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 if (blk_integrity_register(mddev->gendisk,
2005 bdev_get_integrity(reference->bdev)) != 0) {
2006 printk(KERN_ERR "md: failed to register integrity for %s\n",
2010 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2011 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2012 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2018 EXPORT_SYMBOL(md_integrity_register);
2020 /* Disable data integrity if non-capable/non-matching disk is being added */
2021 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2023 struct blk_integrity *bi_rdev;
2024 struct blk_integrity *bi_mddev;
2026 if (!mddev->gendisk)
2029 bi_rdev = bdev_get_integrity(rdev->bdev);
2030 bi_mddev = blk_get_integrity(mddev->gendisk);
2032 if (!bi_mddev) /* nothing to do */
2034 if (rdev->raid_disk < 0) /* skip spares */
2036 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2037 rdev->bdev->bd_disk) >= 0)
2039 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2040 blk_integrity_unregister(mddev->gendisk);
2042 EXPORT_SYMBOL(md_integrity_add_rdev);
2044 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2046 char b[BDEVNAME_SIZE];
2050 /* prevent duplicates */
2051 if (find_rdev(mddev, rdev->bdev->bd_dev))
2054 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2055 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2056 rdev->sectors < mddev->dev_sectors)) {
2058 /* Cannot change size, so fail
2059 * If mddev->level <= 0, then we don't care
2060 * about aligning sizes (e.g. linear)
2062 if (mddev->level > 0)
2065 mddev->dev_sectors = rdev->sectors;
2068 /* Verify rdev->desc_nr is unique.
2069 * If it is -1, assign a free number, else
2070 * check number is not in use
2073 if (rdev->desc_nr < 0) {
2076 choice = mddev->raid_disks;
2077 while (md_find_rdev_nr_rcu(mddev, choice))
2079 rdev->desc_nr = choice;
2081 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2087 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2088 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2089 mdname(mddev), mddev->max_disks);
2092 bdevname(rdev->bdev,b);
2093 strreplace(b, '/', '!');
2095 rdev->mddev = mddev;
2096 printk(KERN_INFO "md: bind<%s>\n", b);
2098 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2101 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2102 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2103 /* failure here is OK */;
2104 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2106 list_add_rcu(&rdev->same_set, &mddev->disks);
2107 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2109 /* May as well allow recovery to be retried once */
2110 mddev->recovery_disabled++;
2115 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2120 static void md_delayed_delete(struct work_struct *ws)
2122 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2123 kobject_del(&rdev->kobj);
2124 kobject_put(&rdev->kobj);
2127 static void unbind_rdev_from_array(struct md_rdev *rdev)
2129 char b[BDEVNAME_SIZE];
2131 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2132 list_del_rcu(&rdev->same_set);
2133 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2135 sysfs_remove_link(&rdev->kobj, "block");
2136 sysfs_put(rdev->sysfs_state);
2137 rdev->sysfs_state = NULL;
2138 rdev->badblocks.count = 0;
2139 /* We need to delay this, otherwise we can deadlock when
2140 * writing to 'remove' to "dev/state". We also need
2141 * to delay it due to rcu usage.
2144 INIT_WORK(&rdev->del_work, md_delayed_delete);
2145 kobject_get(&rdev->kobj);
2146 queue_work(md_misc_wq, &rdev->del_work);
2150 * prevent the device from being mounted, repartitioned or
2151 * otherwise reused by a RAID array (or any other kernel
2152 * subsystem), by bd_claiming the device.
2154 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2157 struct block_device *bdev;
2158 char b[BDEVNAME_SIZE];
2160 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2161 shared ? (struct md_rdev *)lock_rdev : rdev);
2163 printk(KERN_ERR "md: could not open %s.\n",
2164 __bdevname(dev, b));
2165 return PTR_ERR(bdev);
2171 static void unlock_rdev(struct md_rdev *rdev)
2173 struct block_device *bdev = rdev->bdev;
2175 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2178 void md_autodetect_dev(dev_t dev);
2180 static void export_rdev(struct md_rdev *rdev)
2182 char b[BDEVNAME_SIZE];
2184 printk(KERN_INFO "md: export_rdev(%s)\n",
2185 bdevname(rdev->bdev,b));
2186 md_rdev_clear(rdev);
2188 if (test_bit(AutoDetected, &rdev->flags))
2189 md_autodetect_dev(rdev->bdev->bd_dev);
2192 kobject_put(&rdev->kobj);
2195 void md_kick_rdev_from_array(struct md_rdev *rdev)
2197 unbind_rdev_from_array(rdev);
2200 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2202 static void export_array(struct mddev *mddev)
2204 struct md_rdev *rdev;
2206 while (!list_empty(&mddev->disks)) {
2207 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2209 md_kick_rdev_from_array(rdev);
2211 mddev->raid_disks = 0;
2212 mddev->major_version = 0;
2215 static void sync_sbs(struct mddev *mddev, int nospares)
2217 /* Update each superblock (in-memory image), but
2218 * if we are allowed to, skip spares which already
2219 * have the right event counter, or have one earlier
2220 * (which would mean they aren't being marked as dirty
2221 * with the rest of the array)
2223 struct md_rdev *rdev;
2224 rdev_for_each(rdev, mddev) {
2225 if (rdev->sb_events == mddev->events ||
2227 rdev->raid_disk < 0 &&
2228 rdev->sb_events+1 == mddev->events)) {
2229 /* Don't update this superblock */
2230 rdev->sb_loaded = 2;
2232 sync_super(mddev, rdev);
2233 rdev->sb_loaded = 1;
2238 static bool does_sb_need_changing(struct mddev *mddev)
2240 struct md_rdev *rdev;
2241 struct mdp_superblock_1 *sb;
2244 /* Find a good rdev */
2245 rdev_for_each(rdev, mddev)
2246 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2249 /* No good device found. */
2253 sb = page_address(rdev->sb_page);
2254 /* Check if a device has become faulty or a spare become active */
2255 rdev_for_each(rdev, mddev) {
2256 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2257 /* Device activated? */
2258 if (role == 0xffff && rdev->raid_disk >=0 &&
2259 !test_bit(Faulty, &rdev->flags))
2261 /* Device turned faulty? */
2262 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2266 /* Check if any mddev parameters have changed */
2267 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2268 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2269 (mddev->layout != le64_to_cpu(sb->layout)) ||
2270 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2271 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2277 void md_update_sb(struct mddev *mddev, int force_change)
2279 struct md_rdev *rdev;
2282 int any_badblocks_changed = 0;
2287 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2291 if (mddev_is_clustered(mddev)) {
2292 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2294 ret = md_cluster_ops->metadata_update_start(mddev);
2295 /* Has someone else has updated the sb */
2296 if (!does_sb_need_changing(mddev)) {
2298 md_cluster_ops->metadata_update_cancel(mddev);
2299 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2304 /* First make sure individual recovery_offsets are correct */
2305 rdev_for_each(rdev, mddev) {
2306 if (rdev->raid_disk >= 0 &&
2307 mddev->delta_disks >= 0 &&
2308 !test_bit(Journal, &rdev->flags) &&
2309 !test_bit(In_sync, &rdev->flags) &&
2310 mddev->curr_resync_completed > rdev->recovery_offset)
2311 rdev->recovery_offset = mddev->curr_resync_completed;
2314 if (!mddev->persistent) {
2315 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2316 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2317 if (!mddev->external) {
2318 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2319 rdev_for_each(rdev, mddev) {
2320 if (rdev->badblocks.changed) {
2321 rdev->badblocks.changed = 0;
2322 md_ack_all_badblocks(&rdev->badblocks);
2323 md_error(mddev, rdev);
2325 clear_bit(Blocked, &rdev->flags);
2326 clear_bit(BlockedBadBlocks, &rdev->flags);
2327 wake_up(&rdev->blocked_wait);
2330 wake_up(&mddev->sb_wait);
2334 spin_lock(&mddev->lock);
2336 mddev->utime = get_seconds();
2338 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2340 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2341 /* just a clean<-> dirty transition, possibly leave spares alone,
2342 * though if events isn't the right even/odd, we will have to do
2348 if (mddev->degraded)
2349 /* If the array is degraded, then skipping spares is both
2350 * dangerous and fairly pointless.
2351 * Dangerous because a device that was removed from the array
2352 * might have a event_count that still looks up-to-date,
2353 * so it can be re-added without a resync.
2354 * Pointless because if there are any spares to skip,
2355 * then a recovery will happen and soon that array won't
2356 * be degraded any more and the spare can go back to sleep then.
2360 sync_req = mddev->in_sync;
2362 /* If this is just a dirty<->clean transition, and the array is clean
2363 * and 'events' is odd, we can roll back to the previous clean state */
2365 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2366 && mddev->can_decrease_events
2367 && mddev->events != 1) {
2369 mddev->can_decrease_events = 0;
2371 /* otherwise we have to go forward and ... */
2373 mddev->can_decrease_events = nospares;
2377 * This 64-bit counter should never wrap.
2378 * Either we are in around ~1 trillion A.C., assuming
2379 * 1 reboot per second, or we have a bug...
2381 WARN_ON(mddev->events == 0);
2383 rdev_for_each(rdev, mddev) {
2384 if (rdev->badblocks.changed)
2385 any_badblocks_changed++;
2386 if (test_bit(Faulty, &rdev->flags))
2387 set_bit(FaultRecorded, &rdev->flags);
2390 sync_sbs(mddev, nospares);
2391 spin_unlock(&mddev->lock);
2393 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2394 mdname(mddev), mddev->in_sync);
2396 bitmap_update_sb(mddev->bitmap);
2397 rdev_for_each(rdev, mddev) {
2398 char b[BDEVNAME_SIZE];
2400 if (rdev->sb_loaded != 1)
2401 continue; /* no noise on spare devices */
2403 if (!test_bit(Faulty, &rdev->flags)) {
2404 md_super_write(mddev,rdev,
2405 rdev->sb_start, rdev->sb_size,
2407 pr_debug("md: (write) %s's sb offset: %llu\n",
2408 bdevname(rdev->bdev, b),
2409 (unsigned long long)rdev->sb_start);
2410 rdev->sb_events = mddev->events;
2411 if (rdev->badblocks.size) {
2412 md_super_write(mddev, rdev,
2413 rdev->badblocks.sector,
2414 rdev->badblocks.size << 9,
2416 rdev->badblocks.size = 0;
2420 pr_debug("md: %s (skipping faulty)\n",
2421 bdevname(rdev->bdev, b));
2423 if (mddev->level == LEVEL_MULTIPATH)
2424 /* only need to write one superblock... */
2427 md_super_wait(mddev);
2428 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2430 spin_lock(&mddev->lock);
2431 if (mddev->in_sync != sync_req ||
2432 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2433 /* have to write it out again */
2434 spin_unlock(&mddev->lock);
2437 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2438 spin_unlock(&mddev->lock);
2439 wake_up(&mddev->sb_wait);
2440 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2441 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2443 rdev_for_each(rdev, mddev) {
2444 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2445 clear_bit(Blocked, &rdev->flags);
2447 if (any_badblocks_changed)
2448 md_ack_all_badblocks(&rdev->badblocks);
2449 clear_bit(BlockedBadBlocks, &rdev->flags);
2450 wake_up(&rdev->blocked_wait);
2453 if (mddev_is_clustered(mddev) && ret == 0)
2454 md_cluster_ops->metadata_update_finish(mddev);
2456 EXPORT_SYMBOL(md_update_sb);
2458 static int add_bound_rdev(struct md_rdev *rdev)
2460 struct mddev *mddev = rdev->mddev;
2463 if (!mddev->pers->hot_remove_disk) {
2464 /* If there is hot_add_disk but no hot_remove_disk
2465 * then added disks for geometry changes,
2466 * and should be added immediately.
2468 super_types[mddev->major_version].
2469 validate_super(mddev, rdev);
2470 err = mddev->pers->hot_add_disk(mddev, rdev);
2472 unbind_rdev_from_array(rdev);
2477 sysfs_notify_dirent_safe(rdev->sysfs_state);
2479 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2480 if (mddev->degraded)
2481 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2482 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2483 md_new_event(mddev);
2484 md_wakeup_thread(mddev->thread);
2488 /* words written to sysfs files may, or may not, be \n terminated.
2489 * We want to accept with case. For this we use cmd_match.
2491 static int cmd_match(const char *cmd, const char *str)
2493 /* See if cmd, written into a sysfs file, matches
2494 * str. They must either be the same, or cmd can
2495 * have a trailing newline
2497 while (*cmd && *str && *cmd == *str) {
2508 struct rdev_sysfs_entry {
2509 struct attribute attr;
2510 ssize_t (*show)(struct md_rdev *, char *);
2511 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2515 state_show(struct md_rdev *rdev, char *page)
2519 unsigned long flags = ACCESS_ONCE(rdev->flags);
2521 if (test_bit(Faulty, &flags) ||
2522 rdev->badblocks.unacked_exist) {
2523 len+= sprintf(page+len, "%sfaulty",sep);
2526 if (test_bit(In_sync, &flags)) {
2527 len += sprintf(page+len, "%sin_sync",sep);
2530 if (test_bit(Journal, &flags)) {
2531 len += sprintf(page+len, "%sjournal",sep);
2534 if (test_bit(WriteMostly, &flags)) {
2535 len += sprintf(page+len, "%swrite_mostly",sep);
2538 if (test_bit(Blocked, &flags) ||
2539 (rdev->badblocks.unacked_exist
2540 && !test_bit(Faulty, &flags))) {
2541 len += sprintf(page+len, "%sblocked", sep);
2544 if (!test_bit(Faulty, &flags) &&
2545 !test_bit(Journal, &flags) &&
2546 !test_bit(In_sync, &flags)) {
2547 len += sprintf(page+len, "%sspare", sep);
2550 if (test_bit(WriteErrorSeen, &flags)) {
2551 len += sprintf(page+len, "%swrite_error", sep);
2554 if (test_bit(WantReplacement, &flags)) {
2555 len += sprintf(page+len, "%swant_replacement", sep);
2558 if (test_bit(Replacement, &flags)) {
2559 len += sprintf(page+len, "%sreplacement", sep);
2563 return len+sprintf(page+len, "\n");
2567 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2570 * faulty - simulates an error
2571 * remove - disconnects the device
2572 * writemostly - sets write_mostly
2573 * -writemostly - clears write_mostly
2574 * blocked - sets the Blocked flags
2575 * -blocked - clears the Blocked and possibly simulates an error
2576 * insync - sets Insync providing device isn't active
2577 * -insync - clear Insync for a device with a slot assigned,
2578 * so that it gets rebuilt based on bitmap
2579 * write_error - sets WriteErrorSeen
2580 * -write_error - clears WriteErrorSeen
2583 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2584 md_error(rdev->mddev, rdev);
2585 if (test_bit(Faulty, &rdev->flags))
2589 } else if (cmd_match(buf, "remove")) {
2590 if (rdev->raid_disk >= 0)
2593 struct mddev *mddev = rdev->mddev;
2595 if (mddev_is_clustered(mddev))
2596 err = md_cluster_ops->remove_disk(mddev, rdev);
2599 md_kick_rdev_from_array(rdev);
2601 md_update_sb(mddev, 1);
2602 md_new_event(mddev);
2605 } else if (cmd_match(buf, "writemostly")) {
2606 set_bit(WriteMostly, &rdev->flags);
2608 } else if (cmd_match(buf, "-writemostly")) {
2609 clear_bit(WriteMostly, &rdev->flags);
2611 } else if (cmd_match(buf, "blocked")) {
2612 set_bit(Blocked, &rdev->flags);
2614 } else if (cmd_match(buf, "-blocked")) {
2615 if (!test_bit(Faulty, &rdev->flags) &&
2616 rdev->badblocks.unacked_exist) {
2617 /* metadata handler doesn't understand badblocks,
2618 * so we need to fail the device
2620 md_error(rdev->mddev, rdev);
2622 clear_bit(Blocked, &rdev->flags);
2623 clear_bit(BlockedBadBlocks, &rdev->flags);
2624 wake_up(&rdev->blocked_wait);
2625 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2626 md_wakeup_thread(rdev->mddev->thread);
2629 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2630 set_bit(In_sync, &rdev->flags);
2632 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2633 !test_bit(Journal, &rdev->flags)) {
2634 if (rdev->mddev->pers == NULL) {
2635 clear_bit(In_sync, &rdev->flags);
2636 rdev->saved_raid_disk = rdev->raid_disk;
2637 rdev->raid_disk = -1;
2640 } else if (cmd_match(buf, "write_error")) {
2641 set_bit(WriteErrorSeen, &rdev->flags);
2643 } else if (cmd_match(buf, "-write_error")) {
2644 clear_bit(WriteErrorSeen, &rdev->flags);
2646 } else if (cmd_match(buf, "want_replacement")) {
2647 /* Any non-spare device that is not a replacement can
2648 * become want_replacement at any time, but we then need to
2649 * check if recovery is needed.
2651 if (rdev->raid_disk >= 0 &&
2652 !test_bit(Journal, &rdev->flags) &&
2653 !test_bit(Replacement, &rdev->flags))
2654 set_bit(WantReplacement, &rdev->flags);
2655 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2656 md_wakeup_thread(rdev->mddev->thread);
2658 } else if (cmd_match(buf, "-want_replacement")) {
2659 /* Clearing 'want_replacement' is always allowed.
2660 * Once replacements starts it is too late though.
2663 clear_bit(WantReplacement, &rdev->flags);
2664 } else if (cmd_match(buf, "replacement")) {
2665 /* Can only set a device as a replacement when array has not
2666 * yet been started. Once running, replacement is automatic
2667 * from spares, or by assigning 'slot'.
2669 if (rdev->mddev->pers)
2672 set_bit(Replacement, &rdev->flags);
2675 } else if (cmd_match(buf, "-replacement")) {
2676 /* Similarly, can only clear Replacement before start */
2677 if (rdev->mddev->pers)
2680 clear_bit(Replacement, &rdev->flags);
2683 } else if (cmd_match(buf, "re-add")) {
2684 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2685 /* clear_bit is performed _after_ all the devices
2686 * have their local Faulty bit cleared. If any writes
2687 * happen in the meantime in the local node, they
2688 * will land in the local bitmap, which will be synced
2689 * by this node eventually
2691 if (!mddev_is_clustered(rdev->mddev) ||
2692 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2693 clear_bit(Faulty, &rdev->flags);
2694 err = add_bound_rdev(rdev);
2700 sysfs_notify_dirent_safe(rdev->sysfs_state);
2701 return err ? err : len;
2703 static struct rdev_sysfs_entry rdev_state =
2704 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2707 errors_show(struct md_rdev *rdev, char *page)
2709 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2713 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2718 rv = kstrtouint(buf, 10, &n);
2721 atomic_set(&rdev->corrected_errors, n);
2724 static struct rdev_sysfs_entry rdev_errors =
2725 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2728 slot_show(struct md_rdev *rdev, char *page)
2730 if (test_bit(Journal, &rdev->flags))
2731 return sprintf(page, "journal\n");
2732 else if (rdev->raid_disk < 0)
2733 return sprintf(page, "none\n");
2735 return sprintf(page, "%d\n", rdev->raid_disk);
2739 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2744 if (test_bit(Journal, &rdev->flags))
2746 if (strncmp(buf, "none", 4)==0)
2749 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2753 if (rdev->mddev->pers && slot == -1) {
2754 /* Setting 'slot' on an active array requires also
2755 * updating the 'rd%d' link, and communicating
2756 * with the personality with ->hot_*_disk.
2757 * For now we only support removing
2758 * failed/spare devices. This normally happens automatically,
2759 * but not when the metadata is externally managed.
2761 if (rdev->raid_disk == -1)
2763 /* personality does all needed checks */
2764 if (rdev->mddev->pers->hot_remove_disk == NULL)
2766 clear_bit(Blocked, &rdev->flags);
2767 remove_and_add_spares(rdev->mddev, rdev);
2768 if (rdev->raid_disk >= 0)
2770 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2771 md_wakeup_thread(rdev->mddev->thread);
2772 } else if (rdev->mddev->pers) {
2773 /* Activating a spare .. or possibly reactivating
2774 * if we ever get bitmaps working here.
2777 if (rdev->raid_disk != -1)
2780 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2783 if (rdev->mddev->pers->hot_add_disk == NULL)
2786 if (slot >= rdev->mddev->raid_disks &&
2787 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2790 rdev->raid_disk = slot;
2791 if (test_bit(In_sync, &rdev->flags))
2792 rdev->saved_raid_disk = slot;
2794 rdev->saved_raid_disk = -1;
2795 clear_bit(In_sync, &rdev->flags);
2796 clear_bit(Bitmap_sync, &rdev->flags);
2797 remove_and_add_spares(rdev->mddev, rdev);
2798 if (rdev->raid_disk == -1)
2800 /* don't wakeup anyone, leave that to userspace. */
2802 if (slot >= rdev->mddev->raid_disks &&
2803 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2805 rdev->raid_disk = slot;
2806 /* assume it is working */
2807 clear_bit(Faulty, &rdev->flags);
2808 clear_bit(WriteMostly, &rdev->flags);
2809 set_bit(In_sync, &rdev->flags);
2810 sysfs_notify_dirent_safe(rdev->sysfs_state);
2815 static struct rdev_sysfs_entry rdev_slot =
2816 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2819 offset_show(struct md_rdev *rdev, char *page)
2821 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2825 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2827 unsigned long long offset;
2828 if (kstrtoull(buf, 10, &offset) < 0)
2830 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2832 if (rdev->sectors && rdev->mddev->external)
2833 /* Must set offset before size, so overlap checks
2836 rdev->data_offset = offset;
2837 rdev->new_data_offset = offset;
2841 static struct rdev_sysfs_entry rdev_offset =
2842 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2844 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2846 return sprintf(page, "%llu\n",
2847 (unsigned long long)rdev->new_data_offset);
2850 static ssize_t new_offset_store(struct md_rdev *rdev,
2851 const char *buf, size_t len)
2853 unsigned long long new_offset;
2854 struct mddev *mddev = rdev->mddev;
2856 if (kstrtoull(buf, 10, &new_offset) < 0)
2859 if (mddev->sync_thread ||
2860 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2862 if (new_offset == rdev->data_offset)
2863 /* reset is always permitted */
2865 else if (new_offset > rdev->data_offset) {
2866 /* must not push array size beyond rdev_sectors */
2867 if (new_offset - rdev->data_offset
2868 + mddev->dev_sectors > rdev->sectors)
2871 /* Metadata worries about other space details. */
2873 /* decreasing the offset is inconsistent with a backwards
2876 if (new_offset < rdev->data_offset &&
2877 mddev->reshape_backwards)
2879 /* Increasing offset is inconsistent with forwards
2880 * reshape. reshape_direction should be set to
2881 * 'backwards' first.
2883 if (new_offset > rdev->data_offset &&
2884 !mddev->reshape_backwards)
2887 if (mddev->pers && mddev->persistent &&
2888 !super_types[mddev->major_version]
2889 .allow_new_offset(rdev, new_offset))
2891 rdev->new_data_offset = new_offset;
2892 if (new_offset > rdev->data_offset)
2893 mddev->reshape_backwards = 1;
2894 else if (new_offset < rdev->data_offset)
2895 mddev->reshape_backwards = 0;
2899 static struct rdev_sysfs_entry rdev_new_offset =
2900 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2903 rdev_size_show(struct md_rdev *rdev, char *page)
2905 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2908 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2910 /* check if two start/length pairs overlap */
2918 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2920 unsigned long long blocks;
2923 if (kstrtoull(buf, 10, &blocks) < 0)
2926 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2927 return -EINVAL; /* sector conversion overflow */
2930 if (new != blocks * 2)
2931 return -EINVAL; /* unsigned long long to sector_t overflow */
2938 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2940 struct mddev *my_mddev = rdev->mddev;
2941 sector_t oldsectors = rdev->sectors;
2944 if (test_bit(Journal, &rdev->flags))
2946 if (strict_blocks_to_sectors(buf, §ors) < 0)
2948 if (rdev->data_offset != rdev->new_data_offset)
2949 return -EINVAL; /* too confusing */
2950 if (my_mddev->pers && rdev->raid_disk >= 0) {
2951 if (my_mddev->persistent) {
2952 sectors = super_types[my_mddev->major_version].
2953 rdev_size_change(rdev, sectors);
2956 } else if (!sectors)
2957 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2959 if (!my_mddev->pers->resize)
2960 /* Cannot change size for RAID0 or Linear etc */
2963 if (sectors < my_mddev->dev_sectors)
2964 return -EINVAL; /* component must fit device */
2966 rdev->sectors = sectors;
2967 if (sectors > oldsectors && my_mddev->external) {
2968 /* Need to check that all other rdevs with the same
2969 * ->bdev do not overlap. 'rcu' is sufficient to walk
2970 * the rdev lists safely.
2971 * This check does not provide a hard guarantee, it
2972 * just helps avoid dangerous mistakes.
2974 struct mddev *mddev;
2976 struct list_head *tmp;
2979 for_each_mddev(mddev, tmp) {
2980 struct md_rdev *rdev2;
2982 rdev_for_each(rdev2, mddev)
2983 if (rdev->bdev == rdev2->bdev &&
2985 overlaps(rdev->data_offset, rdev->sectors,
2998 /* Someone else could have slipped in a size
2999 * change here, but doing so is just silly.
3000 * We put oldsectors back because we *know* it is
3001 * safe, and trust userspace not to race with
3004 rdev->sectors = oldsectors;
3011 static struct rdev_sysfs_entry rdev_size =
3012 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3014 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3016 unsigned long long recovery_start = rdev->recovery_offset;
3018 if (test_bit(In_sync, &rdev->flags) ||
3019 recovery_start == MaxSector)
3020 return sprintf(page, "none\n");
3022 return sprintf(page, "%llu\n", recovery_start);
3025 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3027 unsigned long long recovery_start;
3029 if (cmd_match(buf, "none"))
3030 recovery_start = MaxSector;
3031 else if (kstrtoull(buf, 10, &recovery_start))
3034 if (rdev->mddev->pers &&
3035 rdev->raid_disk >= 0)
3038 rdev->recovery_offset = recovery_start;
3039 if (recovery_start == MaxSector)
3040 set_bit(In_sync, &rdev->flags);
3042 clear_bit(In_sync, &rdev->flags);
3046 static struct rdev_sysfs_entry rdev_recovery_start =
3047 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3050 badblocks_show(struct badblocks *bb, char *page, int unack);
3052 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3054 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3056 return badblocks_show(&rdev->badblocks, page, 0);
3058 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3060 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3061 /* Maybe that ack was all we needed */
3062 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3063 wake_up(&rdev->blocked_wait);
3066 static struct rdev_sysfs_entry rdev_bad_blocks =
3067 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3069 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3071 return badblocks_show(&rdev->badblocks, page, 1);
3073 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3075 return badblocks_store(&rdev->badblocks, page, len, 1);
3077 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3078 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3080 static struct attribute *rdev_default_attrs[] = {
3085 &rdev_new_offset.attr,
3087 &rdev_recovery_start.attr,
3088 &rdev_bad_blocks.attr,
3089 &rdev_unack_bad_blocks.attr,
3093 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3095 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3096 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3102 return entry->show(rdev, page);
3106 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3107 const char *page, size_t length)
3109 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3110 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3112 struct mddev *mddev = rdev->mddev;
3116 if (!capable(CAP_SYS_ADMIN))
3118 rv = mddev ? mddev_lock(mddev): -EBUSY;
3120 if (rdev->mddev == NULL)
3123 rv = entry->store(rdev, page, length);
3124 mddev_unlock(mddev);
3129 static void rdev_free(struct kobject *ko)
3131 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3134 static const struct sysfs_ops rdev_sysfs_ops = {
3135 .show = rdev_attr_show,
3136 .store = rdev_attr_store,
3138 static struct kobj_type rdev_ktype = {
3139 .release = rdev_free,
3140 .sysfs_ops = &rdev_sysfs_ops,
3141 .default_attrs = rdev_default_attrs,
3144 int md_rdev_init(struct md_rdev *rdev)
3147 rdev->saved_raid_disk = -1;
3148 rdev->raid_disk = -1;
3150 rdev->data_offset = 0;
3151 rdev->new_data_offset = 0;
3152 rdev->sb_events = 0;
3153 rdev->last_read_error.tv_sec = 0;
3154 rdev->last_read_error.tv_nsec = 0;
3155 rdev->sb_loaded = 0;
3156 rdev->bb_page = NULL;
3157 atomic_set(&rdev->nr_pending, 0);
3158 atomic_set(&rdev->read_errors, 0);
3159 atomic_set(&rdev->corrected_errors, 0);
3161 INIT_LIST_HEAD(&rdev->same_set);
3162 init_waitqueue_head(&rdev->blocked_wait);
3164 /* Add space to store bad block list.
3165 * This reserves the space even on arrays where it cannot
3166 * be used - I wonder if that matters
3168 rdev->badblocks.count = 0;
3169 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3170 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3171 seqlock_init(&rdev->badblocks.lock);
3172 if (rdev->badblocks.page == NULL)
3177 EXPORT_SYMBOL_GPL(md_rdev_init);
3179 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3181 * mark the device faulty if:
3183 * - the device is nonexistent (zero size)
3184 * - the device has no valid superblock
3186 * a faulty rdev _never_ has rdev->sb set.
3188 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3190 char b[BDEVNAME_SIZE];
3192 struct md_rdev *rdev;
3195 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3197 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3198 return ERR_PTR(-ENOMEM);
3201 err = md_rdev_init(rdev);
3204 err = alloc_disk_sb(rdev);
3208 err = lock_rdev(rdev, newdev, super_format == -2);
3212 kobject_init(&rdev->kobj, &rdev_ktype);
3214 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3217 "md: %s has zero or unknown size, marking faulty!\n",
3218 bdevname(rdev->bdev,b));
3223 if (super_format >= 0) {
3224 err = super_types[super_format].
3225 load_super(rdev, NULL, super_minor);
3226 if (err == -EINVAL) {
3228 "md: %s does not have a valid v%d.%d "
3229 "superblock, not importing!\n",
3230 bdevname(rdev->bdev,b),
3231 super_format, super_minor);
3236 "md: could not read %s's sb, not importing!\n",
3237 bdevname(rdev->bdev,b));
3247 md_rdev_clear(rdev);
3249 return ERR_PTR(err);
3253 * Check a full RAID array for plausibility
3256 static void analyze_sbs(struct mddev *mddev)
3259 struct md_rdev *rdev, *freshest, *tmp;
3260 char b[BDEVNAME_SIZE];
3263 rdev_for_each_safe(rdev, tmp, mddev)
3264 switch (super_types[mddev->major_version].
3265 load_super(rdev, freshest, mddev->minor_version)) {
3273 "md: fatal superblock inconsistency in %s"
3274 " -- removing from array\n",
3275 bdevname(rdev->bdev,b));
3276 md_kick_rdev_from_array(rdev);
3279 super_types[mddev->major_version].
3280 validate_super(mddev, freshest);
3283 rdev_for_each_safe(rdev, tmp, mddev) {
3284 if (mddev->max_disks &&
3285 (rdev->desc_nr >= mddev->max_disks ||
3286 i > mddev->max_disks)) {
3288 "md: %s: %s: only %d devices permitted\n",
3289 mdname(mddev), bdevname(rdev->bdev, b),
3291 md_kick_rdev_from_array(rdev);
3294 if (rdev != freshest) {
3295 if (super_types[mddev->major_version].
3296 validate_super(mddev, rdev)) {
3297 printk(KERN_WARNING "md: kicking non-fresh %s"
3299 bdevname(rdev->bdev,b));
3300 md_kick_rdev_from_array(rdev);
3304 if (mddev->level == LEVEL_MULTIPATH) {
3305 rdev->desc_nr = i++;
3306 rdev->raid_disk = rdev->desc_nr;
3307 set_bit(In_sync, &rdev->flags);
3308 } else if (rdev->raid_disk >=
3309 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3310 !test_bit(Journal, &rdev->flags)) {
3311 rdev->raid_disk = -1;
3312 clear_bit(In_sync, &rdev->flags);
3317 /* Read a fixed-point number.
3318 * Numbers in sysfs attributes should be in "standard" units where
3319 * possible, so time should be in seconds.
3320 * However we internally use a a much smaller unit such as
3321 * milliseconds or jiffies.
3322 * This function takes a decimal number with a possible fractional
3323 * component, and produces an integer which is the result of
3324 * multiplying that number by 10^'scale'.
3325 * all without any floating-point arithmetic.
3327 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3329 unsigned long result = 0;
3331 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3334 else if (decimals < scale) {
3337 result = result * 10 + value;
3349 while (decimals < scale) {
3358 safe_delay_show(struct mddev *mddev, char *page)
3360 int msec = (mddev->safemode_delay*1000)/HZ;
3361 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3364 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3368 if (mddev_is_clustered(mddev)) {
3369 pr_info("md: Safemode is disabled for clustered mode\n");
3373 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3376 mddev->safemode_delay = 0;
3378 unsigned long old_delay = mddev->safemode_delay;
3379 unsigned long new_delay = (msec*HZ)/1000;
3383 mddev->safemode_delay = new_delay;
3384 if (new_delay < old_delay || old_delay == 0)
3385 mod_timer(&mddev->safemode_timer, jiffies+1);
3389 static struct md_sysfs_entry md_safe_delay =
3390 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3393 level_show(struct mddev *mddev, char *page)
3395 struct md_personality *p;
3397 spin_lock(&mddev->lock);
3400 ret = sprintf(page, "%s\n", p->name);
3401 else if (mddev->clevel[0])
3402 ret = sprintf(page, "%s\n", mddev->clevel);
3403 else if (mddev->level != LEVEL_NONE)
3404 ret = sprintf(page, "%d\n", mddev->level);
3407 spin_unlock(&mddev->lock);
3412 level_store(struct mddev *mddev, const char *buf, size_t len)
3417 struct md_personality *pers, *oldpers;
3419 void *priv, *oldpriv;
3420 struct md_rdev *rdev;
3422 if (slen == 0 || slen >= sizeof(clevel))
3425 rv = mddev_lock(mddev);
3429 if (mddev->pers == NULL) {
3430 strncpy(mddev->clevel, buf, slen);
3431 if (mddev->clevel[slen-1] == '\n')
3433 mddev->clevel[slen] = 0;
3434 mddev->level = LEVEL_NONE;
3442 /* request to change the personality. Need to ensure:
3443 * - array is not engaged in resync/recovery/reshape
3444 * - old personality can be suspended
3445 * - new personality will access other array.
3449 if (mddev->sync_thread ||
3450 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3451 mddev->reshape_position != MaxSector ||
3452 mddev->sysfs_active)
3456 if (!mddev->pers->quiesce) {
3457 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3458 mdname(mddev), mddev->pers->name);
3462 /* Now find the new personality */
3463 strncpy(clevel, buf, slen);
3464 if (clevel[slen-1] == '\n')
3467 if (kstrtol(clevel, 10, &level))
3470 if (request_module("md-%s", clevel) != 0)
3471 request_module("md-level-%s", clevel);
3472 spin_lock(&pers_lock);
3473 pers = find_pers(level, clevel);
3474 if (!pers || !try_module_get(pers->owner)) {
3475 spin_unlock(&pers_lock);
3476 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3480 spin_unlock(&pers_lock);
3482 if (pers == mddev->pers) {
3483 /* Nothing to do! */
3484 module_put(pers->owner);
3488 if (!pers->takeover) {
3489 module_put(pers->owner);
3490 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3491 mdname(mddev), clevel);
3496 rdev_for_each(rdev, mddev)
3497 rdev->new_raid_disk = rdev->raid_disk;
3499 /* ->takeover must set new_* and/or delta_disks
3500 * if it succeeds, and may set them when it fails.
3502 priv = pers->takeover(mddev);
3504 mddev->new_level = mddev->level;
3505 mddev->new_layout = mddev->layout;
3506 mddev->new_chunk_sectors = mddev->chunk_sectors;
3507 mddev->raid_disks -= mddev->delta_disks;
3508 mddev->delta_disks = 0;
3509 mddev->reshape_backwards = 0;
3510 module_put(pers->owner);
3511 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3512 mdname(mddev), clevel);
3517 /* Looks like we have a winner */
3518 mddev_suspend(mddev);
3519 mddev_detach(mddev);
3521 spin_lock(&mddev->lock);
3522 oldpers = mddev->pers;
3523 oldpriv = mddev->private;
3525 mddev->private = priv;
3526 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3527 mddev->level = mddev->new_level;
3528 mddev->layout = mddev->new_layout;
3529 mddev->chunk_sectors = mddev->new_chunk_sectors;
3530 mddev->delta_disks = 0;
3531 mddev->reshape_backwards = 0;
3532 mddev->degraded = 0;
3533 spin_unlock(&mddev->lock);
3535 if (oldpers->sync_request == NULL &&
3537 /* We are converting from a no-redundancy array
3538 * to a redundancy array and metadata is managed
3539 * externally so we need to be sure that writes
3540 * won't block due to a need to transition
3542 * until external management is started.
3545 mddev->safemode_delay = 0;
3546 mddev->safemode = 0;
3549 oldpers->free(mddev, oldpriv);
3551 if (oldpers->sync_request == NULL &&
3552 pers->sync_request != NULL) {
3553 /* need to add the md_redundancy_group */
3554 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3556 "md: cannot register extra attributes for %s\n",
3558 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3560 if (oldpers->sync_request != NULL &&
3561 pers->sync_request == NULL) {
3562 /* need to remove the md_redundancy_group */
3563 if (mddev->to_remove == NULL)
3564 mddev->to_remove = &md_redundancy_group;
3567 rdev_for_each(rdev, mddev) {
3568 if (rdev->raid_disk < 0)
3570 if (rdev->new_raid_disk >= mddev->raid_disks)
3571 rdev->new_raid_disk = -1;
3572 if (rdev->new_raid_disk == rdev->raid_disk)
3574 sysfs_unlink_rdev(mddev, rdev);
3576 rdev_for_each(rdev, mddev) {
3577 if (rdev->raid_disk < 0)
3579 if (rdev->new_raid_disk == rdev->raid_disk)
3581 rdev->raid_disk = rdev->new_raid_disk;
3582 if (rdev->raid_disk < 0)
3583 clear_bit(In_sync, &rdev->flags);
3585 if (sysfs_link_rdev(mddev, rdev))
3586 printk(KERN_WARNING "md: cannot register rd%d"
3587 " for %s after level change\n",
3588 rdev->raid_disk, mdname(mddev));
3592 if (pers->sync_request == NULL) {
3593 /* this is now an array without redundancy, so
3594 * it must always be in_sync
3597 del_timer_sync(&mddev->safemode_timer);
3599 blk_set_stacking_limits(&mddev->queue->limits);
3601 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3602 mddev_resume(mddev);
3604 md_update_sb(mddev, 1);
3605 sysfs_notify(&mddev->kobj, NULL, "level");
3606 md_new_event(mddev);
3609 mddev_unlock(mddev);
3613 static struct md_sysfs_entry md_level =
3614 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3617 layout_show(struct mddev *mddev, char *page)
3619 /* just a number, not meaningful for all levels */
3620 if (mddev->reshape_position != MaxSector &&
3621 mddev->layout != mddev->new_layout)
3622 return sprintf(page, "%d (%d)\n",
3623 mddev->new_layout, mddev->layout);
3624 return sprintf(page, "%d\n", mddev->layout);
3628 layout_store(struct mddev *mddev, const char *buf, size_t len)
3633 err = kstrtouint(buf, 10, &n);
3636 err = mddev_lock(mddev);
3641 if (mddev->pers->check_reshape == NULL)
3646 mddev->new_layout = n;
3647 err = mddev->pers->check_reshape(mddev);
3649 mddev->new_layout = mddev->layout;
3652 mddev->new_layout = n;
3653 if (mddev->reshape_position == MaxSector)
3656 mddev_unlock(mddev);
3659 static struct md_sysfs_entry md_layout =
3660 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3663 raid_disks_show(struct mddev *mddev, char *page)
3665 if (mddev->raid_disks == 0)
3667 if (mddev->reshape_position != MaxSector &&
3668 mddev->delta_disks != 0)
3669 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3670 mddev->raid_disks - mddev->delta_disks);
3671 return sprintf(page, "%d\n", mddev->raid_disks);
3674 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3677 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3682 err = kstrtouint(buf, 10, &n);
3686 err = mddev_lock(mddev);
3690 err = update_raid_disks(mddev, n);
3691 else if (mddev->reshape_position != MaxSector) {
3692 struct md_rdev *rdev;
3693 int olddisks = mddev->raid_disks - mddev->delta_disks;
3696 rdev_for_each(rdev, mddev) {
3698 rdev->data_offset < rdev->new_data_offset)
3701 rdev->data_offset > rdev->new_data_offset)
3705 mddev->delta_disks = n - olddisks;
3706 mddev->raid_disks = n;
3707 mddev->reshape_backwards = (mddev->delta_disks < 0);
3709 mddev->raid_disks = n;
3711 mddev_unlock(mddev);
3712 return err ? err : len;
3714 static struct md_sysfs_entry md_raid_disks =
3715 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3718 chunk_size_show(struct mddev *mddev, char *page)
3720 if (mddev->reshape_position != MaxSector &&
3721 mddev->chunk_sectors != mddev->new_chunk_sectors)
3722 return sprintf(page, "%d (%d)\n",
3723 mddev->new_chunk_sectors << 9,
3724 mddev->chunk_sectors << 9);
3725 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3729 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3734 err = kstrtoul(buf, 10, &n);
3738 err = mddev_lock(mddev);
3742 if (mddev->pers->check_reshape == NULL)
3747 mddev->new_chunk_sectors = n >> 9;
3748 err = mddev->pers->check_reshape(mddev);
3750 mddev->new_chunk_sectors = mddev->chunk_sectors;
3753 mddev->new_chunk_sectors = n >> 9;
3754 if (mddev->reshape_position == MaxSector)
3755 mddev->chunk_sectors = n >> 9;
3757 mddev_unlock(mddev);
3760 static struct md_sysfs_entry md_chunk_size =
3761 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3764 resync_start_show(struct mddev *mddev, char *page)
3766 if (mddev->recovery_cp == MaxSector)
3767 return sprintf(page, "none\n");
3768 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3772 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3774 unsigned long long n;
3777 if (cmd_match(buf, "none"))
3780 err = kstrtoull(buf, 10, &n);
3783 if (n != (sector_t)n)
3787 err = mddev_lock(mddev);
3790 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3794 mddev->recovery_cp = n;
3796 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3798 mddev_unlock(mddev);
3801 static struct md_sysfs_entry md_resync_start =
3802 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3803 resync_start_show, resync_start_store);
3806 * The array state can be:
3809 * No devices, no size, no level
3810 * Equivalent to STOP_ARRAY ioctl
3812 * May have some settings, but array is not active
3813 * all IO results in error
3814 * When written, doesn't tear down array, but just stops it
3815 * suspended (not supported yet)
3816 * All IO requests will block. The array can be reconfigured.
3817 * Writing this, if accepted, will block until array is quiescent
3819 * no resync can happen. no superblocks get written.
3820 * write requests fail
3822 * like readonly, but behaves like 'clean' on a write request.
3824 * clean - no pending writes, but otherwise active.
3825 * When written to inactive array, starts without resync
3826 * If a write request arrives then
3827 * if metadata is known, mark 'dirty' and switch to 'active'.
3828 * if not known, block and switch to write-pending
3829 * If written to an active array that has pending writes, then fails.
3831 * fully active: IO and resync can be happening.
3832 * When written to inactive array, starts with resync
3835 * clean, but writes are blocked waiting for 'active' to be written.
3838 * like active, but no writes have been seen for a while (100msec).
3841 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3842 write_pending, active_idle, bad_word};
3843 static char *array_states[] = {
3844 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3845 "write-pending", "active-idle", NULL };
3847 static int match_word(const char *word, char **list)
3850 for (n=0; list[n]; n++)
3851 if (cmd_match(word, list[n]))
3857 array_state_show(struct mddev *mddev, char *page)
3859 enum array_state st = inactive;
3872 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3874 else if (mddev->safemode)
3880 if (list_empty(&mddev->disks) &&
3881 mddev->raid_disks == 0 &&
3882 mddev->dev_sectors == 0)
3887 return sprintf(page, "%s\n", array_states[st]);
3890 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3891 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3892 static int do_md_run(struct mddev *mddev);
3893 static int restart_array(struct mddev *mddev);
3896 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3899 enum array_state st = match_word(buf, array_states);
3901 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3902 /* don't take reconfig_mutex when toggling between
3905 spin_lock(&mddev->lock);
3907 restart_array(mddev);
3908 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3909 wake_up(&mddev->sb_wait);
3911 } else /* st == clean */ {
3912 restart_array(mddev);
3913 if (atomic_read(&mddev->writes_pending) == 0) {
3914 if (mddev->in_sync == 0) {
3916 if (mddev->safemode == 1)
3917 mddev->safemode = 0;
3918 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3924 spin_unlock(&mddev->lock);
3927 err = mddev_lock(mddev);
3935 /* stopping an active array */
3936 err = do_md_stop(mddev, 0, NULL);
3939 /* stopping an active array */
3941 err = do_md_stop(mddev, 2, NULL);
3943 err = 0; /* already inactive */
3946 break; /* not supported yet */
3949 err = md_set_readonly(mddev, NULL);
3952 set_disk_ro(mddev->gendisk, 1);
3953 err = do_md_run(mddev);
3959 err = md_set_readonly(mddev, NULL);
3960 else if (mddev->ro == 1)
3961 err = restart_array(mddev);
3964 set_disk_ro(mddev->gendisk, 0);
3968 err = do_md_run(mddev);
3973 err = restart_array(mddev);
3976 spin_lock(&mddev->lock);
3977 if (atomic_read(&mddev->writes_pending) == 0) {
3978 if (mddev->in_sync == 0) {
3980 if (mddev->safemode == 1)
3981 mddev->safemode = 0;
3982 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3987 spin_unlock(&mddev->lock);
3993 err = restart_array(mddev);
3996 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3997 wake_up(&mddev->sb_wait);
4001 set_disk_ro(mddev->gendisk, 0);
4002 err = do_md_run(mddev);
4007 /* these cannot be set */
4012 if (mddev->hold_active == UNTIL_IOCTL)
4013 mddev->hold_active = 0;
4014 sysfs_notify_dirent_safe(mddev->sysfs_state);
4016 mddev_unlock(mddev);
4019 static struct md_sysfs_entry md_array_state =
4020 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4023 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4024 return sprintf(page, "%d\n",
4025 atomic_read(&mddev->max_corr_read_errors));
4029 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4034 rv = kstrtouint(buf, 10, &n);
4037 atomic_set(&mddev->max_corr_read_errors, n);
4041 static struct md_sysfs_entry max_corr_read_errors =
4042 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4043 max_corrected_read_errors_store);
4046 null_show(struct mddev *mddev, char *page)
4052 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4054 /* buf must be %d:%d\n? giving major and minor numbers */
4055 /* The new device is added to the array.
4056 * If the array has a persistent superblock, we read the
4057 * superblock to initialise info and check validity.
4058 * Otherwise, only checking done is that in bind_rdev_to_array,
4059 * which mainly checks size.
4062 int major = simple_strtoul(buf, &e, 10);
4065 struct md_rdev *rdev;
4068 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4070 minor = simple_strtoul(e+1, &e, 10);
4071 if (*e && *e != '\n')
4073 dev = MKDEV(major, minor);
4074 if (major != MAJOR(dev) ||
4075 minor != MINOR(dev))
4078 flush_workqueue(md_misc_wq);
4080 err = mddev_lock(mddev);
4083 if (mddev->persistent) {
4084 rdev = md_import_device(dev, mddev->major_version,
4085 mddev->minor_version);
4086 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4087 struct md_rdev *rdev0
4088 = list_entry(mddev->disks.next,
4089 struct md_rdev, same_set);
4090 err = super_types[mddev->major_version]
4091 .load_super(rdev, rdev0, mddev->minor_version);
4095 } else if (mddev->external)
4096 rdev = md_import_device(dev, -2, -1);
4098 rdev = md_import_device(dev, -1, -1);
4101 mddev_unlock(mddev);
4102 return PTR_ERR(rdev);
4104 err = bind_rdev_to_array(rdev, mddev);
4108 mddev_unlock(mddev);
4109 return err ? err : len;
4112 static struct md_sysfs_entry md_new_device =
4113 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4116 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4119 unsigned long chunk, end_chunk;
4122 err = mddev_lock(mddev);
4127 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4129 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4130 if (buf == end) break;
4131 if (*end == '-') { /* range */
4133 end_chunk = simple_strtoul(buf, &end, 0);
4134 if (buf == end) break;
4136 if (*end && !isspace(*end)) break;
4137 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4138 buf = skip_spaces(end);
4140 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4142 mddev_unlock(mddev);
4146 static struct md_sysfs_entry md_bitmap =
4147 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4150 size_show(struct mddev *mddev, char *page)
4152 return sprintf(page, "%llu\n",
4153 (unsigned long long)mddev->dev_sectors / 2);
4156 static int update_size(struct mddev *mddev, sector_t num_sectors);
4159 size_store(struct mddev *mddev, const char *buf, size_t len)
4161 /* If array is inactive, we can reduce the component size, but
4162 * not increase it (except from 0).
4163 * If array is active, we can try an on-line resize
4166 int err = strict_blocks_to_sectors(buf, §ors);
4170 err = mddev_lock(mddev);
4174 err = update_size(mddev, sectors);
4175 md_update_sb(mddev, 1);
4177 if (mddev->dev_sectors == 0 ||
4178 mddev->dev_sectors > sectors)
4179 mddev->dev_sectors = sectors;
4183 mddev_unlock(mddev);
4184 return err ? err : len;
4187 static struct md_sysfs_entry md_size =
4188 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4190 /* Metadata version.
4192 * 'none' for arrays with no metadata (good luck...)
4193 * 'external' for arrays with externally managed metadata,
4194 * or N.M for internally known formats
4197 metadata_show(struct mddev *mddev, char *page)
4199 if (mddev->persistent)
4200 return sprintf(page, "%d.%d\n",
4201 mddev->major_version, mddev->minor_version);
4202 else if (mddev->external)
4203 return sprintf(page, "external:%s\n", mddev->metadata_type);
4205 return sprintf(page, "none\n");
4209 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4214 /* Changing the details of 'external' metadata is
4215 * always permitted. Otherwise there must be
4216 * no devices attached to the array.
4219 err = mddev_lock(mddev);
4223 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4225 else if (!list_empty(&mddev->disks))
4229 if (cmd_match(buf, "none")) {
4230 mddev->persistent = 0;
4231 mddev->external = 0;
4232 mddev->major_version = 0;
4233 mddev->minor_version = 90;
4236 if (strncmp(buf, "external:", 9) == 0) {
4237 size_t namelen = len-9;
4238 if (namelen >= sizeof(mddev->metadata_type))
4239 namelen = sizeof(mddev->metadata_type)-1;
4240 strncpy(mddev->metadata_type, buf+9, namelen);
4241 mddev->metadata_type[namelen] = 0;
4242 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4243 mddev->metadata_type[--namelen] = 0;
4244 mddev->persistent = 0;
4245 mddev->external = 1;
4246 mddev->major_version = 0;
4247 mddev->minor_version = 90;
4250 major = simple_strtoul(buf, &e, 10);
4252 if (e==buf || *e != '.')
4255 minor = simple_strtoul(buf, &e, 10);
4256 if (e==buf || (*e && *e != '\n') )
4259 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4261 mddev->major_version = major;
4262 mddev->minor_version = minor;
4263 mddev->persistent = 1;
4264 mddev->external = 0;
4267 mddev_unlock(mddev);
4271 static struct md_sysfs_entry md_metadata =
4272 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4275 action_show(struct mddev *mddev, char *page)
4277 char *type = "idle";
4278 unsigned long recovery = mddev->recovery;
4279 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4281 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4282 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4283 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4285 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4286 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4288 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4292 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4294 else if (mddev->reshape_position != MaxSector)
4297 return sprintf(page, "%s\n", type);
4301 action_store(struct mddev *mddev, const char *page, size_t len)
4303 if (!mddev->pers || !mddev->pers->sync_request)
4307 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4308 if (cmd_match(page, "frozen"))
4309 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4311 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4312 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4313 mddev_lock(mddev) == 0) {
4314 flush_workqueue(md_misc_wq);
4315 if (mddev->sync_thread) {
4316 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4317 md_reap_sync_thread(mddev);
4319 mddev_unlock(mddev);
4321 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4322 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4324 else if (cmd_match(page, "resync"))
4325 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4326 else if (cmd_match(page, "recover")) {
4327 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4328 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4329 } else if (cmd_match(page, "reshape")) {
4331 if (mddev->pers->start_reshape == NULL)
4333 err = mddev_lock(mddev);
4335 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4336 err = mddev->pers->start_reshape(mddev);
4337 mddev_unlock(mddev);
4341 sysfs_notify(&mddev->kobj, NULL, "degraded");
4343 if (cmd_match(page, "check"))
4344 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4345 else if (!cmd_match(page, "repair"))
4347 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4348 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4349 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4351 if (mddev->ro == 2) {
4352 /* A write to sync_action is enough to justify
4353 * canceling read-auto mode
4356 md_wakeup_thread(mddev->sync_thread);
4358 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4359 md_wakeup_thread(mddev->thread);
4360 sysfs_notify_dirent_safe(mddev->sysfs_action);
4364 static struct md_sysfs_entry md_scan_mode =
4365 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4368 last_sync_action_show(struct mddev *mddev, char *page)
4370 return sprintf(page, "%s\n", mddev->last_sync_action);
4373 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4376 mismatch_cnt_show(struct mddev *mddev, char *page)
4378 return sprintf(page, "%llu\n",
4379 (unsigned long long)
4380 atomic64_read(&mddev->resync_mismatches));
4383 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4386 sync_min_show(struct mddev *mddev, char *page)
4388 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4389 mddev->sync_speed_min ? "local": "system");
4393 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4398 if (strncmp(buf, "system", 6)==0) {
4401 rv = kstrtouint(buf, 10, &min);
4407 mddev->sync_speed_min = min;
4411 static struct md_sysfs_entry md_sync_min =
4412 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4415 sync_max_show(struct mddev *mddev, char *page)
4417 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4418 mddev->sync_speed_max ? "local": "system");
4422 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4427 if (strncmp(buf, "system", 6)==0) {
4430 rv = kstrtouint(buf, 10, &max);
4436 mddev->sync_speed_max = max;
4440 static struct md_sysfs_entry md_sync_max =
4441 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4444 degraded_show(struct mddev *mddev, char *page)
4446 return sprintf(page, "%d\n", mddev->degraded);
4448 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4451 sync_force_parallel_show(struct mddev *mddev, char *page)
4453 return sprintf(page, "%d\n", mddev->parallel_resync);
4457 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4461 if (kstrtol(buf, 10, &n))
4464 if (n != 0 && n != 1)
4467 mddev->parallel_resync = n;
4469 if (mddev->sync_thread)
4470 wake_up(&resync_wait);
4475 /* force parallel resync, even with shared block devices */
4476 static struct md_sysfs_entry md_sync_force_parallel =
4477 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4478 sync_force_parallel_show, sync_force_parallel_store);
4481 sync_speed_show(struct mddev *mddev, char *page)
4483 unsigned long resync, dt, db;
4484 if (mddev->curr_resync == 0)
4485 return sprintf(page, "none\n");
4486 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4487 dt = (jiffies - mddev->resync_mark) / HZ;
4489 db = resync - mddev->resync_mark_cnt;
4490 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4493 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4496 sync_completed_show(struct mddev *mddev, char *page)
4498 unsigned long long max_sectors, resync;
4500 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4501 return sprintf(page, "none\n");
4503 if (mddev->curr_resync == 1 ||
4504 mddev->curr_resync == 2)
4505 return sprintf(page, "delayed\n");
4507 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4508 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4509 max_sectors = mddev->resync_max_sectors;
4511 max_sectors = mddev->dev_sectors;
4513 resync = mddev->curr_resync_completed;
4514 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4517 static struct md_sysfs_entry md_sync_completed =
4518 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4521 min_sync_show(struct mddev *mddev, char *page)
4523 return sprintf(page, "%llu\n",
4524 (unsigned long long)mddev->resync_min);
4527 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4529 unsigned long long min;
4532 if (kstrtoull(buf, 10, &min))
4535 spin_lock(&mddev->lock);
4537 if (min > mddev->resync_max)
4541 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4544 /* Round down to multiple of 4K for safety */
4545 mddev->resync_min = round_down(min, 8);
4549 spin_unlock(&mddev->lock);
4553 static struct md_sysfs_entry md_min_sync =
4554 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4557 max_sync_show(struct mddev *mddev, char *page)
4559 if (mddev->resync_max == MaxSector)
4560 return sprintf(page, "max\n");
4562 return sprintf(page, "%llu\n",
4563 (unsigned long long)mddev->resync_max);
4566 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4569 spin_lock(&mddev->lock);
4570 if (strncmp(buf, "max", 3) == 0)
4571 mddev->resync_max = MaxSector;
4573 unsigned long long max;
4577 if (kstrtoull(buf, 10, &max))
4579 if (max < mddev->resync_min)
4583 if (max < mddev->resync_max &&
4585 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4588 /* Must be a multiple of chunk_size */
4589 chunk = mddev->chunk_sectors;
4591 sector_t temp = max;
4594 if (sector_div(temp, chunk))
4597 mddev->resync_max = max;
4599 wake_up(&mddev->recovery_wait);
4602 spin_unlock(&mddev->lock);
4606 static struct md_sysfs_entry md_max_sync =
4607 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4610 suspend_lo_show(struct mddev *mddev, char *page)
4612 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4616 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4618 unsigned long long old, new;
4621 err = kstrtoull(buf, 10, &new);
4624 if (new != (sector_t)new)
4627 err = mddev_lock(mddev);
4631 if (mddev->pers == NULL ||
4632 mddev->pers->quiesce == NULL)
4634 old = mddev->suspend_lo;
4635 mddev->suspend_lo = new;
4637 /* Shrinking suspended region */
4638 mddev->pers->quiesce(mddev, 2);
4640 /* Expanding suspended region - need to wait */
4641 mddev->pers->quiesce(mddev, 1);
4642 mddev->pers->quiesce(mddev, 0);
4646 mddev_unlock(mddev);
4649 static struct md_sysfs_entry md_suspend_lo =
4650 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4653 suspend_hi_show(struct mddev *mddev, char *page)
4655 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4659 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4661 unsigned long long old, new;
4664 err = kstrtoull(buf, 10, &new);
4667 if (new != (sector_t)new)
4670 err = mddev_lock(mddev);
4674 if (mddev->pers == NULL ||
4675 mddev->pers->quiesce == NULL)
4677 old = mddev->suspend_hi;
4678 mddev->suspend_hi = new;
4680 /* Shrinking suspended region */
4681 mddev->pers->quiesce(mddev, 2);
4683 /* Expanding suspended region - need to wait */
4684 mddev->pers->quiesce(mddev, 1);
4685 mddev->pers->quiesce(mddev, 0);
4689 mddev_unlock(mddev);
4692 static struct md_sysfs_entry md_suspend_hi =
4693 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4696 reshape_position_show(struct mddev *mddev, char *page)
4698 if (mddev->reshape_position != MaxSector)
4699 return sprintf(page, "%llu\n",
4700 (unsigned long long)mddev->reshape_position);
4701 strcpy(page, "none\n");
4706 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4708 struct md_rdev *rdev;
4709 unsigned long long new;
4712 err = kstrtoull(buf, 10, &new);
4715 if (new != (sector_t)new)
4717 err = mddev_lock(mddev);
4723 mddev->reshape_position = new;
4724 mddev->delta_disks = 0;
4725 mddev->reshape_backwards = 0;
4726 mddev->new_level = mddev->level;
4727 mddev->new_layout = mddev->layout;
4728 mddev->new_chunk_sectors = mddev->chunk_sectors;
4729 rdev_for_each(rdev, mddev)
4730 rdev->new_data_offset = rdev->data_offset;
4733 mddev_unlock(mddev);
4737 static struct md_sysfs_entry md_reshape_position =
4738 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4739 reshape_position_store);
4742 reshape_direction_show(struct mddev *mddev, char *page)
4744 return sprintf(page, "%s\n",
4745 mddev->reshape_backwards ? "backwards" : "forwards");
4749 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4754 if (cmd_match(buf, "forwards"))
4756 else if (cmd_match(buf, "backwards"))
4760 if (mddev->reshape_backwards == backwards)
4763 err = mddev_lock(mddev);
4766 /* check if we are allowed to change */
4767 if (mddev->delta_disks)
4769 else if (mddev->persistent &&
4770 mddev->major_version == 0)
4773 mddev->reshape_backwards = backwards;
4774 mddev_unlock(mddev);
4778 static struct md_sysfs_entry md_reshape_direction =
4779 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4780 reshape_direction_store);
4783 array_size_show(struct mddev *mddev, char *page)
4785 if (mddev->external_size)
4786 return sprintf(page, "%llu\n",
4787 (unsigned long long)mddev->array_sectors/2);
4789 return sprintf(page, "default\n");
4793 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4798 err = mddev_lock(mddev);
4802 if (strncmp(buf, "default", 7) == 0) {
4804 sectors = mddev->pers->size(mddev, 0, 0);
4806 sectors = mddev->array_sectors;
4808 mddev->external_size = 0;
4810 if (strict_blocks_to_sectors(buf, §ors) < 0)
4812 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4815 mddev->external_size = 1;
4819 mddev->array_sectors = sectors;
4821 set_capacity(mddev->gendisk, mddev->array_sectors);
4822 revalidate_disk(mddev->gendisk);
4825 mddev_unlock(mddev);
4829 static struct md_sysfs_entry md_array_size =
4830 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4833 static struct attribute *md_default_attrs[] = {
4836 &md_raid_disks.attr,
4837 &md_chunk_size.attr,
4839 &md_resync_start.attr,
4841 &md_new_device.attr,
4842 &md_safe_delay.attr,
4843 &md_array_state.attr,
4844 &md_reshape_position.attr,
4845 &md_reshape_direction.attr,
4846 &md_array_size.attr,
4847 &max_corr_read_errors.attr,
4851 static struct attribute *md_redundancy_attrs[] = {
4853 &md_last_scan_mode.attr,
4854 &md_mismatches.attr,
4857 &md_sync_speed.attr,
4858 &md_sync_force_parallel.attr,
4859 &md_sync_completed.attr,
4862 &md_suspend_lo.attr,
4863 &md_suspend_hi.attr,
4868 static struct attribute_group md_redundancy_group = {
4870 .attrs = md_redundancy_attrs,
4874 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4876 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4877 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4882 spin_lock(&all_mddevs_lock);
4883 if (list_empty(&mddev->all_mddevs)) {
4884 spin_unlock(&all_mddevs_lock);
4888 spin_unlock(&all_mddevs_lock);
4890 rv = entry->show(mddev, page);
4896 md_attr_store(struct kobject *kobj, struct attribute *attr,
4897 const char *page, size_t length)
4899 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4900 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4905 if (!capable(CAP_SYS_ADMIN))
4907 spin_lock(&all_mddevs_lock);
4908 if (list_empty(&mddev->all_mddevs)) {
4909 spin_unlock(&all_mddevs_lock);
4913 spin_unlock(&all_mddevs_lock);
4914 rv = entry->store(mddev, page, length);
4919 static void md_free(struct kobject *ko)
4921 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4923 if (mddev->sysfs_state)
4924 sysfs_put(mddev->sysfs_state);
4927 blk_cleanup_queue(mddev->queue);
4928 if (mddev->gendisk) {
4929 del_gendisk(mddev->gendisk);
4930 put_disk(mddev->gendisk);
4936 static const struct sysfs_ops md_sysfs_ops = {
4937 .show = md_attr_show,
4938 .store = md_attr_store,
4940 static struct kobj_type md_ktype = {
4942 .sysfs_ops = &md_sysfs_ops,
4943 .default_attrs = md_default_attrs,
4948 static void mddev_delayed_delete(struct work_struct *ws)
4950 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4952 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4953 kobject_del(&mddev->kobj);
4954 kobject_put(&mddev->kobj);
4957 static int md_alloc(dev_t dev, char *name)
4959 static DEFINE_MUTEX(disks_mutex);
4960 struct mddev *mddev = mddev_find(dev);
4961 struct gendisk *disk;
4970 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4971 shift = partitioned ? MdpMinorShift : 0;
4972 unit = MINOR(mddev->unit) >> shift;
4974 /* wait for any previous instance of this device to be
4975 * completely removed (mddev_delayed_delete).
4977 flush_workqueue(md_misc_wq);
4979 mutex_lock(&disks_mutex);
4985 /* Need to ensure that 'name' is not a duplicate.
4987 struct mddev *mddev2;
4988 spin_lock(&all_mddevs_lock);
4990 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4991 if (mddev2->gendisk &&
4992 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4993 spin_unlock(&all_mddevs_lock);
4996 spin_unlock(&all_mddevs_lock);
5000 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5003 mddev->queue->queuedata = mddev;
5005 blk_queue_make_request(mddev->queue, md_make_request);
5006 blk_set_stacking_limits(&mddev->queue->limits);
5008 disk = alloc_disk(1 << shift);
5010 blk_cleanup_queue(mddev->queue);
5011 mddev->queue = NULL;
5014 disk->major = MAJOR(mddev->unit);
5015 disk->first_minor = unit << shift;
5017 strcpy(disk->disk_name, name);
5018 else if (partitioned)
5019 sprintf(disk->disk_name, "md_d%d", unit);
5021 sprintf(disk->disk_name, "md%d", unit);
5022 disk->fops = &md_fops;
5023 disk->private_data = mddev;
5024 disk->queue = mddev->queue;
5025 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5026 /* Allow extended partitions. This makes the
5027 * 'mdp' device redundant, but we can't really
5030 disk->flags |= GENHD_FL_EXT_DEVT;
5031 mddev->gendisk = disk;
5032 /* As soon as we call add_disk(), another thread could get
5033 * through to md_open, so make sure it doesn't get too far
5035 mutex_lock(&mddev->open_mutex);
5038 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5039 &disk_to_dev(disk)->kobj, "%s", "md");
5041 /* This isn't possible, but as kobject_init_and_add is marked
5042 * __must_check, we must do something with the result
5044 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5048 if (mddev->kobj.sd &&
5049 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5050 printk(KERN_DEBUG "pointless warning\n");
5051 mutex_unlock(&mddev->open_mutex);
5053 mutex_unlock(&disks_mutex);
5054 if (!error && mddev->kobj.sd) {
5055 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5056 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5062 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5064 md_alloc(dev, NULL);
5068 static int add_named_array(const char *val, struct kernel_param *kp)
5070 /* val must be "md_*" where * is not all digits.
5071 * We allocate an array with a large free minor number, and
5072 * set the name to val. val must not already be an active name.
5074 int len = strlen(val);
5075 char buf[DISK_NAME_LEN];
5077 while (len && val[len-1] == '\n')
5079 if (len >= DISK_NAME_LEN)
5081 strlcpy(buf, val, len+1);
5082 if (strncmp(buf, "md_", 3) != 0)
5084 return md_alloc(0, buf);
5087 static void md_safemode_timeout(unsigned long data)
5089 struct mddev *mddev = (struct mddev *) data;
5091 if (!atomic_read(&mddev->writes_pending)) {
5092 mddev->safemode = 1;
5093 if (mddev->external)
5094 sysfs_notify_dirent_safe(mddev->sysfs_state);
5096 md_wakeup_thread(mddev->thread);
5099 static int start_dirty_degraded;
5101 int md_run(struct mddev *mddev)
5104 struct md_rdev *rdev;
5105 struct md_personality *pers;
5107 if (list_empty(&mddev->disks))
5108 /* cannot run an array with no devices.. */
5113 /* Cannot run until previous stop completes properly */
5114 if (mddev->sysfs_active)
5118 * Analyze all RAID superblock(s)
5120 if (!mddev->raid_disks) {
5121 if (!mddev->persistent)
5126 if (mddev->level != LEVEL_NONE)
5127 request_module("md-level-%d", mddev->level);
5128 else if (mddev->clevel[0])
5129 request_module("md-%s", mddev->clevel);
5132 * Drop all container device buffers, from now on
5133 * the only valid external interface is through the md
5136 rdev_for_each(rdev, mddev) {
5137 if (test_bit(Faulty, &rdev->flags))
5139 sync_blockdev(rdev->bdev);
5140 invalidate_bdev(rdev->bdev);
5142 /* perform some consistency tests on the device.
5143 * We don't want the data to overlap the metadata,
5144 * Internal Bitmap issues have been handled elsewhere.
5146 if (rdev->meta_bdev) {
5147 /* Nothing to check */;
5148 } else if (rdev->data_offset < rdev->sb_start) {
5149 if (mddev->dev_sectors &&
5150 rdev->data_offset + mddev->dev_sectors
5152 printk("md: %s: data overlaps metadata\n",
5157 if (rdev->sb_start + rdev->sb_size/512
5158 > rdev->data_offset) {
5159 printk("md: %s: metadata overlaps data\n",
5164 sysfs_notify_dirent_safe(rdev->sysfs_state);
5167 if (mddev->bio_set == NULL)
5168 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5170 spin_lock(&pers_lock);
5171 pers = find_pers(mddev->level, mddev->clevel);
5172 if (!pers || !try_module_get(pers->owner)) {
5173 spin_unlock(&pers_lock);
5174 if (mddev->level != LEVEL_NONE)
5175 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5178 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5182 spin_unlock(&pers_lock);
5183 if (mddev->level != pers->level) {
5184 mddev->level = pers->level;
5185 mddev->new_level = pers->level;
5187 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5189 if (mddev->reshape_position != MaxSector &&
5190 pers->start_reshape == NULL) {
5191 /* This personality cannot handle reshaping... */
5192 module_put(pers->owner);
5196 if (pers->sync_request) {
5197 /* Warn if this is a potentially silly
5200 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5201 struct md_rdev *rdev2;
5204 rdev_for_each(rdev, mddev)
5205 rdev_for_each(rdev2, mddev) {
5207 rdev->bdev->bd_contains ==
5208 rdev2->bdev->bd_contains) {
5210 "%s: WARNING: %s appears to be"
5211 " on the same physical disk as"
5214 bdevname(rdev->bdev,b),
5215 bdevname(rdev2->bdev,b2));
5222 "True protection against single-disk"
5223 " failure might be compromised.\n");
5226 mddev->recovery = 0;
5227 /* may be over-ridden by personality */
5228 mddev->resync_max_sectors = mddev->dev_sectors;
5230 mddev->ok_start_degraded = start_dirty_degraded;
5232 if (start_readonly && mddev->ro == 0)
5233 mddev->ro = 2; /* read-only, but switch on first write */
5235 err = pers->run(mddev);
5237 printk(KERN_ERR "md: pers->run() failed ...\n");
5238 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5239 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5240 " but 'external_size' not in effect?\n", __func__);
5242 "md: invalid array_size %llu > default size %llu\n",
5243 (unsigned long long)mddev->array_sectors / 2,
5244 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5247 if (err == 0 && pers->sync_request &&
5248 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5249 struct bitmap *bitmap;
5251 bitmap = bitmap_create(mddev, -1);
5252 if (IS_ERR(bitmap)) {
5253 err = PTR_ERR(bitmap);
5254 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5255 mdname(mddev), err);
5257 mddev->bitmap = bitmap;
5261 mddev_detach(mddev);
5263 pers->free(mddev, mddev->private);
5264 mddev->private = NULL;
5265 module_put(pers->owner);
5266 bitmap_destroy(mddev);
5270 mddev->queue->backing_dev_info.congested_data = mddev;
5271 mddev->queue->backing_dev_info.congested_fn = md_congested;
5273 if (pers->sync_request) {
5274 if (mddev->kobj.sd &&
5275 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5277 "md: cannot register extra attributes for %s\n",
5279 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5280 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5283 atomic_set(&mddev->writes_pending,0);
5284 atomic_set(&mddev->max_corr_read_errors,
5285 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5286 mddev->safemode = 0;
5287 if (mddev_is_clustered(mddev))
5288 mddev->safemode_delay = 0;
5290 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5293 spin_lock(&mddev->lock);
5296 spin_unlock(&mddev->lock);
5297 rdev_for_each(rdev, mddev)
5298 if (rdev->raid_disk >= 0)
5299 if (sysfs_link_rdev(mddev, rdev))
5300 /* failure here is OK */;
5302 if (mddev->degraded && !mddev->ro)
5303 /* This ensures that recovering status is reported immediately
5304 * via sysfs - until a lack of spares is confirmed.
5306 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5307 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5309 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5310 md_update_sb(mddev, 0);
5312 md_new_event(mddev);
5313 sysfs_notify_dirent_safe(mddev->sysfs_state);
5314 sysfs_notify_dirent_safe(mddev->sysfs_action);
5315 sysfs_notify(&mddev->kobj, NULL, "degraded");
5318 EXPORT_SYMBOL_GPL(md_run);
5320 static int do_md_run(struct mddev *mddev)
5324 err = md_run(mddev);
5327 err = bitmap_load(mddev);
5329 bitmap_destroy(mddev);
5333 if (mddev_is_clustered(mddev))
5334 md_allow_write(mddev);
5336 md_wakeup_thread(mddev->thread);
5337 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5339 set_capacity(mddev->gendisk, mddev->array_sectors);
5340 revalidate_disk(mddev->gendisk);
5342 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5347 static int restart_array(struct mddev *mddev)
5349 struct gendisk *disk = mddev->gendisk;
5351 /* Complain if it has no devices */
5352 if (list_empty(&mddev->disks))
5358 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5359 struct md_rdev *rdev;
5360 bool has_journal = false;
5363 rdev_for_each_rcu(rdev, mddev) {
5364 if (test_bit(Journal, &rdev->flags) &&
5365 !test_bit(Faulty, &rdev->flags)) {
5372 /* Don't restart rw with journal missing/faulty */
5377 mddev->safemode = 0;
5379 set_disk_ro(disk, 0);
5380 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5382 /* Kick recovery or resync if necessary */
5383 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5384 md_wakeup_thread(mddev->thread);
5385 md_wakeup_thread(mddev->sync_thread);
5386 sysfs_notify_dirent_safe(mddev->sysfs_state);
5390 static void md_clean(struct mddev *mddev)
5392 mddev->array_sectors = 0;
5393 mddev->external_size = 0;
5394 mddev->dev_sectors = 0;
5395 mddev->raid_disks = 0;
5396 mddev->recovery_cp = 0;
5397 mddev->resync_min = 0;
5398 mddev->resync_max = MaxSector;
5399 mddev->reshape_position = MaxSector;
5400 mddev->external = 0;
5401 mddev->persistent = 0;
5402 mddev->level = LEVEL_NONE;
5403 mddev->clevel[0] = 0;
5406 mddev->metadata_type[0] = 0;
5407 mddev->chunk_sectors = 0;
5408 mddev->ctime = mddev->utime = 0;
5410 mddev->max_disks = 0;
5412 mddev->can_decrease_events = 0;
5413 mddev->delta_disks = 0;
5414 mddev->reshape_backwards = 0;
5415 mddev->new_level = LEVEL_NONE;
5416 mddev->new_layout = 0;
5417 mddev->new_chunk_sectors = 0;
5418 mddev->curr_resync = 0;
5419 atomic64_set(&mddev->resync_mismatches, 0);
5420 mddev->suspend_lo = mddev->suspend_hi = 0;
5421 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5422 mddev->recovery = 0;
5425 mddev->degraded = 0;
5426 mddev->safemode = 0;
5427 mddev->private = NULL;
5428 mddev->bitmap_info.offset = 0;
5429 mddev->bitmap_info.default_offset = 0;
5430 mddev->bitmap_info.default_space = 0;
5431 mddev->bitmap_info.chunksize = 0;
5432 mddev->bitmap_info.daemon_sleep = 0;
5433 mddev->bitmap_info.max_write_behind = 0;
5436 static void __md_stop_writes(struct mddev *mddev)
5438 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5439 flush_workqueue(md_misc_wq);
5440 if (mddev->sync_thread) {
5441 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5442 md_reap_sync_thread(mddev);
5445 del_timer_sync(&mddev->safemode_timer);
5447 bitmap_flush(mddev);
5448 md_super_wait(mddev);
5450 if (mddev->ro == 0 &&
5451 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5452 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5453 /* mark array as shutdown cleanly */
5454 if (!mddev_is_clustered(mddev))
5456 md_update_sb(mddev, 1);
5460 void md_stop_writes(struct mddev *mddev)
5462 mddev_lock_nointr(mddev);
5463 __md_stop_writes(mddev);
5464 mddev_unlock(mddev);
5466 EXPORT_SYMBOL_GPL(md_stop_writes);
5468 static void mddev_detach(struct mddev *mddev)
5470 struct bitmap *bitmap = mddev->bitmap;
5471 /* wait for behind writes to complete */
5472 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5473 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5475 /* need to kick something here to make sure I/O goes? */
5476 wait_event(bitmap->behind_wait,
5477 atomic_read(&bitmap->behind_writes) == 0);
5479 if (mddev->pers && mddev->pers->quiesce) {
5480 mddev->pers->quiesce(mddev, 1);
5481 mddev->pers->quiesce(mddev, 0);
5483 md_unregister_thread(&mddev->thread);
5485 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5488 static void __md_stop(struct mddev *mddev)
5490 struct md_personality *pers = mddev->pers;
5491 mddev_detach(mddev);
5492 /* Ensure ->event_work is done */
5493 flush_workqueue(md_misc_wq);
5494 spin_lock(&mddev->lock);
5497 spin_unlock(&mddev->lock);
5498 pers->free(mddev, mddev->private);
5499 mddev->private = NULL;
5500 if (pers->sync_request && mddev->to_remove == NULL)
5501 mddev->to_remove = &md_redundancy_group;
5502 module_put(pers->owner);
5503 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5506 void md_stop(struct mddev *mddev)
5508 /* stop the array and free an attached data structures.
5509 * This is called from dm-raid
5512 bitmap_destroy(mddev);
5514 bioset_free(mddev->bio_set);
5517 EXPORT_SYMBOL_GPL(md_stop);
5519 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5524 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5526 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5527 md_wakeup_thread(mddev->thread);
5529 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5530 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5531 if (mddev->sync_thread)
5532 /* Thread might be blocked waiting for metadata update
5533 * which will now never happen */
5534 wake_up_process(mddev->sync_thread->tsk);
5536 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5538 mddev_unlock(mddev);
5539 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5541 wait_event(mddev->sb_wait,
5542 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5543 mddev_lock_nointr(mddev);
5545 mutex_lock(&mddev->open_mutex);
5546 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5547 mddev->sync_thread ||
5548 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5549 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5550 printk("md: %s still in use.\n",mdname(mddev));
5552 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5553 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5554 md_wakeup_thread(mddev->thread);
5560 __md_stop_writes(mddev);
5566 set_disk_ro(mddev->gendisk, 1);
5567 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5568 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5569 md_wakeup_thread(mddev->thread);
5570 sysfs_notify_dirent_safe(mddev->sysfs_state);
5574 mutex_unlock(&mddev->open_mutex);
5579 * 0 - completely stop and dis-assemble array
5580 * 2 - stop but do not disassemble array
5582 static int do_md_stop(struct mddev *mddev, int mode,
5583 struct block_device *bdev)
5585 struct gendisk *disk = mddev->gendisk;
5586 struct md_rdev *rdev;
5589 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5591 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5592 md_wakeup_thread(mddev->thread);
5594 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5595 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5596 if (mddev->sync_thread)
5597 /* Thread might be blocked waiting for metadata update
5598 * which will now never happen */
5599 wake_up_process(mddev->sync_thread->tsk);
5601 mddev_unlock(mddev);
5602 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5603 !test_bit(MD_RECOVERY_RUNNING,
5604 &mddev->recovery)));
5605 mddev_lock_nointr(mddev);
5607 mutex_lock(&mddev->open_mutex);
5608 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5609 mddev->sysfs_active ||
5610 mddev->sync_thread ||
5611 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5612 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5613 printk("md: %s still in use.\n",mdname(mddev));
5614 mutex_unlock(&mddev->open_mutex);
5616 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5617 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5618 md_wakeup_thread(mddev->thread);
5624 set_disk_ro(disk, 0);
5626 __md_stop_writes(mddev);
5628 mddev->queue->backing_dev_info.congested_fn = NULL;
5630 /* tell userspace to handle 'inactive' */
5631 sysfs_notify_dirent_safe(mddev->sysfs_state);
5633 rdev_for_each(rdev, mddev)
5634 if (rdev->raid_disk >= 0)
5635 sysfs_unlink_rdev(mddev, rdev);
5637 set_capacity(disk, 0);
5638 mutex_unlock(&mddev->open_mutex);
5640 revalidate_disk(disk);
5645 mutex_unlock(&mddev->open_mutex);
5647 * Free resources if final stop
5650 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5652 bitmap_destroy(mddev);
5653 if (mddev->bitmap_info.file) {
5654 struct file *f = mddev->bitmap_info.file;
5655 spin_lock(&mddev->lock);
5656 mddev->bitmap_info.file = NULL;
5657 spin_unlock(&mddev->lock);
5660 mddev->bitmap_info.offset = 0;
5662 export_array(mddev);
5665 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5666 if (mddev->hold_active == UNTIL_STOP)
5667 mddev->hold_active = 0;
5669 blk_integrity_unregister(disk);
5670 md_new_event(mddev);
5671 sysfs_notify_dirent_safe(mddev->sysfs_state);
5676 static void autorun_array(struct mddev *mddev)
5678 struct md_rdev *rdev;
5681 if (list_empty(&mddev->disks))
5684 printk(KERN_INFO "md: running: ");
5686 rdev_for_each(rdev, mddev) {
5687 char b[BDEVNAME_SIZE];
5688 printk("<%s>", bdevname(rdev->bdev,b));
5692 err = do_md_run(mddev);
5694 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5695 do_md_stop(mddev, 0, NULL);
5700 * lets try to run arrays based on all disks that have arrived
5701 * until now. (those are in pending_raid_disks)
5703 * the method: pick the first pending disk, collect all disks with
5704 * the same UUID, remove all from the pending list and put them into
5705 * the 'same_array' list. Then order this list based on superblock
5706 * update time (freshest comes first), kick out 'old' disks and
5707 * compare superblocks. If everything's fine then run it.
5709 * If "unit" is allocated, then bump its reference count
5711 static void autorun_devices(int part)
5713 struct md_rdev *rdev0, *rdev, *tmp;
5714 struct mddev *mddev;
5715 char b[BDEVNAME_SIZE];
5717 printk(KERN_INFO "md: autorun ...\n");
5718 while (!list_empty(&pending_raid_disks)) {
5721 LIST_HEAD(candidates);
5722 rdev0 = list_entry(pending_raid_disks.next,
5723 struct md_rdev, same_set);
5725 printk(KERN_INFO "md: considering %s ...\n",
5726 bdevname(rdev0->bdev,b));
5727 INIT_LIST_HEAD(&candidates);
5728 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5729 if (super_90_load(rdev, rdev0, 0) >= 0) {
5730 printk(KERN_INFO "md: adding %s ...\n",
5731 bdevname(rdev->bdev,b));
5732 list_move(&rdev->same_set, &candidates);
5735 * now we have a set of devices, with all of them having
5736 * mostly sane superblocks. It's time to allocate the
5740 dev = MKDEV(mdp_major,
5741 rdev0->preferred_minor << MdpMinorShift);
5742 unit = MINOR(dev) >> MdpMinorShift;
5744 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5747 if (rdev0->preferred_minor != unit) {
5748 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5749 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5753 md_probe(dev, NULL, NULL);
5754 mddev = mddev_find(dev);
5755 if (!mddev || !mddev->gendisk) {
5759 "md: cannot allocate memory for md drive.\n");
5762 if (mddev_lock(mddev))
5763 printk(KERN_WARNING "md: %s locked, cannot run\n",
5765 else if (mddev->raid_disks || mddev->major_version
5766 || !list_empty(&mddev->disks)) {
5768 "md: %s already running, cannot run %s\n",
5769 mdname(mddev), bdevname(rdev0->bdev,b));
5770 mddev_unlock(mddev);
5772 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5773 mddev->persistent = 1;
5774 rdev_for_each_list(rdev, tmp, &candidates) {
5775 list_del_init(&rdev->same_set);
5776 if (bind_rdev_to_array(rdev, mddev))
5779 autorun_array(mddev);
5780 mddev_unlock(mddev);
5782 /* on success, candidates will be empty, on error
5785 rdev_for_each_list(rdev, tmp, &candidates) {
5786 list_del_init(&rdev->same_set);
5791 printk(KERN_INFO "md: ... autorun DONE.\n");
5793 #endif /* !MODULE */
5795 static int get_version(void __user *arg)
5799 ver.major = MD_MAJOR_VERSION;
5800 ver.minor = MD_MINOR_VERSION;
5801 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5803 if (copy_to_user(arg, &ver, sizeof(ver)))
5809 static int get_array_info(struct mddev *mddev, void __user *arg)
5811 mdu_array_info_t info;
5812 int nr,working,insync,failed,spare;
5813 struct md_rdev *rdev;
5815 nr = working = insync = failed = spare = 0;
5817 rdev_for_each_rcu(rdev, mddev) {
5819 if (test_bit(Faulty, &rdev->flags))
5823 if (test_bit(In_sync, &rdev->flags))
5831 info.major_version = mddev->major_version;
5832 info.minor_version = mddev->minor_version;
5833 info.patch_version = MD_PATCHLEVEL_VERSION;
5834 info.ctime = mddev->ctime;
5835 info.level = mddev->level;
5836 info.size = mddev->dev_sectors / 2;
5837 if (info.size != mddev->dev_sectors / 2) /* overflow */
5840 info.raid_disks = mddev->raid_disks;
5841 info.md_minor = mddev->md_minor;
5842 info.not_persistent= !mddev->persistent;
5844 info.utime = mddev->utime;
5847 info.state = (1<<MD_SB_CLEAN);
5848 if (mddev->bitmap && mddev->bitmap_info.offset)
5849 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5850 if (mddev_is_clustered(mddev))
5851 info.state |= (1<<MD_SB_CLUSTERED);
5852 info.active_disks = insync;
5853 info.working_disks = working;
5854 info.failed_disks = failed;
5855 info.spare_disks = spare;
5857 info.layout = mddev->layout;
5858 info.chunk_size = mddev->chunk_sectors << 9;
5860 if (copy_to_user(arg, &info, sizeof(info)))
5866 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5868 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5872 file = kzalloc(sizeof(*file), GFP_NOIO);
5877 spin_lock(&mddev->lock);
5878 /* bitmap enabled */
5879 if (mddev->bitmap_info.file) {
5880 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5881 sizeof(file->pathname));
5885 memmove(file->pathname, ptr,
5886 sizeof(file->pathname)-(ptr-file->pathname));
5888 spin_unlock(&mddev->lock);
5891 copy_to_user(arg, file, sizeof(*file)))
5898 static int get_disk_info(struct mddev *mddev, void __user * arg)
5900 mdu_disk_info_t info;
5901 struct md_rdev *rdev;
5903 if (copy_from_user(&info, arg, sizeof(info)))
5907 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5909 info.major = MAJOR(rdev->bdev->bd_dev);
5910 info.minor = MINOR(rdev->bdev->bd_dev);
5911 info.raid_disk = rdev->raid_disk;
5913 if (test_bit(Faulty, &rdev->flags))
5914 info.state |= (1<<MD_DISK_FAULTY);
5915 else if (test_bit(In_sync, &rdev->flags)) {
5916 info.state |= (1<<MD_DISK_ACTIVE);
5917 info.state |= (1<<MD_DISK_SYNC);
5919 if (test_bit(Journal, &rdev->flags))
5920 info.state |= (1<<MD_DISK_JOURNAL);
5921 if (test_bit(WriteMostly, &rdev->flags))
5922 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5924 info.major = info.minor = 0;
5925 info.raid_disk = -1;
5926 info.state = (1<<MD_DISK_REMOVED);
5930 if (copy_to_user(arg, &info, sizeof(info)))
5936 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5938 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5939 struct md_rdev *rdev;
5940 dev_t dev = MKDEV(info->major,info->minor);
5942 if (mddev_is_clustered(mddev) &&
5943 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5944 pr_err("%s: Cannot add to clustered mddev.\n",
5949 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5952 if (!mddev->raid_disks) {
5954 /* expecting a device which has a superblock */
5955 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5958 "md: md_import_device returned %ld\n",
5960 return PTR_ERR(rdev);
5962 if (!list_empty(&mddev->disks)) {
5963 struct md_rdev *rdev0
5964 = list_entry(mddev->disks.next,
5965 struct md_rdev, same_set);
5966 err = super_types[mddev->major_version]
5967 .load_super(rdev, rdev0, mddev->minor_version);
5970 "md: %s has different UUID to %s\n",
5971 bdevname(rdev->bdev,b),
5972 bdevname(rdev0->bdev,b2));
5977 err = bind_rdev_to_array(rdev, mddev);
5984 * add_new_disk can be used once the array is assembled
5985 * to add "hot spares". They must already have a superblock
5990 if (!mddev->pers->hot_add_disk) {
5992 "%s: personality does not support diskops!\n",
5996 if (mddev->persistent)
5997 rdev = md_import_device(dev, mddev->major_version,
5998 mddev->minor_version);
6000 rdev = md_import_device(dev, -1, -1);
6003 "md: md_import_device returned %ld\n",
6005 return PTR_ERR(rdev);
6007 /* set saved_raid_disk if appropriate */
6008 if (!mddev->persistent) {
6009 if (info->state & (1<<MD_DISK_SYNC) &&
6010 info->raid_disk < mddev->raid_disks) {
6011 rdev->raid_disk = info->raid_disk;
6012 set_bit(In_sync, &rdev->flags);
6013 clear_bit(Bitmap_sync, &rdev->flags);
6015 rdev->raid_disk = -1;
6016 rdev->saved_raid_disk = rdev->raid_disk;
6018 super_types[mddev->major_version].
6019 validate_super(mddev, rdev);
6020 if ((info->state & (1<<MD_DISK_SYNC)) &&
6021 rdev->raid_disk != info->raid_disk) {
6022 /* This was a hot-add request, but events doesn't
6023 * match, so reject it.
6029 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6030 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6031 set_bit(WriteMostly, &rdev->flags);
6033 clear_bit(WriteMostly, &rdev->flags);
6035 if (info->state & (1<<MD_DISK_JOURNAL))
6036 set_bit(Journal, &rdev->flags);
6038 * check whether the device shows up in other nodes
6040 if (mddev_is_clustered(mddev)) {
6041 if (info->state & (1 << MD_DISK_CANDIDATE))
6042 set_bit(Candidate, &rdev->flags);
6043 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6044 /* --add initiated by this node */
6045 err = md_cluster_ops->add_new_disk(mddev, rdev);
6053 rdev->raid_disk = -1;
6054 err = bind_rdev_to_array(rdev, mddev);
6059 if (mddev_is_clustered(mddev)) {
6060 if (info->state & (1 << MD_DISK_CANDIDATE))
6061 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6064 md_cluster_ops->add_new_disk_cancel(mddev);
6066 err = add_bound_rdev(rdev);
6070 err = add_bound_rdev(rdev);
6075 /* otherwise, add_new_disk is only allowed
6076 * for major_version==0 superblocks
6078 if (mddev->major_version != 0) {
6079 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6084 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6086 rdev = md_import_device(dev, -1, 0);
6089 "md: error, md_import_device() returned %ld\n",
6091 return PTR_ERR(rdev);
6093 rdev->desc_nr = info->number;
6094 if (info->raid_disk < mddev->raid_disks)
6095 rdev->raid_disk = info->raid_disk;
6097 rdev->raid_disk = -1;
6099 if (rdev->raid_disk < mddev->raid_disks)
6100 if (info->state & (1<<MD_DISK_SYNC))
6101 set_bit(In_sync, &rdev->flags);
6103 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6104 set_bit(WriteMostly, &rdev->flags);
6106 if (!mddev->persistent) {
6107 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6108 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6110 rdev->sb_start = calc_dev_sboffset(rdev);
6111 rdev->sectors = rdev->sb_start;
6113 err = bind_rdev_to_array(rdev, mddev);
6123 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6125 char b[BDEVNAME_SIZE];
6126 struct md_rdev *rdev;
6129 rdev = find_rdev(mddev, dev);
6133 if (mddev_is_clustered(mddev))
6134 ret = md_cluster_ops->metadata_update_start(mddev);
6136 if (rdev->raid_disk < 0)
6139 clear_bit(Blocked, &rdev->flags);
6140 remove_and_add_spares(mddev, rdev);
6142 if (rdev->raid_disk >= 0)
6146 if (mddev_is_clustered(mddev) && ret == 0)
6147 md_cluster_ops->remove_disk(mddev, rdev);
6149 md_kick_rdev_from_array(rdev);
6150 md_update_sb(mddev, 1);
6151 md_new_event(mddev);
6155 if (mddev_is_clustered(mddev) && ret == 0)
6156 md_cluster_ops->metadata_update_cancel(mddev);
6158 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6159 bdevname(rdev->bdev,b), mdname(mddev));
6163 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6165 char b[BDEVNAME_SIZE];
6167 struct md_rdev *rdev;
6172 if (mddev->major_version != 0) {
6173 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6174 " version-0 superblocks.\n",
6178 if (!mddev->pers->hot_add_disk) {
6180 "%s: personality does not support diskops!\n",
6185 rdev = md_import_device(dev, -1, 0);
6188 "md: error, md_import_device() returned %ld\n",
6193 if (mddev->persistent)
6194 rdev->sb_start = calc_dev_sboffset(rdev);
6196 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6198 rdev->sectors = rdev->sb_start;
6200 if (test_bit(Faulty, &rdev->flags)) {
6202 "md: can not hot-add faulty %s disk to %s!\n",
6203 bdevname(rdev->bdev,b), mdname(mddev));
6208 clear_bit(In_sync, &rdev->flags);
6210 rdev->saved_raid_disk = -1;
6211 err = bind_rdev_to_array(rdev, mddev);
6216 * The rest should better be atomic, we can have disk failures
6217 * noticed in interrupt contexts ...
6220 rdev->raid_disk = -1;
6222 md_update_sb(mddev, 1);
6224 * Kick recovery, maybe this spare has to be added to the
6225 * array immediately.
6227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6228 md_wakeup_thread(mddev->thread);
6229 md_new_event(mddev);
6237 static int set_bitmap_file(struct mddev *mddev, int fd)
6242 if (!mddev->pers->quiesce || !mddev->thread)
6244 if (mddev->recovery || mddev->sync_thread)
6246 /* we should be able to change the bitmap.. */
6250 struct inode *inode;
6253 if (mddev->bitmap || mddev->bitmap_info.file)
6254 return -EEXIST; /* cannot add when bitmap is present */
6258 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6263 inode = f->f_mapping->host;
6264 if (!S_ISREG(inode->i_mode)) {
6265 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6268 } else if (!(f->f_mode & FMODE_WRITE)) {
6269 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6272 } else if (atomic_read(&inode->i_writecount) != 1) {
6273 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6281 mddev->bitmap_info.file = f;
6282 mddev->bitmap_info.offset = 0; /* file overrides offset */
6283 } else if (mddev->bitmap == NULL)
6284 return -ENOENT; /* cannot remove what isn't there */
6287 mddev->pers->quiesce(mddev, 1);
6289 struct bitmap *bitmap;
6291 bitmap = bitmap_create(mddev, -1);
6292 if (!IS_ERR(bitmap)) {
6293 mddev->bitmap = bitmap;
6294 err = bitmap_load(mddev);
6296 err = PTR_ERR(bitmap);
6298 if (fd < 0 || err) {
6299 bitmap_destroy(mddev);
6300 fd = -1; /* make sure to put the file */
6302 mddev->pers->quiesce(mddev, 0);
6305 struct file *f = mddev->bitmap_info.file;
6307 spin_lock(&mddev->lock);
6308 mddev->bitmap_info.file = NULL;
6309 spin_unlock(&mddev->lock);
6318 * set_array_info is used two different ways
6319 * The original usage is when creating a new array.
6320 * In this usage, raid_disks is > 0 and it together with
6321 * level, size, not_persistent,layout,chunksize determine the
6322 * shape of the array.
6323 * This will always create an array with a type-0.90.0 superblock.
6324 * The newer usage is when assembling an array.
6325 * In this case raid_disks will be 0, and the major_version field is
6326 * use to determine which style super-blocks are to be found on the devices.
6327 * The minor and patch _version numbers are also kept incase the
6328 * super_block handler wishes to interpret them.
6330 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6333 if (info->raid_disks == 0) {
6334 /* just setting version number for superblock loading */
6335 if (info->major_version < 0 ||
6336 info->major_version >= ARRAY_SIZE(super_types) ||
6337 super_types[info->major_version].name == NULL) {
6338 /* maybe try to auto-load a module? */
6340 "md: superblock version %d not known\n",
6341 info->major_version);
6344 mddev->major_version = info->major_version;
6345 mddev->minor_version = info->minor_version;
6346 mddev->patch_version = info->patch_version;
6347 mddev->persistent = !info->not_persistent;
6348 /* ensure mddev_put doesn't delete this now that there
6349 * is some minimal configuration.
6351 mddev->ctime = get_seconds();
6354 mddev->major_version = MD_MAJOR_VERSION;
6355 mddev->minor_version = MD_MINOR_VERSION;
6356 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6357 mddev->ctime = get_seconds();
6359 mddev->level = info->level;
6360 mddev->clevel[0] = 0;
6361 mddev->dev_sectors = 2 * (sector_t)info->size;
6362 mddev->raid_disks = info->raid_disks;
6363 /* don't set md_minor, it is determined by which /dev/md* was
6366 if (info->state & (1<<MD_SB_CLEAN))
6367 mddev->recovery_cp = MaxSector;
6369 mddev->recovery_cp = 0;
6370 mddev->persistent = ! info->not_persistent;
6371 mddev->external = 0;
6373 mddev->layout = info->layout;
6374 mddev->chunk_sectors = info->chunk_size >> 9;
6376 mddev->max_disks = MD_SB_DISKS;
6378 if (mddev->persistent)
6380 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6382 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6383 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6384 mddev->bitmap_info.offset = 0;
6386 mddev->reshape_position = MaxSector;
6389 * Generate a 128 bit UUID
6391 get_random_bytes(mddev->uuid, 16);
6393 mddev->new_level = mddev->level;
6394 mddev->new_chunk_sectors = mddev->chunk_sectors;
6395 mddev->new_layout = mddev->layout;
6396 mddev->delta_disks = 0;
6397 mddev->reshape_backwards = 0;
6402 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6404 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6406 if (mddev->external_size)
6409 mddev->array_sectors = array_sectors;
6411 EXPORT_SYMBOL(md_set_array_sectors);
6413 static int update_size(struct mddev *mddev, sector_t num_sectors)
6415 struct md_rdev *rdev;
6417 int fit = (num_sectors == 0);
6419 if (mddev->pers->resize == NULL)
6421 /* The "num_sectors" is the number of sectors of each device that
6422 * is used. This can only make sense for arrays with redundancy.
6423 * linear and raid0 always use whatever space is available. We can only
6424 * consider changing this number if no resync or reconstruction is
6425 * happening, and if the new size is acceptable. It must fit before the
6426 * sb_start or, if that is <data_offset, it must fit before the size
6427 * of each device. If num_sectors is zero, we find the largest size
6430 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6436 rdev_for_each(rdev, mddev) {
6437 sector_t avail = rdev->sectors;
6439 if (fit && (num_sectors == 0 || num_sectors > avail))
6440 num_sectors = avail;
6441 if (avail < num_sectors)
6444 rv = mddev->pers->resize(mddev, num_sectors);
6446 revalidate_disk(mddev->gendisk);
6450 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6453 struct md_rdev *rdev;
6454 /* change the number of raid disks */
6455 if (mddev->pers->check_reshape == NULL)
6459 if (raid_disks <= 0 ||
6460 (mddev->max_disks && raid_disks >= mddev->max_disks))
6462 if (mddev->sync_thread ||
6463 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6464 mddev->reshape_position != MaxSector)
6467 rdev_for_each(rdev, mddev) {
6468 if (mddev->raid_disks < raid_disks &&
6469 rdev->data_offset < rdev->new_data_offset)
6471 if (mddev->raid_disks > raid_disks &&
6472 rdev->data_offset > rdev->new_data_offset)
6476 mddev->delta_disks = raid_disks - mddev->raid_disks;
6477 if (mddev->delta_disks < 0)
6478 mddev->reshape_backwards = 1;
6479 else if (mddev->delta_disks > 0)
6480 mddev->reshape_backwards = 0;
6482 rv = mddev->pers->check_reshape(mddev);
6484 mddev->delta_disks = 0;
6485 mddev->reshape_backwards = 0;
6491 * update_array_info is used to change the configuration of an
6493 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6494 * fields in the info are checked against the array.
6495 * Any differences that cannot be handled will cause an error.
6496 * Normally, only one change can be managed at a time.
6498 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6504 /* calculate expected state,ignoring low bits */
6505 if (mddev->bitmap && mddev->bitmap_info.offset)
6506 state |= (1 << MD_SB_BITMAP_PRESENT);
6508 if (mddev->major_version != info->major_version ||
6509 mddev->minor_version != info->minor_version ||
6510 /* mddev->patch_version != info->patch_version || */
6511 mddev->ctime != info->ctime ||
6512 mddev->level != info->level ||
6513 /* mddev->layout != info->layout || */
6514 mddev->persistent != !info->not_persistent ||
6515 mddev->chunk_sectors != info->chunk_size >> 9 ||
6516 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6517 ((state^info->state) & 0xfffffe00)
6520 /* Check there is only one change */
6521 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6523 if (mddev->raid_disks != info->raid_disks)
6525 if (mddev->layout != info->layout)
6527 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6534 if (mddev->layout != info->layout) {
6536 * we don't need to do anything at the md level, the
6537 * personality will take care of it all.
6539 if (mddev->pers->check_reshape == NULL)
6542 mddev->new_layout = info->layout;
6543 rv = mddev->pers->check_reshape(mddev);
6545 mddev->new_layout = mddev->layout;
6549 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6550 rv = update_size(mddev, (sector_t)info->size * 2);
6552 if (mddev->raid_disks != info->raid_disks)
6553 rv = update_raid_disks(mddev, info->raid_disks);
6555 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6556 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6560 if (mddev->recovery || mddev->sync_thread) {
6564 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6565 struct bitmap *bitmap;
6566 /* add the bitmap */
6567 if (mddev->bitmap) {
6571 if (mddev->bitmap_info.default_offset == 0) {
6575 mddev->bitmap_info.offset =
6576 mddev->bitmap_info.default_offset;
6577 mddev->bitmap_info.space =
6578 mddev->bitmap_info.default_space;
6579 mddev->pers->quiesce(mddev, 1);
6580 bitmap = bitmap_create(mddev, -1);
6581 if (!IS_ERR(bitmap)) {
6582 mddev->bitmap = bitmap;
6583 rv = bitmap_load(mddev);
6585 rv = PTR_ERR(bitmap);
6587 bitmap_destroy(mddev);
6588 mddev->pers->quiesce(mddev, 0);
6590 /* remove the bitmap */
6591 if (!mddev->bitmap) {
6595 if (mddev->bitmap->storage.file) {
6599 mddev->pers->quiesce(mddev, 1);
6600 bitmap_destroy(mddev);
6601 mddev->pers->quiesce(mddev, 0);
6602 mddev->bitmap_info.offset = 0;
6605 md_update_sb(mddev, 1);
6611 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6613 struct md_rdev *rdev;
6616 if (mddev->pers == NULL)
6620 rdev = find_rdev_rcu(mddev, dev);
6624 md_error(mddev, rdev);
6625 if (!test_bit(Faulty, &rdev->flags))
6633 * We have a problem here : there is no easy way to give a CHS
6634 * virtual geometry. We currently pretend that we have a 2 heads
6635 * 4 sectors (with a BIG number of cylinders...). This drives
6636 * dosfs just mad... ;-)
6638 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6640 struct mddev *mddev = bdev->bd_disk->private_data;
6644 geo->cylinders = mddev->array_sectors / 8;
6648 static inline bool md_ioctl_valid(unsigned int cmd)
6653 case GET_ARRAY_INFO:
6654 case GET_BITMAP_FILE:
6657 case HOT_REMOVE_DISK:
6660 case RESTART_ARRAY_RW:
6662 case SET_ARRAY_INFO:
6663 case SET_BITMAP_FILE:
6664 case SET_DISK_FAULTY:
6667 case CLUSTERED_DISK_NACK:
6674 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6675 unsigned int cmd, unsigned long arg)
6678 void __user *argp = (void __user *)arg;
6679 struct mddev *mddev = NULL;
6682 if (!md_ioctl_valid(cmd))
6687 case GET_ARRAY_INFO:
6691 if (!capable(CAP_SYS_ADMIN))
6696 * Commands dealing with the RAID driver but not any
6701 err = get_version(argp);
6707 autostart_arrays(arg);
6714 * Commands creating/starting a new array:
6717 mddev = bdev->bd_disk->private_data;
6724 /* Some actions do not requires the mutex */
6726 case GET_ARRAY_INFO:
6727 if (!mddev->raid_disks && !mddev->external)
6730 err = get_array_info(mddev, argp);
6734 if (!mddev->raid_disks && !mddev->external)
6737 err = get_disk_info(mddev, argp);
6740 case SET_DISK_FAULTY:
6741 err = set_disk_faulty(mddev, new_decode_dev(arg));
6744 case GET_BITMAP_FILE:
6745 err = get_bitmap_file(mddev, argp);
6750 if (cmd == ADD_NEW_DISK)
6751 /* need to ensure md_delayed_delete() has completed */
6752 flush_workqueue(md_misc_wq);
6754 if (cmd == HOT_REMOVE_DISK)
6755 /* need to ensure recovery thread has run */
6756 wait_event_interruptible_timeout(mddev->sb_wait,
6757 !test_bit(MD_RECOVERY_NEEDED,
6759 msecs_to_jiffies(5000));
6760 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6761 /* Need to flush page cache, and ensure no-one else opens
6764 mutex_lock(&mddev->open_mutex);
6765 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6766 mutex_unlock(&mddev->open_mutex);
6770 set_bit(MD_STILL_CLOSED, &mddev->flags);
6771 mutex_unlock(&mddev->open_mutex);
6772 sync_blockdev(bdev);
6774 err = mddev_lock(mddev);
6777 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6782 if (cmd == SET_ARRAY_INFO) {
6783 mdu_array_info_t info;
6785 memset(&info, 0, sizeof(info));
6786 else if (copy_from_user(&info, argp, sizeof(info))) {
6791 err = update_array_info(mddev, &info);
6793 printk(KERN_WARNING "md: couldn't update"
6794 " array info. %d\n", err);
6799 if (!list_empty(&mddev->disks)) {
6801 "md: array %s already has disks!\n",
6806 if (mddev->raid_disks) {
6808 "md: array %s already initialised!\n",
6813 err = set_array_info(mddev, &info);
6815 printk(KERN_WARNING "md: couldn't set"
6816 " array info. %d\n", err);
6823 * Commands querying/configuring an existing array:
6825 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6826 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6827 if ((!mddev->raid_disks && !mddev->external)
6828 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6829 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6830 && cmd != GET_BITMAP_FILE) {
6836 * Commands even a read-only array can execute:
6839 case RESTART_ARRAY_RW:
6840 err = restart_array(mddev);
6844 err = do_md_stop(mddev, 0, bdev);
6848 err = md_set_readonly(mddev, bdev);
6851 case HOT_REMOVE_DISK:
6852 err = hot_remove_disk(mddev, new_decode_dev(arg));
6856 /* We can support ADD_NEW_DISK on read-only arrays
6857 * on if we are re-adding a preexisting device.
6858 * So require mddev->pers and MD_DISK_SYNC.
6861 mdu_disk_info_t info;
6862 if (copy_from_user(&info, argp, sizeof(info)))
6864 else if (!(info.state & (1<<MD_DISK_SYNC)))
6865 /* Need to clear read-only for this */
6868 err = add_new_disk(mddev, &info);
6874 if (get_user(ro, (int __user *)(arg))) {
6880 /* if the bdev is going readonly the value of mddev->ro
6881 * does not matter, no writes are coming
6886 /* are we are already prepared for writes? */
6890 /* transitioning to readauto need only happen for
6891 * arrays that call md_write_start
6894 err = restart_array(mddev);
6897 set_disk_ro(mddev->gendisk, 0);
6904 * The remaining ioctls are changing the state of the
6905 * superblock, so we do not allow them on read-only arrays.
6907 if (mddev->ro && mddev->pers) {
6908 if (mddev->ro == 2) {
6910 sysfs_notify_dirent_safe(mddev->sysfs_state);
6911 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6912 /* mddev_unlock will wake thread */
6913 /* If a device failed while we were read-only, we
6914 * need to make sure the metadata is updated now.
6916 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6917 mddev_unlock(mddev);
6918 wait_event(mddev->sb_wait,
6919 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6920 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6921 mddev_lock_nointr(mddev);
6932 mdu_disk_info_t info;
6933 if (copy_from_user(&info, argp, sizeof(info)))
6936 err = add_new_disk(mddev, &info);
6940 case CLUSTERED_DISK_NACK:
6941 if (mddev_is_clustered(mddev))
6942 md_cluster_ops->new_disk_ack(mddev, false);
6948 err = hot_add_disk(mddev, new_decode_dev(arg));
6952 err = do_md_run(mddev);
6955 case SET_BITMAP_FILE:
6956 err = set_bitmap_file(mddev, (int)arg);
6965 if (mddev->hold_active == UNTIL_IOCTL &&
6967 mddev->hold_active = 0;
6968 mddev_unlock(mddev);
6972 #ifdef CONFIG_COMPAT
6973 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6974 unsigned int cmd, unsigned long arg)
6977 case HOT_REMOVE_DISK:
6979 case SET_DISK_FAULTY:
6980 case SET_BITMAP_FILE:
6981 /* These take in integer arg, do not convert */
6984 arg = (unsigned long)compat_ptr(arg);
6988 return md_ioctl(bdev, mode, cmd, arg);
6990 #endif /* CONFIG_COMPAT */
6992 static int md_open(struct block_device *bdev, fmode_t mode)
6995 * Succeed if we can lock the mddev, which confirms that
6996 * it isn't being stopped right now.
6998 struct mddev *mddev = mddev_find(bdev->bd_dev);
7004 if (mddev->gendisk != bdev->bd_disk) {
7005 /* we are racing with mddev_put which is discarding this
7009 /* Wait until bdev->bd_disk is definitely gone */
7010 flush_workqueue(md_misc_wq);
7011 /* Then retry the open from the top */
7012 return -ERESTARTSYS;
7014 BUG_ON(mddev != bdev->bd_disk->private_data);
7016 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7020 atomic_inc(&mddev->openers);
7021 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7022 mutex_unlock(&mddev->open_mutex);
7024 check_disk_change(bdev);
7029 static void md_release(struct gendisk *disk, fmode_t mode)
7031 struct mddev *mddev = disk->private_data;
7034 atomic_dec(&mddev->openers);
7038 static int md_media_changed(struct gendisk *disk)
7040 struct mddev *mddev = disk->private_data;
7042 return mddev->changed;
7045 static int md_revalidate(struct gendisk *disk)
7047 struct mddev *mddev = disk->private_data;
7052 static const struct block_device_operations md_fops =
7054 .owner = THIS_MODULE,
7056 .release = md_release,
7058 #ifdef CONFIG_COMPAT
7059 .compat_ioctl = md_compat_ioctl,
7061 .getgeo = md_getgeo,
7062 .media_changed = md_media_changed,
7063 .revalidate_disk= md_revalidate,
7066 static int md_thread(void *arg)
7068 struct md_thread *thread = arg;
7071 * md_thread is a 'system-thread', it's priority should be very
7072 * high. We avoid resource deadlocks individually in each
7073 * raid personality. (RAID5 does preallocation) We also use RR and
7074 * the very same RT priority as kswapd, thus we will never get
7075 * into a priority inversion deadlock.
7077 * we definitely have to have equal or higher priority than
7078 * bdflush, otherwise bdflush will deadlock if there are too
7079 * many dirty RAID5 blocks.
7082 allow_signal(SIGKILL);
7083 while (!kthread_should_stop()) {
7085 /* We need to wait INTERRUPTIBLE so that
7086 * we don't add to the load-average.
7087 * That means we need to be sure no signals are
7090 if (signal_pending(current))
7091 flush_signals(current);
7093 wait_event_interruptible_timeout
7095 test_bit(THREAD_WAKEUP, &thread->flags)
7096 || kthread_should_stop(),
7099 clear_bit(THREAD_WAKEUP, &thread->flags);
7100 if (!kthread_should_stop())
7101 thread->run(thread);
7107 void md_wakeup_thread(struct md_thread *thread)
7110 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7111 set_bit(THREAD_WAKEUP, &thread->flags);
7112 wake_up(&thread->wqueue);
7115 EXPORT_SYMBOL(md_wakeup_thread);
7117 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7118 struct mddev *mddev, const char *name)
7120 struct md_thread *thread;
7122 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7126 init_waitqueue_head(&thread->wqueue);
7129 thread->mddev = mddev;
7130 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7131 thread->tsk = kthread_run(md_thread, thread,
7133 mdname(thread->mddev),
7135 if (IS_ERR(thread->tsk)) {
7141 EXPORT_SYMBOL(md_register_thread);
7143 void md_unregister_thread(struct md_thread **threadp)
7145 struct md_thread *thread = *threadp;
7148 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7149 /* Locking ensures that mddev_unlock does not wake_up a
7150 * non-existent thread
7152 spin_lock(&pers_lock);
7154 spin_unlock(&pers_lock);
7156 kthread_stop(thread->tsk);
7159 EXPORT_SYMBOL(md_unregister_thread);
7161 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7163 if (!rdev || test_bit(Faulty, &rdev->flags))
7166 if (!mddev->pers || !mddev->pers->error_handler)
7168 mddev->pers->error_handler(mddev,rdev);
7169 if (mddev->degraded)
7170 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7171 sysfs_notify_dirent_safe(rdev->sysfs_state);
7172 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7173 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7174 md_wakeup_thread(mddev->thread);
7175 if (mddev->event_work.func)
7176 queue_work(md_misc_wq, &mddev->event_work);
7177 md_new_event_inintr(mddev);
7179 EXPORT_SYMBOL(md_error);
7181 /* seq_file implementation /proc/mdstat */
7183 static void status_unused(struct seq_file *seq)
7186 struct md_rdev *rdev;
7188 seq_printf(seq, "unused devices: ");
7190 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7191 char b[BDEVNAME_SIZE];
7193 seq_printf(seq, "%s ",
7194 bdevname(rdev->bdev,b));
7197 seq_printf(seq, "<none>");
7199 seq_printf(seq, "\n");
7202 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7204 sector_t max_sectors, resync, res;
7205 unsigned long dt, db;
7208 unsigned int per_milli;
7210 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7211 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7212 max_sectors = mddev->resync_max_sectors;
7214 max_sectors = mddev->dev_sectors;
7216 resync = mddev->curr_resync;
7218 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7219 /* Still cleaning up */
7220 resync = max_sectors;
7222 resync -= atomic_read(&mddev->recovery_active);
7225 if (mddev->recovery_cp < MaxSector) {
7226 seq_printf(seq, "\tresync=PENDING");
7232 seq_printf(seq, "\tresync=DELAYED");
7236 WARN_ON(max_sectors == 0);
7237 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7238 * in a sector_t, and (max_sectors>>scale) will fit in a
7239 * u32, as those are the requirements for sector_div.
7240 * Thus 'scale' must be at least 10
7243 if (sizeof(sector_t) > sizeof(unsigned long)) {
7244 while ( max_sectors/2 > (1ULL<<(scale+32)))
7247 res = (resync>>scale)*1000;
7248 sector_div(res, (u32)((max_sectors>>scale)+1));
7252 int i, x = per_milli/50, y = 20-x;
7253 seq_printf(seq, "[");
7254 for (i = 0; i < x; i++)
7255 seq_printf(seq, "=");
7256 seq_printf(seq, ">");
7257 for (i = 0; i < y; i++)
7258 seq_printf(seq, ".");
7259 seq_printf(seq, "] ");
7261 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7262 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7264 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7266 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7267 "resync" : "recovery"))),
7268 per_milli/10, per_milli % 10,
7269 (unsigned long long) resync/2,
7270 (unsigned long long) max_sectors/2);
7273 * dt: time from mark until now
7274 * db: blocks written from mark until now
7275 * rt: remaining time
7277 * rt is a sector_t, so could be 32bit or 64bit.
7278 * So we divide before multiply in case it is 32bit and close
7280 * We scale the divisor (db) by 32 to avoid losing precision
7281 * near the end of resync when the number of remaining sectors
7283 * We then divide rt by 32 after multiplying by db to compensate.
7284 * The '+1' avoids division by zero if db is very small.
7286 dt = ((jiffies - mddev->resync_mark) / HZ);
7288 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7289 - mddev->resync_mark_cnt;
7291 rt = max_sectors - resync; /* number of remaining sectors */
7292 sector_div(rt, db/32+1);
7296 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7297 ((unsigned long)rt % 60)/6);
7299 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7303 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7305 struct list_head *tmp;
7307 struct mddev *mddev;
7315 spin_lock(&all_mddevs_lock);
7316 list_for_each(tmp,&all_mddevs)
7318 mddev = list_entry(tmp, struct mddev, all_mddevs);
7320 spin_unlock(&all_mddevs_lock);
7323 spin_unlock(&all_mddevs_lock);
7325 return (void*)2;/* tail */
7329 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7331 struct list_head *tmp;
7332 struct mddev *next_mddev, *mddev = v;
7338 spin_lock(&all_mddevs_lock);
7340 tmp = all_mddevs.next;
7342 tmp = mddev->all_mddevs.next;
7343 if (tmp != &all_mddevs)
7344 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7346 next_mddev = (void*)2;
7349 spin_unlock(&all_mddevs_lock);
7357 static void md_seq_stop(struct seq_file *seq, void *v)
7359 struct mddev *mddev = v;
7361 if (mddev && v != (void*)1 && v != (void*)2)
7365 static int md_seq_show(struct seq_file *seq, void *v)
7367 struct mddev *mddev = v;
7369 struct md_rdev *rdev;
7371 if (v == (void*)1) {
7372 struct md_personality *pers;
7373 seq_printf(seq, "Personalities : ");
7374 spin_lock(&pers_lock);
7375 list_for_each_entry(pers, &pers_list, list)
7376 seq_printf(seq, "[%s] ", pers->name);
7378 spin_unlock(&pers_lock);
7379 seq_printf(seq, "\n");
7380 seq->poll_event = atomic_read(&md_event_count);
7383 if (v == (void*)2) {
7388 spin_lock(&mddev->lock);
7389 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7390 seq_printf(seq, "%s : %sactive", mdname(mddev),
7391 mddev->pers ? "" : "in");
7394 seq_printf(seq, " (read-only)");
7396 seq_printf(seq, " (auto-read-only)");
7397 seq_printf(seq, " %s", mddev->pers->name);
7402 rdev_for_each_rcu(rdev, mddev) {
7403 char b[BDEVNAME_SIZE];
7404 seq_printf(seq, " %s[%d]",
7405 bdevname(rdev->bdev,b), rdev->desc_nr);
7406 if (test_bit(WriteMostly, &rdev->flags))
7407 seq_printf(seq, "(W)");
7408 if (test_bit(Journal, &rdev->flags))
7409 seq_printf(seq, "(J)");
7410 if (test_bit(Faulty, &rdev->flags)) {
7411 seq_printf(seq, "(F)");
7414 if (rdev->raid_disk < 0)
7415 seq_printf(seq, "(S)"); /* spare */
7416 if (test_bit(Replacement, &rdev->flags))
7417 seq_printf(seq, "(R)");
7418 sectors += rdev->sectors;
7422 if (!list_empty(&mddev->disks)) {
7424 seq_printf(seq, "\n %llu blocks",
7425 (unsigned long long)
7426 mddev->array_sectors / 2);
7428 seq_printf(seq, "\n %llu blocks",
7429 (unsigned long long)sectors / 2);
7431 if (mddev->persistent) {
7432 if (mddev->major_version != 0 ||
7433 mddev->minor_version != 90) {
7434 seq_printf(seq," super %d.%d",
7435 mddev->major_version,
7436 mddev->minor_version);
7438 } else if (mddev->external)
7439 seq_printf(seq, " super external:%s",
7440 mddev->metadata_type);
7442 seq_printf(seq, " super non-persistent");
7445 mddev->pers->status(seq, mddev);
7446 seq_printf(seq, "\n ");
7447 if (mddev->pers->sync_request) {
7448 if (status_resync(seq, mddev))
7449 seq_printf(seq, "\n ");
7452 seq_printf(seq, "\n ");
7454 bitmap_status(seq, mddev->bitmap);
7456 seq_printf(seq, "\n");
7458 spin_unlock(&mddev->lock);
7463 static const struct seq_operations md_seq_ops = {
7464 .start = md_seq_start,
7465 .next = md_seq_next,
7466 .stop = md_seq_stop,
7467 .show = md_seq_show,
7470 static int md_seq_open(struct inode *inode, struct file *file)
7472 struct seq_file *seq;
7475 error = seq_open(file, &md_seq_ops);
7479 seq = file->private_data;
7480 seq->poll_event = atomic_read(&md_event_count);
7484 static int md_unloading;
7485 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7487 struct seq_file *seq = filp->private_data;
7491 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7492 poll_wait(filp, &md_event_waiters, wait);
7494 /* always allow read */
7495 mask = POLLIN | POLLRDNORM;
7497 if (seq->poll_event != atomic_read(&md_event_count))
7498 mask |= POLLERR | POLLPRI;
7502 static const struct file_operations md_seq_fops = {
7503 .owner = THIS_MODULE,
7504 .open = md_seq_open,
7506 .llseek = seq_lseek,
7507 .release = seq_release_private,
7508 .poll = mdstat_poll,
7511 int register_md_personality(struct md_personality *p)
7513 printk(KERN_INFO "md: %s personality registered for level %d\n",
7515 spin_lock(&pers_lock);
7516 list_add_tail(&p->list, &pers_list);
7517 spin_unlock(&pers_lock);
7520 EXPORT_SYMBOL(register_md_personality);
7522 int unregister_md_personality(struct md_personality *p)
7524 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7525 spin_lock(&pers_lock);
7526 list_del_init(&p->list);
7527 spin_unlock(&pers_lock);
7530 EXPORT_SYMBOL(unregister_md_personality);
7532 int register_md_cluster_operations(struct md_cluster_operations *ops,
7533 struct module *module)
7536 spin_lock(&pers_lock);
7537 if (md_cluster_ops != NULL)
7540 md_cluster_ops = ops;
7541 md_cluster_mod = module;
7543 spin_unlock(&pers_lock);
7546 EXPORT_SYMBOL(register_md_cluster_operations);
7548 int unregister_md_cluster_operations(void)
7550 spin_lock(&pers_lock);
7551 md_cluster_ops = NULL;
7552 spin_unlock(&pers_lock);
7555 EXPORT_SYMBOL(unregister_md_cluster_operations);
7557 int md_setup_cluster(struct mddev *mddev, int nodes)
7561 err = request_module("md-cluster");
7563 pr_err("md-cluster module not found.\n");
7567 spin_lock(&pers_lock);
7568 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7569 spin_unlock(&pers_lock);
7572 spin_unlock(&pers_lock);
7574 return md_cluster_ops->join(mddev, nodes);
7577 void md_cluster_stop(struct mddev *mddev)
7579 if (!md_cluster_ops)
7581 md_cluster_ops->leave(mddev);
7582 module_put(md_cluster_mod);
7585 static int is_mddev_idle(struct mddev *mddev, int init)
7587 struct md_rdev *rdev;
7593 rdev_for_each_rcu(rdev, mddev) {
7594 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7595 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7596 (int)part_stat_read(&disk->part0, sectors[1]) -
7597 atomic_read(&disk->sync_io);
7598 /* sync IO will cause sync_io to increase before the disk_stats
7599 * as sync_io is counted when a request starts, and
7600 * disk_stats is counted when it completes.
7601 * So resync activity will cause curr_events to be smaller than
7602 * when there was no such activity.
7603 * non-sync IO will cause disk_stat to increase without
7604 * increasing sync_io so curr_events will (eventually)
7605 * be larger than it was before. Once it becomes
7606 * substantially larger, the test below will cause
7607 * the array to appear non-idle, and resync will slow
7609 * If there is a lot of outstanding resync activity when
7610 * we set last_event to curr_events, then all that activity
7611 * completing might cause the array to appear non-idle
7612 * and resync will be slowed down even though there might
7613 * not have been non-resync activity. This will only
7614 * happen once though. 'last_events' will soon reflect
7615 * the state where there is little or no outstanding
7616 * resync requests, and further resync activity will
7617 * always make curr_events less than last_events.
7620 if (init || curr_events - rdev->last_events > 64) {
7621 rdev->last_events = curr_events;
7629 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7631 /* another "blocks" (512byte) blocks have been synced */
7632 atomic_sub(blocks, &mddev->recovery_active);
7633 wake_up(&mddev->recovery_wait);
7635 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7636 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7637 md_wakeup_thread(mddev->thread);
7638 // stop recovery, signal do_sync ....
7641 EXPORT_SYMBOL(md_done_sync);
7643 /* md_write_start(mddev, bi)
7644 * If we need to update some array metadata (e.g. 'active' flag
7645 * in superblock) before writing, schedule a superblock update
7646 * and wait for it to complete.
7648 void md_write_start(struct mddev *mddev, struct bio *bi)
7651 if (bio_data_dir(bi) != WRITE)
7654 BUG_ON(mddev->ro == 1);
7655 if (mddev->ro == 2) {
7656 /* need to switch to read/write */
7658 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7659 md_wakeup_thread(mddev->thread);
7660 md_wakeup_thread(mddev->sync_thread);
7663 atomic_inc(&mddev->writes_pending);
7664 if (mddev->safemode == 1)
7665 mddev->safemode = 0;
7666 if (mddev->in_sync) {
7667 spin_lock(&mddev->lock);
7668 if (mddev->in_sync) {
7670 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7671 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7672 md_wakeup_thread(mddev->thread);
7675 spin_unlock(&mddev->lock);
7678 sysfs_notify_dirent_safe(mddev->sysfs_state);
7679 wait_event(mddev->sb_wait,
7680 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7682 EXPORT_SYMBOL(md_write_start);
7684 void md_write_end(struct mddev *mddev)
7686 if (atomic_dec_and_test(&mddev->writes_pending)) {
7687 if (mddev->safemode == 2)
7688 md_wakeup_thread(mddev->thread);
7689 else if (mddev->safemode_delay)
7690 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7693 EXPORT_SYMBOL(md_write_end);
7695 /* md_allow_write(mddev)
7696 * Calling this ensures that the array is marked 'active' so that writes
7697 * may proceed without blocking. It is important to call this before
7698 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7699 * Must be called with mddev_lock held.
7701 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7702 * is dropped, so return -EAGAIN after notifying userspace.
7704 int md_allow_write(struct mddev *mddev)
7710 if (!mddev->pers->sync_request)
7713 spin_lock(&mddev->lock);
7714 if (mddev->in_sync) {
7716 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7717 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7718 if (mddev->safemode_delay &&
7719 mddev->safemode == 0)
7720 mddev->safemode = 1;
7721 spin_unlock(&mddev->lock);
7722 md_update_sb(mddev, 0);
7723 sysfs_notify_dirent_safe(mddev->sysfs_state);
7725 spin_unlock(&mddev->lock);
7727 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7732 EXPORT_SYMBOL_GPL(md_allow_write);
7734 #define SYNC_MARKS 10
7735 #define SYNC_MARK_STEP (3*HZ)
7736 #define UPDATE_FREQUENCY (5*60*HZ)
7737 void md_do_sync(struct md_thread *thread)
7739 struct mddev *mddev = thread->mddev;
7740 struct mddev *mddev2;
7741 unsigned int currspeed = 0,
7743 sector_t max_sectors,j, io_sectors, recovery_done;
7744 unsigned long mark[SYNC_MARKS];
7745 unsigned long update_time;
7746 sector_t mark_cnt[SYNC_MARKS];
7748 struct list_head *tmp;
7749 sector_t last_check;
7751 struct md_rdev *rdev;
7752 char *desc, *action = NULL;
7753 struct blk_plug plug;
7754 bool cluster_resync_finished = false;
7756 /* just incase thread restarts... */
7757 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7759 if (mddev->ro) {/* never try to sync a read-only array */
7760 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7764 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7765 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7766 desc = "data-check";
7768 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7769 desc = "requested-resync";
7773 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7778 mddev->last_sync_action = action ?: desc;
7780 /* we overload curr_resync somewhat here.
7781 * 0 == not engaged in resync at all
7782 * 2 == checking that there is no conflict with another sync
7783 * 1 == like 2, but have yielded to allow conflicting resync to
7785 * other == active in resync - this many blocks
7787 * Before starting a resync we must have set curr_resync to
7788 * 2, and then checked that every "conflicting" array has curr_resync
7789 * less than ours. When we find one that is the same or higher
7790 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7791 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7792 * This will mean we have to start checking from the beginning again.
7797 mddev->curr_resync = 2;
7800 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7802 for_each_mddev(mddev2, tmp) {
7803 if (mddev2 == mddev)
7805 if (!mddev->parallel_resync
7806 && mddev2->curr_resync
7807 && match_mddev_units(mddev, mddev2)) {
7809 if (mddev < mddev2 && mddev->curr_resync == 2) {
7810 /* arbitrarily yield */
7811 mddev->curr_resync = 1;
7812 wake_up(&resync_wait);
7814 if (mddev > mddev2 && mddev->curr_resync == 1)
7815 /* no need to wait here, we can wait the next
7816 * time 'round when curr_resync == 2
7819 /* We need to wait 'interruptible' so as not to
7820 * contribute to the load average, and not to
7821 * be caught by 'softlockup'
7823 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7824 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7825 mddev2->curr_resync >= mddev->curr_resync) {
7826 printk(KERN_INFO "md: delaying %s of %s"
7827 " until %s has finished (they"
7828 " share one or more physical units)\n",
7829 desc, mdname(mddev), mdname(mddev2));
7831 if (signal_pending(current))
7832 flush_signals(current);
7834 finish_wait(&resync_wait, &wq);
7837 finish_wait(&resync_wait, &wq);
7840 } while (mddev->curr_resync < 2);
7843 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7844 /* resync follows the size requested by the personality,
7845 * which defaults to physical size, but can be virtual size
7847 max_sectors = mddev->resync_max_sectors;
7848 atomic64_set(&mddev->resync_mismatches, 0);
7849 /* we don't use the checkpoint if there's a bitmap */
7850 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7851 j = mddev->resync_min;
7852 else if (!mddev->bitmap)
7853 j = mddev->recovery_cp;
7855 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7856 max_sectors = mddev->resync_max_sectors;
7858 /* recovery follows the physical size of devices */
7859 max_sectors = mddev->dev_sectors;
7862 rdev_for_each_rcu(rdev, mddev)
7863 if (rdev->raid_disk >= 0 &&
7864 !test_bit(Journal, &rdev->flags) &&
7865 !test_bit(Faulty, &rdev->flags) &&
7866 !test_bit(In_sync, &rdev->flags) &&
7867 rdev->recovery_offset < j)
7868 j = rdev->recovery_offset;
7871 /* If there is a bitmap, we need to make sure all
7872 * writes that started before we added a spare
7873 * complete before we start doing a recovery.
7874 * Otherwise the write might complete and (via
7875 * bitmap_endwrite) set a bit in the bitmap after the
7876 * recovery has checked that bit and skipped that
7879 if (mddev->bitmap) {
7880 mddev->pers->quiesce(mddev, 1);
7881 mddev->pers->quiesce(mddev, 0);
7885 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7886 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7887 " %d KB/sec/disk.\n", speed_min(mddev));
7888 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7889 "(but not more than %d KB/sec) for %s.\n",
7890 speed_max(mddev), desc);
7892 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7895 for (m = 0; m < SYNC_MARKS; m++) {
7897 mark_cnt[m] = io_sectors;
7900 mddev->resync_mark = mark[last_mark];
7901 mddev->resync_mark_cnt = mark_cnt[last_mark];
7904 * Tune reconstruction:
7906 window = 32*(PAGE_SIZE/512);
7907 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7908 window/2, (unsigned long long)max_sectors/2);
7910 atomic_set(&mddev->recovery_active, 0);
7915 "md: resuming %s of %s from checkpoint.\n",
7916 desc, mdname(mddev));
7917 mddev->curr_resync = j;
7919 mddev->curr_resync = 3; /* no longer delayed */
7920 mddev->curr_resync_completed = j;
7921 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7922 md_new_event(mddev);
7923 update_time = jiffies;
7925 blk_start_plug(&plug);
7926 while (j < max_sectors) {
7931 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7932 ((mddev->curr_resync > mddev->curr_resync_completed &&
7933 (mddev->curr_resync - mddev->curr_resync_completed)
7934 > (max_sectors >> 4)) ||
7935 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7936 (j - mddev->curr_resync_completed)*2
7937 >= mddev->resync_max - mddev->curr_resync_completed ||
7938 mddev->curr_resync_completed > mddev->resync_max
7940 /* time to update curr_resync_completed */
7941 wait_event(mddev->recovery_wait,
7942 atomic_read(&mddev->recovery_active) == 0);
7943 mddev->curr_resync_completed = j;
7944 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7945 j > mddev->recovery_cp)
7946 mddev->recovery_cp = j;
7947 update_time = jiffies;
7948 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7949 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7952 while (j >= mddev->resync_max &&
7953 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7954 /* As this condition is controlled by user-space,
7955 * we can block indefinitely, so use '_interruptible'
7956 * to avoid triggering warnings.
7958 flush_signals(current); /* just in case */
7959 wait_event_interruptible(mddev->recovery_wait,
7960 mddev->resync_max > j
7961 || test_bit(MD_RECOVERY_INTR,
7965 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7968 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7970 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7974 if (!skipped) { /* actual IO requested */
7975 io_sectors += sectors;
7976 atomic_add(sectors, &mddev->recovery_active);
7979 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7983 if (j > max_sectors)
7984 /* when skipping, extra large numbers can be returned. */
7987 mddev->curr_resync = j;
7988 mddev->curr_mark_cnt = io_sectors;
7989 if (last_check == 0)
7990 /* this is the earliest that rebuild will be
7991 * visible in /proc/mdstat
7993 md_new_event(mddev);
7995 if (last_check + window > io_sectors || j == max_sectors)
7998 last_check = io_sectors;
8000 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8002 int next = (last_mark+1) % SYNC_MARKS;
8004 mddev->resync_mark = mark[next];
8005 mddev->resync_mark_cnt = mark_cnt[next];
8006 mark[next] = jiffies;
8007 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8011 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8015 * this loop exits only if either when we are slower than
8016 * the 'hard' speed limit, or the system was IO-idle for
8018 * the system might be non-idle CPU-wise, but we only care
8019 * about not overloading the IO subsystem. (things like an
8020 * e2fsck being done on the RAID array should execute fast)
8024 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8025 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8026 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8028 if (currspeed > speed_min(mddev)) {
8029 if (currspeed > speed_max(mddev)) {
8033 if (!is_mddev_idle(mddev, 0)) {
8035 * Give other IO more of a chance.
8036 * The faster the devices, the less we wait.
8038 wait_event(mddev->recovery_wait,
8039 !atomic_read(&mddev->recovery_active));
8043 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8044 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8045 ? "interrupted" : "done");
8047 * this also signals 'finished resyncing' to md_stop
8049 blk_finish_plug(&plug);
8050 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8052 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8053 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8054 mddev->curr_resync > 2) {
8055 mddev->curr_resync_completed = mddev->curr_resync;
8056 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8058 /* tell personality and other nodes that we are finished */
8059 if (mddev_is_clustered(mddev)) {
8060 md_cluster_ops->resync_finish(mddev);
8061 cluster_resync_finished = true;
8063 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8065 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8066 mddev->curr_resync > 2) {
8067 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8068 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8069 if (mddev->curr_resync >= mddev->recovery_cp) {
8071 "md: checkpointing %s of %s.\n",
8072 desc, mdname(mddev));
8073 if (test_bit(MD_RECOVERY_ERROR,
8075 mddev->recovery_cp =
8076 mddev->curr_resync_completed;
8078 mddev->recovery_cp =
8082 mddev->recovery_cp = MaxSector;
8084 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8085 mddev->curr_resync = MaxSector;
8087 rdev_for_each_rcu(rdev, mddev)
8088 if (rdev->raid_disk >= 0 &&
8089 mddev->delta_disks >= 0 &&
8090 !test_bit(Journal, &rdev->flags) &&
8091 !test_bit(Faulty, &rdev->flags) &&
8092 !test_bit(In_sync, &rdev->flags) &&
8093 rdev->recovery_offset < mddev->curr_resync)
8094 rdev->recovery_offset = mddev->curr_resync;
8099 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8101 if (mddev_is_clustered(mddev) &&
8102 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8103 !cluster_resync_finished)
8104 md_cluster_ops->resync_finish(mddev);
8106 spin_lock(&mddev->lock);
8107 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8108 /* We completed so min/max setting can be forgotten if used. */
8109 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8110 mddev->resync_min = 0;
8111 mddev->resync_max = MaxSector;
8112 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8113 mddev->resync_min = mddev->curr_resync_completed;
8114 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8115 mddev->curr_resync = 0;
8116 spin_unlock(&mddev->lock);
8118 wake_up(&resync_wait);
8119 md_wakeup_thread(mddev->thread);
8122 EXPORT_SYMBOL_GPL(md_do_sync);
8124 static int remove_and_add_spares(struct mddev *mddev,
8125 struct md_rdev *this)
8127 struct md_rdev *rdev;
8131 rdev_for_each(rdev, mddev)
8132 if ((this == NULL || rdev == this) &&
8133 rdev->raid_disk >= 0 &&
8134 !test_bit(Blocked, &rdev->flags) &&
8135 (test_bit(Faulty, &rdev->flags) ||
8136 (!test_bit(In_sync, &rdev->flags) &&
8137 !test_bit(Journal, &rdev->flags))) &&
8138 atomic_read(&rdev->nr_pending)==0) {
8139 if (mddev->pers->hot_remove_disk(
8140 mddev, rdev) == 0) {
8141 sysfs_unlink_rdev(mddev, rdev);
8142 rdev->raid_disk = -1;
8146 if (removed && mddev->kobj.sd)
8147 sysfs_notify(&mddev->kobj, NULL, "degraded");
8149 if (this && removed)
8152 rdev_for_each(rdev, mddev) {
8153 if (this && this != rdev)
8155 if (test_bit(Candidate, &rdev->flags))
8157 if (rdev->raid_disk >= 0 &&
8158 !test_bit(In_sync, &rdev->flags) &&
8159 !test_bit(Journal, &rdev->flags) &&
8160 !test_bit(Faulty, &rdev->flags))
8162 if (rdev->raid_disk >= 0)
8164 if (test_bit(Faulty, &rdev->flags))
8166 if (test_bit(Journal, &rdev->flags))
8169 ! (rdev->saved_raid_disk >= 0 &&
8170 !test_bit(Bitmap_sync, &rdev->flags)))
8173 if (rdev->saved_raid_disk < 0)
8174 rdev->recovery_offset = 0;
8176 hot_add_disk(mddev, rdev) == 0) {
8177 if (sysfs_link_rdev(mddev, rdev))
8178 /* failure here is OK */;
8180 md_new_event(mddev);
8181 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8186 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8190 static void md_start_sync(struct work_struct *ws)
8192 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8195 if (mddev_is_clustered(mddev)) {
8196 ret = md_cluster_ops->resync_start(mddev);
8198 mddev->sync_thread = NULL;
8203 mddev->sync_thread = md_register_thread(md_do_sync,
8207 if (!mddev->sync_thread) {
8208 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8209 printk(KERN_ERR "%s: could not start resync"
8212 /* leave the spares where they are, it shouldn't hurt */
8213 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8214 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8215 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8216 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8217 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8218 wake_up(&resync_wait);
8219 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8221 if (mddev->sysfs_action)
8222 sysfs_notify_dirent_safe(mddev->sysfs_action);
8224 md_wakeup_thread(mddev->sync_thread);
8225 sysfs_notify_dirent_safe(mddev->sysfs_action);
8226 md_new_event(mddev);
8230 * This routine is regularly called by all per-raid-array threads to
8231 * deal with generic issues like resync and super-block update.
8232 * Raid personalities that don't have a thread (linear/raid0) do not
8233 * need this as they never do any recovery or update the superblock.
8235 * It does not do any resync itself, but rather "forks" off other threads
8236 * to do that as needed.
8237 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8238 * "->recovery" and create a thread at ->sync_thread.
8239 * When the thread finishes it sets MD_RECOVERY_DONE
8240 * and wakeups up this thread which will reap the thread and finish up.
8241 * This thread also removes any faulty devices (with nr_pending == 0).
8243 * The overall approach is:
8244 * 1/ if the superblock needs updating, update it.
8245 * 2/ If a recovery thread is running, don't do anything else.
8246 * 3/ If recovery has finished, clean up, possibly marking spares active.
8247 * 4/ If there are any faulty devices, remove them.
8248 * 5/ If array is degraded, try to add spares devices
8249 * 6/ If array has spares or is not in-sync, start a resync thread.
8251 void md_check_recovery(struct mddev *mddev)
8253 if (mddev->suspended)
8257 bitmap_daemon_work(mddev);
8259 if (signal_pending(current)) {
8260 if (mddev->pers->sync_request && !mddev->external) {
8261 printk(KERN_INFO "md: %s in immediate safe mode\n",
8263 mddev->safemode = 2;
8265 flush_signals(current);
8268 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8271 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8272 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8273 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8274 (mddev->external == 0 && mddev->safemode == 1) ||
8275 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8276 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8280 if (mddev_trylock(mddev)) {
8284 struct md_rdev *rdev;
8285 if (!mddev->external && mddev->in_sync)
8286 /* 'Blocked' flag not needed as failed devices
8287 * will be recorded if array switched to read/write.
8288 * Leaving it set will prevent the device
8289 * from being removed.
8291 rdev_for_each(rdev, mddev)
8292 clear_bit(Blocked, &rdev->flags);
8293 /* On a read-only array we can:
8294 * - remove failed devices
8295 * - add already-in_sync devices if the array itself
8297 * As we only add devices that are already in-sync,
8298 * we can activate the spares immediately.
8300 remove_and_add_spares(mddev, NULL);
8301 /* There is no thread, but we need to call
8302 * ->spare_active and clear saved_raid_disk
8304 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8305 md_reap_sync_thread(mddev);
8306 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8307 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8308 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8312 if (!mddev->external) {
8314 spin_lock(&mddev->lock);
8315 if (mddev->safemode &&
8316 !atomic_read(&mddev->writes_pending) &&
8318 mddev->recovery_cp == MaxSector) {
8321 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8323 if (mddev->safemode == 1)
8324 mddev->safemode = 0;
8325 spin_unlock(&mddev->lock);
8327 sysfs_notify_dirent_safe(mddev->sysfs_state);
8330 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8331 md_update_sb(mddev, 0);
8333 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8334 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8335 /* resync/recovery still happening */
8336 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8339 if (mddev->sync_thread) {
8340 md_reap_sync_thread(mddev);
8343 /* Set RUNNING before clearing NEEDED to avoid
8344 * any transients in the value of "sync_action".
8346 mddev->curr_resync_completed = 0;
8347 spin_lock(&mddev->lock);
8348 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8349 spin_unlock(&mddev->lock);
8350 /* Clear some bits that don't mean anything, but
8353 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8354 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8356 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8357 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8359 /* no recovery is running.
8360 * remove any failed drives, then
8361 * add spares if possible.
8362 * Spares are also removed and re-added, to allow
8363 * the personality to fail the re-add.
8366 if (mddev->reshape_position != MaxSector) {
8367 if (mddev->pers->check_reshape == NULL ||
8368 mddev->pers->check_reshape(mddev) != 0)
8369 /* Cannot proceed */
8371 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8372 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8373 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8374 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8375 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8376 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8377 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8378 } else if (mddev->recovery_cp < MaxSector) {
8379 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8380 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8381 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8382 /* nothing to be done ... */
8385 if (mddev->pers->sync_request) {
8387 /* We are adding a device or devices to an array
8388 * which has the bitmap stored on all devices.
8389 * So make sure all bitmap pages get written
8391 bitmap_write_all(mddev->bitmap);
8393 INIT_WORK(&mddev->del_work, md_start_sync);
8394 queue_work(md_misc_wq, &mddev->del_work);
8398 if (!mddev->sync_thread) {
8399 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8400 wake_up(&resync_wait);
8401 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8403 if (mddev->sysfs_action)
8404 sysfs_notify_dirent_safe(mddev->sysfs_action);
8407 wake_up(&mddev->sb_wait);
8408 mddev_unlock(mddev);
8411 EXPORT_SYMBOL(md_check_recovery);
8413 void md_reap_sync_thread(struct mddev *mddev)
8415 struct md_rdev *rdev;
8417 /* resync has finished, collect result */
8418 md_unregister_thread(&mddev->sync_thread);
8419 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8420 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8422 /* activate any spares */
8423 if (mddev->pers->spare_active(mddev)) {
8424 sysfs_notify(&mddev->kobj, NULL,
8426 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8429 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8430 mddev->pers->finish_reshape)
8431 mddev->pers->finish_reshape(mddev);
8433 /* If array is no-longer degraded, then any saved_raid_disk
8434 * information must be scrapped.
8436 if (!mddev->degraded)
8437 rdev_for_each(rdev, mddev)
8438 rdev->saved_raid_disk = -1;
8440 md_update_sb(mddev, 1);
8441 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8442 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8443 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8444 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8445 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8446 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8447 wake_up(&resync_wait);
8448 /* flag recovery needed just to double check */
8449 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8450 sysfs_notify_dirent_safe(mddev->sysfs_action);
8451 md_new_event(mddev);
8452 if (mddev->event_work.func)
8453 queue_work(md_misc_wq, &mddev->event_work);
8455 EXPORT_SYMBOL(md_reap_sync_thread);
8457 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8459 sysfs_notify_dirent_safe(rdev->sysfs_state);
8460 wait_event_timeout(rdev->blocked_wait,
8461 !test_bit(Blocked, &rdev->flags) &&
8462 !test_bit(BlockedBadBlocks, &rdev->flags),
8463 msecs_to_jiffies(5000));
8464 rdev_dec_pending(rdev, mddev);
8466 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8468 void md_finish_reshape(struct mddev *mddev)
8470 /* called be personality module when reshape completes. */
8471 struct md_rdev *rdev;
8473 rdev_for_each(rdev, mddev) {
8474 if (rdev->data_offset > rdev->new_data_offset)
8475 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8477 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8478 rdev->data_offset = rdev->new_data_offset;
8481 EXPORT_SYMBOL(md_finish_reshape);
8483 /* Bad block management.
8484 * We can record which blocks on each device are 'bad' and so just
8485 * fail those blocks, or that stripe, rather than the whole device.
8486 * Entries in the bad-block table are 64bits wide. This comprises:
8487 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8488 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8489 * A 'shift' can be set so that larger blocks are tracked and
8490 * consequently larger devices can be covered.
8491 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8493 * Locking of the bad-block table uses a seqlock so md_is_badblock
8494 * might need to retry if it is very unlucky.
8495 * We will sometimes want to check for bad blocks in a bi_end_io function,
8496 * so we use the write_seqlock_irq variant.
8498 * When looking for a bad block we specify a range and want to
8499 * know if any block in the range is bad. So we binary-search
8500 * to the last range that starts at-or-before the given endpoint,
8501 * (or "before the sector after the target range")
8502 * then see if it ends after the given start.
8504 * 0 if there are no known bad blocks in the range
8505 * 1 if there are known bad block which are all acknowledged
8506 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8507 * plus the start/length of the first bad section we overlap.
8509 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8510 sector_t *first_bad, int *bad_sectors)
8516 sector_t target = s + sectors;
8519 if (bb->shift > 0) {
8520 /* round the start down, and the end up */
8522 target += (1<<bb->shift) - 1;
8523 target >>= bb->shift;
8524 sectors = target - s;
8526 /* 'target' is now the first block after the bad range */
8529 seq = read_seqbegin(&bb->lock);
8534 /* Binary search between lo and hi for 'target'
8535 * i.e. for the last range that starts before 'target'
8537 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8538 * are known not to be the last range before target.
8539 * VARIANT: hi-lo is the number of possible
8540 * ranges, and decreases until it reaches 1
8542 while (hi - lo > 1) {
8543 int mid = (lo + hi) / 2;
8544 sector_t a = BB_OFFSET(p[mid]);
8546 /* This could still be the one, earlier ranges
8550 /* This and later ranges are definitely out. */
8553 /* 'lo' might be the last that started before target, but 'hi' isn't */
8555 /* need to check all range that end after 's' to see if
8556 * any are unacknowledged.
8559 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8560 if (BB_OFFSET(p[lo]) < target) {
8561 /* starts before the end, and finishes after
8562 * the start, so they must overlap
8564 if (rv != -1 && BB_ACK(p[lo]))
8568 *first_bad = BB_OFFSET(p[lo]);
8569 *bad_sectors = BB_LEN(p[lo]);
8575 if (read_seqretry(&bb->lock, seq))
8580 EXPORT_SYMBOL_GPL(md_is_badblock);
8583 * Add a range of bad blocks to the table.
8584 * This might extend the table, or might contract it
8585 * if two adjacent ranges can be merged.
8586 * We binary-search to find the 'insertion' point, then
8587 * decide how best to handle it.
8589 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8595 unsigned long flags;
8598 /* badblocks are disabled */
8602 /* round the start down, and the end up */
8603 sector_t next = s + sectors;
8605 next += (1<<bb->shift) - 1;
8610 write_seqlock_irqsave(&bb->lock, flags);
8615 /* Find the last range that starts at-or-before 's' */
8616 while (hi - lo > 1) {
8617 int mid = (lo + hi) / 2;
8618 sector_t a = BB_OFFSET(p[mid]);
8624 if (hi > lo && BB_OFFSET(p[lo]) > s)
8628 /* we found a range that might merge with the start
8631 sector_t a = BB_OFFSET(p[lo]);
8632 sector_t e = a + BB_LEN(p[lo]);
8633 int ack = BB_ACK(p[lo]);
8635 /* Yes, we can merge with a previous range */
8636 if (s == a && s + sectors >= e)
8637 /* new range covers old */
8640 ack = ack && acknowledged;
8642 if (e < s + sectors)
8644 if (e - a <= BB_MAX_LEN) {
8645 p[lo] = BB_MAKE(a, e-a, ack);
8648 /* does not all fit in one range,
8649 * make p[lo] maximal
8651 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8652 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8658 if (sectors && hi < bb->count) {
8659 /* 'hi' points to the first range that starts after 's'.
8660 * Maybe we can merge with the start of that range */
8661 sector_t a = BB_OFFSET(p[hi]);
8662 sector_t e = a + BB_LEN(p[hi]);
8663 int ack = BB_ACK(p[hi]);
8664 if (a <= s + sectors) {
8665 /* merging is possible */
8666 if (e <= s + sectors) {
8671 ack = ack && acknowledged;
8674 if (e - a <= BB_MAX_LEN) {
8675 p[hi] = BB_MAKE(a, e-a, ack);
8678 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8686 if (sectors == 0 && hi < bb->count) {
8687 /* we might be able to combine lo and hi */
8688 /* Note: 's' is at the end of 'lo' */
8689 sector_t a = BB_OFFSET(p[hi]);
8690 int lolen = BB_LEN(p[lo]);
8691 int hilen = BB_LEN(p[hi]);
8692 int newlen = lolen + hilen - (s - a);
8693 if (s >= a && newlen < BB_MAX_LEN) {
8694 /* yes, we can combine them */
8695 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8696 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8697 memmove(p + hi, p + hi + 1,
8698 (bb->count - hi - 1) * 8);
8703 /* didn't merge (it all).
8704 * Need to add a range just before 'hi' */
8705 if (bb->count >= MD_MAX_BADBLOCKS) {
8706 /* No room for more */
8710 int this_sectors = sectors;
8711 memmove(p + hi + 1, p + hi,
8712 (bb->count - hi) * 8);
8715 if (this_sectors > BB_MAX_LEN)
8716 this_sectors = BB_MAX_LEN;
8717 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8718 sectors -= this_sectors;
8725 bb->unacked_exist = 1;
8726 write_sequnlock_irqrestore(&bb->lock, flags);
8731 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8736 s += rdev->new_data_offset;
8738 s += rdev->data_offset;
8739 rv = md_set_badblocks(&rdev->badblocks,
8742 /* Make sure they get written out promptly */
8743 sysfs_notify_dirent_safe(rdev->sysfs_state);
8744 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8745 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8746 md_wakeup_thread(rdev->mddev->thread);
8750 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8753 * Remove a range of bad blocks from the table.
8754 * This may involve extending the table if we spilt a region,
8755 * but it must not fail. So if the table becomes full, we just
8756 * drop the remove request.
8758 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8762 sector_t target = s + sectors;
8765 if (bb->shift > 0) {
8766 /* When clearing we round the start up and the end down.
8767 * This should not matter as the shift should align with
8768 * the block size and no rounding should ever be needed.
8769 * However it is better the think a block is bad when it
8770 * isn't than to think a block is not bad when it is.
8772 s += (1<<bb->shift) - 1;
8774 target >>= bb->shift;
8775 sectors = target - s;
8778 write_seqlock_irq(&bb->lock);
8783 /* Find the last range that starts before 'target' */
8784 while (hi - lo > 1) {
8785 int mid = (lo + hi) / 2;
8786 sector_t a = BB_OFFSET(p[mid]);
8793 /* p[lo] is the last range that could overlap the
8794 * current range. Earlier ranges could also overlap,
8795 * but only this one can overlap the end of the range.
8797 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8798 /* Partial overlap, leave the tail of this range */
8799 int ack = BB_ACK(p[lo]);
8800 sector_t a = BB_OFFSET(p[lo]);
8801 sector_t end = a + BB_LEN(p[lo]);
8804 /* we need to split this range */
8805 if (bb->count >= MD_MAX_BADBLOCKS) {
8809 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8811 p[lo] = BB_MAKE(a, s-a, ack);
8814 p[lo] = BB_MAKE(target, end - target, ack);
8815 /* there is no longer an overlap */
8820 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8821 /* This range does overlap */
8822 if (BB_OFFSET(p[lo]) < s) {
8823 /* Keep the early parts of this range. */
8824 int ack = BB_ACK(p[lo]);
8825 sector_t start = BB_OFFSET(p[lo]);
8826 p[lo] = BB_MAKE(start, s - start, ack);
8827 /* now low doesn't overlap, so.. */
8832 /* 'lo' is strictly before, 'hi' is strictly after,
8833 * anything between needs to be discarded
8836 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8837 bb->count -= (hi - lo - 1);
8843 write_sequnlock_irq(&bb->lock);
8847 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8851 s += rdev->new_data_offset;
8853 s += rdev->data_offset;
8854 return md_clear_badblocks(&rdev->badblocks,
8857 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8860 * Acknowledge all bad blocks in a list.
8861 * This only succeeds if ->changed is clear. It is used by
8862 * in-kernel metadata updates
8864 void md_ack_all_badblocks(struct badblocks *bb)
8866 if (bb->page == NULL || bb->changed)
8867 /* no point even trying */
8869 write_seqlock_irq(&bb->lock);
8871 if (bb->changed == 0 && bb->unacked_exist) {
8874 for (i = 0; i < bb->count ; i++) {
8875 if (!BB_ACK(p[i])) {
8876 sector_t start = BB_OFFSET(p[i]);
8877 int len = BB_LEN(p[i]);
8878 p[i] = BB_MAKE(start, len, 1);
8881 bb->unacked_exist = 0;
8883 write_sequnlock_irq(&bb->lock);
8885 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8887 /* sysfs access to bad-blocks list.
8888 * We present two files.
8889 * 'bad-blocks' lists sector numbers and lengths of ranges that
8890 * are recorded as bad. The list is truncated to fit within
8891 * the one-page limit of sysfs.
8892 * Writing "sector length" to this file adds an acknowledged
8894 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8895 * been acknowledged. Writing to this file adds bad blocks
8896 * without acknowledging them. This is largely for testing.
8900 badblocks_show(struct badblocks *bb, char *page, int unack)
8911 seq = read_seqbegin(&bb->lock);
8916 while (len < PAGE_SIZE && i < bb->count) {
8917 sector_t s = BB_OFFSET(p[i]);
8918 unsigned int length = BB_LEN(p[i]);
8919 int ack = BB_ACK(p[i]);
8925 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8926 (unsigned long long)s << bb->shift,
8927 length << bb->shift);
8929 if (unack && len == 0)
8930 bb->unacked_exist = 0;
8932 if (read_seqretry(&bb->lock, seq))
8941 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8943 unsigned long long sector;
8947 /* Allow clearing via sysfs *only* for testing/debugging.
8948 * Normally only a successful write may clear a badblock
8951 if (page[0] == '-') {
8955 #endif /* DO_DEBUG */
8957 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8959 if (newline != '\n')
8971 md_clear_badblocks(bb, sector, length);
8974 #endif /* DO_DEBUG */
8975 if (md_set_badblocks(bb, sector, length, !unack))
8981 static int md_notify_reboot(struct notifier_block *this,
8982 unsigned long code, void *x)
8984 struct list_head *tmp;
8985 struct mddev *mddev;
8988 for_each_mddev(mddev, tmp) {
8989 if (mddev_trylock(mddev)) {
8991 __md_stop_writes(mddev);
8992 if (mddev->persistent)
8993 mddev->safemode = 2;
8994 mddev_unlock(mddev);
8999 * certain more exotic SCSI devices are known to be
9000 * volatile wrt too early system reboots. While the
9001 * right place to handle this issue is the given
9002 * driver, we do want to have a safe RAID driver ...
9010 static struct notifier_block md_notifier = {
9011 .notifier_call = md_notify_reboot,
9013 .priority = INT_MAX, /* before any real devices */
9016 static void md_geninit(void)
9018 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9020 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9023 static int __init md_init(void)
9027 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9031 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9035 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9038 if ((ret = register_blkdev(0, "mdp")) < 0)
9042 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9043 md_probe, NULL, NULL);
9044 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9045 md_probe, NULL, NULL);
9047 register_reboot_notifier(&md_notifier);
9048 raid_table_header = register_sysctl_table(raid_root_table);
9054 unregister_blkdev(MD_MAJOR, "md");
9056 destroy_workqueue(md_misc_wq);
9058 destroy_workqueue(md_wq);
9063 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9065 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9066 struct md_rdev *rdev2;
9068 char b[BDEVNAME_SIZE];
9070 /* Check for change of roles in the active devices */
9071 rdev_for_each(rdev2, mddev) {
9072 if (test_bit(Faulty, &rdev2->flags))
9075 /* Check if the roles changed */
9076 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9078 if (test_bit(Candidate, &rdev2->flags)) {
9079 if (role == 0xfffe) {
9080 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9081 md_kick_rdev_from_array(rdev2);
9085 clear_bit(Candidate, &rdev2->flags);
9088 if (role != rdev2->raid_disk) {
9090 if (rdev2->raid_disk == -1 && role != 0xffff) {
9091 rdev2->saved_raid_disk = role;
9092 ret = remove_and_add_spares(mddev, rdev2);
9093 pr_info("Activated spare: %s\n",
9094 bdevname(rdev2->bdev,b));
9098 * We just want to do the minimum to mark the disk
9099 * as faulty. The recovery is performed by the
9100 * one who initiated the error.
9102 if ((role == 0xfffe) || (role == 0xfffd)) {
9103 md_error(mddev, rdev2);
9104 clear_bit(Blocked, &rdev2->flags);
9109 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9110 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9112 /* Finally set the event to be up to date */
9113 mddev->events = le64_to_cpu(sb->events);
9116 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9119 struct page *swapout = rdev->sb_page;
9120 struct mdp_superblock_1 *sb;
9122 /* Store the sb page of the rdev in the swapout temporary
9123 * variable in case we err in the future
9125 rdev->sb_page = NULL;
9126 alloc_disk_sb(rdev);
9127 ClearPageUptodate(rdev->sb_page);
9128 rdev->sb_loaded = 0;
9129 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9132 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9133 __func__, __LINE__, rdev->desc_nr, err);
9134 put_page(rdev->sb_page);
9135 rdev->sb_page = swapout;
9136 rdev->sb_loaded = 1;
9140 sb = page_address(rdev->sb_page);
9141 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9145 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9146 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9148 /* The other node finished recovery, call spare_active to set
9149 * device In_sync and mddev->degraded
9151 if (rdev->recovery_offset == MaxSector &&
9152 !test_bit(In_sync, &rdev->flags) &&
9153 mddev->pers->spare_active(mddev))
9154 sysfs_notify(&mddev->kobj, NULL, "degraded");
9160 void md_reload_sb(struct mddev *mddev, int nr)
9162 struct md_rdev *rdev;
9166 rdev_for_each_rcu(rdev, mddev) {
9167 if (rdev->desc_nr == nr)
9171 if (!rdev || rdev->desc_nr != nr) {
9172 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9176 err = read_rdev(mddev, rdev);
9180 check_sb_changes(mddev, rdev);
9182 /* Read all rdev's to update recovery_offset */
9183 rdev_for_each_rcu(rdev, mddev)
9184 read_rdev(mddev, rdev);
9186 EXPORT_SYMBOL(md_reload_sb);
9191 * Searches all registered partitions for autorun RAID arrays
9195 static LIST_HEAD(all_detected_devices);
9196 struct detected_devices_node {
9197 struct list_head list;
9201 void md_autodetect_dev(dev_t dev)
9203 struct detected_devices_node *node_detected_dev;
9205 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9206 if (node_detected_dev) {
9207 node_detected_dev->dev = dev;
9208 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9210 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9211 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9215 static void autostart_arrays(int part)
9217 struct md_rdev *rdev;
9218 struct detected_devices_node *node_detected_dev;
9220 int i_scanned, i_passed;
9225 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
9227 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9229 node_detected_dev = list_entry(all_detected_devices.next,
9230 struct detected_devices_node, list);
9231 list_del(&node_detected_dev->list);
9232 dev = node_detected_dev->dev;
9233 kfree(node_detected_dev);
9234 rdev = md_import_device(dev,0, 90);
9238 if (test_bit(Faulty, &rdev->flags))
9241 set_bit(AutoDetected, &rdev->flags);
9242 list_add(&rdev->same_set, &pending_raid_disks);
9246 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9247 i_scanned, i_passed);
9249 autorun_devices(part);
9252 #endif /* !MODULE */
9254 static __exit void md_exit(void)
9256 struct mddev *mddev;
9257 struct list_head *tmp;
9260 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9261 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9263 unregister_blkdev(MD_MAJOR,"md");
9264 unregister_blkdev(mdp_major, "mdp");
9265 unregister_reboot_notifier(&md_notifier);
9266 unregister_sysctl_table(raid_table_header);
9268 /* We cannot unload the modules while some process is
9269 * waiting for us in select() or poll() - wake them up
9272 while (waitqueue_active(&md_event_waiters)) {
9273 /* not safe to leave yet */
9274 wake_up(&md_event_waiters);
9278 remove_proc_entry("mdstat", NULL);
9280 for_each_mddev(mddev, tmp) {
9281 export_array(mddev);
9282 mddev->hold_active = 0;
9284 destroy_workqueue(md_misc_wq);
9285 destroy_workqueue(md_wq);
9288 subsys_initcall(md_init);
9289 module_exit(md_exit)
9291 static int get_ro(char *buffer, struct kernel_param *kp)
9293 return sprintf(buffer, "%d", start_readonly);
9295 static int set_ro(const char *val, struct kernel_param *kp)
9297 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9300 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9301 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9302 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9304 MODULE_LICENSE("GPL");
9305 MODULE_DESCRIPTION("MD RAID framework");
9307 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);