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]) < 0xfffe)
1612 if (ev1 < mddev->events)
1614 } else if (mddev->bitmap) {
1615 /* If adding to array with a bitmap, then we can accept an
1616 * older device, but not too old.
1618 if (ev1 < mddev->bitmap->events_cleared)
1620 if (ev1 < mddev->events)
1621 set_bit(Bitmap_sync, &rdev->flags);
1623 if (ev1 < mddev->events)
1624 /* just a hot-add of a new device, leave raid_disk at -1 */
1627 if (mddev->level != LEVEL_MULTIPATH) {
1629 if (rdev->desc_nr < 0 ||
1630 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1634 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1636 case 0xffff: /* spare */
1638 case 0xfffe: /* faulty */
1639 set_bit(Faulty, &rdev->flags);
1642 rdev->saved_raid_disk = role;
1643 if ((le32_to_cpu(sb->feature_map) &
1644 MD_FEATURE_RECOVERY_OFFSET)) {
1645 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1646 if (!(le32_to_cpu(sb->feature_map) &
1647 MD_FEATURE_RECOVERY_BITMAP))
1648 rdev->saved_raid_disk = -1;
1650 set_bit(In_sync, &rdev->flags);
1651 rdev->raid_disk = role;
1654 if (sb->devflags & WriteMostly1)
1655 set_bit(WriteMostly, &rdev->flags);
1656 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1657 set_bit(Replacement, &rdev->flags);
1658 } else /* MULTIPATH are always insync */
1659 set_bit(In_sync, &rdev->flags);
1664 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1666 struct mdp_superblock_1 *sb;
1667 struct md_rdev *rdev2;
1669 /* make rdev->sb match mddev and rdev data. */
1671 sb = page_address(rdev->sb_page);
1673 sb->feature_map = 0;
1675 sb->recovery_offset = cpu_to_le64(0);
1676 memset(sb->pad3, 0, sizeof(sb->pad3));
1678 sb->utime = cpu_to_le64((__u64)mddev->utime);
1679 sb->events = cpu_to_le64(mddev->events);
1681 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1683 sb->resync_offset = cpu_to_le64(0);
1685 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1687 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1688 sb->size = cpu_to_le64(mddev->dev_sectors);
1689 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1690 sb->level = cpu_to_le32(mddev->level);
1691 sb->layout = cpu_to_le32(mddev->layout);
1693 if (test_bit(WriteMostly, &rdev->flags))
1694 sb->devflags |= WriteMostly1;
1696 sb->devflags &= ~WriteMostly1;
1697 sb->data_offset = cpu_to_le64(rdev->data_offset);
1698 sb->data_size = cpu_to_le64(rdev->sectors);
1700 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1701 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1702 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1705 if (rdev->raid_disk >= 0 &&
1706 !test_bit(In_sync, &rdev->flags)) {
1708 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1709 sb->recovery_offset =
1710 cpu_to_le64(rdev->recovery_offset);
1711 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1713 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1715 if (test_bit(Replacement, &rdev->flags))
1717 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1719 if (mddev->reshape_position != MaxSector) {
1720 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1721 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1722 sb->new_layout = cpu_to_le32(mddev->new_layout);
1723 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1724 sb->new_level = cpu_to_le32(mddev->new_level);
1725 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1726 if (mddev->delta_disks == 0 &&
1727 mddev->reshape_backwards)
1729 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1730 if (rdev->new_data_offset != rdev->data_offset) {
1732 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1733 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1734 - rdev->data_offset));
1738 if (rdev->badblocks.count == 0)
1739 /* Nothing to do for bad blocks*/ ;
1740 else if (sb->bblog_offset == 0)
1741 /* Cannot record bad blocks on this device */
1742 md_error(mddev, rdev);
1744 struct badblocks *bb = &rdev->badblocks;
1745 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1747 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1752 seq = read_seqbegin(&bb->lock);
1754 memset(bbp, 0xff, PAGE_SIZE);
1756 for (i = 0 ; i < bb->count ; i++) {
1757 u64 internal_bb = p[i];
1758 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1759 | BB_LEN(internal_bb));
1760 bbp[i] = cpu_to_le64(store_bb);
1763 if (read_seqretry(&bb->lock, seq))
1766 bb->sector = (rdev->sb_start +
1767 (int)le32_to_cpu(sb->bblog_offset));
1768 bb->size = le16_to_cpu(sb->bblog_size);
1773 rdev_for_each(rdev2, mddev)
1774 if (rdev2->desc_nr+1 > max_dev)
1775 max_dev = rdev2->desc_nr+1;
1777 if (max_dev > le32_to_cpu(sb->max_dev)) {
1779 sb->max_dev = cpu_to_le32(max_dev);
1780 rdev->sb_size = max_dev * 2 + 256;
1781 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1782 if (rdev->sb_size & bmask)
1783 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1785 max_dev = le32_to_cpu(sb->max_dev);
1787 for (i=0; i<max_dev;i++)
1788 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1790 rdev_for_each(rdev2, mddev) {
1792 if (test_bit(Faulty, &rdev2->flags))
1793 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1794 else if (test_bit(In_sync, &rdev2->flags))
1795 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1796 else if (rdev2->raid_disk >= 0)
1797 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1799 sb->dev_roles[i] = cpu_to_le16(0xffff);
1802 sb->sb_csum = calc_sb_1_csum(sb);
1805 static unsigned long long
1806 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1808 struct mdp_superblock_1 *sb;
1809 sector_t max_sectors;
1810 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1811 return 0; /* component must fit device */
1812 if (rdev->data_offset != rdev->new_data_offset)
1813 return 0; /* too confusing */
1814 if (rdev->sb_start < rdev->data_offset) {
1815 /* minor versions 1 and 2; superblock before data */
1816 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1817 max_sectors -= rdev->data_offset;
1818 if (!num_sectors || num_sectors > max_sectors)
1819 num_sectors = max_sectors;
1820 } else if (rdev->mddev->bitmap_info.offset) {
1821 /* minor version 0 with bitmap we can't move */
1824 /* minor version 0; superblock after data */
1826 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1827 sb_start &= ~(sector_t)(4*2 - 1);
1828 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1829 if (!num_sectors || num_sectors > max_sectors)
1830 num_sectors = max_sectors;
1831 rdev->sb_start = sb_start;
1833 sb = page_address(rdev->sb_page);
1834 sb->data_size = cpu_to_le64(num_sectors);
1835 sb->super_offset = rdev->sb_start;
1836 sb->sb_csum = calc_sb_1_csum(sb);
1837 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1839 md_super_wait(rdev->mddev);
1845 super_1_allow_new_offset(struct md_rdev *rdev,
1846 unsigned long long new_offset)
1848 /* All necessary checks on new >= old have been done */
1849 struct bitmap *bitmap;
1850 if (new_offset >= rdev->data_offset)
1853 /* with 1.0 metadata, there is no metadata to tread on
1854 * so we can always move back */
1855 if (rdev->mddev->minor_version == 0)
1858 /* otherwise we must be sure not to step on
1859 * any metadata, so stay:
1860 * 36K beyond start of superblock
1861 * beyond end of badblocks
1862 * beyond write-intent bitmap
1864 if (rdev->sb_start + (32+4)*2 > new_offset)
1866 bitmap = rdev->mddev->bitmap;
1867 if (bitmap && !rdev->mddev->bitmap_info.file &&
1868 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1869 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1871 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1877 static struct super_type super_types[] = {
1880 .owner = THIS_MODULE,
1881 .load_super = super_90_load,
1882 .validate_super = super_90_validate,
1883 .sync_super = super_90_sync,
1884 .rdev_size_change = super_90_rdev_size_change,
1885 .allow_new_offset = super_90_allow_new_offset,
1889 .owner = THIS_MODULE,
1890 .load_super = super_1_load,
1891 .validate_super = super_1_validate,
1892 .sync_super = super_1_sync,
1893 .rdev_size_change = super_1_rdev_size_change,
1894 .allow_new_offset = super_1_allow_new_offset,
1898 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1900 if (mddev->sync_super) {
1901 mddev->sync_super(mddev, rdev);
1905 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1907 super_types[mddev->major_version].sync_super(mddev, rdev);
1910 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1912 struct md_rdev *rdev, *rdev2;
1915 rdev_for_each_rcu(rdev, mddev1)
1916 rdev_for_each_rcu(rdev2, mddev2)
1917 if (rdev->bdev->bd_contains ==
1918 rdev2->bdev->bd_contains) {
1926 static LIST_HEAD(pending_raid_disks);
1929 * Try to register data integrity profile for an mddev
1931 * This is called when an array is started and after a disk has been kicked
1932 * from the array. It only succeeds if all working and active component devices
1933 * are integrity capable with matching profiles.
1935 int md_integrity_register(struct mddev *mddev)
1937 struct md_rdev *rdev, *reference = NULL;
1939 if (list_empty(&mddev->disks))
1940 return 0; /* nothing to do */
1941 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1942 return 0; /* shouldn't register, or already is */
1943 rdev_for_each(rdev, mddev) {
1944 /* skip spares and non-functional disks */
1945 if (test_bit(Faulty, &rdev->flags))
1947 if (rdev->raid_disk < 0)
1950 /* Use the first rdev as the reference */
1954 /* does this rdev's profile match the reference profile? */
1955 if (blk_integrity_compare(reference->bdev->bd_disk,
1956 rdev->bdev->bd_disk) < 0)
1959 if (!reference || !bdev_get_integrity(reference->bdev))
1962 * All component devices are integrity capable and have matching
1963 * profiles, register the common profile for the md device.
1965 if (blk_integrity_register(mddev->gendisk,
1966 bdev_get_integrity(reference->bdev)) != 0) {
1967 printk(KERN_ERR "md: failed to register integrity for %s\n",
1971 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1972 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1973 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1979 EXPORT_SYMBOL(md_integrity_register);
1981 /* Disable data integrity if non-capable/non-matching disk is being added */
1982 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1984 struct blk_integrity *bi_rdev;
1985 struct blk_integrity *bi_mddev;
1987 if (!mddev->gendisk)
1990 bi_rdev = bdev_get_integrity(rdev->bdev);
1991 bi_mddev = blk_get_integrity(mddev->gendisk);
1993 if (!bi_mddev) /* nothing to do */
1995 if (rdev->raid_disk < 0) /* skip spares */
1997 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1998 rdev->bdev->bd_disk) >= 0)
2000 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2001 blk_integrity_unregister(mddev->gendisk);
2003 EXPORT_SYMBOL(md_integrity_add_rdev);
2005 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2007 char b[BDEVNAME_SIZE];
2011 /* prevent duplicates */
2012 if (find_rdev(mddev, rdev->bdev->bd_dev))
2015 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2016 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2017 rdev->sectors < mddev->dev_sectors)) {
2019 /* Cannot change size, so fail
2020 * If mddev->level <= 0, then we don't care
2021 * about aligning sizes (e.g. linear)
2023 if (mddev->level > 0)
2026 mddev->dev_sectors = rdev->sectors;
2029 /* Verify rdev->desc_nr is unique.
2030 * If it is -1, assign a free number, else
2031 * check number is not in use
2034 if (rdev->desc_nr < 0) {
2037 choice = mddev->raid_disks;
2038 while (md_find_rdev_nr_rcu(mddev, choice))
2040 rdev->desc_nr = choice;
2042 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2048 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2049 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2050 mdname(mddev), mddev->max_disks);
2053 bdevname(rdev->bdev,b);
2054 strreplace(b, '/', '!');
2056 rdev->mddev = mddev;
2057 printk(KERN_INFO "md: bind<%s>\n", b);
2059 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2062 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2063 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2064 /* failure here is OK */;
2065 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2067 list_add_rcu(&rdev->same_set, &mddev->disks);
2068 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2070 /* May as well allow recovery to be retried once */
2071 mddev->recovery_disabled++;
2076 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2081 static void md_delayed_delete(struct work_struct *ws)
2083 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2084 kobject_del(&rdev->kobj);
2085 kobject_put(&rdev->kobj);
2088 static void unbind_rdev_from_array(struct md_rdev *rdev)
2090 char b[BDEVNAME_SIZE];
2092 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2093 list_del_rcu(&rdev->same_set);
2094 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2096 sysfs_remove_link(&rdev->kobj, "block");
2097 sysfs_put(rdev->sysfs_state);
2098 rdev->sysfs_state = NULL;
2099 rdev->badblocks.count = 0;
2100 /* We need to delay this, otherwise we can deadlock when
2101 * writing to 'remove' to "dev/state". We also need
2102 * to delay it due to rcu usage.
2105 INIT_WORK(&rdev->del_work, md_delayed_delete);
2106 kobject_get(&rdev->kobj);
2107 queue_work(md_misc_wq, &rdev->del_work);
2111 * prevent the device from being mounted, repartitioned or
2112 * otherwise reused by a RAID array (or any other kernel
2113 * subsystem), by bd_claiming the device.
2115 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2118 struct block_device *bdev;
2119 char b[BDEVNAME_SIZE];
2121 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2122 shared ? (struct md_rdev *)lock_rdev : rdev);
2124 printk(KERN_ERR "md: could not open %s.\n",
2125 __bdevname(dev, b));
2126 return PTR_ERR(bdev);
2132 static void unlock_rdev(struct md_rdev *rdev)
2134 struct block_device *bdev = rdev->bdev;
2136 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2139 void md_autodetect_dev(dev_t dev);
2141 static void export_rdev(struct md_rdev *rdev)
2143 char b[BDEVNAME_SIZE];
2145 printk(KERN_INFO "md: export_rdev(%s)\n",
2146 bdevname(rdev->bdev,b));
2147 md_rdev_clear(rdev);
2149 if (test_bit(AutoDetected, &rdev->flags))
2150 md_autodetect_dev(rdev->bdev->bd_dev);
2153 kobject_put(&rdev->kobj);
2156 void md_kick_rdev_from_array(struct md_rdev *rdev)
2158 unbind_rdev_from_array(rdev);
2161 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2163 static void export_array(struct mddev *mddev)
2165 struct md_rdev *rdev;
2167 while (!list_empty(&mddev->disks)) {
2168 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2170 md_kick_rdev_from_array(rdev);
2172 mddev->raid_disks = 0;
2173 mddev->major_version = 0;
2176 static void sync_sbs(struct mddev *mddev, int nospares)
2178 /* Update each superblock (in-memory image), but
2179 * if we are allowed to, skip spares which already
2180 * have the right event counter, or have one earlier
2181 * (which would mean they aren't being marked as dirty
2182 * with the rest of the array)
2184 struct md_rdev *rdev;
2185 rdev_for_each(rdev, mddev) {
2186 if (rdev->sb_events == mddev->events ||
2188 rdev->raid_disk < 0 &&
2189 rdev->sb_events+1 == mddev->events)) {
2190 /* Don't update this superblock */
2191 rdev->sb_loaded = 2;
2193 sync_super(mddev, rdev);
2194 rdev->sb_loaded = 1;
2199 void md_update_sb(struct mddev *mddev, int force_change)
2201 struct md_rdev *rdev;
2204 int any_badblocks_changed = 0;
2208 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2212 /* First make sure individual recovery_offsets are correct */
2213 rdev_for_each(rdev, mddev) {
2214 if (rdev->raid_disk >= 0 &&
2215 mddev->delta_disks >= 0 &&
2216 !test_bit(In_sync, &rdev->flags) &&
2217 mddev->curr_resync_completed > rdev->recovery_offset)
2218 rdev->recovery_offset = mddev->curr_resync_completed;
2221 if (!mddev->persistent) {
2222 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2223 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2224 if (!mddev->external) {
2225 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2226 rdev_for_each(rdev, mddev) {
2227 if (rdev->badblocks.changed) {
2228 rdev->badblocks.changed = 0;
2229 md_ack_all_badblocks(&rdev->badblocks);
2230 md_error(mddev, rdev);
2232 clear_bit(Blocked, &rdev->flags);
2233 clear_bit(BlockedBadBlocks, &rdev->flags);
2234 wake_up(&rdev->blocked_wait);
2237 wake_up(&mddev->sb_wait);
2241 spin_lock(&mddev->lock);
2243 mddev->utime = get_seconds();
2245 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2247 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2248 /* just a clean<-> dirty transition, possibly leave spares alone,
2249 * though if events isn't the right even/odd, we will have to do
2255 if (mddev->degraded)
2256 /* If the array is degraded, then skipping spares is both
2257 * dangerous and fairly pointless.
2258 * Dangerous because a device that was removed from the array
2259 * might have a event_count that still looks up-to-date,
2260 * so it can be re-added without a resync.
2261 * Pointless because if there are any spares to skip,
2262 * then a recovery will happen and soon that array won't
2263 * be degraded any more and the spare can go back to sleep then.
2267 sync_req = mddev->in_sync;
2269 /* If this is just a dirty<->clean transition, and the array is clean
2270 * and 'events' is odd, we can roll back to the previous clean state */
2272 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2273 && mddev->can_decrease_events
2274 && mddev->events != 1) {
2276 mddev->can_decrease_events = 0;
2278 /* otherwise we have to go forward and ... */
2280 mddev->can_decrease_events = nospares;
2284 * This 64-bit counter should never wrap.
2285 * Either we are in around ~1 trillion A.C., assuming
2286 * 1 reboot per second, or we have a bug...
2288 WARN_ON(mddev->events == 0);
2290 rdev_for_each(rdev, mddev) {
2291 if (rdev->badblocks.changed)
2292 any_badblocks_changed++;
2293 if (test_bit(Faulty, &rdev->flags))
2294 set_bit(FaultRecorded, &rdev->flags);
2297 sync_sbs(mddev, nospares);
2298 spin_unlock(&mddev->lock);
2300 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2301 mdname(mddev), mddev->in_sync);
2303 bitmap_update_sb(mddev->bitmap);
2304 rdev_for_each(rdev, mddev) {
2305 char b[BDEVNAME_SIZE];
2307 if (rdev->sb_loaded != 1)
2308 continue; /* no noise on spare devices */
2310 if (!test_bit(Faulty, &rdev->flags)) {
2311 md_super_write(mddev,rdev,
2312 rdev->sb_start, rdev->sb_size,
2314 pr_debug("md: (write) %s's sb offset: %llu\n",
2315 bdevname(rdev->bdev, b),
2316 (unsigned long long)rdev->sb_start);
2317 rdev->sb_events = mddev->events;
2318 if (rdev->badblocks.size) {
2319 md_super_write(mddev, rdev,
2320 rdev->badblocks.sector,
2321 rdev->badblocks.size << 9,
2323 rdev->badblocks.size = 0;
2327 pr_debug("md: %s (skipping faulty)\n",
2328 bdevname(rdev->bdev, b));
2330 if (mddev->level == LEVEL_MULTIPATH)
2331 /* only need to write one superblock... */
2334 md_super_wait(mddev);
2335 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2337 spin_lock(&mddev->lock);
2338 if (mddev->in_sync != sync_req ||
2339 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2340 /* have to write it out again */
2341 spin_unlock(&mddev->lock);
2344 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2345 spin_unlock(&mddev->lock);
2346 wake_up(&mddev->sb_wait);
2347 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2348 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2350 rdev_for_each(rdev, mddev) {
2351 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2352 clear_bit(Blocked, &rdev->flags);
2354 if (any_badblocks_changed)
2355 md_ack_all_badblocks(&rdev->badblocks);
2356 clear_bit(BlockedBadBlocks, &rdev->flags);
2357 wake_up(&rdev->blocked_wait);
2360 EXPORT_SYMBOL(md_update_sb);
2362 static int add_bound_rdev(struct md_rdev *rdev)
2364 struct mddev *mddev = rdev->mddev;
2367 if (!mddev->pers->hot_remove_disk) {
2368 /* If there is hot_add_disk but no hot_remove_disk
2369 * then added disks for geometry changes,
2370 * and should be added immediately.
2372 super_types[mddev->major_version].
2373 validate_super(mddev, rdev);
2374 err = mddev->pers->hot_add_disk(mddev, rdev);
2376 unbind_rdev_from_array(rdev);
2381 sysfs_notify_dirent_safe(rdev->sysfs_state);
2383 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2384 if (mddev->degraded)
2385 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2386 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2387 md_new_event(mddev);
2388 md_wakeup_thread(mddev->thread);
2392 /* words written to sysfs files may, or may not, be \n terminated.
2393 * We want to accept with case. For this we use cmd_match.
2395 static int cmd_match(const char *cmd, const char *str)
2397 /* See if cmd, written into a sysfs file, matches
2398 * str. They must either be the same, or cmd can
2399 * have a trailing newline
2401 while (*cmd && *str && *cmd == *str) {
2412 struct rdev_sysfs_entry {
2413 struct attribute attr;
2414 ssize_t (*show)(struct md_rdev *, char *);
2415 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2419 state_show(struct md_rdev *rdev, char *page)
2423 unsigned long flags = ACCESS_ONCE(rdev->flags);
2425 if (test_bit(Faulty, &flags) ||
2426 rdev->badblocks.unacked_exist) {
2427 len+= sprintf(page+len, "%sfaulty",sep);
2430 if (test_bit(In_sync, &flags)) {
2431 len += sprintf(page+len, "%sin_sync",sep);
2434 if (test_bit(WriteMostly, &flags)) {
2435 len += sprintf(page+len, "%swrite_mostly",sep);
2438 if (test_bit(Blocked, &flags) ||
2439 (rdev->badblocks.unacked_exist
2440 && !test_bit(Faulty, &flags))) {
2441 len += sprintf(page+len, "%sblocked", sep);
2444 if (!test_bit(Faulty, &flags) &&
2445 !test_bit(In_sync, &flags)) {
2446 len += sprintf(page+len, "%sspare", sep);
2449 if (test_bit(WriteErrorSeen, &flags)) {
2450 len += sprintf(page+len, "%swrite_error", sep);
2453 if (test_bit(WantReplacement, &flags)) {
2454 len += sprintf(page+len, "%swant_replacement", sep);
2457 if (test_bit(Replacement, &flags)) {
2458 len += sprintf(page+len, "%sreplacement", sep);
2462 return len+sprintf(page+len, "\n");
2466 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2469 * faulty - simulates an error
2470 * remove - disconnects the device
2471 * writemostly - sets write_mostly
2472 * -writemostly - clears write_mostly
2473 * blocked - sets the Blocked flags
2474 * -blocked - clears the Blocked and possibly simulates an error
2475 * insync - sets Insync providing device isn't active
2476 * -insync - clear Insync for a device with a slot assigned,
2477 * so that it gets rebuilt based on bitmap
2478 * write_error - sets WriteErrorSeen
2479 * -write_error - clears WriteErrorSeen
2482 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2483 md_error(rdev->mddev, rdev);
2484 if (test_bit(Faulty, &rdev->flags))
2488 } else if (cmd_match(buf, "remove")) {
2489 if (rdev->raid_disk >= 0)
2492 struct mddev *mddev = rdev->mddev;
2493 if (mddev_is_clustered(mddev))
2494 md_cluster_ops->remove_disk(mddev, rdev);
2495 md_kick_rdev_from_array(rdev);
2496 if (mddev_is_clustered(mddev))
2497 md_cluster_ops->metadata_update_start(mddev);
2499 md_update_sb(mddev, 1);
2500 md_new_event(mddev);
2501 if (mddev_is_clustered(mddev))
2502 md_cluster_ops->metadata_update_finish(mddev);
2505 } else if (cmd_match(buf, "writemostly")) {
2506 set_bit(WriteMostly, &rdev->flags);
2508 } else if (cmd_match(buf, "-writemostly")) {
2509 clear_bit(WriteMostly, &rdev->flags);
2511 } else if (cmd_match(buf, "blocked")) {
2512 set_bit(Blocked, &rdev->flags);
2514 } else if (cmd_match(buf, "-blocked")) {
2515 if (!test_bit(Faulty, &rdev->flags) &&
2516 rdev->badblocks.unacked_exist) {
2517 /* metadata handler doesn't understand badblocks,
2518 * so we need to fail the device
2520 md_error(rdev->mddev, rdev);
2522 clear_bit(Blocked, &rdev->flags);
2523 clear_bit(BlockedBadBlocks, &rdev->flags);
2524 wake_up(&rdev->blocked_wait);
2525 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2526 md_wakeup_thread(rdev->mddev->thread);
2529 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2530 set_bit(In_sync, &rdev->flags);
2532 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2533 if (rdev->mddev->pers == NULL) {
2534 clear_bit(In_sync, &rdev->flags);
2535 rdev->saved_raid_disk = rdev->raid_disk;
2536 rdev->raid_disk = -1;
2539 } else if (cmd_match(buf, "write_error")) {
2540 set_bit(WriteErrorSeen, &rdev->flags);
2542 } else if (cmd_match(buf, "-write_error")) {
2543 clear_bit(WriteErrorSeen, &rdev->flags);
2545 } else if (cmd_match(buf, "want_replacement")) {
2546 /* Any non-spare device that is not a replacement can
2547 * become want_replacement at any time, but we then need to
2548 * check if recovery is needed.
2550 if (rdev->raid_disk >= 0 &&
2551 !test_bit(Replacement, &rdev->flags))
2552 set_bit(WantReplacement, &rdev->flags);
2553 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2554 md_wakeup_thread(rdev->mddev->thread);
2556 } else if (cmd_match(buf, "-want_replacement")) {
2557 /* Clearing 'want_replacement' is always allowed.
2558 * Once replacements starts it is too late though.
2561 clear_bit(WantReplacement, &rdev->flags);
2562 } else if (cmd_match(buf, "replacement")) {
2563 /* Can only set a device as a replacement when array has not
2564 * yet been started. Once running, replacement is automatic
2565 * from spares, or by assigning 'slot'.
2567 if (rdev->mddev->pers)
2570 set_bit(Replacement, &rdev->flags);
2573 } else if (cmd_match(buf, "-replacement")) {
2574 /* Similarly, can only clear Replacement before start */
2575 if (rdev->mddev->pers)
2578 clear_bit(Replacement, &rdev->flags);
2581 } else if (cmd_match(buf, "re-add")) {
2582 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2583 /* clear_bit is performed _after_ all the devices
2584 * have their local Faulty bit cleared. If any writes
2585 * happen in the meantime in the local node, they
2586 * will land in the local bitmap, which will be synced
2587 * by this node eventually
2589 if (!mddev_is_clustered(rdev->mddev) ||
2590 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2591 clear_bit(Faulty, &rdev->flags);
2592 err = add_bound_rdev(rdev);
2598 sysfs_notify_dirent_safe(rdev->sysfs_state);
2599 return err ? err : len;
2601 static struct rdev_sysfs_entry rdev_state =
2602 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2605 errors_show(struct md_rdev *rdev, char *page)
2607 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2611 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2616 rv = kstrtouint(buf, 10, &n);
2619 atomic_set(&rdev->corrected_errors, n);
2622 static struct rdev_sysfs_entry rdev_errors =
2623 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2626 slot_show(struct md_rdev *rdev, char *page)
2628 if (rdev->raid_disk < 0)
2629 return sprintf(page, "none\n");
2631 return sprintf(page, "%d\n", rdev->raid_disk);
2635 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2640 if (strncmp(buf, "none", 4)==0)
2643 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2647 if (rdev->mddev->pers && slot == -1) {
2648 /* Setting 'slot' on an active array requires also
2649 * updating the 'rd%d' link, and communicating
2650 * with the personality with ->hot_*_disk.
2651 * For now we only support removing
2652 * failed/spare devices. This normally happens automatically,
2653 * but not when the metadata is externally managed.
2655 if (rdev->raid_disk == -1)
2657 /* personality does all needed checks */
2658 if (rdev->mddev->pers->hot_remove_disk == NULL)
2660 clear_bit(Blocked, &rdev->flags);
2661 remove_and_add_spares(rdev->mddev, rdev);
2662 if (rdev->raid_disk >= 0)
2664 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2665 md_wakeup_thread(rdev->mddev->thread);
2666 } else if (rdev->mddev->pers) {
2667 /* Activating a spare .. or possibly reactivating
2668 * if we ever get bitmaps working here.
2671 if (rdev->raid_disk != -1)
2674 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2677 if (rdev->mddev->pers->hot_add_disk == NULL)
2680 if (slot >= rdev->mddev->raid_disks &&
2681 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2684 rdev->raid_disk = slot;
2685 if (test_bit(In_sync, &rdev->flags))
2686 rdev->saved_raid_disk = slot;
2688 rdev->saved_raid_disk = -1;
2689 clear_bit(In_sync, &rdev->flags);
2690 clear_bit(Bitmap_sync, &rdev->flags);
2691 err = rdev->mddev->pers->
2692 hot_add_disk(rdev->mddev, rdev);
2694 rdev->raid_disk = -1;
2697 sysfs_notify_dirent_safe(rdev->sysfs_state);
2698 if (sysfs_link_rdev(rdev->mddev, rdev))
2699 /* failure here is OK */;
2700 /* don't wakeup anyone, leave that to userspace. */
2702 if (slot >= rdev->mddev->raid_disks &&
2703 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2705 rdev->raid_disk = slot;
2706 /* assume it is working */
2707 clear_bit(Faulty, &rdev->flags);
2708 clear_bit(WriteMostly, &rdev->flags);
2709 set_bit(In_sync, &rdev->flags);
2710 sysfs_notify_dirent_safe(rdev->sysfs_state);
2715 static struct rdev_sysfs_entry rdev_slot =
2716 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2719 offset_show(struct md_rdev *rdev, char *page)
2721 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2725 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2727 unsigned long long offset;
2728 if (kstrtoull(buf, 10, &offset) < 0)
2730 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2732 if (rdev->sectors && rdev->mddev->external)
2733 /* Must set offset before size, so overlap checks
2736 rdev->data_offset = offset;
2737 rdev->new_data_offset = offset;
2741 static struct rdev_sysfs_entry rdev_offset =
2742 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2744 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2746 return sprintf(page, "%llu\n",
2747 (unsigned long long)rdev->new_data_offset);
2750 static ssize_t new_offset_store(struct md_rdev *rdev,
2751 const char *buf, size_t len)
2753 unsigned long long new_offset;
2754 struct mddev *mddev = rdev->mddev;
2756 if (kstrtoull(buf, 10, &new_offset) < 0)
2759 if (mddev->sync_thread ||
2760 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2762 if (new_offset == rdev->data_offset)
2763 /* reset is always permitted */
2765 else if (new_offset > rdev->data_offset) {
2766 /* must not push array size beyond rdev_sectors */
2767 if (new_offset - rdev->data_offset
2768 + mddev->dev_sectors > rdev->sectors)
2771 /* Metadata worries about other space details. */
2773 /* decreasing the offset is inconsistent with a backwards
2776 if (new_offset < rdev->data_offset &&
2777 mddev->reshape_backwards)
2779 /* Increasing offset is inconsistent with forwards
2780 * reshape. reshape_direction should be set to
2781 * 'backwards' first.
2783 if (new_offset > rdev->data_offset &&
2784 !mddev->reshape_backwards)
2787 if (mddev->pers && mddev->persistent &&
2788 !super_types[mddev->major_version]
2789 .allow_new_offset(rdev, new_offset))
2791 rdev->new_data_offset = new_offset;
2792 if (new_offset > rdev->data_offset)
2793 mddev->reshape_backwards = 1;
2794 else if (new_offset < rdev->data_offset)
2795 mddev->reshape_backwards = 0;
2799 static struct rdev_sysfs_entry rdev_new_offset =
2800 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2803 rdev_size_show(struct md_rdev *rdev, char *page)
2805 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2808 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2810 /* check if two start/length pairs overlap */
2818 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2820 unsigned long long blocks;
2823 if (kstrtoull(buf, 10, &blocks) < 0)
2826 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2827 return -EINVAL; /* sector conversion overflow */
2830 if (new != blocks * 2)
2831 return -EINVAL; /* unsigned long long to sector_t overflow */
2838 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2840 struct mddev *my_mddev = rdev->mddev;
2841 sector_t oldsectors = rdev->sectors;
2844 if (strict_blocks_to_sectors(buf, §ors) < 0)
2846 if (rdev->data_offset != rdev->new_data_offset)
2847 return -EINVAL; /* too confusing */
2848 if (my_mddev->pers && rdev->raid_disk >= 0) {
2849 if (my_mddev->persistent) {
2850 sectors = super_types[my_mddev->major_version].
2851 rdev_size_change(rdev, sectors);
2854 } else if (!sectors)
2855 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2857 if (!my_mddev->pers->resize)
2858 /* Cannot change size for RAID0 or Linear etc */
2861 if (sectors < my_mddev->dev_sectors)
2862 return -EINVAL; /* component must fit device */
2864 rdev->sectors = sectors;
2865 if (sectors > oldsectors && my_mddev->external) {
2866 /* Need to check that all other rdevs with the same
2867 * ->bdev do not overlap. 'rcu' is sufficient to walk
2868 * the rdev lists safely.
2869 * This check does not provide a hard guarantee, it
2870 * just helps avoid dangerous mistakes.
2872 struct mddev *mddev;
2874 struct list_head *tmp;
2877 for_each_mddev(mddev, tmp) {
2878 struct md_rdev *rdev2;
2880 rdev_for_each(rdev2, mddev)
2881 if (rdev->bdev == rdev2->bdev &&
2883 overlaps(rdev->data_offset, rdev->sectors,
2896 /* Someone else could have slipped in a size
2897 * change here, but doing so is just silly.
2898 * We put oldsectors back because we *know* it is
2899 * safe, and trust userspace not to race with
2902 rdev->sectors = oldsectors;
2909 static struct rdev_sysfs_entry rdev_size =
2910 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2912 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2914 unsigned long long recovery_start = rdev->recovery_offset;
2916 if (test_bit(In_sync, &rdev->flags) ||
2917 recovery_start == MaxSector)
2918 return sprintf(page, "none\n");
2920 return sprintf(page, "%llu\n", recovery_start);
2923 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2925 unsigned long long recovery_start;
2927 if (cmd_match(buf, "none"))
2928 recovery_start = MaxSector;
2929 else if (kstrtoull(buf, 10, &recovery_start))
2932 if (rdev->mddev->pers &&
2933 rdev->raid_disk >= 0)
2936 rdev->recovery_offset = recovery_start;
2937 if (recovery_start == MaxSector)
2938 set_bit(In_sync, &rdev->flags);
2940 clear_bit(In_sync, &rdev->flags);
2944 static struct rdev_sysfs_entry rdev_recovery_start =
2945 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2948 badblocks_show(struct badblocks *bb, char *page, int unack);
2950 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2952 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2954 return badblocks_show(&rdev->badblocks, page, 0);
2956 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2958 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2959 /* Maybe that ack was all we needed */
2960 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2961 wake_up(&rdev->blocked_wait);
2964 static struct rdev_sysfs_entry rdev_bad_blocks =
2965 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2967 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2969 return badblocks_show(&rdev->badblocks, page, 1);
2971 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2973 return badblocks_store(&rdev->badblocks, page, len, 1);
2975 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2976 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2978 static struct attribute *rdev_default_attrs[] = {
2983 &rdev_new_offset.attr,
2985 &rdev_recovery_start.attr,
2986 &rdev_bad_blocks.attr,
2987 &rdev_unack_bad_blocks.attr,
2991 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2993 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2994 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3000 return entry->show(rdev, page);
3004 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3005 const char *page, size_t length)
3007 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3008 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3010 struct mddev *mddev = rdev->mddev;
3014 if (!capable(CAP_SYS_ADMIN))
3016 rv = mddev ? mddev_lock(mddev): -EBUSY;
3018 if (rdev->mddev == NULL)
3021 rv = entry->store(rdev, page, length);
3022 mddev_unlock(mddev);
3027 static void rdev_free(struct kobject *ko)
3029 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3032 static const struct sysfs_ops rdev_sysfs_ops = {
3033 .show = rdev_attr_show,
3034 .store = rdev_attr_store,
3036 static struct kobj_type rdev_ktype = {
3037 .release = rdev_free,
3038 .sysfs_ops = &rdev_sysfs_ops,
3039 .default_attrs = rdev_default_attrs,
3042 int md_rdev_init(struct md_rdev *rdev)
3045 rdev->saved_raid_disk = -1;
3046 rdev->raid_disk = -1;
3048 rdev->data_offset = 0;
3049 rdev->new_data_offset = 0;
3050 rdev->sb_events = 0;
3051 rdev->last_read_error.tv_sec = 0;
3052 rdev->last_read_error.tv_nsec = 0;
3053 rdev->sb_loaded = 0;
3054 rdev->bb_page = NULL;
3055 atomic_set(&rdev->nr_pending, 0);
3056 atomic_set(&rdev->read_errors, 0);
3057 atomic_set(&rdev->corrected_errors, 0);
3059 INIT_LIST_HEAD(&rdev->same_set);
3060 init_waitqueue_head(&rdev->blocked_wait);
3062 /* Add space to store bad block list.
3063 * This reserves the space even on arrays where it cannot
3064 * be used - I wonder if that matters
3066 rdev->badblocks.count = 0;
3067 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3068 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3069 seqlock_init(&rdev->badblocks.lock);
3070 if (rdev->badblocks.page == NULL)
3075 EXPORT_SYMBOL_GPL(md_rdev_init);
3077 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3079 * mark the device faulty if:
3081 * - the device is nonexistent (zero size)
3082 * - the device has no valid superblock
3084 * a faulty rdev _never_ has rdev->sb set.
3086 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3088 char b[BDEVNAME_SIZE];
3090 struct md_rdev *rdev;
3093 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3095 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3096 return ERR_PTR(-ENOMEM);
3099 err = md_rdev_init(rdev);
3102 err = alloc_disk_sb(rdev);
3106 err = lock_rdev(rdev, newdev, super_format == -2);
3110 kobject_init(&rdev->kobj, &rdev_ktype);
3112 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3115 "md: %s has zero or unknown size, marking faulty!\n",
3116 bdevname(rdev->bdev,b));
3121 if (super_format >= 0) {
3122 err = super_types[super_format].
3123 load_super(rdev, NULL, super_minor);
3124 if (err == -EINVAL) {
3126 "md: %s does not have a valid v%d.%d "
3127 "superblock, not importing!\n",
3128 bdevname(rdev->bdev,b),
3129 super_format, super_minor);
3134 "md: could not read %s's sb, not importing!\n",
3135 bdevname(rdev->bdev,b));
3145 md_rdev_clear(rdev);
3147 return ERR_PTR(err);
3151 * Check a full RAID array for plausibility
3154 static void analyze_sbs(struct mddev *mddev)
3157 struct md_rdev *rdev, *freshest, *tmp;
3158 char b[BDEVNAME_SIZE];
3161 rdev_for_each_safe(rdev, tmp, mddev)
3162 switch (super_types[mddev->major_version].
3163 load_super(rdev, freshest, mddev->minor_version)) {
3171 "md: fatal superblock inconsistency in %s"
3172 " -- removing from array\n",
3173 bdevname(rdev->bdev,b));
3174 md_kick_rdev_from_array(rdev);
3177 super_types[mddev->major_version].
3178 validate_super(mddev, freshest);
3181 rdev_for_each_safe(rdev, tmp, mddev) {
3182 if (mddev->max_disks &&
3183 (rdev->desc_nr >= mddev->max_disks ||
3184 i > mddev->max_disks)) {
3186 "md: %s: %s: only %d devices permitted\n",
3187 mdname(mddev), bdevname(rdev->bdev, b),
3189 md_kick_rdev_from_array(rdev);
3192 if (rdev != freshest) {
3193 if (super_types[mddev->major_version].
3194 validate_super(mddev, rdev)) {
3195 printk(KERN_WARNING "md: kicking non-fresh %s"
3197 bdevname(rdev->bdev,b));
3198 md_kick_rdev_from_array(rdev);
3201 /* No device should have a Candidate flag
3202 * when reading devices
3204 if (test_bit(Candidate, &rdev->flags)) {
3205 pr_info("md: kicking Cluster Candidate %s from array!\n",
3206 bdevname(rdev->bdev, b));
3207 md_kick_rdev_from_array(rdev);
3210 if (mddev->level == LEVEL_MULTIPATH) {
3211 rdev->desc_nr = i++;
3212 rdev->raid_disk = rdev->desc_nr;
3213 set_bit(In_sync, &rdev->flags);
3214 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3215 rdev->raid_disk = -1;
3216 clear_bit(In_sync, &rdev->flags);
3221 /* Read a fixed-point number.
3222 * Numbers in sysfs attributes should be in "standard" units where
3223 * possible, so time should be in seconds.
3224 * However we internally use a a much smaller unit such as
3225 * milliseconds or jiffies.
3226 * This function takes a decimal number with a possible fractional
3227 * component, and produces an integer which is the result of
3228 * multiplying that number by 10^'scale'.
3229 * all without any floating-point arithmetic.
3231 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3233 unsigned long result = 0;
3235 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3238 else if (decimals < scale) {
3241 result = result * 10 + value;
3253 while (decimals < scale) {
3262 safe_delay_show(struct mddev *mddev, char *page)
3264 int msec = (mddev->safemode_delay*1000)/HZ;
3265 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3268 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3272 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3275 mddev->safemode_delay = 0;
3277 unsigned long old_delay = mddev->safemode_delay;
3278 unsigned long new_delay = (msec*HZ)/1000;
3282 mddev->safemode_delay = new_delay;
3283 if (new_delay < old_delay || old_delay == 0)
3284 mod_timer(&mddev->safemode_timer, jiffies+1);
3288 static struct md_sysfs_entry md_safe_delay =
3289 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3292 level_show(struct mddev *mddev, char *page)
3294 struct md_personality *p;
3296 spin_lock(&mddev->lock);
3299 ret = sprintf(page, "%s\n", p->name);
3300 else if (mddev->clevel[0])
3301 ret = sprintf(page, "%s\n", mddev->clevel);
3302 else if (mddev->level != LEVEL_NONE)
3303 ret = sprintf(page, "%d\n", mddev->level);
3306 spin_unlock(&mddev->lock);
3311 level_store(struct mddev *mddev, const char *buf, size_t len)
3316 struct md_personality *pers, *oldpers;
3318 void *priv, *oldpriv;
3319 struct md_rdev *rdev;
3321 if (slen == 0 || slen >= sizeof(clevel))
3324 rv = mddev_lock(mddev);
3328 if (mddev->pers == NULL) {
3329 strncpy(mddev->clevel, buf, slen);
3330 if (mddev->clevel[slen-1] == '\n')
3332 mddev->clevel[slen] = 0;
3333 mddev->level = LEVEL_NONE;
3341 /* request to change the personality. Need to ensure:
3342 * - array is not engaged in resync/recovery/reshape
3343 * - old personality can be suspended
3344 * - new personality will access other array.
3348 if (mddev->sync_thread ||
3349 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3350 mddev->reshape_position != MaxSector ||
3351 mddev->sysfs_active)
3355 if (!mddev->pers->quiesce) {
3356 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3357 mdname(mddev), mddev->pers->name);
3361 /* Now find the new personality */
3362 strncpy(clevel, buf, slen);
3363 if (clevel[slen-1] == '\n')
3366 if (kstrtol(clevel, 10, &level))
3369 if (request_module("md-%s", clevel) != 0)
3370 request_module("md-level-%s", clevel);
3371 spin_lock(&pers_lock);
3372 pers = find_pers(level, clevel);
3373 if (!pers || !try_module_get(pers->owner)) {
3374 spin_unlock(&pers_lock);
3375 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3379 spin_unlock(&pers_lock);
3381 if (pers == mddev->pers) {
3382 /* Nothing to do! */
3383 module_put(pers->owner);
3387 if (!pers->takeover) {
3388 module_put(pers->owner);
3389 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3390 mdname(mddev), clevel);
3395 rdev_for_each(rdev, mddev)
3396 rdev->new_raid_disk = rdev->raid_disk;
3398 /* ->takeover must set new_* and/or delta_disks
3399 * if it succeeds, and may set them when it fails.
3401 priv = pers->takeover(mddev);
3403 mddev->new_level = mddev->level;
3404 mddev->new_layout = mddev->layout;
3405 mddev->new_chunk_sectors = mddev->chunk_sectors;
3406 mddev->raid_disks -= mddev->delta_disks;
3407 mddev->delta_disks = 0;
3408 mddev->reshape_backwards = 0;
3409 module_put(pers->owner);
3410 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3411 mdname(mddev), clevel);
3416 /* Looks like we have a winner */
3417 mddev_suspend(mddev);
3418 mddev_detach(mddev);
3420 spin_lock(&mddev->lock);
3421 oldpers = mddev->pers;
3422 oldpriv = mddev->private;
3424 mddev->private = priv;
3425 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3426 mddev->level = mddev->new_level;
3427 mddev->layout = mddev->new_layout;
3428 mddev->chunk_sectors = mddev->new_chunk_sectors;
3429 mddev->delta_disks = 0;
3430 mddev->reshape_backwards = 0;
3431 mddev->degraded = 0;
3432 spin_unlock(&mddev->lock);
3434 if (oldpers->sync_request == NULL &&
3436 /* We are converting from a no-redundancy array
3437 * to a redundancy array and metadata is managed
3438 * externally so we need to be sure that writes
3439 * won't block due to a need to transition
3441 * until external management is started.
3444 mddev->safemode_delay = 0;
3445 mddev->safemode = 0;
3448 oldpers->free(mddev, oldpriv);
3450 if (oldpers->sync_request == NULL &&
3451 pers->sync_request != NULL) {
3452 /* need to add the md_redundancy_group */
3453 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3455 "md: cannot register extra attributes for %s\n",
3457 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3459 if (oldpers->sync_request != NULL &&
3460 pers->sync_request == NULL) {
3461 /* need to remove the md_redundancy_group */
3462 if (mddev->to_remove == NULL)
3463 mddev->to_remove = &md_redundancy_group;
3466 rdev_for_each(rdev, mddev) {
3467 if (rdev->raid_disk < 0)
3469 if (rdev->new_raid_disk >= mddev->raid_disks)
3470 rdev->new_raid_disk = -1;
3471 if (rdev->new_raid_disk == rdev->raid_disk)
3473 sysfs_unlink_rdev(mddev, rdev);
3475 rdev_for_each(rdev, mddev) {
3476 if (rdev->raid_disk < 0)
3478 if (rdev->new_raid_disk == rdev->raid_disk)
3480 rdev->raid_disk = rdev->new_raid_disk;
3481 if (rdev->raid_disk < 0)
3482 clear_bit(In_sync, &rdev->flags);
3484 if (sysfs_link_rdev(mddev, rdev))
3485 printk(KERN_WARNING "md: cannot register rd%d"
3486 " for %s after level change\n",
3487 rdev->raid_disk, mdname(mddev));
3491 if (pers->sync_request == NULL) {
3492 /* this is now an array without redundancy, so
3493 * it must always be in_sync
3496 del_timer_sync(&mddev->safemode_timer);
3498 blk_set_stacking_limits(&mddev->queue->limits);
3500 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3501 mddev_resume(mddev);
3503 md_update_sb(mddev, 1);
3504 sysfs_notify(&mddev->kobj, NULL, "level");
3505 md_new_event(mddev);
3508 mddev_unlock(mddev);
3512 static struct md_sysfs_entry md_level =
3513 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3516 layout_show(struct mddev *mddev, char *page)
3518 /* just a number, not meaningful for all levels */
3519 if (mddev->reshape_position != MaxSector &&
3520 mddev->layout != mddev->new_layout)
3521 return sprintf(page, "%d (%d)\n",
3522 mddev->new_layout, mddev->layout);
3523 return sprintf(page, "%d\n", mddev->layout);
3527 layout_store(struct mddev *mddev, const char *buf, size_t len)
3532 err = kstrtouint(buf, 10, &n);
3535 err = mddev_lock(mddev);
3540 if (mddev->pers->check_reshape == NULL)
3545 mddev->new_layout = n;
3546 err = mddev->pers->check_reshape(mddev);
3548 mddev->new_layout = mddev->layout;
3551 mddev->new_layout = n;
3552 if (mddev->reshape_position == MaxSector)
3555 mddev_unlock(mddev);
3558 static struct md_sysfs_entry md_layout =
3559 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3562 raid_disks_show(struct mddev *mddev, char *page)
3564 if (mddev->raid_disks == 0)
3566 if (mddev->reshape_position != MaxSector &&
3567 mddev->delta_disks != 0)
3568 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3569 mddev->raid_disks - mddev->delta_disks);
3570 return sprintf(page, "%d\n", mddev->raid_disks);
3573 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3576 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3581 err = kstrtouint(buf, 10, &n);
3585 err = mddev_lock(mddev);
3589 err = update_raid_disks(mddev, n);
3590 else if (mddev->reshape_position != MaxSector) {
3591 struct md_rdev *rdev;
3592 int olddisks = mddev->raid_disks - mddev->delta_disks;
3595 rdev_for_each(rdev, mddev) {
3597 rdev->data_offset < rdev->new_data_offset)
3600 rdev->data_offset > rdev->new_data_offset)
3604 mddev->delta_disks = n - olddisks;
3605 mddev->raid_disks = n;
3606 mddev->reshape_backwards = (mddev->delta_disks < 0);
3608 mddev->raid_disks = n;
3610 mddev_unlock(mddev);
3611 return err ? err : len;
3613 static struct md_sysfs_entry md_raid_disks =
3614 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3617 chunk_size_show(struct mddev *mddev, char *page)
3619 if (mddev->reshape_position != MaxSector &&
3620 mddev->chunk_sectors != mddev->new_chunk_sectors)
3621 return sprintf(page, "%d (%d)\n",
3622 mddev->new_chunk_sectors << 9,
3623 mddev->chunk_sectors << 9);
3624 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3628 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3633 err = kstrtoul(buf, 10, &n);
3637 err = mddev_lock(mddev);
3641 if (mddev->pers->check_reshape == NULL)
3646 mddev->new_chunk_sectors = n >> 9;
3647 err = mddev->pers->check_reshape(mddev);
3649 mddev->new_chunk_sectors = mddev->chunk_sectors;
3652 mddev->new_chunk_sectors = n >> 9;
3653 if (mddev->reshape_position == MaxSector)
3654 mddev->chunk_sectors = n >> 9;
3656 mddev_unlock(mddev);
3659 static struct md_sysfs_entry md_chunk_size =
3660 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3663 resync_start_show(struct mddev *mddev, char *page)
3665 if (mddev->recovery_cp == MaxSector)
3666 return sprintf(page, "none\n");
3667 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3671 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3673 unsigned long long n;
3676 if (cmd_match(buf, "none"))
3679 err = kstrtoull(buf, 10, &n);
3682 if (n != (sector_t)n)
3686 err = mddev_lock(mddev);
3689 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3693 mddev->recovery_cp = n;
3695 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3697 mddev_unlock(mddev);
3700 static struct md_sysfs_entry md_resync_start =
3701 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3702 resync_start_show, resync_start_store);
3705 * The array state can be:
3708 * No devices, no size, no level
3709 * Equivalent to STOP_ARRAY ioctl
3711 * May have some settings, but array is not active
3712 * all IO results in error
3713 * When written, doesn't tear down array, but just stops it
3714 * suspended (not supported yet)
3715 * All IO requests will block. The array can be reconfigured.
3716 * Writing this, if accepted, will block until array is quiescent
3718 * no resync can happen. no superblocks get written.
3719 * write requests fail
3721 * like readonly, but behaves like 'clean' on a write request.
3723 * clean - no pending writes, but otherwise active.
3724 * When written to inactive array, starts without resync
3725 * If a write request arrives then
3726 * if metadata is known, mark 'dirty' and switch to 'active'.
3727 * if not known, block and switch to write-pending
3728 * If written to an active array that has pending writes, then fails.
3730 * fully active: IO and resync can be happening.
3731 * When written to inactive array, starts with resync
3734 * clean, but writes are blocked waiting for 'active' to be written.
3737 * like active, but no writes have been seen for a while (100msec).
3740 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3741 write_pending, active_idle, bad_word};
3742 static char *array_states[] = {
3743 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3744 "write-pending", "active-idle", NULL };
3746 static int match_word(const char *word, char **list)
3749 for (n=0; list[n]; n++)
3750 if (cmd_match(word, list[n]))
3756 array_state_show(struct mddev *mddev, char *page)
3758 enum array_state st = inactive;
3771 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3773 else if (mddev->safemode)
3779 if (list_empty(&mddev->disks) &&
3780 mddev->raid_disks == 0 &&
3781 mddev->dev_sectors == 0)
3786 return sprintf(page, "%s\n", array_states[st]);
3789 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3790 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3791 static int do_md_run(struct mddev *mddev);
3792 static int restart_array(struct mddev *mddev);
3795 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3798 enum array_state st = match_word(buf, array_states);
3800 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3801 /* don't take reconfig_mutex when toggling between
3804 spin_lock(&mddev->lock);
3806 restart_array(mddev);
3807 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3808 wake_up(&mddev->sb_wait);
3810 } else /* st == clean */ {
3811 restart_array(mddev);
3812 if (atomic_read(&mddev->writes_pending) == 0) {
3813 if (mddev->in_sync == 0) {
3815 if (mddev->safemode == 1)
3816 mddev->safemode = 0;
3817 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3823 spin_unlock(&mddev->lock);
3826 err = mddev_lock(mddev);
3834 /* stopping an active array */
3835 err = do_md_stop(mddev, 0, NULL);
3838 /* stopping an active array */
3840 err = do_md_stop(mddev, 2, NULL);
3842 err = 0; /* already inactive */
3845 break; /* not supported yet */
3848 err = md_set_readonly(mddev, NULL);
3851 set_disk_ro(mddev->gendisk, 1);
3852 err = do_md_run(mddev);
3858 err = md_set_readonly(mddev, NULL);
3859 else if (mddev->ro == 1)
3860 err = restart_array(mddev);
3863 set_disk_ro(mddev->gendisk, 0);
3867 err = do_md_run(mddev);
3872 restart_array(mddev);
3873 spin_lock(&mddev->lock);
3874 if (atomic_read(&mddev->writes_pending) == 0) {
3875 if (mddev->in_sync == 0) {
3877 if (mddev->safemode == 1)
3878 mddev->safemode = 0;
3879 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3884 spin_unlock(&mddev->lock);
3890 restart_array(mddev);
3891 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3892 wake_up(&mddev->sb_wait);
3896 set_disk_ro(mddev->gendisk, 0);
3897 err = do_md_run(mddev);
3902 /* these cannot be set */
3907 if (mddev->hold_active == UNTIL_IOCTL)
3908 mddev->hold_active = 0;
3909 sysfs_notify_dirent_safe(mddev->sysfs_state);
3911 mddev_unlock(mddev);
3914 static struct md_sysfs_entry md_array_state =
3915 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3918 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3919 return sprintf(page, "%d\n",
3920 atomic_read(&mddev->max_corr_read_errors));
3924 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3929 rv = kstrtouint(buf, 10, &n);
3932 atomic_set(&mddev->max_corr_read_errors, n);
3936 static struct md_sysfs_entry max_corr_read_errors =
3937 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3938 max_corrected_read_errors_store);
3941 null_show(struct mddev *mddev, char *page)
3947 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3949 /* buf must be %d:%d\n? giving major and minor numbers */
3950 /* The new device is added to the array.
3951 * If the array has a persistent superblock, we read the
3952 * superblock to initialise info and check validity.
3953 * Otherwise, only checking done is that in bind_rdev_to_array,
3954 * which mainly checks size.
3957 int major = simple_strtoul(buf, &e, 10);
3960 struct md_rdev *rdev;
3963 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3965 minor = simple_strtoul(e+1, &e, 10);
3966 if (*e && *e != '\n')
3968 dev = MKDEV(major, minor);
3969 if (major != MAJOR(dev) ||
3970 minor != MINOR(dev))
3973 flush_workqueue(md_misc_wq);
3975 err = mddev_lock(mddev);
3978 if (mddev->persistent) {
3979 rdev = md_import_device(dev, mddev->major_version,
3980 mddev->minor_version);
3981 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3982 struct md_rdev *rdev0
3983 = list_entry(mddev->disks.next,
3984 struct md_rdev, same_set);
3985 err = super_types[mddev->major_version]
3986 .load_super(rdev, rdev0, mddev->minor_version);
3990 } else if (mddev->external)
3991 rdev = md_import_device(dev, -2, -1);
3993 rdev = md_import_device(dev, -1, -1);
3996 mddev_unlock(mddev);
3997 return PTR_ERR(rdev);
3999 err = bind_rdev_to_array(rdev, mddev);
4003 mddev_unlock(mddev);
4004 return err ? err : len;
4007 static struct md_sysfs_entry md_new_device =
4008 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4011 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4014 unsigned long chunk, end_chunk;
4017 err = mddev_lock(mddev);
4022 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4024 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4025 if (buf == end) break;
4026 if (*end == '-') { /* range */
4028 end_chunk = simple_strtoul(buf, &end, 0);
4029 if (buf == end) break;
4031 if (*end && !isspace(*end)) break;
4032 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4033 buf = skip_spaces(end);
4035 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4037 mddev_unlock(mddev);
4041 static struct md_sysfs_entry md_bitmap =
4042 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4045 size_show(struct mddev *mddev, char *page)
4047 return sprintf(page, "%llu\n",
4048 (unsigned long long)mddev->dev_sectors / 2);
4051 static int update_size(struct mddev *mddev, sector_t num_sectors);
4054 size_store(struct mddev *mddev, const char *buf, size_t len)
4056 /* If array is inactive, we can reduce the component size, but
4057 * not increase it (except from 0).
4058 * If array is active, we can try an on-line resize
4061 int err = strict_blocks_to_sectors(buf, §ors);
4065 err = mddev_lock(mddev);
4069 if (mddev_is_clustered(mddev))
4070 md_cluster_ops->metadata_update_start(mddev);
4071 err = update_size(mddev, sectors);
4072 md_update_sb(mddev, 1);
4073 if (mddev_is_clustered(mddev))
4074 md_cluster_ops->metadata_update_finish(mddev);
4076 if (mddev->dev_sectors == 0 ||
4077 mddev->dev_sectors > sectors)
4078 mddev->dev_sectors = sectors;
4082 mddev_unlock(mddev);
4083 return err ? err : len;
4086 static struct md_sysfs_entry md_size =
4087 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4089 /* Metadata version.
4091 * 'none' for arrays with no metadata (good luck...)
4092 * 'external' for arrays with externally managed metadata,
4093 * or N.M for internally known formats
4096 metadata_show(struct mddev *mddev, char *page)
4098 if (mddev->persistent)
4099 return sprintf(page, "%d.%d\n",
4100 mddev->major_version, mddev->minor_version);
4101 else if (mddev->external)
4102 return sprintf(page, "external:%s\n", mddev->metadata_type);
4104 return sprintf(page, "none\n");
4108 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4113 /* Changing the details of 'external' metadata is
4114 * always permitted. Otherwise there must be
4115 * no devices attached to the array.
4118 err = mddev_lock(mddev);
4122 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4124 else if (!list_empty(&mddev->disks))
4128 if (cmd_match(buf, "none")) {
4129 mddev->persistent = 0;
4130 mddev->external = 0;
4131 mddev->major_version = 0;
4132 mddev->minor_version = 90;
4135 if (strncmp(buf, "external:", 9) == 0) {
4136 size_t namelen = len-9;
4137 if (namelen >= sizeof(mddev->metadata_type))
4138 namelen = sizeof(mddev->metadata_type)-1;
4139 strncpy(mddev->metadata_type, buf+9, namelen);
4140 mddev->metadata_type[namelen] = 0;
4141 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4142 mddev->metadata_type[--namelen] = 0;
4143 mddev->persistent = 0;
4144 mddev->external = 1;
4145 mddev->major_version = 0;
4146 mddev->minor_version = 90;
4149 major = simple_strtoul(buf, &e, 10);
4151 if (e==buf || *e != '.')
4154 minor = simple_strtoul(buf, &e, 10);
4155 if (e==buf || (*e && *e != '\n') )
4158 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4160 mddev->major_version = major;
4161 mddev->minor_version = minor;
4162 mddev->persistent = 1;
4163 mddev->external = 0;
4166 mddev_unlock(mddev);
4170 static struct md_sysfs_entry md_metadata =
4171 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4174 action_show(struct mddev *mddev, char *page)
4176 char *type = "idle";
4177 unsigned long recovery = mddev->recovery;
4178 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4180 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4181 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4182 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4184 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4185 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4187 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4191 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4193 else if (mddev->reshape_position != MaxSector)
4196 return sprintf(page, "%s\n", type);
4200 action_store(struct mddev *mddev, const char *page, size_t len)
4202 if (!mddev->pers || !mddev->pers->sync_request)
4206 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4207 if (cmd_match(page, "frozen"))
4208 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4210 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4211 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4212 mddev_lock(mddev) == 0) {
4213 flush_workqueue(md_misc_wq);
4214 if (mddev->sync_thread) {
4215 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4216 md_reap_sync_thread(mddev);
4218 mddev_unlock(mddev);
4220 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4221 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4223 else if (cmd_match(page, "resync"))
4224 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4225 else if (cmd_match(page, "recover")) {
4226 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4227 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4228 } else if (cmd_match(page, "reshape")) {
4230 if (mddev->pers->start_reshape == NULL)
4232 err = mddev_lock(mddev);
4234 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4235 err = mddev->pers->start_reshape(mddev);
4236 mddev_unlock(mddev);
4240 sysfs_notify(&mddev->kobj, NULL, "degraded");
4242 if (cmd_match(page, "check"))
4243 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4244 else if (!cmd_match(page, "repair"))
4246 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4247 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4248 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4250 if (mddev->ro == 2) {
4251 /* A write to sync_action is enough to justify
4252 * canceling read-auto mode
4255 md_wakeup_thread(mddev->sync_thread);
4257 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4258 md_wakeup_thread(mddev->thread);
4259 sysfs_notify_dirent_safe(mddev->sysfs_action);
4263 static struct md_sysfs_entry md_scan_mode =
4264 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4267 last_sync_action_show(struct mddev *mddev, char *page)
4269 return sprintf(page, "%s\n", mddev->last_sync_action);
4272 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4275 mismatch_cnt_show(struct mddev *mddev, char *page)
4277 return sprintf(page, "%llu\n",
4278 (unsigned long long)
4279 atomic64_read(&mddev->resync_mismatches));
4282 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4285 sync_min_show(struct mddev *mddev, char *page)
4287 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4288 mddev->sync_speed_min ? "local": "system");
4292 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4297 if (strncmp(buf, "system", 6)==0) {
4300 rv = kstrtouint(buf, 10, &min);
4306 mddev->sync_speed_min = min;
4310 static struct md_sysfs_entry md_sync_min =
4311 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4314 sync_max_show(struct mddev *mddev, char *page)
4316 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4317 mddev->sync_speed_max ? "local": "system");
4321 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4326 if (strncmp(buf, "system", 6)==0) {
4329 rv = kstrtouint(buf, 10, &max);
4335 mddev->sync_speed_max = max;
4339 static struct md_sysfs_entry md_sync_max =
4340 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4343 degraded_show(struct mddev *mddev, char *page)
4345 return sprintf(page, "%d\n", mddev->degraded);
4347 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4350 sync_force_parallel_show(struct mddev *mddev, char *page)
4352 return sprintf(page, "%d\n", mddev->parallel_resync);
4356 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4360 if (kstrtol(buf, 10, &n))
4363 if (n != 0 && n != 1)
4366 mddev->parallel_resync = n;
4368 if (mddev->sync_thread)
4369 wake_up(&resync_wait);
4374 /* force parallel resync, even with shared block devices */
4375 static struct md_sysfs_entry md_sync_force_parallel =
4376 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4377 sync_force_parallel_show, sync_force_parallel_store);
4380 sync_speed_show(struct mddev *mddev, char *page)
4382 unsigned long resync, dt, db;
4383 if (mddev->curr_resync == 0)
4384 return sprintf(page, "none\n");
4385 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4386 dt = (jiffies - mddev->resync_mark) / HZ;
4388 db = resync - mddev->resync_mark_cnt;
4389 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4392 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4395 sync_completed_show(struct mddev *mddev, char *page)
4397 unsigned long long max_sectors, resync;
4399 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4400 return sprintf(page, "none\n");
4402 if (mddev->curr_resync == 1 ||
4403 mddev->curr_resync == 2)
4404 return sprintf(page, "delayed\n");
4406 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4407 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4408 max_sectors = mddev->resync_max_sectors;
4410 max_sectors = mddev->dev_sectors;
4412 resync = mddev->curr_resync_completed;
4413 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4416 static struct md_sysfs_entry md_sync_completed =
4417 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4420 min_sync_show(struct mddev *mddev, char *page)
4422 return sprintf(page, "%llu\n",
4423 (unsigned long long)mddev->resync_min);
4426 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4428 unsigned long long min;
4431 if (kstrtoull(buf, 10, &min))
4434 spin_lock(&mddev->lock);
4436 if (min > mddev->resync_max)
4440 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4443 /* Round down to multiple of 4K for safety */
4444 mddev->resync_min = round_down(min, 8);
4448 spin_unlock(&mddev->lock);
4452 static struct md_sysfs_entry md_min_sync =
4453 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4456 max_sync_show(struct mddev *mddev, char *page)
4458 if (mddev->resync_max == MaxSector)
4459 return sprintf(page, "max\n");
4461 return sprintf(page, "%llu\n",
4462 (unsigned long long)mddev->resync_max);
4465 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4468 spin_lock(&mddev->lock);
4469 if (strncmp(buf, "max", 3) == 0)
4470 mddev->resync_max = MaxSector;
4472 unsigned long long max;
4476 if (kstrtoull(buf, 10, &max))
4478 if (max < mddev->resync_min)
4482 if (max < mddev->resync_max &&
4484 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4487 /* Must be a multiple of chunk_size */
4488 chunk = mddev->chunk_sectors;
4490 sector_t temp = max;
4493 if (sector_div(temp, chunk))
4496 mddev->resync_max = max;
4498 wake_up(&mddev->recovery_wait);
4501 spin_unlock(&mddev->lock);
4505 static struct md_sysfs_entry md_max_sync =
4506 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4509 suspend_lo_show(struct mddev *mddev, char *page)
4511 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4515 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4517 unsigned long long old, new;
4520 err = kstrtoull(buf, 10, &new);
4523 if (new != (sector_t)new)
4526 err = mddev_lock(mddev);
4530 if (mddev->pers == NULL ||
4531 mddev->pers->quiesce == NULL)
4533 old = mddev->suspend_lo;
4534 mddev->suspend_lo = new;
4536 /* Shrinking suspended region */
4537 mddev->pers->quiesce(mddev, 2);
4539 /* Expanding suspended region - need to wait */
4540 mddev->pers->quiesce(mddev, 1);
4541 mddev->pers->quiesce(mddev, 0);
4545 mddev_unlock(mddev);
4548 static struct md_sysfs_entry md_suspend_lo =
4549 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4552 suspend_hi_show(struct mddev *mddev, char *page)
4554 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4558 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4560 unsigned long long old, new;
4563 err = kstrtoull(buf, 10, &new);
4566 if (new != (sector_t)new)
4569 err = mddev_lock(mddev);
4573 if (mddev->pers == NULL ||
4574 mddev->pers->quiesce == NULL)
4576 old = mddev->suspend_hi;
4577 mddev->suspend_hi = new;
4579 /* Shrinking suspended region */
4580 mddev->pers->quiesce(mddev, 2);
4582 /* Expanding suspended region - need to wait */
4583 mddev->pers->quiesce(mddev, 1);
4584 mddev->pers->quiesce(mddev, 0);
4588 mddev_unlock(mddev);
4591 static struct md_sysfs_entry md_suspend_hi =
4592 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4595 reshape_position_show(struct mddev *mddev, char *page)
4597 if (mddev->reshape_position != MaxSector)
4598 return sprintf(page, "%llu\n",
4599 (unsigned long long)mddev->reshape_position);
4600 strcpy(page, "none\n");
4605 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4607 struct md_rdev *rdev;
4608 unsigned long long new;
4611 err = kstrtoull(buf, 10, &new);
4614 if (new != (sector_t)new)
4616 err = mddev_lock(mddev);
4622 mddev->reshape_position = new;
4623 mddev->delta_disks = 0;
4624 mddev->reshape_backwards = 0;
4625 mddev->new_level = mddev->level;
4626 mddev->new_layout = mddev->layout;
4627 mddev->new_chunk_sectors = mddev->chunk_sectors;
4628 rdev_for_each(rdev, mddev)
4629 rdev->new_data_offset = rdev->data_offset;
4632 mddev_unlock(mddev);
4636 static struct md_sysfs_entry md_reshape_position =
4637 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4638 reshape_position_store);
4641 reshape_direction_show(struct mddev *mddev, char *page)
4643 return sprintf(page, "%s\n",
4644 mddev->reshape_backwards ? "backwards" : "forwards");
4648 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4653 if (cmd_match(buf, "forwards"))
4655 else if (cmd_match(buf, "backwards"))
4659 if (mddev->reshape_backwards == backwards)
4662 err = mddev_lock(mddev);
4665 /* check if we are allowed to change */
4666 if (mddev->delta_disks)
4668 else if (mddev->persistent &&
4669 mddev->major_version == 0)
4672 mddev->reshape_backwards = backwards;
4673 mddev_unlock(mddev);
4677 static struct md_sysfs_entry md_reshape_direction =
4678 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4679 reshape_direction_store);
4682 array_size_show(struct mddev *mddev, char *page)
4684 if (mddev->external_size)
4685 return sprintf(page, "%llu\n",
4686 (unsigned long long)mddev->array_sectors/2);
4688 return sprintf(page, "default\n");
4692 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4697 err = mddev_lock(mddev);
4701 if (strncmp(buf, "default", 7) == 0) {
4703 sectors = mddev->pers->size(mddev, 0, 0);
4705 sectors = mddev->array_sectors;
4707 mddev->external_size = 0;
4709 if (strict_blocks_to_sectors(buf, §ors) < 0)
4711 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4714 mddev->external_size = 1;
4718 mddev->array_sectors = sectors;
4720 set_capacity(mddev->gendisk, mddev->array_sectors);
4721 revalidate_disk(mddev->gendisk);
4724 mddev_unlock(mddev);
4728 static struct md_sysfs_entry md_array_size =
4729 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4732 static struct attribute *md_default_attrs[] = {
4735 &md_raid_disks.attr,
4736 &md_chunk_size.attr,
4738 &md_resync_start.attr,
4740 &md_new_device.attr,
4741 &md_safe_delay.attr,
4742 &md_array_state.attr,
4743 &md_reshape_position.attr,
4744 &md_reshape_direction.attr,
4745 &md_array_size.attr,
4746 &max_corr_read_errors.attr,
4750 static struct attribute *md_redundancy_attrs[] = {
4752 &md_last_scan_mode.attr,
4753 &md_mismatches.attr,
4756 &md_sync_speed.attr,
4757 &md_sync_force_parallel.attr,
4758 &md_sync_completed.attr,
4761 &md_suspend_lo.attr,
4762 &md_suspend_hi.attr,
4767 static struct attribute_group md_redundancy_group = {
4769 .attrs = md_redundancy_attrs,
4773 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4775 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4776 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4781 spin_lock(&all_mddevs_lock);
4782 if (list_empty(&mddev->all_mddevs)) {
4783 spin_unlock(&all_mddevs_lock);
4787 spin_unlock(&all_mddevs_lock);
4789 rv = entry->show(mddev, page);
4795 md_attr_store(struct kobject *kobj, struct attribute *attr,
4796 const char *page, size_t length)
4798 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4799 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4804 if (!capable(CAP_SYS_ADMIN))
4806 spin_lock(&all_mddevs_lock);
4807 if (list_empty(&mddev->all_mddevs)) {
4808 spin_unlock(&all_mddevs_lock);
4812 spin_unlock(&all_mddevs_lock);
4813 rv = entry->store(mddev, page, length);
4818 static void md_free(struct kobject *ko)
4820 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4822 if (mddev->sysfs_state)
4823 sysfs_put(mddev->sysfs_state);
4826 blk_cleanup_queue(mddev->queue);
4827 if (mddev->gendisk) {
4828 del_gendisk(mddev->gendisk);
4829 put_disk(mddev->gendisk);
4835 static const struct sysfs_ops md_sysfs_ops = {
4836 .show = md_attr_show,
4837 .store = md_attr_store,
4839 static struct kobj_type md_ktype = {
4841 .sysfs_ops = &md_sysfs_ops,
4842 .default_attrs = md_default_attrs,
4847 static void mddev_delayed_delete(struct work_struct *ws)
4849 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4851 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4852 kobject_del(&mddev->kobj);
4853 kobject_put(&mddev->kobj);
4856 static int md_alloc(dev_t dev, char *name)
4858 static DEFINE_MUTEX(disks_mutex);
4859 struct mddev *mddev = mddev_find(dev);
4860 struct gendisk *disk;
4869 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4870 shift = partitioned ? MdpMinorShift : 0;
4871 unit = MINOR(mddev->unit) >> shift;
4873 /* wait for any previous instance of this device to be
4874 * completely removed (mddev_delayed_delete).
4876 flush_workqueue(md_misc_wq);
4878 mutex_lock(&disks_mutex);
4884 /* Need to ensure that 'name' is not a duplicate.
4886 struct mddev *mddev2;
4887 spin_lock(&all_mddevs_lock);
4889 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4890 if (mddev2->gendisk &&
4891 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4892 spin_unlock(&all_mddevs_lock);
4895 spin_unlock(&all_mddevs_lock);
4899 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4902 mddev->queue->queuedata = mddev;
4904 blk_queue_make_request(mddev->queue, md_make_request);
4905 blk_set_stacking_limits(&mddev->queue->limits);
4907 disk = alloc_disk(1 << shift);
4909 blk_cleanup_queue(mddev->queue);
4910 mddev->queue = NULL;
4913 disk->major = MAJOR(mddev->unit);
4914 disk->first_minor = unit << shift;
4916 strcpy(disk->disk_name, name);
4917 else if (partitioned)
4918 sprintf(disk->disk_name, "md_d%d", unit);
4920 sprintf(disk->disk_name, "md%d", unit);
4921 disk->fops = &md_fops;
4922 disk->private_data = mddev;
4923 disk->queue = mddev->queue;
4924 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4925 /* Allow extended partitions. This makes the
4926 * 'mdp' device redundant, but we can't really
4929 disk->flags |= GENHD_FL_EXT_DEVT;
4930 mddev->gendisk = disk;
4931 /* As soon as we call add_disk(), another thread could get
4932 * through to md_open, so make sure it doesn't get too far
4934 mutex_lock(&mddev->open_mutex);
4937 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4938 &disk_to_dev(disk)->kobj, "%s", "md");
4940 /* This isn't possible, but as kobject_init_and_add is marked
4941 * __must_check, we must do something with the result
4943 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4947 if (mddev->kobj.sd &&
4948 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4949 printk(KERN_DEBUG "pointless warning\n");
4950 mutex_unlock(&mddev->open_mutex);
4952 mutex_unlock(&disks_mutex);
4953 if (!error && mddev->kobj.sd) {
4954 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4955 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4961 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4963 md_alloc(dev, NULL);
4967 static int add_named_array(const char *val, struct kernel_param *kp)
4969 /* val must be "md_*" where * is not all digits.
4970 * We allocate an array with a large free minor number, and
4971 * set the name to val. val must not already be an active name.
4973 int len = strlen(val);
4974 char buf[DISK_NAME_LEN];
4976 while (len && val[len-1] == '\n')
4978 if (len >= DISK_NAME_LEN)
4980 strlcpy(buf, val, len+1);
4981 if (strncmp(buf, "md_", 3) != 0)
4983 return md_alloc(0, buf);
4986 static void md_safemode_timeout(unsigned long data)
4988 struct mddev *mddev = (struct mddev *) data;
4990 if (!atomic_read(&mddev->writes_pending)) {
4991 mddev->safemode = 1;
4992 if (mddev->external)
4993 sysfs_notify_dirent_safe(mddev->sysfs_state);
4995 md_wakeup_thread(mddev->thread);
4998 static int start_dirty_degraded;
5000 int md_run(struct mddev *mddev)
5003 struct md_rdev *rdev;
5004 struct md_personality *pers;
5006 if (list_empty(&mddev->disks))
5007 /* cannot run an array with no devices.. */
5012 /* Cannot run until previous stop completes properly */
5013 if (mddev->sysfs_active)
5017 * Analyze all RAID superblock(s)
5019 if (!mddev->raid_disks) {
5020 if (!mddev->persistent)
5025 if (mddev->level != LEVEL_NONE)
5026 request_module("md-level-%d", mddev->level);
5027 else if (mddev->clevel[0])
5028 request_module("md-%s", mddev->clevel);
5031 * Drop all container device buffers, from now on
5032 * the only valid external interface is through the md
5035 rdev_for_each(rdev, mddev) {
5036 if (test_bit(Faulty, &rdev->flags))
5038 sync_blockdev(rdev->bdev);
5039 invalidate_bdev(rdev->bdev);
5041 /* perform some consistency tests on the device.
5042 * We don't want the data to overlap the metadata,
5043 * Internal Bitmap issues have been handled elsewhere.
5045 if (rdev->meta_bdev) {
5046 /* Nothing to check */;
5047 } else if (rdev->data_offset < rdev->sb_start) {
5048 if (mddev->dev_sectors &&
5049 rdev->data_offset + mddev->dev_sectors
5051 printk("md: %s: data overlaps metadata\n",
5056 if (rdev->sb_start + rdev->sb_size/512
5057 > rdev->data_offset) {
5058 printk("md: %s: metadata overlaps data\n",
5063 sysfs_notify_dirent_safe(rdev->sysfs_state);
5066 if (mddev->bio_set == NULL)
5067 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5069 spin_lock(&pers_lock);
5070 pers = find_pers(mddev->level, mddev->clevel);
5071 if (!pers || !try_module_get(pers->owner)) {
5072 spin_unlock(&pers_lock);
5073 if (mddev->level != LEVEL_NONE)
5074 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5077 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5081 spin_unlock(&pers_lock);
5082 if (mddev->level != pers->level) {
5083 mddev->level = pers->level;
5084 mddev->new_level = pers->level;
5086 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5088 if (mddev->reshape_position != MaxSector &&
5089 pers->start_reshape == NULL) {
5090 /* This personality cannot handle reshaping... */
5091 module_put(pers->owner);
5095 if (pers->sync_request) {
5096 /* Warn if this is a potentially silly
5099 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5100 struct md_rdev *rdev2;
5103 rdev_for_each(rdev, mddev)
5104 rdev_for_each(rdev2, mddev) {
5106 rdev->bdev->bd_contains ==
5107 rdev2->bdev->bd_contains) {
5109 "%s: WARNING: %s appears to be"
5110 " on the same physical disk as"
5113 bdevname(rdev->bdev,b),
5114 bdevname(rdev2->bdev,b2));
5121 "True protection against single-disk"
5122 " failure might be compromised.\n");
5125 mddev->recovery = 0;
5126 /* may be over-ridden by personality */
5127 mddev->resync_max_sectors = mddev->dev_sectors;
5129 mddev->ok_start_degraded = start_dirty_degraded;
5131 if (start_readonly && mddev->ro == 0)
5132 mddev->ro = 2; /* read-only, but switch on first write */
5134 err = pers->run(mddev);
5136 printk(KERN_ERR "md: pers->run() failed ...\n");
5137 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5138 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5139 " but 'external_size' not in effect?\n", __func__);
5141 "md: invalid array_size %llu > default size %llu\n",
5142 (unsigned long long)mddev->array_sectors / 2,
5143 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5146 if (err == 0 && pers->sync_request &&
5147 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5148 struct bitmap *bitmap;
5150 bitmap = bitmap_create(mddev, -1);
5151 if (IS_ERR(bitmap)) {
5152 err = PTR_ERR(bitmap);
5153 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5154 mdname(mddev), err);
5156 mddev->bitmap = bitmap;
5160 mddev_detach(mddev);
5162 pers->free(mddev, mddev->private);
5163 mddev->private = NULL;
5164 module_put(pers->owner);
5165 bitmap_destroy(mddev);
5169 mddev->queue->backing_dev_info.congested_data = mddev;
5170 mddev->queue->backing_dev_info.congested_fn = md_congested;
5172 if (pers->sync_request) {
5173 if (mddev->kobj.sd &&
5174 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5176 "md: cannot register extra attributes for %s\n",
5178 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5179 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5182 atomic_set(&mddev->writes_pending,0);
5183 atomic_set(&mddev->max_corr_read_errors,
5184 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5185 mddev->safemode = 0;
5186 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5189 spin_lock(&mddev->lock);
5192 spin_unlock(&mddev->lock);
5193 rdev_for_each(rdev, mddev)
5194 if (rdev->raid_disk >= 0)
5195 if (sysfs_link_rdev(mddev, rdev))
5196 /* failure here is OK */;
5198 if (mddev->degraded && !mddev->ro)
5199 /* This ensures that recovering status is reported immediately
5200 * via sysfs - until a lack of spares is confirmed.
5202 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5203 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5205 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5206 md_update_sb(mddev, 0);
5208 md_new_event(mddev);
5209 sysfs_notify_dirent_safe(mddev->sysfs_state);
5210 sysfs_notify_dirent_safe(mddev->sysfs_action);
5211 sysfs_notify(&mddev->kobj, NULL, "degraded");
5214 EXPORT_SYMBOL_GPL(md_run);
5216 static int do_md_run(struct mddev *mddev)
5220 err = md_run(mddev);
5223 err = bitmap_load(mddev);
5225 bitmap_destroy(mddev);
5229 md_wakeup_thread(mddev->thread);
5230 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5232 set_capacity(mddev->gendisk, mddev->array_sectors);
5233 revalidate_disk(mddev->gendisk);
5235 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5240 static int restart_array(struct mddev *mddev)
5242 struct gendisk *disk = mddev->gendisk;
5244 /* Complain if it has no devices */
5245 if (list_empty(&mddev->disks))
5251 mddev->safemode = 0;
5253 set_disk_ro(disk, 0);
5254 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5256 /* Kick recovery or resync if necessary */
5257 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5258 md_wakeup_thread(mddev->thread);
5259 md_wakeup_thread(mddev->sync_thread);
5260 sysfs_notify_dirent_safe(mddev->sysfs_state);
5264 static void md_clean(struct mddev *mddev)
5266 mddev->array_sectors = 0;
5267 mddev->external_size = 0;
5268 mddev->dev_sectors = 0;
5269 mddev->raid_disks = 0;
5270 mddev->recovery_cp = 0;
5271 mddev->resync_min = 0;
5272 mddev->resync_max = MaxSector;
5273 mddev->reshape_position = MaxSector;
5274 mddev->external = 0;
5275 mddev->persistent = 0;
5276 mddev->level = LEVEL_NONE;
5277 mddev->clevel[0] = 0;
5280 mddev->metadata_type[0] = 0;
5281 mddev->chunk_sectors = 0;
5282 mddev->ctime = mddev->utime = 0;
5284 mddev->max_disks = 0;
5286 mddev->can_decrease_events = 0;
5287 mddev->delta_disks = 0;
5288 mddev->reshape_backwards = 0;
5289 mddev->new_level = LEVEL_NONE;
5290 mddev->new_layout = 0;
5291 mddev->new_chunk_sectors = 0;
5292 mddev->curr_resync = 0;
5293 atomic64_set(&mddev->resync_mismatches, 0);
5294 mddev->suspend_lo = mddev->suspend_hi = 0;
5295 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5296 mddev->recovery = 0;
5299 mddev->degraded = 0;
5300 mddev->safemode = 0;
5301 mddev->private = NULL;
5302 mddev->bitmap_info.offset = 0;
5303 mddev->bitmap_info.default_offset = 0;
5304 mddev->bitmap_info.default_space = 0;
5305 mddev->bitmap_info.chunksize = 0;
5306 mddev->bitmap_info.daemon_sleep = 0;
5307 mddev->bitmap_info.max_write_behind = 0;
5310 static void __md_stop_writes(struct mddev *mddev)
5312 if (mddev_is_clustered(mddev))
5313 md_cluster_ops->metadata_update_start(mddev);
5314 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5315 flush_workqueue(md_misc_wq);
5316 if (mddev->sync_thread) {
5317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5318 md_reap_sync_thread(mddev);
5321 del_timer_sync(&mddev->safemode_timer);
5323 bitmap_flush(mddev);
5324 md_super_wait(mddev);
5326 if (mddev->ro == 0 &&
5327 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5328 /* mark array as shutdown cleanly */
5330 md_update_sb(mddev, 1);
5332 if (mddev_is_clustered(mddev))
5333 md_cluster_ops->metadata_update_finish(mddev);
5336 void md_stop_writes(struct mddev *mddev)
5338 mddev_lock_nointr(mddev);
5339 __md_stop_writes(mddev);
5340 mddev_unlock(mddev);
5342 EXPORT_SYMBOL_GPL(md_stop_writes);
5344 static void mddev_detach(struct mddev *mddev)
5346 struct bitmap *bitmap = mddev->bitmap;
5347 /* wait for behind writes to complete */
5348 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5349 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5351 /* need to kick something here to make sure I/O goes? */
5352 wait_event(bitmap->behind_wait,
5353 atomic_read(&bitmap->behind_writes) == 0);
5355 if (mddev->pers && mddev->pers->quiesce) {
5356 mddev->pers->quiesce(mddev, 1);
5357 mddev->pers->quiesce(mddev, 0);
5359 md_unregister_thread(&mddev->thread);
5361 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5364 static void __md_stop(struct mddev *mddev)
5366 struct md_personality *pers = mddev->pers;
5367 mddev_detach(mddev);
5368 /* Ensure ->event_work is done */
5369 flush_workqueue(md_misc_wq);
5370 spin_lock(&mddev->lock);
5373 spin_unlock(&mddev->lock);
5374 pers->free(mddev, mddev->private);
5375 mddev->private = NULL;
5376 if (pers->sync_request && mddev->to_remove == NULL)
5377 mddev->to_remove = &md_redundancy_group;
5378 module_put(pers->owner);
5379 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5382 void md_stop(struct mddev *mddev)
5384 /* stop the array and free an attached data structures.
5385 * This is called from dm-raid
5388 bitmap_destroy(mddev);
5390 bioset_free(mddev->bio_set);
5393 EXPORT_SYMBOL_GPL(md_stop);
5395 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5400 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5402 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5403 md_wakeup_thread(mddev->thread);
5405 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5406 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5407 if (mddev->sync_thread)
5408 /* Thread might be blocked waiting for metadata update
5409 * which will now never happen */
5410 wake_up_process(mddev->sync_thread->tsk);
5412 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5414 mddev_unlock(mddev);
5415 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5417 wait_event(mddev->sb_wait,
5418 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5419 mddev_lock_nointr(mddev);
5421 mutex_lock(&mddev->open_mutex);
5422 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5423 mddev->sync_thread ||
5424 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5425 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5426 printk("md: %s still in use.\n",mdname(mddev));
5428 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5429 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5430 md_wakeup_thread(mddev->thread);
5436 __md_stop_writes(mddev);
5442 set_disk_ro(mddev->gendisk, 1);
5443 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5444 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5445 md_wakeup_thread(mddev->thread);
5446 sysfs_notify_dirent_safe(mddev->sysfs_state);
5450 mutex_unlock(&mddev->open_mutex);
5455 * 0 - completely stop and dis-assemble array
5456 * 2 - stop but do not disassemble array
5458 static int do_md_stop(struct mddev *mddev, int mode,
5459 struct block_device *bdev)
5461 struct gendisk *disk = mddev->gendisk;
5462 struct md_rdev *rdev;
5465 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5467 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5468 md_wakeup_thread(mddev->thread);
5470 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5471 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5472 if (mddev->sync_thread)
5473 /* Thread might be blocked waiting for metadata update
5474 * which will now never happen */
5475 wake_up_process(mddev->sync_thread->tsk);
5477 mddev_unlock(mddev);
5478 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5479 !test_bit(MD_RECOVERY_RUNNING,
5480 &mddev->recovery)));
5481 mddev_lock_nointr(mddev);
5483 mutex_lock(&mddev->open_mutex);
5484 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5485 mddev->sysfs_active ||
5486 mddev->sync_thread ||
5487 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5488 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5489 printk("md: %s still in use.\n",mdname(mddev));
5490 mutex_unlock(&mddev->open_mutex);
5492 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5493 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5494 md_wakeup_thread(mddev->thread);
5500 set_disk_ro(disk, 0);
5502 __md_stop_writes(mddev);
5504 mddev->queue->backing_dev_info.congested_fn = NULL;
5506 /* tell userspace to handle 'inactive' */
5507 sysfs_notify_dirent_safe(mddev->sysfs_state);
5509 rdev_for_each(rdev, mddev)
5510 if (rdev->raid_disk >= 0)
5511 sysfs_unlink_rdev(mddev, rdev);
5513 set_capacity(disk, 0);
5514 mutex_unlock(&mddev->open_mutex);
5516 revalidate_disk(disk);
5521 mutex_unlock(&mddev->open_mutex);
5523 * Free resources if final stop
5526 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5528 bitmap_destroy(mddev);
5529 if (mddev->bitmap_info.file) {
5530 struct file *f = mddev->bitmap_info.file;
5531 spin_lock(&mddev->lock);
5532 mddev->bitmap_info.file = NULL;
5533 spin_unlock(&mddev->lock);
5536 mddev->bitmap_info.offset = 0;
5538 export_array(mddev);
5541 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5542 if (mddev->hold_active == UNTIL_STOP)
5543 mddev->hold_active = 0;
5545 blk_integrity_unregister(disk);
5546 md_new_event(mddev);
5547 sysfs_notify_dirent_safe(mddev->sysfs_state);
5552 static void autorun_array(struct mddev *mddev)
5554 struct md_rdev *rdev;
5557 if (list_empty(&mddev->disks))
5560 printk(KERN_INFO "md: running: ");
5562 rdev_for_each(rdev, mddev) {
5563 char b[BDEVNAME_SIZE];
5564 printk("<%s>", bdevname(rdev->bdev,b));
5568 err = do_md_run(mddev);
5570 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5571 do_md_stop(mddev, 0, NULL);
5576 * lets try to run arrays based on all disks that have arrived
5577 * until now. (those are in pending_raid_disks)
5579 * the method: pick the first pending disk, collect all disks with
5580 * the same UUID, remove all from the pending list and put them into
5581 * the 'same_array' list. Then order this list based on superblock
5582 * update time (freshest comes first), kick out 'old' disks and
5583 * compare superblocks. If everything's fine then run it.
5585 * If "unit" is allocated, then bump its reference count
5587 static void autorun_devices(int part)
5589 struct md_rdev *rdev0, *rdev, *tmp;
5590 struct mddev *mddev;
5591 char b[BDEVNAME_SIZE];
5593 printk(KERN_INFO "md: autorun ...\n");
5594 while (!list_empty(&pending_raid_disks)) {
5597 LIST_HEAD(candidates);
5598 rdev0 = list_entry(pending_raid_disks.next,
5599 struct md_rdev, same_set);
5601 printk(KERN_INFO "md: considering %s ...\n",
5602 bdevname(rdev0->bdev,b));
5603 INIT_LIST_HEAD(&candidates);
5604 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5605 if (super_90_load(rdev, rdev0, 0) >= 0) {
5606 printk(KERN_INFO "md: adding %s ...\n",
5607 bdevname(rdev->bdev,b));
5608 list_move(&rdev->same_set, &candidates);
5611 * now we have a set of devices, with all of them having
5612 * mostly sane superblocks. It's time to allocate the
5616 dev = MKDEV(mdp_major,
5617 rdev0->preferred_minor << MdpMinorShift);
5618 unit = MINOR(dev) >> MdpMinorShift;
5620 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5623 if (rdev0->preferred_minor != unit) {
5624 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5625 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5629 md_probe(dev, NULL, NULL);
5630 mddev = mddev_find(dev);
5631 if (!mddev || !mddev->gendisk) {
5635 "md: cannot allocate memory for md drive.\n");
5638 if (mddev_lock(mddev))
5639 printk(KERN_WARNING "md: %s locked, cannot run\n",
5641 else if (mddev->raid_disks || mddev->major_version
5642 || !list_empty(&mddev->disks)) {
5644 "md: %s already running, cannot run %s\n",
5645 mdname(mddev), bdevname(rdev0->bdev,b));
5646 mddev_unlock(mddev);
5648 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5649 mddev->persistent = 1;
5650 rdev_for_each_list(rdev, tmp, &candidates) {
5651 list_del_init(&rdev->same_set);
5652 if (bind_rdev_to_array(rdev, mddev))
5655 autorun_array(mddev);
5656 mddev_unlock(mddev);
5658 /* on success, candidates will be empty, on error
5661 rdev_for_each_list(rdev, tmp, &candidates) {
5662 list_del_init(&rdev->same_set);
5667 printk(KERN_INFO "md: ... autorun DONE.\n");
5669 #endif /* !MODULE */
5671 static int get_version(void __user *arg)
5675 ver.major = MD_MAJOR_VERSION;
5676 ver.minor = MD_MINOR_VERSION;
5677 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5679 if (copy_to_user(arg, &ver, sizeof(ver)))
5685 static int get_array_info(struct mddev *mddev, void __user *arg)
5687 mdu_array_info_t info;
5688 int nr,working,insync,failed,spare;
5689 struct md_rdev *rdev;
5691 nr = working = insync = failed = spare = 0;
5693 rdev_for_each_rcu(rdev, mddev) {
5695 if (test_bit(Faulty, &rdev->flags))
5699 if (test_bit(In_sync, &rdev->flags))
5707 info.major_version = mddev->major_version;
5708 info.minor_version = mddev->minor_version;
5709 info.patch_version = MD_PATCHLEVEL_VERSION;
5710 info.ctime = mddev->ctime;
5711 info.level = mddev->level;
5712 info.size = mddev->dev_sectors / 2;
5713 if (info.size != mddev->dev_sectors / 2) /* overflow */
5716 info.raid_disks = mddev->raid_disks;
5717 info.md_minor = mddev->md_minor;
5718 info.not_persistent= !mddev->persistent;
5720 info.utime = mddev->utime;
5723 info.state = (1<<MD_SB_CLEAN);
5724 if (mddev->bitmap && mddev->bitmap_info.offset)
5725 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5726 if (mddev_is_clustered(mddev))
5727 info.state |= (1<<MD_SB_CLUSTERED);
5728 info.active_disks = insync;
5729 info.working_disks = working;
5730 info.failed_disks = failed;
5731 info.spare_disks = spare;
5733 info.layout = mddev->layout;
5734 info.chunk_size = mddev->chunk_sectors << 9;
5736 if (copy_to_user(arg, &info, sizeof(info)))
5742 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5744 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5748 file = kzalloc(sizeof(*file), GFP_NOIO);
5753 spin_lock(&mddev->lock);
5754 /* bitmap enabled */
5755 if (mddev->bitmap_info.file) {
5756 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5757 sizeof(file->pathname));
5761 memmove(file->pathname, ptr,
5762 sizeof(file->pathname)-(ptr-file->pathname));
5764 spin_unlock(&mddev->lock);
5767 copy_to_user(arg, file, sizeof(*file)))
5774 static int get_disk_info(struct mddev *mddev, void __user * arg)
5776 mdu_disk_info_t info;
5777 struct md_rdev *rdev;
5779 if (copy_from_user(&info, arg, sizeof(info)))
5783 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5785 info.major = MAJOR(rdev->bdev->bd_dev);
5786 info.minor = MINOR(rdev->bdev->bd_dev);
5787 info.raid_disk = rdev->raid_disk;
5789 if (test_bit(Faulty, &rdev->flags))
5790 info.state |= (1<<MD_DISK_FAULTY);
5791 else if (test_bit(In_sync, &rdev->flags)) {
5792 info.state |= (1<<MD_DISK_ACTIVE);
5793 info.state |= (1<<MD_DISK_SYNC);
5795 if (test_bit(WriteMostly, &rdev->flags))
5796 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5798 info.major = info.minor = 0;
5799 info.raid_disk = -1;
5800 info.state = (1<<MD_DISK_REMOVED);
5804 if (copy_to_user(arg, &info, sizeof(info)))
5810 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5812 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5813 struct md_rdev *rdev;
5814 dev_t dev = MKDEV(info->major,info->minor);
5816 if (mddev_is_clustered(mddev) &&
5817 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5818 pr_err("%s: Cannot add to clustered mddev.\n",
5823 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5826 if (!mddev->raid_disks) {
5828 /* expecting a device which has a superblock */
5829 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5832 "md: md_import_device returned %ld\n",
5834 return PTR_ERR(rdev);
5836 if (!list_empty(&mddev->disks)) {
5837 struct md_rdev *rdev0
5838 = list_entry(mddev->disks.next,
5839 struct md_rdev, same_set);
5840 err = super_types[mddev->major_version]
5841 .load_super(rdev, rdev0, mddev->minor_version);
5844 "md: %s has different UUID to %s\n",
5845 bdevname(rdev->bdev,b),
5846 bdevname(rdev0->bdev,b2));
5851 err = bind_rdev_to_array(rdev, mddev);
5858 * add_new_disk can be used once the array is assembled
5859 * to add "hot spares". They must already have a superblock
5864 if (!mddev->pers->hot_add_disk) {
5866 "%s: personality does not support diskops!\n",
5870 if (mddev->persistent)
5871 rdev = md_import_device(dev, mddev->major_version,
5872 mddev->minor_version);
5874 rdev = md_import_device(dev, -1, -1);
5877 "md: md_import_device returned %ld\n",
5879 return PTR_ERR(rdev);
5881 /* set saved_raid_disk if appropriate */
5882 if (!mddev->persistent) {
5883 if (info->state & (1<<MD_DISK_SYNC) &&
5884 info->raid_disk < mddev->raid_disks) {
5885 rdev->raid_disk = info->raid_disk;
5886 set_bit(In_sync, &rdev->flags);
5887 clear_bit(Bitmap_sync, &rdev->flags);
5889 rdev->raid_disk = -1;
5890 rdev->saved_raid_disk = rdev->raid_disk;
5892 super_types[mddev->major_version].
5893 validate_super(mddev, rdev);
5894 if ((info->state & (1<<MD_DISK_SYNC)) &&
5895 rdev->raid_disk != info->raid_disk) {
5896 /* This was a hot-add request, but events doesn't
5897 * match, so reject it.
5903 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5904 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5905 set_bit(WriteMostly, &rdev->flags);
5907 clear_bit(WriteMostly, &rdev->flags);
5910 * check whether the device shows up in other nodes
5912 if (mddev_is_clustered(mddev)) {
5913 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5914 /* Through --cluster-confirm */
5915 set_bit(Candidate, &rdev->flags);
5916 err = md_cluster_ops->new_disk_ack(mddev, true);
5921 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5922 /* --add initiated by this node */
5923 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5925 md_cluster_ops->add_new_disk_finish(mddev);
5932 rdev->raid_disk = -1;
5933 err = bind_rdev_to_array(rdev, mddev);
5937 err = add_bound_rdev(rdev);
5938 if (mddev_is_clustered(mddev) &&
5939 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5940 md_cluster_ops->add_new_disk_finish(mddev);
5944 /* otherwise, add_new_disk is only allowed
5945 * for major_version==0 superblocks
5947 if (mddev->major_version != 0) {
5948 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5953 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5955 rdev = md_import_device(dev, -1, 0);
5958 "md: error, md_import_device() returned %ld\n",
5960 return PTR_ERR(rdev);
5962 rdev->desc_nr = info->number;
5963 if (info->raid_disk < mddev->raid_disks)
5964 rdev->raid_disk = info->raid_disk;
5966 rdev->raid_disk = -1;
5968 if (rdev->raid_disk < mddev->raid_disks)
5969 if (info->state & (1<<MD_DISK_SYNC))
5970 set_bit(In_sync, &rdev->flags);
5972 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5973 set_bit(WriteMostly, &rdev->flags);
5975 if (!mddev->persistent) {
5976 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5977 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5979 rdev->sb_start = calc_dev_sboffset(rdev);
5980 rdev->sectors = rdev->sb_start;
5982 err = bind_rdev_to_array(rdev, mddev);
5992 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5994 char b[BDEVNAME_SIZE];
5995 struct md_rdev *rdev;
5997 rdev = find_rdev(mddev, dev);
6001 if (mddev_is_clustered(mddev))
6002 md_cluster_ops->metadata_update_start(mddev);
6004 clear_bit(Blocked, &rdev->flags);
6005 remove_and_add_spares(mddev, rdev);
6007 if (rdev->raid_disk >= 0)
6010 if (mddev_is_clustered(mddev))
6011 md_cluster_ops->remove_disk(mddev, rdev);
6013 md_kick_rdev_from_array(rdev);
6014 md_update_sb(mddev, 1);
6015 md_new_event(mddev);
6017 if (mddev_is_clustered(mddev))
6018 md_cluster_ops->metadata_update_finish(mddev);
6022 if (mddev_is_clustered(mddev))
6023 md_cluster_ops->metadata_update_cancel(mddev);
6024 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6025 bdevname(rdev->bdev,b), mdname(mddev));
6029 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6031 char b[BDEVNAME_SIZE];
6033 struct md_rdev *rdev;
6038 if (mddev->major_version != 0) {
6039 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6040 " version-0 superblocks.\n",
6044 if (!mddev->pers->hot_add_disk) {
6046 "%s: personality does not support diskops!\n",
6051 rdev = md_import_device(dev, -1, 0);
6054 "md: error, md_import_device() returned %ld\n",
6059 if (mddev->persistent)
6060 rdev->sb_start = calc_dev_sboffset(rdev);
6062 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6064 rdev->sectors = rdev->sb_start;
6066 if (test_bit(Faulty, &rdev->flags)) {
6068 "md: can not hot-add faulty %s disk to %s!\n",
6069 bdevname(rdev->bdev,b), mdname(mddev));
6074 if (mddev_is_clustered(mddev))
6075 md_cluster_ops->metadata_update_start(mddev);
6076 clear_bit(In_sync, &rdev->flags);
6078 rdev->saved_raid_disk = -1;
6079 err = bind_rdev_to_array(rdev, mddev);
6081 goto abort_clustered;
6084 * The rest should better be atomic, we can have disk failures
6085 * noticed in interrupt contexts ...
6088 rdev->raid_disk = -1;
6090 md_update_sb(mddev, 1);
6092 if (mddev_is_clustered(mddev))
6093 md_cluster_ops->metadata_update_finish(mddev);
6095 * Kick recovery, maybe this spare has to be added to the
6096 * array immediately.
6098 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6099 md_wakeup_thread(mddev->thread);
6100 md_new_event(mddev);
6104 if (mddev_is_clustered(mddev))
6105 md_cluster_ops->metadata_update_cancel(mddev);
6111 static int set_bitmap_file(struct mddev *mddev, int fd)
6116 if (!mddev->pers->quiesce || !mddev->thread)
6118 if (mddev->recovery || mddev->sync_thread)
6120 /* we should be able to change the bitmap.. */
6124 struct inode *inode;
6127 if (mddev->bitmap || mddev->bitmap_info.file)
6128 return -EEXIST; /* cannot add when bitmap is present */
6132 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6137 inode = f->f_mapping->host;
6138 if (!S_ISREG(inode->i_mode)) {
6139 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6142 } else if (!(f->f_mode & FMODE_WRITE)) {
6143 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6146 } else if (atomic_read(&inode->i_writecount) != 1) {
6147 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6155 mddev->bitmap_info.file = f;
6156 mddev->bitmap_info.offset = 0; /* file overrides offset */
6157 } else if (mddev->bitmap == NULL)
6158 return -ENOENT; /* cannot remove what isn't there */
6161 mddev->pers->quiesce(mddev, 1);
6163 struct bitmap *bitmap;
6165 bitmap = bitmap_create(mddev, -1);
6166 if (!IS_ERR(bitmap)) {
6167 mddev->bitmap = bitmap;
6168 err = bitmap_load(mddev);
6170 err = PTR_ERR(bitmap);
6172 if (fd < 0 || err) {
6173 bitmap_destroy(mddev);
6174 fd = -1; /* make sure to put the file */
6176 mddev->pers->quiesce(mddev, 0);
6179 struct file *f = mddev->bitmap_info.file;
6181 spin_lock(&mddev->lock);
6182 mddev->bitmap_info.file = NULL;
6183 spin_unlock(&mddev->lock);
6192 * set_array_info is used two different ways
6193 * The original usage is when creating a new array.
6194 * In this usage, raid_disks is > 0 and it together with
6195 * level, size, not_persistent,layout,chunksize determine the
6196 * shape of the array.
6197 * This will always create an array with a type-0.90.0 superblock.
6198 * The newer usage is when assembling an array.
6199 * In this case raid_disks will be 0, and the major_version field is
6200 * use to determine which style super-blocks are to be found on the devices.
6201 * The minor and patch _version numbers are also kept incase the
6202 * super_block handler wishes to interpret them.
6204 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6207 if (info->raid_disks == 0) {
6208 /* just setting version number for superblock loading */
6209 if (info->major_version < 0 ||
6210 info->major_version >= ARRAY_SIZE(super_types) ||
6211 super_types[info->major_version].name == NULL) {
6212 /* maybe try to auto-load a module? */
6214 "md: superblock version %d not known\n",
6215 info->major_version);
6218 mddev->major_version = info->major_version;
6219 mddev->minor_version = info->minor_version;
6220 mddev->patch_version = info->patch_version;
6221 mddev->persistent = !info->not_persistent;
6222 /* ensure mddev_put doesn't delete this now that there
6223 * is some minimal configuration.
6225 mddev->ctime = get_seconds();
6228 mddev->major_version = MD_MAJOR_VERSION;
6229 mddev->minor_version = MD_MINOR_VERSION;
6230 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6231 mddev->ctime = get_seconds();
6233 mddev->level = info->level;
6234 mddev->clevel[0] = 0;
6235 mddev->dev_sectors = 2 * (sector_t)info->size;
6236 mddev->raid_disks = info->raid_disks;
6237 /* don't set md_minor, it is determined by which /dev/md* was
6240 if (info->state & (1<<MD_SB_CLEAN))
6241 mddev->recovery_cp = MaxSector;
6243 mddev->recovery_cp = 0;
6244 mddev->persistent = ! info->not_persistent;
6245 mddev->external = 0;
6247 mddev->layout = info->layout;
6248 mddev->chunk_sectors = info->chunk_size >> 9;
6250 mddev->max_disks = MD_SB_DISKS;
6252 if (mddev->persistent)
6254 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6256 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6257 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6258 mddev->bitmap_info.offset = 0;
6260 mddev->reshape_position = MaxSector;
6263 * Generate a 128 bit UUID
6265 get_random_bytes(mddev->uuid, 16);
6267 mddev->new_level = mddev->level;
6268 mddev->new_chunk_sectors = mddev->chunk_sectors;
6269 mddev->new_layout = mddev->layout;
6270 mddev->delta_disks = 0;
6271 mddev->reshape_backwards = 0;
6276 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6278 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6280 if (mddev->external_size)
6283 mddev->array_sectors = array_sectors;
6285 EXPORT_SYMBOL(md_set_array_sectors);
6287 static int update_size(struct mddev *mddev, sector_t num_sectors)
6289 struct md_rdev *rdev;
6291 int fit = (num_sectors == 0);
6293 if (mddev->pers->resize == NULL)
6295 /* The "num_sectors" is the number of sectors of each device that
6296 * is used. This can only make sense for arrays with redundancy.
6297 * linear and raid0 always use whatever space is available. We can only
6298 * consider changing this number if no resync or reconstruction is
6299 * happening, and if the new size is acceptable. It must fit before the
6300 * sb_start or, if that is <data_offset, it must fit before the size
6301 * of each device. If num_sectors is zero, we find the largest size
6304 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6310 rdev_for_each(rdev, mddev) {
6311 sector_t avail = rdev->sectors;
6313 if (fit && (num_sectors == 0 || num_sectors > avail))
6314 num_sectors = avail;
6315 if (avail < num_sectors)
6318 rv = mddev->pers->resize(mddev, num_sectors);
6320 revalidate_disk(mddev->gendisk);
6324 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6327 struct md_rdev *rdev;
6328 /* change the number of raid disks */
6329 if (mddev->pers->check_reshape == NULL)
6333 if (raid_disks <= 0 ||
6334 (mddev->max_disks && raid_disks >= mddev->max_disks))
6336 if (mddev->sync_thread ||
6337 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6338 mddev->reshape_position != MaxSector)
6341 rdev_for_each(rdev, mddev) {
6342 if (mddev->raid_disks < raid_disks &&
6343 rdev->data_offset < rdev->new_data_offset)
6345 if (mddev->raid_disks > raid_disks &&
6346 rdev->data_offset > rdev->new_data_offset)
6350 mddev->delta_disks = raid_disks - mddev->raid_disks;
6351 if (mddev->delta_disks < 0)
6352 mddev->reshape_backwards = 1;
6353 else if (mddev->delta_disks > 0)
6354 mddev->reshape_backwards = 0;
6356 rv = mddev->pers->check_reshape(mddev);
6358 mddev->delta_disks = 0;
6359 mddev->reshape_backwards = 0;
6365 * update_array_info is used to change the configuration of an
6367 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6368 * fields in the info are checked against the array.
6369 * Any differences that cannot be handled will cause an error.
6370 * Normally, only one change can be managed at a time.
6372 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6378 /* calculate expected state,ignoring low bits */
6379 if (mddev->bitmap && mddev->bitmap_info.offset)
6380 state |= (1 << MD_SB_BITMAP_PRESENT);
6382 if (mddev->major_version != info->major_version ||
6383 mddev->minor_version != info->minor_version ||
6384 /* mddev->patch_version != info->patch_version || */
6385 mddev->ctime != info->ctime ||
6386 mddev->level != info->level ||
6387 /* mddev->layout != info->layout || */
6388 mddev->persistent != !info->not_persistent ||
6389 mddev->chunk_sectors != info->chunk_size >> 9 ||
6390 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6391 ((state^info->state) & 0xfffffe00)
6394 /* Check there is only one change */
6395 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6397 if (mddev->raid_disks != info->raid_disks)
6399 if (mddev->layout != info->layout)
6401 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6408 if (mddev->layout != info->layout) {
6410 * we don't need to do anything at the md level, the
6411 * personality will take care of it all.
6413 if (mddev->pers->check_reshape == NULL)
6416 mddev->new_layout = info->layout;
6417 rv = mddev->pers->check_reshape(mddev);
6419 mddev->new_layout = mddev->layout;
6423 if (mddev_is_clustered(mddev))
6424 md_cluster_ops->metadata_update_start(mddev);
6425 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6426 rv = update_size(mddev, (sector_t)info->size * 2);
6428 if (mddev->raid_disks != info->raid_disks)
6429 rv = update_raid_disks(mddev, info->raid_disks);
6431 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6432 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6436 if (mddev->recovery || mddev->sync_thread) {
6440 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6441 struct bitmap *bitmap;
6442 /* add the bitmap */
6443 if (mddev->bitmap) {
6447 if (mddev->bitmap_info.default_offset == 0) {
6451 mddev->bitmap_info.offset =
6452 mddev->bitmap_info.default_offset;
6453 mddev->bitmap_info.space =
6454 mddev->bitmap_info.default_space;
6455 mddev->pers->quiesce(mddev, 1);
6456 bitmap = bitmap_create(mddev, -1);
6457 if (!IS_ERR(bitmap)) {
6458 mddev->bitmap = bitmap;
6459 rv = bitmap_load(mddev);
6461 rv = PTR_ERR(bitmap);
6463 bitmap_destroy(mddev);
6464 mddev->pers->quiesce(mddev, 0);
6466 /* remove the bitmap */
6467 if (!mddev->bitmap) {
6471 if (mddev->bitmap->storage.file) {
6475 mddev->pers->quiesce(mddev, 1);
6476 bitmap_destroy(mddev);
6477 mddev->pers->quiesce(mddev, 0);
6478 mddev->bitmap_info.offset = 0;
6481 md_update_sb(mddev, 1);
6482 if (mddev_is_clustered(mddev))
6483 md_cluster_ops->metadata_update_finish(mddev);
6486 if (mddev_is_clustered(mddev))
6487 md_cluster_ops->metadata_update_cancel(mddev);
6491 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6493 struct md_rdev *rdev;
6496 if (mddev->pers == NULL)
6500 rdev = find_rdev_rcu(mddev, dev);
6504 md_error(mddev, rdev);
6505 if (!test_bit(Faulty, &rdev->flags))
6513 * We have a problem here : there is no easy way to give a CHS
6514 * virtual geometry. We currently pretend that we have a 2 heads
6515 * 4 sectors (with a BIG number of cylinders...). This drives
6516 * dosfs just mad... ;-)
6518 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6520 struct mddev *mddev = bdev->bd_disk->private_data;
6524 geo->cylinders = mddev->array_sectors / 8;
6528 static inline bool md_ioctl_valid(unsigned int cmd)
6533 case GET_ARRAY_INFO:
6534 case GET_BITMAP_FILE:
6537 case HOT_REMOVE_DISK:
6540 case RESTART_ARRAY_RW:
6542 case SET_ARRAY_INFO:
6543 case SET_BITMAP_FILE:
6544 case SET_DISK_FAULTY:
6547 case CLUSTERED_DISK_NACK:
6554 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6555 unsigned int cmd, unsigned long arg)
6558 void __user *argp = (void __user *)arg;
6559 struct mddev *mddev = NULL;
6562 if (!md_ioctl_valid(cmd))
6567 case GET_ARRAY_INFO:
6571 if (!capable(CAP_SYS_ADMIN))
6576 * Commands dealing with the RAID driver but not any
6581 err = get_version(argp);
6587 autostart_arrays(arg);
6594 * Commands creating/starting a new array:
6597 mddev = bdev->bd_disk->private_data;
6604 /* Some actions do not requires the mutex */
6606 case GET_ARRAY_INFO:
6607 if (!mddev->raid_disks && !mddev->external)
6610 err = get_array_info(mddev, argp);
6614 if (!mddev->raid_disks && !mddev->external)
6617 err = get_disk_info(mddev, argp);
6620 case SET_DISK_FAULTY:
6621 err = set_disk_faulty(mddev, new_decode_dev(arg));
6624 case GET_BITMAP_FILE:
6625 err = get_bitmap_file(mddev, argp);
6630 if (cmd == ADD_NEW_DISK)
6631 /* need to ensure md_delayed_delete() has completed */
6632 flush_workqueue(md_misc_wq);
6634 if (cmd == HOT_REMOVE_DISK)
6635 /* need to ensure recovery thread has run */
6636 wait_event_interruptible_timeout(mddev->sb_wait,
6637 !test_bit(MD_RECOVERY_NEEDED,
6639 msecs_to_jiffies(5000));
6640 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6641 /* Need to flush page cache, and ensure no-one else opens
6644 mutex_lock(&mddev->open_mutex);
6645 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6646 mutex_unlock(&mddev->open_mutex);
6650 set_bit(MD_STILL_CLOSED, &mddev->flags);
6651 mutex_unlock(&mddev->open_mutex);
6652 sync_blockdev(bdev);
6654 err = mddev_lock(mddev);
6657 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6662 if (cmd == SET_ARRAY_INFO) {
6663 mdu_array_info_t info;
6665 memset(&info, 0, sizeof(info));
6666 else if (copy_from_user(&info, argp, sizeof(info))) {
6671 err = update_array_info(mddev, &info);
6673 printk(KERN_WARNING "md: couldn't update"
6674 " array info. %d\n", err);
6679 if (!list_empty(&mddev->disks)) {
6681 "md: array %s already has disks!\n",
6686 if (mddev->raid_disks) {
6688 "md: array %s already initialised!\n",
6693 err = set_array_info(mddev, &info);
6695 printk(KERN_WARNING "md: couldn't set"
6696 " array info. %d\n", err);
6703 * Commands querying/configuring an existing array:
6705 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6706 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6707 if ((!mddev->raid_disks && !mddev->external)
6708 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6709 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6710 && cmd != GET_BITMAP_FILE) {
6716 * Commands even a read-only array can execute:
6719 case RESTART_ARRAY_RW:
6720 err = restart_array(mddev);
6724 err = do_md_stop(mddev, 0, bdev);
6728 err = md_set_readonly(mddev, bdev);
6731 case HOT_REMOVE_DISK:
6732 err = hot_remove_disk(mddev, new_decode_dev(arg));
6736 /* We can support ADD_NEW_DISK on read-only arrays
6737 * on if we are re-adding a preexisting device.
6738 * So require mddev->pers and MD_DISK_SYNC.
6741 mdu_disk_info_t info;
6742 if (copy_from_user(&info, argp, sizeof(info)))
6744 else if (!(info.state & (1<<MD_DISK_SYNC)))
6745 /* Need to clear read-only for this */
6748 err = add_new_disk(mddev, &info);
6754 if (get_user(ro, (int __user *)(arg))) {
6760 /* if the bdev is going readonly the value of mddev->ro
6761 * does not matter, no writes are coming
6766 /* are we are already prepared for writes? */
6770 /* transitioning to readauto need only happen for
6771 * arrays that call md_write_start
6774 err = restart_array(mddev);
6777 set_disk_ro(mddev->gendisk, 0);
6784 * The remaining ioctls are changing the state of the
6785 * superblock, so we do not allow them on read-only arrays.
6787 if (mddev->ro && mddev->pers) {
6788 if (mddev->ro == 2) {
6790 sysfs_notify_dirent_safe(mddev->sysfs_state);
6791 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6792 /* mddev_unlock will wake thread */
6793 /* If a device failed while we were read-only, we
6794 * need to make sure the metadata is updated now.
6796 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6797 mddev_unlock(mddev);
6798 wait_event(mddev->sb_wait,
6799 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6800 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6801 mddev_lock_nointr(mddev);
6812 mdu_disk_info_t info;
6813 if (copy_from_user(&info, argp, sizeof(info)))
6816 err = add_new_disk(mddev, &info);
6820 case CLUSTERED_DISK_NACK:
6821 if (mddev_is_clustered(mddev))
6822 md_cluster_ops->new_disk_ack(mddev, false);
6828 err = hot_add_disk(mddev, new_decode_dev(arg));
6832 err = do_md_run(mddev);
6835 case SET_BITMAP_FILE:
6836 err = set_bitmap_file(mddev, (int)arg);
6845 if (mddev->hold_active == UNTIL_IOCTL &&
6847 mddev->hold_active = 0;
6848 mddev_unlock(mddev);
6852 #ifdef CONFIG_COMPAT
6853 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6854 unsigned int cmd, unsigned long arg)
6857 case HOT_REMOVE_DISK:
6859 case SET_DISK_FAULTY:
6860 case SET_BITMAP_FILE:
6861 /* These take in integer arg, do not convert */
6864 arg = (unsigned long)compat_ptr(arg);
6868 return md_ioctl(bdev, mode, cmd, arg);
6870 #endif /* CONFIG_COMPAT */
6872 static int md_open(struct block_device *bdev, fmode_t mode)
6875 * Succeed if we can lock the mddev, which confirms that
6876 * it isn't being stopped right now.
6878 struct mddev *mddev = mddev_find(bdev->bd_dev);
6884 if (mddev->gendisk != bdev->bd_disk) {
6885 /* we are racing with mddev_put which is discarding this
6889 /* Wait until bdev->bd_disk is definitely gone */
6890 flush_workqueue(md_misc_wq);
6891 /* Then retry the open from the top */
6892 return -ERESTARTSYS;
6894 BUG_ON(mddev != bdev->bd_disk->private_data);
6896 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6900 atomic_inc(&mddev->openers);
6901 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6902 mutex_unlock(&mddev->open_mutex);
6904 check_disk_change(bdev);
6909 static void md_release(struct gendisk *disk, fmode_t mode)
6911 struct mddev *mddev = disk->private_data;
6914 atomic_dec(&mddev->openers);
6918 static int md_media_changed(struct gendisk *disk)
6920 struct mddev *mddev = disk->private_data;
6922 return mddev->changed;
6925 static int md_revalidate(struct gendisk *disk)
6927 struct mddev *mddev = disk->private_data;
6932 static const struct block_device_operations md_fops =
6934 .owner = THIS_MODULE,
6936 .release = md_release,
6938 #ifdef CONFIG_COMPAT
6939 .compat_ioctl = md_compat_ioctl,
6941 .getgeo = md_getgeo,
6942 .media_changed = md_media_changed,
6943 .revalidate_disk= md_revalidate,
6946 static int md_thread(void *arg)
6948 struct md_thread *thread = arg;
6951 * md_thread is a 'system-thread', it's priority should be very
6952 * high. We avoid resource deadlocks individually in each
6953 * raid personality. (RAID5 does preallocation) We also use RR and
6954 * the very same RT priority as kswapd, thus we will never get
6955 * into a priority inversion deadlock.
6957 * we definitely have to have equal or higher priority than
6958 * bdflush, otherwise bdflush will deadlock if there are too
6959 * many dirty RAID5 blocks.
6962 allow_signal(SIGKILL);
6963 while (!kthread_should_stop()) {
6965 /* We need to wait INTERRUPTIBLE so that
6966 * we don't add to the load-average.
6967 * That means we need to be sure no signals are
6970 if (signal_pending(current))
6971 flush_signals(current);
6973 wait_event_interruptible_timeout
6975 test_bit(THREAD_WAKEUP, &thread->flags)
6976 || kthread_should_stop(),
6979 clear_bit(THREAD_WAKEUP, &thread->flags);
6980 if (!kthread_should_stop())
6981 thread->run(thread);
6987 void md_wakeup_thread(struct md_thread *thread)
6990 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6991 set_bit(THREAD_WAKEUP, &thread->flags);
6992 wake_up(&thread->wqueue);
6995 EXPORT_SYMBOL(md_wakeup_thread);
6997 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6998 struct mddev *mddev, const char *name)
7000 struct md_thread *thread;
7002 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7006 init_waitqueue_head(&thread->wqueue);
7009 thread->mddev = mddev;
7010 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7011 thread->tsk = kthread_run(md_thread, thread,
7013 mdname(thread->mddev),
7015 if (IS_ERR(thread->tsk)) {
7021 EXPORT_SYMBOL(md_register_thread);
7023 void md_unregister_thread(struct md_thread **threadp)
7025 struct md_thread *thread = *threadp;
7028 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7029 /* Locking ensures that mddev_unlock does not wake_up a
7030 * non-existent thread
7032 spin_lock(&pers_lock);
7034 spin_unlock(&pers_lock);
7036 kthread_stop(thread->tsk);
7039 EXPORT_SYMBOL(md_unregister_thread);
7041 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7043 if (!rdev || test_bit(Faulty, &rdev->flags))
7046 if (!mddev->pers || !mddev->pers->error_handler)
7048 mddev->pers->error_handler(mddev,rdev);
7049 if (mddev->degraded)
7050 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7051 sysfs_notify_dirent_safe(rdev->sysfs_state);
7052 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7053 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7054 md_wakeup_thread(mddev->thread);
7055 if (mddev->event_work.func)
7056 queue_work(md_misc_wq, &mddev->event_work);
7057 md_new_event_inintr(mddev);
7059 EXPORT_SYMBOL(md_error);
7061 /* seq_file implementation /proc/mdstat */
7063 static void status_unused(struct seq_file *seq)
7066 struct md_rdev *rdev;
7068 seq_printf(seq, "unused devices: ");
7070 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7071 char b[BDEVNAME_SIZE];
7073 seq_printf(seq, "%s ",
7074 bdevname(rdev->bdev,b));
7077 seq_printf(seq, "<none>");
7079 seq_printf(seq, "\n");
7082 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7084 sector_t max_sectors, resync, res;
7085 unsigned long dt, db;
7088 unsigned int per_milli;
7090 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7091 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7092 max_sectors = mddev->resync_max_sectors;
7094 max_sectors = mddev->dev_sectors;
7096 resync = mddev->curr_resync;
7098 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7099 /* Still cleaning up */
7100 resync = max_sectors;
7102 resync -= atomic_read(&mddev->recovery_active);
7105 if (mddev->recovery_cp < MaxSector) {
7106 seq_printf(seq, "\tresync=PENDING");
7112 seq_printf(seq, "\tresync=DELAYED");
7116 WARN_ON(max_sectors == 0);
7117 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7118 * in a sector_t, and (max_sectors>>scale) will fit in a
7119 * u32, as those are the requirements for sector_div.
7120 * Thus 'scale' must be at least 10
7123 if (sizeof(sector_t) > sizeof(unsigned long)) {
7124 while ( max_sectors/2 > (1ULL<<(scale+32)))
7127 res = (resync>>scale)*1000;
7128 sector_div(res, (u32)((max_sectors>>scale)+1));
7132 int i, x = per_milli/50, y = 20-x;
7133 seq_printf(seq, "[");
7134 for (i = 0; i < x; i++)
7135 seq_printf(seq, "=");
7136 seq_printf(seq, ">");
7137 for (i = 0; i < y; i++)
7138 seq_printf(seq, ".");
7139 seq_printf(seq, "] ");
7141 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7142 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7144 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7146 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7147 "resync" : "recovery"))),
7148 per_milli/10, per_milli % 10,
7149 (unsigned long long) resync/2,
7150 (unsigned long long) max_sectors/2);
7153 * dt: time from mark until now
7154 * db: blocks written from mark until now
7155 * rt: remaining time
7157 * rt is a sector_t, so could be 32bit or 64bit.
7158 * So we divide before multiply in case it is 32bit and close
7160 * We scale the divisor (db) by 32 to avoid losing precision
7161 * near the end of resync when the number of remaining sectors
7163 * We then divide rt by 32 after multiplying by db to compensate.
7164 * The '+1' avoids division by zero if db is very small.
7166 dt = ((jiffies - mddev->resync_mark) / HZ);
7168 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7169 - mddev->resync_mark_cnt;
7171 rt = max_sectors - resync; /* number of remaining sectors */
7172 sector_div(rt, db/32+1);
7176 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7177 ((unsigned long)rt % 60)/6);
7179 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7183 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7185 struct list_head *tmp;
7187 struct mddev *mddev;
7195 spin_lock(&all_mddevs_lock);
7196 list_for_each(tmp,&all_mddevs)
7198 mddev = list_entry(tmp, struct mddev, all_mddevs);
7200 spin_unlock(&all_mddevs_lock);
7203 spin_unlock(&all_mddevs_lock);
7205 return (void*)2;/* tail */
7209 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7211 struct list_head *tmp;
7212 struct mddev *next_mddev, *mddev = v;
7218 spin_lock(&all_mddevs_lock);
7220 tmp = all_mddevs.next;
7222 tmp = mddev->all_mddevs.next;
7223 if (tmp != &all_mddevs)
7224 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7226 next_mddev = (void*)2;
7229 spin_unlock(&all_mddevs_lock);
7237 static void md_seq_stop(struct seq_file *seq, void *v)
7239 struct mddev *mddev = v;
7241 if (mddev && v != (void*)1 && v != (void*)2)
7245 static int md_seq_show(struct seq_file *seq, void *v)
7247 struct mddev *mddev = v;
7249 struct md_rdev *rdev;
7251 if (v == (void*)1) {
7252 struct md_personality *pers;
7253 seq_printf(seq, "Personalities : ");
7254 spin_lock(&pers_lock);
7255 list_for_each_entry(pers, &pers_list, list)
7256 seq_printf(seq, "[%s] ", pers->name);
7258 spin_unlock(&pers_lock);
7259 seq_printf(seq, "\n");
7260 seq->poll_event = atomic_read(&md_event_count);
7263 if (v == (void*)2) {
7268 spin_lock(&mddev->lock);
7269 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7270 seq_printf(seq, "%s : %sactive", mdname(mddev),
7271 mddev->pers ? "" : "in");
7274 seq_printf(seq, " (read-only)");
7276 seq_printf(seq, " (auto-read-only)");
7277 seq_printf(seq, " %s", mddev->pers->name);
7282 rdev_for_each_rcu(rdev, mddev) {
7283 char b[BDEVNAME_SIZE];
7284 seq_printf(seq, " %s[%d]",
7285 bdevname(rdev->bdev,b), rdev->desc_nr);
7286 if (test_bit(WriteMostly, &rdev->flags))
7287 seq_printf(seq, "(W)");
7288 if (test_bit(Faulty, &rdev->flags)) {
7289 seq_printf(seq, "(F)");
7292 if (rdev->raid_disk < 0)
7293 seq_printf(seq, "(S)"); /* spare */
7294 if (test_bit(Replacement, &rdev->flags))
7295 seq_printf(seq, "(R)");
7296 sectors += rdev->sectors;
7300 if (!list_empty(&mddev->disks)) {
7302 seq_printf(seq, "\n %llu blocks",
7303 (unsigned long long)
7304 mddev->array_sectors / 2);
7306 seq_printf(seq, "\n %llu blocks",
7307 (unsigned long long)sectors / 2);
7309 if (mddev->persistent) {
7310 if (mddev->major_version != 0 ||
7311 mddev->minor_version != 90) {
7312 seq_printf(seq," super %d.%d",
7313 mddev->major_version,
7314 mddev->minor_version);
7316 } else if (mddev->external)
7317 seq_printf(seq, " super external:%s",
7318 mddev->metadata_type);
7320 seq_printf(seq, " super non-persistent");
7323 mddev->pers->status(seq, mddev);
7324 seq_printf(seq, "\n ");
7325 if (mddev->pers->sync_request) {
7326 if (status_resync(seq, mddev))
7327 seq_printf(seq, "\n ");
7330 seq_printf(seq, "\n ");
7332 bitmap_status(seq, mddev->bitmap);
7334 seq_printf(seq, "\n");
7336 spin_unlock(&mddev->lock);
7341 static const struct seq_operations md_seq_ops = {
7342 .start = md_seq_start,
7343 .next = md_seq_next,
7344 .stop = md_seq_stop,
7345 .show = md_seq_show,
7348 static int md_seq_open(struct inode *inode, struct file *file)
7350 struct seq_file *seq;
7353 error = seq_open(file, &md_seq_ops);
7357 seq = file->private_data;
7358 seq->poll_event = atomic_read(&md_event_count);
7362 static int md_unloading;
7363 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7365 struct seq_file *seq = filp->private_data;
7369 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7370 poll_wait(filp, &md_event_waiters, wait);
7372 /* always allow read */
7373 mask = POLLIN | POLLRDNORM;
7375 if (seq->poll_event != atomic_read(&md_event_count))
7376 mask |= POLLERR | POLLPRI;
7380 static const struct file_operations md_seq_fops = {
7381 .owner = THIS_MODULE,
7382 .open = md_seq_open,
7384 .llseek = seq_lseek,
7385 .release = seq_release_private,
7386 .poll = mdstat_poll,
7389 int register_md_personality(struct md_personality *p)
7391 printk(KERN_INFO "md: %s personality registered for level %d\n",
7393 spin_lock(&pers_lock);
7394 list_add_tail(&p->list, &pers_list);
7395 spin_unlock(&pers_lock);
7398 EXPORT_SYMBOL(register_md_personality);
7400 int unregister_md_personality(struct md_personality *p)
7402 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7403 spin_lock(&pers_lock);
7404 list_del_init(&p->list);
7405 spin_unlock(&pers_lock);
7408 EXPORT_SYMBOL(unregister_md_personality);
7410 int register_md_cluster_operations(struct md_cluster_operations *ops,
7411 struct module *module)
7414 spin_lock(&pers_lock);
7415 if (md_cluster_ops != NULL)
7418 md_cluster_ops = ops;
7419 md_cluster_mod = module;
7421 spin_unlock(&pers_lock);
7424 EXPORT_SYMBOL(register_md_cluster_operations);
7426 int unregister_md_cluster_operations(void)
7428 spin_lock(&pers_lock);
7429 md_cluster_ops = NULL;
7430 spin_unlock(&pers_lock);
7433 EXPORT_SYMBOL(unregister_md_cluster_operations);
7435 int md_setup_cluster(struct mddev *mddev, int nodes)
7439 err = request_module("md-cluster");
7441 pr_err("md-cluster module not found.\n");
7445 spin_lock(&pers_lock);
7446 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7447 spin_unlock(&pers_lock);
7450 spin_unlock(&pers_lock);
7452 return md_cluster_ops->join(mddev, nodes);
7455 void md_cluster_stop(struct mddev *mddev)
7457 if (!md_cluster_ops)
7459 md_cluster_ops->leave(mddev);
7460 module_put(md_cluster_mod);
7463 static int is_mddev_idle(struct mddev *mddev, int init)
7465 struct md_rdev *rdev;
7471 rdev_for_each_rcu(rdev, mddev) {
7472 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7473 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7474 (int)part_stat_read(&disk->part0, sectors[1]) -
7475 atomic_read(&disk->sync_io);
7476 /* sync IO will cause sync_io to increase before the disk_stats
7477 * as sync_io is counted when a request starts, and
7478 * disk_stats is counted when it completes.
7479 * So resync activity will cause curr_events to be smaller than
7480 * when there was no such activity.
7481 * non-sync IO will cause disk_stat to increase without
7482 * increasing sync_io so curr_events will (eventually)
7483 * be larger than it was before. Once it becomes
7484 * substantially larger, the test below will cause
7485 * the array to appear non-idle, and resync will slow
7487 * If there is a lot of outstanding resync activity when
7488 * we set last_event to curr_events, then all that activity
7489 * completing might cause the array to appear non-idle
7490 * and resync will be slowed down even though there might
7491 * not have been non-resync activity. This will only
7492 * happen once though. 'last_events' will soon reflect
7493 * the state where there is little or no outstanding
7494 * resync requests, and further resync activity will
7495 * always make curr_events less than last_events.
7498 if (init || curr_events - rdev->last_events > 64) {
7499 rdev->last_events = curr_events;
7507 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7509 /* another "blocks" (512byte) blocks have been synced */
7510 atomic_sub(blocks, &mddev->recovery_active);
7511 wake_up(&mddev->recovery_wait);
7513 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7514 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7515 md_wakeup_thread(mddev->thread);
7516 // stop recovery, signal do_sync ....
7519 EXPORT_SYMBOL(md_done_sync);
7521 /* md_write_start(mddev, bi)
7522 * If we need to update some array metadata (e.g. 'active' flag
7523 * in superblock) before writing, schedule a superblock update
7524 * and wait for it to complete.
7526 void md_write_start(struct mddev *mddev, struct bio *bi)
7529 if (bio_data_dir(bi) != WRITE)
7532 BUG_ON(mddev->ro == 1);
7533 if (mddev->ro == 2) {
7534 /* need to switch to read/write */
7536 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7537 md_wakeup_thread(mddev->thread);
7538 md_wakeup_thread(mddev->sync_thread);
7541 atomic_inc(&mddev->writes_pending);
7542 if (mddev->safemode == 1)
7543 mddev->safemode = 0;
7544 if (mddev->in_sync) {
7545 spin_lock(&mddev->lock);
7546 if (mddev->in_sync) {
7548 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7549 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7550 md_wakeup_thread(mddev->thread);
7553 spin_unlock(&mddev->lock);
7556 sysfs_notify_dirent_safe(mddev->sysfs_state);
7557 wait_event(mddev->sb_wait,
7558 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7560 EXPORT_SYMBOL(md_write_start);
7562 void md_write_end(struct mddev *mddev)
7564 if (atomic_dec_and_test(&mddev->writes_pending)) {
7565 if (mddev->safemode == 2)
7566 md_wakeup_thread(mddev->thread);
7567 else if (mddev->safemode_delay)
7568 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7571 EXPORT_SYMBOL(md_write_end);
7573 /* md_allow_write(mddev)
7574 * Calling this ensures that the array is marked 'active' so that writes
7575 * may proceed without blocking. It is important to call this before
7576 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7577 * Must be called with mddev_lock held.
7579 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7580 * is dropped, so return -EAGAIN after notifying userspace.
7582 int md_allow_write(struct mddev *mddev)
7588 if (!mddev->pers->sync_request)
7591 spin_lock(&mddev->lock);
7592 if (mddev->in_sync) {
7594 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7595 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7596 if (mddev->safemode_delay &&
7597 mddev->safemode == 0)
7598 mddev->safemode = 1;
7599 spin_unlock(&mddev->lock);
7600 if (mddev_is_clustered(mddev))
7601 md_cluster_ops->metadata_update_start(mddev);
7602 md_update_sb(mddev, 0);
7603 if (mddev_is_clustered(mddev))
7604 md_cluster_ops->metadata_update_finish(mddev);
7605 sysfs_notify_dirent_safe(mddev->sysfs_state);
7607 spin_unlock(&mddev->lock);
7609 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7614 EXPORT_SYMBOL_GPL(md_allow_write);
7616 #define SYNC_MARKS 10
7617 #define SYNC_MARK_STEP (3*HZ)
7618 #define UPDATE_FREQUENCY (5*60*HZ)
7619 void md_do_sync(struct md_thread *thread)
7621 struct mddev *mddev = thread->mddev;
7622 struct mddev *mddev2;
7623 unsigned int currspeed = 0,
7625 sector_t max_sectors,j, io_sectors, recovery_done;
7626 unsigned long mark[SYNC_MARKS];
7627 unsigned long update_time;
7628 sector_t mark_cnt[SYNC_MARKS];
7630 struct list_head *tmp;
7631 sector_t last_check;
7633 struct md_rdev *rdev;
7634 char *desc, *action = NULL;
7635 struct blk_plug plug;
7637 /* just incase thread restarts... */
7638 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7640 if (mddev->ro) {/* never try to sync a read-only array */
7641 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7645 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7646 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7647 desc = "data-check";
7649 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7650 desc = "requested-resync";
7654 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7659 mddev->last_sync_action = action ?: desc;
7661 /* we overload curr_resync somewhat here.
7662 * 0 == not engaged in resync at all
7663 * 2 == checking that there is no conflict with another sync
7664 * 1 == like 2, but have yielded to allow conflicting resync to
7666 * other == active in resync - this many blocks
7668 * Before starting a resync we must have set curr_resync to
7669 * 2, and then checked that every "conflicting" array has curr_resync
7670 * less than ours. When we find one that is the same or higher
7671 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7672 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7673 * This will mean we have to start checking from the beginning again.
7678 mddev->curr_resync = 2;
7681 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7683 for_each_mddev(mddev2, tmp) {
7684 if (mddev2 == mddev)
7686 if (!mddev->parallel_resync
7687 && mddev2->curr_resync
7688 && match_mddev_units(mddev, mddev2)) {
7690 if (mddev < mddev2 && mddev->curr_resync == 2) {
7691 /* arbitrarily yield */
7692 mddev->curr_resync = 1;
7693 wake_up(&resync_wait);
7695 if (mddev > mddev2 && mddev->curr_resync == 1)
7696 /* no need to wait here, we can wait the next
7697 * time 'round when curr_resync == 2
7700 /* We need to wait 'interruptible' so as not to
7701 * contribute to the load average, and not to
7702 * be caught by 'softlockup'
7704 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7705 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7706 mddev2->curr_resync >= mddev->curr_resync) {
7707 printk(KERN_INFO "md: delaying %s of %s"
7708 " until %s has finished (they"
7709 " share one or more physical units)\n",
7710 desc, mdname(mddev), mdname(mddev2));
7712 if (signal_pending(current))
7713 flush_signals(current);
7715 finish_wait(&resync_wait, &wq);
7718 finish_wait(&resync_wait, &wq);
7721 } while (mddev->curr_resync < 2);
7724 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7725 /* resync follows the size requested by the personality,
7726 * which defaults to physical size, but can be virtual size
7728 max_sectors = mddev->resync_max_sectors;
7729 atomic64_set(&mddev->resync_mismatches, 0);
7730 /* we don't use the checkpoint if there's a bitmap */
7731 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7732 j = mddev->resync_min;
7733 else if (!mddev->bitmap)
7734 j = mddev->recovery_cp;
7736 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7737 max_sectors = mddev->resync_max_sectors;
7739 /* recovery follows the physical size of devices */
7740 max_sectors = mddev->dev_sectors;
7743 rdev_for_each_rcu(rdev, mddev)
7744 if (rdev->raid_disk >= 0 &&
7745 !test_bit(Faulty, &rdev->flags) &&
7746 !test_bit(In_sync, &rdev->flags) &&
7747 rdev->recovery_offset < j)
7748 j = rdev->recovery_offset;
7751 /* If there is a bitmap, we need to make sure all
7752 * writes that started before we added a spare
7753 * complete before we start doing a recovery.
7754 * Otherwise the write might complete and (via
7755 * bitmap_endwrite) set a bit in the bitmap after the
7756 * recovery has checked that bit and skipped that
7759 if (mddev->bitmap) {
7760 mddev->pers->quiesce(mddev, 1);
7761 mddev->pers->quiesce(mddev, 0);
7765 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7766 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7767 " %d KB/sec/disk.\n", speed_min(mddev));
7768 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7769 "(but not more than %d KB/sec) for %s.\n",
7770 speed_max(mddev), desc);
7772 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7775 for (m = 0; m < SYNC_MARKS; m++) {
7777 mark_cnt[m] = io_sectors;
7780 mddev->resync_mark = mark[last_mark];
7781 mddev->resync_mark_cnt = mark_cnt[last_mark];
7784 * Tune reconstruction:
7786 window = 32*(PAGE_SIZE/512);
7787 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7788 window/2, (unsigned long long)max_sectors/2);
7790 atomic_set(&mddev->recovery_active, 0);
7795 "md: resuming %s of %s from checkpoint.\n",
7796 desc, mdname(mddev));
7797 mddev->curr_resync = j;
7799 mddev->curr_resync = 3; /* no longer delayed */
7800 mddev->curr_resync_completed = j;
7801 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7802 md_new_event(mddev);
7803 update_time = jiffies;
7805 if (mddev_is_clustered(mddev))
7806 md_cluster_ops->resync_start(mddev, j, max_sectors);
7808 blk_start_plug(&plug);
7809 while (j < max_sectors) {
7814 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7815 ((mddev->curr_resync > mddev->curr_resync_completed &&
7816 (mddev->curr_resync - mddev->curr_resync_completed)
7817 > (max_sectors >> 4)) ||
7818 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7819 (j - mddev->curr_resync_completed)*2
7820 >= mddev->resync_max - mddev->curr_resync_completed ||
7821 mddev->curr_resync_completed > mddev->resync_max
7823 /* time to update curr_resync_completed */
7824 wait_event(mddev->recovery_wait,
7825 atomic_read(&mddev->recovery_active) == 0);
7826 mddev->curr_resync_completed = j;
7827 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7828 j > mddev->recovery_cp)
7829 mddev->recovery_cp = j;
7830 update_time = jiffies;
7831 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7832 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7835 while (j >= mddev->resync_max &&
7836 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7837 /* As this condition is controlled by user-space,
7838 * we can block indefinitely, so use '_interruptible'
7839 * to avoid triggering warnings.
7841 flush_signals(current); /* just in case */
7842 wait_event_interruptible(mddev->recovery_wait,
7843 mddev->resync_max > j
7844 || test_bit(MD_RECOVERY_INTR,
7848 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7851 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7853 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7857 if (!skipped) { /* actual IO requested */
7858 io_sectors += sectors;
7859 atomic_add(sectors, &mddev->recovery_active);
7862 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7866 if (j > max_sectors)
7867 /* when skipping, extra large numbers can be returned. */
7870 mddev->curr_resync = j;
7871 if (mddev_is_clustered(mddev))
7872 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7873 mddev->curr_mark_cnt = io_sectors;
7874 if (last_check == 0)
7875 /* this is the earliest that rebuild will be
7876 * visible in /proc/mdstat
7878 md_new_event(mddev);
7880 if (last_check + window > io_sectors || j == max_sectors)
7883 last_check = io_sectors;
7885 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7887 int next = (last_mark+1) % SYNC_MARKS;
7889 mddev->resync_mark = mark[next];
7890 mddev->resync_mark_cnt = mark_cnt[next];
7891 mark[next] = jiffies;
7892 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7896 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7900 * this loop exits only if either when we are slower than
7901 * the 'hard' speed limit, or the system was IO-idle for
7903 * the system might be non-idle CPU-wise, but we only care
7904 * about not overloading the IO subsystem. (things like an
7905 * e2fsck being done on the RAID array should execute fast)
7909 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7910 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7911 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7913 if (currspeed > speed_min(mddev)) {
7914 if (currspeed > speed_max(mddev)) {
7918 if (!is_mddev_idle(mddev, 0)) {
7920 * Give other IO more of a chance.
7921 * The faster the devices, the less we wait.
7923 wait_event(mddev->recovery_wait,
7924 !atomic_read(&mddev->recovery_active));
7928 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7929 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7930 ? "interrupted" : "done");
7932 * this also signals 'finished resyncing' to md_stop
7934 blk_finish_plug(&plug);
7935 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7937 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7938 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7939 mddev->curr_resync > 2) {
7940 mddev->curr_resync_completed = mddev->curr_resync;
7941 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7943 /* tell personality that we are finished */
7944 mddev->pers->sync_request(mddev, max_sectors, &skipped);
7946 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7947 mddev->curr_resync > 2) {
7948 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7949 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7950 if (mddev->curr_resync >= mddev->recovery_cp) {
7952 "md: checkpointing %s of %s.\n",
7953 desc, mdname(mddev));
7954 if (test_bit(MD_RECOVERY_ERROR,
7956 mddev->recovery_cp =
7957 mddev->curr_resync_completed;
7959 mddev->recovery_cp =
7963 mddev->recovery_cp = MaxSector;
7965 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7966 mddev->curr_resync = MaxSector;
7968 rdev_for_each_rcu(rdev, mddev)
7969 if (rdev->raid_disk >= 0 &&
7970 mddev->delta_disks >= 0 &&
7971 !test_bit(Faulty, &rdev->flags) &&
7972 !test_bit(In_sync, &rdev->flags) &&
7973 rdev->recovery_offset < mddev->curr_resync)
7974 rdev->recovery_offset = mddev->curr_resync;
7979 if (mddev_is_clustered(mddev))
7980 md_cluster_ops->resync_finish(mddev);
7982 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7984 spin_lock(&mddev->lock);
7985 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7986 /* We completed so min/max setting can be forgotten if used. */
7987 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7988 mddev->resync_min = 0;
7989 mddev->resync_max = MaxSector;
7990 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7991 mddev->resync_min = mddev->curr_resync_completed;
7992 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7993 mddev->curr_resync = 0;
7994 spin_unlock(&mddev->lock);
7996 wake_up(&resync_wait);
7997 md_wakeup_thread(mddev->thread);
8000 EXPORT_SYMBOL_GPL(md_do_sync);
8002 static int remove_and_add_spares(struct mddev *mddev,
8003 struct md_rdev *this)
8005 struct md_rdev *rdev;
8009 rdev_for_each(rdev, mddev)
8010 if ((this == NULL || rdev == this) &&
8011 rdev->raid_disk >= 0 &&
8012 !test_bit(Blocked, &rdev->flags) &&
8013 (test_bit(Faulty, &rdev->flags) ||
8014 ! test_bit(In_sync, &rdev->flags)) &&
8015 atomic_read(&rdev->nr_pending)==0) {
8016 if (mddev->pers->hot_remove_disk(
8017 mddev, rdev) == 0) {
8018 sysfs_unlink_rdev(mddev, rdev);
8019 rdev->raid_disk = -1;
8023 if (removed && mddev->kobj.sd)
8024 sysfs_notify(&mddev->kobj, NULL, "degraded");
8029 rdev_for_each(rdev, mddev) {
8030 if (rdev->raid_disk >= 0 &&
8031 !test_bit(In_sync, &rdev->flags) &&
8032 !test_bit(Faulty, &rdev->flags))
8034 if (rdev->raid_disk >= 0)
8036 if (test_bit(Faulty, &rdev->flags))
8039 ! (rdev->saved_raid_disk >= 0 &&
8040 !test_bit(Bitmap_sync, &rdev->flags)))
8043 rdev->recovery_offset = 0;
8045 hot_add_disk(mddev, rdev) == 0) {
8046 if (sysfs_link_rdev(mddev, rdev))
8047 /* failure here is OK */;
8049 md_new_event(mddev);
8050 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8055 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8059 static void md_start_sync(struct work_struct *ws)
8061 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8063 mddev->sync_thread = md_register_thread(md_do_sync,
8066 if (!mddev->sync_thread) {
8067 printk(KERN_ERR "%s: could not start resync"
8070 /* leave the spares where they are, it shouldn't hurt */
8071 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8072 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8073 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8074 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8075 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8076 wake_up(&resync_wait);
8077 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8079 if (mddev->sysfs_action)
8080 sysfs_notify_dirent_safe(mddev->sysfs_action);
8082 md_wakeup_thread(mddev->sync_thread);
8083 sysfs_notify_dirent_safe(mddev->sysfs_action);
8084 md_new_event(mddev);
8088 * This routine is regularly called by all per-raid-array threads to
8089 * deal with generic issues like resync and super-block update.
8090 * Raid personalities that don't have a thread (linear/raid0) do not
8091 * need this as they never do any recovery or update the superblock.
8093 * It does not do any resync itself, but rather "forks" off other threads
8094 * to do that as needed.
8095 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8096 * "->recovery" and create a thread at ->sync_thread.
8097 * When the thread finishes it sets MD_RECOVERY_DONE
8098 * and wakeups up this thread which will reap the thread and finish up.
8099 * This thread also removes any faulty devices (with nr_pending == 0).
8101 * The overall approach is:
8102 * 1/ if the superblock needs updating, update it.
8103 * 2/ If a recovery thread is running, don't do anything else.
8104 * 3/ If recovery has finished, clean up, possibly marking spares active.
8105 * 4/ If there are any faulty devices, remove them.
8106 * 5/ If array is degraded, try to add spares devices
8107 * 6/ If array has spares or is not in-sync, start a resync thread.
8109 void md_check_recovery(struct mddev *mddev)
8111 if (mddev->suspended)
8115 bitmap_daemon_work(mddev);
8117 if (signal_pending(current)) {
8118 if (mddev->pers->sync_request && !mddev->external) {
8119 printk(KERN_INFO "md: %s in immediate safe mode\n",
8121 mddev->safemode = 2;
8123 flush_signals(current);
8126 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8129 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8130 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8131 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8132 (mddev->external == 0 && mddev->safemode == 1) ||
8133 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8134 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8138 if (mddev_trylock(mddev)) {
8142 struct md_rdev *rdev;
8143 if (!mddev->external && mddev->in_sync)
8144 /* 'Blocked' flag not needed as failed devices
8145 * will be recorded if array switched to read/write.
8146 * Leaving it set will prevent the device
8147 * from being removed.
8149 rdev_for_each(rdev, mddev)
8150 clear_bit(Blocked, &rdev->flags);
8151 /* On a read-only array we can:
8152 * - remove failed devices
8153 * - add already-in_sync devices if the array itself
8155 * As we only add devices that are already in-sync,
8156 * we can activate the spares immediately.
8158 remove_and_add_spares(mddev, NULL);
8159 /* There is no thread, but we need to call
8160 * ->spare_active and clear saved_raid_disk
8162 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8163 md_reap_sync_thread(mddev);
8164 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8165 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8166 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8170 if (!mddev->external) {
8172 spin_lock(&mddev->lock);
8173 if (mddev->safemode &&
8174 !atomic_read(&mddev->writes_pending) &&
8176 mddev->recovery_cp == MaxSector) {
8179 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8181 if (mddev->safemode == 1)
8182 mddev->safemode = 0;
8183 spin_unlock(&mddev->lock);
8185 sysfs_notify_dirent_safe(mddev->sysfs_state);
8188 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8189 if (mddev_is_clustered(mddev))
8190 md_cluster_ops->metadata_update_start(mddev);
8191 md_update_sb(mddev, 0);
8192 if (mddev_is_clustered(mddev))
8193 md_cluster_ops->metadata_update_finish(mddev);
8196 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8197 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8198 /* resync/recovery still happening */
8199 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8202 if (mddev->sync_thread) {
8203 md_reap_sync_thread(mddev);
8206 /* Set RUNNING before clearing NEEDED to avoid
8207 * any transients in the value of "sync_action".
8209 mddev->curr_resync_completed = 0;
8210 spin_lock(&mddev->lock);
8211 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8212 spin_unlock(&mddev->lock);
8213 /* Clear some bits that don't mean anything, but
8216 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8217 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8219 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8220 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8222 /* no recovery is running.
8223 * remove any failed drives, then
8224 * add spares if possible.
8225 * Spares are also removed and re-added, to allow
8226 * the personality to fail the re-add.
8229 if (mddev->reshape_position != MaxSector) {
8230 if (mddev->pers->check_reshape == NULL ||
8231 mddev->pers->check_reshape(mddev) != 0)
8232 /* Cannot proceed */
8234 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8235 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8236 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8237 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8238 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8239 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8240 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8241 } else if (mddev->recovery_cp < MaxSector) {
8242 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8243 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8244 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8245 /* nothing to be done ... */
8248 if (mddev->pers->sync_request) {
8250 /* We are adding a device or devices to an array
8251 * which has the bitmap stored on all devices.
8252 * So make sure all bitmap pages get written
8254 bitmap_write_all(mddev->bitmap);
8256 INIT_WORK(&mddev->del_work, md_start_sync);
8257 queue_work(md_misc_wq, &mddev->del_work);
8261 if (!mddev->sync_thread) {
8262 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8263 wake_up(&resync_wait);
8264 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8266 if (mddev->sysfs_action)
8267 sysfs_notify_dirent_safe(mddev->sysfs_action);
8270 wake_up(&mddev->sb_wait);
8271 mddev_unlock(mddev);
8274 EXPORT_SYMBOL(md_check_recovery);
8276 void md_reap_sync_thread(struct mddev *mddev)
8278 struct md_rdev *rdev;
8280 /* resync has finished, collect result */
8281 md_unregister_thread(&mddev->sync_thread);
8282 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8283 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8285 /* activate any spares */
8286 if (mddev->pers->spare_active(mddev)) {
8287 sysfs_notify(&mddev->kobj, NULL,
8289 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8292 if (mddev_is_clustered(mddev))
8293 md_cluster_ops->metadata_update_start(mddev);
8294 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8295 mddev->pers->finish_reshape)
8296 mddev->pers->finish_reshape(mddev);
8298 /* If array is no-longer degraded, then any saved_raid_disk
8299 * information must be scrapped.
8301 if (!mddev->degraded)
8302 rdev_for_each(rdev, mddev)
8303 rdev->saved_raid_disk = -1;
8305 md_update_sb(mddev, 1);
8306 if (mddev_is_clustered(mddev))
8307 md_cluster_ops->metadata_update_finish(mddev);
8308 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8309 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8310 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8311 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8312 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8313 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8314 wake_up(&resync_wait);
8315 /* flag recovery needed just to double check */
8316 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8317 sysfs_notify_dirent_safe(mddev->sysfs_action);
8318 md_new_event(mddev);
8319 if (mddev->event_work.func)
8320 queue_work(md_misc_wq, &mddev->event_work);
8322 EXPORT_SYMBOL(md_reap_sync_thread);
8324 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8326 sysfs_notify_dirent_safe(rdev->sysfs_state);
8327 wait_event_timeout(rdev->blocked_wait,
8328 !test_bit(Blocked, &rdev->flags) &&
8329 !test_bit(BlockedBadBlocks, &rdev->flags),
8330 msecs_to_jiffies(5000));
8331 rdev_dec_pending(rdev, mddev);
8333 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8335 void md_finish_reshape(struct mddev *mddev)
8337 /* called be personality module when reshape completes. */
8338 struct md_rdev *rdev;
8340 rdev_for_each(rdev, mddev) {
8341 if (rdev->data_offset > rdev->new_data_offset)
8342 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8344 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8345 rdev->data_offset = rdev->new_data_offset;
8348 EXPORT_SYMBOL(md_finish_reshape);
8350 /* Bad block management.
8351 * We can record which blocks on each device are 'bad' and so just
8352 * fail those blocks, or that stripe, rather than the whole device.
8353 * Entries in the bad-block table are 64bits wide. This comprises:
8354 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8355 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8356 * A 'shift' can be set so that larger blocks are tracked and
8357 * consequently larger devices can be covered.
8358 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8360 * Locking of the bad-block table uses a seqlock so md_is_badblock
8361 * might need to retry if it is very unlucky.
8362 * We will sometimes want to check for bad blocks in a bi_end_io function,
8363 * so we use the write_seqlock_irq variant.
8365 * When looking for a bad block we specify a range and want to
8366 * know if any block in the range is bad. So we binary-search
8367 * to the last range that starts at-or-before the given endpoint,
8368 * (or "before the sector after the target range")
8369 * then see if it ends after the given start.
8371 * 0 if there are no known bad blocks in the range
8372 * 1 if there are known bad block which are all acknowledged
8373 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8374 * plus the start/length of the first bad section we overlap.
8376 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8377 sector_t *first_bad, int *bad_sectors)
8383 sector_t target = s + sectors;
8386 if (bb->shift > 0) {
8387 /* round the start down, and the end up */
8389 target += (1<<bb->shift) - 1;
8390 target >>= bb->shift;
8391 sectors = target - s;
8393 /* 'target' is now the first block after the bad range */
8396 seq = read_seqbegin(&bb->lock);
8401 /* Binary search between lo and hi for 'target'
8402 * i.e. for the last range that starts before 'target'
8404 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8405 * are known not to be the last range before target.
8406 * VARIANT: hi-lo is the number of possible
8407 * ranges, and decreases until it reaches 1
8409 while (hi - lo > 1) {
8410 int mid = (lo + hi) / 2;
8411 sector_t a = BB_OFFSET(p[mid]);
8413 /* This could still be the one, earlier ranges
8417 /* This and later ranges are definitely out. */
8420 /* 'lo' might be the last that started before target, but 'hi' isn't */
8422 /* need to check all range that end after 's' to see if
8423 * any are unacknowledged.
8426 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8427 if (BB_OFFSET(p[lo]) < target) {
8428 /* starts before the end, and finishes after
8429 * the start, so they must overlap
8431 if (rv != -1 && BB_ACK(p[lo]))
8435 *first_bad = BB_OFFSET(p[lo]);
8436 *bad_sectors = BB_LEN(p[lo]);
8442 if (read_seqretry(&bb->lock, seq))
8447 EXPORT_SYMBOL_GPL(md_is_badblock);
8450 * Add a range of bad blocks to the table.
8451 * This might extend the table, or might contract it
8452 * if two adjacent ranges can be merged.
8453 * We binary-search to find the 'insertion' point, then
8454 * decide how best to handle it.
8456 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8462 unsigned long flags;
8465 /* badblocks are disabled */
8469 /* round the start down, and the end up */
8470 sector_t next = s + sectors;
8472 next += (1<<bb->shift) - 1;
8477 write_seqlock_irqsave(&bb->lock, flags);
8482 /* Find the last range that starts at-or-before 's' */
8483 while (hi - lo > 1) {
8484 int mid = (lo + hi) / 2;
8485 sector_t a = BB_OFFSET(p[mid]);
8491 if (hi > lo && BB_OFFSET(p[lo]) > s)
8495 /* we found a range that might merge with the start
8498 sector_t a = BB_OFFSET(p[lo]);
8499 sector_t e = a + BB_LEN(p[lo]);
8500 int ack = BB_ACK(p[lo]);
8502 /* Yes, we can merge with a previous range */
8503 if (s == a && s + sectors >= e)
8504 /* new range covers old */
8507 ack = ack && acknowledged;
8509 if (e < s + sectors)
8511 if (e - a <= BB_MAX_LEN) {
8512 p[lo] = BB_MAKE(a, e-a, ack);
8515 /* does not all fit in one range,
8516 * make p[lo] maximal
8518 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8519 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8525 if (sectors && hi < bb->count) {
8526 /* 'hi' points to the first range that starts after 's'.
8527 * Maybe we can merge with the start of that range */
8528 sector_t a = BB_OFFSET(p[hi]);
8529 sector_t e = a + BB_LEN(p[hi]);
8530 int ack = BB_ACK(p[hi]);
8531 if (a <= s + sectors) {
8532 /* merging is possible */
8533 if (e <= s + sectors) {
8538 ack = ack && acknowledged;
8541 if (e - a <= BB_MAX_LEN) {
8542 p[hi] = BB_MAKE(a, e-a, ack);
8545 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8553 if (sectors == 0 && hi < bb->count) {
8554 /* we might be able to combine lo and hi */
8555 /* Note: 's' is at the end of 'lo' */
8556 sector_t a = BB_OFFSET(p[hi]);
8557 int lolen = BB_LEN(p[lo]);
8558 int hilen = BB_LEN(p[hi]);
8559 int newlen = lolen + hilen - (s - a);
8560 if (s >= a && newlen < BB_MAX_LEN) {
8561 /* yes, we can combine them */
8562 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8563 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8564 memmove(p + hi, p + hi + 1,
8565 (bb->count - hi - 1) * 8);
8570 /* didn't merge (it all).
8571 * Need to add a range just before 'hi' */
8572 if (bb->count >= MD_MAX_BADBLOCKS) {
8573 /* No room for more */
8577 int this_sectors = sectors;
8578 memmove(p + hi + 1, p + hi,
8579 (bb->count - hi) * 8);
8582 if (this_sectors > BB_MAX_LEN)
8583 this_sectors = BB_MAX_LEN;
8584 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8585 sectors -= this_sectors;
8592 bb->unacked_exist = 1;
8593 write_sequnlock_irqrestore(&bb->lock, flags);
8598 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8603 s += rdev->new_data_offset;
8605 s += rdev->data_offset;
8606 rv = md_set_badblocks(&rdev->badblocks,
8609 /* Make sure they get written out promptly */
8610 sysfs_notify_dirent_safe(rdev->sysfs_state);
8611 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8612 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8613 md_wakeup_thread(rdev->mddev->thread);
8617 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8620 * Remove a range of bad blocks from the table.
8621 * This may involve extending the table if we spilt a region,
8622 * but it must not fail. So if the table becomes full, we just
8623 * drop the remove request.
8625 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8629 sector_t target = s + sectors;
8632 if (bb->shift > 0) {
8633 /* When clearing we round the start up and the end down.
8634 * This should not matter as the shift should align with
8635 * the block size and no rounding should ever be needed.
8636 * However it is better the think a block is bad when it
8637 * isn't than to think a block is not bad when it is.
8639 s += (1<<bb->shift) - 1;
8641 target >>= bb->shift;
8642 sectors = target - s;
8645 write_seqlock_irq(&bb->lock);
8650 /* Find the last range that starts before 'target' */
8651 while (hi - lo > 1) {
8652 int mid = (lo + hi) / 2;
8653 sector_t a = BB_OFFSET(p[mid]);
8660 /* p[lo] is the last range that could overlap the
8661 * current range. Earlier ranges could also overlap,
8662 * but only this one can overlap the end of the range.
8664 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8665 /* Partial overlap, leave the tail of this range */
8666 int ack = BB_ACK(p[lo]);
8667 sector_t a = BB_OFFSET(p[lo]);
8668 sector_t end = a + BB_LEN(p[lo]);
8671 /* we need to split this range */
8672 if (bb->count >= MD_MAX_BADBLOCKS) {
8676 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8678 p[lo] = BB_MAKE(a, s-a, ack);
8681 p[lo] = BB_MAKE(target, end - target, ack);
8682 /* there is no longer an overlap */
8687 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8688 /* This range does overlap */
8689 if (BB_OFFSET(p[lo]) < s) {
8690 /* Keep the early parts of this range. */
8691 int ack = BB_ACK(p[lo]);
8692 sector_t start = BB_OFFSET(p[lo]);
8693 p[lo] = BB_MAKE(start, s - start, ack);
8694 /* now low doesn't overlap, so.. */
8699 /* 'lo' is strictly before, 'hi' is strictly after,
8700 * anything between needs to be discarded
8703 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8704 bb->count -= (hi - lo - 1);
8710 write_sequnlock_irq(&bb->lock);
8714 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8718 s += rdev->new_data_offset;
8720 s += rdev->data_offset;
8721 return md_clear_badblocks(&rdev->badblocks,
8724 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8727 * Acknowledge all bad blocks in a list.
8728 * This only succeeds if ->changed is clear. It is used by
8729 * in-kernel metadata updates
8731 void md_ack_all_badblocks(struct badblocks *bb)
8733 if (bb->page == NULL || bb->changed)
8734 /* no point even trying */
8736 write_seqlock_irq(&bb->lock);
8738 if (bb->changed == 0 && bb->unacked_exist) {
8741 for (i = 0; i < bb->count ; i++) {
8742 if (!BB_ACK(p[i])) {
8743 sector_t start = BB_OFFSET(p[i]);
8744 int len = BB_LEN(p[i]);
8745 p[i] = BB_MAKE(start, len, 1);
8748 bb->unacked_exist = 0;
8750 write_sequnlock_irq(&bb->lock);
8752 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8754 /* sysfs access to bad-blocks list.
8755 * We present two files.
8756 * 'bad-blocks' lists sector numbers and lengths of ranges that
8757 * are recorded as bad. The list is truncated to fit within
8758 * the one-page limit of sysfs.
8759 * Writing "sector length" to this file adds an acknowledged
8761 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8762 * been acknowledged. Writing to this file adds bad blocks
8763 * without acknowledging them. This is largely for testing.
8767 badblocks_show(struct badblocks *bb, char *page, int unack)
8778 seq = read_seqbegin(&bb->lock);
8783 while (len < PAGE_SIZE && i < bb->count) {
8784 sector_t s = BB_OFFSET(p[i]);
8785 unsigned int length = BB_LEN(p[i]);
8786 int ack = BB_ACK(p[i]);
8792 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8793 (unsigned long long)s << bb->shift,
8794 length << bb->shift);
8796 if (unack && len == 0)
8797 bb->unacked_exist = 0;
8799 if (read_seqretry(&bb->lock, seq))
8808 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8810 unsigned long long sector;
8814 /* Allow clearing via sysfs *only* for testing/debugging.
8815 * Normally only a successful write may clear a badblock
8818 if (page[0] == '-') {
8822 #endif /* DO_DEBUG */
8824 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8826 if (newline != '\n')
8838 md_clear_badblocks(bb, sector, length);
8841 #endif /* DO_DEBUG */
8842 if (md_set_badblocks(bb, sector, length, !unack))
8848 static int md_notify_reboot(struct notifier_block *this,
8849 unsigned long code, void *x)
8851 struct list_head *tmp;
8852 struct mddev *mddev;
8855 for_each_mddev(mddev, tmp) {
8856 if (mddev_trylock(mddev)) {
8858 __md_stop_writes(mddev);
8859 if (mddev->persistent)
8860 mddev->safemode = 2;
8861 mddev_unlock(mddev);
8866 * certain more exotic SCSI devices are known to be
8867 * volatile wrt too early system reboots. While the
8868 * right place to handle this issue is the given
8869 * driver, we do want to have a safe RAID driver ...
8877 static struct notifier_block md_notifier = {
8878 .notifier_call = md_notify_reboot,
8880 .priority = INT_MAX, /* before any real devices */
8883 static void md_geninit(void)
8885 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8887 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8890 static int __init md_init(void)
8894 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8898 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8902 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8905 if ((ret = register_blkdev(0, "mdp")) < 0)
8909 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8910 md_probe, NULL, NULL);
8911 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8912 md_probe, NULL, NULL);
8914 register_reboot_notifier(&md_notifier);
8915 raid_table_header = register_sysctl_table(raid_root_table);
8921 unregister_blkdev(MD_MAJOR, "md");
8923 destroy_workqueue(md_misc_wq);
8925 destroy_workqueue(md_wq);
8930 void md_reload_sb(struct mddev *mddev)
8932 struct md_rdev *rdev, *tmp;
8934 rdev_for_each_safe(rdev, tmp, mddev) {
8935 rdev->sb_loaded = 0;
8936 ClearPageUptodate(rdev->sb_page);
8938 mddev->raid_disks = 0;
8940 rdev_for_each_safe(rdev, tmp, mddev) {
8941 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8942 /* since we don't write to faulty devices, we figure out if the
8943 * disk is faulty by comparing events
8945 if (mddev->events > sb->events)
8946 set_bit(Faulty, &rdev->flags);
8950 EXPORT_SYMBOL(md_reload_sb);
8955 * Searches all registered partitions for autorun RAID arrays
8959 static LIST_HEAD(all_detected_devices);
8960 struct detected_devices_node {
8961 struct list_head list;
8965 void md_autodetect_dev(dev_t dev)
8967 struct detected_devices_node *node_detected_dev;
8969 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8970 if (node_detected_dev) {
8971 node_detected_dev->dev = dev;
8972 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8974 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8975 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8979 static void autostart_arrays(int part)
8981 struct md_rdev *rdev;
8982 struct detected_devices_node *node_detected_dev;
8984 int i_scanned, i_passed;
8989 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8991 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8993 node_detected_dev = list_entry(all_detected_devices.next,
8994 struct detected_devices_node, list);
8995 list_del(&node_detected_dev->list);
8996 dev = node_detected_dev->dev;
8997 kfree(node_detected_dev);
8998 rdev = md_import_device(dev,0, 90);
9002 if (test_bit(Faulty, &rdev->flags))
9005 set_bit(AutoDetected, &rdev->flags);
9006 list_add(&rdev->same_set, &pending_raid_disks);
9010 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9011 i_scanned, i_passed);
9013 autorun_devices(part);
9016 #endif /* !MODULE */
9018 static __exit void md_exit(void)
9020 struct mddev *mddev;
9021 struct list_head *tmp;
9024 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9025 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9027 unregister_blkdev(MD_MAJOR,"md");
9028 unregister_blkdev(mdp_major, "mdp");
9029 unregister_reboot_notifier(&md_notifier);
9030 unregister_sysctl_table(raid_table_header);
9032 /* We cannot unload the modules while some process is
9033 * waiting for us in select() or poll() - wake them up
9036 while (waitqueue_active(&md_event_waiters)) {
9037 /* not safe to leave yet */
9038 wake_up(&md_event_waiters);
9042 remove_proc_entry("mdstat", NULL);
9044 for_each_mddev(mddev, tmp) {
9045 export_array(mddev);
9046 mddev->hold_active = 0;
9048 destroy_workqueue(md_misc_wq);
9049 destroy_workqueue(md_wq);
9052 subsys_initcall(md_init);
9053 module_exit(md_exit)
9055 static int get_ro(char *buffer, struct kernel_param *kp)
9057 return sprintf(buffer, "%d", start_readonly);
9059 static int set_ro(const char *val, struct kernel_param *kp)
9061 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9064 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9065 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9066 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9068 MODULE_LICENSE("GPL");
9069 MODULE_DESCRIPTION("MD RAID framework");
9071 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);