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 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
311 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
314 smp_rmb(); /* Ensure implications of 'active' are visible */
316 if (mddev->suspended) {
319 prepare_to_wait(&mddev->sb_wait, &__wait,
320 TASK_UNINTERRUPTIBLE);
321 if (!mddev->suspended)
327 finish_wait(&mddev->sb_wait, &__wait);
329 atomic_inc(&mddev->active_io);
333 * save the sectors now since our bio can
334 * go away inside make_request
336 sectors = bio_sectors(bio);
337 mddev->pers->make_request(mddev, bio);
339 cpu = part_stat_lock();
340 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
341 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
344 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
345 wake_up(&mddev->sb_wait);
348 /* mddev_suspend makes sure no new requests are submitted
349 * to the device, and that any requests that have been submitted
350 * are completely handled.
351 * Once ->stop is called and completes, the module will be completely
354 void mddev_suspend(struct mddev *mddev)
356 BUG_ON(mddev->suspended);
357 mddev->suspended = 1;
359 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
360 mddev->pers->quiesce(mddev, 1);
362 del_timer_sync(&mddev->safemode_timer);
364 EXPORT_SYMBOL_GPL(mddev_suspend);
366 void mddev_resume(struct mddev *mddev)
368 mddev->suspended = 0;
369 wake_up(&mddev->sb_wait);
370 mddev->pers->quiesce(mddev, 0);
372 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
373 md_wakeup_thread(mddev->thread);
374 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
376 EXPORT_SYMBOL_GPL(mddev_resume);
378 int mddev_congested(struct mddev *mddev, int bits)
380 return mddev->suspended;
382 EXPORT_SYMBOL(mddev_congested);
385 * Generic flush handling for md
388 static void md_end_flush(struct bio *bio, int err)
390 struct md_rdev *rdev = bio->bi_private;
391 struct mddev *mddev = rdev->mddev;
393 rdev_dec_pending(rdev, mddev);
395 if (atomic_dec_and_test(&mddev->flush_pending)) {
396 /* The pre-request flush has finished */
397 queue_work(md_wq, &mddev->flush_work);
402 static void md_submit_flush_data(struct work_struct *ws);
404 static void submit_flushes(struct work_struct *ws)
406 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
407 struct md_rdev *rdev;
409 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
410 atomic_set(&mddev->flush_pending, 1);
412 rdev_for_each_rcu(rdev, mddev)
413 if (rdev->raid_disk >= 0 &&
414 !test_bit(Faulty, &rdev->flags)) {
415 /* Take two references, one is dropped
416 * when request finishes, one after
417 * we reclaim rcu_read_lock
420 atomic_inc(&rdev->nr_pending);
421 atomic_inc(&rdev->nr_pending);
423 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
424 bi->bi_end_io = md_end_flush;
425 bi->bi_private = rdev;
426 bi->bi_bdev = rdev->bdev;
427 atomic_inc(&mddev->flush_pending);
428 submit_bio(WRITE_FLUSH, bi);
430 rdev_dec_pending(rdev, mddev);
433 if (atomic_dec_and_test(&mddev->flush_pending))
434 queue_work(md_wq, &mddev->flush_work);
437 static void md_submit_flush_data(struct work_struct *ws)
439 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
440 struct bio *bio = mddev->flush_bio;
442 if (bio->bi_size == 0)
443 /* an empty barrier - all done */
446 bio->bi_rw &= ~REQ_FLUSH;
447 mddev->pers->make_request(mddev, bio);
450 mddev->flush_bio = NULL;
451 wake_up(&mddev->sb_wait);
454 void md_flush_request(struct mddev *mddev, struct bio *bio)
456 spin_lock_irq(&mddev->write_lock);
457 wait_event_lock_irq(mddev->sb_wait,
460 mddev->flush_bio = bio;
461 spin_unlock_irq(&mddev->write_lock);
463 INIT_WORK(&mddev->flush_work, submit_flushes);
464 queue_work(md_wq, &mddev->flush_work);
466 EXPORT_SYMBOL(md_flush_request);
468 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
470 struct mddev *mddev = cb->data;
471 md_wakeup_thread(mddev->thread);
474 EXPORT_SYMBOL(md_unplug);
476 static inline struct mddev *mddev_get(struct mddev *mddev)
478 atomic_inc(&mddev->active);
482 static void mddev_delayed_delete(struct work_struct *ws);
484 static void mddev_put(struct mddev *mddev)
486 struct bio_set *bs = NULL;
488 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
490 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
491 mddev->ctime == 0 && !mddev->hold_active) {
492 /* Array is not configured at all, and not held active,
494 list_del_init(&mddev->all_mddevs);
496 mddev->bio_set = NULL;
497 if (mddev->gendisk) {
498 /* We did a probe so need to clean up. Call
499 * queue_work inside the spinlock so that
500 * flush_workqueue() after mddev_find will
501 * succeed in waiting for the work to be done.
503 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
504 queue_work(md_misc_wq, &mddev->del_work);
508 spin_unlock(&all_mddevs_lock);
513 void mddev_init(struct mddev *mddev)
515 mutex_init(&mddev->open_mutex);
516 mutex_init(&mddev->reconfig_mutex);
517 mutex_init(&mddev->bitmap_info.mutex);
518 INIT_LIST_HEAD(&mddev->disks);
519 INIT_LIST_HEAD(&mddev->all_mddevs);
520 init_timer(&mddev->safemode_timer);
521 atomic_set(&mddev->active, 1);
522 atomic_set(&mddev->openers, 0);
523 atomic_set(&mddev->active_io, 0);
524 spin_lock_init(&mddev->write_lock);
525 atomic_set(&mddev->flush_pending, 0);
526 init_waitqueue_head(&mddev->sb_wait);
527 init_waitqueue_head(&mddev->recovery_wait);
528 mddev->reshape_position = MaxSector;
529 mddev->reshape_backwards = 0;
530 mddev->resync_min = 0;
531 mddev->resync_max = MaxSector;
532 mddev->level = LEVEL_NONE;
534 EXPORT_SYMBOL_GPL(mddev_init);
536 static struct mddev * mddev_find(dev_t unit)
538 struct mddev *mddev, *new = NULL;
540 if (unit && MAJOR(unit) != MD_MAJOR)
541 unit &= ~((1<<MdpMinorShift)-1);
544 spin_lock(&all_mddevs_lock);
547 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
548 if (mddev->unit == unit) {
550 spin_unlock(&all_mddevs_lock);
556 list_add(&new->all_mddevs, &all_mddevs);
557 spin_unlock(&all_mddevs_lock);
558 new->hold_active = UNTIL_IOCTL;
562 /* find an unused unit number */
563 static int next_minor = 512;
564 int start = next_minor;
568 dev = MKDEV(MD_MAJOR, next_minor);
570 if (next_minor > MINORMASK)
572 if (next_minor == start) {
573 /* Oh dear, all in use. */
574 spin_unlock(&all_mddevs_lock);
580 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
581 if (mddev->unit == dev) {
587 new->md_minor = MINOR(dev);
588 new->hold_active = UNTIL_STOP;
589 list_add(&new->all_mddevs, &all_mddevs);
590 spin_unlock(&all_mddevs_lock);
593 spin_unlock(&all_mddevs_lock);
595 new = kzalloc(sizeof(*new), GFP_KERNEL);
600 if (MAJOR(unit) == MD_MAJOR)
601 new->md_minor = MINOR(unit);
603 new->md_minor = MINOR(unit) >> MdpMinorShift;
610 static inline int mddev_lock(struct mddev * mddev)
612 return mutex_lock_interruptible(&mddev->reconfig_mutex);
615 static inline int mddev_is_locked(struct mddev *mddev)
617 return mutex_is_locked(&mddev->reconfig_mutex);
620 static inline int mddev_trylock(struct mddev * mddev)
622 return mutex_trylock(&mddev->reconfig_mutex);
625 static struct attribute_group md_redundancy_group;
627 static void mddev_unlock(struct mddev * mddev)
629 if (mddev->to_remove) {
630 /* These cannot be removed under reconfig_mutex as
631 * an access to the files will try to take reconfig_mutex
632 * while holding the file unremovable, which leads to
634 * So hold set sysfs_active while the remove in happeing,
635 * and anything else which might set ->to_remove or my
636 * otherwise change the sysfs namespace will fail with
637 * -EBUSY if sysfs_active is still set.
638 * We set sysfs_active under reconfig_mutex and elsewhere
639 * test it under the same mutex to ensure its correct value
642 struct attribute_group *to_remove = mddev->to_remove;
643 mddev->to_remove = NULL;
644 mddev->sysfs_active = 1;
645 mutex_unlock(&mddev->reconfig_mutex);
647 if (mddev->kobj.sd) {
648 if (to_remove != &md_redundancy_group)
649 sysfs_remove_group(&mddev->kobj, to_remove);
650 if (mddev->pers == NULL ||
651 mddev->pers->sync_request == NULL) {
652 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
653 if (mddev->sysfs_action)
654 sysfs_put(mddev->sysfs_action);
655 mddev->sysfs_action = NULL;
658 mddev->sysfs_active = 0;
660 mutex_unlock(&mddev->reconfig_mutex);
662 /* As we've dropped the mutex we need a spinlock to
663 * make sure the thread doesn't disappear
665 spin_lock(&pers_lock);
666 md_wakeup_thread(mddev->thread);
667 spin_unlock(&pers_lock);
670 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
672 struct md_rdev *rdev;
674 rdev_for_each(rdev, mddev)
675 if (rdev->desc_nr == nr)
681 static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
683 struct md_rdev *rdev;
685 rdev_for_each_rcu(rdev, mddev)
686 if (rdev->desc_nr == nr)
692 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
694 struct md_rdev *rdev;
696 rdev_for_each(rdev, mddev)
697 if (rdev->bdev->bd_dev == dev)
703 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
705 struct md_rdev *rdev;
707 rdev_for_each_rcu(rdev, mddev)
708 if (rdev->bdev->bd_dev == dev)
714 static struct md_personality *find_pers(int level, char *clevel)
716 struct md_personality *pers;
717 list_for_each_entry(pers, &pers_list, list) {
718 if (level != LEVEL_NONE && pers->level == level)
720 if (strcmp(pers->name, clevel)==0)
726 /* return the offset of the super block in 512byte sectors */
727 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
729 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
730 return MD_NEW_SIZE_SECTORS(num_sectors);
733 static int alloc_disk_sb(struct md_rdev * rdev)
738 rdev->sb_page = alloc_page(GFP_KERNEL);
739 if (!rdev->sb_page) {
740 printk(KERN_ALERT "md: out of memory.\n");
747 void md_rdev_clear(struct md_rdev *rdev)
750 put_page(rdev->sb_page);
752 rdev->sb_page = NULL;
757 put_page(rdev->bb_page);
758 rdev->bb_page = NULL;
760 kfree(rdev->badblocks.page);
761 rdev->badblocks.page = NULL;
763 EXPORT_SYMBOL_GPL(md_rdev_clear);
765 static void super_written(struct bio *bio, int error)
767 struct md_rdev *rdev = bio->bi_private;
768 struct mddev *mddev = rdev->mddev;
770 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
771 printk("md: super_written gets error=%d, uptodate=%d\n",
772 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
773 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
774 md_error(mddev, rdev);
777 if (atomic_dec_and_test(&mddev->pending_writes))
778 wake_up(&mddev->sb_wait);
782 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
783 sector_t sector, int size, struct page *page)
785 /* write first size bytes of page to sector of rdev
786 * Increment mddev->pending_writes before returning
787 * and decrement it on completion, waking up sb_wait
788 * if zero is reached.
789 * If an error occurred, call md_error
791 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
793 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
794 bio->bi_sector = sector;
795 bio_add_page(bio, page, size, 0);
796 bio->bi_private = rdev;
797 bio->bi_end_io = super_written;
799 atomic_inc(&mddev->pending_writes);
800 submit_bio(WRITE_FLUSH_FUA, bio);
803 void md_super_wait(struct mddev *mddev)
805 /* wait for all superblock writes that were scheduled to complete */
808 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
809 if (atomic_read(&mddev->pending_writes)==0)
813 finish_wait(&mddev->sb_wait, &wq);
816 static void bi_complete(struct bio *bio, int error)
818 complete((struct completion*)bio->bi_private);
821 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
822 struct page *page, int rw, bool metadata_op)
824 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
825 struct completion event;
830 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
831 rdev->meta_bdev : rdev->bdev;
833 bio->bi_sector = sector + rdev->sb_start;
834 else if (rdev->mddev->reshape_position != MaxSector &&
835 (rdev->mddev->reshape_backwards ==
836 (sector >= rdev->mddev->reshape_position)))
837 bio->bi_sector = sector + rdev->new_data_offset;
839 bio->bi_sector = sector + rdev->data_offset;
840 bio_add_page(bio, page, size, 0);
841 init_completion(&event);
842 bio->bi_private = &event;
843 bio->bi_end_io = bi_complete;
845 wait_for_completion(&event);
847 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
851 EXPORT_SYMBOL_GPL(sync_page_io);
853 static int read_disk_sb(struct md_rdev * rdev, int size)
855 char b[BDEVNAME_SIZE];
856 if (!rdev->sb_page) {
864 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
870 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
871 bdevname(rdev->bdev,b));
875 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
877 return sb1->set_uuid0 == sb2->set_uuid0 &&
878 sb1->set_uuid1 == sb2->set_uuid1 &&
879 sb1->set_uuid2 == sb2->set_uuid2 &&
880 sb1->set_uuid3 == sb2->set_uuid3;
883 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
886 mdp_super_t *tmp1, *tmp2;
888 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
889 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
891 if (!tmp1 || !tmp2) {
893 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
901 * nr_disks is not constant
906 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
914 static u32 md_csum_fold(u32 csum)
916 csum = (csum & 0xffff) + (csum >> 16);
917 return (csum & 0xffff) + (csum >> 16);
920 static unsigned int calc_sb_csum(mdp_super_t * sb)
923 u32 *sb32 = (u32*)sb;
925 unsigned int disk_csum, csum;
927 disk_csum = sb->sb_csum;
930 for (i = 0; i < MD_SB_BYTES/4 ; i++)
932 csum = (newcsum & 0xffffffff) + (newcsum>>32);
936 /* This used to use csum_partial, which was wrong for several
937 * reasons including that different results are returned on
938 * different architectures. It isn't critical that we get exactly
939 * the same return value as before (we always csum_fold before
940 * testing, and that removes any differences). However as we
941 * know that csum_partial always returned a 16bit value on
942 * alphas, do a fold to maximise conformity to previous behaviour.
944 sb->sb_csum = md_csum_fold(disk_csum);
946 sb->sb_csum = disk_csum;
953 * Handle superblock details.
954 * We want to be able to handle multiple superblock formats
955 * so we have a common interface to them all, and an array of
956 * different handlers.
957 * We rely on user-space to write the initial superblock, and support
958 * reading and updating of superblocks.
959 * Interface methods are:
960 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
961 * loads and validates a superblock on dev.
962 * if refdev != NULL, compare superblocks on both devices
964 * 0 - dev has a superblock that is compatible with refdev
965 * 1 - dev has a superblock that is compatible and newer than refdev
966 * so dev should be used as the refdev in future
967 * -EINVAL superblock incompatible or invalid
968 * -othererror e.g. -EIO
970 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
971 * Verify that dev is acceptable into mddev.
972 * The first time, mddev->raid_disks will be 0, and data from
973 * dev should be merged in. Subsequent calls check that dev
974 * is new enough. Return 0 or -EINVAL
976 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
977 * Update the superblock for rdev with data in mddev
978 * This does not write to disc.
984 struct module *owner;
985 int (*load_super)(struct md_rdev *rdev,
986 struct md_rdev *refdev,
988 int (*validate_super)(struct mddev *mddev,
989 struct md_rdev *rdev);
990 void (*sync_super)(struct mddev *mddev,
991 struct md_rdev *rdev);
992 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
993 sector_t num_sectors);
994 int (*allow_new_offset)(struct md_rdev *rdev,
995 unsigned long long new_offset);
999 * Check that the given mddev has no bitmap.
1001 * This function is called from the run method of all personalities that do not
1002 * support bitmaps. It prints an error message and returns non-zero if mddev
1003 * has a bitmap. Otherwise, it returns 0.
1006 int md_check_no_bitmap(struct mddev *mddev)
1008 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1010 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1011 mdname(mddev), mddev->pers->name);
1014 EXPORT_SYMBOL(md_check_no_bitmap);
1017 * load_super for 0.90.0
1019 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1021 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1026 * Calculate the position of the superblock (512byte sectors),
1027 * it's at the end of the disk.
1029 * It also happens to be a multiple of 4Kb.
1031 rdev->sb_start = calc_dev_sboffset(rdev);
1033 ret = read_disk_sb(rdev, MD_SB_BYTES);
1034 if (ret) return ret;
1038 bdevname(rdev->bdev, b);
1039 sb = page_address(rdev->sb_page);
1041 if (sb->md_magic != MD_SB_MAGIC) {
1042 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1047 if (sb->major_version != 0 ||
1048 sb->minor_version < 90 ||
1049 sb->minor_version > 91) {
1050 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1051 sb->major_version, sb->minor_version,
1056 if (sb->raid_disks <= 0)
1059 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1060 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1065 rdev->preferred_minor = sb->md_minor;
1066 rdev->data_offset = 0;
1067 rdev->new_data_offset = 0;
1068 rdev->sb_size = MD_SB_BYTES;
1069 rdev->badblocks.shift = -1;
1071 if (sb->level == LEVEL_MULTIPATH)
1074 rdev->desc_nr = sb->this_disk.number;
1080 mdp_super_t *refsb = page_address(refdev->sb_page);
1081 if (!uuid_equal(refsb, sb)) {
1082 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1083 b, bdevname(refdev->bdev,b2));
1086 if (!sb_equal(refsb, sb)) {
1087 printk(KERN_WARNING "md: %s has same UUID"
1088 " but different superblock to %s\n",
1089 b, bdevname(refdev->bdev, b2));
1093 ev2 = md_event(refsb);
1099 rdev->sectors = rdev->sb_start;
1100 /* Limit to 4TB as metadata cannot record more than that.
1101 * (not needed for Linear and RAID0 as metadata doesn't
1104 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1105 rdev->sectors = (2ULL << 32) - 2;
1107 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1108 /* "this cannot possibly happen" ... */
1116 * validate_super for 0.90.0
1118 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1121 mdp_super_t *sb = page_address(rdev->sb_page);
1122 __u64 ev1 = md_event(sb);
1124 rdev->raid_disk = -1;
1125 clear_bit(Faulty, &rdev->flags);
1126 clear_bit(In_sync, &rdev->flags);
1127 clear_bit(WriteMostly, &rdev->flags);
1129 if (mddev->raid_disks == 0) {
1130 mddev->major_version = 0;
1131 mddev->minor_version = sb->minor_version;
1132 mddev->patch_version = sb->patch_version;
1133 mddev->external = 0;
1134 mddev->chunk_sectors = sb->chunk_size >> 9;
1135 mddev->ctime = sb->ctime;
1136 mddev->utime = sb->utime;
1137 mddev->level = sb->level;
1138 mddev->clevel[0] = 0;
1139 mddev->layout = sb->layout;
1140 mddev->raid_disks = sb->raid_disks;
1141 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1142 mddev->events = ev1;
1143 mddev->bitmap_info.offset = 0;
1144 mddev->bitmap_info.space = 0;
1145 /* bitmap can use 60 K after the 4K superblocks */
1146 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1147 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1148 mddev->reshape_backwards = 0;
1150 if (mddev->minor_version >= 91) {
1151 mddev->reshape_position = sb->reshape_position;
1152 mddev->delta_disks = sb->delta_disks;
1153 mddev->new_level = sb->new_level;
1154 mddev->new_layout = sb->new_layout;
1155 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1156 if (mddev->delta_disks < 0)
1157 mddev->reshape_backwards = 1;
1159 mddev->reshape_position = MaxSector;
1160 mddev->delta_disks = 0;
1161 mddev->new_level = mddev->level;
1162 mddev->new_layout = mddev->layout;
1163 mddev->new_chunk_sectors = mddev->chunk_sectors;
1166 if (sb->state & (1<<MD_SB_CLEAN))
1167 mddev->recovery_cp = MaxSector;
1169 if (sb->events_hi == sb->cp_events_hi &&
1170 sb->events_lo == sb->cp_events_lo) {
1171 mddev->recovery_cp = sb->recovery_cp;
1173 mddev->recovery_cp = 0;
1176 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1177 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1178 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1179 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1181 mddev->max_disks = MD_SB_DISKS;
1183 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1184 mddev->bitmap_info.file == NULL) {
1185 mddev->bitmap_info.offset =
1186 mddev->bitmap_info.default_offset;
1187 mddev->bitmap_info.space =
1188 mddev->bitmap_info.space;
1191 } else if (mddev->pers == NULL) {
1192 /* Insist on good event counter while assembling, except
1193 * for spares (which don't need an event count) */
1195 if (sb->disks[rdev->desc_nr].state & (
1196 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1197 if (ev1 < mddev->events)
1199 } else if (mddev->bitmap) {
1200 /* if adding to array with a bitmap, then we can accept an
1201 * older device ... but not too old.
1203 if (ev1 < mddev->bitmap->events_cleared)
1206 if (ev1 < mddev->events)
1207 /* just a hot-add of a new device, leave raid_disk at -1 */
1211 if (mddev->level != LEVEL_MULTIPATH) {
1212 desc = sb->disks + rdev->desc_nr;
1214 if (desc->state & (1<<MD_DISK_FAULTY))
1215 set_bit(Faulty, &rdev->flags);
1216 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1217 desc->raid_disk < mddev->raid_disks */) {
1218 set_bit(In_sync, &rdev->flags);
1219 rdev->raid_disk = desc->raid_disk;
1220 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1221 /* active but not in sync implies recovery up to
1222 * reshape position. We don't know exactly where
1223 * that is, so set to zero for now */
1224 if (mddev->minor_version >= 91) {
1225 rdev->recovery_offset = 0;
1226 rdev->raid_disk = desc->raid_disk;
1229 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1230 set_bit(WriteMostly, &rdev->flags);
1231 } else /* MULTIPATH are always insync */
1232 set_bit(In_sync, &rdev->flags);
1237 * sync_super for 0.90.0
1239 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1242 struct md_rdev *rdev2;
1243 int next_spare = mddev->raid_disks;
1246 /* make rdev->sb match mddev data..
1249 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1250 * 3/ any empty disks < next_spare become removed
1252 * disks[0] gets initialised to REMOVED because
1253 * we cannot be sure from other fields if it has
1254 * been initialised or not.
1257 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1259 rdev->sb_size = MD_SB_BYTES;
1261 sb = page_address(rdev->sb_page);
1263 memset(sb, 0, sizeof(*sb));
1265 sb->md_magic = MD_SB_MAGIC;
1266 sb->major_version = mddev->major_version;
1267 sb->patch_version = mddev->patch_version;
1268 sb->gvalid_words = 0; /* ignored */
1269 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1270 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1271 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1272 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1274 sb->ctime = mddev->ctime;
1275 sb->level = mddev->level;
1276 sb->size = mddev->dev_sectors / 2;
1277 sb->raid_disks = mddev->raid_disks;
1278 sb->md_minor = mddev->md_minor;
1279 sb->not_persistent = 0;
1280 sb->utime = mddev->utime;
1282 sb->events_hi = (mddev->events>>32);
1283 sb->events_lo = (u32)mddev->events;
1285 if (mddev->reshape_position == MaxSector)
1286 sb->minor_version = 90;
1288 sb->minor_version = 91;
1289 sb->reshape_position = mddev->reshape_position;
1290 sb->new_level = mddev->new_level;
1291 sb->delta_disks = mddev->delta_disks;
1292 sb->new_layout = mddev->new_layout;
1293 sb->new_chunk = mddev->new_chunk_sectors << 9;
1295 mddev->minor_version = sb->minor_version;
1298 sb->recovery_cp = mddev->recovery_cp;
1299 sb->cp_events_hi = (mddev->events>>32);
1300 sb->cp_events_lo = (u32)mddev->events;
1301 if (mddev->recovery_cp == MaxSector)
1302 sb->state = (1<< MD_SB_CLEAN);
1304 sb->recovery_cp = 0;
1306 sb->layout = mddev->layout;
1307 sb->chunk_size = mddev->chunk_sectors << 9;
1309 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1310 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1312 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1313 rdev_for_each(rdev2, mddev) {
1316 int is_active = test_bit(In_sync, &rdev2->flags);
1318 if (rdev2->raid_disk >= 0 &&
1319 sb->minor_version >= 91)
1320 /* we have nowhere to store the recovery_offset,
1321 * but if it is not below the reshape_position,
1322 * we can piggy-back on that.
1325 if (rdev2->raid_disk < 0 ||
1326 test_bit(Faulty, &rdev2->flags))
1329 desc_nr = rdev2->raid_disk;
1331 desc_nr = next_spare++;
1332 rdev2->desc_nr = desc_nr;
1333 d = &sb->disks[rdev2->desc_nr];
1335 d->number = rdev2->desc_nr;
1336 d->major = MAJOR(rdev2->bdev->bd_dev);
1337 d->minor = MINOR(rdev2->bdev->bd_dev);
1339 d->raid_disk = rdev2->raid_disk;
1341 d->raid_disk = rdev2->desc_nr; /* compatibility */
1342 if (test_bit(Faulty, &rdev2->flags))
1343 d->state = (1<<MD_DISK_FAULTY);
1344 else if (is_active) {
1345 d->state = (1<<MD_DISK_ACTIVE);
1346 if (test_bit(In_sync, &rdev2->flags))
1347 d->state |= (1<<MD_DISK_SYNC);
1355 if (test_bit(WriteMostly, &rdev2->flags))
1356 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1358 /* now set the "removed" and "faulty" bits on any missing devices */
1359 for (i=0 ; i < mddev->raid_disks ; i++) {
1360 mdp_disk_t *d = &sb->disks[i];
1361 if (d->state == 0 && d->number == 0) {
1364 d->state = (1<<MD_DISK_REMOVED);
1365 d->state |= (1<<MD_DISK_FAULTY);
1369 sb->nr_disks = nr_disks;
1370 sb->active_disks = active;
1371 sb->working_disks = working;
1372 sb->failed_disks = failed;
1373 sb->spare_disks = spare;
1375 sb->this_disk = sb->disks[rdev->desc_nr];
1376 sb->sb_csum = calc_sb_csum(sb);
1380 * rdev_size_change for 0.90.0
1382 static unsigned long long
1383 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1385 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1386 return 0; /* component must fit device */
1387 if (rdev->mddev->bitmap_info.offset)
1388 return 0; /* can't move bitmap */
1389 rdev->sb_start = calc_dev_sboffset(rdev);
1390 if (!num_sectors || num_sectors > rdev->sb_start)
1391 num_sectors = rdev->sb_start;
1392 /* Limit to 4TB as metadata cannot record more than that.
1393 * 4TB == 2^32 KB, or 2*2^32 sectors.
1395 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1396 num_sectors = (2ULL << 32) - 2;
1397 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1399 md_super_wait(rdev->mddev);
1404 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1406 /* non-zero offset changes not possible with v0.90 */
1407 return new_offset == 0;
1411 * version 1 superblock
1414 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1418 unsigned long long newcsum;
1419 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1420 __le32 *isuper = (__le32*)sb;
1422 disk_csum = sb->sb_csum;
1425 for (; size >= 4; size -= 4)
1426 newcsum += le32_to_cpu(*isuper++);
1429 newcsum += le16_to_cpu(*(__le16*) isuper);
1431 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1432 sb->sb_csum = disk_csum;
1433 return cpu_to_le32(csum);
1436 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1438 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1440 struct mdp_superblock_1 *sb;
1444 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1448 * Calculate the position of the superblock in 512byte sectors.
1449 * It is always aligned to a 4K boundary and
1450 * depeding on minor_version, it can be:
1451 * 0: At least 8K, but less than 12K, from end of device
1452 * 1: At start of device
1453 * 2: 4K from start of device.
1455 switch(minor_version) {
1457 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1459 sb_start &= ~(sector_t)(4*2-1);
1470 rdev->sb_start = sb_start;
1472 /* superblock is rarely larger than 1K, but it can be larger,
1473 * and it is safe to read 4k, so we do that
1475 ret = read_disk_sb(rdev, 4096);
1476 if (ret) return ret;
1479 sb = page_address(rdev->sb_page);
1481 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1482 sb->major_version != cpu_to_le32(1) ||
1483 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1484 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1485 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1488 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1489 printk("md: invalid superblock checksum on %s\n",
1490 bdevname(rdev->bdev,b));
1493 if (le64_to_cpu(sb->data_size) < 10) {
1494 printk("md: data_size too small on %s\n",
1495 bdevname(rdev->bdev,b));
1500 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1501 /* Some padding is non-zero, might be a new feature */
1504 rdev->preferred_minor = 0xffff;
1505 rdev->data_offset = le64_to_cpu(sb->data_offset);
1506 rdev->new_data_offset = rdev->data_offset;
1507 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1508 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1509 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1510 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1512 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1513 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1514 if (rdev->sb_size & bmask)
1515 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1518 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1521 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1524 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1527 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1529 if (!rdev->bb_page) {
1530 rdev->bb_page = alloc_page(GFP_KERNEL);
1534 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1535 rdev->badblocks.count == 0) {
1536 /* need to load the bad block list.
1537 * Currently we limit it to one page.
1543 int sectors = le16_to_cpu(sb->bblog_size);
1544 if (sectors > (PAGE_SIZE / 512))
1546 offset = le32_to_cpu(sb->bblog_offset);
1549 bb_sector = (long long)offset;
1550 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1551 rdev->bb_page, READ, true))
1553 bbp = (u64 *)page_address(rdev->bb_page);
1554 rdev->badblocks.shift = sb->bblog_shift;
1555 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1556 u64 bb = le64_to_cpu(*bbp);
1557 int count = bb & (0x3ff);
1558 u64 sector = bb >> 10;
1559 sector <<= sb->bblog_shift;
1560 count <<= sb->bblog_shift;
1563 if (md_set_badblocks(&rdev->badblocks,
1564 sector, count, 1) == 0)
1567 } else if (sb->bblog_offset == 0)
1568 rdev->badblocks.shift = -1;
1574 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1576 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1577 sb->level != refsb->level ||
1578 sb->layout != refsb->layout ||
1579 sb->chunksize != refsb->chunksize) {
1580 printk(KERN_WARNING "md: %s has strangely different"
1581 " superblock to %s\n",
1582 bdevname(rdev->bdev,b),
1583 bdevname(refdev->bdev,b2));
1586 ev1 = le64_to_cpu(sb->events);
1587 ev2 = le64_to_cpu(refsb->events);
1594 if (minor_version) {
1595 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1596 sectors -= rdev->data_offset;
1598 sectors = rdev->sb_start;
1599 if (sectors < le64_to_cpu(sb->data_size))
1601 rdev->sectors = le64_to_cpu(sb->data_size);
1605 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1607 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1608 __u64 ev1 = le64_to_cpu(sb->events);
1610 rdev->raid_disk = -1;
1611 clear_bit(Faulty, &rdev->flags);
1612 clear_bit(In_sync, &rdev->flags);
1613 clear_bit(WriteMostly, &rdev->flags);
1615 if (mddev->raid_disks == 0) {
1616 mddev->major_version = 1;
1617 mddev->patch_version = 0;
1618 mddev->external = 0;
1619 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1620 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1621 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1622 mddev->level = le32_to_cpu(sb->level);
1623 mddev->clevel[0] = 0;
1624 mddev->layout = le32_to_cpu(sb->layout);
1625 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1626 mddev->dev_sectors = le64_to_cpu(sb->size);
1627 mddev->events = ev1;
1628 mddev->bitmap_info.offset = 0;
1629 mddev->bitmap_info.space = 0;
1630 /* Default location for bitmap is 1K after superblock
1631 * using 3K - total of 4K
1633 mddev->bitmap_info.default_offset = 1024 >> 9;
1634 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1635 mddev->reshape_backwards = 0;
1637 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1638 memcpy(mddev->uuid, sb->set_uuid, 16);
1640 mddev->max_disks = (4096-256)/2;
1642 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1643 mddev->bitmap_info.file == NULL) {
1644 mddev->bitmap_info.offset =
1645 (__s32)le32_to_cpu(sb->bitmap_offset);
1646 /* Metadata doesn't record how much space is available.
1647 * For 1.0, we assume we can use up to the superblock
1648 * if before, else to 4K beyond superblock.
1649 * For others, assume no change is possible.
1651 if (mddev->minor_version > 0)
1652 mddev->bitmap_info.space = 0;
1653 else if (mddev->bitmap_info.offset > 0)
1654 mddev->bitmap_info.space =
1655 8 - mddev->bitmap_info.offset;
1657 mddev->bitmap_info.space =
1658 -mddev->bitmap_info.offset;
1661 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1662 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1663 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1664 mddev->new_level = le32_to_cpu(sb->new_level);
1665 mddev->new_layout = le32_to_cpu(sb->new_layout);
1666 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1667 if (mddev->delta_disks < 0 ||
1668 (mddev->delta_disks == 0 &&
1669 (le32_to_cpu(sb->feature_map)
1670 & MD_FEATURE_RESHAPE_BACKWARDS)))
1671 mddev->reshape_backwards = 1;
1673 mddev->reshape_position = MaxSector;
1674 mddev->delta_disks = 0;
1675 mddev->new_level = mddev->level;
1676 mddev->new_layout = mddev->layout;
1677 mddev->new_chunk_sectors = mddev->chunk_sectors;
1680 } else if (mddev->pers == NULL) {
1681 /* Insist of good event counter while assembling, except for
1682 * spares (which don't need an event count) */
1684 if (rdev->desc_nr >= 0 &&
1685 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1686 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1687 if (ev1 < mddev->events)
1689 } else if (mddev->bitmap) {
1690 /* If adding to array with a bitmap, then we can accept an
1691 * older device, but not too old.
1693 if (ev1 < mddev->bitmap->events_cleared)
1696 if (ev1 < mddev->events)
1697 /* just a hot-add of a new device, leave raid_disk at -1 */
1700 if (mddev->level != LEVEL_MULTIPATH) {
1702 if (rdev->desc_nr < 0 ||
1703 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1707 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1709 case 0xffff: /* spare */
1711 case 0xfffe: /* faulty */
1712 set_bit(Faulty, &rdev->flags);
1715 if ((le32_to_cpu(sb->feature_map) &
1716 MD_FEATURE_RECOVERY_OFFSET))
1717 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1719 set_bit(In_sync, &rdev->flags);
1720 rdev->raid_disk = role;
1723 if (sb->devflags & WriteMostly1)
1724 set_bit(WriteMostly, &rdev->flags);
1725 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1726 set_bit(Replacement, &rdev->flags);
1727 } else /* MULTIPATH are always insync */
1728 set_bit(In_sync, &rdev->flags);
1733 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1735 struct mdp_superblock_1 *sb;
1736 struct md_rdev *rdev2;
1738 /* make rdev->sb match mddev and rdev data. */
1740 sb = page_address(rdev->sb_page);
1742 sb->feature_map = 0;
1744 sb->recovery_offset = cpu_to_le64(0);
1745 memset(sb->pad3, 0, sizeof(sb->pad3));
1747 sb->utime = cpu_to_le64((__u64)mddev->utime);
1748 sb->events = cpu_to_le64(mddev->events);
1750 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1752 sb->resync_offset = cpu_to_le64(0);
1754 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1756 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1757 sb->size = cpu_to_le64(mddev->dev_sectors);
1758 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1759 sb->level = cpu_to_le32(mddev->level);
1760 sb->layout = cpu_to_le32(mddev->layout);
1762 if (test_bit(WriteMostly, &rdev->flags))
1763 sb->devflags |= WriteMostly1;
1765 sb->devflags &= ~WriteMostly1;
1766 sb->data_offset = cpu_to_le64(rdev->data_offset);
1767 sb->data_size = cpu_to_le64(rdev->sectors);
1769 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1770 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1771 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1774 if (rdev->raid_disk >= 0 &&
1775 !test_bit(In_sync, &rdev->flags)) {
1777 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1778 sb->recovery_offset =
1779 cpu_to_le64(rdev->recovery_offset);
1781 if (test_bit(Replacement, &rdev->flags))
1783 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1785 if (mddev->reshape_position != MaxSector) {
1786 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1787 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1788 sb->new_layout = cpu_to_le32(mddev->new_layout);
1789 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1790 sb->new_level = cpu_to_le32(mddev->new_level);
1791 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1792 if (mddev->delta_disks == 0 &&
1793 mddev->reshape_backwards)
1795 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1796 if (rdev->new_data_offset != rdev->data_offset) {
1798 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1799 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1800 - rdev->data_offset));
1804 if (rdev->badblocks.count == 0)
1805 /* Nothing to do for bad blocks*/ ;
1806 else if (sb->bblog_offset == 0)
1807 /* Cannot record bad blocks on this device */
1808 md_error(mddev, rdev);
1810 struct badblocks *bb = &rdev->badblocks;
1811 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1813 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1818 seq = read_seqbegin(&bb->lock);
1820 memset(bbp, 0xff, PAGE_SIZE);
1822 for (i = 0 ; i < bb->count ; i++) {
1823 u64 internal_bb = p[i];
1824 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1825 | BB_LEN(internal_bb));
1826 bbp[i] = cpu_to_le64(store_bb);
1829 if (read_seqretry(&bb->lock, seq))
1832 bb->sector = (rdev->sb_start +
1833 (int)le32_to_cpu(sb->bblog_offset));
1834 bb->size = le16_to_cpu(sb->bblog_size);
1839 rdev_for_each(rdev2, mddev)
1840 if (rdev2->desc_nr+1 > max_dev)
1841 max_dev = rdev2->desc_nr+1;
1843 if (max_dev > le32_to_cpu(sb->max_dev)) {
1845 sb->max_dev = cpu_to_le32(max_dev);
1846 rdev->sb_size = max_dev * 2 + 256;
1847 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1848 if (rdev->sb_size & bmask)
1849 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1851 max_dev = le32_to_cpu(sb->max_dev);
1853 for (i=0; i<max_dev;i++)
1854 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1856 rdev_for_each(rdev2, mddev) {
1858 if (test_bit(Faulty, &rdev2->flags))
1859 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1860 else if (test_bit(In_sync, &rdev2->flags))
1861 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1862 else if (rdev2->raid_disk >= 0)
1863 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1865 sb->dev_roles[i] = cpu_to_le16(0xffff);
1868 sb->sb_csum = calc_sb_1_csum(sb);
1871 static unsigned long long
1872 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1874 struct mdp_superblock_1 *sb;
1875 sector_t max_sectors;
1876 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1877 return 0; /* component must fit device */
1878 if (rdev->data_offset != rdev->new_data_offset)
1879 return 0; /* too confusing */
1880 if (rdev->sb_start < rdev->data_offset) {
1881 /* minor versions 1 and 2; superblock before data */
1882 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1883 max_sectors -= rdev->data_offset;
1884 if (!num_sectors || num_sectors > max_sectors)
1885 num_sectors = max_sectors;
1886 } else if (rdev->mddev->bitmap_info.offset) {
1887 /* minor version 0 with bitmap we can't move */
1890 /* minor version 0; superblock after data */
1892 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1893 sb_start &= ~(sector_t)(4*2 - 1);
1894 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1895 if (!num_sectors || num_sectors > max_sectors)
1896 num_sectors = max_sectors;
1897 rdev->sb_start = sb_start;
1899 sb = page_address(rdev->sb_page);
1900 sb->data_size = cpu_to_le64(num_sectors);
1901 sb->super_offset = rdev->sb_start;
1902 sb->sb_csum = calc_sb_1_csum(sb);
1903 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1905 md_super_wait(rdev->mddev);
1911 super_1_allow_new_offset(struct md_rdev *rdev,
1912 unsigned long long new_offset)
1914 /* All necessary checks on new >= old have been done */
1915 struct bitmap *bitmap;
1916 if (new_offset >= rdev->data_offset)
1919 /* with 1.0 metadata, there is no metadata to tread on
1920 * so we can always move back */
1921 if (rdev->mddev->minor_version == 0)
1924 /* otherwise we must be sure not to step on
1925 * any metadata, so stay:
1926 * 36K beyond start of superblock
1927 * beyond end of badblocks
1928 * beyond write-intent bitmap
1930 if (rdev->sb_start + (32+4)*2 > new_offset)
1932 bitmap = rdev->mddev->bitmap;
1933 if (bitmap && !rdev->mddev->bitmap_info.file &&
1934 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1935 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1937 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1943 static struct super_type super_types[] = {
1946 .owner = THIS_MODULE,
1947 .load_super = super_90_load,
1948 .validate_super = super_90_validate,
1949 .sync_super = super_90_sync,
1950 .rdev_size_change = super_90_rdev_size_change,
1951 .allow_new_offset = super_90_allow_new_offset,
1955 .owner = THIS_MODULE,
1956 .load_super = super_1_load,
1957 .validate_super = super_1_validate,
1958 .sync_super = super_1_sync,
1959 .rdev_size_change = super_1_rdev_size_change,
1960 .allow_new_offset = super_1_allow_new_offset,
1964 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1966 if (mddev->sync_super) {
1967 mddev->sync_super(mddev, rdev);
1971 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1973 super_types[mddev->major_version].sync_super(mddev, rdev);
1976 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1978 struct md_rdev *rdev, *rdev2;
1981 rdev_for_each_rcu(rdev, mddev1)
1982 rdev_for_each_rcu(rdev2, mddev2)
1983 if (rdev->bdev->bd_contains ==
1984 rdev2->bdev->bd_contains) {
1992 static LIST_HEAD(pending_raid_disks);
1995 * Try to register data integrity profile for an mddev
1997 * This is called when an array is started and after a disk has been kicked
1998 * from the array. It only succeeds if all working and active component devices
1999 * are integrity capable with matching profiles.
2001 int md_integrity_register(struct mddev *mddev)
2003 struct md_rdev *rdev, *reference = NULL;
2005 if (list_empty(&mddev->disks))
2006 return 0; /* nothing to do */
2007 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2008 return 0; /* shouldn't register, or already is */
2009 rdev_for_each(rdev, mddev) {
2010 /* skip spares and non-functional disks */
2011 if (test_bit(Faulty, &rdev->flags))
2013 if (rdev->raid_disk < 0)
2016 /* Use the first rdev as the reference */
2020 /* does this rdev's profile match the reference profile? */
2021 if (blk_integrity_compare(reference->bdev->bd_disk,
2022 rdev->bdev->bd_disk) < 0)
2025 if (!reference || !bdev_get_integrity(reference->bdev))
2028 * All component devices are integrity capable and have matching
2029 * profiles, register the common profile for the md device.
2031 if (blk_integrity_register(mddev->gendisk,
2032 bdev_get_integrity(reference->bdev)) != 0) {
2033 printk(KERN_ERR "md: failed to register integrity for %s\n",
2037 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2038 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2039 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2045 EXPORT_SYMBOL(md_integrity_register);
2047 /* Disable data integrity if non-capable/non-matching disk is being added */
2048 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2050 struct blk_integrity *bi_rdev;
2051 struct blk_integrity *bi_mddev;
2053 if (!mddev->gendisk)
2056 bi_rdev = bdev_get_integrity(rdev->bdev);
2057 bi_mddev = blk_get_integrity(mddev->gendisk);
2059 if (!bi_mddev) /* nothing to do */
2061 if (rdev->raid_disk < 0) /* skip spares */
2063 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2064 rdev->bdev->bd_disk) >= 0)
2066 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2067 blk_integrity_unregister(mddev->gendisk);
2069 EXPORT_SYMBOL(md_integrity_add_rdev);
2071 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2073 char b[BDEVNAME_SIZE];
2083 /* prevent duplicates */
2084 if (find_rdev(mddev, rdev->bdev->bd_dev))
2087 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2088 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2089 rdev->sectors < mddev->dev_sectors)) {
2091 /* Cannot change size, so fail
2092 * If mddev->level <= 0, then we don't care
2093 * about aligning sizes (e.g. linear)
2095 if (mddev->level > 0)
2098 mddev->dev_sectors = rdev->sectors;
2101 /* Verify rdev->desc_nr is unique.
2102 * If it is -1, assign a free number, else
2103 * check number is not in use
2105 if (rdev->desc_nr < 0) {
2107 if (mddev->pers) choice = mddev->raid_disks;
2108 while (find_rdev_nr(mddev, choice))
2110 rdev->desc_nr = choice;
2112 if (find_rdev_nr(mddev, rdev->desc_nr))
2115 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2116 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2117 mdname(mddev), mddev->max_disks);
2120 bdevname(rdev->bdev,b);
2121 while ( (s=strchr(b, '/')) != NULL)
2124 rdev->mddev = mddev;
2125 printk(KERN_INFO "md: bind<%s>\n", b);
2127 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2130 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2131 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2132 /* failure here is OK */;
2133 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2135 list_add_rcu(&rdev->same_set, &mddev->disks);
2136 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2138 /* May as well allow recovery to be retried once */
2139 mddev->recovery_disabled++;
2144 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2149 static void md_delayed_delete(struct work_struct *ws)
2151 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2152 kobject_del(&rdev->kobj);
2153 kobject_put(&rdev->kobj);
2156 static void unbind_rdev_from_array(struct md_rdev * rdev)
2158 char b[BDEVNAME_SIZE];
2163 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2164 list_del_rcu(&rdev->same_set);
2165 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2167 sysfs_remove_link(&rdev->kobj, "block");
2168 sysfs_put(rdev->sysfs_state);
2169 rdev->sysfs_state = NULL;
2170 rdev->badblocks.count = 0;
2171 /* We need to delay this, otherwise we can deadlock when
2172 * writing to 'remove' to "dev/state". We also need
2173 * to delay it due to rcu usage.
2176 INIT_WORK(&rdev->del_work, md_delayed_delete);
2177 kobject_get(&rdev->kobj);
2178 queue_work(md_misc_wq, &rdev->del_work);
2182 * prevent the device from being mounted, repartitioned or
2183 * otherwise reused by a RAID array (or any other kernel
2184 * subsystem), by bd_claiming the device.
2186 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2189 struct block_device *bdev;
2190 char b[BDEVNAME_SIZE];
2192 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2193 shared ? (struct md_rdev *)lock_rdev : rdev);
2195 printk(KERN_ERR "md: could not open %s.\n",
2196 __bdevname(dev, b));
2197 return PTR_ERR(bdev);
2203 static void unlock_rdev(struct md_rdev *rdev)
2205 struct block_device *bdev = rdev->bdev;
2209 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2212 void md_autodetect_dev(dev_t dev);
2214 static void export_rdev(struct md_rdev * rdev)
2216 char b[BDEVNAME_SIZE];
2217 printk(KERN_INFO "md: export_rdev(%s)\n",
2218 bdevname(rdev->bdev,b));
2221 md_rdev_clear(rdev);
2223 if (test_bit(AutoDetected, &rdev->flags))
2224 md_autodetect_dev(rdev->bdev->bd_dev);
2227 kobject_put(&rdev->kobj);
2230 static void kick_rdev_from_array(struct md_rdev * rdev)
2232 unbind_rdev_from_array(rdev);
2236 static void export_array(struct mddev *mddev)
2238 struct md_rdev *rdev, *tmp;
2240 rdev_for_each_safe(rdev, tmp, mddev) {
2245 kick_rdev_from_array(rdev);
2247 if (!list_empty(&mddev->disks))
2249 mddev->raid_disks = 0;
2250 mddev->major_version = 0;
2253 static void print_desc(mdp_disk_t *desc)
2255 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2256 desc->major,desc->minor,desc->raid_disk,desc->state);
2259 static void print_sb_90(mdp_super_t *sb)
2264 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2265 sb->major_version, sb->minor_version, sb->patch_version,
2266 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2268 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2269 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2270 sb->md_minor, sb->layout, sb->chunk_size);
2271 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2272 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2273 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2274 sb->failed_disks, sb->spare_disks,
2275 sb->sb_csum, (unsigned long)sb->events_lo);
2278 for (i = 0; i < MD_SB_DISKS; i++) {
2281 desc = sb->disks + i;
2282 if (desc->number || desc->major || desc->minor ||
2283 desc->raid_disk || (desc->state && (desc->state != 4))) {
2284 printk(" D %2d: ", i);
2288 printk(KERN_INFO "md: THIS: ");
2289 print_desc(&sb->this_disk);
2292 static void print_sb_1(struct mdp_superblock_1 *sb)
2296 uuid = sb->set_uuid;
2298 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2299 "md: Name: \"%s\" CT:%llu\n",
2300 le32_to_cpu(sb->major_version),
2301 le32_to_cpu(sb->feature_map),
2304 (unsigned long long)le64_to_cpu(sb->ctime)
2305 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2307 uuid = sb->device_uuid;
2309 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2311 "md: Dev:%08x UUID: %pU\n"
2312 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2313 "md: (MaxDev:%u) \n",
2314 le32_to_cpu(sb->level),
2315 (unsigned long long)le64_to_cpu(sb->size),
2316 le32_to_cpu(sb->raid_disks),
2317 le32_to_cpu(sb->layout),
2318 le32_to_cpu(sb->chunksize),
2319 (unsigned long long)le64_to_cpu(sb->data_offset),
2320 (unsigned long long)le64_to_cpu(sb->data_size),
2321 (unsigned long long)le64_to_cpu(sb->super_offset),
2322 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2323 le32_to_cpu(sb->dev_number),
2326 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2327 (unsigned long long)le64_to_cpu(sb->events),
2328 (unsigned long long)le64_to_cpu(sb->resync_offset),
2329 le32_to_cpu(sb->sb_csum),
2330 le32_to_cpu(sb->max_dev)
2334 static void print_rdev(struct md_rdev *rdev, int major_version)
2336 char b[BDEVNAME_SIZE];
2337 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2338 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2339 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2341 if (rdev->sb_loaded) {
2342 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2343 switch (major_version) {
2345 print_sb_90(page_address(rdev->sb_page));
2348 print_sb_1(page_address(rdev->sb_page));
2352 printk(KERN_INFO "md: no rdev superblock!\n");
2355 static void md_print_devices(void)
2357 struct list_head *tmp;
2358 struct md_rdev *rdev;
2359 struct mddev *mddev;
2360 char b[BDEVNAME_SIZE];
2363 printk("md: **********************************\n");
2364 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2365 printk("md: **********************************\n");
2366 for_each_mddev(mddev, tmp) {
2369 bitmap_print_sb(mddev->bitmap);
2371 printk("%s: ", mdname(mddev));
2372 rdev_for_each(rdev, mddev)
2373 printk("<%s>", bdevname(rdev->bdev,b));
2376 rdev_for_each(rdev, mddev)
2377 print_rdev(rdev, mddev->major_version);
2379 printk("md: **********************************\n");
2384 static void sync_sbs(struct mddev * mddev, int nospares)
2386 /* Update each superblock (in-memory image), but
2387 * if we are allowed to, skip spares which already
2388 * have the right event counter, or have one earlier
2389 * (which would mean they aren't being marked as dirty
2390 * with the rest of the array)
2392 struct md_rdev *rdev;
2393 rdev_for_each(rdev, mddev) {
2394 if (rdev->sb_events == mddev->events ||
2396 rdev->raid_disk < 0 &&
2397 rdev->sb_events+1 == mddev->events)) {
2398 /* Don't update this superblock */
2399 rdev->sb_loaded = 2;
2401 sync_super(mddev, rdev);
2402 rdev->sb_loaded = 1;
2407 static void md_update_sb(struct mddev * mddev, int force_change)
2409 struct md_rdev *rdev;
2412 int any_badblocks_changed = 0;
2416 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2420 /* First make sure individual recovery_offsets are correct */
2421 rdev_for_each(rdev, mddev) {
2422 if (rdev->raid_disk >= 0 &&
2423 mddev->delta_disks >= 0 &&
2424 !test_bit(In_sync, &rdev->flags) &&
2425 mddev->curr_resync_completed > rdev->recovery_offset)
2426 rdev->recovery_offset = mddev->curr_resync_completed;
2429 if (!mddev->persistent) {
2430 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2431 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2432 if (!mddev->external) {
2433 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2434 rdev_for_each(rdev, mddev) {
2435 if (rdev->badblocks.changed) {
2436 rdev->badblocks.changed = 0;
2437 md_ack_all_badblocks(&rdev->badblocks);
2438 md_error(mddev, rdev);
2440 clear_bit(Blocked, &rdev->flags);
2441 clear_bit(BlockedBadBlocks, &rdev->flags);
2442 wake_up(&rdev->blocked_wait);
2445 wake_up(&mddev->sb_wait);
2449 spin_lock_irq(&mddev->write_lock);
2451 mddev->utime = get_seconds();
2453 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2455 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2456 /* just a clean<-> dirty transition, possibly leave spares alone,
2457 * though if events isn't the right even/odd, we will have to do
2463 if (mddev->degraded)
2464 /* If the array is degraded, then skipping spares is both
2465 * dangerous and fairly pointless.
2466 * Dangerous because a device that was removed from the array
2467 * might have a event_count that still looks up-to-date,
2468 * so it can be re-added without a resync.
2469 * Pointless because if there are any spares to skip,
2470 * then a recovery will happen and soon that array won't
2471 * be degraded any more and the spare can go back to sleep then.
2475 sync_req = mddev->in_sync;
2477 /* If this is just a dirty<->clean transition, and the array is clean
2478 * and 'events' is odd, we can roll back to the previous clean state */
2480 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2481 && mddev->can_decrease_events
2482 && mddev->events != 1) {
2484 mddev->can_decrease_events = 0;
2486 /* otherwise we have to go forward and ... */
2488 mddev->can_decrease_events = nospares;
2491 if (!mddev->events) {
2493 * oops, this 64-bit counter should never wrap.
2494 * Either we are in around ~1 trillion A.C., assuming
2495 * 1 reboot per second, or we have a bug:
2501 rdev_for_each(rdev, mddev) {
2502 if (rdev->badblocks.changed)
2503 any_badblocks_changed++;
2504 if (test_bit(Faulty, &rdev->flags))
2505 set_bit(FaultRecorded, &rdev->flags);
2508 sync_sbs(mddev, nospares);
2509 spin_unlock_irq(&mddev->write_lock);
2511 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2512 mdname(mddev), mddev->in_sync);
2514 bitmap_update_sb(mddev->bitmap);
2515 rdev_for_each(rdev, mddev) {
2516 char b[BDEVNAME_SIZE];
2518 if (rdev->sb_loaded != 1)
2519 continue; /* no noise on spare devices */
2521 if (!test_bit(Faulty, &rdev->flags) &&
2522 rdev->saved_raid_disk == -1) {
2523 md_super_write(mddev,rdev,
2524 rdev->sb_start, rdev->sb_size,
2526 pr_debug("md: (write) %s's sb offset: %llu\n",
2527 bdevname(rdev->bdev, b),
2528 (unsigned long long)rdev->sb_start);
2529 rdev->sb_events = mddev->events;
2530 if (rdev->badblocks.size) {
2531 md_super_write(mddev, rdev,
2532 rdev->badblocks.sector,
2533 rdev->badblocks.size << 9,
2535 rdev->badblocks.size = 0;
2538 } else if (test_bit(Faulty, &rdev->flags))
2539 pr_debug("md: %s (skipping faulty)\n",
2540 bdevname(rdev->bdev, b));
2542 pr_debug("(skipping incremental s/r ");
2544 if (mddev->level == LEVEL_MULTIPATH)
2545 /* only need to write one superblock... */
2548 md_super_wait(mddev);
2549 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2551 spin_lock_irq(&mddev->write_lock);
2552 if (mddev->in_sync != sync_req ||
2553 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2554 /* have to write it out again */
2555 spin_unlock_irq(&mddev->write_lock);
2558 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2559 spin_unlock_irq(&mddev->write_lock);
2560 wake_up(&mddev->sb_wait);
2561 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2562 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2564 rdev_for_each(rdev, mddev) {
2565 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2566 clear_bit(Blocked, &rdev->flags);
2568 if (any_badblocks_changed)
2569 md_ack_all_badblocks(&rdev->badblocks);
2570 clear_bit(BlockedBadBlocks, &rdev->flags);
2571 wake_up(&rdev->blocked_wait);
2575 /* words written to sysfs files may, or may not, be \n terminated.
2576 * We want to accept with case. For this we use cmd_match.
2578 static int cmd_match(const char *cmd, const char *str)
2580 /* See if cmd, written into a sysfs file, matches
2581 * str. They must either be the same, or cmd can
2582 * have a trailing newline
2584 while (*cmd && *str && *cmd == *str) {
2595 struct rdev_sysfs_entry {
2596 struct attribute attr;
2597 ssize_t (*show)(struct md_rdev *, char *);
2598 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2602 state_show(struct md_rdev *rdev, char *page)
2607 if (test_bit(Faulty, &rdev->flags) ||
2608 rdev->badblocks.unacked_exist) {
2609 len+= sprintf(page+len, "%sfaulty",sep);
2612 if (test_bit(In_sync, &rdev->flags)) {
2613 len += sprintf(page+len, "%sin_sync",sep);
2616 if (test_bit(WriteMostly, &rdev->flags)) {
2617 len += sprintf(page+len, "%swrite_mostly",sep);
2620 if (test_bit(Blocked, &rdev->flags) ||
2621 (rdev->badblocks.unacked_exist
2622 && !test_bit(Faulty, &rdev->flags))) {
2623 len += sprintf(page+len, "%sblocked", sep);
2626 if (!test_bit(Faulty, &rdev->flags) &&
2627 !test_bit(In_sync, &rdev->flags)) {
2628 len += sprintf(page+len, "%sspare", sep);
2631 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2632 len += sprintf(page+len, "%swrite_error", sep);
2635 if (test_bit(WantReplacement, &rdev->flags)) {
2636 len += sprintf(page+len, "%swant_replacement", sep);
2639 if (test_bit(Replacement, &rdev->flags)) {
2640 len += sprintf(page+len, "%sreplacement", sep);
2644 return len+sprintf(page+len, "\n");
2648 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2651 * faulty - simulates an error
2652 * remove - disconnects the device
2653 * writemostly - sets write_mostly
2654 * -writemostly - clears write_mostly
2655 * blocked - sets the Blocked flags
2656 * -blocked - clears the Blocked and possibly simulates an error
2657 * insync - sets Insync providing device isn't active
2658 * write_error - sets WriteErrorSeen
2659 * -write_error - clears WriteErrorSeen
2662 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2663 md_error(rdev->mddev, rdev);
2664 if (test_bit(Faulty, &rdev->flags))
2668 } else if (cmd_match(buf, "remove")) {
2669 if (rdev->raid_disk >= 0)
2672 struct mddev *mddev = rdev->mddev;
2673 kick_rdev_from_array(rdev);
2675 md_update_sb(mddev, 1);
2676 md_new_event(mddev);
2679 } else if (cmd_match(buf, "writemostly")) {
2680 set_bit(WriteMostly, &rdev->flags);
2682 } else if (cmd_match(buf, "-writemostly")) {
2683 clear_bit(WriteMostly, &rdev->flags);
2685 } else if (cmd_match(buf, "blocked")) {
2686 set_bit(Blocked, &rdev->flags);
2688 } else if (cmd_match(buf, "-blocked")) {
2689 if (!test_bit(Faulty, &rdev->flags) &&
2690 rdev->badblocks.unacked_exist) {
2691 /* metadata handler doesn't understand badblocks,
2692 * so we need to fail the device
2694 md_error(rdev->mddev, rdev);
2696 clear_bit(Blocked, &rdev->flags);
2697 clear_bit(BlockedBadBlocks, &rdev->flags);
2698 wake_up(&rdev->blocked_wait);
2699 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2700 md_wakeup_thread(rdev->mddev->thread);
2703 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2704 set_bit(In_sync, &rdev->flags);
2706 } else if (cmd_match(buf, "write_error")) {
2707 set_bit(WriteErrorSeen, &rdev->flags);
2709 } else if (cmd_match(buf, "-write_error")) {
2710 clear_bit(WriteErrorSeen, &rdev->flags);
2712 } else if (cmd_match(buf, "want_replacement")) {
2713 /* Any non-spare device that is not a replacement can
2714 * become want_replacement at any time, but we then need to
2715 * check if recovery is needed.
2717 if (rdev->raid_disk >= 0 &&
2718 !test_bit(Replacement, &rdev->flags))
2719 set_bit(WantReplacement, &rdev->flags);
2720 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2721 md_wakeup_thread(rdev->mddev->thread);
2723 } else if (cmd_match(buf, "-want_replacement")) {
2724 /* Clearing 'want_replacement' is always allowed.
2725 * Once replacements starts it is too late though.
2728 clear_bit(WantReplacement, &rdev->flags);
2729 } else if (cmd_match(buf, "replacement")) {
2730 /* Can only set a device as a replacement when array has not
2731 * yet been started. Once running, replacement is automatic
2732 * from spares, or by assigning 'slot'.
2734 if (rdev->mddev->pers)
2737 set_bit(Replacement, &rdev->flags);
2740 } else if (cmd_match(buf, "-replacement")) {
2741 /* Similarly, can only clear Replacement before start */
2742 if (rdev->mddev->pers)
2745 clear_bit(Replacement, &rdev->flags);
2750 sysfs_notify_dirent_safe(rdev->sysfs_state);
2751 return err ? err : len;
2753 static struct rdev_sysfs_entry rdev_state =
2754 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2757 errors_show(struct md_rdev *rdev, char *page)
2759 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2763 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2766 unsigned long n = simple_strtoul(buf, &e, 10);
2767 if (*buf && (*e == 0 || *e == '\n')) {
2768 atomic_set(&rdev->corrected_errors, n);
2773 static struct rdev_sysfs_entry rdev_errors =
2774 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2777 slot_show(struct md_rdev *rdev, char *page)
2779 if (rdev->raid_disk < 0)
2780 return sprintf(page, "none\n");
2782 return sprintf(page, "%d\n", rdev->raid_disk);
2786 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2790 int slot = simple_strtoul(buf, &e, 10);
2791 if (strncmp(buf, "none", 4)==0)
2793 else if (e==buf || (*e && *e!= '\n'))
2795 if (rdev->mddev->pers && slot == -1) {
2796 /* Setting 'slot' on an active array requires also
2797 * updating the 'rd%d' link, and communicating
2798 * with the personality with ->hot_*_disk.
2799 * For now we only support removing
2800 * failed/spare devices. This normally happens automatically,
2801 * but not when the metadata is externally managed.
2803 if (rdev->raid_disk == -1)
2805 /* personality does all needed checks */
2806 if (rdev->mddev->pers->hot_remove_disk == NULL)
2808 err = rdev->mddev->pers->
2809 hot_remove_disk(rdev->mddev, rdev);
2812 sysfs_unlink_rdev(rdev->mddev, rdev);
2813 rdev->raid_disk = -1;
2814 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2815 md_wakeup_thread(rdev->mddev->thread);
2816 } else if (rdev->mddev->pers) {
2817 /* Activating a spare .. or possibly reactivating
2818 * if we ever get bitmaps working here.
2821 if (rdev->raid_disk != -1)
2824 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2827 if (rdev->mddev->pers->hot_add_disk == NULL)
2830 if (slot >= rdev->mddev->raid_disks &&
2831 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2834 rdev->raid_disk = slot;
2835 if (test_bit(In_sync, &rdev->flags))
2836 rdev->saved_raid_disk = slot;
2838 rdev->saved_raid_disk = -1;
2839 clear_bit(In_sync, &rdev->flags);
2840 err = rdev->mddev->pers->
2841 hot_add_disk(rdev->mddev, rdev);
2843 rdev->raid_disk = -1;
2846 sysfs_notify_dirent_safe(rdev->sysfs_state);
2847 if (sysfs_link_rdev(rdev->mddev, rdev))
2848 /* failure here is OK */;
2849 /* don't wakeup anyone, leave that to userspace. */
2851 if (slot >= rdev->mddev->raid_disks &&
2852 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2854 rdev->raid_disk = slot;
2855 /* assume it is working */
2856 clear_bit(Faulty, &rdev->flags);
2857 clear_bit(WriteMostly, &rdev->flags);
2858 set_bit(In_sync, &rdev->flags);
2859 sysfs_notify_dirent_safe(rdev->sysfs_state);
2865 static struct rdev_sysfs_entry rdev_slot =
2866 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2869 offset_show(struct md_rdev *rdev, char *page)
2871 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2875 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2877 unsigned long long offset;
2878 if (strict_strtoull(buf, 10, &offset) < 0)
2880 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2882 if (rdev->sectors && rdev->mddev->external)
2883 /* Must set offset before size, so overlap checks
2886 rdev->data_offset = offset;
2887 rdev->new_data_offset = offset;
2891 static struct rdev_sysfs_entry rdev_offset =
2892 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2894 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2896 return sprintf(page, "%llu\n",
2897 (unsigned long long)rdev->new_data_offset);
2900 static ssize_t new_offset_store(struct md_rdev *rdev,
2901 const char *buf, size_t len)
2903 unsigned long long new_offset;
2904 struct mddev *mddev = rdev->mddev;
2906 if (strict_strtoull(buf, 10, &new_offset) < 0)
2909 if (mddev->sync_thread)
2911 if (new_offset == rdev->data_offset)
2912 /* reset is always permitted */
2914 else if (new_offset > rdev->data_offset) {
2915 /* must not push array size beyond rdev_sectors */
2916 if (new_offset - rdev->data_offset
2917 + mddev->dev_sectors > rdev->sectors)
2920 /* Metadata worries about other space details. */
2922 /* decreasing the offset is inconsistent with a backwards
2925 if (new_offset < rdev->data_offset &&
2926 mddev->reshape_backwards)
2928 /* Increasing offset is inconsistent with forwards
2929 * reshape. reshape_direction should be set to
2930 * 'backwards' first.
2932 if (new_offset > rdev->data_offset &&
2933 !mddev->reshape_backwards)
2936 if (mddev->pers && mddev->persistent &&
2937 !super_types[mddev->major_version]
2938 .allow_new_offset(rdev, new_offset))
2940 rdev->new_data_offset = new_offset;
2941 if (new_offset > rdev->data_offset)
2942 mddev->reshape_backwards = 1;
2943 else if (new_offset < rdev->data_offset)
2944 mddev->reshape_backwards = 0;
2948 static struct rdev_sysfs_entry rdev_new_offset =
2949 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2952 rdev_size_show(struct md_rdev *rdev, char *page)
2954 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2957 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2959 /* check if two start/length pairs overlap */
2967 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2969 unsigned long long blocks;
2972 if (strict_strtoull(buf, 10, &blocks) < 0)
2975 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2976 return -EINVAL; /* sector conversion overflow */
2979 if (new != blocks * 2)
2980 return -EINVAL; /* unsigned long long to sector_t overflow */
2987 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2989 struct mddev *my_mddev = rdev->mddev;
2990 sector_t oldsectors = rdev->sectors;
2993 if (strict_blocks_to_sectors(buf, §ors) < 0)
2995 if (rdev->data_offset != rdev->new_data_offset)
2996 return -EINVAL; /* too confusing */
2997 if (my_mddev->pers && rdev->raid_disk >= 0) {
2998 if (my_mddev->persistent) {
2999 sectors = super_types[my_mddev->major_version].
3000 rdev_size_change(rdev, sectors);
3003 } else if (!sectors)
3004 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
3006 if (!my_mddev->pers->resize)
3007 /* Cannot change size for RAID0 or Linear etc */
3010 if (sectors < my_mddev->dev_sectors)
3011 return -EINVAL; /* component must fit device */
3013 rdev->sectors = sectors;
3014 if (sectors > oldsectors && my_mddev->external) {
3015 /* need to check that all other rdevs with the same ->bdev
3016 * do not overlap. We need to unlock the mddev to avoid
3017 * a deadlock. We have already changed rdev->sectors, and if
3018 * we have to change it back, we will have the lock again.
3020 struct mddev *mddev;
3022 struct list_head *tmp;
3024 mddev_unlock(my_mddev);
3025 for_each_mddev(mddev, tmp) {
3026 struct md_rdev *rdev2;
3029 rdev_for_each(rdev2, mddev)
3030 if (rdev->bdev == rdev2->bdev &&
3032 overlaps(rdev->data_offset, rdev->sectors,
3038 mddev_unlock(mddev);
3044 mddev_lock(my_mddev);
3046 /* Someone else could have slipped in a size
3047 * change here, but doing so is just silly.
3048 * We put oldsectors back because we *know* it is
3049 * safe, and trust userspace not to race with
3052 rdev->sectors = oldsectors;
3059 static struct rdev_sysfs_entry rdev_size =
3060 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3063 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3065 unsigned long long recovery_start = rdev->recovery_offset;
3067 if (test_bit(In_sync, &rdev->flags) ||
3068 recovery_start == MaxSector)
3069 return sprintf(page, "none\n");
3071 return sprintf(page, "%llu\n", recovery_start);
3074 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3076 unsigned long long recovery_start;
3078 if (cmd_match(buf, "none"))
3079 recovery_start = MaxSector;
3080 else if (strict_strtoull(buf, 10, &recovery_start))
3083 if (rdev->mddev->pers &&
3084 rdev->raid_disk >= 0)
3087 rdev->recovery_offset = recovery_start;
3088 if (recovery_start == MaxSector)
3089 set_bit(In_sync, &rdev->flags);
3091 clear_bit(In_sync, &rdev->flags);
3095 static struct rdev_sysfs_entry rdev_recovery_start =
3096 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3100 badblocks_show(struct badblocks *bb, char *page, int unack);
3102 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3104 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3106 return badblocks_show(&rdev->badblocks, page, 0);
3108 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3110 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3111 /* Maybe that ack was all we needed */
3112 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3113 wake_up(&rdev->blocked_wait);
3116 static struct rdev_sysfs_entry rdev_bad_blocks =
3117 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3120 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3122 return badblocks_show(&rdev->badblocks, page, 1);
3124 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3126 return badblocks_store(&rdev->badblocks, page, len, 1);
3128 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3129 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3131 static struct attribute *rdev_default_attrs[] = {
3136 &rdev_new_offset.attr,
3138 &rdev_recovery_start.attr,
3139 &rdev_bad_blocks.attr,
3140 &rdev_unack_bad_blocks.attr,
3144 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3146 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3147 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3148 struct mddev *mddev = rdev->mddev;
3154 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3156 if (rdev->mddev == NULL)
3159 rv = entry->show(rdev, page);
3160 mddev_unlock(mddev);
3166 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3167 const char *page, size_t length)
3169 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3170 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3172 struct mddev *mddev = rdev->mddev;
3176 if (!capable(CAP_SYS_ADMIN))
3178 rv = mddev ? mddev_lock(mddev): -EBUSY;
3180 if (rdev->mddev == NULL)
3183 rv = entry->store(rdev, page, length);
3184 mddev_unlock(mddev);
3189 static void rdev_free(struct kobject *ko)
3191 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3194 static const struct sysfs_ops rdev_sysfs_ops = {
3195 .show = rdev_attr_show,
3196 .store = rdev_attr_store,
3198 static struct kobj_type rdev_ktype = {
3199 .release = rdev_free,
3200 .sysfs_ops = &rdev_sysfs_ops,
3201 .default_attrs = rdev_default_attrs,
3204 int md_rdev_init(struct md_rdev *rdev)
3207 rdev->saved_raid_disk = -1;
3208 rdev->raid_disk = -1;
3210 rdev->data_offset = 0;
3211 rdev->new_data_offset = 0;
3212 rdev->sb_events = 0;
3213 rdev->last_read_error.tv_sec = 0;
3214 rdev->last_read_error.tv_nsec = 0;
3215 rdev->sb_loaded = 0;
3216 rdev->bb_page = NULL;
3217 atomic_set(&rdev->nr_pending, 0);
3218 atomic_set(&rdev->read_errors, 0);
3219 atomic_set(&rdev->corrected_errors, 0);
3221 INIT_LIST_HEAD(&rdev->same_set);
3222 init_waitqueue_head(&rdev->blocked_wait);
3224 /* Add space to store bad block list.
3225 * This reserves the space even on arrays where it cannot
3226 * be used - I wonder if that matters
3228 rdev->badblocks.count = 0;
3229 rdev->badblocks.shift = 0;
3230 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3231 seqlock_init(&rdev->badblocks.lock);
3232 if (rdev->badblocks.page == NULL)
3237 EXPORT_SYMBOL_GPL(md_rdev_init);
3239 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3241 * mark the device faulty if:
3243 * - the device is nonexistent (zero size)
3244 * - the device has no valid superblock
3246 * a faulty rdev _never_ has rdev->sb set.
3248 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3250 char b[BDEVNAME_SIZE];
3252 struct md_rdev *rdev;
3255 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3257 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3258 return ERR_PTR(-ENOMEM);
3261 err = md_rdev_init(rdev);
3264 err = alloc_disk_sb(rdev);
3268 err = lock_rdev(rdev, newdev, super_format == -2);
3272 kobject_init(&rdev->kobj, &rdev_ktype);
3274 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3277 "md: %s has zero or unknown size, marking faulty!\n",
3278 bdevname(rdev->bdev,b));
3283 if (super_format >= 0) {
3284 err = super_types[super_format].
3285 load_super(rdev, NULL, super_minor);
3286 if (err == -EINVAL) {
3288 "md: %s does not have a valid v%d.%d "
3289 "superblock, not importing!\n",
3290 bdevname(rdev->bdev,b),
3291 super_format, super_minor);
3296 "md: could not read %s's sb, not importing!\n",
3297 bdevname(rdev->bdev,b));
3301 if (super_format == -1)
3302 /* hot-add for 0.90, or non-persistent: so no badblocks */
3303 rdev->badblocks.shift = -1;
3310 md_rdev_clear(rdev);
3312 return ERR_PTR(err);
3316 * Check a full RAID array for plausibility
3320 static void analyze_sbs(struct mddev * mddev)
3323 struct md_rdev *rdev, *freshest, *tmp;
3324 char b[BDEVNAME_SIZE];
3327 rdev_for_each_safe(rdev, tmp, mddev)
3328 switch (super_types[mddev->major_version].
3329 load_super(rdev, freshest, mddev->minor_version)) {
3337 "md: fatal superblock inconsistency in %s"
3338 " -- removing from array\n",
3339 bdevname(rdev->bdev,b));
3340 kick_rdev_from_array(rdev);
3344 super_types[mddev->major_version].
3345 validate_super(mddev, freshest);
3348 rdev_for_each_safe(rdev, tmp, mddev) {
3349 if (mddev->max_disks &&
3350 (rdev->desc_nr >= mddev->max_disks ||
3351 i > mddev->max_disks)) {
3353 "md: %s: %s: only %d devices permitted\n",
3354 mdname(mddev), bdevname(rdev->bdev, b),
3356 kick_rdev_from_array(rdev);
3359 if (rdev != freshest)
3360 if (super_types[mddev->major_version].
3361 validate_super(mddev, rdev)) {
3362 printk(KERN_WARNING "md: kicking non-fresh %s"
3364 bdevname(rdev->bdev,b));
3365 kick_rdev_from_array(rdev);
3368 if (mddev->level == LEVEL_MULTIPATH) {
3369 rdev->desc_nr = i++;
3370 rdev->raid_disk = rdev->desc_nr;
3371 set_bit(In_sync, &rdev->flags);
3372 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3373 rdev->raid_disk = -1;
3374 clear_bit(In_sync, &rdev->flags);
3379 /* Read a fixed-point number.
3380 * Numbers in sysfs attributes should be in "standard" units where
3381 * possible, so time should be in seconds.
3382 * However we internally use a a much smaller unit such as
3383 * milliseconds or jiffies.
3384 * This function takes a decimal number with a possible fractional
3385 * component, and produces an integer which is the result of
3386 * multiplying that number by 10^'scale'.
3387 * all without any floating-point arithmetic.
3389 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3391 unsigned long result = 0;
3393 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3396 else if (decimals < scale) {
3399 result = result * 10 + value;
3411 while (decimals < scale) {
3420 static void md_safemode_timeout(unsigned long data);
3423 safe_delay_show(struct mddev *mddev, char *page)
3425 int msec = (mddev->safemode_delay*1000)/HZ;
3426 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3429 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3433 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3436 mddev->safemode_delay = 0;
3438 unsigned long old_delay = mddev->safemode_delay;
3439 mddev->safemode_delay = (msec*HZ)/1000;
3440 if (mddev->safemode_delay == 0)
3441 mddev->safemode_delay = 1;
3442 if (mddev->safemode_delay < old_delay)
3443 md_safemode_timeout((unsigned long)mddev);
3447 static struct md_sysfs_entry md_safe_delay =
3448 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3451 level_show(struct mddev *mddev, char *page)
3453 struct md_personality *p = mddev->pers;
3455 return sprintf(page, "%s\n", p->name);
3456 else if (mddev->clevel[0])
3457 return sprintf(page, "%s\n", mddev->clevel);
3458 else if (mddev->level != LEVEL_NONE)
3459 return sprintf(page, "%d\n", mddev->level);
3465 level_store(struct mddev *mddev, const char *buf, size_t len)
3469 struct md_personality *pers;
3472 struct md_rdev *rdev;
3474 if (mddev->pers == NULL) {
3477 if (len >= sizeof(mddev->clevel))
3479 strncpy(mddev->clevel, buf, len);
3480 if (mddev->clevel[len-1] == '\n')
3482 mddev->clevel[len] = 0;
3483 mddev->level = LEVEL_NONE;
3487 /* request to change the personality. Need to ensure:
3488 * - array is not engaged in resync/recovery/reshape
3489 * - old personality can be suspended
3490 * - new personality will access other array.
3493 if (mddev->sync_thread ||
3494 mddev->reshape_position != MaxSector ||
3495 mddev->sysfs_active)
3498 if (!mddev->pers->quiesce) {
3499 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3500 mdname(mddev), mddev->pers->name);
3504 /* Now find the new personality */
3505 if (len == 0 || len >= sizeof(clevel))
3507 strncpy(clevel, buf, len);
3508 if (clevel[len-1] == '\n')
3511 if (strict_strtol(clevel, 10, &level))
3514 if (request_module("md-%s", clevel) != 0)
3515 request_module("md-level-%s", clevel);
3516 spin_lock(&pers_lock);
3517 pers = find_pers(level, clevel);
3518 if (!pers || !try_module_get(pers->owner)) {
3519 spin_unlock(&pers_lock);
3520 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3523 spin_unlock(&pers_lock);
3525 if (pers == mddev->pers) {
3526 /* Nothing to do! */
3527 module_put(pers->owner);
3530 if (!pers->takeover) {
3531 module_put(pers->owner);
3532 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3533 mdname(mddev), clevel);
3537 rdev_for_each(rdev, mddev)
3538 rdev->new_raid_disk = rdev->raid_disk;
3540 /* ->takeover must set new_* and/or delta_disks
3541 * if it succeeds, and may set them when it fails.
3543 priv = pers->takeover(mddev);
3545 mddev->new_level = mddev->level;
3546 mddev->new_layout = mddev->layout;
3547 mddev->new_chunk_sectors = mddev->chunk_sectors;
3548 mddev->raid_disks -= mddev->delta_disks;
3549 mddev->delta_disks = 0;
3550 mddev->reshape_backwards = 0;
3551 module_put(pers->owner);
3552 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3553 mdname(mddev), clevel);
3554 return PTR_ERR(priv);
3557 /* Looks like we have a winner */
3558 mddev_suspend(mddev);
3559 mddev->pers->stop(mddev);
3561 if (mddev->pers->sync_request == NULL &&
3562 pers->sync_request != NULL) {
3563 /* need to add the md_redundancy_group */
3564 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3566 "md: cannot register extra attributes for %s\n",
3568 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3570 if (mddev->pers->sync_request != NULL &&
3571 pers->sync_request == NULL) {
3572 /* need to remove the md_redundancy_group */
3573 if (mddev->to_remove == NULL)
3574 mddev->to_remove = &md_redundancy_group;
3577 if (mddev->pers->sync_request == NULL &&
3579 /* We are converting from a no-redundancy array
3580 * to a redundancy array and metadata is managed
3581 * externally so we need to be sure that writes
3582 * won't block due to a need to transition
3584 * until external management is started.
3587 mddev->safemode_delay = 0;
3588 mddev->safemode = 0;
3591 rdev_for_each(rdev, mddev) {
3592 if (rdev->raid_disk < 0)
3594 if (rdev->new_raid_disk >= mddev->raid_disks)
3595 rdev->new_raid_disk = -1;
3596 if (rdev->new_raid_disk == rdev->raid_disk)
3598 sysfs_unlink_rdev(mddev, rdev);
3600 rdev_for_each(rdev, mddev) {
3601 if (rdev->raid_disk < 0)
3603 if (rdev->new_raid_disk == rdev->raid_disk)
3605 rdev->raid_disk = rdev->new_raid_disk;
3606 if (rdev->raid_disk < 0)
3607 clear_bit(In_sync, &rdev->flags);
3609 if (sysfs_link_rdev(mddev, rdev))
3610 printk(KERN_WARNING "md: cannot register rd%d"
3611 " for %s after level change\n",
3612 rdev->raid_disk, mdname(mddev));
3616 module_put(mddev->pers->owner);
3618 mddev->private = priv;
3619 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3620 mddev->level = mddev->new_level;
3621 mddev->layout = mddev->new_layout;
3622 mddev->chunk_sectors = mddev->new_chunk_sectors;
3623 mddev->delta_disks = 0;
3624 mddev->reshape_backwards = 0;
3625 mddev->degraded = 0;
3626 if (mddev->pers->sync_request == NULL) {
3627 /* this is now an array without redundancy, so
3628 * it must always be in_sync
3631 del_timer_sync(&mddev->safemode_timer);
3634 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3635 mddev_resume(mddev);
3636 sysfs_notify(&mddev->kobj, NULL, "level");
3637 md_new_event(mddev);
3641 static struct md_sysfs_entry md_level =
3642 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3646 layout_show(struct mddev *mddev, char *page)
3648 /* just a number, not meaningful for all levels */
3649 if (mddev->reshape_position != MaxSector &&
3650 mddev->layout != mddev->new_layout)
3651 return sprintf(page, "%d (%d)\n",
3652 mddev->new_layout, mddev->layout);
3653 return sprintf(page, "%d\n", mddev->layout);
3657 layout_store(struct mddev *mddev, const char *buf, size_t len)
3660 unsigned long n = simple_strtoul(buf, &e, 10);
3662 if (!*buf || (*e && *e != '\n'))
3667 if (mddev->pers->check_reshape == NULL)
3669 mddev->new_layout = n;
3670 err = mddev->pers->check_reshape(mddev);
3672 mddev->new_layout = mddev->layout;
3676 mddev->new_layout = n;
3677 if (mddev->reshape_position == MaxSector)
3682 static struct md_sysfs_entry md_layout =
3683 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3687 raid_disks_show(struct mddev *mddev, char *page)
3689 if (mddev->raid_disks == 0)
3691 if (mddev->reshape_position != MaxSector &&
3692 mddev->delta_disks != 0)
3693 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3694 mddev->raid_disks - mddev->delta_disks);
3695 return sprintf(page, "%d\n", mddev->raid_disks);
3698 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3701 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3705 unsigned long n = simple_strtoul(buf, &e, 10);
3707 if (!*buf || (*e && *e != '\n'))
3711 rv = update_raid_disks(mddev, n);
3712 else if (mddev->reshape_position != MaxSector) {
3713 struct md_rdev *rdev;
3714 int olddisks = mddev->raid_disks - mddev->delta_disks;
3716 rdev_for_each(rdev, mddev) {
3718 rdev->data_offset < rdev->new_data_offset)
3721 rdev->data_offset > rdev->new_data_offset)
3724 mddev->delta_disks = n - olddisks;
3725 mddev->raid_disks = n;
3726 mddev->reshape_backwards = (mddev->delta_disks < 0);
3728 mddev->raid_disks = n;
3729 return rv ? rv : len;
3731 static struct md_sysfs_entry md_raid_disks =
3732 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3735 chunk_size_show(struct mddev *mddev, char *page)
3737 if (mddev->reshape_position != MaxSector &&
3738 mddev->chunk_sectors != mddev->new_chunk_sectors)
3739 return sprintf(page, "%d (%d)\n",
3740 mddev->new_chunk_sectors << 9,
3741 mddev->chunk_sectors << 9);
3742 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3746 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3749 unsigned long n = simple_strtoul(buf, &e, 10);
3751 if (!*buf || (*e && *e != '\n'))
3756 if (mddev->pers->check_reshape == NULL)
3758 mddev->new_chunk_sectors = n >> 9;
3759 err = mddev->pers->check_reshape(mddev);
3761 mddev->new_chunk_sectors = mddev->chunk_sectors;
3765 mddev->new_chunk_sectors = n >> 9;
3766 if (mddev->reshape_position == MaxSector)
3767 mddev->chunk_sectors = n >> 9;
3771 static struct md_sysfs_entry md_chunk_size =
3772 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3775 resync_start_show(struct mddev *mddev, char *page)
3777 if (mddev->recovery_cp == MaxSector)
3778 return sprintf(page, "none\n");
3779 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3783 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3786 unsigned long long n = simple_strtoull(buf, &e, 10);
3788 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3790 if (cmd_match(buf, "none"))
3792 else if (!*buf || (*e && *e != '\n'))
3795 mddev->recovery_cp = n;
3797 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3800 static struct md_sysfs_entry md_resync_start =
3801 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3804 * The array state can be:
3807 * No devices, no size, no level
3808 * Equivalent to STOP_ARRAY ioctl
3810 * May have some settings, but array is not active
3811 * all IO results in error
3812 * When written, doesn't tear down array, but just stops it
3813 * suspended (not supported yet)
3814 * All IO requests will block. The array can be reconfigured.
3815 * Writing this, if accepted, will block until array is quiescent
3817 * no resync can happen. no superblocks get written.
3818 * write requests fail
3820 * like readonly, but behaves like 'clean' on a write request.
3822 * clean - no pending writes, but otherwise active.
3823 * When written to inactive array, starts without resync
3824 * If a write request arrives then
3825 * if metadata is known, mark 'dirty' and switch to 'active'.
3826 * if not known, block and switch to write-pending
3827 * If written to an active array that has pending writes, then fails.
3829 * fully active: IO and resync can be happening.
3830 * When written to inactive array, starts with resync
3833 * clean, but writes are blocked waiting for 'active' to be written.
3836 * like active, but no writes have been seen for a while (100msec).
3839 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3840 write_pending, active_idle, bad_word};
3841 static char *array_states[] = {
3842 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3843 "write-pending", "active-idle", NULL };
3845 static int match_word(const char *word, char **list)
3848 for (n=0; list[n]; n++)
3849 if (cmd_match(word, list[n]))
3855 array_state_show(struct mddev *mddev, char *page)
3857 enum array_state st = inactive;
3870 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3872 else if (mddev->safemode)
3878 if (list_empty(&mddev->disks) &&
3879 mddev->raid_disks == 0 &&
3880 mddev->dev_sectors == 0)
3885 return sprintf(page, "%s\n", array_states[st]);
3888 static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3889 static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
3890 static int do_md_run(struct mddev * mddev);
3891 static int restart_array(struct mddev *mddev);
3894 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3897 enum array_state st = match_word(buf, array_states);
3902 /* stopping an active array */
3903 err = do_md_stop(mddev, 0, NULL);
3906 /* stopping an active array */
3908 err = do_md_stop(mddev, 2, NULL);
3910 err = 0; /* already inactive */
3913 break; /* not supported yet */
3916 err = md_set_readonly(mddev, NULL);
3919 set_disk_ro(mddev->gendisk, 1);
3920 err = do_md_run(mddev);
3926 err = md_set_readonly(mddev, NULL);
3927 else if (mddev->ro == 1)
3928 err = restart_array(mddev);
3931 set_disk_ro(mddev->gendisk, 0);
3935 err = do_md_run(mddev);
3940 restart_array(mddev);
3941 spin_lock_irq(&mddev->write_lock);
3942 if (atomic_read(&mddev->writes_pending) == 0) {
3943 if (mddev->in_sync == 0) {
3945 if (mddev->safemode == 1)
3946 mddev->safemode = 0;
3947 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3952 spin_unlock_irq(&mddev->write_lock);
3958 restart_array(mddev);
3959 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3960 wake_up(&mddev->sb_wait);
3964 set_disk_ro(mddev->gendisk, 0);
3965 err = do_md_run(mddev);
3970 /* these cannot be set */
3976 if (mddev->hold_active == UNTIL_IOCTL)
3977 mddev->hold_active = 0;
3978 sysfs_notify_dirent_safe(mddev->sysfs_state);
3982 static struct md_sysfs_entry md_array_state =
3983 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3986 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3987 return sprintf(page, "%d\n",
3988 atomic_read(&mddev->max_corr_read_errors));
3992 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3995 unsigned long n = simple_strtoul(buf, &e, 10);
3997 if (*buf && (*e == 0 || *e == '\n')) {
3998 atomic_set(&mddev->max_corr_read_errors, n);
4004 static struct md_sysfs_entry max_corr_read_errors =
4005 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4006 max_corrected_read_errors_store);
4009 null_show(struct mddev *mddev, char *page)
4015 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4017 /* buf must be %d:%d\n? giving major and minor numbers */
4018 /* The new device is added to the array.
4019 * If the array has a persistent superblock, we read the
4020 * superblock to initialise info and check validity.
4021 * Otherwise, only checking done is that in bind_rdev_to_array,
4022 * which mainly checks size.
4025 int major = simple_strtoul(buf, &e, 10);
4028 struct md_rdev *rdev;
4031 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4033 minor = simple_strtoul(e+1, &e, 10);
4034 if (*e && *e != '\n')
4036 dev = MKDEV(major, minor);
4037 if (major != MAJOR(dev) ||
4038 minor != MINOR(dev))
4042 if (mddev->persistent) {
4043 rdev = md_import_device(dev, mddev->major_version,
4044 mddev->minor_version);
4045 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4046 struct md_rdev *rdev0
4047 = list_entry(mddev->disks.next,
4048 struct md_rdev, same_set);
4049 err = super_types[mddev->major_version]
4050 .load_super(rdev, rdev0, mddev->minor_version);
4054 } else if (mddev->external)
4055 rdev = md_import_device(dev, -2, -1);
4057 rdev = md_import_device(dev, -1, -1);
4060 return PTR_ERR(rdev);
4061 err = bind_rdev_to_array(rdev, mddev);
4065 return err ? err : len;
4068 static struct md_sysfs_entry md_new_device =
4069 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4072 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4075 unsigned long chunk, end_chunk;
4079 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4081 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4082 if (buf == end) break;
4083 if (*end == '-') { /* range */
4085 end_chunk = simple_strtoul(buf, &end, 0);
4086 if (buf == end) break;
4088 if (*end && !isspace(*end)) break;
4089 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4090 buf = skip_spaces(end);
4092 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4097 static struct md_sysfs_entry md_bitmap =
4098 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4101 size_show(struct mddev *mddev, char *page)
4103 return sprintf(page, "%llu\n",
4104 (unsigned long long)mddev->dev_sectors / 2);
4107 static int update_size(struct mddev *mddev, sector_t num_sectors);
4110 size_store(struct mddev *mddev, const char *buf, size_t len)
4112 /* If array is inactive, we can reduce the component size, but
4113 * not increase it (except from 0).
4114 * If array is active, we can try an on-line resize
4117 int err = strict_blocks_to_sectors(buf, §ors);
4122 err = update_size(mddev, sectors);
4123 md_update_sb(mddev, 1);
4125 if (mddev->dev_sectors == 0 ||
4126 mddev->dev_sectors > sectors)
4127 mddev->dev_sectors = sectors;
4131 return err ? err : len;
4134 static struct md_sysfs_entry md_size =
4135 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4138 /* Metadata version.
4140 * 'none' for arrays with no metadata (good luck...)
4141 * 'external' for arrays with externally managed metadata,
4142 * or N.M for internally known formats
4145 metadata_show(struct mddev *mddev, char *page)
4147 if (mddev->persistent)
4148 return sprintf(page, "%d.%d\n",
4149 mddev->major_version, mddev->minor_version);
4150 else if (mddev->external)
4151 return sprintf(page, "external:%s\n", mddev->metadata_type);
4153 return sprintf(page, "none\n");
4157 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4161 /* Changing the details of 'external' metadata is
4162 * always permitted. Otherwise there must be
4163 * no devices attached to the array.
4165 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4167 else if (!list_empty(&mddev->disks))
4170 if (cmd_match(buf, "none")) {
4171 mddev->persistent = 0;
4172 mddev->external = 0;
4173 mddev->major_version = 0;
4174 mddev->minor_version = 90;
4177 if (strncmp(buf, "external:", 9) == 0) {
4178 size_t namelen = len-9;
4179 if (namelen >= sizeof(mddev->metadata_type))
4180 namelen = sizeof(mddev->metadata_type)-1;
4181 strncpy(mddev->metadata_type, buf+9, namelen);
4182 mddev->metadata_type[namelen] = 0;
4183 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4184 mddev->metadata_type[--namelen] = 0;
4185 mddev->persistent = 0;
4186 mddev->external = 1;
4187 mddev->major_version = 0;
4188 mddev->minor_version = 90;
4191 major = simple_strtoul(buf, &e, 10);
4192 if (e==buf || *e != '.')
4195 minor = simple_strtoul(buf, &e, 10);
4196 if (e==buf || (*e && *e != '\n') )
4198 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4200 mddev->major_version = major;
4201 mddev->minor_version = minor;
4202 mddev->persistent = 1;
4203 mddev->external = 0;
4207 static struct md_sysfs_entry md_metadata =
4208 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4211 action_show(struct mddev *mddev, char *page)
4213 char *type = "idle";
4214 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4216 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4217 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4218 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4220 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4221 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4223 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4227 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4230 return sprintf(page, "%s\n", type);
4233 static void reap_sync_thread(struct mddev *mddev);
4236 action_store(struct mddev *mddev, const char *page, size_t len)
4238 if (!mddev->pers || !mddev->pers->sync_request)
4241 if (cmd_match(page, "frozen"))
4242 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4244 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4246 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4247 if (mddev->sync_thread) {
4248 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4249 reap_sync_thread(mddev);
4251 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4252 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4254 else if (cmd_match(page, "resync"))
4255 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4256 else if (cmd_match(page, "recover")) {
4257 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4258 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4259 } else if (cmd_match(page, "reshape")) {
4261 if (mddev->pers->start_reshape == NULL)
4263 err = mddev->pers->start_reshape(mddev);
4266 sysfs_notify(&mddev->kobj, NULL, "degraded");
4268 if (cmd_match(page, "check"))
4269 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4270 else if (!cmd_match(page, "repair"))
4272 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4273 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4275 if (mddev->ro == 2) {
4276 /* A write to sync_action is enough to justify
4277 * canceling read-auto mode
4280 md_wakeup_thread(mddev->sync_thread);
4282 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4283 md_wakeup_thread(mddev->thread);
4284 sysfs_notify_dirent_safe(mddev->sysfs_action);
4289 mismatch_cnt_show(struct mddev *mddev, char *page)
4291 return sprintf(page, "%llu\n",
4292 (unsigned long long)
4293 atomic64_read(&mddev->resync_mismatches));
4296 static struct md_sysfs_entry md_scan_mode =
4297 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4300 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4303 sync_min_show(struct mddev *mddev, char *page)
4305 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4306 mddev->sync_speed_min ? "local": "system");
4310 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4314 if (strncmp(buf, "system", 6)==0) {
4315 mddev->sync_speed_min = 0;
4318 min = simple_strtoul(buf, &e, 10);
4319 if (buf == e || (*e && *e != '\n') || min <= 0)
4321 mddev->sync_speed_min = min;
4325 static struct md_sysfs_entry md_sync_min =
4326 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4329 sync_max_show(struct mddev *mddev, char *page)
4331 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4332 mddev->sync_speed_max ? "local": "system");
4336 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4340 if (strncmp(buf, "system", 6)==0) {
4341 mddev->sync_speed_max = 0;
4344 max = simple_strtoul(buf, &e, 10);
4345 if (buf == e || (*e && *e != '\n') || max <= 0)
4347 mddev->sync_speed_max = max;
4351 static struct md_sysfs_entry md_sync_max =
4352 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4355 degraded_show(struct mddev *mddev, char *page)
4357 return sprintf(page, "%d\n", mddev->degraded);
4359 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4362 sync_force_parallel_show(struct mddev *mddev, char *page)
4364 return sprintf(page, "%d\n", mddev->parallel_resync);
4368 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4372 if (strict_strtol(buf, 10, &n))
4375 if (n != 0 && n != 1)
4378 mddev->parallel_resync = n;
4380 if (mddev->sync_thread)
4381 wake_up(&resync_wait);
4386 /* force parallel resync, even with shared block devices */
4387 static struct md_sysfs_entry md_sync_force_parallel =
4388 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4389 sync_force_parallel_show, sync_force_parallel_store);
4392 sync_speed_show(struct mddev *mddev, char *page)
4394 unsigned long resync, dt, db;
4395 if (mddev->curr_resync == 0)
4396 return sprintf(page, "none\n");
4397 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4398 dt = (jiffies - mddev->resync_mark) / HZ;
4400 db = resync - mddev->resync_mark_cnt;
4401 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4404 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4407 sync_completed_show(struct mddev *mddev, char *page)
4409 unsigned long long max_sectors, resync;
4411 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4412 return sprintf(page, "none\n");
4414 if (mddev->curr_resync == 1 ||
4415 mddev->curr_resync == 2)
4416 return sprintf(page, "delayed\n");
4418 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4419 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4420 max_sectors = mddev->resync_max_sectors;
4422 max_sectors = mddev->dev_sectors;
4424 resync = mddev->curr_resync_completed;
4425 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4428 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4431 min_sync_show(struct mddev *mddev, char *page)
4433 return sprintf(page, "%llu\n",
4434 (unsigned long long)mddev->resync_min);
4437 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4439 unsigned long long min;
4440 if (strict_strtoull(buf, 10, &min))
4442 if (min > mddev->resync_max)
4444 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4447 /* Must be a multiple of chunk_size */
4448 if (mddev->chunk_sectors) {
4449 sector_t temp = min;
4450 if (sector_div(temp, mddev->chunk_sectors))
4453 mddev->resync_min = min;
4458 static struct md_sysfs_entry md_min_sync =
4459 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4462 max_sync_show(struct mddev *mddev, char *page)
4464 if (mddev->resync_max == MaxSector)
4465 return sprintf(page, "max\n");
4467 return sprintf(page, "%llu\n",
4468 (unsigned long long)mddev->resync_max);
4471 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4473 if (strncmp(buf, "max", 3) == 0)
4474 mddev->resync_max = MaxSector;
4476 unsigned long long max;
4477 if (strict_strtoull(buf, 10, &max))
4479 if (max < mddev->resync_min)
4481 if (max < mddev->resync_max &&
4483 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4486 /* Must be a multiple of chunk_size */
4487 if (mddev->chunk_sectors) {
4488 sector_t temp = max;
4489 if (sector_div(temp, mddev->chunk_sectors))
4492 mddev->resync_max = max;
4494 wake_up(&mddev->recovery_wait);
4498 static struct md_sysfs_entry md_max_sync =
4499 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4502 suspend_lo_show(struct mddev *mddev, char *page)
4504 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4508 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4511 unsigned long long new = simple_strtoull(buf, &e, 10);
4512 unsigned long long old = mddev->suspend_lo;
4514 if (mddev->pers == NULL ||
4515 mddev->pers->quiesce == NULL)
4517 if (buf == e || (*e && *e != '\n'))
4520 mddev->suspend_lo = new;
4522 /* Shrinking suspended region */
4523 mddev->pers->quiesce(mddev, 2);
4525 /* Expanding suspended region - need to wait */
4526 mddev->pers->quiesce(mddev, 1);
4527 mddev->pers->quiesce(mddev, 0);
4531 static struct md_sysfs_entry md_suspend_lo =
4532 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4536 suspend_hi_show(struct mddev *mddev, char *page)
4538 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4542 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4545 unsigned long long new = simple_strtoull(buf, &e, 10);
4546 unsigned long long old = mddev->suspend_hi;
4548 if (mddev->pers == NULL ||
4549 mddev->pers->quiesce == NULL)
4551 if (buf == e || (*e && *e != '\n'))
4554 mddev->suspend_hi = new;
4556 /* Shrinking suspended region */
4557 mddev->pers->quiesce(mddev, 2);
4559 /* Expanding suspended region - need to wait */
4560 mddev->pers->quiesce(mddev, 1);
4561 mddev->pers->quiesce(mddev, 0);
4565 static struct md_sysfs_entry md_suspend_hi =
4566 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4569 reshape_position_show(struct mddev *mddev, char *page)
4571 if (mddev->reshape_position != MaxSector)
4572 return sprintf(page, "%llu\n",
4573 (unsigned long long)mddev->reshape_position);
4574 strcpy(page, "none\n");
4579 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4581 struct md_rdev *rdev;
4583 unsigned long long new = simple_strtoull(buf, &e, 10);
4586 if (buf == e || (*e && *e != '\n'))
4588 mddev->reshape_position = new;
4589 mddev->delta_disks = 0;
4590 mddev->reshape_backwards = 0;
4591 mddev->new_level = mddev->level;
4592 mddev->new_layout = mddev->layout;
4593 mddev->new_chunk_sectors = mddev->chunk_sectors;
4594 rdev_for_each(rdev, mddev)
4595 rdev->new_data_offset = rdev->data_offset;
4599 static struct md_sysfs_entry md_reshape_position =
4600 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4601 reshape_position_store);
4604 reshape_direction_show(struct mddev *mddev, char *page)
4606 return sprintf(page, "%s\n",
4607 mddev->reshape_backwards ? "backwards" : "forwards");
4611 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4614 if (cmd_match(buf, "forwards"))
4616 else if (cmd_match(buf, "backwards"))
4620 if (mddev->reshape_backwards == backwards)
4623 /* check if we are allowed to change */
4624 if (mddev->delta_disks)
4627 if (mddev->persistent &&
4628 mddev->major_version == 0)
4631 mddev->reshape_backwards = backwards;
4635 static struct md_sysfs_entry md_reshape_direction =
4636 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4637 reshape_direction_store);
4640 array_size_show(struct mddev *mddev, char *page)
4642 if (mddev->external_size)
4643 return sprintf(page, "%llu\n",
4644 (unsigned long long)mddev->array_sectors/2);
4646 return sprintf(page, "default\n");
4650 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4654 if (strncmp(buf, "default", 7) == 0) {
4656 sectors = mddev->pers->size(mddev, 0, 0);
4658 sectors = mddev->array_sectors;
4660 mddev->external_size = 0;
4662 if (strict_blocks_to_sectors(buf, §ors) < 0)
4664 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4667 mddev->external_size = 1;
4670 mddev->array_sectors = sectors;
4672 set_capacity(mddev->gendisk, mddev->array_sectors);
4673 revalidate_disk(mddev->gendisk);
4678 static struct md_sysfs_entry md_array_size =
4679 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4682 static struct attribute *md_default_attrs[] = {
4685 &md_raid_disks.attr,
4686 &md_chunk_size.attr,
4688 &md_resync_start.attr,
4690 &md_new_device.attr,
4691 &md_safe_delay.attr,
4692 &md_array_state.attr,
4693 &md_reshape_position.attr,
4694 &md_reshape_direction.attr,
4695 &md_array_size.attr,
4696 &max_corr_read_errors.attr,
4700 static struct attribute *md_redundancy_attrs[] = {
4702 &md_mismatches.attr,
4705 &md_sync_speed.attr,
4706 &md_sync_force_parallel.attr,
4707 &md_sync_completed.attr,
4710 &md_suspend_lo.attr,
4711 &md_suspend_hi.attr,
4716 static struct attribute_group md_redundancy_group = {
4718 .attrs = md_redundancy_attrs,
4723 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4725 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4726 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4731 spin_lock(&all_mddevs_lock);
4732 if (list_empty(&mddev->all_mddevs)) {
4733 spin_unlock(&all_mddevs_lock);
4737 spin_unlock(&all_mddevs_lock);
4739 rv = mddev_lock(mddev);
4741 rv = entry->show(mddev, page);
4742 mddev_unlock(mddev);
4749 md_attr_store(struct kobject *kobj, struct attribute *attr,
4750 const char *page, size_t length)
4752 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4753 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4758 if (!capable(CAP_SYS_ADMIN))
4760 spin_lock(&all_mddevs_lock);
4761 if (list_empty(&mddev->all_mddevs)) {
4762 spin_unlock(&all_mddevs_lock);
4766 spin_unlock(&all_mddevs_lock);
4767 if (entry->store == new_dev_store)
4768 flush_workqueue(md_misc_wq);
4769 rv = mddev_lock(mddev);
4771 rv = entry->store(mddev, page, length);
4772 mddev_unlock(mddev);
4778 static void md_free(struct kobject *ko)
4780 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4782 if (mddev->sysfs_state)
4783 sysfs_put(mddev->sysfs_state);
4785 if (mddev->gendisk) {
4786 del_gendisk(mddev->gendisk);
4787 put_disk(mddev->gendisk);
4790 blk_cleanup_queue(mddev->queue);
4795 static const struct sysfs_ops md_sysfs_ops = {
4796 .show = md_attr_show,
4797 .store = md_attr_store,
4799 static struct kobj_type md_ktype = {
4801 .sysfs_ops = &md_sysfs_ops,
4802 .default_attrs = md_default_attrs,
4807 static void mddev_delayed_delete(struct work_struct *ws)
4809 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4811 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4812 kobject_del(&mddev->kobj);
4813 kobject_put(&mddev->kobj);
4816 static int md_alloc(dev_t dev, char *name)
4818 static DEFINE_MUTEX(disks_mutex);
4819 struct mddev *mddev = mddev_find(dev);
4820 struct gendisk *disk;
4829 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4830 shift = partitioned ? MdpMinorShift : 0;
4831 unit = MINOR(mddev->unit) >> shift;
4833 /* wait for any previous instance of this device to be
4834 * completely removed (mddev_delayed_delete).
4836 flush_workqueue(md_misc_wq);
4838 mutex_lock(&disks_mutex);
4844 /* Need to ensure that 'name' is not a duplicate.
4846 struct mddev *mddev2;
4847 spin_lock(&all_mddevs_lock);
4849 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4850 if (mddev2->gendisk &&
4851 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4852 spin_unlock(&all_mddevs_lock);
4855 spin_unlock(&all_mddevs_lock);
4859 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4862 mddev->queue->queuedata = mddev;
4864 blk_queue_make_request(mddev->queue, md_make_request);
4865 blk_set_stacking_limits(&mddev->queue->limits);
4867 disk = alloc_disk(1 << shift);
4869 blk_cleanup_queue(mddev->queue);
4870 mddev->queue = NULL;
4873 disk->major = MAJOR(mddev->unit);
4874 disk->first_minor = unit << shift;
4876 strcpy(disk->disk_name, name);
4877 else if (partitioned)
4878 sprintf(disk->disk_name, "md_d%d", unit);
4880 sprintf(disk->disk_name, "md%d", unit);
4881 disk->fops = &md_fops;
4882 disk->private_data = mddev;
4883 disk->queue = mddev->queue;
4884 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4885 /* Allow extended partitions. This makes the
4886 * 'mdp' device redundant, but we can't really
4889 disk->flags |= GENHD_FL_EXT_DEVT;
4890 mddev->gendisk = disk;
4891 /* As soon as we call add_disk(), another thread could get
4892 * through to md_open, so make sure it doesn't get too far
4894 mutex_lock(&mddev->open_mutex);
4897 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4898 &disk_to_dev(disk)->kobj, "%s", "md");
4900 /* This isn't possible, but as kobject_init_and_add is marked
4901 * __must_check, we must do something with the result
4903 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4907 if (mddev->kobj.sd &&
4908 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4909 printk(KERN_DEBUG "pointless warning\n");
4910 mutex_unlock(&mddev->open_mutex);
4912 mutex_unlock(&disks_mutex);
4913 if (!error && mddev->kobj.sd) {
4914 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4915 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4921 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4923 md_alloc(dev, NULL);
4927 static int add_named_array(const char *val, struct kernel_param *kp)
4929 /* val must be "md_*" where * is not all digits.
4930 * We allocate an array with a large free minor number, and
4931 * set the name to val. val must not already be an active name.
4933 int len = strlen(val);
4934 char buf[DISK_NAME_LEN];
4936 while (len && val[len-1] == '\n')
4938 if (len >= DISK_NAME_LEN)
4940 strlcpy(buf, val, len+1);
4941 if (strncmp(buf, "md_", 3) != 0)
4943 return md_alloc(0, buf);
4946 static void md_safemode_timeout(unsigned long data)
4948 struct mddev *mddev = (struct mddev *) data;
4950 if (!atomic_read(&mddev->writes_pending)) {
4951 mddev->safemode = 1;
4952 if (mddev->external)
4953 sysfs_notify_dirent_safe(mddev->sysfs_state);
4955 md_wakeup_thread(mddev->thread);
4958 static int start_dirty_degraded;
4960 int md_run(struct mddev *mddev)
4963 struct md_rdev *rdev;
4964 struct md_personality *pers;
4966 if (list_empty(&mddev->disks))
4967 /* cannot run an array with no devices.. */
4972 /* Cannot run until previous stop completes properly */
4973 if (mddev->sysfs_active)
4977 * Analyze all RAID superblock(s)
4979 if (!mddev->raid_disks) {
4980 if (!mddev->persistent)
4985 if (mddev->level != LEVEL_NONE)
4986 request_module("md-level-%d", mddev->level);
4987 else if (mddev->clevel[0])
4988 request_module("md-%s", mddev->clevel);
4991 * Drop all container device buffers, from now on
4992 * the only valid external interface is through the md
4995 rdev_for_each(rdev, mddev) {
4996 if (test_bit(Faulty, &rdev->flags))
4998 sync_blockdev(rdev->bdev);
4999 invalidate_bdev(rdev->bdev);
5001 /* perform some consistency tests on the device.
5002 * We don't want the data to overlap the metadata,
5003 * Internal Bitmap issues have been handled elsewhere.
5005 if (rdev->meta_bdev) {
5006 /* Nothing to check */;
5007 } else if (rdev->data_offset < rdev->sb_start) {
5008 if (mddev->dev_sectors &&
5009 rdev->data_offset + mddev->dev_sectors
5011 printk("md: %s: data overlaps metadata\n",
5016 if (rdev->sb_start + rdev->sb_size/512
5017 > rdev->data_offset) {
5018 printk("md: %s: metadata overlaps data\n",
5023 sysfs_notify_dirent_safe(rdev->sysfs_state);
5026 if (mddev->bio_set == NULL)
5027 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5029 spin_lock(&pers_lock);
5030 pers = find_pers(mddev->level, mddev->clevel);
5031 if (!pers || !try_module_get(pers->owner)) {
5032 spin_unlock(&pers_lock);
5033 if (mddev->level != LEVEL_NONE)
5034 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5037 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5042 spin_unlock(&pers_lock);
5043 if (mddev->level != pers->level) {
5044 mddev->level = pers->level;
5045 mddev->new_level = pers->level;
5047 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5049 if (mddev->reshape_position != MaxSector &&
5050 pers->start_reshape == NULL) {
5051 /* This personality cannot handle reshaping... */
5053 module_put(pers->owner);
5057 if (pers->sync_request) {
5058 /* Warn if this is a potentially silly
5061 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5062 struct md_rdev *rdev2;
5065 rdev_for_each(rdev, mddev)
5066 rdev_for_each(rdev2, mddev) {
5068 rdev->bdev->bd_contains ==
5069 rdev2->bdev->bd_contains) {
5071 "%s: WARNING: %s appears to be"
5072 " on the same physical disk as"
5075 bdevname(rdev->bdev,b),
5076 bdevname(rdev2->bdev,b2));
5083 "True protection against single-disk"
5084 " failure might be compromised.\n");
5087 mddev->recovery = 0;
5088 /* may be over-ridden by personality */
5089 mddev->resync_max_sectors = mddev->dev_sectors;
5091 mddev->ok_start_degraded = start_dirty_degraded;
5093 if (start_readonly && mddev->ro == 0)
5094 mddev->ro = 2; /* read-only, but switch on first write */
5096 err = mddev->pers->run(mddev);
5098 printk(KERN_ERR "md: pers->run() failed ...\n");
5099 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5100 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5101 " but 'external_size' not in effect?\n", __func__);
5103 "md: invalid array_size %llu > default size %llu\n",
5104 (unsigned long long)mddev->array_sectors / 2,
5105 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5107 mddev->pers->stop(mddev);
5109 if (err == 0 && mddev->pers->sync_request &&
5110 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5111 err = bitmap_create(mddev);
5113 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5114 mdname(mddev), err);
5115 mddev->pers->stop(mddev);
5119 module_put(mddev->pers->owner);
5121 bitmap_destroy(mddev);
5124 if (mddev->pers->sync_request) {
5125 if (mddev->kobj.sd &&
5126 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5128 "md: cannot register extra attributes for %s\n",
5130 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5131 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5134 atomic_set(&mddev->writes_pending,0);
5135 atomic_set(&mddev->max_corr_read_errors,
5136 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5137 mddev->safemode = 0;
5138 mddev->safemode_timer.function = md_safemode_timeout;
5139 mddev->safemode_timer.data = (unsigned long) mddev;
5140 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5144 rdev_for_each(rdev, mddev)
5145 if (rdev->raid_disk >= 0)
5146 if (sysfs_link_rdev(mddev, rdev))
5147 /* failure here is OK */;
5149 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5152 md_update_sb(mddev, 0);
5154 md_new_event(mddev);
5155 sysfs_notify_dirent_safe(mddev->sysfs_state);
5156 sysfs_notify_dirent_safe(mddev->sysfs_action);
5157 sysfs_notify(&mddev->kobj, NULL, "degraded");
5160 EXPORT_SYMBOL_GPL(md_run);
5162 static int do_md_run(struct mddev *mddev)
5166 err = md_run(mddev);
5169 err = bitmap_load(mddev);
5171 bitmap_destroy(mddev);
5175 md_wakeup_thread(mddev->thread);
5176 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5178 set_capacity(mddev->gendisk, mddev->array_sectors);
5179 revalidate_disk(mddev->gendisk);
5181 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5186 static int restart_array(struct mddev *mddev)
5188 struct gendisk *disk = mddev->gendisk;
5190 /* Complain if it has no devices */
5191 if (list_empty(&mddev->disks))
5197 mddev->safemode = 0;
5199 set_disk_ro(disk, 0);
5200 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5202 /* Kick recovery or resync if necessary */
5203 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5204 md_wakeup_thread(mddev->thread);
5205 md_wakeup_thread(mddev->sync_thread);
5206 sysfs_notify_dirent_safe(mddev->sysfs_state);
5210 /* similar to deny_write_access, but accounts for our holding a reference
5211 * to the file ourselves */
5212 static int deny_bitmap_write_access(struct file * file)
5214 struct inode *inode = file->f_mapping->host;
5216 spin_lock(&inode->i_lock);
5217 if (atomic_read(&inode->i_writecount) > 1) {
5218 spin_unlock(&inode->i_lock);
5221 atomic_set(&inode->i_writecount, -1);
5222 spin_unlock(&inode->i_lock);
5227 void restore_bitmap_write_access(struct file *file)
5229 struct inode *inode = file->f_mapping->host;
5231 spin_lock(&inode->i_lock);
5232 atomic_set(&inode->i_writecount, 1);
5233 spin_unlock(&inode->i_lock);
5236 static void md_clean(struct mddev *mddev)
5238 mddev->array_sectors = 0;
5239 mddev->external_size = 0;
5240 mddev->dev_sectors = 0;
5241 mddev->raid_disks = 0;
5242 mddev->recovery_cp = 0;
5243 mddev->resync_min = 0;
5244 mddev->resync_max = MaxSector;
5245 mddev->reshape_position = MaxSector;
5246 mddev->external = 0;
5247 mddev->persistent = 0;
5248 mddev->level = LEVEL_NONE;
5249 mddev->clevel[0] = 0;
5252 mddev->metadata_type[0] = 0;
5253 mddev->chunk_sectors = 0;
5254 mddev->ctime = mddev->utime = 0;
5256 mddev->max_disks = 0;
5258 mddev->can_decrease_events = 0;
5259 mddev->delta_disks = 0;
5260 mddev->reshape_backwards = 0;
5261 mddev->new_level = LEVEL_NONE;
5262 mddev->new_layout = 0;
5263 mddev->new_chunk_sectors = 0;
5264 mddev->curr_resync = 0;
5265 atomic64_set(&mddev->resync_mismatches, 0);
5266 mddev->suspend_lo = mddev->suspend_hi = 0;
5267 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5268 mddev->recovery = 0;
5271 mddev->degraded = 0;
5272 mddev->safemode = 0;
5273 mddev->merge_check_needed = 0;
5274 mddev->bitmap_info.offset = 0;
5275 mddev->bitmap_info.default_offset = 0;
5276 mddev->bitmap_info.default_space = 0;
5277 mddev->bitmap_info.chunksize = 0;
5278 mddev->bitmap_info.daemon_sleep = 0;
5279 mddev->bitmap_info.max_write_behind = 0;
5282 static void __md_stop_writes(struct mddev *mddev)
5284 if (mddev->sync_thread) {
5285 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5286 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5287 reap_sync_thread(mddev);
5290 del_timer_sync(&mddev->safemode_timer);
5292 bitmap_flush(mddev);
5293 md_super_wait(mddev);
5295 if (!mddev->in_sync || mddev->flags) {
5296 /* mark array as shutdown cleanly */
5298 md_update_sb(mddev, 1);
5302 void md_stop_writes(struct mddev *mddev)
5305 __md_stop_writes(mddev);
5306 mddev_unlock(mddev);
5308 EXPORT_SYMBOL_GPL(md_stop_writes);
5310 static void __md_stop(struct mddev *mddev)
5313 mddev->pers->stop(mddev);
5314 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5315 mddev->to_remove = &md_redundancy_group;
5316 module_put(mddev->pers->owner);
5318 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5321 void md_stop(struct mddev *mddev)
5323 /* stop the array and free an attached data structures.
5324 * This is called from dm-raid
5327 bitmap_destroy(mddev);
5329 bioset_free(mddev->bio_set);
5332 EXPORT_SYMBOL_GPL(md_stop);
5334 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5337 mutex_lock(&mddev->open_mutex);
5338 if (atomic_read(&mddev->openers) > !!bdev) {
5339 printk("md: %s still in use.\n",mdname(mddev));
5344 sync_blockdev(bdev);
5346 __md_stop_writes(mddev);
5352 set_disk_ro(mddev->gendisk, 1);
5353 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5354 sysfs_notify_dirent_safe(mddev->sysfs_state);
5358 mutex_unlock(&mddev->open_mutex);
5363 * 0 - completely stop and dis-assemble array
5364 * 2 - stop but do not disassemble array
5366 static int do_md_stop(struct mddev * mddev, int mode,
5367 struct block_device *bdev)
5369 struct gendisk *disk = mddev->gendisk;
5370 struct md_rdev *rdev;
5372 mutex_lock(&mddev->open_mutex);
5373 if (atomic_read(&mddev->openers) > !!bdev ||
5374 mddev->sysfs_active) {
5375 printk("md: %s still in use.\n",mdname(mddev));
5376 mutex_unlock(&mddev->open_mutex);
5380 /* It is possible IO was issued on some other
5381 * open file which was closed before we took ->open_mutex.
5382 * As that was not the last close __blkdev_put will not
5383 * have called sync_blockdev, so we must.
5385 sync_blockdev(bdev);
5389 set_disk_ro(disk, 0);
5391 __md_stop_writes(mddev);
5393 mddev->queue->merge_bvec_fn = NULL;
5394 mddev->queue->backing_dev_info.congested_fn = NULL;
5396 /* tell userspace to handle 'inactive' */
5397 sysfs_notify_dirent_safe(mddev->sysfs_state);
5399 rdev_for_each(rdev, mddev)
5400 if (rdev->raid_disk >= 0)
5401 sysfs_unlink_rdev(mddev, rdev);
5403 set_capacity(disk, 0);
5404 mutex_unlock(&mddev->open_mutex);
5406 revalidate_disk(disk);
5411 mutex_unlock(&mddev->open_mutex);
5413 * Free resources if final stop
5416 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5418 bitmap_destroy(mddev);
5419 if (mddev->bitmap_info.file) {
5420 restore_bitmap_write_access(mddev->bitmap_info.file);
5421 fput(mddev->bitmap_info.file);
5422 mddev->bitmap_info.file = NULL;
5424 mddev->bitmap_info.offset = 0;
5426 export_array(mddev);
5429 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5430 if (mddev->hold_active == UNTIL_STOP)
5431 mddev->hold_active = 0;
5433 blk_integrity_unregister(disk);
5434 md_new_event(mddev);
5435 sysfs_notify_dirent_safe(mddev->sysfs_state);
5440 static void autorun_array(struct mddev *mddev)
5442 struct md_rdev *rdev;
5445 if (list_empty(&mddev->disks))
5448 printk(KERN_INFO "md: running: ");
5450 rdev_for_each(rdev, mddev) {
5451 char b[BDEVNAME_SIZE];
5452 printk("<%s>", bdevname(rdev->bdev,b));
5456 err = do_md_run(mddev);
5458 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5459 do_md_stop(mddev, 0, NULL);
5464 * lets try to run arrays based on all disks that have arrived
5465 * until now. (those are in pending_raid_disks)
5467 * the method: pick the first pending disk, collect all disks with
5468 * the same UUID, remove all from the pending list and put them into
5469 * the 'same_array' list. Then order this list based on superblock
5470 * update time (freshest comes first), kick out 'old' disks and
5471 * compare superblocks. If everything's fine then run it.
5473 * If "unit" is allocated, then bump its reference count
5475 static void autorun_devices(int part)
5477 struct md_rdev *rdev0, *rdev, *tmp;
5478 struct mddev *mddev;
5479 char b[BDEVNAME_SIZE];
5481 printk(KERN_INFO "md: autorun ...\n");
5482 while (!list_empty(&pending_raid_disks)) {
5485 LIST_HEAD(candidates);
5486 rdev0 = list_entry(pending_raid_disks.next,
5487 struct md_rdev, same_set);
5489 printk(KERN_INFO "md: considering %s ...\n",
5490 bdevname(rdev0->bdev,b));
5491 INIT_LIST_HEAD(&candidates);
5492 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5493 if (super_90_load(rdev, rdev0, 0) >= 0) {
5494 printk(KERN_INFO "md: adding %s ...\n",
5495 bdevname(rdev->bdev,b));
5496 list_move(&rdev->same_set, &candidates);
5499 * now we have a set of devices, with all of them having
5500 * mostly sane superblocks. It's time to allocate the
5504 dev = MKDEV(mdp_major,
5505 rdev0->preferred_minor << MdpMinorShift);
5506 unit = MINOR(dev) >> MdpMinorShift;
5508 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5511 if (rdev0->preferred_minor != unit) {
5512 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5513 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5517 md_probe(dev, NULL, NULL);
5518 mddev = mddev_find(dev);
5519 if (!mddev || !mddev->gendisk) {
5523 "md: cannot allocate memory for md drive.\n");
5526 if (mddev_lock(mddev))
5527 printk(KERN_WARNING "md: %s locked, cannot run\n",
5529 else if (mddev->raid_disks || mddev->major_version
5530 || !list_empty(&mddev->disks)) {
5532 "md: %s already running, cannot run %s\n",
5533 mdname(mddev), bdevname(rdev0->bdev,b));
5534 mddev_unlock(mddev);
5536 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5537 mddev->persistent = 1;
5538 rdev_for_each_list(rdev, tmp, &candidates) {
5539 list_del_init(&rdev->same_set);
5540 if (bind_rdev_to_array(rdev, mddev))
5543 autorun_array(mddev);
5544 mddev_unlock(mddev);
5546 /* on success, candidates will be empty, on error
5549 rdev_for_each_list(rdev, tmp, &candidates) {
5550 list_del_init(&rdev->same_set);
5555 printk(KERN_INFO "md: ... autorun DONE.\n");
5557 #endif /* !MODULE */
5559 static int get_version(void __user * arg)
5563 ver.major = MD_MAJOR_VERSION;
5564 ver.minor = MD_MINOR_VERSION;
5565 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5567 if (copy_to_user(arg, &ver, sizeof(ver)))
5573 static int get_array_info(struct mddev * mddev, void __user * arg)
5575 mdu_array_info_t info;
5576 int nr,working,insync,failed,spare;
5577 struct md_rdev *rdev;
5579 nr = working = insync = failed = spare = 0;
5581 rdev_for_each_rcu(rdev, mddev) {
5583 if (test_bit(Faulty, &rdev->flags))
5587 if (test_bit(In_sync, &rdev->flags))
5595 info.major_version = mddev->major_version;
5596 info.minor_version = mddev->minor_version;
5597 info.patch_version = MD_PATCHLEVEL_VERSION;
5598 info.ctime = mddev->ctime;
5599 info.level = mddev->level;
5600 info.size = mddev->dev_sectors / 2;
5601 if (info.size != mddev->dev_sectors / 2) /* overflow */
5604 info.raid_disks = mddev->raid_disks;
5605 info.md_minor = mddev->md_minor;
5606 info.not_persistent= !mddev->persistent;
5608 info.utime = mddev->utime;
5611 info.state = (1<<MD_SB_CLEAN);
5612 if (mddev->bitmap && mddev->bitmap_info.offset)
5613 info.state = (1<<MD_SB_BITMAP_PRESENT);
5614 info.active_disks = insync;
5615 info.working_disks = working;
5616 info.failed_disks = failed;
5617 info.spare_disks = spare;
5619 info.layout = mddev->layout;
5620 info.chunk_size = mddev->chunk_sectors << 9;
5622 if (copy_to_user(arg, &info, sizeof(info)))
5628 static int get_bitmap_file(struct mddev * mddev, void __user * arg)
5630 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5631 char *ptr, *buf = NULL;
5634 if (md_allow_write(mddev))
5635 file = kmalloc(sizeof(*file), GFP_NOIO);
5637 file = kmalloc(sizeof(*file), GFP_KERNEL);
5642 /* bitmap disabled, zero the first byte and copy out */
5643 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
5644 file->pathname[0] = '\0';
5648 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5652 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5653 buf, sizeof(file->pathname));
5657 strcpy(file->pathname, ptr);
5661 if (copy_to_user(arg, file, sizeof(*file)))
5669 static int get_disk_info(struct mddev * mddev, void __user * arg)
5671 mdu_disk_info_t info;
5672 struct md_rdev *rdev;
5674 if (copy_from_user(&info, arg, sizeof(info)))
5678 rdev = find_rdev_nr_rcu(mddev, info.number);
5680 info.major = MAJOR(rdev->bdev->bd_dev);
5681 info.minor = MINOR(rdev->bdev->bd_dev);
5682 info.raid_disk = rdev->raid_disk;
5684 if (test_bit(Faulty, &rdev->flags))
5685 info.state |= (1<<MD_DISK_FAULTY);
5686 else if (test_bit(In_sync, &rdev->flags)) {
5687 info.state |= (1<<MD_DISK_ACTIVE);
5688 info.state |= (1<<MD_DISK_SYNC);
5690 if (test_bit(WriteMostly, &rdev->flags))
5691 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5693 info.major = info.minor = 0;
5694 info.raid_disk = -1;
5695 info.state = (1<<MD_DISK_REMOVED);
5699 if (copy_to_user(arg, &info, sizeof(info)))
5705 static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
5707 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5708 struct md_rdev *rdev;
5709 dev_t dev = MKDEV(info->major,info->minor);
5711 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5714 if (!mddev->raid_disks) {
5716 /* expecting a device which has a superblock */
5717 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5720 "md: md_import_device returned %ld\n",
5722 return PTR_ERR(rdev);
5724 if (!list_empty(&mddev->disks)) {
5725 struct md_rdev *rdev0
5726 = list_entry(mddev->disks.next,
5727 struct md_rdev, same_set);
5728 err = super_types[mddev->major_version]
5729 .load_super(rdev, rdev0, mddev->minor_version);
5732 "md: %s has different UUID to %s\n",
5733 bdevname(rdev->bdev,b),
5734 bdevname(rdev0->bdev,b2));
5739 err = bind_rdev_to_array(rdev, mddev);
5746 * add_new_disk can be used once the array is assembled
5747 * to add "hot spares". They must already have a superblock
5752 if (!mddev->pers->hot_add_disk) {
5754 "%s: personality does not support diskops!\n",
5758 if (mddev->persistent)
5759 rdev = md_import_device(dev, mddev->major_version,
5760 mddev->minor_version);
5762 rdev = md_import_device(dev, -1, -1);
5765 "md: md_import_device returned %ld\n",
5767 return PTR_ERR(rdev);
5769 /* set saved_raid_disk if appropriate */
5770 if (!mddev->persistent) {
5771 if (info->state & (1<<MD_DISK_SYNC) &&
5772 info->raid_disk < mddev->raid_disks) {
5773 rdev->raid_disk = info->raid_disk;
5774 set_bit(In_sync, &rdev->flags);
5776 rdev->raid_disk = -1;
5778 super_types[mddev->major_version].
5779 validate_super(mddev, rdev);
5780 if ((info->state & (1<<MD_DISK_SYNC)) &&
5781 rdev->raid_disk != info->raid_disk) {
5782 /* This was a hot-add request, but events doesn't
5783 * match, so reject it.
5789 if (test_bit(In_sync, &rdev->flags))
5790 rdev->saved_raid_disk = rdev->raid_disk;
5792 rdev->saved_raid_disk = -1;
5794 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5795 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5796 set_bit(WriteMostly, &rdev->flags);
5798 clear_bit(WriteMostly, &rdev->flags);
5800 rdev->raid_disk = -1;
5801 err = bind_rdev_to_array(rdev, mddev);
5802 if (!err && !mddev->pers->hot_remove_disk) {
5803 /* If there is hot_add_disk but no hot_remove_disk
5804 * then added disks for geometry changes,
5805 * and should be added immediately.
5807 super_types[mddev->major_version].
5808 validate_super(mddev, rdev);
5809 err = mddev->pers->hot_add_disk(mddev, rdev);
5811 unbind_rdev_from_array(rdev);
5816 sysfs_notify_dirent_safe(rdev->sysfs_state);
5818 md_update_sb(mddev, 1);
5819 if (mddev->degraded)
5820 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5821 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5823 md_new_event(mddev);
5824 md_wakeup_thread(mddev->thread);
5828 /* otherwise, add_new_disk is only allowed
5829 * for major_version==0 superblocks
5831 if (mddev->major_version != 0) {
5832 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5837 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5839 rdev = md_import_device(dev, -1, 0);
5842 "md: error, md_import_device() returned %ld\n",
5844 return PTR_ERR(rdev);
5846 rdev->desc_nr = info->number;
5847 if (info->raid_disk < mddev->raid_disks)
5848 rdev->raid_disk = info->raid_disk;
5850 rdev->raid_disk = -1;
5852 if (rdev->raid_disk < mddev->raid_disks)
5853 if (info->state & (1<<MD_DISK_SYNC))
5854 set_bit(In_sync, &rdev->flags);
5856 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5857 set_bit(WriteMostly, &rdev->flags);
5859 if (!mddev->persistent) {
5860 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5861 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5863 rdev->sb_start = calc_dev_sboffset(rdev);
5864 rdev->sectors = rdev->sb_start;
5866 err = bind_rdev_to_array(rdev, mddev);
5876 static int hot_remove_disk(struct mddev * mddev, dev_t dev)
5878 char b[BDEVNAME_SIZE];
5879 struct md_rdev *rdev;
5881 rdev = find_rdev(mddev, dev);
5885 if (rdev->raid_disk >= 0)
5888 kick_rdev_from_array(rdev);
5889 md_update_sb(mddev, 1);
5890 md_new_event(mddev);
5894 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5895 bdevname(rdev->bdev,b), mdname(mddev));
5899 static int hot_add_disk(struct mddev * mddev, dev_t dev)
5901 char b[BDEVNAME_SIZE];
5903 struct md_rdev *rdev;
5908 if (mddev->major_version != 0) {
5909 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5910 " version-0 superblocks.\n",
5914 if (!mddev->pers->hot_add_disk) {
5916 "%s: personality does not support diskops!\n",
5921 rdev = md_import_device(dev, -1, 0);
5924 "md: error, md_import_device() returned %ld\n",
5929 if (mddev->persistent)
5930 rdev->sb_start = calc_dev_sboffset(rdev);
5932 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5934 rdev->sectors = rdev->sb_start;
5936 if (test_bit(Faulty, &rdev->flags)) {
5938 "md: can not hot-add faulty %s disk to %s!\n",
5939 bdevname(rdev->bdev,b), mdname(mddev));
5943 clear_bit(In_sync, &rdev->flags);
5945 rdev->saved_raid_disk = -1;
5946 err = bind_rdev_to_array(rdev, mddev);
5951 * The rest should better be atomic, we can have disk failures
5952 * noticed in interrupt contexts ...
5955 rdev->raid_disk = -1;
5957 md_update_sb(mddev, 1);
5960 * Kick recovery, maybe this spare has to be added to the
5961 * array immediately.
5963 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5964 md_wakeup_thread(mddev->thread);
5965 md_new_event(mddev);
5973 static int set_bitmap_file(struct mddev *mddev, int fd)
5978 if (!mddev->pers->quiesce)
5980 if (mddev->recovery || mddev->sync_thread)
5982 /* we should be able to change the bitmap.. */
5988 return -EEXIST; /* cannot add when bitmap is present */
5989 mddev->bitmap_info.file = fget(fd);
5991 if (mddev->bitmap_info.file == NULL) {
5992 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5997 err = deny_bitmap_write_access(mddev->bitmap_info.file);
5999 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6001 fput(mddev->bitmap_info.file);
6002 mddev->bitmap_info.file = NULL;
6005 mddev->bitmap_info.offset = 0; /* file overrides offset */
6006 } else if (mddev->bitmap == NULL)
6007 return -ENOENT; /* cannot remove what isn't there */
6010 mddev->pers->quiesce(mddev, 1);
6012 err = bitmap_create(mddev);
6014 err = bitmap_load(mddev);
6016 if (fd < 0 || err) {
6017 bitmap_destroy(mddev);
6018 fd = -1; /* make sure to put the file */
6020 mddev->pers->quiesce(mddev, 0);
6023 if (mddev->bitmap_info.file) {
6024 restore_bitmap_write_access(mddev->bitmap_info.file);
6025 fput(mddev->bitmap_info.file);
6027 mddev->bitmap_info.file = NULL;
6034 * set_array_info is used two different ways
6035 * The original usage is when creating a new array.
6036 * In this usage, raid_disks is > 0 and it together with
6037 * level, size, not_persistent,layout,chunksize determine the
6038 * shape of the array.
6039 * This will always create an array with a type-0.90.0 superblock.
6040 * The newer usage is when assembling an array.
6041 * In this case raid_disks will be 0, and the major_version field is
6042 * use to determine which style super-blocks are to be found on the devices.
6043 * The minor and patch _version numbers are also kept incase the
6044 * super_block handler wishes to interpret them.
6046 static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
6049 if (info->raid_disks == 0) {
6050 /* just setting version number for superblock loading */
6051 if (info->major_version < 0 ||
6052 info->major_version >= ARRAY_SIZE(super_types) ||
6053 super_types[info->major_version].name == NULL) {
6054 /* maybe try to auto-load a module? */
6056 "md: superblock version %d not known\n",
6057 info->major_version);
6060 mddev->major_version = info->major_version;
6061 mddev->minor_version = info->minor_version;
6062 mddev->patch_version = info->patch_version;
6063 mddev->persistent = !info->not_persistent;
6064 /* ensure mddev_put doesn't delete this now that there
6065 * is some minimal configuration.
6067 mddev->ctime = get_seconds();
6070 mddev->major_version = MD_MAJOR_VERSION;
6071 mddev->minor_version = MD_MINOR_VERSION;
6072 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6073 mddev->ctime = get_seconds();
6075 mddev->level = info->level;
6076 mddev->clevel[0] = 0;
6077 mddev->dev_sectors = 2 * (sector_t)info->size;
6078 mddev->raid_disks = info->raid_disks;
6079 /* don't set md_minor, it is determined by which /dev/md* was
6082 if (info->state & (1<<MD_SB_CLEAN))
6083 mddev->recovery_cp = MaxSector;
6085 mddev->recovery_cp = 0;
6086 mddev->persistent = ! info->not_persistent;
6087 mddev->external = 0;
6089 mddev->layout = info->layout;
6090 mddev->chunk_sectors = info->chunk_size >> 9;
6092 mddev->max_disks = MD_SB_DISKS;
6094 if (mddev->persistent)
6096 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6098 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6099 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6100 mddev->bitmap_info.offset = 0;
6102 mddev->reshape_position = MaxSector;
6105 * Generate a 128 bit UUID
6107 get_random_bytes(mddev->uuid, 16);
6109 mddev->new_level = mddev->level;
6110 mddev->new_chunk_sectors = mddev->chunk_sectors;
6111 mddev->new_layout = mddev->layout;
6112 mddev->delta_disks = 0;
6113 mddev->reshape_backwards = 0;
6118 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6120 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6122 if (mddev->external_size)
6125 mddev->array_sectors = array_sectors;
6127 EXPORT_SYMBOL(md_set_array_sectors);
6129 static int update_size(struct mddev *mddev, sector_t num_sectors)
6131 struct md_rdev *rdev;
6133 int fit = (num_sectors == 0);
6135 if (mddev->pers->resize == NULL)
6137 /* The "num_sectors" is the number of sectors of each device that
6138 * is used. This can only make sense for arrays with redundancy.
6139 * linear and raid0 always use whatever space is available. We can only
6140 * consider changing this number if no resync or reconstruction is
6141 * happening, and if the new size is acceptable. It must fit before the
6142 * sb_start or, if that is <data_offset, it must fit before the size
6143 * of each device. If num_sectors is zero, we find the largest size
6146 if (mddev->sync_thread)
6149 rdev_for_each(rdev, mddev) {
6150 sector_t avail = rdev->sectors;
6152 if (fit && (num_sectors == 0 || num_sectors > avail))
6153 num_sectors = avail;
6154 if (avail < num_sectors)
6157 rv = mddev->pers->resize(mddev, num_sectors);
6159 revalidate_disk(mddev->gendisk);
6163 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6166 struct md_rdev *rdev;
6167 /* change the number of raid disks */
6168 if (mddev->pers->check_reshape == NULL)
6170 if (raid_disks <= 0 ||
6171 (mddev->max_disks && raid_disks >= mddev->max_disks))
6173 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
6176 rdev_for_each(rdev, mddev) {
6177 if (mddev->raid_disks < raid_disks &&
6178 rdev->data_offset < rdev->new_data_offset)
6180 if (mddev->raid_disks > raid_disks &&
6181 rdev->data_offset > rdev->new_data_offset)
6185 mddev->delta_disks = raid_disks - mddev->raid_disks;
6186 if (mddev->delta_disks < 0)
6187 mddev->reshape_backwards = 1;
6188 else if (mddev->delta_disks > 0)
6189 mddev->reshape_backwards = 0;
6191 rv = mddev->pers->check_reshape(mddev);
6193 mddev->delta_disks = 0;
6194 mddev->reshape_backwards = 0;
6201 * update_array_info is used to change the configuration of an
6203 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6204 * fields in the info are checked against the array.
6205 * Any differences that cannot be handled will cause an error.
6206 * Normally, only one change can be managed at a time.
6208 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6214 /* calculate expected state,ignoring low bits */
6215 if (mddev->bitmap && mddev->bitmap_info.offset)
6216 state |= (1 << MD_SB_BITMAP_PRESENT);
6218 if (mddev->major_version != info->major_version ||
6219 mddev->minor_version != info->minor_version ||
6220 /* mddev->patch_version != info->patch_version || */
6221 mddev->ctime != info->ctime ||
6222 mddev->level != info->level ||
6223 /* mddev->layout != info->layout || */
6224 !mddev->persistent != info->not_persistent||
6225 mddev->chunk_sectors != info->chunk_size >> 9 ||
6226 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6227 ((state^info->state) & 0xfffffe00)
6230 /* Check there is only one change */
6231 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6233 if (mddev->raid_disks != info->raid_disks)
6235 if (mddev->layout != info->layout)
6237 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6244 if (mddev->layout != info->layout) {
6246 * we don't need to do anything at the md level, the
6247 * personality will take care of it all.
6249 if (mddev->pers->check_reshape == NULL)
6252 mddev->new_layout = info->layout;
6253 rv = mddev->pers->check_reshape(mddev);
6255 mddev->new_layout = mddev->layout;
6259 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6260 rv = update_size(mddev, (sector_t)info->size * 2);
6262 if (mddev->raid_disks != info->raid_disks)
6263 rv = update_raid_disks(mddev, info->raid_disks);
6265 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6266 if (mddev->pers->quiesce == NULL)
6268 if (mddev->recovery || mddev->sync_thread)
6270 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6271 /* add the bitmap */
6274 if (mddev->bitmap_info.default_offset == 0)
6276 mddev->bitmap_info.offset =
6277 mddev->bitmap_info.default_offset;
6278 mddev->bitmap_info.space =
6279 mddev->bitmap_info.default_space;
6280 mddev->pers->quiesce(mddev, 1);
6281 rv = bitmap_create(mddev);
6283 rv = bitmap_load(mddev);
6285 bitmap_destroy(mddev);
6286 mddev->pers->quiesce(mddev, 0);
6288 /* remove the bitmap */
6291 if (mddev->bitmap->storage.file)
6293 mddev->pers->quiesce(mddev, 1);
6294 bitmap_destroy(mddev);
6295 mddev->pers->quiesce(mddev, 0);
6296 mddev->bitmap_info.offset = 0;
6299 md_update_sb(mddev, 1);
6303 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6305 struct md_rdev *rdev;
6308 if (mddev->pers == NULL)
6312 rdev = find_rdev_rcu(mddev, dev);
6316 md_error(mddev, rdev);
6317 if (!test_bit(Faulty, &rdev->flags))
6325 * We have a problem here : there is no easy way to give a CHS
6326 * virtual geometry. We currently pretend that we have a 2 heads
6327 * 4 sectors (with a BIG number of cylinders...). This drives
6328 * dosfs just mad... ;-)
6330 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6332 struct mddev *mddev = bdev->bd_disk->private_data;
6336 geo->cylinders = mddev->array_sectors / 8;
6340 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6341 unsigned int cmd, unsigned long arg)
6344 void __user *argp = (void __user *)arg;
6345 struct mddev *mddev = NULL;
6350 case GET_ARRAY_INFO:
6354 if (!capable(CAP_SYS_ADMIN))
6359 * Commands dealing with the RAID driver but not any
6364 err = get_version(argp);
6367 case PRINT_RAID_DEBUG:
6375 autostart_arrays(arg);
6382 * Commands creating/starting a new array:
6385 mddev = bdev->bd_disk->private_data;
6392 /* Some actions do not requires the mutex */
6394 case GET_ARRAY_INFO:
6395 if (!mddev->raid_disks && !mddev->external)
6398 err = get_array_info(mddev, argp);
6402 if (!mddev->raid_disks && !mddev->external)
6405 err = get_disk_info(mddev, argp);
6408 case SET_DISK_FAULTY:
6409 err = set_disk_faulty(mddev, new_decode_dev(arg));
6413 if (cmd == ADD_NEW_DISK)
6414 /* need to ensure md_delayed_delete() has completed */
6415 flush_workqueue(md_misc_wq);
6417 err = mddev_lock(mddev);
6420 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6425 if (cmd == SET_ARRAY_INFO) {
6426 mdu_array_info_t info;
6428 memset(&info, 0, sizeof(info));
6429 else if (copy_from_user(&info, argp, sizeof(info))) {
6434 err = update_array_info(mddev, &info);
6436 printk(KERN_WARNING "md: couldn't update"
6437 " array info. %d\n", err);
6442 if (!list_empty(&mddev->disks)) {
6444 "md: array %s already has disks!\n",
6449 if (mddev->raid_disks) {
6451 "md: array %s already initialised!\n",
6456 err = set_array_info(mddev, &info);
6458 printk(KERN_WARNING "md: couldn't set"
6459 " array info. %d\n", err);
6466 * Commands querying/configuring an existing array:
6468 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6469 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6470 if ((!mddev->raid_disks && !mddev->external)
6471 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6472 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6473 && cmd != GET_BITMAP_FILE) {
6479 * Commands even a read-only array can execute:
6482 case GET_BITMAP_FILE:
6483 err = get_bitmap_file(mddev, argp);
6486 case RESTART_ARRAY_RW:
6487 err = restart_array(mddev);
6491 err = do_md_stop(mddev, 0, bdev);
6495 err = md_set_readonly(mddev, bdev);
6499 if (get_user(ro, (int __user *)(arg))) {
6505 /* if the bdev is going readonly the value of mddev->ro
6506 * does not matter, no writes are coming
6511 /* are we are already prepared for writes? */
6515 /* transitioning to readauto need only happen for
6516 * arrays that call md_write_start
6519 err = restart_array(mddev);
6522 set_disk_ro(mddev->gendisk, 0);
6529 * The remaining ioctls are changing the state of the
6530 * superblock, so we do not allow them on read-only arrays.
6531 * However non-MD ioctls (e.g. get-size) will still come through
6532 * here and hit the 'default' below, so only disallow
6533 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6535 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6536 if (mddev->ro == 2) {
6538 sysfs_notify_dirent_safe(mddev->sysfs_state);
6539 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6540 /* mddev_unlock will wake thread */
6541 /* If a device failed while we were read-only, we
6542 * need to make sure the metadata is updated now.
6544 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6545 mddev_unlock(mddev);
6546 wait_event(mddev->sb_wait,
6547 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6548 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6560 mdu_disk_info_t info;
6561 if (copy_from_user(&info, argp, sizeof(info)))
6564 err = add_new_disk(mddev, &info);
6568 case HOT_REMOVE_DISK:
6569 err = hot_remove_disk(mddev, new_decode_dev(arg));
6573 err = hot_add_disk(mddev, new_decode_dev(arg));
6577 err = do_md_run(mddev);
6580 case SET_BITMAP_FILE:
6581 err = set_bitmap_file(mddev, (int)arg);
6591 if (mddev->hold_active == UNTIL_IOCTL &&
6593 mddev->hold_active = 0;
6594 mddev_unlock(mddev);
6603 #ifdef CONFIG_COMPAT
6604 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6605 unsigned int cmd, unsigned long arg)
6608 case HOT_REMOVE_DISK:
6610 case SET_DISK_FAULTY:
6611 case SET_BITMAP_FILE:
6612 /* These take in integer arg, do not convert */
6615 arg = (unsigned long)compat_ptr(arg);
6619 return md_ioctl(bdev, mode, cmd, arg);
6621 #endif /* CONFIG_COMPAT */
6623 static int md_open(struct block_device *bdev, fmode_t mode)
6626 * Succeed if we can lock the mddev, which confirms that
6627 * it isn't being stopped right now.
6629 struct mddev *mddev = mddev_find(bdev->bd_dev);
6635 if (mddev->gendisk != bdev->bd_disk) {
6636 /* we are racing with mddev_put which is discarding this
6640 /* Wait until bdev->bd_disk is definitely gone */
6641 flush_workqueue(md_misc_wq);
6642 /* Then retry the open from the top */
6643 return -ERESTARTSYS;
6645 BUG_ON(mddev != bdev->bd_disk->private_data);
6647 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6651 atomic_inc(&mddev->openers);
6652 mutex_unlock(&mddev->open_mutex);
6654 check_disk_change(bdev);
6659 static int md_release(struct gendisk *disk, fmode_t mode)
6661 struct mddev *mddev = disk->private_data;
6664 atomic_dec(&mddev->openers);
6670 static int md_media_changed(struct gendisk *disk)
6672 struct mddev *mddev = disk->private_data;
6674 return mddev->changed;
6677 static int md_revalidate(struct gendisk *disk)
6679 struct mddev *mddev = disk->private_data;
6684 static const struct block_device_operations md_fops =
6686 .owner = THIS_MODULE,
6688 .release = md_release,
6690 #ifdef CONFIG_COMPAT
6691 .compat_ioctl = md_compat_ioctl,
6693 .getgeo = md_getgeo,
6694 .media_changed = md_media_changed,
6695 .revalidate_disk= md_revalidate,
6698 static int md_thread(void * arg)
6700 struct md_thread *thread = arg;
6703 * md_thread is a 'system-thread', it's priority should be very
6704 * high. We avoid resource deadlocks individually in each
6705 * raid personality. (RAID5 does preallocation) We also use RR and
6706 * the very same RT priority as kswapd, thus we will never get
6707 * into a priority inversion deadlock.
6709 * we definitely have to have equal or higher priority than
6710 * bdflush, otherwise bdflush will deadlock if there are too
6711 * many dirty RAID5 blocks.
6714 allow_signal(SIGKILL);
6715 while (!kthread_should_stop()) {
6717 /* We need to wait INTERRUPTIBLE so that
6718 * we don't add to the load-average.
6719 * That means we need to be sure no signals are
6722 if (signal_pending(current))
6723 flush_signals(current);
6725 wait_event_interruptible_timeout
6727 test_bit(THREAD_WAKEUP, &thread->flags)
6728 || kthread_should_stop(),
6731 clear_bit(THREAD_WAKEUP, &thread->flags);
6732 if (!kthread_should_stop())
6733 thread->run(thread);
6739 void md_wakeup_thread(struct md_thread *thread)
6742 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6743 set_bit(THREAD_WAKEUP, &thread->flags);
6744 wake_up(&thread->wqueue);
6748 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6749 struct mddev *mddev, const char *name)
6751 struct md_thread *thread;
6753 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6757 init_waitqueue_head(&thread->wqueue);
6760 thread->mddev = mddev;
6761 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6762 thread->tsk = kthread_run(md_thread, thread,
6764 mdname(thread->mddev),
6766 if (IS_ERR(thread->tsk)) {
6773 void md_unregister_thread(struct md_thread **threadp)
6775 struct md_thread *thread = *threadp;
6778 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6779 /* Locking ensures that mddev_unlock does not wake_up a
6780 * non-existent thread
6782 spin_lock(&pers_lock);
6784 spin_unlock(&pers_lock);
6786 kthread_stop(thread->tsk);
6790 void md_error(struct mddev *mddev, struct md_rdev *rdev)
6797 if (!rdev || test_bit(Faulty, &rdev->flags))
6800 if (!mddev->pers || !mddev->pers->error_handler)
6802 mddev->pers->error_handler(mddev,rdev);
6803 if (mddev->degraded)
6804 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6805 sysfs_notify_dirent_safe(rdev->sysfs_state);
6806 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6807 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6808 md_wakeup_thread(mddev->thread);
6809 if (mddev->event_work.func)
6810 queue_work(md_misc_wq, &mddev->event_work);
6811 md_new_event_inintr(mddev);
6814 /* seq_file implementation /proc/mdstat */
6816 static void status_unused(struct seq_file *seq)
6819 struct md_rdev *rdev;
6821 seq_printf(seq, "unused devices: ");
6823 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6824 char b[BDEVNAME_SIZE];
6826 seq_printf(seq, "%s ",
6827 bdevname(rdev->bdev,b));
6830 seq_printf(seq, "<none>");
6832 seq_printf(seq, "\n");
6836 static void status_resync(struct seq_file *seq, struct mddev * mddev)
6838 sector_t max_sectors, resync, res;
6839 unsigned long dt, db;
6842 unsigned int per_milli;
6844 if (mddev->curr_resync <= 3)
6847 resync = mddev->curr_resync
6848 - atomic_read(&mddev->recovery_active);
6850 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6851 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6852 max_sectors = mddev->resync_max_sectors;
6854 max_sectors = mddev->dev_sectors;
6857 * Should not happen.
6863 /* Pick 'scale' such that (resync>>scale)*1000 will fit
6864 * in a sector_t, and (max_sectors>>scale) will fit in a
6865 * u32, as those are the requirements for sector_div.
6866 * Thus 'scale' must be at least 10
6869 if (sizeof(sector_t) > sizeof(unsigned long)) {
6870 while ( max_sectors/2 > (1ULL<<(scale+32)))
6873 res = (resync>>scale)*1000;
6874 sector_div(res, (u32)((max_sectors>>scale)+1));
6878 int i, x = per_milli/50, y = 20-x;
6879 seq_printf(seq, "[");
6880 for (i = 0; i < x; i++)
6881 seq_printf(seq, "=");
6882 seq_printf(seq, ">");
6883 for (i = 0; i < y; i++)
6884 seq_printf(seq, ".");
6885 seq_printf(seq, "] ");
6887 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6888 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6890 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6892 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6893 "resync" : "recovery"))),
6894 per_milli/10, per_milli % 10,
6895 (unsigned long long) resync/2,
6896 (unsigned long long) max_sectors/2);
6899 * dt: time from mark until now
6900 * db: blocks written from mark until now
6901 * rt: remaining time
6903 * rt is a sector_t, so could be 32bit or 64bit.
6904 * So we divide before multiply in case it is 32bit and close
6906 * We scale the divisor (db) by 32 to avoid losing precision
6907 * near the end of resync when the number of remaining sectors
6909 * We then divide rt by 32 after multiplying by db to compensate.
6910 * The '+1' avoids division by zero if db is very small.
6912 dt = ((jiffies - mddev->resync_mark) / HZ);
6914 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6915 - mddev->resync_mark_cnt;
6917 rt = max_sectors - resync; /* number of remaining sectors */
6918 sector_div(rt, db/32+1);
6922 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6923 ((unsigned long)rt % 60)/6);
6925 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6928 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6930 struct list_head *tmp;
6932 struct mddev *mddev;
6940 spin_lock(&all_mddevs_lock);
6941 list_for_each(tmp,&all_mddevs)
6943 mddev = list_entry(tmp, struct mddev, all_mddevs);
6945 spin_unlock(&all_mddevs_lock);
6948 spin_unlock(&all_mddevs_lock);
6950 return (void*)2;/* tail */
6954 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6956 struct list_head *tmp;
6957 struct mddev *next_mddev, *mddev = v;
6963 spin_lock(&all_mddevs_lock);
6965 tmp = all_mddevs.next;
6967 tmp = mddev->all_mddevs.next;
6968 if (tmp != &all_mddevs)
6969 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
6971 next_mddev = (void*)2;
6974 spin_unlock(&all_mddevs_lock);
6982 static void md_seq_stop(struct seq_file *seq, void *v)
6984 struct mddev *mddev = v;
6986 if (mddev && v != (void*)1 && v != (void*)2)
6990 static int md_seq_show(struct seq_file *seq, void *v)
6992 struct mddev *mddev = v;
6994 struct md_rdev *rdev;
6996 if (v == (void*)1) {
6997 struct md_personality *pers;
6998 seq_printf(seq, "Personalities : ");
6999 spin_lock(&pers_lock);
7000 list_for_each_entry(pers, &pers_list, list)
7001 seq_printf(seq, "[%s] ", pers->name);
7003 spin_unlock(&pers_lock);
7004 seq_printf(seq, "\n");
7005 seq->poll_event = atomic_read(&md_event_count);
7008 if (v == (void*)2) {
7013 if (mddev_lock(mddev) < 0)
7016 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7017 seq_printf(seq, "%s : %sactive", mdname(mddev),
7018 mddev->pers ? "" : "in");
7021 seq_printf(seq, " (read-only)");
7023 seq_printf(seq, " (auto-read-only)");
7024 seq_printf(seq, " %s", mddev->pers->name);
7028 rdev_for_each(rdev, mddev) {
7029 char b[BDEVNAME_SIZE];
7030 seq_printf(seq, " %s[%d]",
7031 bdevname(rdev->bdev,b), rdev->desc_nr);
7032 if (test_bit(WriteMostly, &rdev->flags))
7033 seq_printf(seq, "(W)");
7034 if (test_bit(Faulty, &rdev->flags)) {
7035 seq_printf(seq, "(F)");
7038 if (rdev->raid_disk < 0)
7039 seq_printf(seq, "(S)"); /* spare */
7040 if (test_bit(Replacement, &rdev->flags))
7041 seq_printf(seq, "(R)");
7042 sectors += rdev->sectors;
7045 if (!list_empty(&mddev->disks)) {
7047 seq_printf(seq, "\n %llu blocks",
7048 (unsigned long long)
7049 mddev->array_sectors / 2);
7051 seq_printf(seq, "\n %llu blocks",
7052 (unsigned long long)sectors / 2);
7054 if (mddev->persistent) {
7055 if (mddev->major_version != 0 ||
7056 mddev->minor_version != 90) {
7057 seq_printf(seq," super %d.%d",
7058 mddev->major_version,
7059 mddev->minor_version);
7061 } else if (mddev->external)
7062 seq_printf(seq, " super external:%s",
7063 mddev->metadata_type);
7065 seq_printf(seq, " super non-persistent");
7068 mddev->pers->status(seq, mddev);
7069 seq_printf(seq, "\n ");
7070 if (mddev->pers->sync_request) {
7071 if (mddev->curr_resync > 2) {
7072 status_resync(seq, mddev);
7073 seq_printf(seq, "\n ");
7074 } else if (mddev->curr_resync >= 1)
7075 seq_printf(seq, "\tresync=DELAYED\n ");
7076 else if (mddev->recovery_cp < MaxSector)
7077 seq_printf(seq, "\tresync=PENDING\n ");
7080 seq_printf(seq, "\n ");
7082 bitmap_status(seq, mddev->bitmap);
7084 seq_printf(seq, "\n");
7086 mddev_unlock(mddev);
7091 static const struct seq_operations md_seq_ops = {
7092 .start = md_seq_start,
7093 .next = md_seq_next,
7094 .stop = md_seq_stop,
7095 .show = md_seq_show,
7098 static int md_seq_open(struct inode *inode, struct file *file)
7100 struct seq_file *seq;
7103 error = seq_open(file, &md_seq_ops);
7107 seq = file->private_data;
7108 seq->poll_event = atomic_read(&md_event_count);
7112 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7114 struct seq_file *seq = filp->private_data;
7117 poll_wait(filp, &md_event_waiters, wait);
7119 /* always allow read */
7120 mask = POLLIN | POLLRDNORM;
7122 if (seq->poll_event != atomic_read(&md_event_count))
7123 mask |= POLLERR | POLLPRI;
7127 static const struct file_operations md_seq_fops = {
7128 .owner = THIS_MODULE,
7129 .open = md_seq_open,
7131 .llseek = seq_lseek,
7132 .release = seq_release_private,
7133 .poll = mdstat_poll,
7136 int register_md_personality(struct md_personality *p)
7138 spin_lock(&pers_lock);
7139 list_add_tail(&p->list, &pers_list);
7140 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
7141 spin_unlock(&pers_lock);
7145 int unregister_md_personality(struct md_personality *p)
7147 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7148 spin_lock(&pers_lock);
7149 list_del_init(&p->list);
7150 spin_unlock(&pers_lock);
7154 static int is_mddev_idle(struct mddev *mddev, int init)
7156 struct md_rdev * rdev;
7162 rdev_for_each_rcu(rdev, mddev) {
7163 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7164 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7165 (int)part_stat_read(&disk->part0, sectors[1]) -
7166 atomic_read(&disk->sync_io);
7167 /* sync IO will cause sync_io to increase before the disk_stats
7168 * as sync_io is counted when a request starts, and
7169 * disk_stats is counted when it completes.
7170 * So resync activity will cause curr_events to be smaller than
7171 * when there was no such activity.
7172 * non-sync IO will cause disk_stat to increase without
7173 * increasing sync_io so curr_events will (eventually)
7174 * be larger than it was before. Once it becomes
7175 * substantially larger, the test below will cause
7176 * the array to appear non-idle, and resync will slow
7178 * If there is a lot of outstanding resync activity when
7179 * we set last_event to curr_events, then all that activity
7180 * completing might cause the array to appear non-idle
7181 * and resync will be slowed down even though there might
7182 * not have been non-resync activity. This will only
7183 * happen once though. 'last_events' will soon reflect
7184 * the state where there is little or no outstanding
7185 * resync requests, and further resync activity will
7186 * always make curr_events less than last_events.
7189 if (init || curr_events - rdev->last_events > 64) {
7190 rdev->last_events = curr_events;
7198 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7200 /* another "blocks" (512byte) blocks have been synced */
7201 atomic_sub(blocks, &mddev->recovery_active);
7202 wake_up(&mddev->recovery_wait);
7204 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7205 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7206 md_wakeup_thread(mddev->thread);
7207 // stop recovery, signal do_sync ....
7212 /* md_write_start(mddev, bi)
7213 * If we need to update some array metadata (e.g. 'active' flag
7214 * in superblock) before writing, schedule a superblock update
7215 * and wait for it to complete.
7217 void md_write_start(struct mddev *mddev, struct bio *bi)
7220 if (bio_data_dir(bi) != WRITE)
7223 BUG_ON(mddev->ro == 1);
7224 if (mddev->ro == 2) {
7225 /* need to switch to read/write */
7227 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7228 md_wakeup_thread(mddev->thread);
7229 md_wakeup_thread(mddev->sync_thread);
7232 atomic_inc(&mddev->writes_pending);
7233 if (mddev->safemode == 1)
7234 mddev->safemode = 0;
7235 if (mddev->in_sync) {
7236 spin_lock_irq(&mddev->write_lock);
7237 if (mddev->in_sync) {
7239 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7240 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7241 md_wakeup_thread(mddev->thread);
7244 spin_unlock_irq(&mddev->write_lock);
7247 sysfs_notify_dirent_safe(mddev->sysfs_state);
7248 wait_event(mddev->sb_wait,
7249 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7252 void md_write_end(struct mddev *mddev)
7254 if (atomic_dec_and_test(&mddev->writes_pending)) {
7255 if (mddev->safemode == 2)
7256 md_wakeup_thread(mddev->thread);
7257 else if (mddev->safemode_delay)
7258 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7262 /* md_allow_write(mddev)
7263 * Calling this ensures that the array is marked 'active' so that writes
7264 * may proceed without blocking. It is important to call this before
7265 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7266 * Must be called with mddev_lock held.
7268 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7269 * is dropped, so return -EAGAIN after notifying userspace.
7271 int md_allow_write(struct mddev *mddev)
7277 if (!mddev->pers->sync_request)
7280 spin_lock_irq(&mddev->write_lock);
7281 if (mddev->in_sync) {
7283 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7284 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7285 if (mddev->safemode_delay &&
7286 mddev->safemode == 0)
7287 mddev->safemode = 1;
7288 spin_unlock_irq(&mddev->write_lock);
7289 md_update_sb(mddev, 0);
7290 sysfs_notify_dirent_safe(mddev->sysfs_state);
7292 spin_unlock_irq(&mddev->write_lock);
7294 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7299 EXPORT_SYMBOL_GPL(md_allow_write);
7301 #define SYNC_MARKS 10
7302 #define SYNC_MARK_STEP (3*HZ)
7303 #define UPDATE_FREQUENCY (5*60*HZ)
7304 void md_do_sync(struct md_thread *thread)
7306 struct mddev *mddev = thread->mddev;
7307 struct mddev *mddev2;
7308 unsigned int currspeed = 0,
7310 sector_t max_sectors,j, io_sectors;
7311 unsigned long mark[SYNC_MARKS];
7312 unsigned long update_time;
7313 sector_t mark_cnt[SYNC_MARKS];
7315 struct list_head *tmp;
7316 sector_t last_check;
7318 struct md_rdev *rdev;
7320 struct blk_plug plug;
7322 /* just incase thread restarts... */
7323 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7325 if (mddev->ro) /* never try to sync a read-only array */
7328 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7329 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7330 desc = "data-check";
7331 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7332 desc = "requested-resync";
7335 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7340 /* we overload curr_resync somewhat here.
7341 * 0 == not engaged in resync at all
7342 * 2 == checking that there is no conflict with another sync
7343 * 1 == like 2, but have yielded to allow conflicting resync to
7345 * other == active in resync - this many blocks
7347 * Before starting a resync we must have set curr_resync to
7348 * 2, and then checked that every "conflicting" array has curr_resync
7349 * less than ours. When we find one that is the same or higher
7350 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7351 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7352 * This will mean we have to start checking from the beginning again.
7357 mddev->curr_resync = 2;
7360 if (kthread_should_stop())
7361 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7363 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7365 for_each_mddev(mddev2, tmp) {
7366 if (mddev2 == mddev)
7368 if (!mddev->parallel_resync
7369 && mddev2->curr_resync
7370 && match_mddev_units(mddev, mddev2)) {
7372 if (mddev < mddev2 && mddev->curr_resync == 2) {
7373 /* arbitrarily yield */
7374 mddev->curr_resync = 1;
7375 wake_up(&resync_wait);
7377 if (mddev > mddev2 && mddev->curr_resync == 1)
7378 /* no need to wait here, we can wait the next
7379 * time 'round when curr_resync == 2
7382 /* We need to wait 'interruptible' so as not to
7383 * contribute to the load average, and not to
7384 * be caught by 'softlockup'
7386 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7387 if (!kthread_should_stop() &&
7388 mddev2->curr_resync >= mddev->curr_resync) {
7389 printk(KERN_INFO "md: delaying %s of %s"
7390 " until %s has finished (they"
7391 " share one or more physical units)\n",
7392 desc, mdname(mddev), mdname(mddev2));
7394 if (signal_pending(current))
7395 flush_signals(current);
7397 finish_wait(&resync_wait, &wq);
7400 finish_wait(&resync_wait, &wq);
7403 } while (mddev->curr_resync < 2);
7406 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7407 /* resync follows the size requested by the personality,
7408 * which defaults to physical size, but can be virtual size
7410 max_sectors = mddev->resync_max_sectors;
7411 atomic64_set(&mddev->resync_mismatches, 0);
7412 /* we don't use the checkpoint if there's a bitmap */
7413 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7414 j = mddev->resync_min;
7415 else if (!mddev->bitmap)
7416 j = mddev->recovery_cp;
7418 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7419 max_sectors = mddev->resync_max_sectors;
7421 /* recovery follows the physical size of devices */
7422 max_sectors = mddev->dev_sectors;
7425 rdev_for_each_rcu(rdev, mddev)
7426 if (rdev->raid_disk >= 0 &&
7427 !test_bit(Faulty, &rdev->flags) &&
7428 !test_bit(In_sync, &rdev->flags) &&
7429 rdev->recovery_offset < j)
7430 j = rdev->recovery_offset;
7434 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7435 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7436 " %d KB/sec/disk.\n", speed_min(mddev));
7437 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7438 "(but not more than %d KB/sec) for %s.\n",
7439 speed_max(mddev), desc);
7441 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7444 for (m = 0; m < SYNC_MARKS; m++) {
7446 mark_cnt[m] = io_sectors;
7449 mddev->resync_mark = mark[last_mark];
7450 mddev->resync_mark_cnt = mark_cnt[last_mark];
7453 * Tune reconstruction:
7455 window = 32*(PAGE_SIZE/512);
7456 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7457 window/2, (unsigned long long)max_sectors/2);
7459 atomic_set(&mddev->recovery_active, 0);
7464 "md: resuming %s of %s from checkpoint.\n",
7465 desc, mdname(mddev));
7466 mddev->curr_resync = j;
7468 mddev->curr_resync = 3; /* no longer delayed */
7469 mddev->curr_resync_completed = j;
7470 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7471 md_new_event(mddev);
7472 update_time = jiffies;
7474 blk_start_plug(&plug);
7475 while (j < max_sectors) {
7480 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7481 ((mddev->curr_resync > mddev->curr_resync_completed &&
7482 (mddev->curr_resync - mddev->curr_resync_completed)
7483 > (max_sectors >> 4)) ||
7484 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7485 (j - mddev->curr_resync_completed)*2
7486 >= mddev->resync_max - mddev->curr_resync_completed
7488 /* time to update curr_resync_completed */
7489 wait_event(mddev->recovery_wait,
7490 atomic_read(&mddev->recovery_active) == 0);
7491 mddev->curr_resync_completed = j;
7492 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7493 j > mddev->recovery_cp)
7494 mddev->recovery_cp = j;
7495 update_time = jiffies;
7496 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7497 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7500 while (j >= mddev->resync_max && !kthread_should_stop()) {
7501 /* As this condition is controlled by user-space,
7502 * we can block indefinitely, so use '_interruptible'
7503 * to avoid triggering warnings.
7505 flush_signals(current); /* just in case */
7506 wait_event_interruptible(mddev->recovery_wait,
7507 mddev->resync_max > j
7508 || kthread_should_stop());
7511 if (kthread_should_stop())
7514 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7515 currspeed < speed_min(mddev));
7517 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7521 if (!skipped) { /* actual IO requested */
7522 io_sectors += sectors;
7523 atomic_add(sectors, &mddev->recovery_active);
7526 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7531 mddev->curr_resync = j;
7532 mddev->curr_mark_cnt = io_sectors;
7533 if (last_check == 0)
7534 /* this is the earliest that rebuild will be
7535 * visible in /proc/mdstat
7537 md_new_event(mddev);
7539 if (last_check + window > io_sectors || j == max_sectors)
7542 last_check = io_sectors;
7544 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7546 int next = (last_mark+1) % SYNC_MARKS;
7548 mddev->resync_mark = mark[next];
7549 mddev->resync_mark_cnt = mark_cnt[next];
7550 mark[next] = jiffies;
7551 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7556 if (kthread_should_stop())
7561 * this loop exits only if either when we are slower than
7562 * the 'hard' speed limit, or the system was IO-idle for
7564 * the system might be non-idle CPU-wise, but we only care
7565 * about not overloading the IO subsystem. (things like an
7566 * e2fsck being done on the RAID array should execute fast)
7570 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7571 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7573 if (currspeed > speed_min(mddev)) {
7574 if ((currspeed > speed_max(mddev)) ||
7575 !is_mddev_idle(mddev, 0)) {
7581 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7583 * this also signals 'finished resyncing' to md_stop
7586 blk_finish_plug(&plug);
7587 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7589 /* tell personality that we are finished */
7590 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7592 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7593 mddev->curr_resync > 2) {
7594 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7595 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7596 if (mddev->curr_resync >= mddev->recovery_cp) {
7598 "md: checkpointing %s of %s.\n",
7599 desc, mdname(mddev));
7600 if (test_bit(MD_RECOVERY_ERROR,
7602 mddev->recovery_cp =
7603 mddev->curr_resync_completed;
7605 mddev->recovery_cp =
7609 mddev->recovery_cp = MaxSector;
7611 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7612 mddev->curr_resync = MaxSector;
7614 rdev_for_each_rcu(rdev, mddev)
7615 if (rdev->raid_disk >= 0 &&
7616 mddev->delta_disks >= 0 &&
7617 !test_bit(Faulty, &rdev->flags) &&
7618 !test_bit(In_sync, &rdev->flags) &&
7619 rdev->recovery_offset < mddev->curr_resync)
7620 rdev->recovery_offset = mddev->curr_resync;
7625 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7627 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7628 /* We completed so min/max setting can be forgotten if used. */
7629 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7630 mddev->resync_min = 0;
7631 mddev->resync_max = MaxSector;
7632 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7633 mddev->resync_min = mddev->curr_resync_completed;
7634 mddev->curr_resync = 0;
7635 wake_up(&resync_wait);
7636 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7637 md_wakeup_thread(mddev->thread);
7642 * got a signal, exit.
7645 "md: md_do_sync() got signal ... exiting\n");
7646 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7650 EXPORT_SYMBOL_GPL(md_do_sync);
7652 static int remove_and_add_spares(struct mddev *mddev)
7654 struct md_rdev *rdev;
7658 rdev_for_each(rdev, mddev)
7659 if (rdev->raid_disk >= 0 &&
7660 !test_bit(Blocked, &rdev->flags) &&
7661 (test_bit(Faulty, &rdev->flags) ||
7662 ! test_bit(In_sync, &rdev->flags)) &&
7663 atomic_read(&rdev->nr_pending)==0) {
7664 if (mddev->pers->hot_remove_disk(
7665 mddev, rdev) == 0) {
7666 sysfs_unlink_rdev(mddev, rdev);
7667 rdev->raid_disk = -1;
7671 if (removed && mddev->kobj.sd)
7672 sysfs_notify(&mddev->kobj, NULL, "degraded");
7674 rdev_for_each(rdev, mddev) {
7675 if (rdev->raid_disk >= 0 &&
7676 !test_bit(In_sync, &rdev->flags) &&
7677 !test_bit(Faulty, &rdev->flags))
7679 if (rdev->raid_disk < 0
7680 && !test_bit(Faulty, &rdev->flags)) {
7681 rdev->recovery_offset = 0;
7683 hot_add_disk(mddev, rdev) == 0) {
7684 if (sysfs_link_rdev(mddev, rdev))
7685 /* failure here is OK */;
7687 md_new_event(mddev);
7688 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7693 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7697 static void reap_sync_thread(struct mddev *mddev)
7699 struct md_rdev *rdev;
7701 /* resync has finished, collect result */
7702 md_unregister_thread(&mddev->sync_thread);
7703 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7704 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7706 /* activate any spares */
7707 if (mddev->pers->spare_active(mddev)) {
7708 sysfs_notify(&mddev->kobj, NULL,
7710 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7713 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7714 mddev->pers->finish_reshape)
7715 mddev->pers->finish_reshape(mddev);
7717 /* If array is no-longer degraded, then any saved_raid_disk
7718 * information must be scrapped. Also if any device is now
7719 * In_sync we must scrape the saved_raid_disk for that device
7720 * do the superblock for an incrementally recovered device
7723 rdev_for_each(rdev, mddev)
7724 if (!mddev->degraded ||
7725 test_bit(In_sync, &rdev->flags))
7726 rdev->saved_raid_disk = -1;
7728 md_update_sb(mddev, 1);
7729 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7730 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7731 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7732 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7733 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7734 /* flag recovery needed just to double check */
7735 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7736 sysfs_notify_dirent_safe(mddev->sysfs_action);
7737 md_new_event(mddev);
7738 if (mddev->event_work.func)
7739 queue_work(md_misc_wq, &mddev->event_work);
7743 * This routine is regularly called by all per-raid-array threads to
7744 * deal with generic issues like resync and super-block update.
7745 * Raid personalities that don't have a thread (linear/raid0) do not
7746 * need this as they never do any recovery or update the superblock.
7748 * It does not do any resync itself, but rather "forks" off other threads
7749 * to do that as needed.
7750 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7751 * "->recovery" and create a thread at ->sync_thread.
7752 * When the thread finishes it sets MD_RECOVERY_DONE
7753 * and wakeups up this thread which will reap the thread and finish up.
7754 * This thread also removes any faulty devices (with nr_pending == 0).
7756 * The overall approach is:
7757 * 1/ if the superblock needs updating, update it.
7758 * 2/ If a recovery thread is running, don't do anything else.
7759 * 3/ If recovery has finished, clean up, possibly marking spares active.
7760 * 4/ If there are any faulty devices, remove them.
7761 * 5/ If array is degraded, try to add spares devices
7762 * 6/ If array has spares or is not in-sync, start a resync thread.
7764 void md_check_recovery(struct mddev *mddev)
7766 if (mddev->suspended)
7770 bitmap_daemon_work(mddev);
7772 if (signal_pending(current)) {
7773 if (mddev->pers->sync_request && !mddev->external) {
7774 printk(KERN_INFO "md: %s in immediate safe mode\n",
7776 mddev->safemode = 2;
7778 flush_signals(current);
7781 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7784 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7785 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7786 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7787 (mddev->external == 0 && mddev->safemode == 1) ||
7788 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7789 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7793 if (mddev_trylock(mddev)) {
7797 /* Only thing we do on a ro array is remove
7800 struct md_rdev *rdev;
7801 rdev_for_each(rdev, mddev)
7802 if (rdev->raid_disk >= 0 &&
7803 !test_bit(Blocked, &rdev->flags) &&
7804 test_bit(Faulty, &rdev->flags) &&
7805 atomic_read(&rdev->nr_pending)==0) {
7806 if (mddev->pers->hot_remove_disk(
7807 mddev, rdev) == 0) {
7808 sysfs_unlink_rdev(mddev, rdev);
7809 rdev->raid_disk = -1;
7812 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7816 if (!mddev->external) {
7818 spin_lock_irq(&mddev->write_lock);
7819 if (mddev->safemode &&
7820 !atomic_read(&mddev->writes_pending) &&
7822 mddev->recovery_cp == MaxSector) {
7825 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7827 if (mddev->safemode == 1)
7828 mddev->safemode = 0;
7829 spin_unlock_irq(&mddev->write_lock);
7831 sysfs_notify_dirent_safe(mddev->sysfs_state);
7835 md_update_sb(mddev, 0);
7837 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7838 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7839 /* resync/recovery still happening */
7840 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7843 if (mddev->sync_thread) {
7844 reap_sync_thread(mddev);
7847 /* Set RUNNING before clearing NEEDED to avoid
7848 * any transients in the value of "sync_action".
7850 mddev->curr_resync_completed = 0;
7851 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7852 /* Clear some bits that don't mean anything, but
7855 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7856 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7858 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7859 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
7861 /* no recovery is running.
7862 * remove any failed drives, then
7863 * add spares if possible.
7864 * Spares are also removed and re-added, to allow
7865 * the personality to fail the re-add.
7868 if (mddev->reshape_position != MaxSector) {
7869 if (mddev->pers->check_reshape == NULL ||
7870 mddev->pers->check_reshape(mddev) != 0)
7871 /* Cannot proceed */
7873 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7874 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7875 } else if ((spares = remove_and_add_spares(mddev))) {
7876 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7877 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7878 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7879 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7880 } else if (mddev->recovery_cp < MaxSector) {
7881 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7882 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7883 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7884 /* nothing to be done ... */
7887 if (mddev->pers->sync_request) {
7889 /* We are adding a device or devices to an array
7890 * which has the bitmap stored on all devices.
7891 * So make sure all bitmap pages get written
7893 bitmap_write_all(mddev->bitmap);
7895 mddev->sync_thread = md_register_thread(md_do_sync,
7898 if (!mddev->sync_thread) {
7899 printk(KERN_ERR "%s: could not start resync"
7902 /* leave the spares where they are, it shouldn't hurt */
7903 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7904 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7905 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7906 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7907 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7909 md_wakeup_thread(mddev->sync_thread);
7910 sysfs_notify_dirent_safe(mddev->sysfs_action);
7911 md_new_event(mddev);
7914 if (!mddev->sync_thread) {
7915 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7916 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7918 if (mddev->sysfs_action)
7919 sysfs_notify_dirent_safe(mddev->sysfs_action);
7921 mddev_unlock(mddev);
7925 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
7927 sysfs_notify_dirent_safe(rdev->sysfs_state);
7928 wait_event_timeout(rdev->blocked_wait,
7929 !test_bit(Blocked, &rdev->flags) &&
7930 !test_bit(BlockedBadBlocks, &rdev->flags),
7931 msecs_to_jiffies(5000));
7932 rdev_dec_pending(rdev, mddev);
7934 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7936 void md_finish_reshape(struct mddev *mddev)
7938 /* called be personality module when reshape completes. */
7939 struct md_rdev *rdev;
7941 rdev_for_each(rdev, mddev) {
7942 if (rdev->data_offset > rdev->new_data_offset)
7943 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7945 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7946 rdev->data_offset = rdev->new_data_offset;
7949 EXPORT_SYMBOL(md_finish_reshape);
7951 /* Bad block management.
7952 * We can record which blocks on each device are 'bad' and so just
7953 * fail those blocks, or that stripe, rather than the whole device.
7954 * Entries in the bad-block table are 64bits wide. This comprises:
7955 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7956 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7957 * A 'shift' can be set so that larger blocks are tracked and
7958 * consequently larger devices can be covered.
7959 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7961 * Locking of the bad-block table uses a seqlock so md_is_badblock
7962 * might need to retry if it is very unlucky.
7963 * We will sometimes want to check for bad blocks in a bi_end_io function,
7964 * so we use the write_seqlock_irq variant.
7966 * When looking for a bad block we specify a range and want to
7967 * know if any block in the range is bad. So we binary-search
7968 * to the last range that starts at-or-before the given endpoint,
7969 * (or "before the sector after the target range")
7970 * then see if it ends after the given start.
7972 * 0 if there are no known bad blocks in the range
7973 * 1 if there are known bad block which are all acknowledged
7974 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7975 * plus the start/length of the first bad section we overlap.
7977 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7978 sector_t *first_bad, int *bad_sectors)
7984 sector_t target = s + sectors;
7987 if (bb->shift > 0) {
7988 /* round the start down, and the end up */
7990 target += (1<<bb->shift) - 1;
7991 target >>= bb->shift;
7992 sectors = target - s;
7994 /* 'target' is now the first block after the bad range */
7997 seq = read_seqbegin(&bb->lock);
8002 /* Binary search between lo and hi for 'target'
8003 * i.e. for the last range that starts before 'target'
8005 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8006 * are known not to be the last range before target.
8007 * VARIANT: hi-lo is the number of possible
8008 * ranges, and decreases until it reaches 1
8010 while (hi - lo > 1) {
8011 int mid = (lo + hi) / 2;
8012 sector_t a = BB_OFFSET(p[mid]);
8014 /* This could still be the one, earlier ranges
8018 /* This and later ranges are definitely out. */
8021 /* 'lo' might be the last that started before target, but 'hi' isn't */
8023 /* need to check all range that end after 's' to see if
8024 * any are unacknowledged.
8027 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8028 if (BB_OFFSET(p[lo]) < target) {
8029 /* starts before the end, and finishes after
8030 * the start, so they must overlap
8032 if (rv != -1 && BB_ACK(p[lo]))
8036 *first_bad = BB_OFFSET(p[lo]);
8037 *bad_sectors = BB_LEN(p[lo]);
8043 if (read_seqretry(&bb->lock, seq))
8048 EXPORT_SYMBOL_GPL(md_is_badblock);
8051 * Add a range of bad blocks to the table.
8052 * This might extend the table, or might contract it
8053 * if two adjacent ranges can be merged.
8054 * We binary-search to find the 'insertion' point, then
8055 * decide how best to handle it.
8057 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8065 /* badblocks are disabled */
8069 /* round the start down, and the end up */
8070 sector_t next = s + sectors;
8072 next += (1<<bb->shift) - 1;
8077 write_seqlock_irq(&bb->lock);
8082 /* Find the last range that starts at-or-before 's' */
8083 while (hi - lo > 1) {
8084 int mid = (lo + hi) / 2;
8085 sector_t a = BB_OFFSET(p[mid]);
8091 if (hi > lo && BB_OFFSET(p[lo]) > s)
8095 /* we found a range that might merge with the start
8098 sector_t a = BB_OFFSET(p[lo]);
8099 sector_t e = a + BB_LEN(p[lo]);
8100 int ack = BB_ACK(p[lo]);
8102 /* Yes, we can merge with a previous range */
8103 if (s == a && s + sectors >= e)
8104 /* new range covers old */
8107 ack = ack && acknowledged;
8109 if (e < s + sectors)
8111 if (e - a <= BB_MAX_LEN) {
8112 p[lo] = BB_MAKE(a, e-a, ack);
8115 /* does not all fit in one range,
8116 * make p[lo] maximal
8118 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8119 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8125 if (sectors && hi < bb->count) {
8126 /* 'hi' points to the first range that starts after 's'.
8127 * Maybe we can merge with the start of that range */
8128 sector_t a = BB_OFFSET(p[hi]);
8129 sector_t e = a + BB_LEN(p[hi]);
8130 int ack = BB_ACK(p[hi]);
8131 if (a <= s + sectors) {
8132 /* merging is possible */
8133 if (e <= s + sectors) {
8138 ack = ack && acknowledged;
8141 if (e - a <= BB_MAX_LEN) {
8142 p[hi] = BB_MAKE(a, e-a, ack);
8145 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8153 if (sectors == 0 && hi < bb->count) {
8154 /* we might be able to combine lo and hi */
8155 /* Note: 's' is at the end of 'lo' */
8156 sector_t a = BB_OFFSET(p[hi]);
8157 int lolen = BB_LEN(p[lo]);
8158 int hilen = BB_LEN(p[hi]);
8159 int newlen = lolen + hilen - (s - a);
8160 if (s >= a && newlen < BB_MAX_LEN) {
8161 /* yes, we can combine them */
8162 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8163 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8164 memmove(p + hi, p + hi + 1,
8165 (bb->count - hi - 1) * 8);
8170 /* didn't merge (it all).
8171 * Need to add a range just before 'hi' */
8172 if (bb->count >= MD_MAX_BADBLOCKS) {
8173 /* No room for more */
8177 int this_sectors = sectors;
8178 memmove(p + hi + 1, p + hi,
8179 (bb->count - hi) * 8);
8182 if (this_sectors > BB_MAX_LEN)
8183 this_sectors = BB_MAX_LEN;
8184 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8185 sectors -= this_sectors;
8192 bb->unacked_exist = 1;
8193 write_sequnlock_irq(&bb->lock);
8198 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8203 s += rdev->new_data_offset;
8205 s += rdev->data_offset;
8206 rv = md_set_badblocks(&rdev->badblocks,
8209 /* Make sure they get written out promptly */
8210 sysfs_notify_dirent_safe(rdev->sysfs_state);
8211 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8212 md_wakeup_thread(rdev->mddev->thread);
8216 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8219 * Remove a range of bad blocks from the table.
8220 * This may involve extending the table if we spilt a region,
8221 * but it must not fail. So if the table becomes full, we just
8222 * drop the remove request.
8224 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8228 sector_t target = s + sectors;
8231 if (bb->shift > 0) {
8232 /* When clearing we round the start up and the end down.
8233 * This should not matter as the shift should align with
8234 * the block size and no rounding should ever be needed.
8235 * However it is better the think a block is bad when it
8236 * isn't than to think a block is not bad when it is.
8238 s += (1<<bb->shift) - 1;
8240 target >>= bb->shift;
8241 sectors = target - s;
8244 write_seqlock_irq(&bb->lock);
8249 /* Find the last range that starts before 'target' */
8250 while (hi - lo > 1) {
8251 int mid = (lo + hi) / 2;
8252 sector_t a = BB_OFFSET(p[mid]);
8259 /* p[lo] is the last range that could overlap the
8260 * current range. Earlier ranges could also overlap,
8261 * but only this one can overlap the end of the range.
8263 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8264 /* Partial overlap, leave the tail of this range */
8265 int ack = BB_ACK(p[lo]);
8266 sector_t a = BB_OFFSET(p[lo]);
8267 sector_t end = a + BB_LEN(p[lo]);
8270 /* we need to split this range */
8271 if (bb->count >= MD_MAX_BADBLOCKS) {
8275 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8277 p[lo] = BB_MAKE(a, s-a, ack);
8280 p[lo] = BB_MAKE(target, end - target, ack);
8281 /* there is no longer an overlap */
8286 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8287 /* This range does overlap */
8288 if (BB_OFFSET(p[lo]) < s) {
8289 /* Keep the early parts of this range. */
8290 int ack = BB_ACK(p[lo]);
8291 sector_t start = BB_OFFSET(p[lo]);
8292 p[lo] = BB_MAKE(start, s - start, ack);
8293 /* now low doesn't overlap, so.. */
8298 /* 'lo' is strictly before, 'hi' is strictly after,
8299 * anything between needs to be discarded
8302 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8303 bb->count -= (hi - lo - 1);
8309 write_sequnlock_irq(&bb->lock);
8313 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8317 s += rdev->new_data_offset;
8319 s += rdev->data_offset;
8320 return md_clear_badblocks(&rdev->badblocks,
8323 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8326 * Acknowledge all bad blocks in a list.
8327 * This only succeeds if ->changed is clear. It is used by
8328 * in-kernel metadata updates
8330 void md_ack_all_badblocks(struct badblocks *bb)
8332 if (bb->page == NULL || bb->changed)
8333 /* no point even trying */
8335 write_seqlock_irq(&bb->lock);
8337 if (bb->changed == 0 && bb->unacked_exist) {
8340 for (i = 0; i < bb->count ; i++) {
8341 if (!BB_ACK(p[i])) {
8342 sector_t start = BB_OFFSET(p[i]);
8343 int len = BB_LEN(p[i]);
8344 p[i] = BB_MAKE(start, len, 1);
8347 bb->unacked_exist = 0;
8349 write_sequnlock_irq(&bb->lock);
8351 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8353 /* sysfs access to bad-blocks list.
8354 * We present two files.
8355 * 'bad-blocks' lists sector numbers and lengths of ranges that
8356 * are recorded as bad. The list is truncated to fit within
8357 * the one-page limit of sysfs.
8358 * Writing "sector length" to this file adds an acknowledged
8360 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8361 * been acknowledged. Writing to this file adds bad blocks
8362 * without acknowledging them. This is largely for testing.
8366 badblocks_show(struct badblocks *bb, char *page, int unack)
8377 seq = read_seqbegin(&bb->lock);
8382 while (len < PAGE_SIZE && i < bb->count) {
8383 sector_t s = BB_OFFSET(p[i]);
8384 unsigned int length = BB_LEN(p[i]);
8385 int ack = BB_ACK(p[i]);
8391 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8392 (unsigned long long)s << bb->shift,
8393 length << bb->shift);
8395 if (unack && len == 0)
8396 bb->unacked_exist = 0;
8398 if (read_seqretry(&bb->lock, seq))
8407 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8409 unsigned long long sector;
8413 /* Allow clearing via sysfs *only* for testing/debugging.
8414 * Normally only a successful write may clear a badblock
8417 if (page[0] == '-') {
8421 #endif /* DO_DEBUG */
8423 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8425 if (newline != '\n')
8437 md_clear_badblocks(bb, sector, length);
8440 #endif /* DO_DEBUG */
8441 if (md_set_badblocks(bb, sector, length, !unack))
8447 static int md_notify_reboot(struct notifier_block *this,
8448 unsigned long code, void *x)
8450 struct list_head *tmp;
8451 struct mddev *mddev;
8454 for_each_mddev(mddev, tmp) {
8455 if (mddev_trylock(mddev)) {
8457 __md_stop_writes(mddev);
8458 mddev->safemode = 2;
8459 mddev_unlock(mddev);
8464 * certain more exotic SCSI devices are known to be
8465 * volatile wrt too early system reboots. While the
8466 * right place to handle this issue is the given
8467 * driver, we do want to have a safe RAID driver ...
8475 static struct notifier_block md_notifier = {
8476 .notifier_call = md_notify_reboot,
8478 .priority = INT_MAX, /* before any real devices */
8481 static void md_geninit(void)
8483 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8485 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8488 static int __init md_init(void)
8492 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8496 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8500 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8503 if ((ret = register_blkdev(0, "mdp")) < 0)
8507 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8508 md_probe, NULL, NULL);
8509 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8510 md_probe, NULL, NULL);
8512 register_reboot_notifier(&md_notifier);
8513 raid_table_header = register_sysctl_table(raid_root_table);
8519 unregister_blkdev(MD_MAJOR, "md");
8521 destroy_workqueue(md_misc_wq);
8523 destroy_workqueue(md_wq);
8531 * Searches all registered partitions for autorun RAID arrays
8535 static LIST_HEAD(all_detected_devices);
8536 struct detected_devices_node {
8537 struct list_head list;
8541 void md_autodetect_dev(dev_t dev)
8543 struct detected_devices_node *node_detected_dev;
8545 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8546 if (node_detected_dev) {
8547 node_detected_dev->dev = dev;
8548 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8550 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8551 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8556 static void autostart_arrays(int part)
8558 struct md_rdev *rdev;
8559 struct detected_devices_node *node_detected_dev;
8561 int i_scanned, i_passed;
8566 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8568 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8570 node_detected_dev = list_entry(all_detected_devices.next,
8571 struct detected_devices_node, list);
8572 list_del(&node_detected_dev->list);
8573 dev = node_detected_dev->dev;
8574 kfree(node_detected_dev);
8575 rdev = md_import_device(dev,0, 90);
8579 if (test_bit(Faulty, &rdev->flags)) {
8583 set_bit(AutoDetected, &rdev->flags);
8584 list_add(&rdev->same_set, &pending_raid_disks);
8588 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8589 i_scanned, i_passed);
8591 autorun_devices(part);
8594 #endif /* !MODULE */
8596 static __exit void md_exit(void)
8598 struct mddev *mddev;
8599 struct list_head *tmp;
8601 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8602 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8604 unregister_blkdev(MD_MAJOR,"md");
8605 unregister_blkdev(mdp_major, "mdp");
8606 unregister_reboot_notifier(&md_notifier);
8607 unregister_sysctl_table(raid_table_header);
8608 remove_proc_entry("mdstat", NULL);
8609 for_each_mddev(mddev, tmp) {
8610 export_array(mddev);
8611 mddev->hold_active = 0;
8613 destroy_workqueue(md_misc_wq);
8614 destroy_workqueue(md_wq);
8617 subsys_initcall(md_init);
8618 module_exit(md_exit)
8620 static int get_ro(char *buffer, struct kernel_param *kp)
8622 return sprintf(buffer, "%d", start_readonly);
8624 static int set_ro(const char *val, struct kernel_param *kp)
8627 int num = simple_strtoul(val, &e, 10);
8628 if (*val && (*e == '\0' || *e == '\n')) {
8629 start_readonly = num;
8635 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8636 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8638 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8640 EXPORT_SYMBOL(register_md_personality);
8641 EXPORT_SYMBOL(unregister_md_personality);
8642 EXPORT_SYMBOL(md_error);
8643 EXPORT_SYMBOL(md_done_sync);
8644 EXPORT_SYMBOL(md_write_start);
8645 EXPORT_SYMBOL(md_write_end);
8646 EXPORT_SYMBOL(md_register_thread);
8647 EXPORT_SYMBOL(md_unregister_thread);
8648 EXPORT_SYMBOL(md_wakeup_thread);
8649 EXPORT_SYMBOL(md_check_recovery);
8650 MODULE_LICENSE("GPL");
8651 MODULE_DESCRIPTION("MD RAID framework");
8653 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);