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/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
56 #include "md-cluster.h"
59 static void autostart_arrays(int part);
62 /* pers_list is a list of registered personalities protected
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
67 static LIST_HEAD(pers_list);
68 static DEFINE_SPINLOCK(pers_lock);
70 struct md_cluster_operations *md_cluster_ops;
71 EXPORT_SYMBOL(md_cluster_ops);
72 struct module *md_cluster_mod;
73 EXPORT_SYMBOL(md_cluster_mod);
75 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
76 static struct workqueue_struct *md_wq;
77 static struct workqueue_struct *md_misc_wq;
79 static int remove_and_add_spares(struct mddev *mddev,
80 struct md_rdev *this);
81 static void mddev_detach(struct mddev *mddev);
84 * Default number of read corrections we'll attempt on an rdev
85 * before ejecting it from the array. We divide the read error
86 * count by 2 for every hour elapsed between read errors.
88 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
90 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
91 * is 1000 KB/sec, so the extra system load does not show up that much.
92 * Increase it if you want to have more _guaranteed_ speed. Note that
93 * the RAID driver will use the maximum available bandwidth if the IO
94 * subsystem is idle. There is also an 'absolute maximum' reconstruction
95 * speed limit - in case reconstruction slows down your system despite
98 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
99 * or /sys/block/mdX/md/sync_speed_{min,max}
102 static int sysctl_speed_limit_min = 1000;
103 static int sysctl_speed_limit_max = 200000;
104 static inline int speed_min(struct mddev *mddev)
106 return mddev->sync_speed_min ?
107 mddev->sync_speed_min : sysctl_speed_limit_min;
110 static inline int speed_max(struct mddev *mddev)
112 return mddev->sync_speed_max ?
113 mddev->sync_speed_max : sysctl_speed_limit_max;
116 static struct ctl_table_header *raid_table_header;
118 static struct ctl_table raid_table[] = {
120 .procname = "speed_limit_min",
121 .data = &sysctl_speed_limit_min,
122 .maxlen = sizeof(int),
123 .mode = S_IRUGO|S_IWUSR,
124 .proc_handler = proc_dointvec,
127 .procname = "speed_limit_max",
128 .data = &sysctl_speed_limit_max,
129 .maxlen = sizeof(int),
130 .mode = S_IRUGO|S_IWUSR,
131 .proc_handler = proc_dointvec,
136 static struct ctl_table raid_dir_table[] = {
140 .mode = S_IRUGO|S_IXUGO,
146 static struct ctl_table raid_root_table[] = {
151 .child = raid_dir_table,
156 static const struct block_device_operations md_fops;
158 static int start_readonly;
161 * like bio_clone, but with a local bio set
164 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
169 if (!mddev || !mddev->bio_set)
170 return bio_alloc(gfp_mask, nr_iovecs);
172 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
177 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
179 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
182 if (!mddev || !mddev->bio_set)
183 return bio_clone(bio, gfp_mask);
185 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
187 EXPORT_SYMBOL_GPL(bio_clone_mddev);
190 * We have a system wide 'event count' that is incremented
191 * on any 'interesting' event, and readers of /proc/mdstat
192 * can use 'poll' or 'select' to find out when the event
196 * start array, stop array, error, add device, remove device,
197 * start build, activate spare
199 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
200 static atomic_t md_event_count;
201 void md_new_event(struct mddev *mddev)
203 atomic_inc(&md_event_count);
204 wake_up(&md_event_waiters);
206 EXPORT_SYMBOL_GPL(md_new_event);
208 /* Alternate version that can be called from interrupts
209 * when calling sysfs_notify isn't needed.
211 static void md_new_event_inintr(struct mddev *mddev)
213 atomic_inc(&md_event_count);
214 wake_up(&md_event_waiters);
218 * Enables to iterate over all existing md arrays
219 * all_mddevs_lock protects this list.
221 static LIST_HEAD(all_mddevs);
222 static DEFINE_SPINLOCK(all_mddevs_lock);
225 * iterates through all used mddevs in the system.
226 * We take care to grab the all_mddevs_lock whenever navigating
227 * the list, and to always hold a refcount when unlocked.
228 * Any code which breaks out of this loop while own
229 * a reference to the current mddev and must mddev_put it.
231 #define for_each_mddev(_mddev,_tmp) \
233 for (({ spin_lock(&all_mddevs_lock); \
234 _tmp = all_mddevs.next; \
236 ({ if (_tmp != &all_mddevs) \
237 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
238 spin_unlock(&all_mddevs_lock); \
239 if (_mddev) mddev_put(_mddev); \
240 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
241 _tmp != &all_mddevs;}); \
242 ({ spin_lock(&all_mddevs_lock); \
243 _tmp = _tmp->next;}) \
246 /* Rather than calling directly into the personality make_request function,
247 * IO requests come here first so that we can check if the device is
248 * being suspended pending a reconfiguration.
249 * We hold a refcount over the call to ->make_request. By the time that
250 * call has finished, the bio has been linked into some internal structure
251 * and so is visible to ->quiesce(), so we don't need the refcount any more.
253 static void md_make_request(struct request_queue *q, struct bio *bio)
255 const int rw = bio_data_dir(bio);
256 struct mddev *mddev = q->queuedata;
257 unsigned int sectors;
260 blk_queue_split(q, &bio, q->bio_split);
262 if (mddev == NULL || mddev->pers == NULL
267 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
268 if (bio_sectors(bio) != 0)
269 bio->bi_error = -EROFS;
273 smp_rmb(); /* Ensure implications of 'active' are visible */
275 if (mddev->suspended) {
278 prepare_to_wait(&mddev->sb_wait, &__wait,
279 TASK_UNINTERRUPTIBLE);
280 if (!mddev->suspended)
286 finish_wait(&mddev->sb_wait, &__wait);
288 atomic_inc(&mddev->active_io);
292 * save the sectors now since our bio can
293 * go away inside make_request
295 sectors = bio_sectors(bio);
296 mddev->pers->make_request(mddev, bio);
298 cpu = part_stat_lock();
299 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
300 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
303 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
304 wake_up(&mddev->sb_wait);
307 /* mddev_suspend makes sure no new requests are submitted
308 * to the device, and that any requests that have been submitted
309 * are completely handled.
310 * Once mddev_detach() is called and completes, the module will be
313 void mddev_suspend(struct mddev *mddev)
315 BUG_ON(mddev->suspended);
316 mddev->suspended = 1;
318 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
319 mddev->pers->quiesce(mddev, 1);
321 del_timer_sync(&mddev->safemode_timer);
323 EXPORT_SYMBOL_GPL(mddev_suspend);
325 void mddev_resume(struct mddev *mddev)
327 mddev->suspended = 0;
328 wake_up(&mddev->sb_wait);
329 mddev->pers->quiesce(mddev, 0);
331 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
332 md_wakeup_thread(mddev->thread);
333 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
335 EXPORT_SYMBOL_GPL(mddev_resume);
337 int mddev_congested(struct mddev *mddev, int bits)
339 struct md_personality *pers = mddev->pers;
343 if (mddev->suspended)
345 else if (pers && pers->congested)
346 ret = pers->congested(mddev, bits);
350 EXPORT_SYMBOL_GPL(mddev_congested);
351 static int md_congested(void *data, int bits)
353 struct mddev *mddev = data;
354 return mddev_congested(mddev, bits);
358 * Generic flush handling for md
361 static void md_end_flush(struct bio *bio)
363 struct md_rdev *rdev = bio->bi_private;
364 struct mddev *mddev = rdev->mddev;
366 rdev_dec_pending(rdev, mddev);
368 if (atomic_dec_and_test(&mddev->flush_pending)) {
369 /* The pre-request flush has finished */
370 queue_work(md_wq, &mddev->flush_work);
375 static void md_submit_flush_data(struct work_struct *ws);
377 static void submit_flushes(struct work_struct *ws)
379 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
380 struct md_rdev *rdev;
382 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
383 atomic_set(&mddev->flush_pending, 1);
385 rdev_for_each_rcu(rdev, mddev)
386 if (rdev->raid_disk >= 0 &&
387 !test_bit(Faulty, &rdev->flags)) {
388 /* Take two references, one is dropped
389 * when request finishes, one after
390 * we reclaim rcu_read_lock
393 atomic_inc(&rdev->nr_pending);
394 atomic_inc(&rdev->nr_pending);
396 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
397 bi->bi_end_io = md_end_flush;
398 bi->bi_private = rdev;
399 bi->bi_bdev = rdev->bdev;
400 atomic_inc(&mddev->flush_pending);
401 submit_bio(WRITE_FLUSH, bi);
403 rdev_dec_pending(rdev, mddev);
406 if (atomic_dec_and_test(&mddev->flush_pending))
407 queue_work(md_wq, &mddev->flush_work);
410 static void md_submit_flush_data(struct work_struct *ws)
412 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
413 struct bio *bio = mddev->flush_bio;
415 if (bio->bi_iter.bi_size == 0)
416 /* an empty barrier - all done */
419 bio->bi_rw &= ~REQ_FLUSH;
420 mddev->pers->make_request(mddev, bio);
423 mddev->flush_bio = NULL;
424 wake_up(&mddev->sb_wait);
427 void md_flush_request(struct mddev *mddev, struct bio *bio)
429 spin_lock_irq(&mddev->lock);
430 wait_event_lock_irq(mddev->sb_wait,
433 mddev->flush_bio = bio;
434 spin_unlock_irq(&mddev->lock);
436 INIT_WORK(&mddev->flush_work, submit_flushes);
437 queue_work(md_wq, &mddev->flush_work);
439 EXPORT_SYMBOL(md_flush_request);
441 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
443 struct mddev *mddev = cb->data;
444 md_wakeup_thread(mddev->thread);
447 EXPORT_SYMBOL(md_unplug);
449 static inline struct mddev *mddev_get(struct mddev *mddev)
451 atomic_inc(&mddev->active);
455 static void mddev_delayed_delete(struct work_struct *ws);
457 static void mddev_put(struct mddev *mddev)
459 struct bio_set *bs = NULL;
461 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
463 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
464 mddev->ctime == 0 && !mddev->hold_active) {
465 /* Array is not configured at all, and not held active,
467 list_del_init(&mddev->all_mddevs);
469 mddev->bio_set = NULL;
470 if (mddev->gendisk) {
471 /* We did a probe so need to clean up. Call
472 * queue_work inside the spinlock so that
473 * flush_workqueue() after mddev_find will
474 * succeed in waiting for the work to be done.
476 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
477 queue_work(md_misc_wq, &mddev->del_work);
481 spin_unlock(&all_mddevs_lock);
486 static void md_safemode_timeout(unsigned long data);
488 void mddev_init(struct mddev *mddev)
490 mutex_init(&mddev->open_mutex);
491 mutex_init(&mddev->reconfig_mutex);
492 mutex_init(&mddev->bitmap_info.mutex);
493 INIT_LIST_HEAD(&mddev->disks);
494 INIT_LIST_HEAD(&mddev->all_mddevs);
495 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
496 (unsigned long) mddev);
497 atomic_set(&mddev->active, 1);
498 atomic_set(&mddev->openers, 0);
499 atomic_set(&mddev->active_io, 0);
500 spin_lock_init(&mddev->lock);
501 atomic_set(&mddev->flush_pending, 0);
502 init_waitqueue_head(&mddev->sb_wait);
503 init_waitqueue_head(&mddev->recovery_wait);
504 mddev->reshape_position = MaxSector;
505 mddev->reshape_backwards = 0;
506 mddev->last_sync_action = "none";
507 mddev->resync_min = 0;
508 mddev->resync_max = MaxSector;
509 mddev->level = LEVEL_NONE;
511 EXPORT_SYMBOL_GPL(mddev_init);
513 static struct mddev *mddev_find(dev_t unit)
515 struct mddev *mddev, *new = NULL;
517 if (unit && MAJOR(unit) != MD_MAJOR)
518 unit &= ~((1<<MdpMinorShift)-1);
521 spin_lock(&all_mddevs_lock);
524 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
525 if (mddev->unit == unit) {
527 spin_unlock(&all_mddevs_lock);
533 list_add(&new->all_mddevs, &all_mddevs);
534 spin_unlock(&all_mddevs_lock);
535 new->hold_active = UNTIL_IOCTL;
539 /* find an unused unit number */
540 static int next_minor = 512;
541 int start = next_minor;
545 dev = MKDEV(MD_MAJOR, next_minor);
547 if (next_minor > MINORMASK)
549 if (next_minor == start) {
550 /* Oh dear, all in use. */
551 spin_unlock(&all_mddevs_lock);
557 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
558 if (mddev->unit == dev) {
564 new->md_minor = MINOR(dev);
565 new->hold_active = UNTIL_STOP;
566 list_add(&new->all_mddevs, &all_mddevs);
567 spin_unlock(&all_mddevs_lock);
570 spin_unlock(&all_mddevs_lock);
572 new = kzalloc(sizeof(*new), GFP_KERNEL);
577 if (MAJOR(unit) == MD_MAJOR)
578 new->md_minor = MINOR(unit);
580 new->md_minor = MINOR(unit) >> MdpMinorShift;
587 static struct attribute_group md_redundancy_group;
589 void mddev_unlock(struct mddev *mddev)
591 if (mddev->to_remove) {
592 /* These cannot be removed under reconfig_mutex as
593 * an access to the files will try to take reconfig_mutex
594 * while holding the file unremovable, which leads to
596 * So hold set sysfs_active while the remove in happeing,
597 * and anything else which might set ->to_remove or my
598 * otherwise change the sysfs namespace will fail with
599 * -EBUSY if sysfs_active is still set.
600 * We set sysfs_active under reconfig_mutex and elsewhere
601 * test it under the same mutex to ensure its correct value
604 struct attribute_group *to_remove = mddev->to_remove;
605 mddev->to_remove = NULL;
606 mddev->sysfs_active = 1;
607 mutex_unlock(&mddev->reconfig_mutex);
609 if (mddev->kobj.sd) {
610 if (to_remove != &md_redundancy_group)
611 sysfs_remove_group(&mddev->kobj, to_remove);
612 if (mddev->pers == NULL ||
613 mddev->pers->sync_request == NULL) {
614 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
615 if (mddev->sysfs_action)
616 sysfs_put(mddev->sysfs_action);
617 mddev->sysfs_action = NULL;
620 mddev->sysfs_active = 0;
622 mutex_unlock(&mddev->reconfig_mutex);
624 /* As we've dropped the mutex we need a spinlock to
625 * make sure the thread doesn't disappear
627 spin_lock(&pers_lock);
628 md_wakeup_thread(mddev->thread);
629 spin_unlock(&pers_lock);
631 EXPORT_SYMBOL_GPL(mddev_unlock);
633 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
635 struct md_rdev *rdev;
637 rdev_for_each_rcu(rdev, mddev)
638 if (rdev->desc_nr == nr)
643 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
645 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
647 struct md_rdev *rdev;
649 rdev_for_each(rdev, mddev)
650 if (rdev->bdev->bd_dev == dev)
656 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
658 struct md_rdev *rdev;
660 rdev_for_each_rcu(rdev, mddev)
661 if (rdev->bdev->bd_dev == dev)
667 static struct md_personality *find_pers(int level, char *clevel)
669 struct md_personality *pers;
670 list_for_each_entry(pers, &pers_list, list) {
671 if (level != LEVEL_NONE && pers->level == level)
673 if (strcmp(pers->name, clevel)==0)
679 /* return the offset of the super block in 512byte sectors */
680 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
682 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
683 return MD_NEW_SIZE_SECTORS(num_sectors);
686 static int alloc_disk_sb(struct md_rdev *rdev)
688 rdev->sb_page = alloc_page(GFP_KERNEL);
689 if (!rdev->sb_page) {
690 printk(KERN_ALERT "md: out of memory.\n");
697 void md_rdev_clear(struct md_rdev *rdev)
700 put_page(rdev->sb_page);
702 rdev->sb_page = NULL;
707 put_page(rdev->bb_page);
708 rdev->bb_page = NULL;
710 kfree(rdev->badblocks.page);
711 rdev->badblocks.page = NULL;
713 EXPORT_SYMBOL_GPL(md_rdev_clear);
715 static void super_written(struct bio *bio)
717 struct md_rdev *rdev = bio->bi_private;
718 struct mddev *mddev = rdev->mddev;
721 printk("md: super_written gets error=%d\n", bio->bi_error);
722 md_error(mddev, rdev);
725 if (atomic_dec_and_test(&mddev->pending_writes))
726 wake_up(&mddev->sb_wait);
730 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
731 sector_t sector, int size, struct page *page)
733 /* write first size bytes of page to sector of rdev
734 * Increment mddev->pending_writes before returning
735 * and decrement it on completion, waking up sb_wait
736 * if zero is reached.
737 * If an error occurred, call md_error
739 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
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;
747 atomic_inc(&mddev->pending_writes);
748 submit_bio(WRITE_FLUSH_FUA, bio);
751 void md_super_wait(struct mddev *mddev)
753 /* wait for all superblock writes that were scheduled to complete */
754 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
757 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
758 struct page *page, int rw, bool metadata_op)
760 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
763 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
764 rdev->meta_bdev : rdev->bdev;
766 bio->bi_iter.bi_sector = sector + rdev->sb_start;
767 else if (rdev->mddev->reshape_position != MaxSector &&
768 (rdev->mddev->reshape_backwards ==
769 (sector >= rdev->mddev->reshape_position)))
770 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
772 bio->bi_iter.bi_sector = sector + rdev->data_offset;
773 bio_add_page(bio, page, size, 0);
774 submit_bio_wait(rw, bio);
776 ret = !bio->bi_error;
780 EXPORT_SYMBOL_GPL(sync_page_io);
782 static int read_disk_sb(struct md_rdev *rdev, int size)
784 char b[BDEVNAME_SIZE];
789 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
795 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
796 bdevname(rdev->bdev,b));
800 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
802 return sb1->set_uuid0 == sb2->set_uuid0 &&
803 sb1->set_uuid1 == sb2->set_uuid1 &&
804 sb1->set_uuid2 == sb2->set_uuid2 &&
805 sb1->set_uuid3 == sb2->set_uuid3;
808 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
811 mdp_super_t *tmp1, *tmp2;
813 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
814 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
816 if (!tmp1 || !tmp2) {
818 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
826 * nr_disks is not constant
831 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
838 static u32 md_csum_fold(u32 csum)
840 csum = (csum & 0xffff) + (csum >> 16);
841 return (csum & 0xffff) + (csum >> 16);
844 static unsigned int calc_sb_csum(mdp_super_t *sb)
847 u32 *sb32 = (u32*)sb;
849 unsigned int disk_csum, csum;
851 disk_csum = sb->sb_csum;
854 for (i = 0; i < MD_SB_BYTES/4 ; i++)
856 csum = (newcsum & 0xffffffff) + (newcsum>>32);
859 /* This used to use csum_partial, which was wrong for several
860 * reasons including that different results are returned on
861 * different architectures. It isn't critical that we get exactly
862 * the same return value as before (we always csum_fold before
863 * testing, and that removes any differences). However as we
864 * know that csum_partial always returned a 16bit value on
865 * alphas, do a fold to maximise conformity to previous behaviour.
867 sb->sb_csum = md_csum_fold(disk_csum);
869 sb->sb_csum = disk_csum;
875 * Handle superblock details.
876 * We want to be able to handle multiple superblock formats
877 * so we have a common interface to them all, and an array of
878 * different handlers.
879 * We rely on user-space to write the initial superblock, and support
880 * reading and updating of superblocks.
881 * Interface methods are:
882 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
883 * loads and validates a superblock on dev.
884 * if refdev != NULL, compare superblocks on both devices
886 * 0 - dev has a superblock that is compatible with refdev
887 * 1 - dev has a superblock that is compatible and newer than refdev
888 * so dev should be used as the refdev in future
889 * -EINVAL superblock incompatible or invalid
890 * -othererror e.g. -EIO
892 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
893 * Verify that dev is acceptable into mddev.
894 * The first time, mddev->raid_disks will be 0, and data from
895 * dev should be merged in. Subsequent calls check that dev
896 * is new enough. Return 0 or -EINVAL
898 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
899 * Update the superblock for rdev with data in mddev
900 * This does not write to disc.
906 struct module *owner;
907 int (*load_super)(struct md_rdev *rdev,
908 struct md_rdev *refdev,
910 int (*validate_super)(struct mddev *mddev,
911 struct md_rdev *rdev);
912 void (*sync_super)(struct mddev *mddev,
913 struct md_rdev *rdev);
914 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
915 sector_t num_sectors);
916 int (*allow_new_offset)(struct md_rdev *rdev,
917 unsigned long long new_offset);
921 * Check that the given mddev has no bitmap.
923 * This function is called from the run method of all personalities that do not
924 * support bitmaps. It prints an error message and returns non-zero if mddev
925 * has a bitmap. Otherwise, it returns 0.
928 int md_check_no_bitmap(struct mddev *mddev)
930 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
932 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
933 mdname(mddev), mddev->pers->name);
936 EXPORT_SYMBOL(md_check_no_bitmap);
939 * load_super for 0.90.0
941 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
943 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
948 * Calculate the position of the superblock (512byte sectors),
949 * it's at the end of the disk.
951 * It also happens to be a multiple of 4Kb.
953 rdev->sb_start = calc_dev_sboffset(rdev);
955 ret = read_disk_sb(rdev, MD_SB_BYTES);
960 bdevname(rdev->bdev, b);
961 sb = page_address(rdev->sb_page);
963 if (sb->md_magic != MD_SB_MAGIC) {
964 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
969 if (sb->major_version != 0 ||
970 sb->minor_version < 90 ||
971 sb->minor_version > 91) {
972 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
973 sb->major_version, sb->minor_version,
978 if (sb->raid_disks <= 0)
981 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
982 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
987 rdev->preferred_minor = sb->md_minor;
988 rdev->data_offset = 0;
989 rdev->new_data_offset = 0;
990 rdev->sb_size = MD_SB_BYTES;
991 rdev->badblocks.shift = -1;
993 if (sb->level == LEVEL_MULTIPATH)
996 rdev->desc_nr = sb->this_disk.number;
1002 mdp_super_t *refsb = page_address(refdev->sb_page);
1003 if (!uuid_equal(refsb, sb)) {
1004 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1005 b, bdevname(refdev->bdev,b2));
1008 if (!sb_equal(refsb, sb)) {
1009 printk(KERN_WARNING "md: %s has same UUID"
1010 " but different superblock to %s\n",
1011 b, bdevname(refdev->bdev, b2));
1015 ev2 = md_event(refsb);
1021 rdev->sectors = rdev->sb_start;
1022 /* Limit to 4TB as metadata cannot record more than that.
1023 * (not needed for Linear and RAID0 as metadata doesn't
1026 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1027 rdev->sectors = (2ULL << 32) - 2;
1029 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1030 /* "this cannot possibly happen" ... */
1038 * validate_super for 0.90.0
1040 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1043 mdp_super_t *sb = page_address(rdev->sb_page);
1044 __u64 ev1 = md_event(sb);
1046 rdev->raid_disk = -1;
1047 clear_bit(Faulty, &rdev->flags);
1048 clear_bit(In_sync, &rdev->flags);
1049 clear_bit(Bitmap_sync, &rdev->flags);
1050 clear_bit(WriteMostly, &rdev->flags);
1052 if (mddev->raid_disks == 0) {
1053 mddev->major_version = 0;
1054 mddev->minor_version = sb->minor_version;
1055 mddev->patch_version = sb->patch_version;
1056 mddev->external = 0;
1057 mddev->chunk_sectors = sb->chunk_size >> 9;
1058 mddev->ctime = sb->ctime;
1059 mddev->utime = sb->utime;
1060 mddev->level = sb->level;
1061 mddev->clevel[0] = 0;
1062 mddev->layout = sb->layout;
1063 mddev->raid_disks = sb->raid_disks;
1064 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1065 mddev->events = ev1;
1066 mddev->bitmap_info.offset = 0;
1067 mddev->bitmap_info.space = 0;
1068 /* bitmap can use 60 K after the 4K superblocks */
1069 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1070 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1071 mddev->reshape_backwards = 0;
1073 if (mddev->minor_version >= 91) {
1074 mddev->reshape_position = sb->reshape_position;
1075 mddev->delta_disks = sb->delta_disks;
1076 mddev->new_level = sb->new_level;
1077 mddev->new_layout = sb->new_layout;
1078 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1079 if (mddev->delta_disks < 0)
1080 mddev->reshape_backwards = 1;
1082 mddev->reshape_position = MaxSector;
1083 mddev->delta_disks = 0;
1084 mddev->new_level = mddev->level;
1085 mddev->new_layout = mddev->layout;
1086 mddev->new_chunk_sectors = mddev->chunk_sectors;
1089 if (sb->state & (1<<MD_SB_CLEAN))
1090 mddev->recovery_cp = MaxSector;
1092 if (sb->events_hi == sb->cp_events_hi &&
1093 sb->events_lo == sb->cp_events_lo) {
1094 mddev->recovery_cp = sb->recovery_cp;
1096 mddev->recovery_cp = 0;
1099 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1100 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1101 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1102 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1104 mddev->max_disks = MD_SB_DISKS;
1106 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1107 mddev->bitmap_info.file == NULL) {
1108 mddev->bitmap_info.offset =
1109 mddev->bitmap_info.default_offset;
1110 mddev->bitmap_info.space =
1111 mddev->bitmap_info.default_space;
1114 } else if (mddev->pers == NULL) {
1115 /* Insist on good event counter while assembling, except
1116 * for spares (which don't need an event count) */
1118 if (sb->disks[rdev->desc_nr].state & (
1119 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1120 if (ev1 < mddev->events)
1122 } else if (mddev->bitmap) {
1123 /* if adding to array with a bitmap, then we can accept an
1124 * older device ... but not too old.
1126 if (ev1 < mddev->bitmap->events_cleared)
1128 if (ev1 < mddev->events)
1129 set_bit(Bitmap_sync, &rdev->flags);
1131 if (ev1 < mddev->events)
1132 /* just a hot-add of a new device, leave raid_disk at -1 */
1136 if (mddev->level != LEVEL_MULTIPATH) {
1137 desc = sb->disks + rdev->desc_nr;
1139 if (desc->state & (1<<MD_DISK_FAULTY))
1140 set_bit(Faulty, &rdev->flags);
1141 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1142 desc->raid_disk < mddev->raid_disks */) {
1143 set_bit(In_sync, &rdev->flags);
1144 rdev->raid_disk = desc->raid_disk;
1145 rdev->saved_raid_disk = desc->raid_disk;
1146 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1147 /* active but not in sync implies recovery up to
1148 * reshape position. We don't know exactly where
1149 * that is, so set to zero for now */
1150 if (mddev->minor_version >= 91) {
1151 rdev->recovery_offset = 0;
1152 rdev->raid_disk = desc->raid_disk;
1155 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1156 set_bit(WriteMostly, &rdev->flags);
1157 } else /* MULTIPATH are always insync */
1158 set_bit(In_sync, &rdev->flags);
1163 * sync_super for 0.90.0
1165 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1168 struct md_rdev *rdev2;
1169 int next_spare = mddev->raid_disks;
1171 /* make rdev->sb match mddev data..
1174 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1175 * 3/ any empty disks < next_spare become removed
1177 * disks[0] gets initialised to REMOVED because
1178 * we cannot be sure from other fields if it has
1179 * been initialised or not.
1182 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1184 rdev->sb_size = MD_SB_BYTES;
1186 sb = page_address(rdev->sb_page);
1188 memset(sb, 0, sizeof(*sb));
1190 sb->md_magic = MD_SB_MAGIC;
1191 sb->major_version = mddev->major_version;
1192 sb->patch_version = mddev->patch_version;
1193 sb->gvalid_words = 0; /* ignored */
1194 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1195 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1196 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1197 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1199 sb->ctime = mddev->ctime;
1200 sb->level = mddev->level;
1201 sb->size = mddev->dev_sectors / 2;
1202 sb->raid_disks = mddev->raid_disks;
1203 sb->md_minor = mddev->md_minor;
1204 sb->not_persistent = 0;
1205 sb->utime = mddev->utime;
1207 sb->events_hi = (mddev->events>>32);
1208 sb->events_lo = (u32)mddev->events;
1210 if (mddev->reshape_position == MaxSector)
1211 sb->minor_version = 90;
1213 sb->minor_version = 91;
1214 sb->reshape_position = mddev->reshape_position;
1215 sb->new_level = mddev->new_level;
1216 sb->delta_disks = mddev->delta_disks;
1217 sb->new_layout = mddev->new_layout;
1218 sb->new_chunk = mddev->new_chunk_sectors << 9;
1220 mddev->minor_version = sb->minor_version;
1223 sb->recovery_cp = mddev->recovery_cp;
1224 sb->cp_events_hi = (mddev->events>>32);
1225 sb->cp_events_lo = (u32)mddev->events;
1226 if (mddev->recovery_cp == MaxSector)
1227 sb->state = (1<< MD_SB_CLEAN);
1229 sb->recovery_cp = 0;
1231 sb->layout = mddev->layout;
1232 sb->chunk_size = mddev->chunk_sectors << 9;
1234 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1235 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1237 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1238 rdev_for_each(rdev2, mddev) {
1241 int is_active = test_bit(In_sync, &rdev2->flags);
1243 if (rdev2->raid_disk >= 0 &&
1244 sb->minor_version >= 91)
1245 /* we have nowhere to store the recovery_offset,
1246 * but if it is not below the reshape_position,
1247 * we can piggy-back on that.
1250 if (rdev2->raid_disk < 0 ||
1251 test_bit(Faulty, &rdev2->flags))
1254 desc_nr = rdev2->raid_disk;
1256 desc_nr = next_spare++;
1257 rdev2->desc_nr = desc_nr;
1258 d = &sb->disks[rdev2->desc_nr];
1260 d->number = rdev2->desc_nr;
1261 d->major = MAJOR(rdev2->bdev->bd_dev);
1262 d->minor = MINOR(rdev2->bdev->bd_dev);
1264 d->raid_disk = rdev2->raid_disk;
1266 d->raid_disk = rdev2->desc_nr; /* compatibility */
1267 if (test_bit(Faulty, &rdev2->flags))
1268 d->state = (1<<MD_DISK_FAULTY);
1269 else if (is_active) {
1270 d->state = (1<<MD_DISK_ACTIVE);
1271 if (test_bit(In_sync, &rdev2->flags))
1272 d->state |= (1<<MD_DISK_SYNC);
1280 if (test_bit(WriteMostly, &rdev2->flags))
1281 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1283 /* now set the "removed" and "faulty" bits on any missing devices */
1284 for (i=0 ; i < mddev->raid_disks ; i++) {
1285 mdp_disk_t *d = &sb->disks[i];
1286 if (d->state == 0 && d->number == 0) {
1289 d->state = (1<<MD_DISK_REMOVED);
1290 d->state |= (1<<MD_DISK_FAULTY);
1294 sb->nr_disks = nr_disks;
1295 sb->active_disks = active;
1296 sb->working_disks = working;
1297 sb->failed_disks = failed;
1298 sb->spare_disks = spare;
1300 sb->this_disk = sb->disks[rdev->desc_nr];
1301 sb->sb_csum = calc_sb_csum(sb);
1305 * rdev_size_change for 0.90.0
1307 static unsigned long long
1308 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1310 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1311 return 0; /* component must fit device */
1312 if (rdev->mddev->bitmap_info.offset)
1313 return 0; /* can't move bitmap */
1314 rdev->sb_start = calc_dev_sboffset(rdev);
1315 if (!num_sectors || num_sectors > rdev->sb_start)
1316 num_sectors = rdev->sb_start;
1317 /* Limit to 4TB as metadata cannot record more than that.
1318 * 4TB == 2^32 KB, or 2*2^32 sectors.
1320 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1321 num_sectors = (2ULL << 32) - 2;
1322 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1324 md_super_wait(rdev->mddev);
1329 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1331 /* non-zero offset changes not possible with v0.90 */
1332 return new_offset == 0;
1336 * version 1 superblock
1339 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1343 unsigned long long newcsum;
1344 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1345 __le32 *isuper = (__le32*)sb;
1347 disk_csum = sb->sb_csum;
1350 for (; size >= 4; size -= 4)
1351 newcsum += le32_to_cpu(*isuper++);
1354 newcsum += le16_to_cpu(*(__le16*) isuper);
1356 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1357 sb->sb_csum = disk_csum;
1358 return cpu_to_le32(csum);
1361 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1363 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1365 struct mdp_superblock_1 *sb;
1369 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1373 * Calculate the position of the superblock in 512byte sectors.
1374 * It is always aligned to a 4K boundary and
1375 * depeding on minor_version, it can be:
1376 * 0: At least 8K, but less than 12K, from end of device
1377 * 1: At start of device
1378 * 2: 4K from start of device.
1380 switch(minor_version) {
1382 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1384 sb_start &= ~(sector_t)(4*2-1);
1395 rdev->sb_start = sb_start;
1397 /* superblock is rarely larger than 1K, but it can be larger,
1398 * and it is safe to read 4k, so we do that
1400 ret = read_disk_sb(rdev, 4096);
1401 if (ret) return ret;
1403 sb = page_address(rdev->sb_page);
1405 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1406 sb->major_version != cpu_to_le32(1) ||
1407 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1408 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1409 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1412 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1413 printk("md: invalid superblock checksum on %s\n",
1414 bdevname(rdev->bdev,b));
1417 if (le64_to_cpu(sb->data_size) < 10) {
1418 printk("md: data_size too small on %s\n",
1419 bdevname(rdev->bdev,b));
1424 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1425 /* Some padding is non-zero, might be a new feature */
1428 rdev->preferred_minor = 0xffff;
1429 rdev->data_offset = le64_to_cpu(sb->data_offset);
1430 rdev->new_data_offset = rdev->data_offset;
1431 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1432 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1433 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1434 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1436 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1437 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1438 if (rdev->sb_size & bmask)
1439 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1442 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1445 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1448 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1451 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1453 if (!rdev->bb_page) {
1454 rdev->bb_page = alloc_page(GFP_KERNEL);
1458 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1459 rdev->badblocks.count == 0) {
1460 /* need to load the bad block list.
1461 * Currently we limit it to one page.
1467 int sectors = le16_to_cpu(sb->bblog_size);
1468 if (sectors > (PAGE_SIZE / 512))
1470 offset = le32_to_cpu(sb->bblog_offset);
1473 bb_sector = (long long)offset;
1474 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1475 rdev->bb_page, READ, true))
1477 bbp = (u64 *)page_address(rdev->bb_page);
1478 rdev->badblocks.shift = sb->bblog_shift;
1479 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1480 u64 bb = le64_to_cpu(*bbp);
1481 int count = bb & (0x3ff);
1482 u64 sector = bb >> 10;
1483 sector <<= sb->bblog_shift;
1484 count <<= sb->bblog_shift;
1487 if (md_set_badblocks(&rdev->badblocks,
1488 sector, count, 1) == 0)
1491 } else if (sb->bblog_offset != 0)
1492 rdev->badblocks.shift = 0;
1498 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1500 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1501 sb->level != refsb->level ||
1502 sb->layout != refsb->layout ||
1503 sb->chunksize != refsb->chunksize) {
1504 printk(KERN_WARNING "md: %s has strangely different"
1505 " superblock to %s\n",
1506 bdevname(rdev->bdev,b),
1507 bdevname(refdev->bdev,b2));
1510 ev1 = le64_to_cpu(sb->events);
1511 ev2 = le64_to_cpu(refsb->events);
1518 if (minor_version) {
1519 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1520 sectors -= rdev->data_offset;
1522 sectors = rdev->sb_start;
1523 if (sectors < le64_to_cpu(sb->data_size))
1525 rdev->sectors = le64_to_cpu(sb->data_size);
1529 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1531 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1532 __u64 ev1 = le64_to_cpu(sb->events);
1534 rdev->raid_disk = -1;
1535 clear_bit(Faulty, &rdev->flags);
1536 clear_bit(In_sync, &rdev->flags);
1537 clear_bit(Bitmap_sync, &rdev->flags);
1538 clear_bit(WriteMostly, &rdev->flags);
1540 if (mddev->raid_disks == 0) {
1541 mddev->major_version = 1;
1542 mddev->patch_version = 0;
1543 mddev->external = 0;
1544 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1545 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1546 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1547 mddev->level = le32_to_cpu(sb->level);
1548 mddev->clevel[0] = 0;
1549 mddev->layout = le32_to_cpu(sb->layout);
1550 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1551 mddev->dev_sectors = le64_to_cpu(sb->size);
1552 mddev->events = ev1;
1553 mddev->bitmap_info.offset = 0;
1554 mddev->bitmap_info.space = 0;
1555 /* Default location for bitmap is 1K after superblock
1556 * using 3K - total of 4K
1558 mddev->bitmap_info.default_offset = 1024 >> 9;
1559 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1560 mddev->reshape_backwards = 0;
1562 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1563 memcpy(mddev->uuid, sb->set_uuid, 16);
1565 mddev->max_disks = (4096-256)/2;
1567 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1568 mddev->bitmap_info.file == NULL) {
1569 mddev->bitmap_info.offset =
1570 (__s32)le32_to_cpu(sb->bitmap_offset);
1571 /* Metadata doesn't record how much space is available.
1572 * For 1.0, we assume we can use up to the superblock
1573 * if before, else to 4K beyond superblock.
1574 * For others, assume no change is possible.
1576 if (mddev->minor_version > 0)
1577 mddev->bitmap_info.space = 0;
1578 else if (mddev->bitmap_info.offset > 0)
1579 mddev->bitmap_info.space =
1580 8 - mddev->bitmap_info.offset;
1582 mddev->bitmap_info.space =
1583 -mddev->bitmap_info.offset;
1586 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1587 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1588 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1589 mddev->new_level = le32_to_cpu(sb->new_level);
1590 mddev->new_layout = le32_to_cpu(sb->new_layout);
1591 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1592 if (mddev->delta_disks < 0 ||
1593 (mddev->delta_disks == 0 &&
1594 (le32_to_cpu(sb->feature_map)
1595 & MD_FEATURE_RESHAPE_BACKWARDS)))
1596 mddev->reshape_backwards = 1;
1598 mddev->reshape_position = MaxSector;
1599 mddev->delta_disks = 0;
1600 mddev->new_level = mddev->level;
1601 mddev->new_layout = mddev->layout;
1602 mddev->new_chunk_sectors = mddev->chunk_sectors;
1605 } else if (mddev->pers == NULL) {
1606 /* Insist of good event counter while assembling, except for
1607 * spares (which don't need an event count) */
1609 if (rdev->desc_nr >= 0 &&
1610 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1611 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1612 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1613 if (ev1 < mddev->events)
1615 } else if (mddev->bitmap) {
1616 /* If adding to array with a bitmap, then we can accept an
1617 * older device, but not too old.
1619 if (ev1 < mddev->bitmap->events_cleared)
1621 if (ev1 < mddev->events)
1622 set_bit(Bitmap_sync, &rdev->flags);
1624 if (ev1 < mddev->events)
1625 /* just a hot-add of a new device, leave raid_disk at -1 */
1628 if (mddev->level != LEVEL_MULTIPATH) {
1630 if (rdev->desc_nr < 0 ||
1631 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1632 role = MD_DISK_ROLE_SPARE;
1635 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1637 case MD_DISK_ROLE_SPARE: /* spare */
1639 case MD_DISK_ROLE_FAULTY: /* faulty */
1640 set_bit(Faulty, &rdev->flags);
1642 case MD_DISK_ROLE_JOURNAL: /* journal device */
1643 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1644 /* journal device without journal feature */
1646 "md: journal device provided without journal feature, ignoring the device\n");
1649 set_bit(Journal, &rdev->flags);
1650 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1651 if (mddev->recovery_cp == MaxSector)
1652 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1653 rdev->raid_disk = mddev->raid_disks;
1656 rdev->saved_raid_disk = role;
1657 if ((le32_to_cpu(sb->feature_map) &
1658 MD_FEATURE_RECOVERY_OFFSET)) {
1659 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1660 if (!(le32_to_cpu(sb->feature_map) &
1661 MD_FEATURE_RECOVERY_BITMAP))
1662 rdev->saved_raid_disk = -1;
1664 set_bit(In_sync, &rdev->flags);
1665 rdev->raid_disk = role;
1668 if (sb->devflags & WriteMostly1)
1669 set_bit(WriteMostly, &rdev->flags);
1670 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1671 set_bit(Replacement, &rdev->flags);
1672 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1673 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1674 } else /* MULTIPATH are always insync */
1675 set_bit(In_sync, &rdev->flags);
1680 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1682 struct mdp_superblock_1 *sb;
1683 struct md_rdev *rdev2;
1685 /* make rdev->sb match mddev and rdev data. */
1687 sb = page_address(rdev->sb_page);
1689 sb->feature_map = 0;
1691 sb->recovery_offset = cpu_to_le64(0);
1692 memset(sb->pad3, 0, sizeof(sb->pad3));
1694 sb->utime = cpu_to_le64((__u64)mddev->utime);
1695 sb->events = cpu_to_le64(mddev->events);
1697 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1698 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1699 sb->resync_offset = cpu_to_le64(MaxSector);
1701 sb->resync_offset = cpu_to_le64(0);
1703 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1705 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1706 sb->size = cpu_to_le64(mddev->dev_sectors);
1707 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1708 sb->level = cpu_to_le32(mddev->level);
1709 sb->layout = cpu_to_le32(mddev->layout);
1711 if (test_bit(WriteMostly, &rdev->flags))
1712 sb->devflags |= WriteMostly1;
1714 sb->devflags &= ~WriteMostly1;
1715 sb->data_offset = cpu_to_le64(rdev->data_offset);
1716 sb->data_size = cpu_to_le64(rdev->sectors);
1718 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1719 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1720 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1723 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1724 !test_bit(In_sync, &rdev->flags)) {
1726 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1727 sb->recovery_offset =
1728 cpu_to_le64(rdev->recovery_offset);
1729 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1731 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1733 /* Note: recovery_offset and journal_tail share space */
1734 if (test_bit(Journal, &rdev->flags))
1735 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1736 if (test_bit(Replacement, &rdev->flags))
1738 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1740 if (mddev->reshape_position != MaxSector) {
1741 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1742 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1743 sb->new_layout = cpu_to_le32(mddev->new_layout);
1744 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1745 sb->new_level = cpu_to_le32(mddev->new_level);
1746 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1747 if (mddev->delta_disks == 0 &&
1748 mddev->reshape_backwards)
1750 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1751 if (rdev->new_data_offset != rdev->data_offset) {
1753 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1754 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1755 - rdev->data_offset));
1759 if (mddev_is_clustered(mddev))
1760 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1762 if (rdev->badblocks.count == 0)
1763 /* Nothing to do for bad blocks*/ ;
1764 else if (sb->bblog_offset == 0)
1765 /* Cannot record bad blocks on this device */
1766 md_error(mddev, rdev);
1768 struct badblocks *bb = &rdev->badblocks;
1769 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1771 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1776 seq = read_seqbegin(&bb->lock);
1778 memset(bbp, 0xff, PAGE_SIZE);
1780 for (i = 0 ; i < bb->count ; i++) {
1781 u64 internal_bb = p[i];
1782 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1783 | BB_LEN(internal_bb));
1784 bbp[i] = cpu_to_le64(store_bb);
1787 if (read_seqretry(&bb->lock, seq))
1790 bb->sector = (rdev->sb_start +
1791 (int)le32_to_cpu(sb->bblog_offset));
1792 bb->size = le16_to_cpu(sb->bblog_size);
1797 rdev_for_each(rdev2, mddev)
1798 if (rdev2->desc_nr+1 > max_dev)
1799 max_dev = rdev2->desc_nr+1;
1801 if (max_dev > le32_to_cpu(sb->max_dev)) {
1803 sb->max_dev = cpu_to_le32(max_dev);
1804 rdev->sb_size = max_dev * 2 + 256;
1805 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1806 if (rdev->sb_size & bmask)
1807 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1809 max_dev = le32_to_cpu(sb->max_dev);
1811 for (i=0; i<max_dev;i++)
1812 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1814 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1815 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1817 rdev_for_each(rdev2, mddev) {
1819 if (test_bit(Faulty, &rdev2->flags))
1820 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1821 else if (test_bit(In_sync, &rdev2->flags))
1822 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1823 else if (test_bit(Journal, &rdev2->flags))
1824 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1825 else if (rdev2->raid_disk >= 0)
1826 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1828 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1831 sb->sb_csum = calc_sb_1_csum(sb);
1834 static unsigned long long
1835 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1837 struct mdp_superblock_1 *sb;
1838 sector_t max_sectors;
1839 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1840 return 0; /* component must fit device */
1841 if (rdev->data_offset != rdev->new_data_offset)
1842 return 0; /* too confusing */
1843 if (rdev->sb_start < rdev->data_offset) {
1844 /* minor versions 1 and 2; superblock before data */
1845 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1846 max_sectors -= rdev->data_offset;
1847 if (!num_sectors || num_sectors > max_sectors)
1848 num_sectors = max_sectors;
1849 } else if (rdev->mddev->bitmap_info.offset) {
1850 /* minor version 0 with bitmap we can't move */
1853 /* minor version 0; superblock after data */
1855 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1856 sb_start &= ~(sector_t)(4*2 - 1);
1857 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1858 if (!num_sectors || num_sectors > max_sectors)
1859 num_sectors = max_sectors;
1860 rdev->sb_start = sb_start;
1862 sb = page_address(rdev->sb_page);
1863 sb->data_size = cpu_to_le64(num_sectors);
1864 sb->super_offset = rdev->sb_start;
1865 sb->sb_csum = calc_sb_1_csum(sb);
1866 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1868 md_super_wait(rdev->mddev);
1874 super_1_allow_new_offset(struct md_rdev *rdev,
1875 unsigned long long new_offset)
1877 /* All necessary checks on new >= old have been done */
1878 struct bitmap *bitmap;
1879 if (new_offset >= rdev->data_offset)
1882 /* with 1.0 metadata, there is no metadata to tread on
1883 * so we can always move back */
1884 if (rdev->mddev->minor_version == 0)
1887 /* otherwise we must be sure not to step on
1888 * any metadata, so stay:
1889 * 36K beyond start of superblock
1890 * beyond end of badblocks
1891 * beyond write-intent bitmap
1893 if (rdev->sb_start + (32+4)*2 > new_offset)
1895 bitmap = rdev->mddev->bitmap;
1896 if (bitmap && !rdev->mddev->bitmap_info.file &&
1897 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1898 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1900 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1906 static struct super_type super_types[] = {
1909 .owner = THIS_MODULE,
1910 .load_super = super_90_load,
1911 .validate_super = super_90_validate,
1912 .sync_super = super_90_sync,
1913 .rdev_size_change = super_90_rdev_size_change,
1914 .allow_new_offset = super_90_allow_new_offset,
1918 .owner = THIS_MODULE,
1919 .load_super = super_1_load,
1920 .validate_super = super_1_validate,
1921 .sync_super = super_1_sync,
1922 .rdev_size_change = super_1_rdev_size_change,
1923 .allow_new_offset = super_1_allow_new_offset,
1927 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1929 if (mddev->sync_super) {
1930 mddev->sync_super(mddev, rdev);
1934 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1936 super_types[mddev->major_version].sync_super(mddev, rdev);
1939 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1941 struct md_rdev *rdev, *rdev2;
1944 rdev_for_each_rcu(rdev, mddev1) {
1945 if (test_bit(Faulty, &rdev->flags) ||
1946 test_bit(Journal, &rdev->flags) ||
1947 rdev->raid_disk == -1)
1949 rdev_for_each_rcu(rdev2, mddev2) {
1950 if (test_bit(Faulty, &rdev2->flags) ||
1951 test_bit(Journal, &rdev2->flags) ||
1952 rdev2->raid_disk == -1)
1954 if (rdev->bdev->bd_contains ==
1955 rdev2->bdev->bd_contains) {
1965 static LIST_HEAD(pending_raid_disks);
1968 * Try to register data integrity profile for an mddev
1970 * This is called when an array is started and after a disk has been kicked
1971 * from the array. It only succeeds if all working and active component devices
1972 * are integrity capable with matching profiles.
1974 int md_integrity_register(struct mddev *mddev)
1976 struct md_rdev *rdev, *reference = NULL;
1978 if (list_empty(&mddev->disks))
1979 return 0; /* nothing to do */
1980 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1981 return 0; /* shouldn't register, or already is */
1982 rdev_for_each(rdev, mddev) {
1983 /* skip spares and non-functional disks */
1984 if (test_bit(Faulty, &rdev->flags))
1986 if (rdev->raid_disk < 0)
1989 /* Use the first rdev as the reference */
1993 /* does this rdev's profile match the reference profile? */
1994 if (blk_integrity_compare(reference->bdev->bd_disk,
1995 rdev->bdev->bd_disk) < 0)
1998 if (!reference || !bdev_get_integrity(reference->bdev))
2001 * All component devices are integrity capable and have matching
2002 * profiles, register the common profile for the md device.
2004 blk_integrity_register(mddev->gendisk,
2005 bdev_get_integrity(reference->bdev));
2007 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2008 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2009 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2015 EXPORT_SYMBOL(md_integrity_register);
2017 /* Disable data integrity if non-capable/non-matching disk is being added */
2018 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2020 struct blk_integrity *bi_rdev;
2021 struct blk_integrity *bi_mddev;
2023 if (!mddev->gendisk)
2026 bi_rdev = bdev_get_integrity(rdev->bdev);
2027 bi_mddev = blk_get_integrity(mddev->gendisk);
2029 if (!bi_mddev) /* nothing to do */
2031 if (rdev->raid_disk < 0) /* skip spares */
2033 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2034 rdev->bdev->bd_disk) >= 0)
2036 WARN_ON_ONCE(!mddev->suspended);
2037 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2038 blk_integrity_unregister(mddev->gendisk);
2040 EXPORT_SYMBOL(md_integrity_add_rdev);
2042 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2044 char b[BDEVNAME_SIZE];
2048 /* prevent duplicates */
2049 if (find_rdev(mddev, rdev->bdev->bd_dev))
2052 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2053 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2054 rdev->sectors < mddev->dev_sectors)) {
2056 /* Cannot change size, so fail
2057 * If mddev->level <= 0, then we don't care
2058 * about aligning sizes (e.g. linear)
2060 if (mddev->level > 0)
2063 mddev->dev_sectors = rdev->sectors;
2066 /* Verify rdev->desc_nr is unique.
2067 * If it is -1, assign a free number, else
2068 * check number is not in use
2071 if (rdev->desc_nr < 0) {
2074 choice = mddev->raid_disks;
2075 while (md_find_rdev_nr_rcu(mddev, choice))
2077 rdev->desc_nr = choice;
2079 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2085 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2086 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2087 mdname(mddev), mddev->max_disks);
2090 bdevname(rdev->bdev,b);
2091 strreplace(b, '/', '!');
2093 rdev->mddev = mddev;
2094 printk(KERN_INFO "md: bind<%s>\n", b);
2096 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2099 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2100 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2101 /* failure here is OK */;
2102 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2104 list_add_rcu(&rdev->same_set, &mddev->disks);
2105 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2107 /* May as well allow recovery to be retried once */
2108 mddev->recovery_disabled++;
2113 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2118 static void md_delayed_delete(struct work_struct *ws)
2120 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2121 kobject_del(&rdev->kobj);
2122 kobject_put(&rdev->kobj);
2125 static void unbind_rdev_from_array(struct md_rdev *rdev)
2127 char b[BDEVNAME_SIZE];
2129 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2130 list_del_rcu(&rdev->same_set);
2131 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2133 sysfs_remove_link(&rdev->kobj, "block");
2134 sysfs_put(rdev->sysfs_state);
2135 rdev->sysfs_state = NULL;
2136 rdev->badblocks.count = 0;
2137 /* We need to delay this, otherwise we can deadlock when
2138 * writing to 'remove' to "dev/state". We also need
2139 * to delay it due to rcu usage.
2142 INIT_WORK(&rdev->del_work, md_delayed_delete);
2143 kobject_get(&rdev->kobj);
2144 queue_work(md_misc_wq, &rdev->del_work);
2148 * prevent the device from being mounted, repartitioned or
2149 * otherwise reused by a RAID array (or any other kernel
2150 * subsystem), by bd_claiming the device.
2152 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2155 struct block_device *bdev;
2156 char b[BDEVNAME_SIZE];
2158 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2159 shared ? (struct md_rdev *)lock_rdev : rdev);
2161 printk(KERN_ERR "md: could not open %s.\n",
2162 __bdevname(dev, b));
2163 return PTR_ERR(bdev);
2169 static void unlock_rdev(struct md_rdev *rdev)
2171 struct block_device *bdev = rdev->bdev;
2173 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2176 void md_autodetect_dev(dev_t dev);
2178 static void export_rdev(struct md_rdev *rdev)
2180 char b[BDEVNAME_SIZE];
2182 printk(KERN_INFO "md: export_rdev(%s)\n",
2183 bdevname(rdev->bdev,b));
2184 md_rdev_clear(rdev);
2186 if (test_bit(AutoDetected, &rdev->flags))
2187 md_autodetect_dev(rdev->bdev->bd_dev);
2190 kobject_put(&rdev->kobj);
2193 void md_kick_rdev_from_array(struct md_rdev *rdev)
2195 unbind_rdev_from_array(rdev);
2198 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2200 static void export_array(struct mddev *mddev)
2202 struct md_rdev *rdev;
2204 while (!list_empty(&mddev->disks)) {
2205 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2207 md_kick_rdev_from_array(rdev);
2209 mddev->raid_disks = 0;
2210 mddev->major_version = 0;
2213 static void sync_sbs(struct mddev *mddev, int nospares)
2215 /* Update each superblock (in-memory image), but
2216 * if we are allowed to, skip spares which already
2217 * have the right event counter, or have one earlier
2218 * (which would mean they aren't being marked as dirty
2219 * with the rest of the array)
2221 struct md_rdev *rdev;
2222 rdev_for_each(rdev, mddev) {
2223 if (rdev->sb_events == mddev->events ||
2225 rdev->raid_disk < 0 &&
2226 rdev->sb_events+1 == mddev->events)) {
2227 /* Don't update this superblock */
2228 rdev->sb_loaded = 2;
2230 sync_super(mddev, rdev);
2231 rdev->sb_loaded = 1;
2236 static bool does_sb_need_changing(struct mddev *mddev)
2238 struct md_rdev *rdev;
2239 struct mdp_superblock_1 *sb;
2242 /* Find a good rdev */
2243 rdev_for_each(rdev, mddev)
2244 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2247 /* No good device found. */
2251 sb = page_address(rdev->sb_page);
2252 /* Check if a device has become faulty or a spare become active */
2253 rdev_for_each(rdev, mddev) {
2254 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2255 /* Device activated? */
2256 if (role == 0xffff && rdev->raid_disk >=0 &&
2257 !test_bit(Faulty, &rdev->flags))
2259 /* Device turned faulty? */
2260 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2264 /* Check if any mddev parameters have changed */
2265 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2266 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2267 (mddev->layout != le64_to_cpu(sb->layout)) ||
2268 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2269 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2275 void md_update_sb(struct mddev *mddev, int force_change)
2277 struct md_rdev *rdev;
2280 int any_badblocks_changed = 0;
2285 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2289 if (mddev_is_clustered(mddev)) {
2290 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2292 ret = md_cluster_ops->metadata_update_start(mddev);
2293 /* Has someone else has updated the sb */
2294 if (!does_sb_need_changing(mddev)) {
2296 md_cluster_ops->metadata_update_cancel(mddev);
2297 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2302 /* First make sure individual recovery_offsets are correct */
2303 rdev_for_each(rdev, mddev) {
2304 if (rdev->raid_disk >= 0 &&
2305 mddev->delta_disks >= 0 &&
2306 !test_bit(Journal, &rdev->flags) &&
2307 !test_bit(In_sync, &rdev->flags) &&
2308 mddev->curr_resync_completed > rdev->recovery_offset)
2309 rdev->recovery_offset = mddev->curr_resync_completed;
2312 if (!mddev->persistent) {
2313 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2314 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2315 if (!mddev->external) {
2316 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2317 rdev_for_each(rdev, mddev) {
2318 if (rdev->badblocks.changed) {
2319 rdev->badblocks.changed = 0;
2320 md_ack_all_badblocks(&rdev->badblocks);
2321 md_error(mddev, rdev);
2323 clear_bit(Blocked, &rdev->flags);
2324 clear_bit(BlockedBadBlocks, &rdev->flags);
2325 wake_up(&rdev->blocked_wait);
2328 wake_up(&mddev->sb_wait);
2332 spin_lock(&mddev->lock);
2334 mddev->utime = get_seconds();
2336 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2338 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2339 /* just a clean<-> dirty transition, possibly leave spares alone,
2340 * though if events isn't the right even/odd, we will have to do
2346 if (mddev->degraded)
2347 /* If the array is degraded, then skipping spares is both
2348 * dangerous and fairly pointless.
2349 * Dangerous because a device that was removed from the array
2350 * might have a event_count that still looks up-to-date,
2351 * so it can be re-added without a resync.
2352 * Pointless because if there are any spares to skip,
2353 * then a recovery will happen and soon that array won't
2354 * be degraded any more and the spare can go back to sleep then.
2358 sync_req = mddev->in_sync;
2360 /* If this is just a dirty<->clean transition, and the array is clean
2361 * and 'events' is odd, we can roll back to the previous clean state */
2363 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2364 && mddev->can_decrease_events
2365 && mddev->events != 1) {
2367 mddev->can_decrease_events = 0;
2369 /* otherwise we have to go forward and ... */
2371 mddev->can_decrease_events = nospares;
2375 * This 64-bit counter should never wrap.
2376 * Either we are in around ~1 trillion A.C., assuming
2377 * 1 reboot per second, or we have a bug...
2379 WARN_ON(mddev->events == 0);
2381 rdev_for_each(rdev, mddev) {
2382 if (rdev->badblocks.changed)
2383 any_badblocks_changed++;
2384 if (test_bit(Faulty, &rdev->flags))
2385 set_bit(FaultRecorded, &rdev->flags);
2388 sync_sbs(mddev, nospares);
2389 spin_unlock(&mddev->lock);
2391 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2392 mdname(mddev), mddev->in_sync);
2394 bitmap_update_sb(mddev->bitmap);
2395 rdev_for_each(rdev, mddev) {
2396 char b[BDEVNAME_SIZE];
2398 if (rdev->sb_loaded != 1)
2399 continue; /* no noise on spare devices */
2401 if (!test_bit(Faulty, &rdev->flags)) {
2402 md_super_write(mddev,rdev,
2403 rdev->sb_start, rdev->sb_size,
2405 pr_debug("md: (write) %s's sb offset: %llu\n",
2406 bdevname(rdev->bdev, b),
2407 (unsigned long long)rdev->sb_start);
2408 rdev->sb_events = mddev->events;
2409 if (rdev->badblocks.size) {
2410 md_super_write(mddev, rdev,
2411 rdev->badblocks.sector,
2412 rdev->badblocks.size << 9,
2414 rdev->badblocks.size = 0;
2418 pr_debug("md: %s (skipping faulty)\n",
2419 bdevname(rdev->bdev, b));
2421 if (mddev->level == LEVEL_MULTIPATH)
2422 /* only need to write one superblock... */
2425 md_super_wait(mddev);
2426 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2428 spin_lock(&mddev->lock);
2429 if (mddev->in_sync != sync_req ||
2430 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2431 /* have to write it out again */
2432 spin_unlock(&mddev->lock);
2435 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2436 spin_unlock(&mddev->lock);
2437 wake_up(&mddev->sb_wait);
2438 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2439 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2441 rdev_for_each(rdev, mddev) {
2442 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2443 clear_bit(Blocked, &rdev->flags);
2445 if (any_badblocks_changed)
2446 md_ack_all_badblocks(&rdev->badblocks);
2447 clear_bit(BlockedBadBlocks, &rdev->flags);
2448 wake_up(&rdev->blocked_wait);
2451 if (mddev_is_clustered(mddev) && ret == 0)
2452 md_cluster_ops->metadata_update_finish(mddev);
2454 EXPORT_SYMBOL(md_update_sb);
2456 static int add_bound_rdev(struct md_rdev *rdev)
2458 struct mddev *mddev = rdev->mddev;
2461 if (!mddev->pers->hot_remove_disk) {
2462 /* If there is hot_add_disk but no hot_remove_disk
2463 * then added disks for geometry changes,
2464 * and should be added immediately.
2466 super_types[mddev->major_version].
2467 validate_super(mddev, rdev);
2468 err = mddev->pers->hot_add_disk(mddev, rdev);
2470 unbind_rdev_from_array(rdev);
2475 sysfs_notify_dirent_safe(rdev->sysfs_state);
2477 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2478 if (mddev->degraded)
2479 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2480 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2481 md_new_event(mddev);
2482 md_wakeup_thread(mddev->thread);
2486 /* words written to sysfs files may, or may not, be \n terminated.
2487 * We want to accept with case. For this we use cmd_match.
2489 static int cmd_match(const char *cmd, const char *str)
2491 /* See if cmd, written into a sysfs file, matches
2492 * str. They must either be the same, or cmd can
2493 * have a trailing newline
2495 while (*cmd && *str && *cmd == *str) {
2506 struct rdev_sysfs_entry {
2507 struct attribute attr;
2508 ssize_t (*show)(struct md_rdev *, char *);
2509 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2513 state_show(struct md_rdev *rdev, char *page)
2517 unsigned long flags = ACCESS_ONCE(rdev->flags);
2519 if (test_bit(Faulty, &flags) ||
2520 rdev->badblocks.unacked_exist) {
2521 len+= sprintf(page+len, "%sfaulty",sep);
2524 if (test_bit(In_sync, &flags)) {
2525 len += sprintf(page+len, "%sin_sync",sep);
2528 if (test_bit(Journal, &flags)) {
2529 len += sprintf(page+len, "%sjournal",sep);
2532 if (test_bit(WriteMostly, &flags)) {
2533 len += sprintf(page+len, "%swrite_mostly",sep);
2536 if (test_bit(Blocked, &flags) ||
2537 (rdev->badblocks.unacked_exist
2538 && !test_bit(Faulty, &flags))) {
2539 len += sprintf(page+len, "%sblocked", sep);
2542 if (!test_bit(Faulty, &flags) &&
2543 !test_bit(Journal, &flags) &&
2544 !test_bit(In_sync, &flags)) {
2545 len += sprintf(page+len, "%sspare", sep);
2548 if (test_bit(WriteErrorSeen, &flags)) {
2549 len += sprintf(page+len, "%swrite_error", sep);
2552 if (test_bit(WantReplacement, &flags)) {
2553 len += sprintf(page+len, "%swant_replacement", sep);
2556 if (test_bit(Replacement, &flags)) {
2557 len += sprintf(page+len, "%sreplacement", sep);
2561 return len+sprintf(page+len, "\n");
2565 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2568 * faulty - simulates an error
2569 * remove - disconnects the device
2570 * writemostly - sets write_mostly
2571 * -writemostly - clears write_mostly
2572 * blocked - sets the Blocked flags
2573 * -blocked - clears the Blocked and possibly simulates an error
2574 * insync - sets Insync providing device isn't active
2575 * -insync - clear Insync for a device with a slot assigned,
2576 * so that it gets rebuilt based on bitmap
2577 * write_error - sets WriteErrorSeen
2578 * -write_error - clears WriteErrorSeen
2581 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2582 md_error(rdev->mddev, rdev);
2583 if (test_bit(Faulty, &rdev->flags))
2587 } else if (cmd_match(buf, "remove")) {
2588 if (rdev->raid_disk >= 0)
2591 struct mddev *mddev = rdev->mddev;
2593 if (mddev_is_clustered(mddev))
2594 err = md_cluster_ops->remove_disk(mddev, rdev);
2597 md_kick_rdev_from_array(rdev);
2599 md_update_sb(mddev, 1);
2600 md_new_event(mddev);
2603 } else if (cmd_match(buf, "writemostly")) {
2604 set_bit(WriteMostly, &rdev->flags);
2606 } else if (cmd_match(buf, "-writemostly")) {
2607 clear_bit(WriteMostly, &rdev->flags);
2609 } else if (cmd_match(buf, "blocked")) {
2610 set_bit(Blocked, &rdev->flags);
2612 } else if (cmd_match(buf, "-blocked")) {
2613 if (!test_bit(Faulty, &rdev->flags) &&
2614 rdev->badblocks.unacked_exist) {
2615 /* metadata handler doesn't understand badblocks,
2616 * so we need to fail the device
2618 md_error(rdev->mddev, rdev);
2620 clear_bit(Blocked, &rdev->flags);
2621 clear_bit(BlockedBadBlocks, &rdev->flags);
2622 wake_up(&rdev->blocked_wait);
2623 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2624 md_wakeup_thread(rdev->mddev->thread);
2627 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2628 set_bit(In_sync, &rdev->flags);
2630 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2631 !test_bit(Journal, &rdev->flags)) {
2632 if (rdev->mddev->pers == NULL) {
2633 clear_bit(In_sync, &rdev->flags);
2634 rdev->saved_raid_disk = rdev->raid_disk;
2635 rdev->raid_disk = -1;
2638 } else if (cmd_match(buf, "write_error")) {
2639 set_bit(WriteErrorSeen, &rdev->flags);
2641 } else if (cmd_match(buf, "-write_error")) {
2642 clear_bit(WriteErrorSeen, &rdev->flags);
2644 } else if (cmd_match(buf, "want_replacement")) {
2645 /* Any non-spare device that is not a replacement can
2646 * become want_replacement at any time, but we then need to
2647 * check if recovery is needed.
2649 if (rdev->raid_disk >= 0 &&
2650 !test_bit(Journal, &rdev->flags) &&
2651 !test_bit(Replacement, &rdev->flags))
2652 set_bit(WantReplacement, &rdev->flags);
2653 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2654 md_wakeup_thread(rdev->mddev->thread);
2656 } else if (cmd_match(buf, "-want_replacement")) {
2657 /* Clearing 'want_replacement' is always allowed.
2658 * Once replacements starts it is too late though.
2661 clear_bit(WantReplacement, &rdev->flags);
2662 } else if (cmd_match(buf, "replacement")) {
2663 /* Can only set a device as a replacement when array has not
2664 * yet been started. Once running, replacement is automatic
2665 * from spares, or by assigning 'slot'.
2667 if (rdev->mddev->pers)
2670 set_bit(Replacement, &rdev->flags);
2673 } else if (cmd_match(buf, "-replacement")) {
2674 /* Similarly, can only clear Replacement before start */
2675 if (rdev->mddev->pers)
2678 clear_bit(Replacement, &rdev->flags);
2681 } else if (cmd_match(buf, "re-add")) {
2682 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2683 /* clear_bit is performed _after_ all the devices
2684 * have their local Faulty bit cleared. If any writes
2685 * happen in the meantime in the local node, they
2686 * will land in the local bitmap, which will be synced
2687 * by this node eventually
2689 if (!mddev_is_clustered(rdev->mddev) ||
2690 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2691 clear_bit(Faulty, &rdev->flags);
2692 err = add_bound_rdev(rdev);
2698 sysfs_notify_dirent_safe(rdev->sysfs_state);
2699 return err ? err : len;
2701 static struct rdev_sysfs_entry rdev_state =
2702 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2705 errors_show(struct md_rdev *rdev, char *page)
2707 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2711 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2716 rv = kstrtouint(buf, 10, &n);
2719 atomic_set(&rdev->corrected_errors, n);
2722 static struct rdev_sysfs_entry rdev_errors =
2723 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2726 slot_show(struct md_rdev *rdev, char *page)
2728 if (test_bit(Journal, &rdev->flags))
2729 return sprintf(page, "journal\n");
2730 else if (rdev->raid_disk < 0)
2731 return sprintf(page, "none\n");
2733 return sprintf(page, "%d\n", rdev->raid_disk);
2737 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2742 if (test_bit(Journal, &rdev->flags))
2744 if (strncmp(buf, "none", 4)==0)
2747 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2751 if (rdev->mddev->pers && slot == -1) {
2752 /* Setting 'slot' on an active array requires also
2753 * updating the 'rd%d' link, and communicating
2754 * with the personality with ->hot_*_disk.
2755 * For now we only support removing
2756 * failed/spare devices. This normally happens automatically,
2757 * but not when the metadata is externally managed.
2759 if (rdev->raid_disk == -1)
2761 /* personality does all needed checks */
2762 if (rdev->mddev->pers->hot_remove_disk == NULL)
2764 clear_bit(Blocked, &rdev->flags);
2765 remove_and_add_spares(rdev->mddev, rdev);
2766 if (rdev->raid_disk >= 0)
2768 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2769 md_wakeup_thread(rdev->mddev->thread);
2770 } else if (rdev->mddev->pers) {
2771 /* Activating a spare .. or possibly reactivating
2772 * if we ever get bitmaps working here.
2775 if (rdev->raid_disk != -1)
2778 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2781 if (rdev->mddev->pers->hot_add_disk == NULL)
2784 if (slot >= rdev->mddev->raid_disks &&
2785 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2788 rdev->raid_disk = slot;
2789 if (test_bit(In_sync, &rdev->flags))
2790 rdev->saved_raid_disk = slot;
2792 rdev->saved_raid_disk = -1;
2793 clear_bit(In_sync, &rdev->flags);
2794 clear_bit(Bitmap_sync, &rdev->flags);
2795 remove_and_add_spares(rdev->mddev, rdev);
2796 if (rdev->raid_disk == -1)
2798 /* don't wakeup anyone, leave that to userspace. */
2800 if (slot >= rdev->mddev->raid_disks &&
2801 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2803 rdev->raid_disk = slot;
2804 /* assume it is working */
2805 clear_bit(Faulty, &rdev->flags);
2806 clear_bit(WriteMostly, &rdev->flags);
2807 set_bit(In_sync, &rdev->flags);
2808 sysfs_notify_dirent_safe(rdev->sysfs_state);
2813 static struct rdev_sysfs_entry rdev_slot =
2814 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2817 offset_show(struct md_rdev *rdev, char *page)
2819 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2823 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2825 unsigned long long offset;
2826 if (kstrtoull(buf, 10, &offset) < 0)
2828 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2830 if (rdev->sectors && rdev->mddev->external)
2831 /* Must set offset before size, so overlap checks
2834 rdev->data_offset = offset;
2835 rdev->new_data_offset = offset;
2839 static struct rdev_sysfs_entry rdev_offset =
2840 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2842 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2844 return sprintf(page, "%llu\n",
2845 (unsigned long long)rdev->new_data_offset);
2848 static ssize_t new_offset_store(struct md_rdev *rdev,
2849 const char *buf, size_t len)
2851 unsigned long long new_offset;
2852 struct mddev *mddev = rdev->mddev;
2854 if (kstrtoull(buf, 10, &new_offset) < 0)
2857 if (mddev->sync_thread ||
2858 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2860 if (new_offset == rdev->data_offset)
2861 /* reset is always permitted */
2863 else if (new_offset > rdev->data_offset) {
2864 /* must not push array size beyond rdev_sectors */
2865 if (new_offset - rdev->data_offset
2866 + mddev->dev_sectors > rdev->sectors)
2869 /* Metadata worries about other space details. */
2871 /* decreasing the offset is inconsistent with a backwards
2874 if (new_offset < rdev->data_offset &&
2875 mddev->reshape_backwards)
2877 /* Increasing offset is inconsistent with forwards
2878 * reshape. reshape_direction should be set to
2879 * 'backwards' first.
2881 if (new_offset > rdev->data_offset &&
2882 !mddev->reshape_backwards)
2885 if (mddev->pers && mddev->persistent &&
2886 !super_types[mddev->major_version]
2887 .allow_new_offset(rdev, new_offset))
2889 rdev->new_data_offset = new_offset;
2890 if (new_offset > rdev->data_offset)
2891 mddev->reshape_backwards = 1;
2892 else if (new_offset < rdev->data_offset)
2893 mddev->reshape_backwards = 0;
2897 static struct rdev_sysfs_entry rdev_new_offset =
2898 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2901 rdev_size_show(struct md_rdev *rdev, char *page)
2903 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2906 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2908 /* check if two start/length pairs overlap */
2916 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2918 unsigned long long blocks;
2921 if (kstrtoull(buf, 10, &blocks) < 0)
2924 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2925 return -EINVAL; /* sector conversion overflow */
2928 if (new != blocks * 2)
2929 return -EINVAL; /* unsigned long long to sector_t overflow */
2936 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2938 struct mddev *my_mddev = rdev->mddev;
2939 sector_t oldsectors = rdev->sectors;
2942 if (test_bit(Journal, &rdev->flags))
2944 if (strict_blocks_to_sectors(buf, §ors) < 0)
2946 if (rdev->data_offset != rdev->new_data_offset)
2947 return -EINVAL; /* too confusing */
2948 if (my_mddev->pers && rdev->raid_disk >= 0) {
2949 if (my_mddev->persistent) {
2950 sectors = super_types[my_mddev->major_version].
2951 rdev_size_change(rdev, sectors);
2954 } else if (!sectors)
2955 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2957 if (!my_mddev->pers->resize)
2958 /* Cannot change size for RAID0 or Linear etc */
2961 if (sectors < my_mddev->dev_sectors)
2962 return -EINVAL; /* component must fit device */
2964 rdev->sectors = sectors;
2965 if (sectors > oldsectors && my_mddev->external) {
2966 /* Need to check that all other rdevs with the same
2967 * ->bdev do not overlap. 'rcu' is sufficient to walk
2968 * the rdev lists safely.
2969 * This check does not provide a hard guarantee, it
2970 * just helps avoid dangerous mistakes.
2972 struct mddev *mddev;
2974 struct list_head *tmp;
2977 for_each_mddev(mddev, tmp) {
2978 struct md_rdev *rdev2;
2980 rdev_for_each(rdev2, mddev)
2981 if (rdev->bdev == rdev2->bdev &&
2983 overlaps(rdev->data_offset, rdev->sectors,
2996 /* Someone else could have slipped in a size
2997 * change here, but doing so is just silly.
2998 * We put oldsectors back because we *know* it is
2999 * safe, and trust userspace not to race with
3002 rdev->sectors = oldsectors;
3009 static struct rdev_sysfs_entry rdev_size =
3010 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3012 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3014 unsigned long long recovery_start = rdev->recovery_offset;
3016 if (test_bit(In_sync, &rdev->flags) ||
3017 recovery_start == MaxSector)
3018 return sprintf(page, "none\n");
3020 return sprintf(page, "%llu\n", recovery_start);
3023 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3025 unsigned long long recovery_start;
3027 if (cmd_match(buf, "none"))
3028 recovery_start = MaxSector;
3029 else if (kstrtoull(buf, 10, &recovery_start))
3032 if (rdev->mddev->pers &&
3033 rdev->raid_disk >= 0)
3036 rdev->recovery_offset = recovery_start;
3037 if (recovery_start == MaxSector)
3038 set_bit(In_sync, &rdev->flags);
3040 clear_bit(In_sync, &rdev->flags);
3044 static struct rdev_sysfs_entry rdev_recovery_start =
3045 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3048 badblocks_show(struct badblocks *bb, char *page, int unack);
3050 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3052 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3054 return badblocks_show(&rdev->badblocks, page, 0);
3056 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3058 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3059 /* Maybe that ack was all we needed */
3060 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3061 wake_up(&rdev->blocked_wait);
3064 static struct rdev_sysfs_entry rdev_bad_blocks =
3065 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3067 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3069 return badblocks_show(&rdev->badblocks, page, 1);
3071 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3073 return badblocks_store(&rdev->badblocks, page, len, 1);
3075 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3076 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3078 static struct attribute *rdev_default_attrs[] = {
3083 &rdev_new_offset.attr,
3085 &rdev_recovery_start.attr,
3086 &rdev_bad_blocks.attr,
3087 &rdev_unack_bad_blocks.attr,
3091 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3093 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3094 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3100 return entry->show(rdev, page);
3104 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3105 const char *page, size_t length)
3107 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3108 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3110 struct mddev *mddev = rdev->mddev;
3114 if (!capable(CAP_SYS_ADMIN))
3116 rv = mddev ? mddev_lock(mddev): -EBUSY;
3118 if (rdev->mddev == NULL)
3121 rv = entry->store(rdev, page, length);
3122 mddev_unlock(mddev);
3127 static void rdev_free(struct kobject *ko)
3129 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3132 static const struct sysfs_ops rdev_sysfs_ops = {
3133 .show = rdev_attr_show,
3134 .store = rdev_attr_store,
3136 static struct kobj_type rdev_ktype = {
3137 .release = rdev_free,
3138 .sysfs_ops = &rdev_sysfs_ops,
3139 .default_attrs = rdev_default_attrs,
3142 int md_rdev_init(struct md_rdev *rdev)
3145 rdev->saved_raid_disk = -1;
3146 rdev->raid_disk = -1;
3148 rdev->data_offset = 0;
3149 rdev->new_data_offset = 0;
3150 rdev->sb_events = 0;
3151 rdev->last_read_error.tv_sec = 0;
3152 rdev->last_read_error.tv_nsec = 0;
3153 rdev->sb_loaded = 0;
3154 rdev->bb_page = NULL;
3155 atomic_set(&rdev->nr_pending, 0);
3156 atomic_set(&rdev->read_errors, 0);
3157 atomic_set(&rdev->corrected_errors, 0);
3159 INIT_LIST_HEAD(&rdev->same_set);
3160 init_waitqueue_head(&rdev->blocked_wait);
3162 /* Add space to store bad block list.
3163 * This reserves the space even on arrays where it cannot
3164 * be used - I wonder if that matters
3166 rdev->badblocks.count = 0;
3167 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3168 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3169 seqlock_init(&rdev->badblocks.lock);
3170 if (rdev->badblocks.page == NULL)
3175 EXPORT_SYMBOL_GPL(md_rdev_init);
3177 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3179 * mark the device faulty if:
3181 * - the device is nonexistent (zero size)
3182 * - the device has no valid superblock
3184 * a faulty rdev _never_ has rdev->sb set.
3186 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3188 char b[BDEVNAME_SIZE];
3190 struct md_rdev *rdev;
3193 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3195 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3196 return ERR_PTR(-ENOMEM);
3199 err = md_rdev_init(rdev);
3202 err = alloc_disk_sb(rdev);
3206 err = lock_rdev(rdev, newdev, super_format == -2);
3210 kobject_init(&rdev->kobj, &rdev_ktype);
3212 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3215 "md: %s has zero or unknown size, marking faulty!\n",
3216 bdevname(rdev->bdev,b));
3221 if (super_format >= 0) {
3222 err = super_types[super_format].
3223 load_super(rdev, NULL, super_minor);
3224 if (err == -EINVAL) {
3226 "md: %s does not have a valid v%d.%d "
3227 "superblock, not importing!\n",
3228 bdevname(rdev->bdev,b),
3229 super_format, super_minor);
3234 "md: could not read %s's sb, not importing!\n",
3235 bdevname(rdev->bdev,b));
3245 md_rdev_clear(rdev);
3247 return ERR_PTR(err);
3251 * Check a full RAID array for plausibility
3254 static void analyze_sbs(struct mddev *mddev)
3257 struct md_rdev *rdev, *freshest, *tmp;
3258 char b[BDEVNAME_SIZE];
3261 rdev_for_each_safe(rdev, tmp, mddev)
3262 switch (super_types[mddev->major_version].
3263 load_super(rdev, freshest, mddev->minor_version)) {
3271 "md: fatal superblock inconsistency in %s"
3272 " -- removing from array\n",
3273 bdevname(rdev->bdev,b));
3274 md_kick_rdev_from_array(rdev);
3277 super_types[mddev->major_version].
3278 validate_super(mddev, freshest);
3281 rdev_for_each_safe(rdev, tmp, mddev) {
3282 if (mddev->max_disks &&
3283 (rdev->desc_nr >= mddev->max_disks ||
3284 i > mddev->max_disks)) {
3286 "md: %s: %s: only %d devices permitted\n",
3287 mdname(mddev), bdevname(rdev->bdev, b),
3289 md_kick_rdev_from_array(rdev);
3292 if (rdev != freshest) {
3293 if (super_types[mddev->major_version].
3294 validate_super(mddev, rdev)) {
3295 printk(KERN_WARNING "md: kicking non-fresh %s"
3297 bdevname(rdev->bdev,b));
3298 md_kick_rdev_from_array(rdev);
3302 if (mddev->level == LEVEL_MULTIPATH) {
3303 rdev->desc_nr = i++;
3304 rdev->raid_disk = rdev->desc_nr;
3305 set_bit(In_sync, &rdev->flags);
3306 } else if (rdev->raid_disk >=
3307 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3308 !test_bit(Journal, &rdev->flags)) {
3309 rdev->raid_disk = -1;
3310 clear_bit(In_sync, &rdev->flags);
3315 /* Read a fixed-point number.
3316 * Numbers in sysfs attributes should be in "standard" units where
3317 * possible, so time should be in seconds.
3318 * However we internally use a a much smaller unit such as
3319 * milliseconds or jiffies.
3320 * This function takes a decimal number with a possible fractional
3321 * component, and produces an integer which is the result of
3322 * multiplying that number by 10^'scale'.
3323 * all without any floating-point arithmetic.
3325 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3327 unsigned long result = 0;
3329 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3332 else if (decimals < scale) {
3335 result = result * 10 + value;
3347 while (decimals < scale) {
3356 safe_delay_show(struct mddev *mddev, char *page)
3358 int msec = (mddev->safemode_delay*1000)/HZ;
3359 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3362 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3366 if (mddev_is_clustered(mddev)) {
3367 pr_info("md: Safemode is disabled for clustered mode\n");
3371 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3374 mddev->safemode_delay = 0;
3376 unsigned long old_delay = mddev->safemode_delay;
3377 unsigned long new_delay = (msec*HZ)/1000;
3381 mddev->safemode_delay = new_delay;
3382 if (new_delay < old_delay || old_delay == 0)
3383 mod_timer(&mddev->safemode_timer, jiffies+1);
3387 static struct md_sysfs_entry md_safe_delay =
3388 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3391 level_show(struct mddev *mddev, char *page)
3393 struct md_personality *p;
3395 spin_lock(&mddev->lock);
3398 ret = sprintf(page, "%s\n", p->name);
3399 else if (mddev->clevel[0])
3400 ret = sprintf(page, "%s\n", mddev->clevel);
3401 else if (mddev->level != LEVEL_NONE)
3402 ret = sprintf(page, "%d\n", mddev->level);
3405 spin_unlock(&mddev->lock);
3410 level_store(struct mddev *mddev, const char *buf, size_t len)
3415 struct md_personality *pers, *oldpers;
3417 void *priv, *oldpriv;
3418 struct md_rdev *rdev;
3420 if (slen == 0 || slen >= sizeof(clevel))
3423 rv = mddev_lock(mddev);
3427 if (mddev->pers == NULL) {
3428 strncpy(mddev->clevel, buf, slen);
3429 if (mddev->clevel[slen-1] == '\n')
3431 mddev->clevel[slen] = 0;
3432 mddev->level = LEVEL_NONE;
3440 /* request to change the personality. Need to ensure:
3441 * - array is not engaged in resync/recovery/reshape
3442 * - old personality can be suspended
3443 * - new personality will access other array.
3447 if (mddev->sync_thread ||
3448 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3449 mddev->reshape_position != MaxSector ||
3450 mddev->sysfs_active)
3454 if (!mddev->pers->quiesce) {
3455 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3456 mdname(mddev), mddev->pers->name);
3460 /* Now find the new personality */
3461 strncpy(clevel, buf, slen);
3462 if (clevel[slen-1] == '\n')
3465 if (kstrtol(clevel, 10, &level))
3468 if (request_module("md-%s", clevel) != 0)
3469 request_module("md-level-%s", clevel);
3470 spin_lock(&pers_lock);
3471 pers = find_pers(level, clevel);
3472 if (!pers || !try_module_get(pers->owner)) {
3473 spin_unlock(&pers_lock);
3474 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3478 spin_unlock(&pers_lock);
3480 if (pers == mddev->pers) {
3481 /* Nothing to do! */
3482 module_put(pers->owner);
3486 if (!pers->takeover) {
3487 module_put(pers->owner);
3488 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3489 mdname(mddev), clevel);
3494 rdev_for_each(rdev, mddev)
3495 rdev->new_raid_disk = rdev->raid_disk;
3497 /* ->takeover must set new_* and/or delta_disks
3498 * if it succeeds, and may set them when it fails.
3500 priv = pers->takeover(mddev);
3502 mddev->new_level = mddev->level;
3503 mddev->new_layout = mddev->layout;
3504 mddev->new_chunk_sectors = mddev->chunk_sectors;
3505 mddev->raid_disks -= mddev->delta_disks;
3506 mddev->delta_disks = 0;
3507 mddev->reshape_backwards = 0;
3508 module_put(pers->owner);
3509 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3510 mdname(mddev), clevel);
3515 /* Looks like we have a winner */
3516 mddev_suspend(mddev);
3517 mddev_detach(mddev);
3519 spin_lock(&mddev->lock);
3520 oldpers = mddev->pers;
3521 oldpriv = mddev->private;
3523 mddev->private = priv;
3524 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3525 mddev->level = mddev->new_level;
3526 mddev->layout = mddev->new_layout;
3527 mddev->chunk_sectors = mddev->new_chunk_sectors;
3528 mddev->delta_disks = 0;
3529 mddev->reshape_backwards = 0;
3530 mddev->degraded = 0;
3531 spin_unlock(&mddev->lock);
3533 if (oldpers->sync_request == NULL &&
3535 /* We are converting from a no-redundancy array
3536 * to a redundancy array and metadata is managed
3537 * externally so we need to be sure that writes
3538 * won't block due to a need to transition
3540 * until external management is started.
3543 mddev->safemode_delay = 0;
3544 mddev->safemode = 0;
3547 oldpers->free(mddev, oldpriv);
3549 if (oldpers->sync_request == NULL &&
3550 pers->sync_request != NULL) {
3551 /* need to add the md_redundancy_group */
3552 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3554 "md: cannot register extra attributes for %s\n",
3556 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3558 if (oldpers->sync_request != NULL &&
3559 pers->sync_request == NULL) {
3560 /* need to remove the md_redundancy_group */
3561 if (mddev->to_remove == NULL)
3562 mddev->to_remove = &md_redundancy_group;
3565 rdev_for_each(rdev, mddev) {
3566 if (rdev->raid_disk < 0)
3568 if (rdev->new_raid_disk >= mddev->raid_disks)
3569 rdev->new_raid_disk = -1;
3570 if (rdev->new_raid_disk == rdev->raid_disk)
3572 sysfs_unlink_rdev(mddev, rdev);
3574 rdev_for_each(rdev, mddev) {
3575 if (rdev->raid_disk < 0)
3577 if (rdev->new_raid_disk == rdev->raid_disk)
3579 rdev->raid_disk = rdev->new_raid_disk;
3580 if (rdev->raid_disk < 0)
3581 clear_bit(In_sync, &rdev->flags);
3583 if (sysfs_link_rdev(mddev, rdev))
3584 printk(KERN_WARNING "md: cannot register rd%d"
3585 " for %s after level change\n",
3586 rdev->raid_disk, mdname(mddev));
3590 if (pers->sync_request == NULL) {
3591 /* this is now an array without redundancy, so
3592 * it must always be in_sync
3595 del_timer_sync(&mddev->safemode_timer);
3597 blk_set_stacking_limits(&mddev->queue->limits);
3599 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3600 mddev_resume(mddev);
3602 md_update_sb(mddev, 1);
3603 sysfs_notify(&mddev->kobj, NULL, "level");
3604 md_new_event(mddev);
3607 mddev_unlock(mddev);
3611 static struct md_sysfs_entry md_level =
3612 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3615 layout_show(struct mddev *mddev, char *page)
3617 /* just a number, not meaningful for all levels */
3618 if (mddev->reshape_position != MaxSector &&
3619 mddev->layout != mddev->new_layout)
3620 return sprintf(page, "%d (%d)\n",
3621 mddev->new_layout, mddev->layout);
3622 return sprintf(page, "%d\n", mddev->layout);
3626 layout_store(struct mddev *mddev, const char *buf, size_t len)
3631 err = kstrtouint(buf, 10, &n);
3634 err = mddev_lock(mddev);
3639 if (mddev->pers->check_reshape == NULL)
3644 mddev->new_layout = n;
3645 err = mddev->pers->check_reshape(mddev);
3647 mddev->new_layout = mddev->layout;
3650 mddev->new_layout = n;
3651 if (mddev->reshape_position == MaxSector)
3654 mddev_unlock(mddev);
3657 static struct md_sysfs_entry md_layout =
3658 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3661 raid_disks_show(struct mddev *mddev, char *page)
3663 if (mddev->raid_disks == 0)
3665 if (mddev->reshape_position != MaxSector &&
3666 mddev->delta_disks != 0)
3667 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3668 mddev->raid_disks - mddev->delta_disks);
3669 return sprintf(page, "%d\n", mddev->raid_disks);
3672 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3675 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3680 err = kstrtouint(buf, 10, &n);
3684 err = mddev_lock(mddev);
3688 err = update_raid_disks(mddev, n);
3689 else if (mddev->reshape_position != MaxSector) {
3690 struct md_rdev *rdev;
3691 int olddisks = mddev->raid_disks - mddev->delta_disks;
3694 rdev_for_each(rdev, mddev) {
3696 rdev->data_offset < rdev->new_data_offset)
3699 rdev->data_offset > rdev->new_data_offset)
3703 mddev->delta_disks = n - olddisks;
3704 mddev->raid_disks = n;
3705 mddev->reshape_backwards = (mddev->delta_disks < 0);
3707 mddev->raid_disks = n;
3709 mddev_unlock(mddev);
3710 return err ? err : len;
3712 static struct md_sysfs_entry md_raid_disks =
3713 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3716 chunk_size_show(struct mddev *mddev, char *page)
3718 if (mddev->reshape_position != MaxSector &&
3719 mddev->chunk_sectors != mddev->new_chunk_sectors)
3720 return sprintf(page, "%d (%d)\n",
3721 mddev->new_chunk_sectors << 9,
3722 mddev->chunk_sectors << 9);
3723 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3727 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3732 err = kstrtoul(buf, 10, &n);
3736 err = mddev_lock(mddev);
3740 if (mddev->pers->check_reshape == NULL)
3745 mddev->new_chunk_sectors = n >> 9;
3746 err = mddev->pers->check_reshape(mddev);
3748 mddev->new_chunk_sectors = mddev->chunk_sectors;
3751 mddev->new_chunk_sectors = n >> 9;
3752 if (mddev->reshape_position == MaxSector)
3753 mddev->chunk_sectors = n >> 9;
3755 mddev_unlock(mddev);
3758 static struct md_sysfs_entry md_chunk_size =
3759 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3762 resync_start_show(struct mddev *mddev, char *page)
3764 if (mddev->recovery_cp == MaxSector)
3765 return sprintf(page, "none\n");
3766 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3770 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3772 unsigned long long n;
3775 if (cmd_match(buf, "none"))
3778 err = kstrtoull(buf, 10, &n);
3781 if (n != (sector_t)n)
3785 err = mddev_lock(mddev);
3788 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3792 mddev->recovery_cp = n;
3794 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3796 mddev_unlock(mddev);
3799 static struct md_sysfs_entry md_resync_start =
3800 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3801 resync_start_show, resync_start_store);
3804 * The array state can be:
3807 * No devices, no size, no level
3808 * Equivalent to STOP_ARRAY ioctl
3810 * May have some settings, but array is not active
3811 * all IO results in error
3812 * When written, doesn't tear down array, but just stops it
3813 * suspended (not supported yet)
3814 * All IO requests will block. The array can be reconfigured.
3815 * Writing this, if accepted, will block until array is quiescent
3817 * no resync can happen. no superblocks get written.
3818 * write requests fail
3820 * like readonly, but behaves like 'clean' on a write request.
3822 * clean - no pending writes, but otherwise active.
3823 * When written to inactive array, starts without resync
3824 * If a write request arrives then
3825 * if metadata is known, mark 'dirty' and switch to 'active'.
3826 * if not known, block and switch to write-pending
3827 * If written to an active array that has pending writes, then fails.
3829 * fully active: IO and resync can be happening.
3830 * When written to inactive array, starts with resync
3833 * clean, but writes are blocked waiting for 'active' to be written.
3836 * like active, but no writes have been seen for a while (100msec).
3839 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3840 write_pending, active_idle, bad_word};
3841 static char *array_states[] = {
3842 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3843 "write-pending", "active-idle", NULL };
3845 static int match_word(const char *word, char **list)
3848 for (n=0; list[n]; n++)
3849 if (cmd_match(word, list[n]))
3855 array_state_show(struct mddev *mddev, char *page)
3857 enum array_state st = inactive;
3870 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3872 else if (mddev->safemode)
3878 if (list_empty(&mddev->disks) &&
3879 mddev->raid_disks == 0 &&
3880 mddev->dev_sectors == 0)
3885 return sprintf(page, "%s\n", array_states[st]);
3888 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3889 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3890 static int do_md_run(struct mddev *mddev);
3891 static int restart_array(struct mddev *mddev);
3894 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3897 enum array_state st = match_word(buf, array_states);
3899 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3900 /* don't take reconfig_mutex when toggling between
3903 spin_lock(&mddev->lock);
3905 restart_array(mddev);
3906 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3907 wake_up(&mddev->sb_wait);
3909 } else /* st == clean */ {
3910 restart_array(mddev);
3911 if (atomic_read(&mddev->writes_pending) == 0) {
3912 if (mddev->in_sync == 0) {
3914 if (mddev->safemode == 1)
3915 mddev->safemode = 0;
3916 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3922 spin_unlock(&mddev->lock);
3925 err = mddev_lock(mddev);
3933 /* stopping an active array */
3934 err = do_md_stop(mddev, 0, NULL);
3937 /* stopping an active array */
3939 err = do_md_stop(mddev, 2, NULL);
3941 err = 0; /* already inactive */
3944 break; /* not supported yet */
3947 err = md_set_readonly(mddev, NULL);
3950 set_disk_ro(mddev->gendisk, 1);
3951 err = do_md_run(mddev);
3957 err = md_set_readonly(mddev, NULL);
3958 else if (mddev->ro == 1)
3959 err = restart_array(mddev);
3962 set_disk_ro(mddev->gendisk, 0);
3966 err = do_md_run(mddev);
3971 err = restart_array(mddev);
3974 spin_lock(&mddev->lock);
3975 if (atomic_read(&mddev->writes_pending) == 0) {
3976 if (mddev->in_sync == 0) {
3978 if (mddev->safemode == 1)
3979 mddev->safemode = 0;
3980 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3985 spin_unlock(&mddev->lock);
3991 err = restart_array(mddev);
3994 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3995 wake_up(&mddev->sb_wait);
3999 set_disk_ro(mddev->gendisk, 0);
4000 err = do_md_run(mddev);
4005 /* these cannot be set */
4010 if (mddev->hold_active == UNTIL_IOCTL)
4011 mddev->hold_active = 0;
4012 sysfs_notify_dirent_safe(mddev->sysfs_state);
4014 mddev_unlock(mddev);
4017 static struct md_sysfs_entry md_array_state =
4018 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4021 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4022 return sprintf(page, "%d\n",
4023 atomic_read(&mddev->max_corr_read_errors));
4027 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4032 rv = kstrtouint(buf, 10, &n);
4035 atomic_set(&mddev->max_corr_read_errors, n);
4039 static struct md_sysfs_entry max_corr_read_errors =
4040 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4041 max_corrected_read_errors_store);
4044 null_show(struct mddev *mddev, char *page)
4050 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4052 /* buf must be %d:%d\n? giving major and minor numbers */
4053 /* The new device is added to the array.
4054 * If the array has a persistent superblock, we read the
4055 * superblock to initialise info and check validity.
4056 * Otherwise, only checking done is that in bind_rdev_to_array,
4057 * which mainly checks size.
4060 int major = simple_strtoul(buf, &e, 10);
4063 struct md_rdev *rdev;
4066 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4068 minor = simple_strtoul(e+1, &e, 10);
4069 if (*e && *e != '\n')
4071 dev = MKDEV(major, minor);
4072 if (major != MAJOR(dev) ||
4073 minor != MINOR(dev))
4076 flush_workqueue(md_misc_wq);
4078 err = mddev_lock(mddev);
4081 if (mddev->persistent) {
4082 rdev = md_import_device(dev, mddev->major_version,
4083 mddev->minor_version);
4084 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4085 struct md_rdev *rdev0
4086 = list_entry(mddev->disks.next,
4087 struct md_rdev, same_set);
4088 err = super_types[mddev->major_version]
4089 .load_super(rdev, rdev0, mddev->minor_version);
4093 } else if (mddev->external)
4094 rdev = md_import_device(dev, -2, -1);
4096 rdev = md_import_device(dev, -1, -1);
4099 mddev_unlock(mddev);
4100 return PTR_ERR(rdev);
4102 err = bind_rdev_to_array(rdev, mddev);
4106 mddev_unlock(mddev);
4107 return err ? err : len;
4110 static struct md_sysfs_entry md_new_device =
4111 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4114 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4117 unsigned long chunk, end_chunk;
4120 err = mddev_lock(mddev);
4125 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4127 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4128 if (buf == end) break;
4129 if (*end == '-') { /* range */
4131 end_chunk = simple_strtoul(buf, &end, 0);
4132 if (buf == end) break;
4134 if (*end && !isspace(*end)) break;
4135 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4136 buf = skip_spaces(end);
4138 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4140 mddev_unlock(mddev);
4144 static struct md_sysfs_entry md_bitmap =
4145 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4148 size_show(struct mddev *mddev, char *page)
4150 return sprintf(page, "%llu\n",
4151 (unsigned long long)mddev->dev_sectors / 2);
4154 static int update_size(struct mddev *mddev, sector_t num_sectors);
4157 size_store(struct mddev *mddev, const char *buf, size_t len)
4159 /* If array is inactive, we can reduce the component size, but
4160 * not increase it (except from 0).
4161 * If array is active, we can try an on-line resize
4164 int err = strict_blocks_to_sectors(buf, §ors);
4168 err = mddev_lock(mddev);
4172 err = update_size(mddev, sectors);
4173 md_update_sb(mddev, 1);
4175 if (mddev->dev_sectors == 0 ||
4176 mddev->dev_sectors > sectors)
4177 mddev->dev_sectors = sectors;
4181 mddev_unlock(mddev);
4182 return err ? err : len;
4185 static struct md_sysfs_entry md_size =
4186 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4188 /* Metadata version.
4190 * 'none' for arrays with no metadata (good luck...)
4191 * 'external' for arrays with externally managed metadata,
4192 * or N.M for internally known formats
4195 metadata_show(struct mddev *mddev, char *page)
4197 if (mddev->persistent)
4198 return sprintf(page, "%d.%d\n",
4199 mddev->major_version, mddev->minor_version);
4200 else if (mddev->external)
4201 return sprintf(page, "external:%s\n", mddev->metadata_type);
4203 return sprintf(page, "none\n");
4207 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4212 /* Changing the details of 'external' metadata is
4213 * always permitted. Otherwise there must be
4214 * no devices attached to the array.
4217 err = mddev_lock(mddev);
4221 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4223 else if (!list_empty(&mddev->disks))
4227 if (cmd_match(buf, "none")) {
4228 mddev->persistent = 0;
4229 mddev->external = 0;
4230 mddev->major_version = 0;
4231 mddev->minor_version = 90;
4234 if (strncmp(buf, "external:", 9) == 0) {
4235 size_t namelen = len-9;
4236 if (namelen >= sizeof(mddev->metadata_type))
4237 namelen = sizeof(mddev->metadata_type)-1;
4238 strncpy(mddev->metadata_type, buf+9, namelen);
4239 mddev->metadata_type[namelen] = 0;
4240 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4241 mddev->metadata_type[--namelen] = 0;
4242 mddev->persistent = 0;
4243 mddev->external = 1;
4244 mddev->major_version = 0;
4245 mddev->minor_version = 90;
4248 major = simple_strtoul(buf, &e, 10);
4250 if (e==buf || *e != '.')
4253 minor = simple_strtoul(buf, &e, 10);
4254 if (e==buf || (*e && *e != '\n') )
4257 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4259 mddev->major_version = major;
4260 mddev->minor_version = minor;
4261 mddev->persistent = 1;
4262 mddev->external = 0;
4265 mddev_unlock(mddev);
4269 static struct md_sysfs_entry md_metadata =
4270 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4273 action_show(struct mddev *mddev, char *page)
4275 char *type = "idle";
4276 unsigned long recovery = mddev->recovery;
4277 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4279 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4280 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4281 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4283 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4284 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4286 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4290 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4292 else if (mddev->reshape_position != MaxSector)
4295 return sprintf(page, "%s\n", type);
4299 action_store(struct mddev *mddev, const char *page, size_t len)
4301 if (!mddev->pers || !mddev->pers->sync_request)
4305 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4306 if (cmd_match(page, "frozen"))
4307 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4309 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4310 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4311 mddev_lock(mddev) == 0) {
4312 flush_workqueue(md_misc_wq);
4313 if (mddev->sync_thread) {
4314 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4315 md_reap_sync_thread(mddev);
4317 mddev_unlock(mddev);
4319 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4320 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4322 else if (cmd_match(page, "resync"))
4323 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4324 else if (cmd_match(page, "recover")) {
4325 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4326 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4327 } else if (cmd_match(page, "reshape")) {
4329 if (mddev->pers->start_reshape == NULL)
4331 err = mddev_lock(mddev);
4333 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4334 err = mddev->pers->start_reshape(mddev);
4335 mddev_unlock(mddev);
4339 sysfs_notify(&mddev->kobj, NULL, "degraded");
4341 if (cmd_match(page, "check"))
4342 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4343 else if (!cmd_match(page, "repair"))
4345 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4346 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4347 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4349 if (mddev->ro == 2) {
4350 /* A write to sync_action is enough to justify
4351 * canceling read-auto mode
4354 md_wakeup_thread(mddev->sync_thread);
4356 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4357 md_wakeup_thread(mddev->thread);
4358 sysfs_notify_dirent_safe(mddev->sysfs_action);
4362 static struct md_sysfs_entry md_scan_mode =
4363 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4366 last_sync_action_show(struct mddev *mddev, char *page)
4368 return sprintf(page, "%s\n", mddev->last_sync_action);
4371 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4374 mismatch_cnt_show(struct mddev *mddev, char *page)
4376 return sprintf(page, "%llu\n",
4377 (unsigned long long)
4378 atomic64_read(&mddev->resync_mismatches));
4381 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4384 sync_min_show(struct mddev *mddev, char *page)
4386 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4387 mddev->sync_speed_min ? "local": "system");
4391 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4396 if (strncmp(buf, "system", 6)==0) {
4399 rv = kstrtouint(buf, 10, &min);
4405 mddev->sync_speed_min = min;
4409 static struct md_sysfs_entry md_sync_min =
4410 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4413 sync_max_show(struct mddev *mddev, char *page)
4415 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4416 mddev->sync_speed_max ? "local": "system");
4420 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4425 if (strncmp(buf, "system", 6)==0) {
4428 rv = kstrtouint(buf, 10, &max);
4434 mddev->sync_speed_max = max;
4438 static struct md_sysfs_entry md_sync_max =
4439 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4442 degraded_show(struct mddev *mddev, char *page)
4444 return sprintf(page, "%d\n", mddev->degraded);
4446 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4449 sync_force_parallel_show(struct mddev *mddev, char *page)
4451 return sprintf(page, "%d\n", mddev->parallel_resync);
4455 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4459 if (kstrtol(buf, 10, &n))
4462 if (n != 0 && n != 1)
4465 mddev->parallel_resync = n;
4467 if (mddev->sync_thread)
4468 wake_up(&resync_wait);
4473 /* force parallel resync, even with shared block devices */
4474 static struct md_sysfs_entry md_sync_force_parallel =
4475 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4476 sync_force_parallel_show, sync_force_parallel_store);
4479 sync_speed_show(struct mddev *mddev, char *page)
4481 unsigned long resync, dt, db;
4482 if (mddev->curr_resync == 0)
4483 return sprintf(page, "none\n");
4484 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4485 dt = (jiffies - mddev->resync_mark) / HZ;
4487 db = resync - mddev->resync_mark_cnt;
4488 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4491 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4494 sync_completed_show(struct mddev *mddev, char *page)
4496 unsigned long long max_sectors, resync;
4498 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4499 return sprintf(page, "none\n");
4501 if (mddev->curr_resync == 1 ||
4502 mddev->curr_resync == 2)
4503 return sprintf(page, "delayed\n");
4505 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4506 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4507 max_sectors = mddev->resync_max_sectors;
4509 max_sectors = mddev->dev_sectors;
4511 resync = mddev->curr_resync_completed;
4512 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4515 static struct md_sysfs_entry md_sync_completed =
4516 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4519 min_sync_show(struct mddev *mddev, char *page)
4521 return sprintf(page, "%llu\n",
4522 (unsigned long long)mddev->resync_min);
4525 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4527 unsigned long long min;
4530 if (kstrtoull(buf, 10, &min))
4533 spin_lock(&mddev->lock);
4535 if (min > mddev->resync_max)
4539 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4542 /* Round down to multiple of 4K for safety */
4543 mddev->resync_min = round_down(min, 8);
4547 spin_unlock(&mddev->lock);
4551 static struct md_sysfs_entry md_min_sync =
4552 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4555 max_sync_show(struct mddev *mddev, char *page)
4557 if (mddev->resync_max == MaxSector)
4558 return sprintf(page, "max\n");
4560 return sprintf(page, "%llu\n",
4561 (unsigned long long)mddev->resync_max);
4564 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4567 spin_lock(&mddev->lock);
4568 if (strncmp(buf, "max", 3) == 0)
4569 mddev->resync_max = MaxSector;
4571 unsigned long long max;
4575 if (kstrtoull(buf, 10, &max))
4577 if (max < mddev->resync_min)
4581 if (max < mddev->resync_max &&
4583 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4586 /* Must be a multiple of chunk_size */
4587 chunk = mddev->chunk_sectors;
4589 sector_t temp = max;
4592 if (sector_div(temp, chunk))
4595 mddev->resync_max = max;
4597 wake_up(&mddev->recovery_wait);
4600 spin_unlock(&mddev->lock);
4604 static struct md_sysfs_entry md_max_sync =
4605 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4608 suspend_lo_show(struct mddev *mddev, char *page)
4610 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4614 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4616 unsigned long long old, new;
4619 err = kstrtoull(buf, 10, &new);
4622 if (new != (sector_t)new)
4625 err = mddev_lock(mddev);
4629 if (mddev->pers == NULL ||
4630 mddev->pers->quiesce == NULL)
4632 old = mddev->suspend_lo;
4633 mddev->suspend_lo = new;
4635 /* Shrinking suspended region */
4636 mddev->pers->quiesce(mddev, 2);
4638 /* Expanding suspended region - need to wait */
4639 mddev->pers->quiesce(mddev, 1);
4640 mddev->pers->quiesce(mddev, 0);
4644 mddev_unlock(mddev);
4647 static struct md_sysfs_entry md_suspend_lo =
4648 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4651 suspend_hi_show(struct mddev *mddev, char *page)
4653 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4657 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4659 unsigned long long old, new;
4662 err = kstrtoull(buf, 10, &new);
4665 if (new != (sector_t)new)
4668 err = mddev_lock(mddev);
4672 if (mddev->pers == NULL ||
4673 mddev->pers->quiesce == NULL)
4675 old = mddev->suspend_hi;
4676 mddev->suspend_hi = new;
4678 /* Shrinking suspended region */
4679 mddev->pers->quiesce(mddev, 2);
4681 /* Expanding suspended region - need to wait */
4682 mddev->pers->quiesce(mddev, 1);
4683 mddev->pers->quiesce(mddev, 0);
4687 mddev_unlock(mddev);
4690 static struct md_sysfs_entry md_suspend_hi =
4691 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4694 reshape_position_show(struct mddev *mddev, char *page)
4696 if (mddev->reshape_position != MaxSector)
4697 return sprintf(page, "%llu\n",
4698 (unsigned long long)mddev->reshape_position);
4699 strcpy(page, "none\n");
4704 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4706 struct md_rdev *rdev;
4707 unsigned long long new;
4710 err = kstrtoull(buf, 10, &new);
4713 if (new != (sector_t)new)
4715 err = mddev_lock(mddev);
4721 mddev->reshape_position = new;
4722 mddev->delta_disks = 0;
4723 mddev->reshape_backwards = 0;
4724 mddev->new_level = mddev->level;
4725 mddev->new_layout = mddev->layout;
4726 mddev->new_chunk_sectors = mddev->chunk_sectors;
4727 rdev_for_each(rdev, mddev)
4728 rdev->new_data_offset = rdev->data_offset;
4731 mddev_unlock(mddev);
4735 static struct md_sysfs_entry md_reshape_position =
4736 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4737 reshape_position_store);
4740 reshape_direction_show(struct mddev *mddev, char *page)
4742 return sprintf(page, "%s\n",
4743 mddev->reshape_backwards ? "backwards" : "forwards");
4747 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4752 if (cmd_match(buf, "forwards"))
4754 else if (cmd_match(buf, "backwards"))
4758 if (mddev->reshape_backwards == backwards)
4761 err = mddev_lock(mddev);
4764 /* check if we are allowed to change */
4765 if (mddev->delta_disks)
4767 else if (mddev->persistent &&
4768 mddev->major_version == 0)
4771 mddev->reshape_backwards = backwards;
4772 mddev_unlock(mddev);
4776 static struct md_sysfs_entry md_reshape_direction =
4777 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4778 reshape_direction_store);
4781 array_size_show(struct mddev *mddev, char *page)
4783 if (mddev->external_size)
4784 return sprintf(page, "%llu\n",
4785 (unsigned long long)mddev->array_sectors/2);
4787 return sprintf(page, "default\n");
4791 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4796 err = mddev_lock(mddev);
4800 if (strncmp(buf, "default", 7) == 0) {
4802 sectors = mddev->pers->size(mddev, 0, 0);
4804 sectors = mddev->array_sectors;
4806 mddev->external_size = 0;
4808 if (strict_blocks_to_sectors(buf, §ors) < 0)
4810 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4813 mddev->external_size = 1;
4817 mddev->array_sectors = sectors;
4819 set_capacity(mddev->gendisk, mddev->array_sectors);
4820 revalidate_disk(mddev->gendisk);
4823 mddev_unlock(mddev);
4827 static struct md_sysfs_entry md_array_size =
4828 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4831 static struct attribute *md_default_attrs[] = {
4834 &md_raid_disks.attr,
4835 &md_chunk_size.attr,
4837 &md_resync_start.attr,
4839 &md_new_device.attr,
4840 &md_safe_delay.attr,
4841 &md_array_state.attr,
4842 &md_reshape_position.attr,
4843 &md_reshape_direction.attr,
4844 &md_array_size.attr,
4845 &max_corr_read_errors.attr,
4849 static struct attribute *md_redundancy_attrs[] = {
4851 &md_last_scan_mode.attr,
4852 &md_mismatches.attr,
4855 &md_sync_speed.attr,
4856 &md_sync_force_parallel.attr,
4857 &md_sync_completed.attr,
4860 &md_suspend_lo.attr,
4861 &md_suspend_hi.attr,
4866 static struct attribute_group md_redundancy_group = {
4868 .attrs = md_redundancy_attrs,
4872 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4874 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4875 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4880 spin_lock(&all_mddevs_lock);
4881 if (list_empty(&mddev->all_mddevs)) {
4882 spin_unlock(&all_mddevs_lock);
4886 spin_unlock(&all_mddevs_lock);
4888 rv = entry->show(mddev, page);
4894 md_attr_store(struct kobject *kobj, struct attribute *attr,
4895 const char *page, size_t length)
4897 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4898 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4903 if (!capable(CAP_SYS_ADMIN))
4905 spin_lock(&all_mddevs_lock);
4906 if (list_empty(&mddev->all_mddevs)) {
4907 spin_unlock(&all_mddevs_lock);
4911 spin_unlock(&all_mddevs_lock);
4912 rv = entry->store(mddev, page, length);
4917 static void md_free(struct kobject *ko)
4919 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4921 if (mddev->sysfs_state)
4922 sysfs_put(mddev->sysfs_state);
4925 blk_cleanup_queue(mddev->queue);
4926 if (mddev->gendisk) {
4927 del_gendisk(mddev->gendisk);
4928 put_disk(mddev->gendisk);
4934 static const struct sysfs_ops md_sysfs_ops = {
4935 .show = md_attr_show,
4936 .store = md_attr_store,
4938 static struct kobj_type md_ktype = {
4940 .sysfs_ops = &md_sysfs_ops,
4941 .default_attrs = md_default_attrs,
4946 static void mddev_delayed_delete(struct work_struct *ws)
4948 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4950 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4951 kobject_del(&mddev->kobj);
4952 kobject_put(&mddev->kobj);
4955 static int md_alloc(dev_t dev, char *name)
4957 static DEFINE_MUTEX(disks_mutex);
4958 struct mddev *mddev = mddev_find(dev);
4959 struct gendisk *disk;
4968 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4969 shift = partitioned ? MdpMinorShift : 0;
4970 unit = MINOR(mddev->unit) >> shift;
4972 /* wait for any previous instance of this device to be
4973 * completely removed (mddev_delayed_delete).
4975 flush_workqueue(md_misc_wq);
4977 mutex_lock(&disks_mutex);
4983 /* Need to ensure that 'name' is not a duplicate.
4985 struct mddev *mddev2;
4986 spin_lock(&all_mddevs_lock);
4988 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4989 if (mddev2->gendisk &&
4990 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4991 spin_unlock(&all_mddevs_lock);
4994 spin_unlock(&all_mddevs_lock);
4998 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5001 mddev->queue->queuedata = mddev;
5003 blk_queue_make_request(mddev->queue, md_make_request);
5004 blk_set_stacking_limits(&mddev->queue->limits);
5006 disk = alloc_disk(1 << shift);
5008 blk_cleanup_queue(mddev->queue);
5009 mddev->queue = NULL;
5012 disk->major = MAJOR(mddev->unit);
5013 disk->first_minor = unit << shift;
5015 strcpy(disk->disk_name, name);
5016 else if (partitioned)
5017 sprintf(disk->disk_name, "md_d%d", unit);
5019 sprintf(disk->disk_name, "md%d", unit);
5020 disk->fops = &md_fops;
5021 disk->private_data = mddev;
5022 disk->queue = mddev->queue;
5023 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5024 /* Allow extended partitions. This makes the
5025 * 'mdp' device redundant, but we can't really
5028 disk->flags |= GENHD_FL_EXT_DEVT;
5029 mddev->gendisk = disk;
5030 /* As soon as we call add_disk(), another thread could get
5031 * through to md_open, so make sure it doesn't get too far
5033 mutex_lock(&mddev->open_mutex);
5036 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5037 &disk_to_dev(disk)->kobj, "%s", "md");
5039 /* This isn't possible, but as kobject_init_and_add is marked
5040 * __must_check, we must do something with the result
5042 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5046 if (mddev->kobj.sd &&
5047 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5048 printk(KERN_DEBUG "pointless warning\n");
5049 mutex_unlock(&mddev->open_mutex);
5051 mutex_unlock(&disks_mutex);
5052 if (!error && mddev->kobj.sd) {
5053 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5054 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5060 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5062 md_alloc(dev, NULL);
5066 static int add_named_array(const char *val, struct kernel_param *kp)
5068 /* val must be "md_*" where * is not all digits.
5069 * We allocate an array with a large free minor number, and
5070 * set the name to val. val must not already be an active name.
5072 int len = strlen(val);
5073 char buf[DISK_NAME_LEN];
5075 while (len && val[len-1] == '\n')
5077 if (len >= DISK_NAME_LEN)
5079 strlcpy(buf, val, len+1);
5080 if (strncmp(buf, "md_", 3) != 0)
5082 return md_alloc(0, buf);
5085 static void md_safemode_timeout(unsigned long data)
5087 struct mddev *mddev = (struct mddev *) data;
5089 if (!atomic_read(&mddev->writes_pending)) {
5090 mddev->safemode = 1;
5091 if (mddev->external)
5092 sysfs_notify_dirent_safe(mddev->sysfs_state);
5094 md_wakeup_thread(mddev->thread);
5097 static int start_dirty_degraded;
5099 int md_run(struct mddev *mddev)
5102 struct md_rdev *rdev;
5103 struct md_personality *pers;
5105 if (list_empty(&mddev->disks))
5106 /* cannot run an array with no devices.. */
5111 /* Cannot run until previous stop completes properly */
5112 if (mddev->sysfs_active)
5116 * Analyze all RAID superblock(s)
5118 if (!mddev->raid_disks) {
5119 if (!mddev->persistent)
5124 if (mddev->level != LEVEL_NONE)
5125 request_module("md-level-%d", mddev->level);
5126 else if (mddev->clevel[0])
5127 request_module("md-%s", mddev->clevel);
5130 * Drop all container device buffers, from now on
5131 * the only valid external interface is through the md
5134 rdev_for_each(rdev, mddev) {
5135 if (test_bit(Faulty, &rdev->flags))
5137 sync_blockdev(rdev->bdev);
5138 invalidate_bdev(rdev->bdev);
5140 /* perform some consistency tests on the device.
5141 * We don't want the data to overlap the metadata,
5142 * Internal Bitmap issues have been handled elsewhere.
5144 if (rdev->meta_bdev) {
5145 /* Nothing to check */;
5146 } else if (rdev->data_offset < rdev->sb_start) {
5147 if (mddev->dev_sectors &&
5148 rdev->data_offset + mddev->dev_sectors
5150 printk("md: %s: data overlaps metadata\n",
5155 if (rdev->sb_start + rdev->sb_size/512
5156 > rdev->data_offset) {
5157 printk("md: %s: metadata overlaps data\n",
5162 sysfs_notify_dirent_safe(rdev->sysfs_state);
5165 if (mddev->bio_set == NULL)
5166 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5168 spin_lock(&pers_lock);
5169 pers = find_pers(mddev->level, mddev->clevel);
5170 if (!pers || !try_module_get(pers->owner)) {
5171 spin_unlock(&pers_lock);
5172 if (mddev->level != LEVEL_NONE)
5173 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5176 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5180 spin_unlock(&pers_lock);
5181 if (mddev->level != pers->level) {
5182 mddev->level = pers->level;
5183 mddev->new_level = pers->level;
5185 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5187 if (mddev->reshape_position != MaxSector &&
5188 pers->start_reshape == NULL) {
5189 /* This personality cannot handle reshaping... */
5190 module_put(pers->owner);
5194 if (pers->sync_request) {
5195 /* Warn if this is a potentially silly
5198 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5199 struct md_rdev *rdev2;
5202 rdev_for_each(rdev, mddev)
5203 rdev_for_each(rdev2, mddev) {
5205 rdev->bdev->bd_contains ==
5206 rdev2->bdev->bd_contains) {
5208 "%s: WARNING: %s appears to be"
5209 " on the same physical disk as"
5212 bdevname(rdev->bdev,b),
5213 bdevname(rdev2->bdev,b2));
5220 "True protection against single-disk"
5221 " failure might be compromised.\n");
5224 mddev->recovery = 0;
5225 /* may be over-ridden by personality */
5226 mddev->resync_max_sectors = mddev->dev_sectors;
5228 mddev->ok_start_degraded = start_dirty_degraded;
5230 if (start_readonly && mddev->ro == 0)
5231 mddev->ro = 2; /* read-only, but switch on first write */
5233 err = pers->run(mddev);
5235 printk(KERN_ERR "md: pers->run() failed ...\n");
5236 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5237 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5238 " but 'external_size' not in effect?\n", __func__);
5240 "md: invalid array_size %llu > default size %llu\n",
5241 (unsigned long long)mddev->array_sectors / 2,
5242 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5245 if (err == 0 && pers->sync_request &&
5246 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5247 struct bitmap *bitmap;
5249 bitmap = bitmap_create(mddev, -1);
5250 if (IS_ERR(bitmap)) {
5251 err = PTR_ERR(bitmap);
5252 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5253 mdname(mddev), err);
5255 mddev->bitmap = bitmap;
5259 mddev_detach(mddev);
5261 pers->free(mddev, mddev->private);
5262 mddev->private = NULL;
5263 module_put(pers->owner);
5264 bitmap_destroy(mddev);
5268 mddev->queue->backing_dev_info.congested_data = mddev;
5269 mddev->queue->backing_dev_info.congested_fn = md_congested;
5271 if (pers->sync_request) {
5272 if (mddev->kobj.sd &&
5273 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5275 "md: cannot register extra attributes for %s\n",
5277 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5278 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5281 atomic_set(&mddev->writes_pending,0);
5282 atomic_set(&mddev->max_corr_read_errors,
5283 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5284 mddev->safemode = 0;
5285 if (mddev_is_clustered(mddev))
5286 mddev->safemode_delay = 0;
5288 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5291 spin_lock(&mddev->lock);
5294 spin_unlock(&mddev->lock);
5295 rdev_for_each(rdev, mddev)
5296 if (rdev->raid_disk >= 0)
5297 if (sysfs_link_rdev(mddev, rdev))
5298 /* failure here is OK */;
5300 if (mddev->degraded && !mddev->ro)
5301 /* This ensures that recovering status is reported immediately
5302 * via sysfs - until a lack of spares is confirmed.
5304 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5305 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5307 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5308 md_update_sb(mddev, 0);
5310 md_new_event(mddev);
5311 sysfs_notify_dirent_safe(mddev->sysfs_state);
5312 sysfs_notify_dirent_safe(mddev->sysfs_action);
5313 sysfs_notify(&mddev->kobj, NULL, "degraded");
5316 EXPORT_SYMBOL_GPL(md_run);
5318 static int do_md_run(struct mddev *mddev)
5322 err = md_run(mddev);
5325 err = bitmap_load(mddev);
5327 bitmap_destroy(mddev);
5331 if (mddev_is_clustered(mddev))
5332 md_allow_write(mddev);
5334 md_wakeup_thread(mddev->thread);
5335 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5337 set_capacity(mddev->gendisk, mddev->array_sectors);
5338 revalidate_disk(mddev->gendisk);
5340 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5345 static int restart_array(struct mddev *mddev)
5347 struct gendisk *disk = mddev->gendisk;
5349 /* Complain if it has no devices */
5350 if (list_empty(&mddev->disks))
5356 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5357 struct md_rdev *rdev;
5358 bool has_journal = false;
5361 rdev_for_each_rcu(rdev, mddev) {
5362 if (test_bit(Journal, &rdev->flags) &&
5363 !test_bit(Faulty, &rdev->flags)) {
5370 /* Don't restart rw with journal missing/faulty */
5375 mddev->safemode = 0;
5377 set_disk_ro(disk, 0);
5378 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5380 /* Kick recovery or resync if necessary */
5381 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5382 md_wakeup_thread(mddev->thread);
5383 md_wakeup_thread(mddev->sync_thread);
5384 sysfs_notify_dirent_safe(mddev->sysfs_state);
5388 static void md_clean(struct mddev *mddev)
5390 mddev->array_sectors = 0;
5391 mddev->external_size = 0;
5392 mddev->dev_sectors = 0;
5393 mddev->raid_disks = 0;
5394 mddev->recovery_cp = 0;
5395 mddev->resync_min = 0;
5396 mddev->resync_max = MaxSector;
5397 mddev->reshape_position = MaxSector;
5398 mddev->external = 0;
5399 mddev->persistent = 0;
5400 mddev->level = LEVEL_NONE;
5401 mddev->clevel[0] = 0;
5404 mddev->metadata_type[0] = 0;
5405 mddev->chunk_sectors = 0;
5406 mddev->ctime = mddev->utime = 0;
5408 mddev->max_disks = 0;
5410 mddev->can_decrease_events = 0;
5411 mddev->delta_disks = 0;
5412 mddev->reshape_backwards = 0;
5413 mddev->new_level = LEVEL_NONE;
5414 mddev->new_layout = 0;
5415 mddev->new_chunk_sectors = 0;
5416 mddev->curr_resync = 0;
5417 atomic64_set(&mddev->resync_mismatches, 0);
5418 mddev->suspend_lo = mddev->suspend_hi = 0;
5419 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5420 mddev->recovery = 0;
5423 mddev->degraded = 0;
5424 mddev->safemode = 0;
5425 mddev->private = NULL;
5426 mddev->bitmap_info.offset = 0;
5427 mddev->bitmap_info.default_offset = 0;
5428 mddev->bitmap_info.default_space = 0;
5429 mddev->bitmap_info.chunksize = 0;
5430 mddev->bitmap_info.daemon_sleep = 0;
5431 mddev->bitmap_info.max_write_behind = 0;
5434 static void __md_stop_writes(struct mddev *mddev)
5436 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5437 flush_workqueue(md_misc_wq);
5438 if (mddev->sync_thread) {
5439 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5440 md_reap_sync_thread(mddev);
5443 del_timer_sync(&mddev->safemode_timer);
5445 bitmap_flush(mddev);
5446 md_super_wait(mddev);
5448 if (mddev->ro == 0 &&
5449 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5450 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5451 /* mark array as shutdown cleanly */
5452 if (!mddev_is_clustered(mddev))
5454 md_update_sb(mddev, 1);
5458 void md_stop_writes(struct mddev *mddev)
5460 mddev_lock_nointr(mddev);
5461 __md_stop_writes(mddev);
5462 mddev_unlock(mddev);
5464 EXPORT_SYMBOL_GPL(md_stop_writes);
5466 static void mddev_detach(struct mddev *mddev)
5468 struct bitmap *bitmap = mddev->bitmap;
5469 /* wait for behind writes to complete */
5470 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5471 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5473 /* need to kick something here to make sure I/O goes? */
5474 wait_event(bitmap->behind_wait,
5475 atomic_read(&bitmap->behind_writes) == 0);
5477 if (mddev->pers && mddev->pers->quiesce) {
5478 mddev->pers->quiesce(mddev, 1);
5479 mddev->pers->quiesce(mddev, 0);
5481 md_unregister_thread(&mddev->thread);
5483 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5486 static void __md_stop(struct mddev *mddev)
5488 struct md_personality *pers = mddev->pers;
5489 mddev_detach(mddev);
5490 /* Ensure ->event_work is done */
5491 flush_workqueue(md_misc_wq);
5492 spin_lock(&mddev->lock);
5495 spin_unlock(&mddev->lock);
5496 pers->free(mddev, mddev->private);
5497 mddev->private = NULL;
5498 if (pers->sync_request && mddev->to_remove == NULL)
5499 mddev->to_remove = &md_redundancy_group;
5500 module_put(pers->owner);
5501 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5504 void md_stop(struct mddev *mddev)
5506 /* stop the array and free an attached data structures.
5507 * This is called from dm-raid
5510 bitmap_destroy(mddev);
5512 bioset_free(mddev->bio_set);
5515 EXPORT_SYMBOL_GPL(md_stop);
5517 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5522 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5524 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5525 md_wakeup_thread(mddev->thread);
5527 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5528 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5529 if (mddev->sync_thread)
5530 /* Thread might be blocked waiting for metadata update
5531 * which will now never happen */
5532 wake_up_process(mddev->sync_thread->tsk);
5534 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5536 mddev_unlock(mddev);
5537 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5539 wait_event(mddev->sb_wait,
5540 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5541 mddev_lock_nointr(mddev);
5543 mutex_lock(&mddev->open_mutex);
5544 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5545 mddev->sync_thread ||
5546 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5547 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5548 printk("md: %s still in use.\n",mdname(mddev));
5550 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5551 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5552 md_wakeup_thread(mddev->thread);
5558 __md_stop_writes(mddev);
5564 set_disk_ro(mddev->gendisk, 1);
5565 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5566 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5567 md_wakeup_thread(mddev->thread);
5568 sysfs_notify_dirent_safe(mddev->sysfs_state);
5572 mutex_unlock(&mddev->open_mutex);
5577 * 0 - completely stop and dis-assemble array
5578 * 2 - stop but do not disassemble array
5580 static int do_md_stop(struct mddev *mddev, int mode,
5581 struct block_device *bdev)
5583 struct gendisk *disk = mddev->gendisk;
5584 struct md_rdev *rdev;
5587 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5589 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5590 md_wakeup_thread(mddev->thread);
5592 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5593 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5594 if (mddev->sync_thread)
5595 /* Thread might be blocked waiting for metadata update
5596 * which will now never happen */
5597 wake_up_process(mddev->sync_thread->tsk);
5599 mddev_unlock(mddev);
5600 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5601 !test_bit(MD_RECOVERY_RUNNING,
5602 &mddev->recovery)));
5603 mddev_lock_nointr(mddev);
5605 mutex_lock(&mddev->open_mutex);
5606 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5607 mddev->sysfs_active ||
5608 mddev->sync_thread ||
5609 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5610 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5611 printk("md: %s still in use.\n",mdname(mddev));
5612 mutex_unlock(&mddev->open_mutex);
5614 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5615 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5616 md_wakeup_thread(mddev->thread);
5622 set_disk_ro(disk, 0);
5624 __md_stop_writes(mddev);
5626 mddev->queue->backing_dev_info.congested_fn = NULL;
5628 /* tell userspace to handle 'inactive' */
5629 sysfs_notify_dirent_safe(mddev->sysfs_state);
5631 rdev_for_each(rdev, mddev)
5632 if (rdev->raid_disk >= 0)
5633 sysfs_unlink_rdev(mddev, rdev);
5635 set_capacity(disk, 0);
5636 mutex_unlock(&mddev->open_mutex);
5638 revalidate_disk(disk);
5643 mutex_unlock(&mddev->open_mutex);
5645 * Free resources if final stop
5648 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5650 bitmap_destroy(mddev);
5651 if (mddev->bitmap_info.file) {
5652 struct file *f = mddev->bitmap_info.file;
5653 spin_lock(&mddev->lock);
5654 mddev->bitmap_info.file = NULL;
5655 spin_unlock(&mddev->lock);
5658 mddev->bitmap_info.offset = 0;
5660 export_array(mddev);
5663 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5664 if (mddev->hold_active == UNTIL_STOP)
5665 mddev->hold_active = 0;
5667 md_new_event(mddev);
5668 sysfs_notify_dirent_safe(mddev->sysfs_state);
5673 static void autorun_array(struct mddev *mddev)
5675 struct md_rdev *rdev;
5678 if (list_empty(&mddev->disks))
5681 printk(KERN_INFO "md: running: ");
5683 rdev_for_each(rdev, mddev) {
5684 char b[BDEVNAME_SIZE];
5685 printk("<%s>", bdevname(rdev->bdev,b));
5689 err = do_md_run(mddev);
5691 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5692 do_md_stop(mddev, 0, NULL);
5697 * lets try to run arrays based on all disks that have arrived
5698 * until now. (those are in pending_raid_disks)
5700 * the method: pick the first pending disk, collect all disks with
5701 * the same UUID, remove all from the pending list and put them into
5702 * the 'same_array' list. Then order this list based on superblock
5703 * update time (freshest comes first), kick out 'old' disks and
5704 * compare superblocks. If everything's fine then run it.
5706 * If "unit" is allocated, then bump its reference count
5708 static void autorun_devices(int part)
5710 struct md_rdev *rdev0, *rdev, *tmp;
5711 struct mddev *mddev;
5712 char b[BDEVNAME_SIZE];
5714 printk(KERN_INFO "md: autorun ...\n");
5715 while (!list_empty(&pending_raid_disks)) {
5718 LIST_HEAD(candidates);
5719 rdev0 = list_entry(pending_raid_disks.next,
5720 struct md_rdev, same_set);
5722 printk(KERN_INFO "md: considering %s ...\n",
5723 bdevname(rdev0->bdev,b));
5724 INIT_LIST_HEAD(&candidates);
5725 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5726 if (super_90_load(rdev, rdev0, 0) >= 0) {
5727 printk(KERN_INFO "md: adding %s ...\n",
5728 bdevname(rdev->bdev,b));
5729 list_move(&rdev->same_set, &candidates);
5732 * now we have a set of devices, with all of them having
5733 * mostly sane superblocks. It's time to allocate the
5737 dev = MKDEV(mdp_major,
5738 rdev0->preferred_minor << MdpMinorShift);
5739 unit = MINOR(dev) >> MdpMinorShift;
5741 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5744 if (rdev0->preferred_minor != unit) {
5745 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5746 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5750 md_probe(dev, NULL, NULL);
5751 mddev = mddev_find(dev);
5752 if (!mddev || !mddev->gendisk) {
5756 "md: cannot allocate memory for md drive.\n");
5759 if (mddev_lock(mddev))
5760 printk(KERN_WARNING "md: %s locked, cannot run\n",
5762 else if (mddev->raid_disks || mddev->major_version
5763 || !list_empty(&mddev->disks)) {
5765 "md: %s already running, cannot run %s\n",
5766 mdname(mddev), bdevname(rdev0->bdev,b));
5767 mddev_unlock(mddev);
5769 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5770 mddev->persistent = 1;
5771 rdev_for_each_list(rdev, tmp, &candidates) {
5772 list_del_init(&rdev->same_set);
5773 if (bind_rdev_to_array(rdev, mddev))
5776 autorun_array(mddev);
5777 mddev_unlock(mddev);
5779 /* on success, candidates will be empty, on error
5782 rdev_for_each_list(rdev, tmp, &candidates) {
5783 list_del_init(&rdev->same_set);
5788 printk(KERN_INFO "md: ... autorun DONE.\n");
5790 #endif /* !MODULE */
5792 static int get_version(void __user *arg)
5796 ver.major = MD_MAJOR_VERSION;
5797 ver.minor = MD_MINOR_VERSION;
5798 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5800 if (copy_to_user(arg, &ver, sizeof(ver)))
5806 static int get_array_info(struct mddev *mddev, void __user *arg)
5808 mdu_array_info_t info;
5809 int nr,working,insync,failed,spare;
5810 struct md_rdev *rdev;
5812 nr = working = insync = failed = spare = 0;
5814 rdev_for_each_rcu(rdev, mddev) {
5816 if (test_bit(Faulty, &rdev->flags))
5820 if (test_bit(In_sync, &rdev->flags))
5828 info.major_version = mddev->major_version;
5829 info.minor_version = mddev->minor_version;
5830 info.patch_version = MD_PATCHLEVEL_VERSION;
5831 info.ctime = mddev->ctime;
5832 info.level = mddev->level;
5833 info.size = mddev->dev_sectors / 2;
5834 if (info.size != mddev->dev_sectors / 2) /* overflow */
5837 info.raid_disks = mddev->raid_disks;
5838 info.md_minor = mddev->md_minor;
5839 info.not_persistent= !mddev->persistent;
5841 info.utime = mddev->utime;
5844 info.state = (1<<MD_SB_CLEAN);
5845 if (mddev->bitmap && mddev->bitmap_info.offset)
5846 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5847 if (mddev_is_clustered(mddev))
5848 info.state |= (1<<MD_SB_CLUSTERED);
5849 info.active_disks = insync;
5850 info.working_disks = working;
5851 info.failed_disks = failed;
5852 info.spare_disks = spare;
5854 info.layout = mddev->layout;
5855 info.chunk_size = mddev->chunk_sectors << 9;
5857 if (copy_to_user(arg, &info, sizeof(info)))
5863 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5865 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5869 file = kzalloc(sizeof(*file), GFP_NOIO);
5874 spin_lock(&mddev->lock);
5875 /* bitmap enabled */
5876 if (mddev->bitmap_info.file) {
5877 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5878 sizeof(file->pathname));
5882 memmove(file->pathname, ptr,
5883 sizeof(file->pathname)-(ptr-file->pathname));
5885 spin_unlock(&mddev->lock);
5888 copy_to_user(arg, file, sizeof(*file)))
5895 static int get_disk_info(struct mddev *mddev, void __user * arg)
5897 mdu_disk_info_t info;
5898 struct md_rdev *rdev;
5900 if (copy_from_user(&info, arg, sizeof(info)))
5904 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5906 info.major = MAJOR(rdev->bdev->bd_dev);
5907 info.minor = MINOR(rdev->bdev->bd_dev);
5908 info.raid_disk = rdev->raid_disk;
5910 if (test_bit(Faulty, &rdev->flags))
5911 info.state |= (1<<MD_DISK_FAULTY);
5912 else if (test_bit(In_sync, &rdev->flags)) {
5913 info.state |= (1<<MD_DISK_ACTIVE);
5914 info.state |= (1<<MD_DISK_SYNC);
5916 if (test_bit(Journal, &rdev->flags))
5917 info.state |= (1<<MD_DISK_JOURNAL);
5918 if (test_bit(WriteMostly, &rdev->flags))
5919 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5921 info.major = info.minor = 0;
5922 info.raid_disk = -1;
5923 info.state = (1<<MD_DISK_REMOVED);
5927 if (copy_to_user(arg, &info, sizeof(info)))
5933 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5935 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5936 struct md_rdev *rdev;
5937 dev_t dev = MKDEV(info->major,info->minor);
5939 if (mddev_is_clustered(mddev) &&
5940 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5941 pr_err("%s: Cannot add to clustered mddev.\n",
5946 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5949 if (!mddev->raid_disks) {
5951 /* expecting a device which has a superblock */
5952 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5955 "md: md_import_device returned %ld\n",
5957 return PTR_ERR(rdev);
5959 if (!list_empty(&mddev->disks)) {
5960 struct md_rdev *rdev0
5961 = list_entry(mddev->disks.next,
5962 struct md_rdev, same_set);
5963 err = super_types[mddev->major_version]
5964 .load_super(rdev, rdev0, mddev->minor_version);
5967 "md: %s has different UUID to %s\n",
5968 bdevname(rdev->bdev,b),
5969 bdevname(rdev0->bdev,b2));
5974 err = bind_rdev_to_array(rdev, mddev);
5981 * add_new_disk can be used once the array is assembled
5982 * to add "hot spares". They must already have a superblock
5987 if (!mddev->pers->hot_add_disk) {
5989 "%s: personality does not support diskops!\n",
5993 if (mddev->persistent)
5994 rdev = md_import_device(dev, mddev->major_version,
5995 mddev->minor_version);
5997 rdev = md_import_device(dev, -1, -1);
6000 "md: md_import_device returned %ld\n",
6002 return PTR_ERR(rdev);
6004 /* set saved_raid_disk if appropriate */
6005 if (!mddev->persistent) {
6006 if (info->state & (1<<MD_DISK_SYNC) &&
6007 info->raid_disk < mddev->raid_disks) {
6008 rdev->raid_disk = info->raid_disk;
6009 set_bit(In_sync, &rdev->flags);
6010 clear_bit(Bitmap_sync, &rdev->flags);
6012 rdev->raid_disk = -1;
6013 rdev->saved_raid_disk = rdev->raid_disk;
6015 super_types[mddev->major_version].
6016 validate_super(mddev, rdev);
6017 if ((info->state & (1<<MD_DISK_SYNC)) &&
6018 rdev->raid_disk != info->raid_disk) {
6019 /* This was a hot-add request, but events doesn't
6020 * match, so reject it.
6026 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6027 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6028 set_bit(WriteMostly, &rdev->flags);
6030 clear_bit(WriteMostly, &rdev->flags);
6032 if (info->state & (1<<MD_DISK_JOURNAL))
6033 set_bit(Journal, &rdev->flags);
6035 * check whether the device shows up in other nodes
6037 if (mddev_is_clustered(mddev)) {
6038 if (info->state & (1 << MD_DISK_CANDIDATE))
6039 set_bit(Candidate, &rdev->flags);
6040 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6041 /* --add initiated by this node */
6042 err = md_cluster_ops->add_new_disk(mddev, rdev);
6050 rdev->raid_disk = -1;
6051 err = bind_rdev_to_array(rdev, mddev);
6056 if (mddev_is_clustered(mddev)) {
6057 if (info->state & (1 << MD_DISK_CANDIDATE))
6058 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6061 md_cluster_ops->add_new_disk_cancel(mddev);
6063 err = add_bound_rdev(rdev);
6067 err = add_bound_rdev(rdev);
6072 /* otherwise, add_new_disk is only allowed
6073 * for major_version==0 superblocks
6075 if (mddev->major_version != 0) {
6076 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6081 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6083 rdev = md_import_device(dev, -1, 0);
6086 "md: error, md_import_device() returned %ld\n",
6088 return PTR_ERR(rdev);
6090 rdev->desc_nr = info->number;
6091 if (info->raid_disk < mddev->raid_disks)
6092 rdev->raid_disk = info->raid_disk;
6094 rdev->raid_disk = -1;
6096 if (rdev->raid_disk < mddev->raid_disks)
6097 if (info->state & (1<<MD_DISK_SYNC))
6098 set_bit(In_sync, &rdev->flags);
6100 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6101 set_bit(WriteMostly, &rdev->flags);
6103 if (!mddev->persistent) {
6104 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6105 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6107 rdev->sb_start = calc_dev_sboffset(rdev);
6108 rdev->sectors = rdev->sb_start;
6110 err = bind_rdev_to_array(rdev, mddev);
6120 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6122 char b[BDEVNAME_SIZE];
6123 struct md_rdev *rdev;
6126 rdev = find_rdev(mddev, dev);
6130 if (mddev_is_clustered(mddev))
6131 ret = md_cluster_ops->metadata_update_start(mddev);
6133 if (rdev->raid_disk < 0)
6136 clear_bit(Blocked, &rdev->flags);
6137 remove_and_add_spares(mddev, rdev);
6139 if (rdev->raid_disk >= 0)
6143 if (mddev_is_clustered(mddev) && ret == 0)
6144 md_cluster_ops->remove_disk(mddev, rdev);
6146 md_kick_rdev_from_array(rdev);
6147 md_update_sb(mddev, 1);
6148 md_new_event(mddev);
6152 if (mddev_is_clustered(mddev) && ret == 0)
6153 md_cluster_ops->metadata_update_cancel(mddev);
6155 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6156 bdevname(rdev->bdev,b), mdname(mddev));
6160 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6162 char b[BDEVNAME_SIZE];
6164 struct md_rdev *rdev;
6169 if (mddev->major_version != 0) {
6170 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6171 " version-0 superblocks.\n",
6175 if (!mddev->pers->hot_add_disk) {
6177 "%s: personality does not support diskops!\n",
6182 rdev = md_import_device(dev, -1, 0);
6185 "md: error, md_import_device() returned %ld\n",
6190 if (mddev->persistent)
6191 rdev->sb_start = calc_dev_sboffset(rdev);
6193 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6195 rdev->sectors = rdev->sb_start;
6197 if (test_bit(Faulty, &rdev->flags)) {
6199 "md: can not hot-add faulty %s disk to %s!\n",
6200 bdevname(rdev->bdev,b), mdname(mddev));
6205 clear_bit(In_sync, &rdev->flags);
6207 rdev->saved_raid_disk = -1;
6208 err = bind_rdev_to_array(rdev, mddev);
6213 * The rest should better be atomic, we can have disk failures
6214 * noticed in interrupt contexts ...
6217 rdev->raid_disk = -1;
6219 md_update_sb(mddev, 1);
6221 * Kick recovery, maybe this spare has to be added to the
6222 * array immediately.
6224 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6225 md_wakeup_thread(mddev->thread);
6226 md_new_event(mddev);
6234 static int set_bitmap_file(struct mddev *mddev, int fd)
6239 if (!mddev->pers->quiesce || !mddev->thread)
6241 if (mddev->recovery || mddev->sync_thread)
6243 /* we should be able to change the bitmap.. */
6247 struct inode *inode;
6250 if (mddev->bitmap || mddev->bitmap_info.file)
6251 return -EEXIST; /* cannot add when bitmap is present */
6255 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6260 inode = f->f_mapping->host;
6261 if (!S_ISREG(inode->i_mode)) {
6262 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6265 } else if (!(f->f_mode & FMODE_WRITE)) {
6266 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6269 } else if (atomic_read(&inode->i_writecount) != 1) {
6270 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6278 mddev->bitmap_info.file = f;
6279 mddev->bitmap_info.offset = 0; /* file overrides offset */
6280 } else if (mddev->bitmap == NULL)
6281 return -ENOENT; /* cannot remove what isn't there */
6284 mddev->pers->quiesce(mddev, 1);
6286 struct bitmap *bitmap;
6288 bitmap = bitmap_create(mddev, -1);
6289 if (!IS_ERR(bitmap)) {
6290 mddev->bitmap = bitmap;
6291 err = bitmap_load(mddev);
6293 err = PTR_ERR(bitmap);
6295 if (fd < 0 || err) {
6296 bitmap_destroy(mddev);
6297 fd = -1; /* make sure to put the file */
6299 mddev->pers->quiesce(mddev, 0);
6302 struct file *f = mddev->bitmap_info.file;
6304 spin_lock(&mddev->lock);
6305 mddev->bitmap_info.file = NULL;
6306 spin_unlock(&mddev->lock);
6315 * set_array_info is used two different ways
6316 * The original usage is when creating a new array.
6317 * In this usage, raid_disks is > 0 and it together with
6318 * level, size, not_persistent,layout,chunksize determine the
6319 * shape of the array.
6320 * This will always create an array with a type-0.90.0 superblock.
6321 * The newer usage is when assembling an array.
6322 * In this case raid_disks will be 0, and the major_version field is
6323 * use to determine which style super-blocks are to be found on the devices.
6324 * The minor and patch _version numbers are also kept incase the
6325 * super_block handler wishes to interpret them.
6327 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6330 if (info->raid_disks == 0) {
6331 /* just setting version number for superblock loading */
6332 if (info->major_version < 0 ||
6333 info->major_version >= ARRAY_SIZE(super_types) ||
6334 super_types[info->major_version].name == NULL) {
6335 /* maybe try to auto-load a module? */
6337 "md: superblock version %d not known\n",
6338 info->major_version);
6341 mddev->major_version = info->major_version;
6342 mddev->minor_version = info->minor_version;
6343 mddev->patch_version = info->patch_version;
6344 mddev->persistent = !info->not_persistent;
6345 /* ensure mddev_put doesn't delete this now that there
6346 * is some minimal configuration.
6348 mddev->ctime = get_seconds();
6351 mddev->major_version = MD_MAJOR_VERSION;
6352 mddev->minor_version = MD_MINOR_VERSION;
6353 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6354 mddev->ctime = get_seconds();
6356 mddev->level = info->level;
6357 mddev->clevel[0] = 0;
6358 mddev->dev_sectors = 2 * (sector_t)info->size;
6359 mddev->raid_disks = info->raid_disks;
6360 /* don't set md_minor, it is determined by which /dev/md* was
6363 if (info->state & (1<<MD_SB_CLEAN))
6364 mddev->recovery_cp = MaxSector;
6366 mddev->recovery_cp = 0;
6367 mddev->persistent = ! info->not_persistent;
6368 mddev->external = 0;
6370 mddev->layout = info->layout;
6371 mddev->chunk_sectors = info->chunk_size >> 9;
6373 mddev->max_disks = MD_SB_DISKS;
6375 if (mddev->persistent)
6377 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6379 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6380 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6381 mddev->bitmap_info.offset = 0;
6383 mddev->reshape_position = MaxSector;
6386 * Generate a 128 bit UUID
6388 get_random_bytes(mddev->uuid, 16);
6390 mddev->new_level = mddev->level;
6391 mddev->new_chunk_sectors = mddev->chunk_sectors;
6392 mddev->new_layout = mddev->layout;
6393 mddev->delta_disks = 0;
6394 mddev->reshape_backwards = 0;
6399 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6401 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6403 if (mddev->external_size)
6406 mddev->array_sectors = array_sectors;
6408 EXPORT_SYMBOL(md_set_array_sectors);
6410 static int update_size(struct mddev *mddev, sector_t num_sectors)
6412 struct md_rdev *rdev;
6414 int fit = (num_sectors == 0);
6416 if (mddev->pers->resize == NULL)
6418 /* The "num_sectors" is the number of sectors of each device that
6419 * is used. This can only make sense for arrays with redundancy.
6420 * linear and raid0 always use whatever space is available. We can only
6421 * consider changing this number if no resync or reconstruction is
6422 * happening, and if the new size is acceptable. It must fit before the
6423 * sb_start or, if that is <data_offset, it must fit before the size
6424 * of each device. If num_sectors is zero, we find the largest size
6427 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6433 rdev_for_each(rdev, mddev) {
6434 sector_t avail = rdev->sectors;
6436 if (fit && (num_sectors == 0 || num_sectors > avail))
6437 num_sectors = avail;
6438 if (avail < num_sectors)
6441 rv = mddev->pers->resize(mddev, num_sectors);
6443 revalidate_disk(mddev->gendisk);
6447 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6450 struct md_rdev *rdev;
6451 /* change the number of raid disks */
6452 if (mddev->pers->check_reshape == NULL)
6456 if (raid_disks <= 0 ||
6457 (mddev->max_disks && raid_disks >= mddev->max_disks))
6459 if (mddev->sync_thread ||
6460 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6461 mddev->reshape_position != MaxSector)
6464 rdev_for_each(rdev, mddev) {
6465 if (mddev->raid_disks < raid_disks &&
6466 rdev->data_offset < rdev->new_data_offset)
6468 if (mddev->raid_disks > raid_disks &&
6469 rdev->data_offset > rdev->new_data_offset)
6473 mddev->delta_disks = raid_disks - mddev->raid_disks;
6474 if (mddev->delta_disks < 0)
6475 mddev->reshape_backwards = 1;
6476 else if (mddev->delta_disks > 0)
6477 mddev->reshape_backwards = 0;
6479 rv = mddev->pers->check_reshape(mddev);
6481 mddev->delta_disks = 0;
6482 mddev->reshape_backwards = 0;
6488 * update_array_info is used to change the configuration of an
6490 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6491 * fields in the info are checked against the array.
6492 * Any differences that cannot be handled will cause an error.
6493 * Normally, only one change can be managed at a time.
6495 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6501 /* calculate expected state,ignoring low bits */
6502 if (mddev->bitmap && mddev->bitmap_info.offset)
6503 state |= (1 << MD_SB_BITMAP_PRESENT);
6505 if (mddev->major_version != info->major_version ||
6506 mddev->minor_version != info->minor_version ||
6507 /* mddev->patch_version != info->patch_version || */
6508 mddev->ctime != info->ctime ||
6509 mddev->level != info->level ||
6510 /* mddev->layout != info->layout || */
6511 mddev->persistent != !info->not_persistent ||
6512 mddev->chunk_sectors != info->chunk_size >> 9 ||
6513 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6514 ((state^info->state) & 0xfffffe00)
6517 /* Check there is only one change */
6518 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6520 if (mddev->raid_disks != info->raid_disks)
6522 if (mddev->layout != info->layout)
6524 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6531 if (mddev->layout != info->layout) {
6533 * we don't need to do anything at the md level, the
6534 * personality will take care of it all.
6536 if (mddev->pers->check_reshape == NULL)
6539 mddev->new_layout = info->layout;
6540 rv = mddev->pers->check_reshape(mddev);
6542 mddev->new_layout = mddev->layout;
6546 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6547 rv = update_size(mddev, (sector_t)info->size * 2);
6549 if (mddev->raid_disks != info->raid_disks)
6550 rv = update_raid_disks(mddev, info->raid_disks);
6552 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6553 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6557 if (mddev->recovery || mddev->sync_thread) {
6561 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6562 struct bitmap *bitmap;
6563 /* add the bitmap */
6564 if (mddev->bitmap) {
6568 if (mddev->bitmap_info.default_offset == 0) {
6572 mddev->bitmap_info.offset =
6573 mddev->bitmap_info.default_offset;
6574 mddev->bitmap_info.space =
6575 mddev->bitmap_info.default_space;
6576 mddev->pers->quiesce(mddev, 1);
6577 bitmap = bitmap_create(mddev, -1);
6578 if (!IS_ERR(bitmap)) {
6579 mddev->bitmap = bitmap;
6580 rv = bitmap_load(mddev);
6582 rv = PTR_ERR(bitmap);
6584 bitmap_destroy(mddev);
6585 mddev->pers->quiesce(mddev, 0);
6587 /* remove the bitmap */
6588 if (!mddev->bitmap) {
6592 if (mddev->bitmap->storage.file) {
6596 mddev->pers->quiesce(mddev, 1);
6597 bitmap_destroy(mddev);
6598 mddev->pers->quiesce(mddev, 0);
6599 mddev->bitmap_info.offset = 0;
6602 md_update_sb(mddev, 1);
6608 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6610 struct md_rdev *rdev;
6613 if (mddev->pers == NULL)
6617 rdev = find_rdev_rcu(mddev, dev);
6621 md_error(mddev, rdev);
6622 if (!test_bit(Faulty, &rdev->flags))
6630 * We have a problem here : there is no easy way to give a CHS
6631 * virtual geometry. We currently pretend that we have a 2 heads
6632 * 4 sectors (with a BIG number of cylinders...). This drives
6633 * dosfs just mad... ;-)
6635 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6637 struct mddev *mddev = bdev->bd_disk->private_data;
6641 geo->cylinders = mddev->array_sectors / 8;
6645 static inline bool md_ioctl_valid(unsigned int cmd)
6650 case GET_ARRAY_INFO:
6651 case GET_BITMAP_FILE:
6654 case HOT_REMOVE_DISK:
6657 case RESTART_ARRAY_RW:
6659 case SET_ARRAY_INFO:
6660 case SET_BITMAP_FILE:
6661 case SET_DISK_FAULTY:
6664 case CLUSTERED_DISK_NACK:
6671 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6672 unsigned int cmd, unsigned long arg)
6675 void __user *argp = (void __user *)arg;
6676 struct mddev *mddev = NULL;
6679 if (!md_ioctl_valid(cmd))
6684 case GET_ARRAY_INFO:
6688 if (!capable(CAP_SYS_ADMIN))
6693 * Commands dealing with the RAID driver but not any
6698 err = get_version(argp);
6704 autostart_arrays(arg);
6711 * Commands creating/starting a new array:
6714 mddev = bdev->bd_disk->private_data;
6721 /* Some actions do not requires the mutex */
6723 case GET_ARRAY_INFO:
6724 if (!mddev->raid_disks && !mddev->external)
6727 err = get_array_info(mddev, argp);
6731 if (!mddev->raid_disks && !mddev->external)
6734 err = get_disk_info(mddev, argp);
6737 case SET_DISK_FAULTY:
6738 err = set_disk_faulty(mddev, new_decode_dev(arg));
6741 case GET_BITMAP_FILE:
6742 err = get_bitmap_file(mddev, argp);
6747 if (cmd == ADD_NEW_DISK)
6748 /* need to ensure md_delayed_delete() has completed */
6749 flush_workqueue(md_misc_wq);
6751 if (cmd == HOT_REMOVE_DISK)
6752 /* need to ensure recovery thread has run */
6753 wait_event_interruptible_timeout(mddev->sb_wait,
6754 !test_bit(MD_RECOVERY_NEEDED,
6756 msecs_to_jiffies(5000));
6757 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6758 /* Need to flush page cache, and ensure no-one else opens
6761 mutex_lock(&mddev->open_mutex);
6762 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6763 mutex_unlock(&mddev->open_mutex);
6767 set_bit(MD_STILL_CLOSED, &mddev->flags);
6768 mutex_unlock(&mddev->open_mutex);
6769 sync_blockdev(bdev);
6771 err = mddev_lock(mddev);
6774 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6779 if (cmd == SET_ARRAY_INFO) {
6780 mdu_array_info_t info;
6782 memset(&info, 0, sizeof(info));
6783 else if (copy_from_user(&info, argp, sizeof(info))) {
6788 err = update_array_info(mddev, &info);
6790 printk(KERN_WARNING "md: couldn't update"
6791 " array info. %d\n", err);
6796 if (!list_empty(&mddev->disks)) {
6798 "md: array %s already has disks!\n",
6803 if (mddev->raid_disks) {
6805 "md: array %s already initialised!\n",
6810 err = set_array_info(mddev, &info);
6812 printk(KERN_WARNING "md: couldn't set"
6813 " array info. %d\n", err);
6820 * Commands querying/configuring an existing array:
6822 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6823 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6824 if ((!mddev->raid_disks && !mddev->external)
6825 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6826 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6827 && cmd != GET_BITMAP_FILE) {
6833 * Commands even a read-only array can execute:
6836 case RESTART_ARRAY_RW:
6837 err = restart_array(mddev);
6841 err = do_md_stop(mddev, 0, bdev);
6845 err = md_set_readonly(mddev, bdev);
6848 case HOT_REMOVE_DISK:
6849 err = hot_remove_disk(mddev, new_decode_dev(arg));
6853 /* We can support ADD_NEW_DISK on read-only arrays
6854 * on if we are re-adding a preexisting device.
6855 * So require mddev->pers and MD_DISK_SYNC.
6858 mdu_disk_info_t info;
6859 if (copy_from_user(&info, argp, sizeof(info)))
6861 else if (!(info.state & (1<<MD_DISK_SYNC)))
6862 /* Need to clear read-only for this */
6865 err = add_new_disk(mddev, &info);
6871 if (get_user(ro, (int __user *)(arg))) {
6877 /* if the bdev is going readonly the value of mddev->ro
6878 * does not matter, no writes are coming
6883 /* are we are already prepared for writes? */
6887 /* transitioning to readauto need only happen for
6888 * arrays that call md_write_start
6891 err = restart_array(mddev);
6894 set_disk_ro(mddev->gendisk, 0);
6901 * The remaining ioctls are changing the state of the
6902 * superblock, so we do not allow them on read-only arrays.
6904 if (mddev->ro && mddev->pers) {
6905 if (mddev->ro == 2) {
6907 sysfs_notify_dirent_safe(mddev->sysfs_state);
6908 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6909 /* mddev_unlock will wake thread */
6910 /* If a device failed while we were read-only, we
6911 * need to make sure the metadata is updated now.
6913 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6914 mddev_unlock(mddev);
6915 wait_event(mddev->sb_wait,
6916 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6917 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6918 mddev_lock_nointr(mddev);
6929 mdu_disk_info_t info;
6930 if (copy_from_user(&info, argp, sizeof(info)))
6933 err = add_new_disk(mddev, &info);
6937 case CLUSTERED_DISK_NACK:
6938 if (mddev_is_clustered(mddev))
6939 md_cluster_ops->new_disk_ack(mddev, false);
6945 err = hot_add_disk(mddev, new_decode_dev(arg));
6949 err = do_md_run(mddev);
6952 case SET_BITMAP_FILE:
6953 err = set_bitmap_file(mddev, (int)arg);
6962 if (mddev->hold_active == UNTIL_IOCTL &&
6964 mddev->hold_active = 0;
6965 mddev_unlock(mddev);
6969 #ifdef CONFIG_COMPAT
6970 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6971 unsigned int cmd, unsigned long arg)
6974 case HOT_REMOVE_DISK:
6976 case SET_DISK_FAULTY:
6977 case SET_BITMAP_FILE:
6978 /* These take in integer arg, do not convert */
6981 arg = (unsigned long)compat_ptr(arg);
6985 return md_ioctl(bdev, mode, cmd, arg);
6987 #endif /* CONFIG_COMPAT */
6989 static int md_open(struct block_device *bdev, fmode_t mode)
6992 * Succeed if we can lock the mddev, which confirms that
6993 * it isn't being stopped right now.
6995 struct mddev *mddev = mddev_find(bdev->bd_dev);
7001 if (mddev->gendisk != bdev->bd_disk) {
7002 /* we are racing with mddev_put which is discarding this
7006 /* Wait until bdev->bd_disk is definitely gone */
7007 flush_workqueue(md_misc_wq);
7008 /* Then retry the open from the top */
7009 return -ERESTARTSYS;
7011 BUG_ON(mddev != bdev->bd_disk->private_data);
7013 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7017 atomic_inc(&mddev->openers);
7018 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7019 mutex_unlock(&mddev->open_mutex);
7021 check_disk_change(bdev);
7026 static void md_release(struct gendisk *disk, fmode_t mode)
7028 struct mddev *mddev = disk->private_data;
7031 atomic_dec(&mddev->openers);
7035 static int md_media_changed(struct gendisk *disk)
7037 struct mddev *mddev = disk->private_data;
7039 return mddev->changed;
7042 static int md_revalidate(struct gendisk *disk)
7044 struct mddev *mddev = disk->private_data;
7049 static const struct block_device_operations md_fops =
7051 .owner = THIS_MODULE,
7053 .release = md_release,
7055 #ifdef CONFIG_COMPAT
7056 .compat_ioctl = md_compat_ioctl,
7058 .getgeo = md_getgeo,
7059 .media_changed = md_media_changed,
7060 .revalidate_disk= md_revalidate,
7063 static int md_thread(void *arg)
7065 struct md_thread *thread = arg;
7068 * md_thread is a 'system-thread', it's priority should be very
7069 * high. We avoid resource deadlocks individually in each
7070 * raid personality. (RAID5 does preallocation) We also use RR and
7071 * the very same RT priority as kswapd, thus we will never get
7072 * into a priority inversion deadlock.
7074 * we definitely have to have equal or higher priority than
7075 * bdflush, otherwise bdflush will deadlock if there are too
7076 * many dirty RAID5 blocks.
7079 allow_signal(SIGKILL);
7080 while (!kthread_should_stop()) {
7082 /* We need to wait INTERRUPTIBLE so that
7083 * we don't add to the load-average.
7084 * That means we need to be sure no signals are
7087 if (signal_pending(current))
7088 flush_signals(current);
7090 wait_event_interruptible_timeout
7092 test_bit(THREAD_WAKEUP, &thread->flags)
7093 || kthread_should_stop(),
7096 clear_bit(THREAD_WAKEUP, &thread->flags);
7097 if (!kthread_should_stop())
7098 thread->run(thread);
7104 void md_wakeup_thread(struct md_thread *thread)
7107 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7108 set_bit(THREAD_WAKEUP, &thread->flags);
7109 wake_up(&thread->wqueue);
7112 EXPORT_SYMBOL(md_wakeup_thread);
7114 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7115 struct mddev *mddev, const char *name)
7117 struct md_thread *thread;
7119 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7123 init_waitqueue_head(&thread->wqueue);
7126 thread->mddev = mddev;
7127 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7128 thread->tsk = kthread_run(md_thread, thread,
7130 mdname(thread->mddev),
7132 if (IS_ERR(thread->tsk)) {
7138 EXPORT_SYMBOL(md_register_thread);
7140 void md_unregister_thread(struct md_thread **threadp)
7142 struct md_thread *thread = *threadp;
7145 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7146 /* Locking ensures that mddev_unlock does not wake_up a
7147 * non-existent thread
7149 spin_lock(&pers_lock);
7151 spin_unlock(&pers_lock);
7153 kthread_stop(thread->tsk);
7156 EXPORT_SYMBOL(md_unregister_thread);
7158 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7160 if (!rdev || test_bit(Faulty, &rdev->flags))
7163 if (!mddev->pers || !mddev->pers->error_handler)
7165 mddev->pers->error_handler(mddev,rdev);
7166 if (mddev->degraded)
7167 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7168 sysfs_notify_dirent_safe(rdev->sysfs_state);
7169 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7170 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7171 md_wakeup_thread(mddev->thread);
7172 if (mddev->event_work.func)
7173 queue_work(md_misc_wq, &mddev->event_work);
7174 md_new_event_inintr(mddev);
7176 EXPORT_SYMBOL(md_error);
7178 /* seq_file implementation /proc/mdstat */
7180 static void status_unused(struct seq_file *seq)
7183 struct md_rdev *rdev;
7185 seq_printf(seq, "unused devices: ");
7187 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7188 char b[BDEVNAME_SIZE];
7190 seq_printf(seq, "%s ",
7191 bdevname(rdev->bdev,b));
7194 seq_printf(seq, "<none>");
7196 seq_printf(seq, "\n");
7199 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7201 sector_t max_sectors, resync, res;
7202 unsigned long dt, db;
7205 unsigned int per_milli;
7207 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7208 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7209 max_sectors = mddev->resync_max_sectors;
7211 max_sectors = mddev->dev_sectors;
7213 resync = mddev->curr_resync;
7215 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7216 /* Still cleaning up */
7217 resync = max_sectors;
7219 resync -= atomic_read(&mddev->recovery_active);
7222 if (mddev->recovery_cp < MaxSector) {
7223 seq_printf(seq, "\tresync=PENDING");
7229 seq_printf(seq, "\tresync=DELAYED");
7233 WARN_ON(max_sectors == 0);
7234 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7235 * in a sector_t, and (max_sectors>>scale) will fit in a
7236 * u32, as those are the requirements for sector_div.
7237 * Thus 'scale' must be at least 10
7240 if (sizeof(sector_t) > sizeof(unsigned long)) {
7241 while ( max_sectors/2 > (1ULL<<(scale+32)))
7244 res = (resync>>scale)*1000;
7245 sector_div(res, (u32)((max_sectors>>scale)+1));
7249 int i, x = per_milli/50, y = 20-x;
7250 seq_printf(seq, "[");
7251 for (i = 0; i < x; i++)
7252 seq_printf(seq, "=");
7253 seq_printf(seq, ">");
7254 for (i = 0; i < y; i++)
7255 seq_printf(seq, ".");
7256 seq_printf(seq, "] ");
7258 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7259 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7261 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7263 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7264 "resync" : "recovery"))),
7265 per_milli/10, per_milli % 10,
7266 (unsigned long long) resync/2,
7267 (unsigned long long) max_sectors/2);
7270 * dt: time from mark until now
7271 * db: blocks written from mark until now
7272 * rt: remaining time
7274 * rt is a sector_t, so could be 32bit or 64bit.
7275 * So we divide before multiply in case it is 32bit and close
7277 * We scale the divisor (db) by 32 to avoid losing precision
7278 * near the end of resync when the number of remaining sectors
7280 * We then divide rt by 32 after multiplying by db to compensate.
7281 * The '+1' avoids division by zero if db is very small.
7283 dt = ((jiffies - mddev->resync_mark) / HZ);
7285 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7286 - mddev->resync_mark_cnt;
7288 rt = max_sectors - resync; /* number of remaining sectors */
7289 sector_div(rt, db/32+1);
7293 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7294 ((unsigned long)rt % 60)/6);
7296 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7300 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7302 struct list_head *tmp;
7304 struct mddev *mddev;
7312 spin_lock(&all_mddevs_lock);
7313 list_for_each(tmp,&all_mddevs)
7315 mddev = list_entry(tmp, struct mddev, all_mddevs);
7317 spin_unlock(&all_mddevs_lock);
7320 spin_unlock(&all_mddevs_lock);
7322 return (void*)2;/* tail */
7326 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7328 struct list_head *tmp;
7329 struct mddev *next_mddev, *mddev = v;
7335 spin_lock(&all_mddevs_lock);
7337 tmp = all_mddevs.next;
7339 tmp = mddev->all_mddevs.next;
7340 if (tmp != &all_mddevs)
7341 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7343 next_mddev = (void*)2;
7346 spin_unlock(&all_mddevs_lock);
7354 static void md_seq_stop(struct seq_file *seq, void *v)
7356 struct mddev *mddev = v;
7358 if (mddev && v != (void*)1 && v != (void*)2)
7362 static int md_seq_show(struct seq_file *seq, void *v)
7364 struct mddev *mddev = v;
7366 struct md_rdev *rdev;
7368 if (v == (void*)1) {
7369 struct md_personality *pers;
7370 seq_printf(seq, "Personalities : ");
7371 spin_lock(&pers_lock);
7372 list_for_each_entry(pers, &pers_list, list)
7373 seq_printf(seq, "[%s] ", pers->name);
7375 spin_unlock(&pers_lock);
7376 seq_printf(seq, "\n");
7377 seq->poll_event = atomic_read(&md_event_count);
7380 if (v == (void*)2) {
7385 spin_lock(&mddev->lock);
7386 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7387 seq_printf(seq, "%s : %sactive", mdname(mddev),
7388 mddev->pers ? "" : "in");
7391 seq_printf(seq, " (read-only)");
7393 seq_printf(seq, " (auto-read-only)");
7394 seq_printf(seq, " %s", mddev->pers->name);
7399 rdev_for_each_rcu(rdev, mddev) {
7400 char b[BDEVNAME_SIZE];
7401 seq_printf(seq, " %s[%d]",
7402 bdevname(rdev->bdev,b), rdev->desc_nr);
7403 if (test_bit(WriteMostly, &rdev->flags))
7404 seq_printf(seq, "(W)");
7405 if (test_bit(Journal, &rdev->flags))
7406 seq_printf(seq, "(J)");
7407 if (test_bit(Faulty, &rdev->flags)) {
7408 seq_printf(seq, "(F)");
7411 if (rdev->raid_disk < 0)
7412 seq_printf(seq, "(S)"); /* spare */
7413 if (test_bit(Replacement, &rdev->flags))
7414 seq_printf(seq, "(R)");
7415 sectors += rdev->sectors;
7419 if (!list_empty(&mddev->disks)) {
7421 seq_printf(seq, "\n %llu blocks",
7422 (unsigned long long)
7423 mddev->array_sectors / 2);
7425 seq_printf(seq, "\n %llu blocks",
7426 (unsigned long long)sectors / 2);
7428 if (mddev->persistent) {
7429 if (mddev->major_version != 0 ||
7430 mddev->minor_version != 90) {
7431 seq_printf(seq," super %d.%d",
7432 mddev->major_version,
7433 mddev->minor_version);
7435 } else if (mddev->external)
7436 seq_printf(seq, " super external:%s",
7437 mddev->metadata_type);
7439 seq_printf(seq, " super non-persistent");
7442 mddev->pers->status(seq, mddev);
7443 seq_printf(seq, "\n ");
7444 if (mddev->pers->sync_request) {
7445 if (status_resync(seq, mddev))
7446 seq_printf(seq, "\n ");
7449 seq_printf(seq, "\n ");
7451 bitmap_status(seq, mddev->bitmap);
7453 seq_printf(seq, "\n");
7455 spin_unlock(&mddev->lock);
7460 static const struct seq_operations md_seq_ops = {
7461 .start = md_seq_start,
7462 .next = md_seq_next,
7463 .stop = md_seq_stop,
7464 .show = md_seq_show,
7467 static int md_seq_open(struct inode *inode, struct file *file)
7469 struct seq_file *seq;
7472 error = seq_open(file, &md_seq_ops);
7476 seq = file->private_data;
7477 seq->poll_event = atomic_read(&md_event_count);
7481 static int md_unloading;
7482 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7484 struct seq_file *seq = filp->private_data;
7488 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7489 poll_wait(filp, &md_event_waiters, wait);
7491 /* always allow read */
7492 mask = POLLIN | POLLRDNORM;
7494 if (seq->poll_event != atomic_read(&md_event_count))
7495 mask |= POLLERR | POLLPRI;
7499 static const struct file_operations md_seq_fops = {
7500 .owner = THIS_MODULE,
7501 .open = md_seq_open,
7503 .llseek = seq_lseek,
7504 .release = seq_release_private,
7505 .poll = mdstat_poll,
7508 int register_md_personality(struct md_personality *p)
7510 printk(KERN_INFO "md: %s personality registered for level %d\n",
7512 spin_lock(&pers_lock);
7513 list_add_tail(&p->list, &pers_list);
7514 spin_unlock(&pers_lock);
7517 EXPORT_SYMBOL(register_md_personality);
7519 int unregister_md_personality(struct md_personality *p)
7521 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7522 spin_lock(&pers_lock);
7523 list_del_init(&p->list);
7524 spin_unlock(&pers_lock);
7527 EXPORT_SYMBOL(unregister_md_personality);
7529 int register_md_cluster_operations(struct md_cluster_operations *ops,
7530 struct module *module)
7533 spin_lock(&pers_lock);
7534 if (md_cluster_ops != NULL)
7537 md_cluster_ops = ops;
7538 md_cluster_mod = module;
7540 spin_unlock(&pers_lock);
7543 EXPORT_SYMBOL(register_md_cluster_operations);
7545 int unregister_md_cluster_operations(void)
7547 spin_lock(&pers_lock);
7548 md_cluster_ops = NULL;
7549 spin_unlock(&pers_lock);
7552 EXPORT_SYMBOL(unregister_md_cluster_operations);
7554 int md_setup_cluster(struct mddev *mddev, int nodes)
7558 err = request_module("md-cluster");
7560 pr_err("md-cluster module not found.\n");
7564 spin_lock(&pers_lock);
7565 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7566 spin_unlock(&pers_lock);
7569 spin_unlock(&pers_lock);
7571 return md_cluster_ops->join(mddev, nodes);
7574 void md_cluster_stop(struct mddev *mddev)
7576 if (!md_cluster_ops)
7578 md_cluster_ops->leave(mddev);
7579 module_put(md_cluster_mod);
7582 static int is_mddev_idle(struct mddev *mddev, int init)
7584 struct md_rdev *rdev;
7590 rdev_for_each_rcu(rdev, mddev) {
7591 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7592 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7593 (int)part_stat_read(&disk->part0, sectors[1]) -
7594 atomic_read(&disk->sync_io);
7595 /* sync IO will cause sync_io to increase before the disk_stats
7596 * as sync_io is counted when a request starts, and
7597 * disk_stats is counted when it completes.
7598 * So resync activity will cause curr_events to be smaller than
7599 * when there was no such activity.
7600 * non-sync IO will cause disk_stat to increase without
7601 * increasing sync_io so curr_events will (eventually)
7602 * be larger than it was before. Once it becomes
7603 * substantially larger, the test below will cause
7604 * the array to appear non-idle, and resync will slow
7606 * If there is a lot of outstanding resync activity when
7607 * we set last_event to curr_events, then all that activity
7608 * completing might cause the array to appear non-idle
7609 * and resync will be slowed down even though there might
7610 * not have been non-resync activity. This will only
7611 * happen once though. 'last_events' will soon reflect
7612 * the state where there is little or no outstanding
7613 * resync requests, and further resync activity will
7614 * always make curr_events less than last_events.
7617 if (init || curr_events - rdev->last_events > 64) {
7618 rdev->last_events = curr_events;
7626 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7628 /* another "blocks" (512byte) blocks have been synced */
7629 atomic_sub(blocks, &mddev->recovery_active);
7630 wake_up(&mddev->recovery_wait);
7632 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7633 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7634 md_wakeup_thread(mddev->thread);
7635 // stop recovery, signal do_sync ....
7638 EXPORT_SYMBOL(md_done_sync);
7640 /* md_write_start(mddev, bi)
7641 * If we need to update some array metadata (e.g. 'active' flag
7642 * in superblock) before writing, schedule a superblock update
7643 * and wait for it to complete.
7645 void md_write_start(struct mddev *mddev, struct bio *bi)
7648 if (bio_data_dir(bi) != WRITE)
7651 BUG_ON(mddev->ro == 1);
7652 if (mddev->ro == 2) {
7653 /* need to switch to read/write */
7655 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7656 md_wakeup_thread(mddev->thread);
7657 md_wakeup_thread(mddev->sync_thread);
7660 atomic_inc(&mddev->writes_pending);
7661 if (mddev->safemode == 1)
7662 mddev->safemode = 0;
7663 if (mddev->in_sync) {
7664 spin_lock(&mddev->lock);
7665 if (mddev->in_sync) {
7667 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7668 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7669 md_wakeup_thread(mddev->thread);
7672 spin_unlock(&mddev->lock);
7675 sysfs_notify_dirent_safe(mddev->sysfs_state);
7676 wait_event(mddev->sb_wait,
7677 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7679 EXPORT_SYMBOL(md_write_start);
7681 void md_write_end(struct mddev *mddev)
7683 if (atomic_dec_and_test(&mddev->writes_pending)) {
7684 if (mddev->safemode == 2)
7685 md_wakeup_thread(mddev->thread);
7686 else if (mddev->safemode_delay)
7687 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7690 EXPORT_SYMBOL(md_write_end);
7692 /* md_allow_write(mddev)
7693 * Calling this ensures that the array is marked 'active' so that writes
7694 * may proceed without blocking. It is important to call this before
7695 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7696 * Must be called with mddev_lock held.
7698 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7699 * is dropped, so return -EAGAIN after notifying userspace.
7701 int md_allow_write(struct mddev *mddev)
7707 if (!mddev->pers->sync_request)
7710 spin_lock(&mddev->lock);
7711 if (mddev->in_sync) {
7713 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7714 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7715 if (mddev->safemode_delay &&
7716 mddev->safemode == 0)
7717 mddev->safemode = 1;
7718 spin_unlock(&mddev->lock);
7719 md_update_sb(mddev, 0);
7720 sysfs_notify_dirent_safe(mddev->sysfs_state);
7722 spin_unlock(&mddev->lock);
7724 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7729 EXPORT_SYMBOL_GPL(md_allow_write);
7731 #define SYNC_MARKS 10
7732 #define SYNC_MARK_STEP (3*HZ)
7733 #define UPDATE_FREQUENCY (5*60*HZ)
7734 void md_do_sync(struct md_thread *thread)
7736 struct mddev *mddev = thread->mddev;
7737 struct mddev *mddev2;
7738 unsigned int currspeed = 0,
7740 sector_t max_sectors,j, io_sectors, recovery_done;
7741 unsigned long mark[SYNC_MARKS];
7742 unsigned long update_time;
7743 sector_t mark_cnt[SYNC_MARKS];
7745 struct list_head *tmp;
7746 sector_t last_check;
7748 struct md_rdev *rdev;
7749 char *desc, *action = NULL;
7750 struct blk_plug plug;
7751 bool cluster_resync_finished = false;
7753 /* just incase thread restarts... */
7754 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7756 if (mddev->ro) {/* never try to sync a read-only array */
7757 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7761 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7762 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7763 desc = "data-check";
7765 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7766 desc = "requested-resync";
7770 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7775 mddev->last_sync_action = action ?: desc;
7777 /* we overload curr_resync somewhat here.
7778 * 0 == not engaged in resync at all
7779 * 2 == checking that there is no conflict with another sync
7780 * 1 == like 2, but have yielded to allow conflicting resync to
7782 * other == active in resync - this many blocks
7784 * Before starting a resync we must have set curr_resync to
7785 * 2, and then checked that every "conflicting" array has curr_resync
7786 * less than ours. When we find one that is the same or higher
7787 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7788 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7789 * This will mean we have to start checking from the beginning again.
7794 mddev->curr_resync = 2;
7797 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7799 for_each_mddev(mddev2, tmp) {
7800 if (mddev2 == mddev)
7802 if (!mddev->parallel_resync
7803 && mddev2->curr_resync
7804 && match_mddev_units(mddev, mddev2)) {
7806 if (mddev < mddev2 && mddev->curr_resync == 2) {
7807 /* arbitrarily yield */
7808 mddev->curr_resync = 1;
7809 wake_up(&resync_wait);
7811 if (mddev > mddev2 && mddev->curr_resync == 1)
7812 /* no need to wait here, we can wait the next
7813 * time 'round when curr_resync == 2
7816 /* We need to wait 'interruptible' so as not to
7817 * contribute to the load average, and not to
7818 * be caught by 'softlockup'
7820 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7821 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7822 mddev2->curr_resync >= mddev->curr_resync) {
7823 printk(KERN_INFO "md: delaying %s of %s"
7824 " until %s has finished (they"
7825 " share one or more physical units)\n",
7826 desc, mdname(mddev), mdname(mddev2));
7828 if (signal_pending(current))
7829 flush_signals(current);
7831 finish_wait(&resync_wait, &wq);
7834 finish_wait(&resync_wait, &wq);
7837 } while (mddev->curr_resync < 2);
7840 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7841 /* resync follows the size requested by the personality,
7842 * which defaults to physical size, but can be virtual size
7844 max_sectors = mddev->resync_max_sectors;
7845 atomic64_set(&mddev->resync_mismatches, 0);
7846 /* we don't use the checkpoint if there's a bitmap */
7847 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7848 j = mddev->resync_min;
7849 else if (!mddev->bitmap)
7850 j = mddev->recovery_cp;
7852 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7853 max_sectors = mddev->resync_max_sectors;
7855 /* recovery follows the physical size of devices */
7856 max_sectors = mddev->dev_sectors;
7859 rdev_for_each_rcu(rdev, mddev)
7860 if (rdev->raid_disk >= 0 &&
7861 !test_bit(Journal, &rdev->flags) &&
7862 !test_bit(Faulty, &rdev->flags) &&
7863 !test_bit(In_sync, &rdev->flags) &&
7864 rdev->recovery_offset < j)
7865 j = rdev->recovery_offset;
7868 /* If there is a bitmap, we need to make sure all
7869 * writes that started before we added a spare
7870 * complete before we start doing a recovery.
7871 * Otherwise the write might complete and (via
7872 * bitmap_endwrite) set a bit in the bitmap after the
7873 * recovery has checked that bit and skipped that
7876 if (mddev->bitmap) {
7877 mddev->pers->quiesce(mddev, 1);
7878 mddev->pers->quiesce(mddev, 0);
7882 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7883 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7884 " %d KB/sec/disk.\n", speed_min(mddev));
7885 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7886 "(but not more than %d KB/sec) for %s.\n",
7887 speed_max(mddev), desc);
7889 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7892 for (m = 0; m < SYNC_MARKS; m++) {
7894 mark_cnt[m] = io_sectors;
7897 mddev->resync_mark = mark[last_mark];
7898 mddev->resync_mark_cnt = mark_cnt[last_mark];
7901 * Tune reconstruction:
7903 window = 32*(PAGE_SIZE/512);
7904 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7905 window/2, (unsigned long long)max_sectors/2);
7907 atomic_set(&mddev->recovery_active, 0);
7912 "md: resuming %s of %s from checkpoint.\n",
7913 desc, mdname(mddev));
7914 mddev->curr_resync = j;
7916 mddev->curr_resync = 3; /* no longer delayed */
7917 mddev->curr_resync_completed = j;
7918 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7919 md_new_event(mddev);
7920 update_time = jiffies;
7922 blk_start_plug(&plug);
7923 while (j < max_sectors) {
7928 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7929 ((mddev->curr_resync > mddev->curr_resync_completed &&
7930 (mddev->curr_resync - mddev->curr_resync_completed)
7931 > (max_sectors >> 4)) ||
7932 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7933 (j - mddev->curr_resync_completed)*2
7934 >= mddev->resync_max - mddev->curr_resync_completed ||
7935 mddev->curr_resync_completed > mddev->resync_max
7937 /* time to update curr_resync_completed */
7938 wait_event(mddev->recovery_wait,
7939 atomic_read(&mddev->recovery_active) == 0);
7940 mddev->curr_resync_completed = j;
7941 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7942 j > mddev->recovery_cp)
7943 mddev->recovery_cp = j;
7944 update_time = jiffies;
7945 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7946 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7949 while (j >= mddev->resync_max &&
7950 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7951 /* As this condition is controlled by user-space,
7952 * we can block indefinitely, so use '_interruptible'
7953 * to avoid triggering warnings.
7955 flush_signals(current); /* just in case */
7956 wait_event_interruptible(mddev->recovery_wait,
7957 mddev->resync_max > j
7958 || test_bit(MD_RECOVERY_INTR,
7962 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7965 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7967 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7971 if (!skipped) { /* actual IO requested */
7972 io_sectors += sectors;
7973 atomic_add(sectors, &mddev->recovery_active);
7976 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7980 if (j > max_sectors)
7981 /* when skipping, extra large numbers can be returned. */
7984 mddev->curr_resync = j;
7985 mddev->curr_mark_cnt = io_sectors;
7986 if (last_check == 0)
7987 /* this is the earliest that rebuild will be
7988 * visible in /proc/mdstat
7990 md_new_event(mddev);
7992 if (last_check + window > io_sectors || j == max_sectors)
7995 last_check = io_sectors;
7997 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7999 int next = (last_mark+1) % SYNC_MARKS;
8001 mddev->resync_mark = mark[next];
8002 mddev->resync_mark_cnt = mark_cnt[next];
8003 mark[next] = jiffies;
8004 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8008 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8012 * this loop exits only if either when we are slower than
8013 * the 'hard' speed limit, or the system was IO-idle for
8015 * the system might be non-idle CPU-wise, but we only care
8016 * about not overloading the IO subsystem. (things like an
8017 * e2fsck being done on the RAID array should execute fast)
8021 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8022 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8023 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8025 if (currspeed > speed_min(mddev)) {
8026 if (currspeed > speed_max(mddev)) {
8030 if (!is_mddev_idle(mddev, 0)) {
8032 * Give other IO more of a chance.
8033 * The faster the devices, the less we wait.
8035 wait_event(mddev->recovery_wait,
8036 !atomic_read(&mddev->recovery_active));
8040 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8041 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8042 ? "interrupted" : "done");
8044 * this also signals 'finished resyncing' to md_stop
8046 blk_finish_plug(&plug);
8047 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8049 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8050 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8051 mddev->curr_resync > 2) {
8052 mddev->curr_resync_completed = mddev->curr_resync;
8053 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8055 /* tell personality and other nodes that we are finished */
8056 if (mddev_is_clustered(mddev)) {
8057 md_cluster_ops->resync_finish(mddev);
8058 cluster_resync_finished = true;
8060 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8062 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8063 mddev->curr_resync > 2) {
8064 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8065 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8066 if (mddev->curr_resync >= mddev->recovery_cp) {
8068 "md: checkpointing %s of %s.\n",
8069 desc, mdname(mddev));
8070 if (test_bit(MD_RECOVERY_ERROR,
8072 mddev->recovery_cp =
8073 mddev->curr_resync_completed;
8075 mddev->recovery_cp =
8079 mddev->recovery_cp = MaxSector;
8081 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8082 mddev->curr_resync = MaxSector;
8084 rdev_for_each_rcu(rdev, mddev)
8085 if (rdev->raid_disk >= 0 &&
8086 mddev->delta_disks >= 0 &&
8087 !test_bit(Journal, &rdev->flags) &&
8088 !test_bit(Faulty, &rdev->flags) &&
8089 !test_bit(In_sync, &rdev->flags) &&
8090 rdev->recovery_offset < mddev->curr_resync)
8091 rdev->recovery_offset = mddev->curr_resync;
8096 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8098 if (mddev_is_clustered(mddev) &&
8099 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8100 !cluster_resync_finished)
8101 md_cluster_ops->resync_finish(mddev);
8103 spin_lock(&mddev->lock);
8104 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8105 /* We completed so min/max setting can be forgotten if used. */
8106 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8107 mddev->resync_min = 0;
8108 mddev->resync_max = MaxSector;
8109 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8110 mddev->resync_min = mddev->curr_resync_completed;
8111 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8112 mddev->curr_resync = 0;
8113 spin_unlock(&mddev->lock);
8115 wake_up(&resync_wait);
8116 md_wakeup_thread(mddev->thread);
8119 EXPORT_SYMBOL_GPL(md_do_sync);
8121 static int remove_and_add_spares(struct mddev *mddev,
8122 struct md_rdev *this)
8124 struct md_rdev *rdev;
8128 rdev_for_each(rdev, mddev)
8129 if ((this == NULL || rdev == this) &&
8130 rdev->raid_disk >= 0 &&
8131 !test_bit(Blocked, &rdev->flags) &&
8132 (test_bit(Faulty, &rdev->flags) ||
8133 (!test_bit(In_sync, &rdev->flags) &&
8134 !test_bit(Journal, &rdev->flags))) &&
8135 atomic_read(&rdev->nr_pending)==0) {
8136 if (mddev->pers->hot_remove_disk(
8137 mddev, rdev) == 0) {
8138 sysfs_unlink_rdev(mddev, rdev);
8139 rdev->raid_disk = -1;
8143 if (removed && mddev->kobj.sd)
8144 sysfs_notify(&mddev->kobj, NULL, "degraded");
8146 if (this && removed)
8149 rdev_for_each(rdev, mddev) {
8150 if (this && this != rdev)
8152 if (test_bit(Candidate, &rdev->flags))
8154 if (rdev->raid_disk >= 0 &&
8155 !test_bit(In_sync, &rdev->flags) &&
8156 !test_bit(Journal, &rdev->flags) &&
8157 !test_bit(Faulty, &rdev->flags))
8159 if (rdev->raid_disk >= 0)
8161 if (test_bit(Faulty, &rdev->flags))
8163 if (test_bit(Journal, &rdev->flags))
8166 ! (rdev->saved_raid_disk >= 0 &&
8167 !test_bit(Bitmap_sync, &rdev->flags)))
8170 rdev->recovery_offset = 0;
8172 hot_add_disk(mddev, rdev) == 0) {
8173 if (sysfs_link_rdev(mddev, rdev))
8174 /* failure here is OK */;
8176 md_new_event(mddev);
8177 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8182 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8186 static void md_start_sync(struct work_struct *ws)
8188 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8191 if (mddev_is_clustered(mddev)) {
8192 ret = md_cluster_ops->resync_start(mddev);
8194 mddev->sync_thread = NULL;
8199 mddev->sync_thread = md_register_thread(md_do_sync,
8203 if (!mddev->sync_thread) {
8204 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8205 printk(KERN_ERR "%s: could not start resync"
8208 /* leave the spares where they are, it shouldn't hurt */
8209 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8210 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8211 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8212 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8213 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8214 wake_up(&resync_wait);
8215 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8217 if (mddev->sysfs_action)
8218 sysfs_notify_dirent_safe(mddev->sysfs_action);
8220 md_wakeup_thread(mddev->sync_thread);
8221 sysfs_notify_dirent_safe(mddev->sysfs_action);
8222 md_new_event(mddev);
8226 * This routine is regularly called by all per-raid-array threads to
8227 * deal with generic issues like resync and super-block update.
8228 * Raid personalities that don't have a thread (linear/raid0) do not
8229 * need this as they never do any recovery or update the superblock.
8231 * It does not do any resync itself, but rather "forks" off other threads
8232 * to do that as needed.
8233 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8234 * "->recovery" and create a thread at ->sync_thread.
8235 * When the thread finishes it sets MD_RECOVERY_DONE
8236 * and wakeups up this thread which will reap the thread and finish up.
8237 * This thread also removes any faulty devices (with nr_pending == 0).
8239 * The overall approach is:
8240 * 1/ if the superblock needs updating, update it.
8241 * 2/ If a recovery thread is running, don't do anything else.
8242 * 3/ If recovery has finished, clean up, possibly marking spares active.
8243 * 4/ If there are any faulty devices, remove them.
8244 * 5/ If array is degraded, try to add spares devices
8245 * 6/ If array has spares or is not in-sync, start a resync thread.
8247 void md_check_recovery(struct mddev *mddev)
8249 if (mddev->suspended)
8253 bitmap_daemon_work(mddev);
8255 if (signal_pending(current)) {
8256 if (mddev->pers->sync_request && !mddev->external) {
8257 printk(KERN_INFO "md: %s in immediate safe mode\n",
8259 mddev->safemode = 2;
8261 flush_signals(current);
8264 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8267 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8268 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8269 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8270 (mddev->external == 0 && mddev->safemode == 1) ||
8271 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8272 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8276 if (mddev_trylock(mddev)) {
8280 struct md_rdev *rdev;
8281 if (!mddev->external && mddev->in_sync)
8282 /* 'Blocked' flag not needed as failed devices
8283 * will be recorded if array switched to read/write.
8284 * Leaving it set will prevent the device
8285 * from being removed.
8287 rdev_for_each(rdev, mddev)
8288 clear_bit(Blocked, &rdev->flags);
8289 /* On a read-only array we can:
8290 * - remove failed devices
8291 * - add already-in_sync devices if the array itself
8293 * As we only add devices that are already in-sync,
8294 * we can activate the spares immediately.
8296 remove_and_add_spares(mddev, NULL);
8297 /* There is no thread, but we need to call
8298 * ->spare_active and clear saved_raid_disk
8300 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8301 md_reap_sync_thread(mddev);
8302 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8303 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8304 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8308 if (!mddev->external) {
8310 spin_lock(&mddev->lock);
8311 if (mddev->safemode &&
8312 !atomic_read(&mddev->writes_pending) &&
8314 mddev->recovery_cp == MaxSector) {
8317 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8319 if (mddev->safemode == 1)
8320 mddev->safemode = 0;
8321 spin_unlock(&mddev->lock);
8323 sysfs_notify_dirent_safe(mddev->sysfs_state);
8326 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8327 md_update_sb(mddev, 0);
8329 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8330 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8331 /* resync/recovery still happening */
8332 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8335 if (mddev->sync_thread) {
8336 md_reap_sync_thread(mddev);
8339 /* Set RUNNING before clearing NEEDED to avoid
8340 * any transients in the value of "sync_action".
8342 mddev->curr_resync_completed = 0;
8343 spin_lock(&mddev->lock);
8344 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8345 spin_unlock(&mddev->lock);
8346 /* Clear some bits that don't mean anything, but
8349 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8350 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8352 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8353 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8355 /* no recovery is running.
8356 * remove any failed drives, then
8357 * add spares if possible.
8358 * Spares are also removed and re-added, to allow
8359 * the personality to fail the re-add.
8362 if (mddev->reshape_position != MaxSector) {
8363 if (mddev->pers->check_reshape == NULL ||
8364 mddev->pers->check_reshape(mddev) != 0)
8365 /* Cannot proceed */
8367 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8368 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8369 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8370 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8371 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8372 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8373 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8374 } else if (mddev->recovery_cp < MaxSector) {
8375 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8376 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8377 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8378 /* nothing to be done ... */
8381 if (mddev->pers->sync_request) {
8383 /* We are adding a device or devices to an array
8384 * which has the bitmap stored on all devices.
8385 * So make sure all bitmap pages get written
8387 bitmap_write_all(mddev->bitmap);
8389 INIT_WORK(&mddev->del_work, md_start_sync);
8390 queue_work(md_misc_wq, &mddev->del_work);
8394 if (!mddev->sync_thread) {
8395 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8396 wake_up(&resync_wait);
8397 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8399 if (mddev->sysfs_action)
8400 sysfs_notify_dirent_safe(mddev->sysfs_action);
8403 wake_up(&mddev->sb_wait);
8404 mddev_unlock(mddev);
8407 EXPORT_SYMBOL(md_check_recovery);
8409 void md_reap_sync_thread(struct mddev *mddev)
8411 struct md_rdev *rdev;
8413 /* resync has finished, collect result */
8414 md_unregister_thread(&mddev->sync_thread);
8415 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8416 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8418 /* activate any spares */
8419 if (mddev->pers->spare_active(mddev)) {
8420 sysfs_notify(&mddev->kobj, NULL,
8422 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8425 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8426 mddev->pers->finish_reshape)
8427 mddev->pers->finish_reshape(mddev);
8429 /* If array is no-longer degraded, then any saved_raid_disk
8430 * information must be scrapped.
8432 if (!mddev->degraded)
8433 rdev_for_each(rdev, mddev)
8434 rdev->saved_raid_disk = -1;
8436 md_update_sb(mddev, 1);
8437 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8438 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8439 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8440 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8441 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8442 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8443 wake_up(&resync_wait);
8444 /* flag recovery needed just to double check */
8445 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8446 sysfs_notify_dirent_safe(mddev->sysfs_action);
8447 md_new_event(mddev);
8448 if (mddev->event_work.func)
8449 queue_work(md_misc_wq, &mddev->event_work);
8451 EXPORT_SYMBOL(md_reap_sync_thread);
8453 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8455 sysfs_notify_dirent_safe(rdev->sysfs_state);
8456 wait_event_timeout(rdev->blocked_wait,
8457 !test_bit(Blocked, &rdev->flags) &&
8458 !test_bit(BlockedBadBlocks, &rdev->flags),
8459 msecs_to_jiffies(5000));
8460 rdev_dec_pending(rdev, mddev);
8462 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8464 void md_finish_reshape(struct mddev *mddev)
8466 /* called be personality module when reshape completes. */
8467 struct md_rdev *rdev;
8469 rdev_for_each(rdev, mddev) {
8470 if (rdev->data_offset > rdev->new_data_offset)
8471 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8473 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8474 rdev->data_offset = rdev->new_data_offset;
8477 EXPORT_SYMBOL(md_finish_reshape);
8479 /* Bad block management.
8480 * We can record which blocks on each device are 'bad' and so just
8481 * fail those blocks, or that stripe, rather than the whole device.
8482 * Entries in the bad-block table are 64bits wide. This comprises:
8483 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8484 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8485 * A 'shift' can be set so that larger blocks are tracked and
8486 * consequently larger devices can be covered.
8487 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8489 * Locking of the bad-block table uses a seqlock so md_is_badblock
8490 * might need to retry if it is very unlucky.
8491 * We will sometimes want to check for bad blocks in a bi_end_io function,
8492 * so we use the write_seqlock_irq variant.
8494 * When looking for a bad block we specify a range and want to
8495 * know if any block in the range is bad. So we binary-search
8496 * to the last range that starts at-or-before the given endpoint,
8497 * (or "before the sector after the target range")
8498 * then see if it ends after the given start.
8500 * 0 if there are no known bad blocks in the range
8501 * 1 if there are known bad block which are all acknowledged
8502 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8503 * plus the start/length of the first bad section we overlap.
8505 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8506 sector_t *first_bad, int *bad_sectors)
8512 sector_t target = s + sectors;
8515 if (bb->shift > 0) {
8516 /* round the start down, and the end up */
8518 target += (1<<bb->shift) - 1;
8519 target >>= bb->shift;
8520 sectors = target - s;
8522 /* 'target' is now the first block after the bad range */
8525 seq = read_seqbegin(&bb->lock);
8530 /* Binary search between lo and hi for 'target'
8531 * i.e. for the last range that starts before 'target'
8533 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8534 * are known not to be the last range before target.
8535 * VARIANT: hi-lo is the number of possible
8536 * ranges, and decreases until it reaches 1
8538 while (hi - lo > 1) {
8539 int mid = (lo + hi) / 2;
8540 sector_t a = BB_OFFSET(p[mid]);
8542 /* This could still be the one, earlier ranges
8546 /* This and later ranges are definitely out. */
8549 /* 'lo' might be the last that started before target, but 'hi' isn't */
8551 /* need to check all range that end after 's' to see if
8552 * any are unacknowledged.
8555 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8556 if (BB_OFFSET(p[lo]) < target) {
8557 /* starts before the end, and finishes after
8558 * the start, so they must overlap
8560 if (rv != -1 && BB_ACK(p[lo]))
8564 *first_bad = BB_OFFSET(p[lo]);
8565 *bad_sectors = BB_LEN(p[lo]);
8571 if (read_seqretry(&bb->lock, seq))
8576 EXPORT_SYMBOL_GPL(md_is_badblock);
8579 * Add a range of bad blocks to the table.
8580 * This might extend the table, or might contract it
8581 * if two adjacent ranges can be merged.
8582 * We binary-search to find the 'insertion' point, then
8583 * decide how best to handle it.
8585 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8591 unsigned long flags;
8594 /* badblocks are disabled */
8598 /* round the start down, and the end up */
8599 sector_t next = s + sectors;
8601 next += (1<<bb->shift) - 1;
8606 write_seqlock_irqsave(&bb->lock, flags);
8611 /* Find the last range that starts at-or-before 's' */
8612 while (hi - lo > 1) {
8613 int mid = (lo + hi) / 2;
8614 sector_t a = BB_OFFSET(p[mid]);
8620 if (hi > lo && BB_OFFSET(p[lo]) > s)
8624 /* we found a range that might merge with the start
8627 sector_t a = BB_OFFSET(p[lo]);
8628 sector_t e = a + BB_LEN(p[lo]);
8629 int ack = BB_ACK(p[lo]);
8631 /* Yes, we can merge with a previous range */
8632 if (s == a && s + sectors >= e)
8633 /* new range covers old */
8636 ack = ack && acknowledged;
8638 if (e < s + sectors)
8640 if (e - a <= BB_MAX_LEN) {
8641 p[lo] = BB_MAKE(a, e-a, ack);
8644 /* does not all fit in one range,
8645 * make p[lo] maximal
8647 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8648 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8654 if (sectors && hi < bb->count) {
8655 /* 'hi' points to the first range that starts after 's'.
8656 * Maybe we can merge with the start of that range */
8657 sector_t a = BB_OFFSET(p[hi]);
8658 sector_t e = a + BB_LEN(p[hi]);
8659 int ack = BB_ACK(p[hi]);
8660 if (a <= s + sectors) {
8661 /* merging is possible */
8662 if (e <= s + sectors) {
8667 ack = ack && acknowledged;
8670 if (e - a <= BB_MAX_LEN) {
8671 p[hi] = BB_MAKE(a, e-a, ack);
8674 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8682 if (sectors == 0 && hi < bb->count) {
8683 /* we might be able to combine lo and hi */
8684 /* Note: 's' is at the end of 'lo' */
8685 sector_t a = BB_OFFSET(p[hi]);
8686 int lolen = BB_LEN(p[lo]);
8687 int hilen = BB_LEN(p[hi]);
8688 int newlen = lolen + hilen - (s - a);
8689 if (s >= a && newlen < BB_MAX_LEN) {
8690 /* yes, we can combine them */
8691 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8692 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8693 memmove(p + hi, p + hi + 1,
8694 (bb->count - hi - 1) * 8);
8699 /* didn't merge (it all).
8700 * Need to add a range just before 'hi' */
8701 if (bb->count >= MD_MAX_BADBLOCKS) {
8702 /* No room for more */
8706 int this_sectors = sectors;
8707 memmove(p + hi + 1, p + hi,
8708 (bb->count - hi) * 8);
8711 if (this_sectors > BB_MAX_LEN)
8712 this_sectors = BB_MAX_LEN;
8713 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8714 sectors -= this_sectors;
8721 bb->unacked_exist = 1;
8722 write_sequnlock_irqrestore(&bb->lock, flags);
8727 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8732 s += rdev->new_data_offset;
8734 s += rdev->data_offset;
8735 rv = md_set_badblocks(&rdev->badblocks,
8738 /* Make sure they get written out promptly */
8739 sysfs_notify_dirent_safe(rdev->sysfs_state);
8740 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8741 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8742 md_wakeup_thread(rdev->mddev->thread);
8746 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8749 * Remove a range of bad blocks from the table.
8750 * This may involve extending the table if we spilt a region,
8751 * but it must not fail. So if the table becomes full, we just
8752 * drop the remove request.
8754 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8758 sector_t target = s + sectors;
8761 if (bb->shift > 0) {
8762 /* When clearing we round the start up and the end down.
8763 * This should not matter as the shift should align with
8764 * the block size and no rounding should ever be needed.
8765 * However it is better the think a block is bad when it
8766 * isn't than to think a block is not bad when it is.
8768 s += (1<<bb->shift) - 1;
8770 target >>= bb->shift;
8771 sectors = target - s;
8774 write_seqlock_irq(&bb->lock);
8779 /* Find the last range that starts before 'target' */
8780 while (hi - lo > 1) {
8781 int mid = (lo + hi) / 2;
8782 sector_t a = BB_OFFSET(p[mid]);
8789 /* p[lo] is the last range that could overlap the
8790 * current range. Earlier ranges could also overlap,
8791 * but only this one can overlap the end of the range.
8793 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8794 /* Partial overlap, leave the tail of this range */
8795 int ack = BB_ACK(p[lo]);
8796 sector_t a = BB_OFFSET(p[lo]);
8797 sector_t end = a + BB_LEN(p[lo]);
8800 /* we need to split this range */
8801 if (bb->count >= MD_MAX_BADBLOCKS) {
8805 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8807 p[lo] = BB_MAKE(a, s-a, ack);
8810 p[lo] = BB_MAKE(target, end - target, ack);
8811 /* there is no longer an overlap */
8816 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8817 /* This range does overlap */
8818 if (BB_OFFSET(p[lo]) < s) {
8819 /* Keep the early parts of this range. */
8820 int ack = BB_ACK(p[lo]);
8821 sector_t start = BB_OFFSET(p[lo]);
8822 p[lo] = BB_MAKE(start, s - start, ack);
8823 /* now low doesn't overlap, so.. */
8828 /* 'lo' is strictly before, 'hi' is strictly after,
8829 * anything between needs to be discarded
8832 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8833 bb->count -= (hi - lo - 1);
8839 write_sequnlock_irq(&bb->lock);
8843 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8847 s += rdev->new_data_offset;
8849 s += rdev->data_offset;
8850 return md_clear_badblocks(&rdev->badblocks,
8853 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8856 * Acknowledge all bad blocks in a list.
8857 * This only succeeds if ->changed is clear. It is used by
8858 * in-kernel metadata updates
8860 void md_ack_all_badblocks(struct badblocks *bb)
8862 if (bb->page == NULL || bb->changed)
8863 /* no point even trying */
8865 write_seqlock_irq(&bb->lock);
8867 if (bb->changed == 0 && bb->unacked_exist) {
8870 for (i = 0; i < bb->count ; i++) {
8871 if (!BB_ACK(p[i])) {
8872 sector_t start = BB_OFFSET(p[i]);
8873 int len = BB_LEN(p[i]);
8874 p[i] = BB_MAKE(start, len, 1);
8877 bb->unacked_exist = 0;
8879 write_sequnlock_irq(&bb->lock);
8881 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8883 /* sysfs access to bad-blocks list.
8884 * We present two files.
8885 * 'bad-blocks' lists sector numbers and lengths of ranges that
8886 * are recorded as bad. The list is truncated to fit within
8887 * the one-page limit of sysfs.
8888 * Writing "sector length" to this file adds an acknowledged
8890 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8891 * been acknowledged. Writing to this file adds bad blocks
8892 * without acknowledging them. This is largely for testing.
8896 badblocks_show(struct badblocks *bb, char *page, int unack)
8907 seq = read_seqbegin(&bb->lock);
8912 while (len < PAGE_SIZE && i < bb->count) {
8913 sector_t s = BB_OFFSET(p[i]);
8914 unsigned int length = BB_LEN(p[i]);
8915 int ack = BB_ACK(p[i]);
8921 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8922 (unsigned long long)s << bb->shift,
8923 length << bb->shift);
8925 if (unack && len == 0)
8926 bb->unacked_exist = 0;
8928 if (read_seqretry(&bb->lock, seq))
8937 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8939 unsigned long long sector;
8943 /* Allow clearing via sysfs *only* for testing/debugging.
8944 * Normally only a successful write may clear a badblock
8947 if (page[0] == '-') {
8951 #endif /* DO_DEBUG */
8953 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8955 if (newline != '\n')
8967 md_clear_badblocks(bb, sector, length);
8970 #endif /* DO_DEBUG */
8971 if (md_set_badblocks(bb, sector, length, !unack))
8977 static int md_notify_reboot(struct notifier_block *this,
8978 unsigned long code, void *x)
8980 struct list_head *tmp;
8981 struct mddev *mddev;
8984 for_each_mddev(mddev, tmp) {
8985 if (mddev_trylock(mddev)) {
8987 __md_stop_writes(mddev);
8988 if (mddev->persistent)
8989 mddev->safemode = 2;
8990 mddev_unlock(mddev);
8995 * certain more exotic SCSI devices are known to be
8996 * volatile wrt too early system reboots. While the
8997 * right place to handle this issue is the given
8998 * driver, we do want to have a safe RAID driver ...
9006 static struct notifier_block md_notifier = {
9007 .notifier_call = md_notify_reboot,
9009 .priority = INT_MAX, /* before any real devices */
9012 static void md_geninit(void)
9014 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9016 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9019 static int __init md_init(void)
9023 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9027 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9031 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9034 if ((ret = register_blkdev(0, "mdp")) < 0)
9038 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9039 md_probe, NULL, NULL);
9040 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9041 md_probe, NULL, NULL);
9043 register_reboot_notifier(&md_notifier);
9044 raid_table_header = register_sysctl_table(raid_root_table);
9050 unregister_blkdev(MD_MAJOR, "md");
9052 destroy_workqueue(md_misc_wq);
9054 destroy_workqueue(md_wq);
9059 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9061 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9062 struct md_rdev *rdev2;
9064 char b[BDEVNAME_SIZE];
9066 /* Check for change of roles in the active devices */
9067 rdev_for_each(rdev2, mddev) {
9068 if (test_bit(Faulty, &rdev2->flags))
9071 /* Check if the roles changed */
9072 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9074 if (test_bit(Candidate, &rdev2->flags)) {
9075 if (role == 0xfffe) {
9076 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9077 md_kick_rdev_from_array(rdev2);
9081 clear_bit(Candidate, &rdev2->flags);
9084 if (role != rdev2->raid_disk) {
9086 if (rdev2->raid_disk == -1 && role != 0xffff) {
9087 rdev2->saved_raid_disk = role;
9088 ret = remove_and_add_spares(mddev, rdev2);
9089 pr_info("Activated spare: %s\n",
9090 bdevname(rdev2->bdev,b));
9094 * We just want to do the minimum to mark the disk
9095 * as faulty. The recovery is performed by the
9096 * one who initiated the error.
9098 if ((role == 0xfffe) || (role == 0xfffd)) {
9099 md_error(mddev, rdev2);
9100 clear_bit(Blocked, &rdev2->flags);
9105 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9106 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9108 /* Finally set the event to be up to date */
9109 mddev->events = le64_to_cpu(sb->events);
9112 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9115 struct page *swapout = rdev->sb_page;
9116 struct mdp_superblock_1 *sb;
9118 /* Store the sb page of the rdev in the swapout temporary
9119 * variable in case we err in the future
9121 rdev->sb_page = NULL;
9122 alloc_disk_sb(rdev);
9123 ClearPageUptodate(rdev->sb_page);
9124 rdev->sb_loaded = 0;
9125 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9128 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9129 __func__, __LINE__, rdev->desc_nr, err);
9130 put_page(rdev->sb_page);
9131 rdev->sb_page = swapout;
9132 rdev->sb_loaded = 1;
9136 sb = page_address(rdev->sb_page);
9137 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9141 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9142 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9144 /* The other node finished recovery, call spare_active to set
9145 * device In_sync and mddev->degraded
9147 if (rdev->recovery_offset == MaxSector &&
9148 !test_bit(In_sync, &rdev->flags) &&
9149 mddev->pers->spare_active(mddev))
9150 sysfs_notify(&mddev->kobj, NULL, "degraded");
9156 void md_reload_sb(struct mddev *mddev, int nr)
9158 struct md_rdev *rdev;
9162 rdev_for_each_rcu(rdev, mddev) {
9163 if (rdev->desc_nr == nr)
9167 if (!rdev || rdev->desc_nr != nr) {
9168 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9172 err = read_rdev(mddev, rdev);
9176 check_sb_changes(mddev, rdev);
9178 /* Read all rdev's to update recovery_offset */
9179 rdev_for_each_rcu(rdev, mddev)
9180 read_rdev(mddev, rdev);
9182 EXPORT_SYMBOL(md_reload_sb);
9187 * Searches all registered partitions for autorun RAID arrays
9191 static LIST_HEAD(all_detected_devices);
9192 struct detected_devices_node {
9193 struct list_head list;
9197 void md_autodetect_dev(dev_t dev)
9199 struct detected_devices_node *node_detected_dev;
9201 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9202 if (node_detected_dev) {
9203 node_detected_dev->dev = dev;
9204 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9206 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9207 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9211 static void autostart_arrays(int part)
9213 struct md_rdev *rdev;
9214 struct detected_devices_node *node_detected_dev;
9216 int i_scanned, i_passed;
9221 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
9223 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9225 node_detected_dev = list_entry(all_detected_devices.next,
9226 struct detected_devices_node, list);
9227 list_del(&node_detected_dev->list);
9228 dev = node_detected_dev->dev;
9229 kfree(node_detected_dev);
9230 rdev = md_import_device(dev,0, 90);
9234 if (test_bit(Faulty, &rdev->flags))
9237 set_bit(AutoDetected, &rdev->flags);
9238 list_add(&rdev->same_set, &pending_raid_disks);
9242 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9243 i_scanned, i_passed);
9245 autorun_devices(part);
9248 #endif /* !MODULE */
9250 static __exit void md_exit(void)
9252 struct mddev *mddev;
9253 struct list_head *tmp;
9256 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9257 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9259 unregister_blkdev(MD_MAJOR,"md");
9260 unregister_blkdev(mdp_major, "mdp");
9261 unregister_reboot_notifier(&md_notifier);
9262 unregister_sysctl_table(raid_table_header);
9264 /* We cannot unload the modules while some process is
9265 * waiting for us in select() or poll() - wake them up
9268 while (waitqueue_active(&md_event_waiters)) {
9269 /* not safe to leave yet */
9270 wake_up(&md_event_waiters);
9274 remove_proc_entry("mdstat", NULL);
9276 for_each_mddev(mddev, tmp) {
9277 export_array(mddev);
9278 mddev->hold_active = 0;
9280 destroy_workqueue(md_misc_wq);
9281 destroy_workqueue(md_wq);
9284 subsys_initcall(md_init);
9285 module_exit(md_exit)
9287 static int get_ro(char *buffer, struct kernel_param *kp)
9289 return sprintf(buffer, "%d", start_readonly);
9291 static int set_ro(const char *val, struct kernel_param *kp)
9293 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9296 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9297 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9298 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9300 MODULE_LICENSE("GPL");
9301 MODULE_DESCRIPTION("MD RAID framework");
9303 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);