mddev->minor_version;
5862 info.patch_version = MD_PATCHLEVEL_VERSION;
5863 info.ctime = clamp_t(time64_t, mddev->ctime, 0, U32_MAX);
5864 info.level = mddev->level;
5865 info.size = mddev->dev_sectors / 2;
5866 if (info.size != mddev->dev_sectors / 2) /* overflow */
5869 info.raid_disks = mddev->raid_disks;
5870 info.md_minor = mddev->md_minor;
5871 info.not_persistent= !mddev->persistent;
5873 info.utime = clamp_t(time64_t, mddev->utime, 0, U32_MAX);
5876 info.state = (1<<MD_SB_CLEAN);
5877 if (mddev->bitmap && mddev->bitmap_info.offset)
5878 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5879 if (mddev_is_clustered(mddev))
5880 info.state |= (1<<MD_SB_CLUSTERED);
5881 info.active_disks = insync;
5882 info.working_disks = working;
5883 info.failed_disks = failed;
5884 info.spare_disks = spare;
5886 info.layout = mddev->layout;
5887 info.chunk_size = mddev->chunk_sectors << 9;
5889 if (copy_to_user(arg, &info, sizeof(info)))
5895 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5897 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5901 file = kzalloc(sizeof(*file), GFP_NOIO);
5906 spin_lock(&mddev->lock);
5907 /* bitmap enabled */
5908 if (mddev->bitmap_info.file) {
5909 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5910 sizeof(file->pathname));
5914 memmove(file->pathname, ptr,
5915 sizeof(file->pathname)-(ptr-file->pathname));
5917 spin_unlock(&mddev->lock);
5920 copy_to_user(arg, file, sizeof(*file)))
5927 static int get_disk_info(struct mddev *mddev, void __user * arg)
5929 mdu_disk_info_t info;
5930 struct md_rdev *rdev;
5932 if (copy_from_user(&info, arg, sizeof(info)))
5936 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5938 info.major = MAJOR(rdev->bdev->bd_dev);
5939 info.minor = MINOR(rdev->bdev->bd_dev);
5940 info.raid_disk = rdev->raid_disk;
5942 if (test_bit(Faulty, &rdev->flags))
5943 info.state |= (1<<MD_DISK_FAULTY);
5944 else if (test_bit(In_sync, &rdev->flags)) {
5945 info.state |= (1<<MD_DISK_ACTIVE);
5946 info.state |= (1<<MD_DISK_SYNC);
5948 if (test_bit(Journal, &rdev->flags))
5949 info.state |= (1<<MD_DISK_JOURNAL);
5950 if (test_bit(WriteMostly, &rdev->flags))
5951 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5953 info.major = info.minor = 0;
5954 info.raid_disk = -1;
5955 info.state = (1<<MD_DISK_REMOVED);
5959 if (copy_to_user(arg, &info, sizeof(info)))
5965 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5967 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5968 struct md_rdev *rdev;
5969 dev_t dev = MKDEV(info->major,info->minor);
5971 if (mddev_is_clustered(mddev) &&
5972 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5973 pr_err("%s: Cannot add to clustered mddev.\n",
5978 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5981 if (!mddev->raid_disks) {
5983 /* expecting a device which has a superblock */
5984 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5987 "md: md_import_device returned %ld\n",
5989 return PTR_ERR(rdev);
5991 if (!list_empty(&mddev->disks)) {
5992 struct md_rdev *rdev0
5993 = list_entry(mddev->disks.next,
5994 struct md_rdev, same_set);
5995 err = super_types[mddev->major_version]
5996 .load_super(rdev, rdev0, mddev->minor_version);
5999 "md: %s has different UUID to %s\n",
6000 bdevname(rdev->bdev,b),
6001 bdevname(rdev0->bdev,b2));
6006 err = bind_rdev_to_array(rdev, mddev);
6013 * add_new_disk can be used once the array is assembled
6014 * to add "hot spares". They must already have a superblock
6019 if (!mddev->pers->hot_add_disk) {
6021 "%s: personality does not support diskops!\n",
6025 if (mddev->persistent)
6026 rdev = md_import_device(dev, mddev->major_version,
6027 mddev->minor_version);
6029 rdev = md_import_device(dev, -1, -1);
6032 "md: md_import_device returned %ld\n",
6034 return PTR_ERR(rdev);
6036 /* set saved_raid_disk if appropriate */
6037 if (!mddev->persistent) {
6038 if (info->state & (1<<MD_DISK_SYNC) &&
6039 info->raid_disk < mddev->raid_disks) {
6040 rdev->raid_disk = info->raid_disk;
6041 set_bit(In_sync, &rdev->flags);
6042 clear_bit(Bitmap_sync, &rdev->flags);
6044 rdev->raid_disk = -1;
6045 rdev->saved_raid_disk = rdev->raid_disk;
6047 super_types[mddev->major_version].
6048 validate_super(mddev, rdev);
6049 if ((info->state & (1<<MD_DISK_SYNC)) &&
6050 rdev->raid_disk != info->raid_disk) {
6051 /* This was a hot-add request, but events doesn't
6052 * match, so reject it.
6058 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6059 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6060 set_bit(WriteMostly, &rdev->flags);
6062 clear_bit(WriteMostly, &rdev->flags);
6064 if (info->state & (1<<MD_DISK_JOURNAL)) {
6065 struct md_rdev *rdev2;
6066 bool has_journal = false;
6068 /* make sure no existing journal disk */
6069 rdev_for_each(rdev2, mddev) {
6070 if (test_bit(Journal, &rdev2->flags)) {
6079 set_bit(Journal, &rdev->flags);
6082 * check whether the device shows up in other nodes
6084 if (mddev_is_clustered(mddev)) {
6085 if (info->state & (1 << MD_DISK_CANDIDATE))
6086 set_bit(Candidate, &rdev->flags);
6087 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6088 /* --add initiated by this node */
6089 err = md_cluster_ops->add_new_disk(mddev, rdev);
6097 rdev->raid_disk = -1;
6098 err = bind_rdev_to_array(rdev, mddev);
6103 if (mddev_is_clustered(mddev)) {
6104 if (info->state & (1 << MD_DISK_CANDIDATE))
6105 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6108 md_cluster_ops->add_new_disk_cancel(mddev);
6110 err = add_bound_rdev(rdev);
6114 err = add_bound_rdev(rdev);
6119 /* otherwise, add_new_disk is only allowed
6120 * for major_version==0 superblocks
6122 if (mddev->major_version != 0) {
6123 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6128 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6130 rdev = md_import_device(dev, -1, 0);
6133 "md: error, md_import_device() returned %ld\n",
6135 return PTR_ERR(rdev);
6137 rdev->desc_nr = info->number;
6138 if (info->raid_disk < mddev->raid_disks)
6139 rdev->raid_disk = info->raid_disk;
6141 rdev->raid_disk = -1;
6143 if (rdev->raid_disk < mddev->raid_disks)
6144 if (info->state & (1<<MD_DISK_SYNC))
6145 set_bit(In_sync, &rdev->flags);
6147 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6148 set_bit(WriteMostly, &rdev->flags);
6150 if (!mddev->persistent) {
6151 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6152 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6154 rdev->sb_start = calc_dev_sboffset(rdev);
6155 rdev->sectors = rdev->sb_start;
6157 err = bind_rdev_to_array(rdev, mddev);
6167 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6169 char b[BDEVNAME_SIZE];
6170 struct md_rdev *rdev;
6172 rdev = find_rdev(mddev, dev);
6176 if (rdev->raid_disk < 0)
6179 clear_bit(Blocked, &rdev->flags);
6180 remove_and_add_spares(mddev, rdev);
6182 if (rdev->raid_disk >= 0)
6186 if (mddev_is_clustered(mddev))
6187 md_cluster_ops->remove_disk(mddev, rdev);
6189 md_kick_rdev_from_array(rdev);
6190 md_update_sb(mddev, 1);
6191 md_new_event(mddev);
6195 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6196 bdevname(rdev->bdev,b), mdname(mddev));
6200 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6202 char b[BDEVNAME_SIZE];
6204 struct md_rdev *rdev;
6209 if (mddev->major_version != 0) {
6210 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6211 " version-0 superblocks.\n",
6215 if (!mddev->pers->hot_add_disk) {
6217 "%s: personality does not support diskops!\n",
6222 rdev = md_import_device(dev, -1, 0);
6225 "md: error, md_import_device() returned %ld\n",
6230 if (mddev->persistent)
6231 rdev->sb_start = calc_dev_sboffset(rdev);
6233 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6235 rdev->sectors = rdev->sb_start;
6237 if (test_bit(Faulty, &rdev->flags)) {
6239 "md: can not hot-add faulty %s disk to %s!\n",
6240 bdevname(rdev->bdev,b), mdname(mddev));
6245 clear_bit(In_sync, &rdev->flags);
6247 rdev->saved_raid_disk = -1;
6248 err = bind_rdev_to_array(rdev, mddev);
6253 * The rest should better be atomic, we can have disk failures
6254 * noticed in interrupt contexts ...
6257 rdev->raid_disk = -1;
6259 md_update_sb(mddev, 1);
6261 * Kick recovery, maybe this spare has to be added to the
6262 * array immediately.
6264 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6265 md_wakeup_thread(mddev->thread);
6266 md_new_event(mddev);
6274 static int set_bitmap_file(struct mddev *mddev, int fd)
6279 if (!mddev->pers->quiesce || !mddev->thread)
6281 if (mddev->recovery || mddev->sync_thread)
6283 /* we should be able to change the bitmap.. */
6287 struct inode *inode;
6290 if (mddev->bitmap || mddev->bitmap_info.file)
6291 return -EEXIST; /* cannot add when bitmap is present */
6295 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6300 inode = f->f_mapping->host;
6301 if (!S_ISREG(inode->i_mode)) {
6302 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6305 } else if (!(f->f_mode & FMODE_WRITE)) {
6306 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6309 } else if (atomic_read(&inode->i_writecount) != 1) {
6310 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6318 mddev->bitmap_info.file = f;
6319 mddev->bitmap_info.offset = 0; /* file overrides offset */
6320 } else if (mddev->bitmap == NULL)
6321 return -ENOENT; /* cannot remove what isn't there */
6324 mddev->pers->quiesce(mddev, 1);
6326 struct bitmap *bitmap;
6328 bitmap = bitmap_create(mddev, -1);
6329 if (!IS_ERR(bitmap)) {
6330 mddev->bitmap = bitmap;
6331 err = bitmap_load(mddev);
6333 err = PTR_ERR(bitmap);
6335 if (fd < 0 || err) {
6336 bitmap_destroy(mddev);
6337 fd = -1; /* make sure to put the file */
6339 mddev->pers->quiesce(mddev, 0);
6342 struct file *f = mddev->bitmap_info.file;
6344 spin_lock(&mddev->lock);
6345 mddev->bitmap_info.file = NULL;
6346 spin_unlock(&mddev->lock);
6355 * set_array_info is used two different ways
6356 * The original usage is when creating a new array.
6357 * In this usage, raid_disks is > 0 and it together with
6358 * level, size, not_persistent,layout,chunksize determine the
6359 * shape of the array.
6360 * This will always create an array with a type-0.90.0 superblock.
6361 * The newer usage is when assembling an array.
6362 * In this case raid_disks will be 0, and the major_version field is
6363 * use to determine which style super-blocks are to be found on the devices.
6364 * The minor and patch _version numbers are also kept incase the
6365 * super_block handler wishes to interpret them.
6367 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6370 if (info->raid_disks == 0) {
6371 /* just setting version number for superblock loading */
6372 if (info->major_version < 0 ||
6373 info->major_version >= ARRAY_SIZE(super_types) ||
6374 super_types[info->major_version].name == NULL) {
6375 /* maybe try to auto-load a module? */
6377 "md: superblock version %d not known\n",
6378 info->major_version);
6381 mddev->major_version = info->major_version;
6382 mddev->minor_version = info->minor_version;
6383 mddev->patch_version = info->patch_version;
6384 mddev->persistent = !info->not_persistent;
6385 /* ensure mddev_put doesn't delete this now that there
6386 * is some minimal configuration.
6388 mddev->ctime = ktime_get_real_seconds();
6391 mddev->major_version = MD_MAJOR_VERSION;
6392 mddev->minor_version = MD_MINOR_VERSION;
6393 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6394 mddev->ctime = ktime_get_real_seconds();
6396 mddev->level = info->level;
6397 mddev->clevel[0] = 0;
6398 mddev->dev_sectors = 2 * (sector_t)info->size;
6399 mddev->raid_disks = info->raid_disks;
6400 /* don't set md_minor, it is determined by which /dev/md* was
6403 if (info->state & (1<<MD_SB_CLEAN))
6404 mddev->recovery_cp = MaxSector;
6406 mddev->recovery_cp = 0;
6407 mddev->persistent = ! info->not_persistent;
6408 mddev->external = 0;
6410 mddev->layout = info->layout;
6411 mddev->chunk_sectors = info->chunk_size >> 9;
6413 mddev->max_disks = MD_SB_DISKS;
6415 if (mddev->persistent)
6417 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6419 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6420 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6421 mddev->bitmap_info.offset = 0;
6423 mddev->reshape_position = MaxSector;
6426 * Generate a 128 bit UUID
6428 get_random_bytes(mddev->uuid, 16);
6430 mddev->new_level = mddev->level;
6431 mddev->new_chunk_sectors = mddev->chunk_sectors;
6432 mddev->new_layout = mddev->layout;
6433 mddev->delta_disks = 0;
6434 mddev->reshape_backwards = 0;
6439 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6441 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6443 if (mddev->external_size)
6446 mddev->array_sectors = array_sectors;
6448 EXPORT_SYMBOL(md_set_array_sectors);
6450 static int update_size(struct mddev *mddev, sector_t num_sectors)
6452 struct md_rdev *rdev;
6454 int fit = (num_sectors == 0);
6456 /* cluster raid doesn't support update size */
6457 if (mddev_is_clustered(mddev))
6460 if (mddev->pers->resize == NULL)
6462 /* The "num_sectors" is the number of sectors of each device that
6463 * is used. This can only make sense for arrays with redundancy.
6464 * linear and raid0 always use whatever space is available. We can only
6465 * consider changing this number if no resync or reconstruction is
6466 * happening, and if the new size is acceptable. It must fit before the
6467 * sb_start or, if that is <data_offset, it must fit before the size
6468 * of each device. If num_sectors is zero, we find the largest size
6471 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6477 rdev_for_each(rdev, mddev) {
6478 sector_t avail = rdev->sectors;
6480 if (fit && (num_sectors == 0 || num_sectors > avail))
6481 num_sectors = avail;
6482 if (avail < num_sectors)
6485 rv = mddev->pers->resize(mddev, num_sectors);
6487 revalidate_disk(mddev->gendisk);
6491 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6494 struct md_rdev *rdev;
6495 /* change the number of raid disks */
6496 if (mddev->pers->check_reshape == NULL)
6500 if (raid_disks <= 0 ||
6501 (mddev->max_disks && raid_disks >= mddev->max_disks))
6503 if (mddev->sync_thread ||
6504 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6505 mddev->reshape_position != MaxSector)
6508 rdev_for_each(rdev, mddev) {
6509 if (mddev->raid_disks < raid_disks &&
6510 rdev->data_offset < rdev->new_data_offset)
6512 if (mddev->raid_disks > raid_disks &&
6513 rdev->data_offset > rdev->new_data_offset)
6517 mddev->delta_disks = raid_disks - mddev->raid_disks;
6518 if (mddev->delta_disks < 0)
6519 mddev->reshape_backwards = 1;
6520 else if (mddev->delta_disks > 0)
6521 mddev->reshape_backwards = 0;
6523 rv = mddev->pers->check_reshape(mddev);
6525 mddev->delta_disks = 0;
6526 mddev->reshape_backwards = 0;
6532 * update_array_info is used to change the configuration of an
6534 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6535 * fields in the info are checked against the array.
6536 * Any differences that cannot be handled will cause an error.
6537 * Normally, only one change can be managed at a time.
6539 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6545 /* calculate expected state,ignoring low bits */
6546 if (mddev->bitmap && mddev->bitmap_info.offset)
6547 state |= (1 << MD_SB_BITMAP_PRESENT);
6549 if (mddev->major_version != info->major_version ||
6550 mddev->minor_version != info->minor_version ||
6551 /* mddev->patch_version != info->patch_version || */
6552 mddev->ctime != info->ctime ||
6553 mddev->level != info->level ||
6554 /* mddev->layout != info->layout || */
6555 mddev->persistent != !info->not_persistent ||
6556 mddev->chunk_sectors != info->chunk_size >> 9 ||
6557 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6558 ((state^info->state) & 0xfffffe00)
6561 /* Check there is only one change */
6562 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6564 if (mddev->raid_disks != info->raid_disks)
6566 if (mddev->layout != info->layout)
6568 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6575 if (mddev->layout != info->layout) {
6577 * we don't need to do anything at the md level, the
6578 * personality will take care of it all.
6580 if (mddev->pers->check_reshape == NULL)
6583 mddev->new_layout = info->layout;
6584 rv = mddev->pers->check_reshape(mddev);
6586 mddev->new_layout = mddev->layout;
6590 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6591 rv = update_size(mddev, (sector_t)info->size * 2);
6593 if (mddev->raid_disks != info->raid_disks)
6594 rv = update_raid_disks(mddev, info->raid_disks);
6596 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6597 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6601 if (mddev->recovery || mddev->sync_thread) {
6605 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6606 struct bitmap *bitmap;
6607 /* add the bitmap */
6608 if (mddev->bitmap) {
6612 if (mddev->bitmap_info.default_offset == 0) {
6616 mddev->bitmap_info.offset =
6617 mddev->bitmap_info.default_offset;
6618 mddev->bitmap_info.space =
6619 mddev->bitmap_info.default_space;
6620 mddev->pers->quiesce(mddev, 1);
6621 bitmap = bitmap_create(mddev, -1);
6622 if (!IS_ERR(bitmap)) {
6623 mddev->bitmap = bitmap;
6624 rv = bitmap_load(mddev);
6626 rv = PTR_ERR(bitmap);
6628 bitmap_destroy(mddev);
6629 mddev->pers->quiesce(mddev, 0);
6631 /* remove the bitmap */
6632 if (!mddev->bitmap) {
6636 if (mddev->bitmap->storage.file) {
6640 if (mddev->bitmap_info.nodes) {
6641 /* hold PW on all the bitmap lock */
6642 if (md_cluster_ops->lock_all_bitmaps(mddev) <= 0) {
6643 printk("md: can't change bitmap to none since the"
6644 " array is in use by more than one node\n");
6646 md_cluster_ops->unlock_all_bitmaps(mddev);
6650 mddev->bitmap_info.nodes = 0;
6651 md_cluster_ops->leave(mddev);
6653 mddev->pers->quiesce(mddev, 1);
6654 bitmap_destroy(mddev);
6655 mddev->pers->quiesce(mddev, 0);
6656 mddev->bitmap_info.offset = 0;
6659 md_update_sb(mddev, 1);
6665 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6667 struct md_rdev *rdev;
6670 if (mddev->pers == NULL)
6674 rdev = find_rdev_rcu(mddev, dev);
6678 md_error(mddev, rdev);
6679 if (!test_bit(Faulty, &rdev->flags))
6687 * We have a problem here : there is no easy way to give a CHS
6688 * virtual geometry. We currently pretend that we have a 2 heads
6689 * 4 sectors (with a BIG number of cylinders...). This drives
6690 * dosfs just mad... ;-)
6692 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6694 struct mddev *mddev = bdev->bd_disk->private_data;
6698 geo->cylinders = mddev->array_sectors / 8;
6702 static inline bool md_ioctl_valid(unsigned int cmd)
6707 case GET_ARRAY_INFO:
6708 case GET_BITMAP_FILE:
6711 case HOT_REMOVE_DISK:
6714 case RESTART_ARRAY_RW:
6716 case SET_ARRAY_INFO:
6717 case SET_BITMAP_FILE:
6718 case SET_DISK_FAULTY:
6721 case CLUSTERED_DISK_NACK:
6728 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6729 unsigned int cmd, unsigned long arg)
6732 void __user *argp = (void __user *)arg;
6733 struct mddev *mddev = NULL;
6736 if (!md_ioctl_valid(cmd))
6741 case GET_ARRAY_INFO:
6745 if (!capable(CAP_SYS_ADMIN))
6750 * Commands dealing with the RAID driver but not any
6755 err = get_version(argp);
6761 autostart_arrays(arg);
6768 * Commands creating/starting a new array:
6771 mddev = bdev->bd_disk->private_data;
6778 /* Some actions do not requires the mutex */
6780 case GET_ARRAY_INFO:
6781 if (!mddev->raid_disks && !mddev->external)
6784 err = get_array_info(mddev, argp);
6788 if (!mddev->raid_disks && !mddev->external)
6791 err = get_disk_info(mddev, argp);
6794 case SET_DISK_FAULTY:
6795 err = set_disk_faulty(mddev, new_decode_dev(arg));
6798 case GET_BITMAP_FILE:
6799 err = get_bitmap_file(mddev, argp);
6804 if (cmd == ADD_NEW_DISK)
6805 /* need to ensure md_delayed_delete() has completed */
6806 flush_workqueue(md_misc_wq);
6808 if (cmd == HOT_REMOVE_DISK)
6809 /* need to ensure recovery thread has run */
6810 wait_event_interruptible_timeout(mddev->sb_wait,
6811 !test_bit(MD_RECOVERY_NEEDED,
6813 msecs_to_jiffies(5000));
6814 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6815 /* Need to flush page cache, and ensure no-one else opens
6818 mutex_lock(&mddev->open_mutex);
6819 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6820 mutex_unlock(&mddev->open_mutex);
6824 set_bit(MD_STILL_CLOSED, &mddev->flags);
6825 mutex_unlock(&mddev->open_mutex);
6826 sync_blockdev(bdev);
6828 err = mddev_lock(mddev);
6831 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6836 if (cmd == SET_ARRAY_INFO) {
6837 mdu_array_info_t info;
6839 memset(&info, 0, sizeof(info));
6840 else if (copy_from_user(&info, argp, sizeof(info))) {
6845 err = update_array_info(mddev, &info);
6847 printk(KERN_WARNING "md: couldn't update"
6848 " array info. %d\n", err);
6853 if (!list_empty(&mddev->disks)) {
6855 "md: array %s already has disks!\n",
6860 if (mddev->raid_disks) {
6862 "md: array %s already initialised!\n",
6867 err = set_array_info(mddev, &info);
6869 printk(KERN_WARNING "md: couldn't set"
6870 " array info. %d\n", err);
6877 * Commands querying/configuring an existing array:
6879 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6880 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6881 if ((!mddev->raid_disks && !mddev->external)
6882 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6883 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6884 && cmd != GET_BITMAP_FILE) {
6890 * Commands even a read-only array can execute:
6893 case RESTART_ARRAY_RW:
6894 err = restart_array(mddev);
6898 err = do_md_stop(mddev, 0, bdev);
6902 err = md_set_readonly(mddev, bdev);
6905 case HOT_REMOVE_DISK:
6906 err = hot_remove_disk(mddev, new_decode_dev(arg));
6910 /* We can support ADD_NEW_DISK on read-only arrays
6911 * only if we are re-adding a preexisting device.
6912 * So require mddev->pers and MD_DISK_SYNC.
6915 mdu_disk_info_t info;
6916 if (copy_from_user(&info, argp, sizeof(info)))
6918 else if (!(info.state & (1<<MD_DISK_SYNC)))
6919 /* Need to clear read-only for this */
6922 err = add_new_disk(mddev, &info);
6928 if (get_user(ro, (int __user *)(arg))) {
6934 /* if the bdev is going readonly the value of mddev->ro
6935 * does not matter, no writes are coming
6940 /* are we are already prepared for writes? */
6944 /* transitioning to readauto need only happen for
6945 * arrays that call md_write_start
6948 err = restart_array(mddev);
6951 set_disk_ro(mddev->gendisk, 0);
6958 * The remaining ioctls are changing the state of the
6959 * superblock, so we do not allow them on read-only arrays.
6961 if (mddev->ro && mddev->pers) {
6962 if (mddev->ro == 2) {
6964 sysfs_notify_dirent_safe(mddev->sysfs_state);
6965 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6966 /* mddev_unlock will wake thread */
6967 /* If a device failed while we were read-only, we
6968 * need to make sure the metadata is updated now.
6970 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6971 mddev_unlock(mddev);
6972 wait_event(mddev->sb_wait,
6973 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6974 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6975 mddev_lock_nointr(mddev);
6986 mdu_disk_info_t info;
6987 if (copy_from_user(&info, argp, sizeof(info)))
6990 err = add_new_disk(mddev, &info);
6994 case CLUSTERED_DISK_NACK:
6995 if (mddev_is_clustered(mddev))
6996 md_cluster_ops->new_disk_ack(mddev, false);
7002 err = hot_add_disk(mddev, new_decode_dev(arg));
7006 err = do_md_run(mddev);
7009 case SET_BITMAP_FILE:
7010 err = set_bitmap_file(mddev, (int)arg);
7019 if (mddev->hold_active == UNTIL_IOCTL &&
7021 mddev->hold_active = 0;
7022 mddev_unlock(mddev);
7026 #ifdef CONFIG_COMPAT
7027 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
7028 unsigned int cmd, unsigned long arg)
7031 case HOT_REMOVE_DISK:
7033 case SET_DISK_FAULTY:
7034 case SET_BITMAP_FILE:
7035 /* These take in integer arg, do not convert */
7038 arg = (unsigned long)compat_ptr(arg);
7042 return md_ioctl(bdev, mode, cmd, arg);
7044 #endif /* CONFIG_COMPAT */
7046 static int md_open(struct block_device *bdev, fmode_t mode)
7049 * Succeed if we can lock the mddev, which confirms that
7050 * it isn't being stopped right now.
7052 struct mddev *mddev = mddev_find(bdev->bd_dev);
7058 if (mddev->gendisk != bdev->bd_disk) {
7059 /* we are racing with mddev_put which is discarding this
7063 /* Wait until bdev->bd_disk is definitely gone */
7064 flush_workqueue(md_misc_wq);
7065 /* Then retry the open from the top */
7066 return -ERESTARTSYS;
7068 BUG_ON(mddev != bdev->bd_disk->private_data);
7070 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7074 atomic_inc(&mddev->openers);
7075 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7076 mutex_unlock(&mddev->open_mutex);
7078 check_disk_change(bdev);
7083 static void md_release(struct gendisk *disk, fmode_t mode)
7085 struct mddev *mddev = disk->private_data;
7088 atomic_dec(&mddev->openers);
7092 static int md_media_changed(struct gendisk *disk)
7094 struct mddev *mddev = disk->private_data;
7096 return mddev->changed;
7099 static int md_revalidate(struct gendisk *disk)
7101 struct mddev *mddev = disk->private_data;
7106 static const struct block_device_operations md_fops =
7108 .owner = THIS_MODULE,
7110 .release = md_release,
7112 #ifdef CONFIG_COMPAT
7113 .compat_ioctl = md_compat_ioctl,
7115 .getgeo = md_getgeo,
7116 .media_changed = md_media_changed,
7117 .revalidate_disk= md_revalidate,
7120 static int md_thread(void *arg)
7122 struct md_thread *thread = arg;
7125 * md_thread is a 'system-thread', it's priority should be very
7126 * high. We avoid resource deadlocks individually in each
7127 * raid personality. (RAID5 does preallocation) We also use RR and
7128 * the very same RT priority as kswapd, thus we will never get
7129 * into a priority inversion deadlock.
7131 * we definitely have to have equal or higher priority than
7132 * bdflush, otherwise bdflush will deadlock if there are too
7133 * many dirty RAID5 blocks.
7136 allow_signal(SIGKILL);
7137 while (!kthread_should_stop()) {
7139 /* We need to wait INTERRUPTIBLE so that
7140 * we don't add to the load-average.
7141 * That means we need to be sure no signals are
7144 if (signal_pending(current))
7145 flush_signals(current);
7147 wait_event_interruptible_timeout
7149 test_bit(THREAD_WAKEUP, &thread->flags)
7150 || kthread_should_stop(),
7153 clear_bit(THREAD_WAKEUP, &thread->flags);
7154 if (!kthread_should_stop())
7155 thread->run(thread);
7161 void md_wakeup_thread(struct md_thread *thread)
7164 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7165 set_bit(THREAD_WAKEUP, &thread->flags);
7166 wake_up(&thread->wqueue);
7169 EXPORT_SYMBOL(md_wakeup_thread);
7171 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7172 struct mddev *mddev, const char *name)
7174 struct md_thread *thread;
7176 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7180 init_waitqueue_head(&thread->wqueue);
7183 thread->mddev = mddev;
7184 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7185 thread->tsk = kthread_run(md_thread, thread,
7187 mdname(thread->mddev),
7189 if (IS_ERR(thread->tsk)) {
7195 EXPORT_SYMBOL(md_register_thread);
7197 void md_unregister_thread(struct md_thread **threadp)
7199 struct md_thread *thread = *threadp;
7202 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7203 /* Locking ensures that mddev_unlock does not wake_up a
7204 * non-existent thread
7206 spin_lock(&pers_lock);
7208 spin_unlock(&pers_lock);
7210 kthread_stop(thread->tsk);
7213 EXPORT_SYMBOL(md_unregister_thread);
7215 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7217 if (!rdev || test_bit(Faulty, &rdev->flags))
7220 if (!mddev->pers || !mddev->pers->error_handler)
7222 mddev->pers->error_handler(mddev,rdev);
7223 if (mddev->degraded)
7224 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7225 sysfs_notify_dirent_safe(rdev->sysfs_state);
7226 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7228 md_wakeup_thread(mddev->thread);
7229 if (mddev->event_work.func)
7230 queue_work(md_misc_wq, &mddev->event_work);
7231 md_new_event(mddev);
7233 EXPORT_SYMBOL(md_error);
7235 /* seq_file implementation /proc/mdstat */
7237 static void status_unused(struct seq_file *seq)
7240 struct md_rdev *rdev;
7242 seq_printf(seq, "unused devices: ");
7244 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7245 char b[BDEVNAME_SIZE];
7247 seq_printf(seq, "%s ",
7248 bdevname(rdev->bdev,b));
7251 seq_printf(seq, "<none>");
7253 seq_printf(seq, "\n");
7256 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7258 sector_t max_sectors, resync, res;
7259 unsigned long dt, db;
7262 unsigned int per_milli;
7264 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7265 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7266 max_sectors = mddev->resync_max_sectors;
7268 max_sectors = mddev->dev_sectors;
7270 resync = mddev->curr_resync;
7272 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7273 /* Still cleaning up */
7274 resync = max_sectors;
7276 resync -= atomic_read(&mddev->recovery_active);
7279 if (mddev->recovery_cp < MaxSector) {
7280 seq_printf(seq, "\tresync=PENDING");
7286 seq_printf(seq, "\tresync=DELAYED");
7290 WARN_ON(max_sectors == 0);
7291 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7292 * in a sector_t, and (max_sectors>>scale) will fit in a
7293 * u32, as those are the requirements for sector_div.
7294 * Thus 'scale' must be at least 10
7297 if (sizeof(sector_t) > sizeof(unsigned long)) {
7298 while ( max_sectors/2 > (1ULL<<(scale+32)))
7301 res = (resync>>scale)*1000;
7302 sector_div(res, (u32)((max_sectors>>scale)+1));
7306 int i, x = per_milli/50, y = 20-x;
7307 seq_printf(seq, "[");
7308 for (i = 0; i < x; i++)
7309 seq_printf(seq, "=");
7310 seq_printf(seq, ">");
7311 for (i = 0; i < y; i++)
7312 seq_printf(seq, ".");
7313 seq_printf(seq, "] ");
7315 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7316 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7318 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7320 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7321 "resync" : "recovery"))),
7322 per_milli/10, per_milli % 10,
7323 (unsigned long long) resync/2,
7324 (unsigned long long) max_sectors/2);
7327 * dt: time from mark until now
7328 * db: blocks written from mark until now
7329 * rt: remaining time
7331 * rt is a sector_t, so could be 32bit or 64bit.
7332 * So we divide before multiply in case it is 32bit and close
7334 * We scale the divisor (db) by 32 to avoid losing precision
7335 * near the end of resync when the number of remaining sectors
7337 * We then divide rt by 32 after multiplying by db to compensate.
7338 * The '+1' avoids division by zero if db is very small.
7340 dt = ((jiffies - mddev->resync_mark) / HZ);
7342 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7343 - mddev->resync_mark_cnt;
7345 rt = max_sectors - resync; /* number of remaining sectors */
7346 sector_div(rt, db/32+1);
7350 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7351 ((unsigned long)rt % 60)/6);
7353 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7357 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7359 struct list_head *tmp;
7361 struct mddev *mddev;
7369 spin_lock(&all_mddevs_lock);
7370 list_for_each(tmp,&all_mddevs)
7372 mddev = list_entry(tmp, struct mddev, all_mddevs);
7374 spin_unlock(&all_mddevs_lock);
7377 spin_unlock(&all_mddevs_lock);
7379 return (void*)2;/* tail */
7383 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7385 struct list_head *tmp;
7386 struct mddev *next_mddev, *mddev = v;
7392 spin_lock(&all_mddevs_lock);
7394 tmp = all_mddevs.next;
7396 tmp = mddev->all_mddevs.next;
7397 if (tmp != &all_mddevs)
7398 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7400 next_mddev = (void*)2;
7403 spin_unlock(&all_mddevs_lock);
7411 static void md_seq_stop(struct seq_file *seq, void *v)
7413 struct mddev *mddev = v;
7415 if (mddev && v != (void*)1 && v != (void*)2)
7419 static int md_seq_show(struct seq_file *seq, void *v)
7421 struct mddev *mddev = v;
7423 struct md_rdev *rdev;
7425 if (v == (void*)1) {
7426 struct md_personality *pers;
7427 seq_printf(seq, "Personalities : ");
7428 spin_lock(&pers_lock);
7429 list_for_each_entry(pers, &pers_list, list)
7430 seq_printf(seq, "[%s] ", pers->name);
7432 spin_unlock(&pers_lock);
7433 seq_printf(seq, "\n");
7434 seq->poll_event = atomic_read(&md_event_count);
7437 if (v == (void*)2) {
7442 spin_lock(&mddev->lock);
7443 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7444 seq_printf(seq, "%s : %sactive", mdname(mddev),
7445 mddev->pers ? "" : "in");
7448 seq_printf(seq, " (read-only)");
7450 seq_printf(seq, " (auto-read-only)");
7451 seq_printf(seq, " %s", mddev->pers->name);
7456 rdev_for_each_rcu(rdev, mddev) {
7457 char b[BDEVNAME_SIZE];
7458 seq_printf(seq, " %s[%d]",
7459 bdevname(rdev->bdev,b), rdev->desc_nr);
7460 if (test_bit(WriteMostly, &rdev->flags))
7461 seq_printf(seq, "(W)");
7462 if (test_bit(Journal, &rdev->flags))
7463 seq_printf(seq, "(J)");
7464 if (test_bit(Faulty, &rdev->flags)) {
7465 seq_printf(seq, "(F)");
7468 if (rdev->raid_disk < 0)
7469 seq_printf(seq, "(S)"); /* spare */
7470 if (test_bit(Replacement, &rdev->flags))
7471 seq_printf(seq, "(R)");
7472 sectors += rdev->sectors;
7476 if (!list_empty(&mddev->disks)) {
7478 seq_printf(seq, "\n %llu blocks",
7479 (unsigned long long)
7480 mddev->array_sectors / 2);
7482 seq_printf(seq, "\n %llu blocks",
7483 (unsigned long long)sectors / 2);
7485 if (mddev->persistent) {
7486 if (mddev->major_version != 0 ||
7487 mddev->minor_version != 90) {
7488 seq_printf(seq," super %d.%d",
7489 mddev->major_version,
7490 mddev->minor_version);
7492 } else if (mddev->external)
7493 seq_printf(seq, " super external:%s",
7494 mddev->metadata_type);
7496 seq_printf(seq, " super non-persistent");
7499 mddev->pers->status(seq, mddev);
7500 seq_printf(seq, "\n ");
7501 if (mddev->pers->sync_request) {
7502 if (status_resync(seq, mddev))
7503 seq_printf(seq, "\n ");
7506 seq_printf(seq, "\n ");
7508 bitmap_status(seq, mddev->bitmap);
7510 seq_printf(seq, "\n");
7512 spin_unlock(&mddev->lock);
7517 static const struct seq_operations md_seq_ops = {
7518 .start = md_seq_start,
7519 .next = md_seq_next,
7520 .stop = md_seq_stop,
7521 .show = md_seq_show,
7524 static int md_seq_open(struct inode *inode, struct file *file)
7526 struct seq_file *seq;
7529 error = seq_open(file, &md_seq_ops);
7533 seq = file->private_data;
7534 seq->poll_event = atomic_read(&md_event_count);
7538 static int md_unloading;
7539 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7541 struct seq_file *seq = filp->private_data;
7545 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7546 poll_wait(filp, &md_event_waiters, wait);
7548 /* always allow read */
7549 mask = POLLIN | POLLRDNORM;
7551 if (seq->poll_event != atomic_read(&md_event_count))
7552 mask |= POLLERR | POLLPRI;
7556 static const struct file_operations md_seq_fops = {
7557 .owner = THIS_MODULE,
7558 .open = md_seq_open,
7560 .llseek = seq_lseek,
7561 .release = seq_release_private,
7562 .poll = mdstat_poll,
7565 int register_md_personality(struct md_personality *p)
7567 printk(KERN_INFO "md: %s personality registered for level %d\n",
7569 spin_lock(&pers_lock);
7570 list_add_tail(&p->list, &pers_list);
7571 spin_unlock(&pers_lock);
7574 EXPORT_SYMBOL(register_md_personality);
7576 int unregister_md_personality(struct md_personality *p)
7578 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7579 spin_lock(&pers_lock);
7580 list_del_init(&p->list);
7581 spin_unlock(&pers_lock);
7584 EXPORT_SYMBOL(unregister_md_personality);
7586 int register_md_cluster_operations(struct md_cluster_operations *ops,
7587 struct module *module)
7590 spin_lock(&pers_lock);
7591 if (md_cluster_ops != NULL)
7594 md_cluster_ops = ops;
7595 md_cluster_mod = module;
7597 spin_unlock(&pers_lock);
7600 EXPORT_SYMBOL(register_md_cluster_operations);
7602 int unregister_md_cluster_operations(void)
7604 spin_lock(&pers_lock);
7605 md_cluster_ops = NULL;
7606 spin_unlock(&pers_lock);
7609 EXPORT_SYMBOL(unregister_md_cluster_operations);
7611 int md_setup_cluster(struct mddev *mddev, int nodes)
7615 err = request_module("md-cluster");
7617 pr_err("md-cluster module not found.\n");
7621 spin_lock(&pers_lock);
7622 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7623 spin_unlock(&pers_lock);
7626 spin_unlock(&pers_lock);
7628 return md_cluster_ops->join(mddev, nodes);
7631 void md_cluster_stop(struct mddev *mddev)
7633 if (!md_cluster_ops)
7635 md_cluster_ops->leave(mddev);
7636 module_put(md_cluster_mod);
7639 static int is_mddev_idle(struct mddev *mddev, int init)
7641 struct md_rdev *rdev;
7647 rdev_for_each_rcu(rdev, mddev) {
7648 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7649 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7650 (int)part_stat_read(&disk->part0, sectors[1]) -
7651 atomic_read(&disk->sync_io);
7652 /* sync IO will cause sync_io to increase before the disk_stats
7653 * as sync_io is counted when a request starts, and
7654 * disk_stats is counted when it completes.
7655 * So resync activity will cause curr_events to be smaller than
7656 * when there was no such activity.
7657 * non-sync IO will cause disk_stat to increase without
7658 * increasing sync_io so curr_events will (eventually)
7659 * be larger than it was before. Once it becomes
7660 * substantially larger, the test below will cause
7661 * the array to appear non-idle, and resync will slow
7663 * If there is a lot of outstanding resync activity when
7664 * we set last_event to curr_events, then all that activity
7665 * completing might cause the array to appear non-idle
7666 * and resync will be slowed down even though there might
7667 * not have been non-resync activity. This will only
7668 * happen once though. 'last_events' will soon reflect
7669 * the state where there is little or no outstanding
7670 * resync requests, and further resync activity will
7671 * always make curr_events less than last_events.
7674 if (init || curr_events - rdev->last_events > 64) {
7675 rdev->last_events = curr_events;
7683 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7685 /* another "blocks" (512byte) blocks have been synced */
7686 atomic_sub(blocks, &mddev->recovery_active);
7687 wake_up(&mddev->recovery_wait);
7689 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7690 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7691 md_wakeup_thread(mddev->thread);
7692 // stop recovery, signal do_sync ....
7695 EXPORT_SYMBOL(md_done_sync);
7697 /* md_write_start(mddev, bi)
7698 * If we need to update some array metadata (e.g. 'active' flag
7699 * in superblock) before writing, schedule a superblock update
7700 * and wait for it to complete.
7702 void md_write_start(struct mddev *mddev, struct bio *bi)
7705 if (bio_data_dir(bi) != WRITE)
7708 BUG_ON(mddev->ro == 1);
7709 if (mddev->ro == 2) {
7710 /* need to switch to read/write */
7712 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7713 md_wakeup_thread(mddev->thread);
7714 md_wakeup_thread(mddev->sync_thread);
7717 atomic_inc(&mddev->writes_pending);
7718 if (mddev->safemode == 1)
7719 mddev->safemode = 0;
7720 if (mddev->in_sync) {
7721 spin_lock(&mddev->lock);
7722 if (mddev->in_sync) {
7724 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7725 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7726 md_wakeup_thread(mddev->thread);
7729 spin_unlock(&mddev->lock);
7732 sysfs_notify_dirent_safe(mddev->sysfs_state);
7733 wait_event(mddev->sb_wait,
7734 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7736 EXPORT_SYMBOL(md_write_start);
7738 void md_write_end(struct mddev *mddev)
7740 if (atomic_dec_and_test(&mddev->writes_pending)) {
7741 if (mddev->safemode == 2)
7742 md_wakeup_thread(mddev->thread);
7743 else if (mddev->safemode_delay)
7744 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7747 EXPORT_SYMBOL(md_write_end);
7749 /* md_allow_write(mddev)
7750 * Calling this ensures that the array is marked 'active' so that writes
7751 * may proceed without blocking. It is important to call this before
7752 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7753 * Must be called with mddev_lock held.
7755 * In the ->external case MD_CHANGE_PENDING can not be cleared until mddev->lock
7756 * is dropped, so return -EAGAIN after notifying userspace.
7758 int md_allow_write(struct mddev *mddev)
7764 if (!mddev->pers->sync_request)
7767 spin_lock(&mddev->lock);
7768 if (mddev->in_sync) {
7770 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7771 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7772 if (mddev->safemode_delay &&
7773 mddev->safemode == 0)
7774 mddev->safemode = 1;
7775 spin_unlock(&mddev->lock);
7776 md_update_sb(mddev, 0);
7777 sysfs_notify_dirent_safe(mddev->sysfs_state);
7779 spin_unlock(&mddev->lock);
7781 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7786 EXPORT_SYMBOL_GPL(md_allow_write);
7788 #define SYNC_MARKS 10
7789 #define SYNC_MARK_STEP (3*HZ)
7790 #define UPDATE_FREQUENCY (5*60*HZ)
7791 void md_do_sync(struct md_thread *thread)
7793 struct mddev *mddev = thread->mddev;
7794 struct mddev *mddev2;
7795 unsigned int currspeed = 0,
7797 sector_t max_sectors,j, io_sectors, recovery_done;
7798 unsigned long mark[SYNC_MARKS];
7799 unsigned long update_time;
7800 sector_t mark_cnt[SYNC_MARKS];
7802 struct list_head *tmp;
7803 sector_t last_check;
7805 struct md_rdev *rdev;
7806 char *desc, *action = NULL;
7807 struct blk_plug plug;
7810 /* just incase thread restarts... */
7811 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7813 if (mddev->ro) {/* never try to sync a read-only array */
7814 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7818 if (mddev_is_clustered(mddev)) {
7819 ret = md_cluster_ops->resync_start(mddev);
7823 set_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags);
7824 if (!(test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7825 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) ||
7826 test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
7827 && ((unsigned long long)mddev->curr_resync_completed
7828 < (unsigned long long)mddev->resync_max_sectors))
7832 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7833 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7834 desc = "data-check";
7836 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7837 desc = "requested-resync";
7841 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7846 mddev->last_sync_action = action ?: desc;
7848 /* we overload curr_resync somewhat here.
7849 * 0 == not engaged in resync at all
7850 * 2 == checking that there is no conflict with another sync
7851 * 1 == like 2, but have yielded to allow conflicting resync to
7853 * other == active in resync - this many blocks
7855 * Before starting a resync we must have set curr_resync to
7856 * 2, and then checked that every "conflicting" array has curr_resync
7857 * less than ours. When we find one that is the same or higher
7858 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7859 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7860 * This will mean we have to start checking from the beginning again.
7865 mddev->curr_resync = 2;
7868 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7870 for_each_mddev(mddev2, tmp) {
7871 if (mddev2 == mddev)
7873 if (!mddev->parallel_resync
7874 && mddev2->curr_resync
7875 && match_mddev_units(mddev, mddev2)) {
7877 if (mddev < mddev2 && mddev->curr_resync == 2) {
7878 /* arbitrarily yield */
7879 mddev->curr_resync = 1;
7880 wake_up(&resync_wait);
7882 if (mddev > mddev2 && mddev->curr_resync == 1)
7883 /* no need to wait here, we can wait the next
7884 * time 'round when curr_resync == 2
7887 /* We need to wait 'interruptible' so as not to
7888 * contribute to the load average, and not to
7889 * be caught by 'softlockup'
7891 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7892 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7893 mddev2->curr_resync >= mddev->curr_resync) {
7894 printk(KERN_INFO "md: delaying %s of %s"
7895 " until %s has finished (they"
7896 " share one or more physical units)\n",
7897 desc, mdname(mddev), mdname(mddev2));
7899 if (signal_pending(current))
7900 flush_signals(current);
7902 finish_wait(&resync_wait, &wq);
7905 finish_wait(&resync_wait, &wq);
7908 } while (mddev->curr_resync < 2);
7911 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7912 /* resync follows the size requested by the personality,
7913 * which defaults to physical size, but can be virtual size
7915 max_sectors = mddev->resync_max_sectors;
7916 atomic64_set(&mddev->resync_mismatches, 0);
7917 /* we don't use the checkpoint if there's a bitmap */
7918 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7919 j = mddev->resync_min;
7920 else if (!mddev->bitmap)
7921 j = mddev->recovery_cp;
7923 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7924 max_sectors = mddev->resync_max_sectors;
7926 /* recovery follows the physical size of devices */
7927 max_sectors = mddev->dev_sectors;
7930 rdev_for_each_rcu(rdev, mddev)
7931 if (rdev->raid_disk >= 0 &&
7932 !test_bit(Journal, &rdev->flags) &&
7933 !test_bit(Faulty, &rdev->flags) &&
7934 !test_bit(In_sync, &rdev->flags) &&
7935 rdev->recovery_offset < j)
7936 j = rdev->recovery_offset;
7939 /* If there is a bitmap, we need to make sure all
7940 * writes that started before we added a spare
7941 * complete before we start doing a recovery.
7942 * Otherwise the write might complete and (via
7943 * bitmap_endwrite) set a bit in the bitmap after the
7944 * recovery has checked that bit and skipped that
7947 if (mddev->bitmap) {
7948 mddev->pers->quiesce(mddev, 1);
7949 mddev->pers->quiesce(mddev, 0);
7953 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7954 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7955 " %d KB/sec/disk.\n", speed_min(mddev));
7956 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7957 "(but not more than %d KB/sec) for %s.\n",
7958 speed_max(mddev), desc);
7960 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7963 for (m = 0; m < SYNC_MARKS; m++) {
7965 mark_cnt[m] = io_sectors;
7968 mddev->resync_mark = mark[last_mark];
7969 mddev->resync_mark_cnt = mark_cnt[last_mark];
7972 * Tune reconstruction:
7974 window = 32*(PAGE_SIZE/512);
7975 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7976 window/2, (unsigned long long)max_sectors/2);
7978 atomic_set(&mddev->recovery_active, 0);
7983 "md: resuming %s of %s from checkpoint.\n",
7984 desc, mdname(mddev));
7985 mddev->curr_resync = j;
7987 mddev->curr_resync = 3; /* no longer delayed */
7988 mddev->curr_resync_completed = j;
7989 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7990 md_new_event(mddev);
7991 update_time = jiffies;
7993 blk_start_plug(&plug);
7994 while (j < max_sectors) {
7999 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8000 ((mddev->curr_resync > mddev->curr_resync_completed &&
8001 (mddev->curr_resync - mddev->curr_resync_completed)
8002 > (max_sectors >> 4)) ||
8003 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
8004 (j - mddev->curr_resync_completed)*2
8005 >= mddev->resync_max - mddev->curr_resync_completed ||
8006 mddev->curr_resync_completed > mddev->resync_max
8008 /* time to update curr_resync_completed */
8009 wait_event(mddev->recovery_wait,
8010 atomic_read(&mddev->recovery_active) == 0);
8011 mddev->curr_resync_completed = j;
8012 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
8013 j > mddev->recovery_cp)
8014 mddev->recovery_cp = j;
8015 update_time = jiffies;
8016 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8017 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8020 while (j >= mddev->resync_max &&
8021 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8022 /* As this condition is controlled by user-space,
8023 * we can block indefinitely, so use '_interruptible'
8024 * to avoid triggering warnings.
8026 flush_signals(current); /* just in case */
8027 wait_event_interruptible(mddev->recovery_wait,
8028 mddev->resync_max > j
8029 || test_bit(MD_RECOVERY_INTR,
8033 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8036 sectors = mddev->pers->sync_request(mddev, j, &skipped);
8038 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8042 if (!skipped) { /* actual IO requested */
8043 io_sectors += sectors;
8044 atomic_add(sectors, &mddev->recovery_active);
8047 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8051 if (j > max_sectors)
8052 /* when skipping, extra large numbers can be returned. */
8055 mddev->curr_resync = j;
8056 mddev->curr_mark_cnt = io_sectors;
8057 if (last_check == 0)
8058 /* this is the earliest that rebuild will be
8059 * visible in /proc/mdstat
8061 md_new_event(mddev);
8063 if (last_check + window > io_sectors || j == max_sectors)
8066 last_check = io_sectors;
8068 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8070 int next = (last_mark+1) % SYNC_MARKS;
8072 mddev->resync_mark = mark[next];
8073 mddev->resync_mark_cnt = mark_cnt[next];
8074 mark[next] = jiffies;
8075 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8079 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8083 * this loop exits only if either when we are slower than
8084 * the 'hard' speed limit, or the system was IO-idle for
8086 * the system might be non-idle CPU-wise, but we only care
8087 * about not overloading the IO subsystem. (things like an
8088 * e2fsck being done on the RAID array should execute fast)
8092 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8093 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8094 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8096 if (currspeed > speed_min(mddev)) {
8097 if (currspeed > speed_max(mddev)) {
8101 if (!is_mddev_idle(mddev, 0)) {
8103 * Give other IO more of a chance.
8104 * The faster the devices, the less we wait.
8106 wait_event(mddev->recovery_wait,
8107 !atomic_read(&mddev->recovery_active));
8111 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8112 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8113 ? "interrupted" : "done");
8115 * this also signals 'finished resyncing' to md_stop
8117 blk_finish_plug(&plug);
8118 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8120 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8121 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8122 mddev->curr_resync > 2) {
8123 mddev->curr_resync_completed = mddev->curr_resync;
8124 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8126 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8128 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8129 mddev->curr_resync > 2) {
8130 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8131 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8132 if (mddev->curr_resync >= mddev->recovery_cp) {
8134 "md: checkpointing %s of %s.\n",
8135 desc, mdname(mddev));
8136 if (test_bit(MD_RECOVERY_ERROR,
8138 mddev->recovery_cp =
8139 mddev->curr_resync_completed;
8141 mddev->recovery_cp =
8145 mddev->recovery_cp = MaxSector;
8147 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8148 mddev->curr_resync = MaxSector;
8150 rdev_for_each_rcu(rdev, mddev)
8151 if (rdev->raid_disk >= 0 &&
8152 mddev->delta_disks >= 0 &&
8153 !test_bit(Journal, &rdev->flags) &&
8154 !test_bit(Faulty, &rdev->flags) &&
8155 !test_bit(In_sync, &rdev->flags) &&
8156 rdev->recovery_offset < mddev->curr_resync)
8157 rdev->recovery_offset = mddev->curr_resync;
8162 /* set CHANGE_PENDING here since maybe another update is needed,
8163 * so other nodes are informed. It should be harmless for normal
8165 set_mask_bits(&mddev->flags, 0,
8166 BIT(MD_CHANGE_PENDING) | BIT(MD_CHANGE_DEVS));
8168 spin_lock(&mddev->lock);
8169 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8170 /* We completed so min/max setting can be forgotten if used. */
8171 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8172 mddev->resync_min = 0;
8173 mddev->resync_max = MaxSector;
8174 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8175 mddev->resync_min = mddev->curr_resync_completed;
8176 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8177 mddev->curr_resync = 0;
8178 spin_unlock(&mddev->lock);
8180 wake_up(&resync_wait);
8181 md_wakeup_thread(mddev->thread);
8184 EXPORT_SYMBOL_GPL(md_do_sync);
8186 static int remove_and_add_spares(struct mddev *mddev,
8187 struct md_rdev *this)
8189 struct md_rdev *rdev;
8192 bool remove_some = false;
8194 rdev_for_each(rdev, mddev) {
8195 if ((this == NULL || rdev == this) &&
8196 rdev->raid_disk >= 0 &&
8197 !test_bit(Blocked, &rdev->flags) &&
8198 test_bit(Faulty, &rdev->flags) &&
8199 atomic_read(&rdev->nr_pending)==0) {
8200 /* Faulty non-Blocked devices with nr_pending == 0
8201 * never get nr_pending incremented,
8202 * never get Faulty cleared, and never get Blocked set.
8203 * So we can synchronize_rcu now rather than once per device
8206 set_bit(RemoveSynchronized, &rdev->flags);
8212 rdev_for_each(rdev, mddev) {
8213 if ((this == NULL || rdev == this) &&
8214 rdev->raid_disk >= 0 &&
8215 !test_bit(Blocked, &rdev->flags) &&
8216 ((test_bit(RemoveSynchronized, &rdev->flags) ||
8217 (!test_bit(In_sync, &rdev->flags) &&
8218 !test_bit(Journal, &rdev->flags))) &&
8219 atomic_read(&rdev->nr_pending)==0)) {
8220 if (mddev->pers->hot_remove_disk(
8221 mddev, rdev) == 0) {
8222 sysfs_unlink_rdev(mddev, rdev);
8223 rdev->raid_disk = -1;
8227 if (remove_some && test_bit(RemoveSynchronized, &rdev->flags))
8228 clear_bit(RemoveSynchronized, &rdev->flags);
8231 if (removed && mddev->kobj.sd)
8232 sysfs_notify(&mddev->kobj, NULL, "degraded");
8234 if (this && removed)
8237 rdev_for_each(rdev, mddev) {
8238 if (this && this != rdev)
8240 if (test_bit(Candidate, &rdev->flags))
8242 if (rdev->raid_disk >= 0 &&
8243 !test_bit(In_sync, &rdev->flags) &&
8244 !test_bit(Journal, &rdev->flags) &&
8245 !test_bit(Faulty, &rdev->flags))
8247 if (rdev->raid_disk >= 0)
8249 if (test_bit(Faulty, &rdev->flags))
8251 if (!test_bit(Journal, &rdev->flags)) {
8253 ! (rdev->saved_raid_disk >= 0 &&
8254 !test_bit(Bitmap_sync, &rdev->flags)))
8257 rdev->recovery_offset = 0;
8260 hot_add_disk(mddev, rdev) == 0) {
8261 if (sysfs_link_rdev(mddev, rdev))
8262 /* failure here is OK */;
8263 if (!test_bit(Journal, &rdev->flags))
8265 md_new_event(mddev);
8266 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8271 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8275 static void md_start_sync(struct work_struct *ws)
8277 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8280 mddev->sync_thread = md_register_thread(md_do_sync,
8283 if (!mddev->sync_thread) {
8284 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8285 printk(KERN_ERR "%s: could not start resync"
8288 /* leave the spares where they are, it shouldn't hurt */
8289 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8290 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8291 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8292 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8293 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8294 wake_up(&resync_wait);
8295 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8297 if (mddev->sysfs_action)
8298 sysfs_notify_dirent_safe(mddev->sysfs_action);
8300 md_wakeup_thread(mddev->sync_thread);
8301 sysfs_notify_dirent_safe(mddev->sysfs_action);
8302 md_new_event(mddev);
8306 * This routine is regularly called by all per-raid-array threads to
8307 * deal with generic issues like resync and super-block update.
8308 * Raid personalities that don't have a thread (linear/raid0) do not
8309 * need this as they never do any recovery or update the superblock.
8311 * It does not do any resync itself, but rather "forks" off other threads
8312 * to do that as needed.
8313 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8314 * "->recovery" and create a thread at ->sync_thread.
8315 * When the thread finishes it sets MD_RECOVERY_DONE
8316 * and wakeups up this thread which will reap the thread and finish up.
8317 * This thread also removes any faulty devices (with nr_pending == 0).
8319 * The overall approach is:
8320 * 1/ if the superblock needs updating, update it.
8321 * 2/ If a recovery thread is running, don't do anything else.
8322 * 3/ If recovery has finished, clean up, possibly marking spares active.
8323 * 4/ If there are any faulty devices, remove them.
8324 * 5/ If array is degraded, try to add spares devices
8325 * 6/ If array has spares or is not in-sync, start a resync thread.
8327 void md_check_recovery(struct mddev *mddev)
8329 if (mddev->suspended)
8333 bitmap_daemon_work(mddev);
8335 if (signal_pending(current)) {
8336 if (mddev->pers->sync_request && !mddev->external) {
8337 printk(KERN_INFO "md: %s in immediate safe mode\n",
8339 mddev->safemode = 2;
8341 flush_signals(current);
8344 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8347 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8348 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8349 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8350 test_bit(MD_RELOAD_SB, &mddev->flags) ||
8351 (mddev->external == 0 && mddev->safemode == 1) ||
8352 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8353 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8357 if (mddev_trylock(mddev)) {
8361 struct md_rdev *rdev;
8362 if (!mddev->external && mddev->in_sync)
8363 /* 'Blocked' flag not needed as failed devices
8364 * will be recorded if array switched to read/write.
8365 * Leaving it set will prevent the device
8366 * from being removed.
8368 rdev_for_each(rdev, mddev)
8369 clear_bit(Blocked, &rdev->flags);
8370 /* On a read-only array we can:
8371 * - remove failed devices
8372 * - add already-in_sync devices if the array itself
8374 * As we only add devices that are already in-sync,
8375 * we can activate the spares immediately.
8377 remove_and_add_spares(mddev, NULL);
8378 /* There is no thread, but we need to call
8379 * ->spare_active and clear saved_raid_disk
8381 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8382 md_reap_sync_thread(mddev);
8383 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8384 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8385 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8389 if (mddev_is_clustered(mddev)) {
8390 struct md_rdev *rdev;
8391 /* kick the device if another node issued a
8394 rdev_for_each(rdev, mddev) {
8395 if (test_and_clear_bit(ClusterRemove, &rdev->flags) &&
8396 rdev->raid_disk < 0)
8397 md_kick_rdev_from_array(rdev);
8400 if (test_and_clear_bit(MD_RELOAD_SB, &mddev->flags))
8401 md_reload_sb(mddev, mddev->good_device_nr);
8404 if (!mddev->external) {
8406 spin_lock(&mddev->lock);
8407 if (mddev->safemode &&
8408 !atomic_read(&mddev->writes_pending) &&
8410 mddev->recovery_cp == MaxSector) {
8413 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8415 if (mddev->safemode == 1)
8416 mddev->safemode = 0;
8417 spin_unlock(&mddev->lock);
8419 sysfs_notify_dirent_safe(mddev->sysfs_state);
8422 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8423 md_update_sb(mddev, 0);
8425 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8426 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8427 /* resync/recovery still happening */
8428 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8431 if (mddev->sync_thread) {
8432 md_reap_sync_thread(mddev);
8435 /* Set RUNNING before clearing NEEDED to avoid
8436 * any transients in the value of "sync_action".
8438 mddev->curr_resync_completed = 0;
8439 spin_lock(&mddev->lock);
8440 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8441 spin_unlock(&mddev->lock);
8442 /* Clear some bits that don't mean anything, but
8445 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8446 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8448 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8449 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8451 /* no recovery is running.
8452 * remove any failed drives, then
8453 * add spares if possible.
8454 * Spares are also removed and re-added, to allow
8455 * the personality to fail the re-add.
8458 if (mddev->reshape_position != MaxSector) {
8459 if (mddev->pers->check_reshape == NULL ||
8460 mddev->pers->check_reshape(mddev) != 0)
8461 /* Cannot proceed */
8463 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8464 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8465 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8466 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8467 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8468 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8469 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8470 } else if (mddev->recovery_cp < MaxSector) {
8471 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8472 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8473 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8474 /* nothing to be done ... */
8477 if (mddev->pers->sync_request) {
8479 /* We are adding a device or devices to an array
8480 * which has the bitmap stored on all devices.
8481 * So make sure all bitmap pages get written
8483 bitmap_write_all(mddev->bitmap);
8485 INIT_WORK(&mddev->del_work, md_start_sync);
8486 queue_work(md_misc_wq, &mddev->del_work);
8490 if (!mddev->sync_thread) {
8491 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8492 wake_up(&resync_wait);
8493 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8495 if (mddev->sysfs_action)
8496 sysfs_notify_dirent_safe(mddev->sysfs_action);
8499 wake_up(&mddev->sb_wait);
8500 mddev_unlock(mddev);
8503 EXPORT_SYMBOL(md_check_recovery);
8505 void md_reap_sync_thread(struct mddev *mddev)
8507 struct md_rdev *rdev;
8509 /* resync has finished, collect result */
8510 md_unregister_thread(&mddev->sync_thread);
8511 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8512 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8514 /* activate any spares */
8515 if (mddev->pers->spare_active(mddev)) {
8516 sysfs_notify(&mddev->kobj, NULL,
8518 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8521 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8522 mddev->pers->finish_reshape)
8523 mddev->pers->finish_reshape(mddev);
8525 /* If array is no-longer degraded, then any saved_raid_disk
8526 * information must be scrapped.
8528 if (!mddev->degraded)
8529 rdev_for_each(rdev, mddev)
8530 rdev->saved_raid_disk = -1;
8532 md_update_sb(mddev, 1);
8533 /* MD_CHANGE_PENDING should be cleared by md_update_sb, so we can
8534 * call resync_finish here if MD_CLUSTER_RESYNC_LOCKED is set by
8536 if (test_and_clear_bit(MD_CLUSTER_RESYNC_LOCKED, &mddev->flags))
8537 md_cluster_ops->resync_finish(mddev);
8538 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8539 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8540 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8541 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8542 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8543 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8544 wake_up(&resync_wait);
8545 /* flag recovery needed just to double check */
8546 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8547 sysfs_notify_dirent_safe(mddev->sysfs_action);
8548 md_new_event(mddev);
8549 if (mddev->event_work.func)
8550 queue_work(md_misc_wq, &mddev->event_work);
8552 EXPORT_SYMBOL(md_reap_sync_thread);
8554 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8556 sysfs_notify_dirent_safe(rdev->sysfs_state);
8557 wait_event_timeout(rdev->blocked_wait,
8558 !test_bit(Blocked, &rdev->flags) &&
8559 !test_bit(BlockedBadBlocks, &rdev->flags),
8560 msecs_to_jiffies(5000));
8561 rdev_dec_pending(rdev, mddev);
8563 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8565 void md_finish_reshape(struct mddev *mddev)
8567 /* called be personality module when reshape completes. */
8568 struct md_rdev *rdev;
8570 rdev_for_each(rdev, mddev) {
8571 if (rdev->data_offset > rdev->new_data_offset)
8572 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8574 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8575 rdev->data_offset = rdev->new_data_offset;
8578 EXPORT_SYMBOL(md_finish_reshape);
8580 /* Bad block management */
8582 /* Returns 1 on success, 0 on failure */
8583 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8586 struct mddev *mddev = rdev->mddev;
8589 s += rdev->new_data_offset;
8591 s += rdev->data_offset;
8592 rv = badblocks_set(&rdev->badblocks, s, sectors, 0);
8594 /* Make sure they get written out promptly */
8595 sysfs_notify_dirent_safe(rdev->sysfs_state);
8596 set_mask_bits(&mddev->flags, 0,
8597 BIT(MD_CHANGE_CLEAN) | BIT(MD_CHANGE_PENDING));
8598 md_wakeup_thread(rdev->mddev->thread);
8603 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8605 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8609 s += rdev->new_data_offset;
8611 s += rdev->data_offset;
8612 return badblocks_clear(&rdev->badblocks,
8615 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8617 static int md_notify_reboot(struct notifier_block *this,
8618 unsigned long code, void *x)
8620 struct list_head *tmp;
8621 struct mddev *mddev;
8624 for_each_mddev(mddev, tmp) {
8625 if (mddev_trylock(mddev)) {
8627 __md_stop_writes(mddev);
8628 if (mddev->persistent)
8629 mddev->safemode = 2;
8630 mddev_unlock(mddev);
8635 * certain more exotic SCSI devices are known to be
8636 * volatile wrt too early system reboots. While the
8637 * right place to handle this issue is the given
8638 * driver, we do want to have a safe RAID driver ...
8646 static struct notifier_block md_notifier = {
8647 .notifier_call = md_notify_reboot,
8649 .priority = INT_MAX, /* before any real devices */
8652 static void md_geninit(void)
8654 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8656 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8659 static int __init md_init(void)
8663 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8667 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8671 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8674 if ((ret = register_blkdev(0, "mdp")) < 0)
8678 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8679 md_probe, NULL, NULL);
8680 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8681 md_probe, NULL, NULL);
8683 register_reboot_notifier(&md_notifier);
8684 raid_table_header = register_sysctl_table(raid_root_table);
8690 unregister_blkdev(MD_MAJOR, "md");
8692 destroy_workqueue(md_misc_wq);
8694 destroy_workqueue(md_wq);
8699 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
8701 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8702 struct md_rdev *rdev2;
8704 char b[BDEVNAME_SIZE];
8706 /* Check for change of roles in the active devices */
8707 rdev_for_each(rdev2, mddev) {
8708 if (test_bit(Faulty, &rdev2->flags))
8711 /* Check if the roles changed */
8712 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
8714 if (test_bit(Candidate, &rdev2->flags)) {
8715 if (role == 0xfffe) {
8716 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
8717 md_kick_rdev_from_array(rdev2);
8721 clear_bit(Candidate, &rdev2->flags);
8724 if (role != rdev2->raid_disk) {
8726 if (rdev2->raid_disk == -1 && role != 0xffff) {
8727 rdev2->saved_raid_disk = role;
8728 ret = remove_and_add_spares(mddev, rdev2);
8729 pr_info("Activated spare: %s\n",
8730 bdevname(rdev2->bdev,b));
8731 /* wakeup mddev->thread here, so array could
8732 * perform resync with the new activated disk */
8733 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8734 md_wakeup_thread(mddev->thread);
8738 * We just want to do the minimum to mark the disk
8739 * as faulty. The recovery is performed by the
8740 * one who initiated the error.
8742 if ((role == 0xfffe) || (role == 0xfffd)) {
8743 md_error(mddev, rdev2);
8744 clear_bit(Blocked, &rdev2->flags);
8749 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
8750 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
8752 /* Finally set the event to be up to date */
8753 mddev->events = le64_to_cpu(sb->events);
8756 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
8759 struct page *swapout = rdev->sb_page;
8760 struct mdp_superblock_1 *sb;
8762 /* Store the sb page of the rdev in the swapout temporary
8763 * variable in case we err in the future
8765 rdev->sb_page = NULL;
8766 alloc_disk_sb(rdev);
8767 ClearPageUptodate(rdev->sb_page);
8768 rdev->sb_loaded = 0;
8769 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
8772 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
8773 __func__, __LINE__, rdev->desc_nr, err);
8774 put_page(rdev->sb_page);
8775 rdev->sb_page = swapout;
8776 rdev->sb_loaded = 1;
8780 sb = page_address(rdev->sb_page);
8781 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
8785 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
8786 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
8788 /* The other node finished recovery, call spare_active to set
8789 * device In_sync and mddev->degraded
8791 if (rdev->recovery_offset == MaxSector &&
8792 !test_bit(In_sync, &rdev->flags) &&
8793 mddev->pers->spare_active(mddev))
8794 sysfs_notify(&mddev->kobj, NULL, "degraded");
8800 void md_reload_sb(struct mddev *mddev, int nr)
8802 struct md_rdev *rdev;
8806 rdev_for_each_rcu(rdev, mddev) {
8807 if (rdev->desc_nr == nr)
8811 if (!rdev || rdev->desc_nr != nr) {
8812 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
8816 err = read_rdev(mddev, rdev);
8820 check_sb_changes(mddev, rdev);
8822 /* Read all rdev's to update recovery_offset */
8823 rdev_for_each_rcu(rdev, mddev)
8824 read_rdev(mddev, rdev);
8826 EXPORT_SYMBOL(md_reload_sb);
8831 * Searches all registered partitions for autorun RAID arrays
8835 static DEFINE_MUTEX(detected_devices_mutex);
8836 static LIST_HEAD(all_detected_devices);
8837 struct detected_devices_node {
8838 struct list_head list;
8842 void md_autodetect_dev(dev_t dev)
8844 struct detected_devices_node *node_detected_dev;
8846 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8847 if (node_detected_dev) {
8848 node_detected_dev->dev = dev;
8849 mutex_lock(&detected_devices_mutex);
8850 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8851 mutex_unlock(&detected_devices_mutex);
8853 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8854 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8858 static void autostart_arrays(int part)
8860 struct md_rdev *rdev;
8861 struct detected_devices_node *node_detected_dev;
8863 int i_scanned, i_passed;
8868 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8870 mutex_lock(&detected_devices_mutex);
8871 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8873 node_detected_dev = list_entry(all_detected_devices.next,
8874 struct detected_devices_node, list);
8875 list_del(&node_detected_dev->list);
8876 dev = node_detected_dev->dev;
8877 kfree(node_detected_dev);
8878 rdev = md_import_device(dev,0, 90);
8882 if (test_bit(Faulty, &rdev->flags))
8885 set_bit(AutoDetected, &rdev->flags);
8886 list_add(&rdev->same_set, &pending_raid_disks);
8889 mutex_unlock(&detected_devices_mutex);
8891 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8892 i_scanned, i_passed);
8894 autorun_devices(part);
8897 #endif /* !MODULE */
8899 static __exit void md_exit(void)
8901 struct mddev *mddev;
8902 struct list_head *tmp;
8905 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8906 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8908 unregister_blkdev(MD_MAJOR,"md");
8909 unregister_blkdev(mdp_major, "mdp");
8910 unregister_reboot_notifier(&md_notifier);
8911 unregister_sysctl_table(raid_table_header);
8913 /* We cannot unload the modules while some process is
8914 * waiting for us in select() or poll() - wake them up
8917 while (waitqueue_active(&md_event_waiters)) {
8918 /* not safe to leave yet */
8919 wake_up(&md_event_waiters);
8923 remove_proc_entry("mdstat", NULL);
8925 for_each_mddev(mddev, tmp) {
8926 export_array(mddev);
8927 mddev->hold_active = 0;
8929 destroy_workqueue(md_misc_wq);
8930 destroy_workqueue(md_wq);
8933 subsys_initcall(md_init);
8934 module_exit(md_exit)
8936 static int get_ro(char *buffer, struct kernel_param *kp)
8938 return sprintf(buffer, "%d", start_readonly);
8940 static int set_ro(const char *val, struct kernel_param *kp)
8942 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
8945 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8946 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8947 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8949 MODULE_LICENSE("GPL");
8950 MODULE_DESCRIPTION("MD RAID framework");
8952 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);