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 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
84 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
87 * the RAID driver will use the maximum available bandwidth if the IO
88 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93 * or /sys/block/mdX/md/sync_speed_{min,max}
96 static int sysctl_speed_limit_min = 1000;
97 static int sysctl_speed_limit_max = 200000;
98 static inline int speed_min(struct mddev *mddev)
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
104 static inline int speed_max(struct mddev *mddev)
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
110 static struct ctl_table_header *raid_table_header;
112 static ctl_table raid_table[] = {
114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
117 .mode = S_IRUGO|S_IWUSR,
118 .proc_handler = proc_dointvec,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = proc_dointvec,
130 static ctl_table raid_dir_table[] = {
134 .mode = S_IRUGO|S_IXUGO,
140 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static const struct block_device_operations md_fops;
152 static int start_readonly;
155 * like bio_clone, but with a local bio set
158 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
163 if (!mddev || !mddev->bio_set)
164 return bio_alloc(gfp_mask, nr_iovecs);
166 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
171 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
173 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
176 if (!mddev || !mddev->bio_set)
177 return bio_clone(bio, gfp_mask);
179 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
181 EXPORT_SYMBOL_GPL(bio_clone_mddev);
183 void md_trim_bio(struct bio *bio, int offset, int size)
185 /* 'bio' is a cloned bio which we need to trim to match
186 * the given offset and size.
187 * This requires adjusting bi_sector, bi_size, and bi_io_vec
190 struct bio_vec *bvec;
194 if (offset == 0 && size == bio->bi_size)
197 bio->bi_sector += offset;
200 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
202 while (bio->bi_idx < bio->bi_vcnt &&
203 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
204 /* remove this whole bio_vec */
205 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
208 if (bio->bi_idx < bio->bi_vcnt) {
209 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
210 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
212 /* avoid any complications with bi_idx being non-zero*/
214 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
215 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
216 bio->bi_vcnt -= bio->bi_idx;
219 /* Make sure vcnt and last bv are not too big */
220 bio_for_each_segment(bvec, bio, i) {
221 if (sofar + bvec->bv_len > size)
222 bvec->bv_len = size - sofar;
223 if (bvec->bv_len == 0) {
227 sofar += bvec->bv_len;
230 EXPORT_SYMBOL_GPL(md_trim_bio);
233 * We have a system wide 'event count' that is incremented
234 * on any 'interesting' event, and readers of /proc/mdstat
235 * can use 'poll' or 'select' to find out when the event
239 * start array, stop array, error, add device, remove device,
240 * start build, activate spare
242 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
243 static atomic_t md_event_count;
244 void md_new_event(struct mddev *mddev)
246 atomic_inc(&md_event_count);
247 wake_up(&md_event_waiters);
249 EXPORT_SYMBOL_GPL(md_new_event);
251 /* Alternate version that can be called from interrupts
252 * when calling sysfs_notify isn't needed.
254 static void md_new_event_inintr(struct mddev *mddev)
256 atomic_inc(&md_event_count);
257 wake_up(&md_event_waiters);
261 * Enables to iterate over all existing md arrays
262 * all_mddevs_lock protects this list.
264 static LIST_HEAD(all_mddevs);
265 static DEFINE_SPINLOCK(all_mddevs_lock);
269 * iterates through all used mddevs in the system.
270 * We take care to grab the all_mddevs_lock whenever navigating
271 * the list, and to always hold a refcount when unlocked.
272 * Any code which breaks out of this loop while own
273 * a reference to the current mddev and must mddev_put it.
275 #define for_each_mddev(_mddev,_tmp) \
277 for (({ spin_lock(&all_mddevs_lock); \
278 _tmp = all_mddevs.next; \
280 ({ if (_tmp != &all_mddevs) \
281 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
282 spin_unlock(&all_mddevs_lock); \
283 if (_mddev) mddev_put(_mddev); \
284 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
285 _tmp != &all_mddevs;}); \
286 ({ spin_lock(&all_mddevs_lock); \
287 _tmp = _tmp->next;}) \
291 /* Rather than calling directly into the personality make_request function,
292 * IO requests come here first so that we can check if the device is
293 * being suspended pending a reconfiguration.
294 * We hold a refcount over the call to ->make_request. By the time that
295 * call has finished, the bio has been linked into some internal structure
296 * and so is visible to ->quiesce(), so we don't need the refcount any more.
298 static void md_make_request(struct request_queue *q, struct bio *bio)
300 const int rw = bio_data_dir(bio);
301 struct mddev *mddev = q->queuedata;
303 unsigned int sectors;
305 if (mddev == NULL || mddev->pers == NULL
310 smp_rmb(); /* Ensure implications of 'active' are visible */
312 if (mddev->suspended) {
315 prepare_to_wait(&mddev->sb_wait, &__wait,
316 TASK_UNINTERRUPTIBLE);
317 if (!mddev->suspended)
323 finish_wait(&mddev->sb_wait, &__wait);
325 atomic_inc(&mddev->active_io);
329 * save the sectors now since our bio can
330 * go away inside make_request
332 sectors = bio_sectors(bio);
333 mddev->pers->make_request(mddev, bio);
335 cpu = part_stat_lock();
336 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
337 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
340 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
341 wake_up(&mddev->sb_wait);
344 /* mddev_suspend makes sure no new requests are submitted
345 * to the device, and that any requests that have been submitted
346 * are completely handled.
347 * Once ->stop is called and completes, the module will be completely
350 void mddev_suspend(struct mddev *mddev)
352 BUG_ON(mddev->suspended);
353 mddev->suspended = 1;
355 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
356 mddev->pers->quiesce(mddev, 1);
358 del_timer_sync(&mddev->safemode_timer);
360 EXPORT_SYMBOL_GPL(mddev_suspend);
362 void mddev_resume(struct mddev *mddev)
364 mddev->suspended = 0;
365 wake_up(&mddev->sb_wait);
366 mddev->pers->quiesce(mddev, 0);
368 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
369 md_wakeup_thread(mddev->thread);
370 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
372 EXPORT_SYMBOL_GPL(mddev_resume);
374 int mddev_congested(struct mddev *mddev, int bits)
376 return mddev->suspended;
378 EXPORT_SYMBOL(mddev_congested);
381 * Generic flush handling for md
384 static void md_end_flush(struct bio *bio, int err)
386 struct md_rdev *rdev = bio->bi_private;
387 struct mddev *mddev = rdev->mddev;
389 rdev_dec_pending(rdev, mddev);
391 if (atomic_dec_and_test(&mddev->flush_pending)) {
392 /* The pre-request flush has finished */
393 queue_work(md_wq, &mddev->flush_work);
398 static void md_submit_flush_data(struct work_struct *ws);
400 static void submit_flushes(struct work_struct *ws)
402 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
403 struct md_rdev *rdev;
405 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
406 atomic_set(&mddev->flush_pending, 1);
408 rdev_for_each_rcu(rdev, mddev)
409 if (rdev->raid_disk >= 0 &&
410 !test_bit(Faulty, &rdev->flags)) {
411 /* Take two references, one is dropped
412 * when request finishes, one after
413 * we reclaim rcu_read_lock
416 atomic_inc(&rdev->nr_pending);
417 atomic_inc(&rdev->nr_pending);
419 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
420 bi->bi_end_io = md_end_flush;
421 bi->bi_private = rdev;
422 bi->bi_bdev = rdev->bdev;
423 atomic_inc(&mddev->flush_pending);
424 submit_bio(WRITE_FLUSH, bi);
426 rdev_dec_pending(rdev, mddev);
429 if (atomic_dec_and_test(&mddev->flush_pending))
430 queue_work(md_wq, &mddev->flush_work);
433 static void md_submit_flush_data(struct work_struct *ws)
435 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
436 struct bio *bio = mddev->flush_bio;
438 if (bio->bi_size == 0)
439 /* an empty barrier - all done */
442 bio->bi_rw &= ~REQ_FLUSH;
443 mddev->pers->make_request(mddev, bio);
446 mddev->flush_bio = NULL;
447 wake_up(&mddev->sb_wait);
450 void md_flush_request(struct mddev *mddev, struct bio *bio)
452 spin_lock_irq(&mddev->write_lock);
453 wait_event_lock_irq(mddev->sb_wait,
455 mddev->write_lock, /*nothing*/);
456 mddev->flush_bio = bio;
457 spin_unlock_irq(&mddev->write_lock);
459 INIT_WORK(&mddev->flush_work, submit_flushes);
460 queue_work(md_wq, &mddev->flush_work);
462 EXPORT_SYMBOL(md_flush_request);
464 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
466 struct mddev *mddev = cb->data;
467 md_wakeup_thread(mddev->thread);
470 EXPORT_SYMBOL(md_unplug);
472 static inline struct mddev *mddev_get(struct mddev *mddev)
474 atomic_inc(&mddev->active);
478 static void mddev_delayed_delete(struct work_struct *ws);
480 static void mddev_put(struct mddev *mddev)
482 struct bio_set *bs = NULL;
484 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
486 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
487 mddev->ctime == 0 && !mddev->hold_active) {
488 /* Array is not configured at all, and not held active,
490 list_del_init(&mddev->all_mddevs);
492 mddev->bio_set = NULL;
493 if (mddev->gendisk) {
494 /* We did a probe so need to clean up. Call
495 * queue_work inside the spinlock so that
496 * flush_workqueue() after mddev_find will
497 * succeed in waiting for the work to be done.
499 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
500 queue_work(md_misc_wq, &mddev->del_work);
504 spin_unlock(&all_mddevs_lock);
509 void mddev_init(struct mddev *mddev)
511 mutex_init(&mddev->open_mutex);
512 mutex_init(&mddev->reconfig_mutex);
513 mutex_init(&mddev->bitmap_info.mutex);
514 INIT_LIST_HEAD(&mddev->disks);
515 INIT_LIST_HEAD(&mddev->all_mddevs);
516 init_timer(&mddev->safemode_timer);
517 atomic_set(&mddev->active, 1);
518 atomic_set(&mddev->openers, 0);
519 atomic_set(&mddev->active_io, 0);
520 spin_lock_init(&mddev->write_lock);
521 atomic_set(&mddev->flush_pending, 0);
522 init_waitqueue_head(&mddev->sb_wait);
523 init_waitqueue_head(&mddev->recovery_wait);
524 mddev->reshape_position = MaxSector;
525 mddev->reshape_backwards = 0;
526 mddev->resync_min = 0;
527 mddev->resync_max = MaxSector;
528 mddev->level = LEVEL_NONE;
530 EXPORT_SYMBOL_GPL(mddev_init);
532 static struct mddev * mddev_find(dev_t unit)
534 struct mddev *mddev, *new = NULL;
536 if (unit && MAJOR(unit) != MD_MAJOR)
537 unit &= ~((1<<MdpMinorShift)-1);
540 spin_lock(&all_mddevs_lock);
543 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
544 if (mddev->unit == unit) {
546 spin_unlock(&all_mddevs_lock);
552 list_add(&new->all_mddevs, &all_mddevs);
553 spin_unlock(&all_mddevs_lock);
554 new->hold_active = UNTIL_IOCTL;
558 /* find an unused unit number */
559 static int next_minor = 512;
560 int start = next_minor;
564 dev = MKDEV(MD_MAJOR, next_minor);
566 if (next_minor > MINORMASK)
568 if (next_minor == start) {
569 /* Oh dear, all in use. */
570 spin_unlock(&all_mddevs_lock);
576 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
577 if (mddev->unit == dev) {
583 new->md_minor = MINOR(dev);
584 new->hold_active = UNTIL_STOP;
585 list_add(&new->all_mddevs, &all_mddevs);
586 spin_unlock(&all_mddevs_lock);
589 spin_unlock(&all_mddevs_lock);
591 new = kzalloc(sizeof(*new), GFP_KERNEL);
596 if (MAJOR(unit) == MD_MAJOR)
597 new->md_minor = MINOR(unit);
599 new->md_minor = MINOR(unit) >> MdpMinorShift;
606 static inline int mddev_lock(struct mddev * mddev)
608 return mutex_lock_interruptible(&mddev->reconfig_mutex);
611 static inline int mddev_is_locked(struct mddev *mddev)
613 return mutex_is_locked(&mddev->reconfig_mutex);
616 static inline int mddev_trylock(struct mddev * mddev)
618 return mutex_trylock(&mddev->reconfig_mutex);
621 static struct attribute_group md_redundancy_group;
623 static void mddev_unlock(struct mddev * mddev)
625 if (mddev->to_remove) {
626 /* These cannot be removed under reconfig_mutex as
627 * an access to the files will try to take reconfig_mutex
628 * while holding the file unremovable, which leads to
630 * So hold set sysfs_active while the remove in happeing,
631 * and anything else which might set ->to_remove or my
632 * otherwise change the sysfs namespace will fail with
633 * -EBUSY if sysfs_active is still set.
634 * We set sysfs_active under reconfig_mutex and elsewhere
635 * test it under the same mutex to ensure its correct value
638 struct attribute_group *to_remove = mddev->to_remove;
639 mddev->to_remove = NULL;
640 mddev->sysfs_active = 1;
641 mutex_unlock(&mddev->reconfig_mutex);
643 if (mddev->kobj.sd) {
644 if (to_remove != &md_redundancy_group)
645 sysfs_remove_group(&mddev->kobj, to_remove);
646 if (mddev->pers == NULL ||
647 mddev->pers->sync_request == NULL) {
648 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
649 if (mddev->sysfs_action)
650 sysfs_put(mddev->sysfs_action);
651 mddev->sysfs_action = NULL;
654 mddev->sysfs_active = 0;
656 mutex_unlock(&mddev->reconfig_mutex);
658 /* As we've dropped the mutex we need a spinlock to
659 * make sure the thread doesn't disappear
661 spin_lock(&pers_lock);
662 md_wakeup_thread(mddev->thread);
663 spin_unlock(&pers_lock);
666 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
668 struct md_rdev *rdev;
670 rdev_for_each(rdev, mddev)
671 if (rdev->desc_nr == nr)
677 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
679 struct md_rdev *rdev;
681 rdev_for_each_rcu(rdev, mddev)
682 if (rdev->desc_nr == nr)
688 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
690 struct md_rdev *rdev;
692 rdev_for_each(rdev, mddev)
693 if (rdev->bdev->bd_dev == dev)
699 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
701 struct md_rdev *rdev;
703 rdev_for_each_rcu(rdev, mddev)
704 if (rdev->bdev->bd_dev == dev)
710 static struct md_personality *find_pers(int level, char *clevel)
712 struct md_personality *pers;
713 list_for_each_entry(pers, &pers_list, list) {
714 if (level != LEVEL_NONE && pers->level == level)
716 if (strcmp(pers->name, clevel)==0)
722 /* return the offset of the super block in 512byte sectors */
723 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
725 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
726 return MD_NEW_SIZE_SECTORS(num_sectors);
729 static int alloc_disk_sb(struct md_rdev * rdev)
734 rdev->sb_page = alloc_page(GFP_KERNEL);
735 if (!rdev->sb_page) {
736 printk(KERN_ALERT "md: out of memory.\n");
743 void md_rdev_clear(struct md_rdev *rdev)
746 put_page(rdev->sb_page);
748 rdev->sb_page = NULL;
753 put_page(rdev->bb_page);
754 rdev->bb_page = NULL;
756 kfree(rdev->badblocks.page);
757 rdev->badblocks.page = NULL;
759 EXPORT_SYMBOL_GPL(md_rdev_clear);
761 static void super_written(struct bio *bio, int error)
763 struct md_rdev *rdev = bio->bi_private;
764 struct mddev *mddev = rdev->mddev;
766 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
767 printk("md: super_written gets error=%d, uptodate=%d\n",
768 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
769 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
770 md_error(mddev, rdev);
773 if (atomic_dec_and_test(&mddev->pending_writes))
774 wake_up(&mddev->sb_wait);
778 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
779 sector_t sector, int size, struct page *page)
781 /* write first size bytes of page to sector of rdev
782 * Increment mddev->pending_writes before returning
783 * and decrement it on completion, waking up sb_wait
784 * if zero is reached.
785 * If an error occurred, call md_error
787 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
789 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
790 bio->bi_sector = sector;
791 bio_add_page(bio, page, size, 0);
792 bio->bi_private = rdev;
793 bio->bi_end_io = super_written;
795 atomic_inc(&mddev->pending_writes);
796 submit_bio(WRITE_FLUSH_FUA, bio);
799 void md_super_wait(struct mddev *mddev)
801 /* wait for all superblock writes that were scheduled to complete */
804 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
805 if (atomic_read(&mddev->pending_writes)==0)
809 finish_wait(&mddev->sb_wait, &wq);
812 static void bi_complete(struct bio *bio, int error)
814 complete((struct completion*)bio->bi_private);
817 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
818 struct page *page, int rw, bool metadata_op)
820 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
821 struct completion event;
826 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
827 rdev->meta_bdev : rdev->bdev;
829 bio->bi_sector = sector + rdev->sb_start;
830 else if (rdev->mddev->reshape_position != MaxSector &&
831 (rdev->mddev->reshape_backwards ==
832 (sector >= rdev->mddev->reshape_position)))
833 bio->bi_sector = sector + rdev->new_data_offset;
835 bio->bi_sector = sector + rdev->data_offset;
836 bio_add_page(bio, page, size, 0);
837 init_completion(&event);
838 bio->bi_private = &event;
839 bio->bi_end_io = bi_complete;
841 wait_for_completion(&event);
843 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
847 EXPORT_SYMBOL_GPL(sync_page_io);
849 static int read_disk_sb(struct md_rdev * rdev, int size)
851 char b[BDEVNAME_SIZE];
852 if (!rdev->sb_page) {
860 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
866 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
867 bdevname(rdev->bdev,b));
871 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
873 return sb1->set_uuid0 == sb2->set_uuid0 &&
874 sb1->set_uuid1 == sb2->set_uuid1 &&
875 sb1->set_uuid2 == sb2->set_uuid2 &&
876 sb1->set_uuid3 == sb2->set_uuid3;
879 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
882 mdp_super_t *tmp1, *tmp2;
884 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
885 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
887 if (!tmp1 || !tmp2) {
889 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
897 * nr_disks is not constant
902 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
910 static u32 md_csum_fold(u32 csum)
912 csum = (csum & 0xffff) + (csum >> 16);
913 return (csum & 0xffff) + (csum >> 16);
916 static unsigned int calc_sb_csum(mdp_super_t * sb)
919 u32 *sb32 = (u32*)sb;
921 unsigned int disk_csum, csum;
923 disk_csum = sb->sb_csum;
926 for (i = 0; i < MD_SB_BYTES/4 ; i++)
928 csum = (newcsum & 0xffffffff) + (newcsum>>32);
932 /* This used to use csum_partial, which was wrong for several
933 * reasons including that different results are returned on
934 * different architectures. It isn't critical that we get exactly
935 * the same return value as before (we always csum_fold before
936 * testing, and that removes any differences). However as we
937 * know that csum_partial always returned a 16bit value on
938 * alphas, do a fold to maximise conformity to previous behaviour.
940 sb->sb_csum = md_csum_fold(disk_csum);
942 sb->sb_csum = disk_csum;
949 * Handle superblock details.
950 * We want to be able to handle multiple superblock formats
951 * so we have a common interface to them all, and an array of
952 * different handlers.
953 * We rely on user-space to write the initial superblock, and support
954 * reading and updating of superblocks.
955 * Interface methods are:
956 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
957 * loads and validates a superblock on dev.
958 * if refdev != NULL, compare superblocks on both devices
960 * 0 - dev has a superblock that is compatible with refdev
961 * 1 - dev has a superblock that is compatible and newer than refdev
962 * so dev should be used as the refdev in future
963 * -EINVAL superblock incompatible or invalid
964 * -othererror e.g. -EIO
966 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
967 * Verify that dev is acceptable into mddev.
968 * The first time, mddev->raid_disks will be 0, and data from
969 * dev should be merged in. Subsequent calls check that dev
970 * is new enough. Return 0 or -EINVAL
972 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
973 * Update the superblock for rdev with data in mddev
974 * This does not write to disc.
980 struct module *owner;
981 int (*load_super)(struct md_rdev *rdev,
982 struct md_rdev *refdev,
984 int (*validate_super)(struct mddev *mddev,
985 struct md_rdev *rdev);
986 void (*sync_super)(struct mddev *mddev,
987 struct md_rdev *rdev);
988 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
989 sector_t num_sectors);
990 int (*allow_new_offset)(struct md_rdev *rdev,
991 unsigned long long new_offset);
995 * Check that the given mddev has no bitmap.
997 * This function is called from the run method of all personalities that do not
998 * support bitmaps. It prints an error message and returns non-zero if mddev
999 * has a bitmap. Otherwise, it returns 0.
1002 int md_check_no_bitmap(struct mddev *mddev)
1004 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1006 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1007 mdname(mddev), mddev->pers->name);
1010 EXPORT_SYMBOL(md_check_no_bitmap);
1013 * load_super for 0.90.0
1015 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1017 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1022 * Calculate the position of the superblock (512byte sectors),
1023 * it's at the end of the disk.
1025 * It also happens to be a multiple of 4Kb.
1027 rdev->sb_start = calc_dev_sboffset(rdev);
1029 ret = read_disk_sb(rdev, MD_SB_BYTES);
1030 if (ret) return ret;
1034 bdevname(rdev->bdev, b);
1035 sb = page_address(rdev->sb_page);
1037 if (sb->md_magic != MD_SB_MAGIC) {
1038 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1043 if (sb->major_version != 0 ||
1044 sb->minor_version < 90 ||
1045 sb->minor_version > 91) {
1046 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1047 sb->major_version, sb->minor_version,
1052 if (sb->raid_disks <= 0)
1055 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1056 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1061 rdev->preferred_minor = sb->md_minor;
1062 rdev->data_offset = 0;
1063 rdev->new_data_offset = 0;
1064 rdev->sb_size = MD_SB_BYTES;
1065 rdev->badblocks.shift = -1;
1067 if (sb->level == LEVEL_MULTIPATH)
1070 rdev->desc_nr = sb->this_disk.number;
1076 mdp_super_t *refsb = page_address(refdev->sb_page);
1077 if (!uuid_equal(refsb, sb)) {
1078 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1079 b, bdevname(refdev->bdev,b2));
1082 if (!sb_equal(refsb, sb)) {
1083 printk(KERN_WARNING "md: %s has same UUID"
1084 " but different superblock to %s\n",
1085 b, bdevname(refdev->bdev, b2));
1089 ev2 = md_event(refsb);
1095 rdev->sectors = rdev->sb_start;
1096 /* Limit to 4TB as metadata cannot record more than that.
1097 * (not needed for Linear and RAID0 as metadata doesn't
1100 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1101 rdev->sectors = (2ULL << 32) - 2;
1103 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1104 /* "this cannot possibly happen" ... */
1112 * validate_super for 0.90.0
1114 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1117 mdp_super_t *sb = page_address(rdev->sb_page);
1118 __u64 ev1 = md_event(sb);
1120 rdev->raid_disk = -1;
1121 clear_bit(Faulty, &rdev->flags);
1122 clear_bit(In_sync, &rdev->flags);
1123 clear_bit(WriteMostly, &rdev->flags);
1125 if (mddev->raid_disks == 0) {
1126 mddev->major_version = 0;
1127 mddev->minor_version = sb->minor_version;
1128 mddev->patch_version = sb->patch_version;
1129 mddev->external = 0;
1130 mddev->chunk_sectors = sb->chunk_size >> 9;
1131 mddev->ctime = sb->ctime;
1132 mddev->utime = sb->utime;
1133 mddev->level = sb->level;
1134 mddev->clevel[0] = 0;
1135 mddev->layout = sb->layout;
1136 mddev->raid_disks = sb->raid_disks;
1137 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1138 mddev->events = ev1;
1139 mddev->bitmap_info.offset = 0;
1140 mddev->bitmap_info.space = 0;
1141 /* bitmap can use 60 K after the 4K superblocks */
1142 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1143 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1144 mddev->reshape_backwards = 0;
1146 if (mddev->minor_version >= 91) {
1147 mddev->reshape_position = sb->reshape_position;
1148 mddev->delta_disks = sb->delta_disks;
1149 mddev->new_level = sb->new_level;
1150 mddev->new_layout = sb->new_layout;
1151 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1152 if (mddev->delta_disks < 0)
1153 mddev->reshape_backwards = 1;
1155 mddev->reshape_position = MaxSector;
1156 mddev->delta_disks = 0;
1157 mddev->new_level = mddev->level;
1158 mddev->new_layout = mddev->layout;
1159 mddev->new_chunk_sectors = mddev->chunk_sectors;
1162 if (sb->state & (1<<MD_SB_CLEAN))
1163 mddev->recovery_cp = MaxSector;
1165 if (sb->events_hi == sb->cp_events_hi &&
1166 sb->events_lo == sb->cp_events_lo) {
1167 mddev->recovery_cp = sb->recovery_cp;
1169 mddev->recovery_cp = 0;
1172 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1173 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1174 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1175 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1177 mddev->max_disks = MD_SB_DISKS;
1179 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1180 mddev->bitmap_info.file == NULL) {
1181 mddev->bitmap_info.offset =
1182 mddev->bitmap_info.default_offset;
1183 mddev->bitmap_info.space =
1184 mddev->bitmap_info.space;
1187 } else if (mddev->pers == NULL) {
1188 /* Insist on good event counter while assembling, except
1189 * for spares (which don't need an event count) */
1191 if (sb->disks[rdev->desc_nr].state & (
1192 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1193 if (ev1 < mddev->events)
1195 } else if (mddev->bitmap) {
1196 /* if adding to array with a bitmap, then we can accept an
1197 * older device ... but not too old.
1199 if (ev1 < mddev->bitmap->events_cleared)
1202 if (ev1 < mddev->events)
1203 /* just a hot-add of a new device, leave raid_disk at -1 */
1207 if (mddev->level != LEVEL_MULTIPATH) {
1208 desc = sb->disks + rdev->desc_nr;
1210 if (desc->state & (1<<MD_DISK_FAULTY))
1211 set_bit(Faulty, &rdev->flags);
1212 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1213 desc->raid_disk < mddev->raid_disks */) {
1214 set_bit(In_sync, &rdev->flags);
1215 rdev->raid_disk = desc->raid_disk;
1216 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1217 /* active but not in sync implies recovery up to
1218 * reshape position. We don't know exactly where
1219 * that is, so set to zero for now */
1220 if (mddev->minor_version >= 91) {
1221 rdev->recovery_offset = 0;
1222 rdev->raid_disk = desc->raid_disk;
1225 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1226 set_bit(WriteMostly, &rdev->flags);
1227 } else /* MULTIPATH are always insync */
1228 set_bit(In_sync, &rdev->flags);
1233 * sync_super for 0.90.0
1235 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1238 struct md_rdev *rdev2;
1239 int next_spare = mddev->raid_disks;
1242 /* make rdev->sb match mddev data..
1245 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1246 * 3/ any empty disks < next_spare become removed
1248 * disks[0] gets initialised to REMOVED because
1249 * we cannot be sure from other fields if it has
1250 * been initialised or not.
1253 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1255 rdev->sb_size = MD_SB_BYTES;
1257 sb = page_address(rdev->sb_page);
1259 memset(sb, 0, sizeof(*sb));
1261 sb->md_magic = MD_SB_MAGIC;
1262 sb->major_version = mddev->major_version;
1263 sb->patch_version = mddev->patch_version;
1264 sb->gvalid_words = 0; /* ignored */
1265 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1266 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1267 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1268 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1270 sb->ctime = mddev->ctime;
1271 sb->level = mddev->level;
1272 sb->size = mddev->dev_sectors / 2;
1273 sb->raid_disks = mddev->raid_disks;
1274 sb->md_minor = mddev->md_minor;
1275 sb->not_persistent = 0;
1276 sb->utime = mddev->utime;
1278 sb->events_hi = (mddev->events>>32);
1279 sb->events_lo = (u32)mddev->events;
1281 if (mddev->reshape_position == MaxSector)
1282 sb->minor_version = 90;
1284 sb->minor_version = 91;
1285 sb->reshape_position = mddev->reshape_position;
1286 sb->new_level = mddev->new_level;
1287 sb->delta_disks = mddev->delta_disks;
1288 sb->new_layout = mddev->new_layout;
1289 sb->new_chunk = mddev->new_chunk_sectors << 9;
1291 mddev->minor_version = sb->minor_version;
1294 sb->recovery_cp = mddev->recovery_cp;
1295 sb->cp_events_hi = (mddev->events>>32);
1296 sb->cp_events_lo = (u32)mddev->events;
1297 if (mddev->recovery_cp == MaxSector)
1298 sb->state = (1<< MD_SB_CLEAN);
1300 sb->recovery_cp = 0;
1302 sb->layout = mddev->layout;
1303 sb->chunk_size = mddev->chunk_sectors << 9;
1305 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1306 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1308 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1309 rdev_for_each(rdev2, mddev) {
1312 int is_active = test_bit(In_sync, &rdev2->flags);
1314 if (rdev2->raid_disk >= 0 &&
1315 sb->minor_version >= 91)
1316 /* we have nowhere to store the recovery_offset,
1317 * but if it is not below the reshape_position,
1318 * we can piggy-back on that.
1321 if (rdev2->raid_disk < 0 ||
1322 test_bit(Faulty, &rdev2->flags))
1325 desc_nr = rdev2->raid_disk;
1327 desc_nr = next_spare++;
1328 rdev2->desc_nr = desc_nr;
1329 d = &sb->disks[rdev2->desc_nr];
1331 d->number = rdev2->desc_nr;
1332 d->major = MAJOR(rdev2->bdev->bd_dev);
1333 d->minor = MINOR(rdev2->bdev->bd_dev);
1335 d->raid_disk = rdev2->raid_disk;
1337 d->raid_disk = rdev2->desc_nr; /* compatibility */
1338 if (test_bit(Faulty, &rdev2->flags))
1339 d->state = (1<<MD_DISK_FAULTY);
1340 else if (is_active) {
1341 d->state = (1<<MD_DISK_ACTIVE);
1342 if (test_bit(In_sync, &rdev2->flags))
1343 d->state |= (1<<MD_DISK_SYNC);
1351 if (test_bit(WriteMostly, &rdev2->flags))
1352 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1354 /* now set the "removed" and "faulty" bits on any missing devices */
1355 for (i=0 ; i < mddev->raid_disks ; i++) {
1356 mdp_disk_t *d = &sb->disks[i];
1357 if (d->state == 0 && d->number == 0) {
1360 d->state = (1<<MD_DISK_REMOVED);
1361 d->state |= (1<<MD_DISK_FAULTY);
1365 sb->nr_disks = nr_disks;
1366 sb->active_disks = active;
1367 sb->working_disks = working;
1368 sb->failed_disks = failed;
1369 sb->spare_disks = spare;
1371 sb->this_disk = sb->disks[rdev->desc_nr];
1372 sb->sb_csum = calc_sb_csum(sb);
1376 * rdev_size_change for 0.90.0
1378 static unsigned long long
1379 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1381 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1382 return 0; /* component must fit device */
1383 if (rdev->mddev->bitmap_info.offset)
1384 return 0; /* can't move bitmap */
1385 rdev->sb_start = calc_dev_sboffset(rdev);
1386 if (!num_sectors || num_sectors > rdev->sb_start)
1387 num_sectors = rdev->sb_start;
1388 /* Limit to 4TB as metadata cannot record more than that.
1389 * 4TB == 2^32 KB, or 2*2^32 sectors.
1391 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1392 num_sectors = (2ULL << 32) - 2;
1393 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1395 md_super_wait(rdev->mddev);
1400 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1402 /* non-zero offset changes not possible with v0.90 */
1403 return new_offset == 0;
1407 * version 1 superblock
1410 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1414 unsigned long long newcsum;
1415 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1416 __le32 *isuper = (__le32*)sb;
1419 disk_csum = sb->sb_csum;
1422 for (i=0; size>=4; size -= 4 )
1423 newcsum += le32_to_cpu(*isuper++);
1426 newcsum += le16_to_cpu(*(__le16*) isuper);
1428 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1429 sb->sb_csum = disk_csum;
1430 return cpu_to_le32(csum);
1433 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1435 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1437 struct mdp_superblock_1 *sb;
1441 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1445 * Calculate the position of the superblock in 512byte sectors.
1446 * It is always aligned to a 4K boundary and
1447 * depeding on minor_version, it can be:
1448 * 0: At least 8K, but less than 12K, from end of device
1449 * 1: At start of device
1450 * 2: 4K from start of device.
1452 switch(minor_version) {
1454 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1456 sb_start &= ~(sector_t)(4*2-1);
1467 rdev->sb_start = sb_start;
1469 /* superblock is rarely larger than 1K, but it can be larger,
1470 * and it is safe to read 4k, so we do that
1472 ret = read_disk_sb(rdev, 4096);
1473 if (ret) return ret;
1476 sb = page_address(rdev->sb_page);
1478 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1479 sb->major_version != cpu_to_le32(1) ||
1480 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1481 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1482 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1485 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1486 printk("md: invalid superblock checksum on %s\n",
1487 bdevname(rdev->bdev,b));
1490 if (le64_to_cpu(sb->data_size) < 10) {
1491 printk("md: data_size too small on %s\n",
1492 bdevname(rdev->bdev,b));
1497 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1498 /* Some padding is non-zero, might be a new feature */
1501 rdev->preferred_minor = 0xffff;
1502 rdev->data_offset = le64_to_cpu(sb->data_offset);
1503 rdev->new_data_offset = rdev->data_offset;
1504 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1505 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1506 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1507 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1509 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1510 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1511 if (rdev->sb_size & bmask)
1512 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1515 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1518 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1521 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1524 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1526 if (!rdev->bb_page) {
1527 rdev->bb_page = alloc_page(GFP_KERNEL);
1531 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1532 rdev->badblocks.count == 0) {
1533 /* need to load the bad block list.
1534 * Currently we limit it to one page.
1540 int sectors = le16_to_cpu(sb->bblog_size);
1541 if (sectors > (PAGE_SIZE / 512))
1543 offset = le32_to_cpu(sb->bblog_offset);
1546 bb_sector = (long long)offset;
1547 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1548 rdev->bb_page, READ, true))
1550 bbp = (u64 *)page_address(rdev->bb_page);
1551 rdev->badblocks.shift = sb->bblog_shift;
1552 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1553 u64 bb = le64_to_cpu(*bbp);
1554 int count = bb & (0x3ff);
1555 u64 sector = bb >> 10;
1556 sector <<= sb->bblog_shift;
1557 count <<= sb->bblog_shift;
1560 if (md_set_badblocks(&rdev->badblocks,
1561 sector, count, 1) == 0)
1564 } else if (sb->bblog_offset == 0)
1565 rdev->badblocks.shift = -1;
1571 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1573 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1574 sb->level != refsb->level ||
1575 sb->layout != refsb->layout ||
1576 sb->chunksize != refsb->chunksize) {
1577 printk(KERN_WARNING "md: %s has strangely different"
1578 " superblock to %s\n",
1579 bdevname(rdev->bdev,b),
1580 bdevname(refdev->bdev,b2));
1583 ev1 = le64_to_cpu(sb->events);
1584 ev2 = le64_to_cpu(refsb->events);
1591 if (minor_version) {
1592 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1593 sectors -= rdev->data_offset;
1595 sectors = rdev->sb_start;
1596 if (sectors < le64_to_cpu(sb->data_size))
1598 rdev->sectors = le64_to_cpu(sb->data_size);
1602 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1604 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1605 __u64 ev1 = le64_to_cpu(sb->events);
1607 rdev->raid_disk = -1;
1608 clear_bit(Faulty, &rdev->flags);
1609 clear_bit(In_sync, &rdev->flags);
1610 clear_bit(WriteMostly, &rdev->flags);
1612 if (mddev->raid_disks == 0) {
1613 mddev->major_version = 1;
1614 mddev->patch_version = 0;
1615 mddev->external = 0;
1616 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1617 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1618 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1619 mddev->level = le32_to_cpu(sb->level);
1620 mddev->clevel[0] = 0;
1621 mddev->layout = le32_to_cpu(sb->layout);
1622 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1623 mddev->dev_sectors = le64_to_cpu(sb->size);
1624 mddev->events = ev1;
1625 mddev->bitmap_info.offset = 0;
1626 mddev->bitmap_info.space = 0;
1627 /* Default location for bitmap is 1K after superblock
1628 * using 3K - total of 4K
1630 mddev->bitmap_info.default_offset = 1024 >> 9;
1631 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1632 mddev->reshape_backwards = 0;
1634 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1635 memcpy(mddev->uuid, sb->set_uuid, 16);
1637 mddev->max_disks = (4096-256)/2;
1639 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1640 mddev->bitmap_info.file == NULL) {
1641 mddev->bitmap_info.offset =
1642 (__s32)le32_to_cpu(sb->bitmap_offset);
1643 /* Metadata doesn't record how much space is available.
1644 * For 1.0, we assume we can use up to the superblock
1645 * if before, else to 4K beyond superblock.
1646 * For others, assume no change is possible.
1648 if (mddev->minor_version > 0)
1649 mddev->bitmap_info.space = 0;
1650 else if (mddev->bitmap_info.offset > 0)
1651 mddev->bitmap_info.space =
1652 8 - mddev->bitmap_info.offset;
1654 mddev->bitmap_info.space =
1655 -mddev->bitmap_info.offset;
1658 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1659 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1660 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1661 mddev->new_level = le32_to_cpu(sb->new_level);
1662 mddev->new_layout = le32_to_cpu(sb->new_layout);
1663 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1664 if (mddev->delta_disks < 0 ||
1665 (mddev->delta_disks == 0 &&
1666 (le32_to_cpu(sb->feature_map)
1667 & MD_FEATURE_RESHAPE_BACKWARDS)))
1668 mddev->reshape_backwards = 1;
1670 mddev->reshape_position = MaxSector;
1671 mddev->delta_disks = 0;
1672 mddev->new_level = mddev->level;
1673 mddev->new_layout = mddev->layout;
1674 mddev->new_chunk_sectors = mddev->chunk_sectors;
1677 } else if (mddev->pers == NULL) {
1678 /* Insist of good event counter while assembling, except for
1679 * spares (which don't need an event count) */
1681 if (rdev->desc_nr >= 0 &&
1682 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1683 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1684 if (ev1 < mddev->events)
1686 } else if (mddev->bitmap) {
1687 /* If adding to array with a bitmap, then we can accept an
1688 * older device, but not too old.
1690 if (ev1 < mddev->bitmap->events_cleared)
1693 if (ev1 < mddev->events)
1694 /* just a hot-add of a new device, leave raid_disk at -1 */
1697 if (mddev->level != LEVEL_MULTIPATH) {
1699 if (rdev->desc_nr < 0 ||
1700 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1704 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1706 case 0xffff: /* spare */
1708 case 0xfffe: /* faulty */
1709 set_bit(Faulty, &rdev->flags);
1712 if ((le32_to_cpu(sb->feature_map) &
1713 MD_FEATURE_RECOVERY_OFFSET))
1714 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1716 set_bit(In_sync, &rdev->flags);
1717 rdev->raid_disk = role;
1720 if (sb->devflags & WriteMostly1)
1721 set_bit(WriteMostly, &rdev->flags);
1722 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1723 set_bit(Replacement, &rdev->flags);
1724 } else /* MULTIPATH are always insync */
1725 set_bit(In_sync, &rdev->flags);
1730 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1732 struct mdp_superblock_1 *sb;
1733 struct md_rdev *rdev2;
1735 /* make rdev->sb match mddev and rdev data. */
1737 sb = page_address(rdev->sb_page);
1739 sb->feature_map = 0;
1741 sb->recovery_offset = cpu_to_le64(0);
1742 memset(sb->pad3, 0, sizeof(sb->pad3));
1744 sb->utime = cpu_to_le64((__u64)mddev->utime);
1745 sb->events = cpu_to_le64(mddev->events);
1747 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1749 sb->resync_offset = cpu_to_le64(0);
1751 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1753 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1754 sb->size = cpu_to_le64(mddev->dev_sectors);
1755 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1756 sb->level = cpu_to_le32(mddev->level);
1757 sb->layout = cpu_to_le32(mddev->layout);
1759 if (test_bit(WriteMostly, &rdev->flags))
1760 sb->devflags |= WriteMostly1;
1762 sb->devflags &= ~WriteMostly1;
1763 sb->data_offset = cpu_to_le64(rdev->data_offset);
1764 sb->data_size = cpu_to_le64(rdev->sectors);
1766 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1767 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1768 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1771 if (rdev->raid_disk >= 0 &&
1772 !test_bit(In_sync, &rdev->flags)) {
1774 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1775 sb->recovery_offset =
1776 cpu_to_le64(rdev->recovery_offset);
1778 if (test_bit(Replacement, &rdev->flags))
1780 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1782 if (mddev->reshape_position != MaxSector) {
1783 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1784 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1785 sb->new_layout = cpu_to_le32(mddev->new_layout);
1786 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1787 sb->new_level = cpu_to_le32(mddev->new_level);
1788 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1789 if (mddev->delta_disks == 0 &&
1790 mddev->reshape_backwards)
1792 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1793 if (rdev->new_data_offset != rdev->data_offset) {
1795 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1796 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1797 - rdev->data_offset));
1801 if (rdev->badblocks.count == 0)
1802 /* Nothing to do for bad blocks*/ ;
1803 else if (sb->bblog_offset == 0)
1804 /* Cannot record bad blocks on this device */
1805 md_error(mddev, rdev);
1807 struct badblocks *bb = &rdev->badblocks;
1808 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1810 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1815 seq = read_seqbegin(&bb->lock);
1817 memset(bbp, 0xff, PAGE_SIZE);
1819 for (i = 0 ; i < bb->count ; i++) {
1820 u64 internal_bb = p[i];
1821 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1822 | BB_LEN(internal_bb));
1823 bbp[i] = cpu_to_le64(store_bb);
1826 if (read_seqretry(&bb->lock, seq))
1829 bb->sector = (rdev->sb_start +
1830 (int)le32_to_cpu(sb->bblog_offset));
1831 bb->size = le16_to_cpu(sb->bblog_size);
1836 rdev_for_each(rdev2, mddev)
1837 if (rdev2->desc_nr+1 > max_dev)
1838 max_dev = rdev2->desc_nr+1;
1840 if (max_dev > le32_to_cpu(sb->max_dev)) {
1842 sb->max_dev = cpu_to_le32(max_dev);
1843 rdev->sb_size = max_dev * 2 + 256;
1844 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1845 if (rdev->sb_size & bmask)
1846 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1848 max_dev = le32_to_cpu(sb->max_dev);
1850 for (i=0; i<max_dev;i++)
1851 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1853 rdev_for_each(rdev2, mddev) {
1855 if (test_bit(Faulty, &rdev2->flags))
1856 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1857 else if (test_bit(In_sync, &rdev2->flags))
1858 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1859 else if (rdev2->raid_disk >= 0)
1860 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1862 sb->dev_roles[i] = cpu_to_le16(0xffff);
1865 sb->sb_csum = calc_sb_1_csum(sb);
1868 static unsigned long long
1869 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1871 struct mdp_superblock_1 *sb;
1872 sector_t max_sectors;
1873 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1874 return 0; /* component must fit device */
1875 if (rdev->data_offset != rdev->new_data_offset)
1876 return 0; /* too confusing */
1877 if (rdev->sb_start < rdev->data_offset) {
1878 /* minor versions 1 and 2; superblock before data */
1879 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1880 max_sectors -= rdev->data_offset;
1881 if (!num_sectors || num_sectors > max_sectors)
1882 num_sectors = max_sectors;
1883 } else if (rdev->mddev->bitmap_info.offset) {
1884 /* minor version 0 with bitmap we can't move */
1887 /* minor version 0; superblock after data */
1889 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1890 sb_start &= ~(sector_t)(4*2 - 1);
1891 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1892 if (!num_sectors || num_sectors > max_sectors)
1893 num_sectors = max_sectors;
1894 rdev->sb_start = sb_start;
1896 sb = page_address(rdev->sb_page);
1897 sb->data_size = cpu_to_le64(num_sectors);
1898 sb->super_offset = rdev->sb_start;
1899 sb->sb_csum = calc_sb_1_csum(sb);
1900 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1902 md_super_wait(rdev->mddev);
1908 super_1_allow_new_offset(struct md_rdev *rdev,
1909 unsigned long long new_offset)
1911 /* All necessary checks on new >= old have been done */
1912 struct bitmap *bitmap;
1913 if (new_offset >= rdev->data_offset)
1916 /* with 1.0 metadata, there is no metadata to tread on
1917 * so we can always move back */
1918 if (rdev->mddev->minor_version == 0)
1921 /* otherwise we must be sure not to step on
1922 * any metadata, so stay:
1923 * 36K beyond start of superblock
1924 * beyond end of badblocks
1925 * beyond write-intent bitmap
1927 if (rdev->sb_start + (32+4)*2 > new_offset)
1929 bitmap = rdev->mddev->bitmap;
1930 if (bitmap && !rdev->mddev->bitmap_info.file &&
1931 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1932 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1934 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1940 static struct super_type super_types[] = {
1943 .owner = THIS_MODULE,
1944 .load_super = super_90_load,
1945 .validate_super = super_90_validate,
1946 .sync_super = super_90_sync,
1947 .rdev_size_change = super_90_rdev_size_change,
1948 .allow_new_offset = super_90_allow_new_offset,
1952 .owner = THIS_MODULE,
1953 .load_super = super_1_load,
1954 .validate_super = super_1_validate,
1955 .sync_super = super_1_sync,
1956 .rdev_size_change = super_1_rdev_size_change,
1957 .allow_new_offset = super_1_allow_new_offset,
1961 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1963 if (mddev->sync_super) {
1964 mddev->sync_super(mddev, rdev);
1968 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1970 super_types[mddev->major_version].sync_super(mddev, rdev);
1973 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1975 struct md_rdev *rdev, *rdev2;
1978 rdev_for_each_rcu(rdev, mddev1)
1979 rdev_for_each_rcu(rdev2, mddev2)
1980 if (rdev->bdev->bd_contains ==
1981 rdev2->bdev->bd_contains) {
1989 static LIST_HEAD(pending_raid_disks);
1992 * Try to register data integrity profile for an mddev
1994 * This is called when an array is started and after a disk has been kicked
1995 * from the array. It only succeeds if all working and active component devices
1996 * are integrity capable with matching profiles.
1998 int md_integrity_register(struct mddev *mddev)
2000 struct md_rdev *rdev, *reference = NULL;
2002 if (list_empty(&mddev->disks))
2003 return 0; /* nothing to do */
2004 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2005 return 0; /* shouldn't register, or already is */
2006 rdev_for_each(rdev, mddev) {
2007 /* skip spares and non-functional disks */
2008 if (test_bit(Faulty, &rdev->flags))
2010 if (rdev->raid_disk < 0)
2013 /* Use the first rdev as the reference */
2017 /* does this rdev's profile match the reference profile? */
2018 if (blk_integrity_compare(reference->bdev->bd_disk,
2019 rdev->bdev->bd_disk) < 0)
2022 if (!reference || !bdev_get_integrity(reference->bdev))
2025 * All component devices are integrity capable and have matching
2026 * profiles, register the common profile for the md device.
2028 if (blk_integrity_register(mddev->gendisk,
2029 bdev_get_integrity(reference->bdev)) != 0) {
2030 printk(KERN_ERR "md: failed to register integrity for %s\n",
2034 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2035 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2036 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2042 EXPORT_SYMBOL(md_integrity_register);
2044 /* Disable data integrity if non-capable/non-matching disk is being added */
2045 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2047 struct blk_integrity *bi_rdev;
2048 struct blk_integrity *bi_mddev;
2050 if (!mddev->gendisk)
2053 bi_rdev = bdev_get_integrity(rdev->bdev);
2054 bi_mddev = blk_get_integrity(mddev->gendisk);
2056 if (!bi_mddev) /* nothing to do */
2058 if (rdev->raid_disk < 0) /* skip spares */
2060 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2061 rdev->bdev->bd_disk) >= 0)
2063 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2064 blk_integrity_unregister(mddev->gendisk);
2066 EXPORT_SYMBOL(md_integrity_add_rdev);
2068 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2070 char b[BDEVNAME_SIZE];
2080 /* prevent duplicates */
2081 if (find_rdev(mddev, rdev->bdev->bd_dev))
2084 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2085 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2086 rdev->sectors < mddev->dev_sectors)) {
2088 /* Cannot change size, so fail
2089 * If mddev->level <= 0, then we don't care
2090 * about aligning sizes (e.g. linear)
2092 if (mddev->level > 0)
2095 mddev->dev_sectors = rdev->sectors;
2098 /* Verify rdev->desc_nr is unique.
2099 * If it is -1, assign a free number, else
2100 * check number is not in use
2102 if (rdev->desc_nr < 0) {
2104 if (mddev->pers) choice = mddev->raid_disks;
2105 while (find_rdev_nr(mddev, choice))
2107 rdev->desc_nr = choice;
2109 if (find_rdev_nr(mddev, rdev->desc_nr))
2112 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2113 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2114 mdname(mddev), mddev->max_disks);
2117 bdevname(rdev->bdev,b);
2118 while ( (s=strchr(b, '/')) != NULL)
2121 rdev->mddev = mddev;
2122 printk(KERN_INFO "md: bind<%s>\n", b);
2124 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2127 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2128 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2129 /* failure here is OK */;
2130 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2132 list_add_rcu(&rdev->same_set, &mddev->disks);
2133 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2135 /* May as well allow recovery to be retried once */
2136 mddev->recovery_disabled++;
2141 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2146 static void md_delayed_delete(struct work_struct *ws)
2148 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2149 kobject_del(&rdev->kobj);
2150 kobject_put(&rdev->kobj);
2153 static void unbind_rdev_from_array(struct md_rdev * rdev)
2155 char b[BDEVNAME_SIZE];
2160 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2161 list_del_rcu(&rdev->same_set);
2162 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2164 sysfs_remove_link(&rdev->kobj, "block");
2165 sysfs_put(rdev->sysfs_state);
2166 rdev->sysfs_state = NULL;
2167 rdev->badblocks.count = 0;
2168 /* We need to delay this, otherwise we can deadlock when
2169 * writing to 'remove' to "dev/state". We also need
2170 * to delay it due to rcu usage.
2173 INIT_WORK(&rdev->del_work, md_delayed_delete);
2174 kobject_get(&rdev->kobj);
2175 queue_work(md_misc_wq, &rdev->del_work);
2179 * prevent the device from being mounted, repartitioned or
2180 * otherwise reused by a RAID array (or any other kernel
2181 * subsystem), by bd_claiming the device.
2183 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2186 struct block_device *bdev;
2187 char b[BDEVNAME_SIZE];
2189 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2190 shared ? (struct md_rdev *)lock_rdev : rdev);
2192 printk(KERN_ERR "md: could not open %s.\n",
2193 __bdevname(dev, b));
2194 return PTR_ERR(bdev);
2200 static void unlock_rdev(struct md_rdev *rdev)
2202 struct block_device *bdev = rdev->bdev;
2206 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2209 void md_autodetect_dev(dev_t dev);
2211 static void export_rdev(struct md_rdev * rdev)
2213 char b[BDEVNAME_SIZE];
2214 printk(KERN_INFO "md: export_rdev(%s)\n",
2215 bdevname(rdev->bdev,b));
2218 md_rdev_clear(rdev);
2220 if (test_bit(AutoDetected, &rdev->flags))
2221 md_autodetect_dev(rdev->bdev->bd_dev);
2224 kobject_put(&rdev->kobj);
2227 static void kick_rdev_from_array(struct md_rdev * rdev)
2229 unbind_rdev_from_array(rdev);
2233 static void export_array(struct mddev *mddev)
2235 struct md_rdev *rdev, *tmp;
2237 rdev_for_each_safe(rdev, tmp, mddev) {
2242 kick_rdev_from_array(rdev);
2244 if (!list_empty(&mddev->disks))
2246 mddev->raid_disks = 0;
2247 mddev->major_version = 0;
2250 static void print_desc(mdp_disk_t *desc)
2252 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2253 desc->major,desc->minor,desc->raid_disk,desc->state);
2256 static void print_sb_90(mdp_super_t *sb)
2261 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2262 sb->major_version, sb->minor_version, sb->patch_version,
2263 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2265 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2266 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2267 sb->md_minor, sb->layout, sb->chunk_size);
2268 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2269 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2270 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2271 sb->failed_disks, sb->spare_disks,
2272 sb->sb_csum, (unsigned long)sb->events_lo);
2275 for (i = 0; i < MD_SB_DISKS; i++) {
2278 desc = sb->disks + i;
2279 if (desc->number || desc->major || desc->minor ||
2280 desc->raid_disk || (desc->state && (desc->state != 4))) {
2281 printk(" D %2d: ", i);
2285 printk(KERN_INFO "md: THIS: ");
2286 print_desc(&sb->this_disk);
2289 static void print_sb_1(struct mdp_superblock_1 *sb)
2293 uuid = sb->set_uuid;
2295 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2296 "md: Name: \"%s\" CT:%llu\n",
2297 le32_to_cpu(sb->major_version),
2298 le32_to_cpu(sb->feature_map),
2301 (unsigned long long)le64_to_cpu(sb->ctime)
2302 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2304 uuid = sb->device_uuid;
2306 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2308 "md: Dev:%08x UUID: %pU\n"
2309 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2310 "md: (MaxDev:%u) \n",
2311 le32_to_cpu(sb->level),
2312 (unsigned long long)le64_to_cpu(sb->size),
2313 le32_to_cpu(sb->raid_disks),
2314 le32_to_cpu(sb->layout),
2315 le32_to_cpu(sb->chunksize),
2316 (unsigned long long)le64_to_cpu(sb->data_offset),
2317 (unsigned long long)le64_to_cpu(sb->data_size),
2318 (unsigned long long)le64_to_cpu(sb->super_offset),
2319 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2320 le32_to_cpu(sb->dev_number),
2323 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2324 (unsigned long long)le64_to_cpu(sb->events),
2325 (unsigned long long)le64_to_cpu(sb->resync_offset),
2326 le32_to_cpu(sb->sb_csum),
2327 le32_to_cpu(sb->max_dev)
2331 static void print_rdev(struct md_rdev *rdev, int major_version)
2333 char b[BDEVNAME_SIZE];
2334 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2335 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2336 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2338 if (rdev->sb_loaded) {
2339 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2340 switch (major_version) {
2342 print_sb_90(page_address(rdev->sb_page));
2345 print_sb_1(page_address(rdev->sb_page));
2349 printk(KERN_INFO "md: no rdev superblock!\n");
2352 static void md_print_devices(void)
2354 struct list_head *tmp;
2355 struct md_rdev *rdev;
2356 struct mddev *mddev;
2357 char b[BDEVNAME_SIZE];
2360 printk("md: **********************************\n");
2361 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2362 printk("md: **********************************\n");
2363 for_each_mddev(mddev, tmp) {
2366 bitmap_print_sb(mddev->bitmap);
2368 printk("%s: ", mdname(mddev));
2369 rdev_for_each(rdev, mddev)
2370 printk("<%s>", bdevname(rdev->bdev,b));
2373 rdev_for_each(rdev, mddev)
2374 print_rdev(rdev, mddev->major_version);
2376 printk("md: **********************************\n");
2381 static void sync_sbs(struct mddev * mddev, int nospares)
2383 /* Update each superblock (in-memory image), but
2384 * if we are allowed to, skip spares which already
2385 * have the right event counter, or have one earlier
2386 * (which would mean they aren't being marked as dirty
2387 * with the rest of the array)
2389 struct md_rdev *rdev;
2390 rdev_for_each(rdev, mddev) {
2391 if (rdev->sb_events == mddev->events ||
2393 rdev->raid_disk < 0 &&
2394 rdev->sb_events+1 == mddev->events)) {
2395 /* Don't update this superblock */
2396 rdev->sb_loaded = 2;
2398 sync_super(mddev, rdev);
2399 rdev->sb_loaded = 1;
2404 static void md_update_sb(struct mddev * mddev, int force_change)
2406 struct md_rdev *rdev;
2409 int any_badblocks_changed = 0;
2412 /* First make sure individual recovery_offsets are correct */
2413 rdev_for_each(rdev, mddev) {
2414 if (rdev->raid_disk >= 0 &&
2415 mddev->delta_disks >= 0 &&
2416 !test_bit(In_sync, &rdev->flags) &&
2417 mddev->curr_resync_completed > rdev->recovery_offset)
2418 rdev->recovery_offset = mddev->curr_resync_completed;
2421 if (!mddev->persistent) {
2422 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2423 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2424 if (!mddev->external) {
2425 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2426 rdev_for_each(rdev, mddev) {
2427 if (rdev->badblocks.changed) {
2428 rdev->badblocks.changed = 0;
2429 md_ack_all_badblocks(&rdev->badblocks);
2430 md_error(mddev, rdev);
2432 clear_bit(Blocked, &rdev->flags);
2433 clear_bit(BlockedBadBlocks, &rdev->flags);
2434 wake_up(&rdev->blocked_wait);
2437 wake_up(&mddev->sb_wait);
2441 spin_lock_irq(&mddev->write_lock);
2443 mddev->utime = get_seconds();
2445 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2447 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2448 /* just a clean<-> dirty transition, possibly leave spares alone,
2449 * though if events isn't the right even/odd, we will have to do
2455 if (mddev->degraded)
2456 /* If the array is degraded, then skipping spares is both
2457 * dangerous and fairly pointless.
2458 * Dangerous because a device that was removed from the array
2459 * might have a event_count that still looks up-to-date,
2460 * so it can be re-added without a resync.
2461 * Pointless because if there are any spares to skip,
2462 * then a recovery will happen and soon that array won't
2463 * be degraded any more and the spare can go back to sleep then.
2467 sync_req = mddev->in_sync;
2469 /* If this is just a dirty<->clean transition, and the array is clean
2470 * and 'events' is odd, we can roll back to the previous clean state */
2472 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2473 && mddev->can_decrease_events
2474 && mddev->events != 1) {
2476 mddev->can_decrease_events = 0;
2478 /* otherwise we have to go forward and ... */
2480 mddev->can_decrease_events = nospares;
2483 if (!mddev->events) {
2485 * oops, this 64-bit counter should never wrap.
2486 * Either we are in around ~1 trillion A.C., assuming
2487 * 1 reboot per second, or we have a bug:
2493 rdev_for_each(rdev, mddev) {
2494 if (rdev->badblocks.changed)
2495 any_badblocks_changed++;
2496 if (test_bit(Faulty, &rdev->flags))
2497 set_bit(FaultRecorded, &rdev->flags);
2500 sync_sbs(mddev, nospares);
2501 spin_unlock_irq(&mddev->write_lock);
2503 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2504 mdname(mddev), mddev->in_sync);
2506 bitmap_update_sb(mddev->bitmap);
2507 rdev_for_each(rdev, mddev) {
2508 char b[BDEVNAME_SIZE];
2510 if (rdev->sb_loaded != 1)
2511 continue; /* no noise on spare devices */
2513 if (!test_bit(Faulty, &rdev->flags) &&
2514 rdev->saved_raid_disk == -1) {
2515 md_super_write(mddev,rdev,
2516 rdev->sb_start, rdev->sb_size,
2518 pr_debug("md: (write) %s's sb offset: %llu\n",
2519 bdevname(rdev->bdev, b),
2520 (unsigned long long)rdev->sb_start);
2521 rdev->sb_events = mddev->events;
2522 if (rdev->badblocks.size) {
2523 md_super_write(mddev, rdev,
2524 rdev->badblocks.sector,
2525 rdev->badblocks.size << 9,
2527 rdev->badblocks.size = 0;
2530 } else if (test_bit(Faulty, &rdev->flags))
2531 pr_debug("md: %s (skipping faulty)\n",
2532 bdevname(rdev->bdev, b));
2534 pr_debug("(skipping incremental s/r ");
2536 if (mddev->level == LEVEL_MULTIPATH)
2537 /* only need to write one superblock... */
2540 md_super_wait(mddev);
2541 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2543 spin_lock_irq(&mddev->write_lock);
2544 if (mddev->in_sync != sync_req ||
2545 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2546 /* have to write it out again */
2547 spin_unlock_irq(&mddev->write_lock);
2550 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2551 spin_unlock_irq(&mddev->write_lock);
2552 wake_up(&mddev->sb_wait);
2553 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2554 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2556 rdev_for_each(rdev, mddev) {
2557 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2558 clear_bit(Blocked, &rdev->flags);
2560 if (any_badblocks_changed)
2561 md_ack_all_badblocks(&rdev->badblocks);
2562 clear_bit(BlockedBadBlocks, &rdev->flags);
2563 wake_up(&rdev->blocked_wait);
2567 /* words written to sysfs files may, or may not, be \n terminated.
2568 * We want to accept with case. For this we use cmd_match.
2570 static int cmd_match(const char *cmd, const char *str)
2572 /* See if cmd, written into a sysfs file, matches
2573 * str. They must either be the same, or cmd can
2574 * have a trailing newline
2576 while (*cmd && *str && *cmd == *str) {
2587 struct rdev_sysfs_entry {
2588 struct attribute attr;
2589 ssize_t (*show)(struct md_rdev *, char *);
2590 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2594 state_show(struct md_rdev *rdev, char *page)
2599 if (test_bit(Faulty, &rdev->flags) ||
2600 rdev->badblocks.unacked_exist) {
2601 len+= sprintf(page+len, "%sfaulty",sep);
2604 if (test_bit(In_sync, &rdev->flags)) {
2605 len += sprintf(page+len, "%sin_sync",sep);
2608 if (test_bit(WriteMostly, &rdev->flags)) {
2609 len += sprintf(page+len, "%swrite_mostly",sep);
2612 if (test_bit(Blocked, &rdev->flags) ||
2613 (rdev->badblocks.unacked_exist
2614 && !test_bit(Faulty, &rdev->flags))) {
2615 len += sprintf(page+len, "%sblocked", sep);
2618 if (!test_bit(Faulty, &rdev->flags) &&
2619 !test_bit(In_sync, &rdev->flags)) {
2620 len += sprintf(page+len, "%sspare", sep);
2623 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2624 len += sprintf(page+len, "%swrite_error", sep);
2627 if (test_bit(WantReplacement, &rdev->flags)) {
2628 len += sprintf(page+len, "%swant_replacement", sep);
2631 if (test_bit(Replacement, &rdev->flags)) {
2632 len += sprintf(page+len, "%sreplacement", sep);
2636 return len+sprintf(page+len, "\n");
2640 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2643 * faulty - simulates an error
2644 * remove - disconnects the device
2645 * writemostly - sets write_mostly
2646 * -writemostly - clears write_mostly
2647 * blocked - sets the Blocked flags
2648 * -blocked - clears the Blocked and possibly simulates an error
2649 * insync - sets Insync providing device isn't active
2650 * write_error - sets WriteErrorSeen
2651 * -write_error - clears WriteErrorSeen
2654 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2655 md_error(rdev->mddev, rdev);
2656 if (test_bit(Faulty, &rdev->flags))
2660 } else if (cmd_match(buf, "remove")) {
2661 if (rdev->raid_disk >= 0)
2664 struct mddev *mddev = rdev->mddev;
2665 kick_rdev_from_array(rdev);
2667 md_update_sb(mddev, 1);
2668 md_new_event(mddev);
2671 } else if (cmd_match(buf, "writemostly")) {
2672 set_bit(WriteMostly, &rdev->flags);
2674 } else if (cmd_match(buf, "-writemostly")) {
2675 clear_bit(WriteMostly, &rdev->flags);
2677 } else if (cmd_match(buf, "blocked")) {
2678 set_bit(Blocked, &rdev->flags);
2680 } else if (cmd_match(buf, "-blocked")) {
2681 if (!test_bit(Faulty, &rdev->flags) &&
2682 rdev->badblocks.unacked_exist) {
2683 /* metadata handler doesn't understand badblocks,
2684 * so we need to fail the device
2686 md_error(rdev->mddev, rdev);
2688 clear_bit(Blocked, &rdev->flags);
2689 clear_bit(BlockedBadBlocks, &rdev->flags);
2690 wake_up(&rdev->blocked_wait);
2691 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2692 md_wakeup_thread(rdev->mddev->thread);
2695 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2696 set_bit(In_sync, &rdev->flags);
2698 } else if (cmd_match(buf, "write_error")) {
2699 set_bit(WriteErrorSeen, &rdev->flags);
2701 } else if (cmd_match(buf, "-write_error")) {
2702 clear_bit(WriteErrorSeen, &rdev->flags);
2704 } else if (cmd_match(buf, "want_replacement")) {
2705 /* Any non-spare device that is not a replacement can
2706 * become want_replacement at any time, but we then need to
2707 * check if recovery is needed.
2709 if (rdev->raid_disk >= 0 &&
2710 !test_bit(Replacement, &rdev->flags))
2711 set_bit(WantReplacement, &rdev->flags);
2712 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2713 md_wakeup_thread(rdev->mddev->thread);
2715 } else if (cmd_match(buf, "-want_replacement")) {
2716 /* Clearing 'want_replacement' is always allowed.
2717 * Once replacements starts it is too late though.
2720 clear_bit(WantReplacement, &rdev->flags);
2721 } else if (cmd_match(buf, "replacement")) {
2722 /* Can only set a device as a replacement when array has not
2723 * yet been started. Once running, replacement is automatic
2724 * from spares, or by assigning 'slot'.
2726 if (rdev->mddev->pers)
2729 set_bit(Replacement, &rdev->flags);
2732 } else if (cmd_match(buf, "-replacement")) {
2733 /* Similarly, can only clear Replacement before start */
2734 if (rdev->mddev->pers)
2737 clear_bit(Replacement, &rdev->flags);
2742 sysfs_notify_dirent_safe(rdev->sysfs_state);
2743 return err ? err : len;
2745 static struct rdev_sysfs_entry rdev_state =
2746 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2749 errors_show(struct md_rdev *rdev, char *page)
2751 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2755 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2758 unsigned long n = simple_strtoul(buf, &e, 10);
2759 if (*buf && (*e == 0 || *e == '\n')) {
2760 atomic_set(&rdev->corrected_errors, n);
2765 static struct rdev_sysfs_entry rdev_errors =
2766 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2769 slot_show(struct md_rdev *rdev, char *page)
2771 if (rdev->raid_disk < 0)
2772 return sprintf(page, "none\n");
2774 return sprintf(page, "%d\n", rdev->raid_disk);
2778 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2782 int slot = simple_strtoul(buf, &e, 10);
2783 if (strncmp(buf, "none", 4)==0)
2785 else if (e==buf || (*e && *e!= '\n'))
2787 if (rdev->mddev->pers && slot == -1) {
2788 /* Setting 'slot' on an active array requires also
2789 * updating the 'rd%d' link, and communicating
2790 * with the personality with ->hot_*_disk.
2791 * For now we only support removing
2792 * failed/spare devices. This normally happens automatically,
2793 * but not when the metadata is externally managed.
2795 if (rdev->raid_disk == -1)
2797 /* personality does all needed checks */
2798 if (rdev->mddev->pers->hot_remove_disk == NULL)
2800 err = rdev->mddev->pers->
2801 hot_remove_disk(rdev->mddev, rdev);
2804 sysfs_unlink_rdev(rdev->mddev, rdev);
2805 rdev->raid_disk = -1;
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) -
2999 if (sectors < my_mddev->dev_sectors)
3000 return -EINVAL; /* component must fit device */
3002 rdev->sectors = sectors;
3003 if (sectors > oldsectors && my_mddev->external) {
3004 /* need to check that all other rdevs with the same ->bdev
3005 * do not overlap. We need to unlock the mddev to avoid
3006 * a deadlock. We have already changed rdev->sectors, and if
3007 * we have to change it back, we will have the lock again.
3009 struct mddev *mddev;
3011 struct list_head *tmp;
3013 mddev_unlock(my_mddev);
3014 for_each_mddev(mddev, tmp) {
3015 struct md_rdev *rdev2;
3018 rdev_for_each(rdev2, mddev)
3019 if (rdev->bdev == rdev2->bdev &&
3021 overlaps(rdev->data_offset, rdev->sectors,
3027 mddev_unlock(mddev);
3033 mddev_lock(my_mddev);
3035 /* Someone else could have slipped in a size
3036 * change here, but doing so is just silly.
3037 * We put oldsectors back because we *know* it is
3038 * safe, and trust userspace not to race with
3041 rdev->sectors = oldsectors;
3048 static struct rdev_sysfs_entry rdev_size =
3049 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3052 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3054 unsigned long long recovery_start = rdev->recovery_offset;
3056 if (test_bit(In_sync, &rdev->flags) ||
3057 recovery_start == MaxSector)
3058 return sprintf(page, "none\n");
3060 return sprintf(page, "%llu\n", recovery_start);
3063 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3065 unsigned long long recovery_start;
3067 if (cmd_match(buf, "none"))
3068 recovery_start = MaxSector;
3069 else if (strict_strtoull(buf, 10, &recovery_start))
3072 if (rdev->mddev->pers &&
3073 rdev->raid_disk >= 0)
3076 rdev->recovery_offset = recovery_start;
3077 if (recovery_start == MaxSector)
3078 set_bit(In_sync, &rdev->flags);
3080 clear_bit(In_sync, &rdev->flags);
3084 static struct rdev_sysfs_entry rdev_recovery_start =
3085 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3089 badblocks_show(struct badblocks *bb, char *page, int unack);
3091 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3093 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3095 return badblocks_show(&rdev->badblocks, page, 0);
3097 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3099 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3100 /* Maybe that ack was all we needed */
3101 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3102 wake_up(&rdev->blocked_wait);
3105 static struct rdev_sysfs_entry rdev_bad_blocks =
3106 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3109 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3111 return badblocks_show(&rdev->badblocks, page, 1);
3113 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3115 return badblocks_store(&rdev->badblocks, page, len, 1);
3117 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3118 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3120 static struct attribute *rdev_default_attrs[] = {
3125 &rdev_new_offset.attr,
3127 &rdev_recovery_start.attr,
3128 &rdev_bad_blocks.attr,
3129 &rdev_unack_bad_blocks.attr,
3133 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3135 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3136 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3137 struct mddev *mddev = rdev->mddev;
3143 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3145 if (rdev->mddev == NULL)
3148 rv = entry->show(rdev, page);
3149 mddev_unlock(mddev);
3155 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3156 const char *page, size_t length)
3158 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3159 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3161 struct mddev *mddev = rdev->mddev;
3165 if (!capable(CAP_SYS_ADMIN))
3167 rv = mddev ? mddev_lock(mddev): -EBUSY;
3169 if (rdev->mddev == NULL)
3172 rv = entry->store(rdev, page, length);
3173 mddev_unlock(mddev);
3178 static void rdev_free(struct kobject *ko)
3180 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3183 static const struct sysfs_ops rdev_sysfs_ops = {
3184 .show = rdev_attr_show,
3185 .store = rdev_attr_store,
3187 static struct kobj_type rdev_ktype = {
3188 .release = rdev_free,
3189 .sysfs_ops = &rdev_sysfs_ops,
3190 .default_attrs = rdev_default_attrs,
3193 int md_rdev_init(struct md_rdev *rdev)
3196 rdev->saved_raid_disk = -1;
3197 rdev->raid_disk = -1;
3199 rdev->data_offset = 0;
3200 rdev->new_data_offset = 0;
3201 rdev->sb_events = 0;
3202 rdev->last_read_error.tv_sec = 0;
3203 rdev->last_read_error.tv_nsec = 0;
3204 rdev->sb_loaded = 0;
3205 rdev->bb_page = NULL;
3206 atomic_set(&rdev->nr_pending, 0);
3207 atomic_set(&rdev->read_errors, 0);
3208 atomic_set(&rdev->corrected_errors, 0);
3210 INIT_LIST_HEAD(&rdev->same_set);
3211 init_waitqueue_head(&rdev->blocked_wait);
3213 /* Add space to store bad block list.
3214 * This reserves the space even on arrays where it cannot
3215 * be used - I wonder if that matters
3217 rdev->badblocks.count = 0;
3218 rdev->badblocks.shift = 0;
3219 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3220 seqlock_init(&rdev->badblocks.lock);
3221 if (rdev->badblocks.page == NULL)
3226 EXPORT_SYMBOL_GPL(md_rdev_init);
3228 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3230 * mark the device faulty if:
3232 * - the device is nonexistent (zero size)
3233 * - the device has no valid superblock
3235 * a faulty rdev _never_ has rdev->sb set.
3237 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3239 char b[BDEVNAME_SIZE];
3241 struct md_rdev *rdev;
3244 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3246 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3247 return ERR_PTR(-ENOMEM);
3250 err = md_rdev_init(rdev);
3253 err = alloc_disk_sb(rdev);
3257 err = lock_rdev(rdev, newdev, super_format == -2);
3261 kobject_init(&rdev->kobj, &rdev_ktype);
3263 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3266 "md: %s has zero or unknown size, marking faulty!\n",
3267 bdevname(rdev->bdev,b));
3272 if (super_format >= 0) {
3273 err = super_types[super_format].
3274 load_super(rdev, NULL, super_minor);
3275 if (err == -EINVAL) {
3277 "md: %s does not have a valid v%d.%d "
3278 "superblock, not importing!\n",
3279 bdevname(rdev->bdev,b),
3280 super_format, super_minor);
3285 "md: could not read %s's sb, not importing!\n",
3286 bdevname(rdev->bdev,b));
3290 if (super_format == -1)
3291 /* hot-add for 0.90, or non-persistent: so no badblocks */
3292 rdev->badblocks.shift = -1;
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);
3623 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3624 mddev_resume(mddev);
3625 sysfs_notify(&mddev->kobj, NULL, "level");
3626 md_new_event(mddev);
3630 static struct md_sysfs_entry md_level =
3631 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3635 layout_show(struct mddev *mddev, char *page)
3637 /* just a number, not meaningful for all levels */
3638 if (mddev->reshape_position != MaxSector &&
3639 mddev->layout != mddev->new_layout)
3640 return sprintf(page, "%d (%d)\n",
3641 mddev->new_layout, mddev->layout);
3642 return sprintf(page, "%d\n", mddev->layout);
3646 layout_store(struct mddev *mddev, const char *buf, size_t len)
3649 unsigned long n = simple_strtoul(buf, &e, 10);
3651 if (!*buf || (*e && *e != '\n'))
3656 if (mddev->pers->check_reshape == NULL)
3658 mddev->new_layout = n;
3659 err = mddev->pers->check_reshape(mddev);
3661 mddev->new_layout = mddev->layout;
3665 mddev->new_layout = n;
3666 if (mddev->reshape_position == MaxSector)
3671 static struct md_sysfs_entry md_layout =
3672 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3676 raid_disks_show(struct mddev *mddev, char *page)
3678 if (mddev->raid_disks == 0)
3680 if (mddev->reshape_position != MaxSector &&
3681 mddev->delta_disks != 0)
3682 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3683 mddev->raid_disks - mddev->delta_disks);
3684 return sprintf(page, "%d\n", mddev->raid_disks);
3687 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3690 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3694 unsigned long n = simple_strtoul(buf, &e, 10);
3696 if (!*buf || (*e && *e != '\n'))
3700 rv = update_raid_disks(mddev, n);
3701 else if (mddev->reshape_position != MaxSector) {
3702 struct md_rdev *rdev;
3703 int olddisks = mddev->raid_disks - mddev->delta_disks;
3705 rdev_for_each(rdev, mddev) {
3707 rdev->data_offset < rdev->new_data_offset)
3710 rdev->data_offset > rdev->new_data_offset)
3713 mddev->delta_disks = n - olddisks;
3714 mddev->raid_disks = n;
3715 mddev->reshape_backwards = (mddev->delta_disks < 0);
3717 mddev->raid_disks = n;
3718 return rv ? rv : len;
3720 static struct md_sysfs_entry md_raid_disks =
3721 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3724 chunk_size_show(struct mddev *mddev, char *page)
3726 if (mddev->reshape_position != MaxSector &&
3727 mddev->chunk_sectors != mddev->new_chunk_sectors)
3728 return sprintf(page, "%d (%d)\n",
3729 mddev->new_chunk_sectors << 9,
3730 mddev->chunk_sectors << 9);
3731 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3735 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3738 unsigned long n = simple_strtoul(buf, &e, 10);
3740 if (!*buf || (*e && *e != '\n'))
3745 if (mddev->pers->check_reshape == NULL)
3747 mddev->new_chunk_sectors = n >> 9;
3748 err = mddev->pers->check_reshape(mddev);
3750 mddev->new_chunk_sectors = mddev->chunk_sectors;
3754 mddev->new_chunk_sectors = n >> 9;
3755 if (mddev->reshape_position == MaxSector)
3756 mddev->chunk_sectors = n >> 9;
3760 static struct md_sysfs_entry md_chunk_size =
3761 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3764 resync_start_show(struct mddev *mddev, char *page)
3766 if (mddev->recovery_cp == MaxSector)
3767 return sprintf(page, "none\n");
3768 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3772 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3775 unsigned long long n = simple_strtoull(buf, &e, 10);
3777 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3779 if (cmd_match(buf, "none"))
3781 else if (!*buf || (*e && *e != '\n'))
3784 mddev->recovery_cp = n;
3786 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3789 static struct md_sysfs_entry md_resync_start =
3790 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3793 * The array state can be:
3796 * No devices, no size, no level
3797 * Equivalent to STOP_ARRAY ioctl
3799 * May have some settings, but array is not active
3800 * all IO results in error
3801 * When written, doesn't tear down array, but just stops it
3802 * suspended (not supported yet)
3803 * All IO requests will block. The array can be reconfigured.
3804 * Writing this, if accepted, will block until array is quiescent
3806 * no resync can happen. no superblocks get written.
3807 * write requests fail
3809 * like readonly, but behaves like 'clean' on a write request.
3811 * clean - no pending writes, but otherwise active.
3812 * When written to inactive array, starts without resync
3813 * If a write request arrives then
3814 * if metadata is known, mark 'dirty' and switch to 'active'.
3815 * if not known, block and switch to write-pending
3816 * If written to an active array that has pending writes, then fails.
3818 * fully active: IO and resync can be happening.
3819 * When written to inactive array, starts with resync
3822 * clean, but writes are blocked waiting for 'active' to be written.
3825 * like active, but no writes have been seen for a while (100msec).
3828 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3829 write_pending, active_idle, bad_word};
3830 static char *array_states[] = {
3831 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3832 "write-pending", "active-idle", NULL };
3834 static int match_word(const char *word, char **list)
3837 for (n=0; list[n]; n++)
3838 if (cmd_match(word, list[n]))
3844 array_state_show(struct mddev *mddev, char *page)
3846 enum array_state st = inactive;
3859 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3861 else if (mddev->safemode)
3867 if (list_empty(&mddev->disks) &&
3868 mddev->raid_disks == 0 &&
3869 mddev->dev_sectors == 0)
3874 return sprintf(page, "%s\n", array_states[st]);
3877 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3878 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3879 static int do_md_run(struct mddev * mddev);
3880 static int restart_array(struct mddev *mddev);
3883 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3886 enum array_state st = match_word(buf, array_states);
3891 /* stopping an active array */
3892 err = do_md_stop(mddev, 0, NULL);
3895 /* stopping an active array */
3897 err = do_md_stop(mddev, 2, NULL);
3899 err = 0; /* already inactive */
3902 break; /* not supported yet */
3905 err = md_set_readonly(mddev, NULL);
3908 set_disk_ro(mddev->gendisk, 1);
3909 err = do_md_run(mddev);
3915 err = md_set_readonly(mddev, NULL);
3916 else if (mddev->ro == 1)
3917 err = restart_array(mddev);
3920 set_disk_ro(mddev->gendisk, 0);
3924 err = do_md_run(mddev);
3929 restart_array(mddev);
3930 spin_lock_irq(&mddev->write_lock);
3931 if (atomic_read(&mddev->writes_pending) == 0) {
3932 if (mddev->in_sync == 0) {
3934 if (mddev->safemode == 1)
3935 mddev->safemode = 0;
3936 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3941 spin_unlock_irq(&mddev->write_lock);
3947 restart_array(mddev);
3948 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3949 wake_up(&mddev->sb_wait);
3953 set_disk_ro(mddev->gendisk, 0);
3954 err = do_md_run(mddev);
3959 /* these cannot be set */
3965 if (mddev->hold_active == UNTIL_IOCTL)
3966 mddev->hold_active = 0;
3967 sysfs_notify_dirent_safe(mddev->sysfs_state);
3971 static struct md_sysfs_entry md_array_state =
3972 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3975 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3976 return sprintf(page, "%d\n",
3977 atomic_read(&mddev->max_corr_read_errors));
3981 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3984 unsigned long n = simple_strtoul(buf, &e, 10);
3986 if (*buf && (*e == 0 || *e == '\n')) {
3987 atomic_set(&mddev->max_corr_read_errors, n);
3993 static struct md_sysfs_entry max_corr_read_errors =
3994 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3995 max_corrected_read_errors_store);
3998 null_show(struct mddev *mddev, char *page)
4004 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4006 /* buf must be %d:%d\n? giving major and minor numbers */
4007 /* The new device is added to the array.
4008 * If the array has a persistent superblock, we read the
4009 * superblock to initialise info and check validity.
4010 * Otherwise, only checking done is that in bind_rdev_to_array,
4011 * which mainly checks size.
4014 int major = simple_strtoul(buf, &e, 10);
4017 struct md_rdev *rdev;
4020 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4022 minor = simple_strtoul(e+1, &e, 10);
4023 if (*e && *e != '\n')
4025 dev = MKDEV(major, minor);
4026 if (major != MAJOR(dev) ||
4027 minor != MINOR(dev))
4031 if (mddev->persistent) {
4032 rdev = md_import_device(dev, mddev->major_version,
4033 mddev->minor_version);
4034 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4035 struct md_rdev *rdev0
4036 = list_entry(mddev->disks.next,
4037 struct md_rdev, same_set);
4038 err = super_types[mddev->major_version]
4039 .load_super(rdev, rdev0, mddev->minor_version);
4043 } else if (mddev->external)
4044 rdev = md_import_device(dev, -2, -1);
4046 rdev = md_import_device(dev, -1, -1);
4049 return PTR_ERR(rdev);
4050 err = bind_rdev_to_array(rdev, mddev);
4054 return err ? err : len;
4057 static struct md_sysfs_entry md_new_device =
4058 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4061 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4064 unsigned long chunk, end_chunk;
4068 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4070 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4071 if (buf == end) break;
4072 if (*end == '-') { /* range */
4074 end_chunk = simple_strtoul(buf, &end, 0);
4075 if (buf == end) break;
4077 if (*end && !isspace(*end)) break;
4078 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4079 buf = skip_spaces(end);
4081 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4086 static struct md_sysfs_entry md_bitmap =
4087 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4090 size_show(struct mddev *mddev, char *page)
4092 return sprintf(page, "%llu\n",
4093 (unsigned long long)mddev->dev_sectors / 2);
4096 static int update_size(struct mddev *mddev, sector_t num_sectors);
4099 size_store(struct mddev *mddev, const char *buf, size_t len)
4101 /* If array is inactive, we can reduce the component size, but
4102 * not increase it (except from 0).
4103 * If array is active, we can try an on-line resize
4106 int err = strict_blocks_to_sectors(buf, §ors);
4111 err = update_size(mddev, sectors);
4112 md_update_sb(mddev, 1);
4114 if (mddev->dev_sectors == 0 ||
4115 mddev->dev_sectors > sectors)
4116 mddev->dev_sectors = sectors;
4120 return err ? err : len;
4123 static struct md_sysfs_entry md_size =
4124 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4129 * 'none' for arrays with no metadata (good luck...)
4130 * 'external' for arrays with externally managed metadata,
4131 * or N.M for internally known formats
4134 metadata_show(struct mddev *mddev, char *page)
4136 if (mddev->persistent)
4137 return sprintf(page, "%d.%d\n",
4138 mddev->major_version, mddev->minor_version);
4139 else if (mddev->external)
4140 return sprintf(page, "external:%s\n", mddev->metadata_type);
4142 return sprintf(page, "none\n");
4146 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4150 /* Changing the details of 'external' metadata is
4151 * always permitted. Otherwise there must be
4152 * no devices attached to the array.
4154 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4156 else if (!list_empty(&mddev->disks))
4159 if (cmd_match(buf, "none")) {
4160 mddev->persistent = 0;
4161 mddev->external = 0;
4162 mddev->major_version = 0;
4163 mddev->minor_version = 90;
4166 if (strncmp(buf, "external:", 9) == 0) {
4167 size_t namelen = len-9;
4168 if (namelen >= sizeof(mddev->metadata_type))
4169 namelen = sizeof(mddev->metadata_type)-1;
4170 strncpy(mddev->metadata_type, buf+9, namelen);
4171 mddev->metadata_type[namelen] = 0;
4172 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4173 mddev->metadata_type[--namelen] = 0;
4174 mddev->persistent = 0;
4175 mddev->external = 1;
4176 mddev->major_version = 0;
4177 mddev->minor_version = 90;
4180 major = simple_strtoul(buf, &e, 10);
4181 if (e==buf || *e != '.')
4184 minor = simple_strtoul(buf, &e, 10);
4185 if (e==buf || (*e && *e != '\n') )
4187 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4189 mddev->major_version = major;
4190 mddev->minor_version = minor;
4191 mddev->persistent = 1;
4192 mddev->external = 0;
4196 static struct md_sysfs_entry md_metadata =
4197 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4200 action_show(struct mddev *mddev, char *page)
4202 char *type = "idle";
4203 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4205 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4206 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4207 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4209 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4210 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4212 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4216 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4219 return sprintf(page, "%s\n", type);
4222 static void reap_sync_thread(struct mddev *mddev);
4225 action_store(struct mddev *mddev, const char *page, size_t len)
4227 if (!mddev->pers || !mddev->pers->sync_request)
4230 if (cmd_match(page, "frozen"))
4231 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4233 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4235 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4236 if (mddev->sync_thread) {
4237 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4238 reap_sync_thread(mddev);
4240 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4241 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4243 else if (cmd_match(page, "resync"))
4244 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4245 else if (cmd_match(page, "recover")) {
4246 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4247 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4248 } else if (cmd_match(page, "reshape")) {
4250 if (mddev->pers->start_reshape == NULL)
4252 err = mddev->pers->start_reshape(mddev);
4255 sysfs_notify(&mddev->kobj, NULL, "degraded");
4257 if (cmd_match(page, "check"))
4258 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4259 else if (!cmd_match(page, "repair"))
4261 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4262 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4264 if (mddev->ro == 2) {
4265 /* A write to sync_action is enough to justify
4266 * canceling read-auto mode
4269 md_wakeup_thread(mddev->sync_thread);
4271 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4272 md_wakeup_thread(mddev->thread);
4273 sysfs_notify_dirent_safe(mddev->sysfs_action);
4278 mismatch_cnt_show(struct mddev *mddev, char *page)
4280 return sprintf(page, "%llu\n",
4281 (unsigned long long)
4282 atomic64_read(&mddev->resync_mismatches));
4285 static struct md_sysfs_entry md_scan_mode =
4286 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4289 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4292 sync_min_show(struct mddev *mddev, char *page)
4294 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4295 mddev->sync_speed_min ? "local": "system");
4299 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4303 if (strncmp(buf, "system", 6)==0) {
4304 mddev->sync_speed_min = 0;
4307 min = simple_strtoul(buf, &e, 10);
4308 if (buf == e || (*e && *e != '\n') || min <= 0)
4310 mddev->sync_speed_min = min;
4314 static struct md_sysfs_entry md_sync_min =
4315 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4318 sync_max_show(struct mddev *mddev, char *page)
4320 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4321 mddev->sync_speed_max ? "local": "system");
4325 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4329 if (strncmp(buf, "system", 6)==0) {
4330 mddev->sync_speed_max = 0;
4333 max = simple_strtoul(buf, &e, 10);
4334 if (buf == e || (*e && *e != '\n') || max <= 0)
4336 mddev->sync_speed_max = max;
4340 static struct md_sysfs_entry md_sync_max =
4341 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4344 degraded_show(struct mddev *mddev, char *page)
4346 return sprintf(page, "%d\n", mddev->degraded);
4348 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4351 sync_force_parallel_show(struct mddev *mddev, char *page)
4353 return sprintf(page, "%d\n", mddev->parallel_resync);
4357 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4361 if (strict_strtol(buf, 10, &n))
4364 if (n != 0 && n != 1)
4367 mddev->parallel_resync = n;
4369 if (mddev->sync_thread)
4370 wake_up(&resync_wait);
4375 /* force parallel resync, even with shared block devices */
4376 static struct md_sysfs_entry md_sync_force_parallel =
4377 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4378 sync_force_parallel_show, sync_force_parallel_store);
4381 sync_speed_show(struct mddev *mddev, char *page)
4383 unsigned long resync, dt, db;
4384 if (mddev->curr_resync == 0)
4385 return sprintf(page, "none\n");
4386 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4387 dt = (jiffies - mddev->resync_mark) / HZ;
4389 db = resync - mddev->resync_mark_cnt;
4390 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4393 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4396 sync_completed_show(struct mddev *mddev, char *page)
4398 unsigned long long max_sectors, resync;
4400 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4401 return sprintf(page, "none\n");
4403 if (mddev->curr_resync == 1 ||
4404 mddev->curr_resync == 2)
4405 return sprintf(page, "delayed\n");
4407 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4408 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4409 max_sectors = mddev->resync_max_sectors;
4411 max_sectors = mddev->dev_sectors;
4413 resync = mddev->curr_resync_completed;
4414 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4417 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4420 min_sync_show(struct mddev *mddev, char *page)
4422 return sprintf(page, "%llu\n",
4423 (unsigned long long)mddev->resync_min);
4426 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4428 unsigned long long min;
4429 if (strict_strtoull(buf, 10, &min))
4431 if (min > mddev->resync_max)
4433 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4436 /* Must be a multiple of chunk_size */
4437 if (mddev->chunk_sectors) {
4438 sector_t temp = min;
4439 if (sector_div(temp, mddev->chunk_sectors))
4442 mddev->resync_min = min;
4447 static struct md_sysfs_entry md_min_sync =
4448 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4451 max_sync_show(struct mddev *mddev, char *page)
4453 if (mddev->resync_max == MaxSector)
4454 return sprintf(page, "max\n");
4456 return sprintf(page, "%llu\n",
4457 (unsigned long long)mddev->resync_max);
4460 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4462 if (strncmp(buf, "max", 3) == 0)
4463 mddev->resync_max = MaxSector;
4465 unsigned long long max;
4466 if (strict_strtoull(buf, 10, &max))
4468 if (max < mddev->resync_min)
4470 if (max < mddev->resync_max &&
4472 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4475 /* Must be a multiple of chunk_size */
4476 if (mddev->chunk_sectors) {
4477 sector_t temp = max;
4478 if (sector_div(temp, mddev->chunk_sectors))
4481 mddev->resync_max = max;
4483 wake_up(&mddev->recovery_wait);
4487 static struct md_sysfs_entry md_max_sync =
4488 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4491 suspend_lo_show(struct mddev *mddev, char *page)
4493 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4497 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4500 unsigned long long new = simple_strtoull(buf, &e, 10);
4501 unsigned long long old = mddev->suspend_lo;
4503 if (mddev->pers == NULL ||
4504 mddev->pers->quiesce == NULL)
4506 if (buf == e || (*e && *e != '\n'))
4509 mddev->suspend_lo = new;
4511 /* Shrinking suspended region */
4512 mddev->pers->quiesce(mddev, 2);
4514 /* Expanding suspended region - need to wait */
4515 mddev->pers->quiesce(mddev, 1);
4516 mddev->pers->quiesce(mddev, 0);
4520 static struct md_sysfs_entry md_suspend_lo =
4521 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4525 suspend_hi_show(struct mddev *mddev, char *page)
4527 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4531 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4534 unsigned long long new = simple_strtoull(buf, &e, 10);
4535 unsigned long long old = mddev->suspend_hi;
4537 if (mddev->pers == NULL ||
4538 mddev->pers->quiesce == NULL)
4540 if (buf == e || (*e && *e != '\n'))
4543 mddev->suspend_hi = new;
4545 /* Shrinking suspended region */
4546 mddev->pers->quiesce(mddev, 2);
4548 /* Expanding suspended region - need to wait */
4549 mddev->pers->quiesce(mddev, 1);
4550 mddev->pers->quiesce(mddev, 0);
4554 static struct md_sysfs_entry md_suspend_hi =
4555 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4558 reshape_position_show(struct mddev *mddev, char *page)
4560 if (mddev->reshape_position != MaxSector)
4561 return sprintf(page, "%llu\n",
4562 (unsigned long long)mddev->reshape_position);
4563 strcpy(page, "none\n");
4568 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4570 struct md_rdev *rdev;
4572 unsigned long long new = simple_strtoull(buf, &e, 10);
4575 if (buf == e || (*e && *e != '\n'))
4577 mddev->reshape_position = new;
4578 mddev->delta_disks = 0;
4579 mddev->reshape_backwards = 0;
4580 mddev->new_level = mddev->level;
4581 mddev->new_layout = mddev->layout;
4582 mddev->new_chunk_sectors = mddev->chunk_sectors;
4583 rdev_for_each(rdev, mddev)
4584 rdev->new_data_offset = rdev->data_offset;
4588 static struct md_sysfs_entry md_reshape_position =
4589 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4590 reshape_position_store);
4593 reshape_direction_show(struct mddev *mddev, char *page)
4595 return sprintf(page, "%s\n",
4596 mddev->reshape_backwards ? "backwards" : "forwards");
4600 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4603 if (cmd_match(buf, "forwards"))
4605 else if (cmd_match(buf, "backwards"))
4609 if (mddev->reshape_backwards == backwards)
4612 /* check if we are allowed to change */
4613 if (mddev->delta_disks)
4616 if (mddev->persistent &&
4617 mddev->major_version == 0)
4620 mddev->reshape_backwards = backwards;
4624 static struct md_sysfs_entry md_reshape_direction =
4625 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4626 reshape_direction_store);
4629 array_size_show(struct mddev *mddev, char *page)
4631 if (mddev->external_size)
4632 return sprintf(page, "%llu\n",
4633 (unsigned long long)mddev->array_sectors/2);
4635 return sprintf(page, "default\n");
4639 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4643 if (strncmp(buf, "default", 7) == 0) {
4645 sectors = mddev->pers->size(mddev, 0, 0);
4647 sectors = mddev->array_sectors;
4649 mddev->external_size = 0;
4651 if (strict_blocks_to_sectors(buf, §ors) < 0)
4653 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4656 mddev->external_size = 1;
4659 mddev->array_sectors = sectors;
4661 set_capacity(mddev->gendisk, mddev->array_sectors);
4662 revalidate_disk(mddev->gendisk);
4667 static struct md_sysfs_entry md_array_size =
4668 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4671 static struct attribute *md_default_attrs[] = {
4674 &md_raid_disks.attr,
4675 &md_chunk_size.attr,
4677 &md_resync_start.attr,
4679 &md_new_device.attr,
4680 &md_safe_delay.attr,
4681 &md_array_state.attr,
4682 &md_reshape_position.attr,
4683 &md_reshape_direction.attr,
4684 &md_array_size.attr,
4685 &max_corr_read_errors.attr,
4689 static struct attribute *md_redundancy_attrs[] = {
4691 &md_mismatches.attr,
4694 &md_sync_speed.attr,
4695 &md_sync_force_parallel.attr,
4696 &md_sync_completed.attr,
4699 &md_suspend_lo.attr,
4700 &md_suspend_hi.attr,
4705 static struct attribute_group md_redundancy_group = {
4707 .attrs = md_redundancy_attrs,
4712 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4714 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4715 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4720 spin_lock(&all_mddevs_lock);
4721 if (list_empty(&mddev->all_mddevs)) {
4722 spin_unlock(&all_mddevs_lock);
4726 spin_unlock(&all_mddevs_lock);
4728 rv = mddev_lock(mddev);
4730 rv = entry->show(mddev, page);
4731 mddev_unlock(mddev);
4738 md_attr_store(struct kobject *kobj, struct attribute *attr,
4739 const char *page, size_t length)
4741 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4742 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4747 if (!capable(CAP_SYS_ADMIN))
4749 spin_lock(&all_mddevs_lock);
4750 if (list_empty(&mddev->all_mddevs)) {
4751 spin_unlock(&all_mddevs_lock);
4755 spin_unlock(&all_mddevs_lock);
4756 rv = mddev_lock(mddev);
4758 rv = entry->store(mddev, page, length);
4759 mddev_unlock(mddev);
4765 static void md_free(struct kobject *ko)
4767 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4769 if (mddev->sysfs_state)
4770 sysfs_put(mddev->sysfs_state);
4772 if (mddev->gendisk) {
4773 del_gendisk(mddev->gendisk);
4774 put_disk(mddev->gendisk);
4777 blk_cleanup_queue(mddev->queue);
4782 static const struct sysfs_ops md_sysfs_ops = {
4783 .show = md_attr_show,
4784 .store = md_attr_store,
4786 static struct kobj_type md_ktype = {
4788 .sysfs_ops = &md_sysfs_ops,
4789 .default_attrs = md_default_attrs,
4794 static void mddev_delayed_delete(struct work_struct *ws)
4796 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4798 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4799 kobject_del(&mddev->kobj);
4800 kobject_put(&mddev->kobj);
4803 static int md_alloc(dev_t dev, char *name)
4805 static DEFINE_MUTEX(disks_mutex);
4806 struct mddev *mddev = mddev_find(dev);
4807 struct gendisk *disk;
4816 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4817 shift = partitioned ? MdpMinorShift : 0;
4818 unit = MINOR(mddev->unit) >> shift;
4820 /* wait for any previous instance of this device to be
4821 * completely removed (mddev_delayed_delete).
4823 flush_workqueue(md_misc_wq);
4825 mutex_lock(&disks_mutex);
4831 /* Need to ensure that 'name' is not a duplicate.
4833 struct mddev *mddev2;
4834 spin_lock(&all_mddevs_lock);
4836 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4837 if (mddev2->gendisk &&
4838 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4839 spin_unlock(&all_mddevs_lock);
4842 spin_unlock(&all_mddevs_lock);
4846 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4849 mddev->queue->queuedata = mddev;
4851 blk_queue_make_request(mddev->queue, md_make_request);
4852 blk_set_stacking_limits(&mddev->queue->limits);
4854 disk = alloc_disk(1 << shift);
4856 blk_cleanup_queue(mddev->queue);
4857 mddev->queue = NULL;
4860 disk->major = MAJOR(mddev->unit);
4861 disk->first_minor = unit << shift;
4863 strcpy(disk->disk_name, name);
4864 else if (partitioned)
4865 sprintf(disk->disk_name, "md_d%d", unit);
4867 sprintf(disk->disk_name, "md%d", unit);
4868 disk->fops = &md_fops;
4869 disk->private_data = mddev;
4870 disk->queue = mddev->queue;
4871 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4872 /* Allow extended partitions. This makes the
4873 * 'mdp' device redundant, but we can't really
4876 disk->flags |= GENHD_FL_EXT_DEVT;
4877 mddev->gendisk = disk;
4878 /* As soon as we call add_disk(), another thread could get
4879 * through to md_open, so make sure it doesn't get too far
4881 mutex_lock(&mddev->open_mutex);
4884 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4885 &disk_to_dev(disk)->kobj, "%s", "md");
4887 /* This isn't possible, but as kobject_init_and_add is marked
4888 * __must_check, we must do something with the result
4890 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4894 if (mddev->kobj.sd &&
4895 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4896 printk(KERN_DEBUG "pointless warning\n");
4897 mutex_unlock(&mddev->open_mutex);
4899 mutex_unlock(&disks_mutex);
4900 if (!error && mddev->kobj.sd) {
4901 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4902 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4908 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4910 md_alloc(dev, NULL);
4914 static int add_named_array(const char *val, struct kernel_param *kp)
4916 /* val must be "md_*" where * is not all digits.
4917 * We allocate an array with a large free minor number, and
4918 * set the name to val. val must not already be an active name.
4920 int len = strlen(val);
4921 char buf[DISK_NAME_LEN];
4923 while (len && val[len-1] == '\n')
4925 if (len >= DISK_NAME_LEN)
4927 strlcpy(buf, val, len+1);
4928 if (strncmp(buf, "md_", 3) != 0)
4930 return md_alloc(0, buf);
4933 static void md_safemode_timeout(unsigned long data)
4935 struct mddev *mddev = (struct mddev *) data;
4937 if (!atomic_read(&mddev->writes_pending)) {
4938 mddev->safemode = 1;
4939 if (mddev->external)
4940 sysfs_notify_dirent_safe(mddev->sysfs_state);
4942 md_wakeup_thread(mddev->thread);
4945 static int start_dirty_degraded;
4947 int md_run(struct mddev *mddev)
4950 struct md_rdev *rdev;
4951 struct md_personality *pers;
4953 if (list_empty(&mddev->disks))
4954 /* cannot run an array with no devices.. */
4959 /* Cannot run until previous stop completes properly */
4960 if (mddev->sysfs_active)
4964 * Analyze all RAID superblock(s)
4966 if (!mddev->raid_disks) {
4967 if (!mddev->persistent)
4972 if (mddev->level != LEVEL_NONE)
4973 request_module("md-level-%d", mddev->level);
4974 else if (mddev->clevel[0])
4975 request_module("md-%s", mddev->clevel);
4978 * Drop all container device buffers, from now on
4979 * the only valid external interface is through the md
4982 rdev_for_each(rdev, mddev) {
4983 if (test_bit(Faulty, &rdev->flags))
4985 sync_blockdev(rdev->bdev);
4986 invalidate_bdev(rdev->bdev);
4988 /* perform some consistency tests on the device.
4989 * We don't want the data to overlap the metadata,
4990 * Internal Bitmap issues have been handled elsewhere.
4992 if (rdev->meta_bdev) {
4993 /* Nothing to check */;
4994 } else if (rdev->data_offset < rdev->sb_start) {
4995 if (mddev->dev_sectors &&
4996 rdev->data_offset + mddev->dev_sectors
4998 printk("md: %s: data overlaps metadata\n",
5003 if (rdev->sb_start + rdev->sb_size/512
5004 > rdev->data_offset) {
5005 printk("md: %s: metadata overlaps data\n",
5010 sysfs_notify_dirent_safe(rdev->sysfs_state);
5013 if (mddev->bio_set == NULL)
5014 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5016 spin_lock(&pers_lock);
5017 pers = find_pers(mddev->level, mddev->clevel);
5018 if (!pers || !try_module_get(pers->owner)) {
5019 spin_unlock(&pers_lock);
5020 if (mddev->level != LEVEL_NONE)
5021 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5024 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5029 spin_unlock(&pers_lock);
5030 if (mddev->level != pers->level) {
5031 mddev->level = pers->level;
5032 mddev->new_level = pers->level;
5034 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5036 if (mddev->reshape_position != MaxSector &&
5037 pers->start_reshape == NULL) {
5038 /* This personality cannot handle reshaping... */
5040 module_put(pers->owner);
5044 if (pers->sync_request) {
5045 /* Warn if this is a potentially silly
5048 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5049 struct md_rdev *rdev2;
5052 rdev_for_each(rdev, mddev)
5053 rdev_for_each(rdev2, mddev) {
5055 rdev->bdev->bd_contains ==
5056 rdev2->bdev->bd_contains) {
5058 "%s: WARNING: %s appears to be"
5059 " on the same physical disk as"
5062 bdevname(rdev->bdev,b),
5063 bdevname(rdev2->bdev,b2));
5070 "True protection against single-disk"
5071 " failure might be compromised.\n");
5074 mddev->recovery = 0;
5075 /* may be over-ridden by personality */
5076 mddev->resync_max_sectors = mddev->dev_sectors;
5078 mddev->ok_start_degraded = start_dirty_degraded;
5080 if (start_readonly && mddev->ro == 0)
5081 mddev->ro = 2; /* read-only, but switch on first write */
5083 err = mddev->pers->run(mddev);
5085 printk(KERN_ERR "md: pers->run() failed ...\n");
5086 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5087 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5088 " but 'external_size' not in effect?\n", __func__);
5090 "md: invalid array_size %llu > default size %llu\n",
5091 (unsigned long long)mddev->array_sectors / 2,
5092 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5094 mddev->pers->stop(mddev);
5096 if (err == 0 && mddev->pers->sync_request &&
5097 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5098 err = bitmap_create(mddev);
5100 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5101 mdname(mddev), err);
5102 mddev->pers->stop(mddev);
5106 module_put(mddev->pers->owner);
5108 bitmap_destroy(mddev);
5111 if (mddev->pers->sync_request) {
5112 if (mddev->kobj.sd &&
5113 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5115 "md: cannot register extra attributes for %s\n",
5117 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5118 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5121 atomic_set(&mddev->writes_pending,0);
5122 atomic_set(&mddev->max_corr_read_errors,
5123 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5124 mddev->safemode = 0;
5125 mddev->safemode_timer.function = md_safemode_timeout;
5126 mddev->safemode_timer.data = (unsigned long) mddev;
5127 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5131 rdev_for_each(rdev, mddev)
5132 if (rdev->raid_disk >= 0)
5133 if (sysfs_link_rdev(mddev, rdev))
5134 /* failure here is OK */;
5136 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5139 md_update_sb(mddev, 0);
5141 md_new_event(mddev);
5142 sysfs_notify_dirent_safe(mddev->sysfs_state);
5143 sysfs_notify_dirent_safe(mddev->sysfs_action);
5144 sysfs_notify(&mddev->kobj, NULL, "degraded");
5147 EXPORT_SYMBOL_GPL(md_run);
5149 static int do_md_run(struct mddev *mddev)
5153 err = md_run(mddev);
5156 err = bitmap_load(mddev);
5158 bitmap_destroy(mddev);
5162 md_wakeup_thread(mddev->thread);
5163 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5165 set_capacity(mddev->gendisk, mddev->array_sectors);
5166 revalidate_disk(mddev->gendisk);
5168 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5173 static int restart_array(struct mddev *mddev)
5175 struct gendisk *disk = mddev->gendisk;
5177 /* Complain if it has no devices */
5178 if (list_empty(&mddev->disks))
5184 mddev->safemode = 0;
5186 set_disk_ro(disk, 0);
5187 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5189 /* Kick recovery or resync if necessary */
5190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5191 md_wakeup_thread(mddev->thread);
5192 md_wakeup_thread(mddev->sync_thread);
5193 sysfs_notify_dirent_safe(mddev->sysfs_state);
5197 /* similar to deny_write_access, but accounts for our holding a reference
5198 * to the file ourselves */
5199 static int deny_bitmap_write_access(struct file * file)
5201 struct inode *inode = file->f_mapping->host;
5203 spin_lock(&inode->i_lock);
5204 if (atomic_read(&inode->i_writecount) > 1) {
5205 spin_unlock(&inode->i_lock);
5208 atomic_set(&inode->i_writecount, -1);
5209 spin_unlock(&inode->i_lock);
5214 void restore_bitmap_write_access(struct file *file)
5216 struct inode *inode = file->f_mapping->host;
5218 spin_lock(&inode->i_lock);
5219 atomic_set(&inode->i_writecount, 1);
5220 spin_unlock(&inode->i_lock);
5223 static void md_clean(struct mddev *mddev)
5225 mddev->array_sectors = 0;
5226 mddev->external_size = 0;
5227 mddev->dev_sectors = 0;
5228 mddev->raid_disks = 0;
5229 mddev->recovery_cp = 0;
5230 mddev->resync_min = 0;
5231 mddev->resync_max = MaxSector;
5232 mddev->reshape_position = MaxSector;
5233 mddev->external = 0;
5234 mddev->persistent = 0;
5235 mddev->level = LEVEL_NONE;
5236 mddev->clevel[0] = 0;
5239 mddev->metadata_type[0] = 0;
5240 mddev->chunk_sectors = 0;
5241 mddev->ctime = mddev->utime = 0;
5243 mddev->max_disks = 0;
5245 mddev->can_decrease_events = 0;
5246 mddev->delta_disks = 0;
5247 mddev->reshape_backwards = 0;
5248 mddev->new_level = LEVEL_NONE;
5249 mddev->new_layout = 0;
5250 mddev->new_chunk_sectors = 0;
5251 mddev->curr_resync = 0;
5252 atomic64_set(&mddev->resync_mismatches, 0);
5253 mddev->suspend_lo = mddev->suspend_hi = 0;
5254 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5255 mddev->recovery = 0;
5258 mddev->degraded = 0;
5259 mddev->safemode = 0;
5260 mddev->merge_check_needed = 0;
5261 mddev->bitmap_info.offset = 0;
5262 mddev->bitmap_info.default_offset = 0;
5263 mddev->bitmap_info.default_space = 0;
5264 mddev->bitmap_info.chunksize = 0;
5265 mddev->bitmap_info.daemon_sleep = 0;
5266 mddev->bitmap_info.max_write_behind = 0;
5269 static void __md_stop_writes(struct mddev *mddev)
5271 if (mddev->sync_thread) {
5272 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5273 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5274 reap_sync_thread(mddev);
5277 del_timer_sync(&mddev->safemode_timer);
5279 bitmap_flush(mddev);
5280 md_super_wait(mddev);
5282 if (!mddev->in_sync || mddev->flags) {
5283 /* mark array as shutdown cleanly */
5285 md_update_sb(mddev, 1);
5289 void md_stop_writes(struct mddev *mddev)
5292 __md_stop_writes(mddev);
5293 mddev_unlock(mddev);
5295 EXPORT_SYMBOL_GPL(md_stop_writes);
5297 void md_stop(struct mddev *mddev)
5300 mddev->pers->stop(mddev);
5301 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5302 mddev->to_remove = &md_redundancy_group;
5303 module_put(mddev->pers->owner);
5305 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5307 EXPORT_SYMBOL_GPL(md_stop);
5309 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5312 mutex_lock(&mddev->open_mutex);
5313 if (atomic_read(&mddev->openers) > !!bdev) {
5314 printk("md: %s still in use.\n",mdname(mddev));
5319 sync_blockdev(bdev);
5321 __md_stop_writes(mddev);
5327 set_disk_ro(mddev->gendisk, 1);
5328 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5329 sysfs_notify_dirent_safe(mddev->sysfs_state);
5333 mutex_unlock(&mddev->open_mutex);
5338 * 0 - completely stop and dis-assemble array
5339 * 2 - stop but do not disassemble array
5341 static int do_md_stop(struct mddev * mddev, int mode,
5342 struct block_device *bdev)
5344 struct gendisk *disk = mddev->gendisk;
5345 struct md_rdev *rdev;
5347 mutex_lock(&mddev->open_mutex);
5348 if (atomic_read(&mddev->openers) > !!bdev ||
5349 mddev->sysfs_active) {
5350 printk("md: %s still in use.\n",mdname(mddev));
5351 mutex_unlock(&mddev->open_mutex);
5355 /* It is possible IO was issued on some other
5356 * open file which was closed before we took ->open_mutex.
5357 * As that was not the last close __blkdev_put will not
5358 * have called sync_blockdev, so we must.
5360 sync_blockdev(bdev);
5364 set_disk_ro(disk, 0);
5366 __md_stop_writes(mddev);
5368 mddev->queue->merge_bvec_fn = NULL;
5369 mddev->queue->backing_dev_info.congested_fn = NULL;
5371 /* tell userspace to handle 'inactive' */
5372 sysfs_notify_dirent_safe(mddev->sysfs_state);
5374 rdev_for_each(rdev, mddev)
5375 if (rdev->raid_disk >= 0)
5376 sysfs_unlink_rdev(mddev, rdev);
5378 set_capacity(disk, 0);
5379 mutex_unlock(&mddev->open_mutex);
5381 revalidate_disk(disk);
5386 mutex_unlock(&mddev->open_mutex);
5388 * Free resources if final stop
5391 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5393 bitmap_destroy(mddev);
5394 if (mddev->bitmap_info.file) {
5395 restore_bitmap_write_access(mddev->bitmap_info.file);
5396 fput(mddev->bitmap_info.file);
5397 mddev->bitmap_info.file = NULL;
5399 mddev->bitmap_info.offset = 0;
5401 export_array(mddev);
5404 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5405 if (mddev->hold_active == UNTIL_STOP)
5406 mddev->hold_active = 0;
5408 blk_integrity_unregister(disk);
5409 md_new_event(mddev);
5410 sysfs_notify_dirent_safe(mddev->sysfs_state);
5415 static void autorun_array(struct mddev *mddev)
5417 struct md_rdev *rdev;
5420 if (list_empty(&mddev->disks))
5423 printk(KERN_INFO "md: running: ");
5425 rdev_for_each(rdev, mddev) {
5426 char b[BDEVNAME_SIZE];
5427 printk("<%s>", bdevname(rdev->bdev,b));
5431 err = do_md_run(mddev);
5433 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5434 do_md_stop(mddev, 0, NULL);
5439 * lets try to run arrays based on all disks that have arrived
5440 * until now. (those are in pending_raid_disks)
5442 * the method: pick the first pending disk, collect all disks with
5443 * the same UUID, remove all from the pending list and put them into
5444 * the 'same_array' list. Then order this list based on superblock
5445 * update time (freshest comes first), kick out 'old' disks and
5446 * compare superblocks. If everything's fine then run it.
5448 * If "unit" is allocated, then bump its reference count
5450 static void autorun_devices(int part)
5452 struct md_rdev *rdev0, *rdev, *tmp;
5453 struct mddev *mddev;
5454 char b[BDEVNAME_SIZE];
5456 printk(KERN_INFO "md: autorun ...\n");
5457 while (!list_empty(&pending_raid_disks)) {
5460 LIST_HEAD(candidates);
5461 rdev0 = list_entry(pending_raid_disks.next,
5462 struct md_rdev, same_set);
5464 printk(KERN_INFO "md: considering %s ...\n",
5465 bdevname(rdev0->bdev,b));
5466 INIT_LIST_HEAD(&candidates);
5467 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5468 if (super_90_load(rdev, rdev0, 0) >= 0) {
5469 printk(KERN_INFO "md: adding %s ...\n",
5470 bdevname(rdev->bdev,b));
5471 list_move(&rdev->same_set, &candidates);
5474 * now we have a set of devices, with all of them having
5475 * mostly sane superblocks. It's time to allocate the
5479 dev = MKDEV(mdp_major,
5480 rdev0->preferred_minor << MdpMinorShift);
5481 unit = MINOR(dev) >> MdpMinorShift;
5483 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5486 if (rdev0->preferred_minor != unit) {
5487 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5488 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5492 md_probe(dev, NULL, NULL);
5493 mddev = mddev_find(dev);
5494 if (!mddev || !mddev->gendisk) {
5498 "md: cannot allocate memory for md drive.\n");
5501 if (mddev_lock(mddev))
5502 printk(KERN_WARNING "md: %s locked, cannot run\n",
5504 else if (mddev->raid_disks || mddev->major_version
5505 || !list_empty(&mddev->disks)) {
5507 "md: %s already running, cannot run %s\n",
5508 mdname(mddev), bdevname(rdev0->bdev,b));
5509 mddev_unlock(mddev);
5511 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5512 mddev->persistent = 1;
5513 rdev_for_each_list(rdev, tmp, &candidates) {
5514 list_del_init(&rdev->same_set);
5515 if (bind_rdev_to_array(rdev, mddev))
5518 autorun_array(mddev);
5519 mddev_unlock(mddev);
5521 /* on success, candidates will be empty, on error
5524 rdev_for_each_list(rdev, tmp, &candidates) {
5525 list_del_init(&rdev->same_set);
5530 printk(KERN_INFO "md: ... autorun DONE.\n");
5532 #endif /* !MODULE */
5534 static int get_version(void __user * arg)
5538 ver.major = MD_MAJOR_VERSION;
5539 ver.minor = MD_MINOR_VERSION;
5540 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5542 if (copy_to_user(arg, &ver, sizeof(ver)))
5548 static int get_array_info(struct mddev * mddev, void __user * arg)
5550 mdu_array_info_t info;
5551 int nr,working,insync,failed,spare;
5552 struct md_rdev *rdev;
5554 nr = working = insync = failed = spare = 0;
5556 rdev_for_each_rcu(rdev, mddev) {
5558 if (test_bit(Faulty, &rdev->flags))
5562 if (test_bit(In_sync, &rdev->flags))
5570 info.major_version = mddev->major_version;
5571 info.minor_version = mddev->minor_version;
5572 info.patch_version = MD_PATCHLEVEL_VERSION;
5573 info.ctime = mddev->ctime;
5574 info.level = mddev->level;
5575 info.size = mddev->dev_sectors / 2;
5576 if (info.size != mddev->dev_sectors / 2) /* overflow */
5579 info.raid_disks = mddev->raid_disks;
5580 info.md_minor = mddev->md_minor;
5581 info.not_persistent= !mddev->persistent;
5583 info.utime = mddev->utime;
5586 info.state = (1<<MD_SB_CLEAN);
5587 if (mddev->bitmap && mddev->bitmap_info.offset)
5588 info.state = (1<<MD_SB_BITMAP_PRESENT);
5589 info.active_disks = insync;
5590 info.working_disks = working;
5591 info.failed_disks = failed;
5592 info.spare_disks = spare;
5594 info.layout = mddev->layout;
5595 info.chunk_size = mddev->chunk_sectors << 9;
5597 if (copy_to_user(arg, &info, sizeof(info)))
5603 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5605 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5606 char *ptr, *buf = NULL;
5609 if (md_allow_write(mddev))
5610 file = kmalloc(sizeof(*file), GFP_NOIO);
5612 file = kmalloc(sizeof(*file), GFP_KERNEL);
5617 /* bitmap disabled, zero the first byte and copy out */
5618 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5619 file->pathname[0] = '\0';
5623 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5627 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5628 buf, sizeof(file->pathname));
5632 strcpy(file->pathname, ptr);
5636 if (copy_to_user(arg, file, sizeof(*file)))
5644 static int get_disk_info(struct mddev * mddev, void __user * arg)
5646 mdu_disk_info_t info;
5647 struct md_rdev *rdev;
5649 if (copy_from_user(&info, arg, sizeof(info)))
5653 rdev = find_rdev_nr_rcu(mddev, info.number);
5655 info.major = MAJOR(rdev->bdev->bd_dev);
5656 info.minor = MINOR(rdev->bdev->bd_dev);
5657 info.raid_disk = rdev->raid_disk;
5659 if (test_bit(Faulty, &rdev->flags))
5660 info.state |= (1<<MD_DISK_FAULTY);
5661 else if (test_bit(In_sync, &rdev->flags)) {
5662 info.state |= (1<<MD_DISK_ACTIVE);
5663 info.state |= (1<<MD_DISK_SYNC);
5665 if (test_bit(WriteMostly, &rdev->flags))
5666 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5668 info.major = info.minor = 0;
5669 info.raid_disk = -1;
5670 info.state = (1<<MD_DISK_REMOVED);
5674 if (copy_to_user(arg, &info, sizeof(info)))
5680 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5682 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5683 struct md_rdev *rdev;
5684 dev_t dev = MKDEV(info->major,info->minor);
5686 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5689 if (!mddev->raid_disks) {
5691 /* expecting a device which has a superblock */
5692 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5695 "md: md_import_device returned %ld\n",
5697 return PTR_ERR(rdev);
5699 if (!list_empty(&mddev->disks)) {
5700 struct md_rdev *rdev0
5701 = list_entry(mddev->disks.next,
5702 struct md_rdev, same_set);
5703 err = super_types[mddev->major_version]
5704 .load_super(rdev, rdev0, mddev->minor_version);
5707 "md: %s has different UUID to %s\n",
5708 bdevname(rdev->bdev,b),
5709 bdevname(rdev0->bdev,b2));
5714 err = bind_rdev_to_array(rdev, mddev);
5721 * add_new_disk can be used once the array is assembled
5722 * to add "hot spares". They must already have a superblock
5727 if (!mddev->pers->hot_add_disk) {
5729 "%s: personality does not support diskops!\n",
5733 if (mddev->persistent)
5734 rdev = md_import_device(dev, mddev->major_version,
5735 mddev->minor_version);
5737 rdev = md_import_device(dev, -1, -1);
5740 "md: md_import_device returned %ld\n",
5742 return PTR_ERR(rdev);
5744 /* set saved_raid_disk if appropriate */
5745 if (!mddev->persistent) {
5746 if (info->state & (1<<MD_DISK_SYNC) &&
5747 info->raid_disk < mddev->raid_disks) {
5748 rdev->raid_disk = info->raid_disk;
5749 set_bit(In_sync, &rdev->flags);
5751 rdev->raid_disk = -1;
5753 super_types[mddev->major_version].
5754 validate_super(mddev, rdev);
5755 if ((info->state & (1<<MD_DISK_SYNC)) &&
5756 rdev->raid_disk != info->raid_disk) {
5757 /* This was a hot-add request, but events doesn't
5758 * match, so reject it.
5764 if (test_bit(In_sync, &rdev->flags))
5765 rdev->saved_raid_disk = rdev->raid_disk;
5767 rdev->saved_raid_disk = -1;
5769 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5770 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5771 set_bit(WriteMostly, &rdev->flags);
5773 clear_bit(WriteMostly, &rdev->flags);
5775 rdev->raid_disk = -1;
5776 err = bind_rdev_to_array(rdev, mddev);
5777 if (!err && !mddev->pers->hot_remove_disk) {
5778 /* If there is hot_add_disk but no hot_remove_disk
5779 * then added disks for geometry changes,
5780 * and should be added immediately.
5782 super_types[mddev->major_version].
5783 validate_super(mddev, rdev);
5784 err = mddev->pers->hot_add_disk(mddev, rdev);
5786 unbind_rdev_from_array(rdev);
5791 sysfs_notify_dirent_safe(rdev->sysfs_state);
5793 md_update_sb(mddev, 1);
5794 if (mddev->degraded)
5795 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5796 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5798 md_new_event(mddev);
5799 md_wakeup_thread(mddev->thread);
5803 /* otherwise, add_new_disk is only allowed
5804 * for major_version==0 superblocks
5806 if (mddev->major_version != 0) {
5807 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5812 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5814 rdev = md_import_device(dev, -1, 0);
5817 "md: error, md_import_device() returned %ld\n",
5819 return PTR_ERR(rdev);
5821 rdev->desc_nr = info->number;
5822 if (info->raid_disk < mddev->raid_disks)
5823 rdev->raid_disk = info->raid_disk;
5825 rdev->raid_disk = -1;
5827 if (rdev->raid_disk < mddev->raid_disks)
5828 if (info->state & (1<<MD_DISK_SYNC))
5829 set_bit(In_sync, &rdev->flags);
5831 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5832 set_bit(WriteMostly, &rdev->flags);
5834 if (!mddev->persistent) {
5835 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5836 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5838 rdev->sb_start = calc_dev_sboffset(rdev);
5839 rdev->sectors = rdev->sb_start;
5841 err = bind_rdev_to_array(rdev, mddev);
5851 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5853 char b[BDEVNAME_SIZE];
5854 struct md_rdev *rdev;
5856 rdev = find_rdev(mddev, dev);
5860 if (rdev->raid_disk >= 0)
5863 kick_rdev_from_array(rdev);
5864 md_update_sb(mddev, 1);
5865 md_new_event(mddev);
5869 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5870 bdevname(rdev->bdev,b), mdname(mddev));
5874 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5876 char b[BDEVNAME_SIZE];
5878 struct md_rdev *rdev;
5883 if (mddev->major_version != 0) {
5884 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5885 " version-0 superblocks.\n",
5889 if (!mddev->pers->hot_add_disk) {
5891 "%s: personality does not support diskops!\n",
5896 rdev = md_import_device(dev, -1, 0);
5899 "md: error, md_import_device() returned %ld\n",
5904 if (mddev->persistent)
5905 rdev->sb_start = calc_dev_sboffset(rdev);
5907 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5909 rdev->sectors = rdev->sb_start;
5911 if (test_bit(Faulty, &rdev->flags)) {
5913 "md: can not hot-add faulty %s disk to %s!\n",
5914 bdevname(rdev->bdev,b), mdname(mddev));
5918 clear_bit(In_sync, &rdev->flags);
5920 rdev->saved_raid_disk = -1;
5921 err = bind_rdev_to_array(rdev, mddev);
5926 * The rest should better be atomic, we can have disk failures
5927 * noticed in interrupt contexts ...
5930 rdev->raid_disk = -1;
5932 md_update_sb(mddev, 1);
5935 * Kick recovery, maybe this spare has to be added to the
5936 * array immediately.
5938 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5939 md_wakeup_thread(mddev->thread);
5940 md_new_event(mddev);
5948 static int set_bitmap_file(struct mddev *mddev, int fd)
5953 if (!mddev->pers->quiesce)
5955 if (mddev->recovery || mddev->sync_thread)
5957 /* we should be able to change the bitmap.. */
5963 return -EEXIST; /* cannot add when bitmap is present */
5964 mddev->bitmap_info.file = fget(fd);
5966 if (mddev->bitmap_info.file == NULL) {
5967 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5972 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5974 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5976 fput(mddev->bitmap_info.file);
5977 mddev->bitmap_info.file = NULL;
5980 mddev->bitmap_info.offset = 0; /* file overrides offset */
5981 } else if (mddev->bitmap == NULL)
5982 return -ENOENT; /* cannot remove what isn't there */
5985 mddev->pers->quiesce(mddev, 1);
5987 err = bitmap_create(mddev);
5989 err = bitmap_load(mddev);
5991 if (fd < 0 || err) {
5992 bitmap_destroy(mddev);
5993 fd = -1; /* make sure to put the file */
5995 mddev->pers->quiesce(mddev, 0);
5998 if (mddev->bitmap_info.file) {
5999 restore_bitmap_write_access(mddev->bitmap_info.file);
6000 fput(mddev->bitmap_info.file);
6002 mddev->bitmap_info.file = NULL;
6009 * set_array_info is used two different ways
6010 * The original usage is when creating a new array.
6011 * In this usage, raid_disks is > 0 and it together with
6012 * level, size, not_persistent,layout,chunksize determine the
6013 * shape of the array.
6014 * This will always create an array with a type-0.90.0 superblock.
6015 * The newer usage is when assembling an array.
6016 * In this case raid_disks will be 0, and the major_version field is
6017 * use to determine which style super-blocks are to be found on the devices.
6018 * The minor and patch _version numbers are also kept incase the
6019 * super_block handler wishes to interpret them.
6021 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6024 if (info->raid_disks == 0) {
6025 /* just setting version number for superblock loading */
6026 if (info->major_version < 0 ||
6027 info->major_version >= ARRAY_SIZE(super_types) ||
6028 super_types[info->major_version].name == NULL) {
6029 /* maybe try to auto-load a module? */
6031 "md: superblock version %d not known\n",
6032 info->major_version);
6035 mddev->major_version = info->major_version;
6036 mddev->minor_version = info->minor_version;
6037 mddev->patch_version = info->patch_version;
6038 mddev->persistent = !info->not_persistent;
6039 /* ensure mddev_put doesn't delete this now that there
6040 * is some minimal configuration.
6042 mddev->ctime = get_seconds();
6045 mddev->major_version = MD_MAJOR_VERSION;
6046 mddev->minor_version = MD_MINOR_VERSION;
6047 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6048 mddev->ctime = get_seconds();
6050 mddev->level = info->level;
6051 mddev->clevel[0] = 0;
6052 mddev->dev_sectors = 2 * (sector_t)info->size;
6053 mddev->raid_disks = info->raid_disks;
6054 /* don't set md_minor, it is determined by which /dev/md* was
6057 if (info->state & (1<<MD_SB_CLEAN))
6058 mddev->recovery_cp = MaxSector;
6060 mddev->recovery_cp = 0;
6061 mddev->persistent = ! info->not_persistent;
6062 mddev->external = 0;
6064 mddev->layout = info->layout;
6065 mddev->chunk_sectors = info->chunk_size >> 9;
6067 mddev->max_disks = MD_SB_DISKS;
6069 if (mddev->persistent)
6071 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6073 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6074 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6075 mddev->bitmap_info.offset = 0;
6077 mddev->reshape_position = MaxSector;
6080 * Generate a 128 bit UUID
6082 get_random_bytes(mddev->uuid, 16);
6084 mddev->new_level = mddev->level;
6085 mddev->new_chunk_sectors = mddev->chunk_sectors;
6086 mddev->new_layout = mddev->layout;
6087 mddev->delta_disks = 0;
6088 mddev->reshape_backwards = 0;
6093 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6095 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6097 if (mddev->external_size)
6100 mddev->array_sectors = array_sectors;
6102 EXPORT_SYMBOL(md_set_array_sectors);
6104 static int update_size(struct mddev *mddev, sector_t num_sectors)
6106 struct md_rdev *rdev;
6108 int fit = (num_sectors == 0);
6110 if (mddev->pers->resize == NULL)
6112 /* The "num_sectors" is the number of sectors of each device that
6113 * is used. This can only make sense for arrays with redundancy.
6114 * linear and raid0 always use whatever space is available. We can only
6115 * consider changing this number if no resync or reconstruction is
6116 * happening, and if the new size is acceptable. It must fit before the
6117 * sb_start or, if that is <data_offset, it must fit before the size
6118 * of each device. If num_sectors is zero, we find the largest size
6121 if (mddev->sync_thread)
6124 rdev_for_each(rdev, mddev) {
6125 sector_t avail = rdev->sectors;
6127 if (fit && (num_sectors == 0 || num_sectors > avail))
6128 num_sectors = avail;
6129 if (avail < num_sectors)
6132 rv = mddev->pers->resize(mddev, num_sectors);
6134 revalidate_disk(mddev->gendisk);
6138 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6141 struct md_rdev *rdev;
6142 /* change the number of raid disks */
6143 if (mddev->pers->check_reshape == NULL)
6145 if (raid_disks <= 0 ||
6146 (mddev->max_disks && raid_disks >= mddev->max_disks))
6148 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6151 rdev_for_each(rdev, mddev) {
6152 if (mddev->raid_disks < raid_disks &&
6153 rdev->data_offset < rdev->new_data_offset)
6155 if (mddev->raid_disks > raid_disks &&
6156 rdev->data_offset > rdev->new_data_offset)
6160 mddev->delta_disks = raid_disks - mddev->raid_disks;
6161 if (mddev->delta_disks < 0)
6162 mddev->reshape_backwards = 1;
6163 else if (mddev->delta_disks > 0)
6164 mddev->reshape_backwards = 0;
6166 rv = mddev->pers->check_reshape(mddev);
6168 mddev->delta_disks = 0;
6169 mddev->reshape_backwards = 0;
6176 * update_array_info is used to change the configuration of an
6178 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6179 * fields in the info are checked against the array.
6180 * Any differences that cannot be handled will cause an error.
6181 * Normally, only one change can be managed at a time.
6183 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6189 /* calculate expected state,ignoring low bits */
6190 if (mddev->bitmap && mddev->bitmap_info.offset)
6191 state |= (1 << MD_SB_BITMAP_PRESENT);
6193 if (mddev->major_version != info->major_version ||
6194 mddev->minor_version != info->minor_version ||
6195 /* mddev->patch_version != info->patch_version || */
6196 mddev->ctime != info->ctime ||
6197 mddev->level != info->level ||
6198 /* mddev->layout != info->layout || */
6199 !mddev->persistent != info->not_persistent||
6200 mddev->chunk_sectors != info->chunk_size >> 9 ||
6201 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6202 ((state^info->state) & 0xfffffe00)
6205 /* Check there is only one change */
6206 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6208 if (mddev->raid_disks != info->raid_disks)
6210 if (mddev->layout != info->layout)
6212 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6219 if (mddev->layout != info->layout) {
6221 * we don't need to do anything at the md level, the
6222 * personality will take care of it all.
6224 if (mddev->pers->check_reshape == NULL)
6227 mddev->new_layout = info->layout;
6228 rv = mddev->pers->check_reshape(mddev);
6230 mddev->new_layout = mddev->layout;
6234 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6235 rv = update_size(mddev, (sector_t)info->size * 2);
6237 if (mddev->raid_disks != info->raid_disks)
6238 rv = update_raid_disks(mddev, info->raid_disks);
6240 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6241 if (mddev->pers->quiesce == NULL)
6243 if (mddev->recovery || mddev->sync_thread)
6245 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6246 /* add the bitmap */
6249 if (mddev->bitmap_info.default_offset == 0)
6251 mddev->bitmap_info.offset =
6252 mddev->bitmap_info.default_offset;
6253 mddev->bitmap_info.space =
6254 mddev->bitmap_info.default_space;
6255 mddev->pers->quiesce(mddev, 1);
6256 rv = bitmap_create(mddev);
6258 rv = bitmap_load(mddev);
6260 bitmap_destroy(mddev);
6261 mddev->pers->quiesce(mddev, 0);
6263 /* remove the bitmap */
6266 if (mddev->bitmap->storage.file)
6268 mddev->pers->quiesce(mddev, 1);
6269 bitmap_destroy(mddev);
6270 mddev->pers->quiesce(mddev, 0);
6271 mddev->bitmap_info.offset = 0;
6274 md_update_sb(mddev, 1);
6278 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6280 struct md_rdev *rdev;
6283 if (mddev->pers == NULL)
6287 rdev = find_rdev_rcu(mddev, dev);
6291 md_error(mddev, rdev);
6292 if (!test_bit(Faulty, &rdev->flags))
6300 * We have a problem here : there is no easy way to give a CHS
6301 * virtual geometry. We currently pretend that we have a 2 heads
6302 * 4 sectors (with a BIG number of cylinders...). This drives
6303 * dosfs just mad... ;-)
6305 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6307 struct mddev *mddev = bdev->bd_disk->private_data;
6311 geo->cylinders = mddev->array_sectors / 8;
6315 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6316 unsigned int cmd, unsigned long arg)
6319 void __user *argp = (void __user *)arg;
6320 struct mddev *mddev = NULL;
6325 case GET_ARRAY_INFO:
6329 if (!capable(CAP_SYS_ADMIN))
6334 * Commands dealing with the RAID driver but not any
6340 err = get_version(argp);
6343 case PRINT_RAID_DEBUG:
6351 autostart_arrays(arg);
6358 * Commands creating/starting a new array:
6361 mddev = bdev->bd_disk->private_data;
6368 /* Some actions do not requires the mutex */
6370 case GET_ARRAY_INFO:
6371 if (!mddev->raid_disks && !mddev->external)
6374 err = get_array_info(mddev, argp);
6378 if (!mddev->raid_disks && !mddev->external)
6381 err = get_disk_info(mddev, argp);
6384 case SET_DISK_FAULTY:
6385 err = set_disk_faulty(mddev, new_decode_dev(arg));
6389 err = mddev_lock(mddev);
6392 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6399 case SET_ARRAY_INFO:
6401 mdu_array_info_t info;
6403 memset(&info, 0, sizeof(info));
6404 else if (copy_from_user(&info, argp, sizeof(info))) {
6409 err = update_array_info(mddev, &info);
6411 printk(KERN_WARNING "md: couldn't update"
6412 " array info. %d\n", err);
6417 if (!list_empty(&mddev->disks)) {
6419 "md: array %s already has disks!\n",
6424 if (mddev->raid_disks) {
6426 "md: array %s already initialised!\n",
6431 err = set_array_info(mddev, &info);
6433 printk(KERN_WARNING "md: couldn't set"
6434 " array info. %d\n", err);
6444 * Commands querying/configuring an existing array:
6446 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6447 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6448 if ((!mddev->raid_disks && !mddev->external)
6449 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6450 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6451 && cmd != GET_BITMAP_FILE) {
6457 * Commands even a read-only array can execute:
6461 case GET_BITMAP_FILE:
6462 err = get_bitmap_file(mddev, argp);
6465 case RESTART_ARRAY_RW:
6466 err = restart_array(mddev);
6470 err = do_md_stop(mddev, 0, bdev);
6474 err = md_set_readonly(mddev, bdev);
6478 if (get_user(ro, (int __user *)(arg))) {
6484 /* if the bdev is going readonly the value of mddev->ro
6485 * does not matter, no writes are coming
6490 /* are we are already prepared for writes? */
6494 /* transitioning to readauto need only happen for
6495 * arrays that call md_write_start
6498 err = restart_array(mddev);
6501 set_disk_ro(mddev->gendisk, 0);
6508 * The remaining ioctls are changing the state of the
6509 * superblock, so we do not allow them on read-only arrays.
6510 * However non-MD ioctls (e.g. get-size) will still come through
6511 * here and hit the 'default' below, so only disallow
6512 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6514 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6515 if (mddev->ro == 2) {
6517 sysfs_notify_dirent_safe(mddev->sysfs_state);
6518 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6519 md_wakeup_thread(mddev->thread);
6530 mdu_disk_info_t info;
6531 if (copy_from_user(&info, argp, sizeof(info)))
6534 err = add_new_disk(mddev, &info);
6538 case HOT_REMOVE_DISK:
6539 err = hot_remove_disk(mddev, new_decode_dev(arg));
6543 err = hot_add_disk(mddev, new_decode_dev(arg));
6547 err = do_md_run(mddev);
6550 case SET_BITMAP_FILE:
6551 err = set_bitmap_file(mddev, (int)arg);
6561 if (mddev->hold_active == UNTIL_IOCTL &&
6563 mddev->hold_active = 0;
6564 mddev_unlock(mddev);
6573 #ifdef CONFIG_COMPAT
6574 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6575 unsigned int cmd, unsigned long arg)
6578 case HOT_REMOVE_DISK:
6580 case SET_DISK_FAULTY:
6581 case SET_BITMAP_FILE:
6582 /* These take in integer arg, do not convert */
6585 arg = (unsigned long)compat_ptr(arg);
6589 return md_ioctl(bdev, mode, cmd, arg);
6591 #endif /* CONFIG_COMPAT */
6593 static int md_open(struct block_device *bdev, fmode_t mode)
6596 * Succeed if we can lock the mddev, which confirms that
6597 * it isn't being stopped right now.
6599 struct mddev *mddev = mddev_find(bdev->bd_dev);
6605 if (mddev->gendisk != bdev->bd_disk) {
6606 /* we are racing with mddev_put which is discarding this
6610 /* Wait until bdev->bd_disk is definitely gone */
6611 flush_workqueue(md_misc_wq);
6612 /* Then retry the open from the top */
6613 return -ERESTARTSYS;
6615 BUG_ON(mddev != bdev->bd_disk->private_data);
6617 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6621 atomic_inc(&mddev->openers);
6622 mutex_unlock(&mddev->open_mutex);
6624 check_disk_change(bdev);
6629 static int md_release(struct gendisk *disk, fmode_t mode)
6631 struct mddev *mddev = disk->private_data;
6634 atomic_dec(&mddev->openers);
6640 static int md_media_changed(struct gendisk *disk)
6642 struct mddev *mddev = disk->private_data;
6644 return mddev->changed;
6647 static int md_revalidate(struct gendisk *disk)
6649 struct mddev *mddev = disk->private_data;
6654 static const struct block_device_operations md_fops =
6656 .owner = THIS_MODULE,
6658 .release = md_release,
6660 #ifdef CONFIG_COMPAT
6661 .compat_ioctl = md_compat_ioctl,
6663 .getgeo = md_getgeo,
6664 .media_changed = md_media_changed,
6665 .revalidate_disk= md_revalidate,
6668 static int md_thread(void * arg)
6670 struct md_thread *thread = arg;
6673 * md_thread is a 'system-thread', it's priority should be very
6674 * high. We avoid resource deadlocks individually in each
6675 * raid personality. (RAID5 does preallocation) We also use RR and
6676 * the very same RT priority as kswapd, thus we will never get
6677 * into a priority inversion deadlock.
6679 * we definitely have to have equal or higher priority than
6680 * bdflush, otherwise bdflush will deadlock if there are too
6681 * many dirty RAID5 blocks.
6684 allow_signal(SIGKILL);
6685 while (!kthread_should_stop()) {
6687 /* We need to wait INTERRUPTIBLE so that
6688 * we don't add to the load-average.
6689 * That means we need to be sure no signals are
6692 if (signal_pending(current))
6693 flush_signals(current);
6695 wait_event_interruptible_timeout
6697 test_bit(THREAD_WAKEUP, &thread->flags)
6698 || kthread_should_stop(),
6701 clear_bit(THREAD_WAKEUP, &thread->flags);
6702 if (!kthread_should_stop())
6703 thread->run(thread);
6709 void md_wakeup_thread(struct md_thread *thread)
6712 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6713 set_bit(THREAD_WAKEUP, &thread->flags);
6714 wake_up(&thread->wqueue);
6718 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6719 struct mddev *mddev, const char *name)
6721 struct md_thread *thread;
6723 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6727 init_waitqueue_head(&thread->wqueue);
6730 thread->mddev = mddev;
6731 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6732 thread->tsk = kthread_run(md_thread, thread,
6734 mdname(thread->mddev),
6736 if (IS_ERR(thread->tsk)) {
6743 void md_unregister_thread(struct md_thread **threadp)
6745 struct md_thread *thread = *threadp;
6748 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6749 /* Locking ensures that mddev_unlock does not wake_up a
6750 * non-existent thread
6752 spin_lock(&pers_lock);
6754 spin_unlock(&pers_lock);
6756 kthread_stop(thread->tsk);
6760 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6767 if (!rdev || test_bit(Faulty, &rdev->flags))
6770 if (!mddev->pers || !mddev->pers->error_handler)
6772 mddev->pers->error_handler(mddev,rdev);
6773 if (mddev->degraded)
6774 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6775 sysfs_notify_dirent_safe(rdev->sysfs_state);
6776 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6777 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6778 md_wakeup_thread(mddev->thread);
6779 if (mddev->event_work.func)
6780 queue_work(md_misc_wq, &mddev->event_work);
6781 md_new_event_inintr(mddev);
6784 /* seq_file implementation /proc/mdstat */
6786 static void status_unused(struct seq_file *seq)
6789 struct md_rdev *rdev;
6791 seq_printf(seq, "unused devices: ");
6793 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6794 char b[BDEVNAME_SIZE];
6796 seq_printf(seq, "%s ",
6797 bdevname(rdev->bdev,b));
6800 seq_printf(seq, "<none>");
6802 seq_printf(seq, "\n");
6806 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6808 sector_t max_sectors, resync, res;
6809 unsigned long dt, db;
6812 unsigned int per_milli;
6814 if (mddev->curr_resync <= 3)
6817 resync = mddev->curr_resync
6818 - atomic_read(&mddev->recovery_active);
6820 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6821 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6822 max_sectors = mddev->resync_max_sectors;
6824 max_sectors = mddev->dev_sectors;
6827 * Should not happen.
6833 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6834 * in a sector_t, and (max_sectors>>scale) will fit in a
6835 * u32, as those are the requirements for sector_div.
6836 * Thus 'scale' must be at least 10
6839 if (sizeof(sector_t) > sizeof(unsigned long)) {
6840 while ( max_sectors/2 > (1ULL<<(scale+32)))
6843 res = (resync>>scale)*1000;
6844 sector_div(res, (u32)((max_sectors>>scale)+1));
6848 int i, x = per_milli/50, y = 20-x;
6849 seq_printf(seq, "[");
6850 for (i = 0; i < x; i++)
6851 seq_printf(seq, "=");
6852 seq_printf(seq, ">");
6853 for (i = 0; i < y; i++)
6854 seq_printf(seq, ".");
6855 seq_printf(seq, "] ");
6857 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6858 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6860 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6862 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6863 "resync" : "recovery"))),
6864 per_milli/10, per_milli % 10,
6865 (unsigned long long) resync/2,
6866 (unsigned long long) max_sectors/2);
6869 * dt: time from mark until now
6870 * db: blocks written from mark until now
6871 * rt: remaining time
6873 * rt is a sector_t, so could be 32bit or 64bit.
6874 * So we divide before multiply in case it is 32bit and close
6876 * We scale the divisor (db) by 32 to avoid losing precision
6877 * near the end of resync when the number of remaining sectors
6879 * We then divide rt by 32 after multiplying by db to compensate.
6880 * The '+1' avoids division by zero if db is very small.
6882 dt = ((jiffies - mddev->resync_mark) / HZ);
6884 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6885 - mddev->resync_mark_cnt;
6887 rt = max_sectors - resync; /* number of remaining sectors */
6888 sector_div(rt, db/32+1);
6892 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6893 ((unsigned long)rt % 60)/6);
6895 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6898 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6900 struct list_head *tmp;
6902 struct mddev *mddev;
6910 spin_lock(&all_mddevs_lock);
6911 list_for_each(tmp,&all_mddevs)
6913 mddev = list_entry(tmp, struct mddev, all_mddevs);
6915 spin_unlock(&all_mddevs_lock);
6918 spin_unlock(&all_mddevs_lock);
6920 return (void*)2;/* tail */
6924 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6926 struct list_head *tmp;
6927 struct mddev *next_mddev, *mddev = v;
6933 spin_lock(&all_mddevs_lock);
6935 tmp = all_mddevs.next;
6937 tmp = mddev->all_mddevs.next;
6938 if (tmp != &all_mddevs)
6939 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6941 next_mddev = (void*)2;
6944 spin_unlock(&all_mddevs_lock);
6952 static void md_seq_stop(struct seq_file *seq, void *v)
6954 struct mddev *mddev = v;
6956 if (mddev && v != (void*)1 && v != (void*)2)
6960 static int md_seq_show(struct seq_file *seq, void *v)
6962 struct mddev *mddev = v;
6964 struct md_rdev *rdev;
6966 if (v == (void*)1) {
6967 struct md_personality *pers;
6968 seq_printf(seq, "Personalities : ");
6969 spin_lock(&pers_lock);
6970 list_for_each_entry(pers, &pers_list, list)
6971 seq_printf(seq, "[%s] ", pers->name);
6973 spin_unlock(&pers_lock);
6974 seq_printf(seq, "\n");
6975 seq->poll_event = atomic_read(&md_event_count);
6978 if (v == (void*)2) {
6983 if (mddev_lock(mddev) < 0)
6986 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6987 seq_printf(seq, "%s : %sactive", mdname(mddev),
6988 mddev->pers ? "" : "in");
6991 seq_printf(seq, " (read-only)");
6993 seq_printf(seq, " (auto-read-only)");
6994 seq_printf(seq, " %s", mddev->pers->name);
6998 rdev_for_each(rdev, mddev) {
6999 char b[BDEVNAME_SIZE];
7000 seq_printf(seq, " %s[%d]",
7001 bdevname(rdev->bdev,b), rdev->desc_nr);
7002 if (test_bit(WriteMostly, &rdev->flags))
7003 seq_printf(seq, "(W)");
7004 if (test_bit(Faulty, &rdev->flags)) {
7005 seq_printf(seq, "(F)");
7008 if (rdev->raid_disk < 0)
7009 seq_printf(seq, "(S)"); /* spare */
7010 if (test_bit(Replacement, &rdev->flags))
7011 seq_printf(seq, "(R)");
7012 sectors += rdev->sectors;
7015 if (!list_empty(&mddev->disks)) {
7017 seq_printf(seq, "\n %llu blocks",
7018 (unsigned long long)
7019 mddev->array_sectors / 2);
7021 seq_printf(seq, "\n %llu blocks",
7022 (unsigned long long)sectors / 2);
7024 if (mddev->persistent) {
7025 if (mddev->major_version != 0 ||
7026 mddev->minor_version != 90) {
7027 seq_printf(seq," super %d.%d",
7028 mddev->major_version,
7029 mddev->minor_version);
7031 } else if (mddev->external)
7032 seq_printf(seq, " super external:%s",
7033 mddev->metadata_type);
7035 seq_printf(seq, " super non-persistent");
7038 mddev->pers->status(seq, mddev);
7039 seq_printf(seq, "\n ");
7040 if (mddev->pers->sync_request) {
7041 if (mddev->curr_resync > 2) {
7042 status_resync(seq, mddev);
7043 seq_printf(seq, "\n ");
7044 } else if (mddev->curr_resync >= 1)
7045 seq_printf(seq, "\tresync=DELAYED\n ");
7046 else if (mddev->recovery_cp < MaxSector)
7047 seq_printf(seq, "\tresync=PENDING\n ");
7050 seq_printf(seq, "\n ");
7052 bitmap_status(seq, mddev->bitmap);
7054 seq_printf(seq, "\n");
7056 mddev_unlock(mddev);
7061 static const struct seq_operations md_seq_ops = {
7062 .start = md_seq_start,
7063 .next = md_seq_next,
7064 .stop = md_seq_stop,
7065 .show = md_seq_show,
7068 static int md_seq_open(struct inode *inode, struct file *file)
7070 struct seq_file *seq;
7073 error = seq_open(file, &md_seq_ops);
7077 seq = file->private_data;
7078 seq->poll_event = atomic_read(&md_event_count);
7082 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7084 struct seq_file *seq = filp->private_data;
7087 poll_wait(filp, &md_event_waiters, wait);
7089 /* always allow read */
7090 mask = POLLIN | POLLRDNORM;
7092 if (seq->poll_event != atomic_read(&md_event_count))
7093 mask |= POLLERR | POLLPRI;
7097 static const struct file_operations md_seq_fops = {
7098 .owner = THIS_MODULE,
7099 .open = md_seq_open,
7101 .llseek = seq_lseek,
7102 .release = seq_release_private,
7103 .poll = mdstat_poll,
7106 int register_md_personality(struct md_personality *p)
7108 spin_lock(&pers_lock);
7109 list_add_tail(&p->list, &pers_list);
7110 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7111 spin_unlock(&pers_lock);
7115 int unregister_md_personality(struct md_personality *p)
7117 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7118 spin_lock(&pers_lock);
7119 list_del_init(&p->list);
7120 spin_unlock(&pers_lock);
7124 static int is_mddev_idle(struct mddev *mddev, int init)
7126 struct md_rdev * rdev;
7132 rdev_for_each_rcu(rdev, mddev) {
7133 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7134 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7135 (int)part_stat_read(&disk->part0, sectors[1]) -
7136 atomic_read(&disk->sync_io);
7137 /* sync IO will cause sync_io to increase before the disk_stats
7138 * as sync_io is counted when a request starts, and
7139 * disk_stats is counted when it completes.
7140 * So resync activity will cause curr_events to be smaller than
7141 * when there was no such activity.
7142 * non-sync IO will cause disk_stat to increase without
7143 * increasing sync_io so curr_events will (eventually)
7144 * be larger than it was before. Once it becomes
7145 * substantially larger, the test below will cause
7146 * the array to appear non-idle, and resync will slow
7148 * If there is a lot of outstanding resync activity when
7149 * we set last_event to curr_events, then all that activity
7150 * completing might cause the array to appear non-idle
7151 * and resync will be slowed down even though there might
7152 * not have been non-resync activity. This will only
7153 * happen once though. 'last_events' will soon reflect
7154 * the state where there is little or no outstanding
7155 * resync requests, and further resync activity will
7156 * always make curr_events less than last_events.
7159 if (init || curr_events - rdev->last_events > 64) {
7160 rdev->last_events = curr_events;
7168 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7170 /* another "blocks" (512byte) blocks have been synced */
7171 atomic_sub(blocks, &mddev->recovery_active);
7172 wake_up(&mddev->recovery_wait);
7174 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7175 md_wakeup_thread(mddev->thread);
7176 // stop recovery, signal do_sync ....
7181 /* md_write_start(mddev, bi)
7182 * If we need to update some array metadata (e.g. 'active' flag
7183 * in superblock) before writing, schedule a superblock update
7184 * and wait for it to complete.
7186 void md_write_start(struct mddev *mddev, struct bio *bi)
7189 if (bio_data_dir(bi) != WRITE)
7192 BUG_ON(mddev->ro == 1);
7193 if (mddev->ro == 2) {
7194 /* need to switch to read/write */
7196 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7197 md_wakeup_thread(mddev->thread);
7198 md_wakeup_thread(mddev->sync_thread);
7201 atomic_inc(&mddev->writes_pending);
7202 if (mddev->safemode == 1)
7203 mddev->safemode = 0;
7204 if (mddev->in_sync) {
7205 spin_lock_irq(&mddev->write_lock);
7206 if (mddev->in_sync) {
7208 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7209 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7210 md_wakeup_thread(mddev->thread);
7213 spin_unlock_irq(&mddev->write_lock);
7216 sysfs_notify_dirent_safe(mddev->sysfs_state);
7217 wait_event(mddev->sb_wait,
7218 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7221 void md_write_end(struct mddev *mddev)
7223 if (atomic_dec_and_test(&mddev->writes_pending)) {
7224 if (mddev->safemode == 2)
7225 md_wakeup_thread(mddev->thread);
7226 else if (mddev->safemode_delay)
7227 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7231 /* md_allow_write(mddev)
7232 * Calling this ensures that the array is marked 'active' so that writes
7233 * may proceed without blocking. It is important to call this before
7234 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7235 * Must be called with mddev_lock held.
7237 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7238 * is dropped, so return -EAGAIN after notifying userspace.
7240 int md_allow_write(struct mddev *mddev)
7246 if (!mddev->pers->sync_request)
7249 spin_lock_irq(&mddev->write_lock);
7250 if (mddev->in_sync) {
7252 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7253 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7254 if (mddev->safemode_delay &&
7255 mddev->safemode == 0)
7256 mddev->safemode = 1;
7257 spin_unlock_irq(&mddev->write_lock);
7258 md_update_sb(mddev, 0);
7259 sysfs_notify_dirent_safe(mddev->sysfs_state);
7261 spin_unlock_irq(&mddev->write_lock);
7263 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7268 EXPORT_SYMBOL_GPL(md_allow_write);
7270 #define SYNC_MARKS 10
7271 #define SYNC_MARK_STEP (3*HZ)
7272 void md_do_sync(struct md_thread *thread)
7274 struct mddev *mddev = thread->mddev;
7275 struct mddev *mddev2;
7276 unsigned int currspeed = 0,
7278 sector_t max_sectors,j, io_sectors;
7279 unsigned long mark[SYNC_MARKS];
7280 sector_t mark_cnt[SYNC_MARKS];
7282 struct list_head *tmp;
7283 sector_t last_check;
7285 struct md_rdev *rdev;
7287 struct blk_plug plug;
7289 /* just incase thread restarts... */
7290 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7292 if (mddev->ro) /* never try to sync a read-only array */
7295 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7296 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7297 desc = "data-check";
7298 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7299 desc = "requested-resync";
7302 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7307 /* we overload curr_resync somewhat here.
7308 * 0 == not engaged in resync at all
7309 * 2 == checking that there is no conflict with another sync
7310 * 1 == like 2, but have yielded to allow conflicting resync to
7312 * other == active in resync - this many blocks
7314 * Before starting a resync we must have set curr_resync to
7315 * 2, and then checked that every "conflicting" array has curr_resync
7316 * less than ours. When we find one that is the same or higher
7317 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7318 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7319 * This will mean we have to start checking from the beginning again.
7324 mddev->curr_resync = 2;
7327 if (kthread_should_stop())
7328 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7330 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7332 for_each_mddev(mddev2, tmp) {
7333 if (mddev2 == mddev)
7335 if (!mddev->parallel_resync
7336 && mddev2->curr_resync
7337 && match_mddev_units(mddev, mddev2)) {
7339 if (mddev < mddev2 && mddev->curr_resync == 2) {
7340 /* arbitrarily yield */
7341 mddev->curr_resync = 1;
7342 wake_up(&resync_wait);
7344 if (mddev > mddev2 && mddev->curr_resync == 1)
7345 /* no need to wait here, we can wait the next
7346 * time 'round when curr_resync == 2
7349 /* We need to wait 'interruptible' so as not to
7350 * contribute to the load average, and not to
7351 * be caught by 'softlockup'
7353 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7354 if (!kthread_should_stop() &&
7355 mddev2->curr_resync >= mddev->curr_resync) {
7356 printk(KERN_INFO "md: delaying %s of %s"
7357 " until %s has finished (they"
7358 " share one or more physical units)\n",
7359 desc, mdname(mddev), mdname(mddev2));
7361 if (signal_pending(current))
7362 flush_signals(current);
7364 finish_wait(&resync_wait, &wq);
7367 finish_wait(&resync_wait, &wq);
7370 } while (mddev->curr_resync < 2);
7373 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7374 /* resync follows the size requested by the personality,
7375 * which defaults to physical size, but can be virtual size
7377 max_sectors = mddev->resync_max_sectors;
7378 atomic64_set(&mddev->resync_mismatches, 0);
7379 /* we don't use the checkpoint if there's a bitmap */
7380 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7381 j = mddev->resync_min;
7382 else if (!mddev->bitmap)
7383 j = mddev->recovery_cp;
7385 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7386 max_sectors = mddev->resync_max_sectors;
7388 /* recovery follows the physical size of devices */
7389 max_sectors = mddev->dev_sectors;
7392 rdev_for_each_rcu(rdev, mddev)
7393 if (rdev->raid_disk >= 0 &&
7394 !test_bit(Faulty, &rdev->flags) &&
7395 !test_bit(In_sync, &rdev->flags) &&
7396 rdev->recovery_offset < j)
7397 j = rdev->recovery_offset;
7401 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7402 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7403 " %d KB/sec/disk.\n", speed_min(mddev));
7404 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7405 "(but not more than %d KB/sec) for %s.\n",
7406 speed_max(mddev), desc);
7408 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7411 for (m = 0; m < SYNC_MARKS; m++) {
7413 mark_cnt[m] = io_sectors;
7416 mddev->resync_mark = mark[last_mark];
7417 mddev->resync_mark_cnt = mark_cnt[last_mark];
7420 * Tune reconstruction:
7422 window = 32*(PAGE_SIZE/512);
7423 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7424 window/2, (unsigned long long)max_sectors/2);
7426 atomic_set(&mddev->recovery_active, 0);
7431 "md: resuming %s of %s from checkpoint.\n",
7432 desc, mdname(mddev));
7433 mddev->curr_resync = j;
7435 mddev->curr_resync = 3; /* no longer delayed */
7436 mddev->curr_resync_completed = j;
7437 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7438 md_new_event(mddev);
7440 blk_start_plug(&plug);
7441 while (j < max_sectors) {
7446 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7447 ((mddev->curr_resync > mddev->curr_resync_completed &&
7448 (mddev->curr_resync - mddev->curr_resync_completed)
7449 > (max_sectors >> 4)) ||
7450 (j - mddev->curr_resync_completed)*2
7451 >= mddev->resync_max - mddev->curr_resync_completed
7453 /* time to update curr_resync_completed */
7454 wait_event(mddev->recovery_wait,
7455 atomic_read(&mddev->recovery_active) == 0);
7456 mddev->curr_resync_completed = j;
7457 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7458 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7461 while (j >= mddev->resync_max && !kthread_should_stop()) {
7462 /* As this condition is controlled by user-space,
7463 * we can block indefinitely, so use '_interruptible'
7464 * to avoid triggering warnings.
7466 flush_signals(current); /* just in case */
7467 wait_event_interruptible(mddev->recovery_wait,
7468 mddev->resync_max > j
7469 || kthread_should_stop());
7472 if (kthread_should_stop())
7475 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7476 currspeed < speed_min(mddev));
7478 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7482 if (!skipped) { /* actual IO requested */
7483 io_sectors += sectors;
7484 atomic_add(sectors, &mddev->recovery_active);
7487 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7492 mddev->curr_resync = j;
7493 mddev->curr_mark_cnt = io_sectors;
7494 if (last_check == 0)
7495 /* this is the earliest that rebuild will be
7496 * visible in /proc/mdstat
7498 md_new_event(mddev);
7500 if (last_check + window > io_sectors || j == max_sectors)
7503 last_check = io_sectors;
7505 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7507 int next = (last_mark+1) % SYNC_MARKS;
7509 mddev->resync_mark = mark[next];
7510 mddev->resync_mark_cnt = mark_cnt[next];
7511 mark[next] = jiffies;
7512 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7517 if (kthread_should_stop())
7522 * this loop exits only if either when we are slower than
7523 * the 'hard' speed limit, or the system was IO-idle for
7525 * the system might be non-idle CPU-wise, but we only care
7526 * about not overloading the IO subsystem. (things like an
7527 * e2fsck being done on the RAID array should execute fast)
7531 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7532 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7534 if (currspeed > speed_min(mddev)) {
7535 if ((currspeed > speed_max(mddev)) ||
7536 !is_mddev_idle(mddev, 0)) {
7542 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7544 * this also signals 'finished resyncing' to md_stop
7547 blk_finish_plug(&plug);
7548 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7550 /* tell personality that we are finished */
7551 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7553 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7554 mddev->curr_resync > 2) {
7555 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7556 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7557 if (mddev->curr_resync >= mddev->recovery_cp) {
7559 "md: checkpointing %s of %s.\n",
7560 desc, mdname(mddev));
7561 mddev->recovery_cp =
7562 mddev->curr_resync_completed;
7565 mddev->recovery_cp = MaxSector;
7567 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7568 mddev->curr_resync = MaxSector;
7570 rdev_for_each_rcu(rdev, mddev)
7571 if (rdev->raid_disk >= 0 &&
7572 mddev->delta_disks >= 0 &&
7573 !test_bit(Faulty, &rdev->flags) &&
7574 !test_bit(In_sync, &rdev->flags) &&
7575 rdev->recovery_offset < mddev->curr_resync)
7576 rdev->recovery_offset = mddev->curr_resync;
7581 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7583 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7584 /* We completed so min/max setting can be forgotten if used. */
7585 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7586 mddev->resync_min = 0;
7587 mddev->resync_max = MaxSector;
7588 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7589 mddev->resync_min = mddev->curr_resync_completed;
7590 mddev->curr_resync = 0;
7591 wake_up(&resync_wait);
7592 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7593 md_wakeup_thread(mddev->thread);
7598 * got a signal, exit.
7601 "md: md_do_sync() got signal ... exiting\n");
7602 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7606 EXPORT_SYMBOL_GPL(md_do_sync);
7608 static int remove_and_add_spares(struct mddev *mddev)
7610 struct md_rdev *rdev;
7614 rdev_for_each(rdev, mddev)
7615 if (rdev->raid_disk >= 0 &&
7616 !test_bit(Blocked, &rdev->flags) &&
7617 (test_bit(Faulty, &rdev->flags) ||
7618 ! test_bit(In_sync, &rdev->flags)) &&
7619 atomic_read(&rdev->nr_pending)==0) {
7620 if (mddev->pers->hot_remove_disk(
7621 mddev, rdev) == 0) {
7622 sysfs_unlink_rdev(mddev, rdev);
7623 rdev->raid_disk = -1;
7628 sysfs_notify(&mddev->kobj, NULL,
7632 rdev_for_each(rdev, mddev) {
7633 if (rdev->raid_disk >= 0 &&
7634 !test_bit(In_sync, &rdev->flags) &&
7635 !test_bit(Faulty, &rdev->flags))
7637 if (rdev->raid_disk < 0
7638 && !test_bit(Faulty, &rdev->flags)) {
7639 rdev->recovery_offset = 0;
7641 hot_add_disk(mddev, rdev) == 0) {
7642 if (sysfs_link_rdev(mddev, rdev))
7643 /* failure here is OK */;
7645 md_new_event(mddev);
7646 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7651 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7655 static void reap_sync_thread(struct mddev *mddev)
7657 struct md_rdev *rdev;
7659 /* resync has finished, collect result */
7660 md_unregister_thread(&mddev->sync_thread);
7661 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7662 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7664 /* activate any spares */
7665 if (mddev->pers->spare_active(mddev)) {
7666 sysfs_notify(&mddev->kobj, NULL,
7668 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7671 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7672 mddev->pers->finish_reshape)
7673 mddev->pers->finish_reshape(mddev);
7675 /* If array is no-longer degraded, then any saved_raid_disk
7676 * information must be scrapped. Also if any device is now
7677 * In_sync we must scrape the saved_raid_disk for that device
7678 * do the superblock for an incrementally recovered device
7681 rdev_for_each(rdev, mddev)
7682 if (!mddev->degraded ||
7683 test_bit(In_sync, &rdev->flags))
7684 rdev->saved_raid_disk = -1;
7686 md_update_sb(mddev, 1);
7687 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7688 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7689 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7690 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7691 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7692 /* flag recovery needed just to double check */
7693 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7694 sysfs_notify_dirent_safe(mddev->sysfs_action);
7695 md_new_event(mddev);
7696 if (mddev->event_work.func)
7697 queue_work(md_misc_wq, &mddev->event_work);
7701 * This routine is regularly called by all per-raid-array threads to
7702 * deal with generic issues like resync and super-block update.
7703 * Raid personalities that don't have a thread (linear/raid0) do not
7704 * need this as they never do any recovery or update the superblock.
7706 * It does not do any resync itself, but rather "forks" off other threads
7707 * to do that as needed.
7708 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7709 * "->recovery" and create a thread at ->sync_thread.
7710 * When the thread finishes it sets MD_RECOVERY_DONE
7711 * and wakeups up this thread which will reap the thread and finish up.
7712 * This thread also removes any faulty devices (with nr_pending == 0).
7714 * The overall approach is:
7715 * 1/ if the superblock needs updating, update it.
7716 * 2/ If a recovery thread is running, don't do anything else.
7717 * 3/ If recovery has finished, clean up, possibly marking spares active.
7718 * 4/ If there are any faulty devices, remove them.
7719 * 5/ If array is degraded, try to add spares devices
7720 * 6/ If array has spares or is not in-sync, start a resync thread.
7722 void md_check_recovery(struct mddev *mddev)
7724 if (mddev->suspended)
7728 bitmap_daemon_work(mddev);
7730 if (signal_pending(current)) {
7731 if (mddev->pers->sync_request && !mddev->external) {
7732 printk(KERN_INFO "md: %s in immediate safe mode\n",
7734 mddev->safemode = 2;
7736 flush_signals(current);
7739 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7742 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7743 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7744 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7745 (mddev->external == 0 && mddev->safemode == 1) ||
7746 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7747 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7751 if (mddev_trylock(mddev)) {
7755 /* Only thing we do on a ro array is remove
7758 struct md_rdev *rdev;
7759 rdev_for_each(rdev, mddev)
7760 if (rdev->raid_disk >= 0 &&
7761 !test_bit(Blocked, &rdev->flags) &&
7762 test_bit(Faulty, &rdev->flags) &&
7763 atomic_read(&rdev->nr_pending)==0) {
7764 if (mddev->pers->hot_remove_disk(
7765 mddev, rdev) == 0) {
7766 sysfs_unlink_rdev(mddev, rdev);
7767 rdev->raid_disk = -1;
7770 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7774 if (!mddev->external) {
7776 spin_lock_irq(&mddev->write_lock);
7777 if (mddev->safemode &&
7778 !atomic_read(&mddev->writes_pending) &&
7780 mddev->recovery_cp == MaxSector) {
7783 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7785 if (mddev->safemode == 1)
7786 mddev->safemode = 0;
7787 spin_unlock_irq(&mddev->write_lock);
7789 sysfs_notify_dirent_safe(mddev->sysfs_state);
7793 md_update_sb(mddev, 0);
7795 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7796 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7797 /* resync/recovery still happening */
7798 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7801 if (mddev->sync_thread) {
7802 reap_sync_thread(mddev);
7805 /* Set RUNNING before clearing NEEDED to avoid
7806 * any transients in the value of "sync_action".
7808 mddev->curr_resync_completed = 0;
7809 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7810 /* Clear some bits that don't mean anything, but
7813 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7814 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7816 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7817 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7819 /* no recovery is running.
7820 * remove any failed drives, then
7821 * add spares if possible.
7822 * Spares are also removed and re-added, to allow
7823 * the personality to fail the re-add.
7826 if (mddev->reshape_position != MaxSector) {
7827 if (mddev->pers->check_reshape == NULL ||
7828 mddev->pers->check_reshape(mddev) != 0)
7829 /* Cannot proceed */
7831 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7832 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7833 } else if ((spares = remove_and_add_spares(mddev))) {
7834 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7835 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7836 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7837 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7838 } else if (mddev->recovery_cp < MaxSector) {
7839 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7840 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7841 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7842 /* nothing to be done ... */
7845 if (mddev->pers->sync_request) {
7847 /* We are adding a device or devices to an array
7848 * which has the bitmap stored on all devices.
7849 * So make sure all bitmap pages get written
7851 bitmap_write_all(mddev->bitmap);
7853 mddev->sync_thread = md_register_thread(md_do_sync,
7856 if (!mddev->sync_thread) {
7857 printk(KERN_ERR "%s: could not start resync"
7860 /* leave the spares where they are, it shouldn't hurt */
7861 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7862 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7863 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7864 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7865 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7867 md_wakeup_thread(mddev->sync_thread);
7868 sysfs_notify_dirent_safe(mddev->sysfs_action);
7869 md_new_event(mddev);
7872 if (!mddev->sync_thread) {
7873 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7874 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7876 if (mddev->sysfs_action)
7877 sysfs_notify_dirent_safe(mddev->sysfs_action);
7879 mddev_unlock(mddev);
7883 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7885 sysfs_notify_dirent_safe(rdev->sysfs_state);
7886 wait_event_timeout(rdev->blocked_wait,
7887 !test_bit(Blocked, &rdev->flags) &&
7888 !test_bit(BlockedBadBlocks, &rdev->flags),
7889 msecs_to_jiffies(5000));
7890 rdev_dec_pending(rdev, mddev);
7892 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7894 void md_finish_reshape(struct mddev *mddev)
7896 /* called be personality module when reshape completes. */
7897 struct md_rdev *rdev;
7899 rdev_for_each(rdev, mddev) {
7900 if (rdev->data_offset > rdev->new_data_offset)
7901 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7903 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7904 rdev->data_offset = rdev->new_data_offset;
7907 EXPORT_SYMBOL(md_finish_reshape);
7909 /* Bad block management.
7910 * We can record which blocks on each device are 'bad' and so just
7911 * fail those blocks, or that stripe, rather than the whole device.
7912 * Entries in the bad-block table are 64bits wide. This comprises:
7913 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7914 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7915 * A 'shift' can be set so that larger blocks are tracked and
7916 * consequently larger devices can be covered.
7917 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7919 * Locking of the bad-block table uses a seqlock so md_is_badblock
7920 * might need to retry if it is very unlucky.
7921 * We will sometimes want to check for bad blocks in a bi_end_io function,
7922 * so we use the write_seqlock_irq variant.
7924 * When looking for a bad block we specify a range and want to
7925 * know if any block in the range is bad. So we binary-search
7926 * to the last range that starts at-or-before the given endpoint,
7927 * (or "before the sector after the target range")
7928 * then see if it ends after the given start.
7930 * 0 if there are no known bad blocks in the range
7931 * 1 if there are known bad block which are all acknowledged
7932 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7933 * plus the start/length of the first bad section we overlap.
7935 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7936 sector_t *first_bad, int *bad_sectors)
7942 sector_t target = s + sectors;
7945 if (bb->shift > 0) {
7946 /* round the start down, and the end up */
7948 target += (1<<bb->shift) - 1;
7949 target >>= bb->shift;
7950 sectors = target - s;
7952 /* 'target' is now the first block after the bad range */
7955 seq = read_seqbegin(&bb->lock);
7960 /* Binary search between lo and hi for 'target'
7961 * i.e. for the last range that starts before 'target'
7963 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7964 * are known not to be the last range before target.
7965 * VARIANT: hi-lo is the number of possible
7966 * ranges, and decreases until it reaches 1
7968 while (hi - lo > 1) {
7969 int mid = (lo + hi) / 2;
7970 sector_t a = BB_OFFSET(p[mid]);
7972 /* This could still be the one, earlier ranges
7976 /* This and later ranges are definitely out. */
7979 /* 'lo' might be the last that started before target, but 'hi' isn't */
7981 /* need to check all range that end after 's' to see if
7982 * any are unacknowledged.
7985 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7986 if (BB_OFFSET(p[lo]) < target) {
7987 /* starts before the end, and finishes after
7988 * the start, so they must overlap
7990 if (rv != -1 && BB_ACK(p[lo]))
7994 *first_bad = BB_OFFSET(p[lo]);
7995 *bad_sectors = BB_LEN(p[lo]);
8001 if (read_seqretry(&bb->lock, seq))
8006 EXPORT_SYMBOL_GPL(md_is_badblock);
8009 * Add a range of bad blocks to the table.
8010 * This might extend the table, or might contract it
8011 * if two adjacent ranges can be merged.
8012 * We binary-search to find the 'insertion' point, then
8013 * decide how best to handle it.
8015 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8023 /* badblocks are disabled */
8027 /* round the start down, and the end up */
8028 sector_t next = s + sectors;
8030 next += (1<<bb->shift) - 1;
8035 write_seqlock_irq(&bb->lock);
8040 /* Find the last range that starts at-or-before 's' */
8041 while (hi - lo > 1) {
8042 int mid = (lo + hi) / 2;
8043 sector_t a = BB_OFFSET(p[mid]);
8049 if (hi > lo && BB_OFFSET(p[lo]) > s)
8053 /* we found a range that might merge with the start
8056 sector_t a = BB_OFFSET(p[lo]);
8057 sector_t e = a + BB_LEN(p[lo]);
8058 int ack = BB_ACK(p[lo]);
8060 /* Yes, we can merge with a previous range */
8061 if (s == a && s + sectors >= e)
8062 /* new range covers old */
8065 ack = ack && acknowledged;
8067 if (e < s + sectors)
8069 if (e - a <= BB_MAX_LEN) {
8070 p[lo] = BB_MAKE(a, e-a, ack);
8073 /* does not all fit in one range,
8074 * make p[lo] maximal
8076 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8077 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8083 if (sectors && hi < bb->count) {
8084 /* 'hi' points to the first range that starts after 's'.
8085 * Maybe we can merge with the start of that range */
8086 sector_t a = BB_OFFSET(p[hi]);
8087 sector_t e = a + BB_LEN(p[hi]);
8088 int ack = BB_ACK(p[hi]);
8089 if (a <= s + sectors) {
8090 /* merging is possible */
8091 if (e <= s + sectors) {
8096 ack = ack && acknowledged;
8099 if (e - a <= BB_MAX_LEN) {
8100 p[hi] = BB_MAKE(a, e-a, ack);
8103 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8111 if (sectors == 0 && hi < bb->count) {
8112 /* we might be able to combine lo and hi */
8113 /* Note: 's' is at the end of 'lo' */
8114 sector_t a = BB_OFFSET(p[hi]);
8115 int lolen = BB_LEN(p[lo]);
8116 int hilen = BB_LEN(p[hi]);
8117 int newlen = lolen + hilen - (s - a);
8118 if (s >= a && newlen < BB_MAX_LEN) {
8119 /* yes, we can combine them */
8120 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8121 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8122 memmove(p + hi, p + hi + 1,
8123 (bb->count - hi - 1) * 8);
8128 /* didn't merge (it all).
8129 * Need to add a range just before 'hi' */
8130 if (bb->count >= MD_MAX_BADBLOCKS) {
8131 /* No room for more */
8135 int this_sectors = sectors;
8136 memmove(p + hi + 1, p + hi,
8137 (bb->count - hi) * 8);
8140 if (this_sectors > BB_MAX_LEN)
8141 this_sectors = BB_MAX_LEN;
8142 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8143 sectors -= this_sectors;
8150 bb->unacked_exist = 1;
8151 write_sequnlock_irq(&bb->lock);
8156 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8161 s += rdev->new_data_offset;
8163 s += rdev->data_offset;
8164 rv = md_set_badblocks(&rdev->badblocks,
8167 /* Make sure they get written out promptly */
8168 sysfs_notify_dirent_safe(rdev->sysfs_state);
8169 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8170 md_wakeup_thread(rdev->mddev->thread);
8174 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8177 * Remove a range of bad blocks from the table.
8178 * This may involve extending the table if we spilt a region,
8179 * but it must not fail. So if the table becomes full, we just
8180 * drop the remove request.
8182 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8186 sector_t target = s + sectors;
8189 if (bb->shift > 0) {
8190 /* When clearing we round the start up and the end down.
8191 * This should not matter as the shift should align with
8192 * the block size and no rounding should ever be needed.
8193 * However it is better the think a block is bad when it
8194 * isn't than to think a block is not bad when it is.
8196 s += (1<<bb->shift) - 1;
8198 target >>= bb->shift;
8199 sectors = target - s;
8202 write_seqlock_irq(&bb->lock);
8207 /* Find the last range that starts before 'target' */
8208 while (hi - lo > 1) {
8209 int mid = (lo + hi) / 2;
8210 sector_t a = BB_OFFSET(p[mid]);
8217 /* p[lo] is the last range that could overlap the
8218 * current range. Earlier ranges could also overlap,
8219 * but only this one can overlap the end of the range.
8221 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8222 /* Partial overlap, leave the tail of this range */
8223 int ack = BB_ACK(p[lo]);
8224 sector_t a = BB_OFFSET(p[lo]);
8225 sector_t end = a + BB_LEN(p[lo]);
8228 /* we need to split this range */
8229 if (bb->count >= MD_MAX_BADBLOCKS) {
8233 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8235 p[lo] = BB_MAKE(a, s-a, ack);
8238 p[lo] = BB_MAKE(target, end - target, ack);
8239 /* there is no longer an overlap */
8244 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8245 /* This range does overlap */
8246 if (BB_OFFSET(p[lo]) < s) {
8247 /* Keep the early parts of this range. */
8248 int ack = BB_ACK(p[lo]);
8249 sector_t start = BB_OFFSET(p[lo]);
8250 p[lo] = BB_MAKE(start, s - start, ack);
8251 /* now low doesn't overlap, so.. */
8256 /* 'lo' is strictly before, 'hi' is strictly after,
8257 * anything between needs to be discarded
8260 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8261 bb->count -= (hi - lo - 1);
8267 write_sequnlock_irq(&bb->lock);
8271 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8275 s += rdev->new_data_offset;
8277 s += rdev->data_offset;
8278 return md_clear_badblocks(&rdev->badblocks,
8281 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8284 * Acknowledge all bad blocks in a list.
8285 * This only succeeds if ->changed is clear. It is used by
8286 * in-kernel metadata updates
8288 void md_ack_all_badblocks(struct badblocks *bb)
8290 if (bb->page == NULL || bb->changed)
8291 /* no point even trying */
8293 write_seqlock_irq(&bb->lock);
8295 if (bb->changed == 0 && bb->unacked_exist) {
8298 for (i = 0; i < bb->count ; i++) {
8299 if (!BB_ACK(p[i])) {
8300 sector_t start = BB_OFFSET(p[i]);
8301 int len = BB_LEN(p[i]);
8302 p[i] = BB_MAKE(start, len, 1);
8305 bb->unacked_exist = 0;
8307 write_sequnlock_irq(&bb->lock);
8309 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8311 /* sysfs access to bad-blocks list.
8312 * We present two files.
8313 * 'bad-blocks' lists sector numbers and lengths of ranges that
8314 * are recorded as bad. The list is truncated to fit within
8315 * the one-page limit of sysfs.
8316 * Writing "sector length" to this file adds an acknowledged
8318 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8319 * been acknowledged. Writing to this file adds bad blocks
8320 * without acknowledging them. This is largely for testing.
8324 badblocks_show(struct badblocks *bb, char *page, int unack)
8335 seq = read_seqbegin(&bb->lock);
8340 while (len < PAGE_SIZE && i < bb->count) {
8341 sector_t s = BB_OFFSET(p[i]);
8342 unsigned int length = BB_LEN(p[i]);
8343 int ack = BB_ACK(p[i]);
8349 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8350 (unsigned long long)s << bb->shift,
8351 length << bb->shift);
8353 if (unack && len == 0)
8354 bb->unacked_exist = 0;
8356 if (read_seqretry(&bb->lock, seq))
8365 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8367 unsigned long long sector;
8371 /* Allow clearing via sysfs *only* for testing/debugging.
8372 * Normally only a successful write may clear a badblock
8375 if (page[0] == '-') {
8379 #endif /* DO_DEBUG */
8381 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8383 if (newline != '\n')
8395 md_clear_badblocks(bb, sector, length);
8398 #endif /* DO_DEBUG */
8399 if (md_set_badblocks(bb, sector, length, !unack))
8405 static int md_notify_reboot(struct notifier_block *this,
8406 unsigned long code, void *x)
8408 struct list_head *tmp;
8409 struct mddev *mddev;
8412 for_each_mddev(mddev, tmp) {
8413 if (mddev_trylock(mddev)) {
8415 __md_stop_writes(mddev);
8416 mddev->safemode = 2;
8417 mddev_unlock(mddev);
8422 * certain more exotic SCSI devices are known to be
8423 * volatile wrt too early system reboots. While the
8424 * right place to handle this issue is the given
8425 * driver, we do want to have a safe RAID driver ...
8433 static struct notifier_block md_notifier = {
8434 .notifier_call = md_notify_reboot,
8436 .priority = INT_MAX, /* before any real devices */
8439 static void md_geninit(void)
8441 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8443 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8446 static int __init md_init(void)
8450 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8454 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8458 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8461 if ((ret = register_blkdev(0, "mdp")) < 0)
8465 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8466 md_probe, NULL, NULL);
8467 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8468 md_probe, NULL, NULL);
8470 register_reboot_notifier(&md_notifier);
8471 raid_table_header = register_sysctl_table(raid_root_table);
8477 unregister_blkdev(MD_MAJOR, "md");
8479 destroy_workqueue(md_misc_wq);
8481 destroy_workqueue(md_wq);
8489 * Searches all registered partitions for autorun RAID arrays
8493 static LIST_HEAD(all_detected_devices);
8494 struct detected_devices_node {
8495 struct list_head list;
8499 void md_autodetect_dev(dev_t dev)
8501 struct detected_devices_node *node_detected_dev;
8503 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8504 if (node_detected_dev) {
8505 node_detected_dev->dev = dev;
8506 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8508 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8509 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8514 static void autostart_arrays(int part)
8516 struct md_rdev *rdev;
8517 struct detected_devices_node *node_detected_dev;
8519 int i_scanned, i_passed;
8524 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8526 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8528 node_detected_dev = list_entry(all_detected_devices.next,
8529 struct detected_devices_node, list);
8530 list_del(&node_detected_dev->list);
8531 dev = node_detected_dev->dev;
8532 kfree(node_detected_dev);
8533 rdev = md_import_device(dev,0, 90);
8537 if (test_bit(Faulty, &rdev->flags)) {
8541 set_bit(AutoDetected, &rdev->flags);
8542 list_add(&rdev->same_set, &pending_raid_disks);
8546 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8547 i_scanned, i_passed);
8549 autorun_devices(part);
8552 #endif /* !MODULE */
8554 static __exit void md_exit(void)
8556 struct mddev *mddev;
8557 struct list_head *tmp;
8559 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8560 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8562 unregister_blkdev(MD_MAJOR,"md");
8563 unregister_blkdev(mdp_major, "mdp");
8564 unregister_reboot_notifier(&md_notifier);
8565 unregister_sysctl_table(raid_table_header);
8566 remove_proc_entry("mdstat", NULL);
8567 for_each_mddev(mddev, tmp) {
8568 export_array(mddev);
8569 mddev->hold_active = 0;
8571 destroy_workqueue(md_misc_wq);
8572 destroy_workqueue(md_wq);
8575 subsys_initcall(md_init);
8576 module_exit(md_exit)
8578 static int get_ro(char *buffer, struct kernel_param *kp)
8580 return sprintf(buffer, "%d", start_readonly);
8582 static int set_ro(const char *val, struct kernel_param *kp)
8585 int num = simple_strtoul(val, &e, 10);
8586 if (*val && (*e == '\0' || *e == '\n')) {
8587 start_readonly = num;
8593 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8594 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8596 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8598 EXPORT_SYMBOL(register_md_personality);
8599 EXPORT_SYMBOL(unregister_md_personality);
8600 EXPORT_SYMBOL(md_error);
8601 EXPORT_SYMBOL(md_done_sync);
8602 EXPORT_SYMBOL(md_write_start);
8603 EXPORT_SYMBOL(md_write_end);
8604 EXPORT_SYMBOL(md_register_thread);
8605 EXPORT_SYMBOL(md_unregister_thread);
8606 EXPORT_SYMBOL(md_wakeup_thread);
8607 EXPORT_SYMBOL(md_check_recovery);
8608 MODULE_LICENSE("GPL");
8609 MODULE_DESCRIPTION("MD RAID framework");
8611 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);