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>
58 static void autostart_arrays(int part);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
75 static int remove_and_add_spares(struct mddev *mddev,
76 struct md_rdev *this);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
90 * the RAID driver will use the maximum available bandwidth if the IO
91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
96 * or /sys/block/mdX/md/sync_speed_{min,max}
99 static int sysctl_speed_limit_min = 1000;
100 static int sysctl_speed_limit_max = 200000;
101 static inline int speed_min(struct mddev *mddev)
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
107 static inline int speed_max(struct mddev *mddev)
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
113 static struct ctl_table_header *raid_table_header;
115 static struct ctl_table raid_table[] = {
117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
127 .mode = S_IRUGO|S_IWUSR,
128 .proc_handler = proc_dointvec,
133 static struct ctl_table raid_dir_table[] = {
137 .mode = S_IRUGO|S_IXUGO,
143 static struct ctl_table raid_root_table[] = {
148 .child = raid_dir_table,
153 static const struct block_device_operations md_fops;
155 static int start_readonly;
158 * like bio_clone, but with a local bio set
161 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
166 if (!mddev || !mddev->bio_set)
167 return bio_alloc(gfp_mask, nr_iovecs);
169 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
174 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
176 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
179 if (!mddev || !mddev->bio_set)
180 return bio_clone(bio, gfp_mask);
182 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
184 EXPORT_SYMBOL_GPL(bio_clone_mddev);
187 * We have a system wide 'event count' that is incremented
188 * on any 'interesting' event, and readers of /proc/mdstat
189 * can use 'poll' or 'select' to find out when the event
193 * start array, stop array, error, add device, remove device,
194 * start build, activate spare
196 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
197 static atomic_t md_event_count;
198 void md_new_event(struct mddev *mddev)
200 atomic_inc(&md_event_count);
201 wake_up(&md_event_waiters);
203 EXPORT_SYMBOL_GPL(md_new_event);
205 /* Alternate version that can be called from interrupts
206 * when calling sysfs_notify isn't needed.
208 static void md_new_event_inintr(struct mddev *mddev)
210 atomic_inc(&md_event_count);
211 wake_up(&md_event_waiters);
215 * Enables to iterate over all existing md arrays
216 * all_mddevs_lock protects this list.
218 static LIST_HEAD(all_mddevs);
219 static DEFINE_SPINLOCK(all_mddevs_lock);
222 * iterates through all used mddevs in the system.
223 * We take care to grab the all_mddevs_lock whenever navigating
224 * the list, and to always hold a refcount when unlocked.
225 * Any code which breaks out of this loop while own
226 * a reference to the current mddev and must mddev_put it.
228 #define for_each_mddev(_mddev,_tmp) \
230 for (({ spin_lock(&all_mddevs_lock); \
231 _tmp = all_mddevs.next; \
233 ({ if (_tmp != &all_mddevs) \
234 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
235 spin_unlock(&all_mddevs_lock); \
236 if (_mddev) mddev_put(_mddev); \
237 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
238 _tmp != &all_mddevs;}); \
239 ({ spin_lock(&all_mddevs_lock); \
240 _tmp = _tmp->next;}) \
243 /* Rather than calling directly into the personality make_request function,
244 * IO requests come here first so that we can check if the device is
245 * being suspended pending a reconfiguration.
246 * We hold a refcount over the call to ->make_request. By the time that
247 * call has finished, the bio has been linked into some internal structure
248 * and so is visible to ->quiesce(), so we don't need the refcount any more.
250 static void md_make_request(struct request_queue *q, struct bio *bio)
252 const int rw = bio_data_dir(bio);
253 struct mddev *mddev = q->queuedata;
255 unsigned int sectors;
257 if (mddev == NULL || mddev->pers == NULL
262 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
263 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
266 smp_rmb(); /* Ensure implications of 'active' are visible */
268 if (mddev->suspended) {
271 prepare_to_wait(&mddev->sb_wait, &__wait,
272 TASK_UNINTERRUPTIBLE);
273 if (!mddev->suspended)
279 finish_wait(&mddev->sb_wait, &__wait);
281 atomic_inc(&mddev->active_io);
285 * save the sectors now since our bio can
286 * go away inside make_request
288 sectors = bio_sectors(bio);
289 mddev->pers->make_request(mddev, bio);
291 cpu = part_stat_lock();
292 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
293 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
296 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
297 wake_up(&mddev->sb_wait);
300 /* mddev_suspend makes sure no new requests are submitted
301 * to the device, and that any requests that have been submitted
302 * are completely handled.
303 * Once ->stop is called and completes, the module will be completely
306 void mddev_suspend(struct mddev *mddev)
308 BUG_ON(mddev->suspended);
309 mddev->suspended = 1;
311 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
312 mddev->pers->quiesce(mddev, 1);
314 del_timer_sync(&mddev->safemode_timer);
316 EXPORT_SYMBOL_GPL(mddev_suspend);
318 void mddev_resume(struct mddev *mddev)
320 mddev->suspended = 0;
321 wake_up(&mddev->sb_wait);
322 mddev->pers->quiesce(mddev, 0);
324 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
325 md_wakeup_thread(mddev->thread);
326 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
328 EXPORT_SYMBOL_GPL(mddev_resume);
330 int mddev_congested(struct mddev *mddev, int bits)
332 return mddev->suspended;
334 EXPORT_SYMBOL(mddev_congested);
337 * Generic flush handling for md
340 static void md_end_flush(struct bio *bio, int err)
342 struct md_rdev *rdev = bio->bi_private;
343 struct mddev *mddev = rdev->mddev;
345 rdev_dec_pending(rdev, mddev);
347 if (atomic_dec_and_test(&mddev->flush_pending)) {
348 /* The pre-request flush has finished */
349 queue_work(md_wq, &mddev->flush_work);
354 static void md_submit_flush_data(struct work_struct *ws);
356 static void submit_flushes(struct work_struct *ws)
358 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
359 struct md_rdev *rdev;
361 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
362 atomic_set(&mddev->flush_pending, 1);
364 rdev_for_each_rcu(rdev, mddev)
365 if (rdev->raid_disk >= 0 &&
366 !test_bit(Faulty, &rdev->flags)) {
367 /* Take two references, one is dropped
368 * when request finishes, one after
369 * we reclaim rcu_read_lock
372 atomic_inc(&rdev->nr_pending);
373 atomic_inc(&rdev->nr_pending);
375 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
376 bi->bi_end_io = md_end_flush;
377 bi->bi_private = rdev;
378 bi->bi_bdev = rdev->bdev;
379 atomic_inc(&mddev->flush_pending);
380 submit_bio(WRITE_FLUSH, bi);
382 rdev_dec_pending(rdev, mddev);
385 if (atomic_dec_and_test(&mddev->flush_pending))
386 queue_work(md_wq, &mddev->flush_work);
389 static void md_submit_flush_data(struct work_struct *ws)
391 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
392 struct bio *bio = mddev->flush_bio;
394 if (bio->bi_iter.bi_size == 0)
395 /* an empty barrier - all done */
398 bio->bi_rw &= ~REQ_FLUSH;
399 mddev->pers->make_request(mddev, bio);
402 mddev->flush_bio = NULL;
403 wake_up(&mddev->sb_wait);
406 void md_flush_request(struct mddev *mddev, struct bio *bio)
408 spin_lock_irq(&mddev->write_lock);
409 wait_event_lock_irq(mddev->sb_wait,
412 mddev->flush_bio = bio;
413 spin_unlock_irq(&mddev->write_lock);
415 INIT_WORK(&mddev->flush_work, submit_flushes);
416 queue_work(md_wq, &mddev->flush_work);
418 EXPORT_SYMBOL(md_flush_request);
420 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
422 struct mddev *mddev = cb->data;
423 md_wakeup_thread(mddev->thread);
426 EXPORT_SYMBOL(md_unplug);
428 static inline struct mddev *mddev_get(struct mddev *mddev)
430 atomic_inc(&mddev->active);
434 static void mddev_delayed_delete(struct work_struct *ws);
436 static void mddev_put(struct mddev *mddev)
438 struct bio_set *bs = NULL;
440 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
442 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
443 mddev->ctime == 0 && !mddev->hold_active) {
444 /* Array is not configured at all, and not held active,
446 list_del_init(&mddev->all_mddevs);
448 mddev->bio_set = NULL;
449 if (mddev->gendisk) {
450 /* We did a probe so need to clean up. Call
451 * queue_work inside the spinlock so that
452 * flush_workqueue() after mddev_find will
453 * succeed in waiting for the work to be done.
455 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
456 queue_work(md_misc_wq, &mddev->del_work);
460 spin_unlock(&all_mddevs_lock);
465 void mddev_init(struct mddev *mddev)
467 mutex_init(&mddev->open_mutex);
468 mutex_init(&mddev->reconfig_mutex);
469 mutex_init(&mddev->bitmap_info.mutex);
470 INIT_LIST_HEAD(&mddev->disks);
471 INIT_LIST_HEAD(&mddev->all_mddevs);
472 init_timer(&mddev->safemode_timer);
473 atomic_set(&mddev->active, 1);
474 atomic_set(&mddev->openers, 0);
475 atomic_set(&mddev->active_io, 0);
476 spin_lock_init(&mddev->write_lock);
477 atomic_set(&mddev->flush_pending, 0);
478 init_waitqueue_head(&mddev->sb_wait);
479 init_waitqueue_head(&mddev->recovery_wait);
480 mddev->reshape_position = MaxSector;
481 mddev->reshape_backwards = 0;
482 mddev->last_sync_action = "none";
483 mddev->resync_min = 0;
484 mddev->resync_max = MaxSector;
485 mddev->level = LEVEL_NONE;
487 EXPORT_SYMBOL_GPL(mddev_init);
489 static struct mddev *mddev_find(dev_t unit)
491 struct mddev *mddev, *new = NULL;
493 if (unit && MAJOR(unit) != MD_MAJOR)
494 unit &= ~((1<<MdpMinorShift)-1);
497 spin_lock(&all_mddevs_lock);
500 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
501 if (mddev->unit == unit) {
503 spin_unlock(&all_mddevs_lock);
509 list_add(&new->all_mddevs, &all_mddevs);
510 spin_unlock(&all_mddevs_lock);
511 new->hold_active = UNTIL_IOCTL;
515 /* find an unused unit number */
516 static int next_minor = 512;
517 int start = next_minor;
521 dev = MKDEV(MD_MAJOR, next_minor);
523 if (next_minor > MINORMASK)
525 if (next_minor == start) {
526 /* Oh dear, all in use. */
527 spin_unlock(&all_mddevs_lock);
533 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
534 if (mddev->unit == dev) {
540 new->md_minor = MINOR(dev);
541 new->hold_active = UNTIL_STOP;
542 list_add(&new->all_mddevs, &all_mddevs);
543 spin_unlock(&all_mddevs_lock);
546 spin_unlock(&all_mddevs_lock);
548 new = kzalloc(sizeof(*new), GFP_KERNEL);
553 if (MAJOR(unit) == MD_MAJOR)
554 new->md_minor = MINOR(unit);
556 new->md_minor = MINOR(unit) >> MdpMinorShift;
563 static inline int __must_check mddev_lock(struct mddev *mddev)
565 return mutex_lock_interruptible(&mddev->reconfig_mutex);
568 /* Sometimes we need to take the lock in a situation where
569 * failure due to interrupts is not acceptable.
571 static inline void mddev_lock_nointr(struct mddev *mddev)
573 mutex_lock(&mddev->reconfig_mutex);
576 static inline int mddev_is_locked(struct mddev *mddev)
578 return mutex_is_locked(&mddev->reconfig_mutex);
581 static inline int mddev_trylock(struct mddev *mddev)
583 return mutex_trylock(&mddev->reconfig_mutex);
586 static struct attribute_group md_redundancy_group;
588 static void mddev_unlock(struct mddev *mddev)
590 if (mddev->to_remove) {
591 /* These cannot be removed under reconfig_mutex as
592 * an access to the files will try to take reconfig_mutex
593 * while holding the file unremovable, which leads to
595 * So hold set sysfs_active while the remove in happeing,
596 * and anything else which might set ->to_remove or my
597 * otherwise change the sysfs namespace will fail with
598 * -EBUSY if sysfs_active is still set.
599 * We set sysfs_active under reconfig_mutex and elsewhere
600 * test it under the same mutex to ensure its correct value
603 struct attribute_group *to_remove = mddev->to_remove;
604 mddev->to_remove = NULL;
605 mddev->sysfs_active = 1;
606 mutex_unlock(&mddev->reconfig_mutex);
608 if (mddev->kobj.sd) {
609 if (to_remove != &md_redundancy_group)
610 sysfs_remove_group(&mddev->kobj, to_remove);
611 if (mddev->pers == NULL ||
612 mddev->pers->sync_request == NULL) {
613 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
614 if (mddev->sysfs_action)
615 sysfs_put(mddev->sysfs_action);
616 mddev->sysfs_action = NULL;
619 mddev->sysfs_active = 0;
621 mutex_unlock(&mddev->reconfig_mutex);
623 /* As we've dropped the mutex we need a spinlock to
624 * make sure the thread doesn't disappear
626 spin_lock(&pers_lock);
627 md_wakeup_thread(mddev->thread);
628 spin_unlock(&pers_lock);
631 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
633 struct md_rdev *rdev;
635 rdev_for_each_rcu(rdev, mddev)
636 if (rdev->desc_nr == nr)
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)
688 rdev->sb_page = alloc_page(GFP_KERNEL);
689 if (!rdev->sb_page) {
690 printk(KERN_ALERT "md: out of memory.\n");
697 void md_rdev_clear(struct md_rdev *rdev)
700 put_page(rdev->sb_page);
702 rdev->sb_page = NULL;
707 put_page(rdev->bb_page);
708 rdev->bb_page = NULL;
710 kfree(rdev->badblocks.page);
711 rdev->badblocks.page = NULL;
713 EXPORT_SYMBOL_GPL(md_rdev_clear);
715 static void super_written(struct bio *bio, int error)
717 struct md_rdev *rdev = bio->bi_private;
718 struct mddev *mddev = rdev->mddev;
720 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
721 printk("md: super_written gets error=%d, uptodate=%d\n",
722 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
723 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
724 md_error(mddev, rdev);
727 if (atomic_dec_and_test(&mddev->pending_writes))
728 wake_up(&mddev->sb_wait);
732 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
733 sector_t sector, int size, struct page *page)
735 /* write first size bytes of page to sector of rdev
736 * Increment mddev->pending_writes before returning
737 * and decrement it on completion, waking up sb_wait
738 * if zero is reached.
739 * If an error occurred, call md_error
741 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
743 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
744 bio->bi_iter.bi_sector = sector;
745 bio_add_page(bio, page, size, 0);
746 bio->bi_private = rdev;
747 bio->bi_end_io = super_written;
749 atomic_inc(&mddev->pending_writes);
750 submit_bio(WRITE_FLUSH_FUA, bio);
753 void md_super_wait(struct mddev *mddev)
755 /* wait for all superblock writes that were scheduled to complete */
756 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
759 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
760 struct page *page, int rw, bool metadata_op)
762 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
765 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
766 rdev->meta_bdev : rdev->bdev;
768 bio->bi_iter.bi_sector = sector + rdev->sb_start;
769 else if (rdev->mddev->reshape_position != MaxSector &&
770 (rdev->mddev->reshape_backwards ==
771 (sector >= rdev->mddev->reshape_position)))
772 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
774 bio->bi_iter.bi_sector = sector + rdev->data_offset;
775 bio_add_page(bio, page, size, 0);
776 submit_bio_wait(rw, bio);
778 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
782 EXPORT_SYMBOL_GPL(sync_page_io);
784 static int read_disk_sb(struct md_rdev *rdev, int size)
786 char b[BDEVNAME_SIZE];
787 if (!rdev->sb_page) {
794 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
800 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
801 bdevname(rdev->bdev,b));
805 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
807 return sb1->set_uuid0 == sb2->set_uuid0 &&
808 sb1->set_uuid1 == sb2->set_uuid1 &&
809 sb1->set_uuid2 == sb2->set_uuid2 &&
810 sb1->set_uuid3 == sb2->set_uuid3;
813 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
816 mdp_super_t *tmp1, *tmp2;
818 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
819 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
821 if (!tmp1 || !tmp2) {
823 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
831 * nr_disks is not constant
836 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
843 static u32 md_csum_fold(u32 csum)
845 csum = (csum & 0xffff) + (csum >> 16);
846 return (csum & 0xffff) + (csum >> 16);
849 static unsigned int calc_sb_csum(mdp_super_t *sb)
852 u32 *sb32 = (u32*)sb;
854 unsigned int disk_csum, csum;
856 disk_csum = sb->sb_csum;
859 for (i = 0; i < MD_SB_BYTES/4 ; i++)
861 csum = (newcsum & 0xffffffff) + (newcsum>>32);
864 /* This used to use csum_partial, which was wrong for several
865 * reasons including that different results are returned on
866 * different architectures. It isn't critical that we get exactly
867 * the same return value as before (we always csum_fold before
868 * testing, and that removes any differences). However as we
869 * know that csum_partial always returned a 16bit value on
870 * alphas, do a fold to maximise conformity to previous behaviour.
872 sb->sb_csum = md_csum_fold(disk_csum);
874 sb->sb_csum = disk_csum;
880 * Handle superblock details.
881 * We want to be able to handle multiple superblock formats
882 * so we have a common interface to them all, and an array of
883 * different handlers.
884 * We rely on user-space to write the initial superblock, and support
885 * reading and updating of superblocks.
886 * Interface methods are:
887 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
888 * loads and validates a superblock on dev.
889 * if refdev != NULL, compare superblocks on both devices
891 * 0 - dev has a superblock that is compatible with refdev
892 * 1 - dev has a superblock that is compatible and newer than refdev
893 * so dev should be used as the refdev in future
894 * -EINVAL superblock incompatible or invalid
895 * -othererror e.g. -EIO
897 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
898 * Verify that dev is acceptable into mddev.
899 * The first time, mddev->raid_disks will be 0, and data from
900 * dev should be merged in. Subsequent calls check that dev
901 * is new enough. Return 0 or -EINVAL
903 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
904 * Update the superblock for rdev with data in mddev
905 * This does not write to disc.
911 struct module *owner;
912 int (*load_super)(struct md_rdev *rdev,
913 struct md_rdev *refdev,
915 int (*validate_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 void (*sync_super)(struct mddev *mddev,
918 struct md_rdev *rdev);
919 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
920 sector_t num_sectors);
921 int (*allow_new_offset)(struct md_rdev *rdev,
922 unsigned long long new_offset);
926 * Check that the given mddev has no bitmap.
928 * This function is called from the run method of all personalities that do not
929 * support bitmaps. It prints an error message and returns non-zero if mddev
930 * has a bitmap. Otherwise, it returns 0.
933 int md_check_no_bitmap(struct mddev *mddev)
935 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
937 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
938 mdname(mddev), mddev->pers->name);
941 EXPORT_SYMBOL(md_check_no_bitmap);
944 * load_super for 0.90.0
946 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
948 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
953 * Calculate the position of the superblock (512byte sectors),
954 * it's at the end of the disk.
956 * It also happens to be a multiple of 4Kb.
958 rdev->sb_start = calc_dev_sboffset(rdev);
960 ret = read_disk_sb(rdev, MD_SB_BYTES);
965 bdevname(rdev->bdev, b);
966 sb = page_address(rdev->sb_page);
968 if (sb->md_magic != MD_SB_MAGIC) {
969 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
974 if (sb->major_version != 0 ||
975 sb->minor_version < 90 ||
976 sb->minor_version > 91) {
977 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
978 sb->major_version, sb->minor_version,
983 if (sb->raid_disks <= 0)
986 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
987 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
992 rdev->preferred_minor = sb->md_minor;
993 rdev->data_offset = 0;
994 rdev->new_data_offset = 0;
995 rdev->sb_size = MD_SB_BYTES;
996 rdev->badblocks.shift = -1;
998 if (sb->level == LEVEL_MULTIPATH)
1001 rdev->desc_nr = sb->this_disk.number;
1007 mdp_super_t *refsb = page_address(refdev->sb_page);
1008 if (!uuid_equal(refsb, sb)) {
1009 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1010 b, bdevname(refdev->bdev,b2));
1013 if (!sb_equal(refsb, sb)) {
1014 printk(KERN_WARNING "md: %s has same UUID"
1015 " but different superblock to %s\n",
1016 b, bdevname(refdev->bdev, b2));
1020 ev2 = md_event(refsb);
1026 rdev->sectors = rdev->sb_start;
1027 /* Limit to 4TB as metadata cannot record more than that.
1028 * (not needed for Linear and RAID0 as metadata doesn't
1031 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1032 rdev->sectors = (2ULL << 32) - 2;
1034 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1035 /* "this cannot possibly happen" ... */
1043 * validate_super for 0.90.0
1045 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1048 mdp_super_t *sb = page_address(rdev->sb_page);
1049 __u64 ev1 = md_event(sb);
1051 rdev->raid_disk = -1;
1052 clear_bit(Faulty, &rdev->flags);
1053 clear_bit(In_sync, &rdev->flags);
1054 clear_bit(Bitmap_sync, &rdev->flags);
1055 clear_bit(WriteMostly, &rdev->flags);
1057 if (mddev->raid_disks == 0) {
1058 mddev->major_version = 0;
1059 mddev->minor_version = sb->minor_version;
1060 mddev->patch_version = sb->patch_version;
1061 mddev->external = 0;
1062 mddev->chunk_sectors = sb->chunk_size >> 9;
1063 mddev->ctime = sb->ctime;
1064 mddev->utime = sb->utime;
1065 mddev->level = sb->level;
1066 mddev->clevel[0] = 0;
1067 mddev->layout = sb->layout;
1068 mddev->raid_disks = sb->raid_disks;
1069 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1070 mddev->events = ev1;
1071 mddev->bitmap_info.offset = 0;
1072 mddev->bitmap_info.space = 0;
1073 /* bitmap can use 60 K after the 4K superblocks */
1074 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1075 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1076 mddev->reshape_backwards = 0;
1078 if (mddev->minor_version >= 91) {
1079 mddev->reshape_position = sb->reshape_position;
1080 mddev->delta_disks = sb->delta_disks;
1081 mddev->new_level = sb->new_level;
1082 mddev->new_layout = sb->new_layout;
1083 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1084 if (mddev->delta_disks < 0)
1085 mddev->reshape_backwards = 1;
1087 mddev->reshape_position = MaxSector;
1088 mddev->delta_disks = 0;
1089 mddev->new_level = mddev->level;
1090 mddev->new_layout = mddev->layout;
1091 mddev->new_chunk_sectors = mddev->chunk_sectors;
1094 if (sb->state & (1<<MD_SB_CLEAN))
1095 mddev->recovery_cp = MaxSector;
1097 if (sb->events_hi == sb->cp_events_hi &&
1098 sb->events_lo == sb->cp_events_lo) {
1099 mddev->recovery_cp = sb->recovery_cp;
1101 mddev->recovery_cp = 0;
1104 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1105 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1106 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1107 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1109 mddev->max_disks = MD_SB_DISKS;
1111 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1112 mddev->bitmap_info.file == NULL) {
1113 mddev->bitmap_info.offset =
1114 mddev->bitmap_info.default_offset;
1115 mddev->bitmap_info.space =
1116 mddev->bitmap_info.default_space;
1119 } else if (mddev->pers == NULL) {
1120 /* Insist on good event counter while assembling, except
1121 * for spares (which don't need an event count) */
1123 if (sb->disks[rdev->desc_nr].state & (
1124 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1125 if (ev1 < mddev->events)
1127 } else if (mddev->bitmap) {
1128 /* if adding to array with a bitmap, then we can accept an
1129 * older device ... but not too old.
1131 if (ev1 < mddev->bitmap->events_cleared)
1133 if (ev1 < mddev->events)
1134 set_bit(Bitmap_sync, &rdev->flags);
1136 if (ev1 < mddev->events)
1137 /* just a hot-add of a new device, leave raid_disk at -1 */
1141 if (mddev->level != LEVEL_MULTIPATH) {
1142 desc = sb->disks + rdev->desc_nr;
1144 if (desc->state & (1<<MD_DISK_FAULTY))
1145 set_bit(Faulty, &rdev->flags);
1146 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1147 desc->raid_disk < mddev->raid_disks */) {
1148 set_bit(In_sync, &rdev->flags);
1149 rdev->raid_disk = desc->raid_disk;
1150 rdev->saved_raid_disk = desc->raid_disk;
1151 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1152 /* active but not in sync implies recovery up to
1153 * reshape position. We don't know exactly where
1154 * that is, so set to zero for now */
1155 if (mddev->minor_version >= 91) {
1156 rdev->recovery_offset = 0;
1157 rdev->raid_disk = desc->raid_disk;
1160 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1161 set_bit(WriteMostly, &rdev->flags);
1162 } else /* MULTIPATH are always insync */
1163 set_bit(In_sync, &rdev->flags);
1168 * sync_super for 0.90.0
1170 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1173 struct md_rdev *rdev2;
1174 int next_spare = mddev->raid_disks;
1176 /* make rdev->sb match mddev data..
1179 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1180 * 3/ any empty disks < next_spare become removed
1182 * disks[0] gets initialised to REMOVED because
1183 * we cannot be sure from other fields if it has
1184 * been initialised or not.
1187 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1189 rdev->sb_size = MD_SB_BYTES;
1191 sb = page_address(rdev->sb_page);
1193 memset(sb, 0, sizeof(*sb));
1195 sb->md_magic = MD_SB_MAGIC;
1196 sb->major_version = mddev->major_version;
1197 sb->patch_version = mddev->patch_version;
1198 sb->gvalid_words = 0; /* ignored */
1199 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1200 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1201 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1202 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1204 sb->ctime = mddev->ctime;
1205 sb->level = mddev->level;
1206 sb->size = mddev->dev_sectors / 2;
1207 sb->raid_disks = mddev->raid_disks;
1208 sb->md_minor = mddev->md_minor;
1209 sb->not_persistent = 0;
1210 sb->utime = mddev->utime;
1212 sb->events_hi = (mddev->events>>32);
1213 sb->events_lo = (u32)mddev->events;
1215 if (mddev->reshape_position == MaxSector)
1216 sb->minor_version = 90;
1218 sb->minor_version = 91;
1219 sb->reshape_position = mddev->reshape_position;
1220 sb->new_level = mddev->new_level;
1221 sb->delta_disks = mddev->delta_disks;
1222 sb->new_layout = mddev->new_layout;
1223 sb->new_chunk = mddev->new_chunk_sectors << 9;
1225 mddev->minor_version = sb->minor_version;
1228 sb->recovery_cp = mddev->recovery_cp;
1229 sb->cp_events_hi = (mddev->events>>32);
1230 sb->cp_events_lo = (u32)mddev->events;
1231 if (mddev->recovery_cp == MaxSector)
1232 sb->state = (1<< MD_SB_CLEAN);
1234 sb->recovery_cp = 0;
1236 sb->layout = mddev->layout;
1237 sb->chunk_size = mddev->chunk_sectors << 9;
1239 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1240 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1242 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1243 rdev_for_each(rdev2, mddev) {
1246 int is_active = test_bit(In_sync, &rdev2->flags);
1248 if (rdev2->raid_disk >= 0 &&
1249 sb->minor_version >= 91)
1250 /* we have nowhere to store the recovery_offset,
1251 * but if it is not below the reshape_position,
1252 * we can piggy-back on that.
1255 if (rdev2->raid_disk < 0 ||
1256 test_bit(Faulty, &rdev2->flags))
1259 desc_nr = rdev2->raid_disk;
1261 desc_nr = next_spare++;
1262 rdev2->desc_nr = desc_nr;
1263 d = &sb->disks[rdev2->desc_nr];
1265 d->number = rdev2->desc_nr;
1266 d->major = MAJOR(rdev2->bdev->bd_dev);
1267 d->minor = MINOR(rdev2->bdev->bd_dev);
1269 d->raid_disk = rdev2->raid_disk;
1271 d->raid_disk = rdev2->desc_nr; /* compatibility */
1272 if (test_bit(Faulty, &rdev2->flags))
1273 d->state = (1<<MD_DISK_FAULTY);
1274 else if (is_active) {
1275 d->state = (1<<MD_DISK_ACTIVE);
1276 if (test_bit(In_sync, &rdev2->flags))
1277 d->state |= (1<<MD_DISK_SYNC);
1285 if (test_bit(WriteMostly, &rdev2->flags))
1286 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1288 /* now set the "removed" and "faulty" bits on any missing devices */
1289 for (i=0 ; i < mddev->raid_disks ; i++) {
1290 mdp_disk_t *d = &sb->disks[i];
1291 if (d->state == 0 && d->number == 0) {
1294 d->state = (1<<MD_DISK_REMOVED);
1295 d->state |= (1<<MD_DISK_FAULTY);
1299 sb->nr_disks = nr_disks;
1300 sb->active_disks = active;
1301 sb->working_disks = working;
1302 sb->failed_disks = failed;
1303 sb->spare_disks = spare;
1305 sb->this_disk = sb->disks[rdev->desc_nr];
1306 sb->sb_csum = calc_sb_csum(sb);
1310 * rdev_size_change for 0.90.0
1312 static unsigned long long
1313 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1315 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1316 return 0; /* component must fit device */
1317 if (rdev->mddev->bitmap_info.offset)
1318 return 0; /* can't move bitmap */
1319 rdev->sb_start = calc_dev_sboffset(rdev);
1320 if (!num_sectors || num_sectors > rdev->sb_start)
1321 num_sectors = rdev->sb_start;
1322 /* Limit to 4TB as metadata cannot record more than that.
1323 * 4TB == 2^32 KB, or 2*2^32 sectors.
1325 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1326 num_sectors = (2ULL << 32) - 2;
1327 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1329 md_super_wait(rdev->mddev);
1334 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1336 /* non-zero offset changes not possible with v0.90 */
1337 return new_offset == 0;
1341 * version 1 superblock
1344 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1348 unsigned long long newcsum;
1349 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1350 __le32 *isuper = (__le32*)sb;
1352 disk_csum = sb->sb_csum;
1355 for (; size >= 4; size -= 4)
1356 newcsum += le32_to_cpu(*isuper++);
1359 newcsum += le16_to_cpu(*(__le16*) isuper);
1361 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1362 sb->sb_csum = disk_csum;
1363 return cpu_to_le32(csum);
1366 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1368 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1370 struct mdp_superblock_1 *sb;
1374 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1378 * Calculate the position of the superblock in 512byte sectors.
1379 * It is always aligned to a 4K boundary and
1380 * depeding on minor_version, it can be:
1381 * 0: At least 8K, but less than 12K, from end of device
1382 * 1: At start of device
1383 * 2: 4K from start of device.
1385 switch(minor_version) {
1387 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1389 sb_start &= ~(sector_t)(4*2-1);
1400 rdev->sb_start = sb_start;
1402 /* superblock is rarely larger than 1K, but it can be larger,
1403 * and it is safe to read 4k, so we do that
1405 ret = read_disk_sb(rdev, 4096);
1406 if (ret) return ret;
1408 sb = page_address(rdev->sb_page);
1410 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1411 sb->major_version != cpu_to_le32(1) ||
1412 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1413 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1414 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1418 printk("md: invalid superblock checksum on %s\n",
1419 bdevname(rdev->bdev,b));
1422 if (le64_to_cpu(sb->data_size) < 10) {
1423 printk("md: data_size too small on %s\n",
1424 bdevname(rdev->bdev,b));
1429 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1430 /* Some padding is non-zero, might be a new feature */
1433 rdev->preferred_minor = 0xffff;
1434 rdev->data_offset = le64_to_cpu(sb->data_offset);
1435 rdev->new_data_offset = rdev->data_offset;
1436 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1437 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1438 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1439 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1441 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1442 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1443 if (rdev->sb_size & bmask)
1444 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1447 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1450 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1453 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1456 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1458 if (!rdev->bb_page) {
1459 rdev->bb_page = alloc_page(GFP_KERNEL);
1463 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1464 rdev->badblocks.count == 0) {
1465 /* need to load the bad block list.
1466 * Currently we limit it to one page.
1472 int sectors = le16_to_cpu(sb->bblog_size);
1473 if (sectors > (PAGE_SIZE / 512))
1475 offset = le32_to_cpu(sb->bblog_offset);
1478 bb_sector = (long long)offset;
1479 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1480 rdev->bb_page, READ, true))
1482 bbp = (u64 *)page_address(rdev->bb_page);
1483 rdev->badblocks.shift = sb->bblog_shift;
1484 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1485 u64 bb = le64_to_cpu(*bbp);
1486 int count = bb & (0x3ff);
1487 u64 sector = bb >> 10;
1488 sector <<= sb->bblog_shift;
1489 count <<= sb->bblog_shift;
1492 if (md_set_badblocks(&rdev->badblocks,
1493 sector, count, 1) == 0)
1496 } else if (sb->bblog_offset != 0)
1497 rdev->badblocks.shift = 0;
1503 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1505 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1506 sb->level != refsb->level ||
1507 sb->layout != refsb->layout ||
1508 sb->chunksize != refsb->chunksize) {
1509 printk(KERN_WARNING "md: %s has strangely different"
1510 " superblock to %s\n",
1511 bdevname(rdev->bdev,b),
1512 bdevname(refdev->bdev,b2));
1515 ev1 = le64_to_cpu(sb->events);
1516 ev2 = le64_to_cpu(refsb->events);
1523 if (minor_version) {
1524 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1525 sectors -= rdev->data_offset;
1527 sectors = rdev->sb_start;
1528 if (sectors < le64_to_cpu(sb->data_size))
1530 rdev->sectors = le64_to_cpu(sb->data_size);
1534 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1536 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1537 __u64 ev1 = le64_to_cpu(sb->events);
1539 rdev->raid_disk = -1;
1540 clear_bit(Faulty, &rdev->flags);
1541 clear_bit(In_sync, &rdev->flags);
1542 clear_bit(Bitmap_sync, &rdev->flags);
1543 clear_bit(WriteMostly, &rdev->flags);
1545 if (mddev->raid_disks == 0) {
1546 mddev->major_version = 1;
1547 mddev->patch_version = 0;
1548 mddev->external = 0;
1549 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1550 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1551 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1552 mddev->level = le32_to_cpu(sb->level);
1553 mddev->clevel[0] = 0;
1554 mddev->layout = le32_to_cpu(sb->layout);
1555 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1556 mddev->dev_sectors = le64_to_cpu(sb->size);
1557 mddev->events = ev1;
1558 mddev->bitmap_info.offset = 0;
1559 mddev->bitmap_info.space = 0;
1560 /* Default location for bitmap is 1K after superblock
1561 * using 3K - total of 4K
1563 mddev->bitmap_info.default_offset = 1024 >> 9;
1564 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1565 mddev->reshape_backwards = 0;
1567 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1568 memcpy(mddev->uuid, sb->set_uuid, 16);
1570 mddev->max_disks = (4096-256)/2;
1572 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1573 mddev->bitmap_info.file == NULL) {
1574 mddev->bitmap_info.offset =
1575 (__s32)le32_to_cpu(sb->bitmap_offset);
1576 /* Metadata doesn't record how much space is available.
1577 * For 1.0, we assume we can use up to the superblock
1578 * if before, else to 4K beyond superblock.
1579 * For others, assume no change is possible.
1581 if (mddev->minor_version > 0)
1582 mddev->bitmap_info.space = 0;
1583 else if (mddev->bitmap_info.offset > 0)
1584 mddev->bitmap_info.space =
1585 8 - mddev->bitmap_info.offset;
1587 mddev->bitmap_info.space =
1588 -mddev->bitmap_info.offset;
1591 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1592 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1593 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1594 mddev->new_level = le32_to_cpu(sb->new_level);
1595 mddev->new_layout = le32_to_cpu(sb->new_layout);
1596 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1597 if (mddev->delta_disks < 0 ||
1598 (mddev->delta_disks == 0 &&
1599 (le32_to_cpu(sb->feature_map)
1600 & MD_FEATURE_RESHAPE_BACKWARDS)))
1601 mddev->reshape_backwards = 1;
1603 mddev->reshape_position = MaxSector;
1604 mddev->delta_disks = 0;
1605 mddev->new_level = mddev->level;
1606 mddev->new_layout = mddev->layout;
1607 mddev->new_chunk_sectors = mddev->chunk_sectors;
1610 } else if (mddev->pers == NULL) {
1611 /* Insist of good event counter while assembling, except for
1612 * spares (which don't need an event count) */
1614 if (rdev->desc_nr >= 0 &&
1615 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1616 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1617 if (ev1 < mddev->events)
1619 } else if (mddev->bitmap) {
1620 /* If adding to array with a bitmap, then we can accept an
1621 * older device, but not too old.
1623 if (ev1 < mddev->bitmap->events_cleared)
1625 if (ev1 < mddev->events)
1626 set_bit(Bitmap_sync, &rdev->flags);
1628 if (ev1 < mddev->events)
1629 /* just a hot-add of a new device, leave raid_disk at -1 */
1632 if (mddev->level != LEVEL_MULTIPATH) {
1634 if (rdev->desc_nr < 0 ||
1635 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1639 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1641 case 0xffff: /* spare */
1643 case 0xfffe: /* faulty */
1644 set_bit(Faulty, &rdev->flags);
1647 rdev->saved_raid_disk = role;
1648 if ((le32_to_cpu(sb->feature_map) &
1649 MD_FEATURE_RECOVERY_OFFSET)) {
1650 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1651 if (!(le32_to_cpu(sb->feature_map) &
1652 MD_FEATURE_RECOVERY_BITMAP))
1653 rdev->saved_raid_disk = -1;
1655 set_bit(In_sync, &rdev->flags);
1656 rdev->raid_disk = role;
1659 if (sb->devflags & WriteMostly1)
1660 set_bit(WriteMostly, &rdev->flags);
1661 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1662 set_bit(Replacement, &rdev->flags);
1663 } else /* MULTIPATH are always insync */
1664 set_bit(In_sync, &rdev->flags);
1669 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1671 struct mdp_superblock_1 *sb;
1672 struct md_rdev *rdev2;
1674 /* make rdev->sb match mddev and rdev data. */
1676 sb = page_address(rdev->sb_page);
1678 sb->feature_map = 0;
1680 sb->recovery_offset = cpu_to_le64(0);
1681 memset(sb->pad3, 0, sizeof(sb->pad3));
1683 sb->utime = cpu_to_le64((__u64)mddev->utime);
1684 sb->events = cpu_to_le64(mddev->events);
1686 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1688 sb->resync_offset = cpu_to_le64(0);
1690 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1692 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1693 sb->size = cpu_to_le64(mddev->dev_sectors);
1694 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1695 sb->level = cpu_to_le32(mddev->level);
1696 sb->layout = cpu_to_le32(mddev->layout);
1698 if (test_bit(WriteMostly, &rdev->flags))
1699 sb->devflags |= WriteMostly1;
1701 sb->devflags &= ~WriteMostly1;
1702 sb->data_offset = cpu_to_le64(rdev->data_offset);
1703 sb->data_size = cpu_to_le64(rdev->sectors);
1705 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1706 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1707 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1710 if (rdev->raid_disk >= 0 &&
1711 !test_bit(In_sync, &rdev->flags)) {
1713 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1714 sb->recovery_offset =
1715 cpu_to_le64(rdev->recovery_offset);
1716 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1718 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1720 if (test_bit(Replacement, &rdev->flags))
1722 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1724 if (mddev->reshape_position != MaxSector) {
1725 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1726 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1727 sb->new_layout = cpu_to_le32(mddev->new_layout);
1728 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1729 sb->new_level = cpu_to_le32(mddev->new_level);
1730 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1731 if (mddev->delta_disks == 0 &&
1732 mddev->reshape_backwards)
1734 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1735 if (rdev->new_data_offset != rdev->data_offset) {
1737 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1738 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1739 - rdev->data_offset));
1743 if (rdev->badblocks.count == 0)
1744 /* Nothing to do for bad blocks*/ ;
1745 else if (sb->bblog_offset == 0)
1746 /* Cannot record bad blocks on this device */
1747 md_error(mddev, rdev);
1749 struct badblocks *bb = &rdev->badblocks;
1750 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1752 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1757 seq = read_seqbegin(&bb->lock);
1759 memset(bbp, 0xff, PAGE_SIZE);
1761 for (i = 0 ; i < bb->count ; i++) {
1762 u64 internal_bb = p[i];
1763 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1764 | BB_LEN(internal_bb));
1765 bbp[i] = cpu_to_le64(store_bb);
1768 if (read_seqretry(&bb->lock, seq))
1771 bb->sector = (rdev->sb_start +
1772 (int)le32_to_cpu(sb->bblog_offset));
1773 bb->size = le16_to_cpu(sb->bblog_size);
1778 rdev_for_each(rdev2, mddev)
1779 if (rdev2->desc_nr+1 > max_dev)
1780 max_dev = rdev2->desc_nr+1;
1782 if (max_dev > le32_to_cpu(sb->max_dev)) {
1784 sb->max_dev = cpu_to_le32(max_dev);
1785 rdev->sb_size = max_dev * 2 + 256;
1786 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1787 if (rdev->sb_size & bmask)
1788 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1790 max_dev = le32_to_cpu(sb->max_dev);
1792 for (i=0; i<max_dev;i++)
1793 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1795 rdev_for_each(rdev2, mddev) {
1797 if (test_bit(Faulty, &rdev2->flags))
1798 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1799 else if (test_bit(In_sync, &rdev2->flags))
1800 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1801 else if (rdev2->raid_disk >= 0)
1802 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1804 sb->dev_roles[i] = cpu_to_le16(0xffff);
1807 sb->sb_csum = calc_sb_1_csum(sb);
1810 static unsigned long long
1811 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1813 struct mdp_superblock_1 *sb;
1814 sector_t max_sectors;
1815 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1816 return 0; /* component must fit device */
1817 if (rdev->data_offset != rdev->new_data_offset)
1818 return 0; /* too confusing */
1819 if (rdev->sb_start < rdev->data_offset) {
1820 /* minor versions 1 and 2; superblock before data */
1821 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1822 max_sectors -= rdev->data_offset;
1823 if (!num_sectors || num_sectors > max_sectors)
1824 num_sectors = max_sectors;
1825 } else if (rdev->mddev->bitmap_info.offset) {
1826 /* minor version 0 with bitmap we can't move */
1829 /* minor version 0; superblock after data */
1831 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1832 sb_start &= ~(sector_t)(4*2 - 1);
1833 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1834 if (!num_sectors || num_sectors > max_sectors)
1835 num_sectors = max_sectors;
1836 rdev->sb_start = sb_start;
1838 sb = page_address(rdev->sb_page);
1839 sb->data_size = cpu_to_le64(num_sectors);
1840 sb->super_offset = rdev->sb_start;
1841 sb->sb_csum = calc_sb_1_csum(sb);
1842 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1844 md_super_wait(rdev->mddev);
1850 super_1_allow_new_offset(struct md_rdev *rdev,
1851 unsigned long long new_offset)
1853 /* All necessary checks on new >= old have been done */
1854 struct bitmap *bitmap;
1855 if (new_offset >= rdev->data_offset)
1858 /* with 1.0 metadata, there is no metadata to tread on
1859 * so we can always move back */
1860 if (rdev->mddev->minor_version == 0)
1863 /* otherwise we must be sure not to step on
1864 * any metadata, so stay:
1865 * 36K beyond start of superblock
1866 * beyond end of badblocks
1867 * beyond write-intent bitmap
1869 if (rdev->sb_start + (32+4)*2 > new_offset)
1871 bitmap = rdev->mddev->bitmap;
1872 if (bitmap && !rdev->mddev->bitmap_info.file &&
1873 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1874 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1876 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1882 static struct super_type super_types[] = {
1885 .owner = THIS_MODULE,
1886 .load_super = super_90_load,
1887 .validate_super = super_90_validate,
1888 .sync_super = super_90_sync,
1889 .rdev_size_change = super_90_rdev_size_change,
1890 .allow_new_offset = super_90_allow_new_offset,
1894 .owner = THIS_MODULE,
1895 .load_super = super_1_load,
1896 .validate_super = super_1_validate,
1897 .sync_super = super_1_sync,
1898 .rdev_size_change = super_1_rdev_size_change,
1899 .allow_new_offset = super_1_allow_new_offset,
1903 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1905 if (mddev->sync_super) {
1906 mddev->sync_super(mddev, rdev);
1910 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1912 super_types[mddev->major_version].sync_super(mddev, rdev);
1915 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1917 struct md_rdev *rdev, *rdev2;
1920 rdev_for_each_rcu(rdev, mddev1)
1921 rdev_for_each_rcu(rdev2, mddev2)
1922 if (rdev->bdev->bd_contains ==
1923 rdev2->bdev->bd_contains) {
1931 static LIST_HEAD(pending_raid_disks);
1934 * Try to register data integrity profile for an mddev
1936 * This is called when an array is started and after a disk has been kicked
1937 * from the array. It only succeeds if all working and active component devices
1938 * are integrity capable with matching profiles.
1940 int md_integrity_register(struct mddev *mddev)
1942 struct md_rdev *rdev, *reference = NULL;
1944 if (list_empty(&mddev->disks))
1945 return 0; /* nothing to do */
1946 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1947 return 0; /* shouldn't register, or already is */
1948 rdev_for_each(rdev, mddev) {
1949 /* skip spares and non-functional disks */
1950 if (test_bit(Faulty, &rdev->flags))
1952 if (rdev->raid_disk < 0)
1955 /* Use the first rdev as the reference */
1959 /* does this rdev's profile match the reference profile? */
1960 if (blk_integrity_compare(reference->bdev->bd_disk,
1961 rdev->bdev->bd_disk) < 0)
1964 if (!reference || !bdev_get_integrity(reference->bdev))
1967 * All component devices are integrity capable and have matching
1968 * profiles, register the common profile for the md device.
1970 if (blk_integrity_register(mddev->gendisk,
1971 bdev_get_integrity(reference->bdev)) != 0) {
1972 printk(KERN_ERR "md: failed to register integrity for %s\n",
1976 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1977 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1978 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1984 EXPORT_SYMBOL(md_integrity_register);
1986 /* Disable data integrity if non-capable/non-matching disk is being added */
1987 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1989 struct blk_integrity *bi_rdev;
1990 struct blk_integrity *bi_mddev;
1992 if (!mddev->gendisk)
1995 bi_rdev = bdev_get_integrity(rdev->bdev);
1996 bi_mddev = blk_get_integrity(mddev->gendisk);
1998 if (!bi_mddev) /* nothing to do */
2000 if (rdev->raid_disk < 0) /* skip spares */
2002 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2003 rdev->bdev->bd_disk) >= 0)
2005 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2006 blk_integrity_unregister(mddev->gendisk);
2008 EXPORT_SYMBOL(md_integrity_add_rdev);
2010 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2012 char b[BDEVNAME_SIZE];
2022 /* prevent duplicates */
2023 if (find_rdev(mddev, rdev->bdev->bd_dev))
2026 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2027 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2028 rdev->sectors < mddev->dev_sectors)) {
2030 /* Cannot change size, so fail
2031 * If mddev->level <= 0, then we don't care
2032 * about aligning sizes (e.g. linear)
2034 if (mddev->level > 0)
2037 mddev->dev_sectors = rdev->sectors;
2040 /* Verify rdev->desc_nr is unique.
2041 * If it is -1, assign a free number, else
2042 * check number is not in use
2045 if (rdev->desc_nr < 0) {
2048 choice = mddev->raid_disks;
2049 while (find_rdev_nr_rcu(mddev, choice))
2051 rdev->desc_nr = choice;
2053 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2059 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2060 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2061 mdname(mddev), mddev->max_disks);
2064 bdevname(rdev->bdev,b);
2065 while ( (s=strchr(b, '/')) != NULL)
2068 rdev->mddev = mddev;
2069 printk(KERN_INFO "md: bind<%s>\n", b);
2071 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2074 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2075 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2076 /* failure here is OK */;
2077 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2079 list_add_rcu(&rdev->same_set, &mddev->disks);
2080 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2082 /* May as well allow recovery to be retried once */
2083 mddev->recovery_disabled++;
2088 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2093 static void md_delayed_delete(struct work_struct *ws)
2095 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2096 kobject_del(&rdev->kobj);
2097 kobject_put(&rdev->kobj);
2100 static void unbind_rdev_from_array(struct md_rdev *rdev)
2102 char b[BDEVNAME_SIZE];
2107 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2108 list_del_rcu(&rdev->same_set);
2109 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2111 sysfs_remove_link(&rdev->kobj, "block");
2112 sysfs_put(rdev->sysfs_state);
2113 rdev->sysfs_state = NULL;
2114 rdev->badblocks.count = 0;
2115 /* We need to delay this, otherwise we can deadlock when
2116 * writing to 'remove' to "dev/state". We also need
2117 * to delay it due to rcu usage.
2120 INIT_WORK(&rdev->del_work, md_delayed_delete);
2121 kobject_get(&rdev->kobj);
2122 queue_work(md_misc_wq, &rdev->del_work);
2126 * prevent the device from being mounted, repartitioned or
2127 * otherwise reused by a RAID array (or any other kernel
2128 * subsystem), by bd_claiming the device.
2130 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2133 struct block_device *bdev;
2134 char b[BDEVNAME_SIZE];
2136 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2137 shared ? (struct md_rdev *)lock_rdev : rdev);
2139 printk(KERN_ERR "md: could not open %s.\n",
2140 __bdevname(dev, b));
2141 return PTR_ERR(bdev);
2147 static void unlock_rdev(struct md_rdev *rdev)
2149 struct block_device *bdev = rdev->bdev;
2153 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2156 void md_autodetect_dev(dev_t dev);
2158 static void export_rdev(struct md_rdev *rdev)
2160 char b[BDEVNAME_SIZE];
2161 printk(KERN_INFO "md: export_rdev(%s)\n",
2162 bdevname(rdev->bdev,b));
2165 md_rdev_clear(rdev);
2167 if (test_bit(AutoDetected, &rdev->flags))
2168 md_autodetect_dev(rdev->bdev->bd_dev);
2171 kobject_put(&rdev->kobj);
2174 static void kick_rdev_from_array(struct md_rdev *rdev)
2176 unbind_rdev_from_array(rdev);
2180 static void export_array(struct mddev *mddev)
2182 struct md_rdev *rdev;
2184 while (!list_empty(&mddev->disks)) {
2185 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2187 kick_rdev_from_array(rdev);
2189 mddev->raid_disks = 0;
2190 mddev->major_version = 0;
2193 static void print_desc(mdp_disk_t *desc)
2195 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2196 desc->major,desc->minor,desc->raid_disk,desc->state);
2199 static void print_sb_90(mdp_super_t *sb)
2204 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2205 sb->major_version, sb->minor_version, sb->patch_version,
2206 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2208 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2209 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2210 sb->md_minor, sb->layout, sb->chunk_size);
2211 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2212 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2213 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2214 sb->failed_disks, sb->spare_disks,
2215 sb->sb_csum, (unsigned long)sb->events_lo);
2218 for (i = 0; i < MD_SB_DISKS; i++) {
2221 desc = sb->disks + i;
2222 if (desc->number || desc->major || desc->minor ||
2223 desc->raid_disk || (desc->state && (desc->state != 4))) {
2224 printk(" D %2d: ", i);
2228 printk(KERN_INFO "md: THIS: ");
2229 print_desc(&sb->this_disk);
2232 static void print_sb_1(struct mdp_superblock_1 *sb)
2236 uuid = sb->set_uuid;
2238 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2239 "md: Name: \"%s\" CT:%llu\n",
2240 le32_to_cpu(sb->major_version),
2241 le32_to_cpu(sb->feature_map),
2244 (unsigned long long)le64_to_cpu(sb->ctime)
2245 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2247 uuid = sb->device_uuid;
2249 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2251 "md: Dev:%08x UUID: %pU\n"
2252 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2253 "md: (MaxDev:%u) \n",
2254 le32_to_cpu(sb->level),
2255 (unsigned long long)le64_to_cpu(sb->size),
2256 le32_to_cpu(sb->raid_disks),
2257 le32_to_cpu(sb->layout),
2258 le32_to_cpu(sb->chunksize),
2259 (unsigned long long)le64_to_cpu(sb->data_offset),
2260 (unsigned long long)le64_to_cpu(sb->data_size),
2261 (unsigned long long)le64_to_cpu(sb->super_offset),
2262 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2263 le32_to_cpu(sb->dev_number),
2266 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2267 (unsigned long long)le64_to_cpu(sb->events),
2268 (unsigned long long)le64_to_cpu(sb->resync_offset),
2269 le32_to_cpu(sb->sb_csum),
2270 le32_to_cpu(sb->max_dev)
2274 static void print_rdev(struct md_rdev *rdev, int major_version)
2276 char b[BDEVNAME_SIZE];
2277 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2278 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2279 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2281 if (rdev->sb_loaded) {
2282 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2283 switch (major_version) {
2285 print_sb_90(page_address(rdev->sb_page));
2288 print_sb_1(page_address(rdev->sb_page));
2292 printk(KERN_INFO "md: no rdev superblock!\n");
2295 static void md_print_devices(void)
2297 struct list_head *tmp;
2298 struct md_rdev *rdev;
2299 struct mddev *mddev;
2300 char b[BDEVNAME_SIZE];
2303 printk("md: **********************************\n");
2304 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2305 printk("md: **********************************\n");
2306 for_each_mddev(mddev, tmp) {
2309 bitmap_print_sb(mddev->bitmap);
2311 printk("%s: ", mdname(mddev));
2312 rdev_for_each(rdev, mddev)
2313 printk("<%s>", bdevname(rdev->bdev,b));
2316 rdev_for_each(rdev, mddev)
2317 print_rdev(rdev, mddev->major_version);
2319 printk("md: **********************************\n");
2323 static void sync_sbs(struct mddev *mddev, int nospares)
2325 /* Update each superblock (in-memory image), but
2326 * if we are allowed to, skip spares which already
2327 * have the right event counter, or have one earlier
2328 * (which would mean they aren't being marked as dirty
2329 * with the rest of the array)
2331 struct md_rdev *rdev;
2332 rdev_for_each(rdev, mddev) {
2333 if (rdev->sb_events == mddev->events ||
2335 rdev->raid_disk < 0 &&
2336 rdev->sb_events+1 == mddev->events)) {
2337 /* Don't update this superblock */
2338 rdev->sb_loaded = 2;
2340 sync_super(mddev, rdev);
2341 rdev->sb_loaded = 1;
2346 static void md_update_sb(struct mddev *mddev, int force_change)
2348 struct md_rdev *rdev;
2351 int any_badblocks_changed = 0;
2355 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2359 /* First make sure individual recovery_offsets are correct */
2360 rdev_for_each(rdev, mddev) {
2361 if (rdev->raid_disk >= 0 &&
2362 mddev->delta_disks >= 0 &&
2363 !test_bit(In_sync, &rdev->flags) &&
2364 mddev->curr_resync_completed > rdev->recovery_offset)
2365 rdev->recovery_offset = mddev->curr_resync_completed;
2368 if (!mddev->persistent) {
2369 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2370 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2371 if (!mddev->external) {
2372 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2373 rdev_for_each(rdev, mddev) {
2374 if (rdev->badblocks.changed) {
2375 rdev->badblocks.changed = 0;
2376 md_ack_all_badblocks(&rdev->badblocks);
2377 md_error(mddev, rdev);
2379 clear_bit(Blocked, &rdev->flags);
2380 clear_bit(BlockedBadBlocks, &rdev->flags);
2381 wake_up(&rdev->blocked_wait);
2384 wake_up(&mddev->sb_wait);
2388 spin_lock_irq(&mddev->write_lock);
2390 mddev->utime = get_seconds();
2392 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2394 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2395 /* just a clean<-> dirty transition, possibly leave spares alone,
2396 * though if events isn't the right even/odd, we will have to do
2402 if (mddev->degraded)
2403 /* If the array is degraded, then skipping spares is both
2404 * dangerous and fairly pointless.
2405 * Dangerous because a device that was removed from the array
2406 * might have a event_count that still looks up-to-date,
2407 * so it can be re-added without a resync.
2408 * Pointless because if there are any spares to skip,
2409 * then a recovery will happen and soon that array won't
2410 * be degraded any more and the spare can go back to sleep then.
2414 sync_req = mddev->in_sync;
2416 /* If this is just a dirty<->clean transition, and the array is clean
2417 * and 'events' is odd, we can roll back to the previous clean state */
2419 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2420 && mddev->can_decrease_events
2421 && mddev->events != 1) {
2423 mddev->can_decrease_events = 0;
2425 /* otherwise we have to go forward and ... */
2427 mddev->can_decrease_events = nospares;
2430 if (!mddev->events) {
2432 * oops, this 64-bit counter should never wrap.
2433 * Either we are in around ~1 trillion A.C., assuming
2434 * 1 reboot per second, or we have a bug:
2440 rdev_for_each(rdev, mddev) {
2441 if (rdev->badblocks.changed)
2442 any_badblocks_changed++;
2443 if (test_bit(Faulty, &rdev->flags))
2444 set_bit(FaultRecorded, &rdev->flags);
2447 sync_sbs(mddev, nospares);
2448 spin_unlock_irq(&mddev->write_lock);
2450 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2451 mdname(mddev), mddev->in_sync);
2453 bitmap_update_sb(mddev->bitmap);
2454 rdev_for_each(rdev, mddev) {
2455 char b[BDEVNAME_SIZE];
2457 if (rdev->sb_loaded != 1)
2458 continue; /* no noise on spare devices */
2460 if (!test_bit(Faulty, &rdev->flags)) {
2461 md_super_write(mddev,rdev,
2462 rdev->sb_start, rdev->sb_size,
2464 pr_debug("md: (write) %s's sb offset: %llu\n",
2465 bdevname(rdev->bdev, b),
2466 (unsigned long long)rdev->sb_start);
2467 rdev->sb_events = mddev->events;
2468 if (rdev->badblocks.size) {
2469 md_super_write(mddev, rdev,
2470 rdev->badblocks.sector,
2471 rdev->badblocks.size << 9,
2473 rdev->badblocks.size = 0;
2477 pr_debug("md: %s (skipping faulty)\n",
2478 bdevname(rdev->bdev, b));
2480 if (mddev->level == LEVEL_MULTIPATH)
2481 /* only need to write one superblock... */
2484 md_super_wait(mddev);
2485 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2487 spin_lock_irq(&mddev->write_lock);
2488 if (mddev->in_sync != sync_req ||
2489 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2490 /* have to write it out again */
2491 spin_unlock_irq(&mddev->write_lock);
2494 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2495 spin_unlock_irq(&mddev->write_lock);
2496 wake_up(&mddev->sb_wait);
2497 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2498 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2500 rdev_for_each(rdev, mddev) {
2501 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2502 clear_bit(Blocked, &rdev->flags);
2504 if (any_badblocks_changed)
2505 md_ack_all_badblocks(&rdev->badblocks);
2506 clear_bit(BlockedBadBlocks, &rdev->flags);
2507 wake_up(&rdev->blocked_wait);
2511 /* words written to sysfs files may, or may not, be \n terminated.
2512 * We want to accept with case. For this we use cmd_match.
2514 static int cmd_match(const char *cmd, const char *str)
2516 /* See if cmd, written into a sysfs file, matches
2517 * str. They must either be the same, or cmd can
2518 * have a trailing newline
2520 while (*cmd && *str && *cmd == *str) {
2531 struct rdev_sysfs_entry {
2532 struct attribute attr;
2533 ssize_t (*show)(struct md_rdev *, char *);
2534 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2538 state_show(struct md_rdev *rdev, char *page)
2543 if (test_bit(Faulty, &rdev->flags) ||
2544 rdev->badblocks.unacked_exist) {
2545 len+= sprintf(page+len, "%sfaulty",sep);
2548 if (test_bit(In_sync, &rdev->flags)) {
2549 len += sprintf(page+len, "%sin_sync",sep);
2552 if (test_bit(WriteMostly, &rdev->flags)) {
2553 len += sprintf(page+len, "%swrite_mostly",sep);
2556 if (test_bit(Blocked, &rdev->flags) ||
2557 (rdev->badblocks.unacked_exist
2558 && !test_bit(Faulty, &rdev->flags))) {
2559 len += sprintf(page+len, "%sblocked", sep);
2562 if (!test_bit(Faulty, &rdev->flags) &&
2563 !test_bit(In_sync, &rdev->flags)) {
2564 len += sprintf(page+len, "%sspare", sep);
2567 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2568 len += sprintf(page+len, "%swrite_error", sep);
2571 if (test_bit(WantReplacement, &rdev->flags)) {
2572 len += sprintf(page+len, "%swant_replacement", sep);
2575 if (test_bit(Replacement, &rdev->flags)) {
2576 len += sprintf(page+len, "%sreplacement", sep);
2580 return len+sprintf(page+len, "\n");
2584 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2587 * faulty - simulates an error
2588 * remove - disconnects the device
2589 * writemostly - sets write_mostly
2590 * -writemostly - clears write_mostly
2591 * blocked - sets the Blocked flags
2592 * -blocked - clears the Blocked and possibly simulates an error
2593 * insync - sets Insync providing device isn't active
2594 * -insync - clear Insync for a device with a slot assigned,
2595 * so that it gets rebuilt based on bitmap
2596 * write_error - sets WriteErrorSeen
2597 * -write_error - clears WriteErrorSeen
2600 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2601 md_error(rdev->mddev, rdev);
2602 if (test_bit(Faulty, &rdev->flags))
2606 } else if (cmd_match(buf, "remove")) {
2607 if (rdev->raid_disk >= 0)
2610 struct mddev *mddev = rdev->mddev;
2611 kick_rdev_from_array(rdev);
2613 md_update_sb(mddev, 1);
2614 md_new_event(mddev);
2617 } else if (cmd_match(buf, "writemostly")) {
2618 set_bit(WriteMostly, &rdev->flags);
2620 } else if (cmd_match(buf, "-writemostly")) {
2621 clear_bit(WriteMostly, &rdev->flags);
2623 } else if (cmd_match(buf, "blocked")) {
2624 set_bit(Blocked, &rdev->flags);
2626 } else if (cmd_match(buf, "-blocked")) {
2627 if (!test_bit(Faulty, &rdev->flags) &&
2628 rdev->badblocks.unacked_exist) {
2629 /* metadata handler doesn't understand badblocks,
2630 * so we need to fail the device
2632 md_error(rdev->mddev, rdev);
2634 clear_bit(Blocked, &rdev->flags);
2635 clear_bit(BlockedBadBlocks, &rdev->flags);
2636 wake_up(&rdev->blocked_wait);
2637 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2638 md_wakeup_thread(rdev->mddev->thread);
2641 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2642 set_bit(In_sync, &rdev->flags);
2644 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2645 if (rdev->mddev->pers == NULL) {
2646 clear_bit(In_sync, &rdev->flags);
2647 rdev->saved_raid_disk = rdev->raid_disk;
2648 rdev->raid_disk = -1;
2651 } else if (cmd_match(buf, "write_error")) {
2652 set_bit(WriteErrorSeen, &rdev->flags);
2654 } else if (cmd_match(buf, "-write_error")) {
2655 clear_bit(WriteErrorSeen, &rdev->flags);
2657 } else if (cmd_match(buf, "want_replacement")) {
2658 /* Any non-spare device that is not a replacement can
2659 * become want_replacement at any time, but we then need to
2660 * check if recovery is needed.
2662 if (rdev->raid_disk >= 0 &&
2663 !test_bit(Replacement, &rdev->flags))
2664 set_bit(WantReplacement, &rdev->flags);
2665 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2666 md_wakeup_thread(rdev->mddev->thread);
2668 } else if (cmd_match(buf, "-want_replacement")) {
2669 /* Clearing 'want_replacement' is always allowed.
2670 * Once replacements starts it is too late though.
2673 clear_bit(WantReplacement, &rdev->flags);
2674 } else if (cmd_match(buf, "replacement")) {
2675 /* Can only set a device as a replacement when array has not
2676 * yet been started. Once running, replacement is automatic
2677 * from spares, or by assigning 'slot'.
2679 if (rdev->mddev->pers)
2682 set_bit(Replacement, &rdev->flags);
2685 } else if (cmd_match(buf, "-replacement")) {
2686 /* Similarly, can only clear Replacement before start */
2687 if (rdev->mddev->pers)
2690 clear_bit(Replacement, &rdev->flags);
2695 sysfs_notify_dirent_safe(rdev->sysfs_state);
2696 return err ? err : len;
2698 static struct rdev_sysfs_entry rdev_state =
2699 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2702 errors_show(struct md_rdev *rdev, char *page)
2704 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2708 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2711 unsigned long n = simple_strtoul(buf, &e, 10);
2712 if (*buf && (*e == 0 || *e == '\n')) {
2713 atomic_set(&rdev->corrected_errors, n);
2718 static struct rdev_sysfs_entry rdev_errors =
2719 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2722 slot_show(struct md_rdev *rdev, char *page)
2724 if (rdev->raid_disk < 0)
2725 return sprintf(page, "none\n");
2727 return sprintf(page, "%d\n", rdev->raid_disk);
2731 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2735 int slot = simple_strtoul(buf, &e, 10);
2736 if (strncmp(buf, "none", 4)==0)
2738 else if (e==buf || (*e && *e!= '\n'))
2740 if (rdev->mddev->pers && slot == -1) {
2741 /* Setting 'slot' on an active array requires also
2742 * updating the 'rd%d' link, and communicating
2743 * with the personality with ->hot_*_disk.
2744 * For now we only support removing
2745 * failed/spare devices. This normally happens automatically,
2746 * but not when the metadata is externally managed.
2748 if (rdev->raid_disk == -1)
2750 /* personality does all needed checks */
2751 if (rdev->mddev->pers->hot_remove_disk == NULL)
2753 clear_bit(Blocked, &rdev->flags);
2754 remove_and_add_spares(rdev->mddev, rdev);
2755 if (rdev->raid_disk >= 0)
2757 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2758 md_wakeup_thread(rdev->mddev->thread);
2759 } else if (rdev->mddev->pers) {
2760 /* Activating a spare .. or possibly reactivating
2761 * if we ever get bitmaps working here.
2764 if (rdev->raid_disk != -1)
2767 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2770 if (rdev->mddev->pers->hot_add_disk == NULL)
2773 if (slot >= rdev->mddev->raid_disks &&
2774 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2777 rdev->raid_disk = slot;
2778 if (test_bit(In_sync, &rdev->flags))
2779 rdev->saved_raid_disk = slot;
2781 rdev->saved_raid_disk = -1;
2782 clear_bit(In_sync, &rdev->flags);
2783 clear_bit(Bitmap_sync, &rdev->flags);
2784 err = rdev->mddev->pers->
2785 hot_add_disk(rdev->mddev, rdev);
2787 rdev->raid_disk = -1;
2790 sysfs_notify_dirent_safe(rdev->sysfs_state);
2791 if (sysfs_link_rdev(rdev->mddev, rdev))
2792 /* failure here is OK */;
2793 /* don't wakeup anyone, leave that to userspace. */
2795 if (slot >= rdev->mddev->raid_disks &&
2796 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2798 rdev->raid_disk = slot;
2799 /* assume it is working */
2800 clear_bit(Faulty, &rdev->flags);
2801 clear_bit(WriteMostly, &rdev->flags);
2802 set_bit(In_sync, &rdev->flags);
2803 sysfs_notify_dirent_safe(rdev->sysfs_state);
2808 static struct rdev_sysfs_entry rdev_slot =
2809 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2812 offset_show(struct md_rdev *rdev, char *page)
2814 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2818 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2820 unsigned long long offset;
2821 if (kstrtoull(buf, 10, &offset) < 0)
2823 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2825 if (rdev->sectors && rdev->mddev->external)
2826 /* Must set offset before size, so overlap checks
2829 rdev->data_offset = offset;
2830 rdev->new_data_offset = offset;
2834 static struct rdev_sysfs_entry rdev_offset =
2835 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2837 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2839 return sprintf(page, "%llu\n",
2840 (unsigned long long)rdev->new_data_offset);
2843 static ssize_t new_offset_store(struct md_rdev *rdev,
2844 const char *buf, size_t len)
2846 unsigned long long new_offset;
2847 struct mddev *mddev = rdev->mddev;
2849 if (kstrtoull(buf, 10, &new_offset) < 0)
2852 if (mddev->sync_thread)
2854 if (new_offset == rdev->data_offset)
2855 /* reset is always permitted */
2857 else if (new_offset > rdev->data_offset) {
2858 /* must not push array size beyond rdev_sectors */
2859 if (new_offset - rdev->data_offset
2860 + mddev->dev_sectors > rdev->sectors)
2863 /* Metadata worries about other space details. */
2865 /* decreasing the offset is inconsistent with a backwards
2868 if (new_offset < rdev->data_offset &&
2869 mddev->reshape_backwards)
2871 /* Increasing offset is inconsistent with forwards
2872 * reshape. reshape_direction should be set to
2873 * 'backwards' first.
2875 if (new_offset > rdev->data_offset &&
2876 !mddev->reshape_backwards)
2879 if (mddev->pers && mddev->persistent &&
2880 !super_types[mddev->major_version]
2881 .allow_new_offset(rdev, new_offset))
2883 rdev->new_data_offset = new_offset;
2884 if (new_offset > rdev->data_offset)
2885 mddev->reshape_backwards = 1;
2886 else if (new_offset < rdev->data_offset)
2887 mddev->reshape_backwards = 0;
2891 static struct rdev_sysfs_entry rdev_new_offset =
2892 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2895 rdev_size_show(struct md_rdev *rdev, char *page)
2897 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2900 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2902 /* check if two start/length pairs overlap */
2910 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2912 unsigned long long blocks;
2915 if (kstrtoull(buf, 10, &blocks) < 0)
2918 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2919 return -EINVAL; /* sector conversion overflow */
2922 if (new != blocks * 2)
2923 return -EINVAL; /* unsigned long long to sector_t overflow */
2930 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2932 struct mddev *my_mddev = rdev->mddev;
2933 sector_t oldsectors = rdev->sectors;
2936 if (strict_blocks_to_sectors(buf, §ors) < 0)
2938 if (rdev->data_offset != rdev->new_data_offset)
2939 return -EINVAL; /* too confusing */
2940 if (my_mddev->pers && rdev->raid_disk >= 0) {
2941 if (my_mddev->persistent) {
2942 sectors = super_types[my_mddev->major_version].
2943 rdev_size_change(rdev, sectors);
2946 } else if (!sectors)
2947 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2949 if (!my_mddev->pers->resize)
2950 /* Cannot change size for RAID0 or Linear etc */
2953 if (sectors < my_mddev->dev_sectors)
2954 return -EINVAL; /* component must fit device */
2956 rdev->sectors = sectors;
2957 if (sectors > oldsectors && my_mddev->external) {
2958 /* Need to check that all other rdevs with the same
2959 * ->bdev do not overlap. 'rcu' is sufficient to walk
2960 * the rdev lists safely.
2961 * This check does not provide a hard guarantee, it
2962 * just helps avoid dangerous mistakes.
2964 struct mddev *mddev;
2966 struct list_head *tmp;
2969 for_each_mddev(mddev, tmp) {
2970 struct md_rdev *rdev2;
2972 rdev_for_each(rdev2, mddev)
2973 if (rdev->bdev == rdev2->bdev &&
2975 overlaps(rdev->data_offset, rdev->sectors,
2988 /* Someone else could have slipped in a size
2989 * change here, but doing so is just silly.
2990 * We put oldsectors back because we *know* it is
2991 * safe, and trust userspace not to race with
2994 rdev->sectors = oldsectors;
3001 static struct rdev_sysfs_entry rdev_size =
3002 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3004 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3006 unsigned long long recovery_start = rdev->recovery_offset;
3008 if (test_bit(In_sync, &rdev->flags) ||
3009 recovery_start == MaxSector)
3010 return sprintf(page, "none\n");
3012 return sprintf(page, "%llu\n", recovery_start);
3015 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3017 unsigned long long recovery_start;
3019 if (cmd_match(buf, "none"))
3020 recovery_start = MaxSector;
3021 else if (kstrtoull(buf, 10, &recovery_start))
3024 if (rdev->mddev->pers &&
3025 rdev->raid_disk >= 0)
3028 rdev->recovery_offset = recovery_start;
3029 if (recovery_start == MaxSector)
3030 set_bit(In_sync, &rdev->flags);
3032 clear_bit(In_sync, &rdev->flags);
3036 static struct rdev_sysfs_entry rdev_recovery_start =
3037 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3040 badblocks_show(struct badblocks *bb, char *page, int unack);
3042 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3044 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3046 return badblocks_show(&rdev->badblocks, page, 0);
3048 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3050 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3051 /* Maybe that ack was all we needed */
3052 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3053 wake_up(&rdev->blocked_wait);
3056 static struct rdev_sysfs_entry rdev_bad_blocks =
3057 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3059 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3061 return badblocks_show(&rdev->badblocks, page, 1);
3063 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3065 return badblocks_store(&rdev->badblocks, page, len, 1);
3067 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3068 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3070 static struct attribute *rdev_default_attrs[] = {
3075 &rdev_new_offset.attr,
3077 &rdev_recovery_start.attr,
3078 &rdev_bad_blocks.attr,
3079 &rdev_unack_bad_blocks.attr,
3083 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3085 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3086 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3087 struct mddev *mddev = rdev->mddev;
3093 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3095 if (rdev->mddev == NULL)
3098 rv = entry->show(rdev, page);
3099 mddev_unlock(mddev);
3105 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3106 const char *page, size_t length)
3108 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3109 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3111 struct mddev *mddev = rdev->mddev;
3115 if (!capable(CAP_SYS_ADMIN))
3117 rv = mddev ? mddev_lock(mddev): -EBUSY;
3119 if (rdev->mddev == NULL)
3122 rv = entry->store(rdev, page, length);
3123 mddev_unlock(mddev);
3128 static void rdev_free(struct kobject *ko)
3130 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3133 static const struct sysfs_ops rdev_sysfs_ops = {
3134 .show = rdev_attr_show,
3135 .store = rdev_attr_store,
3137 static struct kobj_type rdev_ktype = {
3138 .release = rdev_free,
3139 .sysfs_ops = &rdev_sysfs_ops,
3140 .default_attrs = rdev_default_attrs,
3143 int md_rdev_init(struct md_rdev *rdev)
3146 rdev->saved_raid_disk = -1;
3147 rdev->raid_disk = -1;
3149 rdev->data_offset = 0;
3150 rdev->new_data_offset = 0;
3151 rdev->sb_events = 0;
3152 rdev->last_read_error.tv_sec = 0;
3153 rdev->last_read_error.tv_nsec = 0;
3154 rdev->sb_loaded = 0;
3155 rdev->bb_page = NULL;
3156 atomic_set(&rdev->nr_pending, 0);
3157 atomic_set(&rdev->read_errors, 0);
3158 atomic_set(&rdev->corrected_errors, 0);
3160 INIT_LIST_HEAD(&rdev->same_set);
3161 init_waitqueue_head(&rdev->blocked_wait);
3163 /* Add space to store bad block list.
3164 * This reserves the space even on arrays where it cannot
3165 * be used - I wonder if that matters
3167 rdev->badblocks.count = 0;
3168 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3169 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3170 seqlock_init(&rdev->badblocks.lock);
3171 if (rdev->badblocks.page == NULL)
3176 EXPORT_SYMBOL_GPL(md_rdev_init);
3178 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3180 * mark the device faulty if:
3182 * - the device is nonexistent (zero size)
3183 * - the device has no valid superblock
3185 * a faulty rdev _never_ has rdev->sb set.
3187 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3189 char b[BDEVNAME_SIZE];
3191 struct md_rdev *rdev;
3194 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3196 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3197 return ERR_PTR(-ENOMEM);
3200 err = md_rdev_init(rdev);
3203 err = alloc_disk_sb(rdev);
3207 err = lock_rdev(rdev, newdev, super_format == -2);
3211 kobject_init(&rdev->kobj, &rdev_ktype);
3213 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3216 "md: %s has zero or unknown size, marking faulty!\n",
3217 bdevname(rdev->bdev,b));
3222 if (super_format >= 0) {
3223 err = super_types[super_format].
3224 load_super(rdev, NULL, super_minor);
3225 if (err == -EINVAL) {
3227 "md: %s does not have a valid v%d.%d "
3228 "superblock, not importing!\n",
3229 bdevname(rdev->bdev,b),
3230 super_format, super_minor);
3235 "md: could not read %s's sb, not importing!\n",
3236 bdevname(rdev->bdev,b));
3246 md_rdev_clear(rdev);
3248 return ERR_PTR(err);
3252 * Check a full RAID array for plausibility
3255 static void analyze_sbs(struct mddev *mddev)
3258 struct md_rdev *rdev, *freshest, *tmp;
3259 char b[BDEVNAME_SIZE];
3262 rdev_for_each_safe(rdev, tmp, mddev)
3263 switch (super_types[mddev->major_version].
3264 load_super(rdev, freshest, mddev->minor_version)) {
3272 "md: fatal superblock inconsistency in %s"
3273 " -- removing from array\n",
3274 bdevname(rdev->bdev,b));
3275 kick_rdev_from_array(rdev);
3278 super_types[mddev->major_version].
3279 validate_super(mddev, freshest);
3282 rdev_for_each_safe(rdev, tmp, mddev) {
3283 if (mddev->max_disks &&
3284 (rdev->desc_nr >= mddev->max_disks ||
3285 i > mddev->max_disks)) {
3287 "md: %s: %s: only %d devices permitted\n",
3288 mdname(mddev), bdevname(rdev->bdev, b),
3290 kick_rdev_from_array(rdev);
3293 if (rdev != freshest)
3294 if (super_types[mddev->major_version].
3295 validate_super(mddev, rdev)) {
3296 printk(KERN_WARNING "md: kicking non-fresh %s"
3298 bdevname(rdev->bdev,b));
3299 kick_rdev_from_array(rdev);
3302 if (mddev->level == LEVEL_MULTIPATH) {
3303 rdev->desc_nr = i++;
3304 rdev->raid_disk = rdev->desc_nr;
3305 set_bit(In_sync, &rdev->flags);
3306 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3307 rdev->raid_disk = -1;
3308 clear_bit(In_sync, &rdev->flags);
3313 /* Read a fixed-point number.
3314 * Numbers in sysfs attributes should be in "standard" units where
3315 * possible, so time should be in seconds.
3316 * However we internally use a a much smaller unit such as
3317 * milliseconds or jiffies.
3318 * This function takes a decimal number with a possible fractional
3319 * component, and produces an integer which is the result of
3320 * multiplying that number by 10^'scale'.
3321 * all without any floating-point arithmetic.
3323 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3325 unsigned long result = 0;
3327 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3330 else if (decimals < scale) {
3333 result = result * 10 + value;
3345 while (decimals < scale) {
3353 static void md_safemode_timeout(unsigned long data);
3356 safe_delay_show(struct mddev *mddev, char *page)
3358 int msec = (mddev->safemode_delay*1000)/HZ;
3359 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3362 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3366 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3369 mddev->safemode_delay = 0;
3371 unsigned long old_delay = mddev->safemode_delay;
3372 mddev->safemode_delay = (msec*HZ)/1000;
3373 if (mddev->safemode_delay == 0)
3374 mddev->safemode_delay = 1;
3375 if (mddev->safemode_delay < old_delay || old_delay == 0)
3376 md_safemode_timeout((unsigned long)mddev);
3380 static struct md_sysfs_entry md_safe_delay =
3381 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3384 level_show(struct mddev *mddev, char *page)
3386 struct md_personality *p = mddev->pers;
3388 return sprintf(page, "%s\n", p->name);
3389 else if (mddev->clevel[0])
3390 return sprintf(page, "%s\n", mddev->clevel);
3391 else if (mddev->level != LEVEL_NONE)
3392 return sprintf(page, "%d\n", mddev->level);
3398 level_store(struct mddev *mddev, const char *buf, size_t len)
3402 struct md_personality *pers;
3405 struct md_rdev *rdev;
3407 if (mddev->pers == NULL) {
3410 if (len >= sizeof(mddev->clevel))
3412 strncpy(mddev->clevel, buf, len);
3413 if (mddev->clevel[len-1] == '\n')
3415 mddev->clevel[len] = 0;
3416 mddev->level = LEVEL_NONE;
3422 /* request to change the personality. Need to ensure:
3423 * - array is not engaged in resync/recovery/reshape
3424 * - old personality can be suspended
3425 * - new personality will access other array.
3428 if (mddev->sync_thread ||
3429 mddev->reshape_position != MaxSector ||
3430 mddev->sysfs_active)
3433 if (!mddev->pers->quiesce) {
3434 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3435 mdname(mddev), mddev->pers->name);
3439 /* Now find the new personality */
3440 if (len == 0 || len >= sizeof(clevel))
3442 strncpy(clevel, buf, len);
3443 if (clevel[len-1] == '\n')
3446 if (kstrtol(clevel, 10, &level))
3449 if (request_module("md-%s", clevel) != 0)
3450 request_module("md-level-%s", clevel);
3451 spin_lock(&pers_lock);
3452 pers = find_pers(level, clevel);
3453 if (!pers || !try_module_get(pers->owner)) {
3454 spin_unlock(&pers_lock);
3455 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3458 spin_unlock(&pers_lock);
3460 if (pers == mddev->pers) {
3461 /* Nothing to do! */
3462 module_put(pers->owner);
3465 if (!pers->takeover) {
3466 module_put(pers->owner);
3467 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3468 mdname(mddev), clevel);
3472 rdev_for_each(rdev, mddev)
3473 rdev->new_raid_disk = rdev->raid_disk;
3475 /* ->takeover must set new_* and/or delta_disks
3476 * if it succeeds, and may set them when it fails.
3478 priv = pers->takeover(mddev);
3480 mddev->new_level = mddev->level;
3481 mddev->new_layout = mddev->layout;
3482 mddev->new_chunk_sectors = mddev->chunk_sectors;
3483 mddev->raid_disks -= mddev->delta_disks;
3484 mddev->delta_disks = 0;
3485 mddev->reshape_backwards = 0;
3486 module_put(pers->owner);
3487 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3488 mdname(mddev), clevel);
3489 return PTR_ERR(priv);
3492 /* Looks like we have a winner */
3493 mddev_suspend(mddev);
3494 mddev->pers->stop(mddev);
3496 if (mddev->pers->sync_request == NULL &&
3497 pers->sync_request != NULL) {
3498 /* need to add the md_redundancy_group */
3499 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3501 "md: cannot register extra attributes for %s\n",
3503 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3505 if (mddev->pers->sync_request != NULL &&
3506 pers->sync_request == NULL) {
3507 /* need to remove the md_redundancy_group */
3508 if (mddev->to_remove == NULL)
3509 mddev->to_remove = &md_redundancy_group;
3512 if (mddev->pers->sync_request == NULL &&
3514 /* We are converting from a no-redundancy array
3515 * to a redundancy array and metadata is managed
3516 * externally so we need to be sure that writes
3517 * won't block due to a need to transition
3519 * until external management is started.
3522 mddev->safemode_delay = 0;
3523 mddev->safemode = 0;
3526 rdev_for_each(rdev, mddev) {
3527 if (rdev->raid_disk < 0)
3529 if (rdev->new_raid_disk >= mddev->raid_disks)
3530 rdev->new_raid_disk = -1;
3531 if (rdev->new_raid_disk == rdev->raid_disk)
3533 sysfs_unlink_rdev(mddev, rdev);
3535 rdev_for_each(rdev, mddev) {
3536 if (rdev->raid_disk < 0)
3538 if (rdev->new_raid_disk == rdev->raid_disk)
3540 rdev->raid_disk = rdev->new_raid_disk;
3541 if (rdev->raid_disk < 0)
3542 clear_bit(In_sync, &rdev->flags);
3544 if (sysfs_link_rdev(mddev, rdev))
3545 printk(KERN_WARNING "md: cannot register rd%d"
3546 " for %s after level change\n",
3547 rdev->raid_disk, mdname(mddev));
3551 module_put(mddev->pers->owner);
3553 mddev->private = priv;
3554 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3555 mddev->level = mddev->new_level;
3556 mddev->layout = mddev->new_layout;
3557 mddev->chunk_sectors = mddev->new_chunk_sectors;
3558 mddev->delta_disks = 0;
3559 mddev->reshape_backwards = 0;
3560 mddev->degraded = 0;
3561 if (mddev->pers->sync_request == NULL) {
3562 /* this is now an array without redundancy, so
3563 * it must always be in_sync
3566 del_timer_sync(&mddev->safemode_timer);
3568 blk_set_stacking_limits(&mddev->queue->limits);
3570 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3571 mddev_resume(mddev);
3573 md_update_sb(mddev, 1);
3574 sysfs_notify(&mddev->kobj, NULL, "level");
3575 md_new_event(mddev);
3579 static struct md_sysfs_entry md_level =
3580 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3583 layout_show(struct mddev *mddev, char *page)
3585 /* just a number, not meaningful for all levels */
3586 if (mddev->reshape_position != MaxSector &&
3587 mddev->layout != mddev->new_layout)
3588 return sprintf(page, "%d (%d)\n",
3589 mddev->new_layout, mddev->layout);
3590 return sprintf(page, "%d\n", mddev->layout);
3594 layout_store(struct mddev *mddev, const char *buf, size_t len)
3597 unsigned long n = simple_strtoul(buf, &e, 10);
3599 if (!*buf || (*e && *e != '\n'))
3604 if (mddev->pers->check_reshape == NULL)
3608 mddev->new_layout = n;
3609 err = mddev->pers->check_reshape(mddev);
3611 mddev->new_layout = mddev->layout;
3615 mddev->new_layout = n;
3616 if (mddev->reshape_position == MaxSector)
3621 static struct md_sysfs_entry md_layout =
3622 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3625 raid_disks_show(struct mddev *mddev, char *page)
3627 if (mddev->raid_disks == 0)
3629 if (mddev->reshape_position != MaxSector &&
3630 mddev->delta_disks != 0)
3631 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3632 mddev->raid_disks - mddev->delta_disks);
3633 return sprintf(page, "%d\n", mddev->raid_disks);
3636 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3639 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3643 unsigned long n = simple_strtoul(buf, &e, 10);
3645 if (!*buf || (*e && *e != '\n'))
3649 rv = update_raid_disks(mddev, n);
3650 else if (mddev->reshape_position != MaxSector) {
3651 struct md_rdev *rdev;
3652 int olddisks = mddev->raid_disks - mddev->delta_disks;
3654 rdev_for_each(rdev, mddev) {
3656 rdev->data_offset < rdev->new_data_offset)
3659 rdev->data_offset > rdev->new_data_offset)
3662 mddev->delta_disks = n - olddisks;
3663 mddev->raid_disks = n;
3664 mddev->reshape_backwards = (mddev->delta_disks < 0);
3666 mddev->raid_disks = n;
3667 return rv ? rv : len;
3669 static struct md_sysfs_entry md_raid_disks =
3670 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3673 chunk_size_show(struct mddev *mddev, char *page)
3675 if (mddev->reshape_position != MaxSector &&
3676 mddev->chunk_sectors != mddev->new_chunk_sectors)
3677 return sprintf(page, "%d (%d)\n",
3678 mddev->new_chunk_sectors << 9,
3679 mddev->chunk_sectors << 9);
3680 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3684 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3687 unsigned long n = simple_strtoul(buf, &e, 10);
3689 if (!*buf || (*e && *e != '\n'))
3694 if (mddev->pers->check_reshape == NULL)
3698 mddev->new_chunk_sectors = n >> 9;
3699 err = mddev->pers->check_reshape(mddev);
3701 mddev->new_chunk_sectors = mddev->chunk_sectors;
3705 mddev->new_chunk_sectors = n >> 9;
3706 if (mddev->reshape_position == MaxSector)
3707 mddev->chunk_sectors = n >> 9;
3711 static struct md_sysfs_entry md_chunk_size =
3712 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3715 resync_start_show(struct mddev *mddev, char *page)
3717 if (mddev->recovery_cp == MaxSector)
3718 return sprintf(page, "none\n");
3719 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3723 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3726 unsigned long long n = simple_strtoull(buf, &e, 10);
3728 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3730 if (cmd_match(buf, "none"))
3732 else if (!*buf || (*e && *e != '\n'))
3735 mddev->recovery_cp = n;
3737 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3740 static struct md_sysfs_entry md_resync_start =
3741 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3744 * The array state can be:
3747 * No devices, no size, no level
3748 * Equivalent to STOP_ARRAY ioctl
3750 * May have some settings, but array is not active
3751 * all IO results in error
3752 * When written, doesn't tear down array, but just stops it
3753 * suspended (not supported yet)
3754 * All IO requests will block. The array can be reconfigured.
3755 * Writing this, if accepted, will block until array is quiescent
3757 * no resync can happen. no superblocks get written.
3758 * write requests fail
3760 * like readonly, but behaves like 'clean' on a write request.
3762 * clean - no pending writes, but otherwise active.
3763 * When written to inactive array, starts without resync
3764 * If a write request arrives then
3765 * if metadata is known, mark 'dirty' and switch to 'active'.
3766 * if not known, block and switch to write-pending
3767 * If written to an active array that has pending writes, then fails.
3769 * fully active: IO and resync can be happening.
3770 * When written to inactive array, starts with resync
3773 * clean, but writes are blocked waiting for 'active' to be written.
3776 * like active, but no writes have been seen for a while (100msec).
3779 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3780 write_pending, active_idle, bad_word};
3781 static char *array_states[] = {
3782 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3783 "write-pending", "active-idle", NULL };
3785 static int match_word(const char *word, char **list)
3788 for (n=0; list[n]; n++)
3789 if (cmd_match(word, list[n]))
3795 array_state_show(struct mddev *mddev, char *page)
3797 enum array_state st = inactive;
3810 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3812 else if (mddev->safemode)
3818 if (list_empty(&mddev->disks) &&
3819 mddev->raid_disks == 0 &&
3820 mddev->dev_sectors == 0)
3825 return sprintf(page, "%s\n", array_states[st]);
3828 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3829 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3830 static int do_md_run(struct mddev *mddev);
3831 static int restart_array(struct mddev *mddev);
3834 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3837 enum array_state st = match_word(buf, array_states);
3842 /* stopping an active array */
3843 err = do_md_stop(mddev, 0, NULL);
3846 /* stopping an active array */
3848 err = do_md_stop(mddev, 2, NULL);
3850 err = 0; /* already inactive */
3853 break; /* not supported yet */
3856 err = md_set_readonly(mddev, NULL);
3859 set_disk_ro(mddev->gendisk, 1);
3860 err = do_md_run(mddev);
3866 err = md_set_readonly(mddev, NULL);
3867 else if (mddev->ro == 1)
3868 err = restart_array(mddev);
3871 set_disk_ro(mddev->gendisk, 0);
3875 err = do_md_run(mddev);
3880 restart_array(mddev);
3881 spin_lock_irq(&mddev->write_lock);
3882 if (atomic_read(&mddev->writes_pending) == 0) {
3883 if (mddev->in_sync == 0) {
3885 if (mddev->safemode == 1)
3886 mddev->safemode = 0;
3887 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3892 spin_unlock_irq(&mddev->write_lock);
3898 restart_array(mddev);
3899 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3900 wake_up(&mddev->sb_wait);
3904 set_disk_ro(mddev->gendisk, 0);
3905 err = do_md_run(mddev);
3910 /* these cannot be set */
3916 if (mddev->hold_active == UNTIL_IOCTL)
3917 mddev->hold_active = 0;
3918 sysfs_notify_dirent_safe(mddev->sysfs_state);
3922 static struct md_sysfs_entry md_array_state =
3923 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3926 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3927 return sprintf(page, "%d\n",
3928 atomic_read(&mddev->max_corr_read_errors));
3932 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3935 unsigned long n = simple_strtoul(buf, &e, 10);
3937 if (*buf && (*e == 0 || *e == '\n')) {
3938 atomic_set(&mddev->max_corr_read_errors, n);
3944 static struct md_sysfs_entry max_corr_read_errors =
3945 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3946 max_corrected_read_errors_store);
3949 null_show(struct mddev *mddev, char *page)
3955 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3957 /* buf must be %d:%d\n? giving major and minor numbers */
3958 /* The new device is added to the array.
3959 * If the array has a persistent superblock, we read the
3960 * superblock to initialise info and check validity.
3961 * Otherwise, only checking done is that in bind_rdev_to_array,
3962 * which mainly checks size.
3965 int major = simple_strtoul(buf, &e, 10);
3968 struct md_rdev *rdev;
3971 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3973 minor = simple_strtoul(e+1, &e, 10);
3974 if (*e && *e != '\n')
3976 dev = MKDEV(major, minor);
3977 if (major != MAJOR(dev) ||
3978 minor != MINOR(dev))
3981 if (mddev->persistent) {
3982 rdev = md_import_device(dev, mddev->major_version,
3983 mddev->minor_version);
3984 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3985 struct md_rdev *rdev0
3986 = list_entry(mddev->disks.next,
3987 struct md_rdev, same_set);
3988 err = super_types[mddev->major_version]
3989 .load_super(rdev, rdev0, mddev->minor_version);
3993 } else if (mddev->external)
3994 rdev = md_import_device(dev, -2, -1);
3996 rdev = md_import_device(dev, -1, -1);
3999 return PTR_ERR(rdev);
4000 err = bind_rdev_to_array(rdev, mddev);
4004 return err ? err : len;
4007 static struct md_sysfs_entry md_new_device =
4008 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4011 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4014 unsigned long chunk, end_chunk;
4018 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4020 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4021 if (buf == end) break;
4022 if (*end == '-') { /* range */
4024 end_chunk = simple_strtoul(buf, &end, 0);
4025 if (buf == end) break;
4027 if (*end && !isspace(*end)) break;
4028 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4029 buf = skip_spaces(end);
4031 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4036 static struct md_sysfs_entry md_bitmap =
4037 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4040 size_show(struct mddev *mddev, char *page)
4042 return sprintf(page, "%llu\n",
4043 (unsigned long long)mddev->dev_sectors / 2);
4046 static int update_size(struct mddev *mddev, sector_t num_sectors);
4049 size_store(struct mddev *mddev, const char *buf, size_t len)
4051 /* If array is inactive, we can reduce the component size, but
4052 * not increase it (except from 0).
4053 * If array is active, we can try an on-line resize
4056 int err = strict_blocks_to_sectors(buf, §ors);
4061 err = update_size(mddev, sectors);
4062 md_update_sb(mddev, 1);
4064 if (mddev->dev_sectors == 0 ||
4065 mddev->dev_sectors > sectors)
4066 mddev->dev_sectors = sectors;
4070 return err ? err : len;
4073 static struct md_sysfs_entry md_size =
4074 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4076 /* Metadata version.
4078 * 'none' for arrays with no metadata (good luck...)
4079 * 'external' for arrays with externally managed metadata,
4080 * or N.M for internally known formats
4083 metadata_show(struct mddev *mddev, char *page)
4085 if (mddev->persistent)
4086 return sprintf(page, "%d.%d\n",
4087 mddev->major_version, mddev->minor_version);
4088 else if (mddev->external)
4089 return sprintf(page, "external:%s\n", mddev->metadata_type);
4091 return sprintf(page, "none\n");
4095 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4099 /* Changing the details of 'external' metadata is
4100 * always permitted. Otherwise there must be
4101 * no devices attached to the array.
4103 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4105 else if (!list_empty(&mddev->disks))
4108 if (cmd_match(buf, "none")) {
4109 mddev->persistent = 0;
4110 mddev->external = 0;
4111 mddev->major_version = 0;
4112 mddev->minor_version = 90;
4115 if (strncmp(buf, "external:", 9) == 0) {
4116 size_t namelen = len-9;
4117 if (namelen >= sizeof(mddev->metadata_type))
4118 namelen = sizeof(mddev->metadata_type)-1;
4119 strncpy(mddev->metadata_type, buf+9, namelen);
4120 mddev->metadata_type[namelen] = 0;
4121 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4122 mddev->metadata_type[--namelen] = 0;
4123 mddev->persistent = 0;
4124 mddev->external = 1;
4125 mddev->major_version = 0;
4126 mddev->minor_version = 90;
4129 major = simple_strtoul(buf, &e, 10);
4130 if (e==buf || *e != '.')
4133 minor = simple_strtoul(buf, &e, 10);
4134 if (e==buf || (*e && *e != '\n') )
4136 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4138 mddev->major_version = major;
4139 mddev->minor_version = minor;
4140 mddev->persistent = 1;
4141 mddev->external = 0;
4145 static struct md_sysfs_entry md_metadata =
4146 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4149 action_show(struct mddev *mddev, char *page)
4151 char *type = "idle";
4152 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4154 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4155 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4156 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4158 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4159 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4161 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4165 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4168 return sprintf(page, "%s\n", type);
4172 action_store(struct mddev *mddev, const char *page, size_t len)
4174 if (!mddev->pers || !mddev->pers->sync_request)
4177 if (cmd_match(page, "frozen"))
4178 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4180 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4182 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4183 if (mddev->sync_thread) {
4184 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4185 md_reap_sync_thread(mddev);
4187 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4188 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4190 else if (cmd_match(page, "resync"))
4191 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4192 else if (cmd_match(page, "recover")) {
4193 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4194 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4195 } else if (cmd_match(page, "reshape")) {
4197 if (mddev->pers->start_reshape == NULL)
4199 err = mddev->pers->start_reshape(mddev);
4202 sysfs_notify(&mddev->kobj, NULL, "degraded");
4204 if (cmd_match(page, "check"))
4205 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4206 else if (!cmd_match(page, "repair"))
4208 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4209 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4211 if (mddev->ro == 2) {
4212 /* A write to sync_action is enough to justify
4213 * canceling read-auto mode
4216 md_wakeup_thread(mddev->sync_thread);
4218 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4219 md_wakeup_thread(mddev->thread);
4220 sysfs_notify_dirent_safe(mddev->sysfs_action);
4224 static struct md_sysfs_entry md_scan_mode =
4225 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4228 last_sync_action_show(struct mddev *mddev, char *page)
4230 return sprintf(page, "%s\n", mddev->last_sync_action);
4233 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4236 mismatch_cnt_show(struct mddev *mddev, char *page)
4238 return sprintf(page, "%llu\n",
4239 (unsigned long long)
4240 atomic64_read(&mddev->resync_mismatches));
4243 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4246 sync_min_show(struct mddev *mddev, char *page)
4248 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4249 mddev->sync_speed_min ? "local": "system");
4253 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4257 if (strncmp(buf, "system", 6)==0) {
4258 mddev->sync_speed_min = 0;
4261 min = simple_strtoul(buf, &e, 10);
4262 if (buf == e || (*e && *e != '\n') || min <= 0)
4264 mddev->sync_speed_min = min;
4268 static struct md_sysfs_entry md_sync_min =
4269 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4272 sync_max_show(struct mddev *mddev, char *page)
4274 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4275 mddev->sync_speed_max ? "local": "system");
4279 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4283 if (strncmp(buf, "system", 6)==0) {
4284 mddev->sync_speed_max = 0;
4287 max = simple_strtoul(buf, &e, 10);
4288 if (buf == e || (*e && *e != '\n') || max <= 0)
4290 mddev->sync_speed_max = max;
4294 static struct md_sysfs_entry md_sync_max =
4295 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4298 degraded_show(struct mddev *mddev, char *page)
4300 return sprintf(page, "%d\n", mddev->degraded);
4302 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4305 sync_force_parallel_show(struct mddev *mddev, char *page)
4307 return sprintf(page, "%d\n", mddev->parallel_resync);
4311 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4315 if (kstrtol(buf, 10, &n))
4318 if (n != 0 && n != 1)
4321 mddev->parallel_resync = n;
4323 if (mddev->sync_thread)
4324 wake_up(&resync_wait);
4329 /* force parallel resync, even with shared block devices */
4330 static struct md_sysfs_entry md_sync_force_parallel =
4331 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4332 sync_force_parallel_show, sync_force_parallel_store);
4335 sync_speed_show(struct mddev *mddev, char *page)
4337 unsigned long resync, dt, db;
4338 if (mddev->curr_resync == 0)
4339 return sprintf(page, "none\n");
4340 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4341 dt = (jiffies - mddev->resync_mark) / HZ;
4343 db = resync - mddev->resync_mark_cnt;
4344 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4347 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4350 sync_completed_show(struct mddev *mddev, char *page)
4352 unsigned long long max_sectors, resync;
4354 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4355 return sprintf(page, "none\n");
4357 if (mddev->curr_resync == 1 ||
4358 mddev->curr_resync == 2)
4359 return sprintf(page, "delayed\n");
4361 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4362 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4363 max_sectors = mddev->resync_max_sectors;
4365 max_sectors = mddev->dev_sectors;
4367 resync = mddev->curr_resync_completed;
4368 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4371 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4374 min_sync_show(struct mddev *mddev, char *page)
4376 return sprintf(page, "%llu\n",
4377 (unsigned long long)mddev->resync_min);
4380 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4382 unsigned long long min;
4383 if (kstrtoull(buf, 10, &min))
4385 if (min > mddev->resync_max)
4387 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4390 /* Must be a multiple of chunk_size */
4391 if (mddev->chunk_sectors) {
4392 sector_t temp = min;
4393 if (sector_div(temp, mddev->chunk_sectors))
4396 mddev->resync_min = min;
4401 static struct md_sysfs_entry md_min_sync =
4402 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4405 max_sync_show(struct mddev *mddev, char *page)
4407 if (mddev->resync_max == MaxSector)
4408 return sprintf(page, "max\n");
4410 return sprintf(page, "%llu\n",
4411 (unsigned long long)mddev->resync_max);
4414 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4416 if (strncmp(buf, "max", 3) == 0)
4417 mddev->resync_max = MaxSector;
4419 unsigned long long max;
4420 if (kstrtoull(buf, 10, &max))
4422 if (max < mddev->resync_min)
4424 if (max < mddev->resync_max &&
4426 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4429 /* Must be a multiple of chunk_size */
4430 if (mddev->chunk_sectors) {
4431 sector_t temp = max;
4432 if (sector_div(temp, mddev->chunk_sectors))
4435 mddev->resync_max = max;
4437 wake_up(&mddev->recovery_wait);
4441 static struct md_sysfs_entry md_max_sync =
4442 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4445 suspend_lo_show(struct mddev *mddev, char *page)
4447 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4451 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4454 unsigned long long new = simple_strtoull(buf, &e, 10);
4455 unsigned long long old = mddev->suspend_lo;
4457 if (mddev->pers == NULL ||
4458 mddev->pers->quiesce == NULL)
4460 if (buf == e || (*e && *e != '\n'))
4463 mddev->suspend_lo = new;
4465 /* Shrinking suspended region */
4466 mddev->pers->quiesce(mddev, 2);
4468 /* Expanding suspended region - need to wait */
4469 mddev->pers->quiesce(mddev, 1);
4470 mddev->pers->quiesce(mddev, 0);
4474 static struct md_sysfs_entry md_suspend_lo =
4475 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4478 suspend_hi_show(struct mddev *mddev, char *page)
4480 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4484 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4487 unsigned long long new = simple_strtoull(buf, &e, 10);
4488 unsigned long long old = mddev->suspend_hi;
4490 if (mddev->pers == NULL ||
4491 mddev->pers->quiesce == NULL)
4493 if (buf == e || (*e && *e != '\n'))
4496 mddev->suspend_hi = new;
4498 /* Shrinking suspended region */
4499 mddev->pers->quiesce(mddev, 2);
4501 /* Expanding suspended region - need to wait */
4502 mddev->pers->quiesce(mddev, 1);
4503 mddev->pers->quiesce(mddev, 0);
4507 static struct md_sysfs_entry md_suspend_hi =
4508 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4511 reshape_position_show(struct mddev *mddev, char *page)
4513 if (mddev->reshape_position != MaxSector)
4514 return sprintf(page, "%llu\n",
4515 (unsigned long long)mddev->reshape_position);
4516 strcpy(page, "none\n");
4521 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4523 struct md_rdev *rdev;
4525 unsigned long long new = simple_strtoull(buf, &e, 10);
4528 if (buf == e || (*e && *e != '\n'))
4530 mddev->reshape_position = new;
4531 mddev->delta_disks = 0;
4532 mddev->reshape_backwards = 0;
4533 mddev->new_level = mddev->level;
4534 mddev->new_layout = mddev->layout;
4535 mddev->new_chunk_sectors = mddev->chunk_sectors;
4536 rdev_for_each(rdev, mddev)
4537 rdev->new_data_offset = rdev->data_offset;
4541 static struct md_sysfs_entry md_reshape_position =
4542 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4543 reshape_position_store);
4546 reshape_direction_show(struct mddev *mddev, char *page)
4548 return sprintf(page, "%s\n",
4549 mddev->reshape_backwards ? "backwards" : "forwards");
4553 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4556 if (cmd_match(buf, "forwards"))
4558 else if (cmd_match(buf, "backwards"))
4562 if (mddev->reshape_backwards == backwards)
4565 /* check if we are allowed to change */
4566 if (mddev->delta_disks)
4569 if (mddev->persistent &&
4570 mddev->major_version == 0)
4573 mddev->reshape_backwards = backwards;
4577 static struct md_sysfs_entry md_reshape_direction =
4578 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4579 reshape_direction_store);
4582 array_size_show(struct mddev *mddev, char *page)
4584 if (mddev->external_size)
4585 return sprintf(page, "%llu\n",
4586 (unsigned long long)mddev->array_sectors/2);
4588 return sprintf(page, "default\n");
4592 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4596 if (strncmp(buf, "default", 7) == 0) {
4598 sectors = mddev->pers->size(mddev, 0, 0);
4600 sectors = mddev->array_sectors;
4602 mddev->external_size = 0;
4604 if (strict_blocks_to_sectors(buf, §ors) < 0)
4606 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4609 mddev->external_size = 1;
4612 mddev->array_sectors = sectors;
4614 set_capacity(mddev->gendisk, mddev->array_sectors);
4615 revalidate_disk(mddev->gendisk);
4620 static struct md_sysfs_entry md_array_size =
4621 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4624 static struct attribute *md_default_attrs[] = {
4627 &md_raid_disks.attr,
4628 &md_chunk_size.attr,
4630 &md_resync_start.attr,
4632 &md_new_device.attr,
4633 &md_safe_delay.attr,
4634 &md_array_state.attr,
4635 &md_reshape_position.attr,
4636 &md_reshape_direction.attr,
4637 &md_array_size.attr,
4638 &max_corr_read_errors.attr,
4642 static struct attribute *md_redundancy_attrs[] = {
4644 &md_last_scan_mode.attr,
4645 &md_mismatches.attr,
4648 &md_sync_speed.attr,
4649 &md_sync_force_parallel.attr,
4650 &md_sync_completed.attr,
4653 &md_suspend_lo.attr,
4654 &md_suspend_hi.attr,
4659 static struct attribute_group md_redundancy_group = {
4661 .attrs = md_redundancy_attrs,
4665 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4667 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4668 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4673 spin_lock(&all_mddevs_lock);
4674 if (list_empty(&mddev->all_mddevs)) {
4675 spin_unlock(&all_mddevs_lock);
4679 spin_unlock(&all_mddevs_lock);
4681 rv = mddev_lock(mddev);
4683 rv = entry->show(mddev, page);
4684 mddev_unlock(mddev);
4691 md_attr_store(struct kobject *kobj, struct attribute *attr,
4692 const char *page, size_t length)
4694 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4695 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4700 if (!capable(CAP_SYS_ADMIN))
4702 spin_lock(&all_mddevs_lock);
4703 if (list_empty(&mddev->all_mddevs)) {
4704 spin_unlock(&all_mddevs_lock);
4708 spin_unlock(&all_mddevs_lock);
4709 if (entry->store == new_dev_store)
4710 flush_workqueue(md_misc_wq);
4711 rv = mddev_lock(mddev);
4713 rv = entry->store(mddev, page, length);
4714 mddev_unlock(mddev);
4720 static void md_free(struct kobject *ko)
4722 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4724 if (mddev->sysfs_state)
4725 sysfs_put(mddev->sysfs_state);
4727 if (mddev->gendisk) {
4728 del_gendisk(mddev->gendisk);
4729 put_disk(mddev->gendisk);
4732 blk_cleanup_queue(mddev->queue);
4737 static const struct sysfs_ops md_sysfs_ops = {
4738 .show = md_attr_show,
4739 .store = md_attr_store,
4741 static struct kobj_type md_ktype = {
4743 .sysfs_ops = &md_sysfs_ops,
4744 .default_attrs = md_default_attrs,
4749 static void mddev_delayed_delete(struct work_struct *ws)
4751 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4753 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4754 kobject_del(&mddev->kobj);
4755 kobject_put(&mddev->kobj);
4758 static int md_alloc(dev_t dev, char *name)
4760 static DEFINE_MUTEX(disks_mutex);
4761 struct mddev *mddev = mddev_find(dev);
4762 struct gendisk *disk;
4771 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4772 shift = partitioned ? MdpMinorShift : 0;
4773 unit = MINOR(mddev->unit) >> shift;
4775 /* wait for any previous instance of this device to be
4776 * completely removed (mddev_delayed_delete).
4778 flush_workqueue(md_misc_wq);
4780 mutex_lock(&disks_mutex);
4786 /* Need to ensure that 'name' is not a duplicate.
4788 struct mddev *mddev2;
4789 spin_lock(&all_mddevs_lock);
4791 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4792 if (mddev2->gendisk &&
4793 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4794 spin_unlock(&all_mddevs_lock);
4797 spin_unlock(&all_mddevs_lock);
4801 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4804 mddev->queue->queuedata = mddev;
4806 blk_queue_make_request(mddev->queue, md_make_request);
4807 blk_set_stacking_limits(&mddev->queue->limits);
4809 disk = alloc_disk(1 << shift);
4811 blk_cleanup_queue(mddev->queue);
4812 mddev->queue = NULL;
4815 disk->major = MAJOR(mddev->unit);
4816 disk->first_minor = unit << shift;
4818 strcpy(disk->disk_name, name);
4819 else if (partitioned)
4820 sprintf(disk->disk_name, "md_d%d", unit);
4822 sprintf(disk->disk_name, "md%d", unit);
4823 disk->fops = &md_fops;
4824 disk->private_data = mddev;
4825 disk->queue = mddev->queue;
4826 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4827 /* Allow extended partitions. This makes the
4828 * 'mdp' device redundant, but we can't really
4831 disk->flags |= GENHD_FL_EXT_DEVT;
4832 mddev->gendisk = disk;
4833 /* As soon as we call add_disk(), another thread could get
4834 * through to md_open, so make sure it doesn't get too far
4836 mutex_lock(&mddev->open_mutex);
4839 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4840 &disk_to_dev(disk)->kobj, "%s", "md");
4842 /* This isn't possible, but as kobject_init_and_add is marked
4843 * __must_check, we must do something with the result
4845 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4849 if (mddev->kobj.sd &&
4850 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4851 printk(KERN_DEBUG "pointless warning\n");
4852 mutex_unlock(&mddev->open_mutex);
4854 mutex_unlock(&disks_mutex);
4855 if (!error && mddev->kobj.sd) {
4856 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4857 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4863 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4865 md_alloc(dev, NULL);
4869 static int add_named_array(const char *val, struct kernel_param *kp)
4871 /* val must be "md_*" where * is not all digits.
4872 * We allocate an array with a large free minor number, and
4873 * set the name to val. val must not already be an active name.
4875 int len = strlen(val);
4876 char buf[DISK_NAME_LEN];
4878 while (len && val[len-1] == '\n')
4880 if (len >= DISK_NAME_LEN)
4882 strlcpy(buf, val, len+1);
4883 if (strncmp(buf, "md_", 3) != 0)
4885 return md_alloc(0, buf);
4888 static void md_safemode_timeout(unsigned long data)
4890 struct mddev *mddev = (struct mddev *) data;
4892 if (!atomic_read(&mddev->writes_pending)) {
4893 mddev->safemode = 1;
4894 if (mddev->external)
4895 sysfs_notify_dirent_safe(mddev->sysfs_state);
4897 md_wakeup_thread(mddev->thread);
4900 static int start_dirty_degraded;
4902 int md_run(struct mddev *mddev)
4905 struct md_rdev *rdev;
4906 struct md_personality *pers;
4908 if (list_empty(&mddev->disks))
4909 /* cannot run an array with no devices.. */
4914 /* Cannot run until previous stop completes properly */
4915 if (mddev->sysfs_active)
4919 * Analyze all RAID superblock(s)
4921 if (!mddev->raid_disks) {
4922 if (!mddev->persistent)
4927 if (mddev->level != LEVEL_NONE)
4928 request_module("md-level-%d", mddev->level);
4929 else if (mddev->clevel[0])
4930 request_module("md-%s", mddev->clevel);
4933 * Drop all container device buffers, from now on
4934 * the only valid external interface is through the md
4937 rdev_for_each(rdev, mddev) {
4938 if (test_bit(Faulty, &rdev->flags))
4940 sync_blockdev(rdev->bdev);
4941 invalidate_bdev(rdev->bdev);
4943 /* perform some consistency tests on the device.
4944 * We don't want the data to overlap the metadata,
4945 * Internal Bitmap issues have been handled elsewhere.
4947 if (rdev->meta_bdev) {
4948 /* Nothing to check */;
4949 } else if (rdev->data_offset < rdev->sb_start) {
4950 if (mddev->dev_sectors &&
4951 rdev->data_offset + mddev->dev_sectors
4953 printk("md: %s: data overlaps metadata\n",
4958 if (rdev->sb_start + rdev->sb_size/512
4959 > rdev->data_offset) {
4960 printk("md: %s: metadata overlaps data\n",
4965 sysfs_notify_dirent_safe(rdev->sysfs_state);
4968 if (mddev->bio_set == NULL)
4969 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4971 spin_lock(&pers_lock);
4972 pers = find_pers(mddev->level, mddev->clevel);
4973 if (!pers || !try_module_get(pers->owner)) {
4974 spin_unlock(&pers_lock);
4975 if (mddev->level != LEVEL_NONE)
4976 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4979 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4984 spin_unlock(&pers_lock);
4985 if (mddev->level != pers->level) {
4986 mddev->level = pers->level;
4987 mddev->new_level = pers->level;
4989 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4991 if (mddev->reshape_position != MaxSector &&
4992 pers->start_reshape == NULL) {
4993 /* This personality cannot handle reshaping... */
4995 module_put(pers->owner);
4999 if (pers->sync_request) {
5000 /* Warn if this is a potentially silly
5003 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5004 struct md_rdev *rdev2;
5007 rdev_for_each(rdev, mddev)
5008 rdev_for_each(rdev2, mddev) {
5010 rdev->bdev->bd_contains ==
5011 rdev2->bdev->bd_contains) {
5013 "%s: WARNING: %s appears to be"
5014 " on the same physical disk as"
5017 bdevname(rdev->bdev,b),
5018 bdevname(rdev2->bdev,b2));
5025 "True protection against single-disk"
5026 " failure might be compromised.\n");
5029 mddev->recovery = 0;
5030 /* may be over-ridden by personality */
5031 mddev->resync_max_sectors = mddev->dev_sectors;
5033 mddev->ok_start_degraded = start_dirty_degraded;
5035 if (start_readonly && mddev->ro == 0)
5036 mddev->ro = 2; /* read-only, but switch on first write */
5038 err = mddev->pers->run(mddev);
5040 printk(KERN_ERR "md: pers->run() failed ...\n");
5041 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5042 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5043 " but 'external_size' not in effect?\n", __func__);
5045 "md: invalid array_size %llu > default size %llu\n",
5046 (unsigned long long)mddev->array_sectors / 2,
5047 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5049 mddev->pers->stop(mddev);
5051 if (err == 0 && mddev->pers->sync_request &&
5052 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5053 err = bitmap_create(mddev);
5055 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5056 mdname(mddev), err);
5057 mddev->pers->stop(mddev);
5061 module_put(mddev->pers->owner);
5063 bitmap_destroy(mddev);
5066 if (mddev->pers->sync_request) {
5067 if (mddev->kobj.sd &&
5068 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5070 "md: cannot register extra attributes for %s\n",
5072 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5073 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5076 atomic_set(&mddev->writes_pending,0);
5077 atomic_set(&mddev->max_corr_read_errors,
5078 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5079 mddev->safemode = 0;
5080 mddev->safemode_timer.function = md_safemode_timeout;
5081 mddev->safemode_timer.data = (unsigned long) mddev;
5082 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5086 rdev_for_each(rdev, mddev)
5087 if (rdev->raid_disk >= 0)
5088 if (sysfs_link_rdev(mddev, rdev))
5089 /* failure here is OK */;
5091 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5093 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5094 md_update_sb(mddev, 0);
5096 md_new_event(mddev);
5097 sysfs_notify_dirent_safe(mddev->sysfs_state);
5098 sysfs_notify_dirent_safe(mddev->sysfs_action);
5099 sysfs_notify(&mddev->kobj, NULL, "degraded");
5102 EXPORT_SYMBOL_GPL(md_run);
5104 static int do_md_run(struct mddev *mddev)
5108 err = md_run(mddev);
5111 err = bitmap_load(mddev);
5113 bitmap_destroy(mddev);
5117 md_wakeup_thread(mddev->thread);
5118 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5120 set_capacity(mddev->gendisk, mddev->array_sectors);
5121 revalidate_disk(mddev->gendisk);
5123 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5128 static int restart_array(struct mddev *mddev)
5130 struct gendisk *disk = mddev->gendisk;
5132 /* Complain if it has no devices */
5133 if (list_empty(&mddev->disks))
5139 mddev->safemode = 0;
5141 set_disk_ro(disk, 0);
5142 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5144 /* Kick recovery or resync if necessary */
5145 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5146 md_wakeup_thread(mddev->thread);
5147 md_wakeup_thread(mddev->sync_thread);
5148 sysfs_notify_dirent_safe(mddev->sysfs_state);
5152 static void md_clean(struct mddev *mddev)
5154 mddev->array_sectors = 0;
5155 mddev->external_size = 0;
5156 mddev->dev_sectors = 0;
5157 mddev->raid_disks = 0;
5158 mddev->recovery_cp = 0;
5159 mddev->resync_min = 0;
5160 mddev->resync_max = MaxSector;
5161 mddev->reshape_position = MaxSector;
5162 mddev->external = 0;
5163 mddev->persistent = 0;
5164 mddev->level = LEVEL_NONE;
5165 mddev->clevel[0] = 0;
5168 mddev->metadata_type[0] = 0;
5169 mddev->chunk_sectors = 0;
5170 mddev->ctime = mddev->utime = 0;
5172 mddev->max_disks = 0;
5174 mddev->can_decrease_events = 0;
5175 mddev->delta_disks = 0;
5176 mddev->reshape_backwards = 0;
5177 mddev->new_level = LEVEL_NONE;
5178 mddev->new_layout = 0;
5179 mddev->new_chunk_sectors = 0;
5180 mddev->curr_resync = 0;
5181 atomic64_set(&mddev->resync_mismatches, 0);
5182 mddev->suspend_lo = mddev->suspend_hi = 0;
5183 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5184 mddev->recovery = 0;
5187 mddev->degraded = 0;
5188 mddev->safemode = 0;
5189 mddev->merge_check_needed = 0;
5190 mddev->bitmap_info.offset = 0;
5191 mddev->bitmap_info.default_offset = 0;
5192 mddev->bitmap_info.default_space = 0;
5193 mddev->bitmap_info.chunksize = 0;
5194 mddev->bitmap_info.daemon_sleep = 0;
5195 mddev->bitmap_info.max_write_behind = 0;
5198 static void __md_stop_writes(struct mddev *mddev)
5200 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5201 if (mddev->sync_thread) {
5202 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5203 md_reap_sync_thread(mddev);
5206 del_timer_sync(&mddev->safemode_timer);
5208 bitmap_flush(mddev);
5209 md_super_wait(mddev);
5211 if (mddev->ro == 0 &&
5212 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5213 /* mark array as shutdown cleanly */
5215 md_update_sb(mddev, 1);
5219 void md_stop_writes(struct mddev *mddev)
5221 mddev_lock_nointr(mddev);
5222 __md_stop_writes(mddev);
5223 mddev_unlock(mddev);
5225 EXPORT_SYMBOL_GPL(md_stop_writes);
5227 static void __md_stop(struct mddev *mddev)
5230 mddev->pers->stop(mddev);
5231 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5232 mddev->to_remove = &md_redundancy_group;
5233 module_put(mddev->pers->owner);
5235 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5238 void md_stop(struct mddev *mddev)
5240 /* stop the array and free an attached data structures.
5241 * This is called from dm-raid
5244 bitmap_destroy(mddev);
5246 bioset_free(mddev->bio_set);
5249 EXPORT_SYMBOL_GPL(md_stop);
5251 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5256 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5258 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5259 md_wakeup_thread(mddev->thread);
5261 if (mddev->sync_thread) {
5262 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5263 /* Thread might be blocked waiting for metadata update
5264 * which will now never happen */
5265 wake_up_process(mddev->sync_thread->tsk);
5267 mddev_unlock(mddev);
5268 wait_event(resync_wait, mddev->sync_thread == NULL);
5269 mddev_lock_nointr(mddev);
5271 mutex_lock(&mddev->open_mutex);
5272 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5273 mddev->sync_thread ||
5274 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5275 printk("md: %s still in use.\n",mdname(mddev));
5277 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5278 md_wakeup_thread(mddev->thread);
5284 __md_stop_writes(mddev);
5290 set_disk_ro(mddev->gendisk, 1);
5291 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5292 sysfs_notify_dirent_safe(mddev->sysfs_state);
5296 mutex_unlock(&mddev->open_mutex);
5301 * 0 - completely stop and dis-assemble array
5302 * 2 - stop but do not disassemble array
5304 static int do_md_stop(struct mddev *mddev, int mode,
5305 struct block_device *bdev)
5307 struct gendisk *disk = mddev->gendisk;
5308 struct md_rdev *rdev;
5311 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5313 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5314 md_wakeup_thread(mddev->thread);
5316 if (mddev->sync_thread) {
5317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5318 /* Thread might be blocked waiting for metadata update
5319 * which will now never happen */
5320 wake_up_process(mddev->sync_thread->tsk);
5322 mddev_unlock(mddev);
5323 wait_event(resync_wait, mddev->sync_thread == NULL);
5324 mddev_lock_nointr(mddev);
5326 mutex_lock(&mddev->open_mutex);
5327 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5328 mddev->sysfs_active ||
5329 mddev->sync_thread ||
5330 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5331 printk("md: %s still in use.\n",mdname(mddev));
5332 mutex_unlock(&mddev->open_mutex);
5334 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5335 md_wakeup_thread(mddev->thread);
5341 set_disk_ro(disk, 0);
5343 __md_stop_writes(mddev);
5345 mddev->queue->merge_bvec_fn = NULL;
5346 mddev->queue->backing_dev_info.congested_fn = NULL;
5348 /* tell userspace to handle 'inactive' */
5349 sysfs_notify_dirent_safe(mddev->sysfs_state);
5351 rdev_for_each(rdev, mddev)
5352 if (rdev->raid_disk >= 0)
5353 sysfs_unlink_rdev(mddev, rdev);
5355 set_capacity(disk, 0);
5356 mutex_unlock(&mddev->open_mutex);
5358 revalidate_disk(disk);
5363 mutex_unlock(&mddev->open_mutex);
5365 * Free resources if final stop
5368 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5370 bitmap_destroy(mddev);
5371 if (mddev->bitmap_info.file) {
5372 fput(mddev->bitmap_info.file);
5373 mddev->bitmap_info.file = NULL;
5375 mddev->bitmap_info.offset = 0;
5377 export_array(mddev);
5380 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5381 if (mddev->hold_active == UNTIL_STOP)
5382 mddev->hold_active = 0;
5384 blk_integrity_unregister(disk);
5385 md_new_event(mddev);
5386 sysfs_notify_dirent_safe(mddev->sysfs_state);
5391 static void autorun_array(struct mddev *mddev)
5393 struct md_rdev *rdev;
5396 if (list_empty(&mddev->disks))
5399 printk(KERN_INFO "md: running: ");
5401 rdev_for_each(rdev, mddev) {
5402 char b[BDEVNAME_SIZE];
5403 printk("<%s>", bdevname(rdev->bdev,b));
5407 err = do_md_run(mddev);
5409 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5410 do_md_stop(mddev, 0, NULL);
5415 * lets try to run arrays based on all disks that have arrived
5416 * until now. (those are in pending_raid_disks)
5418 * the method: pick the first pending disk, collect all disks with
5419 * the same UUID, remove all from the pending list and put them into
5420 * the 'same_array' list. Then order this list based on superblock
5421 * update time (freshest comes first), kick out 'old' disks and
5422 * compare superblocks. If everything's fine then run it.
5424 * If "unit" is allocated, then bump its reference count
5426 static void autorun_devices(int part)
5428 struct md_rdev *rdev0, *rdev, *tmp;
5429 struct mddev *mddev;
5430 char b[BDEVNAME_SIZE];
5432 printk(KERN_INFO "md: autorun ...\n");
5433 while (!list_empty(&pending_raid_disks)) {
5436 LIST_HEAD(candidates);
5437 rdev0 = list_entry(pending_raid_disks.next,
5438 struct md_rdev, same_set);
5440 printk(KERN_INFO "md: considering %s ...\n",
5441 bdevname(rdev0->bdev,b));
5442 INIT_LIST_HEAD(&candidates);
5443 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5444 if (super_90_load(rdev, rdev0, 0) >= 0) {
5445 printk(KERN_INFO "md: adding %s ...\n",
5446 bdevname(rdev->bdev,b));
5447 list_move(&rdev->same_set, &candidates);
5450 * now we have a set of devices, with all of them having
5451 * mostly sane superblocks. It's time to allocate the
5455 dev = MKDEV(mdp_major,
5456 rdev0->preferred_minor << MdpMinorShift);
5457 unit = MINOR(dev) >> MdpMinorShift;
5459 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5462 if (rdev0->preferred_minor != unit) {
5463 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5464 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5468 md_probe(dev, NULL, NULL);
5469 mddev = mddev_find(dev);
5470 if (!mddev || !mddev->gendisk) {
5474 "md: cannot allocate memory for md drive.\n");
5477 if (mddev_lock(mddev))
5478 printk(KERN_WARNING "md: %s locked, cannot run\n",
5480 else if (mddev->raid_disks || mddev->major_version
5481 || !list_empty(&mddev->disks)) {
5483 "md: %s already running, cannot run %s\n",
5484 mdname(mddev), bdevname(rdev0->bdev,b));
5485 mddev_unlock(mddev);
5487 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5488 mddev->persistent = 1;
5489 rdev_for_each_list(rdev, tmp, &candidates) {
5490 list_del_init(&rdev->same_set);
5491 if (bind_rdev_to_array(rdev, mddev))
5494 autorun_array(mddev);
5495 mddev_unlock(mddev);
5497 /* on success, candidates will be empty, on error
5500 rdev_for_each_list(rdev, tmp, &candidates) {
5501 list_del_init(&rdev->same_set);
5506 printk(KERN_INFO "md: ... autorun DONE.\n");
5508 #endif /* !MODULE */
5510 static int get_version(void __user *arg)
5514 ver.major = MD_MAJOR_VERSION;
5515 ver.minor = MD_MINOR_VERSION;
5516 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5518 if (copy_to_user(arg, &ver, sizeof(ver)))
5524 static int get_array_info(struct mddev *mddev, void __user *arg)
5526 mdu_array_info_t info;
5527 int nr,working,insync,failed,spare;
5528 struct md_rdev *rdev;
5530 nr = working = insync = failed = spare = 0;
5532 rdev_for_each_rcu(rdev, mddev) {
5534 if (test_bit(Faulty, &rdev->flags))
5538 if (test_bit(In_sync, &rdev->flags))
5546 info.major_version = mddev->major_version;
5547 info.minor_version = mddev->minor_version;
5548 info.patch_version = MD_PATCHLEVEL_VERSION;
5549 info.ctime = mddev->ctime;
5550 info.level = mddev->level;
5551 info.size = mddev->dev_sectors / 2;
5552 if (info.size != mddev->dev_sectors / 2) /* overflow */
5555 info.raid_disks = mddev->raid_disks;
5556 info.md_minor = mddev->md_minor;
5557 info.not_persistent= !mddev->persistent;
5559 info.utime = mddev->utime;
5562 info.state = (1<<MD_SB_CLEAN);
5563 if (mddev->bitmap && mddev->bitmap_info.offset)
5564 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5565 info.active_disks = insync;
5566 info.working_disks = working;
5567 info.failed_disks = failed;
5568 info.spare_disks = spare;
5570 info.layout = mddev->layout;
5571 info.chunk_size = mddev->chunk_sectors << 9;
5573 if (copy_to_user(arg, &info, sizeof(info)))
5579 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5581 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5582 char *ptr, *buf = NULL;
5585 file = kmalloc(sizeof(*file), GFP_NOIO);
5590 /* bitmap disabled, zero the first byte and copy out */
5591 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5592 file->pathname[0] = '\0';
5596 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5600 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5601 buf, sizeof(file->pathname));
5605 strcpy(file->pathname, ptr);
5609 if (copy_to_user(arg, file, sizeof(*file)))
5617 static int get_disk_info(struct mddev *mddev, void __user * arg)
5619 mdu_disk_info_t info;
5620 struct md_rdev *rdev;
5622 if (copy_from_user(&info, arg, sizeof(info)))
5626 rdev = find_rdev_nr_rcu(mddev, info.number);
5628 info.major = MAJOR(rdev->bdev->bd_dev);
5629 info.minor = MINOR(rdev->bdev->bd_dev);
5630 info.raid_disk = rdev->raid_disk;
5632 if (test_bit(Faulty, &rdev->flags))
5633 info.state |= (1<<MD_DISK_FAULTY);
5634 else if (test_bit(In_sync, &rdev->flags)) {
5635 info.state |= (1<<MD_DISK_ACTIVE);
5636 info.state |= (1<<MD_DISK_SYNC);
5638 if (test_bit(WriteMostly, &rdev->flags))
5639 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5641 info.major = info.minor = 0;
5642 info.raid_disk = -1;
5643 info.state = (1<<MD_DISK_REMOVED);
5647 if (copy_to_user(arg, &info, sizeof(info)))
5653 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5655 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5656 struct md_rdev *rdev;
5657 dev_t dev = MKDEV(info->major,info->minor);
5659 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5662 if (!mddev->raid_disks) {
5664 /* expecting a device which has a superblock */
5665 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5668 "md: md_import_device returned %ld\n",
5670 return PTR_ERR(rdev);
5672 if (!list_empty(&mddev->disks)) {
5673 struct md_rdev *rdev0
5674 = list_entry(mddev->disks.next,
5675 struct md_rdev, same_set);
5676 err = super_types[mddev->major_version]
5677 .load_super(rdev, rdev0, mddev->minor_version);
5680 "md: %s has different UUID to %s\n",
5681 bdevname(rdev->bdev,b),
5682 bdevname(rdev0->bdev,b2));
5687 err = bind_rdev_to_array(rdev, mddev);
5694 * add_new_disk can be used once the array is assembled
5695 * to add "hot spares". They must already have a superblock
5700 if (!mddev->pers->hot_add_disk) {
5702 "%s: personality does not support diskops!\n",
5706 if (mddev->persistent)
5707 rdev = md_import_device(dev, mddev->major_version,
5708 mddev->minor_version);
5710 rdev = md_import_device(dev, -1, -1);
5713 "md: md_import_device returned %ld\n",
5715 return PTR_ERR(rdev);
5717 /* set saved_raid_disk if appropriate */
5718 if (!mddev->persistent) {
5719 if (info->state & (1<<MD_DISK_SYNC) &&
5720 info->raid_disk < mddev->raid_disks) {
5721 rdev->raid_disk = info->raid_disk;
5722 set_bit(In_sync, &rdev->flags);
5723 clear_bit(Bitmap_sync, &rdev->flags);
5725 rdev->raid_disk = -1;
5726 rdev->saved_raid_disk = rdev->raid_disk;
5728 super_types[mddev->major_version].
5729 validate_super(mddev, rdev);
5730 if ((info->state & (1<<MD_DISK_SYNC)) &&
5731 rdev->raid_disk != info->raid_disk) {
5732 /* This was a hot-add request, but events doesn't
5733 * match, so reject it.
5739 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5740 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5741 set_bit(WriteMostly, &rdev->flags);
5743 clear_bit(WriteMostly, &rdev->flags);
5745 rdev->raid_disk = -1;
5746 err = bind_rdev_to_array(rdev, mddev);
5747 if (!err && !mddev->pers->hot_remove_disk) {
5748 /* If there is hot_add_disk but no hot_remove_disk
5749 * then added disks for geometry changes,
5750 * and should be added immediately.
5752 super_types[mddev->major_version].
5753 validate_super(mddev, rdev);
5754 err = mddev->pers->hot_add_disk(mddev, rdev);
5756 unbind_rdev_from_array(rdev);
5761 sysfs_notify_dirent_safe(rdev->sysfs_state);
5763 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5764 if (mddev->degraded)
5765 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5766 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5768 md_new_event(mddev);
5769 md_wakeup_thread(mddev->thread);
5773 /* otherwise, add_new_disk is only allowed
5774 * for major_version==0 superblocks
5776 if (mddev->major_version != 0) {
5777 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5782 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5784 rdev = md_import_device(dev, -1, 0);
5787 "md: error, md_import_device() returned %ld\n",
5789 return PTR_ERR(rdev);
5791 rdev->desc_nr = info->number;
5792 if (info->raid_disk < mddev->raid_disks)
5793 rdev->raid_disk = info->raid_disk;
5795 rdev->raid_disk = -1;
5797 if (rdev->raid_disk < mddev->raid_disks)
5798 if (info->state & (1<<MD_DISK_SYNC))
5799 set_bit(In_sync, &rdev->flags);
5801 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5802 set_bit(WriteMostly, &rdev->flags);
5804 if (!mddev->persistent) {
5805 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5806 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5808 rdev->sb_start = calc_dev_sboffset(rdev);
5809 rdev->sectors = rdev->sb_start;
5811 err = bind_rdev_to_array(rdev, mddev);
5821 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5823 char b[BDEVNAME_SIZE];
5824 struct md_rdev *rdev;
5826 rdev = find_rdev(mddev, dev);
5830 clear_bit(Blocked, &rdev->flags);
5831 remove_and_add_spares(mddev, rdev);
5833 if (rdev->raid_disk >= 0)
5836 kick_rdev_from_array(rdev);
5837 md_update_sb(mddev, 1);
5838 md_new_event(mddev);
5842 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5843 bdevname(rdev->bdev,b), mdname(mddev));
5847 static int hot_add_disk(struct mddev *mddev, dev_t dev)
5849 char b[BDEVNAME_SIZE];
5851 struct md_rdev *rdev;
5856 if (mddev->major_version != 0) {
5857 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5858 " version-0 superblocks.\n",
5862 if (!mddev->pers->hot_add_disk) {
5864 "%s: personality does not support diskops!\n",
5869 rdev = md_import_device(dev, -1, 0);
5872 "md: error, md_import_device() returned %ld\n",
5877 if (mddev->persistent)
5878 rdev->sb_start = calc_dev_sboffset(rdev);
5880 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5882 rdev->sectors = rdev->sb_start;
5884 if (test_bit(Faulty, &rdev->flags)) {
5886 "md: can not hot-add faulty %s disk to %s!\n",
5887 bdevname(rdev->bdev,b), mdname(mddev));
5891 clear_bit(In_sync, &rdev->flags);
5893 rdev->saved_raid_disk = -1;
5894 err = bind_rdev_to_array(rdev, mddev);
5899 * The rest should better be atomic, we can have disk failures
5900 * noticed in interrupt contexts ...
5903 rdev->raid_disk = -1;
5905 md_update_sb(mddev, 1);
5908 * Kick recovery, maybe this spare has to be added to the
5909 * array immediately.
5911 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5912 md_wakeup_thread(mddev->thread);
5913 md_new_event(mddev);
5921 static int set_bitmap_file(struct mddev *mddev, int fd)
5926 if (!mddev->pers->quiesce || !mddev->thread)
5928 if (mddev->recovery || mddev->sync_thread)
5930 /* we should be able to change the bitmap.. */
5934 struct inode *inode;
5936 return -EEXIST; /* cannot add when bitmap is present */
5937 mddev->bitmap_info.file = fget(fd);
5939 if (mddev->bitmap_info.file == NULL) {
5940 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5945 inode = mddev->bitmap_info.file->f_mapping->host;
5946 if (!S_ISREG(inode->i_mode)) {
5947 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
5950 } else if (!(mddev->bitmap_info.file->f_mode & FMODE_WRITE)) {
5951 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
5954 } else if (atomic_read(&inode->i_writecount) != 1) {
5955 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5960 fput(mddev->bitmap_info.file);
5961 mddev->bitmap_info.file = NULL;
5964 mddev->bitmap_info.offset = 0; /* file overrides offset */
5965 } else if (mddev->bitmap == NULL)
5966 return -ENOENT; /* cannot remove what isn't there */
5969 mddev->pers->quiesce(mddev, 1);
5971 err = bitmap_create(mddev);
5973 err = bitmap_load(mddev);
5975 if (fd < 0 || err) {
5976 bitmap_destroy(mddev);
5977 fd = -1; /* make sure to put the file */
5979 mddev->pers->quiesce(mddev, 0);
5982 if (mddev->bitmap_info.file)
5983 fput(mddev->bitmap_info.file);
5984 mddev->bitmap_info.file = NULL;
5991 * set_array_info is used two different ways
5992 * The original usage is when creating a new array.
5993 * In this usage, raid_disks is > 0 and it together with
5994 * level, size, not_persistent,layout,chunksize determine the
5995 * shape of the array.
5996 * This will always create an array with a type-0.90.0 superblock.
5997 * The newer usage is when assembling an array.
5998 * In this case raid_disks will be 0, and the major_version field is
5999 * use to determine which style super-blocks are to be found on the devices.
6000 * The minor and patch _version numbers are also kept incase the
6001 * super_block handler wishes to interpret them.
6003 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6006 if (info->raid_disks == 0) {
6007 /* just setting version number for superblock loading */
6008 if (info->major_version < 0 ||
6009 info->major_version >= ARRAY_SIZE(super_types) ||
6010 super_types[info->major_version].name == NULL) {
6011 /* maybe try to auto-load a module? */
6013 "md: superblock version %d not known\n",
6014 info->major_version);
6017 mddev->major_version = info->major_version;
6018 mddev->minor_version = info->minor_version;
6019 mddev->patch_version = info->patch_version;
6020 mddev->persistent = !info->not_persistent;
6021 /* ensure mddev_put doesn't delete this now that there
6022 * is some minimal configuration.
6024 mddev->ctime = get_seconds();
6027 mddev->major_version = MD_MAJOR_VERSION;
6028 mddev->minor_version = MD_MINOR_VERSION;
6029 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6030 mddev->ctime = get_seconds();
6032 mddev->level = info->level;
6033 mddev->clevel[0] = 0;
6034 mddev->dev_sectors = 2 * (sector_t)info->size;
6035 mddev->raid_disks = info->raid_disks;
6036 /* don't set md_minor, it is determined by which /dev/md* was
6039 if (info->state & (1<<MD_SB_CLEAN))
6040 mddev->recovery_cp = MaxSector;
6042 mddev->recovery_cp = 0;
6043 mddev->persistent = ! info->not_persistent;
6044 mddev->external = 0;
6046 mddev->layout = info->layout;
6047 mddev->chunk_sectors = info->chunk_size >> 9;
6049 mddev->max_disks = MD_SB_DISKS;
6051 if (mddev->persistent)
6053 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6055 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6056 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6057 mddev->bitmap_info.offset = 0;
6059 mddev->reshape_position = MaxSector;
6062 * Generate a 128 bit UUID
6064 get_random_bytes(mddev->uuid, 16);
6066 mddev->new_level = mddev->level;
6067 mddev->new_chunk_sectors = mddev->chunk_sectors;
6068 mddev->new_layout = mddev->layout;
6069 mddev->delta_disks = 0;
6070 mddev->reshape_backwards = 0;
6075 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6077 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6079 if (mddev->external_size)
6082 mddev->array_sectors = array_sectors;
6084 EXPORT_SYMBOL(md_set_array_sectors);
6086 static int update_size(struct mddev *mddev, sector_t num_sectors)
6088 struct md_rdev *rdev;
6090 int fit = (num_sectors == 0);
6092 if (mddev->pers->resize == NULL)
6094 /* The "num_sectors" is the number of sectors of each device that
6095 * is used. This can only make sense for arrays with redundancy.
6096 * linear and raid0 always use whatever space is available. We can only
6097 * consider changing this number if no resync or reconstruction is
6098 * happening, and if the new size is acceptable. It must fit before the
6099 * sb_start or, if that is <data_offset, it must fit before the size
6100 * of each device. If num_sectors is zero, we find the largest size
6103 if (mddev->sync_thread)
6108 rdev_for_each(rdev, mddev) {
6109 sector_t avail = rdev->sectors;
6111 if (fit && (num_sectors == 0 || num_sectors > avail))
6112 num_sectors = avail;
6113 if (avail < num_sectors)
6116 rv = mddev->pers->resize(mddev, num_sectors);
6118 revalidate_disk(mddev->gendisk);
6122 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6125 struct md_rdev *rdev;
6126 /* change the number of raid disks */
6127 if (mddev->pers->check_reshape == NULL)
6131 if (raid_disks <= 0 ||
6132 (mddev->max_disks && raid_disks >= mddev->max_disks))
6134 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6137 rdev_for_each(rdev, mddev) {
6138 if (mddev->raid_disks < raid_disks &&
6139 rdev->data_offset < rdev->new_data_offset)
6141 if (mddev->raid_disks > raid_disks &&
6142 rdev->data_offset > rdev->new_data_offset)
6146 mddev->delta_disks = raid_disks - mddev->raid_disks;
6147 if (mddev->delta_disks < 0)
6148 mddev->reshape_backwards = 1;
6149 else if (mddev->delta_disks > 0)
6150 mddev->reshape_backwards = 0;
6152 rv = mddev->pers->check_reshape(mddev);
6154 mddev->delta_disks = 0;
6155 mddev->reshape_backwards = 0;
6161 * update_array_info is used to change the configuration of an
6163 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6164 * fields in the info are checked against the array.
6165 * Any differences that cannot be handled will cause an error.
6166 * Normally, only one change can be managed at a time.
6168 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6174 /* calculate expected state,ignoring low bits */
6175 if (mddev->bitmap && mddev->bitmap_info.offset)
6176 state |= (1 << MD_SB_BITMAP_PRESENT);
6178 if (mddev->major_version != info->major_version ||
6179 mddev->minor_version != info->minor_version ||
6180 /* mddev->patch_version != info->patch_version || */
6181 mddev->ctime != info->ctime ||
6182 mddev->level != info->level ||
6183 /* mddev->layout != info->layout || */
6184 !mddev->persistent != info->not_persistent||
6185 mddev->chunk_sectors != info->chunk_size >> 9 ||
6186 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6187 ((state^info->state) & 0xfffffe00)
6190 /* Check there is only one change */
6191 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6193 if (mddev->raid_disks != info->raid_disks)
6195 if (mddev->layout != info->layout)
6197 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6204 if (mddev->layout != info->layout) {
6206 * we don't need to do anything at the md level, the
6207 * personality will take care of it all.
6209 if (mddev->pers->check_reshape == NULL)
6212 mddev->new_layout = info->layout;
6213 rv = mddev->pers->check_reshape(mddev);
6215 mddev->new_layout = mddev->layout;
6219 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6220 rv = update_size(mddev, (sector_t)info->size * 2);
6222 if (mddev->raid_disks != info->raid_disks)
6223 rv = update_raid_disks(mddev, info->raid_disks);
6225 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6226 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
6228 if (mddev->recovery || mddev->sync_thread)
6230 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6231 /* add the bitmap */
6234 if (mddev->bitmap_info.default_offset == 0)
6236 mddev->bitmap_info.offset =
6237 mddev->bitmap_info.default_offset;
6238 mddev->bitmap_info.space =
6239 mddev->bitmap_info.default_space;
6240 mddev->pers->quiesce(mddev, 1);
6241 rv = bitmap_create(mddev);
6243 rv = bitmap_load(mddev);
6245 bitmap_destroy(mddev);
6246 mddev->pers->quiesce(mddev, 0);
6248 /* remove the bitmap */
6251 if (mddev->bitmap->storage.file)
6253 mddev->pers->quiesce(mddev, 1);
6254 bitmap_destroy(mddev);
6255 mddev->pers->quiesce(mddev, 0);
6256 mddev->bitmap_info.offset = 0;
6259 md_update_sb(mddev, 1);
6263 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6265 struct md_rdev *rdev;
6268 if (mddev->pers == NULL)
6272 rdev = find_rdev_rcu(mddev, dev);
6276 md_error(mddev, rdev);
6277 if (!test_bit(Faulty, &rdev->flags))
6285 * We have a problem here : there is no easy way to give a CHS
6286 * virtual geometry. We currently pretend that we have a 2 heads
6287 * 4 sectors (with a BIG number of cylinders...). This drives
6288 * dosfs just mad... ;-)
6290 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6292 struct mddev *mddev = bdev->bd_disk->private_data;
6296 geo->cylinders = mddev->array_sectors / 8;
6300 static inline bool md_ioctl_valid(unsigned int cmd)
6305 case GET_ARRAY_INFO:
6306 case GET_BITMAP_FILE:
6309 case HOT_REMOVE_DISK:
6310 case PRINT_RAID_DEBUG:
6313 case RESTART_ARRAY_RW:
6315 case SET_ARRAY_INFO:
6316 case SET_BITMAP_FILE:
6317 case SET_DISK_FAULTY:
6326 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6327 unsigned int cmd, unsigned long arg)
6330 void __user *argp = (void __user *)arg;
6331 struct mddev *mddev = NULL;
6334 if (!md_ioctl_valid(cmd))
6339 case GET_ARRAY_INFO:
6343 if (!capable(CAP_SYS_ADMIN))
6348 * Commands dealing with the RAID driver but not any
6353 err = get_version(argp);
6356 case PRINT_RAID_DEBUG:
6364 autostart_arrays(arg);
6371 * Commands creating/starting a new array:
6374 mddev = bdev->bd_disk->private_data;
6381 /* Some actions do not requires the mutex */
6383 case GET_ARRAY_INFO:
6384 if (!mddev->raid_disks && !mddev->external)
6387 err = get_array_info(mddev, argp);
6391 if (!mddev->raid_disks && !mddev->external)
6394 err = get_disk_info(mddev, argp);
6397 case SET_DISK_FAULTY:
6398 err = set_disk_faulty(mddev, new_decode_dev(arg));
6402 if (cmd == ADD_NEW_DISK)
6403 /* need to ensure md_delayed_delete() has completed */
6404 flush_workqueue(md_misc_wq);
6406 if (cmd == HOT_REMOVE_DISK)
6407 /* need to ensure recovery thread has run */
6408 wait_event_interruptible_timeout(mddev->sb_wait,
6409 !test_bit(MD_RECOVERY_NEEDED,
6411 msecs_to_jiffies(5000));
6412 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6413 /* Need to flush page cache, and ensure no-one else opens
6416 mutex_lock(&mddev->open_mutex);
6417 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6418 mutex_unlock(&mddev->open_mutex);
6422 set_bit(MD_STILL_CLOSED, &mddev->flags);
6423 mutex_unlock(&mddev->open_mutex);
6424 sync_blockdev(bdev);
6426 err = mddev_lock(mddev);
6429 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6434 if (cmd == SET_ARRAY_INFO) {
6435 mdu_array_info_t info;
6437 memset(&info, 0, sizeof(info));
6438 else if (copy_from_user(&info, argp, sizeof(info))) {
6443 err = update_array_info(mddev, &info);
6445 printk(KERN_WARNING "md: couldn't update"
6446 " array info. %d\n", err);
6451 if (!list_empty(&mddev->disks)) {
6453 "md: array %s already has disks!\n",
6458 if (mddev->raid_disks) {
6460 "md: array %s already initialised!\n",
6465 err = set_array_info(mddev, &info);
6467 printk(KERN_WARNING "md: couldn't set"
6468 " array info. %d\n", err);
6475 * Commands querying/configuring an existing array:
6477 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6478 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6479 if ((!mddev->raid_disks && !mddev->external)
6480 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6481 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6482 && cmd != GET_BITMAP_FILE) {
6488 * Commands even a read-only array can execute:
6491 case GET_BITMAP_FILE:
6492 err = get_bitmap_file(mddev, argp);
6495 case RESTART_ARRAY_RW:
6496 err = restart_array(mddev);
6500 err = do_md_stop(mddev, 0, bdev);
6504 err = md_set_readonly(mddev, bdev);
6507 case HOT_REMOVE_DISK:
6508 err = hot_remove_disk(mddev, new_decode_dev(arg));
6512 /* We can support ADD_NEW_DISK on read-only arrays
6513 * on if we are re-adding a preexisting device.
6514 * So require mddev->pers and MD_DISK_SYNC.
6517 mdu_disk_info_t info;
6518 if (copy_from_user(&info, argp, sizeof(info)))
6520 else if (!(info.state & (1<<MD_DISK_SYNC)))
6521 /* Need to clear read-only for this */
6524 err = add_new_disk(mddev, &info);
6530 if (get_user(ro, (int __user *)(arg))) {
6536 /* if the bdev is going readonly the value of mddev->ro
6537 * does not matter, no writes are coming
6542 /* are we are already prepared for writes? */
6546 /* transitioning to readauto need only happen for
6547 * arrays that call md_write_start
6550 err = restart_array(mddev);
6553 set_disk_ro(mddev->gendisk, 0);
6560 * The remaining ioctls are changing the state of the
6561 * superblock, so we do not allow them on read-only arrays.
6562 * However non-MD ioctls (e.g. get-size) will still come through
6563 * here and hit the 'default' below, so only disallow
6564 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6566 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6567 if (mddev->ro == 2) {
6569 sysfs_notify_dirent_safe(mddev->sysfs_state);
6570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6571 /* mddev_unlock will wake thread */
6572 /* If a device failed while we were read-only, we
6573 * need to make sure the metadata is updated now.
6575 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6576 mddev_unlock(mddev);
6577 wait_event(mddev->sb_wait,
6578 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6579 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6580 mddev_lock_nointr(mddev);
6591 mdu_disk_info_t info;
6592 if (copy_from_user(&info, argp, sizeof(info)))
6595 err = add_new_disk(mddev, &info);
6600 err = hot_add_disk(mddev, new_decode_dev(arg));
6604 err = do_md_run(mddev);
6607 case SET_BITMAP_FILE:
6608 err = set_bitmap_file(mddev, (int)arg);
6618 if (mddev->hold_active == UNTIL_IOCTL &&
6620 mddev->hold_active = 0;
6621 mddev_unlock(mddev);
6630 #ifdef CONFIG_COMPAT
6631 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6632 unsigned int cmd, unsigned long arg)
6635 case HOT_REMOVE_DISK:
6637 case SET_DISK_FAULTY:
6638 case SET_BITMAP_FILE:
6639 /* These take in integer arg, do not convert */
6642 arg = (unsigned long)compat_ptr(arg);
6646 return md_ioctl(bdev, mode, cmd, arg);
6648 #endif /* CONFIG_COMPAT */
6650 static int md_open(struct block_device *bdev, fmode_t mode)
6653 * Succeed if we can lock the mddev, which confirms that
6654 * it isn't being stopped right now.
6656 struct mddev *mddev = mddev_find(bdev->bd_dev);
6662 if (mddev->gendisk != bdev->bd_disk) {
6663 /* we are racing with mddev_put which is discarding this
6667 /* Wait until bdev->bd_disk is definitely gone */
6668 flush_workqueue(md_misc_wq);
6669 /* Then retry the open from the top */
6670 return -ERESTARTSYS;
6672 BUG_ON(mddev != bdev->bd_disk->private_data);
6674 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6678 atomic_inc(&mddev->openers);
6679 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6680 mutex_unlock(&mddev->open_mutex);
6682 check_disk_change(bdev);
6687 static void md_release(struct gendisk *disk, fmode_t mode)
6689 struct mddev *mddev = disk->private_data;
6692 atomic_dec(&mddev->openers);
6696 static int md_media_changed(struct gendisk *disk)
6698 struct mddev *mddev = disk->private_data;
6700 return mddev->changed;
6703 static int md_revalidate(struct gendisk *disk)
6705 struct mddev *mddev = disk->private_data;
6710 static const struct block_device_operations md_fops =
6712 .owner = THIS_MODULE,
6714 .release = md_release,
6716 #ifdef CONFIG_COMPAT
6717 .compat_ioctl = md_compat_ioctl,
6719 .getgeo = md_getgeo,
6720 .media_changed = md_media_changed,
6721 .revalidate_disk= md_revalidate,
6724 static int md_thread(void *arg)
6726 struct md_thread *thread = arg;
6729 * md_thread is a 'system-thread', it's priority should be very
6730 * high. We avoid resource deadlocks individually in each
6731 * raid personality. (RAID5 does preallocation) We also use RR and
6732 * the very same RT priority as kswapd, thus we will never get
6733 * into a priority inversion deadlock.
6735 * we definitely have to have equal or higher priority than
6736 * bdflush, otherwise bdflush will deadlock if there are too
6737 * many dirty RAID5 blocks.
6740 allow_signal(SIGKILL);
6741 while (!kthread_should_stop()) {
6743 /* We need to wait INTERRUPTIBLE so that
6744 * we don't add to the load-average.
6745 * That means we need to be sure no signals are
6748 if (signal_pending(current))
6749 flush_signals(current);
6751 wait_event_interruptible_timeout
6753 test_bit(THREAD_WAKEUP, &thread->flags)
6754 || kthread_should_stop(),
6757 clear_bit(THREAD_WAKEUP, &thread->flags);
6758 if (!kthread_should_stop())
6759 thread->run(thread);
6765 void md_wakeup_thread(struct md_thread *thread)
6768 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6769 set_bit(THREAD_WAKEUP, &thread->flags);
6770 wake_up(&thread->wqueue);
6774 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6775 struct mddev *mddev, const char *name)
6777 struct md_thread *thread;
6779 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6783 init_waitqueue_head(&thread->wqueue);
6786 thread->mddev = mddev;
6787 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6788 thread->tsk = kthread_run(md_thread, thread,
6790 mdname(thread->mddev),
6792 if (IS_ERR(thread->tsk)) {
6799 void md_unregister_thread(struct md_thread **threadp)
6801 struct md_thread *thread = *threadp;
6804 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6805 /* Locking ensures that mddev_unlock does not wake_up a
6806 * non-existent thread
6808 spin_lock(&pers_lock);
6810 spin_unlock(&pers_lock);
6812 kthread_stop(thread->tsk);
6816 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6823 if (!rdev || test_bit(Faulty, &rdev->flags))
6826 if (!mddev->pers || !mddev->pers->error_handler)
6828 mddev->pers->error_handler(mddev,rdev);
6829 if (mddev->degraded)
6830 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6831 sysfs_notify_dirent_safe(rdev->sysfs_state);
6832 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6833 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6834 md_wakeup_thread(mddev->thread);
6835 if (mddev->event_work.func)
6836 queue_work(md_misc_wq, &mddev->event_work);
6837 md_new_event_inintr(mddev);
6840 /* seq_file implementation /proc/mdstat */
6842 static void status_unused(struct seq_file *seq)
6845 struct md_rdev *rdev;
6847 seq_printf(seq, "unused devices: ");
6849 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6850 char b[BDEVNAME_SIZE];
6852 seq_printf(seq, "%s ",
6853 bdevname(rdev->bdev,b));
6856 seq_printf(seq, "<none>");
6858 seq_printf(seq, "\n");
6861 static void status_resync(struct seq_file *seq, struct mddev *mddev)
6863 sector_t max_sectors, resync, res;
6864 unsigned long dt, db;
6867 unsigned int per_milli;
6869 if (mddev->curr_resync <= 3)
6872 resync = mddev->curr_resync
6873 - atomic_read(&mddev->recovery_active);
6875 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6876 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6877 max_sectors = mddev->resync_max_sectors;
6879 max_sectors = mddev->dev_sectors;
6882 * Should not happen.
6888 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6889 * in a sector_t, and (max_sectors>>scale) will fit in a
6890 * u32, as those are the requirements for sector_div.
6891 * Thus 'scale' must be at least 10
6894 if (sizeof(sector_t) > sizeof(unsigned long)) {
6895 while ( max_sectors/2 > (1ULL<<(scale+32)))
6898 res = (resync>>scale)*1000;
6899 sector_div(res, (u32)((max_sectors>>scale)+1));
6903 int i, x = per_milli/50, y = 20-x;
6904 seq_printf(seq, "[");
6905 for (i = 0; i < x; i++)
6906 seq_printf(seq, "=");
6907 seq_printf(seq, ">");
6908 for (i = 0; i < y; i++)
6909 seq_printf(seq, ".");
6910 seq_printf(seq, "] ");
6912 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6913 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6915 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6917 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6918 "resync" : "recovery"))),
6919 per_milli/10, per_milli % 10,
6920 (unsigned long long) resync/2,
6921 (unsigned long long) max_sectors/2);
6924 * dt: time from mark until now
6925 * db: blocks written from mark until now
6926 * rt: remaining time
6928 * rt is a sector_t, so could be 32bit or 64bit.
6929 * So we divide before multiply in case it is 32bit and close
6931 * We scale the divisor (db) by 32 to avoid losing precision
6932 * near the end of resync when the number of remaining sectors
6934 * We then divide rt by 32 after multiplying by db to compensate.
6935 * The '+1' avoids division by zero if db is very small.
6937 dt = ((jiffies - mddev->resync_mark) / HZ);
6939 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6940 - mddev->resync_mark_cnt;
6942 rt = max_sectors - resync; /* number of remaining sectors */
6943 sector_div(rt, db/32+1);
6947 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6948 ((unsigned long)rt % 60)/6);
6950 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6953 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6955 struct list_head *tmp;
6957 struct mddev *mddev;
6965 spin_lock(&all_mddevs_lock);
6966 list_for_each(tmp,&all_mddevs)
6968 mddev = list_entry(tmp, struct mddev, all_mddevs);
6970 spin_unlock(&all_mddevs_lock);
6973 spin_unlock(&all_mddevs_lock);
6975 return (void*)2;/* tail */
6979 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6981 struct list_head *tmp;
6982 struct mddev *next_mddev, *mddev = v;
6988 spin_lock(&all_mddevs_lock);
6990 tmp = all_mddevs.next;
6992 tmp = mddev->all_mddevs.next;
6993 if (tmp != &all_mddevs)
6994 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6996 next_mddev = (void*)2;
6999 spin_unlock(&all_mddevs_lock);
7007 static void md_seq_stop(struct seq_file *seq, void *v)
7009 struct mddev *mddev = v;
7011 if (mddev && v != (void*)1 && v != (void*)2)
7015 static int md_seq_show(struct seq_file *seq, void *v)
7017 struct mddev *mddev = v;
7019 struct md_rdev *rdev;
7021 if (v == (void*)1) {
7022 struct md_personality *pers;
7023 seq_printf(seq, "Personalities : ");
7024 spin_lock(&pers_lock);
7025 list_for_each_entry(pers, &pers_list, list)
7026 seq_printf(seq, "[%s] ", pers->name);
7028 spin_unlock(&pers_lock);
7029 seq_printf(seq, "\n");
7030 seq->poll_event = atomic_read(&md_event_count);
7033 if (v == (void*)2) {
7038 if (mddev_lock(mddev) < 0)
7041 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7042 seq_printf(seq, "%s : %sactive", mdname(mddev),
7043 mddev->pers ? "" : "in");
7046 seq_printf(seq, " (read-only)");
7048 seq_printf(seq, " (auto-read-only)");
7049 seq_printf(seq, " %s", mddev->pers->name);
7053 rdev_for_each(rdev, mddev) {
7054 char b[BDEVNAME_SIZE];
7055 seq_printf(seq, " %s[%d]",
7056 bdevname(rdev->bdev,b), rdev->desc_nr);
7057 if (test_bit(WriteMostly, &rdev->flags))
7058 seq_printf(seq, "(W)");
7059 if (test_bit(Faulty, &rdev->flags)) {
7060 seq_printf(seq, "(F)");
7063 if (rdev->raid_disk < 0)
7064 seq_printf(seq, "(S)"); /* spare */
7065 if (test_bit(Replacement, &rdev->flags))
7066 seq_printf(seq, "(R)");
7067 sectors += rdev->sectors;
7070 if (!list_empty(&mddev->disks)) {
7072 seq_printf(seq, "\n %llu blocks",
7073 (unsigned long long)
7074 mddev->array_sectors / 2);
7076 seq_printf(seq, "\n %llu blocks",
7077 (unsigned long long)sectors / 2);
7079 if (mddev->persistent) {
7080 if (mddev->major_version != 0 ||
7081 mddev->minor_version != 90) {
7082 seq_printf(seq," super %d.%d",
7083 mddev->major_version,
7084 mddev->minor_version);
7086 } else if (mddev->external)
7087 seq_printf(seq, " super external:%s",
7088 mddev->metadata_type);
7090 seq_printf(seq, " super non-persistent");
7093 mddev->pers->status(seq, mddev);
7094 seq_printf(seq, "\n ");
7095 if (mddev->pers->sync_request) {
7096 if (mddev->curr_resync > 2) {
7097 status_resync(seq, mddev);
7098 seq_printf(seq, "\n ");
7099 } else if (mddev->curr_resync >= 1)
7100 seq_printf(seq, "\tresync=DELAYED\n ");
7101 else if (mddev->recovery_cp < MaxSector)
7102 seq_printf(seq, "\tresync=PENDING\n ");
7105 seq_printf(seq, "\n ");
7107 bitmap_status(seq, mddev->bitmap);
7109 seq_printf(seq, "\n");
7111 mddev_unlock(mddev);
7116 static const struct seq_operations md_seq_ops = {
7117 .start = md_seq_start,
7118 .next = md_seq_next,
7119 .stop = md_seq_stop,
7120 .show = md_seq_show,
7123 static int md_seq_open(struct inode *inode, struct file *file)
7125 struct seq_file *seq;
7128 error = seq_open(file, &md_seq_ops);
7132 seq = file->private_data;
7133 seq->poll_event = atomic_read(&md_event_count);
7137 static int md_unloading;
7138 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7140 struct seq_file *seq = filp->private_data;
7144 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;;
7145 poll_wait(filp, &md_event_waiters, wait);
7147 /* always allow read */
7148 mask = POLLIN | POLLRDNORM;
7150 if (seq->poll_event != atomic_read(&md_event_count))
7151 mask |= POLLERR | POLLPRI;
7155 static const struct file_operations md_seq_fops = {
7156 .owner = THIS_MODULE,
7157 .open = md_seq_open,
7159 .llseek = seq_lseek,
7160 .release = seq_release_private,
7161 .poll = mdstat_poll,
7164 int register_md_personality(struct md_personality *p)
7166 printk(KERN_INFO "md: %s personality registered for level %d\n",
7168 spin_lock(&pers_lock);
7169 list_add_tail(&p->list, &pers_list);
7170 spin_unlock(&pers_lock);
7174 int unregister_md_personality(struct md_personality *p)
7176 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7177 spin_lock(&pers_lock);
7178 list_del_init(&p->list);
7179 spin_unlock(&pers_lock);
7183 static int is_mddev_idle(struct mddev *mddev, int init)
7185 struct md_rdev *rdev;
7191 rdev_for_each_rcu(rdev, mddev) {
7192 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7193 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7194 (int)part_stat_read(&disk->part0, sectors[1]) -
7195 atomic_read(&disk->sync_io);
7196 /* sync IO will cause sync_io to increase before the disk_stats
7197 * as sync_io is counted when a request starts, and
7198 * disk_stats is counted when it completes.
7199 * So resync activity will cause curr_events to be smaller than
7200 * when there was no such activity.
7201 * non-sync IO will cause disk_stat to increase without
7202 * increasing sync_io so curr_events will (eventually)
7203 * be larger than it was before. Once it becomes
7204 * substantially larger, the test below will cause
7205 * the array to appear non-idle, and resync will slow
7207 * If there is a lot of outstanding resync activity when
7208 * we set last_event to curr_events, then all that activity
7209 * completing might cause the array to appear non-idle
7210 * and resync will be slowed down even though there might
7211 * not have been non-resync activity. This will only
7212 * happen once though. 'last_events' will soon reflect
7213 * the state where there is little or no outstanding
7214 * resync requests, and further resync activity will
7215 * always make curr_events less than last_events.
7218 if (init || curr_events - rdev->last_events > 64) {
7219 rdev->last_events = curr_events;
7227 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7229 /* another "blocks" (512byte) blocks have been synced */
7230 atomic_sub(blocks, &mddev->recovery_active);
7231 wake_up(&mddev->recovery_wait);
7233 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7234 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7235 md_wakeup_thread(mddev->thread);
7236 // stop recovery, signal do_sync ....
7240 /* md_write_start(mddev, bi)
7241 * If we need to update some array metadata (e.g. 'active' flag
7242 * in superblock) before writing, schedule a superblock update
7243 * and wait for it to complete.
7245 void md_write_start(struct mddev *mddev, struct bio *bi)
7248 if (bio_data_dir(bi) != WRITE)
7251 BUG_ON(mddev->ro == 1);
7252 if (mddev->ro == 2) {
7253 /* need to switch to read/write */
7255 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7256 md_wakeup_thread(mddev->thread);
7257 md_wakeup_thread(mddev->sync_thread);
7260 atomic_inc(&mddev->writes_pending);
7261 if (mddev->safemode == 1)
7262 mddev->safemode = 0;
7263 if (mddev->in_sync) {
7264 spin_lock_irq(&mddev->write_lock);
7265 if (mddev->in_sync) {
7267 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7268 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7269 md_wakeup_thread(mddev->thread);
7272 spin_unlock_irq(&mddev->write_lock);
7275 sysfs_notify_dirent_safe(mddev->sysfs_state);
7276 wait_event(mddev->sb_wait,
7277 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7280 void md_write_end(struct mddev *mddev)
7282 if (atomic_dec_and_test(&mddev->writes_pending)) {
7283 if (mddev->safemode == 2)
7284 md_wakeup_thread(mddev->thread);
7285 else if (mddev->safemode_delay)
7286 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7290 /* md_allow_write(mddev)
7291 * Calling this ensures that the array is marked 'active' so that writes
7292 * may proceed without blocking. It is important to call this before
7293 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7294 * Must be called with mddev_lock held.
7296 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7297 * is dropped, so return -EAGAIN after notifying userspace.
7299 int md_allow_write(struct mddev *mddev)
7305 if (!mddev->pers->sync_request)
7308 spin_lock_irq(&mddev->write_lock);
7309 if (mddev->in_sync) {
7311 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7312 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7313 if (mddev->safemode_delay &&
7314 mddev->safemode == 0)
7315 mddev->safemode = 1;
7316 spin_unlock_irq(&mddev->write_lock);
7317 md_update_sb(mddev, 0);
7318 sysfs_notify_dirent_safe(mddev->sysfs_state);
7320 spin_unlock_irq(&mddev->write_lock);
7322 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7327 EXPORT_SYMBOL_GPL(md_allow_write);
7329 #define SYNC_MARKS 10
7330 #define SYNC_MARK_STEP (3*HZ)
7331 #define UPDATE_FREQUENCY (5*60*HZ)
7332 void md_do_sync(struct md_thread *thread)
7334 struct mddev *mddev = thread->mddev;
7335 struct mddev *mddev2;
7336 unsigned int currspeed = 0,
7338 sector_t max_sectors,j, io_sectors, recovery_done;
7339 unsigned long mark[SYNC_MARKS];
7340 unsigned long update_time;
7341 sector_t mark_cnt[SYNC_MARKS];
7343 struct list_head *tmp;
7344 sector_t last_check;
7346 struct md_rdev *rdev;
7347 char *desc, *action = NULL;
7348 struct blk_plug plug;
7350 /* just incase thread restarts... */
7351 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7353 if (mddev->ro) {/* never try to sync a read-only array */
7354 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7358 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7359 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7360 desc = "data-check";
7362 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7363 desc = "requested-resync";
7367 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7372 mddev->last_sync_action = action ?: desc;
7374 /* we overload curr_resync somewhat here.
7375 * 0 == not engaged in resync at all
7376 * 2 == checking that there is no conflict with another sync
7377 * 1 == like 2, but have yielded to allow conflicting resync to
7379 * other == active in resync - this many blocks
7381 * Before starting a resync we must have set curr_resync to
7382 * 2, and then checked that every "conflicting" array has curr_resync
7383 * less than ours. When we find one that is the same or higher
7384 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7385 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7386 * This will mean we have to start checking from the beginning again.
7391 mddev->curr_resync = 2;
7394 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7396 for_each_mddev(mddev2, tmp) {
7397 if (mddev2 == mddev)
7399 if (!mddev->parallel_resync
7400 && mddev2->curr_resync
7401 && match_mddev_units(mddev, mddev2)) {
7403 if (mddev < mddev2 && mddev->curr_resync == 2) {
7404 /* arbitrarily yield */
7405 mddev->curr_resync = 1;
7406 wake_up(&resync_wait);
7408 if (mddev > mddev2 && mddev->curr_resync == 1)
7409 /* no need to wait here, we can wait the next
7410 * time 'round when curr_resync == 2
7413 /* We need to wait 'interruptible' so as not to
7414 * contribute to the load average, and not to
7415 * be caught by 'softlockup'
7417 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7418 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7419 mddev2->curr_resync >= mddev->curr_resync) {
7420 printk(KERN_INFO "md: delaying %s of %s"
7421 " until %s has finished (they"
7422 " share one or more physical units)\n",
7423 desc, mdname(mddev), mdname(mddev2));
7425 if (signal_pending(current))
7426 flush_signals(current);
7428 finish_wait(&resync_wait, &wq);
7431 finish_wait(&resync_wait, &wq);
7434 } while (mddev->curr_resync < 2);
7437 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7438 /* resync follows the size requested by the personality,
7439 * which defaults to physical size, but can be virtual size
7441 max_sectors = mddev->resync_max_sectors;
7442 atomic64_set(&mddev->resync_mismatches, 0);
7443 /* we don't use the checkpoint if there's a bitmap */
7444 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7445 j = mddev->resync_min;
7446 else if (!mddev->bitmap)
7447 j = mddev->recovery_cp;
7449 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7450 max_sectors = mddev->resync_max_sectors;
7452 /* recovery follows the physical size of devices */
7453 max_sectors = mddev->dev_sectors;
7456 rdev_for_each_rcu(rdev, mddev)
7457 if (rdev->raid_disk >= 0 &&
7458 !test_bit(Faulty, &rdev->flags) &&
7459 !test_bit(In_sync, &rdev->flags) &&
7460 rdev->recovery_offset < j)
7461 j = rdev->recovery_offset;
7464 /* If there is a bitmap, we need to make sure all
7465 * writes that started before we added a spare
7466 * complete before we start doing a recovery.
7467 * Otherwise the write might complete and (via
7468 * bitmap_endwrite) set a bit in the bitmap after the
7469 * recovery has checked that bit and skipped that
7472 if (mddev->bitmap) {
7473 mddev->pers->quiesce(mddev, 1);
7474 mddev->pers->quiesce(mddev, 0);
7478 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7479 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7480 " %d KB/sec/disk.\n", speed_min(mddev));
7481 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7482 "(but not more than %d KB/sec) for %s.\n",
7483 speed_max(mddev), desc);
7485 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7488 for (m = 0; m < SYNC_MARKS; m++) {
7490 mark_cnt[m] = io_sectors;
7493 mddev->resync_mark = mark[last_mark];
7494 mddev->resync_mark_cnt = mark_cnt[last_mark];
7497 * Tune reconstruction:
7499 window = 32*(PAGE_SIZE/512);
7500 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7501 window/2, (unsigned long long)max_sectors/2);
7503 atomic_set(&mddev->recovery_active, 0);
7508 "md: resuming %s of %s from checkpoint.\n",
7509 desc, mdname(mddev));
7510 mddev->curr_resync = j;
7512 mddev->curr_resync = 3; /* no longer delayed */
7513 mddev->curr_resync_completed = j;
7514 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7515 md_new_event(mddev);
7516 update_time = jiffies;
7518 blk_start_plug(&plug);
7519 while (j < max_sectors) {
7524 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7525 ((mddev->curr_resync > mddev->curr_resync_completed &&
7526 (mddev->curr_resync - mddev->curr_resync_completed)
7527 > (max_sectors >> 4)) ||
7528 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7529 (j - mddev->curr_resync_completed)*2
7530 >= mddev->resync_max - mddev->curr_resync_completed
7532 /* time to update curr_resync_completed */
7533 wait_event(mddev->recovery_wait,
7534 atomic_read(&mddev->recovery_active) == 0);
7535 mddev->curr_resync_completed = j;
7536 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7537 j > mddev->recovery_cp)
7538 mddev->recovery_cp = j;
7539 update_time = jiffies;
7540 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7541 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7544 while (j >= mddev->resync_max &&
7545 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7546 /* As this condition is controlled by user-space,
7547 * we can block indefinitely, so use '_interruptible'
7548 * to avoid triggering warnings.
7550 flush_signals(current); /* just in case */
7551 wait_event_interruptible(mddev->recovery_wait,
7552 mddev->resync_max > j
7553 || test_bit(MD_RECOVERY_INTR,
7557 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7560 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7561 currspeed < speed_min(mddev));
7563 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7567 if (!skipped) { /* actual IO requested */
7568 io_sectors += sectors;
7569 atomic_add(sectors, &mddev->recovery_active);
7572 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7577 mddev->curr_resync = j;
7578 mddev->curr_mark_cnt = io_sectors;
7579 if (last_check == 0)
7580 /* this is the earliest that rebuild will be
7581 * visible in /proc/mdstat
7583 md_new_event(mddev);
7585 if (last_check + window > io_sectors || j == max_sectors)
7588 last_check = io_sectors;
7590 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7592 int next = (last_mark+1) % SYNC_MARKS;
7594 mddev->resync_mark = mark[next];
7595 mddev->resync_mark_cnt = mark_cnt[next];
7596 mark[next] = jiffies;
7597 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7601 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7605 * this loop exits only if either when we are slower than
7606 * the 'hard' speed limit, or the system was IO-idle for
7608 * the system might be non-idle CPU-wise, but we only care
7609 * about not overloading the IO subsystem. (things like an
7610 * e2fsck being done on the RAID array should execute fast)
7614 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7615 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7616 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7618 if (currspeed > speed_min(mddev)) {
7619 if ((currspeed > speed_max(mddev)) ||
7620 !is_mddev_idle(mddev, 0)) {
7626 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7627 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7628 ? "interrupted" : "done");
7630 * this also signals 'finished resyncing' to md_stop
7632 blk_finish_plug(&plug);
7633 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7635 /* tell personality that we are finished */
7636 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7638 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7639 mddev->curr_resync > 2) {
7640 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7641 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7642 if (mddev->curr_resync >= mddev->recovery_cp) {
7644 "md: checkpointing %s of %s.\n",
7645 desc, mdname(mddev));
7646 if (test_bit(MD_RECOVERY_ERROR,
7648 mddev->recovery_cp =
7649 mddev->curr_resync_completed;
7651 mddev->recovery_cp =
7655 mddev->recovery_cp = MaxSector;
7657 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7658 mddev->curr_resync = MaxSector;
7660 rdev_for_each_rcu(rdev, mddev)
7661 if (rdev->raid_disk >= 0 &&
7662 mddev->delta_disks >= 0 &&
7663 !test_bit(Faulty, &rdev->flags) &&
7664 !test_bit(In_sync, &rdev->flags) &&
7665 rdev->recovery_offset < mddev->curr_resync)
7666 rdev->recovery_offset = mddev->curr_resync;
7671 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7673 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7674 /* We completed so min/max setting can be forgotten if used. */
7675 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7676 mddev->resync_min = 0;
7677 mddev->resync_max = MaxSector;
7678 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7679 mddev->resync_min = mddev->curr_resync_completed;
7680 mddev->curr_resync = 0;
7681 wake_up(&resync_wait);
7682 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7683 md_wakeup_thread(mddev->thread);
7686 EXPORT_SYMBOL_GPL(md_do_sync);
7688 static int remove_and_add_spares(struct mddev *mddev,
7689 struct md_rdev *this)
7691 struct md_rdev *rdev;
7695 rdev_for_each(rdev, mddev)
7696 if ((this == NULL || rdev == this) &&
7697 rdev->raid_disk >= 0 &&
7698 !test_bit(Blocked, &rdev->flags) &&
7699 (test_bit(Faulty, &rdev->flags) ||
7700 ! test_bit(In_sync, &rdev->flags)) &&
7701 atomic_read(&rdev->nr_pending)==0) {
7702 if (mddev->pers->hot_remove_disk(
7703 mddev, rdev) == 0) {
7704 sysfs_unlink_rdev(mddev, rdev);
7705 rdev->raid_disk = -1;
7709 if (removed && mddev->kobj.sd)
7710 sysfs_notify(&mddev->kobj, NULL, "degraded");
7715 rdev_for_each(rdev, mddev) {
7716 if (rdev->raid_disk >= 0 &&
7717 !test_bit(In_sync, &rdev->flags) &&
7718 !test_bit(Faulty, &rdev->flags))
7720 if (rdev->raid_disk >= 0)
7722 if (test_bit(Faulty, &rdev->flags))
7725 ! (rdev->saved_raid_disk >= 0 &&
7726 !test_bit(Bitmap_sync, &rdev->flags)))
7729 if (rdev->saved_raid_disk < 0)
7730 rdev->recovery_offset = 0;
7732 hot_add_disk(mddev, rdev) == 0) {
7733 if (sysfs_link_rdev(mddev, rdev))
7734 /* failure here is OK */;
7736 md_new_event(mddev);
7737 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7742 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7746 static void md_start_sync(struct work_struct *ws)
7748 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7750 mddev->sync_thread = md_register_thread(md_do_sync,
7753 if (!mddev->sync_thread) {
7754 printk(KERN_ERR "%s: could not start resync"
7757 /* leave the spares where they are, it shouldn't hurt */
7758 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7759 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7760 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7761 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7762 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7763 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7765 if (mddev->sysfs_action)
7766 sysfs_notify_dirent_safe(mddev->sysfs_action);
7768 md_wakeup_thread(mddev->sync_thread);
7769 sysfs_notify_dirent_safe(mddev->sysfs_action);
7770 md_new_event(mddev);
7774 * This routine is regularly called by all per-raid-array threads to
7775 * deal with generic issues like resync and super-block update.
7776 * Raid personalities that don't have a thread (linear/raid0) do not
7777 * need this as they never do any recovery or update the superblock.
7779 * It does not do any resync itself, but rather "forks" off other threads
7780 * to do that as needed.
7781 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7782 * "->recovery" and create a thread at ->sync_thread.
7783 * When the thread finishes it sets MD_RECOVERY_DONE
7784 * and wakeups up this thread which will reap the thread and finish up.
7785 * This thread also removes any faulty devices (with nr_pending == 0).
7787 * The overall approach is:
7788 * 1/ if the superblock needs updating, update it.
7789 * 2/ If a recovery thread is running, don't do anything else.
7790 * 3/ If recovery has finished, clean up, possibly marking spares active.
7791 * 4/ If there are any faulty devices, remove them.
7792 * 5/ If array is degraded, try to add spares devices
7793 * 6/ If array has spares or is not in-sync, start a resync thread.
7795 void md_check_recovery(struct mddev *mddev)
7797 if (mddev->suspended)
7801 bitmap_daemon_work(mddev);
7803 if (signal_pending(current)) {
7804 if (mddev->pers->sync_request && !mddev->external) {
7805 printk(KERN_INFO "md: %s in immediate safe mode\n",
7807 mddev->safemode = 2;
7809 flush_signals(current);
7812 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7815 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7816 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7817 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7818 (mddev->external == 0 && mddev->safemode == 1) ||
7819 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7820 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7824 if (mddev_trylock(mddev)) {
7828 /* On a read-only array we can:
7829 * - remove failed devices
7830 * - add already-in_sync devices if the array itself
7832 * As we only add devices that are already in-sync,
7833 * we can activate the spares immediately.
7835 remove_and_add_spares(mddev, NULL);
7836 /* There is no thread, but we need to call
7837 * ->spare_active and clear saved_raid_disk
7839 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7840 md_reap_sync_thread(mddev);
7841 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7845 if (!mddev->external) {
7847 spin_lock_irq(&mddev->write_lock);
7848 if (mddev->safemode &&
7849 !atomic_read(&mddev->writes_pending) &&
7851 mddev->recovery_cp == MaxSector) {
7854 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7856 if (mddev->safemode == 1)
7857 mddev->safemode = 0;
7858 spin_unlock_irq(&mddev->write_lock);
7860 sysfs_notify_dirent_safe(mddev->sysfs_state);
7863 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7864 md_update_sb(mddev, 0);
7866 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7867 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7868 /* resync/recovery still happening */
7869 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7872 if (mddev->sync_thread) {
7873 md_reap_sync_thread(mddev);
7876 /* Set RUNNING before clearing NEEDED to avoid
7877 * any transients in the value of "sync_action".
7879 mddev->curr_resync_completed = 0;
7880 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7881 /* Clear some bits that don't mean anything, but
7884 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7885 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7887 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7888 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7890 /* no recovery is running.
7891 * remove any failed drives, then
7892 * add spares if possible.
7893 * Spares are also removed and re-added, to allow
7894 * the personality to fail the re-add.
7897 if (mddev->reshape_position != MaxSector) {
7898 if (mddev->pers->check_reshape == NULL ||
7899 mddev->pers->check_reshape(mddev) != 0)
7900 /* Cannot proceed */
7902 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7903 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7904 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7905 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7906 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7907 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7908 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7909 } else if (mddev->recovery_cp < MaxSector) {
7910 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7911 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7912 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7913 /* nothing to be done ... */
7916 if (mddev->pers->sync_request) {
7918 /* We are adding a device or devices to an array
7919 * which has the bitmap stored on all devices.
7920 * So make sure all bitmap pages get written
7922 bitmap_write_all(mddev->bitmap);
7924 INIT_WORK(&mddev->del_work, md_start_sync);
7925 queue_work(md_misc_wq, &mddev->del_work);
7929 if (!mddev->sync_thread) {
7930 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7931 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7933 if (mddev->sysfs_action)
7934 sysfs_notify_dirent_safe(mddev->sysfs_action);
7937 wake_up(&mddev->sb_wait);
7938 mddev_unlock(mddev);
7942 void md_reap_sync_thread(struct mddev *mddev)
7944 struct md_rdev *rdev;
7946 /* resync has finished, collect result */
7947 md_unregister_thread(&mddev->sync_thread);
7948 wake_up(&resync_wait);
7949 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7950 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7952 /* activate any spares */
7953 if (mddev->pers->spare_active(mddev)) {
7954 sysfs_notify(&mddev->kobj, NULL,
7956 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7959 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7960 mddev->pers->finish_reshape)
7961 mddev->pers->finish_reshape(mddev);
7963 /* If array is no-longer degraded, then any saved_raid_disk
7964 * information must be scrapped.
7966 if (!mddev->degraded)
7967 rdev_for_each(rdev, mddev)
7968 rdev->saved_raid_disk = -1;
7970 md_update_sb(mddev, 1);
7971 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7972 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7973 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7974 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7975 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7976 /* flag recovery needed just to double check */
7977 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7978 sysfs_notify_dirent_safe(mddev->sysfs_action);
7979 md_new_event(mddev);
7980 if (mddev->event_work.func)
7981 queue_work(md_misc_wq, &mddev->event_work);
7984 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7986 sysfs_notify_dirent_safe(rdev->sysfs_state);
7987 wait_event_timeout(rdev->blocked_wait,
7988 !test_bit(Blocked, &rdev->flags) &&
7989 !test_bit(BlockedBadBlocks, &rdev->flags),
7990 msecs_to_jiffies(5000));
7991 rdev_dec_pending(rdev, mddev);
7993 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7995 void md_finish_reshape(struct mddev *mddev)
7997 /* called be personality module when reshape completes. */
7998 struct md_rdev *rdev;
8000 rdev_for_each(rdev, mddev) {
8001 if (rdev->data_offset > rdev->new_data_offset)
8002 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8004 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8005 rdev->data_offset = rdev->new_data_offset;
8008 EXPORT_SYMBOL(md_finish_reshape);
8010 /* Bad block management.
8011 * We can record which blocks on each device are 'bad' and so just
8012 * fail those blocks, or that stripe, rather than the whole device.
8013 * Entries in the bad-block table are 64bits wide. This comprises:
8014 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8015 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8016 * A 'shift' can be set so that larger blocks are tracked and
8017 * consequently larger devices can be covered.
8018 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8020 * Locking of the bad-block table uses a seqlock so md_is_badblock
8021 * might need to retry if it is very unlucky.
8022 * We will sometimes want to check for bad blocks in a bi_end_io function,
8023 * so we use the write_seqlock_irq variant.
8025 * When looking for a bad block we specify a range and want to
8026 * know if any block in the range is bad. So we binary-search
8027 * to the last range that starts at-or-before the given endpoint,
8028 * (or "before the sector after the target range")
8029 * then see if it ends after the given start.
8031 * 0 if there are no known bad blocks in the range
8032 * 1 if there are known bad block which are all acknowledged
8033 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8034 * plus the start/length of the first bad section we overlap.
8036 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8037 sector_t *first_bad, int *bad_sectors)
8043 sector_t target = s + sectors;
8046 if (bb->shift > 0) {
8047 /* round the start down, and the end up */
8049 target += (1<<bb->shift) - 1;
8050 target >>= bb->shift;
8051 sectors = target - s;
8053 /* 'target' is now the first block after the bad range */
8056 seq = read_seqbegin(&bb->lock);
8061 /* Binary search between lo and hi for 'target'
8062 * i.e. for the last range that starts before 'target'
8064 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8065 * are known not to be the last range before target.
8066 * VARIANT: hi-lo is the number of possible
8067 * ranges, and decreases until it reaches 1
8069 while (hi - lo > 1) {
8070 int mid = (lo + hi) / 2;
8071 sector_t a = BB_OFFSET(p[mid]);
8073 /* This could still be the one, earlier ranges
8077 /* This and later ranges are definitely out. */
8080 /* 'lo' might be the last that started before target, but 'hi' isn't */
8082 /* need to check all range that end after 's' to see if
8083 * any are unacknowledged.
8086 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8087 if (BB_OFFSET(p[lo]) < target) {
8088 /* starts before the end, and finishes after
8089 * the start, so they must overlap
8091 if (rv != -1 && BB_ACK(p[lo]))
8095 *first_bad = BB_OFFSET(p[lo]);
8096 *bad_sectors = BB_LEN(p[lo]);
8102 if (read_seqretry(&bb->lock, seq))
8107 EXPORT_SYMBOL_GPL(md_is_badblock);
8110 * Add a range of bad blocks to the table.
8111 * This might extend the table, or might contract it
8112 * if two adjacent ranges can be merged.
8113 * We binary-search to find the 'insertion' point, then
8114 * decide how best to handle it.
8116 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8122 unsigned long flags;
8125 /* badblocks are disabled */
8129 /* round the start down, and the end up */
8130 sector_t next = s + sectors;
8132 next += (1<<bb->shift) - 1;
8137 write_seqlock_irqsave(&bb->lock, flags);
8142 /* Find the last range that starts at-or-before 's' */
8143 while (hi - lo > 1) {
8144 int mid = (lo + hi) / 2;
8145 sector_t a = BB_OFFSET(p[mid]);
8151 if (hi > lo && BB_OFFSET(p[lo]) > s)
8155 /* we found a range that might merge with the start
8158 sector_t a = BB_OFFSET(p[lo]);
8159 sector_t e = a + BB_LEN(p[lo]);
8160 int ack = BB_ACK(p[lo]);
8162 /* Yes, we can merge with a previous range */
8163 if (s == a && s + sectors >= e)
8164 /* new range covers old */
8167 ack = ack && acknowledged;
8169 if (e < s + sectors)
8171 if (e - a <= BB_MAX_LEN) {
8172 p[lo] = BB_MAKE(a, e-a, ack);
8175 /* does not all fit in one range,
8176 * make p[lo] maximal
8178 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8179 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8185 if (sectors && hi < bb->count) {
8186 /* 'hi' points to the first range that starts after 's'.
8187 * Maybe we can merge with the start of that range */
8188 sector_t a = BB_OFFSET(p[hi]);
8189 sector_t e = a + BB_LEN(p[hi]);
8190 int ack = BB_ACK(p[hi]);
8191 if (a <= s + sectors) {
8192 /* merging is possible */
8193 if (e <= s + sectors) {
8198 ack = ack && acknowledged;
8201 if (e - a <= BB_MAX_LEN) {
8202 p[hi] = BB_MAKE(a, e-a, ack);
8205 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8213 if (sectors == 0 && hi < bb->count) {
8214 /* we might be able to combine lo and hi */
8215 /* Note: 's' is at the end of 'lo' */
8216 sector_t a = BB_OFFSET(p[hi]);
8217 int lolen = BB_LEN(p[lo]);
8218 int hilen = BB_LEN(p[hi]);
8219 int newlen = lolen + hilen - (s - a);
8220 if (s >= a && newlen < BB_MAX_LEN) {
8221 /* yes, we can combine them */
8222 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8223 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8224 memmove(p + hi, p + hi + 1,
8225 (bb->count - hi - 1) * 8);
8230 /* didn't merge (it all).
8231 * Need to add a range just before 'hi' */
8232 if (bb->count >= MD_MAX_BADBLOCKS) {
8233 /* No room for more */
8237 int this_sectors = sectors;
8238 memmove(p + hi + 1, p + hi,
8239 (bb->count - hi) * 8);
8242 if (this_sectors > BB_MAX_LEN)
8243 this_sectors = BB_MAX_LEN;
8244 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8245 sectors -= this_sectors;
8252 bb->unacked_exist = 1;
8253 write_sequnlock_irqrestore(&bb->lock, flags);
8258 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8263 s += rdev->new_data_offset;
8265 s += rdev->data_offset;
8266 rv = md_set_badblocks(&rdev->badblocks,
8269 /* Make sure they get written out promptly */
8270 sysfs_notify_dirent_safe(rdev->sysfs_state);
8271 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8272 md_wakeup_thread(rdev->mddev->thread);
8276 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8279 * Remove a range of bad blocks from the table.
8280 * This may involve extending the table if we spilt a region,
8281 * but it must not fail. So if the table becomes full, we just
8282 * drop the remove request.
8284 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8288 sector_t target = s + sectors;
8291 if (bb->shift > 0) {
8292 /* When clearing we round the start up and the end down.
8293 * This should not matter as the shift should align with
8294 * the block size and no rounding should ever be needed.
8295 * However it is better the think a block is bad when it
8296 * isn't than to think a block is not bad when it is.
8298 s += (1<<bb->shift) - 1;
8300 target >>= bb->shift;
8301 sectors = target - s;
8304 write_seqlock_irq(&bb->lock);
8309 /* Find the last range that starts before 'target' */
8310 while (hi - lo > 1) {
8311 int mid = (lo + hi) / 2;
8312 sector_t a = BB_OFFSET(p[mid]);
8319 /* p[lo] is the last range that could overlap the
8320 * current range. Earlier ranges could also overlap,
8321 * but only this one can overlap the end of the range.
8323 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8324 /* Partial overlap, leave the tail of this range */
8325 int ack = BB_ACK(p[lo]);
8326 sector_t a = BB_OFFSET(p[lo]);
8327 sector_t end = a + BB_LEN(p[lo]);
8330 /* we need to split this range */
8331 if (bb->count >= MD_MAX_BADBLOCKS) {
8335 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8337 p[lo] = BB_MAKE(a, s-a, ack);
8340 p[lo] = BB_MAKE(target, end - target, ack);
8341 /* there is no longer an overlap */
8346 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8347 /* This range does overlap */
8348 if (BB_OFFSET(p[lo]) < s) {
8349 /* Keep the early parts of this range. */
8350 int ack = BB_ACK(p[lo]);
8351 sector_t start = BB_OFFSET(p[lo]);
8352 p[lo] = BB_MAKE(start, s - start, ack);
8353 /* now low doesn't overlap, so.. */
8358 /* 'lo' is strictly before, 'hi' is strictly after,
8359 * anything between needs to be discarded
8362 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8363 bb->count -= (hi - lo - 1);
8369 write_sequnlock_irq(&bb->lock);
8373 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8377 s += rdev->new_data_offset;
8379 s += rdev->data_offset;
8380 return md_clear_badblocks(&rdev->badblocks,
8383 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8386 * Acknowledge all bad blocks in a list.
8387 * This only succeeds if ->changed is clear. It is used by
8388 * in-kernel metadata updates
8390 void md_ack_all_badblocks(struct badblocks *bb)
8392 if (bb->page == NULL || bb->changed)
8393 /* no point even trying */
8395 write_seqlock_irq(&bb->lock);
8397 if (bb->changed == 0 && bb->unacked_exist) {
8400 for (i = 0; i < bb->count ; i++) {
8401 if (!BB_ACK(p[i])) {
8402 sector_t start = BB_OFFSET(p[i]);
8403 int len = BB_LEN(p[i]);
8404 p[i] = BB_MAKE(start, len, 1);
8407 bb->unacked_exist = 0;
8409 write_sequnlock_irq(&bb->lock);
8411 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8413 /* sysfs access to bad-blocks list.
8414 * We present two files.
8415 * 'bad-blocks' lists sector numbers and lengths of ranges that
8416 * are recorded as bad. The list is truncated to fit within
8417 * the one-page limit of sysfs.
8418 * Writing "sector length" to this file adds an acknowledged
8420 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8421 * been acknowledged. Writing to this file adds bad blocks
8422 * without acknowledging them. This is largely for testing.
8426 badblocks_show(struct badblocks *bb, char *page, int unack)
8437 seq = read_seqbegin(&bb->lock);
8442 while (len < PAGE_SIZE && i < bb->count) {
8443 sector_t s = BB_OFFSET(p[i]);
8444 unsigned int length = BB_LEN(p[i]);
8445 int ack = BB_ACK(p[i]);
8451 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8452 (unsigned long long)s << bb->shift,
8453 length << bb->shift);
8455 if (unack && len == 0)
8456 bb->unacked_exist = 0;
8458 if (read_seqretry(&bb->lock, seq))
8467 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8469 unsigned long long sector;
8473 /* Allow clearing via sysfs *only* for testing/debugging.
8474 * Normally only a successful write may clear a badblock
8477 if (page[0] == '-') {
8481 #endif /* DO_DEBUG */
8483 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8485 if (newline != '\n')
8497 md_clear_badblocks(bb, sector, length);
8500 #endif /* DO_DEBUG */
8501 if (md_set_badblocks(bb, sector, length, !unack))
8507 static int md_notify_reboot(struct notifier_block *this,
8508 unsigned long code, void *x)
8510 struct list_head *tmp;
8511 struct mddev *mddev;
8514 for_each_mddev(mddev, tmp) {
8515 if (mddev_trylock(mddev)) {
8517 __md_stop_writes(mddev);
8518 if (mddev->persistent)
8519 mddev->safemode = 2;
8520 mddev_unlock(mddev);
8525 * certain more exotic SCSI devices are known to be
8526 * volatile wrt too early system reboots. While the
8527 * right place to handle this issue is the given
8528 * driver, we do want to have a safe RAID driver ...
8536 static struct notifier_block md_notifier = {
8537 .notifier_call = md_notify_reboot,
8539 .priority = INT_MAX, /* before any real devices */
8542 static void md_geninit(void)
8544 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8546 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8549 static int __init md_init(void)
8553 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8557 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8561 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8564 if ((ret = register_blkdev(0, "mdp")) < 0)
8568 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8569 md_probe, NULL, NULL);
8570 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8571 md_probe, NULL, NULL);
8573 register_reboot_notifier(&md_notifier);
8574 raid_table_header = register_sysctl_table(raid_root_table);
8580 unregister_blkdev(MD_MAJOR, "md");
8582 destroy_workqueue(md_misc_wq);
8584 destroy_workqueue(md_wq);
8592 * Searches all registered partitions for autorun RAID arrays
8596 static LIST_HEAD(all_detected_devices);
8597 struct detected_devices_node {
8598 struct list_head list;
8602 void md_autodetect_dev(dev_t dev)
8604 struct detected_devices_node *node_detected_dev;
8606 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8607 if (node_detected_dev) {
8608 node_detected_dev->dev = dev;
8609 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8611 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8612 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8616 static void autostart_arrays(int part)
8618 struct md_rdev *rdev;
8619 struct detected_devices_node *node_detected_dev;
8621 int i_scanned, i_passed;
8626 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8628 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8630 node_detected_dev = list_entry(all_detected_devices.next,
8631 struct detected_devices_node, list);
8632 list_del(&node_detected_dev->list);
8633 dev = node_detected_dev->dev;
8634 kfree(node_detected_dev);
8635 rdev = md_import_device(dev,0, 90);
8639 if (test_bit(Faulty, &rdev->flags)) {
8643 set_bit(AutoDetected, &rdev->flags);
8644 list_add(&rdev->same_set, &pending_raid_disks);
8648 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8649 i_scanned, i_passed);
8651 autorun_devices(part);
8654 #endif /* !MODULE */
8656 static __exit void md_exit(void)
8658 struct mddev *mddev;
8659 struct list_head *tmp;
8662 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8663 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8665 unregister_blkdev(MD_MAJOR,"md");
8666 unregister_blkdev(mdp_major, "mdp");
8667 unregister_reboot_notifier(&md_notifier);
8668 unregister_sysctl_table(raid_table_header);
8670 /* We cannot unload the modules while some process is
8671 * waiting for us in select() or poll() - wake them up
8674 while (waitqueue_active(&md_event_waiters)) {
8675 /* not safe to leave yet */
8676 wake_up(&md_event_waiters);
8680 remove_proc_entry("mdstat", NULL);
8682 for_each_mddev(mddev, tmp) {
8683 export_array(mddev);
8684 mddev->hold_active = 0;
8686 destroy_workqueue(md_misc_wq);
8687 destroy_workqueue(md_wq);
8690 subsys_initcall(md_init);
8691 module_exit(md_exit)
8693 static int get_ro(char *buffer, struct kernel_param *kp)
8695 return sprintf(buffer, "%d", start_readonly);
8697 static int set_ro(const char *val, struct kernel_param *kp)
8700 int num = simple_strtoul(val, &e, 10);
8701 if (*val && (*e == '\0' || *e == '\n')) {
8702 start_readonly = num;
8708 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8709 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8711 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8713 EXPORT_SYMBOL(register_md_personality);
8714 EXPORT_SYMBOL(unregister_md_personality);
8715 EXPORT_SYMBOL(md_error);
8716 EXPORT_SYMBOL(md_done_sync);
8717 EXPORT_SYMBOL(md_write_start);
8718 EXPORT_SYMBOL(md_write_end);
8719 EXPORT_SYMBOL(md_register_thread);
8720 EXPORT_SYMBOL(md_unregister_thread);
8721 EXPORT_SYMBOL(md_wakeup_thread);
8722 EXPORT_SYMBOL(md_check_recovery);
8723 EXPORT_SYMBOL(md_reap_sync_thread);
8724 MODULE_LICENSE("GPL");
8725 MODULE_DESCRIPTION("MD RAID framework");
8727 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);