2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
58 static void autostart_arrays(int part);
61 /* pers_list is a list of registered personalities protected
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
69 static void md_print_devices(void);
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
75 static int remove_and_add_spares(struct mddev *mddev,
76 struct md_rdev *this);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Default number of read corrections we'll attempt on an rdev
82 * before ejecting it from the array. We divide the read error
83 * count by 2 for every hour elapsed between read errors.
85 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
88 * is 1000 KB/sec, so the extra system load does not show up that much.
89 * Increase it if you want to have more _guaranteed_ speed. Note that
90 * the RAID driver will use the maximum available bandwidth if the IO
91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
92 * speed limit - in case reconstruction slows down your system despite
95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
96 * or /sys/block/mdX/md/sync_speed_{min,max}
99 static int sysctl_speed_limit_min = 1000;
100 static int sysctl_speed_limit_max = 200000;
101 static inline int speed_min(struct mddev *mddev)
103 return mddev->sync_speed_min ?
104 mddev->sync_speed_min : sysctl_speed_limit_min;
107 static inline int speed_max(struct mddev *mddev)
109 return mddev->sync_speed_max ?
110 mddev->sync_speed_max : sysctl_speed_limit_max;
113 static struct ctl_table_header *raid_table_header;
115 static ctl_table raid_table[] = {
117 .procname = "speed_limit_min",
118 .data = &sysctl_speed_limit_min,
119 .maxlen = sizeof(int),
120 .mode = S_IRUGO|S_IWUSR,
121 .proc_handler = proc_dointvec,
124 .procname = "speed_limit_max",
125 .data = &sysctl_speed_limit_max,
126 .maxlen = sizeof(int),
127 .mode = S_IRUGO|S_IWUSR,
128 .proc_handler = proc_dointvec,
133 static ctl_table raid_dir_table[] = {
137 .mode = S_IRUGO|S_IXUGO,
143 static ctl_table raid_root_table[] = {
148 .child = raid_dir_table,
153 static const struct block_device_operations md_fops;
155 static int start_readonly;
158 * like bio_clone, but with a local bio set
161 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
166 if (!mddev || !mddev->bio_set)
167 return bio_alloc(gfp_mask, nr_iovecs);
169 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
174 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
176 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
179 if (!mddev || !mddev->bio_set)
180 return bio_clone(bio, gfp_mask);
182 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
184 EXPORT_SYMBOL_GPL(bio_clone_mddev);
186 void md_trim_bio(struct bio *bio, int offset, int size)
188 /* 'bio' is a cloned bio which we need to trim to match
189 * the given offset and size.
190 * This requires adjusting bi_sector, bi_size, and bi_io_vec
193 struct bio_vec *bvec;
197 if (offset == 0 && size == bio->bi_size)
200 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
202 bio_advance(bio, offset << 9);
206 /* avoid any complications with bi_idx being non-zero*/
208 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
209 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
210 bio->bi_vcnt -= bio->bi_idx;
213 /* Make sure vcnt and last bv are not too big */
214 bio_for_each_segment(bvec, bio, i) {
215 if (sofar + bvec->bv_len > size)
216 bvec->bv_len = size - sofar;
217 if (bvec->bv_len == 0) {
221 sofar += bvec->bv_len;
224 EXPORT_SYMBOL_GPL(md_trim_bio);
227 * We have a system wide 'event count' that is incremented
228 * on any 'interesting' event, and readers of /proc/mdstat
229 * can use 'poll' or 'select' to find out when the event
233 * start array, stop array, error, add device, remove device,
234 * start build, activate spare
236 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
237 static atomic_t md_event_count;
238 void md_new_event(struct mddev *mddev)
240 atomic_inc(&md_event_count);
241 wake_up(&md_event_waiters);
243 EXPORT_SYMBOL_GPL(md_new_event);
245 /* Alternate version that can be called from interrupts
246 * when calling sysfs_notify isn't needed.
248 static void md_new_event_inintr(struct mddev *mddev)
250 atomic_inc(&md_event_count);
251 wake_up(&md_event_waiters);
255 * Enables to iterate over all existing md arrays
256 * all_mddevs_lock protects this list.
258 static LIST_HEAD(all_mddevs);
259 static DEFINE_SPINLOCK(all_mddevs_lock);
263 * iterates through all used mddevs in the system.
264 * We take care to grab the all_mddevs_lock whenever navigating
265 * the list, and to always hold a refcount when unlocked.
266 * Any code which breaks out of this loop while own
267 * a reference to the current mddev and must mddev_put it.
269 #define for_each_mddev(_mddev,_tmp) \
271 for (({ spin_lock(&all_mddevs_lock); \
272 _tmp = all_mddevs.next; \
274 ({ if (_tmp != &all_mddevs) \
275 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
276 spin_unlock(&all_mddevs_lock); \
277 if (_mddev) mddev_put(_mddev); \
278 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
279 _tmp != &all_mddevs;}); \
280 ({ spin_lock(&all_mddevs_lock); \
281 _tmp = _tmp->next;}) \
285 /* Rather than calling directly into the personality make_request function,
286 * IO requests come here first so that we can check if the device is
287 * being suspended pending a reconfiguration.
288 * We hold a refcount over the call to ->make_request. By the time that
289 * call has finished, the bio has been linked into some internal structure
290 * and so is visible to ->quiesce(), so we don't need the refcount any more.
292 static void md_make_request(struct request_queue *q, struct bio *bio)
294 const int rw = bio_data_dir(bio);
295 struct mddev *mddev = q->queuedata;
297 unsigned int sectors;
299 if (mddev == NULL || mddev->pers == NULL
304 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
305 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
308 smp_rmb(); /* Ensure implications of 'active' are visible */
310 if (mddev->suspended) {
313 prepare_to_wait(&mddev->sb_wait, &__wait,
314 TASK_UNINTERRUPTIBLE);
315 if (!mddev->suspended)
321 finish_wait(&mddev->sb_wait, &__wait);
323 atomic_inc(&mddev->active_io);
327 * save the sectors now since our bio can
328 * go away inside make_request
330 sectors = bio_sectors(bio);
331 mddev->pers->make_request(mddev, bio);
333 cpu = part_stat_lock();
334 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
335 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
338 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
339 wake_up(&mddev->sb_wait);
342 /* mddev_suspend makes sure no new requests are submitted
343 * to the device, and that any requests that have been submitted
344 * are completely handled.
345 * Once ->stop is called and completes, the module will be completely
348 void mddev_suspend(struct mddev *mddev)
350 BUG_ON(mddev->suspended);
351 mddev->suspended = 1;
353 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
354 mddev->pers->quiesce(mddev, 1);
356 del_timer_sync(&mddev->safemode_timer);
358 EXPORT_SYMBOL_GPL(mddev_suspend);
360 void mddev_resume(struct mddev *mddev)
362 mddev->suspended = 0;
363 wake_up(&mddev->sb_wait);
364 mddev->pers->quiesce(mddev, 0);
366 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
367 md_wakeup_thread(mddev->thread);
368 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
370 EXPORT_SYMBOL_GPL(mddev_resume);
372 int mddev_congested(struct mddev *mddev, int bits)
374 return mddev->suspended;
376 EXPORT_SYMBOL(mddev_congested);
379 * Generic flush handling for md
382 static void md_end_flush(struct bio *bio, int err)
384 struct md_rdev *rdev = bio->bi_private;
385 struct mddev *mddev = rdev->mddev;
387 rdev_dec_pending(rdev, mddev);
389 if (atomic_dec_and_test(&mddev->flush_pending)) {
390 /* The pre-request flush has finished */
391 queue_work(md_wq, &mddev->flush_work);
396 static void md_submit_flush_data(struct work_struct *ws);
398 static void submit_flushes(struct work_struct *ws)
400 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
401 struct md_rdev *rdev;
403 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
404 atomic_set(&mddev->flush_pending, 1);
406 rdev_for_each_rcu(rdev, mddev)
407 if (rdev->raid_disk >= 0 &&
408 !test_bit(Faulty, &rdev->flags)) {
409 /* Take two references, one is dropped
410 * when request finishes, one after
411 * we reclaim rcu_read_lock
414 atomic_inc(&rdev->nr_pending);
415 atomic_inc(&rdev->nr_pending);
417 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
418 bi->bi_end_io = md_end_flush;
419 bi->bi_private = rdev;
420 bi->bi_bdev = rdev->bdev;
421 atomic_inc(&mddev->flush_pending);
422 submit_bio(WRITE_FLUSH, bi);
424 rdev_dec_pending(rdev, mddev);
427 if (atomic_dec_and_test(&mddev->flush_pending))
428 queue_work(md_wq, &mddev->flush_work);
431 static void md_submit_flush_data(struct work_struct *ws)
433 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
434 struct bio *bio = mddev->flush_bio;
436 if (bio->bi_size == 0)
437 /* an empty barrier - all done */
440 bio->bi_rw &= ~REQ_FLUSH;
441 mddev->pers->make_request(mddev, bio);
444 mddev->flush_bio = NULL;
445 wake_up(&mddev->sb_wait);
448 void md_flush_request(struct mddev *mddev, struct bio *bio)
450 spin_lock_irq(&mddev->write_lock);
451 wait_event_lock_irq(mddev->sb_wait,
454 mddev->flush_bio = bio;
455 spin_unlock_irq(&mddev->write_lock);
457 INIT_WORK(&mddev->flush_work, submit_flushes);
458 queue_work(md_wq, &mddev->flush_work);
460 EXPORT_SYMBOL(md_flush_request);
462 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
464 struct mddev *mddev = cb->data;
465 md_wakeup_thread(mddev->thread);
468 EXPORT_SYMBOL(md_unplug);
470 static inline struct mddev *mddev_get(struct mddev *mddev)
472 atomic_inc(&mddev->active);
476 static void mddev_delayed_delete(struct work_struct *ws);
478 static void mddev_put(struct mddev *mddev)
480 struct bio_set *bs = NULL;
482 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
484 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
485 mddev->ctime == 0 && !mddev->hold_active) {
486 /* Array is not configured at all, and not held active,
488 list_del_init(&mddev->all_mddevs);
490 mddev->bio_set = NULL;
491 if (mddev->gendisk) {
492 /* We did a probe so need to clean up. Call
493 * queue_work inside the spinlock so that
494 * flush_workqueue() after mddev_find will
495 * succeed in waiting for the work to be done.
497 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
498 queue_work(md_misc_wq, &mddev->del_work);
502 spin_unlock(&all_mddevs_lock);
507 void mddev_init(struct mddev *mddev)
509 mutex_init(&mddev->open_mutex);
510 mutex_init(&mddev->reconfig_mutex);
511 mutex_init(&mddev->bitmap_info.mutex);
512 INIT_LIST_HEAD(&mddev->disks);
513 INIT_LIST_HEAD(&mddev->all_mddevs);
514 init_timer(&mddev->safemode_timer);
515 atomic_set(&mddev->active, 1);
516 atomic_set(&mddev->openers, 0);
517 atomic_set(&mddev->active_io, 0);
518 spin_lock_init(&mddev->write_lock);
519 atomic_set(&mddev->flush_pending, 0);
520 init_waitqueue_head(&mddev->sb_wait);
521 init_waitqueue_head(&mddev->recovery_wait);
522 mddev->reshape_position = MaxSector;
523 mddev->reshape_backwards = 0;
524 mddev->resync_min = 0;
525 mddev->resync_max = MaxSector;
526 mddev->level = LEVEL_NONE;
528 EXPORT_SYMBOL_GPL(mddev_init);
530 static struct mddev * mddev_find(dev_t unit)
532 struct mddev *mddev, *new = NULL;
534 if (unit && MAJOR(unit) != MD_MAJOR)
535 unit &= ~((1<<MdpMinorShift)-1);
538 spin_lock(&all_mddevs_lock);
541 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
542 if (mddev->unit == unit) {
544 spin_unlock(&all_mddevs_lock);
550 list_add(&new->all_mddevs, &all_mddevs);
551 spin_unlock(&all_mddevs_lock);
552 new->hold_active = UNTIL_IOCTL;
556 /* find an unused unit number */
557 static int next_minor = 512;
558 int start = next_minor;
562 dev = MKDEV(MD_MAJOR, next_minor);
564 if (next_minor > MINORMASK)
566 if (next_minor == start) {
567 /* Oh dear, all in use. */
568 spin_unlock(&all_mddevs_lock);
574 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
575 if (mddev->unit == dev) {
581 new->md_minor = MINOR(dev);
582 new->hold_active = UNTIL_STOP;
583 list_add(&new->all_mddevs, &all_mddevs);
584 spin_unlock(&all_mddevs_lock);
587 spin_unlock(&all_mddevs_lock);
589 new = kzalloc(sizeof(*new), GFP_KERNEL);
594 if (MAJOR(unit) == MD_MAJOR)
595 new->md_minor = MINOR(unit);
597 new->md_minor = MINOR(unit) >> MdpMinorShift;
604 static inline int mddev_lock(struct mddev * mddev)
606 return mutex_lock_interruptible(&mddev->reconfig_mutex);
609 static inline int mddev_is_locked(struct mddev *mddev)
611 return mutex_is_locked(&mddev->reconfig_mutex);
614 static inline int mddev_trylock(struct mddev * mddev)
616 return mutex_trylock(&mddev->reconfig_mutex);
619 static struct attribute_group md_redundancy_group;
621 static void mddev_unlock(struct mddev * mddev)
623 if (mddev->to_remove) {
624 /* These cannot be removed under reconfig_mutex as
625 * an access to the files will try to take reconfig_mutex
626 * while holding the file unremovable, which leads to
628 * So hold set sysfs_active while the remove in happeing,
629 * and anything else which might set ->to_remove or my
630 * otherwise change the sysfs namespace will fail with
631 * -EBUSY if sysfs_active is still set.
632 * We set sysfs_active under reconfig_mutex and elsewhere
633 * test it under the same mutex to ensure its correct value
636 struct attribute_group *to_remove = mddev->to_remove;
637 mddev->to_remove = NULL;
638 mddev->sysfs_active = 1;
639 mutex_unlock(&mddev->reconfig_mutex);
641 if (mddev->kobj.sd) {
642 if (to_remove != &md_redundancy_group)
643 sysfs_remove_group(&mddev->kobj, to_remove);
644 if (mddev->pers == NULL ||
645 mddev->pers->sync_request == NULL) {
646 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
647 if (mddev->sysfs_action)
648 sysfs_put(mddev->sysfs_action);
649 mddev->sysfs_action = NULL;
652 mddev->sysfs_active = 0;
654 mutex_unlock(&mddev->reconfig_mutex);
656 /* As we've dropped the mutex we need a spinlock to
657 * make sure the thread doesn't disappear
659 spin_lock(&pers_lock);
660 md_wakeup_thread(mddev->thread);
661 spin_unlock(&pers_lock);
664 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
666 struct md_rdev *rdev;
668 rdev_for_each(rdev, mddev)
669 if (rdev->desc_nr == nr)
675 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
677 struct md_rdev *rdev;
679 rdev_for_each_rcu(rdev, mddev)
680 if (rdev->desc_nr == nr)
686 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
688 struct md_rdev *rdev;
690 rdev_for_each(rdev, mddev)
691 if (rdev->bdev->bd_dev == dev)
697 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
699 struct md_rdev *rdev;
701 rdev_for_each_rcu(rdev, mddev)
702 if (rdev->bdev->bd_dev == dev)
708 static struct md_personality *find_pers(int level, char *clevel)
710 struct md_personality *pers;
711 list_for_each_entry(pers, &pers_list, list) {
712 if (level != LEVEL_NONE && pers->level == level)
714 if (strcmp(pers->name, clevel)==0)
720 /* return the offset of the super block in 512byte sectors */
721 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
723 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
724 return MD_NEW_SIZE_SECTORS(num_sectors);
727 static int alloc_disk_sb(struct md_rdev * rdev)
732 rdev->sb_page = alloc_page(GFP_KERNEL);
733 if (!rdev->sb_page) {
734 printk(KERN_ALERT "md: out of memory.\n");
741 void md_rdev_clear(struct md_rdev *rdev)
744 put_page(rdev->sb_page);
746 rdev->sb_page = NULL;
751 put_page(rdev->bb_page);
752 rdev->bb_page = NULL;
754 kfree(rdev->badblocks.page);
755 rdev->badblocks.page = NULL;
757 EXPORT_SYMBOL_GPL(md_rdev_clear);
759 static void super_written(struct bio *bio, int error)
761 struct md_rdev *rdev = bio->bi_private;
762 struct mddev *mddev = rdev->mddev;
764 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
765 printk("md: super_written gets error=%d, uptodate=%d\n",
766 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
767 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
768 md_error(mddev, rdev);
771 if (atomic_dec_and_test(&mddev->pending_writes))
772 wake_up(&mddev->sb_wait);
776 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
777 sector_t sector, int size, struct page *page)
779 /* write first size bytes of page to sector of rdev
780 * Increment mddev->pending_writes before returning
781 * and decrement it on completion, waking up sb_wait
782 * if zero is reached.
783 * If an error occurred, call md_error
785 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
787 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
788 bio->bi_sector = sector;
789 bio_add_page(bio, page, size, 0);
790 bio->bi_private = rdev;
791 bio->bi_end_io = super_written;
793 atomic_inc(&mddev->pending_writes);
794 submit_bio(WRITE_FLUSH_FUA, bio);
797 void md_super_wait(struct mddev *mddev)
799 /* wait for all superblock writes that were scheduled to complete */
802 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
803 if (atomic_read(&mddev->pending_writes)==0)
807 finish_wait(&mddev->sb_wait, &wq);
810 static void bi_complete(struct bio *bio, int error)
812 complete((struct completion*)bio->bi_private);
815 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
816 struct page *page, int rw, bool metadata_op)
818 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
819 struct completion event;
824 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
825 rdev->meta_bdev : rdev->bdev;
827 bio->bi_sector = sector + rdev->sb_start;
828 else if (rdev->mddev->reshape_position != MaxSector &&
829 (rdev->mddev->reshape_backwards ==
830 (sector >= rdev->mddev->reshape_position)))
831 bio->bi_sector = sector + rdev->new_data_offset;
833 bio->bi_sector = sector + rdev->data_offset;
834 bio_add_page(bio, page, size, 0);
835 init_completion(&event);
836 bio->bi_private = &event;
837 bio->bi_end_io = bi_complete;
839 wait_for_completion(&event);
841 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
845 EXPORT_SYMBOL_GPL(sync_page_io);
847 static int read_disk_sb(struct md_rdev * rdev, int size)
849 char b[BDEVNAME_SIZE];
850 if (!rdev->sb_page) {
858 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
864 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
865 bdevname(rdev->bdev,b));
869 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
871 return sb1->set_uuid0 == sb2->set_uuid0 &&
872 sb1->set_uuid1 == sb2->set_uuid1 &&
873 sb1->set_uuid2 == sb2->set_uuid2 &&
874 sb1->set_uuid3 == sb2->set_uuid3;
877 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
880 mdp_super_t *tmp1, *tmp2;
882 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
883 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
885 if (!tmp1 || !tmp2) {
887 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
895 * nr_disks is not constant
900 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
908 static u32 md_csum_fold(u32 csum)
910 csum = (csum & 0xffff) + (csum >> 16);
911 return (csum & 0xffff) + (csum >> 16);
914 static unsigned int calc_sb_csum(mdp_super_t * sb)
917 u32 *sb32 = (u32*)sb;
919 unsigned int disk_csum, csum;
921 disk_csum = sb->sb_csum;
924 for (i = 0; i < MD_SB_BYTES/4 ; i++)
926 csum = (newcsum & 0xffffffff) + (newcsum>>32);
930 /* This used to use csum_partial, which was wrong for several
931 * reasons including that different results are returned on
932 * different architectures. It isn't critical that we get exactly
933 * the same return value as before (we always csum_fold before
934 * testing, and that removes any differences). However as we
935 * know that csum_partial always returned a 16bit value on
936 * alphas, do a fold to maximise conformity to previous behaviour.
938 sb->sb_csum = md_csum_fold(disk_csum);
940 sb->sb_csum = disk_csum;
947 * Handle superblock details.
948 * We want to be able to handle multiple superblock formats
949 * so we have a common interface to them all, and an array of
950 * different handlers.
951 * We rely on user-space to write the initial superblock, and support
952 * reading and updating of superblocks.
953 * Interface methods are:
954 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
955 * loads and validates a superblock on dev.
956 * if refdev != NULL, compare superblocks on both devices
958 * 0 - dev has a superblock that is compatible with refdev
959 * 1 - dev has a superblock that is compatible and newer than refdev
960 * so dev should be used as the refdev in future
961 * -EINVAL superblock incompatible or invalid
962 * -othererror e.g. -EIO
964 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
965 * Verify that dev is acceptable into mddev.
966 * The first time, mddev->raid_disks will be 0, and data from
967 * dev should be merged in. Subsequent calls check that dev
968 * is new enough. Return 0 or -EINVAL
970 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
971 * Update the superblock for rdev with data in mddev
972 * This does not write to disc.
978 struct module *owner;
979 int (*load_super)(struct md_rdev *rdev,
980 struct md_rdev *refdev,
982 int (*validate_super)(struct mddev *mddev,
983 struct md_rdev *rdev);
984 void (*sync_super)(struct mddev *mddev,
985 struct md_rdev *rdev);
986 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
987 sector_t num_sectors);
988 int (*allow_new_offset)(struct md_rdev *rdev,
989 unsigned long long new_offset);
993 * Check that the given mddev has no bitmap.
995 * This function is called from the run method of all personalities that do not
996 * support bitmaps. It prints an error message and returns non-zero if mddev
997 * has a bitmap. Otherwise, it returns 0.
1000 int md_check_no_bitmap(struct mddev *mddev)
1002 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1004 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1005 mdname(mddev), mddev->pers->name);
1008 EXPORT_SYMBOL(md_check_no_bitmap);
1011 * load_super for 0.90.0
1013 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1015 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1020 * Calculate the position of the superblock (512byte sectors),
1021 * it's at the end of the disk.
1023 * It also happens to be a multiple of 4Kb.
1025 rdev->sb_start = calc_dev_sboffset(rdev);
1027 ret = read_disk_sb(rdev, MD_SB_BYTES);
1028 if (ret) return ret;
1032 bdevname(rdev->bdev, b);
1033 sb = page_address(rdev->sb_page);
1035 if (sb->md_magic != MD_SB_MAGIC) {
1036 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1041 if (sb->major_version != 0 ||
1042 sb->minor_version < 90 ||
1043 sb->minor_version > 91) {
1044 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1045 sb->major_version, sb->minor_version,
1050 if (sb->raid_disks <= 0)
1053 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1054 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1059 rdev->preferred_minor = sb->md_minor;
1060 rdev->data_offset = 0;
1061 rdev->new_data_offset = 0;
1062 rdev->sb_size = MD_SB_BYTES;
1063 rdev->badblocks.shift = -1;
1065 if (sb->level == LEVEL_MULTIPATH)
1068 rdev->desc_nr = sb->this_disk.number;
1074 mdp_super_t *refsb = page_address(refdev->sb_page);
1075 if (!uuid_equal(refsb, sb)) {
1076 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1077 b, bdevname(refdev->bdev,b2));
1080 if (!sb_equal(refsb, sb)) {
1081 printk(KERN_WARNING "md: %s has same UUID"
1082 " but different superblock to %s\n",
1083 b, bdevname(refdev->bdev, b2));
1087 ev2 = md_event(refsb);
1093 rdev->sectors = rdev->sb_start;
1094 /* Limit to 4TB as metadata cannot record more than that.
1095 * (not needed for Linear and RAID0 as metadata doesn't
1098 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1099 rdev->sectors = (2ULL << 32) - 2;
1101 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1102 /* "this cannot possibly happen" ... */
1110 * validate_super for 0.90.0
1112 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1115 mdp_super_t *sb = page_address(rdev->sb_page);
1116 __u64 ev1 = md_event(sb);
1118 rdev->raid_disk = -1;
1119 clear_bit(Faulty, &rdev->flags);
1120 clear_bit(In_sync, &rdev->flags);
1121 clear_bit(Bitmap_sync, &rdev->flags);
1122 clear_bit(WriteMostly, &rdev->flags);
1124 if (mddev->raid_disks == 0) {
1125 mddev->major_version = 0;
1126 mddev->minor_version = sb->minor_version;
1127 mddev->patch_version = sb->patch_version;
1128 mddev->external = 0;
1129 mddev->chunk_sectors = sb->chunk_size >> 9;
1130 mddev->ctime = sb->ctime;
1131 mddev->utime = sb->utime;
1132 mddev->level = sb->level;
1133 mddev->clevel[0] = 0;
1134 mddev->layout = sb->layout;
1135 mddev->raid_disks = sb->raid_disks;
1136 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1137 mddev->events = ev1;
1138 mddev->bitmap_info.offset = 0;
1139 mddev->bitmap_info.space = 0;
1140 /* bitmap can use 60 K after the 4K superblocks */
1141 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1142 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1143 mddev->reshape_backwards = 0;
1145 if (mddev->minor_version >= 91) {
1146 mddev->reshape_position = sb->reshape_position;
1147 mddev->delta_disks = sb->delta_disks;
1148 mddev->new_level = sb->new_level;
1149 mddev->new_layout = sb->new_layout;
1150 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1151 if (mddev->delta_disks < 0)
1152 mddev->reshape_backwards = 1;
1154 mddev->reshape_position = MaxSector;
1155 mddev->delta_disks = 0;
1156 mddev->new_level = mddev->level;
1157 mddev->new_layout = mddev->layout;
1158 mddev->new_chunk_sectors = mddev->chunk_sectors;
1161 if (sb->state & (1<<MD_SB_CLEAN))
1162 mddev->recovery_cp = MaxSector;
1164 if (sb->events_hi == sb->cp_events_hi &&
1165 sb->events_lo == sb->cp_events_lo) {
1166 mddev->recovery_cp = sb->recovery_cp;
1168 mddev->recovery_cp = 0;
1171 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1172 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1173 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1174 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1176 mddev->max_disks = MD_SB_DISKS;
1178 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1179 mddev->bitmap_info.file == NULL) {
1180 mddev->bitmap_info.offset =
1181 mddev->bitmap_info.default_offset;
1182 mddev->bitmap_info.space =
1183 mddev->bitmap_info.space;
1186 } else if (mddev->pers == NULL) {
1187 /* Insist on good event counter while assembling, except
1188 * for spares (which don't need an event count) */
1190 if (sb->disks[rdev->desc_nr].state & (
1191 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1192 if (ev1 < mddev->events)
1194 } else if (mddev->bitmap) {
1195 /* if adding to array with a bitmap, then we can accept an
1196 * older device ... but not too old.
1198 if (ev1 < mddev->bitmap->events_cleared)
1200 if (ev1 < mddev->events)
1201 set_bit(Bitmap_sync, &rdev->flags);
1203 if (ev1 < mddev->events)
1204 /* just a hot-add of a new device, leave raid_disk at -1 */
1208 if (mddev->level != LEVEL_MULTIPATH) {
1209 desc = sb->disks + rdev->desc_nr;
1211 if (desc->state & (1<<MD_DISK_FAULTY))
1212 set_bit(Faulty, &rdev->flags);
1213 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1214 desc->raid_disk < mddev->raid_disks */) {
1215 set_bit(In_sync, &rdev->flags);
1216 rdev->raid_disk = desc->raid_disk;
1217 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1218 /* active but not in sync implies recovery up to
1219 * reshape position. We don't know exactly where
1220 * that is, so set to zero for now */
1221 if (mddev->minor_version >= 91) {
1222 rdev->recovery_offset = 0;
1223 rdev->raid_disk = desc->raid_disk;
1226 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1227 set_bit(WriteMostly, &rdev->flags);
1228 } else /* MULTIPATH are always insync */
1229 set_bit(In_sync, &rdev->flags);
1234 * sync_super for 0.90.0
1236 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1239 struct md_rdev *rdev2;
1240 int next_spare = mddev->raid_disks;
1243 /* make rdev->sb match mddev data..
1246 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1247 * 3/ any empty disks < next_spare become removed
1249 * disks[0] gets initialised to REMOVED because
1250 * we cannot be sure from other fields if it has
1251 * been initialised or not.
1254 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1256 rdev->sb_size = MD_SB_BYTES;
1258 sb = page_address(rdev->sb_page);
1260 memset(sb, 0, sizeof(*sb));
1262 sb->md_magic = MD_SB_MAGIC;
1263 sb->major_version = mddev->major_version;
1264 sb->patch_version = mddev->patch_version;
1265 sb->gvalid_words = 0; /* ignored */
1266 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1267 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1268 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1269 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1271 sb->ctime = mddev->ctime;
1272 sb->level = mddev->level;
1273 sb->size = mddev->dev_sectors / 2;
1274 sb->raid_disks = mddev->raid_disks;
1275 sb->md_minor = mddev->md_minor;
1276 sb->not_persistent = 0;
1277 sb->utime = mddev->utime;
1279 sb->events_hi = (mddev->events>>32);
1280 sb->events_lo = (u32)mddev->events;
1282 if (mddev->reshape_position == MaxSector)
1283 sb->minor_version = 90;
1285 sb->minor_version = 91;
1286 sb->reshape_position = mddev->reshape_position;
1287 sb->new_level = mddev->new_level;
1288 sb->delta_disks = mddev->delta_disks;
1289 sb->new_layout = mddev->new_layout;
1290 sb->new_chunk = mddev->new_chunk_sectors << 9;
1292 mddev->minor_version = sb->minor_version;
1295 sb->recovery_cp = mddev->recovery_cp;
1296 sb->cp_events_hi = (mddev->events>>32);
1297 sb->cp_events_lo = (u32)mddev->events;
1298 if (mddev->recovery_cp == MaxSector)
1299 sb->state = (1<< MD_SB_CLEAN);
1301 sb->recovery_cp = 0;
1303 sb->layout = mddev->layout;
1304 sb->chunk_size = mddev->chunk_sectors << 9;
1306 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1307 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1309 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1310 rdev_for_each(rdev2, mddev) {
1313 int is_active = test_bit(In_sync, &rdev2->flags);
1315 if (rdev2->raid_disk >= 0 &&
1316 sb->minor_version >= 91)
1317 /* we have nowhere to store the recovery_offset,
1318 * but if it is not below the reshape_position,
1319 * we can piggy-back on that.
1322 if (rdev2->raid_disk < 0 ||
1323 test_bit(Faulty, &rdev2->flags))
1326 desc_nr = rdev2->raid_disk;
1328 desc_nr = next_spare++;
1329 rdev2->desc_nr = desc_nr;
1330 d = &sb->disks[rdev2->desc_nr];
1332 d->number = rdev2->desc_nr;
1333 d->major = MAJOR(rdev2->bdev->bd_dev);
1334 d->minor = MINOR(rdev2->bdev->bd_dev);
1336 d->raid_disk = rdev2->raid_disk;
1338 d->raid_disk = rdev2->desc_nr; /* compatibility */
1339 if (test_bit(Faulty, &rdev2->flags))
1340 d->state = (1<<MD_DISK_FAULTY);
1341 else if (is_active) {
1342 d->state = (1<<MD_DISK_ACTIVE);
1343 if (test_bit(In_sync, &rdev2->flags))
1344 d->state |= (1<<MD_DISK_SYNC);
1352 if (test_bit(WriteMostly, &rdev2->flags))
1353 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1355 /* now set the "removed" and "faulty" bits on any missing devices */
1356 for (i=0 ; i < mddev->raid_disks ; i++) {
1357 mdp_disk_t *d = &sb->disks[i];
1358 if (d->state == 0 && d->number == 0) {
1361 d->state = (1<<MD_DISK_REMOVED);
1362 d->state |= (1<<MD_DISK_FAULTY);
1366 sb->nr_disks = nr_disks;
1367 sb->active_disks = active;
1368 sb->working_disks = working;
1369 sb->failed_disks = failed;
1370 sb->spare_disks = spare;
1372 sb->this_disk = sb->disks[rdev->desc_nr];
1373 sb->sb_csum = calc_sb_csum(sb);
1377 * rdev_size_change for 0.90.0
1379 static unsigned long long
1380 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1382 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1383 return 0; /* component must fit device */
1384 if (rdev->mddev->bitmap_info.offset)
1385 return 0; /* can't move bitmap */
1386 rdev->sb_start = calc_dev_sboffset(rdev);
1387 if (!num_sectors || num_sectors > rdev->sb_start)
1388 num_sectors = rdev->sb_start;
1389 /* Limit to 4TB as metadata cannot record more than that.
1390 * 4TB == 2^32 KB, or 2*2^32 sectors.
1392 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1393 num_sectors = (2ULL << 32) - 2;
1394 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1396 md_super_wait(rdev->mddev);
1401 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1403 /* non-zero offset changes not possible with v0.90 */
1404 return new_offset == 0;
1408 * version 1 superblock
1411 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1415 unsigned long long newcsum;
1416 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1417 __le32 *isuper = (__le32*)sb;
1419 disk_csum = sb->sb_csum;
1422 for (; size >= 4; size -= 4)
1423 newcsum += le32_to_cpu(*isuper++);
1426 newcsum += le16_to_cpu(*(__le16*) isuper);
1428 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1429 sb->sb_csum = disk_csum;
1430 return cpu_to_le32(csum);
1433 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1435 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1437 struct mdp_superblock_1 *sb;
1441 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1445 * Calculate the position of the superblock in 512byte sectors.
1446 * It is always aligned to a 4K boundary and
1447 * depeding on minor_version, it can be:
1448 * 0: At least 8K, but less than 12K, from end of device
1449 * 1: At start of device
1450 * 2: 4K from start of device.
1452 switch(minor_version) {
1454 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1456 sb_start &= ~(sector_t)(4*2-1);
1467 rdev->sb_start = sb_start;
1469 /* superblock is rarely larger than 1K, but it can be larger,
1470 * and it is safe to read 4k, so we do that
1472 ret = read_disk_sb(rdev, 4096);
1473 if (ret) return ret;
1476 sb = page_address(rdev->sb_page);
1478 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1479 sb->major_version != cpu_to_le32(1) ||
1480 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1481 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1482 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1485 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1486 printk("md: invalid superblock checksum on %s\n",
1487 bdevname(rdev->bdev,b));
1490 if (le64_to_cpu(sb->data_size) < 10) {
1491 printk("md: data_size too small on %s\n",
1492 bdevname(rdev->bdev,b));
1497 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1498 /* Some padding is non-zero, might be a new feature */
1501 rdev->preferred_minor = 0xffff;
1502 rdev->data_offset = le64_to_cpu(sb->data_offset);
1503 rdev->new_data_offset = rdev->data_offset;
1504 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1505 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1506 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1507 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1509 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1510 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1511 if (rdev->sb_size & bmask)
1512 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1515 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1518 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1521 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1524 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1526 if (!rdev->bb_page) {
1527 rdev->bb_page = alloc_page(GFP_KERNEL);
1531 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1532 rdev->badblocks.count == 0) {
1533 /* need to load the bad block list.
1534 * Currently we limit it to one page.
1540 int sectors = le16_to_cpu(sb->bblog_size);
1541 if (sectors > (PAGE_SIZE / 512))
1543 offset = le32_to_cpu(sb->bblog_offset);
1546 bb_sector = (long long)offset;
1547 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1548 rdev->bb_page, READ, true))
1550 bbp = (u64 *)page_address(rdev->bb_page);
1551 rdev->badblocks.shift = sb->bblog_shift;
1552 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1553 u64 bb = le64_to_cpu(*bbp);
1554 int count = bb & (0x3ff);
1555 u64 sector = bb >> 10;
1556 sector <<= sb->bblog_shift;
1557 count <<= sb->bblog_shift;
1560 if (md_set_badblocks(&rdev->badblocks,
1561 sector, count, 1) == 0)
1564 } else if (sb->bblog_offset != 0)
1565 rdev->badblocks.shift = 0;
1571 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1573 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1574 sb->level != refsb->level ||
1575 sb->layout != refsb->layout ||
1576 sb->chunksize != refsb->chunksize) {
1577 printk(KERN_WARNING "md: %s has strangely different"
1578 " superblock to %s\n",
1579 bdevname(rdev->bdev,b),
1580 bdevname(refdev->bdev,b2));
1583 ev1 = le64_to_cpu(sb->events);
1584 ev2 = le64_to_cpu(refsb->events);
1591 if (minor_version) {
1592 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1593 sectors -= rdev->data_offset;
1595 sectors = rdev->sb_start;
1596 if (sectors < le64_to_cpu(sb->data_size))
1598 rdev->sectors = le64_to_cpu(sb->data_size);
1602 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1604 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1605 __u64 ev1 = le64_to_cpu(sb->events);
1607 rdev->raid_disk = -1;
1608 clear_bit(Faulty, &rdev->flags);
1609 clear_bit(In_sync, &rdev->flags);
1610 clear_bit(Bitmap_sync, &rdev->flags);
1611 clear_bit(WriteMostly, &rdev->flags);
1613 if (mddev->raid_disks == 0) {
1614 mddev->major_version = 1;
1615 mddev->patch_version = 0;
1616 mddev->external = 0;
1617 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1618 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1619 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1620 mddev->level = le32_to_cpu(sb->level);
1621 mddev->clevel[0] = 0;
1622 mddev->layout = le32_to_cpu(sb->layout);
1623 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1624 mddev->dev_sectors = le64_to_cpu(sb->size);
1625 mddev->events = ev1;
1626 mddev->bitmap_info.offset = 0;
1627 mddev->bitmap_info.space = 0;
1628 /* Default location for bitmap is 1K after superblock
1629 * using 3K - total of 4K
1631 mddev->bitmap_info.default_offset = 1024 >> 9;
1632 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1633 mddev->reshape_backwards = 0;
1635 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1636 memcpy(mddev->uuid, sb->set_uuid, 16);
1638 mddev->max_disks = (4096-256)/2;
1640 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1641 mddev->bitmap_info.file == NULL) {
1642 mddev->bitmap_info.offset =
1643 (__s32)le32_to_cpu(sb->bitmap_offset);
1644 /* Metadata doesn't record how much space is available.
1645 * For 1.0, we assume we can use up to the superblock
1646 * if before, else to 4K beyond superblock.
1647 * For others, assume no change is possible.
1649 if (mddev->minor_version > 0)
1650 mddev->bitmap_info.space = 0;
1651 else if (mddev->bitmap_info.offset > 0)
1652 mddev->bitmap_info.space =
1653 8 - mddev->bitmap_info.offset;
1655 mddev->bitmap_info.space =
1656 -mddev->bitmap_info.offset;
1659 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1660 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1661 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1662 mddev->new_level = le32_to_cpu(sb->new_level);
1663 mddev->new_layout = le32_to_cpu(sb->new_layout);
1664 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1665 if (mddev->delta_disks < 0 ||
1666 (mddev->delta_disks == 0 &&
1667 (le32_to_cpu(sb->feature_map)
1668 & MD_FEATURE_RESHAPE_BACKWARDS)))
1669 mddev->reshape_backwards = 1;
1671 mddev->reshape_position = MaxSector;
1672 mddev->delta_disks = 0;
1673 mddev->new_level = mddev->level;
1674 mddev->new_layout = mddev->layout;
1675 mddev->new_chunk_sectors = mddev->chunk_sectors;
1678 } else if (mddev->pers == NULL) {
1679 /* Insist of good event counter while assembling, except for
1680 * spares (which don't need an event count) */
1682 if (rdev->desc_nr >= 0 &&
1683 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1684 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1685 if (ev1 < mddev->events)
1687 } else if (mddev->bitmap) {
1688 /* If adding to array with a bitmap, then we can accept an
1689 * older device, but not too old.
1691 if (ev1 < mddev->bitmap->events_cleared)
1693 if (ev1 < mddev->events)
1694 set_bit(Bitmap_sync, &rdev->flags);
1696 if (ev1 < mddev->events)
1697 /* just a hot-add of a new device, leave raid_disk at -1 */
1700 if (mddev->level != LEVEL_MULTIPATH) {
1702 if (rdev->desc_nr < 0 ||
1703 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1707 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1709 case 0xffff: /* spare */
1711 case 0xfffe: /* faulty */
1712 set_bit(Faulty, &rdev->flags);
1715 if ((le32_to_cpu(sb->feature_map) &
1716 MD_FEATURE_RECOVERY_OFFSET))
1717 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1719 set_bit(In_sync, &rdev->flags);
1720 rdev->raid_disk = role;
1723 if (sb->devflags & WriteMostly1)
1724 set_bit(WriteMostly, &rdev->flags);
1725 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1726 set_bit(Replacement, &rdev->flags);
1727 } else /* MULTIPATH are always insync */
1728 set_bit(In_sync, &rdev->flags);
1733 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1735 struct mdp_superblock_1 *sb;
1736 struct md_rdev *rdev2;
1738 /* make rdev->sb match mddev and rdev data. */
1740 sb = page_address(rdev->sb_page);
1742 sb->feature_map = 0;
1744 sb->recovery_offset = cpu_to_le64(0);
1745 memset(sb->pad3, 0, sizeof(sb->pad3));
1747 sb->utime = cpu_to_le64((__u64)mddev->utime);
1748 sb->events = cpu_to_le64(mddev->events);
1750 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1752 sb->resync_offset = cpu_to_le64(0);
1754 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1756 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1757 sb->size = cpu_to_le64(mddev->dev_sectors);
1758 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1759 sb->level = cpu_to_le32(mddev->level);
1760 sb->layout = cpu_to_le32(mddev->layout);
1762 if (test_bit(WriteMostly, &rdev->flags))
1763 sb->devflags |= WriteMostly1;
1765 sb->devflags &= ~WriteMostly1;
1766 sb->data_offset = cpu_to_le64(rdev->data_offset);
1767 sb->data_size = cpu_to_le64(rdev->sectors);
1769 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1770 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1771 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1774 if (rdev->raid_disk >= 0 &&
1775 !test_bit(In_sync, &rdev->flags)) {
1777 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1778 sb->recovery_offset =
1779 cpu_to_le64(rdev->recovery_offset);
1781 if (test_bit(Replacement, &rdev->flags))
1783 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1785 if (mddev->reshape_position != MaxSector) {
1786 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1787 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1788 sb->new_layout = cpu_to_le32(mddev->new_layout);
1789 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1790 sb->new_level = cpu_to_le32(mddev->new_level);
1791 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1792 if (mddev->delta_disks == 0 &&
1793 mddev->reshape_backwards)
1795 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1796 if (rdev->new_data_offset != rdev->data_offset) {
1798 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1799 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1800 - rdev->data_offset));
1804 if (rdev->badblocks.count == 0)
1805 /* Nothing to do for bad blocks*/ ;
1806 else if (sb->bblog_offset == 0)
1807 /* Cannot record bad blocks on this device */
1808 md_error(mddev, rdev);
1810 struct badblocks *bb = &rdev->badblocks;
1811 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1813 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1818 seq = read_seqbegin(&bb->lock);
1820 memset(bbp, 0xff, PAGE_SIZE);
1822 for (i = 0 ; i < bb->count ; i++) {
1823 u64 internal_bb = p[i];
1824 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1825 | BB_LEN(internal_bb));
1826 bbp[i] = cpu_to_le64(store_bb);
1829 if (read_seqretry(&bb->lock, seq))
1832 bb->sector = (rdev->sb_start +
1833 (int)le32_to_cpu(sb->bblog_offset));
1834 bb->size = le16_to_cpu(sb->bblog_size);
1839 rdev_for_each(rdev2, mddev)
1840 if (rdev2->desc_nr+1 > max_dev)
1841 max_dev = rdev2->desc_nr+1;
1843 if (max_dev > le32_to_cpu(sb->max_dev)) {
1845 sb->max_dev = cpu_to_le32(max_dev);
1846 rdev->sb_size = max_dev * 2 + 256;
1847 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1848 if (rdev->sb_size & bmask)
1849 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1851 max_dev = le32_to_cpu(sb->max_dev);
1853 for (i=0; i<max_dev;i++)
1854 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1856 rdev_for_each(rdev2, mddev) {
1858 if (test_bit(Faulty, &rdev2->flags))
1859 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1860 else if (test_bit(In_sync, &rdev2->flags))
1861 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1862 else if (rdev2->raid_disk >= 0)
1863 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1865 sb->dev_roles[i] = cpu_to_le16(0xffff);
1868 sb->sb_csum = calc_sb_1_csum(sb);
1871 static unsigned long long
1872 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1874 struct mdp_superblock_1 *sb;
1875 sector_t max_sectors;
1876 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1877 return 0; /* component must fit device */
1878 if (rdev->data_offset != rdev->new_data_offset)
1879 return 0; /* too confusing */
1880 if (rdev->sb_start < rdev->data_offset) {
1881 /* minor versions 1 and 2; superblock before data */
1882 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1883 max_sectors -= rdev->data_offset;
1884 if (!num_sectors || num_sectors > max_sectors)
1885 num_sectors = max_sectors;
1886 } else if (rdev->mddev->bitmap_info.offset) {
1887 /* minor version 0 with bitmap we can't move */
1890 /* minor version 0; superblock after data */
1892 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1893 sb_start &= ~(sector_t)(4*2 - 1);
1894 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1895 if (!num_sectors || num_sectors > max_sectors)
1896 num_sectors = max_sectors;
1897 rdev->sb_start = sb_start;
1899 sb = page_address(rdev->sb_page);
1900 sb->data_size = cpu_to_le64(num_sectors);
1901 sb->super_offset = rdev->sb_start;
1902 sb->sb_csum = calc_sb_1_csum(sb);
1903 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1905 md_super_wait(rdev->mddev);
1911 super_1_allow_new_offset(struct md_rdev *rdev,
1912 unsigned long long new_offset)
1914 /* All necessary checks on new >= old have been done */
1915 struct bitmap *bitmap;
1916 if (new_offset >= rdev->data_offset)
1919 /* with 1.0 metadata, there is no metadata to tread on
1920 * so we can always move back */
1921 if (rdev->mddev->minor_version == 0)
1924 /* otherwise we must be sure not to step on
1925 * any metadata, so stay:
1926 * 36K beyond start of superblock
1927 * beyond end of badblocks
1928 * beyond write-intent bitmap
1930 if (rdev->sb_start + (32+4)*2 > new_offset)
1932 bitmap = rdev->mddev->bitmap;
1933 if (bitmap && !rdev->mddev->bitmap_info.file &&
1934 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1935 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1937 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1943 static struct super_type super_types[] = {
1946 .owner = THIS_MODULE,
1947 .load_super = super_90_load,
1948 .validate_super = super_90_validate,
1949 .sync_super = super_90_sync,
1950 .rdev_size_change = super_90_rdev_size_change,
1951 .allow_new_offset = super_90_allow_new_offset,
1955 .owner = THIS_MODULE,
1956 .load_super = super_1_load,
1957 .validate_super = super_1_validate,
1958 .sync_super = super_1_sync,
1959 .rdev_size_change = super_1_rdev_size_change,
1960 .allow_new_offset = super_1_allow_new_offset,
1964 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1966 if (mddev->sync_super) {
1967 mddev->sync_super(mddev, rdev);
1971 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1973 super_types[mddev->major_version].sync_super(mddev, rdev);
1976 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1978 struct md_rdev *rdev, *rdev2;
1981 rdev_for_each_rcu(rdev, mddev1)
1982 rdev_for_each_rcu(rdev2, mddev2)
1983 if (rdev->bdev->bd_contains ==
1984 rdev2->bdev->bd_contains) {
1992 static LIST_HEAD(pending_raid_disks);
1995 * Try to register data integrity profile for an mddev
1997 * This is called when an array is started and after a disk has been kicked
1998 * from the array. It only succeeds if all working and active component devices
1999 * are integrity capable with matching profiles.
2001 int md_integrity_register(struct mddev *mddev)
2003 struct md_rdev *rdev, *reference = NULL;
2005 if (list_empty(&mddev->disks))
2006 return 0; /* nothing to do */
2007 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2008 return 0; /* shouldn't register, or already is */
2009 rdev_for_each(rdev, mddev) {
2010 /* skip spares and non-functional disks */
2011 if (test_bit(Faulty, &rdev->flags))
2013 if (rdev->raid_disk < 0)
2016 /* Use the first rdev as the reference */
2020 /* does this rdev's profile match the reference profile? */
2021 if (blk_integrity_compare(reference->bdev->bd_disk,
2022 rdev->bdev->bd_disk) < 0)
2025 if (!reference || !bdev_get_integrity(reference->bdev))
2028 * All component devices are integrity capable and have matching
2029 * profiles, register the common profile for the md device.
2031 if (blk_integrity_register(mddev->gendisk,
2032 bdev_get_integrity(reference->bdev)) != 0) {
2033 printk(KERN_ERR "md: failed to register integrity for %s\n",
2037 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2038 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2039 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2045 EXPORT_SYMBOL(md_integrity_register);
2047 /* Disable data integrity if non-capable/non-matching disk is being added */
2048 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2050 struct blk_integrity *bi_rdev;
2051 struct blk_integrity *bi_mddev;
2053 if (!mddev->gendisk)
2056 bi_rdev = bdev_get_integrity(rdev->bdev);
2057 bi_mddev = blk_get_integrity(mddev->gendisk);
2059 if (!bi_mddev) /* nothing to do */
2061 if (rdev->raid_disk < 0) /* skip spares */
2063 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2064 rdev->bdev->bd_disk) >= 0)
2066 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2067 blk_integrity_unregister(mddev->gendisk);
2069 EXPORT_SYMBOL(md_integrity_add_rdev);
2071 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2073 char b[BDEVNAME_SIZE];
2083 /* prevent duplicates */
2084 if (find_rdev(mddev, rdev->bdev->bd_dev))
2087 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2088 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2089 rdev->sectors < mddev->dev_sectors)) {
2091 /* Cannot change size, so fail
2092 * If mddev->level <= 0, then we don't care
2093 * about aligning sizes (e.g. linear)
2095 if (mddev->level > 0)
2098 mddev->dev_sectors = rdev->sectors;
2101 /* Verify rdev->desc_nr is unique.
2102 * If it is -1, assign a free number, else
2103 * check number is not in use
2105 if (rdev->desc_nr < 0) {
2107 if (mddev->pers) choice = mddev->raid_disks;
2108 while (find_rdev_nr(mddev, choice))
2110 rdev->desc_nr = choice;
2112 if (find_rdev_nr(mddev, rdev->desc_nr))
2115 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2116 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2117 mdname(mddev), mddev->max_disks);
2120 bdevname(rdev->bdev,b);
2121 while ( (s=strchr(b, '/')) != NULL)
2124 rdev->mddev = mddev;
2125 printk(KERN_INFO "md: bind<%s>\n", b);
2127 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2130 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2131 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2132 /* failure here is OK */;
2133 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2135 list_add_rcu(&rdev->same_set, &mddev->disks);
2136 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2138 /* May as well allow recovery to be retried once */
2139 mddev->recovery_disabled++;
2144 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2149 static void md_delayed_delete(struct work_struct *ws)
2151 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2152 kobject_del(&rdev->kobj);
2153 kobject_put(&rdev->kobj);
2156 static void unbind_rdev_from_array(struct md_rdev * rdev)
2158 char b[BDEVNAME_SIZE];
2163 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2164 list_del_rcu(&rdev->same_set);
2165 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2167 sysfs_remove_link(&rdev->kobj, "block");
2168 sysfs_put(rdev->sysfs_state);
2169 rdev->sysfs_state = NULL;
2170 rdev->badblocks.count = 0;
2171 /* We need to delay this, otherwise we can deadlock when
2172 * writing to 'remove' to "dev/state". We also need
2173 * to delay it due to rcu usage.
2176 INIT_WORK(&rdev->del_work, md_delayed_delete);
2177 kobject_get(&rdev->kobj);
2178 queue_work(md_misc_wq, &rdev->del_work);
2182 * prevent the device from being mounted, repartitioned or
2183 * otherwise reused by a RAID array (or any other kernel
2184 * subsystem), by bd_claiming the device.
2186 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2189 struct block_device *bdev;
2190 char b[BDEVNAME_SIZE];
2192 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2193 shared ? (struct md_rdev *)lock_rdev : rdev);
2195 printk(KERN_ERR "md: could not open %s.\n",
2196 __bdevname(dev, b));
2197 return PTR_ERR(bdev);
2203 static void unlock_rdev(struct md_rdev *rdev)
2205 struct block_device *bdev = rdev->bdev;
2209 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2212 void md_autodetect_dev(dev_t dev);
2214 static void export_rdev(struct md_rdev * rdev)
2216 char b[BDEVNAME_SIZE];
2217 printk(KERN_INFO "md: export_rdev(%s)\n",
2218 bdevname(rdev->bdev,b));
2221 md_rdev_clear(rdev);
2223 if (test_bit(AutoDetected, &rdev->flags))
2224 md_autodetect_dev(rdev->bdev->bd_dev);
2227 kobject_put(&rdev->kobj);
2230 static void kick_rdev_from_array(struct md_rdev * rdev)
2232 unbind_rdev_from_array(rdev);
2236 static void export_array(struct mddev *mddev)
2238 struct md_rdev *rdev, *tmp;
2240 rdev_for_each_safe(rdev, tmp, mddev) {
2245 kick_rdev_from_array(rdev);
2247 if (!list_empty(&mddev->disks))
2249 mddev->raid_disks = 0;
2250 mddev->major_version = 0;
2253 static void print_desc(mdp_disk_t *desc)
2255 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2256 desc->major,desc->minor,desc->raid_disk,desc->state);
2259 static void print_sb_90(mdp_super_t *sb)
2264 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2265 sb->major_version, sb->minor_version, sb->patch_version,
2266 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2268 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2269 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2270 sb->md_minor, sb->layout, sb->chunk_size);
2271 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2272 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2273 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2274 sb->failed_disks, sb->spare_disks,
2275 sb->sb_csum, (unsigned long)sb->events_lo);
2278 for (i = 0; i < MD_SB_DISKS; i++) {
2281 desc = sb->disks + i;
2282 if (desc->number || desc->major || desc->minor ||
2283 desc->raid_disk || (desc->state && (desc->state != 4))) {
2284 printk(" D %2d: ", i);
2288 printk(KERN_INFO "md: THIS: ");
2289 print_desc(&sb->this_disk);
2292 static void print_sb_1(struct mdp_superblock_1 *sb)
2296 uuid = sb->set_uuid;
2298 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2299 "md: Name: \"%s\" CT:%llu\n",
2300 le32_to_cpu(sb->major_version),
2301 le32_to_cpu(sb->feature_map),
2304 (unsigned long long)le64_to_cpu(sb->ctime)
2305 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2307 uuid = sb->device_uuid;
2309 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2311 "md: Dev:%08x UUID: %pU\n"
2312 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2313 "md: (MaxDev:%u) \n",
2314 le32_to_cpu(sb->level),
2315 (unsigned long long)le64_to_cpu(sb->size),
2316 le32_to_cpu(sb->raid_disks),
2317 le32_to_cpu(sb->layout),
2318 le32_to_cpu(sb->chunksize),
2319 (unsigned long long)le64_to_cpu(sb->data_offset),
2320 (unsigned long long)le64_to_cpu(sb->data_size),
2321 (unsigned long long)le64_to_cpu(sb->super_offset),
2322 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2323 le32_to_cpu(sb->dev_number),
2326 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2327 (unsigned long long)le64_to_cpu(sb->events),
2328 (unsigned long long)le64_to_cpu(sb->resync_offset),
2329 le32_to_cpu(sb->sb_csum),
2330 le32_to_cpu(sb->max_dev)
2334 static void print_rdev(struct md_rdev *rdev, int major_version)
2336 char b[BDEVNAME_SIZE];
2337 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2338 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2339 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2341 if (rdev->sb_loaded) {
2342 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2343 switch (major_version) {
2345 print_sb_90(page_address(rdev->sb_page));
2348 print_sb_1(page_address(rdev->sb_page));
2352 printk(KERN_INFO "md: no rdev superblock!\n");
2355 static void md_print_devices(void)
2357 struct list_head *tmp;
2358 struct md_rdev *rdev;
2359 struct mddev *mddev;
2360 char b[BDEVNAME_SIZE];
2363 printk("md: **********************************\n");
2364 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2365 printk("md: **********************************\n");
2366 for_each_mddev(mddev, tmp) {
2369 bitmap_print_sb(mddev->bitmap);
2371 printk("%s: ", mdname(mddev));
2372 rdev_for_each(rdev, mddev)
2373 printk("<%s>", bdevname(rdev->bdev,b));
2376 rdev_for_each(rdev, mddev)
2377 print_rdev(rdev, mddev->major_version);
2379 printk("md: **********************************\n");
2384 static void sync_sbs(struct mddev * mddev, int nospares)
2386 /* Update each superblock (in-memory image), but
2387 * if we are allowed to, skip spares which already
2388 * have the right event counter, or have one earlier
2389 * (which would mean they aren't being marked as dirty
2390 * with the rest of the array)
2392 struct md_rdev *rdev;
2393 rdev_for_each(rdev, mddev) {
2394 if (rdev->sb_events == mddev->events ||
2396 rdev->raid_disk < 0 &&
2397 rdev->sb_events+1 == mddev->events)) {
2398 /* Don't update this superblock */
2399 rdev->sb_loaded = 2;
2401 sync_super(mddev, rdev);
2402 rdev->sb_loaded = 1;
2407 static void md_update_sb(struct mddev * mddev, int force_change)
2409 struct md_rdev *rdev;
2412 int any_badblocks_changed = 0;
2416 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2420 /* First make sure individual recovery_offsets are correct */
2421 rdev_for_each(rdev, mddev) {
2422 if (rdev->raid_disk >= 0 &&
2423 mddev->delta_disks >= 0 &&
2424 !test_bit(In_sync, &rdev->flags) &&
2425 mddev->curr_resync_completed > rdev->recovery_offset)
2426 rdev->recovery_offset = mddev->curr_resync_completed;
2429 if (!mddev->persistent) {
2430 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2431 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2432 if (!mddev->external) {
2433 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2434 rdev_for_each(rdev, mddev) {
2435 if (rdev->badblocks.changed) {
2436 rdev->badblocks.changed = 0;
2437 md_ack_all_badblocks(&rdev->badblocks);
2438 md_error(mddev, rdev);
2440 clear_bit(Blocked, &rdev->flags);
2441 clear_bit(BlockedBadBlocks, &rdev->flags);
2442 wake_up(&rdev->blocked_wait);
2445 wake_up(&mddev->sb_wait);
2449 spin_lock_irq(&mddev->write_lock);
2451 mddev->utime = get_seconds();
2453 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2455 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2456 /* just a clean<-> dirty transition, possibly leave spares alone,
2457 * though if events isn't the right even/odd, we will have to do
2463 if (mddev->degraded)
2464 /* If the array is degraded, then skipping spares is both
2465 * dangerous and fairly pointless.
2466 * Dangerous because a device that was removed from the array
2467 * might have a event_count that still looks up-to-date,
2468 * so it can be re-added without a resync.
2469 * Pointless because if there are any spares to skip,
2470 * then a recovery will happen and soon that array won't
2471 * be degraded any more and the spare can go back to sleep then.
2475 sync_req = mddev->in_sync;
2477 /* If this is just a dirty<->clean transition, and the array is clean
2478 * and 'events' is odd, we can roll back to the previous clean state */
2480 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2481 && mddev->can_decrease_events
2482 && mddev->events != 1) {
2484 mddev->can_decrease_events = 0;
2486 /* otherwise we have to go forward and ... */
2488 mddev->can_decrease_events = nospares;
2491 if (!mddev->events) {
2493 * oops, this 64-bit counter should never wrap.
2494 * Either we are in around ~1 trillion A.C., assuming
2495 * 1 reboot per second, or we have a bug:
2501 rdev_for_each(rdev, mddev) {
2502 if (rdev->badblocks.changed)
2503 any_badblocks_changed++;
2504 if (test_bit(Faulty, &rdev->flags))
2505 set_bit(FaultRecorded, &rdev->flags);
2508 sync_sbs(mddev, nospares);
2509 spin_unlock_irq(&mddev->write_lock);
2511 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2512 mdname(mddev), mddev->in_sync);
2514 bitmap_update_sb(mddev->bitmap);
2515 rdev_for_each(rdev, mddev) {
2516 char b[BDEVNAME_SIZE];
2518 if (rdev->sb_loaded != 1)
2519 continue; /* no noise on spare devices */
2521 if (!test_bit(Faulty, &rdev->flags) &&
2522 rdev->saved_raid_disk == -1) {
2523 md_super_write(mddev,rdev,
2524 rdev->sb_start, rdev->sb_size,
2526 pr_debug("md: (write) %s's sb offset: %llu\n",
2527 bdevname(rdev->bdev, b),
2528 (unsigned long long)rdev->sb_start);
2529 rdev->sb_events = mddev->events;
2530 if (rdev->badblocks.size) {
2531 md_super_write(mddev, rdev,
2532 rdev->badblocks.sector,
2533 rdev->badblocks.size << 9,
2535 rdev->badblocks.size = 0;
2538 } else if (test_bit(Faulty, &rdev->flags))
2539 pr_debug("md: %s (skipping faulty)\n",
2540 bdevname(rdev->bdev, b));
2542 pr_debug("(skipping incremental s/r ");
2544 if (mddev->level == LEVEL_MULTIPATH)
2545 /* only need to write one superblock... */
2548 md_super_wait(mddev);
2549 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2551 spin_lock_irq(&mddev->write_lock);
2552 if (mddev->in_sync != sync_req ||
2553 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2554 /* have to write it out again */
2555 spin_unlock_irq(&mddev->write_lock);
2558 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2559 spin_unlock_irq(&mddev->write_lock);
2560 wake_up(&mddev->sb_wait);
2561 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2562 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2564 rdev_for_each(rdev, mddev) {
2565 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2566 clear_bit(Blocked, &rdev->flags);
2568 if (any_badblocks_changed)
2569 md_ack_all_badblocks(&rdev->badblocks);
2570 clear_bit(BlockedBadBlocks, &rdev->flags);
2571 wake_up(&rdev->blocked_wait);
2575 /* words written to sysfs files may, or may not, be \n terminated.
2576 * We want to accept with case. For this we use cmd_match.
2578 static int cmd_match(const char *cmd, const char *str)
2580 /* See if cmd, written into a sysfs file, matches
2581 * str. They must either be the same, or cmd can
2582 * have a trailing newline
2584 while (*cmd && *str && *cmd == *str) {
2595 struct rdev_sysfs_entry {
2596 struct attribute attr;
2597 ssize_t (*show)(struct md_rdev *, char *);
2598 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2602 state_show(struct md_rdev *rdev, char *page)
2607 if (test_bit(Faulty, &rdev->flags) ||
2608 rdev->badblocks.unacked_exist) {
2609 len+= sprintf(page+len, "%sfaulty",sep);
2612 if (test_bit(In_sync, &rdev->flags)) {
2613 len += sprintf(page+len, "%sin_sync",sep);
2616 if (test_bit(WriteMostly, &rdev->flags)) {
2617 len += sprintf(page+len, "%swrite_mostly",sep);
2620 if (test_bit(Blocked, &rdev->flags) ||
2621 (rdev->badblocks.unacked_exist
2622 && !test_bit(Faulty, &rdev->flags))) {
2623 len += sprintf(page+len, "%sblocked", sep);
2626 if (!test_bit(Faulty, &rdev->flags) &&
2627 !test_bit(In_sync, &rdev->flags)) {
2628 len += sprintf(page+len, "%sspare", sep);
2631 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2632 len += sprintf(page+len, "%swrite_error", sep);
2635 if (test_bit(WantReplacement, &rdev->flags)) {
2636 len += sprintf(page+len, "%swant_replacement", sep);
2639 if (test_bit(Replacement, &rdev->flags)) {
2640 len += sprintf(page+len, "%sreplacement", sep);
2644 return len+sprintf(page+len, "\n");
2648 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2651 * faulty - simulates an error
2652 * remove - disconnects the device
2653 * writemostly - sets write_mostly
2654 * -writemostly - clears write_mostly
2655 * blocked - sets the Blocked flags
2656 * -blocked - clears the Blocked and possibly simulates an error
2657 * insync - sets Insync providing device isn't active
2658 * write_error - sets WriteErrorSeen
2659 * -write_error - clears WriteErrorSeen
2662 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2663 md_error(rdev->mddev, rdev);
2664 if (test_bit(Faulty, &rdev->flags))
2668 } else if (cmd_match(buf, "remove")) {
2669 if (rdev->raid_disk >= 0)
2672 struct mddev *mddev = rdev->mddev;
2673 kick_rdev_from_array(rdev);
2675 md_update_sb(mddev, 1);
2676 md_new_event(mddev);
2679 } else if (cmd_match(buf, "writemostly")) {
2680 set_bit(WriteMostly, &rdev->flags);
2682 } else if (cmd_match(buf, "-writemostly")) {
2683 clear_bit(WriteMostly, &rdev->flags);
2685 } else if (cmd_match(buf, "blocked")) {
2686 set_bit(Blocked, &rdev->flags);
2688 } else if (cmd_match(buf, "-blocked")) {
2689 if (!test_bit(Faulty, &rdev->flags) &&
2690 rdev->badblocks.unacked_exist) {
2691 /* metadata handler doesn't understand badblocks,
2692 * so we need to fail the device
2694 md_error(rdev->mddev, rdev);
2696 clear_bit(Blocked, &rdev->flags);
2697 clear_bit(BlockedBadBlocks, &rdev->flags);
2698 wake_up(&rdev->blocked_wait);
2699 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2700 md_wakeup_thread(rdev->mddev->thread);
2703 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2704 set_bit(In_sync, &rdev->flags);
2706 } else if (cmd_match(buf, "write_error")) {
2707 set_bit(WriteErrorSeen, &rdev->flags);
2709 } else if (cmd_match(buf, "-write_error")) {
2710 clear_bit(WriteErrorSeen, &rdev->flags);
2712 } else if (cmd_match(buf, "want_replacement")) {
2713 /* Any non-spare device that is not a replacement can
2714 * become want_replacement at any time, but we then need to
2715 * check if recovery is needed.
2717 if (rdev->raid_disk >= 0 &&
2718 !test_bit(Replacement, &rdev->flags))
2719 set_bit(WantReplacement, &rdev->flags);
2720 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2721 md_wakeup_thread(rdev->mddev->thread);
2723 } else if (cmd_match(buf, "-want_replacement")) {
2724 /* Clearing 'want_replacement' is always allowed.
2725 * Once replacements starts it is too late though.
2728 clear_bit(WantReplacement, &rdev->flags);
2729 } else if (cmd_match(buf, "replacement")) {
2730 /* Can only set a device as a replacement when array has not
2731 * yet been started. Once running, replacement is automatic
2732 * from spares, or by assigning 'slot'.
2734 if (rdev->mddev->pers)
2737 set_bit(Replacement, &rdev->flags);
2740 } else if (cmd_match(buf, "-replacement")) {
2741 /* Similarly, can only clear Replacement before start */
2742 if (rdev->mddev->pers)
2745 clear_bit(Replacement, &rdev->flags);
2750 sysfs_notify_dirent_safe(rdev->sysfs_state);
2751 return err ? err : len;
2753 static struct rdev_sysfs_entry rdev_state =
2754 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2757 errors_show(struct md_rdev *rdev, char *page)
2759 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2763 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2766 unsigned long n = simple_strtoul(buf, &e, 10);
2767 if (*buf && (*e == 0 || *e == '\n')) {
2768 atomic_set(&rdev->corrected_errors, n);
2773 static struct rdev_sysfs_entry rdev_errors =
2774 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2777 slot_show(struct md_rdev *rdev, char *page)
2779 if (rdev->raid_disk < 0)
2780 return sprintf(page, "none\n");
2782 return sprintf(page, "%d\n", rdev->raid_disk);
2786 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2790 int slot = simple_strtoul(buf, &e, 10);
2791 if (strncmp(buf, "none", 4)==0)
2793 else if (e==buf || (*e && *e!= '\n'))
2795 if (rdev->mddev->pers && slot == -1) {
2796 /* Setting 'slot' on an active array requires also
2797 * updating the 'rd%d' link, and communicating
2798 * with the personality with ->hot_*_disk.
2799 * For now we only support removing
2800 * failed/spare devices. This normally happens automatically,
2801 * but not when the metadata is externally managed.
2803 if (rdev->raid_disk == -1)
2805 /* personality does all needed checks */
2806 if (rdev->mddev->pers->hot_remove_disk == NULL)
2808 clear_bit(Blocked, &rdev->flags);
2809 remove_and_add_spares(rdev->mddev, rdev);
2810 if (rdev->raid_disk >= 0)
2812 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2813 md_wakeup_thread(rdev->mddev->thread);
2814 } else if (rdev->mddev->pers) {
2815 /* Activating a spare .. or possibly reactivating
2816 * if we ever get bitmaps working here.
2819 if (rdev->raid_disk != -1)
2822 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2825 if (rdev->mddev->pers->hot_add_disk == NULL)
2828 if (slot >= rdev->mddev->raid_disks &&
2829 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2832 rdev->raid_disk = slot;
2833 if (test_bit(In_sync, &rdev->flags))
2834 rdev->saved_raid_disk = slot;
2836 rdev->saved_raid_disk = -1;
2837 clear_bit(In_sync, &rdev->flags);
2838 clear_bit(Bitmap_sync, &rdev->flags);
2839 err = rdev->mddev->pers->
2840 hot_add_disk(rdev->mddev, rdev);
2842 rdev->raid_disk = -1;
2845 sysfs_notify_dirent_safe(rdev->sysfs_state);
2846 if (sysfs_link_rdev(rdev->mddev, rdev))
2847 /* failure here is OK */;
2848 /* don't wakeup anyone, leave that to userspace. */
2850 if (slot >= rdev->mddev->raid_disks &&
2851 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2853 rdev->raid_disk = slot;
2854 /* assume it is working */
2855 clear_bit(Faulty, &rdev->flags);
2856 clear_bit(WriteMostly, &rdev->flags);
2857 set_bit(In_sync, &rdev->flags);
2858 sysfs_notify_dirent_safe(rdev->sysfs_state);
2864 static struct rdev_sysfs_entry rdev_slot =
2865 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2868 offset_show(struct md_rdev *rdev, char *page)
2870 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2874 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2876 unsigned long long offset;
2877 if (strict_strtoull(buf, 10, &offset) < 0)
2879 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2881 if (rdev->sectors && rdev->mddev->external)
2882 /* Must set offset before size, so overlap checks
2885 rdev->data_offset = offset;
2886 rdev->new_data_offset = offset;
2890 static struct rdev_sysfs_entry rdev_offset =
2891 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2893 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2895 return sprintf(page, "%llu\n",
2896 (unsigned long long)rdev->new_data_offset);
2899 static ssize_t new_offset_store(struct md_rdev *rdev,
2900 const char *buf, size_t len)
2902 unsigned long long new_offset;
2903 struct mddev *mddev = rdev->mddev;
2905 if (strict_strtoull(buf, 10, &new_offset) < 0)
2908 if (mddev->sync_thread)
2910 if (new_offset == rdev->data_offset)
2911 /* reset is always permitted */
2913 else if (new_offset > rdev->data_offset) {
2914 /* must not push array size beyond rdev_sectors */
2915 if (new_offset - rdev->data_offset
2916 + mddev->dev_sectors > rdev->sectors)
2919 /* Metadata worries about other space details. */
2921 /* decreasing the offset is inconsistent with a backwards
2924 if (new_offset < rdev->data_offset &&
2925 mddev->reshape_backwards)
2927 /* Increasing offset is inconsistent with forwards
2928 * reshape. reshape_direction should be set to
2929 * 'backwards' first.
2931 if (new_offset > rdev->data_offset &&
2932 !mddev->reshape_backwards)
2935 if (mddev->pers && mddev->persistent &&
2936 !super_types[mddev->major_version]
2937 .allow_new_offset(rdev, new_offset))
2939 rdev->new_data_offset = new_offset;
2940 if (new_offset > rdev->data_offset)
2941 mddev->reshape_backwards = 1;
2942 else if (new_offset < rdev->data_offset)
2943 mddev->reshape_backwards = 0;
2947 static struct rdev_sysfs_entry rdev_new_offset =
2948 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2951 rdev_size_show(struct md_rdev *rdev, char *page)
2953 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2956 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2958 /* check if two start/length pairs overlap */
2966 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2968 unsigned long long blocks;
2971 if (strict_strtoull(buf, 10, &blocks) < 0)
2974 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2975 return -EINVAL; /* sector conversion overflow */
2978 if (new != blocks * 2)
2979 return -EINVAL; /* unsigned long long to sector_t overflow */
2986 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2988 struct mddev *my_mddev = rdev->mddev;
2989 sector_t oldsectors = rdev->sectors;
2992 if (strict_blocks_to_sectors(buf, §ors) < 0)
2994 if (rdev->data_offset != rdev->new_data_offset)
2995 return -EINVAL; /* too confusing */
2996 if (my_mddev->pers && rdev->raid_disk >= 0) {
2997 if (my_mddev->persistent) {
2998 sectors = super_types[my_mddev->major_version].
2999 rdev_size_change(rdev, sectors);
3002 } else if (!sectors)
3003 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3005 if (!my_mddev->pers->resize)
3006 /* Cannot change size for RAID0 or Linear etc */
3009 if (sectors < my_mddev->dev_sectors)
3010 return -EINVAL; /* component must fit device */
3012 rdev->sectors = sectors;
3013 if (sectors > oldsectors && my_mddev->external) {
3014 /* need to check that all other rdevs with the same ->bdev
3015 * do not overlap. We need to unlock the mddev to avoid
3016 * a deadlock. We have already changed rdev->sectors, and if
3017 * we have to change it back, we will have the lock again.
3019 struct mddev *mddev;
3021 struct list_head *tmp;
3023 mddev_unlock(my_mddev);
3024 for_each_mddev(mddev, tmp) {
3025 struct md_rdev *rdev2;
3028 rdev_for_each(rdev2, mddev)
3029 if (rdev->bdev == rdev2->bdev &&
3031 overlaps(rdev->data_offset, rdev->sectors,
3037 mddev_unlock(mddev);
3043 mddev_lock(my_mddev);
3045 /* Someone else could have slipped in a size
3046 * change here, but doing so is just silly.
3047 * We put oldsectors back because we *know* it is
3048 * safe, and trust userspace not to race with
3051 rdev->sectors = oldsectors;
3058 static struct rdev_sysfs_entry rdev_size =
3059 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3062 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3064 unsigned long long recovery_start = rdev->recovery_offset;
3066 if (test_bit(In_sync, &rdev->flags) ||
3067 recovery_start == MaxSector)
3068 return sprintf(page, "none\n");
3070 return sprintf(page, "%llu\n", recovery_start);
3073 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3075 unsigned long long recovery_start;
3077 if (cmd_match(buf, "none"))
3078 recovery_start = MaxSector;
3079 else if (strict_strtoull(buf, 10, &recovery_start))
3082 if (rdev->mddev->pers &&
3083 rdev->raid_disk >= 0)
3086 rdev->recovery_offset = recovery_start;
3087 if (recovery_start == MaxSector)
3088 set_bit(In_sync, &rdev->flags);
3090 clear_bit(In_sync, &rdev->flags);
3094 static struct rdev_sysfs_entry rdev_recovery_start =
3095 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3099 badblocks_show(struct badblocks *bb, char *page, int unack);
3101 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3103 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3105 return badblocks_show(&rdev->badblocks, page, 0);
3107 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3109 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3110 /* Maybe that ack was all we needed */
3111 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3112 wake_up(&rdev->blocked_wait);
3115 static struct rdev_sysfs_entry rdev_bad_blocks =
3116 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3119 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3121 return badblocks_show(&rdev->badblocks, page, 1);
3123 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3125 return badblocks_store(&rdev->badblocks, page, len, 1);
3127 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3128 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3130 static struct attribute *rdev_default_attrs[] = {
3135 &rdev_new_offset.attr,
3137 &rdev_recovery_start.attr,
3138 &rdev_bad_blocks.attr,
3139 &rdev_unack_bad_blocks.attr,
3143 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3145 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3146 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3147 struct mddev *mddev = rdev->mddev;
3153 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3155 if (rdev->mddev == NULL)
3158 rv = entry->show(rdev, page);
3159 mddev_unlock(mddev);
3165 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3166 const char *page, size_t length)
3168 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3169 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3171 struct mddev *mddev = rdev->mddev;
3175 if (!capable(CAP_SYS_ADMIN))
3177 rv = mddev ? mddev_lock(mddev): -EBUSY;
3179 if (rdev->mddev == NULL)
3182 rv = entry->store(rdev, page, length);
3183 mddev_unlock(mddev);
3188 static void rdev_free(struct kobject *ko)
3190 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3193 static const struct sysfs_ops rdev_sysfs_ops = {
3194 .show = rdev_attr_show,
3195 .store = rdev_attr_store,
3197 static struct kobj_type rdev_ktype = {
3198 .release = rdev_free,
3199 .sysfs_ops = &rdev_sysfs_ops,
3200 .default_attrs = rdev_default_attrs,
3203 int md_rdev_init(struct md_rdev *rdev)
3206 rdev->saved_raid_disk = -1;
3207 rdev->raid_disk = -1;
3209 rdev->data_offset = 0;
3210 rdev->new_data_offset = 0;
3211 rdev->sb_events = 0;
3212 rdev->last_read_error.tv_sec = 0;
3213 rdev->last_read_error.tv_nsec = 0;
3214 rdev->sb_loaded = 0;
3215 rdev->bb_page = NULL;
3216 atomic_set(&rdev->nr_pending, 0);
3217 atomic_set(&rdev->read_errors, 0);
3218 atomic_set(&rdev->corrected_errors, 0);
3220 INIT_LIST_HEAD(&rdev->same_set);
3221 init_waitqueue_head(&rdev->blocked_wait);
3223 /* Add space to store bad block list.
3224 * This reserves the space even on arrays where it cannot
3225 * be used - I wonder if that matters
3227 rdev->badblocks.count = 0;
3228 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3229 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3230 seqlock_init(&rdev->badblocks.lock);
3231 if (rdev->badblocks.page == NULL)
3236 EXPORT_SYMBOL_GPL(md_rdev_init);
3238 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3240 * mark the device faulty if:
3242 * - the device is nonexistent (zero size)
3243 * - the device has no valid superblock
3245 * a faulty rdev _never_ has rdev->sb set.
3247 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3249 char b[BDEVNAME_SIZE];
3251 struct md_rdev *rdev;
3254 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3256 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3257 return ERR_PTR(-ENOMEM);
3260 err = md_rdev_init(rdev);
3263 err = alloc_disk_sb(rdev);
3267 err = lock_rdev(rdev, newdev, super_format == -2);
3271 kobject_init(&rdev->kobj, &rdev_ktype);
3273 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3276 "md: %s has zero or unknown size, marking faulty!\n",
3277 bdevname(rdev->bdev,b));
3282 if (super_format >= 0) {
3283 err = super_types[super_format].
3284 load_super(rdev, NULL, super_minor);
3285 if (err == -EINVAL) {
3287 "md: %s does not have a valid v%d.%d "
3288 "superblock, not importing!\n",
3289 bdevname(rdev->bdev,b),
3290 super_format, super_minor);
3295 "md: could not read %s's sb, not importing!\n",
3296 bdevname(rdev->bdev,b));
3306 md_rdev_clear(rdev);
3308 return ERR_PTR(err);
3312 * Check a full RAID array for plausibility
3316 static void analyze_sbs(struct mddev * mddev)
3319 struct md_rdev *rdev, *freshest, *tmp;
3320 char b[BDEVNAME_SIZE];
3323 rdev_for_each_safe(rdev, tmp, mddev)
3324 switch (super_types[mddev->major_version].
3325 load_super(rdev, freshest, mddev->minor_version)) {
3333 "md: fatal superblock inconsistency in %s"
3334 " -- removing from array\n",
3335 bdevname(rdev->bdev,b));
3336 kick_rdev_from_array(rdev);
3340 super_types[mddev->major_version].
3341 validate_super(mddev, freshest);
3344 rdev_for_each_safe(rdev, tmp, mddev) {
3345 if (mddev->max_disks &&
3346 (rdev->desc_nr >= mddev->max_disks ||
3347 i > mddev->max_disks)) {
3349 "md: %s: %s: only %d devices permitted\n",
3350 mdname(mddev), bdevname(rdev->bdev, b),
3352 kick_rdev_from_array(rdev);
3355 if (rdev != freshest)
3356 if (super_types[mddev->major_version].
3357 validate_super(mddev, rdev)) {
3358 printk(KERN_WARNING "md: kicking non-fresh %s"
3360 bdevname(rdev->bdev,b));
3361 kick_rdev_from_array(rdev);
3364 if (mddev->level == LEVEL_MULTIPATH) {
3365 rdev->desc_nr = i++;
3366 rdev->raid_disk = rdev->desc_nr;
3367 set_bit(In_sync, &rdev->flags);
3368 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3369 rdev->raid_disk = -1;
3370 clear_bit(In_sync, &rdev->flags);
3375 /* Read a fixed-point number.
3376 * Numbers in sysfs attributes should be in "standard" units where
3377 * possible, so time should be in seconds.
3378 * However we internally use a a much smaller unit such as
3379 * milliseconds or jiffies.
3380 * This function takes a decimal number with a possible fractional
3381 * component, and produces an integer which is the result of
3382 * multiplying that number by 10^'scale'.
3383 * all without any floating-point arithmetic.
3385 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3387 unsigned long result = 0;
3389 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3392 else if (decimals < scale) {
3395 result = result * 10 + value;
3407 while (decimals < scale) {
3416 static void md_safemode_timeout(unsigned long data);
3419 safe_delay_show(struct mddev *mddev, char *page)
3421 int msec = (mddev->safemode_delay*1000)/HZ;
3422 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3425 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3429 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3432 mddev->safemode_delay = 0;
3434 unsigned long old_delay = mddev->safemode_delay;
3435 mddev->safemode_delay = (msec*HZ)/1000;
3436 if (mddev->safemode_delay == 0)
3437 mddev->safemode_delay = 1;
3438 if (mddev->safemode_delay < old_delay)
3439 md_safemode_timeout((unsigned long)mddev);
3443 static struct md_sysfs_entry md_safe_delay =
3444 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3447 level_show(struct mddev *mddev, char *page)
3449 struct md_personality *p = mddev->pers;
3451 return sprintf(page, "%s\n", p->name);
3452 else if (mddev->clevel[0])
3453 return sprintf(page, "%s\n", mddev->clevel);
3454 else if (mddev->level != LEVEL_NONE)
3455 return sprintf(page, "%d\n", mddev->level);
3461 level_store(struct mddev *mddev, const char *buf, size_t len)
3465 struct md_personality *pers;
3468 struct md_rdev *rdev;
3470 if (mddev->pers == NULL) {
3473 if (len >= sizeof(mddev->clevel))
3475 strncpy(mddev->clevel, buf, len);
3476 if (mddev->clevel[len-1] == '\n')
3478 mddev->clevel[len] = 0;
3479 mddev->level = LEVEL_NONE;
3483 /* request to change the personality. Need to ensure:
3484 * - array is not engaged in resync/recovery/reshape
3485 * - old personality can be suspended
3486 * - new personality will access other array.
3489 if (mddev->sync_thread ||
3490 mddev->reshape_position != MaxSector ||
3491 mddev->sysfs_active)
3494 if (!mddev->pers->quiesce) {
3495 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3496 mdname(mddev), mddev->pers->name);
3500 /* Now find the new personality */
3501 if (len == 0 || len >= sizeof(clevel))
3503 strncpy(clevel, buf, len);
3504 if (clevel[len-1] == '\n')
3507 if (strict_strtol(clevel, 10, &level))
3510 if (request_module("md-%s", clevel) != 0)
3511 request_module("md-level-%s", clevel);
3512 spin_lock(&pers_lock);
3513 pers = find_pers(level, clevel);
3514 if (!pers || !try_module_get(pers->owner)) {
3515 spin_unlock(&pers_lock);
3516 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3519 spin_unlock(&pers_lock);
3521 if (pers == mddev->pers) {
3522 /* Nothing to do! */
3523 module_put(pers->owner);
3526 if (!pers->takeover) {
3527 module_put(pers->owner);
3528 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3529 mdname(mddev), clevel);
3533 rdev_for_each(rdev, mddev)
3534 rdev->new_raid_disk = rdev->raid_disk;
3536 /* ->takeover must set new_* and/or delta_disks
3537 * if it succeeds, and may set them when it fails.
3539 priv = pers->takeover(mddev);
3541 mddev->new_level = mddev->level;
3542 mddev->new_layout = mddev->layout;
3543 mddev->new_chunk_sectors = mddev->chunk_sectors;
3544 mddev->raid_disks -= mddev->delta_disks;
3545 mddev->delta_disks = 0;
3546 mddev->reshape_backwards = 0;
3547 module_put(pers->owner);
3548 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3549 mdname(mddev), clevel);
3550 return PTR_ERR(priv);
3553 /* Looks like we have a winner */
3554 mddev_suspend(mddev);
3555 mddev->pers->stop(mddev);
3557 if (mddev->pers->sync_request == NULL &&
3558 pers->sync_request != NULL) {
3559 /* need to add the md_redundancy_group */
3560 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3562 "md: cannot register extra attributes for %s\n",
3564 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3566 if (mddev->pers->sync_request != NULL &&
3567 pers->sync_request == NULL) {
3568 /* need to remove the md_redundancy_group */
3569 if (mddev->to_remove == NULL)
3570 mddev->to_remove = &md_redundancy_group;
3573 if (mddev->pers->sync_request == NULL &&
3575 /* We are converting from a no-redundancy array
3576 * to a redundancy array and metadata is managed
3577 * externally so we need to be sure that writes
3578 * won't block due to a need to transition
3580 * until external management is started.
3583 mddev->safemode_delay = 0;
3584 mddev->safemode = 0;
3587 rdev_for_each(rdev, mddev) {
3588 if (rdev->raid_disk < 0)
3590 if (rdev->new_raid_disk >= mddev->raid_disks)
3591 rdev->new_raid_disk = -1;
3592 if (rdev->new_raid_disk == rdev->raid_disk)
3594 sysfs_unlink_rdev(mddev, rdev);
3596 rdev_for_each(rdev, mddev) {
3597 if (rdev->raid_disk < 0)
3599 if (rdev->new_raid_disk == rdev->raid_disk)
3601 rdev->raid_disk = rdev->new_raid_disk;
3602 if (rdev->raid_disk < 0)
3603 clear_bit(In_sync, &rdev->flags);
3605 if (sysfs_link_rdev(mddev, rdev))
3606 printk(KERN_WARNING "md: cannot register rd%d"
3607 " for %s after level change\n",
3608 rdev->raid_disk, mdname(mddev));
3612 module_put(mddev->pers->owner);
3614 mddev->private = priv;
3615 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3616 mddev->level = mddev->new_level;
3617 mddev->layout = mddev->new_layout;
3618 mddev->chunk_sectors = mddev->new_chunk_sectors;
3619 mddev->delta_disks = 0;
3620 mddev->reshape_backwards = 0;
3621 mddev->degraded = 0;
3622 if (mddev->pers->sync_request == NULL) {
3623 /* this is now an array without redundancy, so
3624 * it must always be in_sync
3627 del_timer_sync(&mddev->safemode_timer);
3629 blk_set_stacking_limits(&mddev->queue->limits);
3631 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3632 mddev_resume(mddev);
3633 sysfs_notify(&mddev->kobj, NULL, "level");
3634 md_new_event(mddev);
3638 static struct md_sysfs_entry md_level =
3639 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3643 layout_show(struct mddev *mddev, char *page)
3645 /* just a number, not meaningful for all levels */
3646 if (mddev->reshape_position != MaxSector &&
3647 mddev->layout != mddev->new_layout)
3648 return sprintf(page, "%d (%d)\n",
3649 mddev->new_layout, mddev->layout);
3650 return sprintf(page, "%d\n", mddev->layout);
3654 layout_store(struct mddev *mddev, const char *buf, size_t len)
3657 unsigned long n = simple_strtoul(buf, &e, 10);
3659 if (!*buf || (*e && *e != '\n'))
3664 if (mddev->pers->check_reshape == NULL)
3666 mddev->new_layout = n;
3667 err = mddev->pers->check_reshape(mddev);
3669 mddev->new_layout = mddev->layout;
3673 mddev->new_layout = n;
3674 if (mddev->reshape_position == MaxSector)
3679 static struct md_sysfs_entry md_layout =
3680 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3684 raid_disks_show(struct mddev *mddev, char *page)
3686 if (mddev->raid_disks == 0)
3688 if (mddev->reshape_position != MaxSector &&
3689 mddev->delta_disks != 0)
3690 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3691 mddev->raid_disks - mddev->delta_disks);
3692 return sprintf(page, "%d\n", mddev->raid_disks);
3695 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3698 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3702 unsigned long n = simple_strtoul(buf, &e, 10);
3704 if (!*buf || (*e && *e != '\n'))
3708 rv = update_raid_disks(mddev, n);
3709 else if (mddev->reshape_position != MaxSector) {
3710 struct md_rdev *rdev;
3711 int olddisks = mddev->raid_disks - mddev->delta_disks;
3713 rdev_for_each(rdev, mddev) {
3715 rdev->data_offset < rdev->new_data_offset)
3718 rdev->data_offset > rdev->new_data_offset)
3721 mddev->delta_disks = n - olddisks;
3722 mddev->raid_disks = n;
3723 mddev->reshape_backwards = (mddev->delta_disks < 0);
3725 mddev->raid_disks = n;
3726 return rv ? rv : len;
3728 static struct md_sysfs_entry md_raid_disks =
3729 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3732 chunk_size_show(struct mddev *mddev, char *page)
3734 if (mddev->reshape_position != MaxSector &&
3735 mddev->chunk_sectors != mddev->new_chunk_sectors)
3736 return sprintf(page, "%d (%d)\n",
3737 mddev->new_chunk_sectors << 9,
3738 mddev->chunk_sectors << 9);
3739 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3743 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3746 unsigned long n = simple_strtoul(buf, &e, 10);
3748 if (!*buf || (*e && *e != '\n'))
3753 if (mddev->pers->check_reshape == NULL)
3755 mddev->new_chunk_sectors = n >> 9;
3756 err = mddev->pers->check_reshape(mddev);
3758 mddev->new_chunk_sectors = mddev->chunk_sectors;
3762 mddev->new_chunk_sectors = n >> 9;
3763 if (mddev->reshape_position == MaxSector)
3764 mddev->chunk_sectors = n >> 9;
3768 static struct md_sysfs_entry md_chunk_size =
3769 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3772 resync_start_show(struct mddev *mddev, char *page)
3774 if (mddev->recovery_cp == MaxSector)
3775 return sprintf(page, "none\n");
3776 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3780 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3783 unsigned long long n = simple_strtoull(buf, &e, 10);
3785 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3787 if (cmd_match(buf, "none"))
3789 else if (!*buf || (*e && *e != '\n'))
3792 mddev->recovery_cp = n;
3794 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3797 static struct md_sysfs_entry md_resync_start =
3798 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3801 * The array state can be:
3804 * No devices, no size, no level
3805 * Equivalent to STOP_ARRAY ioctl
3807 * May have some settings, but array is not active
3808 * all IO results in error
3809 * When written, doesn't tear down array, but just stops it
3810 * suspended (not supported yet)
3811 * All IO requests will block. The array can be reconfigured.
3812 * Writing this, if accepted, will block until array is quiescent
3814 * no resync can happen. no superblocks get written.
3815 * write requests fail
3817 * like readonly, but behaves like 'clean' on a write request.
3819 * clean - no pending writes, but otherwise active.
3820 * When written to inactive array, starts without resync
3821 * If a write request arrives then
3822 * if metadata is known, mark 'dirty' and switch to 'active'.
3823 * if not known, block and switch to write-pending
3824 * If written to an active array that has pending writes, then fails.
3826 * fully active: IO and resync can be happening.
3827 * When written to inactive array, starts with resync
3830 * clean, but writes are blocked waiting for 'active' to be written.
3833 * like active, but no writes have been seen for a while (100msec).
3836 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3837 write_pending, active_idle, bad_word};
3838 static char *array_states[] = {
3839 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3840 "write-pending", "active-idle", NULL };
3842 static int match_word(const char *word, char **list)
3845 for (n=0; list[n]; n++)
3846 if (cmd_match(word, list[n]))
3852 array_state_show(struct mddev *mddev, char *page)
3854 enum array_state st = inactive;
3867 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3869 else if (mddev->safemode)
3875 if (list_empty(&mddev->disks) &&
3876 mddev->raid_disks == 0 &&
3877 mddev->dev_sectors == 0)
3882 return sprintf(page, "%s\n", array_states[st]);
3885 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3886 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3887 static int do_md_run(struct mddev * mddev);
3888 static int restart_array(struct mddev *mddev);
3891 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3894 enum array_state st = match_word(buf, array_states);
3899 /* stopping an active array */
3900 err = do_md_stop(mddev, 0, NULL);
3903 /* stopping an active array */
3905 err = do_md_stop(mddev, 2, NULL);
3907 err = 0; /* already inactive */
3910 break; /* not supported yet */
3913 err = md_set_readonly(mddev, NULL);
3916 set_disk_ro(mddev->gendisk, 1);
3917 err = do_md_run(mddev);
3923 err = md_set_readonly(mddev, NULL);
3924 else if (mddev->ro == 1)
3925 err = restart_array(mddev);
3928 set_disk_ro(mddev->gendisk, 0);
3932 err = do_md_run(mddev);
3937 restart_array(mddev);
3938 spin_lock_irq(&mddev->write_lock);
3939 if (atomic_read(&mddev->writes_pending) == 0) {
3940 if (mddev->in_sync == 0) {
3942 if (mddev->safemode == 1)
3943 mddev->safemode = 0;
3944 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3949 spin_unlock_irq(&mddev->write_lock);
3955 restart_array(mddev);
3956 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3957 wake_up(&mddev->sb_wait);
3961 set_disk_ro(mddev->gendisk, 0);
3962 err = do_md_run(mddev);
3967 /* these cannot be set */
3973 if (mddev->hold_active == UNTIL_IOCTL)
3974 mddev->hold_active = 0;
3975 sysfs_notify_dirent_safe(mddev->sysfs_state);
3979 static struct md_sysfs_entry md_array_state =
3980 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3983 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3984 return sprintf(page, "%d\n",
3985 atomic_read(&mddev->max_corr_read_errors));
3989 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3992 unsigned long n = simple_strtoul(buf, &e, 10);
3994 if (*buf && (*e == 0 || *e == '\n')) {
3995 atomic_set(&mddev->max_corr_read_errors, n);
4001 static struct md_sysfs_entry max_corr_read_errors =
4002 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4003 max_corrected_read_errors_store);
4006 null_show(struct mddev *mddev, char *page)
4012 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4014 /* buf must be %d:%d\n? giving major and minor numbers */
4015 /* The new device is added to the array.
4016 * If the array has a persistent superblock, we read the
4017 * superblock to initialise info and check validity.
4018 * Otherwise, only checking done is that in bind_rdev_to_array,
4019 * which mainly checks size.
4022 int major = simple_strtoul(buf, &e, 10);
4025 struct md_rdev *rdev;
4028 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4030 minor = simple_strtoul(e+1, &e, 10);
4031 if (*e && *e != '\n')
4033 dev = MKDEV(major, minor);
4034 if (major != MAJOR(dev) ||
4035 minor != MINOR(dev))
4039 if (mddev->persistent) {
4040 rdev = md_import_device(dev, mddev->major_version,
4041 mddev->minor_version);
4042 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4043 struct md_rdev *rdev0
4044 = list_entry(mddev->disks.next,
4045 struct md_rdev, same_set);
4046 err = super_types[mddev->major_version]
4047 .load_super(rdev, rdev0, mddev->minor_version);
4051 } else if (mddev->external)
4052 rdev = md_import_device(dev, -2, -1);
4054 rdev = md_import_device(dev, -1, -1);
4057 return PTR_ERR(rdev);
4058 err = bind_rdev_to_array(rdev, mddev);
4062 return err ? err : len;
4065 static struct md_sysfs_entry md_new_device =
4066 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4069 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4072 unsigned long chunk, end_chunk;
4076 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4078 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4079 if (buf == end) break;
4080 if (*end == '-') { /* range */
4082 end_chunk = simple_strtoul(buf, &end, 0);
4083 if (buf == end) break;
4085 if (*end && !isspace(*end)) break;
4086 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4087 buf = skip_spaces(end);
4089 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4094 static struct md_sysfs_entry md_bitmap =
4095 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4098 size_show(struct mddev *mddev, char *page)
4100 return sprintf(page, "%llu\n",
4101 (unsigned long long)mddev->dev_sectors / 2);
4104 static int update_size(struct mddev *mddev, sector_t num_sectors);
4107 size_store(struct mddev *mddev, const char *buf, size_t len)
4109 /* If array is inactive, we can reduce the component size, but
4110 * not increase it (except from 0).
4111 * If array is active, we can try an on-line resize
4114 int err = strict_blocks_to_sectors(buf, §ors);
4119 err = update_size(mddev, sectors);
4120 md_update_sb(mddev, 1);
4122 if (mddev->dev_sectors == 0 ||
4123 mddev->dev_sectors > sectors)
4124 mddev->dev_sectors = sectors;
4128 return err ? err : len;
4131 static struct md_sysfs_entry md_size =
4132 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4135 /* Metadata version.
4137 * 'none' for arrays with no metadata (good luck...)
4138 * 'external' for arrays with externally managed metadata,
4139 * or N.M for internally known formats
4142 metadata_show(struct mddev *mddev, char *page)
4144 if (mddev->persistent)
4145 return sprintf(page, "%d.%d\n",
4146 mddev->major_version, mddev->minor_version);
4147 else if (mddev->external)
4148 return sprintf(page, "external:%s\n", mddev->metadata_type);
4150 return sprintf(page, "none\n");
4154 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4158 /* Changing the details of 'external' metadata is
4159 * always permitted. Otherwise there must be
4160 * no devices attached to the array.
4162 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4164 else if (!list_empty(&mddev->disks))
4167 if (cmd_match(buf, "none")) {
4168 mddev->persistent = 0;
4169 mddev->external = 0;
4170 mddev->major_version = 0;
4171 mddev->minor_version = 90;
4174 if (strncmp(buf, "external:", 9) == 0) {
4175 size_t namelen = len-9;
4176 if (namelen >= sizeof(mddev->metadata_type))
4177 namelen = sizeof(mddev->metadata_type)-1;
4178 strncpy(mddev->metadata_type, buf+9, namelen);
4179 mddev->metadata_type[namelen] = 0;
4180 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4181 mddev->metadata_type[--namelen] = 0;
4182 mddev->persistent = 0;
4183 mddev->external = 1;
4184 mddev->major_version = 0;
4185 mddev->minor_version = 90;
4188 major = simple_strtoul(buf, &e, 10);
4189 if (e==buf || *e != '.')
4192 minor = simple_strtoul(buf, &e, 10);
4193 if (e==buf || (*e && *e != '\n') )
4195 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4197 mddev->major_version = major;
4198 mddev->minor_version = minor;
4199 mddev->persistent = 1;
4200 mddev->external = 0;
4204 static struct md_sysfs_entry md_metadata =
4205 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4208 action_show(struct mddev *mddev, char *page)
4210 char *type = "idle";
4211 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4213 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4214 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4215 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4217 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4218 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4220 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4224 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4227 return sprintf(page, "%s\n", type);
4231 action_store(struct mddev *mddev, const char *page, size_t len)
4233 if (!mddev->pers || !mddev->pers->sync_request)
4236 if (cmd_match(page, "frozen"))
4237 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4239 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4241 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4242 if (mddev->sync_thread) {
4243 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4244 md_reap_sync_thread(mddev);
4246 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4247 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4249 else if (cmd_match(page, "resync"))
4250 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4251 else if (cmd_match(page, "recover")) {
4252 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4253 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4254 } else if (cmd_match(page, "reshape")) {
4256 if (mddev->pers->start_reshape == NULL)
4258 err = mddev->pers->start_reshape(mddev);
4261 sysfs_notify(&mddev->kobj, NULL, "degraded");
4263 if (cmd_match(page, "check"))
4264 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4265 else if (!cmd_match(page, "repair"))
4267 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4268 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4270 if (mddev->ro == 2) {
4271 /* A write to sync_action is enough to justify
4272 * canceling read-auto mode
4275 md_wakeup_thread(mddev->sync_thread);
4277 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4278 md_wakeup_thread(mddev->thread);
4279 sysfs_notify_dirent_safe(mddev->sysfs_action);
4284 mismatch_cnt_show(struct mddev *mddev, char *page)
4286 return sprintf(page, "%llu\n",
4287 (unsigned long long)
4288 atomic64_read(&mddev->resync_mismatches));
4291 static struct md_sysfs_entry md_scan_mode =
4292 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4295 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4298 sync_min_show(struct mddev *mddev, char *page)
4300 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4301 mddev->sync_speed_min ? "local": "system");
4305 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4309 if (strncmp(buf, "system", 6)==0) {
4310 mddev->sync_speed_min = 0;
4313 min = simple_strtoul(buf, &e, 10);
4314 if (buf == e || (*e && *e != '\n') || min <= 0)
4316 mddev->sync_speed_min = min;
4320 static struct md_sysfs_entry md_sync_min =
4321 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4324 sync_max_show(struct mddev *mddev, char *page)
4326 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4327 mddev->sync_speed_max ? "local": "system");
4331 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4335 if (strncmp(buf, "system", 6)==0) {
4336 mddev->sync_speed_max = 0;
4339 max = simple_strtoul(buf, &e, 10);
4340 if (buf == e || (*e && *e != '\n') || max <= 0)
4342 mddev->sync_speed_max = max;
4346 static struct md_sysfs_entry md_sync_max =
4347 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4350 degraded_show(struct mddev *mddev, char *page)
4352 return sprintf(page, "%d\n", mddev->degraded);
4354 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4357 sync_force_parallel_show(struct mddev *mddev, char *page)
4359 return sprintf(page, "%d\n", mddev->parallel_resync);
4363 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4367 if (strict_strtol(buf, 10, &n))
4370 if (n != 0 && n != 1)
4373 mddev->parallel_resync = n;
4375 if (mddev->sync_thread)
4376 wake_up(&resync_wait);
4381 /* force parallel resync, even with shared block devices */
4382 static struct md_sysfs_entry md_sync_force_parallel =
4383 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4384 sync_force_parallel_show, sync_force_parallel_store);
4387 sync_speed_show(struct mddev *mddev, char *page)
4389 unsigned long resync, dt, db;
4390 if (mddev->curr_resync == 0)
4391 return sprintf(page, "none\n");
4392 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4393 dt = (jiffies - mddev->resync_mark) / HZ;
4395 db = resync - mddev->resync_mark_cnt;
4396 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4399 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4402 sync_completed_show(struct mddev *mddev, char *page)
4404 unsigned long long max_sectors, resync;
4406 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4407 return sprintf(page, "none\n");
4409 if (mddev->curr_resync == 1 ||
4410 mddev->curr_resync == 2)
4411 return sprintf(page, "delayed\n");
4413 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4414 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4415 max_sectors = mddev->resync_max_sectors;
4417 max_sectors = mddev->dev_sectors;
4419 resync = mddev->curr_resync_completed;
4420 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4423 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4426 min_sync_show(struct mddev *mddev, char *page)
4428 return sprintf(page, "%llu\n",
4429 (unsigned long long)mddev->resync_min);
4432 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4434 unsigned long long min;
4435 if (strict_strtoull(buf, 10, &min))
4437 if (min > mddev->resync_max)
4439 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4442 /* Must be a multiple of chunk_size */
4443 if (mddev->chunk_sectors) {
4444 sector_t temp = min;
4445 if (sector_div(temp, mddev->chunk_sectors))
4448 mddev->resync_min = min;
4453 static struct md_sysfs_entry md_min_sync =
4454 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4457 max_sync_show(struct mddev *mddev, char *page)
4459 if (mddev->resync_max == MaxSector)
4460 return sprintf(page, "max\n");
4462 return sprintf(page, "%llu\n",
4463 (unsigned long long)mddev->resync_max);
4466 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4468 if (strncmp(buf, "max", 3) == 0)
4469 mddev->resync_max = MaxSector;
4471 unsigned long long max;
4472 if (strict_strtoull(buf, 10, &max))
4474 if (max < mddev->resync_min)
4476 if (max < mddev->resync_max &&
4478 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4481 /* Must be a multiple of chunk_size */
4482 if (mddev->chunk_sectors) {
4483 sector_t temp = max;
4484 if (sector_div(temp, mddev->chunk_sectors))
4487 mddev->resync_max = max;
4489 wake_up(&mddev->recovery_wait);
4493 static struct md_sysfs_entry md_max_sync =
4494 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4497 suspend_lo_show(struct mddev *mddev, char *page)
4499 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4503 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4506 unsigned long long new = simple_strtoull(buf, &e, 10);
4507 unsigned long long old = mddev->suspend_lo;
4509 if (mddev->pers == NULL ||
4510 mddev->pers->quiesce == NULL)
4512 if (buf == e || (*e && *e != '\n'))
4515 mddev->suspend_lo = new;
4517 /* Shrinking suspended region */
4518 mddev->pers->quiesce(mddev, 2);
4520 /* Expanding suspended region - need to wait */
4521 mddev->pers->quiesce(mddev, 1);
4522 mddev->pers->quiesce(mddev, 0);
4526 static struct md_sysfs_entry md_suspend_lo =
4527 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4531 suspend_hi_show(struct mddev *mddev, char *page)
4533 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4537 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4540 unsigned long long new = simple_strtoull(buf, &e, 10);
4541 unsigned long long old = mddev->suspend_hi;
4543 if (mddev->pers == NULL ||
4544 mddev->pers->quiesce == NULL)
4546 if (buf == e || (*e && *e != '\n'))
4549 mddev->suspend_hi = new;
4551 /* Shrinking suspended region */
4552 mddev->pers->quiesce(mddev, 2);
4554 /* Expanding suspended region - need to wait */
4555 mddev->pers->quiesce(mddev, 1);
4556 mddev->pers->quiesce(mddev, 0);
4560 static struct md_sysfs_entry md_suspend_hi =
4561 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4564 reshape_position_show(struct mddev *mddev, char *page)
4566 if (mddev->reshape_position != MaxSector)
4567 return sprintf(page, "%llu\n",
4568 (unsigned long long)mddev->reshape_position);
4569 strcpy(page, "none\n");
4574 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4576 struct md_rdev *rdev;
4578 unsigned long long new = simple_strtoull(buf, &e, 10);
4581 if (buf == e || (*e && *e != '\n'))
4583 mddev->reshape_position = new;
4584 mddev->delta_disks = 0;
4585 mddev->reshape_backwards = 0;
4586 mddev->new_level = mddev->level;
4587 mddev->new_layout = mddev->layout;
4588 mddev->new_chunk_sectors = mddev->chunk_sectors;
4589 rdev_for_each(rdev, mddev)
4590 rdev->new_data_offset = rdev->data_offset;
4594 static struct md_sysfs_entry md_reshape_position =
4595 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4596 reshape_position_store);
4599 reshape_direction_show(struct mddev *mddev, char *page)
4601 return sprintf(page, "%s\n",
4602 mddev->reshape_backwards ? "backwards" : "forwards");
4606 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4609 if (cmd_match(buf, "forwards"))
4611 else if (cmd_match(buf, "backwards"))
4615 if (mddev->reshape_backwards == backwards)
4618 /* check if we are allowed to change */
4619 if (mddev->delta_disks)
4622 if (mddev->persistent &&
4623 mddev->major_version == 0)
4626 mddev->reshape_backwards = backwards;
4630 static struct md_sysfs_entry md_reshape_direction =
4631 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4632 reshape_direction_store);
4635 array_size_show(struct mddev *mddev, char *page)
4637 if (mddev->external_size)
4638 return sprintf(page, "%llu\n",
4639 (unsigned long long)mddev->array_sectors/2);
4641 return sprintf(page, "default\n");
4645 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4649 if (strncmp(buf, "default", 7) == 0) {
4651 sectors = mddev->pers->size(mddev, 0, 0);
4653 sectors = mddev->array_sectors;
4655 mddev->external_size = 0;
4657 if (strict_blocks_to_sectors(buf, §ors) < 0)
4659 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4662 mddev->external_size = 1;
4665 mddev->array_sectors = sectors;
4667 set_capacity(mddev->gendisk, mddev->array_sectors);
4668 revalidate_disk(mddev->gendisk);
4673 static struct md_sysfs_entry md_array_size =
4674 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4677 static struct attribute *md_default_attrs[] = {
4680 &md_raid_disks.attr,
4681 &md_chunk_size.attr,
4683 &md_resync_start.attr,
4685 &md_new_device.attr,
4686 &md_safe_delay.attr,
4687 &md_array_state.attr,
4688 &md_reshape_position.attr,
4689 &md_reshape_direction.attr,
4690 &md_array_size.attr,
4691 &max_corr_read_errors.attr,
4695 static struct attribute *md_redundancy_attrs[] = {
4697 &md_mismatches.attr,
4700 &md_sync_speed.attr,
4701 &md_sync_force_parallel.attr,
4702 &md_sync_completed.attr,
4705 &md_suspend_lo.attr,
4706 &md_suspend_hi.attr,
4711 static struct attribute_group md_redundancy_group = {
4713 .attrs = md_redundancy_attrs,
4718 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4720 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4721 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4726 spin_lock(&all_mddevs_lock);
4727 if (list_empty(&mddev->all_mddevs)) {
4728 spin_unlock(&all_mddevs_lock);
4732 spin_unlock(&all_mddevs_lock);
4734 rv = mddev_lock(mddev);
4736 rv = entry->show(mddev, page);
4737 mddev_unlock(mddev);
4744 md_attr_store(struct kobject *kobj, struct attribute *attr,
4745 const char *page, size_t length)
4747 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4748 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4753 if (!capable(CAP_SYS_ADMIN))
4755 spin_lock(&all_mddevs_lock);
4756 if (list_empty(&mddev->all_mddevs)) {
4757 spin_unlock(&all_mddevs_lock);
4761 spin_unlock(&all_mddevs_lock);
4762 if (entry->store == new_dev_store)
4763 flush_workqueue(md_misc_wq);
4764 rv = mddev_lock(mddev);
4766 rv = entry->store(mddev, page, length);
4767 mddev_unlock(mddev);
4773 static void md_free(struct kobject *ko)
4775 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4777 if (mddev->sysfs_state)
4778 sysfs_put(mddev->sysfs_state);
4780 if (mddev->gendisk) {
4781 del_gendisk(mddev->gendisk);
4782 put_disk(mddev->gendisk);
4785 blk_cleanup_queue(mddev->queue);
4790 static const struct sysfs_ops md_sysfs_ops = {
4791 .show = md_attr_show,
4792 .store = md_attr_store,
4794 static struct kobj_type md_ktype = {
4796 .sysfs_ops = &md_sysfs_ops,
4797 .default_attrs = md_default_attrs,
4802 static void mddev_delayed_delete(struct work_struct *ws)
4804 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4806 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4807 kobject_del(&mddev->kobj);
4808 kobject_put(&mddev->kobj);
4811 static int md_alloc(dev_t dev, char *name)
4813 static DEFINE_MUTEX(disks_mutex);
4814 struct mddev *mddev = mddev_find(dev);
4815 struct gendisk *disk;
4824 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4825 shift = partitioned ? MdpMinorShift : 0;
4826 unit = MINOR(mddev->unit) >> shift;
4828 /* wait for any previous instance of this device to be
4829 * completely removed (mddev_delayed_delete).
4831 flush_workqueue(md_misc_wq);
4833 mutex_lock(&disks_mutex);
4839 /* Need to ensure that 'name' is not a duplicate.
4841 struct mddev *mddev2;
4842 spin_lock(&all_mddevs_lock);
4844 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4845 if (mddev2->gendisk &&
4846 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4847 spin_unlock(&all_mddevs_lock);
4850 spin_unlock(&all_mddevs_lock);
4854 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4857 mddev->queue->queuedata = mddev;
4859 blk_queue_make_request(mddev->queue, md_make_request);
4860 blk_set_stacking_limits(&mddev->queue->limits);
4862 disk = alloc_disk(1 << shift);
4864 blk_cleanup_queue(mddev->queue);
4865 mddev->queue = NULL;
4868 disk->major = MAJOR(mddev->unit);
4869 disk->first_minor = unit << shift;
4871 strcpy(disk->disk_name, name);
4872 else if (partitioned)
4873 sprintf(disk->disk_name, "md_d%d", unit);
4875 sprintf(disk->disk_name, "md%d", unit);
4876 disk->fops = &md_fops;
4877 disk->private_data = mddev;
4878 disk->queue = mddev->queue;
4879 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4880 /* Allow extended partitions. This makes the
4881 * 'mdp' device redundant, but we can't really
4884 disk->flags |= GENHD_FL_EXT_DEVT;
4885 mddev->gendisk = disk;
4886 /* As soon as we call add_disk(), another thread could get
4887 * through to md_open, so make sure it doesn't get too far
4889 mutex_lock(&mddev->open_mutex);
4892 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4893 &disk_to_dev(disk)->kobj, "%s", "md");
4895 /* This isn't possible, but as kobject_init_and_add is marked
4896 * __must_check, we must do something with the result
4898 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4902 if (mddev->kobj.sd &&
4903 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4904 printk(KERN_DEBUG "pointless warning\n");
4905 mutex_unlock(&mddev->open_mutex);
4907 mutex_unlock(&disks_mutex);
4908 if (!error && mddev->kobj.sd) {
4909 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4910 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4916 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4918 md_alloc(dev, NULL);
4922 static int add_named_array(const char *val, struct kernel_param *kp)
4924 /* val must be "md_*" where * is not all digits.
4925 * We allocate an array with a large free minor number, and
4926 * set the name to val. val must not already be an active name.
4928 int len = strlen(val);
4929 char buf[DISK_NAME_LEN];
4931 while (len && val[len-1] == '\n')
4933 if (len >= DISK_NAME_LEN)
4935 strlcpy(buf, val, len+1);
4936 if (strncmp(buf, "md_", 3) != 0)
4938 return md_alloc(0, buf);
4941 static void md_safemode_timeout(unsigned long data)
4943 struct mddev *mddev = (struct mddev *) data;
4945 if (!atomic_read(&mddev->writes_pending)) {
4946 mddev->safemode = 1;
4947 if (mddev->external)
4948 sysfs_notify_dirent_safe(mddev->sysfs_state);
4950 md_wakeup_thread(mddev->thread);
4953 static int start_dirty_degraded;
4955 int md_run(struct mddev *mddev)
4958 struct md_rdev *rdev;
4959 struct md_personality *pers;
4961 if (list_empty(&mddev->disks))
4962 /* cannot run an array with no devices.. */
4967 /* Cannot run until previous stop completes properly */
4968 if (mddev->sysfs_active)
4972 * Analyze all RAID superblock(s)
4974 if (!mddev->raid_disks) {
4975 if (!mddev->persistent)
4980 if (mddev->level != LEVEL_NONE)
4981 request_module("md-level-%d", mddev->level);
4982 else if (mddev->clevel[0])
4983 request_module("md-%s", mddev->clevel);
4986 * Drop all container device buffers, from now on
4987 * the only valid external interface is through the md
4990 rdev_for_each(rdev, mddev) {
4991 if (test_bit(Faulty, &rdev->flags))
4993 sync_blockdev(rdev->bdev);
4994 invalidate_bdev(rdev->bdev);
4996 /* perform some consistency tests on the device.
4997 * We don't want the data to overlap the metadata,
4998 * Internal Bitmap issues have been handled elsewhere.
5000 if (rdev->meta_bdev) {
5001 /* Nothing to check */;
5002 } else if (rdev->data_offset < rdev->sb_start) {
5003 if (mddev->dev_sectors &&
5004 rdev->data_offset + mddev->dev_sectors
5006 printk("md: %s: data overlaps metadata\n",
5011 if (rdev->sb_start + rdev->sb_size/512
5012 > rdev->data_offset) {
5013 printk("md: %s: metadata overlaps data\n",
5018 sysfs_notify_dirent_safe(rdev->sysfs_state);
5021 if (mddev->bio_set == NULL)
5022 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5024 spin_lock(&pers_lock);
5025 pers = find_pers(mddev->level, mddev->clevel);
5026 if (!pers || !try_module_get(pers->owner)) {
5027 spin_unlock(&pers_lock);
5028 if (mddev->level != LEVEL_NONE)
5029 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5032 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5037 spin_unlock(&pers_lock);
5038 if (mddev->level != pers->level) {
5039 mddev->level = pers->level;
5040 mddev->new_level = pers->level;
5042 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5044 if (mddev->reshape_position != MaxSector &&
5045 pers->start_reshape == NULL) {
5046 /* This personality cannot handle reshaping... */
5048 module_put(pers->owner);
5052 if (pers->sync_request) {
5053 /* Warn if this is a potentially silly
5056 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5057 struct md_rdev *rdev2;
5060 rdev_for_each(rdev, mddev)
5061 rdev_for_each(rdev2, mddev) {
5063 rdev->bdev->bd_contains ==
5064 rdev2->bdev->bd_contains) {
5066 "%s: WARNING: %s appears to be"
5067 " on the same physical disk as"
5070 bdevname(rdev->bdev,b),
5071 bdevname(rdev2->bdev,b2));
5078 "True protection against single-disk"
5079 " failure might be compromised.\n");
5082 mddev->recovery = 0;
5083 /* may be over-ridden by personality */
5084 mddev->resync_max_sectors = mddev->dev_sectors;
5086 mddev->ok_start_degraded = start_dirty_degraded;
5088 if (start_readonly && mddev->ro == 0)
5089 mddev->ro = 2; /* read-only, but switch on first write */
5091 err = mddev->pers->run(mddev);
5093 printk(KERN_ERR "md: pers->run() failed ...\n");
5094 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5095 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5096 " but 'external_size' not in effect?\n", __func__);
5098 "md: invalid array_size %llu > default size %llu\n",
5099 (unsigned long long)mddev->array_sectors / 2,
5100 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5102 mddev->pers->stop(mddev);
5104 if (err == 0 && mddev->pers->sync_request &&
5105 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5106 err = bitmap_create(mddev);
5108 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5109 mdname(mddev), err);
5110 mddev->pers->stop(mddev);
5114 module_put(mddev->pers->owner);
5116 bitmap_destroy(mddev);
5119 if (mddev->pers->sync_request) {
5120 if (mddev->kobj.sd &&
5121 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5123 "md: cannot register extra attributes for %s\n",
5125 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5126 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5129 atomic_set(&mddev->writes_pending,0);
5130 atomic_set(&mddev->max_corr_read_errors,
5131 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5132 mddev->safemode = 0;
5133 mddev->safemode_timer.function = md_safemode_timeout;
5134 mddev->safemode_timer.data = (unsigned long) mddev;
5135 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5139 rdev_for_each(rdev, mddev)
5140 if (rdev->raid_disk >= 0)
5141 if (sysfs_link_rdev(mddev, rdev))
5142 /* failure here is OK */;
5144 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5147 md_update_sb(mddev, 0);
5149 md_new_event(mddev);
5150 sysfs_notify_dirent_safe(mddev->sysfs_state);
5151 sysfs_notify_dirent_safe(mddev->sysfs_action);
5152 sysfs_notify(&mddev->kobj, NULL, "degraded");
5155 EXPORT_SYMBOL_GPL(md_run);
5157 static int do_md_run(struct mddev *mddev)
5161 err = md_run(mddev);
5164 err = bitmap_load(mddev);
5166 bitmap_destroy(mddev);
5170 md_wakeup_thread(mddev->thread);
5171 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5173 set_capacity(mddev->gendisk, mddev->array_sectors);
5174 revalidate_disk(mddev->gendisk);
5176 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5181 static int restart_array(struct mddev *mddev)
5183 struct gendisk *disk = mddev->gendisk;
5185 /* Complain if it has no devices */
5186 if (list_empty(&mddev->disks))
5192 mddev->safemode = 0;
5194 set_disk_ro(disk, 0);
5195 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5197 /* Kick recovery or resync if necessary */
5198 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5199 md_wakeup_thread(mddev->thread);
5200 md_wakeup_thread(mddev->sync_thread);
5201 sysfs_notify_dirent_safe(mddev->sysfs_state);
5205 /* similar to deny_write_access, but accounts for our holding a reference
5206 * to the file ourselves */
5207 static int deny_bitmap_write_access(struct file * file)
5209 struct inode *inode = file->f_mapping->host;
5211 spin_lock(&inode->i_lock);
5212 if (atomic_read(&inode->i_writecount) > 1) {
5213 spin_unlock(&inode->i_lock);
5216 atomic_set(&inode->i_writecount, -1);
5217 spin_unlock(&inode->i_lock);
5222 void restore_bitmap_write_access(struct file *file)
5224 struct inode *inode = file->f_mapping->host;
5226 spin_lock(&inode->i_lock);
5227 atomic_set(&inode->i_writecount, 1);
5228 spin_unlock(&inode->i_lock);
5231 static void md_clean(struct mddev *mddev)
5233 mddev->array_sectors = 0;
5234 mddev->external_size = 0;
5235 mddev->dev_sectors = 0;
5236 mddev->raid_disks = 0;
5237 mddev->recovery_cp = 0;
5238 mddev->resync_min = 0;
5239 mddev->resync_max = MaxSector;
5240 mddev->reshape_position = MaxSector;
5241 mddev->external = 0;
5242 mddev->persistent = 0;
5243 mddev->level = LEVEL_NONE;
5244 mddev->clevel[0] = 0;
5247 mddev->metadata_type[0] = 0;
5248 mddev->chunk_sectors = 0;
5249 mddev->ctime = mddev->utime = 0;
5251 mddev->max_disks = 0;
5253 mddev->can_decrease_events = 0;
5254 mddev->delta_disks = 0;
5255 mddev->reshape_backwards = 0;
5256 mddev->new_level = LEVEL_NONE;
5257 mddev->new_layout = 0;
5258 mddev->new_chunk_sectors = 0;
5259 mddev->curr_resync = 0;
5260 atomic64_set(&mddev->resync_mismatches, 0);
5261 mddev->suspend_lo = mddev->suspend_hi = 0;
5262 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5263 mddev->recovery = 0;
5266 mddev->degraded = 0;
5267 mddev->safemode = 0;
5268 mddev->merge_check_needed = 0;
5269 mddev->bitmap_info.offset = 0;
5270 mddev->bitmap_info.default_offset = 0;
5271 mddev->bitmap_info.default_space = 0;
5272 mddev->bitmap_info.chunksize = 0;
5273 mddev->bitmap_info.daemon_sleep = 0;
5274 mddev->bitmap_info.max_write_behind = 0;
5277 static void __md_stop_writes(struct mddev *mddev)
5279 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5280 if (mddev->sync_thread) {
5281 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5282 md_reap_sync_thread(mddev);
5285 del_timer_sync(&mddev->safemode_timer);
5287 bitmap_flush(mddev);
5288 md_super_wait(mddev);
5290 if (mddev->ro == 0 &&
5291 (!mddev->in_sync || mddev->flags)) {
5292 /* mark array as shutdown cleanly */
5294 md_update_sb(mddev, 1);
5298 void md_stop_writes(struct mddev *mddev)
5301 __md_stop_writes(mddev);
5302 mddev_unlock(mddev);
5304 EXPORT_SYMBOL_GPL(md_stop_writes);
5306 static void __md_stop(struct mddev *mddev)
5309 mddev->pers->stop(mddev);
5310 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5311 mddev->to_remove = &md_redundancy_group;
5312 module_put(mddev->pers->owner);
5314 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5317 void md_stop(struct mddev *mddev)
5319 /* stop the array and free an attached data structures.
5320 * This is called from dm-raid
5323 bitmap_destroy(mddev);
5325 bioset_free(mddev->bio_set);
5328 EXPORT_SYMBOL_GPL(md_stop);
5330 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5333 mutex_lock(&mddev->open_mutex);
5334 if (atomic_read(&mddev->openers) > !!bdev) {
5335 printk("md: %s still in use.\n",mdname(mddev));
5340 sync_blockdev(bdev);
5342 __md_stop_writes(mddev);
5348 set_disk_ro(mddev->gendisk, 1);
5349 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5350 sysfs_notify_dirent_safe(mddev->sysfs_state);
5354 mutex_unlock(&mddev->open_mutex);
5359 * 0 - completely stop and dis-assemble array
5360 * 2 - stop but do not disassemble array
5362 static int do_md_stop(struct mddev * mddev, int mode,
5363 struct block_device *bdev)
5365 struct gendisk *disk = mddev->gendisk;
5366 struct md_rdev *rdev;
5368 mutex_lock(&mddev->open_mutex);
5369 if (atomic_read(&mddev->openers) > !!bdev ||
5370 mddev->sysfs_active) {
5371 printk("md: %s still in use.\n",mdname(mddev));
5372 mutex_unlock(&mddev->open_mutex);
5376 /* It is possible IO was issued on some other
5377 * open file which was closed before we took ->open_mutex.
5378 * As that was not the last close __blkdev_put will not
5379 * have called sync_blockdev, so we must.
5381 sync_blockdev(bdev);
5385 set_disk_ro(disk, 0);
5387 __md_stop_writes(mddev);
5389 mddev->queue->merge_bvec_fn = NULL;
5390 mddev->queue->backing_dev_info.congested_fn = NULL;
5392 /* tell userspace to handle 'inactive' */
5393 sysfs_notify_dirent_safe(mddev->sysfs_state);
5395 rdev_for_each(rdev, mddev)
5396 if (rdev->raid_disk >= 0)
5397 sysfs_unlink_rdev(mddev, rdev);
5399 set_capacity(disk, 0);
5400 mutex_unlock(&mddev->open_mutex);
5402 revalidate_disk(disk);
5407 mutex_unlock(&mddev->open_mutex);
5409 * Free resources if final stop
5412 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5414 bitmap_destroy(mddev);
5415 if (mddev->bitmap_info.file) {
5416 restore_bitmap_write_access(mddev->bitmap_info.file);
5417 fput(mddev->bitmap_info.file);
5418 mddev->bitmap_info.file = NULL;
5420 mddev->bitmap_info.offset = 0;
5422 export_array(mddev);
5425 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5426 if (mddev->hold_active == UNTIL_STOP)
5427 mddev->hold_active = 0;
5429 blk_integrity_unregister(disk);
5430 md_new_event(mddev);
5431 sysfs_notify_dirent_safe(mddev->sysfs_state);
5436 static void autorun_array(struct mddev *mddev)
5438 struct md_rdev *rdev;
5441 if (list_empty(&mddev->disks))
5444 printk(KERN_INFO "md: running: ");
5446 rdev_for_each(rdev, mddev) {
5447 char b[BDEVNAME_SIZE];
5448 printk("<%s>", bdevname(rdev->bdev,b));
5452 err = do_md_run(mddev);
5454 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5455 do_md_stop(mddev, 0, NULL);
5460 * lets try to run arrays based on all disks that have arrived
5461 * until now. (those are in pending_raid_disks)
5463 * the method: pick the first pending disk, collect all disks with
5464 * the same UUID, remove all from the pending list and put them into
5465 * the 'same_array' list. Then order this list based on superblock
5466 * update time (freshest comes first), kick out 'old' disks and
5467 * compare superblocks. If everything's fine then run it.
5469 * If "unit" is allocated, then bump its reference count
5471 static void autorun_devices(int part)
5473 struct md_rdev *rdev0, *rdev, *tmp;
5474 struct mddev *mddev;
5475 char b[BDEVNAME_SIZE];
5477 printk(KERN_INFO "md: autorun ...\n");
5478 while (!list_empty(&pending_raid_disks)) {
5481 LIST_HEAD(candidates);
5482 rdev0 = list_entry(pending_raid_disks.next,
5483 struct md_rdev, same_set);
5485 printk(KERN_INFO "md: considering %s ...\n",
5486 bdevname(rdev0->bdev,b));
5487 INIT_LIST_HEAD(&candidates);
5488 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5489 if (super_90_load(rdev, rdev0, 0) >= 0) {
5490 printk(KERN_INFO "md: adding %s ...\n",
5491 bdevname(rdev->bdev,b));
5492 list_move(&rdev->same_set, &candidates);
5495 * now we have a set of devices, with all of them having
5496 * mostly sane superblocks. It's time to allocate the
5500 dev = MKDEV(mdp_major,
5501 rdev0->preferred_minor << MdpMinorShift);
5502 unit = MINOR(dev) >> MdpMinorShift;
5504 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5507 if (rdev0->preferred_minor != unit) {
5508 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5509 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5513 md_probe(dev, NULL, NULL);
5514 mddev = mddev_find(dev);
5515 if (!mddev || !mddev->gendisk) {
5519 "md: cannot allocate memory for md drive.\n");
5522 if (mddev_lock(mddev))
5523 printk(KERN_WARNING "md: %s locked, cannot run\n",
5525 else if (mddev->raid_disks || mddev->major_version
5526 || !list_empty(&mddev->disks)) {
5528 "md: %s already running, cannot run %s\n",
5529 mdname(mddev), bdevname(rdev0->bdev,b));
5530 mddev_unlock(mddev);
5532 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5533 mddev->persistent = 1;
5534 rdev_for_each_list(rdev, tmp, &candidates) {
5535 list_del_init(&rdev->same_set);
5536 if (bind_rdev_to_array(rdev, mddev))
5539 autorun_array(mddev);
5540 mddev_unlock(mddev);
5542 /* on success, candidates will be empty, on error
5545 rdev_for_each_list(rdev, tmp, &candidates) {
5546 list_del_init(&rdev->same_set);
5551 printk(KERN_INFO "md: ... autorun DONE.\n");
5553 #endif /* !MODULE */
5555 static int get_version(void __user * arg)
5559 ver.major = MD_MAJOR_VERSION;
5560 ver.minor = MD_MINOR_VERSION;
5561 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5563 if (copy_to_user(arg, &ver, sizeof(ver)))
5569 static int get_array_info(struct mddev * mddev, void __user * arg)
5571 mdu_array_info_t info;
5572 int nr,working,insync,failed,spare;
5573 struct md_rdev *rdev;
5575 nr = working = insync = failed = spare = 0;
5577 rdev_for_each_rcu(rdev, mddev) {
5579 if (test_bit(Faulty, &rdev->flags))
5583 if (test_bit(In_sync, &rdev->flags))
5591 info.major_version = mddev->major_version;
5592 info.minor_version = mddev->minor_version;
5593 info.patch_version = MD_PATCHLEVEL_VERSION;
5594 info.ctime = mddev->ctime;
5595 info.level = mddev->level;
5596 info.size = mddev->dev_sectors / 2;
5597 if (info.size != mddev->dev_sectors / 2) /* overflow */
5600 info.raid_disks = mddev->raid_disks;
5601 info.md_minor = mddev->md_minor;
5602 info.not_persistent= !mddev->persistent;
5604 info.utime = mddev->utime;
5607 info.state = (1<<MD_SB_CLEAN);
5608 if (mddev->bitmap && mddev->bitmap_info.offset)
5609 info.state = (1<<MD_SB_BITMAP_PRESENT);
5610 info.active_disks = insync;
5611 info.working_disks = working;
5612 info.failed_disks = failed;
5613 info.spare_disks = spare;
5615 info.layout = mddev->layout;
5616 info.chunk_size = mddev->chunk_sectors << 9;
5618 if (copy_to_user(arg, &info, sizeof(info)))
5624 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5626 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5627 char *ptr, *buf = NULL;
5630 if (md_allow_write(mddev))
5631 file = kmalloc(sizeof(*file), GFP_NOIO);
5633 file = kmalloc(sizeof(*file), GFP_KERNEL);
5638 /* bitmap disabled, zero the first byte and copy out */
5639 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5640 file->pathname[0] = '\0';
5644 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5648 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5649 buf, sizeof(file->pathname));
5653 strcpy(file->pathname, ptr);
5657 if (copy_to_user(arg, file, sizeof(*file)))
5665 static int get_disk_info(struct mddev * mddev, void __user * arg)
5667 mdu_disk_info_t info;
5668 struct md_rdev *rdev;
5670 if (copy_from_user(&info, arg, sizeof(info)))
5674 rdev = find_rdev_nr_rcu(mddev, info.number);
5676 info.major = MAJOR(rdev->bdev->bd_dev);
5677 info.minor = MINOR(rdev->bdev->bd_dev);
5678 info.raid_disk = rdev->raid_disk;
5680 if (test_bit(Faulty, &rdev->flags))
5681 info.state |= (1<<MD_DISK_FAULTY);
5682 else if (test_bit(In_sync, &rdev->flags)) {
5683 info.state |= (1<<MD_DISK_ACTIVE);
5684 info.state |= (1<<MD_DISK_SYNC);
5686 if (test_bit(WriteMostly, &rdev->flags))
5687 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5689 info.major = info.minor = 0;
5690 info.raid_disk = -1;
5691 info.state = (1<<MD_DISK_REMOVED);
5695 if (copy_to_user(arg, &info, sizeof(info)))
5701 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5703 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5704 struct md_rdev *rdev;
5705 dev_t dev = MKDEV(info->major,info->minor);
5707 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5710 if (!mddev->raid_disks) {
5712 /* expecting a device which has a superblock */
5713 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5716 "md: md_import_device returned %ld\n",
5718 return PTR_ERR(rdev);
5720 if (!list_empty(&mddev->disks)) {
5721 struct md_rdev *rdev0
5722 = list_entry(mddev->disks.next,
5723 struct md_rdev, same_set);
5724 err = super_types[mddev->major_version]
5725 .load_super(rdev, rdev0, mddev->minor_version);
5728 "md: %s has different UUID to %s\n",
5729 bdevname(rdev->bdev,b),
5730 bdevname(rdev0->bdev,b2));
5735 err = bind_rdev_to_array(rdev, mddev);
5742 * add_new_disk can be used once the array is assembled
5743 * to add "hot spares". They must already have a superblock
5748 if (!mddev->pers->hot_add_disk) {
5750 "%s: personality does not support diskops!\n",
5754 if (mddev->persistent)
5755 rdev = md_import_device(dev, mddev->major_version,
5756 mddev->minor_version);
5758 rdev = md_import_device(dev, -1, -1);
5761 "md: md_import_device returned %ld\n",
5763 return PTR_ERR(rdev);
5765 /* set saved_raid_disk if appropriate */
5766 if (!mddev->persistent) {
5767 if (info->state & (1<<MD_DISK_SYNC) &&
5768 info->raid_disk < mddev->raid_disks) {
5769 rdev->raid_disk = info->raid_disk;
5770 set_bit(In_sync, &rdev->flags);
5771 clear_bit(Bitmap_sync, &rdev->flags);
5773 rdev->raid_disk = -1;
5775 super_types[mddev->major_version].
5776 validate_super(mddev, rdev);
5777 if ((info->state & (1<<MD_DISK_SYNC)) &&
5778 rdev->raid_disk != info->raid_disk) {
5779 /* This was a hot-add request, but events doesn't
5780 * match, so reject it.
5786 if (test_bit(In_sync, &rdev->flags))
5787 rdev->saved_raid_disk = rdev->raid_disk;
5789 rdev->saved_raid_disk = -1;
5791 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5792 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5793 set_bit(WriteMostly, &rdev->flags);
5795 clear_bit(WriteMostly, &rdev->flags);
5797 rdev->raid_disk = -1;
5798 err = bind_rdev_to_array(rdev, mddev);
5799 if (!err && !mddev->pers->hot_remove_disk) {
5800 /* If there is hot_add_disk but no hot_remove_disk
5801 * then added disks for geometry changes,
5802 * and should be added immediately.
5804 super_types[mddev->major_version].
5805 validate_super(mddev, rdev);
5806 err = mddev->pers->hot_add_disk(mddev, rdev);
5808 unbind_rdev_from_array(rdev);
5813 sysfs_notify_dirent_safe(rdev->sysfs_state);
5815 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5816 if (mddev->degraded)
5817 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5818 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5820 md_new_event(mddev);
5821 md_wakeup_thread(mddev->thread);
5825 /* otherwise, add_new_disk is only allowed
5826 * for major_version==0 superblocks
5828 if (mddev->major_version != 0) {
5829 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5834 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5836 rdev = md_import_device(dev, -1, 0);
5839 "md: error, md_import_device() returned %ld\n",
5841 return PTR_ERR(rdev);
5843 rdev->desc_nr = info->number;
5844 if (info->raid_disk < mddev->raid_disks)
5845 rdev->raid_disk = info->raid_disk;
5847 rdev->raid_disk = -1;
5849 if (rdev->raid_disk < mddev->raid_disks)
5850 if (info->state & (1<<MD_DISK_SYNC))
5851 set_bit(In_sync, &rdev->flags);
5853 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5854 set_bit(WriteMostly, &rdev->flags);
5856 if (!mddev->persistent) {
5857 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5858 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5860 rdev->sb_start = calc_dev_sboffset(rdev);
5861 rdev->sectors = rdev->sb_start;
5863 err = bind_rdev_to_array(rdev, mddev);
5873 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5875 char b[BDEVNAME_SIZE];
5876 struct md_rdev *rdev;
5878 rdev = find_rdev(mddev, dev);
5882 clear_bit(Blocked, &rdev->flags);
5883 remove_and_add_spares(mddev, rdev);
5885 if (rdev->raid_disk >= 0)
5888 kick_rdev_from_array(rdev);
5889 md_update_sb(mddev, 1);
5890 md_new_event(mddev);
5894 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5895 bdevname(rdev->bdev,b), mdname(mddev));
5899 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5901 char b[BDEVNAME_SIZE];
5903 struct md_rdev *rdev;
5908 if (mddev->major_version != 0) {
5909 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5910 " version-0 superblocks.\n",
5914 if (!mddev->pers->hot_add_disk) {
5916 "%s: personality does not support diskops!\n",
5921 rdev = md_import_device(dev, -1, 0);
5924 "md: error, md_import_device() returned %ld\n",
5929 if (mddev->persistent)
5930 rdev->sb_start = calc_dev_sboffset(rdev);
5932 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5934 rdev->sectors = rdev->sb_start;
5936 if (test_bit(Faulty, &rdev->flags)) {
5938 "md: can not hot-add faulty %s disk to %s!\n",
5939 bdevname(rdev->bdev,b), mdname(mddev));
5943 clear_bit(In_sync, &rdev->flags);
5945 rdev->saved_raid_disk = -1;
5946 err = bind_rdev_to_array(rdev, mddev);
5951 * The rest should better be atomic, we can have disk failures
5952 * noticed in interrupt contexts ...
5955 rdev->raid_disk = -1;
5957 md_update_sb(mddev, 1);
5960 * Kick recovery, maybe this spare has to be added to the
5961 * array immediately.
5963 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5964 md_wakeup_thread(mddev->thread);
5965 md_new_event(mddev);
5973 static int set_bitmap_file(struct mddev *mddev, int fd)
5978 if (!mddev->pers->quiesce)
5980 if (mddev->recovery || mddev->sync_thread)
5982 /* we should be able to change the bitmap.. */
5988 return -EEXIST; /* cannot add when bitmap is present */
5989 mddev->bitmap_info.file = fget(fd);
5991 if (mddev->bitmap_info.file == NULL) {
5992 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5997 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5999 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6001 fput(mddev->bitmap_info.file);
6002 mddev->bitmap_info.file = NULL;
6005 mddev->bitmap_info.offset = 0; /* file overrides offset */
6006 } else if (mddev->bitmap == NULL)
6007 return -ENOENT; /* cannot remove what isn't there */
6010 mddev->pers->quiesce(mddev, 1);
6012 err = bitmap_create(mddev);
6014 err = bitmap_load(mddev);
6016 if (fd < 0 || err) {
6017 bitmap_destroy(mddev);
6018 fd = -1; /* make sure to put the file */
6020 mddev->pers->quiesce(mddev, 0);
6023 if (mddev->bitmap_info.file) {
6024 restore_bitmap_write_access(mddev->bitmap_info.file);
6025 fput(mddev->bitmap_info.file);
6027 mddev->bitmap_info.file = NULL;
6034 * set_array_info is used two different ways
6035 * The original usage is when creating a new array.
6036 * In this usage, raid_disks is > 0 and it together with
6037 * level, size, not_persistent,layout,chunksize determine the
6038 * shape of the array.
6039 * This will always create an array with a type-0.90.0 superblock.
6040 * The newer usage is when assembling an array.
6041 * In this case raid_disks will be 0, and the major_version field is
6042 * use to determine which style super-blocks are to be found on the devices.
6043 * The minor and patch _version numbers are also kept incase the
6044 * super_block handler wishes to interpret them.
6046 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6049 if (info->raid_disks == 0) {
6050 /* just setting version number for superblock loading */
6051 if (info->major_version < 0 ||
6052 info->major_version >= ARRAY_SIZE(super_types) ||
6053 super_types[info->major_version].name == NULL) {
6054 /* maybe try to auto-load a module? */
6056 "md: superblock version %d not known\n",
6057 info->major_version);
6060 mddev->major_version = info->major_version;
6061 mddev->minor_version = info->minor_version;
6062 mddev->patch_version = info->patch_version;
6063 mddev->persistent = !info->not_persistent;
6064 /* ensure mddev_put doesn't delete this now that there
6065 * is some minimal configuration.
6067 mddev->ctime = get_seconds();
6070 mddev->major_version = MD_MAJOR_VERSION;
6071 mddev->minor_version = MD_MINOR_VERSION;
6072 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6073 mddev->ctime = get_seconds();
6075 mddev->level = info->level;
6076 mddev->clevel[0] = 0;
6077 mddev->dev_sectors = 2 * (sector_t)info->size;
6078 mddev->raid_disks = info->raid_disks;
6079 /* don't set md_minor, it is determined by which /dev/md* was
6082 if (info->state & (1<<MD_SB_CLEAN))
6083 mddev->recovery_cp = MaxSector;
6085 mddev->recovery_cp = 0;
6086 mddev->persistent = ! info->not_persistent;
6087 mddev->external = 0;
6089 mddev->layout = info->layout;
6090 mddev->chunk_sectors = info->chunk_size >> 9;
6092 mddev->max_disks = MD_SB_DISKS;
6094 if (mddev->persistent)
6096 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6098 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6099 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6100 mddev->bitmap_info.offset = 0;
6102 mddev->reshape_position = MaxSector;
6105 * Generate a 128 bit UUID
6107 get_random_bytes(mddev->uuid, 16);
6109 mddev->new_level = mddev->level;
6110 mddev->new_chunk_sectors = mddev->chunk_sectors;
6111 mddev->new_layout = mddev->layout;
6112 mddev->delta_disks = 0;
6113 mddev->reshape_backwards = 0;
6118 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6120 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6122 if (mddev->external_size)
6125 mddev->array_sectors = array_sectors;
6127 EXPORT_SYMBOL(md_set_array_sectors);
6129 static int update_size(struct mddev *mddev, sector_t num_sectors)
6131 struct md_rdev *rdev;
6133 int fit = (num_sectors == 0);
6135 if (mddev->pers->resize == NULL)
6137 /* The "num_sectors" is the number of sectors of each device that
6138 * is used. This can only make sense for arrays with redundancy.
6139 * linear and raid0 always use whatever space is available. We can only
6140 * consider changing this number if no resync or reconstruction is
6141 * happening, and if the new size is acceptable. It must fit before the
6142 * sb_start or, if that is <data_offset, it must fit before the size
6143 * of each device. If num_sectors is zero, we find the largest size
6146 if (mddev->sync_thread)
6149 rdev_for_each(rdev, mddev) {
6150 sector_t avail = rdev->sectors;
6152 if (fit && (num_sectors == 0 || num_sectors > avail))
6153 num_sectors = avail;
6154 if (avail < num_sectors)
6157 rv = mddev->pers->resize(mddev, num_sectors);
6159 revalidate_disk(mddev->gendisk);
6163 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6166 struct md_rdev *rdev;
6167 /* change the number of raid disks */
6168 if (mddev->pers->check_reshape == NULL)
6170 if (raid_disks <= 0 ||
6171 (mddev->max_disks && raid_disks >= mddev->max_disks))
6173 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6176 rdev_for_each(rdev, mddev) {
6177 if (mddev->raid_disks < raid_disks &&
6178 rdev->data_offset < rdev->new_data_offset)
6180 if (mddev->raid_disks > raid_disks &&
6181 rdev->data_offset > rdev->new_data_offset)
6185 mddev->delta_disks = raid_disks - mddev->raid_disks;
6186 if (mddev->delta_disks < 0)
6187 mddev->reshape_backwards = 1;
6188 else if (mddev->delta_disks > 0)
6189 mddev->reshape_backwards = 0;
6191 rv = mddev->pers->check_reshape(mddev);
6193 mddev->delta_disks = 0;
6194 mddev->reshape_backwards = 0;
6201 * update_array_info is used to change the configuration of an
6203 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6204 * fields in the info are checked against the array.
6205 * Any differences that cannot be handled will cause an error.
6206 * Normally, only one change can be managed at a time.
6208 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6214 /* calculate expected state,ignoring low bits */
6215 if (mddev->bitmap && mddev->bitmap_info.offset)
6216 state |= (1 << MD_SB_BITMAP_PRESENT);
6218 if (mddev->major_version != info->major_version ||
6219 mddev->minor_version != info->minor_version ||
6220 /* mddev->patch_version != info->patch_version || */
6221 mddev->ctime != info->ctime ||
6222 mddev->level != info->level ||
6223 /* mddev->layout != info->layout || */
6224 !mddev->persistent != info->not_persistent||
6225 mddev->chunk_sectors != info->chunk_size >> 9 ||
6226 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6227 ((state^info->state) & 0xfffffe00)
6230 /* Check there is only one change */
6231 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6233 if (mddev->raid_disks != info->raid_disks)
6235 if (mddev->layout != info->layout)
6237 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6244 if (mddev->layout != info->layout) {
6246 * we don't need to do anything at the md level, the
6247 * personality will take care of it all.
6249 if (mddev->pers->check_reshape == NULL)
6252 mddev->new_layout = info->layout;
6253 rv = mddev->pers->check_reshape(mddev);
6255 mddev->new_layout = mddev->layout;
6259 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6260 rv = update_size(mddev, (sector_t)info->size * 2);
6262 if (mddev->raid_disks != info->raid_disks)
6263 rv = update_raid_disks(mddev, info->raid_disks);
6265 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6266 if (mddev->pers->quiesce == NULL)
6268 if (mddev->recovery || mddev->sync_thread)
6270 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6271 /* add the bitmap */
6274 if (mddev->bitmap_info.default_offset == 0)
6276 mddev->bitmap_info.offset =
6277 mddev->bitmap_info.default_offset;
6278 mddev->bitmap_info.space =
6279 mddev->bitmap_info.default_space;
6280 mddev->pers->quiesce(mddev, 1);
6281 rv = bitmap_create(mddev);
6283 rv = bitmap_load(mddev);
6285 bitmap_destroy(mddev);
6286 mddev->pers->quiesce(mddev, 0);
6288 /* remove the bitmap */
6291 if (mddev->bitmap->storage.file)
6293 mddev->pers->quiesce(mddev, 1);
6294 bitmap_destroy(mddev);
6295 mddev->pers->quiesce(mddev, 0);
6296 mddev->bitmap_info.offset = 0;
6299 md_update_sb(mddev, 1);
6303 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6305 struct md_rdev *rdev;
6308 if (mddev->pers == NULL)
6312 rdev = find_rdev_rcu(mddev, dev);
6316 md_error(mddev, rdev);
6317 if (!test_bit(Faulty, &rdev->flags))
6325 * We have a problem here : there is no easy way to give a CHS
6326 * virtual geometry. We currently pretend that we have a 2 heads
6327 * 4 sectors (with a BIG number of cylinders...). This drives
6328 * dosfs just mad... ;-)
6330 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6332 struct mddev *mddev = bdev->bd_disk->private_data;
6336 geo->cylinders = mddev->array_sectors / 8;
6340 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6341 unsigned int cmd, unsigned long arg)
6344 void __user *argp = (void __user *)arg;
6345 struct mddev *mddev = NULL;
6350 case GET_ARRAY_INFO:
6354 if (!capable(CAP_SYS_ADMIN))
6359 * Commands dealing with the RAID driver but not any
6364 err = get_version(argp);
6367 case PRINT_RAID_DEBUG:
6375 autostart_arrays(arg);
6382 * Commands creating/starting a new array:
6385 mddev = bdev->bd_disk->private_data;
6392 /* Some actions do not requires the mutex */
6394 case GET_ARRAY_INFO:
6395 if (!mddev->raid_disks && !mddev->external)
6398 err = get_array_info(mddev, argp);
6402 if (!mddev->raid_disks && !mddev->external)
6405 err = get_disk_info(mddev, argp);
6408 case SET_DISK_FAULTY:
6409 err = set_disk_faulty(mddev, new_decode_dev(arg));
6413 if (cmd == ADD_NEW_DISK)
6414 /* need to ensure md_delayed_delete() has completed */
6415 flush_workqueue(md_misc_wq);
6417 err = mddev_lock(mddev);
6420 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6425 if (cmd == SET_ARRAY_INFO) {
6426 mdu_array_info_t info;
6428 memset(&info, 0, sizeof(info));
6429 else if (copy_from_user(&info, argp, sizeof(info))) {
6434 err = update_array_info(mddev, &info);
6436 printk(KERN_WARNING "md: couldn't update"
6437 " array info. %d\n", err);
6442 if (!list_empty(&mddev->disks)) {
6444 "md: array %s already has disks!\n",
6449 if (mddev->raid_disks) {
6451 "md: array %s already initialised!\n",
6456 err = set_array_info(mddev, &info);
6458 printk(KERN_WARNING "md: couldn't set"
6459 " array info. %d\n", err);
6466 * Commands querying/configuring an existing array:
6468 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6469 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6470 if ((!mddev->raid_disks && !mddev->external)
6471 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6472 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6473 && cmd != GET_BITMAP_FILE) {
6479 * Commands even a read-only array can execute:
6482 case GET_BITMAP_FILE:
6483 err = get_bitmap_file(mddev, argp);
6486 case RESTART_ARRAY_RW:
6487 err = restart_array(mddev);
6491 err = do_md_stop(mddev, 0, bdev);
6495 err = md_set_readonly(mddev, bdev);
6498 case HOT_REMOVE_DISK:
6499 err = hot_remove_disk(mddev, new_decode_dev(arg));
6503 /* We can support ADD_NEW_DISK on read-only arrays
6504 * on if we are re-adding a preexisting device.
6505 * So require mddev->pers and MD_DISK_SYNC.
6508 mdu_disk_info_t info;
6509 if (copy_from_user(&info, argp, sizeof(info)))
6511 else if (!(info.state & (1<<MD_DISK_SYNC)))
6512 /* Need to clear read-only for this */
6515 err = add_new_disk(mddev, &info);
6521 if (get_user(ro, (int __user *)(arg))) {
6527 /* if the bdev is going readonly the value of mddev->ro
6528 * does not matter, no writes are coming
6533 /* are we are already prepared for writes? */
6537 /* transitioning to readauto need only happen for
6538 * arrays that call md_write_start
6541 err = restart_array(mddev);
6544 set_disk_ro(mddev->gendisk, 0);
6551 * The remaining ioctls are changing the state of the
6552 * superblock, so we do not allow them on read-only arrays.
6553 * However non-MD ioctls (e.g. get-size) will still come through
6554 * here and hit the 'default' below, so only disallow
6555 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6557 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6558 if (mddev->ro == 2) {
6560 sysfs_notify_dirent_safe(mddev->sysfs_state);
6561 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6562 /* mddev_unlock will wake thread */
6563 /* If a device failed while we were read-only, we
6564 * need to make sure the metadata is updated now.
6566 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6567 mddev_unlock(mddev);
6568 wait_event(mddev->sb_wait,
6569 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6570 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6582 mdu_disk_info_t info;
6583 if (copy_from_user(&info, argp, sizeof(info)))
6586 err = add_new_disk(mddev, &info);
6591 err = hot_add_disk(mddev, new_decode_dev(arg));
6595 err = do_md_run(mddev);
6598 case SET_BITMAP_FILE:
6599 err = set_bitmap_file(mddev, (int)arg);
6609 if (mddev->hold_active == UNTIL_IOCTL &&
6611 mddev->hold_active = 0;
6612 mddev_unlock(mddev);
6621 #ifdef CONFIG_COMPAT
6622 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6623 unsigned int cmd, unsigned long arg)
6626 case HOT_REMOVE_DISK:
6628 case SET_DISK_FAULTY:
6629 case SET_BITMAP_FILE:
6630 /* These take in integer arg, do not convert */
6633 arg = (unsigned long)compat_ptr(arg);
6637 return md_ioctl(bdev, mode, cmd, arg);
6639 #endif /* CONFIG_COMPAT */
6641 static int md_open(struct block_device *bdev, fmode_t mode)
6644 * Succeed if we can lock the mddev, which confirms that
6645 * it isn't being stopped right now.
6647 struct mddev *mddev = mddev_find(bdev->bd_dev);
6653 if (mddev->gendisk != bdev->bd_disk) {
6654 /* we are racing with mddev_put which is discarding this
6658 /* Wait until bdev->bd_disk is definitely gone */
6659 flush_workqueue(md_misc_wq);
6660 /* Then retry the open from the top */
6661 return -ERESTARTSYS;
6663 BUG_ON(mddev != bdev->bd_disk->private_data);
6665 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6669 atomic_inc(&mddev->openers);
6670 mutex_unlock(&mddev->open_mutex);
6672 check_disk_change(bdev);
6677 static void md_release(struct gendisk *disk, fmode_t mode)
6679 struct mddev *mddev = disk->private_data;
6682 atomic_dec(&mddev->openers);
6686 static int md_media_changed(struct gendisk *disk)
6688 struct mddev *mddev = disk->private_data;
6690 return mddev->changed;
6693 static int md_revalidate(struct gendisk *disk)
6695 struct mddev *mddev = disk->private_data;
6700 static const struct block_device_operations md_fops =
6702 .owner = THIS_MODULE,
6704 .release = md_release,
6706 #ifdef CONFIG_COMPAT
6707 .compat_ioctl = md_compat_ioctl,
6709 .getgeo = md_getgeo,
6710 .media_changed = md_media_changed,
6711 .revalidate_disk= md_revalidate,
6714 static int md_thread(void * arg)
6716 struct md_thread *thread = arg;
6719 * md_thread is a 'system-thread', it's priority should be very
6720 * high. We avoid resource deadlocks individually in each
6721 * raid personality. (RAID5 does preallocation) We also use RR and
6722 * the very same RT priority as kswapd, thus we will never get
6723 * into a priority inversion deadlock.
6725 * we definitely have to have equal or higher priority than
6726 * bdflush, otherwise bdflush will deadlock if there are too
6727 * many dirty RAID5 blocks.
6730 allow_signal(SIGKILL);
6731 while (!kthread_should_stop()) {
6733 /* We need to wait INTERRUPTIBLE so that
6734 * we don't add to the load-average.
6735 * That means we need to be sure no signals are
6738 if (signal_pending(current))
6739 flush_signals(current);
6741 wait_event_interruptible_timeout
6743 test_bit(THREAD_WAKEUP, &thread->flags)
6744 || kthread_should_stop(),
6747 clear_bit(THREAD_WAKEUP, &thread->flags);
6748 if (!kthread_should_stop())
6749 thread->run(thread);
6755 void md_wakeup_thread(struct md_thread *thread)
6758 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6759 set_bit(THREAD_WAKEUP, &thread->flags);
6760 wake_up(&thread->wqueue);
6764 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6765 struct mddev *mddev, const char *name)
6767 struct md_thread *thread;
6769 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6773 init_waitqueue_head(&thread->wqueue);
6776 thread->mddev = mddev;
6777 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6778 thread->tsk = kthread_run(md_thread, thread,
6780 mdname(thread->mddev),
6782 if (IS_ERR(thread->tsk)) {
6789 void md_unregister_thread(struct md_thread **threadp)
6791 struct md_thread *thread = *threadp;
6794 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6795 /* Locking ensures that mddev_unlock does not wake_up a
6796 * non-existent thread
6798 spin_lock(&pers_lock);
6800 spin_unlock(&pers_lock);
6802 kthread_stop(thread->tsk);
6806 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6813 if (!rdev || test_bit(Faulty, &rdev->flags))
6816 if (!mddev->pers || !mddev->pers->error_handler)
6818 mddev->pers->error_handler(mddev,rdev);
6819 if (mddev->degraded)
6820 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6821 sysfs_notify_dirent_safe(rdev->sysfs_state);
6822 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6823 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6824 md_wakeup_thread(mddev->thread);
6825 if (mddev->event_work.func)
6826 queue_work(md_misc_wq, &mddev->event_work);
6827 md_new_event_inintr(mddev);
6830 /* seq_file implementation /proc/mdstat */
6832 static void status_unused(struct seq_file *seq)
6835 struct md_rdev *rdev;
6837 seq_printf(seq, "unused devices: ");
6839 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6840 char b[BDEVNAME_SIZE];
6842 seq_printf(seq, "%s ",
6843 bdevname(rdev->bdev,b));
6846 seq_printf(seq, "<none>");
6848 seq_printf(seq, "\n");
6852 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6854 sector_t max_sectors, resync, res;
6855 unsigned long dt, db;
6858 unsigned int per_milli;
6860 if (mddev->curr_resync <= 3)
6863 resync = mddev->curr_resync
6864 - atomic_read(&mddev->recovery_active);
6866 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6867 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6868 max_sectors = mddev->resync_max_sectors;
6870 max_sectors = mddev->dev_sectors;
6873 * Should not happen.
6879 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6880 * in a sector_t, and (max_sectors>>scale) will fit in a
6881 * u32, as those are the requirements for sector_div.
6882 * Thus 'scale' must be at least 10
6885 if (sizeof(sector_t) > sizeof(unsigned long)) {
6886 while ( max_sectors/2 > (1ULL<<(scale+32)))
6889 res = (resync>>scale)*1000;
6890 sector_div(res, (u32)((max_sectors>>scale)+1));
6894 int i, x = per_milli/50, y = 20-x;
6895 seq_printf(seq, "[");
6896 for (i = 0; i < x; i++)
6897 seq_printf(seq, "=");
6898 seq_printf(seq, ">");
6899 for (i = 0; i < y; i++)
6900 seq_printf(seq, ".");
6901 seq_printf(seq, "] ");
6903 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6904 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6906 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6908 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6909 "resync" : "recovery"))),
6910 per_milli/10, per_milli % 10,
6911 (unsigned long long) resync/2,
6912 (unsigned long long) max_sectors/2);
6915 * dt: time from mark until now
6916 * db: blocks written from mark until now
6917 * rt: remaining time
6919 * rt is a sector_t, so could be 32bit or 64bit.
6920 * So we divide before multiply in case it is 32bit and close
6922 * We scale the divisor (db) by 32 to avoid losing precision
6923 * near the end of resync when the number of remaining sectors
6925 * We then divide rt by 32 after multiplying by db to compensate.
6926 * The '+1' avoids division by zero if db is very small.
6928 dt = ((jiffies - mddev->resync_mark) / HZ);
6930 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6931 - mddev->resync_mark_cnt;
6933 rt = max_sectors - resync; /* number of remaining sectors */
6934 sector_div(rt, db/32+1);
6938 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6939 ((unsigned long)rt % 60)/6);
6941 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6944 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6946 struct list_head *tmp;
6948 struct mddev *mddev;
6956 spin_lock(&all_mddevs_lock);
6957 list_for_each(tmp,&all_mddevs)
6959 mddev = list_entry(tmp, struct mddev, all_mddevs);
6961 spin_unlock(&all_mddevs_lock);
6964 spin_unlock(&all_mddevs_lock);
6966 return (void*)2;/* tail */
6970 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6972 struct list_head *tmp;
6973 struct mddev *next_mddev, *mddev = v;
6979 spin_lock(&all_mddevs_lock);
6981 tmp = all_mddevs.next;
6983 tmp = mddev->all_mddevs.next;
6984 if (tmp != &all_mddevs)
6985 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6987 next_mddev = (void*)2;
6990 spin_unlock(&all_mddevs_lock);
6998 static void md_seq_stop(struct seq_file *seq, void *v)
7000 struct mddev *mddev = v;
7002 if (mddev && v != (void*)1 && v != (void*)2)
7006 static int md_seq_show(struct seq_file *seq, void *v)
7008 struct mddev *mddev = v;
7010 struct md_rdev *rdev;
7012 if (v == (void*)1) {
7013 struct md_personality *pers;
7014 seq_printf(seq, "Personalities : ");
7015 spin_lock(&pers_lock);
7016 list_for_each_entry(pers, &pers_list, list)
7017 seq_printf(seq, "[%s] ", pers->name);
7019 spin_unlock(&pers_lock);
7020 seq_printf(seq, "\n");
7021 seq->poll_event = atomic_read(&md_event_count);
7024 if (v == (void*)2) {
7029 if (mddev_lock(mddev) < 0)
7032 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7033 seq_printf(seq, "%s : %sactive", mdname(mddev),
7034 mddev->pers ? "" : "in");
7037 seq_printf(seq, " (read-only)");
7039 seq_printf(seq, " (auto-read-only)");
7040 seq_printf(seq, " %s", mddev->pers->name);
7044 rdev_for_each(rdev, mddev) {
7045 char b[BDEVNAME_SIZE];
7046 seq_printf(seq, " %s[%d]",
7047 bdevname(rdev->bdev,b), rdev->desc_nr);
7048 if (test_bit(WriteMostly, &rdev->flags))
7049 seq_printf(seq, "(W)");
7050 if (test_bit(Faulty, &rdev->flags)) {
7051 seq_printf(seq, "(F)");
7054 if (rdev->raid_disk < 0)
7055 seq_printf(seq, "(S)"); /* spare */
7056 if (test_bit(Replacement, &rdev->flags))
7057 seq_printf(seq, "(R)");
7058 sectors += rdev->sectors;
7061 if (!list_empty(&mddev->disks)) {
7063 seq_printf(seq, "\n %llu blocks",
7064 (unsigned long long)
7065 mddev->array_sectors / 2);
7067 seq_printf(seq, "\n %llu blocks",
7068 (unsigned long long)sectors / 2);
7070 if (mddev->persistent) {
7071 if (mddev->major_version != 0 ||
7072 mddev->minor_version != 90) {
7073 seq_printf(seq," super %d.%d",
7074 mddev->major_version,
7075 mddev->minor_version);
7077 } else if (mddev->external)
7078 seq_printf(seq, " super external:%s",
7079 mddev->metadata_type);
7081 seq_printf(seq, " super non-persistent");
7084 mddev->pers->status(seq, mddev);
7085 seq_printf(seq, "\n ");
7086 if (mddev->pers->sync_request) {
7087 if (mddev->curr_resync > 2) {
7088 status_resync(seq, mddev);
7089 seq_printf(seq, "\n ");
7090 } else if (mddev->curr_resync >= 1)
7091 seq_printf(seq, "\tresync=DELAYED\n ");
7092 else if (mddev->recovery_cp < MaxSector)
7093 seq_printf(seq, "\tresync=PENDING\n ");
7096 seq_printf(seq, "\n ");
7098 bitmap_status(seq, mddev->bitmap);
7100 seq_printf(seq, "\n");
7102 mddev_unlock(mddev);
7107 static const struct seq_operations md_seq_ops = {
7108 .start = md_seq_start,
7109 .next = md_seq_next,
7110 .stop = md_seq_stop,
7111 .show = md_seq_show,
7114 static int md_seq_open(struct inode *inode, struct file *file)
7116 struct seq_file *seq;
7119 error = seq_open(file, &md_seq_ops);
7123 seq = file->private_data;
7124 seq->poll_event = atomic_read(&md_event_count);
7128 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7130 struct seq_file *seq = filp->private_data;
7133 poll_wait(filp, &md_event_waiters, wait);
7135 /* always allow read */
7136 mask = POLLIN | POLLRDNORM;
7138 if (seq->poll_event != atomic_read(&md_event_count))
7139 mask |= POLLERR | POLLPRI;
7143 static const struct file_operations md_seq_fops = {
7144 .owner = THIS_MODULE,
7145 .open = md_seq_open,
7147 .llseek = seq_lseek,
7148 .release = seq_release_private,
7149 .poll = mdstat_poll,
7152 int register_md_personality(struct md_personality *p)
7154 spin_lock(&pers_lock);
7155 list_add_tail(&p->list, &pers_list);
7156 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7157 spin_unlock(&pers_lock);
7161 int unregister_md_personality(struct md_personality *p)
7163 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7164 spin_lock(&pers_lock);
7165 list_del_init(&p->list);
7166 spin_unlock(&pers_lock);
7170 static int is_mddev_idle(struct mddev *mddev, int init)
7172 struct md_rdev * rdev;
7178 rdev_for_each_rcu(rdev, mddev) {
7179 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7180 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7181 (int)part_stat_read(&disk->part0, sectors[1]) -
7182 atomic_read(&disk->sync_io);
7183 /* sync IO will cause sync_io to increase before the disk_stats
7184 * as sync_io is counted when a request starts, and
7185 * disk_stats is counted when it completes.
7186 * So resync activity will cause curr_events to be smaller than
7187 * when there was no such activity.
7188 * non-sync IO will cause disk_stat to increase without
7189 * increasing sync_io so curr_events will (eventually)
7190 * be larger than it was before. Once it becomes
7191 * substantially larger, the test below will cause
7192 * the array to appear non-idle, and resync will slow
7194 * If there is a lot of outstanding resync activity when
7195 * we set last_event to curr_events, then all that activity
7196 * completing might cause the array to appear non-idle
7197 * and resync will be slowed down even though there might
7198 * not have been non-resync activity. This will only
7199 * happen once though. 'last_events' will soon reflect
7200 * the state where there is little or no outstanding
7201 * resync requests, and further resync activity will
7202 * always make curr_events less than last_events.
7205 if (init || curr_events - rdev->last_events > 64) {
7206 rdev->last_events = curr_events;
7214 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7216 /* another "blocks" (512byte) blocks have been synced */
7217 atomic_sub(blocks, &mddev->recovery_active);
7218 wake_up(&mddev->recovery_wait);
7220 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7221 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7222 md_wakeup_thread(mddev->thread);
7223 // stop recovery, signal do_sync ....
7228 /* md_write_start(mddev, bi)
7229 * If we need to update some array metadata (e.g. 'active' flag
7230 * in superblock) before writing, schedule a superblock update
7231 * and wait for it to complete.
7233 void md_write_start(struct mddev *mddev, struct bio *bi)
7236 if (bio_data_dir(bi) != WRITE)
7239 BUG_ON(mddev->ro == 1);
7240 if (mddev->ro == 2) {
7241 /* need to switch to read/write */
7243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7244 md_wakeup_thread(mddev->thread);
7245 md_wakeup_thread(mddev->sync_thread);
7248 atomic_inc(&mddev->writes_pending);
7249 if (mddev->safemode == 1)
7250 mddev->safemode = 0;
7251 if (mddev->in_sync) {
7252 spin_lock_irq(&mddev->write_lock);
7253 if (mddev->in_sync) {
7255 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7256 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7257 md_wakeup_thread(mddev->thread);
7260 spin_unlock_irq(&mddev->write_lock);
7263 sysfs_notify_dirent_safe(mddev->sysfs_state);
7264 wait_event(mddev->sb_wait,
7265 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7268 void md_write_end(struct mddev *mddev)
7270 if (atomic_dec_and_test(&mddev->writes_pending)) {
7271 if (mddev->safemode == 2)
7272 md_wakeup_thread(mddev->thread);
7273 else if (mddev->safemode_delay)
7274 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7278 /* md_allow_write(mddev)
7279 * Calling this ensures that the array is marked 'active' so that writes
7280 * may proceed without blocking. It is important to call this before
7281 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7282 * Must be called with mddev_lock held.
7284 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7285 * is dropped, so return -EAGAIN after notifying userspace.
7287 int md_allow_write(struct mddev *mddev)
7293 if (!mddev->pers->sync_request)
7296 spin_lock_irq(&mddev->write_lock);
7297 if (mddev->in_sync) {
7299 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7300 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7301 if (mddev->safemode_delay &&
7302 mddev->safemode == 0)
7303 mddev->safemode = 1;
7304 spin_unlock_irq(&mddev->write_lock);
7305 md_update_sb(mddev, 0);
7306 sysfs_notify_dirent_safe(mddev->sysfs_state);
7308 spin_unlock_irq(&mddev->write_lock);
7310 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7315 EXPORT_SYMBOL_GPL(md_allow_write);
7317 #define SYNC_MARKS 10
7318 #define SYNC_MARK_STEP (3*HZ)
7319 #define UPDATE_FREQUENCY (5*60*HZ)
7320 void md_do_sync(struct md_thread *thread)
7322 struct mddev *mddev = thread->mddev;
7323 struct mddev *mddev2;
7324 unsigned int currspeed = 0,
7326 sector_t max_sectors,j, io_sectors;
7327 unsigned long mark[SYNC_MARKS];
7328 unsigned long update_time;
7329 sector_t mark_cnt[SYNC_MARKS];
7331 struct list_head *tmp;
7332 sector_t last_check;
7334 struct md_rdev *rdev;
7336 struct blk_plug plug;
7338 /* just incase thread restarts... */
7339 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7341 if (mddev->ro) {/* never try to sync a read-only array */
7342 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7346 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7347 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7348 desc = "data-check";
7349 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7350 desc = "requested-resync";
7353 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7358 /* we overload curr_resync somewhat here.
7359 * 0 == not engaged in resync at all
7360 * 2 == checking that there is no conflict with another sync
7361 * 1 == like 2, but have yielded to allow conflicting resync to
7363 * other == active in resync - this many blocks
7365 * Before starting a resync we must have set curr_resync to
7366 * 2, and then checked that every "conflicting" array has curr_resync
7367 * less than ours. When we find one that is the same or higher
7368 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7369 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7370 * This will mean we have to start checking from the beginning again.
7375 mddev->curr_resync = 2;
7378 if (kthread_should_stop())
7379 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7381 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7383 for_each_mddev(mddev2, tmp) {
7384 if (mddev2 == mddev)
7386 if (!mddev->parallel_resync
7387 && mddev2->curr_resync
7388 && match_mddev_units(mddev, mddev2)) {
7390 if (mddev < mddev2 && mddev->curr_resync == 2) {
7391 /* arbitrarily yield */
7392 mddev->curr_resync = 1;
7393 wake_up(&resync_wait);
7395 if (mddev > mddev2 && mddev->curr_resync == 1)
7396 /* no need to wait here, we can wait the next
7397 * time 'round when curr_resync == 2
7400 /* We need to wait 'interruptible' so as not to
7401 * contribute to the load average, and not to
7402 * be caught by 'softlockup'
7404 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7405 if (!kthread_should_stop() &&
7406 mddev2->curr_resync >= mddev->curr_resync) {
7407 printk(KERN_INFO "md: delaying %s of %s"
7408 " until %s has finished (they"
7409 " share one or more physical units)\n",
7410 desc, mdname(mddev), mdname(mddev2));
7412 if (signal_pending(current))
7413 flush_signals(current);
7415 finish_wait(&resync_wait, &wq);
7418 finish_wait(&resync_wait, &wq);
7421 } while (mddev->curr_resync < 2);
7424 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7425 /* resync follows the size requested by the personality,
7426 * which defaults to physical size, but can be virtual size
7428 max_sectors = mddev->resync_max_sectors;
7429 atomic64_set(&mddev->resync_mismatches, 0);
7430 /* we don't use the checkpoint if there's a bitmap */
7431 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7432 j = mddev->resync_min;
7433 else if (!mddev->bitmap)
7434 j = mddev->recovery_cp;
7436 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7437 max_sectors = mddev->resync_max_sectors;
7439 /* recovery follows the physical size of devices */
7440 max_sectors = mddev->dev_sectors;
7443 rdev_for_each_rcu(rdev, mddev)
7444 if (rdev->raid_disk >= 0 &&
7445 !test_bit(Faulty, &rdev->flags) &&
7446 !test_bit(In_sync, &rdev->flags) &&
7447 rdev->recovery_offset < j)
7448 j = rdev->recovery_offset;
7451 /* If there is a bitmap, we need to make sure all
7452 * writes that started before we added a spare
7453 * complete before we start doing a recovery.
7454 * Otherwise the write might complete and (via
7455 * bitmap_endwrite) set a bit in the bitmap after the
7456 * recovery has checked that bit and skipped that
7459 if (mddev->bitmap) {
7460 mddev->pers->quiesce(mddev, 1);
7461 mddev->pers->quiesce(mddev, 0);
7465 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7466 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7467 " %d KB/sec/disk.\n", speed_min(mddev));
7468 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7469 "(but not more than %d KB/sec) for %s.\n",
7470 speed_max(mddev), desc);
7472 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7475 for (m = 0; m < SYNC_MARKS; m++) {
7477 mark_cnt[m] = io_sectors;
7480 mddev->resync_mark = mark[last_mark];
7481 mddev->resync_mark_cnt = mark_cnt[last_mark];
7484 * Tune reconstruction:
7486 window = 32*(PAGE_SIZE/512);
7487 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7488 window/2, (unsigned long long)max_sectors/2);
7490 atomic_set(&mddev->recovery_active, 0);
7495 "md: resuming %s of %s from checkpoint.\n",
7496 desc, mdname(mddev));
7497 mddev->curr_resync = j;
7499 mddev->curr_resync = 3; /* no longer delayed */
7500 mddev->curr_resync_completed = j;
7501 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7502 md_new_event(mddev);
7503 update_time = jiffies;
7505 blk_start_plug(&plug);
7506 while (j < max_sectors) {
7511 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7512 ((mddev->curr_resync > mddev->curr_resync_completed &&
7513 (mddev->curr_resync - mddev->curr_resync_completed)
7514 > (max_sectors >> 4)) ||
7515 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7516 (j - mddev->curr_resync_completed)*2
7517 >= mddev->resync_max - mddev->curr_resync_completed
7519 /* time to update curr_resync_completed */
7520 wait_event(mddev->recovery_wait,
7521 atomic_read(&mddev->recovery_active) == 0);
7522 mddev->curr_resync_completed = j;
7523 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7524 j > mddev->recovery_cp)
7525 mddev->recovery_cp = j;
7526 update_time = jiffies;
7527 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7528 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7531 while (j >= mddev->resync_max && !kthread_should_stop()) {
7532 /* As this condition is controlled by user-space,
7533 * we can block indefinitely, so use '_interruptible'
7534 * to avoid triggering warnings.
7536 flush_signals(current); /* just in case */
7537 wait_event_interruptible(mddev->recovery_wait,
7538 mddev->resync_max > j
7539 || kthread_should_stop());
7542 if (kthread_should_stop())
7545 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7546 currspeed < speed_min(mddev));
7548 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7552 if (!skipped) { /* actual IO requested */
7553 io_sectors += sectors;
7554 atomic_add(sectors, &mddev->recovery_active);
7557 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7562 mddev->curr_resync = j;
7563 mddev->curr_mark_cnt = io_sectors;
7564 if (last_check == 0)
7565 /* this is the earliest that rebuild will be
7566 * visible in /proc/mdstat
7568 md_new_event(mddev);
7570 if (last_check + window > io_sectors || j == max_sectors)
7573 last_check = io_sectors;
7575 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7577 int next = (last_mark+1) % SYNC_MARKS;
7579 mddev->resync_mark = mark[next];
7580 mddev->resync_mark_cnt = mark_cnt[next];
7581 mark[next] = jiffies;
7582 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7587 if (kthread_should_stop())
7592 * this loop exits only if either when we are slower than
7593 * the 'hard' speed limit, or the system was IO-idle for
7595 * the system might be non-idle CPU-wise, but we only care
7596 * about not overloading the IO subsystem. (things like an
7597 * e2fsck being done on the RAID array should execute fast)
7601 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7602 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7604 if (currspeed > speed_min(mddev)) {
7605 if ((currspeed > speed_max(mddev)) ||
7606 !is_mddev_idle(mddev, 0)) {
7612 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7614 * this also signals 'finished resyncing' to md_stop
7617 blk_finish_plug(&plug);
7618 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7620 /* tell personality that we are finished */
7621 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7623 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7624 mddev->curr_resync > 2) {
7625 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7626 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7627 if (mddev->curr_resync >= mddev->recovery_cp) {
7629 "md: checkpointing %s of %s.\n",
7630 desc, mdname(mddev));
7631 if (test_bit(MD_RECOVERY_ERROR,
7633 mddev->recovery_cp =
7634 mddev->curr_resync_completed;
7636 mddev->recovery_cp =
7640 mddev->recovery_cp = MaxSector;
7642 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7643 mddev->curr_resync = MaxSector;
7645 rdev_for_each_rcu(rdev, mddev)
7646 if (rdev->raid_disk >= 0 &&
7647 mddev->delta_disks >= 0 &&
7648 !test_bit(Faulty, &rdev->flags) &&
7649 !test_bit(In_sync, &rdev->flags) &&
7650 rdev->recovery_offset < mddev->curr_resync)
7651 rdev->recovery_offset = mddev->curr_resync;
7656 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7658 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7659 /* We completed so min/max setting can be forgotten if used. */
7660 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7661 mddev->resync_min = 0;
7662 mddev->resync_max = MaxSector;
7663 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7664 mddev->resync_min = mddev->curr_resync_completed;
7665 mddev->curr_resync = 0;
7666 wake_up(&resync_wait);
7667 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7668 md_wakeup_thread(mddev->thread);
7673 * got a signal, exit.
7676 "md: md_do_sync() got signal ... exiting\n");
7677 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7681 EXPORT_SYMBOL_GPL(md_do_sync);
7683 static int remove_and_add_spares(struct mddev *mddev,
7684 struct md_rdev *this)
7686 struct md_rdev *rdev;
7690 rdev_for_each(rdev, mddev)
7691 if ((this == NULL || rdev == this) &&
7692 rdev->raid_disk >= 0 &&
7693 !test_bit(Blocked, &rdev->flags) &&
7694 (test_bit(Faulty, &rdev->flags) ||
7695 ! test_bit(In_sync, &rdev->flags)) &&
7696 atomic_read(&rdev->nr_pending)==0) {
7697 if (mddev->pers->hot_remove_disk(
7698 mddev, rdev) == 0) {
7699 sysfs_unlink_rdev(mddev, rdev);
7700 rdev->raid_disk = -1;
7704 if (removed && mddev->kobj.sd)
7705 sysfs_notify(&mddev->kobj, NULL, "degraded");
7710 rdev_for_each(rdev, mddev) {
7711 if (rdev->raid_disk >= 0 &&
7712 !test_bit(In_sync, &rdev->flags) &&
7713 !test_bit(Faulty, &rdev->flags))
7715 if (rdev->raid_disk >= 0)
7717 if (test_bit(Faulty, &rdev->flags))
7720 ! (rdev->saved_raid_disk >= 0 &&
7721 !test_bit(Bitmap_sync, &rdev->flags)))
7724 rdev->recovery_offset = 0;
7726 hot_add_disk(mddev, rdev) == 0) {
7727 if (sysfs_link_rdev(mddev, rdev))
7728 /* failure here is OK */;
7730 md_new_event(mddev);
7731 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7736 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7741 * This routine is regularly called by all per-raid-array threads to
7742 * deal with generic issues like resync and super-block update.
7743 * Raid personalities that don't have a thread (linear/raid0) do not
7744 * need this as they never do any recovery or update the superblock.
7746 * It does not do any resync itself, but rather "forks" off other threads
7747 * to do that as needed.
7748 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7749 * "->recovery" and create a thread at ->sync_thread.
7750 * When the thread finishes it sets MD_RECOVERY_DONE
7751 * and wakeups up this thread which will reap the thread and finish up.
7752 * This thread also removes any faulty devices (with nr_pending == 0).
7754 * The overall approach is:
7755 * 1/ if the superblock needs updating, update it.
7756 * 2/ If a recovery thread is running, don't do anything else.
7757 * 3/ If recovery has finished, clean up, possibly marking spares active.
7758 * 4/ If there are any faulty devices, remove them.
7759 * 5/ If array is degraded, try to add spares devices
7760 * 6/ If array has spares or is not in-sync, start a resync thread.
7762 void md_check_recovery(struct mddev *mddev)
7764 if (mddev->suspended)
7768 bitmap_daemon_work(mddev);
7770 if (signal_pending(current)) {
7771 if (mddev->pers->sync_request && !mddev->external) {
7772 printk(KERN_INFO "md: %s in immediate safe mode\n",
7774 mddev->safemode = 2;
7776 flush_signals(current);
7779 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7782 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7783 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7784 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7785 (mddev->external == 0 && mddev->safemode == 1) ||
7786 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7787 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7791 if (mddev_trylock(mddev)) {
7795 /* On a read-only array we can:
7796 * - remove failed devices
7797 * - add already-in_sync devices if the array itself
7799 * As we only add devices that are already in-sync,
7800 * we can activate the spares immediately.
7802 remove_and_add_spares(mddev, NULL);
7803 /* There is no thread, but we need to call
7804 * ->spare_active and clear saved_raid_disk
7806 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7807 md_reap_sync_thread(mddev);
7808 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7812 if (!mddev->external) {
7814 spin_lock_irq(&mddev->write_lock);
7815 if (mddev->safemode &&
7816 !atomic_read(&mddev->writes_pending) &&
7818 mddev->recovery_cp == MaxSector) {
7821 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7823 if (mddev->safemode == 1)
7824 mddev->safemode = 0;
7825 spin_unlock_irq(&mddev->write_lock);
7827 sysfs_notify_dirent_safe(mddev->sysfs_state);
7831 md_update_sb(mddev, 0);
7833 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7834 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7835 /* resync/recovery still happening */
7836 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7839 if (mddev->sync_thread) {
7840 md_reap_sync_thread(mddev);
7843 /* Set RUNNING before clearing NEEDED to avoid
7844 * any transients in the value of "sync_action".
7846 mddev->curr_resync_completed = 0;
7847 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7848 /* Clear some bits that don't mean anything, but
7851 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7852 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7854 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7855 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7857 /* no recovery is running.
7858 * remove any failed drives, then
7859 * add spares if possible.
7860 * Spares are also removed and re-added, to allow
7861 * the personality to fail the re-add.
7864 if (mddev->reshape_position != MaxSector) {
7865 if (mddev->pers->check_reshape == NULL ||
7866 mddev->pers->check_reshape(mddev) != 0)
7867 /* Cannot proceed */
7869 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7870 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7871 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7872 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7873 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7874 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7875 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7876 } else if (mddev->recovery_cp < MaxSector) {
7877 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7878 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7879 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7880 /* nothing to be done ... */
7883 if (mddev->pers->sync_request) {
7885 /* We are adding a device or devices to an array
7886 * which has the bitmap stored on all devices.
7887 * So make sure all bitmap pages get written
7889 bitmap_write_all(mddev->bitmap);
7891 mddev->sync_thread = md_register_thread(md_do_sync,
7894 if (!mddev->sync_thread) {
7895 printk(KERN_ERR "%s: could not start resync"
7898 /* leave the spares where they are, it shouldn't hurt */
7899 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7900 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7901 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7902 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7903 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7905 md_wakeup_thread(mddev->sync_thread);
7906 sysfs_notify_dirent_safe(mddev->sysfs_action);
7907 md_new_event(mddev);
7910 if (!mddev->sync_thread) {
7911 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7912 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7914 if (mddev->sysfs_action)
7915 sysfs_notify_dirent_safe(mddev->sysfs_action);
7917 mddev_unlock(mddev);
7921 void md_reap_sync_thread(struct mddev *mddev)
7923 struct md_rdev *rdev;
7925 /* resync has finished, collect result */
7926 md_unregister_thread(&mddev->sync_thread);
7927 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7928 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7930 /* activate any spares */
7931 if (mddev->pers->spare_active(mddev)) {
7932 sysfs_notify(&mddev->kobj, NULL,
7934 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7937 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7938 mddev->pers->finish_reshape)
7939 mddev->pers->finish_reshape(mddev);
7941 /* If array is no-longer degraded, then any saved_raid_disk
7942 * information must be scrapped. Also if any device is now
7943 * In_sync we must scrape the saved_raid_disk for that device
7944 * do the superblock for an incrementally recovered device
7947 rdev_for_each(rdev, mddev)
7948 if (!mddev->degraded ||
7949 test_bit(In_sync, &rdev->flags))
7950 rdev->saved_raid_disk = -1;
7952 md_update_sb(mddev, 1);
7953 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7954 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7955 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7956 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7957 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7958 /* flag recovery needed just to double check */
7959 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7960 sysfs_notify_dirent_safe(mddev->sysfs_action);
7961 md_new_event(mddev);
7962 if (mddev->event_work.func)
7963 queue_work(md_misc_wq, &mddev->event_work);
7966 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7968 sysfs_notify_dirent_safe(rdev->sysfs_state);
7969 wait_event_timeout(rdev->blocked_wait,
7970 !test_bit(Blocked, &rdev->flags) &&
7971 !test_bit(BlockedBadBlocks, &rdev->flags),
7972 msecs_to_jiffies(5000));
7973 rdev_dec_pending(rdev, mddev);
7975 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7977 void md_finish_reshape(struct mddev *mddev)
7979 /* called be personality module when reshape completes. */
7980 struct md_rdev *rdev;
7982 rdev_for_each(rdev, mddev) {
7983 if (rdev->data_offset > rdev->new_data_offset)
7984 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7986 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7987 rdev->data_offset = rdev->new_data_offset;
7990 EXPORT_SYMBOL(md_finish_reshape);
7992 /* Bad block management.
7993 * We can record which blocks on each device are 'bad' and so just
7994 * fail those blocks, or that stripe, rather than the whole device.
7995 * Entries in the bad-block table are 64bits wide. This comprises:
7996 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7997 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7998 * A 'shift' can be set so that larger blocks are tracked and
7999 * consequently larger devices can be covered.
8000 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8002 * Locking of the bad-block table uses a seqlock so md_is_badblock
8003 * might need to retry if it is very unlucky.
8004 * We will sometimes want to check for bad blocks in a bi_end_io function,
8005 * so we use the write_seqlock_irq variant.
8007 * When looking for a bad block we specify a range and want to
8008 * know if any block in the range is bad. So we binary-search
8009 * to the last range that starts at-or-before the given endpoint,
8010 * (or "before the sector after the target range")
8011 * then see if it ends after the given start.
8013 * 0 if there are no known bad blocks in the range
8014 * 1 if there are known bad block which are all acknowledged
8015 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8016 * plus the start/length of the first bad section we overlap.
8018 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8019 sector_t *first_bad, int *bad_sectors)
8025 sector_t target = s + sectors;
8028 if (bb->shift > 0) {
8029 /* round the start down, and the end up */
8031 target += (1<<bb->shift) - 1;
8032 target >>= bb->shift;
8033 sectors = target - s;
8035 /* 'target' is now the first block after the bad range */
8038 seq = read_seqbegin(&bb->lock);
8043 /* Binary search between lo and hi for 'target'
8044 * i.e. for the last range that starts before 'target'
8046 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8047 * are known not to be the last range before target.
8048 * VARIANT: hi-lo is the number of possible
8049 * ranges, and decreases until it reaches 1
8051 while (hi - lo > 1) {
8052 int mid = (lo + hi) / 2;
8053 sector_t a = BB_OFFSET(p[mid]);
8055 /* This could still be the one, earlier ranges
8059 /* This and later ranges are definitely out. */
8062 /* 'lo' might be the last that started before target, but 'hi' isn't */
8064 /* need to check all range that end after 's' to see if
8065 * any are unacknowledged.
8068 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8069 if (BB_OFFSET(p[lo]) < target) {
8070 /* starts before the end, and finishes after
8071 * the start, so they must overlap
8073 if (rv != -1 && BB_ACK(p[lo]))
8077 *first_bad = BB_OFFSET(p[lo]);
8078 *bad_sectors = BB_LEN(p[lo]);
8084 if (read_seqretry(&bb->lock, seq))
8089 EXPORT_SYMBOL_GPL(md_is_badblock);
8092 * Add a range of bad blocks to the table.
8093 * This might extend the table, or might contract it
8094 * if two adjacent ranges can be merged.
8095 * We binary-search to find the 'insertion' point, then
8096 * decide how best to handle it.
8098 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8104 unsigned long flags;
8107 /* badblocks are disabled */
8111 /* round the start down, and the end up */
8112 sector_t next = s + sectors;
8114 next += (1<<bb->shift) - 1;
8119 write_seqlock_irqsave(&bb->lock, flags);
8124 /* Find the last range that starts at-or-before 's' */
8125 while (hi - lo > 1) {
8126 int mid = (lo + hi) / 2;
8127 sector_t a = BB_OFFSET(p[mid]);
8133 if (hi > lo && BB_OFFSET(p[lo]) > s)
8137 /* we found a range that might merge with the start
8140 sector_t a = BB_OFFSET(p[lo]);
8141 sector_t e = a + BB_LEN(p[lo]);
8142 int ack = BB_ACK(p[lo]);
8144 /* Yes, we can merge with a previous range */
8145 if (s == a && s + sectors >= e)
8146 /* new range covers old */
8149 ack = ack && acknowledged;
8151 if (e < s + sectors)
8153 if (e - a <= BB_MAX_LEN) {
8154 p[lo] = BB_MAKE(a, e-a, ack);
8157 /* does not all fit in one range,
8158 * make p[lo] maximal
8160 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8161 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8167 if (sectors && hi < bb->count) {
8168 /* 'hi' points to the first range that starts after 's'.
8169 * Maybe we can merge with the start of that range */
8170 sector_t a = BB_OFFSET(p[hi]);
8171 sector_t e = a + BB_LEN(p[hi]);
8172 int ack = BB_ACK(p[hi]);
8173 if (a <= s + sectors) {
8174 /* merging is possible */
8175 if (e <= s + sectors) {
8180 ack = ack && acknowledged;
8183 if (e - a <= BB_MAX_LEN) {
8184 p[hi] = BB_MAKE(a, e-a, ack);
8187 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8195 if (sectors == 0 && hi < bb->count) {
8196 /* we might be able to combine lo and hi */
8197 /* Note: 's' is at the end of 'lo' */
8198 sector_t a = BB_OFFSET(p[hi]);
8199 int lolen = BB_LEN(p[lo]);
8200 int hilen = BB_LEN(p[hi]);
8201 int newlen = lolen + hilen - (s - a);
8202 if (s >= a && newlen < BB_MAX_LEN) {
8203 /* yes, we can combine them */
8204 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8205 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8206 memmove(p + hi, p + hi + 1,
8207 (bb->count - hi - 1) * 8);
8212 /* didn't merge (it all).
8213 * Need to add a range just before 'hi' */
8214 if (bb->count >= MD_MAX_BADBLOCKS) {
8215 /* No room for more */
8219 int this_sectors = sectors;
8220 memmove(p + hi + 1, p + hi,
8221 (bb->count - hi) * 8);
8224 if (this_sectors > BB_MAX_LEN)
8225 this_sectors = BB_MAX_LEN;
8226 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8227 sectors -= this_sectors;
8234 bb->unacked_exist = 1;
8235 write_sequnlock_irqrestore(&bb->lock, flags);
8240 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8245 s += rdev->new_data_offset;
8247 s += rdev->data_offset;
8248 rv = md_set_badblocks(&rdev->badblocks,
8251 /* Make sure they get written out promptly */
8252 sysfs_notify_dirent_safe(rdev->sysfs_state);
8253 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8254 md_wakeup_thread(rdev->mddev->thread);
8258 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8261 * Remove a range of bad blocks from the table.
8262 * This may involve extending the table if we spilt a region,
8263 * but it must not fail. So if the table becomes full, we just
8264 * drop the remove request.
8266 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8270 sector_t target = s + sectors;
8273 if (bb->shift > 0) {
8274 /* When clearing we round the start up and the end down.
8275 * This should not matter as the shift should align with
8276 * the block size and no rounding should ever be needed.
8277 * However it is better the think a block is bad when it
8278 * isn't than to think a block is not bad when it is.
8280 s += (1<<bb->shift) - 1;
8282 target >>= bb->shift;
8283 sectors = target - s;
8286 write_seqlock_irq(&bb->lock);
8291 /* Find the last range that starts before 'target' */
8292 while (hi - lo > 1) {
8293 int mid = (lo + hi) / 2;
8294 sector_t a = BB_OFFSET(p[mid]);
8301 /* p[lo] is the last range that could overlap the
8302 * current range. Earlier ranges could also overlap,
8303 * but only this one can overlap the end of the range.
8305 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8306 /* Partial overlap, leave the tail of this range */
8307 int ack = BB_ACK(p[lo]);
8308 sector_t a = BB_OFFSET(p[lo]);
8309 sector_t end = a + BB_LEN(p[lo]);
8312 /* we need to split this range */
8313 if (bb->count >= MD_MAX_BADBLOCKS) {
8317 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8319 p[lo] = BB_MAKE(a, s-a, ack);
8322 p[lo] = BB_MAKE(target, end - target, ack);
8323 /* there is no longer an overlap */
8328 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8329 /* This range does overlap */
8330 if (BB_OFFSET(p[lo]) < s) {
8331 /* Keep the early parts of this range. */
8332 int ack = BB_ACK(p[lo]);
8333 sector_t start = BB_OFFSET(p[lo]);
8334 p[lo] = BB_MAKE(start, s - start, ack);
8335 /* now low doesn't overlap, so.. */
8340 /* 'lo' is strictly before, 'hi' is strictly after,
8341 * anything between needs to be discarded
8344 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8345 bb->count -= (hi - lo - 1);
8351 write_sequnlock_irq(&bb->lock);
8355 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8359 s += rdev->new_data_offset;
8361 s += rdev->data_offset;
8362 return md_clear_badblocks(&rdev->badblocks,
8365 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8368 * Acknowledge all bad blocks in a list.
8369 * This only succeeds if ->changed is clear. It is used by
8370 * in-kernel metadata updates
8372 void md_ack_all_badblocks(struct badblocks *bb)
8374 if (bb->page == NULL || bb->changed)
8375 /* no point even trying */
8377 write_seqlock_irq(&bb->lock);
8379 if (bb->changed == 0 && bb->unacked_exist) {
8382 for (i = 0; i < bb->count ; i++) {
8383 if (!BB_ACK(p[i])) {
8384 sector_t start = BB_OFFSET(p[i]);
8385 int len = BB_LEN(p[i]);
8386 p[i] = BB_MAKE(start, len, 1);
8389 bb->unacked_exist = 0;
8391 write_sequnlock_irq(&bb->lock);
8393 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8395 /* sysfs access to bad-blocks list.
8396 * We present two files.
8397 * 'bad-blocks' lists sector numbers and lengths of ranges that
8398 * are recorded as bad. The list is truncated to fit within
8399 * the one-page limit of sysfs.
8400 * Writing "sector length" to this file adds an acknowledged
8402 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8403 * been acknowledged. Writing to this file adds bad blocks
8404 * without acknowledging them. This is largely for testing.
8408 badblocks_show(struct badblocks *bb, char *page, int unack)
8419 seq = read_seqbegin(&bb->lock);
8424 while (len < PAGE_SIZE && i < bb->count) {
8425 sector_t s = BB_OFFSET(p[i]);
8426 unsigned int length = BB_LEN(p[i]);
8427 int ack = BB_ACK(p[i]);
8433 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8434 (unsigned long long)s << bb->shift,
8435 length << bb->shift);
8437 if (unack && len == 0)
8438 bb->unacked_exist = 0;
8440 if (read_seqretry(&bb->lock, seq))
8449 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8451 unsigned long long sector;
8455 /* Allow clearing via sysfs *only* for testing/debugging.
8456 * Normally only a successful write may clear a badblock
8459 if (page[0] == '-') {
8463 #endif /* DO_DEBUG */
8465 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8467 if (newline != '\n')
8479 md_clear_badblocks(bb, sector, length);
8482 #endif /* DO_DEBUG */
8483 if (md_set_badblocks(bb, sector, length, !unack))
8489 static int md_notify_reboot(struct notifier_block *this,
8490 unsigned long code, void *x)
8492 struct list_head *tmp;
8493 struct mddev *mddev;
8496 for_each_mddev(mddev, tmp) {
8497 if (mddev_trylock(mddev)) {
8499 __md_stop_writes(mddev);
8500 if (mddev->persistent)
8501 mddev->safemode = 2;
8502 mddev_unlock(mddev);
8507 * certain more exotic SCSI devices are known to be
8508 * volatile wrt too early system reboots. While the
8509 * right place to handle this issue is the given
8510 * driver, we do want to have a safe RAID driver ...
8518 static struct notifier_block md_notifier = {
8519 .notifier_call = md_notify_reboot,
8521 .priority = INT_MAX, /* before any real devices */
8524 static void md_geninit(void)
8526 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8528 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8531 static int __init md_init(void)
8535 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8539 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8543 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8546 if ((ret = register_blkdev(0, "mdp")) < 0)
8550 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8551 md_probe, NULL, NULL);
8552 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8553 md_probe, NULL, NULL);
8555 register_reboot_notifier(&md_notifier);
8556 raid_table_header = register_sysctl_table(raid_root_table);
8562 unregister_blkdev(MD_MAJOR, "md");
8564 destroy_workqueue(md_misc_wq);
8566 destroy_workqueue(md_wq);
8574 * Searches all registered partitions for autorun RAID arrays
8578 static LIST_HEAD(all_detected_devices);
8579 struct detected_devices_node {
8580 struct list_head list;
8584 void md_autodetect_dev(dev_t dev)
8586 struct detected_devices_node *node_detected_dev;
8588 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8589 if (node_detected_dev) {
8590 node_detected_dev->dev = dev;
8591 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8593 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8594 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8599 static void autostart_arrays(int part)
8601 struct md_rdev *rdev;
8602 struct detected_devices_node *node_detected_dev;
8604 int i_scanned, i_passed;
8609 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8611 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8613 node_detected_dev = list_entry(all_detected_devices.next,
8614 struct detected_devices_node, list);
8615 list_del(&node_detected_dev->list);
8616 dev = node_detected_dev->dev;
8617 kfree(node_detected_dev);
8618 rdev = md_import_device(dev,0, 90);
8622 if (test_bit(Faulty, &rdev->flags)) {
8626 set_bit(AutoDetected, &rdev->flags);
8627 list_add(&rdev->same_set, &pending_raid_disks);
8631 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8632 i_scanned, i_passed);
8634 autorun_devices(part);
8637 #endif /* !MODULE */
8639 static __exit void md_exit(void)
8641 struct mddev *mddev;
8642 struct list_head *tmp;
8644 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8645 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8647 unregister_blkdev(MD_MAJOR,"md");
8648 unregister_blkdev(mdp_major, "mdp");
8649 unregister_reboot_notifier(&md_notifier);
8650 unregister_sysctl_table(raid_table_header);
8651 remove_proc_entry("mdstat", NULL);
8652 for_each_mddev(mddev, tmp) {
8653 export_array(mddev);
8654 mddev->hold_active = 0;
8656 destroy_workqueue(md_misc_wq);
8657 destroy_workqueue(md_wq);
8660 subsys_initcall(md_init);
8661 module_exit(md_exit)
8663 static int get_ro(char *buffer, struct kernel_param *kp)
8665 return sprintf(buffer, "%d", start_readonly);
8667 static int set_ro(const char *val, struct kernel_param *kp)
8670 int num = simple_strtoul(val, &e, 10);
8671 if (*val && (*e == '\0' || *e == '\n')) {
8672 start_readonly = num;
8678 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8679 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8681 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8683 EXPORT_SYMBOL(register_md_personality);
8684 EXPORT_SYMBOL(unregister_md_personality);
8685 EXPORT_SYMBOL(md_error);
8686 EXPORT_SYMBOL(md_done_sync);
8687 EXPORT_SYMBOL(md_write_start);
8688 EXPORT_SYMBOL(md_write_end);
8689 EXPORT_SYMBOL(md_register_thread);
8690 EXPORT_SYMBOL(md_unregister_thread);
8691 EXPORT_SYMBOL(md_wakeup_thread);
8692 EXPORT_SYMBOL(md_check_recovery);
8693 EXPORT_SYMBOL(md_reap_sync_thread);
8694 MODULE_LICENSE("GPL");
8695 MODULE_DESCRIPTION("MD RAID framework");
8697 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);