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(WriteMostly, &rdev->flags);
1123 if (mddev->raid_disks == 0) {
1124 mddev->major_version = 0;
1125 mddev->minor_version = sb->minor_version;
1126 mddev->patch_version = sb->patch_version;
1127 mddev->external = 0;
1128 mddev->chunk_sectors = sb->chunk_size >> 9;
1129 mddev->ctime = sb->ctime;
1130 mddev->utime = sb->utime;
1131 mddev->level = sb->level;
1132 mddev->clevel[0] = 0;
1133 mddev->layout = sb->layout;
1134 mddev->raid_disks = sb->raid_disks;
1135 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1136 mddev->events = ev1;
1137 mddev->bitmap_info.offset = 0;
1138 mddev->bitmap_info.space = 0;
1139 /* bitmap can use 60 K after the 4K superblocks */
1140 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1141 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1142 mddev->reshape_backwards = 0;
1144 if (mddev->minor_version >= 91) {
1145 mddev->reshape_position = sb->reshape_position;
1146 mddev->delta_disks = sb->delta_disks;
1147 mddev->new_level = sb->new_level;
1148 mddev->new_layout = sb->new_layout;
1149 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1150 if (mddev->delta_disks < 0)
1151 mddev->reshape_backwards = 1;
1153 mddev->reshape_position = MaxSector;
1154 mddev->delta_disks = 0;
1155 mddev->new_level = mddev->level;
1156 mddev->new_layout = mddev->layout;
1157 mddev->new_chunk_sectors = mddev->chunk_sectors;
1160 if (sb->state & (1<<MD_SB_CLEAN))
1161 mddev->recovery_cp = MaxSector;
1163 if (sb->events_hi == sb->cp_events_hi &&
1164 sb->events_lo == sb->cp_events_lo) {
1165 mddev->recovery_cp = sb->recovery_cp;
1167 mddev->recovery_cp = 0;
1170 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1171 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1172 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1173 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1175 mddev->max_disks = MD_SB_DISKS;
1177 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1178 mddev->bitmap_info.file == NULL) {
1179 mddev->bitmap_info.offset =
1180 mddev->bitmap_info.default_offset;
1181 mddev->bitmap_info.space =
1182 mddev->bitmap_info.space;
1185 } else if (mddev->pers == NULL) {
1186 /* Insist on good event counter while assembling, except
1187 * for spares (which don't need an event count) */
1189 if (sb->disks[rdev->desc_nr].state & (
1190 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1191 if (ev1 < mddev->events)
1193 } else if (mddev->bitmap) {
1194 /* if adding to array with a bitmap, then we can accept an
1195 * older device ... but not too old.
1197 if (ev1 < mddev->bitmap->events_cleared)
1200 if (ev1 < mddev->events)
1201 /* just a hot-add of a new device, leave raid_disk at -1 */
1205 if (mddev->level != LEVEL_MULTIPATH) {
1206 desc = sb->disks + rdev->desc_nr;
1208 if (desc->state & (1<<MD_DISK_FAULTY))
1209 set_bit(Faulty, &rdev->flags);
1210 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1211 desc->raid_disk < mddev->raid_disks */) {
1212 set_bit(In_sync, &rdev->flags);
1213 rdev->raid_disk = desc->raid_disk;
1214 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1215 /* active but not in sync implies recovery up to
1216 * reshape position. We don't know exactly where
1217 * that is, so set to zero for now */
1218 if (mddev->minor_version >= 91) {
1219 rdev->recovery_offset = 0;
1220 rdev->raid_disk = desc->raid_disk;
1223 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1224 set_bit(WriteMostly, &rdev->flags);
1225 } else /* MULTIPATH are always insync */
1226 set_bit(In_sync, &rdev->flags);
1231 * sync_super for 0.90.0
1233 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1236 struct md_rdev *rdev2;
1237 int next_spare = mddev->raid_disks;
1240 /* make rdev->sb match mddev data..
1243 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1244 * 3/ any empty disks < next_spare become removed
1246 * disks[0] gets initialised to REMOVED because
1247 * we cannot be sure from other fields if it has
1248 * been initialised or not.
1251 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1253 rdev->sb_size = MD_SB_BYTES;
1255 sb = page_address(rdev->sb_page);
1257 memset(sb, 0, sizeof(*sb));
1259 sb->md_magic = MD_SB_MAGIC;
1260 sb->major_version = mddev->major_version;
1261 sb->patch_version = mddev->patch_version;
1262 sb->gvalid_words = 0; /* ignored */
1263 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1264 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1265 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1266 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1268 sb->ctime = mddev->ctime;
1269 sb->level = mddev->level;
1270 sb->size = mddev->dev_sectors / 2;
1271 sb->raid_disks = mddev->raid_disks;
1272 sb->md_minor = mddev->md_minor;
1273 sb->not_persistent = 0;
1274 sb->utime = mddev->utime;
1276 sb->events_hi = (mddev->events>>32);
1277 sb->events_lo = (u32)mddev->events;
1279 if (mddev->reshape_position == MaxSector)
1280 sb->minor_version = 90;
1282 sb->minor_version = 91;
1283 sb->reshape_position = mddev->reshape_position;
1284 sb->new_level = mddev->new_level;
1285 sb->delta_disks = mddev->delta_disks;
1286 sb->new_layout = mddev->new_layout;
1287 sb->new_chunk = mddev->new_chunk_sectors << 9;
1289 mddev->minor_version = sb->minor_version;
1292 sb->recovery_cp = mddev->recovery_cp;
1293 sb->cp_events_hi = (mddev->events>>32);
1294 sb->cp_events_lo = (u32)mddev->events;
1295 if (mddev->recovery_cp == MaxSector)
1296 sb->state = (1<< MD_SB_CLEAN);
1298 sb->recovery_cp = 0;
1300 sb->layout = mddev->layout;
1301 sb->chunk_size = mddev->chunk_sectors << 9;
1303 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1304 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1306 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1307 rdev_for_each(rdev2, mddev) {
1310 int is_active = test_bit(In_sync, &rdev2->flags);
1312 if (rdev2->raid_disk >= 0 &&
1313 sb->minor_version >= 91)
1314 /* we have nowhere to store the recovery_offset,
1315 * but if it is not below the reshape_position,
1316 * we can piggy-back on that.
1319 if (rdev2->raid_disk < 0 ||
1320 test_bit(Faulty, &rdev2->flags))
1323 desc_nr = rdev2->raid_disk;
1325 desc_nr = next_spare++;
1326 rdev2->desc_nr = desc_nr;
1327 d = &sb->disks[rdev2->desc_nr];
1329 d->number = rdev2->desc_nr;
1330 d->major = MAJOR(rdev2->bdev->bd_dev);
1331 d->minor = MINOR(rdev2->bdev->bd_dev);
1333 d->raid_disk = rdev2->raid_disk;
1335 d->raid_disk = rdev2->desc_nr; /* compatibility */
1336 if (test_bit(Faulty, &rdev2->flags))
1337 d->state = (1<<MD_DISK_FAULTY);
1338 else if (is_active) {
1339 d->state = (1<<MD_DISK_ACTIVE);
1340 if (test_bit(In_sync, &rdev2->flags))
1341 d->state |= (1<<MD_DISK_SYNC);
1349 if (test_bit(WriteMostly, &rdev2->flags))
1350 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1352 /* now set the "removed" and "faulty" bits on any missing devices */
1353 for (i=0 ; i < mddev->raid_disks ; i++) {
1354 mdp_disk_t *d = &sb->disks[i];
1355 if (d->state == 0 && d->number == 0) {
1358 d->state = (1<<MD_DISK_REMOVED);
1359 d->state |= (1<<MD_DISK_FAULTY);
1363 sb->nr_disks = nr_disks;
1364 sb->active_disks = active;
1365 sb->working_disks = working;
1366 sb->failed_disks = failed;
1367 sb->spare_disks = spare;
1369 sb->this_disk = sb->disks[rdev->desc_nr];
1370 sb->sb_csum = calc_sb_csum(sb);
1374 * rdev_size_change for 0.90.0
1376 static unsigned long long
1377 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1379 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1380 return 0; /* component must fit device */
1381 if (rdev->mddev->bitmap_info.offset)
1382 return 0; /* can't move bitmap */
1383 rdev->sb_start = calc_dev_sboffset(rdev);
1384 if (!num_sectors || num_sectors > rdev->sb_start)
1385 num_sectors = rdev->sb_start;
1386 /* Limit to 4TB as metadata cannot record more than that.
1387 * 4TB == 2^32 KB, or 2*2^32 sectors.
1389 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1390 num_sectors = (2ULL << 32) - 2;
1391 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1393 md_super_wait(rdev->mddev);
1398 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1400 /* non-zero offset changes not possible with v0.90 */
1401 return new_offset == 0;
1405 * version 1 superblock
1408 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1412 unsigned long long newcsum;
1413 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1414 __le32 *isuper = (__le32*)sb;
1416 disk_csum = sb->sb_csum;
1419 for (; size >= 4; size -= 4)
1420 newcsum += le32_to_cpu(*isuper++);
1423 newcsum += le16_to_cpu(*(__le16*) isuper);
1425 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1426 sb->sb_csum = disk_csum;
1427 return cpu_to_le32(csum);
1430 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1432 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1434 struct mdp_superblock_1 *sb;
1438 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1442 * Calculate the position of the superblock in 512byte sectors.
1443 * It is always aligned to a 4K boundary and
1444 * depeding on minor_version, it can be:
1445 * 0: At least 8K, but less than 12K, from end of device
1446 * 1: At start of device
1447 * 2: 4K from start of device.
1449 switch(minor_version) {
1451 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1453 sb_start &= ~(sector_t)(4*2-1);
1464 rdev->sb_start = sb_start;
1466 /* superblock is rarely larger than 1K, but it can be larger,
1467 * and it is safe to read 4k, so we do that
1469 ret = read_disk_sb(rdev, 4096);
1470 if (ret) return ret;
1473 sb = page_address(rdev->sb_page);
1475 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1476 sb->major_version != cpu_to_le32(1) ||
1477 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1478 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1479 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1482 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1483 printk("md: invalid superblock checksum on %s\n",
1484 bdevname(rdev->bdev,b));
1487 if (le64_to_cpu(sb->data_size) < 10) {
1488 printk("md: data_size too small on %s\n",
1489 bdevname(rdev->bdev,b));
1494 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1495 /* Some padding is non-zero, might be a new feature */
1498 rdev->preferred_minor = 0xffff;
1499 rdev->data_offset = le64_to_cpu(sb->data_offset);
1500 rdev->new_data_offset = rdev->data_offset;
1501 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1502 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1503 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1504 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1506 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1507 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1508 if (rdev->sb_size & bmask)
1509 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1512 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1515 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1518 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1521 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1523 if (!rdev->bb_page) {
1524 rdev->bb_page = alloc_page(GFP_KERNEL);
1528 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1529 rdev->badblocks.count == 0) {
1530 /* need to load the bad block list.
1531 * Currently we limit it to one page.
1537 int sectors = le16_to_cpu(sb->bblog_size);
1538 if (sectors > (PAGE_SIZE / 512))
1540 offset = le32_to_cpu(sb->bblog_offset);
1543 bb_sector = (long long)offset;
1544 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1545 rdev->bb_page, READ, true))
1547 bbp = (u64 *)page_address(rdev->bb_page);
1548 rdev->badblocks.shift = sb->bblog_shift;
1549 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1550 u64 bb = le64_to_cpu(*bbp);
1551 int count = bb & (0x3ff);
1552 u64 sector = bb >> 10;
1553 sector <<= sb->bblog_shift;
1554 count <<= sb->bblog_shift;
1557 if (md_set_badblocks(&rdev->badblocks,
1558 sector, count, 1) == 0)
1561 } else if (sb->bblog_offset != 0)
1562 rdev->badblocks.shift = 0;
1568 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1570 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1571 sb->level != refsb->level ||
1572 sb->layout != refsb->layout ||
1573 sb->chunksize != refsb->chunksize) {
1574 printk(KERN_WARNING "md: %s has strangely different"
1575 " superblock to %s\n",
1576 bdevname(rdev->bdev,b),
1577 bdevname(refdev->bdev,b2));
1580 ev1 = le64_to_cpu(sb->events);
1581 ev2 = le64_to_cpu(refsb->events);
1588 if (minor_version) {
1589 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1590 sectors -= rdev->data_offset;
1592 sectors = rdev->sb_start;
1593 if (sectors < le64_to_cpu(sb->data_size))
1595 rdev->sectors = le64_to_cpu(sb->data_size);
1599 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1601 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1602 __u64 ev1 = le64_to_cpu(sb->events);
1604 rdev->raid_disk = -1;
1605 clear_bit(Faulty, &rdev->flags);
1606 clear_bit(In_sync, &rdev->flags);
1607 clear_bit(WriteMostly, &rdev->flags);
1609 if (mddev->raid_disks == 0) {
1610 mddev->major_version = 1;
1611 mddev->patch_version = 0;
1612 mddev->external = 0;
1613 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1614 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1615 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1616 mddev->level = le32_to_cpu(sb->level);
1617 mddev->clevel[0] = 0;
1618 mddev->layout = le32_to_cpu(sb->layout);
1619 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1620 mddev->dev_sectors = le64_to_cpu(sb->size);
1621 mddev->events = ev1;
1622 mddev->bitmap_info.offset = 0;
1623 mddev->bitmap_info.space = 0;
1624 /* Default location for bitmap is 1K after superblock
1625 * using 3K - total of 4K
1627 mddev->bitmap_info.default_offset = 1024 >> 9;
1628 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1629 mddev->reshape_backwards = 0;
1631 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1632 memcpy(mddev->uuid, sb->set_uuid, 16);
1634 mddev->max_disks = (4096-256)/2;
1636 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1637 mddev->bitmap_info.file == NULL) {
1638 mddev->bitmap_info.offset =
1639 (__s32)le32_to_cpu(sb->bitmap_offset);
1640 /* Metadata doesn't record how much space is available.
1641 * For 1.0, we assume we can use up to the superblock
1642 * if before, else to 4K beyond superblock.
1643 * For others, assume no change is possible.
1645 if (mddev->minor_version > 0)
1646 mddev->bitmap_info.space = 0;
1647 else if (mddev->bitmap_info.offset > 0)
1648 mddev->bitmap_info.space =
1649 8 - mddev->bitmap_info.offset;
1651 mddev->bitmap_info.space =
1652 -mddev->bitmap_info.offset;
1655 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1656 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1657 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1658 mddev->new_level = le32_to_cpu(sb->new_level);
1659 mddev->new_layout = le32_to_cpu(sb->new_layout);
1660 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1661 if (mddev->delta_disks < 0 ||
1662 (mddev->delta_disks == 0 &&
1663 (le32_to_cpu(sb->feature_map)
1664 & MD_FEATURE_RESHAPE_BACKWARDS)))
1665 mddev->reshape_backwards = 1;
1667 mddev->reshape_position = MaxSector;
1668 mddev->delta_disks = 0;
1669 mddev->new_level = mddev->level;
1670 mddev->new_layout = mddev->layout;
1671 mddev->new_chunk_sectors = mddev->chunk_sectors;
1674 } else if (mddev->pers == NULL) {
1675 /* Insist of good event counter while assembling, except for
1676 * spares (which don't need an event count) */
1678 if (rdev->desc_nr >= 0 &&
1679 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1680 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1681 if (ev1 < mddev->events)
1683 } else if (mddev->bitmap) {
1684 /* If adding to array with a bitmap, then we can accept an
1685 * older device, but not too old.
1687 if (ev1 < mddev->bitmap->events_cleared)
1690 if (ev1 < mddev->events)
1691 /* just a hot-add of a new device, leave raid_disk at -1 */
1694 if (mddev->level != LEVEL_MULTIPATH) {
1696 if (rdev->desc_nr < 0 ||
1697 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1701 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1703 case 0xffff: /* spare */
1705 case 0xfffe: /* faulty */
1706 set_bit(Faulty, &rdev->flags);
1709 if ((le32_to_cpu(sb->feature_map) &
1710 MD_FEATURE_RECOVERY_OFFSET))
1711 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1713 set_bit(In_sync, &rdev->flags);
1714 rdev->raid_disk = role;
1717 if (sb->devflags & WriteMostly1)
1718 set_bit(WriteMostly, &rdev->flags);
1719 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1720 set_bit(Replacement, &rdev->flags);
1721 } else /* MULTIPATH are always insync */
1722 set_bit(In_sync, &rdev->flags);
1727 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1729 struct mdp_superblock_1 *sb;
1730 struct md_rdev *rdev2;
1732 /* make rdev->sb match mddev and rdev data. */
1734 sb = page_address(rdev->sb_page);
1736 sb->feature_map = 0;
1738 sb->recovery_offset = cpu_to_le64(0);
1739 memset(sb->pad3, 0, sizeof(sb->pad3));
1741 sb->utime = cpu_to_le64((__u64)mddev->utime);
1742 sb->events = cpu_to_le64(mddev->events);
1744 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1746 sb->resync_offset = cpu_to_le64(0);
1748 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1750 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1751 sb->size = cpu_to_le64(mddev->dev_sectors);
1752 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1753 sb->level = cpu_to_le32(mddev->level);
1754 sb->layout = cpu_to_le32(mddev->layout);
1756 if (test_bit(WriteMostly, &rdev->flags))
1757 sb->devflags |= WriteMostly1;
1759 sb->devflags &= ~WriteMostly1;
1760 sb->data_offset = cpu_to_le64(rdev->data_offset);
1761 sb->data_size = cpu_to_le64(rdev->sectors);
1763 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1764 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1765 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1768 if (rdev->raid_disk >= 0 &&
1769 !test_bit(In_sync, &rdev->flags)) {
1771 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1772 sb->recovery_offset =
1773 cpu_to_le64(rdev->recovery_offset);
1775 if (test_bit(Replacement, &rdev->flags))
1777 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1779 if (mddev->reshape_position != MaxSector) {
1780 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1781 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1782 sb->new_layout = cpu_to_le32(mddev->new_layout);
1783 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1784 sb->new_level = cpu_to_le32(mddev->new_level);
1785 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1786 if (mddev->delta_disks == 0 &&
1787 mddev->reshape_backwards)
1789 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1790 if (rdev->new_data_offset != rdev->data_offset) {
1792 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1793 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1794 - rdev->data_offset));
1798 if (rdev->badblocks.count == 0)
1799 /* Nothing to do for bad blocks*/ ;
1800 else if (sb->bblog_offset == 0)
1801 /* Cannot record bad blocks on this device */
1802 md_error(mddev, rdev);
1804 struct badblocks *bb = &rdev->badblocks;
1805 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1807 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1812 seq = read_seqbegin(&bb->lock);
1814 memset(bbp, 0xff, PAGE_SIZE);
1816 for (i = 0 ; i < bb->count ; i++) {
1817 u64 internal_bb = p[i];
1818 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1819 | BB_LEN(internal_bb));
1820 bbp[i] = cpu_to_le64(store_bb);
1823 if (read_seqretry(&bb->lock, seq))
1826 bb->sector = (rdev->sb_start +
1827 (int)le32_to_cpu(sb->bblog_offset));
1828 bb->size = le16_to_cpu(sb->bblog_size);
1833 rdev_for_each(rdev2, mddev)
1834 if (rdev2->desc_nr+1 > max_dev)
1835 max_dev = rdev2->desc_nr+1;
1837 if (max_dev > le32_to_cpu(sb->max_dev)) {
1839 sb->max_dev = cpu_to_le32(max_dev);
1840 rdev->sb_size = max_dev * 2 + 256;
1841 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1842 if (rdev->sb_size & bmask)
1843 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1845 max_dev = le32_to_cpu(sb->max_dev);
1847 for (i=0; i<max_dev;i++)
1848 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1850 rdev_for_each(rdev2, mddev) {
1852 if (test_bit(Faulty, &rdev2->flags))
1853 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1854 else if (test_bit(In_sync, &rdev2->flags))
1855 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1856 else if (rdev2->raid_disk >= 0)
1857 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1859 sb->dev_roles[i] = cpu_to_le16(0xffff);
1862 sb->sb_csum = calc_sb_1_csum(sb);
1865 static unsigned long long
1866 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1868 struct mdp_superblock_1 *sb;
1869 sector_t max_sectors;
1870 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1871 return 0; /* component must fit device */
1872 if (rdev->data_offset != rdev->new_data_offset)
1873 return 0; /* too confusing */
1874 if (rdev->sb_start < rdev->data_offset) {
1875 /* minor versions 1 and 2; superblock before data */
1876 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1877 max_sectors -= rdev->data_offset;
1878 if (!num_sectors || num_sectors > max_sectors)
1879 num_sectors = max_sectors;
1880 } else if (rdev->mddev->bitmap_info.offset) {
1881 /* minor version 0 with bitmap we can't move */
1884 /* minor version 0; superblock after data */
1886 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1887 sb_start &= ~(sector_t)(4*2 - 1);
1888 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1889 if (!num_sectors || num_sectors > max_sectors)
1890 num_sectors = max_sectors;
1891 rdev->sb_start = sb_start;
1893 sb = page_address(rdev->sb_page);
1894 sb->data_size = cpu_to_le64(num_sectors);
1895 sb->super_offset = rdev->sb_start;
1896 sb->sb_csum = calc_sb_1_csum(sb);
1897 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1899 md_super_wait(rdev->mddev);
1905 super_1_allow_new_offset(struct md_rdev *rdev,
1906 unsigned long long new_offset)
1908 /* All necessary checks on new >= old have been done */
1909 struct bitmap *bitmap;
1910 if (new_offset >= rdev->data_offset)
1913 /* with 1.0 metadata, there is no metadata to tread on
1914 * so we can always move back */
1915 if (rdev->mddev->minor_version == 0)
1918 /* otherwise we must be sure not to step on
1919 * any metadata, so stay:
1920 * 36K beyond start of superblock
1921 * beyond end of badblocks
1922 * beyond write-intent bitmap
1924 if (rdev->sb_start + (32+4)*2 > new_offset)
1926 bitmap = rdev->mddev->bitmap;
1927 if (bitmap && !rdev->mddev->bitmap_info.file &&
1928 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1929 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1931 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1937 static struct super_type super_types[] = {
1940 .owner = THIS_MODULE,
1941 .load_super = super_90_load,
1942 .validate_super = super_90_validate,
1943 .sync_super = super_90_sync,
1944 .rdev_size_change = super_90_rdev_size_change,
1945 .allow_new_offset = super_90_allow_new_offset,
1949 .owner = THIS_MODULE,
1950 .load_super = super_1_load,
1951 .validate_super = super_1_validate,
1952 .sync_super = super_1_sync,
1953 .rdev_size_change = super_1_rdev_size_change,
1954 .allow_new_offset = super_1_allow_new_offset,
1958 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1960 if (mddev->sync_super) {
1961 mddev->sync_super(mddev, rdev);
1965 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1967 super_types[mddev->major_version].sync_super(mddev, rdev);
1970 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1972 struct md_rdev *rdev, *rdev2;
1975 rdev_for_each_rcu(rdev, mddev1)
1976 rdev_for_each_rcu(rdev2, mddev2)
1977 if (rdev->bdev->bd_contains ==
1978 rdev2->bdev->bd_contains) {
1986 static LIST_HEAD(pending_raid_disks);
1989 * Try to register data integrity profile for an mddev
1991 * This is called when an array is started and after a disk has been kicked
1992 * from the array. It only succeeds if all working and active component devices
1993 * are integrity capable with matching profiles.
1995 int md_integrity_register(struct mddev *mddev)
1997 struct md_rdev *rdev, *reference = NULL;
1999 if (list_empty(&mddev->disks))
2000 return 0; /* nothing to do */
2001 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2002 return 0; /* shouldn't register, or already is */
2003 rdev_for_each(rdev, mddev) {
2004 /* skip spares and non-functional disks */
2005 if (test_bit(Faulty, &rdev->flags))
2007 if (rdev->raid_disk < 0)
2010 /* Use the first rdev as the reference */
2014 /* does this rdev's profile match the reference profile? */
2015 if (blk_integrity_compare(reference->bdev->bd_disk,
2016 rdev->bdev->bd_disk) < 0)
2019 if (!reference || !bdev_get_integrity(reference->bdev))
2022 * All component devices are integrity capable and have matching
2023 * profiles, register the common profile for the md device.
2025 if (blk_integrity_register(mddev->gendisk,
2026 bdev_get_integrity(reference->bdev)) != 0) {
2027 printk(KERN_ERR "md: failed to register integrity for %s\n",
2031 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2032 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2033 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2039 EXPORT_SYMBOL(md_integrity_register);
2041 /* Disable data integrity if non-capable/non-matching disk is being added */
2042 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2044 struct blk_integrity *bi_rdev;
2045 struct blk_integrity *bi_mddev;
2047 if (!mddev->gendisk)
2050 bi_rdev = bdev_get_integrity(rdev->bdev);
2051 bi_mddev = blk_get_integrity(mddev->gendisk);
2053 if (!bi_mddev) /* nothing to do */
2055 if (rdev->raid_disk < 0) /* skip spares */
2057 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2058 rdev->bdev->bd_disk) >= 0)
2060 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2061 blk_integrity_unregister(mddev->gendisk);
2063 EXPORT_SYMBOL(md_integrity_add_rdev);
2065 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2067 char b[BDEVNAME_SIZE];
2077 /* prevent duplicates */
2078 if (find_rdev(mddev, rdev->bdev->bd_dev))
2081 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2082 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2083 rdev->sectors < mddev->dev_sectors)) {
2085 /* Cannot change size, so fail
2086 * If mddev->level <= 0, then we don't care
2087 * about aligning sizes (e.g. linear)
2089 if (mddev->level > 0)
2092 mddev->dev_sectors = rdev->sectors;
2095 /* Verify rdev->desc_nr is unique.
2096 * If it is -1, assign a free number, else
2097 * check number is not in use
2099 if (rdev->desc_nr < 0) {
2101 if (mddev->pers) choice = mddev->raid_disks;
2102 while (find_rdev_nr(mddev, choice))
2104 rdev->desc_nr = choice;
2106 if (find_rdev_nr(mddev, rdev->desc_nr))
2109 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2110 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2111 mdname(mddev), mddev->max_disks);
2114 bdevname(rdev->bdev,b);
2115 while ( (s=strchr(b, '/')) != NULL)
2118 rdev->mddev = mddev;
2119 printk(KERN_INFO "md: bind<%s>\n", b);
2121 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2124 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2125 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2126 /* failure here is OK */;
2127 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2129 list_add_rcu(&rdev->same_set, &mddev->disks);
2130 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2132 /* May as well allow recovery to be retried once */
2133 mddev->recovery_disabled++;
2138 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2143 static void md_delayed_delete(struct work_struct *ws)
2145 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2146 kobject_del(&rdev->kobj);
2147 kobject_put(&rdev->kobj);
2150 static void unbind_rdev_from_array(struct md_rdev * rdev)
2152 char b[BDEVNAME_SIZE];
2157 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2158 list_del_rcu(&rdev->same_set);
2159 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2161 sysfs_remove_link(&rdev->kobj, "block");
2162 sysfs_put(rdev->sysfs_state);
2163 rdev->sysfs_state = NULL;
2164 rdev->badblocks.count = 0;
2165 /* We need to delay this, otherwise we can deadlock when
2166 * writing to 'remove' to "dev/state". We also need
2167 * to delay it due to rcu usage.
2170 INIT_WORK(&rdev->del_work, md_delayed_delete);
2171 kobject_get(&rdev->kobj);
2172 queue_work(md_misc_wq, &rdev->del_work);
2176 * prevent the device from being mounted, repartitioned or
2177 * otherwise reused by a RAID array (or any other kernel
2178 * subsystem), by bd_claiming the device.
2180 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2183 struct block_device *bdev;
2184 char b[BDEVNAME_SIZE];
2186 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2187 shared ? (struct md_rdev *)lock_rdev : rdev);
2189 printk(KERN_ERR "md: could not open %s.\n",
2190 __bdevname(dev, b));
2191 return PTR_ERR(bdev);
2197 static void unlock_rdev(struct md_rdev *rdev)
2199 struct block_device *bdev = rdev->bdev;
2203 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2206 void md_autodetect_dev(dev_t dev);
2208 static void export_rdev(struct md_rdev * rdev)
2210 char b[BDEVNAME_SIZE];
2211 printk(KERN_INFO "md: export_rdev(%s)\n",
2212 bdevname(rdev->bdev,b));
2215 md_rdev_clear(rdev);
2217 if (test_bit(AutoDetected, &rdev->flags))
2218 md_autodetect_dev(rdev->bdev->bd_dev);
2221 kobject_put(&rdev->kobj);
2224 static void kick_rdev_from_array(struct md_rdev * rdev)
2226 unbind_rdev_from_array(rdev);
2230 static void export_array(struct mddev *mddev)
2232 struct md_rdev *rdev, *tmp;
2234 rdev_for_each_safe(rdev, tmp, mddev) {
2239 kick_rdev_from_array(rdev);
2241 if (!list_empty(&mddev->disks))
2243 mddev->raid_disks = 0;
2244 mddev->major_version = 0;
2247 static void print_desc(mdp_disk_t *desc)
2249 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2250 desc->major,desc->minor,desc->raid_disk,desc->state);
2253 static void print_sb_90(mdp_super_t *sb)
2258 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2259 sb->major_version, sb->minor_version, sb->patch_version,
2260 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2262 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2263 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2264 sb->md_minor, sb->layout, sb->chunk_size);
2265 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2266 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2267 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2268 sb->failed_disks, sb->spare_disks,
2269 sb->sb_csum, (unsigned long)sb->events_lo);
2272 for (i = 0; i < MD_SB_DISKS; i++) {
2275 desc = sb->disks + i;
2276 if (desc->number || desc->major || desc->minor ||
2277 desc->raid_disk || (desc->state && (desc->state != 4))) {
2278 printk(" D %2d: ", i);
2282 printk(KERN_INFO "md: THIS: ");
2283 print_desc(&sb->this_disk);
2286 static void print_sb_1(struct mdp_superblock_1 *sb)
2290 uuid = sb->set_uuid;
2292 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2293 "md: Name: \"%s\" CT:%llu\n",
2294 le32_to_cpu(sb->major_version),
2295 le32_to_cpu(sb->feature_map),
2298 (unsigned long long)le64_to_cpu(sb->ctime)
2299 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2301 uuid = sb->device_uuid;
2303 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2305 "md: Dev:%08x UUID: %pU\n"
2306 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2307 "md: (MaxDev:%u) \n",
2308 le32_to_cpu(sb->level),
2309 (unsigned long long)le64_to_cpu(sb->size),
2310 le32_to_cpu(sb->raid_disks),
2311 le32_to_cpu(sb->layout),
2312 le32_to_cpu(sb->chunksize),
2313 (unsigned long long)le64_to_cpu(sb->data_offset),
2314 (unsigned long long)le64_to_cpu(sb->data_size),
2315 (unsigned long long)le64_to_cpu(sb->super_offset),
2316 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2317 le32_to_cpu(sb->dev_number),
2320 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2321 (unsigned long long)le64_to_cpu(sb->events),
2322 (unsigned long long)le64_to_cpu(sb->resync_offset),
2323 le32_to_cpu(sb->sb_csum),
2324 le32_to_cpu(sb->max_dev)
2328 static void print_rdev(struct md_rdev *rdev, int major_version)
2330 char b[BDEVNAME_SIZE];
2331 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2332 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2333 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2335 if (rdev->sb_loaded) {
2336 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2337 switch (major_version) {
2339 print_sb_90(page_address(rdev->sb_page));
2342 print_sb_1(page_address(rdev->sb_page));
2346 printk(KERN_INFO "md: no rdev superblock!\n");
2349 static void md_print_devices(void)
2351 struct list_head *tmp;
2352 struct md_rdev *rdev;
2353 struct mddev *mddev;
2354 char b[BDEVNAME_SIZE];
2357 printk("md: **********************************\n");
2358 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2359 printk("md: **********************************\n");
2360 for_each_mddev(mddev, tmp) {
2363 bitmap_print_sb(mddev->bitmap);
2365 printk("%s: ", mdname(mddev));
2366 rdev_for_each(rdev, mddev)
2367 printk("<%s>", bdevname(rdev->bdev,b));
2370 rdev_for_each(rdev, mddev)
2371 print_rdev(rdev, mddev->major_version);
2373 printk("md: **********************************\n");
2378 static void sync_sbs(struct mddev * mddev, int nospares)
2380 /* Update each superblock (in-memory image), but
2381 * if we are allowed to, skip spares which already
2382 * have the right event counter, or have one earlier
2383 * (which would mean they aren't being marked as dirty
2384 * with the rest of the array)
2386 struct md_rdev *rdev;
2387 rdev_for_each(rdev, mddev) {
2388 if (rdev->sb_events == mddev->events ||
2390 rdev->raid_disk < 0 &&
2391 rdev->sb_events+1 == mddev->events)) {
2392 /* Don't update this superblock */
2393 rdev->sb_loaded = 2;
2395 sync_super(mddev, rdev);
2396 rdev->sb_loaded = 1;
2401 static void md_update_sb(struct mddev * mddev, int force_change)
2403 struct md_rdev *rdev;
2406 int any_badblocks_changed = 0;
2410 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2414 /* First make sure individual recovery_offsets are correct */
2415 rdev_for_each(rdev, mddev) {
2416 if (rdev->raid_disk >= 0 &&
2417 mddev->delta_disks >= 0 &&
2418 !test_bit(In_sync, &rdev->flags) &&
2419 mddev->curr_resync_completed > rdev->recovery_offset)
2420 rdev->recovery_offset = mddev->curr_resync_completed;
2423 if (!mddev->persistent) {
2424 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2425 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2426 if (!mddev->external) {
2427 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2428 rdev_for_each(rdev, mddev) {
2429 if (rdev->badblocks.changed) {
2430 rdev->badblocks.changed = 0;
2431 md_ack_all_badblocks(&rdev->badblocks);
2432 md_error(mddev, rdev);
2434 clear_bit(Blocked, &rdev->flags);
2435 clear_bit(BlockedBadBlocks, &rdev->flags);
2436 wake_up(&rdev->blocked_wait);
2439 wake_up(&mddev->sb_wait);
2443 spin_lock_irq(&mddev->write_lock);
2445 mddev->utime = get_seconds();
2447 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2449 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2450 /* just a clean<-> dirty transition, possibly leave spares alone,
2451 * though if events isn't the right even/odd, we will have to do
2457 if (mddev->degraded)
2458 /* If the array is degraded, then skipping spares is both
2459 * dangerous and fairly pointless.
2460 * Dangerous because a device that was removed from the array
2461 * might have a event_count that still looks up-to-date,
2462 * so it can be re-added without a resync.
2463 * Pointless because if there are any spares to skip,
2464 * then a recovery will happen and soon that array won't
2465 * be degraded any more and the spare can go back to sleep then.
2469 sync_req = mddev->in_sync;
2471 /* If this is just a dirty<->clean transition, and the array is clean
2472 * and 'events' is odd, we can roll back to the previous clean state */
2474 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2475 && mddev->can_decrease_events
2476 && mddev->events != 1) {
2478 mddev->can_decrease_events = 0;
2480 /* otherwise we have to go forward and ... */
2482 mddev->can_decrease_events = nospares;
2485 if (!mddev->events) {
2487 * oops, this 64-bit counter should never wrap.
2488 * Either we are in around ~1 trillion A.C., assuming
2489 * 1 reboot per second, or we have a bug:
2495 rdev_for_each(rdev, mddev) {
2496 if (rdev->badblocks.changed)
2497 any_badblocks_changed++;
2498 if (test_bit(Faulty, &rdev->flags))
2499 set_bit(FaultRecorded, &rdev->flags);
2502 sync_sbs(mddev, nospares);
2503 spin_unlock_irq(&mddev->write_lock);
2505 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2506 mdname(mddev), mddev->in_sync);
2508 bitmap_update_sb(mddev->bitmap);
2509 rdev_for_each(rdev, mddev) {
2510 char b[BDEVNAME_SIZE];
2512 if (rdev->sb_loaded != 1)
2513 continue; /* no noise on spare devices */
2515 if (!test_bit(Faulty, &rdev->flags) &&
2516 rdev->saved_raid_disk == -1) {
2517 md_super_write(mddev,rdev,
2518 rdev->sb_start, rdev->sb_size,
2520 pr_debug("md: (write) %s's sb offset: %llu\n",
2521 bdevname(rdev->bdev, b),
2522 (unsigned long long)rdev->sb_start);
2523 rdev->sb_events = mddev->events;
2524 if (rdev->badblocks.size) {
2525 md_super_write(mddev, rdev,
2526 rdev->badblocks.sector,
2527 rdev->badblocks.size << 9,
2529 rdev->badblocks.size = 0;
2532 } else if (test_bit(Faulty, &rdev->flags))
2533 pr_debug("md: %s (skipping faulty)\n",
2534 bdevname(rdev->bdev, b));
2536 pr_debug("(skipping incremental s/r ");
2538 if (mddev->level == LEVEL_MULTIPATH)
2539 /* only need to write one superblock... */
2542 md_super_wait(mddev);
2543 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2545 spin_lock_irq(&mddev->write_lock);
2546 if (mddev->in_sync != sync_req ||
2547 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2548 /* have to write it out again */
2549 spin_unlock_irq(&mddev->write_lock);
2552 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2553 spin_unlock_irq(&mddev->write_lock);
2554 wake_up(&mddev->sb_wait);
2555 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2556 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2558 rdev_for_each(rdev, mddev) {
2559 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2560 clear_bit(Blocked, &rdev->flags);
2562 if (any_badblocks_changed)
2563 md_ack_all_badblocks(&rdev->badblocks);
2564 clear_bit(BlockedBadBlocks, &rdev->flags);
2565 wake_up(&rdev->blocked_wait);
2569 /* words written to sysfs files may, or may not, be \n terminated.
2570 * We want to accept with case. For this we use cmd_match.
2572 static int cmd_match(const char *cmd, const char *str)
2574 /* See if cmd, written into a sysfs file, matches
2575 * str. They must either be the same, or cmd can
2576 * have a trailing newline
2578 while (*cmd && *str && *cmd == *str) {
2589 struct rdev_sysfs_entry {
2590 struct attribute attr;
2591 ssize_t (*show)(struct md_rdev *, char *);
2592 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2596 state_show(struct md_rdev *rdev, char *page)
2601 if (test_bit(Faulty, &rdev->flags) ||
2602 rdev->badblocks.unacked_exist) {
2603 len+= sprintf(page+len, "%sfaulty",sep);
2606 if (test_bit(In_sync, &rdev->flags)) {
2607 len += sprintf(page+len, "%sin_sync",sep);
2610 if (test_bit(WriteMostly, &rdev->flags)) {
2611 len += sprintf(page+len, "%swrite_mostly",sep);
2614 if (test_bit(Blocked, &rdev->flags) ||
2615 (rdev->badblocks.unacked_exist
2616 && !test_bit(Faulty, &rdev->flags))) {
2617 len += sprintf(page+len, "%sblocked", sep);
2620 if (!test_bit(Faulty, &rdev->flags) &&
2621 !test_bit(In_sync, &rdev->flags)) {
2622 len += sprintf(page+len, "%sspare", sep);
2625 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2626 len += sprintf(page+len, "%swrite_error", sep);
2629 if (test_bit(WantReplacement, &rdev->flags)) {
2630 len += sprintf(page+len, "%swant_replacement", sep);
2633 if (test_bit(Replacement, &rdev->flags)) {
2634 len += sprintf(page+len, "%sreplacement", sep);
2638 return len+sprintf(page+len, "\n");
2642 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2645 * faulty - simulates an error
2646 * remove - disconnects the device
2647 * writemostly - sets write_mostly
2648 * -writemostly - clears write_mostly
2649 * blocked - sets the Blocked flags
2650 * -blocked - clears the Blocked and possibly simulates an error
2651 * insync - sets Insync providing device isn't active
2652 * write_error - sets WriteErrorSeen
2653 * -write_error - clears WriteErrorSeen
2656 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2657 md_error(rdev->mddev, rdev);
2658 if (test_bit(Faulty, &rdev->flags))
2662 } else if (cmd_match(buf, "remove")) {
2663 if (rdev->raid_disk >= 0)
2666 struct mddev *mddev = rdev->mddev;
2667 kick_rdev_from_array(rdev);
2669 md_update_sb(mddev, 1);
2670 md_new_event(mddev);
2673 } else if (cmd_match(buf, "writemostly")) {
2674 set_bit(WriteMostly, &rdev->flags);
2676 } else if (cmd_match(buf, "-writemostly")) {
2677 clear_bit(WriteMostly, &rdev->flags);
2679 } else if (cmd_match(buf, "blocked")) {
2680 set_bit(Blocked, &rdev->flags);
2682 } else if (cmd_match(buf, "-blocked")) {
2683 if (!test_bit(Faulty, &rdev->flags) &&
2684 rdev->badblocks.unacked_exist) {
2685 /* metadata handler doesn't understand badblocks,
2686 * so we need to fail the device
2688 md_error(rdev->mddev, rdev);
2690 clear_bit(Blocked, &rdev->flags);
2691 clear_bit(BlockedBadBlocks, &rdev->flags);
2692 wake_up(&rdev->blocked_wait);
2693 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2694 md_wakeup_thread(rdev->mddev->thread);
2697 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2698 set_bit(In_sync, &rdev->flags);
2700 } else if (cmd_match(buf, "write_error")) {
2701 set_bit(WriteErrorSeen, &rdev->flags);
2703 } else if (cmd_match(buf, "-write_error")) {
2704 clear_bit(WriteErrorSeen, &rdev->flags);
2706 } else if (cmd_match(buf, "want_replacement")) {
2707 /* Any non-spare device that is not a replacement can
2708 * become want_replacement at any time, but we then need to
2709 * check if recovery is needed.
2711 if (rdev->raid_disk >= 0 &&
2712 !test_bit(Replacement, &rdev->flags))
2713 set_bit(WantReplacement, &rdev->flags);
2714 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2715 md_wakeup_thread(rdev->mddev->thread);
2717 } else if (cmd_match(buf, "-want_replacement")) {
2718 /* Clearing 'want_replacement' is always allowed.
2719 * Once replacements starts it is too late though.
2722 clear_bit(WantReplacement, &rdev->flags);
2723 } else if (cmd_match(buf, "replacement")) {
2724 /* Can only set a device as a replacement when array has not
2725 * yet been started. Once running, replacement is automatic
2726 * from spares, or by assigning 'slot'.
2728 if (rdev->mddev->pers)
2731 set_bit(Replacement, &rdev->flags);
2734 } else if (cmd_match(buf, "-replacement")) {
2735 /* Similarly, can only clear Replacement before start */
2736 if (rdev->mddev->pers)
2739 clear_bit(Replacement, &rdev->flags);
2744 sysfs_notify_dirent_safe(rdev->sysfs_state);
2745 return err ? err : len;
2747 static struct rdev_sysfs_entry rdev_state =
2748 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2751 errors_show(struct md_rdev *rdev, char *page)
2753 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2757 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2760 unsigned long n = simple_strtoul(buf, &e, 10);
2761 if (*buf && (*e == 0 || *e == '\n')) {
2762 atomic_set(&rdev->corrected_errors, n);
2767 static struct rdev_sysfs_entry rdev_errors =
2768 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2771 slot_show(struct md_rdev *rdev, char *page)
2773 if (rdev->raid_disk < 0)
2774 return sprintf(page, "none\n");
2776 return sprintf(page, "%d\n", rdev->raid_disk);
2780 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2784 int slot = simple_strtoul(buf, &e, 10);
2785 if (strncmp(buf, "none", 4)==0)
2787 else if (e==buf || (*e && *e!= '\n'))
2789 if (rdev->mddev->pers && slot == -1) {
2790 /* Setting 'slot' on an active array requires also
2791 * updating the 'rd%d' link, and communicating
2792 * with the personality with ->hot_*_disk.
2793 * For now we only support removing
2794 * failed/spare devices. This normally happens automatically,
2795 * but not when the metadata is externally managed.
2797 if (rdev->raid_disk == -1)
2799 /* personality does all needed checks */
2800 if (rdev->mddev->pers->hot_remove_disk == NULL)
2802 clear_bit(Blocked, &rdev->flags);
2803 remove_and_add_spares(rdev->mddev, rdev);
2804 if (rdev->raid_disk >= 0)
2806 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2807 md_wakeup_thread(rdev->mddev->thread);
2808 } else if (rdev->mddev->pers) {
2809 /* Activating a spare .. or possibly reactivating
2810 * if we ever get bitmaps working here.
2813 if (rdev->raid_disk != -1)
2816 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2819 if (rdev->mddev->pers->hot_add_disk == NULL)
2822 if (slot >= rdev->mddev->raid_disks &&
2823 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2826 rdev->raid_disk = slot;
2827 if (test_bit(In_sync, &rdev->flags))
2828 rdev->saved_raid_disk = slot;
2830 rdev->saved_raid_disk = -1;
2831 clear_bit(In_sync, &rdev->flags);
2832 err = rdev->mddev->pers->
2833 hot_add_disk(rdev->mddev, rdev);
2835 rdev->raid_disk = -1;
2838 sysfs_notify_dirent_safe(rdev->sysfs_state);
2839 if (sysfs_link_rdev(rdev->mddev, rdev))
2840 /* failure here is OK */;
2841 /* don't wakeup anyone, leave that to userspace. */
2843 if (slot >= rdev->mddev->raid_disks &&
2844 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2846 rdev->raid_disk = slot;
2847 /* assume it is working */
2848 clear_bit(Faulty, &rdev->flags);
2849 clear_bit(WriteMostly, &rdev->flags);
2850 set_bit(In_sync, &rdev->flags);
2851 sysfs_notify_dirent_safe(rdev->sysfs_state);
2857 static struct rdev_sysfs_entry rdev_slot =
2858 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2861 offset_show(struct md_rdev *rdev, char *page)
2863 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2867 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2869 unsigned long long offset;
2870 if (strict_strtoull(buf, 10, &offset) < 0)
2872 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2874 if (rdev->sectors && rdev->mddev->external)
2875 /* Must set offset before size, so overlap checks
2878 rdev->data_offset = offset;
2879 rdev->new_data_offset = offset;
2883 static struct rdev_sysfs_entry rdev_offset =
2884 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2886 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2888 return sprintf(page, "%llu\n",
2889 (unsigned long long)rdev->new_data_offset);
2892 static ssize_t new_offset_store(struct md_rdev *rdev,
2893 const char *buf, size_t len)
2895 unsigned long long new_offset;
2896 struct mddev *mddev = rdev->mddev;
2898 if (strict_strtoull(buf, 10, &new_offset) < 0)
2901 if (mddev->sync_thread)
2903 if (new_offset == rdev->data_offset)
2904 /* reset is always permitted */
2906 else if (new_offset > rdev->data_offset) {
2907 /* must not push array size beyond rdev_sectors */
2908 if (new_offset - rdev->data_offset
2909 + mddev->dev_sectors > rdev->sectors)
2912 /* Metadata worries about other space details. */
2914 /* decreasing the offset is inconsistent with a backwards
2917 if (new_offset < rdev->data_offset &&
2918 mddev->reshape_backwards)
2920 /* Increasing offset is inconsistent with forwards
2921 * reshape. reshape_direction should be set to
2922 * 'backwards' first.
2924 if (new_offset > rdev->data_offset &&
2925 !mddev->reshape_backwards)
2928 if (mddev->pers && mddev->persistent &&
2929 !super_types[mddev->major_version]
2930 .allow_new_offset(rdev, new_offset))
2932 rdev->new_data_offset = new_offset;
2933 if (new_offset > rdev->data_offset)
2934 mddev->reshape_backwards = 1;
2935 else if (new_offset < rdev->data_offset)
2936 mddev->reshape_backwards = 0;
2940 static struct rdev_sysfs_entry rdev_new_offset =
2941 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2944 rdev_size_show(struct md_rdev *rdev, char *page)
2946 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2949 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2951 /* check if two start/length pairs overlap */
2959 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2961 unsigned long long blocks;
2964 if (strict_strtoull(buf, 10, &blocks) < 0)
2967 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2968 return -EINVAL; /* sector conversion overflow */
2971 if (new != blocks * 2)
2972 return -EINVAL; /* unsigned long long to sector_t overflow */
2979 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2981 struct mddev *my_mddev = rdev->mddev;
2982 sector_t oldsectors = rdev->sectors;
2985 if (strict_blocks_to_sectors(buf, §ors) < 0)
2987 if (rdev->data_offset != rdev->new_data_offset)
2988 return -EINVAL; /* too confusing */
2989 if (my_mddev->pers && rdev->raid_disk >= 0) {
2990 if (my_mddev->persistent) {
2991 sectors = super_types[my_mddev->major_version].
2992 rdev_size_change(rdev, sectors);
2995 } else if (!sectors)
2996 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2998 if (!my_mddev->pers->resize)
2999 /* Cannot change size for RAID0 or Linear etc */
3002 if (sectors < my_mddev->dev_sectors)
3003 return -EINVAL; /* component must fit device */
3005 rdev->sectors = sectors;
3006 if (sectors > oldsectors && my_mddev->external) {
3007 /* need to check that all other rdevs with the same ->bdev
3008 * do not overlap. We need to unlock the mddev to avoid
3009 * a deadlock. We have already changed rdev->sectors, and if
3010 * we have to change it back, we will have the lock again.
3012 struct mddev *mddev;
3014 struct list_head *tmp;
3016 mddev_unlock(my_mddev);
3017 for_each_mddev(mddev, tmp) {
3018 struct md_rdev *rdev2;
3021 rdev_for_each(rdev2, mddev)
3022 if (rdev->bdev == rdev2->bdev &&
3024 overlaps(rdev->data_offset, rdev->sectors,
3030 mddev_unlock(mddev);
3036 mddev_lock(my_mddev);
3038 /* Someone else could have slipped in a size
3039 * change here, but doing so is just silly.
3040 * We put oldsectors back because we *know* it is
3041 * safe, and trust userspace not to race with
3044 rdev->sectors = oldsectors;
3051 static struct rdev_sysfs_entry rdev_size =
3052 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3055 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3057 unsigned long long recovery_start = rdev->recovery_offset;
3059 if (test_bit(In_sync, &rdev->flags) ||
3060 recovery_start == MaxSector)
3061 return sprintf(page, "none\n");
3063 return sprintf(page, "%llu\n", recovery_start);
3066 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3068 unsigned long long recovery_start;
3070 if (cmd_match(buf, "none"))
3071 recovery_start = MaxSector;
3072 else if (strict_strtoull(buf, 10, &recovery_start))
3075 if (rdev->mddev->pers &&
3076 rdev->raid_disk >= 0)
3079 rdev->recovery_offset = recovery_start;
3080 if (recovery_start == MaxSector)
3081 set_bit(In_sync, &rdev->flags);
3083 clear_bit(In_sync, &rdev->flags);
3087 static struct rdev_sysfs_entry rdev_recovery_start =
3088 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3092 badblocks_show(struct badblocks *bb, char *page, int unack);
3094 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3096 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3098 return badblocks_show(&rdev->badblocks, page, 0);
3100 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3102 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3103 /* Maybe that ack was all we needed */
3104 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3105 wake_up(&rdev->blocked_wait);
3108 static struct rdev_sysfs_entry rdev_bad_blocks =
3109 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3112 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3114 return badblocks_show(&rdev->badblocks, page, 1);
3116 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3118 return badblocks_store(&rdev->badblocks, page, len, 1);
3120 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3121 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3123 static struct attribute *rdev_default_attrs[] = {
3128 &rdev_new_offset.attr,
3130 &rdev_recovery_start.attr,
3131 &rdev_bad_blocks.attr,
3132 &rdev_unack_bad_blocks.attr,
3136 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3138 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3139 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3140 struct mddev *mddev = rdev->mddev;
3146 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3148 if (rdev->mddev == NULL)
3151 rv = entry->show(rdev, page);
3152 mddev_unlock(mddev);
3158 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3159 const char *page, size_t length)
3161 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3162 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3164 struct mddev *mddev = rdev->mddev;
3168 if (!capable(CAP_SYS_ADMIN))
3170 rv = mddev ? mddev_lock(mddev): -EBUSY;
3172 if (rdev->mddev == NULL)
3175 rv = entry->store(rdev, page, length);
3176 mddev_unlock(mddev);
3181 static void rdev_free(struct kobject *ko)
3183 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3186 static const struct sysfs_ops rdev_sysfs_ops = {
3187 .show = rdev_attr_show,
3188 .store = rdev_attr_store,
3190 static struct kobj_type rdev_ktype = {
3191 .release = rdev_free,
3192 .sysfs_ops = &rdev_sysfs_ops,
3193 .default_attrs = rdev_default_attrs,
3196 int md_rdev_init(struct md_rdev *rdev)
3199 rdev->saved_raid_disk = -1;
3200 rdev->raid_disk = -1;
3202 rdev->data_offset = 0;
3203 rdev->new_data_offset = 0;
3204 rdev->sb_events = 0;
3205 rdev->last_read_error.tv_sec = 0;
3206 rdev->last_read_error.tv_nsec = 0;
3207 rdev->sb_loaded = 0;
3208 rdev->bb_page = NULL;
3209 atomic_set(&rdev->nr_pending, 0);
3210 atomic_set(&rdev->read_errors, 0);
3211 atomic_set(&rdev->corrected_errors, 0);
3213 INIT_LIST_HEAD(&rdev->same_set);
3214 init_waitqueue_head(&rdev->blocked_wait);
3216 /* Add space to store bad block list.
3217 * This reserves the space even on arrays where it cannot
3218 * be used - I wonder if that matters
3220 rdev->badblocks.count = 0;
3221 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3222 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3223 seqlock_init(&rdev->badblocks.lock);
3224 if (rdev->badblocks.page == NULL)
3229 EXPORT_SYMBOL_GPL(md_rdev_init);
3231 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3233 * mark the device faulty if:
3235 * - the device is nonexistent (zero size)
3236 * - the device has no valid superblock
3238 * a faulty rdev _never_ has rdev->sb set.
3240 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3242 char b[BDEVNAME_SIZE];
3244 struct md_rdev *rdev;
3247 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3249 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3250 return ERR_PTR(-ENOMEM);
3253 err = md_rdev_init(rdev);
3256 err = alloc_disk_sb(rdev);
3260 err = lock_rdev(rdev, newdev, super_format == -2);
3264 kobject_init(&rdev->kobj, &rdev_ktype);
3266 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3269 "md: %s has zero or unknown size, marking faulty!\n",
3270 bdevname(rdev->bdev,b));
3275 if (super_format >= 0) {
3276 err = super_types[super_format].
3277 load_super(rdev, NULL, super_minor);
3278 if (err == -EINVAL) {
3280 "md: %s does not have a valid v%d.%d "
3281 "superblock, not importing!\n",
3282 bdevname(rdev->bdev,b),
3283 super_format, super_minor);
3288 "md: could not read %s's sb, not importing!\n",
3289 bdevname(rdev->bdev,b));
3299 md_rdev_clear(rdev);
3301 return ERR_PTR(err);
3305 * Check a full RAID array for plausibility
3309 static void analyze_sbs(struct mddev * mddev)
3312 struct md_rdev *rdev, *freshest, *tmp;
3313 char b[BDEVNAME_SIZE];
3316 rdev_for_each_safe(rdev, tmp, mddev)
3317 switch (super_types[mddev->major_version].
3318 load_super(rdev, freshest, mddev->minor_version)) {
3326 "md: fatal superblock inconsistency in %s"
3327 " -- removing from array\n",
3328 bdevname(rdev->bdev,b));
3329 kick_rdev_from_array(rdev);
3333 super_types[mddev->major_version].
3334 validate_super(mddev, freshest);
3337 rdev_for_each_safe(rdev, tmp, mddev) {
3338 if (mddev->max_disks &&
3339 (rdev->desc_nr >= mddev->max_disks ||
3340 i > mddev->max_disks)) {
3342 "md: %s: %s: only %d devices permitted\n",
3343 mdname(mddev), bdevname(rdev->bdev, b),
3345 kick_rdev_from_array(rdev);
3348 if (rdev != freshest)
3349 if (super_types[mddev->major_version].
3350 validate_super(mddev, rdev)) {
3351 printk(KERN_WARNING "md: kicking non-fresh %s"
3353 bdevname(rdev->bdev,b));
3354 kick_rdev_from_array(rdev);
3357 if (mddev->level == LEVEL_MULTIPATH) {
3358 rdev->desc_nr = i++;
3359 rdev->raid_disk = rdev->desc_nr;
3360 set_bit(In_sync, &rdev->flags);
3361 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3362 rdev->raid_disk = -1;
3363 clear_bit(In_sync, &rdev->flags);
3368 /* Read a fixed-point number.
3369 * Numbers in sysfs attributes should be in "standard" units where
3370 * possible, so time should be in seconds.
3371 * However we internally use a a much smaller unit such as
3372 * milliseconds or jiffies.
3373 * This function takes a decimal number with a possible fractional
3374 * component, and produces an integer which is the result of
3375 * multiplying that number by 10^'scale'.
3376 * all without any floating-point arithmetic.
3378 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3380 unsigned long result = 0;
3382 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3385 else if (decimals < scale) {
3388 result = result * 10 + value;
3400 while (decimals < scale) {
3409 static void md_safemode_timeout(unsigned long data);
3412 safe_delay_show(struct mddev *mddev, char *page)
3414 int msec = (mddev->safemode_delay*1000)/HZ;
3415 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3418 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3422 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3425 mddev->safemode_delay = 0;
3427 unsigned long old_delay = mddev->safemode_delay;
3428 mddev->safemode_delay = (msec*HZ)/1000;
3429 if (mddev->safemode_delay == 0)
3430 mddev->safemode_delay = 1;
3431 if (mddev->safemode_delay < old_delay)
3432 md_safemode_timeout((unsigned long)mddev);
3436 static struct md_sysfs_entry md_safe_delay =
3437 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3440 level_show(struct mddev *mddev, char *page)
3442 struct md_personality *p = mddev->pers;
3444 return sprintf(page, "%s\n", p->name);
3445 else if (mddev->clevel[0])
3446 return sprintf(page, "%s\n", mddev->clevel);
3447 else if (mddev->level != LEVEL_NONE)
3448 return sprintf(page, "%d\n", mddev->level);
3454 level_store(struct mddev *mddev, const char *buf, size_t len)
3458 struct md_personality *pers;
3461 struct md_rdev *rdev;
3463 if (mddev->pers == NULL) {
3466 if (len >= sizeof(mddev->clevel))
3468 strncpy(mddev->clevel, buf, len);
3469 if (mddev->clevel[len-1] == '\n')
3471 mddev->clevel[len] = 0;
3472 mddev->level = LEVEL_NONE;
3476 /* request to change the personality. Need to ensure:
3477 * - array is not engaged in resync/recovery/reshape
3478 * - old personality can be suspended
3479 * - new personality will access other array.
3482 if (mddev->sync_thread ||
3483 mddev->reshape_position != MaxSector ||
3484 mddev->sysfs_active)
3487 if (!mddev->pers->quiesce) {
3488 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3489 mdname(mddev), mddev->pers->name);
3493 /* Now find the new personality */
3494 if (len == 0 || len >= sizeof(clevel))
3496 strncpy(clevel, buf, len);
3497 if (clevel[len-1] == '\n')
3500 if (strict_strtol(clevel, 10, &level))
3503 if (request_module("md-%s", clevel) != 0)
3504 request_module("md-level-%s", clevel);
3505 spin_lock(&pers_lock);
3506 pers = find_pers(level, clevel);
3507 if (!pers || !try_module_get(pers->owner)) {
3508 spin_unlock(&pers_lock);
3509 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3512 spin_unlock(&pers_lock);
3514 if (pers == mddev->pers) {
3515 /* Nothing to do! */
3516 module_put(pers->owner);
3519 if (!pers->takeover) {
3520 module_put(pers->owner);
3521 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3522 mdname(mddev), clevel);
3526 rdev_for_each(rdev, mddev)
3527 rdev->new_raid_disk = rdev->raid_disk;
3529 /* ->takeover must set new_* and/or delta_disks
3530 * if it succeeds, and may set them when it fails.
3532 priv = pers->takeover(mddev);
3534 mddev->new_level = mddev->level;
3535 mddev->new_layout = mddev->layout;
3536 mddev->new_chunk_sectors = mddev->chunk_sectors;
3537 mddev->raid_disks -= mddev->delta_disks;
3538 mddev->delta_disks = 0;
3539 mddev->reshape_backwards = 0;
3540 module_put(pers->owner);
3541 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3542 mdname(mddev), clevel);
3543 return PTR_ERR(priv);
3546 /* Looks like we have a winner */
3547 mddev_suspend(mddev);
3548 mddev->pers->stop(mddev);
3550 if (mddev->pers->sync_request == NULL &&
3551 pers->sync_request != NULL) {
3552 /* need to add the md_redundancy_group */
3553 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3555 "md: cannot register extra attributes for %s\n",
3557 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3559 if (mddev->pers->sync_request != NULL &&
3560 pers->sync_request == NULL) {
3561 /* need to remove the md_redundancy_group */
3562 if (mddev->to_remove == NULL)
3563 mddev->to_remove = &md_redundancy_group;
3566 if (mddev->pers->sync_request == NULL &&
3568 /* We are converting from a no-redundancy array
3569 * to a redundancy array and metadata is managed
3570 * externally so we need to be sure that writes
3571 * won't block due to a need to transition
3573 * until external management is started.
3576 mddev->safemode_delay = 0;
3577 mddev->safemode = 0;
3580 rdev_for_each(rdev, mddev) {
3581 if (rdev->raid_disk < 0)
3583 if (rdev->new_raid_disk >= mddev->raid_disks)
3584 rdev->new_raid_disk = -1;
3585 if (rdev->new_raid_disk == rdev->raid_disk)
3587 sysfs_unlink_rdev(mddev, rdev);
3589 rdev_for_each(rdev, mddev) {
3590 if (rdev->raid_disk < 0)
3592 if (rdev->new_raid_disk == rdev->raid_disk)
3594 rdev->raid_disk = rdev->new_raid_disk;
3595 if (rdev->raid_disk < 0)
3596 clear_bit(In_sync, &rdev->flags);
3598 if (sysfs_link_rdev(mddev, rdev))
3599 printk(KERN_WARNING "md: cannot register rd%d"
3600 " for %s after level change\n",
3601 rdev->raid_disk, mdname(mddev));
3605 module_put(mddev->pers->owner);
3607 mddev->private = priv;
3608 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3609 mddev->level = mddev->new_level;
3610 mddev->layout = mddev->new_layout;
3611 mddev->chunk_sectors = mddev->new_chunk_sectors;
3612 mddev->delta_disks = 0;
3613 mddev->reshape_backwards = 0;
3614 mddev->degraded = 0;
3615 if (mddev->pers->sync_request == NULL) {
3616 /* this is now an array without redundancy, so
3617 * it must always be in_sync
3620 del_timer_sync(&mddev->safemode_timer);
3622 blk_set_stacking_limits(&mddev->queue->limits);
3624 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3625 mddev_resume(mddev);
3626 sysfs_notify(&mddev->kobj, NULL, "level");
3627 md_new_event(mddev);
3631 static struct md_sysfs_entry md_level =
3632 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3636 layout_show(struct mddev *mddev, char *page)
3638 /* just a number, not meaningful for all levels */
3639 if (mddev->reshape_position != MaxSector &&
3640 mddev->layout != mddev->new_layout)
3641 return sprintf(page, "%d (%d)\n",
3642 mddev->new_layout, mddev->layout);
3643 return sprintf(page, "%d\n", mddev->layout);
3647 layout_store(struct mddev *mddev, const char *buf, size_t len)
3650 unsigned long n = simple_strtoul(buf, &e, 10);
3652 if (!*buf || (*e && *e != '\n'))
3657 if (mddev->pers->check_reshape == NULL)
3659 mddev->new_layout = n;
3660 err = mddev->pers->check_reshape(mddev);
3662 mddev->new_layout = mddev->layout;
3666 mddev->new_layout = n;
3667 if (mddev->reshape_position == MaxSector)
3672 static struct md_sysfs_entry md_layout =
3673 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3677 raid_disks_show(struct mddev *mddev, char *page)
3679 if (mddev->raid_disks == 0)
3681 if (mddev->reshape_position != MaxSector &&
3682 mddev->delta_disks != 0)
3683 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3684 mddev->raid_disks - mddev->delta_disks);
3685 return sprintf(page, "%d\n", mddev->raid_disks);
3688 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3691 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3695 unsigned long n = simple_strtoul(buf, &e, 10);
3697 if (!*buf || (*e && *e != '\n'))
3701 rv = update_raid_disks(mddev, n);
3702 else if (mddev->reshape_position != MaxSector) {
3703 struct md_rdev *rdev;
3704 int olddisks = mddev->raid_disks - mddev->delta_disks;
3706 rdev_for_each(rdev, mddev) {
3708 rdev->data_offset < rdev->new_data_offset)
3711 rdev->data_offset > rdev->new_data_offset)
3714 mddev->delta_disks = n - olddisks;
3715 mddev->raid_disks = n;
3716 mddev->reshape_backwards = (mddev->delta_disks < 0);
3718 mddev->raid_disks = n;
3719 return rv ? rv : len;
3721 static struct md_sysfs_entry md_raid_disks =
3722 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3725 chunk_size_show(struct mddev *mddev, char *page)
3727 if (mddev->reshape_position != MaxSector &&
3728 mddev->chunk_sectors != mddev->new_chunk_sectors)
3729 return sprintf(page, "%d (%d)\n",
3730 mddev->new_chunk_sectors << 9,
3731 mddev->chunk_sectors << 9);
3732 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3736 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3739 unsigned long n = simple_strtoul(buf, &e, 10);
3741 if (!*buf || (*e && *e != '\n'))
3746 if (mddev->pers->check_reshape == NULL)
3748 mddev->new_chunk_sectors = n >> 9;
3749 err = mddev->pers->check_reshape(mddev);
3751 mddev->new_chunk_sectors = mddev->chunk_sectors;
3755 mddev->new_chunk_sectors = n >> 9;
3756 if (mddev->reshape_position == MaxSector)
3757 mddev->chunk_sectors = n >> 9;
3761 static struct md_sysfs_entry md_chunk_size =
3762 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3765 resync_start_show(struct mddev *mddev, char *page)
3767 if (mddev->recovery_cp == MaxSector)
3768 return sprintf(page, "none\n");
3769 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3773 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3776 unsigned long long n = simple_strtoull(buf, &e, 10);
3778 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3780 if (cmd_match(buf, "none"))
3782 else if (!*buf || (*e && *e != '\n'))
3785 mddev->recovery_cp = n;
3787 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3790 static struct md_sysfs_entry md_resync_start =
3791 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3794 * The array state can be:
3797 * No devices, no size, no level
3798 * Equivalent to STOP_ARRAY ioctl
3800 * May have some settings, but array is not active
3801 * all IO results in error
3802 * When written, doesn't tear down array, but just stops it
3803 * suspended (not supported yet)
3804 * All IO requests will block. The array can be reconfigured.
3805 * Writing this, if accepted, will block until array is quiescent
3807 * no resync can happen. no superblocks get written.
3808 * write requests fail
3810 * like readonly, but behaves like 'clean' on a write request.
3812 * clean - no pending writes, but otherwise active.
3813 * When written to inactive array, starts without resync
3814 * If a write request arrives then
3815 * if metadata is known, mark 'dirty' and switch to 'active'.
3816 * if not known, block and switch to write-pending
3817 * If written to an active array that has pending writes, then fails.
3819 * fully active: IO and resync can be happening.
3820 * When written to inactive array, starts with resync
3823 * clean, but writes are blocked waiting for 'active' to be written.
3826 * like active, but no writes have been seen for a while (100msec).
3829 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3830 write_pending, active_idle, bad_word};
3831 static char *array_states[] = {
3832 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3833 "write-pending", "active-idle", NULL };
3835 static int match_word(const char *word, char **list)
3838 for (n=0; list[n]; n++)
3839 if (cmd_match(word, list[n]))
3845 array_state_show(struct mddev *mddev, char *page)
3847 enum array_state st = inactive;
3860 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3862 else if (mddev->safemode)
3868 if (list_empty(&mddev->disks) &&
3869 mddev->raid_disks == 0 &&
3870 mddev->dev_sectors == 0)
3875 return sprintf(page, "%s\n", array_states[st]);
3878 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3879 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3880 static int do_md_run(struct mddev * mddev);
3881 static int restart_array(struct mddev *mddev);
3884 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3887 enum array_state st = match_word(buf, array_states);
3892 /* stopping an active array */
3893 err = do_md_stop(mddev, 0, NULL);
3896 /* stopping an active array */
3898 err = do_md_stop(mddev, 2, NULL);
3900 err = 0; /* already inactive */
3903 break; /* not supported yet */
3906 err = md_set_readonly(mddev, NULL);
3909 set_disk_ro(mddev->gendisk, 1);
3910 err = do_md_run(mddev);
3916 err = md_set_readonly(mddev, NULL);
3917 else if (mddev->ro == 1)
3918 err = restart_array(mddev);
3921 set_disk_ro(mddev->gendisk, 0);
3925 err = do_md_run(mddev);
3930 restart_array(mddev);
3931 spin_lock_irq(&mddev->write_lock);
3932 if (atomic_read(&mddev->writes_pending) == 0) {
3933 if (mddev->in_sync == 0) {
3935 if (mddev->safemode == 1)
3936 mddev->safemode = 0;
3937 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3942 spin_unlock_irq(&mddev->write_lock);
3948 restart_array(mddev);
3949 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3950 wake_up(&mddev->sb_wait);
3954 set_disk_ro(mddev->gendisk, 0);
3955 err = do_md_run(mddev);
3960 /* these cannot be set */
3966 if (mddev->hold_active == UNTIL_IOCTL)
3967 mddev->hold_active = 0;
3968 sysfs_notify_dirent_safe(mddev->sysfs_state);
3972 static struct md_sysfs_entry md_array_state =
3973 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3976 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3977 return sprintf(page, "%d\n",
3978 atomic_read(&mddev->max_corr_read_errors));
3982 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3985 unsigned long n = simple_strtoul(buf, &e, 10);
3987 if (*buf && (*e == 0 || *e == '\n')) {
3988 atomic_set(&mddev->max_corr_read_errors, n);
3994 static struct md_sysfs_entry max_corr_read_errors =
3995 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3996 max_corrected_read_errors_store);
3999 null_show(struct mddev *mddev, char *page)
4005 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4007 /* buf must be %d:%d\n? giving major and minor numbers */
4008 /* The new device is added to the array.
4009 * If the array has a persistent superblock, we read the
4010 * superblock to initialise info and check validity.
4011 * Otherwise, only checking done is that in bind_rdev_to_array,
4012 * which mainly checks size.
4015 int major = simple_strtoul(buf, &e, 10);
4018 struct md_rdev *rdev;
4021 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4023 minor = simple_strtoul(e+1, &e, 10);
4024 if (*e && *e != '\n')
4026 dev = MKDEV(major, minor);
4027 if (major != MAJOR(dev) ||
4028 minor != MINOR(dev))
4032 if (mddev->persistent) {
4033 rdev = md_import_device(dev, mddev->major_version,
4034 mddev->minor_version);
4035 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4036 struct md_rdev *rdev0
4037 = list_entry(mddev->disks.next,
4038 struct md_rdev, same_set);
4039 err = super_types[mddev->major_version]
4040 .load_super(rdev, rdev0, mddev->minor_version);
4044 } else if (mddev->external)
4045 rdev = md_import_device(dev, -2, -1);
4047 rdev = md_import_device(dev, -1, -1);
4050 return PTR_ERR(rdev);
4051 err = bind_rdev_to_array(rdev, mddev);
4055 return err ? err : len;
4058 static struct md_sysfs_entry md_new_device =
4059 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4062 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4065 unsigned long chunk, end_chunk;
4069 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4071 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4072 if (buf == end) break;
4073 if (*end == '-') { /* range */
4075 end_chunk = simple_strtoul(buf, &end, 0);
4076 if (buf == end) break;
4078 if (*end && !isspace(*end)) break;
4079 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4080 buf = skip_spaces(end);
4082 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4087 static struct md_sysfs_entry md_bitmap =
4088 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4091 size_show(struct mddev *mddev, char *page)
4093 return sprintf(page, "%llu\n",
4094 (unsigned long long)mddev->dev_sectors / 2);
4097 static int update_size(struct mddev *mddev, sector_t num_sectors);
4100 size_store(struct mddev *mddev, const char *buf, size_t len)
4102 /* If array is inactive, we can reduce the component size, but
4103 * not increase it (except from 0).
4104 * If array is active, we can try an on-line resize
4107 int err = strict_blocks_to_sectors(buf, §ors);
4112 err = update_size(mddev, sectors);
4113 md_update_sb(mddev, 1);
4115 if (mddev->dev_sectors == 0 ||
4116 mddev->dev_sectors > sectors)
4117 mddev->dev_sectors = sectors;
4121 return err ? err : len;
4124 static struct md_sysfs_entry md_size =
4125 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4128 /* Metadata version.
4130 * 'none' for arrays with no metadata (good luck...)
4131 * 'external' for arrays with externally managed metadata,
4132 * or N.M for internally known formats
4135 metadata_show(struct mddev *mddev, char *page)
4137 if (mddev->persistent)
4138 return sprintf(page, "%d.%d\n",
4139 mddev->major_version, mddev->minor_version);
4140 else if (mddev->external)
4141 return sprintf(page, "external:%s\n", mddev->metadata_type);
4143 return sprintf(page, "none\n");
4147 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4151 /* Changing the details of 'external' metadata is
4152 * always permitted. Otherwise there must be
4153 * no devices attached to the array.
4155 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4157 else if (!list_empty(&mddev->disks))
4160 if (cmd_match(buf, "none")) {
4161 mddev->persistent = 0;
4162 mddev->external = 0;
4163 mddev->major_version = 0;
4164 mddev->minor_version = 90;
4167 if (strncmp(buf, "external:", 9) == 0) {
4168 size_t namelen = len-9;
4169 if (namelen >= sizeof(mddev->metadata_type))
4170 namelen = sizeof(mddev->metadata_type)-1;
4171 strncpy(mddev->metadata_type, buf+9, namelen);
4172 mddev->metadata_type[namelen] = 0;
4173 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4174 mddev->metadata_type[--namelen] = 0;
4175 mddev->persistent = 0;
4176 mddev->external = 1;
4177 mddev->major_version = 0;
4178 mddev->minor_version = 90;
4181 major = simple_strtoul(buf, &e, 10);
4182 if (e==buf || *e != '.')
4185 minor = simple_strtoul(buf, &e, 10);
4186 if (e==buf || (*e && *e != '\n') )
4188 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4190 mddev->major_version = major;
4191 mddev->minor_version = minor;
4192 mddev->persistent = 1;
4193 mddev->external = 0;
4197 static struct md_sysfs_entry md_metadata =
4198 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4201 action_show(struct mddev *mddev, char *page)
4203 char *type = "idle";
4204 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4206 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4207 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4208 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4210 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4211 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4213 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4217 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4220 return sprintf(page, "%s\n", type);
4224 action_store(struct mddev *mddev, const char *page, size_t len)
4226 if (!mddev->pers || !mddev->pers->sync_request)
4229 if (cmd_match(page, "frozen"))
4230 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4232 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4234 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4235 if (mddev->sync_thread) {
4236 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4237 md_reap_sync_thread(mddev);
4239 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4240 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4242 else if (cmd_match(page, "resync"))
4243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4244 else if (cmd_match(page, "recover")) {
4245 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4246 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4247 } else if (cmd_match(page, "reshape")) {
4249 if (mddev->pers->start_reshape == NULL)
4251 err = mddev->pers->start_reshape(mddev);
4254 sysfs_notify(&mddev->kobj, NULL, "degraded");
4256 if (cmd_match(page, "check"))
4257 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4258 else if (!cmd_match(page, "repair"))
4260 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4261 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4263 if (mddev->ro == 2) {
4264 /* A write to sync_action is enough to justify
4265 * canceling read-auto mode
4268 md_wakeup_thread(mddev->sync_thread);
4270 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4271 md_wakeup_thread(mddev->thread);
4272 sysfs_notify_dirent_safe(mddev->sysfs_action);
4277 mismatch_cnt_show(struct mddev *mddev, char *page)
4279 return sprintf(page, "%llu\n",
4280 (unsigned long long)
4281 atomic64_read(&mddev->resync_mismatches));
4284 static struct md_sysfs_entry md_scan_mode =
4285 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4288 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4291 sync_min_show(struct mddev *mddev, char *page)
4293 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4294 mddev->sync_speed_min ? "local": "system");
4298 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4302 if (strncmp(buf, "system", 6)==0) {
4303 mddev->sync_speed_min = 0;
4306 min = simple_strtoul(buf, &e, 10);
4307 if (buf == e || (*e && *e != '\n') || min <= 0)
4309 mddev->sync_speed_min = min;
4313 static struct md_sysfs_entry md_sync_min =
4314 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4317 sync_max_show(struct mddev *mddev, char *page)
4319 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4320 mddev->sync_speed_max ? "local": "system");
4324 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4328 if (strncmp(buf, "system", 6)==0) {
4329 mddev->sync_speed_max = 0;
4332 max = simple_strtoul(buf, &e, 10);
4333 if (buf == e || (*e && *e != '\n') || max <= 0)
4335 mddev->sync_speed_max = max;
4339 static struct md_sysfs_entry md_sync_max =
4340 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4343 degraded_show(struct mddev *mddev, char *page)
4345 return sprintf(page, "%d\n", mddev->degraded);
4347 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4350 sync_force_parallel_show(struct mddev *mddev, char *page)
4352 return sprintf(page, "%d\n", mddev->parallel_resync);
4356 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4360 if (strict_strtol(buf, 10, &n))
4363 if (n != 0 && n != 1)
4366 mddev->parallel_resync = n;
4368 if (mddev->sync_thread)
4369 wake_up(&resync_wait);
4374 /* force parallel resync, even with shared block devices */
4375 static struct md_sysfs_entry md_sync_force_parallel =
4376 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4377 sync_force_parallel_show, sync_force_parallel_store);
4380 sync_speed_show(struct mddev *mddev, char *page)
4382 unsigned long resync, dt, db;
4383 if (mddev->curr_resync == 0)
4384 return sprintf(page, "none\n");
4385 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4386 dt = (jiffies - mddev->resync_mark) / HZ;
4388 db = resync - mddev->resync_mark_cnt;
4389 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4392 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4395 sync_completed_show(struct mddev *mddev, char *page)
4397 unsigned long long max_sectors, resync;
4399 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4400 return sprintf(page, "none\n");
4402 if (mddev->curr_resync == 1 ||
4403 mddev->curr_resync == 2)
4404 return sprintf(page, "delayed\n");
4406 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4407 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4408 max_sectors = mddev->resync_max_sectors;
4410 max_sectors = mddev->dev_sectors;
4412 resync = mddev->curr_resync_completed;
4413 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4416 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4419 min_sync_show(struct mddev *mddev, char *page)
4421 return sprintf(page, "%llu\n",
4422 (unsigned long long)mddev->resync_min);
4425 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4427 unsigned long long min;
4428 if (strict_strtoull(buf, 10, &min))
4430 if (min > mddev->resync_max)
4432 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4435 /* Must be a multiple of chunk_size */
4436 if (mddev->chunk_sectors) {
4437 sector_t temp = min;
4438 if (sector_div(temp, mddev->chunk_sectors))
4441 mddev->resync_min = min;
4446 static struct md_sysfs_entry md_min_sync =
4447 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4450 max_sync_show(struct mddev *mddev, char *page)
4452 if (mddev->resync_max == MaxSector)
4453 return sprintf(page, "max\n");
4455 return sprintf(page, "%llu\n",
4456 (unsigned long long)mddev->resync_max);
4459 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4461 if (strncmp(buf, "max", 3) == 0)
4462 mddev->resync_max = MaxSector;
4464 unsigned long long max;
4465 if (strict_strtoull(buf, 10, &max))
4467 if (max < mddev->resync_min)
4469 if (max < mddev->resync_max &&
4471 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4474 /* Must be a multiple of chunk_size */
4475 if (mddev->chunk_sectors) {
4476 sector_t temp = max;
4477 if (sector_div(temp, mddev->chunk_sectors))
4480 mddev->resync_max = max;
4482 wake_up(&mddev->recovery_wait);
4486 static struct md_sysfs_entry md_max_sync =
4487 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4490 suspend_lo_show(struct mddev *mddev, char *page)
4492 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4496 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4499 unsigned long long new = simple_strtoull(buf, &e, 10);
4500 unsigned long long old = mddev->suspend_lo;
4502 if (mddev->pers == NULL ||
4503 mddev->pers->quiesce == NULL)
4505 if (buf == e || (*e && *e != '\n'))
4508 mddev->suspend_lo = new;
4510 /* Shrinking suspended region */
4511 mddev->pers->quiesce(mddev, 2);
4513 /* Expanding suspended region - need to wait */
4514 mddev->pers->quiesce(mddev, 1);
4515 mddev->pers->quiesce(mddev, 0);
4519 static struct md_sysfs_entry md_suspend_lo =
4520 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4524 suspend_hi_show(struct mddev *mddev, char *page)
4526 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4530 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4533 unsigned long long new = simple_strtoull(buf, &e, 10);
4534 unsigned long long old = mddev->suspend_hi;
4536 if (mddev->pers == NULL ||
4537 mddev->pers->quiesce == NULL)
4539 if (buf == e || (*e && *e != '\n'))
4542 mddev->suspend_hi = new;
4544 /* Shrinking suspended region */
4545 mddev->pers->quiesce(mddev, 2);
4547 /* Expanding suspended region - need to wait */
4548 mddev->pers->quiesce(mddev, 1);
4549 mddev->pers->quiesce(mddev, 0);
4553 static struct md_sysfs_entry md_suspend_hi =
4554 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4557 reshape_position_show(struct mddev *mddev, char *page)
4559 if (mddev->reshape_position != MaxSector)
4560 return sprintf(page, "%llu\n",
4561 (unsigned long long)mddev->reshape_position);
4562 strcpy(page, "none\n");
4567 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4569 struct md_rdev *rdev;
4571 unsigned long long new = simple_strtoull(buf, &e, 10);
4574 if (buf == e || (*e && *e != '\n'))
4576 mddev->reshape_position = new;
4577 mddev->delta_disks = 0;
4578 mddev->reshape_backwards = 0;
4579 mddev->new_level = mddev->level;
4580 mddev->new_layout = mddev->layout;
4581 mddev->new_chunk_sectors = mddev->chunk_sectors;
4582 rdev_for_each(rdev, mddev)
4583 rdev->new_data_offset = rdev->data_offset;
4587 static struct md_sysfs_entry md_reshape_position =
4588 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4589 reshape_position_store);
4592 reshape_direction_show(struct mddev *mddev, char *page)
4594 return sprintf(page, "%s\n",
4595 mddev->reshape_backwards ? "backwards" : "forwards");
4599 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4602 if (cmd_match(buf, "forwards"))
4604 else if (cmd_match(buf, "backwards"))
4608 if (mddev->reshape_backwards == backwards)
4611 /* check if we are allowed to change */
4612 if (mddev->delta_disks)
4615 if (mddev->persistent &&
4616 mddev->major_version == 0)
4619 mddev->reshape_backwards = backwards;
4623 static struct md_sysfs_entry md_reshape_direction =
4624 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4625 reshape_direction_store);
4628 array_size_show(struct mddev *mddev, char *page)
4630 if (mddev->external_size)
4631 return sprintf(page, "%llu\n",
4632 (unsigned long long)mddev->array_sectors/2);
4634 return sprintf(page, "default\n");
4638 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4642 if (strncmp(buf, "default", 7) == 0) {
4644 sectors = mddev->pers->size(mddev, 0, 0);
4646 sectors = mddev->array_sectors;
4648 mddev->external_size = 0;
4650 if (strict_blocks_to_sectors(buf, §ors) < 0)
4652 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4655 mddev->external_size = 1;
4658 mddev->array_sectors = sectors;
4660 set_capacity(mddev->gendisk, mddev->array_sectors);
4661 revalidate_disk(mddev->gendisk);
4666 static struct md_sysfs_entry md_array_size =
4667 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4670 static struct attribute *md_default_attrs[] = {
4673 &md_raid_disks.attr,
4674 &md_chunk_size.attr,
4676 &md_resync_start.attr,
4678 &md_new_device.attr,
4679 &md_safe_delay.attr,
4680 &md_array_state.attr,
4681 &md_reshape_position.attr,
4682 &md_reshape_direction.attr,
4683 &md_array_size.attr,
4684 &max_corr_read_errors.attr,
4688 static struct attribute *md_redundancy_attrs[] = {
4690 &md_mismatches.attr,
4693 &md_sync_speed.attr,
4694 &md_sync_force_parallel.attr,
4695 &md_sync_completed.attr,
4698 &md_suspend_lo.attr,
4699 &md_suspend_hi.attr,
4704 static struct attribute_group md_redundancy_group = {
4706 .attrs = md_redundancy_attrs,
4711 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4713 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4714 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4719 spin_lock(&all_mddevs_lock);
4720 if (list_empty(&mddev->all_mddevs)) {
4721 spin_unlock(&all_mddevs_lock);
4725 spin_unlock(&all_mddevs_lock);
4727 rv = mddev_lock(mddev);
4729 rv = entry->show(mddev, page);
4730 mddev_unlock(mddev);
4737 md_attr_store(struct kobject *kobj, struct attribute *attr,
4738 const char *page, size_t length)
4740 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4741 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4746 if (!capable(CAP_SYS_ADMIN))
4748 spin_lock(&all_mddevs_lock);
4749 if (list_empty(&mddev->all_mddevs)) {
4750 spin_unlock(&all_mddevs_lock);
4754 spin_unlock(&all_mddevs_lock);
4755 if (entry->store == new_dev_store)
4756 flush_workqueue(md_misc_wq);
4757 rv = mddev_lock(mddev);
4759 rv = entry->store(mddev, page, length);
4760 mddev_unlock(mddev);
4766 static void md_free(struct kobject *ko)
4768 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4770 if (mddev->sysfs_state)
4771 sysfs_put(mddev->sysfs_state);
4773 if (mddev->gendisk) {
4774 del_gendisk(mddev->gendisk);
4775 put_disk(mddev->gendisk);
4778 blk_cleanup_queue(mddev->queue);
4783 static const struct sysfs_ops md_sysfs_ops = {
4784 .show = md_attr_show,
4785 .store = md_attr_store,
4787 static struct kobj_type md_ktype = {
4789 .sysfs_ops = &md_sysfs_ops,
4790 .default_attrs = md_default_attrs,
4795 static void mddev_delayed_delete(struct work_struct *ws)
4797 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4799 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4800 kobject_del(&mddev->kobj);
4801 kobject_put(&mddev->kobj);
4804 static int md_alloc(dev_t dev, char *name)
4806 static DEFINE_MUTEX(disks_mutex);
4807 struct mddev *mddev = mddev_find(dev);
4808 struct gendisk *disk;
4817 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4818 shift = partitioned ? MdpMinorShift : 0;
4819 unit = MINOR(mddev->unit) >> shift;
4821 /* wait for any previous instance of this device to be
4822 * completely removed (mddev_delayed_delete).
4824 flush_workqueue(md_misc_wq);
4826 mutex_lock(&disks_mutex);
4832 /* Need to ensure that 'name' is not a duplicate.
4834 struct mddev *mddev2;
4835 spin_lock(&all_mddevs_lock);
4837 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4838 if (mddev2->gendisk &&
4839 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4840 spin_unlock(&all_mddevs_lock);
4843 spin_unlock(&all_mddevs_lock);
4847 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4850 mddev->queue->queuedata = mddev;
4852 blk_queue_make_request(mddev->queue, md_make_request);
4853 blk_set_stacking_limits(&mddev->queue->limits);
4855 disk = alloc_disk(1 << shift);
4857 blk_cleanup_queue(mddev->queue);
4858 mddev->queue = NULL;
4861 disk->major = MAJOR(mddev->unit);
4862 disk->first_minor = unit << shift;
4864 strcpy(disk->disk_name, name);
4865 else if (partitioned)
4866 sprintf(disk->disk_name, "md_d%d", unit);
4868 sprintf(disk->disk_name, "md%d", unit);
4869 disk->fops = &md_fops;
4870 disk->private_data = mddev;
4871 disk->queue = mddev->queue;
4872 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4873 /* Allow extended partitions. This makes the
4874 * 'mdp' device redundant, but we can't really
4877 disk->flags |= GENHD_FL_EXT_DEVT;
4878 mddev->gendisk = disk;
4879 /* As soon as we call add_disk(), another thread could get
4880 * through to md_open, so make sure it doesn't get too far
4882 mutex_lock(&mddev->open_mutex);
4885 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4886 &disk_to_dev(disk)->kobj, "%s", "md");
4888 /* This isn't possible, but as kobject_init_and_add is marked
4889 * __must_check, we must do something with the result
4891 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4895 if (mddev->kobj.sd &&
4896 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4897 printk(KERN_DEBUG "pointless warning\n");
4898 mutex_unlock(&mddev->open_mutex);
4900 mutex_unlock(&disks_mutex);
4901 if (!error && mddev->kobj.sd) {
4902 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4903 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4909 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4911 md_alloc(dev, NULL);
4915 static int add_named_array(const char *val, struct kernel_param *kp)
4917 /* val must be "md_*" where * is not all digits.
4918 * We allocate an array with a large free minor number, and
4919 * set the name to val. val must not already be an active name.
4921 int len = strlen(val);
4922 char buf[DISK_NAME_LEN];
4924 while (len && val[len-1] == '\n')
4926 if (len >= DISK_NAME_LEN)
4928 strlcpy(buf, val, len+1);
4929 if (strncmp(buf, "md_", 3) != 0)
4931 return md_alloc(0, buf);
4934 static void md_safemode_timeout(unsigned long data)
4936 struct mddev *mddev = (struct mddev *) data;
4938 if (!atomic_read(&mddev->writes_pending)) {
4939 mddev->safemode = 1;
4940 if (mddev->external)
4941 sysfs_notify_dirent_safe(mddev->sysfs_state);
4943 md_wakeup_thread(mddev->thread);
4946 static int start_dirty_degraded;
4948 int md_run(struct mddev *mddev)
4951 struct md_rdev *rdev;
4952 struct md_personality *pers;
4954 if (list_empty(&mddev->disks))
4955 /* cannot run an array with no devices.. */
4960 /* Cannot run until previous stop completes properly */
4961 if (mddev->sysfs_active)
4965 * Analyze all RAID superblock(s)
4967 if (!mddev->raid_disks) {
4968 if (!mddev->persistent)
4973 if (mddev->level != LEVEL_NONE)
4974 request_module("md-level-%d", mddev->level);
4975 else if (mddev->clevel[0])
4976 request_module("md-%s", mddev->clevel);
4979 * Drop all container device buffers, from now on
4980 * the only valid external interface is through the md
4983 rdev_for_each(rdev, mddev) {
4984 if (test_bit(Faulty, &rdev->flags))
4986 sync_blockdev(rdev->bdev);
4987 invalidate_bdev(rdev->bdev);
4989 /* perform some consistency tests on the device.
4990 * We don't want the data to overlap the metadata,
4991 * Internal Bitmap issues have been handled elsewhere.
4993 if (rdev->meta_bdev) {
4994 /* Nothing to check */;
4995 } else if (rdev->data_offset < rdev->sb_start) {
4996 if (mddev->dev_sectors &&
4997 rdev->data_offset + mddev->dev_sectors
4999 printk("md: %s: data overlaps metadata\n",
5004 if (rdev->sb_start + rdev->sb_size/512
5005 > rdev->data_offset) {
5006 printk("md: %s: metadata overlaps data\n",
5011 sysfs_notify_dirent_safe(rdev->sysfs_state);
5014 if (mddev->bio_set == NULL)
5015 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5017 spin_lock(&pers_lock);
5018 pers = find_pers(mddev->level, mddev->clevel);
5019 if (!pers || !try_module_get(pers->owner)) {
5020 spin_unlock(&pers_lock);
5021 if (mddev->level != LEVEL_NONE)
5022 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5025 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5030 spin_unlock(&pers_lock);
5031 if (mddev->level != pers->level) {
5032 mddev->level = pers->level;
5033 mddev->new_level = pers->level;
5035 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5037 if (mddev->reshape_position != MaxSector &&
5038 pers->start_reshape == NULL) {
5039 /* This personality cannot handle reshaping... */
5041 module_put(pers->owner);
5045 if (pers->sync_request) {
5046 /* Warn if this is a potentially silly
5049 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5050 struct md_rdev *rdev2;
5053 rdev_for_each(rdev, mddev)
5054 rdev_for_each(rdev2, mddev) {
5056 rdev->bdev->bd_contains ==
5057 rdev2->bdev->bd_contains) {
5059 "%s: WARNING: %s appears to be"
5060 " on the same physical disk as"
5063 bdevname(rdev->bdev,b),
5064 bdevname(rdev2->bdev,b2));
5071 "True protection against single-disk"
5072 " failure might be compromised.\n");
5075 mddev->recovery = 0;
5076 /* may be over-ridden by personality */
5077 mddev->resync_max_sectors = mddev->dev_sectors;
5079 mddev->ok_start_degraded = start_dirty_degraded;
5081 if (start_readonly && mddev->ro == 0)
5082 mddev->ro = 2; /* read-only, but switch on first write */
5084 err = mddev->pers->run(mddev);
5086 printk(KERN_ERR "md: pers->run() failed ...\n");
5087 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5088 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5089 " but 'external_size' not in effect?\n", __func__);
5091 "md: invalid array_size %llu > default size %llu\n",
5092 (unsigned long long)mddev->array_sectors / 2,
5093 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5095 mddev->pers->stop(mddev);
5097 if (err == 0 && mddev->pers->sync_request &&
5098 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5099 err = bitmap_create(mddev);
5101 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5102 mdname(mddev), err);
5103 mddev->pers->stop(mddev);
5107 module_put(mddev->pers->owner);
5109 bitmap_destroy(mddev);
5112 if (mddev->pers->sync_request) {
5113 if (mddev->kobj.sd &&
5114 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5116 "md: cannot register extra attributes for %s\n",
5118 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5119 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5122 atomic_set(&mddev->writes_pending,0);
5123 atomic_set(&mddev->max_corr_read_errors,
5124 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5125 mddev->safemode = 0;
5126 mddev->safemode_timer.function = md_safemode_timeout;
5127 mddev->safemode_timer.data = (unsigned long) mddev;
5128 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5132 rdev_for_each(rdev, mddev)
5133 if (rdev->raid_disk >= 0)
5134 if (sysfs_link_rdev(mddev, rdev))
5135 /* failure here is OK */;
5137 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5140 md_update_sb(mddev, 0);
5142 md_new_event(mddev);
5143 sysfs_notify_dirent_safe(mddev->sysfs_state);
5144 sysfs_notify_dirent_safe(mddev->sysfs_action);
5145 sysfs_notify(&mddev->kobj, NULL, "degraded");
5148 EXPORT_SYMBOL_GPL(md_run);
5150 static int do_md_run(struct mddev *mddev)
5154 err = md_run(mddev);
5157 err = bitmap_load(mddev);
5159 bitmap_destroy(mddev);
5163 md_wakeup_thread(mddev->thread);
5164 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5166 set_capacity(mddev->gendisk, mddev->array_sectors);
5167 revalidate_disk(mddev->gendisk);
5169 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5174 static int restart_array(struct mddev *mddev)
5176 struct gendisk *disk = mddev->gendisk;
5178 /* Complain if it has no devices */
5179 if (list_empty(&mddev->disks))
5185 mddev->safemode = 0;
5187 set_disk_ro(disk, 0);
5188 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5190 /* Kick recovery or resync if necessary */
5191 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5192 md_wakeup_thread(mddev->thread);
5193 md_wakeup_thread(mddev->sync_thread);
5194 sysfs_notify_dirent_safe(mddev->sysfs_state);
5198 /* similar to deny_write_access, but accounts for our holding a reference
5199 * to the file ourselves */
5200 static int deny_bitmap_write_access(struct file * file)
5202 struct inode *inode = file->f_mapping->host;
5204 spin_lock(&inode->i_lock);
5205 if (atomic_read(&inode->i_writecount) > 1) {
5206 spin_unlock(&inode->i_lock);
5209 atomic_set(&inode->i_writecount, -1);
5210 spin_unlock(&inode->i_lock);
5215 void restore_bitmap_write_access(struct file *file)
5217 struct inode *inode = file->f_mapping->host;
5219 spin_lock(&inode->i_lock);
5220 atomic_set(&inode->i_writecount, 1);
5221 spin_unlock(&inode->i_lock);
5224 static void md_clean(struct mddev *mddev)
5226 mddev->array_sectors = 0;
5227 mddev->external_size = 0;
5228 mddev->dev_sectors = 0;
5229 mddev->raid_disks = 0;
5230 mddev->recovery_cp = 0;
5231 mddev->resync_min = 0;
5232 mddev->resync_max = MaxSector;
5233 mddev->reshape_position = MaxSector;
5234 mddev->external = 0;
5235 mddev->persistent = 0;
5236 mddev->level = LEVEL_NONE;
5237 mddev->clevel[0] = 0;
5240 mddev->metadata_type[0] = 0;
5241 mddev->chunk_sectors = 0;
5242 mddev->ctime = mddev->utime = 0;
5244 mddev->max_disks = 0;
5246 mddev->can_decrease_events = 0;
5247 mddev->delta_disks = 0;
5248 mddev->reshape_backwards = 0;
5249 mddev->new_level = LEVEL_NONE;
5250 mddev->new_layout = 0;
5251 mddev->new_chunk_sectors = 0;
5252 mddev->curr_resync = 0;
5253 atomic64_set(&mddev->resync_mismatches, 0);
5254 mddev->suspend_lo = mddev->suspend_hi = 0;
5255 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5256 mddev->recovery = 0;
5259 mddev->degraded = 0;
5260 mddev->safemode = 0;
5261 mddev->merge_check_needed = 0;
5262 mddev->bitmap_info.offset = 0;
5263 mddev->bitmap_info.default_offset = 0;
5264 mddev->bitmap_info.default_space = 0;
5265 mddev->bitmap_info.chunksize = 0;
5266 mddev->bitmap_info.daemon_sleep = 0;
5267 mddev->bitmap_info.max_write_behind = 0;
5270 static void __md_stop_writes(struct mddev *mddev)
5272 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5273 if (mddev->sync_thread) {
5274 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5275 md_reap_sync_thread(mddev);
5278 del_timer_sync(&mddev->safemode_timer);
5280 bitmap_flush(mddev);
5281 md_super_wait(mddev);
5283 if (mddev->ro == 0 &&
5284 (!mddev->in_sync || mddev->flags)) {
5285 /* mark array as shutdown cleanly */
5287 md_update_sb(mddev, 1);
5291 void md_stop_writes(struct mddev *mddev)
5294 __md_stop_writes(mddev);
5295 mddev_unlock(mddev);
5297 EXPORT_SYMBOL_GPL(md_stop_writes);
5299 static void __md_stop(struct mddev *mddev)
5302 mddev->pers->stop(mddev);
5303 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5304 mddev->to_remove = &md_redundancy_group;
5305 module_put(mddev->pers->owner);
5307 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5310 void md_stop(struct mddev *mddev)
5312 /* stop the array and free an attached data structures.
5313 * This is called from dm-raid
5316 bitmap_destroy(mddev);
5318 bioset_free(mddev->bio_set);
5321 EXPORT_SYMBOL_GPL(md_stop);
5323 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5326 mutex_lock(&mddev->open_mutex);
5327 if (atomic_read(&mddev->openers) > !!bdev) {
5328 printk("md: %s still in use.\n",mdname(mddev));
5333 sync_blockdev(bdev);
5335 __md_stop_writes(mddev);
5341 set_disk_ro(mddev->gendisk, 1);
5342 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5343 sysfs_notify_dirent_safe(mddev->sysfs_state);
5347 mutex_unlock(&mddev->open_mutex);
5352 * 0 - completely stop and dis-assemble array
5353 * 2 - stop but do not disassemble array
5355 static int do_md_stop(struct mddev * mddev, int mode,
5356 struct block_device *bdev)
5358 struct gendisk *disk = mddev->gendisk;
5359 struct md_rdev *rdev;
5361 mutex_lock(&mddev->open_mutex);
5362 if (atomic_read(&mddev->openers) > !!bdev ||
5363 mddev->sysfs_active) {
5364 printk("md: %s still in use.\n",mdname(mddev));
5365 mutex_unlock(&mddev->open_mutex);
5369 /* It is possible IO was issued on some other
5370 * open file which was closed before we took ->open_mutex.
5371 * As that was not the last close __blkdev_put will not
5372 * have called sync_blockdev, so we must.
5374 sync_blockdev(bdev);
5378 set_disk_ro(disk, 0);
5380 __md_stop_writes(mddev);
5382 mddev->queue->merge_bvec_fn = NULL;
5383 mddev->queue->backing_dev_info.congested_fn = NULL;
5385 /* tell userspace to handle 'inactive' */
5386 sysfs_notify_dirent_safe(mddev->sysfs_state);
5388 rdev_for_each(rdev, mddev)
5389 if (rdev->raid_disk >= 0)
5390 sysfs_unlink_rdev(mddev, rdev);
5392 set_capacity(disk, 0);
5393 mutex_unlock(&mddev->open_mutex);
5395 revalidate_disk(disk);
5400 mutex_unlock(&mddev->open_mutex);
5402 * Free resources if final stop
5405 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5407 bitmap_destroy(mddev);
5408 if (mddev->bitmap_info.file) {
5409 restore_bitmap_write_access(mddev->bitmap_info.file);
5410 fput(mddev->bitmap_info.file);
5411 mddev->bitmap_info.file = NULL;
5413 mddev->bitmap_info.offset = 0;
5415 export_array(mddev);
5418 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5419 if (mddev->hold_active == UNTIL_STOP)
5420 mddev->hold_active = 0;
5422 blk_integrity_unregister(disk);
5423 md_new_event(mddev);
5424 sysfs_notify_dirent_safe(mddev->sysfs_state);
5429 static void autorun_array(struct mddev *mddev)
5431 struct md_rdev *rdev;
5434 if (list_empty(&mddev->disks))
5437 printk(KERN_INFO "md: running: ");
5439 rdev_for_each(rdev, mddev) {
5440 char b[BDEVNAME_SIZE];
5441 printk("<%s>", bdevname(rdev->bdev,b));
5445 err = do_md_run(mddev);
5447 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5448 do_md_stop(mddev, 0, NULL);
5453 * lets try to run arrays based on all disks that have arrived
5454 * until now. (those are in pending_raid_disks)
5456 * the method: pick the first pending disk, collect all disks with
5457 * the same UUID, remove all from the pending list and put them into
5458 * the 'same_array' list. Then order this list based on superblock
5459 * update time (freshest comes first), kick out 'old' disks and
5460 * compare superblocks. If everything's fine then run it.
5462 * If "unit" is allocated, then bump its reference count
5464 static void autorun_devices(int part)
5466 struct md_rdev *rdev0, *rdev, *tmp;
5467 struct mddev *mddev;
5468 char b[BDEVNAME_SIZE];
5470 printk(KERN_INFO "md: autorun ...\n");
5471 while (!list_empty(&pending_raid_disks)) {
5474 LIST_HEAD(candidates);
5475 rdev0 = list_entry(pending_raid_disks.next,
5476 struct md_rdev, same_set);
5478 printk(KERN_INFO "md: considering %s ...\n",
5479 bdevname(rdev0->bdev,b));
5480 INIT_LIST_HEAD(&candidates);
5481 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5482 if (super_90_load(rdev, rdev0, 0) >= 0) {
5483 printk(KERN_INFO "md: adding %s ...\n",
5484 bdevname(rdev->bdev,b));
5485 list_move(&rdev->same_set, &candidates);
5488 * now we have a set of devices, with all of them having
5489 * mostly sane superblocks. It's time to allocate the
5493 dev = MKDEV(mdp_major,
5494 rdev0->preferred_minor << MdpMinorShift);
5495 unit = MINOR(dev) >> MdpMinorShift;
5497 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5500 if (rdev0->preferred_minor != unit) {
5501 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5502 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5506 md_probe(dev, NULL, NULL);
5507 mddev = mddev_find(dev);
5508 if (!mddev || !mddev->gendisk) {
5512 "md: cannot allocate memory for md drive.\n");
5515 if (mddev_lock(mddev))
5516 printk(KERN_WARNING "md: %s locked, cannot run\n",
5518 else if (mddev->raid_disks || mddev->major_version
5519 || !list_empty(&mddev->disks)) {
5521 "md: %s already running, cannot run %s\n",
5522 mdname(mddev), bdevname(rdev0->bdev,b));
5523 mddev_unlock(mddev);
5525 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5526 mddev->persistent = 1;
5527 rdev_for_each_list(rdev, tmp, &candidates) {
5528 list_del_init(&rdev->same_set);
5529 if (bind_rdev_to_array(rdev, mddev))
5532 autorun_array(mddev);
5533 mddev_unlock(mddev);
5535 /* on success, candidates will be empty, on error
5538 rdev_for_each_list(rdev, tmp, &candidates) {
5539 list_del_init(&rdev->same_set);
5544 printk(KERN_INFO "md: ... autorun DONE.\n");
5546 #endif /* !MODULE */
5548 static int get_version(void __user * arg)
5552 ver.major = MD_MAJOR_VERSION;
5553 ver.minor = MD_MINOR_VERSION;
5554 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5556 if (copy_to_user(arg, &ver, sizeof(ver)))
5562 static int get_array_info(struct mddev * mddev, void __user * arg)
5564 mdu_array_info_t info;
5565 int nr,working,insync,failed,spare;
5566 struct md_rdev *rdev;
5568 nr = working = insync = failed = spare = 0;
5570 rdev_for_each_rcu(rdev, mddev) {
5572 if (test_bit(Faulty, &rdev->flags))
5576 if (test_bit(In_sync, &rdev->flags))
5584 info.major_version = mddev->major_version;
5585 info.minor_version = mddev->minor_version;
5586 info.patch_version = MD_PATCHLEVEL_VERSION;
5587 info.ctime = mddev->ctime;
5588 info.level = mddev->level;
5589 info.size = mddev->dev_sectors / 2;
5590 if (info.size != mddev->dev_sectors / 2) /* overflow */
5593 info.raid_disks = mddev->raid_disks;
5594 info.md_minor = mddev->md_minor;
5595 info.not_persistent= !mddev->persistent;
5597 info.utime = mddev->utime;
5600 info.state = (1<<MD_SB_CLEAN);
5601 if (mddev->bitmap && mddev->bitmap_info.offset)
5602 info.state = (1<<MD_SB_BITMAP_PRESENT);
5603 info.active_disks = insync;
5604 info.working_disks = working;
5605 info.failed_disks = failed;
5606 info.spare_disks = spare;
5608 info.layout = mddev->layout;
5609 info.chunk_size = mddev->chunk_sectors << 9;
5611 if (copy_to_user(arg, &info, sizeof(info)))
5617 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5619 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5620 char *ptr, *buf = NULL;
5623 if (md_allow_write(mddev))
5624 file = kmalloc(sizeof(*file), GFP_NOIO);
5626 file = kmalloc(sizeof(*file), GFP_KERNEL);
5631 /* bitmap disabled, zero the first byte and copy out */
5632 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5633 file->pathname[0] = '\0';
5637 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5641 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5642 buf, sizeof(file->pathname));
5646 strcpy(file->pathname, ptr);
5650 if (copy_to_user(arg, file, sizeof(*file)))
5658 static int get_disk_info(struct mddev * mddev, void __user * arg)
5660 mdu_disk_info_t info;
5661 struct md_rdev *rdev;
5663 if (copy_from_user(&info, arg, sizeof(info)))
5667 rdev = find_rdev_nr_rcu(mddev, info.number);
5669 info.major = MAJOR(rdev->bdev->bd_dev);
5670 info.minor = MINOR(rdev->bdev->bd_dev);
5671 info.raid_disk = rdev->raid_disk;
5673 if (test_bit(Faulty, &rdev->flags))
5674 info.state |= (1<<MD_DISK_FAULTY);
5675 else if (test_bit(In_sync, &rdev->flags)) {
5676 info.state |= (1<<MD_DISK_ACTIVE);
5677 info.state |= (1<<MD_DISK_SYNC);
5679 if (test_bit(WriteMostly, &rdev->flags))
5680 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5682 info.major = info.minor = 0;
5683 info.raid_disk = -1;
5684 info.state = (1<<MD_DISK_REMOVED);
5688 if (copy_to_user(arg, &info, sizeof(info)))
5694 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5696 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5697 struct md_rdev *rdev;
5698 dev_t dev = MKDEV(info->major,info->minor);
5700 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5703 if (!mddev->raid_disks) {
5705 /* expecting a device which has a superblock */
5706 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5709 "md: md_import_device returned %ld\n",
5711 return PTR_ERR(rdev);
5713 if (!list_empty(&mddev->disks)) {
5714 struct md_rdev *rdev0
5715 = list_entry(mddev->disks.next,
5716 struct md_rdev, same_set);
5717 err = super_types[mddev->major_version]
5718 .load_super(rdev, rdev0, mddev->minor_version);
5721 "md: %s has different UUID to %s\n",
5722 bdevname(rdev->bdev,b),
5723 bdevname(rdev0->bdev,b2));
5728 err = bind_rdev_to_array(rdev, mddev);
5735 * add_new_disk can be used once the array is assembled
5736 * to add "hot spares". They must already have a superblock
5741 if (!mddev->pers->hot_add_disk) {
5743 "%s: personality does not support diskops!\n",
5747 if (mddev->persistent)
5748 rdev = md_import_device(dev, mddev->major_version,
5749 mddev->minor_version);
5751 rdev = md_import_device(dev, -1, -1);
5754 "md: md_import_device returned %ld\n",
5756 return PTR_ERR(rdev);
5758 /* set saved_raid_disk if appropriate */
5759 if (!mddev->persistent) {
5760 if (info->state & (1<<MD_DISK_SYNC) &&
5761 info->raid_disk < mddev->raid_disks) {
5762 rdev->raid_disk = info->raid_disk;
5763 set_bit(In_sync, &rdev->flags);
5765 rdev->raid_disk = -1;
5767 super_types[mddev->major_version].
5768 validate_super(mddev, rdev);
5769 if ((info->state & (1<<MD_DISK_SYNC)) &&
5770 rdev->raid_disk != info->raid_disk) {
5771 /* This was a hot-add request, but events doesn't
5772 * match, so reject it.
5778 if (test_bit(In_sync, &rdev->flags))
5779 rdev->saved_raid_disk = rdev->raid_disk;
5781 rdev->saved_raid_disk = -1;
5783 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5784 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5785 set_bit(WriteMostly, &rdev->flags);
5787 clear_bit(WriteMostly, &rdev->flags);
5789 rdev->raid_disk = -1;
5790 err = bind_rdev_to_array(rdev, mddev);
5791 if (!err && !mddev->pers->hot_remove_disk) {
5792 /* If there is hot_add_disk but no hot_remove_disk
5793 * then added disks for geometry changes,
5794 * and should be added immediately.
5796 super_types[mddev->major_version].
5797 validate_super(mddev, rdev);
5798 err = mddev->pers->hot_add_disk(mddev, rdev);
5800 unbind_rdev_from_array(rdev);
5805 sysfs_notify_dirent_safe(rdev->sysfs_state);
5807 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5808 if (mddev->degraded)
5809 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5810 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5812 md_new_event(mddev);
5813 md_wakeup_thread(mddev->thread);
5817 /* otherwise, add_new_disk is only allowed
5818 * for major_version==0 superblocks
5820 if (mddev->major_version != 0) {
5821 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5826 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5828 rdev = md_import_device(dev, -1, 0);
5831 "md: error, md_import_device() returned %ld\n",
5833 return PTR_ERR(rdev);
5835 rdev->desc_nr = info->number;
5836 if (info->raid_disk < mddev->raid_disks)
5837 rdev->raid_disk = info->raid_disk;
5839 rdev->raid_disk = -1;
5841 if (rdev->raid_disk < mddev->raid_disks)
5842 if (info->state & (1<<MD_DISK_SYNC))
5843 set_bit(In_sync, &rdev->flags);
5845 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5846 set_bit(WriteMostly, &rdev->flags);
5848 if (!mddev->persistent) {
5849 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5850 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5852 rdev->sb_start = calc_dev_sboffset(rdev);
5853 rdev->sectors = rdev->sb_start;
5855 err = bind_rdev_to_array(rdev, mddev);
5865 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5867 char b[BDEVNAME_SIZE];
5868 struct md_rdev *rdev;
5870 rdev = find_rdev(mddev, dev);
5874 clear_bit(Blocked, &rdev->flags);
5875 remove_and_add_spares(mddev, rdev);
5877 if (rdev->raid_disk >= 0)
5880 kick_rdev_from_array(rdev);
5881 md_update_sb(mddev, 1);
5882 md_new_event(mddev);
5886 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5887 bdevname(rdev->bdev,b), mdname(mddev));
5891 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5893 char b[BDEVNAME_SIZE];
5895 struct md_rdev *rdev;
5900 if (mddev->major_version != 0) {
5901 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5902 " version-0 superblocks.\n",
5906 if (!mddev->pers->hot_add_disk) {
5908 "%s: personality does not support diskops!\n",
5913 rdev = md_import_device(dev, -1, 0);
5916 "md: error, md_import_device() returned %ld\n",
5921 if (mddev->persistent)
5922 rdev->sb_start = calc_dev_sboffset(rdev);
5924 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5926 rdev->sectors = rdev->sb_start;
5928 if (test_bit(Faulty, &rdev->flags)) {
5930 "md: can not hot-add faulty %s disk to %s!\n",
5931 bdevname(rdev->bdev,b), mdname(mddev));
5935 clear_bit(In_sync, &rdev->flags);
5937 rdev->saved_raid_disk = -1;
5938 err = bind_rdev_to_array(rdev, mddev);
5943 * The rest should better be atomic, we can have disk failures
5944 * noticed in interrupt contexts ...
5947 rdev->raid_disk = -1;
5949 md_update_sb(mddev, 1);
5952 * Kick recovery, maybe this spare has to be added to the
5953 * array immediately.
5955 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5956 md_wakeup_thread(mddev->thread);
5957 md_new_event(mddev);
5965 static int set_bitmap_file(struct mddev *mddev, int fd)
5970 if (!mddev->pers->quiesce)
5972 if (mddev->recovery || mddev->sync_thread)
5974 /* we should be able to change the bitmap.. */
5980 return -EEXIST; /* cannot add when bitmap is present */
5981 mddev->bitmap_info.file = fget(fd);
5983 if (mddev->bitmap_info.file == NULL) {
5984 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5989 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5991 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5993 fput(mddev->bitmap_info.file);
5994 mddev->bitmap_info.file = NULL;
5997 mddev->bitmap_info.offset = 0; /* file overrides offset */
5998 } else if (mddev->bitmap == NULL)
5999 return -ENOENT; /* cannot remove what isn't there */
6002 mddev->pers->quiesce(mddev, 1);
6004 err = bitmap_create(mddev);
6006 err = bitmap_load(mddev);
6008 if (fd < 0 || err) {
6009 bitmap_destroy(mddev);
6010 fd = -1; /* make sure to put the file */
6012 mddev->pers->quiesce(mddev, 0);
6015 if (mddev->bitmap_info.file) {
6016 restore_bitmap_write_access(mddev->bitmap_info.file);
6017 fput(mddev->bitmap_info.file);
6019 mddev->bitmap_info.file = NULL;
6026 * set_array_info is used two different ways
6027 * The original usage is when creating a new array.
6028 * In this usage, raid_disks is > 0 and it together with
6029 * level, size, not_persistent,layout,chunksize determine the
6030 * shape of the array.
6031 * This will always create an array with a type-0.90.0 superblock.
6032 * The newer usage is when assembling an array.
6033 * In this case raid_disks will be 0, and the major_version field is
6034 * use to determine which style super-blocks are to be found on the devices.
6035 * The minor and patch _version numbers are also kept incase the
6036 * super_block handler wishes to interpret them.
6038 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6041 if (info->raid_disks == 0) {
6042 /* just setting version number for superblock loading */
6043 if (info->major_version < 0 ||
6044 info->major_version >= ARRAY_SIZE(super_types) ||
6045 super_types[info->major_version].name == NULL) {
6046 /* maybe try to auto-load a module? */
6048 "md: superblock version %d not known\n",
6049 info->major_version);
6052 mddev->major_version = info->major_version;
6053 mddev->minor_version = info->minor_version;
6054 mddev->patch_version = info->patch_version;
6055 mddev->persistent = !info->not_persistent;
6056 /* ensure mddev_put doesn't delete this now that there
6057 * is some minimal configuration.
6059 mddev->ctime = get_seconds();
6062 mddev->major_version = MD_MAJOR_VERSION;
6063 mddev->minor_version = MD_MINOR_VERSION;
6064 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6065 mddev->ctime = get_seconds();
6067 mddev->level = info->level;
6068 mddev->clevel[0] = 0;
6069 mddev->dev_sectors = 2 * (sector_t)info->size;
6070 mddev->raid_disks = info->raid_disks;
6071 /* don't set md_minor, it is determined by which /dev/md* was
6074 if (info->state & (1<<MD_SB_CLEAN))
6075 mddev->recovery_cp = MaxSector;
6077 mddev->recovery_cp = 0;
6078 mddev->persistent = ! info->not_persistent;
6079 mddev->external = 0;
6081 mddev->layout = info->layout;
6082 mddev->chunk_sectors = info->chunk_size >> 9;
6084 mddev->max_disks = MD_SB_DISKS;
6086 if (mddev->persistent)
6088 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6090 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6091 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6092 mddev->bitmap_info.offset = 0;
6094 mddev->reshape_position = MaxSector;
6097 * Generate a 128 bit UUID
6099 get_random_bytes(mddev->uuid, 16);
6101 mddev->new_level = mddev->level;
6102 mddev->new_chunk_sectors = mddev->chunk_sectors;
6103 mddev->new_layout = mddev->layout;
6104 mddev->delta_disks = 0;
6105 mddev->reshape_backwards = 0;
6110 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6112 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6114 if (mddev->external_size)
6117 mddev->array_sectors = array_sectors;
6119 EXPORT_SYMBOL(md_set_array_sectors);
6121 static int update_size(struct mddev *mddev, sector_t num_sectors)
6123 struct md_rdev *rdev;
6125 int fit = (num_sectors == 0);
6127 if (mddev->pers->resize == NULL)
6129 /* The "num_sectors" is the number of sectors of each device that
6130 * is used. This can only make sense for arrays with redundancy.
6131 * linear and raid0 always use whatever space is available. We can only
6132 * consider changing this number if no resync or reconstruction is
6133 * happening, and if the new size is acceptable. It must fit before the
6134 * sb_start or, if that is <data_offset, it must fit before the size
6135 * of each device. If num_sectors is zero, we find the largest size
6138 if (mddev->sync_thread)
6141 rdev_for_each(rdev, mddev) {
6142 sector_t avail = rdev->sectors;
6144 if (fit && (num_sectors == 0 || num_sectors > avail))
6145 num_sectors = avail;
6146 if (avail < num_sectors)
6149 rv = mddev->pers->resize(mddev, num_sectors);
6151 revalidate_disk(mddev->gendisk);
6155 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6158 struct md_rdev *rdev;
6159 /* change the number of raid disks */
6160 if (mddev->pers->check_reshape == NULL)
6162 if (raid_disks <= 0 ||
6163 (mddev->max_disks && raid_disks >= mddev->max_disks))
6165 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6168 rdev_for_each(rdev, mddev) {
6169 if (mddev->raid_disks < raid_disks &&
6170 rdev->data_offset < rdev->new_data_offset)
6172 if (mddev->raid_disks > raid_disks &&
6173 rdev->data_offset > rdev->new_data_offset)
6177 mddev->delta_disks = raid_disks - mddev->raid_disks;
6178 if (mddev->delta_disks < 0)
6179 mddev->reshape_backwards = 1;
6180 else if (mddev->delta_disks > 0)
6181 mddev->reshape_backwards = 0;
6183 rv = mddev->pers->check_reshape(mddev);
6185 mddev->delta_disks = 0;
6186 mddev->reshape_backwards = 0;
6193 * update_array_info is used to change the configuration of an
6195 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6196 * fields in the info are checked against the array.
6197 * Any differences that cannot be handled will cause an error.
6198 * Normally, only one change can be managed at a time.
6200 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6206 /* calculate expected state,ignoring low bits */
6207 if (mddev->bitmap && mddev->bitmap_info.offset)
6208 state |= (1 << MD_SB_BITMAP_PRESENT);
6210 if (mddev->major_version != info->major_version ||
6211 mddev->minor_version != info->minor_version ||
6212 /* mddev->patch_version != info->patch_version || */
6213 mddev->ctime != info->ctime ||
6214 mddev->level != info->level ||
6215 /* mddev->layout != info->layout || */
6216 !mddev->persistent != info->not_persistent||
6217 mddev->chunk_sectors != info->chunk_size >> 9 ||
6218 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6219 ((state^info->state) & 0xfffffe00)
6222 /* Check there is only one change */
6223 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6225 if (mddev->raid_disks != info->raid_disks)
6227 if (mddev->layout != info->layout)
6229 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6236 if (mddev->layout != info->layout) {
6238 * we don't need to do anything at the md level, the
6239 * personality will take care of it all.
6241 if (mddev->pers->check_reshape == NULL)
6244 mddev->new_layout = info->layout;
6245 rv = mddev->pers->check_reshape(mddev);
6247 mddev->new_layout = mddev->layout;
6251 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6252 rv = update_size(mddev, (sector_t)info->size * 2);
6254 if (mddev->raid_disks != info->raid_disks)
6255 rv = update_raid_disks(mddev, info->raid_disks);
6257 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6258 if (mddev->pers->quiesce == NULL)
6260 if (mddev->recovery || mddev->sync_thread)
6262 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6263 /* add the bitmap */
6266 if (mddev->bitmap_info.default_offset == 0)
6268 mddev->bitmap_info.offset =
6269 mddev->bitmap_info.default_offset;
6270 mddev->bitmap_info.space =
6271 mddev->bitmap_info.default_space;
6272 mddev->pers->quiesce(mddev, 1);
6273 rv = bitmap_create(mddev);
6275 rv = bitmap_load(mddev);
6277 bitmap_destroy(mddev);
6278 mddev->pers->quiesce(mddev, 0);
6280 /* remove the bitmap */
6283 if (mddev->bitmap->storage.file)
6285 mddev->pers->quiesce(mddev, 1);
6286 bitmap_destroy(mddev);
6287 mddev->pers->quiesce(mddev, 0);
6288 mddev->bitmap_info.offset = 0;
6291 md_update_sb(mddev, 1);
6295 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6297 struct md_rdev *rdev;
6300 if (mddev->pers == NULL)
6304 rdev = find_rdev_rcu(mddev, dev);
6308 md_error(mddev, rdev);
6309 if (!test_bit(Faulty, &rdev->flags))
6317 * We have a problem here : there is no easy way to give a CHS
6318 * virtual geometry. We currently pretend that we have a 2 heads
6319 * 4 sectors (with a BIG number of cylinders...). This drives
6320 * dosfs just mad... ;-)
6322 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6324 struct mddev *mddev = bdev->bd_disk->private_data;
6328 geo->cylinders = mddev->array_sectors / 8;
6332 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6333 unsigned int cmd, unsigned long arg)
6336 void __user *argp = (void __user *)arg;
6337 struct mddev *mddev = NULL;
6342 case GET_ARRAY_INFO:
6346 if (!capable(CAP_SYS_ADMIN))
6351 * Commands dealing with the RAID driver but not any
6356 err = get_version(argp);
6359 case PRINT_RAID_DEBUG:
6367 autostart_arrays(arg);
6374 * Commands creating/starting a new array:
6377 mddev = bdev->bd_disk->private_data;
6384 /* Some actions do not requires the mutex */
6386 case GET_ARRAY_INFO:
6387 if (!mddev->raid_disks && !mddev->external)
6390 err = get_array_info(mddev, argp);
6394 if (!mddev->raid_disks && !mddev->external)
6397 err = get_disk_info(mddev, argp);
6400 case SET_DISK_FAULTY:
6401 err = set_disk_faulty(mddev, new_decode_dev(arg));
6405 if (cmd == ADD_NEW_DISK)
6406 /* need to ensure md_delayed_delete() has completed */
6407 flush_workqueue(md_misc_wq);
6409 err = mddev_lock(mddev);
6412 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6417 if (cmd == SET_ARRAY_INFO) {
6418 mdu_array_info_t info;
6420 memset(&info, 0, sizeof(info));
6421 else if (copy_from_user(&info, argp, sizeof(info))) {
6426 err = update_array_info(mddev, &info);
6428 printk(KERN_WARNING "md: couldn't update"
6429 " array info. %d\n", err);
6434 if (!list_empty(&mddev->disks)) {
6436 "md: array %s already has disks!\n",
6441 if (mddev->raid_disks) {
6443 "md: array %s already initialised!\n",
6448 err = set_array_info(mddev, &info);
6450 printk(KERN_WARNING "md: couldn't set"
6451 " array info. %d\n", err);
6458 * Commands querying/configuring an existing array:
6460 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6461 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6462 if ((!mddev->raid_disks && !mddev->external)
6463 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6464 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6465 && cmd != GET_BITMAP_FILE) {
6471 * Commands even a read-only array can execute:
6474 case GET_BITMAP_FILE:
6475 err = get_bitmap_file(mddev, argp);
6478 case RESTART_ARRAY_RW:
6479 err = restart_array(mddev);
6483 err = do_md_stop(mddev, 0, bdev);
6487 err = md_set_readonly(mddev, bdev);
6490 case HOT_REMOVE_DISK:
6491 err = hot_remove_disk(mddev, new_decode_dev(arg));
6495 /* We can support ADD_NEW_DISK on read-only arrays
6496 * on if we are re-adding a preexisting device.
6497 * So require mddev->pers and MD_DISK_SYNC.
6500 mdu_disk_info_t info;
6501 if (copy_from_user(&info, argp, sizeof(info)))
6503 else if (!(info.state & (1<<MD_DISK_SYNC)))
6504 /* Need to clear read-only for this */
6507 err = add_new_disk(mddev, &info);
6513 if (get_user(ro, (int __user *)(arg))) {
6519 /* if the bdev is going readonly the value of mddev->ro
6520 * does not matter, no writes are coming
6525 /* are we are already prepared for writes? */
6529 /* transitioning to readauto need only happen for
6530 * arrays that call md_write_start
6533 err = restart_array(mddev);
6536 set_disk_ro(mddev->gendisk, 0);
6543 * The remaining ioctls are changing the state of the
6544 * superblock, so we do not allow them on read-only arrays.
6545 * However non-MD ioctls (e.g. get-size) will still come through
6546 * here and hit the 'default' below, so only disallow
6547 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6549 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6550 if (mddev->ro == 2) {
6552 sysfs_notify_dirent_safe(mddev->sysfs_state);
6553 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6554 /* mddev_unlock will wake thread */
6555 /* If a device failed while we were read-only, we
6556 * need to make sure the metadata is updated now.
6558 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6559 mddev_unlock(mddev);
6560 wait_event(mddev->sb_wait,
6561 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6562 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6574 mdu_disk_info_t info;
6575 if (copy_from_user(&info, argp, sizeof(info)))
6578 err = add_new_disk(mddev, &info);
6583 err = hot_add_disk(mddev, new_decode_dev(arg));
6587 err = do_md_run(mddev);
6590 case SET_BITMAP_FILE:
6591 err = set_bitmap_file(mddev, (int)arg);
6601 if (mddev->hold_active == UNTIL_IOCTL &&
6603 mddev->hold_active = 0;
6604 mddev_unlock(mddev);
6613 #ifdef CONFIG_COMPAT
6614 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6615 unsigned int cmd, unsigned long arg)
6618 case HOT_REMOVE_DISK:
6620 case SET_DISK_FAULTY:
6621 case SET_BITMAP_FILE:
6622 /* These take in integer arg, do not convert */
6625 arg = (unsigned long)compat_ptr(arg);
6629 return md_ioctl(bdev, mode, cmd, arg);
6631 #endif /* CONFIG_COMPAT */
6633 static int md_open(struct block_device *bdev, fmode_t mode)
6636 * Succeed if we can lock the mddev, which confirms that
6637 * it isn't being stopped right now.
6639 struct mddev *mddev = mddev_find(bdev->bd_dev);
6645 if (mddev->gendisk != bdev->bd_disk) {
6646 /* we are racing with mddev_put which is discarding this
6650 /* Wait until bdev->bd_disk is definitely gone */
6651 flush_workqueue(md_misc_wq);
6652 /* Then retry the open from the top */
6653 return -ERESTARTSYS;
6655 BUG_ON(mddev != bdev->bd_disk->private_data);
6657 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6661 atomic_inc(&mddev->openers);
6662 mutex_unlock(&mddev->open_mutex);
6664 check_disk_change(bdev);
6669 static void md_release(struct gendisk *disk, fmode_t mode)
6671 struct mddev *mddev = disk->private_data;
6674 atomic_dec(&mddev->openers);
6678 static int md_media_changed(struct gendisk *disk)
6680 struct mddev *mddev = disk->private_data;
6682 return mddev->changed;
6685 static int md_revalidate(struct gendisk *disk)
6687 struct mddev *mddev = disk->private_data;
6692 static const struct block_device_operations md_fops =
6694 .owner = THIS_MODULE,
6696 .release = md_release,
6698 #ifdef CONFIG_COMPAT
6699 .compat_ioctl = md_compat_ioctl,
6701 .getgeo = md_getgeo,
6702 .media_changed = md_media_changed,
6703 .revalidate_disk= md_revalidate,
6706 static int md_thread(void * arg)
6708 struct md_thread *thread = arg;
6711 * md_thread is a 'system-thread', it's priority should be very
6712 * high. We avoid resource deadlocks individually in each
6713 * raid personality. (RAID5 does preallocation) We also use RR and
6714 * the very same RT priority as kswapd, thus we will never get
6715 * into a priority inversion deadlock.
6717 * we definitely have to have equal or higher priority than
6718 * bdflush, otherwise bdflush will deadlock if there are too
6719 * many dirty RAID5 blocks.
6722 allow_signal(SIGKILL);
6723 while (!kthread_should_stop()) {
6725 /* We need to wait INTERRUPTIBLE so that
6726 * we don't add to the load-average.
6727 * That means we need to be sure no signals are
6730 if (signal_pending(current))
6731 flush_signals(current);
6733 wait_event_interruptible_timeout
6735 test_bit(THREAD_WAKEUP, &thread->flags)
6736 || kthread_should_stop(),
6739 clear_bit(THREAD_WAKEUP, &thread->flags);
6740 if (!kthread_should_stop())
6741 thread->run(thread);
6747 void md_wakeup_thread(struct md_thread *thread)
6750 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6751 set_bit(THREAD_WAKEUP, &thread->flags);
6752 wake_up(&thread->wqueue);
6756 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6757 struct mddev *mddev, const char *name)
6759 struct md_thread *thread;
6761 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6765 init_waitqueue_head(&thread->wqueue);
6768 thread->mddev = mddev;
6769 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6770 thread->tsk = kthread_run(md_thread, thread,
6772 mdname(thread->mddev),
6774 if (IS_ERR(thread->tsk)) {
6781 void md_unregister_thread(struct md_thread **threadp)
6783 struct md_thread *thread = *threadp;
6786 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6787 /* Locking ensures that mddev_unlock does not wake_up a
6788 * non-existent thread
6790 spin_lock(&pers_lock);
6792 spin_unlock(&pers_lock);
6794 kthread_stop(thread->tsk);
6798 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6805 if (!rdev || test_bit(Faulty, &rdev->flags))
6808 if (!mddev->pers || !mddev->pers->error_handler)
6810 mddev->pers->error_handler(mddev,rdev);
6811 if (mddev->degraded)
6812 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6813 sysfs_notify_dirent_safe(rdev->sysfs_state);
6814 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6815 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6816 md_wakeup_thread(mddev->thread);
6817 if (mddev->event_work.func)
6818 queue_work(md_misc_wq, &mddev->event_work);
6819 md_new_event_inintr(mddev);
6822 /* seq_file implementation /proc/mdstat */
6824 static void status_unused(struct seq_file *seq)
6827 struct md_rdev *rdev;
6829 seq_printf(seq, "unused devices: ");
6831 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6832 char b[BDEVNAME_SIZE];
6834 seq_printf(seq, "%s ",
6835 bdevname(rdev->bdev,b));
6838 seq_printf(seq, "<none>");
6840 seq_printf(seq, "\n");
6844 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6846 sector_t max_sectors, resync, res;
6847 unsigned long dt, db;
6850 unsigned int per_milli;
6852 if (mddev->curr_resync <= 3)
6855 resync = mddev->curr_resync
6856 - atomic_read(&mddev->recovery_active);
6858 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6859 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6860 max_sectors = mddev->resync_max_sectors;
6862 max_sectors = mddev->dev_sectors;
6865 * Should not happen.
6871 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6872 * in a sector_t, and (max_sectors>>scale) will fit in a
6873 * u32, as those are the requirements for sector_div.
6874 * Thus 'scale' must be at least 10
6877 if (sizeof(sector_t) > sizeof(unsigned long)) {
6878 while ( max_sectors/2 > (1ULL<<(scale+32)))
6881 res = (resync>>scale)*1000;
6882 sector_div(res, (u32)((max_sectors>>scale)+1));
6886 int i, x = per_milli/50, y = 20-x;
6887 seq_printf(seq, "[");
6888 for (i = 0; i < x; i++)
6889 seq_printf(seq, "=");
6890 seq_printf(seq, ">");
6891 for (i = 0; i < y; i++)
6892 seq_printf(seq, ".");
6893 seq_printf(seq, "] ");
6895 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6896 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6898 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6900 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6901 "resync" : "recovery"))),
6902 per_milli/10, per_milli % 10,
6903 (unsigned long long) resync/2,
6904 (unsigned long long) max_sectors/2);
6907 * dt: time from mark until now
6908 * db: blocks written from mark until now
6909 * rt: remaining time
6911 * rt is a sector_t, so could be 32bit or 64bit.
6912 * So we divide before multiply in case it is 32bit and close
6914 * We scale the divisor (db) by 32 to avoid losing precision
6915 * near the end of resync when the number of remaining sectors
6917 * We then divide rt by 32 after multiplying by db to compensate.
6918 * The '+1' avoids division by zero if db is very small.
6920 dt = ((jiffies - mddev->resync_mark) / HZ);
6922 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6923 - mddev->resync_mark_cnt;
6925 rt = max_sectors - resync; /* number of remaining sectors */
6926 sector_div(rt, db/32+1);
6930 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6931 ((unsigned long)rt % 60)/6);
6933 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6936 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6938 struct list_head *tmp;
6940 struct mddev *mddev;
6948 spin_lock(&all_mddevs_lock);
6949 list_for_each(tmp,&all_mddevs)
6951 mddev = list_entry(tmp, struct mddev, all_mddevs);
6953 spin_unlock(&all_mddevs_lock);
6956 spin_unlock(&all_mddevs_lock);
6958 return (void*)2;/* tail */
6962 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6964 struct list_head *tmp;
6965 struct mddev *next_mddev, *mddev = v;
6971 spin_lock(&all_mddevs_lock);
6973 tmp = all_mddevs.next;
6975 tmp = mddev->all_mddevs.next;
6976 if (tmp != &all_mddevs)
6977 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6979 next_mddev = (void*)2;
6982 spin_unlock(&all_mddevs_lock);
6990 static void md_seq_stop(struct seq_file *seq, void *v)
6992 struct mddev *mddev = v;
6994 if (mddev && v != (void*)1 && v != (void*)2)
6998 static int md_seq_show(struct seq_file *seq, void *v)
7000 struct mddev *mddev = v;
7002 struct md_rdev *rdev;
7004 if (v == (void*)1) {
7005 struct md_personality *pers;
7006 seq_printf(seq, "Personalities : ");
7007 spin_lock(&pers_lock);
7008 list_for_each_entry(pers, &pers_list, list)
7009 seq_printf(seq, "[%s] ", pers->name);
7011 spin_unlock(&pers_lock);
7012 seq_printf(seq, "\n");
7013 seq->poll_event = atomic_read(&md_event_count);
7016 if (v == (void*)2) {
7021 if (mddev_lock(mddev) < 0)
7024 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7025 seq_printf(seq, "%s : %sactive", mdname(mddev),
7026 mddev->pers ? "" : "in");
7029 seq_printf(seq, " (read-only)");
7031 seq_printf(seq, " (auto-read-only)");
7032 seq_printf(seq, " %s", mddev->pers->name);
7036 rdev_for_each(rdev, mddev) {
7037 char b[BDEVNAME_SIZE];
7038 seq_printf(seq, " %s[%d]",
7039 bdevname(rdev->bdev,b), rdev->desc_nr);
7040 if (test_bit(WriteMostly, &rdev->flags))
7041 seq_printf(seq, "(W)");
7042 if (test_bit(Faulty, &rdev->flags)) {
7043 seq_printf(seq, "(F)");
7046 if (rdev->raid_disk < 0)
7047 seq_printf(seq, "(S)"); /* spare */
7048 if (test_bit(Replacement, &rdev->flags))
7049 seq_printf(seq, "(R)");
7050 sectors += rdev->sectors;
7053 if (!list_empty(&mddev->disks)) {
7055 seq_printf(seq, "\n %llu blocks",
7056 (unsigned long long)
7057 mddev->array_sectors / 2);
7059 seq_printf(seq, "\n %llu blocks",
7060 (unsigned long long)sectors / 2);
7062 if (mddev->persistent) {
7063 if (mddev->major_version != 0 ||
7064 mddev->minor_version != 90) {
7065 seq_printf(seq," super %d.%d",
7066 mddev->major_version,
7067 mddev->minor_version);
7069 } else if (mddev->external)
7070 seq_printf(seq, " super external:%s",
7071 mddev->metadata_type);
7073 seq_printf(seq, " super non-persistent");
7076 mddev->pers->status(seq, mddev);
7077 seq_printf(seq, "\n ");
7078 if (mddev->pers->sync_request) {
7079 if (mddev->curr_resync > 2) {
7080 status_resync(seq, mddev);
7081 seq_printf(seq, "\n ");
7082 } else if (mddev->curr_resync >= 1)
7083 seq_printf(seq, "\tresync=DELAYED\n ");
7084 else if (mddev->recovery_cp < MaxSector)
7085 seq_printf(seq, "\tresync=PENDING\n ");
7088 seq_printf(seq, "\n ");
7090 bitmap_status(seq, mddev->bitmap);
7092 seq_printf(seq, "\n");
7094 mddev_unlock(mddev);
7099 static const struct seq_operations md_seq_ops = {
7100 .start = md_seq_start,
7101 .next = md_seq_next,
7102 .stop = md_seq_stop,
7103 .show = md_seq_show,
7106 static int md_seq_open(struct inode *inode, struct file *file)
7108 struct seq_file *seq;
7111 error = seq_open(file, &md_seq_ops);
7115 seq = file->private_data;
7116 seq->poll_event = atomic_read(&md_event_count);
7120 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7122 struct seq_file *seq = filp->private_data;
7125 poll_wait(filp, &md_event_waiters, wait);
7127 /* always allow read */
7128 mask = POLLIN | POLLRDNORM;
7130 if (seq->poll_event != atomic_read(&md_event_count))
7131 mask |= POLLERR | POLLPRI;
7135 static const struct file_operations md_seq_fops = {
7136 .owner = THIS_MODULE,
7137 .open = md_seq_open,
7139 .llseek = seq_lseek,
7140 .release = seq_release_private,
7141 .poll = mdstat_poll,
7144 int register_md_personality(struct md_personality *p)
7146 spin_lock(&pers_lock);
7147 list_add_tail(&p->list, &pers_list);
7148 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7149 spin_unlock(&pers_lock);
7153 int unregister_md_personality(struct md_personality *p)
7155 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7156 spin_lock(&pers_lock);
7157 list_del_init(&p->list);
7158 spin_unlock(&pers_lock);
7162 static int is_mddev_idle(struct mddev *mddev, int init)
7164 struct md_rdev * rdev;
7170 rdev_for_each_rcu(rdev, mddev) {
7171 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7172 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7173 (int)part_stat_read(&disk->part0, sectors[1]) -
7174 atomic_read(&disk->sync_io);
7175 /* sync IO will cause sync_io to increase before the disk_stats
7176 * as sync_io is counted when a request starts, and
7177 * disk_stats is counted when it completes.
7178 * So resync activity will cause curr_events to be smaller than
7179 * when there was no such activity.
7180 * non-sync IO will cause disk_stat to increase without
7181 * increasing sync_io so curr_events will (eventually)
7182 * be larger than it was before. Once it becomes
7183 * substantially larger, the test below will cause
7184 * the array to appear non-idle, and resync will slow
7186 * If there is a lot of outstanding resync activity when
7187 * we set last_event to curr_events, then all that activity
7188 * completing might cause the array to appear non-idle
7189 * and resync will be slowed down even though there might
7190 * not have been non-resync activity. This will only
7191 * happen once though. 'last_events' will soon reflect
7192 * the state where there is little or no outstanding
7193 * resync requests, and further resync activity will
7194 * always make curr_events less than last_events.
7197 if (init || curr_events - rdev->last_events > 64) {
7198 rdev->last_events = curr_events;
7206 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7208 /* another "blocks" (512byte) blocks have been synced */
7209 atomic_sub(blocks, &mddev->recovery_active);
7210 wake_up(&mddev->recovery_wait);
7212 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7213 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7214 md_wakeup_thread(mddev->thread);
7215 // stop recovery, signal do_sync ....
7220 /* md_write_start(mddev, bi)
7221 * If we need to update some array metadata (e.g. 'active' flag
7222 * in superblock) before writing, schedule a superblock update
7223 * and wait for it to complete.
7225 void md_write_start(struct mddev *mddev, struct bio *bi)
7228 if (bio_data_dir(bi) != WRITE)
7231 BUG_ON(mddev->ro == 1);
7232 if (mddev->ro == 2) {
7233 /* need to switch to read/write */
7235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7236 md_wakeup_thread(mddev->thread);
7237 md_wakeup_thread(mddev->sync_thread);
7240 atomic_inc(&mddev->writes_pending);
7241 if (mddev->safemode == 1)
7242 mddev->safemode = 0;
7243 if (mddev->in_sync) {
7244 spin_lock_irq(&mddev->write_lock);
7245 if (mddev->in_sync) {
7247 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7248 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7249 md_wakeup_thread(mddev->thread);
7252 spin_unlock_irq(&mddev->write_lock);
7255 sysfs_notify_dirent_safe(mddev->sysfs_state);
7256 wait_event(mddev->sb_wait,
7257 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7260 void md_write_end(struct mddev *mddev)
7262 if (atomic_dec_and_test(&mddev->writes_pending)) {
7263 if (mddev->safemode == 2)
7264 md_wakeup_thread(mddev->thread);
7265 else if (mddev->safemode_delay)
7266 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7270 /* md_allow_write(mddev)
7271 * Calling this ensures that the array is marked 'active' so that writes
7272 * may proceed without blocking. It is important to call this before
7273 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7274 * Must be called with mddev_lock held.
7276 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7277 * is dropped, so return -EAGAIN after notifying userspace.
7279 int md_allow_write(struct mddev *mddev)
7285 if (!mddev->pers->sync_request)
7288 spin_lock_irq(&mddev->write_lock);
7289 if (mddev->in_sync) {
7291 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7292 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7293 if (mddev->safemode_delay &&
7294 mddev->safemode == 0)
7295 mddev->safemode = 1;
7296 spin_unlock_irq(&mddev->write_lock);
7297 md_update_sb(mddev, 0);
7298 sysfs_notify_dirent_safe(mddev->sysfs_state);
7300 spin_unlock_irq(&mddev->write_lock);
7302 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7307 EXPORT_SYMBOL_GPL(md_allow_write);
7309 #define SYNC_MARKS 10
7310 #define SYNC_MARK_STEP (3*HZ)
7311 #define UPDATE_FREQUENCY (5*60*HZ)
7312 void md_do_sync(struct md_thread *thread)
7314 struct mddev *mddev = thread->mddev;
7315 struct mddev *mddev2;
7316 unsigned int currspeed = 0,
7318 sector_t max_sectors,j, io_sectors;
7319 unsigned long mark[SYNC_MARKS];
7320 unsigned long update_time;
7321 sector_t mark_cnt[SYNC_MARKS];
7323 struct list_head *tmp;
7324 sector_t last_check;
7326 struct md_rdev *rdev;
7328 struct blk_plug plug;
7330 /* just incase thread restarts... */
7331 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7333 if (mddev->ro) /* never try to sync a read-only array */
7336 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7337 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7338 desc = "data-check";
7339 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7340 desc = "requested-resync";
7343 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7348 /* we overload curr_resync somewhat here.
7349 * 0 == not engaged in resync at all
7350 * 2 == checking that there is no conflict with another sync
7351 * 1 == like 2, but have yielded to allow conflicting resync to
7353 * other == active in resync - this many blocks
7355 * Before starting a resync we must have set curr_resync to
7356 * 2, and then checked that every "conflicting" array has curr_resync
7357 * less than ours. When we find one that is the same or higher
7358 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7359 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7360 * This will mean we have to start checking from the beginning again.
7365 mddev->curr_resync = 2;
7368 if (kthread_should_stop())
7369 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7371 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7373 for_each_mddev(mddev2, tmp) {
7374 if (mddev2 == mddev)
7376 if (!mddev->parallel_resync
7377 && mddev2->curr_resync
7378 && match_mddev_units(mddev, mddev2)) {
7380 if (mddev < mddev2 && mddev->curr_resync == 2) {
7381 /* arbitrarily yield */
7382 mddev->curr_resync = 1;
7383 wake_up(&resync_wait);
7385 if (mddev > mddev2 && mddev->curr_resync == 1)
7386 /* no need to wait here, we can wait the next
7387 * time 'round when curr_resync == 2
7390 /* We need to wait 'interruptible' so as not to
7391 * contribute to the load average, and not to
7392 * be caught by 'softlockup'
7394 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7395 if (!kthread_should_stop() &&
7396 mddev2->curr_resync >= mddev->curr_resync) {
7397 printk(KERN_INFO "md: delaying %s of %s"
7398 " until %s has finished (they"
7399 " share one or more physical units)\n",
7400 desc, mdname(mddev), mdname(mddev2));
7402 if (signal_pending(current))
7403 flush_signals(current);
7405 finish_wait(&resync_wait, &wq);
7408 finish_wait(&resync_wait, &wq);
7411 } while (mddev->curr_resync < 2);
7414 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7415 /* resync follows the size requested by the personality,
7416 * which defaults to physical size, but can be virtual size
7418 max_sectors = mddev->resync_max_sectors;
7419 atomic64_set(&mddev->resync_mismatches, 0);
7420 /* we don't use the checkpoint if there's a bitmap */
7421 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7422 j = mddev->resync_min;
7423 else if (!mddev->bitmap)
7424 j = mddev->recovery_cp;
7426 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7427 max_sectors = mddev->resync_max_sectors;
7429 /* recovery follows the physical size of devices */
7430 max_sectors = mddev->dev_sectors;
7433 rdev_for_each_rcu(rdev, mddev)
7434 if (rdev->raid_disk >= 0 &&
7435 !test_bit(Faulty, &rdev->flags) &&
7436 !test_bit(In_sync, &rdev->flags) &&
7437 rdev->recovery_offset < j)
7438 j = rdev->recovery_offset;
7442 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7443 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7444 " %d KB/sec/disk.\n", speed_min(mddev));
7445 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7446 "(but not more than %d KB/sec) for %s.\n",
7447 speed_max(mddev), desc);
7449 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7452 for (m = 0; m < SYNC_MARKS; m++) {
7454 mark_cnt[m] = io_sectors;
7457 mddev->resync_mark = mark[last_mark];
7458 mddev->resync_mark_cnt = mark_cnt[last_mark];
7461 * Tune reconstruction:
7463 window = 32*(PAGE_SIZE/512);
7464 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7465 window/2, (unsigned long long)max_sectors/2);
7467 atomic_set(&mddev->recovery_active, 0);
7472 "md: resuming %s of %s from checkpoint.\n",
7473 desc, mdname(mddev));
7474 mddev->curr_resync = j;
7476 mddev->curr_resync = 3; /* no longer delayed */
7477 mddev->curr_resync_completed = j;
7478 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7479 md_new_event(mddev);
7480 update_time = jiffies;
7482 blk_start_plug(&plug);
7483 while (j < max_sectors) {
7488 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7489 ((mddev->curr_resync > mddev->curr_resync_completed &&
7490 (mddev->curr_resync - mddev->curr_resync_completed)
7491 > (max_sectors >> 4)) ||
7492 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7493 (j - mddev->curr_resync_completed)*2
7494 >= mddev->resync_max - mddev->curr_resync_completed
7496 /* time to update curr_resync_completed */
7497 wait_event(mddev->recovery_wait,
7498 atomic_read(&mddev->recovery_active) == 0);
7499 mddev->curr_resync_completed = j;
7500 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7501 j > mddev->recovery_cp)
7502 mddev->recovery_cp = j;
7503 update_time = jiffies;
7504 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7505 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7508 while (j >= mddev->resync_max && !kthread_should_stop()) {
7509 /* As this condition is controlled by user-space,
7510 * we can block indefinitely, so use '_interruptible'
7511 * to avoid triggering warnings.
7513 flush_signals(current); /* just in case */
7514 wait_event_interruptible(mddev->recovery_wait,
7515 mddev->resync_max > j
7516 || kthread_should_stop());
7519 if (kthread_should_stop())
7522 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7523 currspeed < speed_min(mddev));
7525 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7529 if (!skipped) { /* actual IO requested */
7530 io_sectors += sectors;
7531 atomic_add(sectors, &mddev->recovery_active);
7534 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7539 mddev->curr_resync = j;
7540 mddev->curr_mark_cnt = io_sectors;
7541 if (last_check == 0)
7542 /* this is the earliest that rebuild will be
7543 * visible in /proc/mdstat
7545 md_new_event(mddev);
7547 if (last_check + window > io_sectors || j == max_sectors)
7550 last_check = io_sectors;
7552 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7554 int next = (last_mark+1) % SYNC_MARKS;
7556 mddev->resync_mark = mark[next];
7557 mddev->resync_mark_cnt = mark_cnt[next];
7558 mark[next] = jiffies;
7559 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7564 if (kthread_should_stop())
7569 * this loop exits only if either when we are slower than
7570 * the 'hard' speed limit, or the system was IO-idle for
7572 * the system might be non-idle CPU-wise, but we only care
7573 * about not overloading the IO subsystem. (things like an
7574 * e2fsck being done on the RAID array should execute fast)
7578 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7579 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7581 if (currspeed > speed_min(mddev)) {
7582 if ((currspeed > speed_max(mddev)) ||
7583 !is_mddev_idle(mddev, 0)) {
7589 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7591 * this also signals 'finished resyncing' to md_stop
7594 blk_finish_plug(&plug);
7595 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7597 /* tell personality that we are finished */
7598 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7600 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7601 mddev->curr_resync > 2) {
7602 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7603 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7604 if (mddev->curr_resync >= mddev->recovery_cp) {
7606 "md: checkpointing %s of %s.\n",
7607 desc, mdname(mddev));
7608 if (test_bit(MD_RECOVERY_ERROR,
7610 mddev->recovery_cp =
7611 mddev->curr_resync_completed;
7613 mddev->recovery_cp =
7617 mddev->recovery_cp = MaxSector;
7619 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7620 mddev->curr_resync = MaxSector;
7622 rdev_for_each_rcu(rdev, mddev)
7623 if (rdev->raid_disk >= 0 &&
7624 mddev->delta_disks >= 0 &&
7625 !test_bit(Faulty, &rdev->flags) &&
7626 !test_bit(In_sync, &rdev->flags) &&
7627 rdev->recovery_offset < mddev->curr_resync)
7628 rdev->recovery_offset = mddev->curr_resync;
7633 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7635 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7636 /* We completed so min/max setting can be forgotten if used. */
7637 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7638 mddev->resync_min = 0;
7639 mddev->resync_max = MaxSector;
7640 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7641 mddev->resync_min = mddev->curr_resync_completed;
7642 mddev->curr_resync = 0;
7643 wake_up(&resync_wait);
7644 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7645 md_wakeup_thread(mddev->thread);
7650 * got a signal, exit.
7653 "md: md_do_sync() got signal ... exiting\n");
7654 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7658 EXPORT_SYMBOL_GPL(md_do_sync);
7660 static int remove_and_add_spares(struct mddev *mddev,
7661 struct md_rdev *this)
7663 struct md_rdev *rdev;
7667 rdev_for_each(rdev, mddev)
7668 if ((this == NULL || rdev == this) &&
7669 rdev->raid_disk >= 0 &&
7670 !test_bit(Blocked, &rdev->flags) &&
7671 (test_bit(Faulty, &rdev->flags) ||
7672 ! test_bit(In_sync, &rdev->flags)) &&
7673 atomic_read(&rdev->nr_pending)==0) {
7674 if (mddev->pers->hot_remove_disk(
7675 mddev, rdev) == 0) {
7676 sysfs_unlink_rdev(mddev, rdev);
7677 rdev->raid_disk = -1;
7681 if (removed && mddev->kobj.sd)
7682 sysfs_notify(&mddev->kobj, NULL, "degraded");
7687 rdev_for_each(rdev, mddev) {
7688 if (rdev->raid_disk >= 0 &&
7689 !test_bit(In_sync, &rdev->flags) &&
7690 !test_bit(Faulty, &rdev->flags))
7692 if (rdev->raid_disk >= 0)
7694 if (test_bit(Faulty, &rdev->flags))
7697 rdev->saved_raid_disk < 0)
7700 rdev->recovery_offset = 0;
7702 hot_add_disk(mddev, rdev) == 0) {
7703 if (sysfs_link_rdev(mddev, rdev))
7704 /* failure here is OK */;
7706 md_new_event(mddev);
7707 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7712 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7717 * This routine is regularly called by all per-raid-array threads to
7718 * deal with generic issues like resync and super-block update.
7719 * Raid personalities that don't have a thread (linear/raid0) do not
7720 * need this as they never do any recovery or update the superblock.
7722 * It does not do any resync itself, but rather "forks" off other threads
7723 * to do that as needed.
7724 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7725 * "->recovery" and create a thread at ->sync_thread.
7726 * When the thread finishes it sets MD_RECOVERY_DONE
7727 * and wakeups up this thread which will reap the thread and finish up.
7728 * This thread also removes any faulty devices (with nr_pending == 0).
7730 * The overall approach is:
7731 * 1/ if the superblock needs updating, update it.
7732 * 2/ If a recovery thread is running, don't do anything else.
7733 * 3/ If recovery has finished, clean up, possibly marking spares active.
7734 * 4/ If there are any faulty devices, remove them.
7735 * 5/ If array is degraded, try to add spares devices
7736 * 6/ If array has spares or is not in-sync, start a resync thread.
7738 void md_check_recovery(struct mddev *mddev)
7740 if (mddev->suspended)
7744 bitmap_daemon_work(mddev);
7746 if (signal_pending(current)) {
7747 if (mddev->pers->sync_request && !mddev->external) {
7748 printk(KERN_INFO "md: %s in immediate safe mode\n",
7750 mddev->safemode = 2;
7752 flush_signals(current);
7755 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7758 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7759 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7760 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7761 (mddev->external == 0 && mddev->safemode == 1) ||
7762 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7763 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7767 if (mddev_trylock(mddev)) {
7771 /* On a read-only array we can:
7772 * - remove failed devices
7773 * - add already-in_sync devices if the array itself
7775 * As we only add devices that are already in-sync,
7776 * we can activate the spares immediately.
7778 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7779 remove_and_add_spares(mddev, NULL);
7780 mddev->pers->spare_active(mddev);
7784 if (!mddev->external) {
7786 spin_lock_irq(&mddev->write_lock);
7787 if (mddev->safemode &&
7788 !atomic_read(&mddev->writes_pending) &&
7790 mddev->recovery_cp == MaxSector) {
7793 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7795 if (mddev->safemode == 1)
7796 mddev->safemode = 0;
7797 spin_unlock_irq(&mddev->write_lock);
7799 sysfs_notify_dirent_safe(mddev->sysfs_state);
7803 md_update_sb(mddev, 0);
7805 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7806 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7807 /* resync/recovery still happening */
7808 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7811 if (mddev->sync_thread) {
7812 md_reap_sync_thread(mddev);
7815 /* Set RUNNING before clearing NEEDED to avoid
7816 * any transients in the value of "sync_action".
7818 mddev->curr_resync_completed = 0;
7819 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7820 /* Clear some bits that don't mean anything, but
7823 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7824 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7826 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7827 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7829 /* no recovery is running.
7830 * remove any failed drives, then
7831 * add spares if possible.
7832 * Spares are also removed and re-added, to allow
7833 * the personality to fail the re-add.
7836 if (mddev->reshape_position != MaxSector) {
7837 if (mddev->pers->check_reshape == NULL ||
7838 mddev->pers->check_reshape(mddev) != 0)
7839 /* Cannot proceed */
7841 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7842 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7843 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
7844 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7845 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7846 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7847 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7848 } else if (mddev->recovery_cp < MaxSector) {
7849 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7850 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7851 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7852 /* nothing to be done ... */
7855 if (mddev->pers->sync_request) {
7857 /* We are adding a device or devices to an array
7858 * which has the bitmap stored on all devices.
7859 * So make sure all bitmap pages get written
7861 bitmap_write_all(mddev->bitmap);
7863 mddev->sync_thread = md_register_thread(md_do_sync,
7866 if (!mddev->sync_thread) {
7867 printk(KERN_ERR "%s: could not start resync"
7870 /* leave the spares where they are, it shouldn't hurt */
7871 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7872 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7873 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7874 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7875 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7877 md_wakeup_thread(mddev->sync_thread);
7878 sysfs_notify_dirent_safe(mddev->sysfs_action);
7879 md_new_event(mddev);
7882 if (!mddev->sync_thread) {
7883 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7884 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7886 if (mddev->sysfs_action)
7887 sysfs_notify_dirent_safe(mddev->sysfs_action);
7889 mddev_unlock(mddev);
7893 void md_reap_sync_thread(struct mddev *mddev)
7895 struct md_rdev *rdev;
7897 /* resync has finished, collect result */
7898 md_unregister_thread(&mddev->sync_thread);
7899 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7900 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7902 /* activate any spares */
7903 if (mddev->pers->spare_active(mddev)) {
7904 sysfs_notify(&mddev->kobj, NULL,
7906 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7909 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7910 mddev->pers->finish_reshape)
7911 mddev->pers->finish_reshape(mddev);
7913 /* If array is no-longer degraded, then any saved_raid_disk
7914 * information must be scrapped. Also if any device is now
7915 * In_sync we must scrape the saved_raid_disk for that device
7916 * do the superblock for an incrementally recovered device
7919 rdev_for_each(rdev, mddev)
7920 if (!mddev->degraded ||
7921 test_bit(In_sync, &rdev->flags))
7922 rdev->saved_raid_disk = -1;
7924 md_update_sb(mddev, 1);
7925 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7926 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7927 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7928 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7929 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7930 /* flag recovery needed just to double check */
7931 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7932 sysfs_notify_dirent_safe(mddev->sysfs_action);
7933 md_new_event(mddev);
7934 if (mddev->event_work.func)
7935 queue_work(md_misc_wq, &mddev->event_work);
7938 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7940 sysfs_notify_dirent_safe(rdev->sysfs_state);
7941 wait_event_timeout(rdev->blocked_wait,
7942 !test_bit(Blocked, &rdev->flags) &&
7943 !test_bit(BlockedBadBlocks, &rdev->flags),
7944 msecs_to_jiffies(5000));
7945 rdev_dec_pending(rdev, mddev);
7947 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7949 void md_finish_reshape(struct mddev *mddev)
7951 /* called be personality module when reshape completes. */
7952 struct md_rdev *rdev;
7954 rdev_for_each(rdev, mddev) {
7955 if (rdev->data_offset > rdev->new_data_offset)
7956 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7958 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7959 rdev->data_offset = rdev->new_data_offset;
7962 EXPORT_SYMBOL(md_finish_reshape);
7964 /* Bad block management.
7965 * We can record which blocks on each device are 'bad' and so just
7966 * fail those blocks, or that stripe, rather than the whole device.
7967 * Entries in the bad-block table are 64bits wide. This comprises:
7968 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7969 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7970 * A 'shift' can be set so that larger blocks are tracked and
7971 * consequently larger devices can be covered.
7972 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7974 * Locking of the bad-block table uses a seqlock so md_is_badblock
7975 * might need to retry if it is very unlucky.
7976 * We will sometimes want to check for bad blocks in a bi_end_io function,
7977 * so we use the write_seqlock_irq variant.
7979 * When looking for a bad block we specify a range and want to
7980 * know if any block in the range is bad. So we binary-search
7981 * to the last range that starts at-or-before the given endpoint,
7982 * (or "before the sector after the target range")
7983 * then see if it ends after the given start.
7985 * 0 if there are no known bad blocks in the range
7986 * 1 if there are known bad block which are all acknowledged
7987 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7988 * plus the start/length of the first bad section we overlap.
7990 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7991 sector_t *first_bad, int *bad_sectors)
7997 sector_t target = s + sectors;
8000 if (bb->shift > 0) {
8001 /* round the start down, and the end up */
8003 target += (1<<bb->shift) - 1;
8004 target >>= bb->shift;
8005 sectors = target - s;
8007 /* 'target' is now the first block after the bad range */
8010 seq = read_seqbegin(&bb->lock);
8015 /* Binary search between lo and hi for 'target'
8016 * i.e. for the last range that starts before 'target'
8018 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8019 * are known not to be the last range before target.
8020 * VARIANT: hi-lo is the number of possible
8021 * ranges, and decreases until it reaches 1
8023 while (hi - lo > 1) {
8024 int mid = (lo + hi) / 2;
8025 sector_t a = BB_OFFSET(p[mid]);
8027 /* This could still be the one, earlier ranges
8031 /* This and later ranges are definitely out. */
8034 /* 'lo' might be the last that started before target, but 'hi' isn't */
8036 /* need to check all range that end after 's' to see if
8037 * any are unacknowledged.
8040 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8041 if (BB_OFFSET(p[lo]) < target) {
8042 /* starts before the end, and finishes after
8043 * the start, so they must overlap
8045 if (rv != -1 && BB_ACK(p[lo]))
8049 *first_bad = BB_OFFSET(p[lo]);
8050 *bad_sectors = BB_LEN(p[lo]);
8056 if (read_seqretry(&bb->lock, seq))
8061 EXPORT_SYMBOL_GPL(md_is_badblock);
8064 * Add a range of bad blocks to the table.
8065 * This might extend the table, or might contract it
8066 * if two adjacent ranges can be merged.
8067 * We binary-search to find the 'insertion' point, then
8068 * decide how best to handle it.
8070 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8076 unsigned long flags;
8079 /* badblocks are disabled */
8083 /* round the start down, and the end up */
8084 sector_t next = s + sectors;
8086 next += (1<<bb->shift) - 1;
8091 write_seqlock_irqsave(&bb->lock, flags);
8096 /* Find the last range that starts at-or-before 's' */
8097 while (hi - lo > 1) {
8098 int mid = (lo + hi) / 2;
8099 sector_t a = BB_OFFSET(p[mid]);
8105 if (hi > lo && BB_OFFSET(p[lo]) > s)
8109 /* we found a range that might merge with the start
8112 sector_t a = BB_OFFSET(p[lo]);
8113 sector_t e = a + BB_LEN(p[lo]);
8114 int ack = BB_ACK(p[lo]);
8116 /* Yes, we can merge with a previous range */
8117 if (s == a && s + sectors >= e)
8118 /* new range covers old */
8121 ack = ack && acknowledged;
8123 if (e < s + sectors)
8125 if (e - a <= BB_MAX_LEN) {
8126 p[lo] = BB_MAKE(a, e-a, ack);
8129 /* does not all fit in one range,
8130 * make p[lo] maximal
8132 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8133 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8139 if (sectors && hi < bb->count) {
8140 /* 'hi' points to the first range that starts after 's'.
8141 * Maybe we can merge with the start of that range */
8142 sector_t a = BB_OFFSET(p[hi]);
8143 sector_t e = a + BB_LEN(p[hi]);
8144 int ack = BB_ACK(p[hi]);
8145 if (a <= s + sectors) {
8146 /* merging is possible */
8147 if (e <= s + sectors) {
8152 ack = ack && acknowledged;
8155 if (e - a <= BB_MAX_LEN) {
8156 p[hi] = BB_MAKE(a, e-a, ack);
8159 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8167 if (sectors == 0 && hi < bb->count) {
8168 /* we might be able to combine lo and hi */
8169 /* Note: 's' is at the end of 'lo' */
8170 sector_t a = BB_OFFSET(p[hi]);
8171 int lolen = BB_LEN(p[lo]);
8172 int hilen = BB_LEN(p[hi]);
8173 int newlen = lolen + hilen - (s - a);
8174 if (s >= a && newlen < BB_MAX_LEN) {
8175 /* yes, we can combine them */
8176 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8177 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8178 memmove(p + hi, p + hi + 1,
8179 (bb->count - hi - 1) * 8);
8184 /* didn't merge (it all).
8185 * Need to add a range just before 'hi' */
8186 if (bb->count >= MD_MAX_BADBLOCKS) {
8187 /* No room for more */
8191 int this_sectors = sectors;
8192 memmove(p + hi + 1, p + hi,
8193 (bb->count - hi) * 8);
8196 if (this_sectors > BB_MAX_LEN)
8197 this_sectors = BB_MAX_LEN;
8198 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8199 sectors -= this_sectors;
8206 bb->unacked_exist = 1;
8207 write_sequnlock_irqrestore(&bb->lock, flags);
8212 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8217 s += rdev->new_data_offset;
8219 s += rdev->data_offset;
8220 rv = md_set_badblocks(&rdev->badblocks,
8223 /* Make sure they get written out promptly */
8224 sysfs_notify_dirent_safe(rdev->sysfs_state);
8225 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8226 md_wakeup_thread(rdev->mddev->thread);
8230 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8233 * Remove a range of bad blocks from the table.
8234 * This may involve extending the table if we spilt a region,
8235 * but it must not fail. So if the table becomes full, we just
8236 * drop the remove request.
8238 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8242 sector_t target = s + sectors;
8245 if (bb->shift > 0) {
8246 /* When clearing we round the start up and the end down.
8247 * This should not matter as the shift should align with
8248 * the block size and no rounding should ever be needed.
8249 * However it is better the think a block is bad when it
8250 * isn't than to think a block is not bad when it is.
8252 s += (1<<bb->shift) - 1;
8254 target >>= bb->shift;
8255 sectors = target - s;
8258 write_seqlock_irq(&bb->lock);
8263 /* Find the last range that starts before 'target' */
8264 while (hi - lo > 1) {
8265 int mid = (lo + hi) / 2;
8266 sector_t a = BB_OFFSET(p[mid]);
8273 /* p[lo] is the last range that could overlap the
8274 * current range. Earlier ranges could also overlap,
8275 * but only this one can overlap the end of the range.
8277 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8278 /* Partial overlap, leave the tail of this range */
8279 int ack = BB_ACK(p[lo]);
8280 sector_t a = BB_OFFSET(p[lo]);
8281 sector_t end = a + BB_LEN(p[lo]);
8284 /* we need to split this range */
8285 if (bb->count >= MD_MAX_BADBLOCKS) {
8289 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8291 p[lo] = BB_MAKE(a, s-a, ack);
8294 p[lo] = BB_MAKE(target, end - target, ack);
8295 /* there is no longer an overlap */
8300 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8301 /* This range does overlap */
8302 if (BB_OFFSET(p[lo]) < s) {
8303 /* Keep the early parts of this range. */
8304 int ack = BB_ACK(p[lo]);
8305 sector_t start = BB_OFFSET(p[lo]);
8306 p[lo] = BB_MAKE(start, s - start, ack);
8307 /* now low doesn't overlap, so.. */
8312 /* 'lo' is strictly before, 'hi' is strictly after,
8313 * anything between needs to be discarded
8316 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8317 bb->count -= (hi - lo - 1);
8323 write_sequnlock_irq(&bb->lock);
8327 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8331 s += rdev->new_data_offset;
8333 s += rdev->data_offset;
8334 return md_clear_badblocks(&rdev->badblocks,
8337 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8340 * Acknowledge all bad blocks in a list.
8341 * This only succeeds if ->changed is clear. It is used by
8342 * in-kernel metadata updates
8344 void md_ack_all_badblocks(struct badblocks *bb)
8346 if (bb->page == NULL || bb->changed)
8347 /* no point even trying */
8349 write_seqlock_irq(&bb->lock);
8351 if (bb->changed == 0 && bb->unacked_exist) {
8354 for (i = 0; i < bb->count ; i++) {
8355 if (!BB_ACK(p[i])) {
8356 sector_t start = BB_OFFSET(p[i]);
8357 int len = BB_LEN(p[i]);
8358 p[i] = BB_MAKE(start, len, 1);
8361 bb->unacked_exist = 0;
8363 write_sequnlock_irq(&bb->lock);
8365 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8367 /* sysfs access to bad-blocks list.
8368 * We present two files.
8369 * 'bad-blocks' lists sector numbers and lengths of ranges that
8370 * are recorded as bad. The list is truncated to fit within
8371 * the one-page limit of sysfs.
8372 * Writing "sector length" to this file adds an acknowledged
8374 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8375 * been acknowledged. Writing to this file adds bad blocks
8376 * without acknowledging them. This is largely for testing.
8380 badblocks_show(struct badblocks *bb, char *page, int unack)
8391 seq = read_seqbegin(&bb->lock);
8396 while (len < PAGE_SIZE && i < bb->count) {
8397 sector_t s = BB_OFFSET(p[i]);
8398 unsigned int length = BB_LEN(p[i]);
8399 int ack = BB_ACK(p[i]);
8405 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8406 (unsigned long long)s << bb->shift,
8407 length << bb->shift);
8409 if (unack && len == 0)
8410 bb->unacked_exist = 0;
8412 if (read_seqretry(&bb->lock, seq))
8421 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8423 unsigned long long sector;
8427 /* Allow clearing via sysfs *only* for testing/debugging.
8428 * Normally only a successful write may clear a badblock
8431 if (page[0] == '-') {
8435 #endif /* DO_DEBUG */
8437 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8439 if (newline != '\n')
8451 md_clear_badblocks(bb, sector, length);
8454 #endif /* DO_DEBUG */
8455 if (md_set_badblocks(bb, sector, length, !unack))
8461 static int md_notify_reboot(struct notifier_block *this,
8462 unsigned long code, void *x)
8464 struct list_head *tmp;
8465 struct mddev *mddev;
8468 for_each_mddev(mddev, tmp) {
8469 if (mddev_trylock(mddev)) {
8471 __md_stop_writes(mddev);
8472 mddev->safemode = 2;
8473 mddev_unlock(mddev);
8478 * certain more exotic SCSI devices are known to be
8479 * volatile wrt too early system reboots. While the
8480 * right place to handle this issue is the given
8481 * driver, we do want to have a safe RAID driver ...
8489 static struct notifier_block md_notifier = {
8490 .notifier_call = md_notify_reboot,
8492 .priority = INT_MAX, /* before any real devices */
8495 static void md_geninit(void)
8497 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8499 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8502 static int __init md_init(void)
8506 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8510 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8514 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8517 if ((ret = register_blkdev(0, "mdp")) < 0)
8521 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8522 md_probe, NULL, NULL);
8523 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8524 md_probe, NULL, NULL);
8526 register_reboot_notifier(&md_notifier);
8527 raid_table_header = register_sysctl_table(raid_root_table);
8533 unregister_blkdev(MD_MAJOR, "md");
8535 destroy_workqueue(md_misc_wq);
8537 destroy_workqueue(md_wq);
8545 * Searches all registered partitions for autorun RAID arrays
8549 static LIST_HEAD(all_detected_devices);
8550 struct detected_devices_node {
8551 struct list_head list;
8555 void md_autodetect_dev(dev_t dev)
8557 struct detected_devices_node *node_detected_dev;
8559 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8560 if (node_detected_dev) {
8561 node_detected_dev->dev = dev;
8562 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8564 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8565 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8570 static void autostart_arrays(int part)
8572 struct md_rdev *rdev;
8573 struct detected_devices_node *node_detected_dev;
8575 int i_scanned, i_passed;
8580 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8582 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8584 node_detected_dev = list_entry(all_detected_devices.next,
8585 struct detected_devices_node, list);
8586 list_del(&node_detected_dev->list);
8587 dev = node_detected_dev->dev;
8588 kfree(node_detected_dev);
8589 rdev = md_import_device(dev,0, 90);
8593 if (test_bit(Faulty, &rdev->flags)) {
8597 set_bit(AutoDetected, &rdev->flags);
8598 list_add(&rdev->same_set, &pending_raid_disks);
8602 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8603 i_scanned, i_passed);
8605 autorun_devices(part);
8608 #endif /* !MODULE */
8610 static __exit void md_exit(void)
8612 struct mddev *mddev;
8613 struct list_head *tmp;
8615 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8616 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8618 unregister_blkdev(MD_MAJOR,"md");
8619 unregister_blkdev(mdp_major, "mdp");
8620 unregister_reboot_notifier(&md_notifier);
8621 unregister_sysctl_table(raid_table_header);
8622 remove_proc_entry("mdstat", NULL);
8623 for_each_mddev(mddev, tmp) {
8624 export_array(mddev);
8625 mddev->hold_active = 0;
8627 destroy_workqueue(md_misc_wq);
8628 destroy_workqueue(md_wq);
8631 subsys_initcall(md_init);
8632 module_exit(md_exit)
8634 static int get_ro(char *buffer, struct kernel_param *kp)
8636 return sprintf(buffer, "%d", start_readonly);
8638 static int set_ro(const char *val, struct kernel_param *kp)
8641 int num = simple_strtoul(val, &e, 10);
8642 if (*val && (*e == '\0' || *e == '\n')) {
8643 start_readonly = num;
8649 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8650 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8652 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8654 EXPORT_SYMBOL(register_md_personality);
8655 EXPORT_SYMBOL(unregister_md_personality);
8656 EXPORT_SYMBOL(md_error);
8657 EXPORT_SYMBOL(md_done_sync);
8658 EXPORT_SYMBOL(md_write_start);
8659 EXPORT_SYMBOL(md_write_end);
8660 EXPORT_SYMBOL(md_register_thread);
8661 EXPORT_SYMBOL(md_unregister_thread);
8662 EXPORT_SYMBOL(md_wakeup_thread);
8663 EXPORT_SYMBOL(md_check_recovery);
8664 EXPORT_SYMBOL(md_reap_sync_thread);
8665 MODULE_LICENSE("GPL");
8666 MODULE_DESCRIPTION("MD RAID framework");
8668 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);