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 DECLARE_WAIT_QUEUE_HEAD(resync_wait);
70 static struct workqueue_struct *md_wq;
71 static struct workqueue_struct *md_misc_wq;
73 static int remove_and_add_spares(struct mddev *mddev,
74 struct md_rdev *this);
75 static void mddev_detach(struct mddev *mddev);
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min = 1000;
97 static int sysctl_speed_limit_max = 200000;
98 static inline int speed_min(struct mddev *mddev)
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
104 static inline int speed_max(struct mddev *mddev)
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
110 static struct ctl_table_header *raid_table_header;
112 static struct ctl_table raid_table[] = {
114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
117 .mode = S_IRUGO|S_IWUSR,
118 .proc_handler = proc_dointvec,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
130 static struct ctl_table raid_dir_table[] = {
134 .mode = S_IRUGO|S_IXUGO,
140 static struct ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static const struct block_device_operations md_fops;
152 static int start_readonly;
155 * like bio_clone, but with a local bio set
158 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
163 if (!mddev || !mddev->bio_set)
164 return bio_alloc(gfp_mask, nr_iovecs);
166 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
171 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
173 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
176 if (!mddev || !mddev->bio_set)
177 return bio_clone(bio, gfp_mask);
179 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
181 EXPORT_SYMBOL_GPL(bio_clone_mddev);
184 * We have a system wide 'event count' that is incremented
185 * on any 'interesting' event, and readers of /proc/mdstat
186 * can use 'poll' or 'select' to find out when the event
190 * start array, stop array, error, add device, remove device,
191 * start build, activate spare
193 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
194 static atomic_t md_event_count;
195 void md_new_event(struct mddev *mddev)
197 atomic_inc(&md_event_count);
198 wake_up(&md_event_waiters);
200 EXPORT_SYMBOL_GPL(md_new_event);
202 /* Alternate version that can be called from interrupts
203 * when calling sysfs_notify isn't needed.
205 static void md_new_event_inintr(struct mddev *mddev)
207 atomic_inc(&md_event_count);
208 wake_up(&md_event_waiters);
212 * Enables to iterate over all existing md arrays
213 * all_mddevs_lock protects this list.
215 static LIST_HEAD(all_mddevs);
216 static DEFINE_SPINLOCK(all_mddevs_lock);
219 * iterates through all used mddevs in the system.
220 * We take care to grab the all_mddevs_lock whenever navigating
221 * the list, and to always hold a refcount when unlocked.
222 * Any code which breaks out of this loop while own
223 * a reference to the current mddev and must mddev_put it.
225 #define for_each_mddev(_mddev,_tmp) \
227 for (({ spin_lock(&all_mddevs_lock); \
228 _tmp = all_mddevs.next; \
230 ({ if (_tmp != &all_mddevs) \
231 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
232 spin_unlock(&all_mddevs_lock); \
233 if (_mddev) mddev_put(_mddev); \
234 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
235 _tmp != &all_mddevs;}); \
236 ({ spin_lock(&all_mddevs_lock); \
237 _tmp = _tmp->next;}) \
240 /* Rather than calling directly into the personality make_request function,
241 * IO requests come here first so that we can check if the device is
242 * being suspended pending a reconfiguration.
243 * We hold a refcount over the call to ->make_request. By the time that
244 * call has finished, the bio has been linked into some internal structure
245 * and so is visible to ->quiesce(), so we don't need the refcount any more.
247 static void md_make_request(struct request_queue *q, struct bio *bio)
249 const int rw = bio_data_dir(bio);
250 struct mddev *mddev = q->queuedata;
251 unsigned int sectors;
253 if (mddev == NULL || mddev->pers == NULL
258 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
259 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
262 smp_rmb(); /* Ensure implications of 'active' are visible */
264 if (mddev->suspended) {
267 prepare_to_wait(&mddev->sb_wait, &__wait,
268 TASK_UNINTERRUPTIBLE);
269 if (!mddev->suspended)
275 finish_wait(&mddev->sb_wait, &__wait);
277 atomic_inc(&mddev->active_io);
281 * save the sectors now since our bio can
282 * go away inside make_request
284 sectors = bio_sectors(bio);
285 mddev->pers->make_request(mddev, bio);
287 generic_start_io_acct(rw, sectors, &mddev->gendisk->part0);
289 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
290 wake_up(&mddev->sb_wait);
293 /* mddev_suspend makes sure no new requests are submitted
294 * to the device, and that any requests that have been submitted
295 * are completely handled.
296 * Once mddev_detach() is called and completes, the module will be
299 void mddev_suspend(struct mddev *mddev)
301 BUG_ON(mddev->suspended);
302 mddev->suspended = 1;
304 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
305 mddev->pers->quiesce(mddev, 1);
307 del_timer_sync(&mddev->safemode_timer);
309 EXPORT_SYMBOL_GPL(mddev_suspend);
311 void mddev_resume(struct mddev *mddev)
313 mddev->suspended = 0;
314 wake_up(&mddev->sb_wait);
315 mddev->pers->quiesce(mddev, 0);
317 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
318 md_wakeup_thread(mddev->thread);
319 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
321 EXPORT_SYMBOL_GPL(mddev_resume);
323 int mddev_congested(struct mddev *mddev, int bits)
325 struct md_personality *pers = mddev->pers;
329 if (mddev->suspended)
331 else if (pers && pers->congested)
332 ret = pers->congested(mddev, bits);
336 EXPORT_SYMBOL_GPL(mddev_congested);
337 static int md_congested(void *data, int bits)
339 struct mddev *mddev = data;
340 return mddev_congested(mddev, bits);
343 static int md_mergeable_bvec(struct request_queue *q,
344 struct bvec_merge_data *bvm,
345 struct bio_vec *biovec)
347 struct mddev *mddev = q->queuedata;
350 if (mddev->suspended) {
351 /* Must always allow one vec */
352 if (bvm->bi_size == 0)
353 ret = biovec->bv_len;
357 struct md_personality *pers = mddev->pers;
358 if (pers && pers->mergeable_bvec)
359 ret = pers->mergeable_bvec(mddev, bvm, biovec);
361 ret = biovec->bv_len;
367 * Generic flush handling for md
370 static void md_end_flush(struct bio *bio, int err)
372 struct md_rdev *rdev = bio->bi_private;
373 struct mddev *mddev = rdev->mddev;
375 rdev_dec_pending(rdev, mddev);
377 if (atomic_dec_and_test(&mddev->flush_pending)) {
378 /* The pre-request flush has finished */
379 queue_work(md_wq, &mddev->flush_work);
384 static void md_submit_flush_data(struct work_struct *ws);
386 static void submit_flushes(struct work_struct *ws)
388 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
389 struct md_rdev *rdev;
391 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
392 atomic_set(&mddev->flush_pending, 1);
394 rdev_for_each_rcu(rdev, mddev)
395 if (rdev->raid_disk >= 0 &&
396 !test_bit(Faulty, &rdev->flags)) {
397 /* Take two references, one is dropped
398 * when request finishes, one after
399 * we reclaim rcu_read_lock
402 atomic_inc(&rdev->nr_pending);
403 atomic_inc(&rdev->nr_pending);
405 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
406 bi->bi_end_io = md_end_flush;
407 bi->bi_private = rdev;
408 bi->bi_bdev = rdev->bdev;
409 atomic_inc(&mddev->flush_pending);
410 submit_bio(WRITE_FLUSH, bi);
412 rdev_dec_pending(rdev, mddev);
415 if (atomic_dec_and_test(&mddev->flush_pending))
416 queue_work(md_wq, &mddev->flush_work);
419 static void md_submit_flush_data(struct work_struct *ws)
421 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
422 struct bio *bio = mddev->flush_bio;
424 if (bio->bi_iter.bi_size == 0)
425 /* an empty barrier - all done */
428 bio->bi_rw &= ~REQ_FLUSH;
429 mddev->pers->make_request(mddev, bio);
432 mddev->flush_bio = NULL;
433 wake_up(&mddev->sb_wait);
436 void md_flush_request(struct mddev *mddev, struct bio *bio)
438 spin_lock_irq(&mddev->lock);
439 wait_event_lock_irq(mddev->sb_wait,
442 mddev->flush_bio = bio;
443 spin_unlock_irq(&mddev->lock);
445 INIT_WORK(&mddev->flush_work, submit_flushes);
446 queue_work(md_wq, &mddev->flush_work);
448 EXPORT_SYMBOL(md_flush_request);
450 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
452 struct mddev *mddev = cb->data;
453 md_wakeup_thread(mddev->thread);
456 EXPORT_SYMBOL(md_unplug);
458 static inline struct mddev *mddev_get(struct mddev *mddev)
460 atomic_inc(&mddev->active);
464 static void mddev_delayed_delete(struct work_struct *ws);
466 static void mddev_put(struct mddev *mddev)
468 struct bio_set *bs = NULL;
470 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
472 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
473 mddev->ctime == 0 && !mddev->hold_active) {
474 /* Array is not configured at all, and not held active,
476 list_del_init(&mddev->all_mddevs);
478 mddev->bio_set = NULL;
479 if (mddev->gendisk) {
480 /* We did a probe so need to clean up. Call
481 * queue_work inside the spinlock so that
482 * flush_workqueue() after mddev_find will
483 * succeed in waiting for the work to be done.
485 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
486 queue_work(md_misc_wq, &mddev->del_work);
490 spin_unlock(&all_mddevs_lock);
495 void mddev_init(struct mddev *mddev)
497 mutex_init(&mddev->open_mutex);
498 mutex_init(&mddev->reconfig_mutex);
499 mutex_init(&mddev->bitmap_info.mutex);
500 INIT_LIST_HEAD(&mddev->disks);
501 INIT_LIST_HEAD(&mddev->all_mddevs);
502 init_timer(&mddev->safemode_timer);
503 atomic_set(&mddev->active, 1);
504 atomic_set(&mddev->openers, 0);
505 atomic_set(&mddev->active_io, 0);
506 spin_lock_init(&mddev->lock);
507 atomic_set(&mddev->flush_pending, 0);
508 init_waitqueue_head(&mddev->sb_wait);
509 init_waitqueue_head(&mddev->recovery_wait);
510 mddev->reshape_position = MaxSector;
511 mddev->reshape_backwards = 0;
512 mddev->last_sync_action = "none";
513 mddev->resync_min = 0;
514 mddev->resync_max = MaxSector;
515 mddev->level = LEVEL_NONE;
517 EXPORT_SYMBOL_GPL(mddev_init);
519 static struct mddev *mddev_find(dev_t unit)
521 struct mddev *mddev, *new = NULL;
523 if (unit && MAJOR(unit) != MD_MAJOR)
524 unit &= ~((1<<MdpMinorShift)-1);
527 spin_lock(&all_mddevs_lock);
530 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
531 if (mddev->unit == unit) {
533 spin_unlock(&all_mddevs_lock);
539 list_add(&new->all_mddevs, &all_mddevs);
540 spin_unlock(&all_mddevs_lock);
541 new->hold_active = UNTIL_IOCTL;
545 /* find an unused unit number */
546 static int next_minor = 512;
547 int start = next_minor;
551 dev = MKDEV(MD_MAJOR, next_minor);
553 if (next_minor > MINORMASK)
555 if (next_minor == start) {
556 /* Oh dear, all in use. */
557 spin_unlock(&all_mddevs_lock);
563 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
564 if (mddev->unit == dev) {
570 new->md_minor = MINOR(dev);
571 new->hold_active = UNTIL_STOP;
572 list_add(&new->all_mddevs, &all_mddevs);
573 spin_unlock(&all_mddevs_lock);
576 spin_unlock(&all_mddevs_lock);
578 new = kzalloc(sizeof(*new), GFP_KERNEL);
583 if (MAJOR(unit) == MD_MAJOR)
584 new->md_minor = MINOR(unit);
586 new->md_minor = MINOR(unit) >> MdpMinorShift;
593 static struct attribute_group md_redundancy_group;
595 void mddev_unlock(struct mddev *mddev)
597 if (mddev->to_remove) {
598 /* These cannot be removed under reconfig_mutex as
599 * an access to the files will try to take reconfig_mutex
600 * while holding the file unremovable, which leads to
602 * So hold set sysfs_active while the remove in happeing,
603 * and anything else which might set ->to_remove or my
604 * otherwise change the sysfs namespace will fail with
605 * -EBUSY if sysfs_active is still set.
606 * We set sysfs_active under reconfig_mutex and elsewhere
607 * test it under the same mutex to ensure its correct value
610 struct attribute_group *to_remove = mddev->to_remove;
611 mddev->to_remove = NULL;
612 mddev->sysfs_active = 1;
613 mutex_unlock(&mddev->reconfig_mutex);
615 if (mddev->kobj.sd) {
616 if (to_remove != &md_redundancy_group)
617 sysfs_remove_group(&mddev->kobj, to_remove);
618 if (mddev->pers == NULL ||
619 mddev->pers->sync_request == NULL) {
620 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
621 if (mddev->sysfs_action)
622 sysfs_put(mddev->sysfs_action);
623 mddev->sysfs_action = NULL;
626 mddev->sysfs_active = 0;
628 mutex_unlock(&mddev->reconfig_mutex);
630 /* As we've dropped the mutex we need a spinlock to
631 * make sure the thread doesn't disappear
633 spin_lock(&pers_lock);
634 md_wakeup_thread(mddev->thread);
635 spin_unlock(&pers_lock);
637 EXPORT_SYMBOL_GPL(mddev_unlock);
639 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
641 struct md_rdev *rdev;
643 rdev_for_each_rcu(rdev, mddev)
644 if (rdev->desc_nr == nr)
650 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
652 struct md_rdev *rdev;
654 rdev_for_each(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
661 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
663 struct md_rdev *rdev;
665 rdev_for_each_rcu(rdev, mddev)
666 if (rdev->bdev->bd_dev == dev)
672 static struct md_personality *find_pers(int level, char *clevel)
674 struct md_personality *pers;
675 list_for_each_entry(pers, &pers_list, list) {
676 if (level != LEVEL_NONE && pers->level == level)
678 if (strcmp(pers->name, clevel)==0)
684 /* return the offset of the super block in 512byte sectors */
685 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
687 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
688 return MD_NEW_SIZE_SECTORS(num_sectors);
691 static int alloc_disk_sb(struct md_rdev *rdev)
693 rdev->sb_page = alloc_page(GFP_KERNEL);
694 if (!rdev->sb_page) {
695 printk(KERN_ALERT "md: out of memory.\n");
702 void md_rdev_clear(struct md_rdev *rdev)
705 put_page(rdev->sb_page);
707 rdev->sb_page = NULL;
712 put_page(rdev->bb_page);
713 rdev->bb_page = NULL;
715 kfree(rdev->badblocks.page);
716 rdev->badblocks.page = NULL;
718 EXPORT_SYMBOL_GPL(md_rdev_clear);
720 static void super_written(struct bio *bio, int error)
722 struct md_rdev *rdev = bio->bi_private;
723 struct mddev *mddev = rdev->mddev;
725 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
726 printk("md: super_written gets error=%d, uptodate=%d\n",
727 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
728 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
729 md_error(mddev, rdev);
732 if (atomic_dec_and_test(&mddev->pending_writes))
733 wake_up(&mddev->sb_wait);
737 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
738 sector_t sector, int size, struct page *page)
740 /* write first size bytes of page to sector of rdev
741 * Increment mddev->pending_writes before returning
742 * and decrement it on completion, waking up sb_wait
743 * if zero is reached.
744 * If an error occurred, call md_error
746 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
748 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
749 bio->bi_iter.bi_sector = sector;
750 bio_add_page(bio, page, size, 0);
751 bio->bi_private = rdev;
752 bio->bi_end_io = super_written;
754 atomic_inc(&mddev->pending_writes);
755 submit_bio(WRITE_FLUSH_FUA, bio);
758 void md_super_wait(struct mddev *mddev)
760 /* wait for all superblock writes that were scheduled to complete */
761 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
764 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
765 struct page *page, int rw, bool metadata_op)
767 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
770 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
771 rdev->meta_bdev : rdev->bdev;
773 bio->bi_iter.bi_sector = sector + rdev->sb_start;
774 else if (rdev->mddev->reshape_position != MaxSector &&
775 (rdev->mddev->reshape_backwards ==
776 (sector >= rdev->mddev->reshape_position)))
777 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
779 bio->bi_iter.bi_sector = sector + rdev->data_offset;
780 bio_add_page(bio, page, size, 0);
781 submit_bio_wait(rw, bio);
783 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
787 EXPORT_SYMBOL_GPL(sync_page_io);
789 static int read_disk_sb(struct md_rdev *rdev, int size)
791 char b[BDEVNAME_SIZE];
796 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
802 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
803 bdevname(rdev->bdev,b));
807 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
809 return sb1->set_uuid0 == sb2->set_uuid0 &&
810 sb1->set_uuid1 == sb2->set_uuid1 &&
811 sb1->set_uuid2 == sb2->set_uuid2 &&
812 sb1->set_uuid3 == sb2->set_uuid3;
815 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
818 mdp_super_t *tmp1, *tmp2;
820 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
821 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
823 if (!tmp1 || !tmp2) {
825 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
833 * nr_disks is not constant
838 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
845 static u32 md_csum_fold(u32 csum)
847 csum = (csum & 0xffff) + (csum >> 16);
848 return (csum & 0xffff) + (csum >> 16);
851 static unsigned int calc_sb_csum(mdp_super_t *sb)
854 u32 *sb32 = (u32*)sb;
856 unsigned int disk_csum, csum;
858 disk_csum = sb->sb_csum;
861 for (i = 0; i < MD_SB_BYTES/4 ; i++)
863 csum = (newcsum & 0xffffffff) + (newcsum>>32);
866 /* This used to use csum_partial, which was wrong for several
867 * reasons including that different results are returned on
868 * different architectures. It isn't critical that we get exactly
869 * the same return value as before (we always csum_fold before
870 * testing, and that removes any differences). However as we
871 * know that csum_partial always returned a 16bit value on
872 * alphas, do a fold to maximise conformity to previous behaviour.
874 sb->sb_csum = md_csum_fold(disk_csum);
876 sb->sb_csum = disk_csum;
882 * Handle superblock details.
883 * We want to be able to handle multiple superblock formats
884 * so we have a common interface to them all, and an array of
885 * different handlers.
886 * We rely on user-space to write the initial superblock, and support
887 * reading and updating of superblocks.
888 * Interface methods are:
889 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
890 * loads and validates a superblock on dev.
891 * if refdev != NULL, compare superblocks on both devices
893 * 0 - dev has a superblock that is compatible with refdev
894 * 1 - dev has a superblock that is compatible and newer than refdev
895 * so dev should be used as the refdev in future
896 * -EINVAL superblock incompatible or invalid
897 * -othererror e.g. -EIO
899 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
900 * Verify that dev is acceptable into mddev.
901 * The first time, mddev->raid_disks will be 0, and data from
902 * dev should be merged in. Subsequent calls check that dev
903 * is new enough. Return 0 or -EINVAL
905 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
906 * Update the superblock for rdev with data in mddev
907 * This does not write to disc.
913 struct module *owner;
914 int (*load_super)(struct md_rdev *rdev,
915 struct md_rdev *refdev,
917 int (*validate_super)(struct mddev *mddev,
918 struct md_rdev *rdev);
919 void (*sync_super)(struct mddev *mddev,
920 struct md_rdev *rdev);
921 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
922 sector_t num_sectors);
923 int (*allow_new_offset)(struct md_rdev *rdev,
924 unsigned long long new_offset);
928 * Check that the given mddev has no bitmap.
930 * This function is called from the run method of all personalities that do not
931 * support bitmaps. It prints an error message and returns non-zero if mddev
932 * has a bitmap. Otherwise, it returns 0.
935 int md_check_no_bitmap(struct mddev *mddev)
937 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
939 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
940 mdname(mddev), mddev->pers->name);
943 EXPORT_SYMBOL(md_check_no_bitmap);
946 * load_super for 0.90.0
948 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
950 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
955 * Calculate the position of the superblock (512byte sectors),
956 * it's at the end of the disk.
958 * It also happens to be a multiple of 4Kb.
960 rdev->sb_start = calc_dev_sboffset(rdev);
962 ret = read_disk_sb(rdev, MD_SB_BYTES);
967 bdevname(rdev->bdev, b);
968 sb = page_address(rdev->sb_page);
970 if (sb->md_magic != MD_SB_MAGIC) {
971 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
976 if (sb->major_version != 0 ||
977 sb->minor_version < 90 ||
978 sb->minor_version > 91) {
979 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
980 sb->major_version, sb->minor_version,
985 if (sb->raid_disks <= 0)
988 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
989 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
994 rdev->preferred_minor = sb->md_minor;
995 rdev->data_offset = 0;
996 rdev->new_data_offset = 0;
997 rdev->sb_size = MD_SB_BYTES;
998 rdev->badblocks.shift = -1;
1000 if (sb->level == LEVEL_MULTIPATH)
1003 rdev->desc_nr = sb->this_disk.number;
1009 mdp_super_t *refsb = page_address(refdev->sb_page);
1010 if (!uuid_equal(refsb, sb)) {
1011 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1012 b, bdevname(refdev->bdev,b2));
1015 if (!sb_equal(refsb, sb)) {
1016 printk(KERN_WARNING "md: %s has same UUID"
1017 " but different superblock to %s\n",
1018 b, bdevname(refdev->bdev, b2));
1022 ev2 = md_event(refsb);
1028 rdev->sectors = rdev->sb_start;
1029 /* Limit to 4TB as metadata cannot record more than that.
1030 * (not needed for Linear and RAID0 as metadata doesn't
1033 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1034 rdev->sectors = (2ULL << 32) - 2;
1036 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1037 /* "this cannot possibly happen" ... */
1045 * validate_super for 0.90.0
1047 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1050 mdp_super_t *sb = page_address(rdev->sb_page);
1051 __u64 ev1 = md_event(sb);
1053 rdev->raid_disk = -1;
1054 clear_bit(Faulty, &rdev->flags);
1055 clear_bit(In_sync, &rdev->flags);
1056 clear_bit(Bitmap_sync, &rdev->flags);
1057 clear_bit(WriteMostly, &rdev->flags);
1059 if (mddev->raid_disks == 0) {
1060 mddev->major_version = 0;
1061 mddev->minor_version = sb->minor_version;
1062 mddev->patch_version = sb->patch_version;
1063 mddev->external = 0;
1064 mddev->chunk_sectors = sb->chunk_size >> 9;
1065 mddev->ctime = sb->ctime;
1066 mddev->utime = sb->utime;
1067 mddev->level = sb->level;
1068 mddev->clevel[0] = 0;
1069 mddev->layout = sb->layout;
1070 mddev->raid_disks = sb->raid_disks;
1071 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1072 mddev->events = ev1;
1073 mddev->bitmap_info.offset = 0;
1074 mddev->bitmap_info.space = 0;
1075 /* bitmap can use 60 K after the 4K superblocks */
1076 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1077 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1078 mddev->reshape_backwards = 0;
1080 if (mddev->minor_version >= 91) {
1081 mddev->reshape_position = sb->reshape_position;
1082 mddev->delta_disks = sb->delta_disks;
1083 mddev->new_level = sb->new_level;
1084 mddev->new_layout = sb->new_layout;
1085 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1086 if (mddev->delta_disks < 0)
1087 mddev->reshape_backwards = 1;
1089 mddev->reshape_position = MaxSector;
1090 mddev->delta_disks = 0;
1091 mddev->new_level = mddev->level;
1092 mddev->new_layout = mddev->layout;
1093 mddev->new_chunk_sectors = mddev->chunk_sectors;
1096 if (sb->state & (1<<MD_SB_CLEAN))
1097 mddev->recovery_cp = MaxSector;
1099 if (sb->events_hi == sb->cp_events_hi &&
1100 sb->events_lo == sb->cp_events_lo) {
1101 mddev->recovery_cp = sb->recovery_cp;
1103 mddev->recovery_cp = 0;
1106 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1107 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1108 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1109 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1111 mddev->max_disks = MD_SB_DISKS;
1113 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1114 mddev->bitmap_info.file == NULL) {
1115 mddev->bitmap_info.offset =
1116 mddev->bitmap_info.default_offset;
1117 mddev->bitmap_info.space =
1118 mddev->bitmap_info.default_space;
1121 } else if (mddev->pers == NULL) {
1122 /* Insist on good event counter while assembling, except
1123 * for spares (which don't need an event count) */
1125 if (sb->disks[rdev->desc_nr].state & (
1126 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1127 if (ev1 < mddev->events)
1129 } else if (mddev->bitmap) {
1130 /* if adding to array with a bitmap, then we can accept an
1131 * older device ... but not too old.
1133 if (ev1 < mddev->bitmap->events_cleared)
1135 if (ev1 < mddev->events)
1136 set_bit(Bitmap_sync, &rdev->flags);
1138 if (ev1 < mddev->events)
1139 /* just a hot-add of a new device, leave raid_disk at -1 */
1143 if (mddev->level != LEVEL_MULTIPATH) {
1144 desc = sb->disks + rdev->desc_nr;
1146 if (desc->state & (1<<MD_DISK_FAULTY))
1147 set_bit(Faulty, &rdev->flags);
1148 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1149 desc->raid_disk < mddev->raid_disks */) {
1150 set_bit(In_sync, &rdev->flags);
1151 rdev->raid_disk = desc->raid_disk;
1152 rdev->saved_raid_disk = desc->raid_disk;
1153 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1154 /* active but not in sync implies recovery up to
1155 * reshape position. We don't know exactly where
1156 * that is, so set to zero for now */
1157 if (mddev->minor_version >= 91) {
1158 rdev->recovery_offset = 0;
1159 rdev->raid_disk = desc->raid_disk;
1162 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1163 set_bit(WriteMostly, &rdev->flags);
1164 } else /* MULTIPATH are always insync */
1165 set_bit(In_sync, &rdev->flags);
1170 * sync_super for 0.90.0
1172 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1175 struct md_rdev *rdev2;
1176 int next_spare = mddev->raid_disks;
1178 /* make rdev->sb match mddev data..
1181 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1182 * 3/ any empty disks < next_spare become removed
1184 * disks[0] gets initialised to REMOVED because
1185 * we cannot be sure from other fields if it has
1186 * been initialised or not.
1189 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1191 rdev->sb_size = MD_SB_BYTES;
1193 sb = page_address(rdev->sb_page);
1195 memset(sb, 0, sizeof(*sb));
1197 sb->md_magic = MD_SB_MAGIC;
1198 sb->major_version = mddev->major_version;
1199 sb->patch_version = mddev->patch_version;
1200 sb->gvalid_words = 0; /* ignored */
1201 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1202 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1203 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1204 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1206 sb->ctime = mddev->ctime;
1207 sb->level = mddev->level;
1208 sb->size = mddev->dev_sectors / 2;
1209 sb->raid_disks = mddev->raid_disks;
1210 sb->md_minor = mddev->md_minor;
1211 sb->not_persistent = 0;
1212 sb->utime = mddev->utime;
1214 sb->events_hi = (mddev->events>>32);
1215 sb->events_lo = (u32)mddev->events;
1217 if (mddev->reshape_position == MaxSector)
1218 sb->minor_version = 90;
1220 sb->minor_version = 91;
1221 sb->reshape_position = mddev->reshape_position;
1222 sb->new_level = mddev->new_level;
1223 sb->delta_disks = mddev->delta_disks;
1224 sb->new_layout = mddev->new_layout;
1225 sb->new_chunk = mddev->new_chunk_sectors << 9;
1227 mddev->minor_version = sb->minor_version;
1230 sb->recovery_cp = mddev->recovery_cp;
1231 sb->cp_events_hi = (mddev->events>>32);
1232 sb->cp_events_lo = (u32)mddev->events;
1233 if (mddev->recovery_cp == MaxSector)
1234 sb->state = (1<< MD_SB_CLEAN);
1236 sb->recovery_cp = 0;
1238 sb->layout = mddev->layout;
1239 sb->chunk_size = mddev->chunk_sectors << 9;
1241 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1242 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1244 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1245 rdev_for_each(rdev2, mddev) {
1248 int is_active = test_bit(In_sync, &rdev2->flags);
1250 if (rdev2->raid_disk >= 0 &&
1251 sb->minor_version >= 91)
1252 /* we have nowhere to store the recovery_offset,
1253 * but if it is not below the reshape_position,
1254 * we can piggy-back on that.
1257 if (rdev2->raid_disk < 0 ||
1258 test_bit(Faulty, &rdev2->flags))
1261 desc_nr = rdev2->raid_disk;
1263 desc_nr = next_spare++;
1264 rdev2->desc_nr = desc_nr;
1265 d = &sb->disks[rdev2->desc_nr];
1267 d->number = rdev2->desc_nr;
1268 d->major = MAJOR(rdev2->bdev->bd_dev);
1269 d->minor = MINOR(rdev2->bdev->bd_dev);
1271 d->raid_disk = rdev2->raid_disk;
1273 d->raid_disk = rdev2->desc_nr; /* compatibility */
1274 if (test_bit(Faulty, &rdev2->flags))
1275 d->state = (1<<MD_DISK_FAULTY);
1276 else if (is_active) {
1277 d->state = (1<<MD_DISK_ACTIVE);
1278 if (test_bit(In_sync, &rdev2->flags))
1279 d->state |= (1<<MD_DISK_SYNC);
1287 if (test_bit(WriteMostly, &rdev2->flags))
1288 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1290 /* now set the "removed" and "faulty" bits on any missing devices */
1291 for (i=0 ; i < mddev->raid_disks ; i++) {
1292 mdp_disk_t *d = &sb->disks[i];
1293 if (d->state == 0 && d->number == 0) {
1296 d->state = (1<<MD_DISK_REMOVED);
1297 d->state |= (1<<MD_DISK_FAULTY);
1301 sb->nr_disks = nr_disks;
1302 sb->active_disks = active;
1303 sb->working_disks = working;
1304 sb->failed_disks = failed;
1305 sb->spare_disks = spare;
1307 sb->this_disk = sb->disks[rdev->desc_nr];
1308 sb->sb_csum = calc_sb_csum(sb);
1312 * rdev_size_change for 0.90.0
1314 static unsigned long long
1315 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1317 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1318 return 0; /* component must fit device */
1319 if (rdev->mddev->bitmap_info.offset)
1320 return 0; /* can't move bitmap */
1321 rdev->sb_start = calc_dev_sboffset(rdev);
1322 if (!num_sectors || num_sectors > rdev->sb_start)
1323 num_sectors = rdev->sb_start;
1324 /* Limit to 4TB as metadata cannot record more than that.
1325 * 4TB == 2^32 KB, or 2*2^32 sectors.
1327 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1328 num_sectors = (2ULL << 32) - 2;
1329 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1331 md_super_wait(rdev->mddev);
1336 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1338 /* non-zero offset changes not possible with v0.90 */
1339 return new_offset == 0;
1343 * version 1 superblock
1346 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1350 unsigned long long newcsum;
1351 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1352 __le32 *isuper = (__le32*)sb;
1354 disk_csum = sb->sb_csum;
1357 for (; size >= 4; size -= 4)
1358 newcsum += le32_to_cpu(*isuper++);
1361 newcsum += le16_to_cpu(*(__le16*) isuper);
1363 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1364 sb->sb_csum = disk_csum;
1365 return cpu_to_le32(csum);
1368 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1370 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1372 struct mdp_superblock_1 *sb;
1376 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1380 * Calculate the position of the superblock in 512byte sectors.
1381 * It is always aligned to a 4K boundary and
1382 * depeding on minor_version, it can be:
1383 * 0: At least 8K, but less than 12K, from end of device
1384 * 1: At start of device
1385 * 2: 4K from start of device.
1387 switch(minor_version) {
1389 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1391 sb_start &= ~(sector_t)(4*2-1);
1402 rdev->sb_start = sb_start;
1404 /* superblock is rarely larger than 1K, but it can be larger,
1405 * and it is safe to read 4k, so we do that
1407 ret = read_disk_sb(rdev, 4096);
1408 if (ret) return ret;
1410 sb = page_address(rdev->sb_page);
1412 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1413 sb->major_version != cpu_to_le32(1) ||
1414 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1415 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1416 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1419 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1420 printk("md: invalid superblock checksum on %s\n",
1421 bdevname(rdev->bdev,b));
1424 if (le64_to_cpu(sb->data_size) < 10) {
1425 printk("md: data_size too small on %s\n",
1426 bdevname(rdev->bdev,b));
1431 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1432 /* Some padding is non-zero, might be a new feature */
1435 rdev->preferred_minor = 0xffff;
1436 rdev->data_offset = le64_to_cpu(sb->data_offset);
1437 rdev->new_data_offset = rdev->data_offset;
1438 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1439 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1440 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1441 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1443 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1444 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1445 if (rdev->sb_size & bmask)
1446 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1449 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1452 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1455 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1458 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1460 if (!rdev->bb_page) {
1461 rdev->bb_page = alloc_page(GFP_KERNEL);
1465 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1466 rdev->badblocks.count == 0) {
1467 /* need to load the bad block list.
1468 * Currently we limit it to one page.
1474 int sectors = le16_to_cpu(sb->bblog_size);
1475 if (sectors > (PAGE_SIZE / 512))
1477 offset = le32_to_cpu(sb->bblog_offset);
1480 bb_sector = (long long)offset;
1481 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1482 rdev->bb_page, READ, true))
1484 bbp = (u64 *)page_address(rdev->bb_page);
1485 rdev->badblocks.shift = sb->bblog_shift;
1486 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1487 u64 bb = le64_to_cpu(*bbp);
1488 int count = bb & (0x3ff);
1489 u64 sector = bb >> 10;
1490 sector <<= sb->bblog_shift;
1491 count <<= sb->bblog_shift;
1494 if (md_set_badblocks(&rdev->badblocks,
1495 sector, count, 1) == 0)
1498 } else if (sb->bblog_offset != 0)
1499 rdev->badblocks.shift = 0;
1505 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1507 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1508 sb->level != refsb->level ||
1509 sb->layout != refsb->layout ||
1510 sb->chunksize != refsb->chunksize) {
1511 printk(KERN_WARNING "md: %s has strangely different"
1512 " superblock to %s\n",
1513 bdevname(rdev->bdev,b),
1514 bdevname(refdev->bdev,b2));
1517 ev1 = le64_to_cpu(sb->events);
1518 ev2 = le64_to_cpu(refsb->events);
1525 if (minor_version) {
1526 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1527 sectors -= rdev->data_offset;
1529 sectors = rdev->sb_start;
1530 if (sectors < le64_to_cpu(sb->data_size))
1532 rdev->sectors = le64_to_cpu(sb->data_size);
1536 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1538 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1539 __u64 ev1 = le64_to_cpu(sb->events);
1541 rdev->raid_disk = -1;
1542 clear_bit(Faulty, &rdev->flags);
1543 clear_bit(In_sync, &rdev->flags);
1544 clear_bit(Bitmap_sync, &rdev->flags);
1545 clear_bit(WriteMostly, &rdev->flags);
1547 if (mddev->raid_disks == 0) {
1548 mddev->major_version = 1;
1549 mddev->patch_version = 0;
1550 mddev->external = 0;
1551 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1552 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1553 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1554 mddev->level = le32_to_cpu(sb->level);
1555 mddev->clevel[0] = 0;
1556 mddev->layout = le32_to_cpu(sb->layout);
1557 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1558 mddev->dev_sectors = le64_to_cpu(sb->size);
1559 mddev->events = ev1;
1560 mddev->bitmap_info.offset = 0;
1561 mddev->bitmap_info.space = 0;
1562 /* Default location for bitmap is 1K after superblock
1563 * using 3K - total of 4K
1565 mddev->bitmap_info.default_offset = 1024 >> 9;
1566 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1567 mddev->reshape_backwards = 0;
1569 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1570 memcpy(mddev->uuid, sb->set_uuid, 16);
1572 mddev->max_disks = (4096-256)/2;
1574 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1575 mddev->bitmap_info.file == NULL) {
1576 mddev->bitmap_info.offset =
1577 (__s32)le32_to_cpu(sb->bitmap_offset);
1578 /* Metadata doesn't record how much space is available.
1579 * For 1.0, we assume we can use up to the superblock
1580 * if before, else to 4K beyond superblock.
1581 * For others, assume no change is possible.
1583 if (mddev->minor_version > 0)
1584 mddev->bitmap_info.space = 0;
1585 else if (mddev->bitmap_info.offset > 0)
1586 mddev->bitmap_info.space =
1587 8 - mddev->bitmap_info.offset;
1589 mddev->bitmap_info.space =
1590 -mddev->bitmap_info.offset;
1593 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1594 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1595 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1596 mddev->new_level = le32_to_cpu(sb->new_level);
1597 mddev->new_layout = le32_to_cpu(sb->new_layout);
1598 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1599 if (mddev->delta_disks < 0 ||
1600 (mddev->delta_disks == 0 &&
1601 (le32_to_cpu(sb->feature_map)
1602 & MD_FEATURE_RESHAPE_BACKWARDS)))
1603 mddev->reshape_backwards = 1;
1605 mddev->reshape_position = MaxSector;
1606 mddev->delta_disks = 0;
1607 mddev->new_level = mddev->level;
1608 mddev->new_layout = mddev->layout;
1609 mddev->new_chunk_sectors = mddev->chunk_sectors;
1612 } else if (mddev->pers == NULL) {
1613 /* Insist of good event counter while assembling, except for
1614 * spares (which don't need an event count) */
1616 if (rdev->desc_nr >= 0 &&
1617 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1618 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1619 if (ev1 < mddev->events)
1621 } else if (mddev->bitmap) {
1622 /* If adding to array with a bitmap, then we can accept an
1623 * older device, but not too old.
1625 if (ev1 < mddev->bitmap->events_cleared)
1627 if (ev1 < mddev->events)
1628 set_bit(Bitmap_sync, &rdev->flags);
1630 if (ev1 < mddev->events)
1631 /* just a hot-add of a new device, leave raid_disk at -1 */
1634 if (mddev->level != LEVEL_MULTIPATH) {
1636 if (rdev->desc_nr < 0 ||
1637 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1641 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1643 case 0xffff: /* spare */
1645 case 0xfffe: /* faulty */
1646 set_bit(Faulty, &rdev->flags);
1649 rdev->saved_raid_disk = role;
1650 if ((le32_to_cpu(sb->feature_map) &
1651 MD_FEATURE_RECOVERY_OFFSET)) {
1652 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1653 if (!(le32_to_cpu(sb->feature_map) &
1654 MD_FEATURE_RECOVERY_BITMAP))
1655 rdev->saved_raid_disk = -1;
1657 set_bit(In_sync, &rdev->flags);
1658 rdev->raid_disk = role;
1661 if (sb->devflags & WriteMostly1)
1662 set_bit(WriteMostly, &rdev->flags);
1663 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1664 set_bit(Replacement, &rdev->flags);
1665 } else /* MULTIPATH are always insync */
1666 set_bit(In_sync, &rdev->flags);
1671 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1673 struct mdp_superblock_1 *sb;
1674 struct md_rdev *rdev2;
1676 /* make rdev->sb match mddev and rdev data. */
1678 sb = page_address(rdev->sb_page);
1680 sb->feature_map = 0;
1682 sb->recovery_offset = cpu_to_le64(0);
1683 memset(sb->pad3, 0, sizeof(sb->pad3));
1685 sb->utime = cpu_to_le64((__u64)mddev->utime);
1686 sb->events = cpu_to_le64(mddev->events);
1688 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1690 sb->resync_offset = cpu_to_le64(0);
1692 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1694 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1695 sb->size = cpu_to_le64(mddev->dev_sectors);
1696 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1697 sb->level = cpu_to_le32(mddev->level);
1698 sb->layout = cpu_to_le32(mddev->layout);
1700 if (test_bit(WriteMostly, &rdev->flags))
1701 sb->devflags |= WriteMostly1;
1703 sb->devflags &= ~WriteMostly1;
1704 sb->data_offset = cpu_to_le64(rdev->data_offset);
1705 sb->data_size = cpu_to_le64(rdev->sectors);
1707 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1708 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1709 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1712 if (rdev->raid_disk >= 0 &&
1713 !test_bit(In_sync, &rdev->flags)) {
1715 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1716 sb->recovery_offset =
1717 cpu_to_le64(rdev->recovery_offset);
1718 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1720 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1722 if (test_bit(Replacement, &rdev->flags))
1724 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1726 if (mddev->reshape_position != MaxSector) {
1727 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1728 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1729 sb->new_layout = cpu_to_le32(mddev->new_layout);
1730 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1731 sb->new_level = cpu_to_le32(mddev->new_level);
1732 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1733 if (mddev->delta_disks == 0 &&
1734 mddev->reshape_backwards)
1736 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1737 if (rdev->new_data_offset != rdev->data_offset) {
1739 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1740 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1741 - rdev->data_offset));
1745 if (rdev->badblocks.count == 0)
1746 /* Nothing to do for bad blocks*/ ;
1747 else if (sb->bblog_offset == 0)
1748 /* Cannot record bad blocks on this device */
1749 md_error(mddev, rdev);
1751 struct badblocks *bb = &rdev->badblocks;
1752 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1754 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1759 seq = read_seqbegin(&bb->lock);
1761 memset(bbp, 0xff, PAGE_SIZE);
1763 for (i = 0 ; i < bb->count ; i++) {
1764 u64 internal_bb = p[i];
1765 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1766 | BB_LEN(internal_bb));
1767 bbp[i] = cpu_to_le64(store_bb);
1770 if (read_seqretry(&bb->lock, seq))
1773 bb->sector = (rdev->sb_start +
1774 (int)le32_to_cpu(sb->bblog_offset));
1775 bb->size = le16_to_cpu(sb->bblog_size);
1780 rdev_for_each(rdev2, mddev)
1781 if (rdev2->desc_nr+1 > max_dev)
1782 max_dev = rdev2->desc_nr+1;
1784 if (max_dev > le32_to_cpu(sb->max_dev)) {
1786 sb->max_dev = cpu_to_le32(max_dev);
1787 rdev->sb_size = max_dev * 2 + 256;
1788 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1789 if (rdev->sb_size & bmask)
1790 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1792 max_dev = le32_to_cpu(sb->max_dev);
1794 for (i=0; i<max_dev;i++)
1795 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1797 rdev_for_each(rdev2, mddev) {
1799 if (test_bit(Faulty, &rdev2->flags))
1800 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1801 else if (test_bit(In_sync, &rdev2->flags))
1802 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1803 else if (rdev2->raid_disk >= 0)
1804 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1806 sb->dev_roles[i] = cpu_to_le16(0xffff);
1809 sb->sb_csum = calc_sb_1_csum(sb);
1812 static unsigned long long
1813 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1815 struct mdp_superblock_1 *sb;
1816 sector_t max_sectors;
1817 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1818 return 0; /* component must fit device */
1819 if (rdev->data_offset != rdev->new_data_offset)
1820 return 0; /* too confusing */
1821 if (rdev->sb_start < rdev->data_offset) {
1822 /* minor versions 1 and 2; superblock before data */
1823 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1824 max_sectors -= rdev->data_offset;
1825 if (!num_sectors || num_sectors > max_sectors)
1826 num_sectors = max_sectors;
1827 } else if (rdev->mddev->bitmap_info.offset) {
1828 /* minor version 0 with bitmap we can't move */
1831 /* minor version 0; superblock after data */
1833 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1834 sb_start &= ~(sector_t)(4*2 - 1);
1835 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1836 if (!num_sectors || num_sectors > max_sectors)
1837 num_sectors = max_sectors;
1838 rdev->sb_start = sb_start;
1840 sb = page_address(rdev->sb_page);
1841 sb->data_size = cpu_to_le64(num_sectors);
1842 sb->super_offset = rdev->sb_start;
1843 sb->sb_csum = calc_sb_1_csum(sb);
1844 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1846 md_super_wait(rdev->mddev);
1852 super_1_allow_new_offset(struct md_rdev *rdev,
1853 unsigned long long new_offset)
1855 /* All necessary checks on new >= old have been done */
1856 struct bitmap *bitmap;
1857 if (new_offset >= rdev->data_offset)
1860 /* with 1.0 metadata, there is no metadata to tread on
1861 * so we can always move back */
1862 if (rdev->mddev->minor_version == 0)
1865 /* otherwise we must be sure not to step on
1866 * any metadata, so stay:
1867 * 36K beyond start of superblock
1868 * beyond end of badblocks
1869 * beyond write-intent bitmap
1871 if (rdev->sb_start + (32+4)*2 > new_offset)
1873 bitmap = rdev->mddev->bitmap;
1874 if (bitmap && !rdev->mddev->bitmap_info.file &&
1875 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1876 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1878 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1884 static struct super_type super_types[] = {
1887 .owner = THIS_MODULE,
1888 .load_super = super_90_load,
1889 .validate_super = super_90_validate,
1890 .sync_super = super_90_sync,
1891 .rdev_size_change = super_90_rdev_size_change,
1892 .allow_new_offset = super_90_allow_new_offset,
1896 .owner = THIS_MODULE,
1897 .load_super = super_1_load,
1898 .validate_super = super_1_validate,
1899 .sync_super = super_1_sync,
1900 .rdev_size_change = super_1_rdev_size_change,
1901 .allow_new_offset = super_1_allow_new_offset,
1905 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1907 if (mddev->sync_super) {
1908 mddev->sync_super(mddev, rdev);
1912 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1914 super_types[mddev->major_version].sync_super(mddev, rdev);
1917 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1919 struct md_rdev *rdev, *rdev2;
1922 rdev_for_each_rcu(rdev, mddev1)
1923 rdev_for_each_rcu(rdev2, mddev2)
1924 if (rdev->bdev->bd_contains ==
1925 rdev2->bdev->bd_contains) {
1933 static LIST_HEAD(pending_raid_disks);
1936 * Try to register data integrity profile for an mddev
1938 * This is called when an array is started and after a disk has been kicked
1939 * from the array. It only succeeds if all working and active component devices
1940 * are integrity capable with matching profiles.
1942 int md_integrity_register(struct mddev *mddev)
1944 struct md_rdev *rdev, *reference = NULL;
1946 if (list_empty(&mddev->disks))
1947 return 0; /* nothing to do */
1948 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1949 return 0; /* shouldn't register, or already is */
1950 rdev_for_each(rdev, mddev) {
1951 /* skip spares and non-functional disks */
1952 if (test_bit(Faulty, &rdev->flags))
1954 if (rdev->raid_disk < 0)
1957 /* Use the first rdev as the reference */
1961 /* does this rdev's profile match the reference profile? */
1962 if (blk_integrity_compare(reference->bdev->bd_disk,
1963 rdev->bdev->bd_disk) < 0)
1966 if (!reference || !bdev_get_integrity(reference->bdev))
1969 * All component devices are integrity capable and have matching
1970 * profiles, register the common profile for the md device.
1972 if (blk_integrity_register(mddev->gendisk,
1973 bdev_get_integrity(reference->bdev)) != 0) {
1974 printk(KERN_ERR "md: failed to register integrity for %s\n",
1978 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1979 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1980 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1986 EXPORT_SYMBOL(md_integrity_register);
1988 /* Disable data integrity if non-capable/non-matching disk is being added */
1989 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1991 struct blk_integrity *bi_rdev;
1992 struct blk_integrity *bi_mddev;
1994 if (!mddev->gendisk)
1997 bi_rdev = bdev_get_integrity(rdev->bdev);
1998 bi_mddev = blk_get_integrity(mddev->gendisk);
2000 if (!bi_mddev) /* nothing to do */
2002 if (rdev->raid_disk < 0) /* skip spares */
2004 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2005 rdev->bdev->bd_disk) >= 0)
2007 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2008 blk_integrity_unregister(mddev->gendisk);
2010 EXPORT_SYMBOL(md_integrity_add_rdev);
2012 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2014 char b[BDEVNAME_SIZE];
2019 /* prevent duplicates */
2020 if (find_rdev(mddev, rdev->bdev->bd_dev))
2023 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2024 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2025 rdev->sectors < mddev->dev_sectors)) {
2027 /* Cannot change size, so fail
2028 * If mddev->level <= 0, then we don't care
2029 * about aligning sizes (e.g. linear)
2031 if (mddev->level > 0)
2034 mddev->dev_sectors = rdev->sectors;
2037 /* Verify rdev->desc_nr is unique.
2038 * If it is -1, assign a free number, else
2039 * check number is not in use
2042 if (rdev->desc_nr < 0) {
2045 choice = mddev->raid_disks;
2046 while (find_rdev_nr_rcu(mddev, choice))
2048 rdev->desc_nr = choice;
2050 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2056 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2057 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2058 mdname(mddev), mddev->max_disks);
2061 bdevname(rdev->bdev,b);
2062 while ( (s=strchr(b, '/')) != NULL)
2065 rdev->mddev = mddev;
2066 printk(KERN_INFO "md: bind<%s>\n", b);
2068 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2071 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2072 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2073 /* failure here is OK */;
2074 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2076 list_add_rcu(&rdev->same_set, &mddev->disks);
2077 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2079 /* May as well allow recovery to be retried once */
2080 mddev->recovery_disabled++;
2085 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2090 static void md_delayed_delete(struct work_struct *ws)
2092 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2093 kobject_del(&rdev->kobj);
2094 kobject_put(&rdev->kobj);
2097 static void unbind_rdev_from_array(struct md_rdev *rdev)
2099 char b[BDEVNAME_SIZE];
2101 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2102 list_del_rcu(&rdev->same_set);
2103 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2105 sysfs_remove_link(&rdev->kobj, "block");
2106 sysfs_put(rdev->sysfs_state);
2107 rdev->sysfs_state = NULL;
2108 rdev->badblocks.count = 0;
2109 /* We need to delay this, otherwise we can deadlock when
2110 * writing to 'remove' to "dev/state". We also need
2111 * to delay it due to rcu usage.
2114 INIT_WORK(&rdev->del_work, md_delayed_delete);
2115 kobject_get(&rdev->kobj);
2116 queue_work(md_misc_wq, &rdev->del_work);
2120 * prevent the device from being mounted, repartitioned or
2121 * otherwise reused by a RAID array (or any other kernel
2122 * subsystem), by bd_claiming the device.
2124 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2127 struct block_device *bdev;
2128 char b[BDEVNAME_SIZE];
2130 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2131 shared ? (struct md_rdev *)lock_rdev : rdev);
2133 printk(KERN_ERR "md: could not open %s.\n",
2134 __bdevname(dev, b));
2135 return PTR_ERR(bdev);
2141 static void unlock_rdev(struct md_rdev *rdev)
2143 struct block_device *bdev = rdev->bdev;
2145 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2148 void md_autodetect_dev(dev_t dev);
2150 static void export_rdev(struct md_rdev *rdev)
2152 char b[BDEVNAME_SIZE];
2154 printk(KERN_INFO "md: export_rdev(%s)\n",
2155 bdevname(rdev->bdev,b));
2156 md_rdev_clear(rdev);
2158 if (test_bit(AutoDetected, &rdev->flags))
2159 md_autodetect_dev(rdev->bdev->bd_dev);
2162 kobject_put(&rdev->kobj);
2165 static void kick_rdev_from_array(struct md_rdev *rdev)
2167 unbind_rdev_from_array(rdev);
2171 static void export_array(struct mddev *mddev)
2173 struct md_rdev *rdev;
2175 while (!list_empty(&mddev->disks)) {
2176 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2178 kick_rdev_from_array(rdev);
2180 mddev->raid_disks = 0;
2181 mddev->major_version = 0;
2184 static void sync_sbs(struct mddev *mddev, int nospares)
2186 /* Update each superblock (in-memory image), but
2187 * if we are allowed to, skip spares which already
2188 * have the right event counter, or have one earlier
2189 * (which would mean they aren't being marked as dirty
2190 * with the rest of the array)
2192 struct md_rdev *rdev;
2193 rdev_for_each(rdev, mddev) {
2194 if (rdev->sb_events == mddev->events ||
2196 rdev->raid_disk < 0 &&
2197 rdev->sb_events+1 == mddev->events)) {
2198 /* Don't update this superblock */
2199 rdev->sb_loaded = 2;
2201 sync_super(mddev, rdev);
2202 rdev->sb_loaded = 1;
2207 static void md_update_sb(struct mddev *mddev, int force_change)
2209 struct md_rdev *rdev;
2212 int any_badblocks_changed = 0;
2216 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2220 /* First make sure individual recovery_offsets are correct */
2221 rdev_for_each(rdev, mddev) {
2222 if (rdev->raid_disk >= 0 &&
2223 mddev->delta_disks >= 0 &&
2224 !test_bit(In_sync, &rdev->flags) &&
2225 mddev->curr_resync_completed > rdev->recovery_offset)
2226 rdev->recovery_offset = mddev->curr_resync_completed;
2229 if (!mddev->persistent) {
2230 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2231 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2232 if (!mddev->external) {
2233 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2234 rdev_for_each(rdev, mddev) {
2235 if (rdev->badblocks.changed) {
2236 rdev->badblocks.changed = 0;
2237 md_ack_all_badblocks(&rdev->badblocks);
2238 md_error(mddev, rdev);
2240 clear_bit(Blocked, &rdev->flags);
2241 clear_bit(BlockedBadBlocks, &rdev->flags);
2242 wake_up(&rdev->blocked_wait);
2245 wake_up(&mddev->sb_wait);
2249 spin_lock(&mddev->lock);
2251 mddev->utime = get_seconds();
2253 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2255 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2256 /* just a clean<-> dirty transition, possibly leave spares alone,
2257 * though if events isn't the right even/odd, we will have to do
2263 if (mddev->degraded)
2264 /* If the array is degraded, then skipping spares is both
2265 * dangerous and fairly pointless.
2266 * Dangerous because a device that was removed from the array
2267 * might have a event_count that still looks up-to-date,
2268 * so it can be re-added without a resync.
2269 * Pointless because if there are any spares to skip,
2270 * then a recovery will happen and soon that array won't
2271 * be degraded any more and the spare can go back to sleep then.
2275 sync_req = mddev->in_sync;
2277 /* If this is just a dirty<->clean transition, and the array is clean
2278 * and 'events' is odd, we can roll back to the previous clean state */
2280 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2281 && mddev->can_decrease_events
2282 && mddev->events != 1) {
2284 mddev->can_decrease_events = 0;
2286 /* otherwise we have to go forward and ... */
2288 mddev->can_decrease_events = nospares;
2292 * This 64-bit counter should never wrap.
2293 * Either we are in around ~1 trillion A.C., assuming
2294 * 1 reboot per second, or we have a bug...
2296 WARN_ON(mddev->events == 0);
2298 rdev_for_each(rdev, mddev) {
2299 if (rdev->badblocks.changed)
2300 any_badblocks_changed++;
2301 if (test_bit(Faulty, &rdev->flags))
2302 set_bit(FaultRecorded, &rdev->flags);
2305 sync_sbs(mddev, nospares);
2306 spin_unlock(&mddev->lock);
2308 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2309 mdname(mddev), mddev->in_sync);
2311 bitmap_update_sb(mddev->bitmap);
2312 rdev_for_each(rdev, mddev) {
2313 char b[BDEVNAME_SIZE];
2315 if (rdev->sb_loaded != 1)
2316 continue; /* no noise on spare devices */
2318 if (!test_bit(Faulty, &rdev->flags)) {
2319 md_super_write(mddev,rdev,
2320 rdev->sb_start, rdev->sb_size,
2322 pr_debug("md: (write) %s's sb offset: %llu\n",
2323 bdevname(rdev->bdev, b),
2324 (unsigned long long)rdev->sb_start);
2325 rdev->sb_events = mddev->events;
2326 if (rdev->badblocks.size) {
2327 md_super_write(mddev, rdev,
2328 rdev->badblocks.sector,
2329 rdev->badblocks.size << 9,
2331 rdev->badblocks.size = 0;
2335 pr_debug("md: %s (skipping faulty)\n",
2336 bdevname(rdev->bdev, b));
2338 if (mddev->level == LEVEL_MULTIPATH)
2339 /* only need to write one superblock... */
2342 md_super_wait(mddev);
2343 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2345 spin_lock(&mddev->lock);
2346 if (mddev->in_sync != sync_req ||
2347 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2348 /* have to write it out again */
2349 spin_unlock(&mddev->lock);
2352 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2353 spin_unlock(&mddev->lock);
2354 wake_up(&mddev->sb_wait);
2355 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2356 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2358 rdev_for_each(rdev, mddev) {
2359 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2360 clear_bit(Blocked, &rdev->flags);
2362 if (any_badblocks_changed)
2363 md_ack_all_badblocks(&rdev->badblocks);
2364 clear_bit(BlockedBadBlocks, &rdev->flags);
2365 wake_up(&rdev->blocked_wait);
2369 /* words written to sysfs files may, or may not, be \n terminated.
2370 * We want to accept with case. For this we use cmd_match.
2372 static int cmd_match(const char *cmd, const char *str)
2374 /* See if cmd, written into a sysfs file, matches
2375 * str. They must either be the same, or cmd can
2376 * have a trailing newline
2378 while (*cmd && *str && *cmd == *str) {
2389 struct rdev_sysfs_entry {
2390 struct attribute attr;
2391 ssize_t (*show)(struct md_rdev *, char *);
2392 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2396 state_show(struct md_rdev *rdev, char *page)
2400 unsigned long flags = ACCESS_ONCE(rdev->flags);
2402 if (test_bit(Faulty, &flags) ||
2403 rdev->badblocks.unacked_exist) {
2404 len+= sprintf(page+len, "%sfaulty",sep);
2407 if (test_bit(In_sync, &flags)) {
2408 len += sprintf(page+len, "%sin_sync",sep);
2411 if (test_bit(WriteMostly, &flags)) {
2412 len += sprintf(page+len, "%swrite_mostly",sep);
2415 if (test_bit(Blocked, &flags) ||
2416 (rdev->badblocks.unacked_exist
2417 && !test_bit(Faulty, &flags))) {
2418 len += sprintf(page+len, "%sblocked", sep);
2421 if (!test_bit(Faulty, &flags) &&
2422 !test_bit(In_sync, &flags)) {
2423 len += sprintf(page+len, "%sspare", sep);
2426 if (test_bit(WriteErrorSeen, &flags)) {
2427 len += sprintf(page+len, "%swrite_error", sep);
2430 if (test_bit(WantReplacement, &flags)) {
2431 len += sprintf(page+len, "%swant_replacement", sep);
2434 if (test_bit(Replacement, &flags)) {
2435 len += sprintf(page+len, "%sreplacement", sep);
2439 return len+sprintf(page+len, "\n");
2443 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2446 * faulty - simulates an error
2447 * remove - disconnects the device
2448 * writemostly - sets write_mostly
2449 * -writemostly - clears write_mostly
2450 * blocked - sets the Blocked flags
2451 * -blocked - clears the Blocked and possibly simulates an error
2452 * insync - sets Insync providing device isn't active
2453 * -insync - clear Insync for a device with a slot assigned,
2454 * so that it gets rebuilt based on bitmap
2455 * write_error - sets WriteErrorSeen
2456 * -write_error - clears WriteErrorSeen
2459 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2460 md_error(rdev->mddev, rdev);
2461 if (test_bit(Faulty, &rdev->flags))
2465 } else if (cmd_match(buf, "remove")) {
2466 if (rdev->raid_disk >= 0)
2469 struct mddev *mddev = rdev->mddev;
2470 kick_rdev_from_array(rdev);
2472 md_update_sb(mddev, 1);
2473 md_new_event(mddev);
2476 } else if (cmd_match(buf, "writemostly")) {
2477 set_bit(WriteMostly, &rdev->flags);
2479 } else if (cmd_match(buf, "-writemostly")) {
2480 clear_bit(WriteMostly, &rdev->flags);
2482 } else if (cmd_match(buf, "blocked")) {
2483 set_bit(Blocked, &rdev->flags);
2485 } else if (cmd_match(buf, "-blocked")) {
2486 if (!test_bit(Faulty, &rdev->flags) &&
2487 rdev->badblocks.unacked_exist) {
2488 /* metadata handler doesn't understand badblocks,
2489 * so we need to fail the device
2491 md_error(rdev->mddev, rdev);
2493 clear_bit(Blocked, &rdev->flags);
2494 clear_bit(BlockedBadBlocks, &rdev->flags);
2495 wake_up(&rdev->blocked_wait);
2496 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2497 md_wakeup_thread(rdev->mddev->thread);
2500 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2501 set_bit(In_sync, &rdev->flags);
2503 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2504 if (rdev->mddev->pers == NULL) {
2505 clear_bit(In_sync, &rdev->flags);
2506 rdev->saved_raid_disk = rdev->raid_disk;
2507 rdev->raid_disk = -1;
2510 } else if (cmd_match(buf, "write_error")) {
2511 set_bit(WriteErrorSeen, &rdev->flags);
2513 } else if (cmd_match(buf, "-write_error")) {
2514 clear_bit(WriteErrorSeen, &rdev->flags);
2516 } else if (cmd_match(buf, "want_replacement")) {
2517 /* Any non-spare device that is not a replacement can
2518 * become want_replacement at any time, but we then need to
2519 * check if recovery is needed.
2521 if (rdev->raid_disk >= 0 &&
2522 !test_bit(Replacement, &rdev->flags))
2523 set_bit(WantReplacement, &rdev->flags);
2524 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2525 md_wakeup_thread(rdev->mddev->thread);
2527 } else if (cmd_match(buf, "-want_replacement")) {
2528 /* Clearing 'want_replacement' is always allowed.
2529 * Once replacements starts it is too late though.
2532 clear_bit(WantReplacement, &rdev->flags);
2533 } else if (cmd_match(buf, "replacement")) {
2534 /* Can only set a device as a replacement when array has not
2535 * yet been started. Once running, replacement is automatic
2536 * from spares, or by assigning 'slot'.
2538 if (rdev->mddev->pers)
2541 set_bit(Replacement, &rdev->flags);
2544 } else if (cmd_match(buf, "-replacement")) {
2545 /* Similarly, can only clear Replacement before start */
2546 if (rdev->mddev->pers)
2549 clear_bit(Replacement, &rdev->flags);
2554 sysfs_notify_dirent_safe(rdev->sysfs_state);
2555 return err ? err : len;
2557 static struct rdev_sysfs_entry rdev_state =
2558 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2561 errors_show(struct md_rdev *rdev, char *page)
2563 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2567 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2570 unsigned long n = simple_strtoul(buf, &e, 10);
2571 if (*buf && (*e == 0 || *e == '\n')) {
2572 atomic_set(&rdev->corrected_errors, n);
2577 static struct rdev_sysfs_entry rdev_errors =
2578 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2581 slot_show(struct md_rdev *rdev, char *page)
2583 if (rdev->raid_disk < 0)
2584 return sprintf(page, "none\n");
2586 return sprintf(page, "%d\n", rdev->raid_disk);
2590 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2594 int slot = simple_strtoul(buf, &e, 10);
2595 if (strncmp(buf, "none", 4)==0)
2597 else if (e==buf || (*e && *e!= '\n'))
2599 if (rdev->mddev->pers && slot == -1) {
2600 /* Setting 'slot' on an active array requires also
2601 * updating the 'rd%d' link, and communicating
2602 * with the personality with ->hot_*_disk.
2603 * For now we only support removing
2604 * failed/spare devices. This normally happens automatically,
2605 * but not when the metadata is externally managed.
2607 if (rdev->raid_disk == -1)
2609 /* personality does all needed checks */
2610 if (rdev->mddev->pers->hot_remove_disk == NULL)
2612 clear_bit(Blocked, &rdev->flags);
2613 remove_and_add_spares(rdev->mddev, rdev);
2614 if (rdev->raid_disk >= 0)
2616 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2617 md_wakeup_thread(rdev->mddev->thread);
2618 } else if (rdev->mddev->pers) {
2619 /* Activating a spare .. or possibly reactivating
2620 * if we ever get bitmaps working here.
2623 if (rdev->raid_disk != -1)
2626 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2629 if (rdev->mddev->pers->hot_add_disk == NULL)
2632 if (slot >= rdev->mddev->raid_disks &&
2633 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2636 rdev->raid_disk = slot;
2637 if (test_bit(In_sync, &rdev->flags))
2638 rdev->saved_raid_disk = slot;
2640 rdev->saved_raid_disk = -1;
2641 clear_bit(In_sync, &rdev->flags);
2642 clear_bit(Bitmap_sync, &rdev->flags);
2643 err = rdev->mddev->pers->
2644 hot_add_disk(rdev->mddev, rdev);
2646 rdev->raid_disk = -1;
2649 sysfs_notify_dirent_safe(rdev->sysfs_state);
2650 if (sysfs_link_rdev(rdev->mddev, rdev))
2651 /* failure here is OK */;
2652 /* don't wakeup anyone, leave that to userspace. */
2654 if (slot >= rdev->mddev->raid_disks &&
2655 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2657 rdev->raid_disk = slot;
2658 /* assume it is working */
2659 clear_bit(Faulty, &rdev->flags);
2660 clear_bit(WriteMostly, &rdev->flags);
2661 set_bit(In_sync, &rdev->flags);
2662 sysfs_notify_dirent_safe(rdev->sysfs_state);
2667 static struct rdev_sysfs_entry rdev_slot =
2668 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2671 offset_show(struct md_rdev *rdev, char *page)
2673 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2677 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2679 unsigned long long offset;
2680 if (kstrtoull(buf, 10, &offset) < 0)
2682 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2684 if (rdev->sectors && rdev->mddev->external)
2685 /* Must set offset before size, so overlap checks
2688 rdev->data_offset = offset;
2689 rdev->new_data_offset = offset;
2693 static struct rdev_sysfs_entry rdev_offset =
2694 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2696 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2698 return sprintf(page, "%llu\n",
2699 (unsigned long long)rdev->new_data_offset);
2702 static ssize_t new_offset_store(struct md_rdev *rdev,
2703 const char *buf, size_t len)
2705 unsigned long long new_offset;
2706 struct mddev *mddev = rdev->mddev;
2708 if (kstrtoull(buf, 10, &new_offset) < 0)
2711 if (mddev->sync_thread ||
2712 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2714 if (new_offset == rdev->data_offset)
2715 /* reset is always permitted */
2717 else if (new_offset > rdev->data_offset) {
2718 /* must not push array size beyond rdev_sectors */
2719 if (new_offset - rdev->data_offset
2720 + mddev->dev_sectors > rdev->sectors)
2723 /* Metadata worries about other space details. */
2725 /* decreasing the offset is inconsistent with a backwards
2728 if (new_offset < rdev->data_offset &&
2729 mddev->reshape_backwards)
2731 /* Increasing offset is inconsistent with forwards
2732 * reshape. reshape_direction should be set to
2733 * 'backwards' first.
2735 if (new_offset > rdev->data_offset &&
2736 !mddev->reshape_backwards)
2739 if (mddev->pers && mddev->persistent &&
2740 !super_types[mddev->major_version]
2741 .allow_new_offset(rdev, new_offset))
2743 rdev->new_data_offset = new_offset;
2744 if (new_offset > rdev->data_offset)
2745 mddev->reshape_backwards = 1;
2746 else if (new_offset < rdev->data_offset)
2747 mddev->reshape_backwards = 0;
2751 static struct rdev_sysfs_entry rdev_new_offset =
2752 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2755 rdev_size_show(struct md_rdev *rdev, char *page)
2757 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2760 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2762 /* check if two start/length pairs overlap */
2770 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2772 unsigned long long blocks;
2775 if (kstrtoull(buf, 10, &blocks) < 0)
2778 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2779 return -EINVAL; /* sector conversion overflow */
2782 if (new != blocks * 2)
2783 return -EINVAL; /* unsigned long long to sector_t overflow */
2790 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2792 struct mddev *my_mddev = rdev->mddev;
2793 sector_t oldsectors = rdev->sectors;
2796 if (strict_blocks_to_sectors(buf, §ors) < 0)
2798 if (rdev->data_offset != rdev->new_data_offset)
2799 return -EINVAL; /* too confusing */
2800 if (my_mddev->pers && rdev->raid_disk >= 0) {
2801 if (my_mddev->persistent) {
2802 sectors = super_types[my_mddev->major_version].
2803 rdev_size_change(rdev, sectors);
2806 } else if (!sectors)
2807 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2809 if (!my_mddev->pers->resize)
2810 /* Cannot change size for RAID0 or Linear etc */
2813 if (sectors < my_mddev->dev_sectors)
2814 return -EINVAL; /* component must fit device */
2816 rdev->sectors = sectors;
2817 if (sectors > oldsectors && my_mddev->external) {
2818 /* Need to check that all other rdevs with the same
2819 * ->bdev do not overlap. 'rcu' is sufficient to walk
2820 * the rdev lists safely.
2821 * This check does not provide a hard guarantee, it
2822 * just helps avoid dangerous mistakes.
2824 struct mddev *mddev;
2826 struct list_head *tmp;
2829 for_each_mddev(mddev, tmp) {
2830 struct md_rdev *rdev2;
2832 rdev_for_each(rdev2, mddev)
2833 if (rdev->bdev == rdev2->bdev &&
2835 overlaps(rdev->data_offset, rdev->sectors,
2848 /* Someone else could have slipped in a size
2849 * change here, but doing so is just silly.
2850 * We put oldsectors back because we *know* it is
2851 * safe, and trust userspace not to race with
2854 rdev->sectors = oldsectors;
2861 static struct rdev_sysfs_entry rdev_size =
2862 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2864 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2866 unsigned long long recovery_start = rdev->recovery_offset;
2868 if (test_bit(In_sync, &rdev->flags) ||
2869 recovery_start == MaxSector)
2870 return sprintf(page, "none\n");
2872 return sprintf(page, "%llu\n", recovery_start);
2875 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2877 unsigned long long recovery_start;
2879 if (cmd_match(buf, "none"))
2880 recovery_start = MaxSector;
2881 else if (kstrtoull(buf, 10, &recovery_start))
2884 if (rdev->mddev->pers &&
2885 rdev->raid_disk >= 0)
2888 rdev->recovery_offset = recovery_start;
2889 if (recovery_start == MaxSector)
2890 set_bit(In_sync, &rdev->flags);
2892 clear_bit(In_sync, &rdev->flags);
2896 static struct rdev_sysfs_entry rdev_recovery_start =
2897 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2900 badblocks_show(struct badblocks *bb, char *page, int unack);
2902 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2904 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2906 return badblocks_show(&rdev->badblocks, page, 0);
2908 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2910 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2911 /* Maybe that ack was all we needed */
2912 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2913 wake_up(&rdev->blocked_wait);
2916 static struct rdev_sysfs_entry rdev_bad_blocks =
2917 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2919 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2921 return badblocks_show(&rdev->badblocks, page, 1);
2923 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2925 return badblocks_store(&rdev->badblocks, page, len, 1);
2927 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2928 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2930 static struct attribute *rdev_default_attrs[] = {
2935 &rdev_new_offset.attr,
2937 &rdev_recovery_start.attr,
2938 &rdev_bad_blocks.attr,
2939 &rdev_unack_bad_blocks.attr,
2943 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2945 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2946 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2952 return entry->show(rdev, page);
2956 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2957 const char *page, size_t length)
2959 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2960 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2962 struct mddev *mddev = rdev->mddev;
2966 if (!capable(CAP_SYS_ADMIN))
2968 rv = mddev ? mddev_lock(mddev): -EBUSY;
2970 if (rdev->mddev == NULL)
2973 rv = entry->store(rdev, page, length);
2974 mddev_unlock(mddev);
2979 static void rdev_free(struct kobject *ko)
2981 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
2984 static const struct sysfs_ops rdev_sysfs_ops = {
2985 .show = rdev_attr_show,
2986 .store = rdev_attr_store,
2988 static struct kobj_type rdev_ktype = {
2989 .release = rdev_free,
2990 .sysfs_ops = &rdev_sysfs_ops,
2991 .default_attrs = rdev_default_attrs,
2994 int md_rdev_init(struct md_rdev *rdev)
2997 rdev->saved_raid_disk = -1;
2998 rdev->raid_disk = -1;
3000 rdev->data_offset = 0;
3001 rdev->new_data_offset = 0;
3002 rdev->sb_events = 0;
3003 rdev->last_read_error.tv_sec = 0;
3004 rdev->last_read_error.tv_nsec = 0;
3005 rdev->sb_loaded = 0;
3006 rdev->bb_page = NULL;
3007 atomic_set(&rdev->nr_pending, 0);
3008 atomic_set(&rdev->read_errors, 0);
3009 atomic_set(&rdev->corrected_errors, 0);
3011 INIT_LIST_HEAD(&rdev->same_set);
3012 init_waitqueue_head(&rdev->blocked_wait);
3014 /* Add space to store bad block list.
3015 * This reserves the space even on arrays where it cannot
3016 * be used - I wonder if that matters
3018 rdev->badblocks.count = 0;
3019 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3020 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3021 seqlock_init(&rdev->badblocks.lock);
3022 if (rdev->badblocks.page == NULL)
3027 EXPORT_SYMBOL_GPL(md_rdev_init);
3029 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3031 * mark the device faulty if:
3033 * - the device is nonexistent (zero size)
3034 * - the device has no valid superblock
3036 * a faulty rdev _never_ has rdev->sb set.
3038 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3040 char b[BDEVNAME_SIZE];
3042 struct md_rdev *rdev;
3045 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3047 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3048 return ERR_PTR(-ENOMEM);
3051 err = md_rdev_init(rdev);
3054 err = alloc_disk_sb(rdev);
3058 err = lock_rdev(rdev, newdev, super_format == -2);
3062 kobject_init(&rdev->kobj, &rdev_ktype);
3064 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3067 "md: %s has zero or unknown size, marking faulty!\n",
3068 bdevname(rdev->bdev,b));
3073 if (super_format >= 0) {
3074 err = super_types[super_format].
3075 load_super(rdev, NULL, super_minor);
3076 if (err == -EINVAL) {
3078 "md: %s does not have a valid v%d.%d "
3079 "superblock, not importing!\n",
3080 bdevname(rdev->bdev,b),
3081 super_format, super_minor);
3086 "md: could not read %s's sb, not importing!\n",
3087 bdevname(rdev->bdev,b));
3097 md_rdev_clear(rdev);
3099 return ERR_PTR(err);
3103 * Check a full RAID array for plausibility
3106 static void analyze_sbs(struct mddev *mddev)
3109 struct md_rdev *rdev, *freshest, *tmp;
3110 char b[BDEVNAME_SIZE];
3113 rdev_for_each_safe(rdev, tmp, mddev)
3114 switch (super_types[mddev->major_version].
3115 load_super(rdev, freshest, mddev->minor_version)) {
3123 "md: fatal superblock inconsistency in %s"
3124 " -- removing from array\n",
3125 bdevname(rdev->bdev,b));
3126 kick_rdev_from_array(rdev);
3129 super_types[mddev->major_version].
3130 validate_super(mddev, freshest);
3133 rdev_for_each_safe(rdev, tmp, mddev) {
3134 if (mddev->max_disks &&
3135 (rdev->desc_nr >= mddev->max_disks ||
3136 i > mddev->max_disks)) {
3138 "md: %s: %s: only %d devices permitted\n",
3139 mdname(mddev), bdevname(rdev->bdev, b),
3141 kick_rdev_from_array(rdev);
3144 if (rdev != freshest)
3145 if (super_types[mddev->major_version].
3146 validate_super(mddev, rdev)) {
3147 printk(KERN_WARNING "md: kicking non-fresh %s"
3149 bdevname(rdev->bdev,b));
3150 kick_rdev_from_array(rdev);
3153 if (mddev->level == LEVEL_MULTIPATH) {
3154 rdev->desc_nr = i++;
3155 rdev->raid_disk = rdev->desc_nr;
3156 set_bit(In_sync, &rdev->flags);
3157 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3158 rdev->raid_disk = -1;
3159 clear_bit(In_sync, &rdev->flags);
3164 /* Read a fixed-point number.
3165 * Numbers in sysfs attributes should be in "standard" units where
3166 * possible, so time should be in seconds.
3167 * However we internally use a a much smaller unit such as
3168 * milliseconds or jiffies.
3169 * This function takes a decimal number with a possible fractional
3170 * component, and produces an integer which is the result of
3171 * multiplying that number by 10^'scale'.
3172 * all without any floating-point arithmetic.
3174 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3176 unsigned long result = 0;
3178 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3181 else if (decimals < scale) {
3184 result = result * 10 + value;
3196 while (decimals < scale) {
3204 static void md_safemode_timeout(unsigned long data);
3207 safe_delay_show(struct mddev *mddev, char *page)
3209 int msec = (mddev->safemode_delay*1000)/HZ;
3210 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3213 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3217 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3220 mddev->safemode_delay = 0;
3222 unsigned long old_delay = mddev->safemode_delay;
3223 unsigned long new_delay = (msec*HZ)/1000;
3227 mddev->safemode_delay = new_delay;
3228 if (new_delay < old_delay || old_delay == 0)
3229 mod_timer(&mddev->safemode_timer, jiffies+1);
3233 static struct md_sysfs_entry md_safe_delay =
3234 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3237 level_show(struct mddev *mddev, char *page)
3239 struct md_personality *p;
3241 spin_lock(&mddev->lock);
3244 ret = sprintf(page, "%s\n", p->name);
3245 else if (mddev->clevel[0])
3246 ret = sprintf(page, "%s\n", mddev->clevel);
3247 else if (mddev->level != LEVEL_NONE)
3248 ret = sprintf(page, "%d\n", mddev->level);
3251 spin_unlock(&mddev->lock);
3256 level_store(struct mddev *mddev, const char *buf, size_t len)
3261 struct md_personality *pers, *oldpers;
3263 void *priv, *oldpriv;
3264 struct md_rdev *rdev;
3266 if (slen == 0 || slen >= sizeof(clevel))
3269 rv = mddev_lock(mddev);
3273 if (mddev->pers == NULL) {
3274 strncpy(mddev->clevel, buf, slen);
3275 if (mddev->clevel[slen-1] == '\n')
3277 mddev->clevel[slen] = 0;
3278 mddev->level = LEVEL_NONE;
3286 /* request to change the personality. Need to ensure:
3287 * - array is not engaged in resync/recovery/reshape
3288 * - old personality can be suspended
3289 * - new personality will access other array.
3293 if (mddev->sync_thread ||
3294 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3295 mddev->reshape_position != MaxSector ||
3296 mddev->sysfs_active)
3300 if (!mddev->pers->quiesce) {
3301 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3302 mdname(mddev), mddev->pers->name);
3306 /* Now find the new personality */
3307 strncpy(clevel, buf, slen);
3308 if (clevel[slen-1] == '\n')
3311 if (kstrtol(clevel, 10, &level))
3314 if (request_module("md-%s", clevel) != 0)
3315 request_module("md-level-%s", clevel);
3316 spin_lock(&pers_lock);
3317 pers = find_pers(level, clevel);
3318 if (!pers || !try_module_get(pers->owner)) {
3319 spin_unlock(&pers_lock);
3320 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3324 spin_unlock(&pers_lock);
3326 if (pers == mddev->pers) {
3327 /* Nothing to do! */
3328 module_put(pers->owner);
3332 if (!pers->takeover) {
3333 module_put(pers->owner);
3334 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3335 mdname(mddev), clevel);
3340 rdev_for_each(rdev, mddev)
3341 rdev->new_raid_disk = rdev->raid_disk;
3343 /* ->takeover must set new_* and/or delta_disks
3344 * if it succeeds, and may set them when it fails.
3346 priv = pers->takeover(mddev);
3348 mddev->new_level = mddev->level;
3349 mddev->new_layout = mddev->layout;
3350 mddev->new_chunk_sectors = mddev->chunk_sectors;
3351 mddev->raid_disks -= mddev->delta_disks;
3352 mddev->delta_disks = 0;
3353 mddev->reshape_backwards = 0;
3354 module_put(pers->owner);
3355 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3356 mdname(mddev), clevel);
3361 /* Looks like we have a winner */
3362 mddev_suspend(mddev);
3363 mddev_detach(mddev);
3365 spin_lock(&mddev->lock);
3366 oldpers = mddev->pers;
3367 oldpriv = mddev->private;
3369 mddev->private = priv;
3370 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3371 mddev->level = mddev->new_level;
3372 mddev->layout = mddev->new_layout;
3373 mddev->chunk_sectors = mddev->new_chunk_sectors;
3374 mddev->delta_disks = 0;
3375 mddev->reshape_backwards = 0;
3376 mddev->degraded = 0;
3377 spin_unlock(&mddev->lock);
3379 if (oldpers->sync_request == NULL &&
3381 /* We are converting from a no-redundancy array
3382 * to a redundancy array and metadata is managed
3383 * externally so we need to be sure that writes
3384 * won't block due to a need to transition
3386 * until external management is started.
3389 mddev->safemode_delay = 0;
3390 mddev->safemode = 0;
3393 oldpers->free(mddev, oldpriv);
3395 if (oldpers->sync_request == NULL &&
3396 pers->sync_request != NULL) {
3397 /* need to add the md_redundancy_group */
3398 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3400 "md: cannot register extra attributes for %s\n",
3402 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3404 if (oldpers->sync_request != NULL &&
3405 pers->sync_request == NULL) {
3406 /* need to remove the md_redundancy_group */
3407 if (mddev->to_remove == NULL)
3408 mddev->to_remove = &md_redundancy_group;
3411 rdev_for_each(rdev, mddev) {
3412 if (rdev->raid_disk < 0)
3414 if (rdev->new_raid_disk >= mddev->raid_disks)
3415 rdev->new_raid_disk = -1;
3416 if (rdev->new_raid_disk == rdev->raid_disk)
3418 sysfs_unlink_rdev(mddev, rdev);
3420 rdev_for_each(rdev, mddev) {
3421 if (rdev->raid_disk < 0)
3423 if (rdev->new_raid_disk == rdev->raid_disk)
3425 rdev->raid_disk = rdev->new_raid_disk;
3426 if (rdev->raid_disk < 0)
3427 clear_bit(In_sync, &rdev->flags);
3429 if (sysfs_link_rdev(mddev, rdev))
3430 printk(KERN_WARNING "md: cannot register rd%d"
3431 " for %s after level change\n",
3432 rdev->raid_disk, mdname(mddev));
3436 if (pers->sync_request == NULL) {
3437 /* this is now an array without redundancy, so
3438 * it must always be in_sync
3441 del_timer_sync(&mddev->safemode_timer);
3443 blk_set_stacking_limits(&mddev->queue->limits);
3445 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3446 mddev_resume(mddev);
3448 md_update_sb(mddev, 1);
3449 sysfs_notify(&mddev->kobj, NULL, "level");
3450 md_new_event(mddev);
3453 mddev_unlock(mddev);
3457 static struct md_sysfs_entry md_level =
3458 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3461 layout_show(struct mddev *mddev, char *page)
3463 /* just a number, not meaningful for all levels */
3464 if (mddev->reshape_position != MaxSector &&
3465 mddev->layout != mddev->new_layout)
3466 return sprintf(page, "%d (%d)\n",
3467 mddev->new_layout, mddev->layout);
3468 return sprintf(page, "%d\n", mddev->layout);
3472 layout_store(struct mddev *mddev, const char *buf, size_t len)
3475 unsigned long n = simple_strtoul(buf, &e, 10);
3478 if (!*buf || (*e && *e != '\n'))
3480 err = mddev_lock(mddev);
3485 if (mddev->pers->check_reshape == NULL)
3490 mddev->new_layout = n;
3491 err = mddev->pers->check_reshape(mddev);
3493 mddev->new_layout = mddev->layout;
3496 mddev->new_layout = n;
3497 if (mddev->reshape_position == MaxSector)
3500 mddev_unlock(mddev);
3503 static struct md_sysfs_entry md_layout =
3504 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3507 raid_disks_show(struct mddev *mddev, char *page)
3509 if (mddev->raid_disks == 0)
3511 if (mddev->reshape_position != MaxSector &&
3512 mddev->delta_disks != 0)
3513 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3514 mddev->raid_disks - mddev->delta_disks);
3515 return sprintf(page, "%d\n", mddev->raid_disks);
3518 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3521 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3525 unsigned long n = simple_strtoul(buf, &e, 10);
3527 if (!*buf || (*e && *e != '\n'))
3530 err = mddev_lock(mddev);
3534 err = update_raid_disks(mddev, n);
3535 else if (mddev->reshape_position != MaxSector) {
3536 struct md_rdev *rdev;
3537 int olddisks = mddev->raid_disks - mddev->delta_disks;
3540 rdev_for_each(rdev, mddev) {
3542 rdev->data_offset < rdev->new_data_offset)
3545 rdev->data_offset > rdev->new_data_offset)
3549 mddev->delta_disks = n - olddisks;
3550 mddev->raid_disks = n;
3551 mddev->reshape_backwards = (mddev->delta_disks < 0);
3553 mddev->raid_disks = n;
3555 mddev_unlock(mddev);
3556 return err ? err : len;
3558 static struct md_sysfs_entry md_raid_disks =
3559 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3562 chunk_size_show(struct mddev *mddev, char *page)
3564 if (mddev->reshape_position != MaxSector &&
3565 mddev->chunk_sectors != mddev->new_chunk_sectors)
3566 return sprintf(page, "%d (%d)\n",
3567 mddev->new_chunk_sectors << 9,
3568 mddev->chunk_sectors << 9);
3569 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3573 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3577 unsigned long n = simple_strtoul(buf, &e, 10);
3579 if (!*buf || (*e && *e != '\n'))
3582 err = mddev_lock(mddev);
3586 if (mddev->pers->check_reshape == NULL)
3591 mddev->new_chunk_sectors = n >> 9;
3592 err = mddev->pers->check_reshape(mddev);
3594 mddev->new_chunk_sectors = mddev->chunk_sectors;
3597 mddev->new_chunk_sectors = n >> 9;
3598 if (mddev->reshape_position == MaxSector)
3599 mddev->chunk_sectors = n >> 9;
3601 mddev_unlock(mddev);
3604 static struct md_sysfs_entry md_chunk_size =
3605 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3608 resync_start_show(struct mddev *mddev, char *page)
3610 if (mddev->recovery_cp == MaxSector)
3611 return sprintf(page, "none\n");
3612 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3616 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3620 unsigned long long n = simple_strtoull(buf, &e, 10);
3622 err = mddev_lock(mddev);
3625 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3627 else if (cmd_match(buf, "none"))
3629 else if (!*buf || (*e && *e != '\n'))
3633 mddev->recovery_cp = n;
3635 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3637 mddev_unlock(mddev);
3640 static struct md_sysfs_entry md_resync_start =
3641 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3644 * The array state can be:
3647 * No devices, no size, no level
3648 * Equivalent to STOP_ARRAY ioctl
3650 * May have some settings, but array is not active
3651 * all IO results in error
3652 * When written, doesn't tear down array, but just stops it
3653 * suspended (not supported yet)
3654 * All IO requests will block. The array can be reconfigured.
3655 * Writing this, if accepted, will block until array is quiescent
3657 * no resync can happen. no superblocks get written.
3658 * write requests fail
3660 * like readonly, but behaves like 'clean' on a write request.
3662 * clean - no pending writes, but otherwise active.
3663 * When written to inactive array, starts without resync
3664 * If a write request arrives then
3665 * if metadata is known, mark 'dirty' and switch to 'active'.
3666 * if not known, block and switch to write-pending
3667 * If written to an active array that has pending writes, then fails.
3669 * fully active: IO and resync can be happening.
3670 * When written to inactive array, starts with resync
3673 * clean, but writes are blocked waiting for 'active' to be written.
3676 * like active, but no writes have been seen for a while (100msec).
3679 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3680 write_pending, active_idle, bad_word};
3681 static char *array_states[] = {
3682 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3683 "write-pending", "active-idle", NULL };
3685 static int match_word(const char *word, char **list)
3688 for (n=0; list[n]; n++)
3689 if (cmd_match(word, list[n]))
3695 array_state_show(struct mddev *mddev, char *page)
3697 enum array_state st = inactive;
3710 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3712 else if (mddev->safemode)
3718 if (list_empty(&mddev->disks) &&
3719 mddev->raid_disks == 0 &&
3720 mddev->dev_sectors == 0)
3725 return sprintf(page, "%s\n", array_states[st]);
3728 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3729 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3730 static int do_md_run(struct mddev *mddev);
3731 static int restart_array(struct mddev *mddev);
3734 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3737 enum array_state st = match_word(buf, array_states);
3739 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3740 /* don't take reconfig_mutex when toggling between
3743 spin_lock(&mddev->lock);
3745 restart_array(mddev);
3746 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3747 wake_up(&mddev->sb_wait);
3749 } else /* st == clean */ {
3750 restart_array(mddev);
3751 if (atomic_read(&mddev->writes_pending) == 0) {
3752 if (mddev->in_sync == 0) {
3754 if (mddev->safemode == 1)
3755 mddev->safemode = 0;
3756 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3762 spin_unlock(&mddev->lock);
3765 err = mddev_lock(mddev);
3773 /* stopping an active array */
3774 err = do_md_stop(mddev, 0, NULL);
3777 /* stopping an active array */
3779 err = do_md_stop(mddev, 2, NULL);
3781 err = 0; /* already inactive */
3784 break; /* not supported yet */
3787 err = md_set_readonly(mddev, NULL);
3790 set_disk_ro(mddev->gendisk, 1);
3791 err = do_md_run(mddev);
3797 err = md_set_readonly(mddev, NULL);
3798 else if (mddev->ro == 1)
3799 err = restart_array(mddev);
3802 set_disk_ro(mddev->gendisk, 0);
3806 err = do_md_run(mddev);
3811 restart_array(mddev);
3812 spin_lock(&mddev->lock);
3813 if (atomic_read(&mddev->writes_pending) == 0) {
3814 if (mddev->in_sync == 0) {
3816 if (mddev->safemode == 1)
3817 mddev->safemode = 0;
3818 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3823 spin_unlock(&mddev->lock);
3829 restart_array(mddev);
3830 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3831 wake_up(&mddev->sb_wait);
3835 set_disk_ro(mddev->gendisk, 0);
3836 err = do_md_run(mddev);
3841 /* these cannot be set */
3846 if (mddev->hold_active == UNTIL_IOCTL)
3847 mddev->hold_active = 0;
3848 sysfs_notify_dirent_safe(mddev->sysfs_state);
3850 mddev_unlock(mddev);
3853 static struct md_sysfs_entry md_array_state =
3854 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3857 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3858 return sprintf(page, "%d\n",
3859 atomic_read(&mddev->max_corr_read_errors));
3863 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3866 unsigned long n = simple_strtoul(buf, &e, 10);
3868 if (*buf && (*e == 0 || *e == '\n')) {
3869 atomic_set(&mddev->max_corr_read_errors, n);
3875 static struct md_sysfs_entry max_corr_read_errors =
3876 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3877 max_corrected_read_errors_store);
3880 null_show(struct mddev *mddev, char *page)
3886 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3888 /* buf must be %d:%d\n? giving major and minor numbers */
3889 /* The new device is added to the array.
3890 * If the array has a persistent superblock, we read the
3891 * superblock to initialise info and check validity.
3892 * Otherwise, only checking done is that in bind_rdev_to_array,
3893 * which mainly checks size.
3896 int major = simple_strtoul(buf, &e, 10);
3899 struct md_rdev *rdev;
3902 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3904 minor = simple_strtoul(e+1, &e, 10);
3905 if (*e && *e != '\n')
3907 dev = MKDEV(major, minor);
3908 if (major != MAJOR(dev) ||
3909 minor != MINOR(dev))
3912 flush_workqueue(md_misc_wq);
3914 err = mddev_lock(mddev);
3917 if (mddev->persistent) {
3918 rdev = md_import_device(dev, mddev->major_version,
3919 mddev->minor_version);
3920 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3921 struct md_rdev *rdev0
3922 = list_entry(mddev->disks.next,
3923 struct md_rdev, same_set);
3924 err = super_types[mddev->major_version]
3925 .load_super(rdev, rdev0, mddev->minor_version);
3929 } else if (mddev->external)
3930 rdev = md_import_device(dev, -2, -1);
3932 rdev = md_import_device(dev, -1, -1);
3935 return PTR_ERR(rdev);
3936 err = bind_rdev_to_array(rdev, mddev);
3940 mddev_unlock(mddev);
3941 return err ? err : len;
3944 static struct md_sysfs_entry md_new_device =
3945 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3948 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3951 unsigned long chunk, end_chunk;
3954 err = mddev_lock(mddev);
3959 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3961 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3962 if (buf == end) break;
3963 if (*end == '-') { /* range */
3965 end_chunk = simple_strtoul(buf, &end, 0);
3966 if (buf == end) break;
3968 if (*end && !isspace(*end)) break;
3969 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3970 buf = skip_spaces(end);
3972 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3974 mddev_unlock(mddev);
3978 static struct md_sysfs_entry md_bitmap =
3979 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3982 size_show(struct mddev *mddev, char *page)
3984 return sprintf(page, "%llu\n",
3985 (unsigned long long)mddev->dev_sectors / 2);
3988 static int update_size(struct mddev *mddev, sector_t num_sectors);
3991 size_store(struct mddev *mddev, const char *buf, size_t len)
3993 /* If array is inactive, we can reduce the component size, but
3994 * not increase it (except from 0).
3995 * If array is active, we can try an on-line resize
3998 int err = strict_blocks_to_sectors(buf, §ors);
4002 err = mddev_lock(mddev);
4006 err = update_size(mddev, sectors);
4007 md_update_sb(mddev, 1);
4009 if (mddev->dev_sectors == 0 ||
4010 mddev->dev_sectors > sectors)
4011 mddev->dev_sectors = sectors;
4015 mddev_unlock(mddev);
4016 return err ? err : len;
4019 static struct md_sysfs_entry md_size =
4020 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4022 /* Metadata version.
4024 * 'none' for arrays with no metadata (good luck...)
4025 * 'external' for arrays with externally managed metadata,
4026 * or N.M for internally known formats
4029 metadata_show(struct mddev *mddev, char *page)
4031 if (mddev->persistent)
4032 return sprintf(page, "%d.%d\n",
4033 mddev->major_version, mddev->minor_version);
4034 else if (mddev->external)
4035 return sprintf(page, "external:%s\n", mddev->metadata_type);
4037 return sprintf(page, "none\n");
4041 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4046 /* Changing the details of 'external' metadata is
4047 * always permitted. Otherwise there must be
4048 * no devices attached to the array.
4051 err = mddev_lock(mddev);
4055 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4057 else if (!list_empty(&mddev->disks))
4061 if (cmd_match(buf, "none")) {
4062 mddev->persistent = 0;
4063 mddev->external = 0;
4064 mddev->major_version = 0;
4065 mddev->minor_version = 90;
4068 if (strncmp(buf, "external:", 9) == 0) {
4069 size_t namelen = len-9;
4070 if (namelen >= sizeof(mddev->metadata_type))
4071 namelen = sizeof(mddev->metadata_type)-1;
4072 strncpy(mddev->metadata_type, buf+9, namelen);
4073 mddev->metadata_type[namelen] = 0;
4074 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4075 mddev->metadata_type[--namelen] = 0;
4076 mddev->persistent = 0;
4077 mddev->external = 1;
4078 mddev->major_version = 0;
4079 mddev->minor_version = 90;
4082 major = simple_strtoul(buf, &e, 10);
4084 if (e==buf || *e != '.')
4087 minor = simple_strtoul(buf, &e, 10);
4088 if (e==buf || (*e && *e != '\n') )
4091 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4093 mddev->major_version = major;
4094 mddev->minor_version = minor;
4095 mddev->persistent = 1;
4096 mddev->external = 0;
4099 mddev_unlock(mddev);
4103 static struct md_sysfs_entry md_metadata =
4104 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4107 action_show(struct mddev *mddev, char *page)
4109 char *type = "idle";
4110 unsigned long recovery = mddev->recovery;
4111 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4113 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4114 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4115 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4117 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4118 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4120 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4124 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4127 return sprintf(page, "%s\n", type);
4131 action_store(struct mddev *mddev, const char *page, size_t len)
4133 if (!mddev->pers || !mddev->pers->sync_request)
4136 if (cmd_match(page, "frozen"))
4137 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4139 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4141 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4142 flush_workqueue(md_misc_wq);
4143 if (mddev->sync_thread) {
4144 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4145 if (mddev_lock(mddev) == 0) {
4146 md_reap_sync_thread(mddev);
4147 mddev_unlock(mddev);
4150 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4151 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4153 else if (cmd_match(page, "resync"))
4154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4155 else if (cmd_match(page, "recover")) {
4156 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4157 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4158 } else if (cmd_match(page, "reshape")) {
4160 if (mddev->pers->start_reshape == NULL)
4162 err = mddev_lock(mddev);
4164 err = mddev->pers->start_reshape(mddev);
4165 mddev_unlock(mddev);
4169 sysfs_notify(&mddev->kobj, NULL, "degraded");
4171 if (cmd_match(page, "check"))
4172 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4173 else if (!cmd_match(page, "repair"))
4175 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4176 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4178 if (mddev->ro == 2) {
4179 /* A write to sync_action is enough to justify
4180 * canceling read-auto mode
4183 md_wakeup_thread(mddev->sync_thread);
4185 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4186 md_wakeup_thread(mddev->thread);
4187 sysfs_notify_dirent_safe(mddev->sysfs_action);
4191 static struct md_sysfs_entry md_scan_mode =
4192 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4195 last_sync_action_show(struct mddev *mddev, char *page)
4197 return sprintf(page, "%s\n", mddev->last_sync_action);
4200 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4203 mismatch_cnt_show(struct mddev *mddev, char *page)
4205 return sprintf(page, "%llu\n",
4206 (unsigned long long)
4207 atomic64_read(&mddev->resync_mismatches));
4210 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4213 sync_min_show(struct mddev *mddev, char *page)
4215 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4216 mddev->sync_speed_min ? "local": "system");
4220 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4224 if (strncmp(buf, "system", 6)==0) {
4225 mddev->sync_speed_min = 0;
4228 min = simple_strtoul(buf, &e, 10);
4229 if (buf == e || (*e && *e != '\n') || min <= 0)
4231 mddev->sync_speed_min = min;
4235 static struct md_sysfs_entry md_sync_min =
4236 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4239 sync_max_show(struct mddev *mddev, char *page)
4241 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4242 mddev->sync_speed_max ? "local": "system");
4246 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4250 if (strncmp(buf, "system", 6)==0) {
4251 mddev->sync_speed_max = 0;
4254 max = simple_strtoul(buf, &e, 10);
4255 if (buf == e || (*e && *e != '\n') || max <= 0)
4257 mddev->sync_speed_max = max;
4261 static struct md_sysfs_entry md_sync_max =
4262 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4265 degraded_show(struct mddev *mddev, char *page)
4267 return sprintf(page, "%d\n", mddev->degraded);
4269 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4272 sync_force_parallel_show(struct mddev *mddev, char *page)
4274 return sprintf(page, "%d\n", mddev->parallel_resync);
4278 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4282 if (kstrtol(buf, 10, &n))
4285 if (n != 0 && n != 1)
4288 mddev->parallel_resync = n;
4290 if (mddev->sync_thread)
4291 wake_up(&resync_wait);
4296 /* force parallel resync, even with shared block devices */
4297 static struct md_sysfs_entry md_sync_force_parallel =
4298 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4299 sync_force_parallel_show, sync_force_parallel_store);
4302 sync_speed_show(struct mddev *mddev, char *page)
4304 unsigned long resync, dt, db;
4305 if (mddev->curr_resync == 0)
4306 return sprintf(page, "none\n");
4307 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4308 dt = (jiffies - mddev->resync_mark) / HZ;
4310 db = resync - mddev->resync_mark_cnt;
4311 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4314 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4317 sync_completed_show(struct mddev *mddev, char *page)
4319 unsigned long long max_sectors, resync;
4321 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4322 return sprintf(page, "none\n");
4324 if (mddev->curr_resync == 1 ||
4325 mddev->curr_resync == 2)
4326 return sprintf(page, "delayed\n");
4328 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4329 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4330 max_sectors = mddev->resync_max_sectors;
4332 max_sectors = mddev->dev_sectors;
4334 resync = mddev->curr_resync_completed;
4335 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4338 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4341 min_sync_show(struct mddev *mddev, char *page)
4343 return sprintf(page, "%llu\n",
4344 (unsigned long long)mddev->resync_min);
4347 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4349 unsigned long long min;
4353 if (kstrtoull(buf, 10, &min))
4356 spin_lock(&mddev->lock);
4358 if (min > mddev->resync_max)
4362 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4365 /* Must be a multiple of chunk_size */
4366 chunk = mddev->chunk_sectors;
4368 sector_t temp = min;
4371 if (sector_div(temp, chunk))
4374 mddev->resync_min = min;
4378 spin_unlock(&mddev->lock);
4382 static struct md_sysfs_entry md_min_sync =
4383 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4386 max_sync_show(struct mddev *mddev, char *page)
4388 if (mddev->resync_max == MaxSector)
4389 return sprintf(page, "max\n");
4391 return sprintf(page, "%llu\n",
4392 (unsigned long long)mddev->resync_max);
4395 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4398 spin_lock(&mddev->lock);
4399 if (strncmp(buf, "max", 3) == 0)
4400 mddev->resync_max = MaxSector;
4402 unsigned long long max;
4406 if (kstrtoull(buf, 10, &max))
4408 if (max < mddev->resync_min)
4412 if (max < mddev->resync_max &&
4414 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4417 /* Must be a multiple of chunk_size */
4418 chunk = mddev->chunk_sectors;
4420 sector_t temp = max;
4423 if (sector_div(temp, chunk))
4426 mddev->resync_max = max;
4428 wake_up(&mddev->recovery_wait);
4431 spin_unlock(&mddev->lock);
4435 static struct md_sysfs_entry md_max_sync =
4436 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4439 suspend_lo_show(struct mddev *mddev, char *page)
4441 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4445 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4448 unsigned long long new = simple_strtoull(buf, &e, 10);
4449 unsigned long long old;
4452 if (buf == e || (*e && *e != '\n'))
4455 err = mddev_lock(mddev);
4459 if (mddev->pers == NULL ||
4460 mddev->pers->quiesce == NULL)
4462 old = mddev->suspend_lo;
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 mddev_unlock(mddev);
4477 static struct md_sysfs_entry md_suspend_lo =
4478 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4481 suspend_hi_show(struct mddev *mddev, char *page)
4483 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4487 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4490 unsigned long long new = simple_strtoull(buf, &e, 10);
4491 unsigned long long old;
4494 if (buf == e || (*e && *e != '\n'))
4497 err = mddev_lock(mddev);
4501 if (mddev->pers == NULL ||
4502 mddev->pers->quiesce == NULL)
4504 old = mddev->suspend_hi;
4505 mddev->suspend_hi = new;
4507 /* Shrinking suspended region */
4508 mddev->pers->quiesce(mddev, 2);
4510 /* Expanding suspended region - need to wait */
4511 mddev->pers->quiesce(mddev, 1);
4512 mddev->pers->quiesce(mddev, 0);
4516 mddev_unlock(mddev);
4519 static struct md_sysfs_entry md_suspend_hi =
4520 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4523 reshape_position_show(struct mddev *mddev, char *page)
4525 if (mddev->reshape_position != MaxSector)
4526 return sprintf(page, "%llu\n",
4527 (unsigned long long)mddev->reshape_position);
4528 strcpy(page, "none\n");
4533 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4535 struct md_rdev *rdev;
4538 unsigned long long new = simple_strtoull(buf, &e, 10);
4540 if (buf == e || (*e && *e != '\n'))
4542 err = mddev_lock(mddev);
4548 mddev->reshape_position = new;
4549 mddev->delta_disks = 0;
4550 mddev->reshape_backwards = 0;
4551 mddev->new_level = mddev->level;
4552 mddev->new_layout = mddev->layout;
4553 mddev->new_chunk_sectors = mddev->chunk_sectors;
4554 rdev_for_each(rdev, mddev)
4555 rdev->new_data_offset = rdev->data_offset;
4558 mddev_unlock(mddev);
4562 static struct md_sysfs_entry md_reshape_position =
4563 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4564 reshape_position_store);
4567 reshape_direction_show(struct mddev *mddev, char *page)
4569 return sprintf(page, "%s\n",
4570 mddev->reshape_backwards ? "backwards" : "forwards");
4574 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4579 if (cmd_match(buf, "forwards"))
4581 else if (cmd_match(buf, "backwards"))
4585 if (mddev->reshape_backwards == backwards)
4588 err = mddev_lock(mddev);
4591 /* check if we are allowed to change */
4592 if (mddev->delta_disks)
4594 else if (mddev->persistent &&
4595 mddev->major_version == 0)
4598 mddev->reshape_backwards = backwards;
4599 mddev_unlock(mddev);
4603 static struct md_sysfs_entry md_reshape_direction =
4604 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4605 reshape_direction_store);
4608 array_size_show(struct mddev *mddev, char *page)
4610 if (mddev->external_size)
4611 return sprintf(page, "%llu\n",
4612 (unsigned long long)mddev->array_sectors/2);
4614 return sprintf(page, "default\n");
4618 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4623 err = mddev_lock(mddev);
4627 if (strncmp(buf, "default", 7) == 0) {
4629 sectors = mddev->pers->size(mddev, 0, 0);
4631 sectors = mddev->array_sectors;
4633 mddev->external_size = 0;
4635 if (strict_blocks_to_sectors(buf, §ors) < 0)
4637 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4640 mddev->external_size = 1;
4644 mddev->array_sectors = sectors;
4646 set_capacity(mddev->gendisk, mddev->array_sectors);
4647 revalidate_disk(mddev->gendisk);
4650 mddev_unlock(mddev);
4654 static struct md_sysfs_entry md_array_size =
4655 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4658 static struct attribute *md_default_attrs[] = {
4661 &md_raid_disks.attr,
4662 &md_chunk_size.attr,
4664 &md_resync_start.attr,
4666 &md_new_device.attr,
4667 &md_safe_delay.attr,
4668 &md_array_state.attr,
4669 &md_reshape_position.attr,
4670 &md_reshape_direction.attr,
4671 &md_array_size.attr,
4672 &max_corr_read_errors.attr,
4676 static struct attribute *md_redundancy_attrs[] = {
4678 &md_last_scan_mode.attr,
4679 &md_mismatches.attr,
4682 &md_sync_speed.attr,
4683 &md_sync_force_parallel.attr,
4684 &md_sync_completed.attr,
4687 &md_suspend_lo.attr,
4688 &md_suspend_hi.attr,
4693 static struct attribute_group md_redundancy_group = {
4695 .attrs = md_redundancy_attrs,
4699 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4701 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4702 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4707 spin_lock(&all_mddevs_lock);
4708 if (list_empty(&mddev->all_mddevs)) {
4709 spin_unlock(&all_mddevs_lock);
4713 spin_unlock(&all_mddevs_lock);
4715 rv = entry->show(mddev, page);
4721 md_attr_store(struct kobject *kobj, struct attribute *attr,
4722 const char *page, size_t length)
4724 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4725 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4730 if (!capable(CAP_SYS_ADMIN))
4732 spin_lock(&all_mddevs_lock);
4733 if (list_empty(&mddev->all_mddevs)) {
4734 spin_unlock(&all_mddevs_lock);
4738 spin_unlock(&all_mddevs_lock);
4739 rv = entry->store(mddev, page, length);
4744 static void md_free(struct kobject *ko)
4746 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4748 if (mddev->sysfs_state)
4749 sysfs_put(mddev->sysfs_state);
4751 if (mddev->gendisk) {
4752 del_gendisk(mddev->gendisk);
4753 put_disk(mddev->gendisk);
4756 blk_cleanup_queue(mddev->queue);
4761 static const struct sysfs_ops md_sysfs_ops = {
4762 .show = md_attr_show,
4763 .store = md_attr_store,
4765 static struct kobj_type md_ktype = {
4767 .sysfs_ops = &md_sysfs_ops,
4768 .default_attrs = md_default_attrs,
4773 static void mddev_delayed_delete(struct work_struct *ws)
4775 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4777 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4778 kobject_del(&mddev->kobj);
4779 kobject_put(&mddev->kobj);
4782 static int md_alloc(dev_t dev, char *name)
4784 static DEFINE_MUTEX(disks_mutex);
4785 struct mddev *mddev = mddev_find(dev);
4786 struct gendisk *disk;
4795 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4796 shift = partitioned ? MdpMinorShift : 0;
4797 unit = MINOR(mddev->unit) >> shift;
4799 /* wait for any previous instance of this device to be
4800 * completely removed (mddev_delayed_delete).
4802 flush_workqueue(md_misc_wq);
4804 mutex_lock(&disks_mutex);
4810 /* Need to ensure that 'name' is not a duplicate.
4812 struct mddev *mddev2;
4813 spin_lock(&all_mddevs_lock);
4815 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4816 if (mddev2->gendisk &&
4817 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4818 spin_unlock(&all_mddevs_lock);
4821 spin_unlock(&all_mddevs_lock);
4825 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4828 mddev->queue->queuedata = mddev;
4830 blk_queue_make_request(mddev->queue, md_make_request);
4831 blk_set_stacking_limits(&mddev->queue->limits);
4833 disk = alloc_disk(1 << shift);
4835 blk_cleanup_queue(mddev->queue);
4836 mddev->queue = NULL;
4839 disk->major = MAJOR(mddev->unit);
4840 disk->first_minor = unit << shift;
4842 strcpy(disk->disk_name, name);
4843 else if (partitioned)
4844 sprintf(disk->disk_name, "md_d%d", unit);
4846 sprintf(disk->disk_name, "md%d", unit);
4847 disk->fops = &md_fops;
4848 disk->private_data = mddev;
4849 disk->queue = mddev->queue;
4850 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4851 /* Allow extended partitions. This makes the
4852 * 'mdp' device redundant, but we can't really
4855 disk->flags |= GENHD_FL_EXT_DEVT;
4856 mddev->gendisk = disk;
4857 /* As soon as we call add_disk(), another thread could get
4858 * through to md_open, so make sure it doesn't get too far
4860 mutex_lock(&mddev->open_mutex);
4863 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4864 &disk_to_dev(disk)->kobj, "%s", "md");
4866 /* This isn't possible, but as kobject_init_and_add is marked
4867 * __must_check, we must do something with the result
4869 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4873 if (mddev->kobj.sd &&
4874 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4875 printk(KERN_DEBUG "pointless warning\n");
4876 mutex_unlock(&mddev->open_mutex);
4878 mutex_unlock(&disks_mutex);
4879 if (!error && mddev->kobj.sd) {
4880 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4881 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4887 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4889 md_alloc(dev, NULL);
4893 static int add_named_array(const char *val, struct kernel_param *kp)
4895 /* val must be "md_*" where * is not all digits.
4896 * We allocate an array with a large free minor number, and
4897 * set the name to val. val must not already be an active name.
4899 int len = strlen(val);
4900 char buf[DISK_NAME_LEN];
4902 while (len && val[len-1] == '\n')
4904 if (len >= DISK_NAME_LEN)
4906 strlcpy(buf, val, len+1);
4907 if (strncmp(buf, "md_", 3) != 0)
4909 return md_alloc(0, buf);
4912 static void md_safemode_timeout(unsigned long data)
4914 struct mddev *mddev = (struct mddev *) data;
4916 if (!atomic_read(&mddev->writes_pending)) {
4917 mddev->safemode = 1;
4918 if (mddev->external)
4919 sysfs_notify_dirent_safe(mddev->sysfs_state);
4921 md_wakeup_thread(mddev->thread);
4924 static int start_dirty_degraded;
4926 int md_run(struct mddev *mddev)
4929 struct md_rdev *rdev;
4930 struct md_personality *pers;
4932 if (list_empty(&mddev->disks))
4933 /* cannot run an array with no devices.. */
4938 /* Cannot run until previous stop completes properly */
4939 if (mddev->sysfs_active)
4943 * Analyze all RAID superblock(s)
4945 if (!mddev->raid_disks) {
4946 if (!mddev->persistent)
4951 if (mddev->level != LEVEL_NONE)
4952 request_module("md-level-%d", mddev->level);
4953 else if (mddev->clevel[0])
4954 request_module("md-%s", mddev->clevel);
4957 * Drop all container device buffers, from now on
4958 * the only valid external interface is through the md
4961 rdev_for_each(rdev, mddev) {
4962 if (test_bit(Faulty, &rdev->flags))
4964 sync_blockdev(rdev->bdev);
4965 invalidate_bdev(rdev->bdev);
4967 /* perform some consistency tests on the device.
4968 * We don't want the data to overlap the metadata,
4969 * Internal Bitmap issues have been handled elsewhere.
4971 if (rdev->meta_bdev) {
4972 /* Nothing to check */;
4973 } else if (rdev->data_offset < rdev->sb_start) {
4974 if (mddev->dev_sectors &&
4975 rdev->data_offset + mddev->dev_sectors
4977 printk("md: %s: data overlaps metadata\n",
4982 if (rdev->sb_start + rdev->sb_size/512
4983 > rdev->data_offset) {
4984 printk("md: %s: metadata overlaps data\n",
4989 sysfs_notify_dirent_safe(rdev->sysfs_state);
4992 if (mddev->bio_set == NULL)
4993 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
4995 spin_lock(&pers_lock);
4996 pers = find_pers(mddev->level, mddev->clevel);
4997 if (!pers || !try_module_get(pers->owner)) {
4998 spin_unlock(&pers_lock);
4999 if (mddev->level != LEVEL_NONE)
5000 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5003 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5007 spin_unlock(&pers_lock);
5008 if (mddev->level != pers->level) {
5009 mddev->level = pers->level;
5010 mddev->new_level = pers->level;
5012 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5014 if (mddev->reshape_position != MaxSector &&
5015 pers->start_reshape == NULL) {
5016 /* This personality cannot handle reshaping... */
5017 module_put(pers->owner);
5021 if (pers->sync_request) {
5022 /* Warn if this is a potentially silly
5025 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5026 struct md_rdev *rdev2;
5029 rdev_for_each(rdev, mddev)
5030 rdev_for_each(rdev2, mddev) {
5032 rdev->bdev->bd_contains ==
5033 rdev2->bdev->bd_contains) {
5035 "%s: WARNING: %s appears to be"
5036 " on the same physical disk as"
5039 bdevname(rdev->bdev,b),
5040 bdevname(rdev2->bdev,b2));
5047 "True protection against single-disk"
5048 " failure might be compromised.\n");
5051 mddev->recovery = 0;
5052 /* may be over-ridden by personality */
5053 mddev->resync_max_sectors = mddev->dev_sectors;
5055 mddev->ok_start_degraded = start_dirty_degraded;
5057 if (start_readonly && mddev->ro == 0)
5058 mddev->ro = 2; /* read-only, but switch on first write */
5060 err = pers->run(mddev);
5062 printk(KERN_ERR "md: pers->run() failed ...\n");
5063 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5064 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5065 " but 'external_size' not in effect?\n", __func__);
5067 "md: invalid array_size %llu > default size %llu\n",
5068 (unsigned long long)mddev->array_sectors / 2,
5069 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5072 if (err == 0 && pers->sync_request &&
5073 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5074 err = bitmap_create(mddev);
5076 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5077 mdname(mddev), err);
5080 mddev_detach(mddev);
5081 pers->free(mddev, mddev->private);
5082 module_put(pers->owner);
5083 bitmap_destroy(mddev);
5087 mddev->queue->backing_dev_info.congested_data = mddev;
5088 mddev->queue->backing_dev_info.congested_fn = md_congested;
5089 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5091 if (pers->sync_request) {
5092 if (mddev->kobj.sd &&
5093 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5095 "md: cannot register extra attributes for %s\n",
5097 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5098 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5101 atomic_set(&mddev->writes_pending,0);
5102 atomic_set(&mddev->max_corr_read_errors,
5103 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5104 mddev->safemode = 0;
5105 mddev->safemode_timer.function = md_safemode_timeout;
5106 mddev->safemode_timer.data = (unsigned long) mddev;
5107 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5110 spin_lock(&mddev->lock);
5113 spin_unlock(&mddev->lock);
5114 rdev_for_each(rdev, mddev)
5115 if (rdev->raid_disk >= 0)
5116 if (sysfs_link_rdev(mddev, rdev))
5117 /* failure here is OK */;
5119 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5121 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5122 md_update_sb(mddev, 0);
5124 md_new_event(mddev);
5125 sysfs_notify_dirent_safe(mddev->sysfs_state);
5126 sysfs_notify_dirent_safe(mddev->sysfs_action);
5127 sysfs_notify(&mddev->kobj, NULL, "degraded");
5130 EXPORT_SYMBOL_GPL(md_run);
5132 static int do_md_run(struct mddev *mddev)
5136 err = md_run(mddev);
5139 err = bitmap_load(mddev);
5141 bitmap_destroy(mddev);
5145 md_wakeup_thread(mddev->thread);
5146 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5148 set_capacity(mddev->gendisk, mddev->array_sectors);
5149 revalidate_disk(mddev->gendisk);
5151 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5156 static int restart_array(struct mddev *mddev)
5158 struct gendisk *disk = mddev->gendisk;
5160 /* Complain if it has no devices */
5161 if (list_empty(&mddev->disks))
5167 mddev->safemode = 0;
5169 set_disk_ro(disk, 0);
5170 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5172 /* Kick recovery or resync if necessary */
5173 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5174 md_wakeup_thread(mddev->thread);
5175 md_wakeup_thread(mddev->sync_thread);
5176 sysfs_notify_dirent_safe(mddev->sysfs_state);
5180 static void md_clean(struct mddev *mddev)
5182 mddev->array_sectors = 0;
5183 mddev->external_size = 0;
5184 mddev->dev_sectors = 0;
5185 mddev->raid_disks = 0;
5186 mddev->recovery_cp = 0;
5187 mddev->resync_min = 0;
5188 mddev->resync_max = MaxSector;
5189 mddev->reshape_position = MaxSector;
5190 mddev->external = 0;
5191 mddev->persistent = 0;
5192 mddev->level = LEVEL_NONE;
5193 mddev->clevel[0] = 0;
5196 mddev->metadata_type[0] = 0;
5197 mddev->chunk_sectors = 0;
5198 mddev->ctime = mddev->utime = 0;
5200 mddev->max_disks = 0;
5202 mddev->can_decrease_events = 0;
5203 mddev->delta_disks = 0;
5204 mddev->reshape_backwards = 0;
5205 mddev->new_level = LEVEL_NONE;
5206 mddev->new_layout = 0;
5207 mddev->new_chunk_sectors = 0;
5208 mddev->curr_resync = 0;
5209 atomic64_set(&mddev->resync_mismatches, 0);
5210 mddev->suspend_lo = mddev->suspend_hi = 0;
5211 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5212 mddev->recovery = 0;
5215 mddev->degraded = 0;
5216 mddev->safemode = 0;
5217 mddev->merge_check_needed = 0;
5218 mddev->bitmap_info.offset = 0;
5219 mddev->bitmap_info.default_offset = 0;
5220 mddev->bitmap_info.default_space = 0;
5221 mddev->bitmap_info.chunksize = 0;
5222 mddev->bitmap_info.daemon_sleep = 0;
5223 mddev->bitmap_info.max_write_behind = 0;
5226 static void __md_stop_writes(struct mddev *mddev)
5228 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5229 flush_workqueue(md_misc_wq);
5230 if (mddev->sync_thread) {
5231 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5232 md_reap_sync_thread(mddev);
5235 del_timer_sync(&mddev->safemode_timer);
5237 bitmap_flush(mddev);
5238 md_super_wait(mddev);
5240 if (mddev->ro == 0 &&
5241 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5242 /* mark array as shutdown cleanly */
5244 md_update_sb(mddev, 1);
5248 void md_stop_writes(struct mddev *mddev)
5250 mddev_lock_nointr(mddev);
5251 __md_stop_writes(mddev);
5252 mddev_unlock(mddev);
5254 EXPORT_SYMBOL_GPL(md_stop_writes);
5256 static void mddev_detach(struct mddev *mddev)
5258 struct bitmap *bitmap = mddev->bitmap;
5259 /* wait for behind writes to complete */
5260 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5261 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5263 /* need to kick something here to make sure I/O goes? */
5264 wait_event(bitmap->behind_wait,
5265 atomic_read(&bitmap->behind_writes) == 0);
5267 if (mddev->pers && mddev->pers->quiesce) {
5268 mddev->pers->quiesce(mddev, 1);
5269 mddev->pers->quiesce(mddev, 0);
5271 md_unregister_thread(&mddev->thread);
5273 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5276 static void __md_stop(struct mddev *mddev)
5278 struct md_personality *pers = mddev->pers;
5279 mddev_detach(mddev);
5280 spin_lock(&mddev->lock);
5283 spin_unlock(&mddev->lock);
5284 pers->free(mddev, mddev->private);
5285 if (pers->sync_request && mddev->to_remove == NULL)
5286 mddev->to_remove = &md_redundancy_group;
5287 module_put(pers->owner);
5288 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5291 void md_stop(struct mddev *mddev)
5293 /* stop the array and free an attached data structures.
5294 * This is called from dm-raid
5297 bitmap_destroy(mddev);
5299 bioset_free(mddev->bio_set);
5302 EXPORT_SYMBOL_GPL(md_stop);
5304 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5309 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5311 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5312 md_wakeup_thread(mddev->thread);
5314 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5315 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5316 if (mddev->sync_thread)
5317 /* Thread might be blocked waiting for metadata update
5318 * which will now never happen */
5319 wake_up_process(mddev->sync_thread->tsk);
5321 mddev_unlock(mddev);
5322 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5324 mddev_lock_nointr(mddev);
5326 mutex_lock(&mddev->open_mutex);
5327 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5328 mddev->sync_thread ||
5329 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5330 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5331 printk("md: %s still in use.\n",mdname(mddev));
5333 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5334 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5335 md_wakeup_thread(mddev->thread);
5341 __md_stop_writes(mddev);
5347 set_disk_ro(mddev->gendisk, 1);
5348 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5349 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5350 md_wakeup_thread(mddev->thread);
5351 sysfs_notify_dirent_safe(mddev->sysfs_state);
5355 mutex_unlock(&mddev->open_mutex);
5360 * 0 - completely stop and dis-assemble array
5361 * 2 - stop but do not disassemble array
5363 static int do_md_stop(struct mddev *mddev, int mode,
5364 struct block_device *bdev)
5366 struct gendisk *disk = mddev->gendisk;
5367 struct md_rdev *rdev;
5370 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5372 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5373 md_wakeup_thread(mddev->thread);
5375 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5376 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5377 if (mddev->sync_thread)
5378 /* Thread might be blocked waiting for metadata update
5379 * which will now never happen */
5380 wake_up_process(mddev->sync_thread->tsk);
5382 mddev_unlock(mddev);
5383 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5384 !test_bit(MD_RECOVERY_RUNNING,
5385 &mddev->recovery)));
5386 mddev_lock_nointr(mddev);
5388 mutex_lock(&mddev->open_mutex);
5389 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5390 mddev->sysfs_active ||
5391 mddev->sync_thread ||
5392 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5393 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5394 printk("md: %s still in use.\n",mdname(mddev));
5395 mutex_unlock(&mddev->open_mutex);
5397 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5398 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5399 md_wakeup_thread(mddev->thread);
5405 set_disk_ro(disk, 0);
5407 __md_stop_writes(mddev);
5409 mddev->queue->merge_bvec_fn = NULL;
5410 mddev->queue->backing_dev_info.congested_fn = NULL;
5412 /* tell userspace to handle 'inactive' */
5413 sysfs_notify_dirent_safe(mddev->sysfs_state);
5415 rdev_for_each(rdev, mddev)
5416 if (rdev->raid_disk >= 0)
5417 sysfs_unlink_rdev(mddev, rdev);
5419 set_capacity(disk, 0);
5420 mutex_unlock(&mddev->open_mutex);
5422 revalidate_disk(disk);
5427 mutex_unlock(&mddev->open_mutex);
5429 * Free resources if final stop
5432 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5434 bitmap_destroy(mddev);
5435 if (mddev->bitmap_info.file) {
5436 struct file *f = mddev->bitmap_info.file;
5437 spin_lock(&mddev->lock);
5438 mddev->bitmap_info.file = NULL;
5439 spin_unlock(&mddev->lock);
5442 mddev->bitmap_info.offset = 0;
5444 export_array(mddev);
5447 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5448 if (mddev->hold_active == UNTIL_STOP)
5449 mddev->hold_active = 0;
5451 blk_integrity_unregister(disk);
5452 md_new_event(mddev);
5453 sysfs_notify_dirent_safe(mddev->sysfs_state);
5458 static void autorun_array(struct mddev *mddev)
5460 struct md_rdev *rdev;
5463 if (list_empty(&mddev->disks))
5466 printk(KERN_INFO "md: running: ");
5468 rdev_for_each(rdev, mddev) {
5469 char b[BDEVNAME_SIZE];
5470 printk("<%s>", bdevname(rdev->bdev,b));
5474 err = do_md_run(mddev);
5476 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5477 do_md_stop(mddev, 0, NULL);
5482 * lets try to run arrays based on all disks that have arrived
5483 * until now. (those are in pending_raid_disks)
5485 * the method: pick the first pending disk, collect all disks with
5486 * the same UUID, remove all from the pending list and put them into
5487 * the 'same_array' list. Then order this list based on superblock
5488 * update time (freshest comes first), kick out 'old' disks and
5489 * compare superblocks. If everything's fine then run it.
5491 * If "unit" is allocated, then bump its reference count
5493 static void autorun_devices(int part)
5495 struct md_rdev *rdev0, *rdev, *tmp;
5496 struct mddev *mddev;
5497 char b[BDEVNAME_SIZE];
5499 printk(KERN_INFO "md: autorun ...\n");
5500 while (!list_empty(&pending_raid_disks)) {
5503 LIST_HEAD(candidates);
5504 rdev0 = list_entry(pending_raid_disks.next,
5505 struct md_rdev, same_set);
5507 printk(KERN_INFO "md: considering %s ...\n",
5508 bdevname(rdev0->bdev,b));
5509 INIT_LIST_HEAD(&candidates);
5510 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5511 if (super_90_load(rdev, rdev0, 0) >= 0) {
5512 printk(KERN_INFO "md: adding %s ...\n",
5513 bdevname(rdev->bdev,b));
5514 list_move(&rdev->same_set, &candidates);
5517 * now we have a set of devices, with all of them having
5518 * mostly sane superblocks. It's time to allocate the
5522 dev = MKDEV(mdp_major,
5523 rdev0->preferred_minor << MdpMinorShift);
5524 unit = MINOR(dev) >> MdpMinorShift;
5526 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5529 if (rdev0->preferred_minor != unit) {
5530 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5531 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5535 md_probe(dev, NULL, NULL);
5536 mddev = mddev_find(dev);
5537 if (!mddev || !mddev->gendisk) {
5541 "md: cannot allocate memory for md drive.\n");
5544 if (mddev_lock(mddev))
5545 printk(KERN_WARNING "md: %s locked, cannot run\n",
5547 else if (mddev->raid_disks || mddev->major_version
5548 || !list_empty(&mddev->disks)) {
5550 "md: %s already running, cannot run %s\n",
5551 mdname(mddev), bdevname(rdev0->bdev,b));
5552 mddev_unlock(mddev);
5554 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5555 mddev->persistent = 1;
5556 rdev_for_each_list(rdev, tmp, &candidates) {
5557 list_del_init(&rdev->same_set);
5558 if (bind_rdev_to_array(rdev, mddev))
5561 autorun_array(mddev);
5562 mddev_unlock(mddev);
5564 /* on success, candidates will be empty, on error
5567 rdev_for_each_list(rdev, tmp, &candidates) {
5568 list_del_init(&rdev->same_set);
5573 printk(KERN_INFO "md: ... autorun DONE.\n");
5575 #endif /* !MODULE */
5577 static int get_version(void __user *arg)
5581 ver.major = MD_MAJOR_VERSION;
5582 ver.minor = MD_MINOR_VERSION;
5583 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5585 if (copy_to_user(arg, &ver, sizeof(ver)))
5591 static int get_array_info(struct mddev *mddev, void __user *arg)
5593 mdu_array_info_t info;
5594 int nr,working,insync,failed,spare;
5595 struct md_rdev *rdev;
5597 nr = working = insync = failed = spare = 0;
5599 rdev_for_each_rcu(rdev, mddev) {
5601 if (test_bit(Faulty, &rdev->flags))
5605 if (test_bit(In_sync, &rdev->flags))
5613 info.major_version = mddev->major_version;
5614 info.minor_version = mddev->minor_version;
5615 info.patch_version = MD_PATCHLEVEL_VERSION;
5616 info.ctime = mddev->ctime;
5617 info.level = mddev->level;
5618 info.size = mddev->dev_sectors / 2;
5619 if (info.size != mddev->dev_sectors / 2) /* overflow */
5622 info.raid_disks = mddev->raid_disks;
5623 info.md_minor = mddev->md_minor;
5624 info.not_persistent= !mddev->persistent;
5626 info.utime = mddev->utime;
5629 info.state = (1<<MD_SB_CLEAN);
5630 if (mddev->bitmap && mddev->bitmap_info.offset)
5631 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5632 info.active_disks = insync;
5633 info.working_disks = working;
5634 info.failed_disks = failed;
5635 info.spare_disks = spare;
5637 info.layout = mddev->layout;
5638 info.chunk_size = mddev->chunk_sectors << 9;
5640 if (copy_to_user(arg, &info, sizeof(info)))
5646 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5648 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5652 file = kmalloc(sizeof(*file), GFP_NOIO);
5657 spin_lock(&mddev->lock);
5658 /* bitmap disabled, zero the first byte and copy out */
5659 if (!mddev->bitmap_info.file)
5660 file->pathname[0] = '\0';
5661 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5662 file->pathname, sizeof(file->pathname))),
5666 memmove(file->pathname, ptr,
5667 sizeof(file->pathname)-(ptr-file->pathname));
5668 spin_unlock(&mddev->lock);
5671 copy_to_user(arg, file, sizeof(*file)))
5678 static int get_disk_info(struct mddev *mddev, void __user * arg)
5680 mdu_disk_info_t info;
5681 struct md_rdev *rdev;
5683 if (copy_from_user(&info, arg, sizeof(info)))
5687 rdev = find_rdev_nr_rcu(mddev, info.number);
5689 info.major = MAJOR(rdev->bdev->bd_dev);
5690 info.minor = MINOR(rdev->bdev->bd_dev);
5691 info.raid_disk = rdev->raid_disk;
5693 if (test_bit(Faulty, &rdev->flags))
5694 info.state |= (1<<MD_DISK_FAULTY);
5695 else if (test_bit(In_sync, &rdev->flags)) {
5696 info.state |= (1<<MD_DISK_ACTIVE);
5697 info.state |= (1<<MD_DISK_SYNC);
5699 if (test_bit(WriteMostly, &rdev->flags))
5700 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5702 info.major = info.minor = 0;
5703 info.raid_disk = -1;
5704 info.state = (1<<MD_DISK_REMOVED);
5708 if (copy_to_user(arg, &info, sizeof(info)))
5714 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5716 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5717 struct md_rdev *rdev;
5718 dev_t dev = MKDEV(info->major,info->minor);
5720 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5723 if (!mddev->raid_disks) {
5725 /* expecting a device which has a superblock */
5726 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5729 "md: md_import_device returned %ld\n",
5731 return PTR_ERR(rdev);
5733 if (!list_empty(&mddev->disks)) {
5734 struct md_rdev *rdev0
5735 = list_entry(mddev->disks.next,
5736 struct md_rdev, same_set);
5737 err = super_types[mddev->major_version]
5738 .load_super(rdev, rdev0, mddev->minor_version);
5741 "md: %s has different UUID to %s\n",
5742 bdevname(rdev->bdev,b),
5743 bdevname(rdev0->bdev,b2));
5748 err = bind_rdev_to_array(rdev, mddev);
5755 * add_new_disk can be used once the array is assembled
5756 * to add "hot spares". They must already have a superblock
5761 if (!mddev->pers->hot_add_disk) {
5763 "%s: personality does not support diskops!\n",
5767 if (mddev->persistent)
5768 rdev = md_import_device(dev, mddev->major_version,
5769 mddev->minor_version);
5771 rdev = md_import_device(dev, -1, -1);
5774 "md: md_import_device returned %ld\n",
5776 return PTR_ERR(rdev);
5778 /* set saved_raid_disk if appropriate */
5779 if (!mddev->persistent) {
5780 if (info->state & (1<<MD_DISK_SYNC) &&
5781 info->raid_disk < mddev->raid_disks) {
5782 rdev->raid_disk = info->raid_disk;
5783 set_bit(In_sync, &rdev->flags);
5784 clear_bit(Bitmap_sync, &rdev->flags);
5786 rdev->raid_disk = -1;
5787 rdev->saved_raid_disk = rdev->raid_disk;
5789 super_types[mddev->major_version].
5790 validate_super(mddev, rdev);
5791 if ((info->state & (1<<MD_DISK_SYNC)) &&
5792 rdev->raid_disk != info->raid_disk) {
5793 /* This was a hot-add request, but events doesn't
5794 * match, so reject it.
5800 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5801 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5802 set_bit(WriteMostly, &rdev->flags);
5804 clear_bit(WriteMostly, &rdev->flags);
5806 rdev->raid_disk = -1;
5807 err = bind_rdev_to_array(rdev, mddev);
5808 if (!err && !mddev->pers->hot_remove_disk) {
5809 /* If there is hot_add_disk but no hot_remove_disk
5810 * then added disks for geometry changes,
5811 * and should be added immediately.
5813 super_types[mddev->major_version].
5814 validate_super(mddev, rdev);
5815 err = mddev->pers->hot_add_disk(mddev, rdev);
5817 unbind_rdev_from_array(rdev);
5822 sysfs_notify_dirent_safe(rdev->sysfs_state);
5824 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5825 if (mddev->degraded)
5826 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5827 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5829 md_new_event(mddev);
5830 md_wakeup_thread(mddev->thread);
5834 /* otherwise, add_new_disk is only allowed
5835 * for major_version==0 superblocks
5837 if (mddev->major_version != 0) {
5838 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5843 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5845 rdev = md_import_device(dev, -1, 0);
5848 "md: error, md_import_device() returned %ld\n",
5850 return PTR_ERR(rdev);
5852 rdev->desc_nr = info->number;
5853 if (info->raid_disk < mddev->raid_disks)
5854 rdev->raid_disk = info->raid_disk;
5856 rdev->raid_disk = -1;
5858 if (rdev->raid_disk < mddev->raid_disks)
5859 if (info->state & (1<<MD_DISK_SYNC))
5860 set_bit(In_sync, &rdev->flags);
5862 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5863 set_bit(WriteMostly, &rdev->flags);
5865 if (!mddev->persistent) {
5866 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5867 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5869 rdev->sb_start = calc_dev_sboffset(rdev);
5870 rdev->sectors = rdev->sb_start;
5872 err = bind_rdev_to_array(rdev, mddev);
5882 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5884 char b[BDEVNAME_SIZE];
5885 struct md_rdev *rdev;
5887 rdev = find_rdev(mddev, dev);
5891 clear_bit(Blocked, &rdev->flags);
5892 remove_and_add_spares(mddev, rdev);
5894 if (rdev->raid_disk >= 0)
5897 kick_rdev_from_array(rdev);
5898 md_update_sb(mddev, 1);
5899 md_new_event(mddev);
5903 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5904 bdevname(rdev->bdev,b), mdname(mddev));
5908 static int hot_add_disk(struct mddev *mddev, dev_t dev)
5910 char b[BDEVNAME_SIZE];
5912 struct md_rdev *rdev;
5917 if (mddev->major_version != 0) {
5918 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5919 " version-0 superblocks.\n",
5923 if (!mddev->pers->hot_add_disk) {
5925 "%s: personality does not support diskops!\n",
5930 rdev = md_import_device(dev, -1, 0);
5933 "md: error, md_import_device() returned %ld\n",
5938 if (mddev->persistent)
5939 rdev->sb_start = calc_dev_sboffset(rdev);
5941 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5943 rdev->sectors = rdev->sb_start;
5945 if (test_bit(Faulty, &rdev->flags)) {
5947 "md: can not hot-add faulty %s disk to %s!\n",
5948 bdevname(rdev->bdev,b), mdname(mddev));
5952 clear_bit(In_sync, &rdev->flags);
5954 rdev->saved_raid_disk = -1;
5955 err = bind_rdev_to_array(rdev, mddev);
5960 * The rest should better be atomic, we can have disk failures
5961 * noticed in interrupt contexts ...
5964 rdev->raid_disk = -1;
5966 md_update_sb(mddev, 1);
5969 * Kick recovery, maybe this spare has to be added to the
5970 * array immediately.
5972 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5973 md_wakeup_thread(mddev->thread);
5974 md_new_event(mddev);
5982 static int set_bitmap_file(struct mddev *mddev, int fd)
5987 if (!mddev->pers->quiesce || !mddev->thread)
5989 if (mddev->recovery || mddev->sync_thread)
5991 /* we should be able to change the bitmap.. */
5995 struct inode *inode;
5998 if (mddev->bitmap || mddev->bitmap_info.file)
5999 return -EEXIST; /* cannot add when bitmap is present */
6003 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6008 inode = f->f_mapping->host;
6009 if (!S_ISREG(inode->i_mode)) {
6010 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6013 } else if (!(f->f_mode & FMODE_WRITE)) {
6014 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6017 } else if (atomic_read(&inode->i_writecount) != 1) {
6018 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6026 mddev->bitmap_info.file = f;
6027 mddev->bitmap_info.offset = 0; /* file overrides offset */
6028 } else if (mddev->bitmap == NULL)
6029 return -ENOENT; /* cannot remove what isn't there */
6032 mddev->pers->quiesce(mddev, 1);
6034 err = bitmap_create(mddev);
6036 err = bitmap_load(mddev);
6038 if (fd < 0 || err) {
6039 bitmap_destroy(mddev);
6040 fd = -1; /* make sure to put the file */
6042 mddev->pers->quiesce(mddev, 0);
6045 struct file *f = mddev->bitmap_info.file;
6047 spin_lock(&mddev->lock);
6048 mddev->bitmap_info.file = NULL;
6049 spin_unlock(&mddev->lock);
6058 * set_array_info is used two different ways
6059 * The original usage is when creating a new array.
6060 * In this usage, raid_disks is > 0 and it together with
6061 * level, size, not_persistent,layout,chunksize determine the
6062 * shape of the array.
6063 * This will always create an array with a type-0.90.0 superblock.
6064 * The newer usage is when assembling an array.
6065 * In this case raid_disks will be 0, and the major_version field is
6066 * use to determine which style super-blocks are to be found on the devices.
6067 * The minor and patch _version numbers are also kept incase the
6068 * super_block handler wishes to interpret them.
6070 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6073 if (info->raid_disks == 0) {
6074 /* just setting version number for superblock loading */
6075 if (info->major_version < 0 ||
6076 info->major_version >= ARRAY_SIZE(super_types) ||
6077 super_types[info->major_version].name == NULL) {
6078 /* maybe try to auto-load a module? */
6080 "md: superblock version %d not known\n",
6081 info->major_version);
6084 mddev->major_version = info->major_version;
6085 mddev->minor_version = info->minor_version;
6086 mddev->patch_version = info->patch_version;
6087 mddev->persistent = !info->not_persistent;
6088 /* ensure mddev_put doesn't delete this now that there
6089 * is some minimal configuration.
6091 mddev->ctime = get_seconds();
6094 mddev->major_version = MD_MAJOR_VERSION;
6095 mddev->minor_version = MD_MINOR_VERSION;
6096 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6097 mddev->ctime = get_seconds();
6099 mddev->level = info->level;
6100 mddev->clevel[0] = 0;
6101 mddev->dev_sectors = 2 * (sector_t)info->size;
6102 mddev->raid_disks = info->raid_disks;
6103 /* don't set md_minor, it is determined by which /dev/md* was
6106 if (info->state & (1<<MD_SB_CLEAN))
6107 mddev->recovery_cp = MaxSector;
6109 mddev->recovery_cp = 0;
6110 mddev->persistent = ! info->not_persistent;
6111 mddev->external = 0;
6113 mddev->layout = info->layout;
6114 mddev->chunk_sectors = info->chunk_size >> 9;
6116 mddev->max_disks = MD_SB_DISKS;
6118 if (mddev->persistent)
6120 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6122 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6123 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6124 mddev->bitmap_info.offset = 0;
6126 mddev->reshape_position = MaxSector;
6129 * Generate a 128 bit UUID
6131 get_random_bytes(mddev->uuid, 16);
6133 mddev->new_level = mddev->level;
6134 mddev->new_chunk_sectors = mddev->chunk_sectors;
6135 mddev->new_layout = mddev->layout;
6136 mddev->delta_disks = 0;
6137 mddev->reshape_backwards = 0;
6142 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6144 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6146 if (mddev->external_size)
6149 mddev->array_sectors = array_sectors;
6151 EXPORT_SYMBOL(md_set_array_sectors);
6153 static int update_size(struct mddev *mddev, sector_t num_sectors)
6155 struct md_rdev *rdev;
6157 int fit = (num_sectors == 0);
6159 if (mddev->pers->resize == NULL)
6161 /* The "num_sectors" is the number of sectors of each device that
6162 * is used. This can only make sense for arrays with redundancy.
6163 * linear and raid0 always use whatever space is available. We can only
6164 * consider changing this number if no resync or reconstruction is
6165 * happening, and if the new size is acceptable. It must fit before the
6166 * sb_start or, if that is <data_offset, it must fit before the size
6167 * of each device. If num_sectors is zero, we find the largest size
6170 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6176 rdev_for_each(rdev, mddev) {
6177 sector_t avail = rdev->sectors;
6179 if (fit && (num_sectors == 0 || num_sectors > avail))
6180 num_sectors = avail;
6181 if (avail < num_sectors)
6184 rv = mddev->pers->resize(mddev, num_sectors);
6186 revalidate_disk(mddev->gendisk);
6190 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6193 struct md_rdev *rdev;
6194 /* change the number of raid disks */
6195 if (mddev->pers->check_reshape == NULL)
6199 if (raid_disks <= 0 ||
6200 (mddev->max_disks && raid_disks >= mddev->max_disks))
6202 if (mddev->sync_thread ||
6203 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6204 mddev->reshape_position != MaxSector)
6207 rdev_for_each(rdev, mddev) {
6208 if (mddev->raid_disks < raid_disks &&
6209 rdev->data_offset < rdev->new_data_offset)
6211 if (mddev->raid_disks > raid_disks &&
6212 rdev->data_offset > rdev->new_data_offset)
6216 mddev->delta_disks = raid_disks - mddev->raid_disks;
6217 if (mddev->delta_disks < 0)
6218 mddev->reshape_backwards = 1;
6219 else if (mddev->delta_disks > 0)
6220 mddev->reshape_backwards = 0;
6222 rv = mddev->pers->check_reshape(mddev);
6224 mddev->delta_disks = 0;
6225 mddev->reshape_backwards = 0;
6231 * update_array_info is used to change the configuration of an
6233 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6234 * fields in the info are checked against the array.
6235 * Any differences that cannot be handled will cause an error.
6236 * Normally, only one change can be managed at a time.
6238 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6244 /* calculate expected state,ignoring low bits */
6245 if (mddev->bitmap && mddev->bitmap_info.offset)
6246 state |= (1 << MD_SB_BITMAP_PRESENT);
6248 if (mddev->major_version != info->major_version ||
6249 mddev->minor_version != info->minor_version ||
6250 /* mddev->patch_version != info->patch_version || */
6251 mddev->ctime != info->ctime ||
6252 mddev->level != info->level ||
6253 /* mddev->layout != info->layout || */
6254 !mddev->persistent != info->not_persistent||
6255 mddev->chunk_sectors != info->chunk_size >> 9 ||
6256 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6257 ((state^info->state) & 0xfffffe00)
6260 /* Check there is only one change */
6261 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6263 if (mddev->raid_disks != info->raid_disks)
6265 if (mddev->layout != info->layout)
6267 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6274 if (mddev->layout != info->layout) {
6276 * we don't need to do anything at the md level, the
6277 * personality will take care of it all.
6279 if (mddev->pers->check_reshape == NULL)
6282 mddev->new_layout = info->layout;
6283 rv = mddev->pers->check_reshape(mddev);
6285 mddev->new_layout = mddev->layout;
6289 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6290 rv = update_size(mddev, (sector_t)info->size * 2);
6292 if (mddev->raid_disks != info->raid_disks)
6293 rv = update_raid_disks(mddev, info->raid_disks);
6295 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6296 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
6298 if (mddev->recovery || mddev->sync_thread)
6300 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6301 /* add the bitmap */
6304 if (mddev->bitmap_info.default_offset == 0)
6306 mddev->bitmap_info.offset =
6307 mddev->bitmap_info.default_offset;
6308 mddev->bitmap_info.space =
6309 mddev->bitmap_info.default_space;
6310 mddev->pers->quiesce(mddev, 1);
6311 rv = bitmap_create(mddev);
6313 rv = bitmap_load(mddev);
6315 bitmap_destroy(mddev);
6316 mddev->pers->quiesce(mddev, 0);
6318 /* remove the bitmap */
6321 if (mddev->bitmap->storage.file)
6323 mddev->pers->quiesce(mddev, 1);
6324 bitmap_destroy(mddev);
6325 mddev->pers->quiesce(mddev, 0);
6326 mddev->bitmap_info.offset = 0;
6329 md_update_sb(mddev, 1);
6333 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6335 struct md_rdev *rdev;
6338 if (mddev->pers == NULL)
6342 rdev = find_rdev_rcu(mddev, dev);
6346 md_error(mddev, rdev);
6347 if (!test_bit(Faulty, &rdev->flags))
6355 * We have a problem here : there is no easy way to give a CHS
6356 * virtual geometry. We currently pretend that we have a 2 heads
6357 * 4 sectors (with a BIG number of cylinders...). This drives
6358 * dosfs just mad... ;-)
6360 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6362 struct mddev *mddev = bdev->bd_disk->private_data;
6366 geo->cylinders = mddev->array_sectors / 8;
6370 static inline bool md_ioctl_valid(unsigned int cmd)
6375 case GET_ARRAY_INFO:
6376 case GET_BITMAP_FILE:
6379 case HOT_REMOVE_DISK:
6382 case RESTART_ARRAY_RW:
6384 case SET_ARRAY_INFO:
6385 case SET_BITMAP_FILE:
6386 case SET_DISK_FAULTY:
6395 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6396 unsigned int cmd, unsigned long arg)
6399 void __user *argp = (void __user *)arg;
6400 struct mddev *mddev = NULL;
6403 if (!md_ioctl_valid(cmd))
6408 case GET_ARRAY_INFO:
6412 if (!capable(CAP_SYS_ADMIN))
6417 * Commands dealing with the RAID driver but not any
6422 err = get_version(argp);
6428 autostart_arrays(arg);
6435 * Commands creating/starting a new array:
6438 mddev = bdev->bd_disk->private_data;
6445 /* Some actions do not requires the mutex */
6447 case GET_ARRAY_INFO:
6448 if (!mddev->raid_disks && !mddev->external)
6451 err = get_array_info(mddev, argp);
6455 if (!mddev->raid_disks && !mddev->external)
6458 err = get_disk_info(mddev, argp);
6461 case SET_DISK_FAULTY:
6462 err = set_disk_faulty(mddev, new_decode_dev(arg));
6465 case GET_BITMAP_FILE:
6466 err = get_bitmap_file(mddev, argp);
6471 if (cmd == ADD_NEW_DISK)
6472 /* need to ensure md_delayed_delete() has completed */
6473 flush_workqueue(md_misc_wq);
6475 if (cmd == HOT_REMOVE_DISK)
6476 /* need to ensure recovery thread has run */
6477 wait_event_interruptible_timeout(mddev->sb_wait,
6478 !test_bit(MD_RECOVERY_NEEDED,
6480 msecs_to_jiffies(5000));
6481 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6482 /* Need to flush page cache, and ensure no-one else opens
6485 mutex_lock(&mddev->open_mutex);
6486 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6487 mutex_unlock(&mddev->open_mutex);
6491 set_bit(MD_STILL_CLOSED, &mddev->flags);
6492 mutex_unlock(&mddev->open_mutex);
6493 sync_blockdev(bdev);
6495 err = mddev_lock(mddev);
6498 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6503 if (cmd == SET_ARRAY_INFO) {
6504 mdu_array_info_t info;
6506 memset(&info, 0, sizeof(info));
6507 else if (copy_from_user(&info, argp, sizeof(info))) {
6512 err = update_array_info(mddev, &info);
6514 printk(KERN_WARNING "md: couldn't update"
6515 " array info. %d\n", err);
6520 if (!list_empty(&mddev->disks)) {
6522 "md: array %s already has disks!\n",
6527 if (mddev->raid_disks) {
6529 "md: array %s already initialised!\n",
6534 err = set_array_info(mddev, &info);
6536 printk(KERN_WARNING "md: couldn't set"
6537 " array info. %d\n", err);
6544 * Commands querying/configuring an existing array:
6546 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6547 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6548 if ((!mddev->raid_disks && !mddev->external)
6549 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6550 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6551 && cmd != GET_BITMAP_FILE) {
6557 * Commands even a read-only array can execute:
6560 case RESTART_ARRAY_RW:
6561 err = restart_array(mddev);
6565 err = do_md_stop(mddev, 0, bdev);
6569 err = md_set_readonly(mddev, bdev);
6572 case HOT_REMOVE_DISK:
6573 err = hot_remove_disk(mddev, new_decode_dev(arg));
6577 /* We can support ADD_NEW_DISK on read-only arrays
6578 * on if we are re-adding a preexisting device.
6579 * So require mddev->pers and MD_DISK_SYNC.
6582 mdu_disk_info_t info;
6583 if (copy_from_user(&info, argp, sizeof(info)))
6585 else if (!(info.state & (1<<MD_DISK_SYNC)))
6586 /* Need to clear read-only for this */
6589 err = add_new_disk(mddev, &info);
6595 if (get_user(ro, (int __user *)(arg))) {
6601 /* if the bdev is going readonly the value of mddev->ro
6602 * does not matter, no writes are coming
6607 /* are we are already prepared for writes? */
6611 /* transitioning to readauto need only happen for
6612 * arrays that call md_write_start
6615 err = restart_array(mddev);
6618 set_disk_ro(mddev->gendisk, 0);
6625 * The remaining ioctls are changing the state of the
6626 * superblock, so we do not allow them on read-only arrays.
6628 if (mddev->ro && mddev->pers) {
6629 if (mddev->ro == 2) {
6631 sysfs_notify_dirent_safe(mddev->sysfs_state);
6632 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6633 /* mddev_unlock will wake thread */
6634 /* If a device failed while we were read-only, we
6635 * need to make sure the metadata is updated now.
6637 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6638 mddev_unlock(mddev);
6639 wait_event(mddev->sb_wait,
6640 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6641 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6642 mddev_lock_nointr(mddev);
6653 mdu_disk_info_t info;
6654 if (copy_from_user(&info, argp, sizeof(info)))
6657 err = add_new_disk(mddev, &info);
6662 err = hot_add_disk(mddev, new_decode_dev(arg));
6666 err = do_md_run(mddev);
6669 case SET_BITMAP_FILE:
6670 err = set_bitmap_file(mddev, (int)arg);
6679 if (mddev->hold_active == UNTIL_IOCTL &&
6681 mddev->hold_active = 0;
6682 mddev_unlock(mddev);
6686 #ifdef CONFIG_COMPAT
6687 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6688 unsigned int cmd, unsigned long arg)
6691 case HOT_REMOVE_DISK:
6693 case SET_DISK_FAULTY:
6694 case SET_BITMAP_FILE:
6695 /* These take in integer arg, do not convert */
6698 arg = (unsigned long)compat_ptr(arg);
6702 return md_ioctl(bdev, mode, cmd, arg);
6704 #endif /* CONFIG_COMPAT */
6706 static int md_open(struct block_device *bdev, fmode_t mode)
6709 * Succeed if we can lock the mddev, which confirms that
6710 * it isn't being stopped right now.
6712 struct mddev *mddev = mddev_find(bdev->bd_dev);
6718 if (mddev->gendisk != bdev->bd_disk) {
6719 /* we are racing with mddev_put which is discarding this
6723 /* Wait until bdev->bd_disk is definitely gone */
6724 flush_workqueue(md_misc_wq);
6725 /* Then retry the open from the top */
6726 return -ERESTARTSYS;
6728 BUG_ON(mddev != bdev->bd_disk->private_data);
6730 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6734 atomic_inc(&mddev->openers);
6735 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6736 mutex_unlock(&mddev->open_mutex);
6738 check_disk_change(bdev);
6743 static void md_release(struct gendisk *disk, fmode_t mode)
6745 struct mddev *mddev = disk->private_data;
6748 atomic_dec(&mddev->openers);
6752 static int md_media_changed(struct gendisk *disk)
6754 struct mddev *mddev = disk->private_data;
6756 return mddev->changed;
6759 static int md_revalidate(struct gendisk *disk)
6761 struct mddev *mddev = disk->private_data;
6766 static const struct block_device_operations md_fops =
6768 .owner = THIS_MODULE,
6770 .release = md_release,
6772 #ifdef CONFIG_COMPAT
6773 .compat_ioctl = md_compat_ioctl,
6775 .getgeo = md_getgeo,
6776 .media_changed = md_media_changed,
6777 .revalidate_disk= md_revalidate,
6780 static int md_thread(void *arg)
6782 struct md_thread *thread = arg;
6785 * md_thread is a 'system-thread', it's priority should be very
6786 * high. We avoid resource deadlocks individually in each
6787 * raid personality. (RAID5 does preallocation) We also use RR and
6788 * the very same RT priority as kswapd, thus we will never get
6789 * into a priority inversion deadlock.
6791 * we definitely have to have equal or higher priority than
6792 * bdflush, otherwise bdflush will deadlock if there are too
6793 * many dirty RAID5 blocks.
6796 allow_signal(SIGKILL);
6797 while (!kthread_should_stop()) {
6799 /* We need to wait INTERRUPTIBLE so that
6800 * we don't add to the load-average.
6801 * That means we need to be sure no signals are
6804 if (signal_pending(current))
6805 flush_signals(current);
6807 wait_event_interruptible_timeout
6809 test_bit(THREAD_WAKEUP, &thread->flags)
6810 || kthread_should_stop(),
6813 clear_bit(THREAD_WAKEUP, &thread->flags);
6814 if (!kthread_should_stop())
6815 thread->run(thread);
6821 void md_wakeup_thread(struct md_thread *thread)
6824 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6825 set_bit(THREAD_WAKEUP, &thread->flags);
6826 wake_up(&thread->wqueue);
6829 EXPORT_SYMBOL(md_wakeup_thread);
6831 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6832 struct mddev *mddev, const char *name)
6834 struct md_thread *thread;
6836 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6840 init_waitqueue_head(&thread->wqueue);
6843 thread->mddev = mddev;
6844 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6845 thread->tsk = kthread_run(md_thread, thread,
6847 mdname(thread->mddev),
6849 if (IS_ERR(thread->tsk)) {
6855 EXPORT_SYMBOL(md_register_thread);
6857 void md_unregister_thread(struct md_thread **threadp)
6859 struct md_thread *thread = *threadp;
6862 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6863 /* Locking ensures that mddev_unlock does not wake_up a
6864 * non-existent thread
6866 spin_lock(&pers_lock);
6868 spin_unlock(&pers_lock);
6870 kthread_stop(thread->tsk);
6873 EXPORT_SYMBOL(md_unregister_thread);
6875 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6877 if (!rdev || test_bit(Faulty, &rdev->flags))
6880 if (!mddev->pers || !mddev->pers->error_handler)
6882 mddev->pers->error_handler(mddev,rdev);
6883 if (mddev->degraded)
6884 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6885 sysfs_notify_dirent_safe(rdev->sysfs_state);
6886 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6887 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6888 md_wakeup_thread(mddev->thread);
6889 if (mddev->event_work.func)
6890 queue_work(md_misc_wq, &mddev->event_work);
6891 md_new_event_inintr(mddev);
6893 EXPORT_SYMBOL(md_error);
6895 /* seq_file implementation /proc/mdstat */
6897 static void status_unused(struct seq_file *seq)
6900 struct md_rdev *rdev;
6902 seq_printf(seq, "unused devices: ");
6904 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6905 char b[BDEVNAME_SIZE];
6907 seq_printf(seq, "%s ",
6908 bdevname(rdev->bdev,b));
6911 seq_printf(seq, "<none>");
6913 seq_printf(seq, "\n");
6916 static void status_resync(struct seq_file *seq, struct mddev *mddev)
6918 sector_t max_sectors, resync, res;
6919 unsigned long dt, db;
6922 unsigned int per_milli;
6924 if (mddev->curr_resync <= 3)
6927 resync = mddev->curr_resync
6928 - atomic_read(&mddev->recovery_active);
6930 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6931 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6932 max_sectors = mddev->resync_max_sectors;
6934 max_sectors = mddev->dev_sectors;
6936 WARN_ON(max_sectors == 0);
6937 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6938 * in a sector_t, and (max_sectors>>scale) will fit in a
6939 * u32, as those are the requirements for sector_div.
6940 * Thus 'scale' must be at least 10
6943 if (sizeof(sector_t) > sizeof(unsigned long)) {
6944 while ( max_sectors/2 > (1ULL<<(scale+32)))
6947 res = (resync>>scale)*1000;
6948 sector_div(res, (u32)((max_sectors>>scale)+1));
6952 int i, x = per_milli/50, y = 20-x;
6953 seq_printf(seq, "[");
6954 for (i = 0; i < x; i++)
6955 seq_printf(seq, "=");
6956 seq_printf(seq, ">");
6957 for (i = 0; i < y; i++)
6958 seq_printf(seq, ".");
6959 seq_printf(seq, "] ");
6961 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6962 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6964 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6966 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6967 "resync" : "recovery"))),
6968 per_milli/10, per_milli % 10,
6969 (unsigned long long) resync/2,
6970 (unsigned long long) max_sectors/2);
6973 * dt: time from mark until now
6974 * db: blocks written from mark until now
6975 * rt: remaining time
6977 * rt is a sector_t, so could be 32bit or 64bit.
6978 * So we divide before multiply in case it is 32bit and close
6980 * We scale the divisor (db) by 32 to avoid losing precision
6981 * near the end of resync when the number of remaining sectors
6983 * We then divide rt by 32 after multiplying by db to compensate.
6984 * The '+1' avoids division by zero if db is very small.
6986 dt = ((jiffies - mddev->resync_mark) / HZ);
6988 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6989 - mddev->resync_mark_cnt;
6991 rt = max_sectors - resync; /* number of remaining sectors */
6992 sector_div(rt, db/32+1);
6996 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6997 ((unsigned long)rt % 60)/6);
6999 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7002 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7004 struct list_head *tmp;
7006 struct mddev *mddev;
7014 spin_lock(&all_mddevs_lock);
7015 list_for_each(tmp,&all_mddevs)
7017 mddev = list_entry(tmp, struct mddev, all_mddevs);
7019 spin_unlock(&all_mddevs_lock);
7022 spin_unlock(&all_mddevs_lock);
7024 return (void*)2;/* tail */
7028 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7030 struct list_head *tmp;
7031 struct mddev *next_mddev, *mddev = v;
7037 spin_lock(&all_mddevs_lock);
7039 tmp = all_mddevs.next;
7041 tmp = mddev->all_mddevs.next;
7042 if (tmp != &all_mddevs)
7043 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7045 next_mddev = (void*)2;
7048 spin_unlock(&all_mddevs_lock);
7056 static void md_seq_stop(struct seq_file *seq, void *v)
7058 struct mddev *mddev = v;
7060 if (mddev && v != (void*)1 && v != (void*)2)
7064 static int md_seq_show(struct seq_file *seq, void *v)
7066 struct mddev *mddev = v;
7068 struct md_rdev *rdev;
7070 if (v == (void*)1) {
7071 struct md_personality *pers;
7072 seq_printf(seq, "Personalities : ");
7073 spin_lock(&pers_lock);
7074 list_for_each_entry(pers, &pers_list, list)
7075 seq_printf(seq, "[%s] ", pers->name);
7077 spin_unlock(&pers_lock);
7078 seq_printf(seq, "\n");
7079 seq->poll_event = atomic_read(&md_event_count);
7082 if (v == (void*)2) {
7087 spin_lock(&mddev->lock);
7088 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7089 seq_printf(seq, "%s : %sactive", mdname(mddev),
7090 mddev->pers ? "" : "in");
7093 seq_printf(seq, " (read-only)");
7095 seq_printf(seq, " (auto-read-only)");
7096 seq_printf(seq, " %s", mddev->pers->name);
7101 rdev_for_each_rcu(rdev, mddev) {
7102 char b[BDEVNAME_SIZE];
7103 seq_printf(seq, " %s[%d]",
7104 bdevname(rdev->bdev,b), rdev->desc_nr);
7105 if (test_bit(WriteMostly, &rdev->flags))
7106 seq_printf(seq, "(W)");
7107 if (test_bit(Faulty, &rdev->flags)) {
7108 seq_printf(seq, "(F)");
7111 if (rdev->raid_disk < 0)
7112 seq_printf(seq, "(S)"); /* spare */
7113 if (test_bit(Replacement, &rdev->flags))
7114 seq_printf(seq, "(R)");
7115 sectors += rdev->sectors;
7119 if (!list_empty(&mddev->disks)) {
7121 seq_printf(seq, "\n %llu blocks",
7122 (unsigned long long)
7123 mddev->array_sectors / 2);
7125 seq_printf(seq, "\n %llu blocks",
7126 (unsigned long long)sectors / 2);
7128 if (mddev->persistent) {
7129 if (mddev->major_version != 0 ||
7130 mddev->minor_version != 90) {
7131 seq_printf(seq," super %d.%d",
7132 mddev->major_version,
7133 mddev->minor_version);
7135 } else if (mddev->external)
7136 seq_printf(seq, " super external:%s",
7137 mddev->metadata_type);
7139 seq_printf(seq, " super non-persistent");
7142 mddev->pers->status(seq, mddev);
7143 seq_printf(seq, "\n ");
7144 if (mddev->pers->sync_request) {
7145 if (mddev->curr_resync > 2) {
7146 status_resync(seq, mddev);
7147 seq_printf(seq, "\n ");
7148 } else if (mddev->curr_resync >= 1)
7149 seq_printf(seq, "\tresync=DELAYED\n ");
7150 else if (mddev->recovery_cp < MaxSector)
7151 seq_printf(seq, "\tresync=PENDING\n ");
7154 seq_printf(seq, "\n ");
7156 bitmap_status(seq, mddev->bitmap);
7158 seq_printf(seq, "\n");
7160 spin_unlock(&mddev->lock);
7165 static const struct seq_operations md_seq_ops = {
7166 .start = md_seq_start,
7167 .next = md_seq_next,
7168 .stop = md_seq_stop,
7169 .show = md_seq_show,
7172 static int md_seq_open(struct inode *inode, struct file *file)
7174 struct seq_file *seq;
7177 error = seq_open(file, &md_seq_ops);
7181 seq = file->private_data;
7182 seq->poll_event = atomic_read(&md_event_count);
7186 static int md_unloading;
7187 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7189 struct seq_file *seq = filp->private_data;
7193 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7194 poll_wait(filp, &md_event_waiters, wait);
7196 /* always allow read */
7197 mask = POLLIN | POLLRDNORM;
7199 if (seq->poll_event != atomic_read(&md_event_count))
7200 mask |= POLLERR | POLLPRI;
7204 static const struct file_operations md_seq_fops = {
7205 .owner = THIS_MODULE,
7206 .open = md_seq_open,
7208 .llseek = seq_lseek,
7209 .release = seq_release_private,
7210 .poll = mdstat_poll,
7213 int register_md_personality(struct md_personality *p)
7215 printk(KERN_INFO "md: %s personality registered for level %d\n",
7217 spin_lock(&pers_lock);
7218 list_add_tail(&p->list, &pers_list);
7219 spin_unlock(&pers_lock);
7222 EXPORT_SYMBOL(register_md_personality);
7224 int unregister_md_personality(struct md_personality *p)
7226 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7227 spin_lock(&pers_lock);
7228 list_del_init(&p->list);
7229 spin_unlock(&pers_lock);
7232 EXPORT_SYMBOL(unregister_md_personality);
7234 static int is_mddev_idle(struct mddev *mddev, int init)
7236 struct md_rdev *rdev;
7242 rdev_for_each_rcu(rdev, mddev) {
7243 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7244 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7245 (int)part_stat_read(&disk->part0, sectors[1]) -
7246 atomic_read(&disk->sync_io);
7247 /* sync IO will cause sync_io to increase before the disk_stats
7248 * as sync_io is counted when a request starts, and
7249 * disk_stats is counted when it completes.
7250 * So resync activity will cause curr_events to be smaller than
7251 * when there was no such activity.
7252 * non-sync IO will cause disk_stat to increase without
7253 * increasing sync_io so curr_events will (eventually)
7254 * be larger than it was before. Once it becomes
7255 * substantially larger, the test below will cause
7256 * the array to appear non-idle, and resync will slow
7258 * If there is a lot of outstanding resync activity when
7259 * we set last_event to curr_events, then all that activity
7260 * completing might cause the array to appear non-idle
7261 * and resync will be slowed down even though there might
7262 * not have been non-resync activity. This will only
7263 * happen once though. 'last_events' will soon reflect
7264 * the state where there is little or no outstanding
7265 * resync requests, and further resync activity will
7266 * always make curr_events less than last_events.
7269 if (init || curr_events - rdev->last_events > 64) {
7270 rdev->last_events = curr_events;
7278 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7280 /* another "blocks" (512byte) blocks have been synced */
7281 atomic_sub(blocks, &mddev->recovery_active);
7282 wake_up(&mddev->recovery_wait);
7284 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7285 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7286 md_wakeup_thread(mddev->thread);
7287 // stop recovery, signal do_sync ....
7290 EXPORT_SYMBOL(md_done_sync);
7292 /* md_write_start(mddev, bi)
7293 * If we need to update some array metadata (e.g. 'active' flag
7294 * in superblock) before writing, schedule a superblock update
7295 * and wait for it to complete.
7297 void md_write_start(struct mddev *mddev, struct bio *bi)
7300 if (bio_data_dir(bi) != WRITE)
7303 BUG_ON(mddev->ro == 1);
7304 if (mddev->ro == 2) {
7305 /* need to switch to read/write */
7307 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7308 md_wakeup_thread(mddev->thread);
7309 md_wakeup_thread(mddev->sync_thread);
7312 atomic_inc(&mddev->writes_pending);
7313 if (mddev->safemode == 1)
7314 mddev->safemode = 0;
7315 if (mddev->in_sync) {
7316 spin_lock(&mddev->lock);
7317 if (mddev->in_sync) {
7319 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7320 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7321 md_wakeup_thread(mddev->thread);
7324 spin_unlock(&mddev->lock);
7327 sysfs_notify_dirent_safe(mddev->sysfs_state);
7328 wait_event(mddev->sb_wait,
7329 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7331 EXPORT_SYMBOL(md_write_start);
7333 void md_write_end(struct mddev *mddev)
7335 if (atomic_dec_and_test(&mddev->writes_pending)) {
7336 if (mddev->safemode == 2)
7337 md_wakeup_thread(mddev->thread);
7338 else if (mddev->safemode_delay)
7339 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7342 EXPORT_SYMBOL(md_write_end);
7344 /* md_allow_write(mddev)
7345 * Calling this ensures that the array is marked 'active' so that writes
7346 * may proceed without blocking. It is important to call this before
7347 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7348 * Must be called with mddev_lock held.
7350 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7351 * is dropped, so return -EAGAIN after notifying userspace.
7353 int md_allow_write(struct mddev *mddev)
7359 if (!mddev->pers->sync_request)
7362 spin_lock(&mddev->lock);
7363 if (mddev->in_sync) {
7365 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7366 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7367 if (mddev->safemode_delay &&
7368 mddev->safemode == 0)
7369 mddev->safemode = 1;
7370 spin_unlock(&mddev->lock);
7371 md_update_sb(mddev, 0);
7372 sysfs_notify_dirent_safe(mddev->sysfs_state);
7374 spin_unlock(&mddev->lock);
7376 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7381 EXPORT_SYMBOL_GPL(md_allow_write);
7383 #define SYNC_MARKS 10
7384 #define SYNC_MARK_STEP (3*HZ)
7385 #define UPDATE_FREQUENCY (5*60*HZ)
7386 void md_do_sync(struct md_thread *thread)
7388 struct mddev *mddev = thread->mddev;
7389 struct mddev *mddev2;
7390 unsigned int currspeed = 0,
7392 sector_t max_sectors,j, io_sectors, recovery_done;
7393 unsigned long mark[SYNC_MARKS];
7394 unsigned long update_time;
7395 sector_t mark_cnt[SYNC_MARKS];
7397 struct list_head *tmp;
7398 sector_t last_check;
7400 struct md_rdev *rdev;
7401 char *desc, *action = NULL;
7402 struct blk_plug plug;
7404 /* just incase thread restarts... */
7405 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7407 if (mddev->ro) {/* never try to sync a read-only array */
7408 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7412 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7413 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7414 desc = "data-check";
7416 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7417 desc = "requested-resync";
7421 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7426 mddev->last_sync_action = action ?: desc;
7428 /* we overload curr_resync somewhat here.
7429 * 0 == not engaged in resync at all
7430 * 2 == checking that there is no conflict with another sync
7431 * 1 == like 2, but have yielded to allow conflicting resync to
7433 * other == active in resync - this many blocks
7435 * Before starting a resync we must have set curr_resync to
7436 * 2, and then checked that every "conflicting" array has curr_resync
7437 * less than ours. When we find one that is the same or higher
7438 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7439 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7440 * This will mean we have to start checking from the beginning again.
7445 mddev->curr_resync = 2;
7448 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7450 for_each_mddev(mddev2, tmp) {
7451 if (mddev2 == mddev)
7453 if (!mddev->parallel_resync
7454 && mddev2->curr_resync
7455 && match_mddev_units(mddev, mddev2)) {
7457 if (mddev < mddev2 && mddev->curr_resync == 2) {
7458 /* arbitrarily yield */
7459 mddev->curr_resync = 1;
7460 wake_up(&resync_wait);
7462 if (mddev > mddev2 && mddev->curr_resync == 1)
7463 /* no need to wait here, we can wait the next
7464 * time 'round when curr_resync == 2
7467 /* We need to wait 'interruptible' so as not to
7468 * contribute to the load average, and not to
7469 * be caught by 'softlockup'
7471 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7472 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7473 mddev2->curr_resync >= mddev->curr_resync) {
7474 printk(KERN_INFO "md: delaying %s of %s"
7475 " until %s has finished (they"
7476 " share one or more physical units)\n",
7477 desc, mdname(mddev), mdname(mddev2));
7479 if (signal_pending(current))
7480 flush_signals(current);
7482 finish_wait(&resync_wait, &wq);
7485 finish_wait(&resync_wait, &wq);
7488 } while (mddev->curr_resync < 2);
7491 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7492 /* resync follows the size requested by the personality,
7493 * which defaults to physical size, but can be virtual size
7495 max_sectors = mddev->resync_max_sectors;
7496 atomic64_set(&mddev->resync_mismatches, 0);
7497 /* we don't use the checkpoint if there's a bitmap */
7498 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7499 j = mddev->resync_min;
7500 else if (!mddev->bitmap)
7501 j = mddev->recovery_cp;
7503 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7504 max_sectors = mddev->resync_max_sectors;
7506 /* recovery follows the physical size of devices */
7507 max_sectors = mddev->dev_sectors;
7510 rdev_for_each_rcu(rdev, mddev)
7511 if (rdev->raid_disk >= 0 &&
7512 !test_bit(Faulty, &rdev->flags) &&
7513 !test_bit(In_sync, &rdev->flags) &&
7514 rdev->recovery_offset < j)
7515 j = rdev->recovery_offset;
7518 /* If there is a bitmap, we need to make sure all
7519 * writes that started before we added a spare
7520 * complete before we start doing a recovery.
7521 * Otherwise the write might complete and (via
7522 * bitmap_endwrite) set a bit in the bitmap after the
7523 * recovery has checked that bit and skipped that
7526 if (mddev->bitmap) {
7527 mddev->pers->quiesce(mddev, 1);
7528 mddev->pers->quiesce(mddev, 0);
7532 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7533 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7534 " %d KB/sec/disk.\n", speed_min(mddev));
7535 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7536 "(but not more than %d KB/sec) for %s.\n",
7537 speed_max(mddev), desc);
7539 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7542 for (m = 0; m < SYNC_MARKS; m++) {
7544 mark_cnt[m] = io_sectors;
7547 mddev->resync_mark = mark[last_mark];
7548 mddev->resync_mark_cnt = mark_cnt[last_mark];
7551 * Tune reconstruction:
7553 window = 32*(PAGE_SIZE/512);
7554 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7555 window/2, (unsigned long long)max_sectors/2);
7557 atomic_set(&mddev->recovery_active, 0);
7562 "md: resuming %s of %s from checkpoint.\n",
7563 desc, mdname(mddev));
7564 mddev->curr_resync = j;
7566 mddev->curr_resync = 3; /* no longer delayed */
7567 mddev->curr_resync_completed = j;
7568 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7569 md_new_event(mddev);
7570 update_time = jiffies;
7572 blk_start_plug(&plug);
7573 while (j < max_sectors) {
7578 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7579 ((mddev->curr_resync > mddev->curr_resync_completed &&
7580 (mddev->curr_resync - mddev->curr_resync_completed)
7581 > (max_sectors >> 4)) ||
7582 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7583 (j - mddev->curr_resync_completed)*2
7584 >= mddev->resync_max - mddev->curr_resync_completed
7586 /* time to update curr_resync_completed */
7587 wait_event(mddev->recovery_wait,
7588 atomic_read(&mddev->recovery_active) == 0);
7589 mddev->curr_resync_completed = j;
7590 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7591 j > mddev->recovery_cp)
7592 mddev->recovery_cp = j;
7593 update_time = jiffies;
7594 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7595 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7598 while (j >= mddev->resync_max &&
7599 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7600 /* As this condition is controlled by user-space,
7601 * we can block indefinitely, so use '_interruptible'
7602 * to avoid triggering warnings.
7604 flush_signals(current); /* just in case */
7605 wait_event_interruptible(mddev->recovery_wait,
7606 mddev->resync_max > j
7607 || test_bit(MD_RECOVERY_INTR,
7611 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7614 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7615 currspeed < speed_min(mddev));
7617 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7621 if (!skipped) { /* actual IO requested */
7622 io_sectors += sectors;
7623 atomic_add(sectors, &mddev->recovery_active);
7626 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7631 mddev->curr_resync = j;
7632 mddev->curr_mark_cnt = io_sectors;
7633 if (last_check == 0)
7634 /* this is the earliest that rebuild will be
7635 * visible in /proc/mdstat
7637 md_new_event(mddev);
7639 if (last_check + window > io_sectors || j == max_sectors)
7642 last_check = io_sectors;
7644 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7646 int next = (last_mark+1) % SYNC_MARKS;
7648 mddev->resync_mark = mark[next];
7649 mddev->resync_mark_cnt = mark_cnt[next];
7650 mark[next] = jiffies;
7651 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7655 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7659 * this loop exits only if either when we are slower than
7660 * the 'hard' speed limit, or the system was IO-idle for
7662 * the system might be non-idle CPU-wise, but we only care
7663 * about not overloading the IO subsystem. (things like an
7664 * e2fsck being done on the RAID array should execute fast)
7668 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7669 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7670 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7672 if (currspeed > speed_min(mddev)) {
7673 if ((currspeed > speed_max(mddev)) ||
7674 !is_mddev_idle(mddev, 0)) {
7680 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7681 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7682 ? "interrupted" : "done");
7684 * this also signals 'finished resyncing' to md_stop
7686 blk_finish_plug(&plug);
7687 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7689 /* tell personality that we are finished */
7690 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7692 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7693 mddev->curr_resync > 2) {
7694 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7695 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7696 if (mddev->curr_resync >= mddev->recovery_cp) {
7698 "md: checkpointing %s of %s.\n",
7699 desc, mdname(mddev));
7700 if (test_bit(MD_RECOVERY_ERROR,
7702 mddev->recovery_cp =
7703 mddev->curr_resync_completed;
7705 mddev->recovery_cp =
7709 mddev->recovery_cp = MaxSector;
7711 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7712 mddev->curr_resync = MaxSector;
7714 rdev_for_each_rcu(rdev, mddev)
7715 if (rdev->raid_disk >= 0 &&
7716 mddev->delta_disks >= 0 &&
7717 !test_bit(Faulty, &rdev->flags) &&
7718 !test_bit(In_sync, &rdev->flags) &&
7719 rdev->recovery_offset < mddev->curr_resync)
7720 rdev->recovery_offset = mddev->curr_resync;
7725 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7727 spin_lock(&mddev->lock);
7728 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7729 /* We completed so min/max setting can be forgotten if used. */
7730 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7731 mddev->resync_min = 0;
7732 mddev->resync_max = MaxSector;
7733 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7734 mddev->resync_min = mddev->curr_resync_completed;
7735 mddev->curr_resync = 0;
7736 spin_unlock(&mddev->lock);
7738 wake_up(&resync_wait);
7739 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7740 md_wakeup_thread(mddev->thread);
7743 EXPORT_SYMBOL_GPL(md_do_sync);
7745 static int remove_and_add_spares(struct mddev *mddev,
7746 struct md_rdev *this)
7748 struct md_rdev *rdev;
7752 rdev_for_each(rdev, mddev)
7753 if ((this == NULL || rdev == this) &&
7754 rdev->raid_disk >= 0 &&
7755 !test_bit(Blocked, &rdev->flags) &&
7756 (test_bit(Faulty, &rdev->flags) ||
7757 ! test_bit(In_sync, &rdev->flags)) &&
7758 atomic_read(&rdev->nr_pending)==0) {
7759 if (mddev->pers->hot_remove_disk(
7760 mddev, rdev) == 0) {
7761 sysfs_unlink_rdev(mddev, rdev);
7762 rdev->raid_disk = -1;
7766 if (removed && mddev->kobj.sd)
7767 sysfs_notify(&mddev->kobj, NULL, "degraded");
7772 rdev_for_each(rdev, mddev) {
7773 if (rdev->raid_disk >= 0 &&
7774 !test_bit(In_sync, &rdev->flags) &&
7775 !test_bit(Faulty, &rdev->flags))
7777 if (rdev->raid_disk >= 0)
7779 if (test_bit(Faulty, &rdev->flags))
7782 ! (rdev->saved_raid_disk >= 0 &&
7783 !test_bit(Bitmap_sync, &rdev->flags)))
7786 if (rdev->saved_raid_disk < 0)
7787 rdev->recovery_offset = 0;
7789 hot_add_disk(mddev, rdev) == 0) {
7790 if (sysfs_link_rdev(mddev, rdev))
7791 /* failure here is OK */;
7793 md_new_event(mddev);
7794 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7799 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7803 static void md_start_sync(struct work_struct *ws)
7805 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7807 mddev->sync_thread = md_register_thread(md_do_sync,
7810 if (!mddev->sync_thread) {
7811 printk(KERN_ERR "%s: could not start resync"
7814 /* leave the spares where they are, it shouldn't hurt */
7815 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7816 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7817 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7818 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7819 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7820 wake_up(&resync_wait);
7821 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7823 if (mddev->sysfs_action)
7824 sysfs_notify_dirent_safe(mddev->sysfs_action);
7826 md_wakeup_thread(mddev->sync_thread);
7827 sysfs_notify_dirent_safe(mddev->sysfs_action);
7828 md_new_event(mddev);
7832 * This routine is regularly called by all per-raid-array threads to
7833 * deal with generic issues like resync and super-block update.
7834 * Raid personalities that don't have a thread (linear/raid0) do not
7835 * need this as they never do any recovery or update the superblock.
7837 * It does not do any resync itself, but rather "forks" off other threads
7838 * to do that as needed.
7839 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7840 * "->recovery" and create a thread at ->sync_thread.
7841 * When the thread finishes it sets MD_RECOVERY_DONE
7842 * and wakeups up this thread which will reap the thread and finish up.
7843 * This thread also removes any faulty devices (with nr_pending == 0).
7845 * The overall approach is:
7846 * 1/ if the superblock needs updating, update it.
7847 * 2/ If a recovery thread is running, don't do anything else.
7848 * 3/ If recovery has finished, clean up, possibly marking spares active.
7849 * 4/ If there are any faulty devices, remove them.
7850 * 5/ If array is degraded, try to add spares devices
7851 * 6/ If array has spares or is not in-sync, start a resync thread.
7853 void md_check_recovery(struct mddev *mddev)
7855 if (mddev->suspended)
7859 bitmap_daemon_work(mddev);
7861 if (signal_pending(current)) {
7862 if (mddev->pers->sync_request && !mddev->external) {
7863 printk(KERN_INFO "md: %s in immediate safe mode\n",
7865 mddev->safemode = 2;
7867 flush_signals(current);
7870 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7873 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
7874 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7875 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7876 (mddev->external == 0 && mddev->safemode == 1) ||
7877 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7878 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7882 if (mddev_trylock(mddev)) {
7886 /* On a read-only array we can:
7887 * - remove failed devices
7888 * - add already-in_sync devices if the array itself
7890 * As we only add devices that are already in-sync,
7891 * we can activate the spares immediately.
7893 remove_and_add_spares(mddev, NULL);
7894 /* There is no thread, but we need to call
7895 * ->spare_active and clear saved_raid_disk
7897 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7898 md_reap_sync_thread(mddev);
7899 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7903 if (!mddev->external) {
7905 spin_lock(&mddev->lock);
7906 if (mddev->safemode &&
7907 !atomic_read(&mddev->writes_pending) &&
7909 mddev->recovery_cp == MaxSector) {
7912 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7914 if (mddev->safemode == 1)
7915 mddev->safemode = 0;
7916 spin_unlock(&mddev->lock);
7918 sysfs_notify_dirent_safe(mddev->sysfs_state);
7921 if (mddev->flags & MD_UPDATE_SB_FLAGS)
7922 md_update_sb(mddev, 0);
7924 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7925 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7926 /* resync/recovery still happening */
7927 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7930 if (mddev->sync_thread) {
7931 md_reap_sync_thread(mddev);
7934 /* Set RUNNING before clearing NEEDED to avoid
7935 * any transients in the value of "sync_action".
7937 mddev->curr_resync_completed = 0;
7938 spin_lock(&mddev->lock);
7939 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7940 spin_unlock(&mddev->lock);
7941 /* Clear some bits that don't mean anything, but
7944 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7945 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7947 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7948 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7950 /* no recovery is running.
7951 * remove any failed drives, then
7952 * add spares if possible.
7953 * Spares are also removed and re-added, to allow
7954 * the personality to fail the re-add.
7957 if (mddev->reshape_position != MaxSector) {
7958 if (mddev->pers->check_reshape == NULL ||
7959 mddev->pers->check_reshape(mddev) != 0)
7960 /* Cannot proceed */
7962 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7963 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7964 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7965 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7966 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7967 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7968 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7969 } else if (mddev->recovery_cp < MaxSector) {
7970 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7971 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7972 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7973 /* nothing to be done ... */
7976 if (mddev->pers->sync_request) {
7978 /* We are adding a device or devices to an array
7979 * which has the bitmap stored on all devices.
7980 * So make sure all bitmap pages get written
7982 bitmap_write_all(mddev->bitmap);
7984 INIT_WORK(&mddev->del_work, md_start_sync);
7985 queue_work(md_misc_wq, &mddev->del_work);
7989 if (!mddev->sync_thread) {
7990 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7991 wake_up(&resync_wait);
7992 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7994 if (mddev->sysfs_action)
7995 sysfs_notify_dirent_safe(mddev->sysfs_action);
7998 wake_up(&mddev->sb_wait);
7999 mddev_unlock(mddev);
8002 EXPORT_SYMBOL(md_check_recovery);
8004 void md_reap_sync_thread(struct mddev *mddev)
8006 struct md_rdev *rdev;
8008 /* resync has finished, collect result */
8009 md_unregister_thread(&mddev->sync_thread);
8010 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8011 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8013 /* activate any spares */
8014 if (mddev->pers->spare_active(mddev)) {
8015 sysfs_notify(&mddev->kobj, NULL,
8017 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8020 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8021 mddev->pers->finish_reshape)
8022 mddev->pers->finish_reshape(mddev);
8024 /* If array is no-longer degraded, then any saved_raid_disk
8025 * information must be scrapped.
8027 if (!mddev->degraded)
8028 rdev_for_each(rdev, mddev)
8029 rdev->saved_raid_disk = -1;
8031 md_update_sb(mddev, 1);
8032 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8033 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8034 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8035 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8036 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8037 wake_up(&resync_wait);
8038 /* flag recovery needed just to double check */
8039 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8040 sysfs_notify_dirent_safe(mddev->sysfs_action);
8041 md_new_event(mddev);
8042 if (mddev->event_work.func)
8043 queue_work(md_misc_wq, &mddev->event_work);
8045 EXPORT_SYMBOL(md_reap_sync_thread);
8047 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8049 sysfs_notify_dirent_safe(rdev->sysfs_state);
8050 wait_event_timeout(rdev->blocked_wait,
8051 !test_bit(Blocked, &rdev->flags) &&
8052 !test_bit(BlockedBadBlocks, &rdev->flags),
8053 msecs_to_jiffies(5000));
8054 rdev_dec_pending(rdev, mddev);
8056 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8058 void md_finish_reshape(struct mddev *mddev)
8060 /* called be personality module when reshape completes. */
8061 struct md_rdev *rdev;
8063 rdev_for_each(rdev, mddev) {
8064 if (rdev->data_offset > rdev->new_data_offset)
8065 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8067 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8068 rdev->data_offset = rdev->new_data_offset;
8071 EXPORT_SYMBOL(md_finish_reshape);
8073 /* Bad block management.
8074 * We can record which blocks on each device are 'bad' and so just
8075 * fail those blocks, or that stripe, rather than the whole device.
8076 * Entries in the bad-block table are 64bits wide. This comprises:
8077 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8078 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8079 * A 'shift' can be set so that larger blocks are tracked and
8080 * consequently larger devices can be covered.
8081 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8083 * Locking of the bad-block table uses a seqlock so md_is_badblock
8084 * might need to retry if it is very unlucky.
8085 * We will sometimes want to check for bad blocks in a bi_end_io function,
8086 * so we use the write_seqlock_irq variant.
8088 * When looking for a bad block we specify a range and want to
8089 * know if any block in the range is bad. So we binary-search
8090 * to the last range that starts at-or-before the given endpoint,
8091 * (or "before the sector after the target range")
8092 * then see if it ends after the given start.
8094 * 0 if there are no known bad blocks in the range
8095 * 1 if there are known bad block which are all acknowledged
8096 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8097 * plus the start/length of the first bad section we overlap.
8099 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8100 sector_t *first_bad, int *bad_sectors)
8106 sector_t target = s + sectors;
8109 if (bb->shift > 0) {
8110 /* round the start down, and the end up */
8112 target += (1<<bb->shift) - 1;
8113 target >>= bb->shift;
8114 sectors = target - s;
8116 /* 'target' is now the first block after the bad range */
8119 seq = read_seqbegin(&bb->lock);
8124 /* Binary search between lo and hi for 'target'
8125 * i.e. for the last range that starts before 'target'
8127 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8128 * are known not to be the last range before target.
8129 * VARIANT: hi-lo is the number of possible
8130 * ranges, and decreases until it reaches 1
8132 while (hi - lo > 1) {
8133 int mid = (lo + hi) / 2;
8134 sector_t a = BB_OFFSET(p[mid]);
8136 /* This could still be the one, earlier ranges
8140 /* This and later ranges are definitely out. */
8143 /* 'lo' might be the last that started before target, but 'hi' isn't */
8145 /* need to check all range that end after 's' to see if
8146 * any are unacknowledged.
8149 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8150 if (BB_OFFSET(p[lo]) < target) {
8151 /* starts before the end, and finishes after
8152 * the start, so they must overlap
8154 if (rv != -1 && BB_ACK(p[lo]))
8158 *first_bad = BB_OFFSET(p[lo]);
8159 *bad_sectors = BB_LEN(p[lo]);
8165 if (read_seqretry(&bb->lock, seq))
8170 EXPORT_SYMBOL_GPL(md_is_badblock);
8173 * Add a range of bad blocks to the table.
8174 * This might extend the table, or might contract it
8175 * if two adjacent ranges can be merged.
8176 * We binary-search to find the 'insertion' point, then
8177 * decide how best to handle it.
8179 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8185 unsigned long flags;
8188 /* badblocks are disabled */
8192 /* round the start down, and the end up */
8193 sector_t next = s + sectors;
8195 next += (1<<bb->shift) - 1;
8200 write_seqlock_irqsave(&bb->lock, flags);
8205 /* Find the last range that starts at-or-before 's' */
8206 while (hi - lo > 1) {
8207 int mid = (lo + hi) / 2;
8208 sector_t a = BB_OFFSET(p[mid]);
8214 if (hi > lo && BB_OFFSET(p[lo]) > s)
8218 /* we found a range that might merge with the start
8221 sector_t a = BB_OFFSET(p[lo]);
8222 sector_t e = a + BB_LEN(p[lo]);
8223 int ack = BB_ACK(p[lo]);
8225 /* Yes, we can merge with a previous range */
8226 if (s == a && s + sectors >= e)
8227 /* new range covers old */
8230 ack = ack && acknowledged;
8232 if (e < s + sectors)
8234 if (e - a <= BB_MAX_LEN) {
8235 p[lo] = BB_MAKE(a, e-a, ack);
8238 /* does not all fit in one range,
8239 * make p[lo] maximal
8241 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8242 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8248 if (sectors && hi < bb->count) {
8249 /* 'hi' points to the first range that starts after 's'.
8250 * Maybe we can merge with the start of that range */
8251 sector_t a = BB_OFFSET(p[hi]);
8252 sector_t e = a + BB_LEN(p[hi]);
8253 int ack = BB_ACK(p[hi]);
8254 if (a <= s + sectors) {
8255 /* merging is possible */
8256 if (e <= s + sectors) {
8261 ack = ack && acknowledged;
8264 if (e - a <= BB_MAX_LEN) {
8265 p[hi] = BB_MAKE(a, e-a, ack);
8268 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8276 if (sectors == 0 && hi < bb->count) {
8277 /* we might be able to combine lo and hi */
8278 /* Note: 's' is at the end of 'lo' */
8279 sector_t a = BB_OFFSET(p[hi]);
8280 int lolen = BB_LEN(p[lo]);
8281 int hilen = BB_LEN(p[hi]);
8282 int newlen = lolen + hilen - (s - a);
8283 if (s >= a && newlen < BB_MAX_LEN) {
8284 /* yes, we can combine them */
8285 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8286 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8287 memmove(p + hi, p + hi + 1,
8288 (bb->count - hi - 1) * 8);
8293 /* didn't merge (it all).
8294 * Need to add a range just before 'hi' */
8295 if (bb->count >= MD_MAX_BADBLOCKS) {
8296 /* No room for more */
8300 int this_sectors = sectors;
8301 memmove(p + hi + 1, p + hi,
8302 (bb->count - hi) * 8);
8305 if (this_sectors > BB_MAX_LEN)
8306 this_sectors = BB_MAX_LEN;
8307 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8308 sectors -= this_sectors;
8315 bb->unacked_exist = 1;
8316 write_sequnlock_irqrestore(&bb->lock, flags);
8321 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8326 s += rdev->new_data_offset;
8328 s += rdev->data_offset;
8329 rv = md_set_badblocks(&rdev->badblocks,
8332 /* Make sure they get written out promptly */
8333 sysfs_notify_dirent_safe(rdev->sysfs_state);
8334 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8335 md_wakeup_thread(rdev->mddev->thread);
8339 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8342 * Remove a range of bad blocks from the table.
8343 * This may involve extending the table if we spilt a region,
8344 * but it must not fail. So if the table becomes full, we just
8345 * drop the remove request.
8347 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8351 sector_t target = s + sectors;
8354 if (bb->shift > 0) {
8355 /* When clearing we round the start up and the end down.
8356 * This should not matter as the shift should align with
8357 * the block size and no rounding should ever be needed.
8358 * However it is better the think a block is bad when it
8359 * isn't than to think a block is not bad when it is.
8361 s += (1<<bb->shift) - 1;
8363 target >>= bb->shift;
8364 sectors = target - s;
8367 write_seqlock_irq(&bb->lock);
8372 /* Find the last range that starts before 'target' */
8373 while (hi - lo > 1) {
8374 int mid = (lo + hi) / 2;
8375 sector_t a = BB_OFFSET(p[mid]);
8382 /* p[lo] is the last range that could overlap the
8383 * current range. Earlier ranges could also overlap,
8384 * but only this one can overlap the end of the range.
8386 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8387 /* Partial overlap, leave the tail of this range */
8388 int ack = BB_ACK(p[lo]);
8389 sector_t a = BB_OFFSET(p[lo]);
8390 sector_t end = a + BB_LEN(p[lo]);
8393 /* we need to split this range */
8394 if (bb->count >= MD_MAX_BADBLOCKS) {
8398 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8400 p[lo] = BB_MAKE(a, s-a, ack);
8403 p[lo] = BB_MAKE(target, end - target, ack);
8404 /* there is no longer an overlap */
8409 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8410 /* This range does overlap */
8411 if (BB_OFFSET(p[lo]) < s) {
8412 /* Keep the early parts of this range. */
8413 int ack = BB_ACK(p[lo]);
8414 sector_t start = BB_OFFSET(p[lo]);
8415 p[lo] = BB_MAKE(start, s - start, ack);
8416 /* now low doesn't overlap, so.. */
8421 /* 'lo' is strictly before, 'hi' is strictly after,
8422 * anything between needs to be discarded
8425 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8426 bb->count -= (hi - lo - 1);
8432 write_sequnlock_irq(&bb->lock);
8436 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8440 s += rdev->new_data_offset;
8442 s += rdev->data_offset;
8443 return md_clear_badblocks(&rdev->badblocks,
8446 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8449 * Acknowledge all bad blocks in a list.
8450 * This only succeeds if ->changed is clear. It is used by
8451 * in-kernel metadata updates
8453 void md_ack_all_badblocks(struct badblocks *bb)
8455 if (bb->page == NULL || bb->changed)
8456 /* no point even trying */
8458 write_seqlock_irq(&bb->lock);
8460 if (bb->changed == 0 && bb->unacked_exist) {
8463 for (i = 0; i < bb->count ; i++) {
8464 if (!BB_ACK(p[i])) {
8465 sector_t start = BB_OFFSET(p[i]);
8466 int len = BB_LEN(p[i]);
8467 p[i] = BB_MAKE(start, len, 1);
8470 bb->unacked_exist = 0;
8472 write_sequnlock_irq(&bb->lock);
8474 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8476 /* sysfs access to bad-blocks list.
8477 * We present two files.
8478 * 'bad-blocks' lists sector numbers and lengths of ranges that
8479 * are recorded as bad. The list is truncated to fit within
8480 * the one-page limit of sysfs.
8481 * Writing "sector length" to this file adds an acknowledged
8483 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8484 * been acknowledged. Writing to this file adds bad blocks
8485 * without acknowledging them. This is largely for testing.
8489 badblocks_show(struct badblocks *bb, char *page, int unack)
8500 seq = read_seqbegin(&bb->lock);
8505 while (len < PAGE_SIZE && i < bb->count) {
8506 sector_t s = BB_OFFSET(p[i]);
8507 unsigned int length = BB_LEN(p[i]);
8508 int ack = BB_ACK(p[i]);
8514 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8515 (unsigned long long)s << bb->shift,
8516 length << bb->shift);
8518 if (unack && len == 0)
8519 bb->unacked_exist = 0;
8521 if (read_seqretry(&bb->lock, seq))
8530 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8532 unsigned long long sector;
8536 /* Allow clearing via sysfs *only* for testing/debugging.
8537 * Normally only a successful write may clear a badblock
8540 if (page[0] == '-') {
8544 #endif /* DO_DEBUG */
8546 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8548 if (newline != '\n')
8560 md_clear_badblocks(bb, sector, length);
8563 #endif /* DO_DEBUG */
8564 if (md_set_badblocks(bb, sector, length, !unack))
8570 static int md_notify_reboot(struct notifier_block *this,
8571 unsigned long code, void *x)
8573 struct list_head *tmp;
8574 struct mddev *mddev;
8577 for_each_mddev(mddev, tmp) {
8578 if (mddev_trylock(mddev)) {
8580 __md_stop_writes(mddev);
8581 if (mddev->persistent)
8582 mddev->safemode = 2;
8583 mddev_unlock(mddev);
8588 * certain more exotic SCSI devices are known to be
8589 * volatile wrt too early system reboots. While the
8590 * right place to handle this issue is the given
8591 * driver, we do want to have a safe RAID driver ...
8599 static struct notifier_block md_notifier = {
8600 .notifier_call = md_notify_reboot,
8602 .priority = INT_MAX, /* before any real devices */
8605 static void md_geninit(void)
8607 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8609 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8612 static int __init md_init(void)
8616 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8620 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8624 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8627 if ((ret = register_blkdev(0, "mdp")) < 0)
8631 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8632 md_probe, NULL, NULL);
8633 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8634 md_probe, NULL, NULL);
8636 register_reboot_notifier(&md_notifier);
8637 raid_table_header = register_sysctl_table(raid_root_table);
8643 unregister_blkdev(MD_MAJOR, "md");
8645 destroy_workqueue(md_misc_wq);
8647 destroy_workqueue(md_wq);
8655 * Searches all registered partitions for autorun RAID arrays
8659 static LIST_HEAD(all_detected_devices);
8660 struct detected_devices_node {
8661 struct list_head list;
8665 void md_autodetect_dev(dev_t dev)
8667 struct detected_devices_node *node_detected_dev;
8669 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8670 if (node_detected_dev) {
8671 node_detected_dev->dev = dev;
8672 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8674 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8675 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8679 static void autostart_arrays(int part)
8681 struct md_rdev *rdev;
8682 struct detected_devices_node *node_detected_dev;
8684 int i_scanned, i_passed;
8689 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8691 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8693 node_detected_dev = list_entry(all_detected_devices.next,
8694 struct detected_devices_node, list);
8695 list_del(&node_detected_dev->list);
8696 dev = node_detected_dev->dev;
8697 kfree(node_detected_dev);
8698 rdev = md_import_device(dev,0, 90);
8702 if (test_bit(Faulty, &rdev->flags))
8705 set_bit(AutoDetected, &rdev->flags);
8706 list_add(&rdev->same_set, &pending_raid_disks);
8710 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8711 i_scanned, i_passed);
8713 autorun_devices(part);
8716 #endif /* !MODULE */
8718 static __exit void md_exit(void)
8720 struct mddev *mddev;
8721 struct list_head *tmp;
8724 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8725 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8727 unregister_blkdev(MD_MAJOR,"md");
8728 unregister_blkdev(mdp_major, "mdp");
8729 unregister_reboot_notifier(&md_notifier);
8730 unregister_sysctl_table(raid_table_header);
8732 /* We cannot unload the modules while some process is
8733 * waiting for us in select() or poll() - wake them up
8736 while (waitqueue_active(&md_event_waiters)) {
8737 /* not safe to leave yet */
8738 wake_up(&md_event_waiters);
8742 remove_proc_entry("mdstat", NULL);
8744 for_each_mddev(mddev, tmp) {
8745 export_array(mddev);
8746 mddev->hold_active = 0;
8748 destroy_workqueue(md_misc_wq);
8749 destroy_workqueue(md_wq);
8752 subsys_initcall(md_init);
8753 module_exit(md_exit)
8755 static int get_ro(char *buffer, struct kernel_param *kp)
8757 return sprintf(buffer, "%d", start_readonly);
8759 static int set_ro(const char *val, struct kernel_param *kp)
8762 int num = simple_strtoul(val, &e, 10);
8763 if (*val && (*e == '\0' || *e == '\n')) {
8764 start_readonly = num;
8770 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8771 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8772 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8774 MODULE_LICENSE("GPL");
8775 MODULE_DESCRIPTION("MD RAID framework");
8777 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);