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 void mddev_init(struct mddev *mddev)
488 mutex_init(&mddev->open_mutex);
489 mutex_init(&mddev->reconfig_mutex);
490 mutex_init(&mddev->bitmap_info.mutex);
491 INIT_LIST_HEAD(&mddev->disks);
492 INIT_LIST_HEAD(&mddev->all_mddevs);
493 init_timer(&mddev->safemode_timer);
494 atomic_set(&mddev->active, 1);
495 atomic_set(&mddev->openers, 0);
496 atomic_set(&mddev->active_io, 0);
497 spin_lock_init(&mddev->lock);
498 atomic_set(&mddev->flush_pending, 0);
499 init_waitqueue_head(&mddev->sb_wait);
500 init_waitqueue_head(&mddev->recovery_wait);
501 mddev->reshape_position = MaxSector;
502 mddev->reshape_backwards = 0;
503 mddev->last_sync_action = "none";
504 mddev->resync_min = 0;
505 mddev->resync_max = MaxSector;
506 mddev->level = LEVEL_NONE;
508 EXPORT_SYMBOL_GPL(mddev_init);
510 static struct mddev *mddev_find(dev_t unit)
512 struct mddev *mddev, *new = NULL;
514 if (unit && MAJOR(unit) != MD_MAJOR)
515 unit &= ~((1<<MdpMinorShift)-1);
518 spin_lock(&all_mddevs_lock);
521 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
522 if (mddev->unit == unit) {
524 spin_unlock(&all_mddevs_lock);
530 list_add(&new->all_mddevs, &all_mddevs);
531 spin_unlock(&all_mddevs_lock);
532 new->hold_active = UNTIL_IOCTL;
536 /* find an unused unit number */
537 static int next_minor = 512;
538 int start = next_minor;
542 dev = MKDEV(MD_MAJOR, next_minor);
544 if (next_minor > MINORMASK)
546 if (next_minor == start) {
547 /* Oh dear, all in use. */
548 spin_unlock(&all_mddevs_lock);
554 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
555 if (mddev->unit == dev) {
561 new->md_minor = MINOR(dev);
562 new->hold_active = UNTIL_STOP;
563 list_add(&new->all_mddevs, &all_mddevs);
564 spin_unlock(&all_mddevs_lock);
567 spin_unlock(&all_mddevs_lock);
569 new = kzalloc(sizeof(*new), GFP_KERNEL);
574 if (MAJOR(unit) == MD_MAJOR)
575 new->md_minor = MINOR(unit);
577 new->md_minor = MINOR(unit) >> MdpMinorShift;
584 static struct attribute_group md_redundancy_group;
586 void mddev_unlock(struct mddev *mddev)
588 if (mddev->to_remove) {
589 /* These cannot be removed under reconfig_mutex as
590 * an access to the files will try to take reconfig_mutex
591 * while holding the file unremovable, which leads to
593 * So hold set sysfs_active while the remove in happeing,
594 * and anything else which might set ->to_remove or my
595 * otherwise change the sysfs namespace will fail with
596 * -EBUSY if sysfs_active is still set.
597 * We set sysfs_active under reconfig_mutex and elsewhere
598 * test it under the same mutex to ensure its correct value
601 struct attribute_group *to_remove = mddev->to_remove;
602 mddev->to_remove = NULL;
603 mddev->sysfs_active = 1;
604 mutex_unlock(&mddev->reconfig_mutex);
606 if (mddev->kobj.sd) {
607 if (to_remove != &md_redundancy_group)
608 sysfs_remove_group(&mddev->kobj, to_remove);
609 if (mddev->pers == NULL ||
610 mddev->pers->sync_request == NULL) {
611 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
612 if (mddev->sysfs_action)
613 sysfs_put(mddev->sysfs_action);
614 mddev->sysfs_action = NULL;
617 mddev->sysfs_active = 0;
619 mutex_unlock(&mddev->reconfig_mutex);
621 /* As we've dropped the mutex we need a spinlock to
622 * make sure the thread doesn't disappear
624 spin_lock(&pers_lock);
625 md_wakeup_thread(mddev->thread);
626 spin_unlock(&pers_lock);
628 EXPORT_SYMBOL_GPL(mddev_unlock);
630 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
632 struct md_rdev *rdev;
634 rdev_for_each_rcu(rdev, mddev)
635 if (rdev->desc_nr == nr)
640 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
642 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
644 struct md_rdev *rdev;
646 rdev_for_each(rdev, mddev)
647 if (rdev->bdev->bd_dev == dev)
653 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
655 struct md_rdev *rdev;
657 rdev_for_each_rcu(rdev, mddev)
658 if (rdev->bdev->bd_dev == dev)
664 static struct md_personality *find_pers(int level, char *clevel)
666 struct md_personality *pers;
667 list_for_each_entry(pers, &pers_list, list) {
668 if (level != LEVEL_NONE && pers->level == level)
670 if (strcmp(pers->name, clevel)==0)
676 /* return the offset of the super block in 512byte sectors */
677 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
679 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
680 return MD_NEW_SIZE_SECTORS(num_sectors);
683 static int alloc_disk_sb(struct md_rdev *rdev)
685 rdev->sb_page = alloc_page(GFP_KERNEL);
686 if (!rdev->sb_page) {
687 printk(KERN_ALERT "md: out of memory.\n");
694 void md_rdev_clear(struct md_rdev *rdev)
697 put_page(rdev->sb_page);
699 rdev->sb_page = NULL;
704 put_page(rdev->bb_page);
705 rdev->bb_page = NULL;
707 kfree(rdev->badblocks.page);
708 rdev->badblocks.page = NULL;
710 EXPORT_SYMBOL_GPL(md_rdev_clear);
712 static void super_written(struct bio *bio)
714 struct md_rdev *rdev = bio->bi_private;
715 struct mddev *mddev = rdev->mddev;
718 printk("md: super_written gets error=%d\n", bio->bi_error);
719 md_error(mddev, rdev);
722 if (atomic_dec_and_test(&mddev->pending_writes))
723 wake_up(&mddev->sb_wait);
727 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
728 sector_t sector, int size, struct page *page)
730 /* write first size bytes of page to sector of rdev
731 * Increment mddev->pending_writes before returning
732 * and decrement it on completion, waking up sb_wait
733 * if zero is reached.
734 * If an error occurred, call md_error
736 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
738 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
739 bio->bi_iter.bi_sector = sector;
740 bio_add_page(bio, page, size, 0);
741 bio->bi_private = rdev;
742 bio->bi_end_io = super_written;
744 atomic_inc(&mddev->pending_writes);
745 submit_bio(WRITE_FLUSH_FUA, bio);
748 void md_super_wait(struct mddev *mddev)
750 /* wait for all superblock writes that were scheduled to complete */
751 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
754 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
755 struct page *page, int rw, bool metadata_op)
757 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
760 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
761 rdev->meta_bdev : rdev->bdev;
763 bio->bi_iter.bi_sector = sector + rdev->sb_start;
764 else if (rdev->mddev->reshape_position != MaxSector &&
765 (rdev->mddev->reshape_backwards ==
766 (sector >= rdev->mddev->reshape_position)))
767 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
769 bio->bi_iter.bi_sector = sector + rdev->data_offset;
770 bio_add_page(bio, page, size, 0);
771 submit_bio_wait(rw, bio);
773 ret = !bio->bi_error;
777 EXPORT_SYMBOL_GPL(sync_page_io);
779 static int read_disk_sb(struct md_rdev *rdev, int size)
781 char b[BDEVNAME_SIZE];
786 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
792 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
793 bdevname(rdev->bdev,b));
797 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
799 return sb1->set_uuid0 == sb2->set_uuid0 &&
800 sb1->set_uuid1 == sb2->set_uuid1 &&
801 sb1->set_uuid2 == sb2->set_uuid2 &&
802 sb1->set_uuid3 == sb2->set_uuid3;
805 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
808 mdp_super_t *tmp1, *tmp2;
810 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
811 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
813 if (!tmp1 || !tmp2) {
815 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
823 * nr_disks is not constant
828 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
835 static u32 md_csum_fold(u32 csum)
837 csum = (csum & 0xffff) + (csum >> 16);
838 return (csum & 0xffff) + (csum >> 16);
841 static unsigned int calc_sb_csum(mdp_super_t *sb)
844 u32 *sb32 = (u32*)sb;
846 unsigned int disk_csum, csum;
848 disk_csum = sb->sb_csum;
851 for (i = 0; i < MD_SB_BYTES/4 ; i++)
853 csum = (newcsum & 0xffffffff) + (newcsum>>32);
856 /* This used to use csum_partial, which was wrong for several
857 * reasons including that different results are returned on
858 * different architectures. It isn't critical that we get exactly
859 * the same return value as before (we always csum_fold before
860 * testing, and that removes any differences). However as we
861 * know that csum_partial always returned a 16bit value on
862 * alphas, do a fold to maximise conformity to previous behaviour.
864 sb->sb_csum = md_csum_fold(disk_csum);
866 sb->sb_csum = disk_csum;
872 * Handle superblock details.
873 * We want to be able to handle multiple superblock formats
874 * so we have a common interface to them all, and an array of
875 * different handlers.
876 * We rely on user-space to write the initial superblock, and support
877 * reading and updating of superblocks.
878 * Interface methods are:
879 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
880 * loads and validates a superblock on dev.
881 * if refdev != NULL, compare superblocks on both devices
883 * 0 - dev has a superblock that is compatible with refdev
884 * 1 - dev has a superblock that is compatible and newer than refdev
885 * so dev should be used as the refdev in future
886 * -EINVAL superblock incompatible or invalid
887 * -othererror e.g. -EIO
889 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
890 * Verify that dev is acceptable into mddev.
891 * The first time, mddev->raid_disks will be 0, and data from
892 * dev should be merged in. Subsequent calls check that dev
893 * is new enough. Return 0 or -EINVAL
895 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
896 * Update the superblock for rdev with data in mddev
897 * This does not write to disc.
903 struct module *owner;
904 int (*load_super)(struct md_rdev *rdev,
905 struct md_rdev *refdev,
907 int (*validate_super)(struct mddev *mddev,
908 struct md_rdev *rdev);
909 void (*sync_super)(struct mddev *mddev,
910 struct md_rdev *rdev);
911 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
912 sector_t num_sectors);
913 int (*allow_new_offset)(struct md_rdev *rdev,
914 unsigned long long new_offset);
918 * Check that the given mddev has no bitmap.
920 * This function is called from the run method of all personalities that do not
921 * support bitmaps. It prints an error message and returns non-zero if mddev
922 * has a bitmap. Otherwise, it returns 0.
925 int md_check_no_bitmap(struct mddev *mddev)
927 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
929 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
930 mdname(mddev), mddev->pers->name);
933 EXPORT_SYMBOL(md_check_no_bitmap);
936 * load_super for 0.90.0
938 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
940 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
945 * Calculate the position of the superblock (512byte sectors),
946 * it's at the end of the disk.
948 * It also happens to be a multiple of 4Kb.
950 rdev->sb_start = calc_dev_sboffset(rdev);
952 ret = read_disk_sb(rdev, MD_SB_BYTES);
957 bdevname(rdev->bdev, b);
958 sb = page_address(rdev->sb_page);
960 if (sb->md_magic != MD_SB_MAGIC) {
961 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
966 if (sb->major_version != 0 ||
967 sb->minor_version < 90 ||
968 sb->minor_version > 91) {
969 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
970 sb->major_version, sb->minor_version,
975 if (sb->raid_disks <= 0)
978 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
979 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
984 rdev->preferred_minor = sb->md_minor;
985 rdev->data_offset = 0;
986 rdev->new_data_offset = 0;
987 rdev->sb_size = MD_SB_BYTES;
988 rdev->badblocks.shift = -1;
990 if (sb->level == LEVEL_MULTIPATH)
993 rdev->desc_nr = sb->this_disk.number;
999 mdp_super_t *refsb = page_address(refdev->sb_page);
1000 if (!uuid_equal(refsb, sb)) {
1001 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1002 b, bdevname(refdev->bdev,b2));
1005 if (!sb_equal(refsb, sb)) {
1006 printk(KERN_WARNING "md: %s has same UUID"
1007 " but different superblock to %s\n",
1008 b, bdevname(refdev->bdev, b2));
1012 ev2 = md_event(refsb);
1018 rdev->sectors = rdev->sb_start;
1019 /* Limit to 4TB as metadata cannot record more than that.
1020 * (not needed for Linear and RAID0 as metadata doesn't
1023 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1024 rdev->sectors = (2ULL << 32) - 2;
1026 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1027 /* "this cannot possibly happen" ... */
1035 * validate_super for 0.90.0
1037 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1040 mdp_super_t *sb = page_address(rdev->sb_page);
1041 __u64 ev1 = md_event(sb);
1043 rdev->raid_disk = -1;
1044 clear_bit(Faulty, &rdev->flags);
1045 clear_bit(In_sync, &rdev->flags);
1046 clear_bit(Bitmap_sync, &rdev->flags);
1047 clear_bit(WriteMostly, &rdev->flags);
1049 if (mddev->raid_disks == 0) {
1050 mddev->major_version = 0;
1051 mddev->minor_version = sb->minor_version;
1052 mddev->patch_version = sb->patch_version;
1053 mddev->external = 0;
1054 mddev->chunk_sectors = sb->chunk_size >> 9;
1055 mddev->ctime = sb->ctime;
1056 mddev->utime = sb->utime;
1057 mddev->level = sb->level;
1058 mddev->clevel[0] = 0;
1059 mddev->layout = sb->layout;
1060 mddev->raid_disks = sb->raid_disks;
1061 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1062 mddev->events = ev1;
1063 mddev->bitmap_info.offset = 0;
1064 mddev->bitmap_info.space = 0;
1065 /* bitmap can use 60 K after the 4K superblocks */
1066 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1067 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1068 mddev->reshape_backwards = 0;
1070 if (mddev->minor_version >= 91) {
1071 mddev->reshape_position = sb->reshape_position;
1072 mddev->delta_disks = sb->delta_disks;
1073 mddev->new_level = sb->new_level;
1074 mddev->new_layout = sb->new_layout;
1075 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1076 if (mddev->delta_disks < 0)
1077 mddev->reshape_backwards = 1;
1079 mddev->reshape_position = MaxSector;
1080 mddev->delta_disks = 0;
1081 mddev->new_level = mddev->level;
1082 mddev->new_layout = mddev->layout;
1083 mddev->new_chunk_sectors = mddev->chunk_sectors;
1086 if (sb->state & (1<<MD_SB_CLEAN))
1087 mddev->recovery_cp = MaxSector;
1089 if (sb->events_hi == sb->cp_events_hi &&
1090 sb->events_lo == sb->cp_events_lo) {
1091 mddev->recovery_cp = sb->recovery_cp;
1093 mddev->recovery_cp = 0;
1096 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1097 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1098 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1099 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1101 mddev->max_disks = MD_SB_DISKS;
1103 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1104 mddev->bitmap_info.file == NULL) {
1105 mddev->bitmap_info.offset =
1106 mddev->bitmap_info.default_offset;
1107 mddev->bitmap_info.space =
1108 mddev->bitmap_info.default_space;
1111 } else if (mddev->pers == NULL) {
1112 /* Insist on good event counter while assembling, except
1113 * for spares (which don't need an event count) */
1115 if (sb->disks[rdev->desc_nr].state & (
1116 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1117 if (ev1 < mddev->events)
1119 } else if (mddev->bitmap) {
1120 /* if adding to array with a bitmap, then we can accept an
1121 * older device ... but not too old.
1123 if (ev1 < mddev->bitmap->events_cleared)
1125 if (ev1 < mddev->events)
1126 set_bit(Bitmap_sync, &rdev->flags);
1128 if (ev1 < mddev->events)
1129 /* just a hot-add of a new device, leave raid_disk at -1 */
1133 if (mddev->level != LEVEL_MULTIPATH) {
1134 desc = sb->disks + rdev->desc_nr;
1136 if (desc->state & (1<<MD_DISK_FAULTY))
1137 set_bit(Faulty, &rdev->flags);
1138 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1139 desc->raid_disk < mddev->raid_disks */) {
1140 set_bit(In_sync, &rdev->flags);
1141 rdev->raid_disk = desc->raid_disk;
1142 rdev->saved_raid_disk = desc->raid_disk;
1143 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1144 /* active but not in sync implies recovery up to
1145 * reshape position. We don't know exactly where
1146 * that is, so set to zero for now */
1147 if (mddev->minor_version >= 91) {
1148 rdev->recovery_offset = 0;
1149 rdev->raid_disk = desc->raid_disk;
1152 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1153 set_bit(WriteMostly, &rdev->flags);
1154 } else /* MULTIPATH are always insync */
1155 set_bit(In_sync, &rdev->flags);
1160 * sync_super for 0.90.0
1162 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1165 struct md_rdev *rdev2;
1166 int next_spare = mddev->raid_disks;
1168 /* make rdev->sb match mddev data..
1171 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1172 * 3/ any empty disks < next_spare become removed
1174 * disks[0] gets initialised to REMOVED because
1175 * we cannot be sure from other fields if it has
1176 * been initialised or not.
1179 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1181 rdev->sb_size = MD_SB_BYTES;
1183 sb = page_address(rdev->sb_page);
1185 memset(sb, 0, sizeof(*sb));
1187 sb->md_magic = MD_SB_MAGIC;
1188 sb->major_version = mddev->major_version;
1189 sb->patch_version = mddev->patch_version;
1190 sb->gvalid_words = 0; /* ignored */
1191 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1192 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1193 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1194 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1196 sb->ctime = mddev->ctime;
1197 sb->level = mddev->level;
1198 sb->size = mddev->dev_sectors / 2;
1199 sb->raid_disks = mddev->raid_disks;
1200 sb->md_minor = mddev->md_minor;
1201 sb->not_persistent = 0;
1202 sb->utime = mddev->utime;
1204 sb->events_hi = (mddev->events>>32);
1205 sb->events_lo = (u32)mddev->events;
1207 if (mddev->reshape_position == MaxSector)
1208 sb->minor_version = 90;
1210 sb->minor_version = 91;
1211 sb->reshape_position = mddev->reshape_position;
1212 sb->new_level = mddev->new_level;
1213 sb->delta_disks = mddev->delta_disks;
1214 sb->new_layout = mddev->new_layout;
1215 sb->new_chunk = mddev->new_chunk_sectors << 9;
1217 mddev->minor_version = sb->minor_version;
1220 sb->recovery_cp = mddev->recovery_cp;
1221 sb->cp_events_hi = (mddev->events>>32);
1222 sb->cp_events_lo = (u32)mddev->events;
1223 if (mddev->recovery_cp == MaxSector)
1224 sb->state = (1<< MD_SB_CLEAN);
1226 sb->recovery_cp = 0;
1228 sb->layout = mddev->layout;
1229 sb->chunk_size = mddev->chunk_sectors << 9;
1231 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1232 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1234 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1235 rdev_for_each(rdev2, mddev) {
1238 int is_active = test_bit(In_sync, &rdev2->flags);
1240 if (rdev2->raid_disk >= 0 &&
1241 sb->minor_version >= 91)
1242 /* we have nowhere to store the recovery_offset,
1243 * but if it is not below the reshape_position,
1244 * we can piggy-back on that.
1247 if (rdev2->raid_disk < 0 ||
1248 test_bit(Faulty, &rdev2->flags))
1251 desc_nr = rdev2->raid_disk;
1253 desc_nr = next_spare++;
1254 rdev2->desc_nr = desc_nr;
1255 d = &sb->disks[rdev2->desc_nr];
1257 d->number = rdev2->desc_nr;
1258 d->major = MAJOR(rdev2->bdev->bd_dev);
1259 d->minor = MINOR(rdev2->bdev->bd_dev);
1261 d->raid_disk = rdev2->raid_disk;
1263 d->raid_disk = rdev2->desc_nr; /* compatibility */
1264 if (test_bit(Faulty, &rdev2->flags))
1265 d->state = (1<<MD_DISK_FAULTY);
1266 else if (is_active) {
1267 d->state = (1<<MD_DISK_ACTIVE);
1268 if (test_bit(In_sync, &rdev2->flags))
1269 d->state |= (1<<MD_DISK_SYNC);
1277 if (test_bit(WriteMostly, &rdev2->flags))
1278 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1280 /* now set the "removed" and "faulty" bits on any missing devices */
1281 for (i=0 ; i < mddev->raid_disks ; i++) {
1282 mdp_disk_t *d = &sb->disks[i];
1283 if (d->state == 0 && d->number == 0) {
1286 d->state = (1<<MD_DISK_REMOVED);
1287 d->state |= (1<<MD_DISK_FAULTY);
1291 sb->nr_disks = nr_disks;
1292 sb->active_disks = active;
1293 sb->working_disks = working;
1294 sb->failed_disks = failed;
1295 sb->spare_disks = spare;
1297 sb->this_disk = sb->disks[rdev->desc_nr];
1298 sb->sb_csum = calc_sb_csum(sb);
1302 * rdev_size_change for 0.90.0
1304 static unsigned long long
1305 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1307 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1308 return 0; /* component must fit device */
1309 if (rdev->mddev->bitmap_info.offset)
1310 return 0; /* can't move bitmap */
1311 rdev->sb_start = calc_dev_sboffset(rdev);
1312 if (!num_sectors || num_sectors > rdev->sb_start)
1313 num_sectors = rdev->sb_start;
1314 /* Limit to 4TB as metadata cannot record more than that.
1315 * 4TB == 2^32 KB, or 2*2^32 sectors.
1317 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1318 num_sectors = (2ULL << 32) - 2;
1319 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1321 md_super_wait(rdev->mddev);
1326 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1328 /* non-zero offset changes not possible with v0.90 */
1329 return new_offset == 0;
1333 * version 1 superblock
1336 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1340 unsigned long long newcsum;
1341 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1342 __le32 *isuper = (__le32*)sb;
1344 disk_csum = sb->sb_csum;
1347 for (; size >= 4; size -= 4)
1348 newcsum += le32_to_cpu(*isuper++);
1351 newcsum += le16_to_cpu(*(__le16*) isuper);
1353 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1354 sb->sb_csum = disk_csum;
1355 return cpu_to_le32(csum);
1358 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1360 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1362 struct mdp_superblock_1 *sb;
1366 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1370 * Calculate the position of the superblock in 512byte sectors.
1371 * It is always aligned to a 4K boundary and
1372 * depeding on minor_version, it can be:
1373 * 0: At least 8K, but less than 12K, from end of device
1374 * 1: At start of device
1375 * 2: 4K from start of device.
1377 switch(minor_version) {
1379 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1381 sb_start &= ~(sector_t)(4*2-1);
1392 rdev->sb_start = sb_start;
1394 /* superblock is rarely larger than 1K, but it can be larger,
1395 * and it is safe to read 4k, so we do that
1397 ret = read_disk_sb(rdev, 4096);
1398 if (ret) return ret;
1400 sb = page_address(rdev->sb_page);
1402 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1403 sb->major_version != cpu_to_le32(1) ||
1404 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1405 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1406 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1409 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1410 printk("md: invalid superblock checksum on %s\n",
1411 bdevname(rdev->bdev,b));
1414 if (le64_to_cpu(sb->data_size) < 10) {
1415 printk("md: data_size too small on %s\n",
1416 bdevname(rdev->bdev,b));
1421 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1422 /* Some padding is non-zero, might be a new feature */
1425 rdev->preferred_minor = 0xffff;
1426 rdev->data_offset = le64_to_cpu(sb->data_offset);
1427 rdev->new_data_offset = rdev->data_offset;
1428 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1429 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1430 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1431 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1433 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1434 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1435 if (rdev->sb_size & bmask)
1436 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1439 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1442 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1445 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1448 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1450 if (!rdev->bb_page) {
1451 rdev->bb_page = alloc_page(GFP_KERNEL);
1455 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1456 rdev->badblocks.count == 0) {
1457 /* need to load the bad block list.
1458 * Currently we limit it to one page.
1464 int sectors = le16_to_cpu(sb->bblog_size);
1465 if (sectors > (PAGE_SIZE / 512))
1467 offset = le32_to_cpu(sb->bblog_offset);
1470 bb_sector = (long long)offset;
1471 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1472 rdev->bb_page, READ, true))
1474 bbp = (u64 *)page_address(rdev->bb_page);
1475 rdev->badblocks.shift = sb->bblog_shift;
1476 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1477 u64 bb = le64_to_cpu(*bbp);
1478 int count = bb & (0x3ff);
1479 u64 sector = bb >> 10;
1480 sector <<= sb->bblog_shift;
1481 count <<= sb->bblog_shift;
1484 if (md_set_badblocks(&rdev->badblocks,
1485 sector, count, 1) == 0)
1488 } else if (sb->bblog_offset != 0)
1489 rdev->badblocks.shift = 0;
1495 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1497 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1498 sb->level != refsb->level ||
1499 sb->layout != refsb->layout ||
1500 sb->chunksize != refsb->chunksize) {
1501 printk(KERN_WARNING "md: %s has strangely different"
1502 " superblock to %s\n",
1503 bdevname(rdev->bdev,b),
1504 bdevname(refdev->bdev,b2));
1507 ev1 = le64_to_cpu(sb->events);
1508 ev2 = le64_to_cpu(refsb->events);
1515 if (minor_version) {
1516 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1517 sectors -= rdev->data_offset;
1519 sectors = rdev->sb_start;
1520 if (sectors < le64_to_cpu(sb->data_size))
1522 rdev->sectors = le64_to_cpu(sb->data_size);
1526 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1528 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1529 __u64 ev1 = le64_to_cpu(sb->events);
1531 rdev->raid_disk = -1;
1532 clear_bit(Faulty, &rdev->flags);
1533 clear_bit(In_sync, &rdev->flags);
1534 clear_bit(Bitmap_sync, &rdev->flags);
1535 clear_bit(WriteMostly, &rdev->flags);
1537 if (mddev->raid_disks == 0) {
1538 mddev->major_version = 1;
1539 mddev->patch_version = 0;
1540 mddev->external = 0;
1541 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1542 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1543 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1544 mddev->level = le32_to_cpu(sb->level);
1545 mddev->clevel[0] = 0;
1546 mddev->layout = le32_to_cpu(sb->layout);
1547 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1548 mddev->dev_sectors = le64_to_cpu(sb->size);
1549 mddev->events = ev1;
1550 mddev->bitmap_info.offset = 0;
1551 mddev->bitmap_info.space = 0;
1552 /* Default location for bitmap is 1K after superblock
1553 * using 3K - total of 4K
1555 mddev->bitmap_info.default_offset = 1024 >> 9;
1556 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1557 mddev->reshape_backwards = 0;
1559 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1560 memcpy(mddev->uuid, sb->set_uuid, 16);
1562 mddev->max_disks = (4096-256)/2;
1564 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1565 mddev->bitmap_info.file == NULL) {
1566 mddev->bitmap_info.offset =
1567 (__s32)le32_to_cpu(sb->bitmap_offset);
1568 /* Metadata doesn't record how much space is available.
1569 * For 1.0, we assume we can use up to the superblock
1570 * if before, else to 4K beyond superblock.
1571 * For others, assume no change is possible.
1573 if (mddev->minor_version > 0)
1574 mddev->bitmap_info.space = 0;
1575 else if (mddev->bitmap_info.offset > 0)
1576 mddev->bitmap_info.space =
1577 8 - mddev->bitmap_info.offset;
1579 mddev->bitmap_info.space =
1580 -mddev->bitmap_info.offset;
1583 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1584 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1585 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1586 mddev->new_level = le32_to_cpu(sb->new_level);
1587 mddev->new_layout = le32_to_cpu(sb->new_layout);
1588 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1589 if (mddev->delta_disks < 0 ||
1590 (mddev->delta_disks == 0 &&
1591 (le32_to_cpu(sb->feature_map)
1592 & MD_FEATURE_RESHAPE_BACKWARDS)))
1593 mddev->reshape_backwards = 1;
1595 mddev->reshape_position = MaxSector;
1596 mddev->delta_disks = 0;
1597 mddev->new_level = mddev->level;
1598 mddev->new_layout = mddev->layout;
1599 mddev->new_chunk_sectors = mddev->chunk_sectors;
1602 } else if (mddev->pers == NULL) {
1603 /* Insist of good event counter while assembling, except for
1604 * spares (which don't need an event count) */
1606 if (rdev->desc_nr >= 0 &&
1607 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1608 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1609 if (ev1 < mddev->events)
1611 } else if (mddev->bitmap) {
1612 /* If adding to array with a bitmap, then we can accept an
1613 * older device, but not too old.
1615 if (ev1 < mddev->bitmap->events_cleared)
1617 if (ev1 < mddev->events)
1618 set_bit(Bitmap_sync, &rdev->flags);
1620 if (ev1 < mddev->events)
1621 /* just a hot-add of a new device, leave raid_disk at -1 */
1624 if (mddev->level != LEVEL_MULTIPATH) {
1626 if (rdev->desc_nr < 0 ||
1627 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1631 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1633 case 0xffff: /* spare */
1635 case 0xfffe: /* faulty */
1636 set_bit(Faulty, &rdev->flags);
1639 rdev->saved_raid_disk = role;
1640 if ((le32_to_cpu(sb->feature_map) &
1641 MD_FEATURE_RECOVERY_OFFSET)) {
1642 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1643 if (!(le32_to_cpu(sb->feature_map) &
1644 MD_FEATURE_RECOVERY_BITMAP))
1645 rdev->saved_raid_disk = -1;
1647 set_bit(In_sync, &rdev->flags);
1648 rdev->raid_disk = role;
1651 if (sb->devflags & WriteMostly1)
1652 set_bit(WriteMostly, &rdev->flags);
1653 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1654 set_bit(Replacement, &rdev->flags);
1655 } else /* MULTIPATH are always insync */
1656 set_bit(In_sync, &rdev->flags);
1661 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1663 struct mdp_superblock_1 *sb;
1664 struct md_rdev *rdev2;
1666 /* make rdev->sb match mddev and rdev data. */
1668 sb = page_address(rdev->sb_page);
1670 sb->feature_map = 0;
1672 sb->recovery_offset = cpu_to_le64(0);
1673 memset(sb->pad3, 0, sizeof(sb->pad3));
1675 sb->utime = cpu_to_le64((__u64)mddev->utime);
1676 sb->events = cpu_to_le64(mddev->events);
1678 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1680 sb->resync_offset = cpu_to_le64(0);
1682 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1684 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1685 sb->size = cpu_to_le64(mddev->dev_sectors);
1686 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1687 sb->level = cpu_to_le32(mddev->level);
1688 sb->layout = cpu_to_le32(mddev->layout);
1690 if (test_bit(WriteMostly, &rdev->flags))
1691 sb->devflags |= WriteMostly1;
1693 sb->devflags &= ~WriteMostly1;
1694 sb->data_offset = cpu_to_le64(rdev->data_offset);
1695 sb->data_size = cpu_to_le64(rdev->sectors);
1697 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1698 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1699 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1702 if (rdev->raid_disk >= 0 &&
1703 !test_bit(In_sync, &rdev->flags)) {
1705 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1706 sb->recovery_offset =
1707 cpu_to_le64(rdev->recovery_offset);
1708 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1710 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1712 if (test_bit(Replacement, &rdev->flags))
1714 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1716 if (mddev->reshape_position != MaxSector) {
1717 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1718 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1719 sb->new_layout = cpu_to_le32(mddev->new_layout);
1720 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1721 sb->new_level = cpu_to_le32(mddev->new_level);
1722 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1723 if (mddev->delta_disks == 0 &&
1724 mddev->reshape_backwards)
1726 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1727 if (rdev->new_data_offset != rdev->data_offset) {
1729 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1730 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1731 - rdev->data_offset));
1735 if (rdev->badblocks.count == 0)
1736 /* Nothing to do for bad blocks*/ ;
1737 else if (sb->bblog_offset == 0)
1738 /* Cannot record bad blocks on this device */
1739 md_error(mddev, rdev);
1741 struct badblocks *bb = &rdev->badblocks;
1742 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1744 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1749 seq = read_seqbegin(&bb->lock);
1751 memset(bbp, 0xff, PAGE_SIZE);
1753 for (i = 0 ; i < bb->count ; i++) {
1754 u64 internal_bb = p[i];
1755 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1756 | BB_LEN(internal_bb));
1757 bbp[i] = cpu_to_le64(store_bb);
1760 if (read_seqretry(&bb->lock, seq))
1763 bb->sector = (rdev->sb_start +
1764 (int)le32_to_cpu(sb->bblog_offset));
1765 bb->size = le16_to_cpu(sb->bblog_size);
1770 rdev_for_each(rdev2, mddev)
1771 if (rdev2->desc_nr+1 > max_dev)
1772 max_dev = rdev2->desc_nr+1;
1774 if (max_dev > le32_to_cpu(sb->max_dev)) {
1776 sb->max_dev = cpu_to_le32(max_dev);
1777 rdev->sb_size = max_dev * 2 + 256;
1778 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1779 if (rdev->sb_size & bmask)
1780 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1782 max_dev = le32_to_cpu(sb->max_dev);
1784 for (i=0; i<max_dev;i++)
1785 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1787 rdev_for_each(rdev2, mddev) {
1789 if (test_bit(Faulty, &rdev2->flags))
1790 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1791 else if (test_bit(In_sync, &rdev2->flags))
1792 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1793 else if (rdev2->raid_disk >= 0)
1794 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1796 sb->dev_roles[i] = cpu_to_le16(0xffff);
1799 sb->sb_csum = calc_sb_1_csum(sb);
1802 static unsigned long long
1803 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1805 struct mdp_superblock_1 *sb;
1806 sector_t max_sectors;
1807 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1808 return 0; /* component must fit device */
1809 if (rdev->data_offset != rdev->new_data_offset)
1810 return 0; /* too confusing */
1811 if (rdev->sb_start < rdev->data_offset) {
1812 /* minor versions 1 and 2; superblock before data */
1813 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1814 max_sectors -= rdev->data_offset;
1815 if (!num_sectors || num_sectors > max_sectors)
1816 num_sectors = max_sectors;
1817 } else if (rdev->mddev->bitmap_info.offset) {
1818 /* minor version 0 with bitmap we can't move */
1821 /* minor version 0; superblock after data */
1823 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1824 sb_start &= ~(sector_t)(4*2 - 1);
1825 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1826 if (!num_sectors || num_sectors > max_sectors)
1827 num_sectors = max_sectors;
1828 rdev->sb_start = sb_start;
1830 sb = page_address(rdev->sb_page);
1831 sb->data_size = cpu_to_le64(num_sectors);
1832 sb->super_offset = rdev->sb_start;
1833 sb->sb_csum = calc_sb_1_csum(sb);
1834 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1836 md_super_wait(rdev->mddev);
1842 super_1_allow_new_offset(struct md_rdev *rdev,
1843 unsigned long long new_offset)
1845 /* All necessary checks on new >= old have been done */
1846 struct bitmap *bitmap;
1847 if (new_offset >= rdev->data_offset)
1850 /* with 1.0 metadata, there is no metadata to tread on
1851 * so we can always move back */
1852 if (rdev->mddev->minor_version == 0)
1855 /* otherwise we must be sure not to step on
1856 * any metadata, so stay:
1857 * 36K beyond start of superblock
1858 * beyond end of badblocks
1859 * beyond write-intent bitmap
1861 if (rdev->sb_start + (32+4)*2 > new_offset)
1863 bitmap = rdev->mddev->bitmap;
1864 if (bitmap && !rdev->mddev->bitmap_info.file &&
1865 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1866 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1868 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1874 static struct super_type super_types[] = {
1877 .owner = THIS_MODULE,
1878 .load_super = super_90_load,
1879 .validate_super = super_90_validate,
1880 .sync_super = super_90_sync,
1881 .rdev_size_change = super_90_rdev_size_change,
1882 .allow_new_offset = super_90_allow_new_offset,
1886 .owner = THIS_MODULE,
1887 .load_super = super_1_load,
1888 .validate_super = super_1_validate,
1889 .sync_super = super_1_sync,
1890 .rdev_size_change = super_1_rdev_size_change,
1891 .allow_new_offset = super_1_allow_new_offset,
1895 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1897 if (mddev->sync_super) {
1898 mddev->sync_super(mddev, rdev);
1902 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1904 super_types[mddev->major_version].sync_super(mddev, rdev);
1907 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1909 struct md_rdev *rdev, *rdev2;
1912 rdev_for_each_rcu(rdev, mddev1)
1913 rdev_for_each_rcu(rdev2, mddev2)
1914 if (rdev->bdev->bd_contains ==
1915 rdev2->bdev->bd_contains) {
1923 static LIST_HEAD(pending_raid_disks);
1926 * Try to register data integrity profile for an mddev
1928 * This is called when an array is started and after a disk has been kicked
1929 * from the array. It only succeeds if all working and active component devices
1930 * are integrity capable with matching profiles.
1932 int md_integrity_register(struct mddev *mddev)
1934 struct md_rdev *rdev, *reference = NULL;
1936 if (list_empty(&mddev->disks))
1937 return 0; /* nothing to do */
1938 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1939 return 0; /* shouldn't register, or already is */
1940 rdev_for_each(rdev, mddev) {
1941 /* skip spares and non-functional disks */
1942 if (test_bit(Faulty, &rdev->flags))
1944 if (rdev->raid_disk < 0)
1947 /* Use the first rdev as the reference */
1951 /* does this rdev's profile match the reference profile? */
1952 if (blk_integrity_compare(reference->bdev->bd_disk,
1953 rdev->bdev->bd_disk) < 0)
1956 if (!reference || !bdev_get_integrity(reference->bdev))
1959 * All component devices are integrity capable and have matching
1960 * profiles, register the common profile for the md device.
1962 if (blk_integrity_register(mddev->gendisk,
1963 bdev_get_integrity(reference->bdev)) != 0) {
1964 printk(KERN_ERR "md: failed to register integrity for %s\n",
1968 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1969 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1970 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1976 EXPORT_SYMBOL(md_integrity_register);
1978 /* Disable data integrity if non-capable/non-matching disk is being added */
1979 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1981 struct blk_integrity *bi_rdev;
1982 struct blk_integrity *bi_mddev;
1984 if (!mddev->gendisk)
1987 bi_rdev = bdev_get_integrity(rdev->bdev);
1988 bi_mddev = blk_get_integrity(mddev->gendisk);
1990 if (!bi_mddev) /* nothing to do */
1992 if (rdev->raid_disk < 0) /* skip spares */
1994 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1995 rdev->bdev->bd_disk) >= 0)
1997 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1998 blk_integrity_unregister(mddev->gendisk);
2000 EXPORT_SYMBOL(md_integrity_add_rdev);
2002 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2004 char b[BDEVNAME_SIZE];
2008 /* prevent duplicates */
2009 if (find_rdev(mddev, rdev->bdev->bd_dev))
2012 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2013 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2014 rdev->sectors < mddev->dev_sectors)) {
2016 /* Cannot change size, so fail
2017 * If mddev->level <= 0, then we don't care
2018 * about aligning sizes (e.g. linear)
2020 if (mddev->level > 0)
2023 mddev->dev_sectors = rdev->sectors;
2026 /* Verify rdev->desc_nr is unique.
2027 * If it is -1, assign a free number, else
2028 * check number is not in use
2031 if (rdev->desc_nr < 0) {
2034 choice = mddev->raid_disks;
2035 while (md_find_rdev_nr_rcu(mddev, choice))
2037 rdev->desc_nr = choice;
2039 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2045 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2046 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2047 mdname(mddev), mddev->max_disks);
2050 bdevname(rdev->bdev,b);
2051 strreplace(b, '/', '!');
2053 rdev->mddev = mddev;
2054 printk(KERN_INFO "md: bind<%s>\n", b);
2056 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2059 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2060 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2061 /* failure here is OK */;
2062 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2064 list_add_rcu(&rdev->same_set, &mddev->disks);
2065 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2067 /* May as well allow recovery to be retried once */
2068 mddev->recovery_disabled++;
2073 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2078 static void md_delayed_delete(struct work_struct *ws)
2080 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2081 kobject_del(&rdev->kobj);
2082 kobject_put(&rdev->kobj);
2085 static void unbind_rdev_from_array(struct md_rdev *rdev)
2087 char b[BDEVNAME_SIZE];
2089 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2090 list_del_rcu(&rdev->same_set);
2091 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2093 sysfs_remove_link(&rdev->kobj, "block");
2094 sysfs_put(rdev->sysfs_state);
2095 rdev->sysfs_state = NULL;
2096 rdev->badblocks.count = 0;
2097 /* We need to delay this, otherwise we can deadlock when
2098 * writing to 'remove' to "dev/state". We also need
2099 * to delay it due to rcu usage.
2102 INIT_WORK(&rdev->del_work, md_delayed_delete);
2103 kobject_get(&rdev->kobj);
2104 queue_work(md_misc_wq, &rdev->del_work);
2108 * prevent the device from being mounted, repartitioned or
2109 * otherwise reused by a RAID array (or any other kernel
2110 * subsystem), by bd_claiming the device.
2112 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2115 struct block_device *bdev;
2116 char b[BDEVNAME_SIZE];
2118 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2119 shared ? (struct md_rdev *)lock_rdev : rdev);
2121 printk(KERN_ERR "md: could not open %s.\n",
2122 __bdevname(dev, b));
2123 return PTR_ERR(bdev);
2129 static void unlock_rdev(struct md_rdev *rdev)
2131 struct block_device *bdev = rdev->bdev;
2133 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2136 void md_autodetect_dev(dev_t dev);
2138 static void export_rdev(struct md_rdev *rdev)
2140 char b[BDEVNAME_SIZE];
2142 printk(KERN_INFO "md: export_rdev(%s)\n",
2143 bdevname(rdev->bdev,b));
2144 md_rdev_clear(rdev);
2146 if (test_bit(AutoDetected, &rdev->flags))
2147 md_autodetect_dev(rdev->bdev->bd_dev);
2150 kobject_put(&rdev->kobj);
2153 void md_kick_rdev_from_array(struct md_rdev *rdev)
2155 unbind_rdev_from_array(rdev);
2158 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2160 static void export_array(struct mddev *mddev)
2162 struct md_rdev *rdev;
2164 while (!list_empty(&mddev->disks)) {
2165 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2167 md_kick_rdev_from_array(rdev);
2169 mddev->raid_disks = 0;
2170 mddev->major_version = 0;
2173 static void sync_sbs(struct mddev *mddev, int nospares)
2175 /* Update each superblock (in-memory image), but
2176 * if we are allowed to, skip spares which already
2177 * have the right event counter, or have one earlier
2178 * (which would mean they aren't being marked as dirty
2179 * with the rest of the array)
2181 struct md_rdev *rdev;
2182 rdev_for_each(rdev, mddev) {
2183 if (rdev->sb_events == mddev->events ||
2185 rdev->raid_disk < 0 &&
2186 rdev->sb_events+1 == mddev->events)) {
2187 /* Don't update this superblock */
2188 rdev->sb_loaded = 2;
2190 sync_super(mddev, rdev);
2191 rdev->sb_loaded = 1;
2196 void md_update_sb(struct mddev *mddev, int force_change)
2198 struct md_rdev *rdev;
2201 int any_badblocks_changed = 0;
2205 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2209 /* First make sure individual recovery_offsets are correct */
2210 rdev_for_each(rdev, mddev) {
2211 if (rdev->raid_disk >= 0 &&
2212 mddev->delta_disks >= 0 &&
2213 !test_bit(In_sync, &rdev->flags) &&
2214 mddev->curr_resync_completed > rdev->recovery_offset)
2215 rdev->recovery_offset = mddev->curr_resync_completed;
2218 if (!mddev->persistent) {
2219 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2220 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2221 if (!mddev->external) {
2222 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2223 rdev_for_each(rdev, mddev) {
2224 if (rdev->badblocks.changed) {
2225 rdev->badblocks.changed = 0;
2226 md_ack_all_badblocks(&rdev->badblocks);
2227 md_error(mddev, rdev);
2229 clear_bit(Blocked, &rdev->flags);
2230 clear_bit(BlockedBadBlocks, &rdev->flags);
2231 wake_up(&rdev->blocked_wait);
2234 wake_up(&mddev->sb_wait);
2238 spin_lock(&mddev->lock);
2240 mddev->utime = get_seconds();
2242 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2244 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2245 /* just a clean<-> dirty transition, possibly leave spares alone,
2246 * though if events isn't the right even/odd, we will have to do
2252 if (mddev->degraded)
2253 /* If the array is degraded, then skipping spares is both
2254 * dangerous and fairly pointless.
2255 * Dangerous because a device that was removed from the array
2256 * might have a event_count that still looks up-to-date,
2257 * so it can be re-added without a resync.
2258 * Pointless because if there are any spares to skip,
2259 * then a recovery will happen and soon that array won't
2260 * be degraded any more and the spare can go back to sleep then.
2264 sync_req = mddev->in_sync;
2266 /* If this is just a dirty<->clean transition, and the array is clean
2267 * and 'events' is odd, we can roll back to the previous clean state */
2269 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2270 && mddev->can_decrease_events
2271 && mddev->events != 1) {
2273 mddev->can_decrease_events = 0;
2275 /* otherwise we have to go forward and ... */
2277 mddev->can_decrease_events = nospares;
2281 * This 64-bit counter should never wrap.
2282 * Either we are in around ~1 trillion A.C., assuming
2283 * 1 reboot per second, or we have a bug...
2285 WARN_ON(mddev->events == 0);
2287 rdev_for_each(rdev, mddev) {
2288 if (rdev->badblocks.changed)
2289 any_badblocks_changed++;
2290 if (test_bit(Faulty, &rdev->flags))
2291 set_bit(FaultRecorded, &rdev->flags);
2294 sync_sbs(mddev, nospares);
2295 spin_unlock(&mddev->lock);
2297 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2298 mdname(mddev), mddev->in_sync);
2300 bitmap_update_sb(mddev->bitmap);
2301 rdev_for_each(rdev, mddev) {
2302 char b[BDEVNAME_SIZE];
2304 if (rdev->sb_loaded != 1)
2305 continue; /* no noise on spare devices */
2307 if (!test_bit(Faulty, &rdev->flags)) {
2308 md_super_write(mddev,rdev,
2309 rdev->sb_start, rdev->sb_size,
2311 pr_debug("md: (write) %s's sb offset: %llu\n",
2312 bdevname(rdev->bdev, b),
2313 (unsigned long long)rdev->sb_start);
2314 rdev->sb_events = mddev->events;
2315 if (rdev->badblocks.size) {
2316 md_super_write(mddev, rdev,
2317 rdev->badblocks.sector,
2318 rdev->badblocks.size << 9,
2320 rdev->badblocks.size = 0;
2324 pr_debug("md: %s (skipping faulty)\n",
2325 bdevname(rdev->bdev, b));
2327 if (mddev->level == LEVEL_MULTIPATH)
2328 /* only need to write one superblock... */
2331 md_super_wait(mddev);
2332 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2334 spin_lock(&mddev->lock);
2335 if (mddev->in_sync != sync_req ||
2336 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2337 /* have to write it out again */
2338 spin_unlock(&mddev->lock);
2341 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2342 spin_unlock(&mddev->lock);
2343 wake_up(&mddev->sb_wait);
2344 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2345 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2347 rdev_for_each(rdev, mddev) {
2348 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2349 clear_bit(Blocked, &rdev->flags);
2351 if (any_badblocks_changed)
2352 md_ack_all_badblocks(&rdev->badblocks);
2353 clear_bit(BlockedBadBlocks, &rdev->flags);
2354 wake_up(&rdev->blocked_wait);
2357 EXPORT_SYMBOL(md_update_sb);
2359 static int add_bound_rdev(struct md_rdev *rdev)
2361 struct mddev *mddev = rdev->mddev;
2364 if (!mddev->pers->hot_remove_disk) {
2365 /* If there is hot_add_disk but no hot_remove_disk
2366 * then added disks for geometry changes,
2367 * and should be added immediately.
2369 super_types[mddev->major_version].
2370 validate_super(mddev, rdev);
2371 err = mddev->pers->hot_add_disk(mddev, rdev);
2373 unbind_rdev_from_array(rdev);
2378 sysfs_notify_dirent_safe(rdev->sysfs_state);
2380 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2381 if (mddev->degraded)
2382 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2383 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2384 md_new_event(mddev);
2385 md_wakeup_thread(mddev->thread);
2389 /* words written to sysfs files may, or may not, be \n terminated.
2390 * We want to accept with case. For this we use cmd_match.
2392 static int cmd_match(const char *cmd, const char *str)
2394 /* See if cmd, written into a sysfs file, matches
2395 * str. They must either be the same, or cmd can
2396 * have a trailing newline
2398 while (*cmd && *str && *cmd == *str) {
2409 struct rdev_sysfs_entry {
2410 struct attribute attr;
2411 ssize_t (*show)(struct md_rdev *, char *);
2412 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2416 state_show(struct md_rdev *rdev, char *page)
2420 unsigned long flags = ACCESS_ONCE(rdev->flags);
2422 if (test_bit(Faulty, &flags) ||
2423 rdev->badblocks.unacked_exist) {
2424 len+= sprintf(page+len, "%sfaulty",sep);
2427 if (test_bit(In_sync, &flags)) {
2428 len += sprintf(page+len, "%sin_sync",sep);
2431 if (test_bit(WriteMostly, &flags)) {
2432 len += sprintf(page+len, "%swrite_mostly",sep);
2435 if (test_bit(Blocked, &flags) ||
2436 (rdev->badblocks.unacked_exist
2437 && !test_bit(Faulty, &flags))) {
2438 len += sprintf(page+len, "%sblocked", sep);
2441 if (!test_bit(Faulty, &flags) &&
2442 !test_bit(In_sync, &flags)) {
2443 len += sprintf(page+len, "%sspare", sep);
2446 if (test_bit(WriteErrorSeen, &flags)) {
2447 len += sprintf(page+len, "%swrite_error", sep);
2450 if (test_bit(WantReplacement, &flags)) {
2451 len += sprintf(page+len, "%swant_replacement", sep);
2454 if (test_bit(Replacement, &flags)) {
2455 len += sprintf(page+len, "%sreplacement", sep);
2459 return len+sprintf(page+len, "\n");
2463 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2466 * faulty - simulates an error
2467 * remove - disconnects the device
2468 * writemostly - sets write_mostly
2469 * -writemostly - clears write_mostly
2470 * blocked - sets the Blocked flags
2471 * -blocked - clears the Blocked and possibly simulates an error
2472 * insync - sets Insync providing device isn't active
2473 * -insync - clear Insync for a device with a slot assigned,
2474 * so that it gets rebuilt based on bitmap
2475 * write_error - sets WriteErrorSeen
2476 * -write_error - clears WriteErrorSeen
2479 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2480 md_error(rdev->mddev, rdev);
2481 if (test_bit(Faulty, &rdev->flags))
2485 } else if (cmd_match(buf, "remove")) {
2486 if (rdev->raid_disk >= 0)
2489 struct mddev *mddev = rdev->mddev;
2490 if (mddev_is_clustered(mddev))
2491 md_cluster_ops->remove_disk(mddev, rdev);
2492 md_kick_rdev_from_array(rdev);
2493 if (mddev_is_clustered(mddev))
2494 md_cluster_ops->metadata_update_start(mddev);
2496 md_update_sb(mddev, 1);
2497 md_new_event(mddev);
2498 if (mddev_is_clustered(mddev))
2499 md_cluster_ops->metadata_update_finish(mddev);
2502 } else if (cmd_match(buf, "writemostly")) {
2503 set_bit(WriteMostly, &rdev->flags);
2505 } else if (cmd_match(buf, "-writemostly")) {
2506 clear_bit(WriteMostly, &rdev->flags);
2508 } else if (cmd_match(buf, "blocked")) {
2509 set_bit(Blocked, &rdev->flags);
2511 } else if (cmd_match(buf, "-blocked")) {
2512 if (!test_bit(Faulty, &rdev->flags) &&
2513 rdev->badblocks.unacked_exist) {
2514 /* metadata handler doesn't understand badblocks,
2515 * so we need to fail the device
2517 md_error(rdev->mddev, rdev);
2519 clear_bit(Blocked, &rdev->flags);
2520 clear_bit(BlockedBadBlocks, &rdev->flags);
2521 wake_up(&rdev->blocked_wait);
2522 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2523 md_wakeup_thread(rdev->mddev->thread);
2526 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2527 set_bit(In_sync, &rdev->flags);
2529 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2530 if (rdev->mddev->pers == NULL) {
2531 clear_bit(In_sync, &rdev->flags);
2532 rdev->saved_raid_disk = rdev->raid_disk;
2533 rdev->raid_disk = -1;
2536 } else if (cmd_match(buf, "write_error")) {
2537 set_bit(WriteErrorSeen, &rdev->flags);
2539 } else if (cmd_match(buf, "-write_error")) {
2540 clear_bit(WriteErrorSeen, &rdev->flags);
2542 } else if (cmd_match(buf, "want_replacement")) {
2543 /* Any non-spare device that is not a replacement can
2544 * become want_replacement at any time, but we then need to
2545 * check if recovery is needed.
2547 if (rdev->raid_disk >= 0 &&
2548 !test_bit(Replacement, &rdev->flags))
2549 set_bit(WantReplacement, &rdev->flags);
2550 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2551 md_wakeup_thread(rdev->mddev->thread);
2553 } else if (cmd_match(buf, "-want_replacement")) {
2554 /* Clearing 'want_replacement' is always allowed.
2555 * Once replacements starts it is too late though.
2558 clear_bit(WantReplacement, &rdev->flags);
2559 } else if (cmd_match(buf, "replacement")) {
2560 /* Can only set a device as a replacement when array has not
2561 * yet been started. Once running, replacement is automatic
2562 * from spares, or by assigning 'slot'.
2564 if (rdev->mddev->pers)
2567 set_bit(Replacement, &rdev->flags);
2570 } else if (cmd_match(buf, "-replacement")) {
2571 /* Similarly, can only clear Replacement before start */
2572 if (rdev->mddev->pers)
2575 clear_bit(Replacement, &rdev->flags);
2578 } else if (cmd_match(buf, "re-add")) {
2579 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2580 /* clear_bit is performed _after_ all the devices
2581 * have their local Faulty bit cleared. If any writes
2582 * happen in the meantime in the local node, they
2583 * will land in the local bitmap, which will be synced
2584 * by this node eventually
2586 if (!mddev_is_clustered(rdev->mddev) ||
2587 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2588 clear_bit(Faulty, &rdev->flags);
2589 err = add_bound_rdev(rdev);
2595 sysfs_notify_dirent_safe(rdev->sysfs_state);
2596 return err ? err : len;
2598 static struct rdev_sysfs_entry rdev_state =
2599 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2602 errors_show(struct md_rdev *rdev, char *page)
2604 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2608 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2613 rv = kstrtouint(buf, 10, &n);
2616 atomic_set(&rdev->corrected_errors, n);
2619 static struct rdev_sysfs_entry rdev_errors =
2620 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2623 slot_show(struct md_rdev *rdev, char *page)
2625 if (rdev->raid_disk < 0)
2626 return sprintf(page, "none\n");
2628 return sprintf(page, "%d\n", rdev->raid_disk);
2632 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2637 if (strncmp(buf, "none", 4)==0)
2640 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2644 if (rdev->mddev->pers && slot == -1) {
2645 /* Setting 'slot' on an active array requires also
2646 * updating the 'rd%d' link, and communicating
2647 * with the personality with ->hot_*_disk.
2648 * For now we only support removing
2649 * failed/spare devices. This normally happens automatically,
2650 * but not when the metadata is externally managed.
2652 if (rdev->raid_disk == -1)
2654 /* personality does all needed checks */
2655 if (rdev->mddev->pers->hot_remove_disk == NULL)
2657 clear_bit(Blocked, &rdev->flags);
2658 remove_and_add_spares(rdev->mddev, rdev);
2659 if (rdev->raid_disk >= 0)
2661 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2662 md_wakeup_thread(rdev->mddev->thread);
2663 } else if (rdev->mddev->pers) {
2664 /* Activating a spare .. or possibly reactivating
2665 * if we ever get bitmaps working here.
2668 if (rdev->raid_disk != -1)
2671 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2674 if (rdev->mddev->pers->hot_add_disk == NULL)
2677 if (slot >= rdev->mddev->raid_disks &&
2678 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2681 rdev->raid_disk = slot;
2682 if (test_bit(In_sync, &rdev->flags))
2683 rdev->saved_raid_disk = slot;
2685 rdev->saved_raid_disk = -1;
2686 clear_bit(In_sync, &rdev->flags);
2687 clear_bit(Bitmap_sync, &rdev->flags);
2688 err = rdev->mddev->pers->
2689 hot_add_disk(rdev->mddev, rdev);
2691 rdev->raid_disk = -1;
2694 sysfs_notify_dirent_safe(rdev->sysfs_state);
2695 if (sysfs_link_rdev(rdev->mddev, rdev))
2696 /* failure here is OK */;
2697 /* don't wakeup anyone, leave that to userspace. */
2699 if (slot >= rdev->mddev->raid_disks &&
2700 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2702 rdev->raid_disk = slot;
2703 /* assume it is working */
2704 clear_bit(Faulty, &rdev->flags);
2705 clear_bit(WriteMostly, &rdev->flags);
2706 set_bit(In_sync, &rdev->flags);
2707 sysfs_notify_dirent_safe(rdev->sysfs_state);
2712 static struct rdev_sysfs_entry rdev_slot =
2713 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2716 offset_show(struct md_rdev *rdev, char *page)
2718 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2722 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2724 unsigned long long offset;
2725 if (kstrtoull(buf, 10, &offset) < 0)
2727 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2729 if (rdev->sectors && rdev->mddev->external)
2730 /* Must set offset before size, so overlap checks
2733 rdev->data_offset = offset;
2734 rdev->new_data_offset = offset;
2738 static struct rdev_sysfs_entry rdev_offset =
2739 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2741 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2743 return sprintf(page, "%llu\n",
2744 (unsigned long long)rdev->new_data_offset);
2747 static ssize_t new_offset_store(struct md_rdev *rdev,
2748 const char *buf, size_t len)
2750 unsigned long long new_offset;
2751 struct mddev *mddev = rdev->mddev;
2753 if (kstrtoull(buf, 10, &new_offset) < 0)
2756 if (mddev->sync_thread ||
2757 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2759 if (new_offset == rdev->data_offset)
2760 /* reset is always permitted */
2762 else if (new_offset > rdev->data_offset) {
2763 /* must not push array size beyond rdev_sectors */
2764 if (new_offset - rdev->data_offset
2765 + mddev->dev_sectors > rdev->sectors)
2768 /* Metadata worries about other space details. */
2770 /* decreasing the offset is inconsistent with a backwards
2773 if (new_offset < rdev->data_offset &&
2774 mddev->reshape_backwards)
2776 /* Increasing offset is inconsistent with forwards
2777 * reshape. reshape_direction should be set to
2778 * 'backwards' first.
2780 if (new_offset > rdev->data_offset &&
2781 !mddev->reshape_backwards)
2784 if (mddev->pers && mddev->persistent &&
2785 !super_types[mddev->major_version]
2786 .allow_new_offset(rdev, new_offset))
2788 rdev->new_data_offset = new_offset;
2789 if (new_offset > rdev->data_offset)
2790 mddev->reshape_backwards = 1;
2791 else if (new_offset < rdev->data_offset)
2792 mddev->reshape_backwards = 0;
2796 static struct rdev_sysfs_entry rdev_new_offset =
2797 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2800 rdev_size_show(struct md_rdev *rdev, char *page)
2802 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2805 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2807 /* check if two start/length pairs overlap */
2815 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2817 unsigned long long blocks;
2820 if (kstrtoull(buf, 10, &blocks) < 0)
2823 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2824 return -EINVAL; /* sector conversion overflow */
2827 if (new != blocks * 2)
2828 return -EINVAL; /* unsigned long long to sector_t overflow */
2835 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2837 struct mddev *my_mddev = rdev->mddev;
2838 sector_t oldsectors = rdev->sectors;
2841 if (strict_blocks_to_sectors(buf, §ors) < 0)
2843 if (rdev->data_offset != rdev->new_data_offset)
2844 return -EINVAL; /* too confusing */
2845 if (my_mddev->pers && rdev->raid_disk >= 0) {
2846 if (my_mddev->persistent) {
2847 sectors = super_types[my_mddev->major_version].
2848 rdev_size_change(rdev, sectors);
2851 } else if (!sectors)
2852 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2854 if (!my_mddev->pers->resize)
2855 /* Cannot change size for RAID0 or Linear etc */
2858 if (sectors < my_mddev->dev_sectors)
2859 return -EINVAL; /* component must fit device */
2861 rdev->sectors = sectors;
2862 if (sectors > oldsectors && my_mddev->external) {
2863 /* Need to check that all other rdevs with the same
2864 * ->bdev do not overlap. 'rcu' is sufficient to walk
2865 * the rdev lists safely.
2866 * This check does not provide a hard guarantee, it
2867 * just helps avoid dangerous mistakes.
2869 struct mddev *mddev;
2871 struct list_head *tmp;
2874 for_each_mddev(mddev, tmp) {
2875 struct md_rdev *rdev2;
2877 rdev_for_each(rdev2, mddev)
2878 if (rdev->bdev == rdev2->bdev &&
2880 overlaps(rdev->data_offset, rdev->sectors,
2893 /* Someone else could have slipped in a size
2894 * change here, but doing so is just silly.
2895 * We put oldsectors back because we *know* it is
2896 * safe, and trust userspace not to race with
2899 rdev->sectors = oldsectors;
2906 static struct rdev_sysfs_entry rdev_size =
2907 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2909 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2911 unsigned long long recovery_start = rdev->recovery_offset;
2913 if (test_bit(In_sync, &rdev->flags) ||
2914 recovery_start == MaxSector)
2915 return sprintf(page, "none\n");
2917 return sprintf(page, "%llu\n", recovery_start);
2920 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2922 unsigned long long recovery_start;
2924 if (cmd_match(buf, "none"))
2925 recovery_start = MaxSector;
2926 else if (kstrtoull(buf, 10, &recovery_start))
2929 if (rdev->mddev->pers &&
2930 rdev->raid_disk >= 0)
2933 rdev->recovery_offset = recovery_start;
2934 if (recovery_start == MaxSector)
2935 set_bit(In_sync, &rdev->flags);
2937 clear_bit(In_sync, &rdev->flags);
2941 static struct rdev_sysfs_entry rdev_recovery_start =
2942 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2945 badblocks_show(struct badblocks *bb, char *page, int unack);
2947 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2949 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2951 return badblocks_show(&rdev->badblocks, page, 0);
2953 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2955 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2956 /* Maybe that ack was all we needed */
2957 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2958 wake_up(&rdev->blocked_wait);
2961 static struct rdev_sysfs_entry rdev_bad_blocks =
2962 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2964 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2966 return badblocks_show(&rdev->badblocks, page, 1);
2968 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2970 return badblocks_store(&rdev->badblocks, page, len, 1);
2972 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2973 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2975 static struct attribute *rdev_default_attrs[] = {
2980 &rdev_new_offset.attr,
2982 &rdev_recovery_start.attr,
2983 &rdev_bad_blocks.attr,
2984 &rdev_unack_bad_blocks.attr,
2988 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2990 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2991 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2997 return entry->show(rdev, page);
3001 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3002 const char *page, size_t length)
3004 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3005 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3007 struct mddev *mddev = rdev->mddev;
3011 if (!capable(CAP_SYS_ADMIN))
3013 rv = mddev ? mddev_lock(mddev): -EBUSY;
3015 if (rdev->mddev == NULL)
3018 rv = entry->store(rdev, page, length);
3019 mddev_unlock(mddev);
3024 static void rdev_free(struct kobject *ko)
3026 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3029 static const struct sysfs_ops rdev_sysfs_ops = {
3030 .show = rdev_attr_show,
3031 .store = rdev_attr_store,
3033 static struct kobj_type rdev_ktype = {
3034 .release = rdev_free,
3035 .sysfs_ops = &rdev_sysfs_ops,
3036 .default_attrs = rdev_default_attrs,
3039 int md_rdev_init(struct md_rdev *rdev)
3042 rdev->saved_raid_disk = -1;
3043 rdev->raid_disk = -1;
3045 rdev->data_offset = 0;
3046 rdev->new_data_offset = 0;
3047 rdev->sb_events = 0;
3048 rdev->last_read_error.tv_sec = 0;
3049 rdev->last_read_error.tv_nsec = 0;
3050 rdev->sb_loaded = 0;
3051 rdev->bb_page = NULL;
3052 atomic_set(&rdev->nr_pending, 0);
3053 atomic_set(&rdev->read_errors, 0);
3054 atomic_set(&rdev->corrected_errors, 0);
3056 INIT_LIST_HEAD(&rdev->same_set);
3057 init_waitqueue_head(&rdev->blocked_wait);
3059 /* Add space to store bad block list.
3060 * This reserves the space even on arrays where it cannot
3061 * be used - I wonder if that matters
3063 rdev->badblocks.count = 0;
3064 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3065 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3066 seqlock_init(&rdev->badblocks.lock);
3067 if (rdev->badblocks.page == NULL)
3072 EXPORT_SYMBOL_GPL(md_rdev_init);
3074 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3076 * mark the device faulty if:
3078 * - the device is nonexistent (zero size)
3079 * - the device has no valid superblock
3081 * a faulty rdev _never_ has rdev->sb set.
3083 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3085 char b[BDEVNAME_SIZE];
3087 struct md_rdev *rdev;
3090 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3092 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3093 return ERR_PTR(-ENOMEM);
3096 err = md_rdev_init(rdev);
3099 err = alloc_disk_sb(rdev);
3103 err = lock_rdev(rdev, newdev, super_format == -2);
3107 kobject_init(&rdev->kobj, &rdev_ktype);
3109 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3112 "md: %s has zero or unknown size, marking faulty!\n",
3113 bdevname(rdev->bdev,b));
3118 if (super_format >= 0) {
3119 err = super_types[super_format].
3120 load_super(rdev, NULL, super_minor);
3121 if (err == -EINVAL) {
3123 "md: %s does not have a valid v%d.%d "
3124 "superblock, not importing!\n",
3125 bdevname(rdev->bdev,b),
3126 super_format, super_minor);
3131 "md: could not read %s's sb, not importing!\n",
3132 bdevname(rdev->bdev,b));
3142 md_rdev_clear(rdev);
3144 return ERR_PTR(err);
3148 * Check a full RAID array for plausibility
3151 static void analyze_sbs(struct mddev *mddev)
3154 struct md_rdev *rdev, *freshest, *tmp;
3155 char b[BDEVNAME_SIZE];
3158 rdev_for_each_safe(rdev, tmp, mddev)
3159 switch (super_types[mddev->major_version].
3160 load_super(rdev, freshest, mddev->minor_version)) {
3168 "md: fatal superblock inconsistency in %s"
3169 " -- removing from array\n",
3170 bdevname(rdev->bdev,b));
3171 md_kick_rdev_from_array(rdev);
3174 super_types[mddev->major_version].
3175 validate_super(mddev, freshest);
3178 rdev_for_each_safe(rdev, tmp, mddev) {
3179 if (mddev->max_disks &&
3180 (rdev->desc_nr >= mddev->max_disks ||
3181 i > mddev->max_disks)) {
3183 "md: %s: %s: only %d devices permitted\n",
3184 mdname(mddev), bdevname(rdev->bdev, b),
3186 md_kick_rdev_from_array(rdev);
3189 if (rdev != freshest) {
3190 if (super_types[mddev->major_version].
3191 validate_super(mddev, rdev)) {
3192 printk(KERN_WARNING "md: kicking non-fresh %s"
3194 bdevname(rdev->bdev,b));
3195 md_kick_rdev_from_array(rdev);
3198 /* No device should have a Candidate flag
3199 * when reading devices
3201 if (test_bit(Candidate, &rdev->flags)) {
3202 pr_info("md: kicking Cluster Candidate %s from array!\n",
3203 bdevname(rdev->bdev, b));
3204 md_kick_rdev_from_array(rdev);
3207 if (mddev->level == LEVEL_MULTIPATH) {
3208 rdev->desc_nr = i++;
3209 rdev->raid_disk = rdev->desc_nr;
3210 set_bit(In_sync, &rdev->flags);
3211 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3212 rdev->raid_disk = -1;
3213 clear_bit(In_sync, &rdev->flags);
3218 /* Read a fixed-point number.
3219 * Numbers in sysfs attributes should be in "standard" units where
3220 * possible, so time should be in seconds.
3221 * However we internally use a a much smaller unit such as
3222 * milliseconds or jiffies.
3223 * This function takes a decimal number with a possible fractional
3224 * component, and produces an integer which is the result of
3225 * multiplying that number by 10^'scale'.
3226 * all without any floating-point arithmetic.
3228 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3230 unsigned long result = 0;
3232 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3235 else if (decimals < scale) {
3238 result = result * 10 + value;
3250 while (decimals < scale) {
3258 static void md_safemode_timeout(unsigned long data);
3261 safe_delay_show(struct mddev *mddev, char *page)
3263 int msec = (mddev->safemode_delay*1000)/HZ;
3264 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3267 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3271 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3274 mddev->safemode_delay = 0;
3276 unsigned long old_delay = mddev->safemode_delay;
3277 unsigned long new_delay = (msec*HZ)/1000;
3281 mddev->safemode_delay = new_delay;
3282 if (new_delay < old_delay || old_delay == 0)
3283 mod_timer(&mddev->safemode_timer, jiffies+1);
3287 static struct md_sysfs_entry md_safe_delay =
3288 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3291 level_show(struct mddev *mddev, char *page)
3293 struct md_personality *p;
3295 spin_lock(&mddev->lock);
3298 ret = sprintf(page, "%s\n", p->name);
3299 else if (mddev->clevel[0])
3300 ret = sprintf(page, "%s\n", mddev->clevel);
3301 else if (mddev->level != LEVEL_NONE)
3302 ret = sprintf(page, "%d\n", mddev->level);
3305 spin_unlock(&mddev->lock);
3310 level_store(struct mddev *mddev, const char *buf, size_t len)
3315 struct md_personality *pers, *oldpers;
3317 void *priv, *oldpriv;
3318 struct md_rdev *rdev;
3320 if (slen == 0 || slen >= sizeof(clevel))
3323 rv = mddev_lock(mddev);
3327 if (mddev->pers == NULL) {
3328 strncpy(mddev->clevel, buf, slen);
3329 if (mddev->clevel[slen-1] == '\n')
3331 mddev->clevel[slen] = 0;
3332 mddev->level = LEVEL_NONE;
3340 /* request to change the personality. Need to ensure:
3341 * - array is not engaged in resync/recovery/reshape
3342 * - old personality can be suspended
3343 * - new personality will access other array.
3347 if (mddev->sync_thread ||
3348 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3349 mddev->reshape_position != MaxSector ||
3350 mddev->sysfs_active)
3354 if (!mddev->pers->quiesce) {
3355 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3356 mdname(mddev), mddev->pers->name);
3360 /* Now find the new personality */
3361 strncpy(clevel, buf, slen);
3362 if (clevel[slen-1] == '\n')
3365 if (kstrtol(clevel, 10, &level))
3368 if (request_module("md-%s", clevel) != 0)
3369 request_module("md-level-%s", clevel);
3370 spin_lock(&pers_lock);
3371 pers = find_pers(level, clevel);
3372 if (!pers || !try_module_get(pers->owner)) {
3373 spin_unlock(&pers_lock);
3374 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3378 spin_unlock(&pers_lock);
3380 if (pers == mddev->pers) {
3381 /* Nothing to do! */
3382 module_put(pers->owner);
3386 if (!pers->takeover) {
3387 module_put(pers->owner);
3388 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3389 mdname(mddev), clevel);
3394 rdev_for_each(rdev, mddev)
3395 rdev->new_raid_disk = rdev->raid_disk;
3397 /* ->takeover must set new_* and/or delta_disks
3398 * if it succeeds, and may set them when it fails.
3400 priv = pers->takeover(mddev);
3402 mddev->new_level = mddev->level;
3403 mddev->new_layout = mddev->layout;
3404 mddev->new_chunk_sectors = mddev->chunk_sectors;
3405 mddev->raid_disks -= mddev->delta_disks;
3406 mddev->delta_disks = 0;
3407 mddev->reshape_backwards = 0;
3408 module_put(pers->owner);
3409 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3410 mdname(mddev), clevel);
3415 /* Looks like we have a winner */
3416 mddev_suspend(mddev);
3417 mddev_detach(mddev);
3419 spin_lock(&mddev->lock);
3420 oldpers = mddev->pers;
3421 oldpriv = mddev->private;
3423 mddev->private = priv;
3424 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3425 mddev->level = mddev->new_level;
3426 mddev->layout = mddev->new_layout;
3427 mddev->chunk_sectors = mddev->new_chunk_sectors;
3428 mddev->delta_disks = 0;
3429 mddev->reshape_backwards = 0;
3430 mddev->degraded = 0;
3431 spin_unlock(&mddev->lock);
3433 if (oldpers->sync_request == NULL &&
3435 /* We are converting from a no-redundancy array
3436 * to a redundancy array and metadata is managed
3437 * externally so we need to be sure that writes
3438 * won't block due to a need to transition
3440 * until external management is started.
3443 mddev->safemode_delay = 0;
3444 mddev->safemode = 0;
3447 oldpers->free(mddev, oldpriv);
3449 if (oldpers->sync_request == NULL &&
3450 pers->sync_request != NULL) {
3451 /* need to add the md_redundancy_group */
3452 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3454 "md: cannot register extra attributes for %s\n",
3456 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3458 if (oldpers->sync_request != NULL &&
3459 pers->sync_request == NULL) {
3460 /* need to remove the md_redundancy_group */
3461 if (mddev->to_remove == NULL)
3462 mddev->to_remove = &md_redundancy_group;
3465 rdev_for_each(rdev, mddev) {
3466 if (rdev->raid_disk < 0)
3468 if (rdev->new_raid_disk >= mddev->raid_disks)
3469 rdev->new_raid_disk = -1;
3470 if (rdev->new_raid_disk == rdev->raid_disk)
3472 sysfs_unlink_rdev(mddev, rdev);
3474 rdev_for_each(rdev, mddev) {
3475 if (rdev->raid_disk < 0)
3477 if (rdev->new_raid_disk == rdev->raid_disk)
3479 rdev->raid_disk = rdev->new_raid_disk;
3480 if (rdev->raid_disk < 0)
3481 clear_bit(In_sync, &rdev->flags);
3483 if (sysfs_link_rdev(mddev, rdev))
3484 printk(KERN_WARNING "md: cannot register rd%d"
3485 " for %s after level change\n",
3486 rdev->raid_disk, mdname(mddev));
3490 if (pers->sync_request == NULL) {
3491 /* this is now an array without redundancy, so
3492 * it must always be in_sync
3495 del_timer_sync(&mddev->safemode_timer);
3497 blk_set_stacking_limits(&mddev->queue->limits);
3499 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3500 mddev_resume(mddev);
3502 md_update_sb(mddev, 1);
3503 sysfs_notify(&mddev->kobj, NULL, "level");
3504 md_new_event(mddev);
3507 mddev_unlock(mddev);
3511 static struct md_sysfs_entry md_level =
3512 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3515 layout_show(struct mddev *mddev, char *page)
3517 /* just a number, not meaningful for all levels */
3518 if (mddev->reshape_position != MaxSector &&
3519 mddev->layout != mddev->new_layout)
3520 return sprintf(page, "%d (%d)\n",
3521 mddev->new_layout, mddev->layout);
3522 return sprintf(page, "%d\n", mddev->layout);
3526 layout_store(struct mddev *mddev, const char *buf, size_t len)
3531 err = kstrtouint(buf, 10, &n);
3534 err = mddev_lock(mddev);
3539 if (mddev->pers->check_reshape == NULL)
3544 mddev->new_layout = n;
3545 err = mddev->pers->check_reshape(mddev);
3547 mddev->new_layout = mddev->layout;
3550 mddev->new_layout = n;
3551 if (mddev->reshape_position == MaxSector)
3554 mddev_unlock(mddev);
3557 static struct md_sysfs_entry md_layout =
3558 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3561 raid_disks_show(struct mddev *mddev, char *page)
3563 if (mddev->raid_disks == 0)
3565 if (mddev->reshape_position != MaxSector &&
3566 mddev->delta_disks != 0)
3567 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3568 mddev->raid_disks - mddev->delta_disks);
3569 return sprintf(page, "%d\n", mddev->raid_disks);
3572 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3575 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3580 err = kstrtouint(buf, 10, &n);
3584 err = mddev_lock(mddev);
3588 err = update_raid_disks(mddev, n);
3589 else if (mddev->reshape_position != MaxSector) {
3590 struct md_rdev *rdev;
3591 int olddisks = mddev->raid_disks - mddev->delta_disks;
3594 rdev_for_each(rdev, mddev) {
3596 rdev->data_offset < rdev->new_data_offset)
3599 rdev->data_offset > rdev->new_data_offset)
3603 mddev->delta_disks = n - olddisks;
3604 mddev->raid_disks = n;
3605 mddev->reshape_backwards = (mddev->delta_disks < 0);
3607 mddev->raid_disks = n;
3609 mddev_unlock(mddev);
3610 return err ? err : len;
3612 static struct md_sysfs_entry md_raid_disks =
3613 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3616 chunk_size_show(struct mddev *mddev, char *page)
3618 if (mddev->reshape_position != MaxSector &&
3619 mddev->chunk_sectors != mddev->new_chunk_sectors)
3620 return sprintf(page, "%d (%d)\n",
3621 mddev->new_chunk_sectors << 9,
3622 mddev->chunk_sectors << 9);
3623 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3627 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3632 err = kstrtoul(buf, 10, &n);
3636 err = mddev_lock(mddev);
3640 if (mddev->pers->check_reshape == NULL)
3645 mddev->new_chunk_sectors = n >> 9;
3646 err = mddev->pers->check_reshape(mddev);
3648 mddev->new_chunk_sectors = mddev->chunk_sectors;
3651 mddev->new_chunk_sectors = n >> 9;
3652 if (mddev->reshape_position == MaxSector)
3653 mddev->chunk_sectors = n >> 9;
3655 mddev_unlock(mddev);
3658 static struct md_sysfs_entry md_chunk_size =
3659 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3662 resync_start_show(struct mddev *mddev, char *page)
3664 if (mddev->recovery_cp == MaxSector)
3665 return sprintf(page, "none\n");
3666 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3670 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3672 unsigned long long n;
3675 if (cmd_match(buf, "none"))
3678 err = kstrtoull(buf, 10, &n);
3681 if (n != (sector_t)n)
3685 err = mddev_lock(mddev);
3688 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3692 mddev->recovery_cp = n;
3694 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3696 mddev_unlock(mddev);
3699 static struct md_sysfs_entry md_resync_start =
3700 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3701 resync_start_show, resync_start_store);
3704 * The array state can be:
3707 * No devices, no size, no level
3708 * Equivalent to STOP_ARRAY ioctl
3710 * May have some settings, but array is not active
3711 * all IO results in error
3712 * When written, doesn't tear down array, but just stops it
3713 * suspended (not supported yet)
3714 * All IO requests will block. The array can be reconfigured.
3715 * Writing this, if accepted, will block until array is quiescent
3717 * no resync can happen. no superblocks get written.
3718 * write requests fail
3720 * like readonly, but behaves like 'clean' on a write request.
3722 * clean - no pending writes, but otherwise active.
3723 * When written to inactive array, starts without resync
3724 * If a write request arrives then
3725 * if metadata is known, mark 'dirty' and switch to 'active'.
3726 * if not known, block and switch to write-pending
3727 * If written to an active array that has pending writes, then fails.
3729 * fully active: IO and resync can be happening.
3730 * When written to inactive array, starts with resync
3733 * clean, but writes are blocked waiting for 'active' to be written.
3736 * like active, but no writes have been seen for a while (100msec).
3739 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3740 write_pending, active_idle, bad_word};
3741 static char *array_states[] = {
3742 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3743 "write-pending", "active-idle", NULL };
3745 static int match_word(const char *word, char **list)
3748 for (n=0; list[n]; n++)
3749 if (cmd_match(word, list[n]))
3755 array_state_show(struct mddev *mddev, char *page)
3757 enum array_state st = inactive;
3770 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3772 else if (mddev->safemode)
3778 if (list_empty(&mddev->disks) &&
3779 mddev->raid_disks == 0 &&
3780 mddev->dev_sectors == 0)
3785 return sprintf(page, "%s\n", array_states[st]);
3788 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3789 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3790 static int do_md_run(struct mddev *mddev);
3791 static int restart_array(struct mddev *mddev);
3794 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3797 enum array_state st = match_word(buf, array_states);
3799 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3800 /* don't take reconfig_mutex when toggling between
3803 spin_lock(&mddev->lock);
3805 restart_array(mddev);
3806 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3807 wake_up(&mddev->sb_wait);
3809 } else /* st == clean */ {
3810 restart_array(mddev);
3811 if (atomic_read(&mddev->writes_pending) == 0) {
3812 if (mddev->in_sync == 0) {
3814 if (mddev->safemode == 1)
3815 mddev->safemode = 0;
3816 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3822 spin_unlock(&mddev->lock);
3825 err = mddev_lock(mddev);
3833 /* stopping an active array */
3834 err = do_md_stop(mddev, 0, NULL);
3837 /* stopping an active array */
3839 err = do_md_stop(mddev, 2, NULL);
3841 err = 0; /* already inactive */
3844 break; /* not supported yet */
3847 err = md_set_readonly(mddev, NULL);
3850 set_disk_ro(mddev->gendisk, 1);
3851 err = do_md_run(mddev);
3857 err = md_set_readonly(mddev, NULL);
3858 else if (mddev->ro == 1)
3859 err = restart_array(mddev);
3862 set_disk_ro(mddev->gendisk, 0);
3866 err = do_md_run(mddev);
3871 restart_array(mddev);
3872 spin_lock(&mddev->lock);
3873 if (atomic_read(&mddev->writes_pending) == 0) {
3874 if (mddev->in_sync == 0) {
3876 if (mddev->safemode == 1)
3877 mddev->safemode = 0;
3878 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3883 spin_unlock(&mddev->lock);
3889 restart_array(mddev);
3890 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3891 wake_up(&mddev->sb_wait);
3895 set_disk_ro(mddev->gendisk, 0);
3896 err = do_md_run(mddev);
3901 /* these cannot be set */
3906 if (mddev->hold_active == UNTIL_IOCTL)
3907 mddev->hold_active = 0;
3908 sysfs_notify_dirent_safe(mddev->sysfs_state);
3910 mddev_unlock(mddev);
3913 static struct md_sysfs_entry md_array_state =
3914 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3917 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3918 return sprintf(page, "%d\n",
3919 atomic_read(&mddev->max_corr_read_errors));
3923 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3928 rv = kstrtouint(buf, 10, &n);
3931 atomic_set(&mddev->max_corr_read_errors, n);
3935 static struct md_sysfs_entry max_corr_read_errors =
3936 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3937 max_corrected_read_errors_store);
3940 null_show(struct mddev *mddev, char *page)
3946 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3948 /* buf must be %d:%d\n? giving major and minor numbers */
3949 /* The new device is added to the array.
3950 * If the array has a persistent superblock, we read the
3951 * superblock to initialise info and check validity.
3952 * Otherwise, only checking done is that in bind_rdev_to_array,
3953 * which mainly checks size.
3956 int major = simple_strtoul(buf, &e, 10);
3959 struct md_rdev *rdev;
3962 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3964 minor = simple_strtoul(e+1, &e, 10);
3965 if (*e && *e != '\n')
3967 dev = MKDEV(major, minor);
3968 if (major != MAJOR(dev) ||
3969 minor != MINOR(dev))
3972 flush_workqueue(md_misc_wq);
3974 err = mddev_lock(mddev);
3977 if (mddev->persistent) {
3978 rdev = md_import_device(dev, mddev->major_version,
3979 mddev->minor_version);
3980 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3981 struct md_rdev *rdev0
3982 = list_entry(mddev->disks.next,
3983 struct md_rdev, same_set);
3984 err = super_types[mddev->major_version]
3985 .load_super(rdev, rdev0, mddev->minor_version);
3989 } else if (mddev->external)
3990 rdev = md_import_device(dev, -2, -1);
3992 rdev = md_import_device(dev, -1, -1);
3995 mddev_unlock(mddev);
3996 return PTR_ERR(rdev);
3998 err = bind_rdev_to_array(rdev, mddev);
4002 mddev_unlock(mddev);
4003 return err ? err : len;
4006 static struct md_sysfs_entry md_new_device =
4007 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4010 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4013 unsigned long chunk, end_chunk;
4016 err = mddev_lock(mddev);
4021 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4023 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4024 if (buf == end) break;
4025 if (*end == '-') { /* range */
4027 end_chunk = simple_strtoul(buf, &end, 0);
4028 if (buf == end) break;
4030 if (*end && !isspace(*end)) break;
4031 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4032 buf = skip_spaces(end);
4034 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4036 mddev_unlock(mddev);
4040 static struct md_sysfs_entry md_bitmap =
4041 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4044 size_show(struct mddev *mddev, char *page)
4046 return sprintf(page, "%llu\n",
4047 (unsigned long long)mddev->dev_sectors / 2);
4050 static int update_size(struct mddev *mddev, sector_t num_sectors);
4053 size_store(struct mddev *mddev, const char *buf, size_t len)
4055 /* If array is inactive, we can reduce the component size, but
4056 * not increase it (except from 0).
4057 * If array is active, we can try an on-line resize
4060 int err = strict_blocks_to_sectors(buf, §ors);
4064 err = mddev_lock(mddev);
4068 if (mddev_is_clustered(mddev))
4069 md_cluster_ops->metadata_update_start(mddev);
4070 err = update_size(mddev, sectors);
4071 md_update_sb(mddev, 1);
4072 if (mddev_is_clustered(mddev))
4073 md_cluster_ops->metadata_update_finish(mddev);
4075 if (mddev->dev_sectors == 0 ||
4076 mddev->dev_sectors > sectors)
4077 mddev->dev_sectors = sectors;
4081 mddev_unlock(mddev);
4082 return err ? err : len;
4085 static struct md_sysfs_entry md_size =
4086 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4088 /* Metadata version.
4090 * 'none' for arrays with no metadata (good luck...)
4091 * 'external' for arrays with externally managed metadata,
4092 * or N.M for internally known formats
4095 metadata_show(struct mddev *mddev, char *page)
4097 if (mddev->persistent)
4098 return sprintf(page, "%d.%d\n",
4099 mddev->major_version, mddev->minor_version);
4100 else if (mddev->external)
4101 return sprintf(page, "external:%s\n", mddev->metadata_type);
4103 return sprintf(page, "none\n");
4107 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4112 /* Changing the details of 'external' metadata is
4113 * always permitted. Otherwise there must be
4114 * no devices attached to the array.
4117 err = mddev_lock(mddev);
4121 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4123 else if (!list_empty(&mddev->disks))
4127 if (cmd_match(buf, "none")) {
4128 mddev->persistent = 0;
4129 mddev->external = 0;
4130 mddev->major_version = 0;
4131 mddev->minor_version = 90;
4134 if (strncmp(buf, "external:", 9) == 0) {
4135 size_t namelen = len-9;
4136 if (namelen >= sizeof(mddev->metadata_type))
4137 namelen = sizeof(mddev->metadata_type)-1;
4138 strncpy(mddev->metadata_type, buf+9, namelen);
4139 mddev->metadata_type[namelen] = 0;
4140 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4141 mddev->metadata_type[--namelen] = 0;
4142 mddev->persistent = 0;
4143 mddev->external = 1;
4144 mddev->major_version = 0;
4145 mddev->minor_version = 90;
4148 major = simple_strtoul(buf, &e, 10);
4150 if (e==buf || *e != '.')
4153 minor = simple_strtoul(buf, &e, 10);
4154 if (e==buf || (*e && *e != '\n') )
4157 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4159 mddev->major_version = major;
4160 mddev->minor_version = minor;
4161 mddev->persistent = 1;
4162 mddev->external = 0;
4165 mddev_unlock(mddev);
4169 static struct md_sysfs_entry md_metadata =
4170 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4173 action_show(struct mddev *mddev, char *page)
4175 char *type = "idle";
4176 unsigned long recovery = mddev->recovery;
4177 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4179 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4180 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4181 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4183 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4184 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4186 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4190 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4193 return sprintf(page, "%s\n", type);
4197 action_store(struct mddev *mddev, const char *page, size_t len)
4199 if (!mddev->pers || !mddev->pers->sync_request)
4203 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4204 if (cmd_match(page, "frozen"))
4205 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4207 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4208 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4209 mddev_lock(mddev) == 0) {
4210 flush_workqueue(md_misc_wq);
4211 if (mddev->sync_thread) {
4212 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4213 md_reap_sync_thread(mddev);
4215 mddev_unlock(mddev);
4217 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4218 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4220 else if (cmd_match(page, "resync"))
4221 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4222 else if (cmd_match(page, "recover")) {
4223 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4224 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4225 } else if (cmd_match(page, "reshape")) {
4227 if (mddev->pers->start_reshape == NULL)
4229 err = mddev_lock(mddev);
4231 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4232 err = mddev->pers->start_reshape(mddev);
4233 mddev_unlock(mddev);
4237 sysfs_notify(&mddev->kobj, NULL, "degraded");
4239 if (cmd_match(page, "check"))
4240 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4241 else if (!cmd_match(page, "repair"))
4243 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4244 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4245 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4247 if (mddev->ro == 2) {
4248 /* A write to sync_action is enough to justify
4249 * canceling read-auto mode
4252 md_wakeup_thread(mddev->sync_thread);
4254 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4255 md_wakeup_thread(mddev->thread);
4256 sysfs_notify_dirent_safe(mddev->sysfs_action);
4260 static struct md_sysfs_entry md_scan_mode =
4261 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4264 last_sync_action_show(struct mddev *mddev, char *page)
4266 return sprintf(page, "%s\n", mddev->last_sync_action);
4269 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4272 mismatch_cnt_show(struct mddev *mddev, char *page)
4274 return sprintf(page, "%llu\n",
4275 (unsigned long long)
4276 atomic64_read(&mddev->resync_mismatches));
4279 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4282 sync_min_show(struct mddev *mddev, char *page)
4284 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4285 mddev->sync_speed_min ? "local": "system");
4289 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4294 if (strncmp(buf, "system", 6)==0) {
4297 rv = kstrtouint(buf, 10, &min);
4303 mddev->sync_speed_min = min;
4307 static struct md_sysfs_entry md_sync_min =
4308 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4311 sync_max_show(struct mddev *mddev, char *page)
4313 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4314 mddev->sync_speed_max ? "local": "system");
4318 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4323 if (strncmp(buf, "system", 6)==0) {
4326 rv = kstrtouint(buf, 10, &max);
4332 mddev->sync_speed_max = max;
4336 static struct md_sysfs_entry md_sync_max =
4337 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4340 degraded_show(struct mddev *mddev, char *page)
4342 return sprintf(page, "%d\n", mddev->degraded);
4344 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4347 sync_force_parallel_show(struct mddev *mddev, char *page)
4349 return sprintf(page, "%d\n", mddev->parallel_resync);
4353 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4357 if (kstrtol(buf, 10, &n))
4360 if (n != 0 && n != 1)
4363 mddev->parallel_resync = n;
4365 if (mddev->sync_thread)
4366 wake_up(&resync_wait);
4371 /* force parallel resync, even with shared block devices */
4372 static struct md_sysfs_entry md_sync_force_parallel =
4373 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4374 sync_force_parallel_show, sync_force_parallel_store);
4377 sync_speed_show(struct mddev *mddev, char *page)
4379 unsigned long resync, dt, db;
4380 if (mddev->curr_resync == 0)
4381 return sprintf(page, "none\n");
4382 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4383 dt = (jiffies - mddev->resync_mark) / HZ;
4385 db = resync - mddev->resync_mark_cnt;
4386 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4389 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4392 sync_completed_show(struct mddev *mddev, char *page)
4394 unsigned long long max_sectors, resync;
4396 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4397 return sprintf(page, "none\n");
4399 if (mddev->curr_resync == 1 ||
4400 mddev->curr_resync == 2)
4401 return sprintf(page, "delayed\n");
4403 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4404 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4405 max_sectors = mddev->resync_max_sectors;
4407 max_sectors = mddev->dev_sectors;
4409 resync = mddev->curr_resync_completed;
4410 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4413 static struct md_sysfs_entry md_sync_completed =
4414 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4417 min_sync_show(struct mddev *mddev, char *page)
4419 return sprintf(page, "%llu\n",
4420 (unsigned long long)mddev->resync_min);
4423 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4425 unsigned long long min;
4428 if (kstrtoull(buf, 10, &min))
4431 spin_lock(&mddev->lock);
4433 if (min > mddev->resync_max)
4437 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4440 /* Round down to multiple of 4K for safety */
4441 mddev->resync_min = round_down(min, 8);
4445 spin_unlock(&mddev->lock);
4449 static struct md_sysfs_entry md_min_sync =
4450 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4453 max_sync_show(struct mddev *mddev, char *page)
4455 if (mddev->resync_max == MaxSector)
4456 return sprintf(page, "max\n");
4458 return sprintf(page, "%llu\n",
4459 (unsigned long long)mddev->resync_max);
4462 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4465 spin_lock(&mddev->lock);
4466 if (strncmp(buf, "max", 3) == 0)
4467 mddev->resync_max = MaxSector;
4469 unsigned long long max;
4473 if (kstrtoull(buf, 10, &max))
4475 if (max < mddev->resync_min)
4479 if (max < mddev->resync_max &&
4481 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4484 /* Must be a multiple of chunk_size */
4485 chunk = mddev->chunk_sectors;
4487 sector_t temp = max;
4490 if (sector_div(temp, chunk))
4493 mddev->resync_max = max;
4495 wake_up(&mddev->recovery_wait);
4498 spin_unlock(&mddev->lock);
4502 static struct md_sysfs_entry md_max_sync =
4503 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4506 suspend_lo_show(struct mddev *mddev, char *page)
4508 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4512 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4514 unsigned long long old, new;
4517 err = kstrtoull(buf, 10, &new);
4520 if (new != (sector_t)new)
4523 err = mddev_lock(mddev);
4527 if (mddev->pers == NULL ||
4528 mddev->pers->quiesce == NULL)
4530 old = mddev->suspend_lo;
4531 mddev->suspend_lo = new;
4533 /* Shrinking suspended region */
4534 mddev->pers->quiesce(mddev, 2);
4536 /* Expanding suspended region - need to wait */
4537 mddev->pers->quiesce(mddev, 1);
4538 mddev->pers->quiesce(mddev, 0);
4542 mddev_unlock(mddev);
4545 static struct md_sysfs_entry md_suspend_lo =
4546 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4549 suspend_hi_show(struct mddev *mddev, char *page)
4551 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4555 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4557 unsigned long long old, new;
4560 err = kstrtoull(buf, 10, &new);
4563 if (new != (sector_t)new)
4566 err = mddev_lock(mddev);
4570 if (mddev->pers == NULL ||
4571 mddev->pers->quiesce == NULL)
4573 old = mddev->suspend_hi;
4574 mddev->suspend_hi = new;
4576 /* Shrinking suspended region */
4577 mddev->pers->quiesce(mddev, 2);
4579 /* Expanding suspended region - need to wait */
4580 mddev->pers->quiesce(mddev, 1);
4581 mddev->pers->quiesce(mddev, 0);
4585 mddev_unlock(mddev);
4588 static struct md_sysfs_entry md_suspend_hi =
4589 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4592 reshape_position_show(struct mddev *mddev, char *page)
4594 if (mddev->reshape_position != MaxSector)
4595 return sprintf(page, "%llu\n",
4596 (unsigned long long)mddev->reshape_position);
4597 strcpy(page, "none\n");
4602 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4604 struct md_rdev *rdev;
4605 unsigned long long new;
4608 err = kstrtoull(buf, 10, &new);
4611 if (new != (sector_t)new)
4613 err = mddev_lock(mddev);
4619 mddev->reshape_position = new;
4620 mddev->delta_disks = 0;
4621 mddev->reshape_backwards = 0;
4622 mddev->new_level = mddev->level;
4623 mddev->new_layout = mddev->layout;
4624 mddev->new_chunk_sectors = mddev->chunk_sectors;
4625 rdev_for_each(rdev, mddev)
4626 rdev->new_data_offset = rdev->data_offset;
4629 mddev_unlock(mddev);
4633 static struct md_sysfs_entry md_reshape_position =
4634 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4635 reshape_position_store);
4638 reshape_direction_show(struct mddev *mddev, char *page)
4640 return sprintf(page, "%s\n",
4641 mddev->reshape_backwards ? "backwards" : "forwards");
4645 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4650 if (cmd_match(buf, "forwards"))
4652 else if (cmd_match(buf, "backwards"))
4656 if (mddev->reshape_backwards == backwards)
4659 err = mddev_lock(mddev);
4662 /* check if we are allowed to change */
4663 if (mddev->delta_disks)
4665 else if (mddev->persistent &&
4666 mddev->major_version == 0)
4669 mddev->reshape_backwards = backwards;
4670 mddev_unlock(mddev);
4674 static struct md_sysfs_entry md_reshape_direction =
4675 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4676 reshape_direction_store);
4679 array_size_show(struct mddev *mddev, char *page)
4681 if (mddev->external_size)
4682 return sprintf(page, "%llu\n",
4683 (unsigned long long)mddev->array_sectors/2);
4685 return sprintf(page, "default\n");
4689 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4694 err = mddev_lock(mddev);
4698 if (strncmp(buf, "default", 7) == 0) {
4700 sectors = mddev->pers->size(mddev, 0, 0);
4702 sectors = mddev->array_sectors;
4704 mddev->external_size = 0;
4706 if (strict_blocks_to_sectors(buf, §ors) < 0)
4708 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4711 mddev->external_size = 1;
4715 mddev->array_sectors = sectors;
4717 set_capacity(mddev->gendisk, mddev->array_sectors);
4718 revalidate_disk(mddev->gendisk);
4721 mddev_unlock(mddev);
4725 static struct md_sysfs_entry md_array_size =
4726 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4729 static struct attribute *md_default_attrs[] = {
4732 &md_raid_disks.attr,
4733 &md_chunk_size.attr,
4735 &md_resync_start.attr,
4737 &md_new_device.attr,
4738 &md_safe_delay.attr,
4739 &md_array_state.attr,
4740 &md_reshape_position.attr,
4741 &md_reshape_direction.attr,
4742 &md_array_size.attr,
4743 &max_corr_read_errors.attr,
4747 static struct attribute *md_redundancy_attrs[] = {
4749 &md_last_scan_mode.attr,
4750 &md_mismatches.attr,
4753 &md_sync_speed.attr,
4754 &md_sync_force_parallel.attr,
4755 &md_sync_completed.attr,
4758 &md_suspend_lo.attr,
4759 &md_suspend_hi.attr,
4764 static struct attribute_group md_redundancy_group = {
4766 .attrs = md_redundancy_attrs,
4770 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4772 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4773 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4778 spin_lock(&all_mddevs_lock);
4779 if (list_empty(&mddev->all_mddevs)) {
4780 spin_unlock(&all_mddevs_lock);
4784 spin_unlock(&all_mddevs_lock);
4786 rv = entry->show(mddev, page);
4792 md_attr_store(struct kobject *kobj, struct attribute *attr,
4793 const char *page, size_t length)
4795 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4796 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4801 if (!capable(CAP_SYS_ADMIN))
4803 spin_lock(&all_mddevs_lock);
4804 if (list_empty(&mddev->all_mddevs)) {
4805 spin_unlock(&all_mddevs_lock);
4809 spin_unlock(&all_mddevs_lock);
4810 rv = entry->store(mddev, page, length);
4815 static void md_free(struct kobject *ko)
4817 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4819 if (mddev->sysfs_state)
4820 sysfs_put(mddev->sysfs_state);
4823 blk_cleanup_queue(mddev->queue);
4824 if (mddev->gendisk) {
4825 del_gendisk(mddev->gendisk);
4826 put_disk(mddev->gendisk);
4832 static const struct sysfs_ops md_sysfs_ops = {
4833 .show = md_attr_show,
4834 .store = md_attr_store,
4836 static struct kobj_type md_ktype = {
4838 .sysfs_ops = &md_sysfs_ops,
4839 .default_attrs = md_default_attrs,
4844 static void mddev_delayed_delete(struct work_struct *ws)
4846 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4848 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4849 kobject_del(&mddev->kobj);
4850 kobject_put(&mddev->kobj);
4853 static int md_alloc(dev_t dev, char *name)
4855 static DEFINE_MUTEX(disks_mutex);
4856 struct mddev *mddev = mddev_find(dev);
4857 struct gendisk *disk;
4866 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4867 shift = partitioned ? MdpMinorShift : 0;
4868 unit = MINOR(mddev->unit) >> shift;
4870 /* wait for any previous instance of this device to be
4871 * completely removed (mddev_delayed_delete).
4873 flush_workqueue(md_misc_wq);
4875 mutex_lock(&disks_mutex);
4881 /* Need to ensure that 'name' is not a duplicate.
4883 struct mddev *mddev2;
4884 spin_lock(&all_mddevs_lock);
4886 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4887 if (mddev2->gendisk &&
4888 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4889 spin_unlock(&all_mddevs_lock);
4892 spin_unlock(&all_mddevs_lock);
4896 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4899 mddev->queue->queuedata = mddev;
4901 blk_queue_make_request(mddev->queue, md_make_request);
4902 blk_set_stacking_limits(&mddev->queue->limits);
4904 disk = alloc_disk(1 << shift);
4906 blk_cleanup_queue(mddev->queue);
4907 mddev->queue = NULL;
4910 disk->major = MAJOR(mddev->unit);
4911 disk->first_minor = unit << shift;
4913 strcpy(disk->disk_name, name);
4914 else if (partitioned)
4915 sprintf(disk->disk_name, "md_d%d", unit);
4917 sprintf(disk->disk_name, "md%d", unit);
4918 disk->fops = &md_fops;
4919 disk->private_data = mddev;
4920 disk->queue = mddev->queue;
4921 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4922 /* Allow extended partitions. This makes the
4923 * 'mdp' device redundant, but we can't really
4926 disk->flags |= GENHD_FL_EXT_DEVT;
4927 mddev->gendisk = disk;
4928 /* As soon as we call add_disk(), another thread could get
4929 * through to md_open, so make sure it doesn't get too far
4931 mutex_lock(&mddev->open_mutex);
4934 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4935 &disk_to_dev(disk)->kobj, "%s", "md");
4937 /* This isn't possible, but as kobject_init_and_add is marked
4938 * __must_check, we must do something with the result
4940 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4944 if (mddev->kobj.sd &&
4945 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4946 printk(KERN_DEBUG "pointless warning\n");
4947 mutex_unlock(&mddev->open_mutex);
4949 mutex_unlock(&disks_mutex);
4950 if (!error && mddev->kobj.sd) {
4951 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4952 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4958 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4960 md_alloc(dev, NULL);
4964 static int add_named_array(const char *val, struct kernel_param *kp)
4966 /* val must be "md_*" where * is not all digits.
4967 * We allocate an array with a large free minor number, and
4968 * set the name to val. val must not already be an active name.
4970 int len = strlen(val);
4971 char buf[DISK_NAME_LEN];
4973 while (len && val[len-1] == '\n')
4975 if (len >= DISK_NAME_LEN)
4977 strlcpy(buf, val, len+1);
4978 if (strncmp(buf, "md_", 3) != 0)
4980 return md_alloc(0, buf);
4983 static void md_safemode_timeout(unsigned long data)
4985 struct mddev *mddev = (struct mddev *) data;
4987 if (!atomic_read(&mddev->writes_pending)) {
4988 mddev->safemode = 1;
4989 if (mddev->external)
4990 sysfs_notify_dirent_safe(mddev->sysfs_state);
4992 md_wakeup_thread(mddev->thread);
4995 static int start_dirty_degraded;
4997 int md_run(struct mddev *mddev)
5000 struct md_rdev *rdev;
5001 struct md_personality *pers;
5003 if (list_empty(&mddev->disks))
5004 /* cannot run an array with no devices.. */
5009 /* Cannot run until previous stop completes properly */
5010 if (mddev->sysfs_active)
5014 * Analyze all RAID superblock(s)
5016 if (!mddev->raid_disks) {
5017 if (!mddev->persistent)
5022 if (mddev->level != LEVEL_NONE)
5023 request_module("md-level-%d", mddev->level);
5024 else if (mddev->clevel[0])
5025 request_module("md-%s", mddev->clevel);
5028 * Drop all container device buffers, from now on
5029 * the only valid external interface is through the md
5032 rdev_for_each(rdev, mddev) {
5033 if (test_bit(Faulty, &rdev->flags))
5035 sync_blockdev(rdev->bdev);
5036 invalidate_bdev(rdev->bdev);
5038 /* perform some consistency tests on the device.
5039 * We don't want the data to overlap the metadata,
5040 * Internal Bitmap issues have been handled elsewhere.
5042 if (rdev->meta_bdev) {
5043 /* Nothing to check */;
5044 } else if (rdev->data_offset < rdev->sb_start) {
5045 if (mddev->dev_sectors &&
5046 rdev->data_offset + mddev->dev_sectors
5048 printk("md: %s: data overlaps metadata\n",
5053 if (rdev->sb_start + rdev->sb_size/512
5054 > rdev->data_offset) {
5055 printk("md: %s: metadata overlaps data\n",
5060 sysfs_notify_dirent_safe(rdev->sysfs_state);
5063 if (mddev->bio_set == NULL)
5064 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5066 spin_lock(&pers_lock);
5067 pers = find_pers(mddev->level, mddev->clevel);
5068 if (!pers || !try_module_get(pers->owner)) {
5069 spin_unlock(&pers_lock);
5070 if (mddev->level != LEVEL_NONE)
5071 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5074 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5078 spin_unlock(&pers_lock);
5079 if (mddev->level != pers->level) {
5080 mddev->level = pers->level;
5081 mddev->new_level = pers->level;
5083 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5085 if (mddev->reshape_position != MaxSector &&
5086 pers->start_reshape == NULL) {
5087 /* This personality cannot handle reshaping... */
5088 module_put(pers->owner);
5092 if (pers->sync_request) {
5093 /* Warn if this is a potentially silly
5096 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5097 struct md_rdev *rdev2;
5100 rdev_for_each(rdev, mddev)
5101 rdev_for_each(rdev2, mddev) {
5103 rdev->bdev->bd_contains ==
5104 rdev2->bdev->bd_contains) {
5106 "%s: WARNING: %s appears to be"
5107 " on the same physical disk as"
5110 bdevname(rdev->bdev,b),
5111 bdevname(rdev2->bdev,b2));
5118 "True protection against single-disk"
5119 " failure might be compromised.\n");
5122 mddev->recovery = 0;
5123 /* may be over-ridden by personality */
5124 mddev->resync_max_sectors = mddev->dev_sectors;
5126 mddev->ok_start_degraded = start_dirty_degraded;
5128 if (start_readonly && mddev->ro == 0)
5129 mddev->ro = 2; /* read-only, but switch on first write */
5131 err = pers->run(mddev);
5133 printk(KERN_ERR "md: pers->run() failed ...\n");
5134 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5135 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5136 " but 'external_size' not in effect?\n", __func__);
5138 "md: invalid array_size %llu > default size %llu\n",
5139 (unsigned long long)mddev->array_sectors / 2,
5140 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5143 if (err == 0 && pers->sync_request &&
5144 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5145 struct bitmap *bitmap;
5147 bitmap = bitmap_create(mddev, -1);
5148 if (IS_ERR(bitmap)) {
5149 err = PTR_ERR(bitmap);
5150 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5151 mdname(mddev), err);
5153 mddev->bitmap = bitmap;
5157 mddev_detach(mddev);
5159 pers->free(mddev, mddev->private);
5160 mddev->private = NULL;
5161 module_put(pers->owner);
5162 bitmap_destroy(mddev);
5166 mddev->queue->backing_dev_info.congested_data = mddev;
5167 mddev->queue->backing_dev_info.congested_fn = md_congested;
5169 if (pers->sync_request) {
5170 if (mddev->kobj.sd &&
5171 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5173 "md: cannot register extra attributes for %s\n",
5175 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5176 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5179 atomic_set(&mddev->writes_pending,0);
5180 atomic_set(&mddev->max_corr_read_errors,
5181 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5182 mddev->safemode = 0;
5183 mddev->safemode_timer.function = md_safemode_timeout;
5184 mddev->safemode_timer.data = (unsigned long) mddev;
5185 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5188 spin_lock(&mddev->lock);
5191 spin_unlock(&mddev->lock);
5192 rdev_for_each(rdev, mddev)
5193 if (rdev->raid_disk >= 0)
5194 if (sysfs_link_rdev(mddev, rdev))
5195 /* failure here is OK */;
5197 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5199 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5200 md_update_sb(mddev, 0);
5202 md_new_event(mddev);
5203 sysfs_notify_dirent_safe(mddev->sysfs_state);
5204 sysfs_notify_dirent_safe(mddev->sysfs_action);
5205 sysfs_notify(&mddev->kobj, NULL, "degraded");
5208 EXPORT_SYMBOL_GPL(md_run);
5210 static int do_md_run(struct mddev *mddev)
5214 err = md_run(mddev);
5217 err = bitmap_load(mddev);
5219 bitmap_destroy(mddev);
5223 md_wakeup_thread(mddev->thread);
5224 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5226 set_capacity(mddev->gendisk, mddev->array_sectors);
5227 revalidate_disk(mddev->gendisk);
5229 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5234 static int restart_array(struct mddev *mddev)
5236 struct gendisk *disk = mddev->gendisk;
5238 /* Complain if it has no devices */
5239 if (list_empty(&mddev->disks))
5245 mddev->safemode = 0;
5247 set_disk_ro(disk, 0);
5248 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5250 /* Kick recovery or resync if necessary */
5251 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5252 md_wakeup_thread(mddev->thread);
5253 md_wakeup_thread(mddev->sync_thread);
5254 sysfs_notify_dirent_safe(mddev->sysfs_state);
5258 static void md_clean(struct mddev *mddev)
5260 mddev->array_sectors = 0;
5261 mddev->external_size = 0;
5262 mddev->dev_sectors = 0;
5263 mddev->raid_disks = 0;
5264 mddev->recovery_cp = 0;
5265 mddev->resync_min = 0;
5266 mddev->resync_max = MaxSector;
5267 mddev->reshape_position = MaxSector;
5268 mddev->external = 0;
5269 mddev->persistent = 0;
5270 mddev->level = LEVEL_NONE;
5271 mddev->clevel[0] = 0;
5274 mddev->metadata_type[0] = 0;
5275 mddev->chunk_sectors = 0;
5276 mddev->ctime = mddev->utime = 0;
5278 mddev->max_disks = 0;
5280 mddev->can_decrease_events = 0;
5281 mddev->delta_disks = 0;
5282 mddev->reshape_backwards = 0;
5283 mddev->new_level = LEVEL_NONE;
5284 mddev->new_layout = 0;
5285 mddev->new_chunk_sectors = 0;
5286 mddev->curr_resync = 0;
5287 atomic64_set(&mddev->resync_mismatches, 0);
5288 mddev->suspend_lo = mddev->suspend_hi = 0;
5289 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5290 mddev->recovery = 0;
5293 mddev->degraded = 0;
5294 mddev->safemode = 0;
5295 mddev->private = NULL;
5296 mddev->bitmap_info.offset = 0;
5297 mddev->bitmap_info.default_offset = 0;
5298 mddev->bitmap_info.default_space = 0;
5299 mddev->bitmap_info.chunksize = 0;
5300 mddev->bitmap_info.daemon_sleep = 0;
5301 mddev->bitmap_info.max_write_behind = 0;
5304 static void __md_stop_writes(struct mddev *mddev)
5306 if (mddev_is_clustered(mddev))
5307 md_cluster_ops->metadata_update_start(mddev);
5308 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5309 flush_workqueue(md_misc_wq);
5310 if (mddev->sync_thread) {
5311 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5312 md_reap_sync_thread(mddev);
5315 del_timer_sync(&mddev->safemode_timer);
5317 bitmap_flush(mddev);
5318 md_super_wait(mddev);
5320 if (mddev->ro == 0 &&
5321 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5322 /* mark array as shutdown cleanly */
5324 md_update_sb(mddev, 1);
5326 if (mddev_is_clustered(mddev))
5327 md_cluster_ops->metadata_update_finish(mddev);
5330 void md_stop_writes(struct mddev *mddev)
5332 mddev_lock_nointr(mddev);
5333 __md_stop_writes(mddev);
5334 mddev_unlock(mddev);
5336 EXPORT_SYMBOL_GPL(md_stop_writes);
5338 static void mddev_detach(struct mddev *mddev)
5340 struct bitmap *bitmap = mddev->bitmap;
5341 /* wait for behind writes to complete */
5342 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5343 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5345 /* need to kick something here to make sure I/O goes? */
5346 wait_event(bitmap->behind_wait,
5347 atomic_read(&bitmap->behind_writes) == 0);
5349 if (mddev->pers && mddev->pers->quiesce) {
5350 mddev->pers->quiesce(mddev, 1);
5351 mddev->pers->quiesce(mddev, 0);
5353 md_unregister_thread(&mddev->thread);
5355 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5358 static void __md_stop(struct mddev *mddev)
5360 struct md_personality *pers = mddev->pers;
5361 mddev_detach(mddev);
5362 /* Ensure ->event_work is done */
5363 flush_workqueue(md_misc_wq);
5364 spin_lock(&mddev->lock);
5367 spin_unlock(&mddev->lock);
5368 pers->free(mddev, mddev->private);
5369 mddev->private = NULL;
5370 if (pers->sync_request && mddev->to_remove == NULL)
5371 mddev->to_remove = &md_redundancy_group;
5372 module_put(pers->owner);
5373 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5376 void md_stop(struct mddev *mddev)
5378 /* stop the array and free an attached data structures.
5379 * This is called from dm-raid
5382 bitmap_destroy(mddev);
5384 bioset_free(mddev->bio_set);
5387 EXPORT_SYMBOL_GPL(md_stop);
5389 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5394 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5396 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5397 md_wakeup_thread(mddev->thread);
5399 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5400 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5401 if (mddev->sync_thread)
5402 /* Thread might be blocked waiting for metadata update
5403 * which will now never happen */
5404 wake_up_process(mddev->sync_thread->tsk);
5406 mddev_unlock(mddev);
5407 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5409 mddev_lock_nointr(mddev);
5411 mutex_lock(&mddev->open_mutex);
5412 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5413 mddev->sync_thread ||
5414 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5415 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5416 printk("md: %s still in use.\n",mdname(mddev));
5418 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5419 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5420 md_wakeup_thread(mddev->thread);
5426 __md_stop_writes(mddev);
5432 set_disk_ro(mddev->gendisk, 1);
5433 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5434 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5435 md_wakeup_thread(mddev->thread);
5436 sysfs_notify_dirent_safe(mddev->sysfs_state);
5440 mutex_unlock(&mddev->open_mutex);
5445 * 0 - completely stop and dis-assemble array
5446 * 2 - stop but do not disassemble array
5448 static int do_md_stop(struct mddev *mddev, int mode,
5449 struct block_device *bdev)
5451 struct gendisk *disk = mddev->gendisk;
5452 struct md_rdev *rdev;
5455 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5457 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5458 md_wakeup_thread(mddev->thread);
5460 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5461 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5462 if (mddev->sync_thread)
5463 /* Thread might be blocked waiting for metadata update
5464 * which will now never happen */
5465 wake_up_process(mddev->sync_thread->tsk);
5467 mddev_unlock(mddev);
5468 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5469 !test_bit(MD_RECOVERY_RUNNING,
5470 &mddev->recovery)));
5471 mddev_lock_nointr(mddev);
5473 mutex_lock(&mddev->open_mutex);
5474 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5475 mddev->sysfs_active ||
5476 mddev->sync_thread ||
5477 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5478 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5479 printk("md: %s still in use.\n",mdname(mddev));
5480 mutex_unlock(&mddev->open_mutex);
5482 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5483 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5484 md_wakeup_thread(mddev->thread);
5490 set_disk_ro(disk, 0);
5492 __md_stop_writes(mddev);
5494 mddev->queue->backing_dev_info.congested_fn = NULL;
5496 /* tell userspace to handle 'inactive' */
5497 sysfs_notify_dirent_safe(mddev->sysfs_state);
5499 rdev_for_each(rdev, mddev)
5500 if (rdev->raid_disk >= 0)
5501 sysfs_unlink_rdev(mddev, rdev);
5503 set_capacity(disk, 0);
5504 mutex_unlock(&mddev->open_mutex);
5506 revalidate_disk(disk);
5511 mutex_unlock(&mddev->open_mutex);
5513 * Free resources if final stop
5516 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5518 bitmap_destroy(mddev);
5519 if (mddev->bitmap_info.file) {
5520 struct file *f = mddev->bitmap_info.file;
5521 spin_lock(&mddev->lock);
5522 mddev->bitmap_info.file = NULL;
5523 spin_unlock(&mddev->lock);
5526 mddev->bitmap_info.offset = 0;
5528 export_array(mddev);
5531 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5532 if (mddev->hold_active == UNTIL_STOP)
5533 mddev->hold_active = 0;
5535 blk_integrity_unregister(disk);
5536 md_new_event(mddev);
5537 sysfs_notify_dirent_safe(mddev->sysfs_state);
5542 static void autorun_array(struct mddev *mddev)
5544 struct md_rdev *rdev;
5547 if (list_empty(&mddev->disks))
5550 printk(KERN_INFO "md: running: ");
5552 rdev_for_each(rdev, mddev) {
5553 char b[BDEVNAME_SIZE];
5554 printk("<%s>", bdevname(rdev->bdev,b));
5558 err = do_md_run(mddev);
5560 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5561 do_md_stop(mddev, 0, NULL);
5566 * lets try to run arrays based on all disks that have arrived
5567 * until now. (those are in pending_raid_disks)
5569 * the method: pick the first pending disk, collect all disks with
5570 * the same UUID, remove all from the pending list and put them into
5571 * the 'same_array' list. Then order this list based on superblock
5572 * update time (freshest comes first), kick out 'old' disks and
5573 * compare superblocks. If everything's fine then run it.
5575 * If "unit" is allocated, then bump its reference count
5577 static void autorun_devices(int part)
5579 struct md_rdev *rdev0, *rdev, *tmp;
5580 struct mddev *mddev;
5581 char b[BDEVNAME_SIZE];
5583 printk(KERN_INFO "md: autorun ...\n");
5584 while (!list_empty(&pending_raid_disks)) {
5587 LIST_HEAD(candidates);
5588 rdev0 = list_entry(pending_raid_disks.next,
5589 struct md_rdev, same_set);
5591 printk(KERN_INFO "md: considering %s ...\n",
5592 bdevname(rdev0->bdev,b));
5593 INIT_LIST_HEAD(&candidates);
5594 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5595 if (super_90_load(rdev, rdev0, 0) >= 0) {
5596 printk(KERN_INFO "md: adding %s ...\n",
5597 bdevname(rdev->bdev,b));
5598 list_move(&rdev->same_set, &candidates);
5601 * now we have a set of devices, with all of them having
5602 * mostly sane superblocks. It's time to allocate the
5606 dev = MKDEV(mdp_major,
5607 rdev0->preferred_minor << MdpMinorShift);
5608 unit = MINOR(dev) >> MdpMinorShift;
5610 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5613 if (rdev0->preferred_minor != unit) {
5614 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5615 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5619 md_probe(dev, NULL, NULL);
5620 mddev = mddev_find(dev);
5621 if (!mddev || !mddev->gendisk) {
5625 "md: cannot allocate memory for md drive.\n");
5628 if (mddev_lock(mddev))
5629 printk(KERN_WARNING "md: %s locked, cannot run\n",
5631 else if (mddev->raid_disks || mddev->major_version
5632 || !list_empty(&mddev->disks)) {
5634 "md: %s already running, cannot run %s\n",
5635 mdname(mddev), bdevname(rdev0->bdev,b));
5636 mddev_unlock(mddev);
5638 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5639 mddev->persistent = 1;
5640 rdev_for_each_list(rdev, tmp, &candidates) {
5641 list_del_init(&rdev->same_set);
5642 if (bind_rdev_to_array(rdev, mddev))
5645 autorun_array(mddev);
5646 mddev_unlock(mddev);
5648 /* on success, candidates will be empty, on error
5651 rdev_for_each_list(rdev, tmp, &candidates) {
5652 list_del_init(&rdev->same_set);
5657 printk(KERN_INFO "md: ... autorun DONE.\n");
5659 #endif /* !MODULE */
5661 static int get_version(void __user *arg)
5665 ver.major = MD_MAJOR_VERSION;
5666 ver.minor = MD_MINOR_VERSION;
5667 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5669 if (copy_to_user(arg, &ver, sizeof(ver)))
5675 static int get_array_info(struct mddev *mddev, void __user *arg)
5677 mdu_array_info_t info;
5678 int nr,working,insync,failed,spare;
5679 struct md_rdev *rdev;
5681 nr = working = insync = failed = spare = 0;
5683 rdev_for_each_rcu(rdev, mddev) {
5685 if (test_bit(Faulty, &rdev->flags))
5689 if (test_bit(In_sync, &rdev->flags))
5697 info.major_version = mddev->major_version;
5698 info.minor_version = mddev->minor_version;
5699 info.patch_version = MD_PATCHLEVEL_VERSION;
5700 info.ctime = mddev->ctime;
5701 info.level = mddev->level;
5702 info.size = mddev->dev_sectors / 2;
5703 if (info.size != mddev->dev_sectors / 2) /* overflow */
5706 info.raid_disks = mddev->raid_disks;
5707 info.md_minor = mddev->md_minor;
5708 info.not_persistent= !mddev->persistent;
5710 info.utime = mddev->utime;
5713 info.state = (1<<MD_SB_CLEAN);
5714 if (mddev->bitmap && mddev->bitmap_info.offset)
5715 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5716 if (mddev_is_clustered(mddev))
5717 info.state |= (1<<MD_SB_CLUSTERED);
5718 info.active_disks = insync;
5719 info.working_disks = working;
5720 info.failed_disks = failed;
5721 info.spare_disks = spare;
5723 info.layout = mddev->layout;
5724 info.chunk_size = mddev->chunk_sectors << 9;
5726 if (copy_to_user(arg, &info, sizeof(info)))
5732 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5734 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5738 file = kzalloc(sizeof(*file), GFP_NOIO);
5743 spin_lock(&mddev->lock);
5744 /* bitmap disabled, zero the first byte and copy out */
5745 if (!mddev->bitmap_info.file)
5746 file->pathname[0] = '\0';
5747 else if ((ptr = file_path(mddev->bitmap_info.file,
5748 file->pathname, sizeof(file->pathname))),
5752 memmove(file->pathname, ptr,
5753 sizeof(file->pathname)-(ptr-file->pathname));
5754 spin_unlock(&mddev->lock);
5757 copy_to_user(arg, file, sizeof(*file)))
5764 static int get_disk_info(struct mddev *mddev, void __user * arg)
5766 mdu_disk_info_t info;
5767 struct md_rdev *rdev;
5769 if (copy_from_user(&info, arg, sizeof(info)))
5773 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5775 info.major = MAJOR(rdev->bdev->bd_dev);
5776 info.minor = MINOR(rdev->bdev->bd_dev);
5777 info.raid_disk = rdev->raid_disk;
5779 if (test_bit(Faulty, &rdev->flags))
5780 info.state |= (1<<MD_DISK_FAULTY);
5781 else if (test_bit(In_sync, &rdev->flags)) {
5782 info.state |= (1<<MD_DISK_ACTIVE);
5783 info.state |= (1<<MD_DISK_SYNC);
5785 if (test_bit(WriteMostly, &rdev->flags))
5786 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5788 info.major = info.minor = 0;
5789 info.raid_disk = -1;
5790 info.state = (1<<MD_DISK_REMOVED);
5794 if (copy_to_user(arg, &info, sizeof(info)))
5800 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5802 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5803 struct md_rdev *rdev;
5804 dev_t dev = MKDEV(info->major,info->minor);
5806 if (mddev_is_clustered(mddev) &&
5807 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5808 pr_err("%s: Cannot add to clustered mddev.\n",
5813 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5816 if (!mddev->raid_disks) {
5818 /* expecting a device which has a superblock */
5819 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5822 "md: md_import_device returned %ld\n",
5824 return PTR_ERR(rdev);
5826 if (!list_empty(&mddev->disks)) {
5827 struct md_rdev *rdev0
5828 = list_entry(mddev->disks.next,
5829 struct md_rdev, same_set);
5830 err = super_types[mddev->major_version]
5831 .load_super(rdev, rdev0, mddev->minor_version);
5834 "md: %s has different UUID to %s\n",
5835 bdevname(rdev->bdev,b),
5836 bdevname(rdev0->bdev,b2));
5841 err = bind_rdev_to_array(rdev, mddev);
5848 * add_new_disk can be used once the array is assembled
5849 * to add "hot spares". They must already have a superblock
5854 if (!mddev->pers->hot_add_disk) {
5856 "%s: personality does not support diskops!\n",
5860 if (mddev->persistent)
5861 rdev = md_import_device(dev, mddev->major_version,
5862 mddev->minor_version);
5864 rdev = md_import_device(dev, -1, -1);
5867 "md: md_import_device returned %ld\n",
5869 return PTR_ERR(rdev);
5871 /* set saved_raid_disk if appropriate */
5872 if (!mddev->persistent) {
5873 if (info->state & (1<<MD_DISK_SYNC) &&
5874 info->raid_disk < mddev->raid_disks) {
5875 rdev->raid_disk = info->raid_disk;
5876 set_bit(In_sync, &rdev->flags);
5877 clear_bit(Bitmap_sync, &rdev->flags);
5879 rdev->raid_disk = -1;
5880 rdev->saved_raid_disk = rdev->raid_disk;
5882 super_types[mddev->major_version].
5883 validate_super(mddev, rdev);
5884 if ((info->state & (1<<MD_DISK_SYNC)) &&
5885 rdev->raid_disk != info->raid_disk) {
5886 /* This was a hot-add request, but events doesn't
5887 * match, so reject it.
5893 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5894 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5895 set_bit(WriteMostly, &rdev->flags);
5897 clear_bit(WriteMostly, &rdev->flags);
5900 * check whether the device shows up in other nodes
5902 if (mddev_is_clustered(mddev)) {
5903 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5904 /* Through --cluster-confirm */
5905 set_bit(Candidate, &rdev->flags);
5906 err = md_cluster_ops->new_disk_ack(mddev, true);
5911 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5912 /* --add initiated by this node */
5913 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5915 md_cluster_ops->add_new_disk_finish(mddev);
5922 rdev->raid_disk = -1;
5923 err = bind_rdev_to_array(rdev, mddev);
5927 err = add_bound_rdev(rdev);
5928 if (mddev_is_clustered(mddev) &&
5929 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5930 md_cluster_ops->add_new_disk_finish(mddev);
5934 /* otherwise, add_new_disk is only allowed
5935 * for major_version==0 superblocks
5937 if (mddev->major_version != 0) {
5938 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5943 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5945 rdev = md_import_device(dev, -1, 0);
5948 "md: error, md_import_device() returned %ld\n",
5950 return PTR_ERR(rdev);
5952 rdev->desc_nr = info->number;
5953 if (info->raid_disk < mddev->raid_disks)
5954 rdev->raid_disk = info->raid_disk;
5956 rdev->raid_disk = -1;
5958 if (rdev->raid_disk < mddev->raid_disks)
5959 if (info->state & (1<<MD_DISK_SYNC))
5960 set_bit(In_sync, &rdev->flags);
5962 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5963 set_bit(WriteMostly, &rdev->flags);
5965 if (!mddev->persistent) {
5966 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5967 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5969 rdev->sb_start = calc_dev_sboffset(rdev);
5970 rdev->sectors = rdev->sb_start;
5972 err = bind_rdev_to_array(rdev, mddev);
5982 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5984 char b[BDEVNAME_SIZE];
5985 struct md_rdev *rdev;
5987 rdev = find_rdev(mddev, dev);
5991 if (mddev_is_clustered(mddev))
5992 md_cluster_ops->metadata_update_start(mddev);
5994 clear_bit(Blocked, &rdev->flags);
5995 remove_and_add_spares(mddev, rdev);
5997 if (rdev->raid_disk >= 0)
6000 if (mddev_is_clustered(mddev))
6001 md_cluster_ops->remove_disk(mddev, rdev);
6003 md_kick_rdev_from_array(rdev);
6004 md_update_sb(mddev, 1);
6005 md_new_event(mddev);
6007 if (mddev_is_clustered(mddev))
6008 md_cluster_ops->metadata_update_finish(mddev);
6012 if (mddev_is_clustered(mddev))
6013 md_cluster_ops->metadata_update_cancel(mddev);
6014 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6015 bdevname(rdev->bdev,b), mdname(mddev));
6019 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6021 char b[BDEVNAME_SIZE];
6023 struct md_rdev *rdev;
6028 if (mddev->major_version != 0) {
6029 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6030 " version-0 superblocks.\n",
6034 if (!mddev->pers->hot_add_disk) {
6036 "%s: personality does not support diskops!\n",
6041 rdev = md_import_device(dev, -1, 0);
6044 "md: error, md_import_device() returned %ld\n",
6049 if (mddev->persistent)
6050 rdev->sb_start = calc_dev_sboffset(rdev);
6052 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6054 rdev->sectors = rdev->sb_start;
6056 if (test_bit(Faulty, &rdev->flags)) {
6058 "md: can not hot-add faulty %s disk to %s!\n",
6059 bdevname(rdev->bdev,b), mdname(mddev));
6064 if (mddev_is_clustered(mddev))
6065 md_cluster_ops->metadata_update_start(mddev);
6066 clear_bit(In_sync, &rdev->flags);
6068 rdev->saved_raid_disk = -1;
6069 err = bind_rdev_to_array(rdev, mddev);
6071 goto abort_clustered;
6074 * The rest should better be atomic, we can have disk failures
6075 * noticed in interrupt contexts ...
6078 rdev->raid_disk = -1;
6080 md_update_sb(mddev, 1);
6082 if (mddev_is_clustered(mddev))
6083 md_cluster_ops->metadata_update_finish(mddev);
6085 * Kick recovery, maybe this spare has to be added to the
6086 * array immediately.
6088 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6089 md_wakeup_thread(mddev->thread);
6090 md_new_event(mddev);
6094 if (mddev_is_clustered(mddev))
6095 md_cluster_ops->metadata_update_cancel(mddev);
6101 static int set_bitmap_file(struct mddev *mddev, int fd)
6106 if (!mddev->pers->quiesce || !mddev->thread)
6108 if (mddev->recovery || mddev->sync_thread)
6110 /* we should be able to change the bitmap.. */
6114 struct inode *inode;
6117 if (mddev->bitmap || mddev->bitmap_info.file)
6118 return -EEXIST; /* cannot add when bitmap is present */
6122 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6127 inode = f->f_mapping->host;
6128 if (!S_ISREG(inode->i_mode)) {
6129 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6132 } else if (!(f->f_mode & FMODE_WRITE)) {
6133 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6136 } else if (atomic_read(&inode->i_writecount) != 1) {
6137 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6145 mddev->bitmap_info.file = f;
6146 mddev->bitmap_info.offset = 0; /* file overrides offset */
6147 } else if (mddev->bitmap == NULL)
6148 return -ENOENT; /* cannot remove what isn't there */
6151 mddev->pers->quiesce(mddev, 1);
6153 struct bitmap *bitmap;
6155 bitmap = bitmap_create(mddev, -1);
6156 if (!IS_ERR(bitmap)) {
6157 mddev->bitmap = bitmap;
6158 err = bitmap_load(mddev);
6160 err = PTR_ERR(bitmap);
6162 if (fd < 0 || err) {
6163 bitmap_destroy(mddev);
6164 fd = -1; /* make sure to put the file */
6166 mddev->pers->quiesce(mddev, 0);
6169 struct file *f = mddev->bitmap_info.file;
6171 spin_lock(&mddev->lock);
6172 mddev->bitmap_info.file = NULL;
6173 spin_unlock(&mddev->lock);
6182 * set_array_info is used two different ways
6183 * The original usage is when creating a new array.
6184 * In this usage, raid_disks is > 0 and it together with
6185 * level, size, not_persistent,layout,chunksize determine the
6186 * shape of the array.
6187 * This will always create an array with a type-0.90.0 superblock.
6188 * The newer usage is when assembling an array.
6189 * In this case raid_disks will be 0, and the major_version field is
6190 * use to determine which style super-blocks are to be found on the devices.
6191 * The minor and patch _version numbers are also kept incase the
6192 * super_block handler wishes to interpret them.
6194 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6197 if (info->raid_disks == 0) {
6198 /* just setting version number for superblock loading */
6199 if (info->major_version < 0 ||
6200 info->major_version >= ARRAY_SIZE(super_types) ||
6201 super_types[info->major_version].name == NULL) {
6202 /* maybe try to auto-load a module? */
6204 "md: superblock version %d not known\n",
6205 info->major_version);
6208 mddev->major_version = info->major_version;
6209 mddev->minor_version = info->minor_version;
6210 mddev->patch_version = info->patch_version;
6211 mddev->persistent = !info->not_persistent;
6212 /* ensure mddev_put doesn't delete this now that there
6213 * is some minimal configuration.
6215 mddev->ctime = get_seconds();
6218 mddev->major_version = MD_MAJOR_VERSION;
6219 mddev->minor_version = MD_MINOR_VERSION;
6220 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6221 mddev->ctime = get_seconds();
6223 mddev->level = info->level;
6224 mddev->clevel[0] = 0;
6225 mddev->dev_sectors = 2 * (sector_t)info->size;
6226 mddev->raid_disks = info->raid_disks;
6227 /* don't set md_minor, it is determined by which /dev/md* was
6230 if (info->state & (1<<MD_SB_CLEAN))
6231 mddev->recovery_cp = MaxSector;
6233 mddev->recovery_cp = 0;
6234 mddev->persistent = ! info->not_persistent;
6235 mddev->external = 0;
6237 mddev->layout = info->layout;
6238 mddev->chunk_sectors = info->chunk_size >> 9;
6240 mddev->max_disks = MD_SB_DISKS;
6242 if (mddev->persistent)
6244 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6246 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6247 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6248 mddev->bitmap_info.offset = 0;
6250 mddev->reshape_position = MaxSector;
6253 * Generate a 128 bit UUID
6255 get_random_bytes(mddev->uuid, 16);
6257 mddev->new_level = mddev->level;
6258 mddev->new_chunk_sectors = mddev->chunk_sectors;
6259 mddev->new_layout = mddev->layout;
6260 mddev->delta_disks = 0;
6261 mddev->reshape_backwards = 0;
6266 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6268 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6270 if (mddev->external_size)
6273 mddev->array_sectors = array_sectors;
6275 EXPORT_SYMBOL(md_set_array_sectors);
6277 static int update_size(struct mddev *mddev, sector_t num_sectors)
6279 struct md_rdev *rdev;
6281 int fit = (num_sectors == 0);
6283 if (mddev->pers->resize == NULL)
6285 /* The "num_sectors" is the number of sectors of each device that
6286 * is used. This can only make sense for arrays with redundancy.
6287 * linear and raid0 always use whatever space is available. We can only
6288 * consider changing this number if no resync or reconstruction is
6289 * happening, and if the new size is acceptable. It must fit before the
6290 * sb_start or, if that is <data_offset, it must fit before the size
6291 * of each device. If num_sectors is zero, we find the largest size
6294 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6300 rdev_for_each(rdev, mddev) {
6301 sector_t avail = rdev->sectors;
6303 if (fit && (num_sectors == 0 || num_sectors > avail))
6304 num_sectors = avail;
6305 if (avail < num_sectors)
6308 rv = mddev->pers->resize(mddev, num_sectors);
6310 revalidate_disk(mddev->gendisk);
6314 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6317 struct md_rdev *rdev;
6318 /* change the number of raid disks */
6319 if (mddev->pers->check_reshape == NULL)
6323 if (raid_disks <= 0 ||
6324 (mddev->max_disks && raid_disks >= mddev->max_disks))
6326 if (mddev->sync_thread ||
6327 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6328 mddev->reshape_position != MaxSector)
6331 rdev_for_each(rdev, mddev) {
6332 if (mddev->raid_disks < raid_disks &&
6333 rdev->data_offset < rdev->new_data_offset)
6335 if (mddev->raid_disks > raid_disks &&
6336 rdev->data_offset > rdev->new_data_offset)
6340 mddev->delta_disks = raid_disks - mddev->raid_disks;
6341 if (mddev->delta_disks < 0)
6342 mddev->reshape_backwards = 1;
6343 else if (mddev->delta_disks > 0)
6344 mddev->reshape_backwards = 0;
6346 rv = mddev->pers->check_reshape(mddev);
6348 mddev->delta_disks = 0;
6349 mddev->reshape_backwards = 0;
6355 * update_array_info is used to change the configuration of an
6357 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6358 * fields in the info are checked against the array.
6359 * Any differences that cannot be handled will cause an error.
6360 * Normally, only one change can be managed at a time.
6362 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6368 /* calculate expected state,ignoring low bits */
6369 if (mddev->bitmap && mddev->bitmap_info.offset)
6370 state |= (1 << MD_SB_BITMAP_PRESENT);
6372 if (mddev->major_version != info->major_version ||
6373 mddev->minor_version != info->minor_version ||
6374 /* mddev->patch_version != info->patch_version || */
6375 mddev->ctime != info->ctime ||
6376 mddev->level != info->level ||
6377 /* mddev->layout != info->layout || */
6378 mddev->persistent != !info->not_persistent ||
6379 mddev->chunk_sectors != info->chunk_size >> 9 ||
6380 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6381 ((state^info->state) & 0xfffffe00)
6384 /* Check there is only one change */
6385 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6387 if (mddev->raid_disks != info->raid_disks)
6389 if (mddev->layout != info->layout)
6391 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6398 if (mddev->layout != info->layout) {
6400 * we don't need to do anything at the md level, the
6401 * personality will take care of it all.
6403 if (mddev->pers->check_reshape == NULL)
6406 mddev->new_layout = info->layout;
6407 rv = mddev->pers->check_reshape(mddev);
6409 mddev->new_layout = mddev->layout;
6413 if (mddev_is_clustered(mddev))
6414 md_cluster_ops->metadata_update_start(mddev);
6415 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6416 rv = update_size(mddev, (sector_t)info->size * 2);
6418 if (mddev->raid_disks != info->raid_disks)
6419 rv = update_raid_disks(mddev, info->raid_disks);
6421 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6422 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6426 if (mddev->recovery || mddev->sync_thread) {
6430 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6431 struct bitmap *bitmap;
6432 /* add the bitmap */
6433 if (mddev->bitmap) {
6437 if (mddev->bitmap_info.default_offset == 0) {
6441 mddev->bitmap_info.offset =
6442 mddev->bitmap_info.default_offset;
6443 mddev->bitmap_info.space =
6444 mddev->bitmap_info.default_space;
6445 mddev->pers->quiesce(mddev, 1);
6446 bitmap = bitmap_create(mddev, -1);
6447 if (!IS_ERR(bitmap)) {
6448 mddev->bitmap = bitmap;
6449 rv = bitmap_load(mddev);
6451 rv = PTR_ERR(bitmap);
6453 bitmap_destroy(mddev);
6454 mddev->pers->quiesce(mddev, 0);
6456 /* remove the bitmap */
6457 if (!mddev->bitmap) {
6461 if (mddev->bitmap->storage.file) {
6465 mddev->pers->quiesce(mddev, 1);
6466 bitmap_destroy(mddev);
6467 mddev->pers->quiesce(mddev, 0);
6468 mddev->bitmap_info.offset = 0;
6471 md_update_sb(mddev, 1);
6472 if (mddev_is_clustered(mddev))
6473 md_cluster_ops->metadata_update_finish(mddev);
6476 if (mddev_is_clustered(mddev))
6477 md_cluster_ops->metadata_update_cancel(mddev);
6481 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6483 struct md_rdev *rdev;
6486 if (mddev->pers == NULL)
6490 rdev = find_rdev_rcu(mddev, dev);
6494 md_error(mddev, rdev);
6495 if (!test_bit(Faulty, &rdev->flags))
6503 * We have a problem here : there is no easy way to give a CHS
6504 * virtual geometry. We currently pretend that we have a 2 heads
6505 * 4 sectors (with a BIG number of cylinders...). This drives
6506 * dosfs just mad... ;-)
6508 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6510 struct mddev *mddev = bdev->bd_disk->private_data;
6514 geo->cylinders = mddev->array_sectors / 8;
6518 static inline bool md_ioctl_valid(unsigned int cmd)
6523 case GET_ARRAY_INFO:
6524 case GET_BITMAP_FILE:
6527 case HOT_REMOVE_DISK:
6530 case RESTART_ARRAY_RW:
6532 case SET_ARRAY_INFO:
6533 case SET_BITMAP_FILE:
6534 case SET_DISK_FAULTY:
6537 case CLUSTERED_DISK_NACK:
6544 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6545 unsigned int cmd, unsigned long arg)
6548 void __user *argp = (void __user *)arg;
6549 struct mddev *mddev = NULL;
6552 if (!md_ioctl_valid(cmd))
6557 case GET_ARRAY_INFO:
6561 if (!capable(CAP_SYS_ADMIN))
6566 * Commands dealing with the RAID driver but not any
6571 err = get_version(argp);
6577 autostart_arrays(arg);
6584 * Commands creating/starting a new array:
6587 mddev = bdev->bd_disk->private_data;
6594 /* Some actions do not requires the mutex */
6596 case GET_ARRAY_INFO:
6597 if (!mddev->raid_disks && !mddev->external)
6600 err = get_array_info(mddev, argp);
6604 if (!mddev->raid_disks && !mddev->external)
6607 err = get_disk_info(mddev, argp);
6610 case SET_DISK_FAULTY:
6611 err = set_disk_faulty(mddev, new_decode_dev(arg));
6614 case GET_BITMAP_FILE:
6615 err = get_bitmap_file(mddev, argp);
6620 if (cmd == ADD_NEW_DISK)
6621 /* need to ensure md_delayed_delete() has completed */
6622 flush_workqueue(md_misc_wq);
6624 if (cmd == HOT_REMOVE_DISK)
6625 /* need to ensure recovery thread has run */
6626 wait_event_interruptible_timeout(mddev->sb_wait,
6627 !test_bit(MD_RECOVERY_NEEDED,
6629 msecs_to_jiffies(5000));
6630 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6631 /* Need to flush page cache, and ensure no-one else opens
6634 mutex_lock(&mddev->open_mutex);
6635 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6636 mutex_unlock(&mddev->open_mutex);
6640 set_bit(MD_STILL_CLOSED, &mddev->flags);
6641 mutex_unlock(&mddev->open_mutex);
6642 sync_blockdev(bdev);
6644 err = mddev_lock(mddev);
6647 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6652 if (cmd == SET_ARRAY_INFO) {
6653 mdu_array_info_t info;
6655 memset(&info, 0, sizeof(info));
6656 else if (copy_from_user(&info, argp, sizeof(info))) {
6661 err = update_array_info(mddev, &info);
6663 printk(KERN_WARNING "md: couldn't update"
6664 " array info. %d\n", err);
6669 if (!list_empty(&mddev->disks)) {
6671 "md: array %s already has disks!\n",
6676 if (mddev->raid_disks) {
6678 "md: array %s already initialised!\n",
6683 err = set_array_info(mddev, &info);
6685 printk(KERN_WARNING "md: couldn't set"
6686 " array info. %d\n", err);
6693 * Commands querying/configuring an existing array:
6695 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6696 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6697 if ((!mddev->raid_disks && !mddev->external)
6698 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6699 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6700 && cmd != GET_BITMAP_FILE) {
6706 * Commands even a read-only array can execute:
6709 case RESTART_ARRAY_RW:
6710 err = restart_array(mddev);
6714 err = do_md_stop(mddev, 0, bdev);
6718 err = md_set_readonly(mddev, bdev);
6721 case HOT_REMOVE_DISK:
6722 err = hot_remove_disk(mddev, new_decode_dev(arg));
6726 /* We can support ADD_NEW_DISK on read-only arrays
6727 * on if we are re-adding a preexisting device.
6728 * So require mddev->pers and MD_DISK_SYNC.
6731 mdu_disk_info_t info;
6732 if (copy_from_user(&info, argp, sizeof(info)))
6734 else if (!(info.state & (1<<MD_DISK_SYNC)))
6735 /* Need to clear read-only for this */
6738 err = add_new_disk(mddev, &info);
6744 if (get_user(ro, (int __user *)(arg))) {
6750 /* if the bdev is going readonly the value of mddev->ro
6751 * does not matter, no writes are coming
6756 /* are we are already prepared for writes? */
6760 /* transitioning to readauto need only happen for
6761 * arrays that call md_write_start
6764 err = restart_array(mddev);
6767 set_disk_ro(mddev->gendisk, 0);
6774 * The remaining ioctls are changing the state of the
6775 * superblock, so we do not allow them on read-only arrays.
6777 if (mddev->ro && mddev->pers) {
6778 if (mddev->ro == 2) {
6780 sysfs_notify_dirent_safe(mddev->sysfs_state);
6781 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6782 /* mddev_unlock will wake thread */
6783 /* If a device failed while we were read-only, we
6784 * need to make sure the metadata is updated now.
6786 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6787 mddev_unlock(mddev);
6788 wait_event(mddev->sb_wait,
6789 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6790 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6791 mddev_lock_nointr(mddev);
6802 mdu_disk_info_t info;
6803 if (copy_from_user(&info, argp, sizeof(info)))
6806 err = add_new_disk(mddev, &info);
6810 case CLUSTERED_DISK_NACK:
6811 if (mddev_is_clustered(mddev))
6812 md_cluster_ops->new_disk_ack(mddev, false);
6818 err = hot_add_disk(mddev, new_decode_dev(arg));
6822 err = do_md_run(mddev);
6825 case SET_BITMAP_FILE:
6826 err = set_bitmap_file(mddev, (int)arg);
6835 if (mddev->hold_active == UNTIL_IOCTL &&
6837 mddev->hold_active = 0;
6838 mddev_unlock(mddev);
6842 #ifdef CONFIG_COMPAT
6843 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6844 unsigned int cmd, unsigned long arg)
6847 case HOT_REMOVE_DISK:
6849 case SET_DISK_FAULTY:
6850 case SET_BITMAP_FILE:
6851 /* These take in integer arg, do not convert */
6854 arg = (unsigned long)compat_ptr(arg);
6858 return md_ioctl(bdev, mode, cmd, arg);
6860 #endif /* CONFIG_COMPAT */
6862 static int md_open(struct block_device *bdev, fmode_t mode)
6865 * Succeed if we can lock the mddev, which confirms that
6866 * it isn't being stopped right now.
6868 struct mddev *mddev = mddev_find(bdev->bd_dev);
6874 if (mddev->gendisk != bdev->bd_disk) {
6875 /* we are racing with mddev_put which is discarding this
6879 /* Wait until bdev->bd_disk is definitely gone */
6880 flush_workqueue(md_misc_wq);
6881 /* Then retry the open from the top */
6882 return -ERESTARTSYS;
6884 BUG_ON(mddev != bdev->bd_disk->private_data);
6886 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6890 atomic_inc(&mddev->openers);
6891 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6892 mutex_unlock(&mddev->open_mutex);
6894 check_disk_change(bdev);
6899 static void md_release(struct gendisk *disk, fmode_t mode)
6901 struct mddev *mddev = disk->private_data;
6904 atomic_dec(&mddev->openers);
6908 static int md_media_changed(struct gendisk *disk)
6910 struct mddev *mddev = disk->private_data;
6912 return mddev->changed;
6915 static int md_revalidate(struct gendisk *disk)
6917 struct mddev *mddev = disk->private_data;
6922 static const struct block_device_operations md_fops =
6924 .owner = THIS_MODULE,
6926 .release = md_release,
6928 #ifdef CONFIG_COMPAT
6929 .compat_ioctl = md_compat_ioctl,
6931 .getgeo = md_getgeo,
6932 .media_changed = md_media_changed,
6933 .revalidate_disk= md_revalidate,
6936 static int md_thread(void *arg)
6938 struct md_thread *thread = arg;
6941 * md_thread is a 'system-thread', it's priority should be very
6942 * high. We avoid resource deadlocks individually in each
6943 * raid personality. (RAID5 does preallocation) We also use RR and
6944 * the very same RT priority as kswapd, thus we will never get
6945 * into a priority inversion deadlock.
6947 * we definitely have to have equal or higher priority than
6948 * bdflush, otherwise bdflush will deadlock if there are too
6949 * many dirty RAID5 blocks.
6952 allow_signal(SIGKILL);
6953 while (!kthread_should_stop()) {
6955 /* We need to wait INTERRUPTIBLE so that
6956 * we don't add to the load-average.
6957 * That means we need to be sure no signals are
6960 if (signal_pending(current))
6961 flush_signals(current);
6963 wait_event_interruptible_timeout
6965 test_bit(THREAD_WAKEUP, &thread->flags)
6966 || kthread_should_stop(),
6969 clear_bit(THREAD_WAKEUP, &thread->flags);
6970 if (!kthread_should_stop())
6971 thread->run(thread);
6977 void md_wakeup_thread(struct md_thread *thread)
6980 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6981 set_bit(THREAD_WAKEUP, &thread->flags);
6982 wake_up(&thread->wqueue);
6985 EXPORT_SYMBOL(md_wakeup_thread);
6987 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6988 struct mddev *mddev, const char *name)
6990 struct md_thread *thread;
6992 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6996 init_waitqueue_head(&thread->wqueue);
6999 thread->mddev = mddev;
7000 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7001 thread->tsk = kthread_run(md_thread, thread,
7003 mdname(thread->mddev),
7005 if (IS_ERR(thread->tsk)) {
7011 EXPORT_SYMBOL(md_register_thread);
7013 void md_unregister_thread(struct md_thread **threadp)
7015 struct md_thread *thread = *threadp;
7018 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7019 /* Locking ensures that mddev_unlock does not wake_up a
7020 * non-existent thread
7022 spin_lock(&pers_lock);
7024 spin_unlock(&pers_lock);
7026 kthread_stop(thread->tsk);
7029 EXPORT_SYMBOL(md_unregister_thread);
7031 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7033 if (!rdev || test_bit(Faulty, &rdev->flags))
7036 if (!mddev->pers || !mddev->pers->error_handler)
7038 mddev->pers->error_handler(mddev,rdev);
7039 if (mddev->degraded)
7040 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7041 sysfs_notify_dirent_safe(rdev->sysfs_state);
7042 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7043 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7044 md_wakeup_thread(mddev->thread);
7045 if (mddev->event_work.func)
7046 queue_work(md_misc_wq, &mddev->event_work);
7047 md_new_event_inintr(mddev);
7049 EXPORT_SYMBOL(md_error);
7051 /* seq_file implementation /proc/mdstat */
7053 static void status_unused(struct seq_file *seq)
7056 struct md_rdev *rdev;
7058 seq_printf(seq, "unused devices: ");
7060 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7061 char b[BDEVNAME_SIZE];
7063 seq_printf(seq, "%s ",
7064 bdevname(rdev->bdev,b));
7067 seq_printf(seq, "<none>");
7069 seq_printf(seq, "\n");
7072 static void status_resync(struct seq_file *seq, struct mddev *mddev)
7074 sector_t max_sectors, resync, res;
7075 unsigned long dt, db;
7078 unsigned int per_milli;
7080 if (mddev->curr_resync <= 3)
7083 resync = mddev->curr_resync
7084 - atomic_read(&mddev->recovery_active);
7086 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7087 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7088 max_sectors = mddev->resync_max_sectors;
7090 max_sectors = mddev->dev_sectors;
7092 WARN_ON(max_sectors == 0);
7093 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7094 * in a sector_t, and (max_sectors>>scale) will fit in a
7095 * u32, as those are the requirements for sector_div.
7096 * Thus 'scale' must be at least 10
7099 if (sizeof(sector_t) > sizeof(unsigned long)) {
7100 while ( max_sectors/2 > (1ULL<<(scale+32)))
7103 res = (resync>>scale)*1000;
7104 sector_div(res, (u32)((max_sectors>>scale)+1));
7108 int i, x = per_milli/50, y = 20-x;
7109 seq_printf(seq, "[");
7110 for (i = 0; i < x; i++)
7111 seq_printf(seq, "=");
7112 seq_printf(seq, ">");
7113 for (i = 0; i < y; i++)
7114 seq_printf(seq, ".");
7115 seq_printf(seq, "] ");
7117 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7118 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7120 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7122 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7123 "resync" : "recovery"))),
7124 per_milli/10, per_milli % 10,
7125 (unsigned long long) resync/2,
7126 (unsigned long long) max_sectors/2);
7129 * dt: time from mark until now
7130 * db: blocks written from mark until now
7131 * rt: remaining time
7133 * rt is a sector_t, so could be 32bit or 64bit.
7134 * So we divide before multiply in case it is 32bit and close
7136 * We scale the divisor (db) by 32 to avoid losing precision
7137 * near the end of resync when the number of remaining sectors
7139 * We then divide rt by 32 after multiplying by db to compensate.
7140 * The '+1' avoids division by zero if db is very small.
7142 dt = ((jiffies - mddev->resync_mark) / HZ);
7144 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7145 - mddev->resync_mark_cnt;
7147 rt = max_sectors - resync; /* number of remaining sectors */
7148 sector_div(rt, db/32+1);
7152 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7153 ((unsigned long)rt % 60)/6);
7155 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7158 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7160 struct list_head *tmp;
7162 struct mddev *mddev;
7170 spin_lock(&all_mddevs_lock);
7171 list_for_each(tmp,&all_mddevs)
7173 mddev = list_entry(tmp, struct mddev, all_mddevs);
7175 spin_unlock(&all_mddevs_lock);
7178 spin_unlock(&all_mddevs_lock);
7180 return (void*)2;/* tail */
7184 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7186 struct list_head *tmp;
7187 struct mddev *next_mddev, *mddev = v;
7193 spin_lock(&all_mddevs_lock);
7195 tmp = all_mddevs.next;
7197 tmp = mddev->all_mddevs.next;
7198 if (tmp != &all_mddevs)
7199 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7201 next_mddev = (void*)2;
7204 spin_unlock(&all_mddevs_lock);
7212 static void md_seq_stop(struct seq_file *seq, void *v)
7214 struct mddev *mddev = v;
7216 if (mddev && v != (void*)1 && v != (void*)2)
7220 static int md_seq_show(struct seq_file *seq, void *v)
7222 struct mddev *mddev = v;
7224 struct md_rdev *rdev;
7226 if (v == (void*)1) {
7227 struct md_personality *pers;
7228 seq_printf(seq, "Personalities : ");
7229 spin_lock(&pers_lock);
7230 list_for_each_entry(pers, &pers_list, list)
7231 seq_printf(seq, "[%s] ", pers->name);
7233 spin_unlock(&pers_lock);
7234 seq_printf(seq, "\n");
7235 seq->poll_event = atomic_read(&md_event_count);
7238 if (v == (void*)2) {
7243 spin_lock(&mddev->lock);
7244 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7245 seq_printf(seq, "%s : %sactive", mdname(mddev),
7246 mddev->pers ? "" : "in");
7249 seq_printf(seq, " (read-only)");
7251 seq_printf(seq, " (auto-read-only)");
7252 seq_printf(seq, " %s", mddev->pers->name);
7257 rdev_for_each_rcu(rdev, mddev) {
7258 char b[BDEVNAME_SIZE];
7259 seq_printf(seq, " %s[%d]",
7260 bdevname(rdev->bdev,b), rdev->desc_nr);
7261 if (test_bit(WriteMostly, &rdev->flags))
7262 seq_printf(seq, "(W)");
7263 if (test_bit(Faulty, &rdev->flags)) {
7264 seq_printf(seq, "(F)");
7267 if (rdev->raid_disk < 0)
7268 seq_printf(seq, "(S)"); /* spare */
7269 if (test_bit(Replacement, &rdev->flags))
7270 seq_printf(seq, "(R)");
7271 sectors += rdev->sectors;
7275 if (!list_empty(&mddev->disks)) {
7277 seq_printf(seq, "\n %llu blocks",
7278 (unsigned long long)
7279 mddev->array_sectors / 2);
7281 seq_printf(seq, "\n %llu blocks",
7282 (unsigned long long)sectors / 2);
7284 if (mddev->persistent) {
7285 if (mddev->major_version != 0 ||
7286 mddev->minor_version != 90) {
7287 seq_printf(seq," super %d.%d",
7288 mddev->major_version,
7289 mddev->minor_version);
7291 } else if (mddev->external)
7292 seq_printf(seq, " super external:%s",
7293 mddev->metadata_type);
7295 seq_printf(seq, " super non-persistent");
7298 mddev->pers->status(seq, mddev);
7299 seq_printf(seq, "\n ");
7300 if (mddev->pers->sync_request) {
7301 if (mddev->curr_resync > 2) {
7302 status_resync(seq, mddev);
7303 seq_printf(seq, "\n ");
7304 } else if (mddev->curr_resync >= 1)
7305 seq_printf(seq, "\tresync=DELAYED\n ");
7306 else if (mddev->recovery_cp < MaxSector)
7307 seq_printf(seq, "\tresync=PENDING\n ");
7310 seq_printf(seq, "\n ");
7312 bitmap_status(seq, mddev->bitmap);
7314 seq_printf(seq, "\n");
7316 spin_unlock(&mddev->lock);
7321 static const struct seq_operations md_seq_ops = {
7322 .start = md_seq_start,
7323 .next = md_seq_next,
7324 .stop = md_seq_stop,
7325 .show = md_seq_show,
7328 static int md_seq_open(struct inode *inode, struct file *file)
7330 struct seq_file *seq;
7333 error = seq_open(file, &md_seq_ops);
7337 seq = file->private_data;
7338 seq->poll_event = atomic_read(&md_event_count);
7342 static int md_unloading;
7343 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7345 struct seq_file *seq = filp->private_data;
7349 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7350 poll_wait(filp, &md_event_waiters, wait);
7352 /* always allow read */
7353 mask = POLLIN | POLLRDNORM;
7355 if (seq->poll_event != atomic_read(&md_event_count))
7356 mask |= POLLERR | POLLPRI;
7360 static const struct file_operations md_seq_fops = {
7361 .owner = THIS_MODULE,
7362 .open = md_seq_open,
7364 .llseek = seq_lseek,
7365 .release = seq_release_private,
7366 .poll = mdstat_poll,
7369 int register_md_personality(struct md_personality *p)
7371 printk(KERN_INFO "md: %s personality registered for level %d\n",
7373 spin_lock(&pers_lock);
7374 list_add_tail(&p->list, &pers_list);
7375 spin_unlock(&pers_lock);
7378 EXPORT_SYMBOL(register_md_personality);
7380 int unregister_md_personality(struct md_personality *p)
7382 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7383 spin_lock(&pers_lock);
7384 list_del_init(&p->list);
7385 spin_unlock(&pers_lock);
7388 EXPORT_SYMBOL(unregister_md_personality);
7390 int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7392 if (md_cluster_ops != NULL)
7394 spin_lock(&pers_lock);
7395 md_cluster_ops = ops;
7396 md_cluster_mod = module;
7397 spin_unlock(&pers_lock);
7400 EXPORT_SYMBOL(register_md_cluster_operations);
7402 int unregister_md_cluster_operations(void)
7404 spin_lock(&pers_lock);
7405 md_cluster_ops = NULL;
7406 spin_unlock(&pers_lock);
7409 EXPORT_SYMBOL(unregister_md_cluster_operations);
7411 int md_setup_cluster(struct mddev *mddev, int nodes)
7415 err = request_module("md-cluster");
7417 pr_err("md-cluster module not found.\n");
7421 spin_lock(&pers_lock);
7422 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7423 spin_unlock(&pers_lock);
7426 spin_unlock(&pers_lock);
7428 return md_cluster_ops->join(mddev, nodes);
7431 void md_cluster_stop(struct mddev *mddev)
7433 if (!md_cluster_ops)
7435 md_cluster_ops->leave(mddev);
7436 module_put(md_cluster_mod);
7439 static int is_mddev_idle(struct mddev *mddev, int init)
7441 struct md_rdev *rdev;
7447 rdev_for_each_rcu(rdev, mddev) {
7448 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7449 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7450 (int)part_stat_read(&disk->part0, sectors[1]) -
7451 atomic_read(&disk->sync_io);
7452 /* sync IO will cause sync_io to increase before the disk_stats
7453 * as sync_io is counted when a request starts, and
7454 * disk_stats is counted when it completes.
7455 * So resync activity will cause curr_events to be smaller than
7456 * when there was no such activity.
7457 * non-sync IO will cause disk_stat to increase without
7458 * increasing sync_io so curr_events will (eventually)
7459 * be larger than it was before. Once it becomes
7460 * substantially larger, the test below will cause
7461 * the array to appear non-idle, and resync will slow
7463 * If there is a lot of outstanding resync activity when
7464 * we set last_event to curr_events, then all that activity
7465 * completing might cause the array to appear non-idle
7466 * and resync will be slowed down even though there might
7467 * not have been non-resync activity. This will only
7468 * happen once though. 'last_events' will soon reflect
7469 * the state where there is little or no outstanding
7470 * resync requests, and further resync activity will
7471 * always make curr_events less than last_events.
7474 if (init || curr_events - rdev->last_events > 64) {
7475 rdev->last_events = curr_events;
7483 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7485 /* another "blocks" (512byte) blocks have been synced */
7486 atomic_sub(blocks, &mddev->recovery_active);
7487 wake_up(&mddev->recovery_wait);
7489 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7490 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7491 md_wakeup_thread(mddev->thread);
7492 // stop recovery, signal do_sync ....
7495 EXPORT_SYMBOL(md_done_sync);
7497 /* md_write_start(mddev, bi)
7498 * If we need to update some array metadata (e.g. 'active' flag
7499 * in superblock) before writing, schedule a superblock update
7500 * and wait for it to complete.
7502 void md_write_start(struct mddev *mddev, struct bio *bi)
7505 if (bio_data_dir(bi) != WRITE)
7508 BUG_ON(mddev->ro == 1);
7509 if (mddev->ro == 2) {
7510 /* need to switch to read/write */
7512 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7513 md_wakeup_thread(mddev->thread);
7514 md_wakeup_thread(mddev->sync_thread);
7517 atomic_inc(&mddev->writes_pending);
7518 if (mddev->safemode == 1)
7519 mddev->safemode = 0;
7520 if (mddev->in_sync) {
7521 spin_lock(&mddev->lock);
7522 if (mddev->in_sync) {
7524 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7525 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7526 md_wakeup_thread(mddev->thread);
7529 spin_unlock(&mddev->lock);
7532 sysfs_notify_dirent_safe(mddev->sysfs_state);
7533 wait_event(mddev->sb_wait,
7534 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7536 EXPORT_SYMBOL(md_write_start);
7538 void md_write_end(struct mddev *mddev)
7540 if (atomic_dec_and_test(&mddev->writes_pending)) {
7541 if (mddev->safemode == 2)
7542 md_wakeup_thread(mddev->thread);
7543 else if (mddev->safemode_delay)
7544 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7547 EXPORT_SYMBOL(md_write_end);
7549 /* md_allow_write(mddev)
7550 * Calling this ensures that the array is marked 'active' so that writes
7551 * may proceed without blocking. It is important to call this before
7552 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7553 * Must be called with mddev_lock held.
7555 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7556 * is dropped, so return -EAGAIN after notifying userspace.
7558 int md_allow_write(struct mddev *mddev)
7564 if (!mddev->pers->sync_request)
7567 spin_lock(&mddev->lock);
7568 if (mddev->in_sync) {
7570 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7571 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7572 if (mddev->safemode_delay &&
7573 mddev->safemode == 0)
7574 mddev->safemode = 1;
7575 spin_unlock(&mddev->lock);
7576 if (mddev_is_clustered(mddev))
7577 md_cluster_ops->metadata_update_start(mddev);
7578 md_update_sb(mddev, 0);
7579 if (mddev_is_clustered(mddev))
7580 md_cluster_ops->metadata_update_finish(mddev);
7581 sysfs_notify_dirent_safe(mddev->sysfs_state);
7583 spin_unlock(&mddev->lock);
7585 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7590 EXPORT_SYMBOL_GPL(md_allow_write);
7592 #define SYNC_MARKS 10
7593 #define SYNC_MARK_STEP (3*HZ)
7594 #define UPDATE_FREQUENCY (5*60*HZ)
7595 void md_do_sync(struct md_thread *thread)
7597 struct mddev *mddev = thread->mddev;
7598 struct mddev *mddev2;
7599 unsigned int currspeed = 0,
7601 sector_t max_sectors,j, io_sectors, recovery_done;
7602 unsigned long mark[SYNC_MARKS];
7603 unsigned long update_time;
7604 sector_t mark_cnt[SYNC_MARKS];
7606 struct list_head *tmp;
7607 sector_t last_check;
7609 struct md_rdev *rdev;
7610 char *desc, *action = NULL;
7611 struct blk_plug plug;
7613 /* just incase thread restarts... */
7614 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7616 if (mddev->ro) {/* never try to sync a read-only array */
7617 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7621 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7622 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7623 desc = "data-check";
7625 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7626 desc = "requested-resync";
7630 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7635 mddev->last_sync_action = action ?: desc;
7637 /* we overload curr_resync somewhat here.
7638 * 0 == not engaged in resync at all
7639 * 2 == checking that there is no conflict with another sync
7640 * 1 == like 2, but have yielded to allow conflicting resync to
7642 * other == active in resync - this many blocks
7644 * Before starting a resync we must have set curr_resync to
7645 * 2, and then checked that every "conflicting" array has curr_resync
7646 * less than ours. When we find one that is the same or higher
7647 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7648 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7649 * This will mean we have to start checking from the beginning again.
7654 mddev->curr_resync = 2;
7657 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7659 for_each_mddev(mddev2, tmp) {
7660 if (mddev2 == mddev)
7662 if (!mddev->parallel_resync
7663 && mddev2->curr_resync
7664 && match_mddev_units(mddev, mddev2)) {
7666 if (mddev < mddev2 && mddev->curr_resync == 2) {
7667 /* arbitrarily yield */
7668 mddev->curr_resync = 1;
7669 wake_up(&resync_wait);
7671 if (mddev > mddev2 && mddev->curr_resync == 1)
7672 /* no need to wait here, we can wait the next
7673 * time 'round when curr_resync == 2
7676 /* We need to wait 'interruptible' so as not to
7677 * contribute to the load average, and not to
7678 * be caught by 'softlockup'
7680 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7681 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7682 mddev2->curr_resync >= mddev->curr_resync) {
7683 printk(KERN_INFO "md: delaying %s of %s"
7684 " until %s has finished (they"
7685 " share one or more physical units)\n",
7686 desc, mdname(mddev), mdname(mddev2));
7688 if (signal_pending(current))
7689 flush_signals(current);
7691 finish_wait(&resync_wait, &wq);
7694 finish_wait(&resync_wait, &wq);
7697 } while (mddev->curr_resync < 2);
7700 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7701 /* resync follows the size requested by the personality,
7702 * which defaults to physical size, but can be virtual size
7704 max_sectors = mddev->resync_max_sectors;
7705 atomic64_set(&mddev->resync_mismatches, 0);
7706 /* we don't use the checkpoint if there's a bitmap */
7707 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7708 j = mddev->resync_min;
7709 else if (!mddev->bitmap)
7710 j = mddev->recovery_cp;
7712 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7713 max_sectors = mddev->resync_max_sectors;
7715 /* recovery follows the physical size of devices */
7716 max_sectors = mddev->dev_sectors;
7719 rdev_for_each_rcu(rdev, mddev)
7720 if (rdev->raid_disk >= 0 &&
7721 !test_bit(Faulty, &rdev->flags) &&
7722 !test_bit(In_sync, &rdev->flags) &&
7723 rdev->recovery_offset < j)
7724 j = rdev->recovery_offset;
7727 /* If there is a bitmap, we need to make sure all
7728 * writes that started before we added a spare
7729 * complete before we start doing a recovery.
7730 * Otherwise the write might complete and (via
7731 * bitmap_endwrite) set a bit in the bitmap after the
7732 * recovery has checked that bit and skipped that
7735 if (mddev->bitmap) {
7736 mddev->pers->quiesce(mddev, 1);
7737 mddev->pers->quiesce(mddev, 0);
7741 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7742 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7743 " %d KB/sec/disk.\n", speed_min(mddev));
7744 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7745 "(but not more than %d KB/sec) for %s.\n",
7746 speed_max(mddev), desc);
7748 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7751 for (m = 0; m < SYNC_MARKS; m++) {
7753 mark_cnt[m] = io_sectors;
7756 mddev->resync_mark = mark[last_mark];
7757 mddev->resync_mark_cnt = mark_cnt[last_mark];
7760 * Tune reconstruction:
7762 window = 32*(PAGE_SIZE/512);
7763 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7764 window/2, (unsigned long long)max_sectors/2);
7766 atomic_set(&mddev->recovery_active, 0);
7771 "md: resuming %s of %s from checkpoint.\n",
7772 desc, mdname(mddev));
7773 mddev->curr_resync = j;
7775 mddev->curr_resync = 3; /* no longer delayed */
7776 mddev->curr_resync_completed = j;
7777 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7778 md_new_event(mddev);
7779 update_time = jiffies;
7781 if (mddev_is_clustered(mddev))
7782 md_cluster_ops->resync_start(mddev, j, max_sectors);
7784 blk_start_plug(&plug);
7785 while (j < max_sectors) {
7790 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7791 ((mddev->curr_resync > mddev->curr_resync_completed &&
7792 (mddev->curr_resync - mddev->curr_resync_completed)
7793 > (max_sectors >> 4)) ||
7794 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7795 (j - mddev->curr_resync_completed)*2
7796 >= mddev->resync_max - mddev->curr_resync_completed
7798 /* time to update curr_resync_completed */
7799 wait_event(mddev->recovery_wait,
7800 atomic_read(&mddev->recovery_active) == 0);
7801 mddev->curr_resync_completed = j;
7802 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7803 j > mddev->recovery_cp)
7804 mddev->recovery_cp = j;
7805 update_time = jiffies;
7806 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7807 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7810 while (j >= mddev->resync_max &&
7811 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7812 /* As this condition is controlled by user-space,
7813 * we can block indefinitely, so use '_interruptible'
7814 * to avoid triggering warnings.
7816 flush_signals(current); /* just in case */
7817 wait_event_interruptible(mddev->recovery_wait,
7818 mddev->resync_max > j
7819 || test_bit(MD_RECOVERY_INTR,
7823 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7826 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7828 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7832 if (!skipped) { /* actual IO requested */
7833 io_sectors += sectors;
7834 atomic_add(sectors, &mddev->recovery_active);
7837 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7842 mddev->curr_resync = j;
7843 if (mddev_is_clustered(mddev))
7844 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7845 mddev->curr_mark_cnt = io_sectors;
7846 if (last_check == 0)
7847 /* this is the earliest that rebuild will be
7848 * visible in /proc/mdstat
7850 md_new_event(mddev);
7852 if (last_check + window > io_sectors || j == max_sectors)
7855 last_check = io_sectors;
7857 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7859 int next = (last_mark+1) % SYNC_MARKS;
7861 mddev->resync_mark = mark[next];
7862 mddev->resync_mark_cnt = mark_cnt[next];
7863 mark[next] = jiffies;
7864 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7868 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7872 * this loop exits only if either when we are slower than
7873 * the 'hard' speed limit, or the system was IO-idle for
7875 * the system might be non-idle CPU-wise, but we only care
7876 * about not overloading the IO subsystem. (things like an
7877 * e2fsck being done on the RAID array should execute fast)
7881 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7882 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7883 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7885 if (currspeed > speed_min(mddev)) {
7886 if (currspeed > speed_max(mddev)) {
7890 if (!is_mddev_idle(mddev, 0)) {
7892 * Give other IO more of a chance.
7893 * The faster the devices, the less we wait.
7895 wait_event(mddev->recovery_wait,
7896 !atomic_read(&mddev->recovery_active));
7900 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7901 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7902 ? "interrupted" : "done");
7904 * this also signals 'finished resyncing' to md_stop
7906 blk_finish_plug(&plug);
7907 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7909 /* tell personality that we are finished */
7910 mddev->pers->sync_request(mddev, max_sectors, &skipped);
7912 if (mddev_is_clustered(mddev))
7913 md_cluster_ops->resync_finish(mddev);
7915 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7916 mddev->curr_resync > 2) {
7917 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7918 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7919 if (mddev->curr_resync >= mddev->recovery_cp) {
7921 "md: checkpointing %s of %s.\n",
7922 desc, mdname(mddev));
7923 if (test_bit(MD_RECOVERY_ERROR,
7925 mddev->recovery_cp =
7926 mddev->curr_resync_completed;
7928 mddev->recovery_cp =
7932 mddev->recovery_cp = MaxSector;
7934 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7935 mddev->curr_resync = MaxSector;
7937 rdev_for_each_rcu(rdev, mddev)
7938 if (rdev->raid_disk >= 0 &&
7939 mddev->delta_disks >= 0 &&
7940 !test_bit(Faulty, &rdev->flags) &&
7941 !test_bit(In_sync, &rdev->flags) &&
7942 rdev->recovery_offset < mddev->curr_resync)
7943 rdev->recovery_offset = mddev->curr_resync;
7948 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7950 spin_lock(&mddev->lock);
7951 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7952 /* We completed so min/max setting can be forgotten if used. */
7953 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7954 mddev->resync_min = 0;
7955 mddev->resync_max = MaxSector;
7956 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7957 mddev->resync_min = mddev->curr_resync_completed;
7958 mddev->curr_resync = 0;
7959 spin_unlock(&mddev->lock);
7961 wake_up(&resync_wait);
7962 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7963 md_wakeup_thread(mddev->thread);
7966 EXPORT_SYMBOL_GPL(md_do_sync);
7968 static int remove_and_add_spares(struct mddev *mddev,
7969 struct md_rdev *this)
7971 struct md_rdev *rdev;
7975 rdev_for_each(rdev, mddev)
7976 if ((this == NULL || rdev == this) &&
7977 rdev->raid_disk >= 0 &&
7978 !test_bit(Blocked, &rdev->flags) &&
7979 (test_bit(Faulty, &rdev->flags) ||
7980 ! test_bit(In_sync, &rdev->flags)) &&
7981 atomic_read(&rdev->nr_pending)==0) {
7982 if (mddev->pers->hot_remove_disk(
7983 mddev, rdev) == 0) {
7984 sysfs_unlink_rdev(mddev, rdev);
7985 rdev->raid_disk = -1;
7989 if (removed && mddev->kobj.sd)
7990 sysfs_notify(&mddev->kobj, NULL, "degraded");
7995 rdev_for_each(rdev, mddev) {
7996 if (rdev->raid_disk >= 0 &&
7997 !test_bit(In_sync, &rdev->flags) &&
7998 !test_bit(Faulty, &rdev->flags))
8000 if (rdev->raid_disk >= 0)
8002 if (test_bit(Faulty, &rdev->flags))
8005 ! (rdev->saved_raid_disk >= 0 &&
8006 !test_bit(Bitmap_sync, &rdev->flags)))
8009 if (rdev->saved_raid_disk < 0)
8010 rdev->recovery_offset = 0;
8012 hot_add_disk(mddev, rdev) == 0) {
8013 if (sysfs_link_rdev(mddev, rdev))
8014 /* failure here is OK */;
8016 md_new_event(mddev);
8017 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8022 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8026 static void md_start_sync(struct work_struct *ws)
8028 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8030 mddev->sync_thread = md_register_thread(md_do_sync,
8033 if (!mddev->sync_thread) {
8034 printk(KERN_ERR "%s: could not start resync"
8037 /* leave the spares where they are, it shouldn't hurt */
8038 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8039 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8040 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8041 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8042 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8043 wake_up(&resync_wait);
8044 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8046 if (mddev->sysfs_action)
8047 sysfs_notify_dirent_safe(mddev->sysfs_action);
8049 md_wakeup_thread(mddev->sync_thread);
8050 sysfs_notify_dirent_safe(mddev->sysfs_action);
8051 md_new_event(mddev);
8055 * This routine is regularly called by all per-raid-array threads to
8056 * deal with generic issues like resync and super-block update.
8057 * Raid personalities that don't have a thread (linear/raid0) do not
8058 * need this as they never do any recovery or update the superblock.
8060 * It does not do any resync itself, but rather "forks" off other threads
8061 * to do that as needed.
8062 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8063 * "->recovery" and create a thread at ->sync_thread.
8064 * When the thread finishes it sets MD_RECOVERY_DONE
8065 * and wakeups up this thread which will reap the thread and finish up.
8066 * This thread also removes any faulty devices (with nr_pending == 0).
8068 * The overall approach is:
8069 * 1/ if the superblock needs updating, update it.
8070 * 2/ If a recovery thread is running, don't do anything else.
8071 * 3/ If recovery has finished, clean up, possibly marking spares active.
8072 * 4/ If there are any faulty devices, remove them.
8073 * 5/ If array is degraded, try to add spares devices
8074 * 6/ If array has spares or is not in-sync, start a resync thread.
8076 void md_check_recovery(struct mddev *mddev)
8078 if (mddev->suspended)
8082 bitmap_daemon_work(mddev);
8084 if (signal_pending(current)) {
8085 if (mddev->pers->sync_request && !mddev->external) {
8086 printk(KERN_INFO "md: %s in immediate safe mode\n",
8088 mddev->safemode = 2;
8090 flush_signals(current);
8093 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8096 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8097 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8098 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8099 (mddev->external == 0 && mddev->safemode == 1) ||
8100 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8101 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8105 if (mddev_trylock(mddev)) {
8109 struct md_rdev *rdev;
8110 if (!mddev->external && mddev->in_sync)
8111 /* 'Blocked' flag not needed as failed devices
8112 * will be recorded if array switched to read/write.
8113 * Leaving it set will prevent the device
8114 * from being removed.
8116 rdev_for_each(rdev, mddev)
8117 clear_bit(Blocked, &rdev->flags);
8118 /* On a read-only array we can:
8119 * - remove failed devices
8120 * - add already-in_sync devices if the array itself
8122 * As we only add devices that are already in-sync,
8123 * we can activate the spares immediately.
8125 remove_and_add_spares(mddev, NULL);
8126 /* There is no thread, but we need to call
8127 * ->spare_active and clear saved_raid_disk
8129 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8130 md_reap_sync_thread(mddev);
8131 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8135 if (!mddev->external) {
8137 spin_lock(&mddev->lock);
8138 if (mddev->safemode &&
8139 !atomic_read(&mddev->writes_pending) &&
8141 mddev->recovery_cp == MaxSector) {
8144 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8146 if (mddev->safemode == 1)
8147 mddev->safemode = 0;
8148 spin_unlock(&mddev->lock);
8150 sysfs_notify_dirent_safe(mddev->sysfs_state);
8153 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8154 if (mddev_is_clustered(mddev))
8155 md_cluster_ops->metadata_update_start(mddev);
8156 md_update_sb(mddev, 0);
8157 if (mddev_is_clustered(mddev))
8158 md_cluster_ops->metadata_update_finish(mddev);
8161 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8162 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8163 /* resync/recovery still happening */
8164 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8167 if (mddev->sync_thread) {
8168 md_reap_sync_thread(mddev);
8171 /* Set RUNNING before clearing NEEDED to avoid
8172 * any transients in the value of "sync_action".
8174 mddev->curr_resync_completed = 0;
8175 spin_lock(&mddev->lock);
8176 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8177 spin_unlock(&mddev->lock);
8178 /* Clear some bits that don't mean anything, but
8181 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8182 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8184 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8185 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8187 /* no recovery is running.
8188 * remove any failed drives, then
8189 * add spares if possible.
8190 * Spares are also removed and re-added, to allow
8191 * the personality to fail the re-add.
8194 if (mddev->reshape_position != MaxSector) {
8195 if (mddev->pers->check_reshape == NULL ||
8196 mddev->pers->check_reshape(mddev) != 0)
8197 /* Cannot proceed */
8199 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8200 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8201 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8202 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8203 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8204 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8205 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8206 } else if (mddev->recovery_cp < MaxSector) {
8207 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8208 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8209 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8210 /* nothing to be done ... */
8213 if (mddev->pers->sync_request) {
8215 /* We are adding a device or devices to an array
8216 * which has the bitmap stored on all devices.
8217 * So make sure all bitmap pages get written
8219 bitmap_write_all(mddev->bitmap);
8221 INIT_WORK(&mddev->del_work, md_start_sync);
8222 queue_work(md_misc_wq, &mddev->del_work);
8226 if (!mddev->sync_thread) {
8227 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8228 wake_up(&resync_wait);
8229 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8231 if (mddev->sysfs_action)
8232 sysfs_notify_dirent_safe(mddev->sysfs_action);
8235 wake_up(&mddev->sb_wait);
8236 mddev_unlock(mddev);
8239 EXPORT_SYMBOL(md_check_recovery);
8241 void md_reap_sync_thread(struct mddev *mddev)
8243 struct md_rdev *rdev;
8245 /* resync has finished, collect result */
8246 md_unregister_thread(&mddev->sync_thread);
8247 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8248 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8250 /* activate any spares */
8251 if (mddev->pers->spare_active(mddev)) {
8252 sysfs_notify(&mddev->kobj, NULL,
8254 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8257 if (mddev_is_clustered(mddev))
8258 md_cluster_ops->metadata_update_start(mddev);
8259 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8260 mddev->pers->finish_reshape)
8261 mddev->pers->finish_reshape(mddev);
8263 /* If array is no-longer degraded, then any saved_raid_disk
8264 * information must be scrapped.
8266 if (!mddev->degraded)
8267 rdev_for_each(rdev, mddev)
8268 rdev->saved_raid_disk = -1;
8270 md_update_sb(mddev, 1);
8271 if (mddev_is_clustered(mddev))
8272 md_cluster_ops->metadata_update_finish(mddev);
8273 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8274 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8275 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8276 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8277 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8278 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8279 wake_up(&resync_wait);
8280 /* flag recovery needed just to double check */
8281 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8282 sysfs_notify_dirent_safe(mddev->sysfs_action);
8283 md_new_event(mddev);
8284 if (mddev->event_work.func)
8285 queue_work(md_misc_wq, &mddev->event_work);
8287 EXPORT_SYMBOL(md_reap_sync_thread);
8289 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8291 sysfs_notify_dirent_safe(rdev->sysfs_state);
8292 wait_event_timeout(rdev->blocked_wait,
8293 !test_bit(Blocked, &rdev->flags) &&
8294 !test_bit(BlockedBadBlocks, &rdev->flags),
8295 msecs_to_jiffies(5000));
8296 rdev_dec_pending(rdev, mddev);
8298 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8300 void md_finish_reshape(struct mddev *mddev)
8302 /* called be personality module when reshape completes. */
8303 struct md_rdev *rdev;
8305 rdev_for_each(rdev, mddev) {
8306 if (rdev->data_offset > rdev->new_data_offset)
8307 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8309 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8310 rdev->data_offset = rdev->new_data_offset;
8313 EXPORT_SYMBOL(md_finish_reshape);
8315 /* Bad block management.
8316 * We can record which blocks on each device are 'bad' and so just
8317 * fail those blocks, or that stripe, rather than the whole device.
8318 * Entries in the bad-block table are 64bits wide. This comprises:
8319 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8320 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8321 * A 'shift' can be set so that larger blocks are tracked and
8322 * consequently larger devices can be covered.
8323 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8325 * Locking of the bad-block table uses a seqlock so md_is_badblock
8326 * might need to retry if it is very unlucky.
8327 * We will sometimes want to check for bad blocks in a bi_end_io function,
8328 * so we use the write_seqlock_irq variant.
8330 * When looking for a bad block we specify a range and want to
8331 * know if any block in the range is bad. So we binary-search
8332 * to the last range that starts at-or-before the given endpoint,
8333 * (or "before the sector after the target range")
8334 * then see if it ends after the given start.
8336 * 0 if there are no known bad blocks in the range
8337 * 1 if there are known bad block which are all acknowledged
8338 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8339 * plus the start/length of the first bad section we overlap.
8341 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8342 sector_t *first_bad, int *bad_sectors)
8348 sector_t target = s + sectors;
8351 if (bb->shift > 0) {
8352 /* round the start down, and the end up */
8354 target += (1<<bb->shift) - 1;
8355 target >>= bb->shift;
8356 sectors = target - s;
8358 /* 'target' is now the first block after the bad range */
8361 seq = read_seqbegin(&bb->lock);
8366 /* Binary search between lo and hi for 'target'
8367 * i.e. for the last range that starts before 'target'
8369 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8370 * are known not to be the last range before target.
8371 * VARIANT: hi-lo is the number of possible
8372 * ranges, and decreases until it reaches 1
8374 while (hi - lo > 1) {
8375 int mid = (lo + hi) / 2;
8376 sector_t a = BB_OFFSET(p[mid]);
8378 /* This could still be the one, earlier ranges
8382 /* This and later ranges are definitely out. */
8385 /* 'lo' might be the last that started before target, but 'hi' isn't */
8387 /* need to check all range that end after 's' to see if
8388 * any are unacknowledged.
8391 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8392 if (BB_OFFSET(p[lo]) < target) {
8393 /* starts before the end, and finishes after
8394 * the start, so they must overlap
8396 if (rv != -1 && BB_ACK(p[lo]))
8400 *first_bad = BB_OFFSET(p[lo]);
8401 *bad_sectors = BB_LEN(p[lo]);
8407 if (read_seqretry(&bb->lock, seq))
8412 EXPORT_SYMBOL_GPL(md_is_badblock);
8415 * Add a range of bad blocks to the table.
8416 * This might extend the table, or might contract it
8417 * if two adjacent ranges can be merged.
8418 * We binary-search to find the 'insertion' point, then
8419 * decide how best to handle it.
8421 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8427 unsigned long flags;
8430 /* badblocks are disabled */
8434 /* round the start down, and the end up */
8435 sector_t next = s + sectors;
8437 next += (1<<bb->shift) - 1;
8442 write_seqlock_irqsave(&bb->lock, flags);
8447 /* Find the last range that starts at-or-before 's' */
8448 while (hi - lo > 1) {
8449 int mid = (lo + hi) / 2;
8450 sector_t a = BB_OFFSET(p[mid]);
8456 if (hi > lo && BB_OFFSET(p[lo]) > s)
8460 /* we found a range that might merge with the start
8463 sector_t a = BB_OFFSET(p[lo]);
8464 sector_t e = a + BB_LEN(p[lo]);
8465 int ack = BB_ACK(p[lo]);
8467 /* Yes, we can merge with a previous range */
8468 if (s == a && s + sectors >= e)
8469 /* new range covers old */
8472 ack = ack && acknowledged;
8474 if (e < s + sectors)
8476 if (e - a <= BB_MAX_LEN) {
8477 p[lo] = BB_MAKE(a, e-a, ack);
8480 /* does not all fit in one range,
8481 * make p[lo] maximal
8483 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8484 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8490 if (sectors && hi < bb->count) {
8491 /* 'hi' points to the first range that starts after 's'.
8492 * Maybe we can merge with the start of that range */
8493 sector_t a = BB_OFFSET(p[hi]);
8494 sector_t e = a + BB_LEN(p[hi]);
8495 int ack = BB_ACK(p[hi]);
8496 if (a <= s + sectors) {
8497 /* merging is possible */
8498 if (e <= s + sectors) {
8503 ack = ack && acknowledged;
8506 if (e - a <= BB_MAX_LEN) {
8507 p[hi] = BB_MAKE(a, e-a, ack);
8510 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8518 if (sectors == 0 && hi < bb->count) {
8519 /* we might be able to combine lo and hi */
8520 /* Note: 's' is at the end of 'lo' */
8521 sector_t a = BB_OFFSET(p[hi]);
8522 int lolen = BB_LEN(p[lo]);
8523 int hilen = BB_LEN(p[hi]);
8524 int newlen = lolen + hilen - (s - a);
8525 if (s >= a && newlen < BB_MAX_LEN) {
8526 /* yes, we can combine them */
8527 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8528 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8529 memmove(p + hi, p + hi + 1,
8530 (bb->count - hi - 1) * 8);
8535 /* didn't merge (it all).
8536 * Need to add a range just before 'hi' */
8537 if (bb->count >= MD_MAX_BADBLOCKS) {
8538 /* No room for more */
8542 int this_sectors = sectors;
8543 memmove(p + hi + 1, p + hi,
8544 (bb->count - hi) * 8);
8547 if (this_sectors > BB_MAX_LEN)
8548 this_sectors = BB_MAX_LEN;
8549 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8550 sectors -= this_sectors;
8557 bb->unacked_exist = 1;
8558 write_sequnlock_irqrestore(&bb->lock, flags);
8563 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8568 s += rdev->new_data_offset;
8570 s += rdev->data_offset;
8571 rv = md_set_badblocks(&rdev->badblocks,
8574 /* Make sure they get written out promptly */
8575 sysfs_notify_dirent_safe(rdev->sysfs_state);
8576 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8577 md_wakeup_thread(rdev->mddev->thread);
8581 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8584 * Remove a range of bad blocks from the table.
8585 * This may involve extending the table if we spilt a region,
8586 * but it must not fail. So if the table becomes full, we just
8587 * drop the remove request.
8589 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8593 sector_t target = s + sectors;
8596 if (bb->shift > 0) {
8597 /* When clearing we round the start up and the end down.
8598 * This should not matter as the shift should align with
8599 * the block size and no rounding should ever be needed.
8600 * However it is better the think a block is bad when it
8601 * isn't than to think a block is not bad when it is.
8603 s += (1<<bb->shift) - 1;
8605 target >>= bb->shift;
8606 sectors = target - s;
8609 write_seqlock_irq(&bb->lock);
8614 /* Find the last range that starts before 'target' */
8615 while (hi - lo > 1) {
8616 int mid = (lo + hi) / 2;
8617 sector_t a = BB_OFFSET(p[mid]);
8624 /* p[lo] is the last range that could overlap the
8625 * current range. Earlier ranges could also overlap,
8626 * but only this one can overlap the end of the range.
8628 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8629 /* Partial overlap, leave the tail of this range */
8630 int ack = BB_ACK(p[lo]);
8631 sector_t a = BB_OFFSET(p[lo]);
8632 sector_t end = a + BB_LEN(p[lo]);
8635 /* we need to split this range */
8636 if (bb->count >= MD_MAX_BADBLOCKS) {
8640 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8642 p[lo] = BB_MAKE(a, s-a, ack);
8645 p[lo] = BB_MAKE(target, end - target, ack);
8646 /* there is no longer an overlap */
8651 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8652 /* This range does overlap */
8653 if (BB_OFFSET(p[lo]) < s) {
8654 /* Keep the early parts of this range. */
8655 int ack = BB_ACK(p[lo]);
8656 sector_t start = BB_OFFSET(p[lo]);
8657 p[lo] = BB_MAKE(start, s - start, ack);
8658 /* now low doesn't overlap, so.. */
8663 /* 'lo' is strictly before, 'hi' is strictly after,
8664 * anything between needs to be discarded
8667 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8668 bb->count -= (hi - lo - 1);
8674 write_sequnlock_irq(&bb->lock);
8678 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8682 s += rdev->new_data_offset;
8684 s += rdev->data_offset;
8685 return md_clear_badblocks(&rdev->badblocks,
8688 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8691 * Acknowledge all bad blocks in a list.
8692 * This only succeeds if ->changed is clear. It is used by
8693 * in-kernel metadata updates
8695 void md_ack_all_badblocks(struct badblocks *bb)
8697 if (bb->page == NULL || bb->changed)
8698 /* no point even trying */
8700 write_seqlock_irq(&bb->lock);
8702 if (bb->changed == 0 && bb->unacked_exist) {
8705 for (i = 0; i < bb->count ; i++) {
8706 if (!BB_ACK(p[i])) {
8707 sector_t start = BB_OFFSET(p[i]);
8708 int len = BB_LEN(p[i]);
8709 p[i] = BB_MAKE(start, len, 1);
8712 bb->unacked_exist = 0;
8714 write_sequnlock_irq(&bb->lock);
8716 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8718 /* sysfs access to bad-blocks list.
8719 * We present two files.
8720 * 'bad-blocks' lists sector numbers and lengths of ranges that
8721 * are recorded as bad. The list is truncated to fit within
8722 * the one-page limit of sysfs.
8723 * Writing "sector length" to this file adds an acknowledged
8725 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8726 * been acknowledged. Writing to this file adds bad blocks
8727 * without acknowledging them. This is largely for testing.
8731 badblocks_show(struct badblocks *bb, char *page, int unack)
8742 seq = read_seqbegin(&bb->lock);
8747 while (len < PAGE_SIZE && i < bb->count) {
8748 sector_t s = BB_OFFSET(p[i]);
8749 unsigned int length = BB_LEN(p[i]);
8750 int ack = BB_ACK(p[i]);
8756 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8757 (unsigned long long)s << bb->shift,
8758 length << bb->shift);
8760 if (unack && len == 0)
8761 bb->unacked_exist = 0;
8763 if (read_seqretry(&bb->lock, seq))
8772 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8774 unsigned long long sector;
8778 /* Allow clearing via sysfs *only* for testing/debugging.
8779 * Normally only a successful write may clear a badblock
8782 if (page[0] == '-') {
8786 #endif /* DO_DEBUG */
8788 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8790 if (newline != '\n')
8802 md_clear_badblocks(bb, sector, length);
8805 #endif /* DO_DEBUG */
8806 if (md_set_badblocks(bb, sector, length, !unack))
8812 static int md_notify_reboot(struct notifier_block *this,
8813 unsigned long code, void *x)
8815 struct list_head *tmp;
8816 struct mddev *mddev;
8819 for_each_mddev(mddev, tmp) {
8820 if (mddev_trylock(mddev)) {
8822 __md_stop_writes(mddev);
8823 if (mddev->persistent)
8824 mddev->safemode = 2;
8825 mddev_unlock(mddev);
8830 * certain more exotic SCSI devices are known to be
8831 * volatile wrt too early system reboots. While the
8832 * right place to handle this issue is the given
8833 * driver, we do want to have a safe RAID driver ...
8841 static struct notifier_block md_notifier = {
8842 .notifier_call = md_notify_reboot,
8844 .priority = INT_MAX, /* before any real devices */
8847 static void md_geninit(void)
8849 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8851 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8854 static int __init md_init(void)
8858 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8862 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8866 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8869 if ((ret = register_blkdev(0, "mdp")) < 0)
8873 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8874 md_probe, NULL, NULL);
8875 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8876 md_probe, NULL, NULL);
8878 register_reboot_notifier(&md_notifier);
8879 raid_table_header = register_sysctl_table(raid_root_table);
8885 unregister_blkdev(MD_MAJOR, "md");
8887 destroy_workqueue(md_misc_wq);
8889 destroy_workqueue(md_wq);
8894 void md_reload_sb(struct mddev *mddev)
8896 struct md_rdev *rdev, *tmp;
8898 rdev_for_each_safe(rdev, tmp, mddev) {
8899 rdev->sb_loaded = 0;
8900 ClearPageUptodate(rdev->sb_page);
8902 mddev->raid_disks = 0;
8904 rdev_for_each_safe(rdev, tmp, mddev) {
8905 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8906 /* since we don't write to faulty devices, we figure out if the
8907 * disk is faulty by comparing events
8909 if (mddev->events > sb->events)
8910 set_bit(Faulty, &rdev->flags);
8914 EXPORT_SYMBOL(md_reload_sb);
8919 * Searches all registered partitions for autorun RAID arrays
8923 static LIST_HEAD(all_detected_devices);
8924 struct detected_devices_node {
8925 struct list_head list;
8929 void md_autodetect_dev(dev_t dev)
8931 struct detected_devices_node *node_detected_dev;
8933 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8934 if (node_detected_dev) {
8935 node_detected_dev->dev = dev;
8936 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8938 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8939 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8943 static void autostart_arrays(int part)
8945 struct md_rdev *rdev;
8946 struct detected_devices_node *node_detected_dev;
8948 int i_scanned, i_passed;
8953 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8955 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8957 node_detected_dev = list_entry(all_detected_devices.next,
8958 struct detected_devices_node, list);
8959 list_del(&node_detected_dev->list);
8960 dev = node_detected_dev->dev;
8961 kfree(node_detected_dev);
8962 rdev = md_import_device(dev,0, 90);
8966 if (test_bit(Faulty, &rdev->flags))
8969 set_bit(AutoDetected, &rdev->flags);
8970 list_add(&rdev->same_set, &pending_raid_disks);
8974 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8975 i_scanned, i_passed);
8977 autorun_devices(part);
8980 #endif /* !MODULE */
8982 static __exit void md_exit(void)
8984 struct mddev *mddev;
8985 struct list_head *tmp;
8988 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8989 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8991 unregister_blkdev(MD_MAJOR,"md");
8992 unregister_blkdev(mdp_major, "mdp");
8993 unregister_reboot_notifier(&md_notifier);
8994 unregister_sysctl_table(raid_table_header);
8996 /* We cannot unload the modules while some process is
8997 * waiting for us in select() or poll() - wake them up
9000 while (waitqueue_active(&md_event_waiters)) {
9001 /* not safe to leave yet */
9002 wake_up(&md_event_waiters);
9006 remove_proc_entry("mdstat", NULL);
9008 for_each_mddev(mddev, tmp) {
9009 export_array(mddev);
9010 mddev->hold_active = 0;
9012 destroy_workqueue(md_misc_wq);
9013 destroy_workqueue(md_wq);
9016 subsys_initcall(md_init);
9017 module_exit(md_exit)
9019 static int get_ro(char *buffer, struct kernel_param *kp)
9021 return sprintf(buffer, "%d", start_readonly);
9023 static int set_ro(const char *val, struct kernel_param *kp)
9025 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9028 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9029 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9030 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9032 MODULE_LICENSE("GPL");
9033 MODULE_DESCRIPTION("MD RAID framework");
9035 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);