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>
56 #include "md-cluster.h"
59 static void autostart_arrays(int part);
62 /* pers_list is a list of registered personalities protected
64 * pers_lock does extra service to protect accesses to
65 * mddev->thread when the mutex cannot be held.
67 static LIST_HEAD(pers_list);
68 static DEFINE_SPINLOCK(pers_lock);
70 struct md_cluster_operations *md_cluster_ops;
71 EXPORT_SYMBOL(md_cluster_ops);
72 struct module *md_cluster_mod;
73 EXPORT_SYMBOL(md_cluster_mod);
75 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
76 static struct workqueue_struct *md_wq;
77 static struct workqueue_struct *md_misc_wq;
79 static int remove_and_add_spares(struct mddev *mddev,
80 struct md_rdev *this);
81 static void mddev_detach(struct mddev *mddev);
84 * Default number of read corrections we'll attempt on an rdev
85 * before ejecting it from the array. We divide the read error
86 * count by 2 for every hour elapsed between read errors.
88 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
90 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
91 * is 1000 KB/sec, so the extra system load does not show up that much.
92 * Increase it if you want to have more _guaranteed_ speed. Note that
93 * the RAID driver will use the maximum available bandwidth if the IO
94 * subsystem is idle. There is also an 'absolute maximum' reconstruction
95 * speed limit - in case reconstruction slows down your system despite
98 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
99 * or /sys/block/mdX/md/sync_speed_{min,max}
102 static int sysctl_speed_limit_min = 1000;
103 static int sysctl_speed_limit_max = 200000;
104 static inline int speed_min(struct mddev *mddev)
106 return mddev->sync_speed_min ?
107 mddev->sync_speed_min : sysctl_speed_limit_min;
110 static inline int speed_max(struct mddev *mddev)
112 return mddev->sync_speed_max ?
113 mddev->sync_speed_max : sysctl_speed_limit_max;
116 static struct ctl_table_header *raid_table_header;
118 static struct ctl_table raid_table[] = {
120 .procname = "speed_limit_min",
121 .data = &sysctl_speed_limit_min,
122 .maxlen = sizeof(int),
123 .mode = S_IRUGO|S_IWUSR,
124 .proc_handler = proc_dointvec,
127 .procname = "speed_limit_max",
128 .data = &sysctl_speed_limit_max,
129 .maxlen = sizeof(int),
130 .mode = S_IRUGO|S_IWUSR,
131 .proc_handler = proc_dointvec,
136 static struct ctl_table raid_dir_table[] = {
140 .mode = S_IRUGO|S_IXUGO,
146 static struct ctl_table raid_root_table[] = {
151 .child = raid_dir_table,
156 static const struct block_device_operations md_fops;
158 static int start_readonly;
161 * like bio_clone, but with a local bio set
164 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
169 if (!mddev || !mddev->bio_set)
170 return bio_alloc(gfp_mask, nr_iovecs);
172 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
177 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
179 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
182 if (!mddev || !mddev->bio_set)
183 return bio_clone(bio, gfp_mask);
185 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
187 EXPORT_SYMBOL_GPL(bio_clone_mddev);
190 * We have a system wide 'event count' that is incremented
191 * on any 'interesting' event, and readers of /proc/mdstat
192 * can use 'poll' or 'select' to find out when the event
196 * start array, stop array, error, add device, remove device,
197 * start build, activate spare
199 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
200 static atomic_t md_event_count;
201 void md_new_event(struct mddev *mddev)
203 atomic_inc(&md_event_count);
204 wake_up(&md_event_waiters);
206 EXPORT_SYMBOL_GPL(md_new_event);
208 /* Alternate version that can be called from interrupts
209 * when calling sysfs_notify isn't needed.
211 static void md_new_event_inintr(struct mddev *mddev)
213 atomic_inc(&md_event_count);
214 wake_up(&md_event_waiters);
218 * Enables to iterate over all existing md arrays
219 * all_mddevs_lock protects this list.
221 static LIST_HEAD(all_mddevs);
222 static DEFINE_SPINLOCK(all_mddevs_lock);
225 * iterates through all used mddevs in the system.
226 * We take care to grab the all_mddevs_lock whenever navigating
227 * the list, and to always hold a refcount when unlocked.
228 * Any code which breaks out of this loop while own
229 * a reference to the current mddev and must mddev_put it.
231 #define for_each_mddev(_mddev,_tmp) \
233 for (({ spin_lock(&all_mddevs_lock); \
234 _tmp = all_mddevs.next; \
236 ({ if (_tmp != &all_mddevs) \
237 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
238 spin_unlock(&all_mddevs_lock); \
239 if (_mddev) mddev_put(_mddev); \
240 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
241 _tmp != &all_mddevs;}); \
242 ({ spin_lock(&all_mddevs_lock); \
243 _tmp = _tmp->next;}) \
246 /* Rather than calling directly into the personality make_request function,
247 * IO requests come here first so that we can check if the device is
248 * being suspended pending a reconfiguration.
249 * We hold a refcount over the call to ->make_request. By the time that
250 * call has finished, the bio has been linked into some internal structure
251 * and so is visible to ->quiesce(), so we don't need the refcount any more.
253 static blk_qc_t md_make_request(struct request_queue *q, struct bio *bio)
255 const int rw = bio_data_dir(bio);
256 struct mddev *mddev = q->queuedata;
257 unsigned int sectors;
260 blk_queue_split(q, &bio, q->bio_split);
262 if (mddev == NULL || mddev->pers == NULL
265 return BLK_QC_T_NONE;
267 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
268 if (bio_sectors(bio) != 0)
269 bio->bi_error = -EROFS;
271 return BLK_QC_T_NONE;
273 smp_rmb(); /* Ensure implications of 'active' are visible */
275 if (mddev->suspended) {
278 prepare_to_wait(&mddev->sb_wait, &__wait,
279 TASK_UNINTERRUPTIBLE);
280 if (!mddev->suspended)
286 finish_wait(&mddev->sb_wait, &__wait);
288 atomic_inc(&mddev->active_io);
292 * save the sectors now since our bio can
293 * go away inside make_request
295 sectors = bio_sectors(bio);
296 mddev->pers->make_request(mddev, bio);
298 cpu = part_stat_lock();
299 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
300 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
303 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
304 wake_up(&mddev->sb_wait);
306 return BLK_QC_T_NONE;
309 /* mddev_suspend makes sure no new requests are submitted
310 * to the device, and that any requests that have been submitted
311 * are completely handled.
312 * Once mddev_detach() is called and completes, the module will be
315 void mddev_suspend(struct mddev *mddev)
317 if (mddev->suspended++)
320 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
321 mddev->pers->quiesce(mddev, 1);
323 del_timer_sync(&mddev->safemode_timer);
325 EXPORT_SYMBOL_GPL(mddev_suspend);
327 void mddev_resume(struct mddev *mddev)
329 if (--mddev->suspended)
331 wake_up(&mddev->sb_wait);
332 mddev->pers->quiesce(mddev, 0);
334 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
335 md_wakeup_thread(mddev->thread);
336 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
338 EXPORT_SYMBOL_GPL(mddev_resume);
340 int mddev_congested(struct mddev *mddev, int bits)
342 struct md_personality *pers = mddev->pers;
346 if (mddev->suspended)
348 else if (pers && pers->congested)
349 ret = pers->congested(mddev, bits);
353 EXPORT_SYMBOL_GPL(mddev_congested);
354 static int md_congested(void *data, int bits)
356 struct mddev *mddev = data;
357 return mddev_congested(mddev, bits);
361 * Generic flush handling for md
364 static void md_end_flush(struct bio *bio)
366 struct md_rdev *rdev = bio->bi_private;
367 struct mddev *mddev = rdev->mddev;
369 rdev_dec_pending(rdev, mddev);
371 if (atomic_dec_and_test(&mddev->flush_pending)) {
372 /* The pre-request flush has finished */
373 queue_work(md_wq, &mddev->flush_work);
378 static void md_submit_flush_data(struct work_struct *ws);
380 static void submit_flushes(struct work_struct *ws)
382 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
383 struct md_rdev *rdev;
385 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
386 atomic_set(&mddev->flush_pending, 1);
388 rdev_for_each_rcu(rdev, mddev)
389 if (rdev->raid_disk >= 0 &&
390 !test_bit(Faulty, &rdev->flags)) {
391 /* Take two references, one is dropped
392 * when request finishes, one after
393 * we reclaim rcu_read_lock
396 atomic_inc(&rdev->nr_pending);
397 atomic_inc(&rdev->nr_pending);
399 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
400 bi->bi_end_io = md_end_flush;
401 bi->bi_private = rdev;
402 bi->bi_bdev = rdev->bdev;
403 atomic_inc(&mddev->flush_pending);
404 submit_bio(WRITE_FLUSH, bi);
406 rdev_dec_pending(rdev, mddev);
409 if (atomic_dec_and_test(&mddev->flush_pending))
410 queue_work(md_wq, &mddev->flush_work);
413 static void md_submit_flush_data(struct work_struct *ws)
415 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
416 struct bio *bio = mddev->flush_bio;
418 if (bio->bi_iter.bi_size == 0)
419 /* an empty barrier - all done */
422 bio->bi_rw &= ~REQ_FLUSH;
423 mddev->pers->make_request(mddev, bio);
426 mddev->flush_bio = NULL;
427 wake_up(&mddev->sb_wait);
430 void md_flush_request(struct mddev *mddev, struct bio *bio)
432 spin_lock_irq(&mddev->lock);
433 wait_event_lock_irq(mddev->sb_wait,
436 mddev->flush_bio = bio;
437 spin_unlock_irq(&mddev->lock);
439 INIT_WORK(&mddev->flush_work, submit_flushes);
440 queue_work(md_wq, &mddev->flush_work);
442 EXPORT_SYMBOL(md_flush_request);
444 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
446 struct mddev *mddev = cb->data;
447 md_wakeup_thread(mddev->thread);
450 EXPORT_SYMBOL(md_unplug);
452 static inline struct mddev *mddev_get(struct mddev *mddev)
454 atomic_inc(&mddev->active);
458 static void mddev_delayed_delete(struct work_struct *ws);
460 static void mddev_put(struct mddev *mddev)
462 struct bio_set *bs = NULL;
464 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
466 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
467 mddev->ctime == 0 && !mddev->hold_active) {
468 /* Array is not configured at all, and not held active,
470 list_del_init(&mddev->all_mddevs);
472 mddev->bio_set = NULL;
473 if (mddev->gendisk) {
474 /* We did a probe so need to clean up. Call
475 * queue_work inside the spinlock so that
476 * flush_workqueue() after mddev_find will
477 * succeed in waiting for the work to be done.
479 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
480 queue_work(md_misc_wq, &mddev->del_work);
484 spin_unlock(&all_mddevs_lock);
489 static void md_safemode_timeout(unsigned long data);
491 void mddev_init(struct mddev *mddev)
493 mutex_init(&mddev->open_mutex);
494 mutex_init(&mddev->reconfig_mutex);
495 mutex_init(&mddev->bitmap_info.mutex);
496 INIT_LIST_HEAD(&mddev->disks);
497 INIT_LIST_HEAD(&mddev->all_mddevs);
498 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
499 (unsigned long) mddev);
500 atomic_set(&mddev->active, 1);
501 atomic_set(&mddev->openers, 0);
502 atomic_set(&mddev->active_io, 0);
503 spin_lock_init(&mddev->lock);
504 atomic_set(&mddev->flush_pending, 0);
505 init_waitqueue_head(&mddev->sb_wait);
506 init_waitqueue_head(&mddev->recovery_wait);
507 mddev->reshape_position = MaxSector;
508 mddev->reshape_backwards = 0;
509 mddev->last_sync_action = "none";
510 mddev->resync_min = 0;
511 mddev->resync_max = MaxSector;
512 mddev->level = LEVEL_NONE;
514 EXPORT_SYMBOL_GPL(mddev_init);
516 static struct mddev *mddev_find(dev_t unit)
518 struct mddev *mddev, *new = NULL;
520 if (unit && MAJOR(unit) != MD_MAJOR)
521 unit &= ~((1<<MdpMinorShift)-1);
524 spin_lock(&all_mddevs_lock);
527 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
528 if (mddev->unit == unit) {
530 spin_unlock(&all_mddevs_lock);
536 list_add(&new->all_mddevs, &all_mddevs);
537 spin_unlock(&all_mddevs_lock);
538 new->hold_active = UNTIL_IOCTL;
542 /* find an unused unit number */
543 static int next_minor = 512;
544 int start = next_minor;
548 dev = MKDEV(MD_MAJOR, next_minor);
550 if (next_minor > MINORMASK)
552 if (next_minor == start) {
553 /* Oh dear, all in use. */
554 spin_unlock(&all_mddevs_lock);
560 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
561 if (mddev->unit == dev) {
567 new->md_minor = MINOR(dev);
568 new->hold_active = UNTIL_STOP;
569 list_add(&new->all_mddevs, &all_mddevs);
570 spin_unlock(&all_mddevs_lock);
573 spin_unlock(&all_mddevs_lock);
575 new = kzalloc(sizeof(*new), GFP_KERNEL);
580 if (MAJOR(unit) == MD_MAJOR)
581 new->md_minor = MINOR(unit);
583 new->md_minor = MINOR(unit) >> MdpMinorShift;
590 static struct attribute_group md_redundancy_group;
592 void mddev_unlock(struct mddev *mddev)
594 if (mddev->to_remove) {
595 /* These cannot be removed under reconfig_mutex as
596 * an access to the files will try to take reconfig_mutex
597 * while holding the file unremovable, which leads to
599 * So hold set sysfs_active while the remove in happeing,
600 * and anything else which might set ->to_remove or my
601 * otherwise change the sysfs namespace will fail with
602 * -EBUSY if sysfs_active is still set.
603 * We set sysfs_active under reconfig_mutex and elsewhere
604 * test it under the same mutex to ensure its correct value
607 struct attribute_group *to_remove = mddev->to_remove;
608 mddev->to_remove = NULL;
609 mddev->sysfs_active = 1;
610 mutex_unlock(&mddev->reconfig_mutex);
612 if (mddev->kobj.sd) {
613 if (to_remove != &md_redundancy_group)
614 sysfs_remove_group(&mddev->kobj, to_remove);
615 if (mddev->pers == NULL ||
616 mddev->pers->sync_request == NULL) {
617 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
618 if (mddev->sysfs_action)
619 sysfs_put(mddev->sysfs_action);
620 mddev->sysfs_action = NULL;
623 mddev->sysfs_active = 0;
625 mutex_unlock(&mddev->reconfig_mutex);
627 /* As we've dropped the mutex we need a spinlock to
628 * make sure the thread doesn't disappear
630 spin_lock(&pers_lock);
631 md_wakeup_thread(mddev->thread);
632 spin_unlock(&pers_lock);
634 EXPORT_SYMBOL_GPL(mddev_unlock);
636 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
638 struct md_rdev *rdev;
640 rdev_for_each_rcu(rdev, mddev)
641 if (rdev->desc_nr == nr)
646 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
648 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
650 struct md_rdev *rdev;
652 rdev_for_each(rdev, mddev)
653 if (rdev->bdev->bd_dev == dev)
659 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
661 struct md_rdev *rdev;
663 rdev_for_each_rcu(rdev, mddev)
664 if (rdev->bdev->bd_dev == dev)
670 static struct md_personality *find_pers(int level, char *clevel)
672 struct md_personality *pers;
673 list_for_each_entry(pers, &pers_list, list) {
674 if (level != LEVEL_NONE && pers->level == level)
676 if (strcmp(pers->name, clevel)==0)
682 /* return the offset of the super block in 512byte sectors */
683 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
685 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
686 return MD_NEW_SIZE_SECTORS(num_sectors);
689 static int alloc_disk_sb(struct md_rdev *rdev)
691 rdev->sb_page = alloc_page(GFP_KERNEL);
692 if (!rdev->sb_page) {
693 printk(KERN_ALERT "md: out of memory.\n");
700 void md_rdev_clear(struct md_rdev *rdev)
703 put_page(rdev->sb_page);
705 rdev->sb_page = NULL;
710 put_page(rdev->bb_page);
711 rdev->bb_page = NULL;
713 kfree(rdev->badblocks.page);
714 rdev->badblocks.page = NULL;
716 EXPORT_SYMBOL_GPL(md_rdev_clear);
718 static void super_written(struct bio *bio)
720 struct md_rdev *rdev = bio->bi_private;
721 struct mddev *mddev = rdev->mddev;
724 printk("md: super_written gets error=%d\n", bio->bi_error);
725 md_error(mddev, rdev);
728 if (atomic_dec_and_test(&mddev->pending_writes))
729 wake_up(&mddev->sb_wait);
733 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
734 sector_t sector, int size, struct page *page)
736 /* write first size bytes of page to sector of rdev
737 * Increment mddev->pending_writes before returning
738 * and decrement it on completion, waking up sb_wait
739 * if zero is reached.
740 * If an error occurred, call md_error
742 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
744 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
745 bio->bi_iter.bi_sector = sector;
746 bio_add_page(bio, page, size, 0);
747 bio->bi_private = rdev;
748 bio->bi_end_io = super_written;
750 atomic_inc(&mddev->pending_writes);
751 submit_bio(WRITE_FLUSH_FUA, bio);
754 void md_super_wait(struct mddev *mddev)
756 /* wait for all superblock writes that were scheduled to complete */
757 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
760 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
761 struct page *page, int rw, bool metadata_op)
763 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
766 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
767 rdev->meta_bdev : rdev->bdev;
769 bio->bi_iter.bi_sector = sector + rdev->sb_start;
770 else if (rdev->mddev->reshape_position != MaxSector &&
771 (rdev->mddev->reshape_backwards ==
772 (sector >= rdev->mddev->reshape_position)))
773 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
775 bio->bi_iter.bi_sector = sector + rdev->data_offset;
776 bio_add_page(bio, page, size, 0);
777 submit_bio_wait(rw, bio);
779 ret = !bio->bi_error;
783 EXPORT_SYMBOL_GPL(sync_page_io);
785 static int read_disk_sb(struct md_rdev *rdev, int size)
787 char b[BDEVNAME_SIZE];
792 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
798 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
799 bdevname(rdev->bdev,b));
803 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
805 return sb1->set_uuid0 == sb2->set_uuid0 &&
806 sb1->set_uuid1 == sb2->set_uuid1 &&
807 sb1->set_uuid2 == sb2->set_uuid2 &&
808 sb1->set_uuid3 == sb2->set_uuid3;
811 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
814 mdp_super_t *tmp1, *tmp2;
816 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
817 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
819 if (!tmp1 || !tmp2) {
821 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
829 * nr_disks is not constant
834 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
841 static u32 md_csum_fold(u32 csum)
843 csum = (csum & 0xffff) + (csum >> 16);
844 return (csum & 0xffff) + (csum >> 16);
847 static unsigned int calc_sb_csum(mdp_super_t *sb)
850 u32 *sb32 = (u32*)sb;
852 unsigned int disk_csum, csum;
854 disk_csum = sb->sb_csum;
857 for (i = 0; i < MD_SB_BYTES/4 ; i++)
859 csum = (newcsum & 0xffffffff) + (newcsum>>32);
862 /* This used to use csum_partial, which was wrong for several
863 * reasons including that different results are returned on
864 * different architectures. It isn't critical that we get exactly
865 * the same return value as before (we always csum_fold before
866 * testing, and that removes any differences). However as we
867 * know that csum_partial always returned a 16bit value on
868 * alphas, do a fold to maximise conformity to previous behaviour.
870 sb->sb_csum = md_csum_fold(disk_csum);
872 sb->sb_csum = disk_csum;
878 * Handle superblock details.
879 * We want to be able to handle multiple superblock formats
880 * so we have a common interface to them all, and an array of
881 * different handlers.
882 * We rely on user-space to write the initial superblock, and support
883 * reading and updating of superblocks.
884 * Interface methods are:
885 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
886 * loads and validates a superblock on dev.
887 * if refdev != NULL, compare superblocks on both devices
889 * 0 - dev has a superblock that is compatible with refdev
890 * 1 - dev has a superblock that is compatible and newer than refdev
891 * so dev should be used as the refdev in future
892 * -EINVAL superblock incompatible or invalid
893 * -othererror e.g. -EIO
895 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
896 * Verify that dev is acceptable into mddev.
897 * The first time, mddev->raid_disks will be 0, and data from
898 * dev should be merged in. Subsequent calls check that dev
899 * is new enough. Return 0 or -EINVAL
901 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
902 * Update the superblock for rdev with data in mddev
903 * This does not write to disc.
909 struct module *owner;
910 int (*load_super)(struct md_rdev *rdev,
911 struct md_rdev *refdev,
913 int (*validate_super)(struct mddev *mddev,
914 struct md_rdev *rdev);
915 void (*sync_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
918 sector_t num_sectors);
919 int (*allow_new_offset)(struct md_rdev *rdev,
920 unsigned long long new_offset);
924 * Check that the given mddev has no bitmap.
926 * This function is called from the run method of all personalities that do not
927 * support bitmaps. It prints an error message and returns non-zero if mddev
928 * has a bitmap. Otherwise, it returns 0.
931 int md_check_no_bitmap(struct mddev *mddev)
933 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
935 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
936 mdname(mddev), mddev->pers->name);
939 EXPORT_SYMBOL(md_check_no_bitmap);
942 * load_super for 0.90.0
944 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
946 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
951 * Calculate the position of the superblock (512byte sectors),
952 * it's at the end of the disk.
954 * It also happens to be a multiple of 4Kb.
956 rdev->sb_start = calc_dev_sboffset(rdev);
958 ret = read_disk_sb(rdev, MD_SB_BYTES);
963 bdevname(rdev->bdev, b);
964 sb = page_address(rdev->sb_page);
966 if (sb->md_magic != MD_SB_MAGIC) {
967 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
972 if (sb->major_version != 0 ||
973 sb->minor_version < 90 ||
974 sb->minor_version > 91) {
975 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
976 sb->major_version, sb->minor_version,
981 if (sb->raid_disks <= 0)
984 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
985 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
990 rdev->preferred_minor = sb->md_minor;
991 rdev->data_offset = 0;
992 rdev->new_data_offset = 0;
993 rdev->sb_size = MD_SB_BYTES;
994 rdev->badblocks.shift = -1;
996 if (sb->level == LEVEL_MULTIPATH)
999 rdev->desc_nr = sb->this_disk.number;
1005 mdp_super_t *refsb = page_address(refdev->sb_page);
1006 if (!uuid_equal(refsb, sb)) {
1007 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1008 b, bdevname(refdev->bdev,b2));
1011 if (!sb_equal(refsb, sb)) {
1012 printk(KERN_WARNING "md: %s has same UUID"
1013 " but different superblock to %s\n",
1014 b, bdevname(refdev->bdev, b2));
1018 ev2 = md_event(refsb);
1024 rdev->sectors = rdev->sb_start;
1025 /* Limit to 4TB as metadata cannot record more than that.
1026 * (not needed for Linear and RAID0 as metadata doesn't
1029 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1030 rdev->sectors = (2ULL << 32) - 2;
1032 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1033 /* "this cannot possibly happen" ... */
1041 * validate_super for 0.90.0
1043 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1046 mdp_super_t *sb = page_address(rdev->sb_page);
1047 __u64 ev1 = md_event(sb);
1049 rdev->raid_disk = -1;
1050 clear_bit(Faulty, &rdev->flags);
1051 clear_bit(In_sync, &rdev->flags);
1052 clear_bit(Bitmap_sync, &rdev->flags);
1053 clear_bit(WriteMostly, &rdev->flags);
1055 if (mddev->raid_disks == 0) {
1056 mddev->major_version = 0;
1057 mddev->minor_version = sb->minor_version;
1058 mddev->patch_version = sb->patch_version;
1059 mddev->external = 0;
1060 mddev->chunk_sectors = sb->chunk_size >> 9;
1061 mddev->ctime = sb->ctime;
1062 mddev->utime = sb->utime;
1063 mddev->level = sb->level;
1064 mddev->clevel[0] = 0;
1065 mddev->layout = sb->layout;
1066 mddev->raid_disks = sb->raid_disks;
1067 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1068 mddev->events = ev1;
1069 mddev->bitmap_info.offset = 0;
1070 mddev->bitmap_info.space = 0;
1071 /* bitmap can use 60 K after the 4K superblocks */
1072 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1073 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1074 mddev->reshape_backwards = 0;
1076 if (mddev->minor_version >= 91) {
1077 mddev->reshape_position = sb->reshape_position;
1078 mddev->delta_disks = sb->delta_disks;
1079 mddev->new_level = sb->new_level;
1080 mddev->new_layout = sb->new_layout;
1081 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1082 if (mddev->delta_disks < 0)
1083 mddev->reshape_backwards = 1;
1085 mddev->reshape_position = MaxSector;
1086 mddev->delta_disks = 0;
1087 mddev->new_level = mddev->level;
1088 mddev->new_layout = mddev->layout;
1089 mddev->new_chunk_sectors = mddev->chunk_sectors;
1092 if (sb->state & (1<<MD_SB_CLEAN))
1093 mddev->recovery_cp = MaxSector;
1095 if (sb->events_hi == sb->cp_events_hi &&
1096 sb->events_lo == sb->cp_events_lo) {
1097 mddev->recovery_cp = sb->recovery_cp;
1099 mddev->recovery_cp = 0;
1102 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1103 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1104 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1105 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1107 mddev->max_disks = MD_SB_DISKS;
1109 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1110 mddev->bitmap_info.file == NULL) {
1111 mddev->bitmap_info.offset =
1112 mddev->bitmap_info.default_offset;
1113 mddev->bitmap_info.space =
1114 mddev->bitmap_info.default_space;
1117 } else if (mddev->pers == NULL) {
1118 /* Insist on good event counter while assembling, except
1119 * for spares (which don't need an event count) */
1121 if (sb->disks[rdev->desc_nr].state & (
1122 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1123 if (ev1 < mddev->events)
1125 } else if (mddev->bitmap) {
1126 /* if adding to array with a bitmap, then we can accept an
1127 * older device ... but not too old.
1129 if (ev1 < mddev->bitmap->events_cleared)
1131 if (ev1 < mddev->events)
1132 set_bit(Bitmap_sync, &rdev->flags);
1134 if (ev1 < mddev->events)
1135 /* just a hot-add of a new device, leave raid_disk at -1 */
1139 if (mddev->level != LEVEL_MULTIPATH) {
1140 desc = sb->disks + rdev->desc_nr;
1142 if (desc->state & (1<<MD_DISK_FAULTY))
1143 set_bit(Faulty, &rdev->flags);
1144 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1145 desc->raid_disk < mddev->raid_disks */) {
1146 set_bit(In_sync, &rdev->flags);
1147 rdev->raid_disk = desc->raid_disk;
1148 rdev->saved_raid_disk = desc->raid_disk;
1149 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1150 /* active but not in sync implies recovery up to
1151 * reshape position. We don't know exactly where
1152 * that is, so set to zero for now */
1153 if (mddev->minor_version >= 91) {
1154 rdev->recovery_offset = 0;
1155 rdev->raid_disk = desc->raid_disk;
1158 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1159 set_bit(WriteMostly, &rdev->flags);
1160 } else /* MULTIPATH are always insync */
1161 set_bit(In_sync, &rdev->flags);
1166 * sync_super for 0.90.0
1168 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1171 struct md_rdev *rdev2;
1172 int next_spare = mddev->raid_disks;
1174 /* make rdev->sb match mddev data..
1177 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1178 * 3/ any empty disks < next_spare become removed
1180 * disks[0] gets initialised to REMOVED because
1181 * we cannot be sure from other fields if it has
1182 * been initialised or not.
1185 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1187 rdev->sb_size = MD_SB_BYTES;
1189 sb = page_address(rdev->sb_page);
1191 memset(sb, 0, sizeof(*sb));
1193 sb->md_magic = MD_SB_MAGIC;
1194 sb->major_version = mddev->major_version;
1195 sb->patch_version = mddev->patch_version;
1196 sb->gvalid_words = 0; /* ignored */
1197 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1198 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1199 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1200 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1202 sb->ctime = mddev->ctime;
1203 sb->level = mddev->level;
1204 sb->size = mddev->dev_sectors / 2;
1205 sb->raid_disks = mddev->raid_disks;
1206 sb->md_minor = mddev->md_minor;
1207 sb->not_persistent = 0;
1208 sb->utime = mddev->utime;
1210 sb->events_hi = (mddev->events>>32);
1211 sb->events_lo = (u32)mddev->events;
1213 if (mddev->reshape_position == MaxSector)
1214 sb->minor_version = 90;
1216 sb->minor_version = 91;
1217 sb->reshape_position = mddev->reshape_position;
1218 sb->new_level = mddev->new_level;
1219 sb->delta_disks = mddev->delta_disks;
1220 sb->new_layout = mddev->new_layout;
1221 sb->new_chunk = mddev->new_chunk_sectors << 9;
1223 mddev->minor_version = sb->minor_version;
1226 sb->recovery_cp = mddev->recovery_cp;
1227 sb->cp_events_hi = (mddev->events>>32);
1228 sb->cp_events_lo = (u32)mddev->events;
1229 if (mddev->recovery_cp == MaxSector)
1230 sb->state = (1<< MD_SB_CLEAN);
1232 sb->recovery_cp = 0;
1234 sb->layout = mddev->layout;
1235 sb->chunk_size = mddev->chunk_sectors << 9;
1237 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1238 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1240 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1241 rdev_for_each(rdev2, mddev) {
1244 int is_active = test_bit(In_sync, &rdev2->flags);
1246 if (rdev2->raid_disk >= 0 &&
1247 sb->minor_version >= 91)
1248 /* we have nowhere to store the recovery_offset,
1249 * but if it is not below the reshape_position,
1250 * we can piggy-back on that.
1253 if (rdev2->raid_disk < 0 ||
1254 test_bit(Faulty, &rdev2->flags))
1257 desc_nr = rdev2->raid_disk;
1259 desc_nr = next_spare++;
1260 rdev2->desc_nr = desc_nr;
1261 d = &sb->disks[rdev2->desc_nr];
1263 d->number = rdev2->desc_nr;
1264 d->major = MAJOR(rdev2->bdev->bd_dev);
1265 d->minor = MINOR(rdev2->bdev->bd_dev);
1267 d->raid_disk = rdev2->raid_disk;
1269 d->raid_disk = rdev2->desc_nr; /* compatibility */
1270 if (test_bit(Faulty, &rdev2->flags))
1271 d->state = (1<<MD_DISK_FAULTY);
1272 else if (is_active) {
1273 d->state = (1<<MD_DISK_ACTIVE);
1274 if (test_bit(In_sync, &rdev2->flags))
1275 d->state |= (1<<MD_DISK_SYNC);
1283 if (test_bit(WriteMostly, &rdev2->flags))
1284 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1286 /* now set the "removed" and "faulty" bits on any missing devices */
1287 for (i=0 ; i < mddev->raid_disks ; i++) {
1288 mdp_disk_t *d = &sb->disks[i];
1289 if (d->state == 0 && d->number == 0) {
1292 d->state = (1<<MD_DISK_REMOVED);
1293 d->state |= (1<<MD_DISK_FAULTY);
1297 sb->nr_disks = nr_disks;
1298 sb->active_disks = active;
1299 sb->working_disks = working;
1300 sb->failed_disks = failed;
1301 sb->spare_disks = spare;
1303 sb->this_disk = sb->disks[rdev->desc_nr];
1304 sb->sb_csum = calc_sb_csum(sb);
1308 * rdev_size_change for 0.90.0
1310 static unsigned long long
1311 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1313 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1314 return 0; /* component must fit device */
1315 if (rdev->mddev->bitmap_info.offset)
1316 return 0; /* can't move bitmap */
1317 rdev->sb_start = calc_dev_sboffset(rdev);
1318 if (!num_sectors || num_sectors > rdev->sb_start)
1319 num_sectors = rdev->sb_start;
1320 /* Limit to 4TB as metadata cannot record more than that.
1321 * 4TB == 2^32 KB, or 2*2^32 sectors.
1323 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1324 num_sectors = (2ULL << 32) - 2;
1325 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1327 md_super_wait(rdev->mddev);
1332 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1334 /* non-zero offset changes not possible with v0.90 */
1335 return new_offset == 0;
1339 * version 1 superblock
1342 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1346 unsigned long long newcsum;
1347 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1348 __le32 *isuper = (__le32*)sb;
1350 disk_csum = sb->sb_csum;
1353 for (; size >= 4; size -= 4)
1354 newcsum += le32_to_cpu(*isuper++);
1357 newcsum += le16_to_cpu(*(__le16*) isuper);
1359 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1360 sb->sb_csum = disk_csum;
1361 return cpu_to_le32(csum);
1364 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1366 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1368 struct mdp_superblock_1 *sb;
1372 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1376 * Calculate the position of the superblock in 512byte sectors.
1377 * It is always aligned to a 4K boundary and
1378 * depeding on minor_version, it can be:
1379 * 0: At least 8K, but less than 12K, from end of device
1380 * 1: At start of device
1381 * 2: 4K from start of device.
1383 switch(minor_version) {
1385 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1387 sb_start &= ~(sector_t)(4*2-1);
1398 rdev->sb_start = sb_start;
1400 /* superblock is rarely larger than 1K, but it can be larger,
1401 * and it is safe to read 4k, so we do that
1403 ret = read_disk_sb(rdev, 4096);
1404 if (ret) return ret;
1406 sb = page_address(rdev->sb_page);
1408 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1409 sb->major_version != cpu_to_le32(1) ||
1410 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1411 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1412 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1415 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1416 printk("md: invalid superblock checksum on %s\n",
1417 bdevname(rdev->bdev,b));
1420 if (le64_to_cpu(sb->data_size) < 10) {
1421 printk("md: data_size too small on %s\n",
1422 bdevname(rdev->bdev,b));
1427 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1428 /* Some padding is non-zero, might be a new feature */
1431 rdev->preferred_minor = 0xffff;
1432 rdev->data_offset = le64_to_cpu(sb->data_offset);
1433 rdev->new_data_offset = rdev->data_offset;
1434 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1435 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1436 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1437 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1439 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1440 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1441 if (rdev->sb_size & bmask)
1442 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1445 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1448 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1451 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1454 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1456 if (!rdev->bb_page) {
1457 rdev->bb_page = alloc_page(GFP_KERNEL);
1461 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1462 rdev->badblocks.count == 0) {
1463 /* need to load the bad block list.
1464 * Currently we limit it to one page.
1470 int sectors = le16_to_cpu(sb->bblog_size);
1471 if (sectors > (PAGE_SIZE / 512))
1473 offset = le32_to_cpu(sb->bblog_offset);
1476 bb_sector = (long long)offset;
1477 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1478 rdev->bb_page, READ, true))
1480 bbp = (u64 *)page_address(rdev->bb_page);
1481 rdev->badblocks.shift = sb->bblog_shift;
1482 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1483 u64 bb = le64_to_cpu(*bbp);
1484 int count = bb & (0x3ff);
1485 u64 sector = bb >> 10;
1486 sector <<= sb->bblog_shift;
1487 count <<= sb->bblog_shift;
1490 if (md_set_badblocks(&rdev->badblocks,
1491 sector, count, 1) == 0)
1494 } else if (sb->bblog_offset != 0)
1495 rdev->badblocks.shift = 0;
1501 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1503 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1504 sb->level != refsb->level ||
1505 sb->layout != refsb->layout ||
1506 sb->chunksize != refsb->chunksize) {
1507 printk(KERN_WARNING "md: %s has strangely different"
1508 " superblock to %s\n",
1509 bdevname(rdev->bdev,b),
1510 bdevname(refdev->bdev,b2));
1513 ev1 = le64_to_cpu(sb->events);
1514 ev2 = le64_to_cpu(refsb->events);
1521 if (minor_version) {
1522 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1523 sectors -= rdev->data_offset;
1525 sectors = rdev->sb_start;
1526 if (sectors < le64_to_cpu(sb->data_size))
1528 rdev->sectors = le64_to_cpu(sb->data_size);
1532 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1534 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1535 __u64 ev1 = le64_to_cpu(sb->events);
1537 rdev->raid_disk = -1;
1538 clear_bit(Faulty, &rdev->flags);
1539 clear_bit(In_sync, &rdev->flags);
1540 clear_bit(Bitmap_sync, &rdev->flags);
1541 clear_bit(WriteMostly, &rdev->flags);
1543 if (mddev->raid_disks == 0) {
1544 mddev->major_version = 1;
1545 mddev->patch_version = 0;
1546 mddev->external = 0;
1547 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1548 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1549 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1550 mddev->level = le32_to_cpu(sb->level);
1551 mddev->clevel[0] = 0;
1552 mddev->layout = le32_to_cpu(sb->layout);
1553 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1554 mddev->dev_sectors = le64_to_cpu(sb->size);
1555 mddev->events = ev1;
1556 mddev->bitmap_info.offset = 0;
1557 mddev->bitmap_info.space = 0;
1558 /* Default location for bitmap is 1K after superblock
1559 * using 3K - total of 4K
1561 mddev->bitmap_info.default_offset = 1024 >> 9;
1562 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1563 mddev->reshape_backwards = 0;
1565 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1566 memcpy(mddev->uuid, sb->set_uuid, 16);
1568 mddev->max_disks = (4096-256)/2;
1570 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1571 mddev->bitmap_info.file == NULL) {
1572 mddev->bitmap_info.offset =
1573 (__s32)le32_to_cpu(sb->bitmap_offset);
1574 /* Metadata doesn't record how much space is available.
1575 * For 1.0, we assume we can use up to the superblock
1576 * if before, else to 4K beyond superblock.
1577 * For others, assume no change is possible.
1579 if (mddev->minor_version > 0)
1580 mddev->bitmap_info.space = 0;
1581 else if (mddev->bitmap_info.offset > 0)
1582 mddev->bitmap_info.space =
1583 8 - mddev->bitmap_info.offset;
1585 mddev->bitmap_info.space =
1586 -mddev->bitmap_info.offset;
1589 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1590 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1591 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1592 mddev->new_level = le32_to_cpu(sb->new_level);
1593 mddev->new_layout = le32_to_cpu(sb->new_layout);
1594 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1595 if (mddev->delta_disks < 0 ||
1596 (mddev->delta_disks == 0 &&
1597 (le32_to_cpu(sb->feature_map)
1598 & MD_FEATURE_RESHAPE_BACKWARDS)))
1599 mddev->reshape_backwards = 1;
1601 mddev->reshape_position = MaxSector;
1602 mddev->delta_disks = 0;
1603 mddev->new_level = mddev->level;
1604 mddev->new_layout = mddev->layout;
1605 mddev->new_chunk_sectors = mddev->chunk_sectors;
1608 } else if (mddev->pers == NULL) {
1609 /* Insist of good event counter while assembling, except for
1610 * spares (which don't need an event count) */
1612 if (rdev->desc_nr >= 0 &&
1613 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1614 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1615 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1616 if (ev1 < mddev->events)
1618 } else if (mddev->bitmap) {
1619 /* If adding to array with a bitmap, then we can accept an
1620 * older device, but not too old.
1622 if (ev1 < mddev->bitmap->events_cleared)
1624 if (ev1 < mddev->events)
1625 set_bit(Bitmap_sync, &rdev->flags);
1627 if (ev1 < mddev->events)
1628 /* just a hot-add of a new device, leave raid_disk at -1 */
1631 if (mddev->level != LEVEL_MULTIPATH) {
1633 if (rdev->desc_nr < 0 ||
1634 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1635 role = MD_DISK_ROLE_SPARE;
1638 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1640 case MD_DISK_ROLE_SPARE: /* spare */
1642 case MD_DISK_ROLE_FAULTY: /* faulty */
1643 set_bit(Faulty, &rdev->flags);
1645 case MD_DISK_ROLE_JOURNAL: /* journal device */
1646 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1647 /* journal device without journal feature */
1649 "md: journal device provided without journal feature, ignoring the device\n");
1652 set_bit(Journal, &rdev->flags);
1653 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1654 if (mddev->recovery_cp == MaxSector)
1655 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1656 rdev->raid_disk = 0;
1659 rdev->saved_raid_disk = role;
1660 if ((le32_to_cpu(sb->feature_map) &
1661 MD_FEATURE_RECOVERY_OFFSET)) {
1662 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1663 if (!(le32_to_cpu(sb->feature_map) &
1664 MD_FEATURE_RECOVERY_BITMAP))
1665 rdev->saved_raid_disk = -1;
1667 set_bit(In_sync, &rdev->flags);
1668 rdev->raid_disk = role;
1671 if (sb->devflags & WriteMostly1)
1672 set_bit(WriteMostly, &rdev->flags);
1673 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1674 set_bit(Replacement, &rdev->flags);
1675 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1676 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1677 } else /* MULTIPATH are always insync */
1678 set_bit(In_sync, &rdev->flags);
1683 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1685 struct mdp_superblock_1 *sb;
1686 struct md_rdev *rdev2;
1688 /* make rdev->sb match mddev and rdev data. */
1690 sb = page_address(rdev->sb_page);
1692 sb->feature_map = 0;
1694 sb->recovery_offset = cpu_to_le64(0);
1695 memset(sb->pad3, 0, sizeof(sb->pad3));
1697 sb->utime = cpu_to_le64((__u64)mddev->utime);
1698 sb->events = cpu_to_le64(mddev->events);
1700 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1701 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1702 sb->resync_offset = cpu_to_le64(MaxSector);
1704 sb->resync_offset = cpu_to_le64(0);
1706 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1708 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1709 sb->size = cpu_to_le64(mddev->dev_sectors);
1710 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1711 sb->level = cpu_to_le32(mddev->level);
1712 sb->layout = cpu_to_le32(mddev->layout);
1714 if (test_bit(WriteMostly, &rdev->flags))
1715 sb->devflags |= WriteMostly1;
1717 sb->devflags &= ~WriteMostly1;
1718 sb->data_offset = cpu_to_le64(rdev->data_offset);
1719 sb->data_size = cpu_to_le64(rdev->sectors);
1721 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1722 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1723 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1726 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1727 !test_bit(In_sync, &rdev->flags)) {
1729 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1730 sb->recovery_offset =
1731 cpu_to_le64(rdev->recovery_offset);
1732 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1734 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1736 /* Note: recovery_offset and journal_tail share space */
1737 if (test_bit(Journal, &rdev->flags))
1738 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1739 if (test_bit(Replacement, &rdev->flags))
1741 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1743 if (mddev->reshape_position != MaxSector) {
1744 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1745 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1746 sb->new_layout = cpu_to_le32(mddev->new_layout);
1747 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1748 sb->new_level = cpu_to_le32(mddev->new_level);
1749 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1750 if (mddev->delta_disks == 0 &&
1751 mddev->reshape_backwards)
1753 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1754 if (rdev->new_data_offset != rdev->data_offset) {
1756 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1757 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1758 - rdev->data_offset));
1762 if (mddev_is_clustered(mddev))
1763 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1765 if (rdev->badblocks.count == 0)
1766 /* Nothing to do for bad blocks*/ ;
1767 else if (sb->bblog_offset == 0)
1768 /* Cannot record bad blocks on this device */
1769 md_error(mddev, rdev);
1771 struct badblocks *bb = &rdev->badblocks;
1772 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1774 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1779 seq = read_seqbegin(&bb->lock);
1781 memset(bbp, 0xff, PAGE_SIZE);
1783 for (i = 0 ; i < bb->count ; i++) {
1784 u64 internal_bb = p[i];
1785 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1786 | BB_LEN(internal_bb));
1787 bbp[i] = cpu_to_le64(store_bb);
1790 if (read_seqretry(&bb->lock, seq))
1793 bb->sector = (rdev->sb_start +
1794 (int)le32_to_cpu(sb->bblog_offset));
1795 bb->size = le16_to_cpu(sb->bblog_size);
1800 rdev_for_each(rdev2, mddev)
1801 if (rdev2->desc_nr+1 > max_dev)
1802 max_dev = rdev2->desc_nr+1;
1804 if (max_dev > le32_to_cpu(sb->max_dev)) {
1806 sb->max_dev = cpu_to_le32(max_dev);
1807 rdev->sb_size = max_dev * 2 + 256;
1808 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1809 if (rdev->sb_size & bmask)
1810 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1812 max_dev = le32_to_cpu(sb->max_dev);
1814 for (i=0; i<max_dev;i++)
1815 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1817 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1818 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1820 rdev_for_each(rdev2, mddev) {
1822 if (test_bit(Faulty, &rdev2->flags))
1823 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1824 else if (test_bit(In_sync, &rdev2->flags))
1825 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1826 else if (test_bit(Journal, &rdev2->flags))
1827 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1828 else if (rdev2->raid_disk >= 0)
1829 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1831 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1834 sb->sb_csum = calc_sb_1_csum(sb);
1837 static unsigned long long
1838 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1840 struct mdp_superblock_1 *sb;
1841 sector_t max_sectors;
1842 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1843 return 0; /* component must fit device */
1844 if (rdev->data_offset != rdev->new_data_offset)
1845 return 0; /* too confusing */
1846 if (rdev->sb_start < rdev->data_offset) {
1847 /* minor versions 1 and 2; superblock before data */
1848 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1849 max_sectors -= rdev->data_offset;
1850 if (!num_sectors || num_sectors > max_sectors)
1851 num_sectors = max_sectors;
1852 } else if (rdev->mddev->bitmap_info.offset) {
1853 /* minor version 0 with bitmap we can't move */
1856 /* minor version 0; superblock after data */
1858 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1859 sb_start &= ~(sector_t)(4*2 - 1);
1860 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1861 if (!num_sectors || num_sectors > max_sectors)
1862 num_sectors = max_sectors;
1863 rdev->sb_start = sb_start;
1865 sb = page_address(rdev->sb_page);
1866 sb->data_size = cpu_to_le64(num_sectors);
1867 sb->super_offset = rdev->sb_start;
1868 sb->sb_csum = calc_sb_1_csum(sb);
1869 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1871 md_super_wait(rdev->mddev);
1877 super_1_allow_new_offset(struct md_rdev *rdev,
1878 unsigned long long new_offset)
1880 /* All necessary checks on new >= old have been done */
1881 struct bitmap *bitmap;
1882 if (new_offset >= rdev->data_offset)
1885 /* with 1.0 metadata, there is no metadata to tread on
1886 * so we can always move back */
1887 if (rdev->mddev->minor_version == 0)
1890 /* otherwise we must be sure not to step on
1891 * any metadata, so stay:
1892 * 36K beyond start of superblock
1893 * beyond end of badblocks
1894 * beyond write-intent bitmap
1896 if (rdev->sb_start + (32+4)*2 > new_offset)
1898 bitmap = rdev->mddev->bitmap;
1899 if (bitmap && !rdev->mddev->bitmap_info.file &&
1900 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1901 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1903 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1909 static struct super_type super_types[] = {
1912 .owner = THIS_MODULE,
1913 .load_super = super_90_load,
1914 .validate_super = super_90_validate,
1915 .sync_super = super_90_sync,
1916 .rdev_size_change = super_90_rdev_size_change,
1917 .allow_new_offset = super_90_allow_new_offset,
1921 .owner = THIS_MODULE,
1922 .load_super = super_1_load,
1923 .validate_super = super_1_validate,
1924 .sync_super = super_1_sync,
1925 .rdev_size_change = super_1_rdev_size_change,
1926 .allow_new_offset = super_1_allow_new_offset,
1930 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1932 if (mddev->sync_super) {
1933 mddev->sync_super(mddev, rdev);
1937 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1939 super_types[mddev->major_version].sync_super(mddev, rdev);
1942 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1944 struct md_rdev *rdev, *rdev2;
1947 rdev_for_each_rcu(rdev, mddev1) {
1948 if (test_bit(Faulty, &rdev->flags) ||
1949 test_bit(Journal, &rdev->flags) ||
1950 rdev->raid_disk == -1)
1952 rdev_for_each_rcu(rdev2, mddev2) {
1953 if (test_bit(Faulty, &rdev2->flags) ||
1954 test_bit(Journal, &rdev2->flags) ||
1955 rdev2->raid_disk == -1)
1957 if (rdev->bdev->bd_contains ==
1958 rdev2->bdev->bd_contains) {
1968 static LIST_HEAD(pending_raid_disks);
1971 * Try to register data integrity profile for an mddev
1973 * This is called when an array is started and after a disk has been kicked
1974 * from the array. It only succeeds if all working and active component devices
1975 * are integrity capable with matching profiles.
1977 int md_integrity_register(struct mddev *mddev)
1979 struct md_rdev *rdev, *reference = NULL;
1981 if (list_empty(&mddev->disks))
1982 return 0; /* nothing to do */
1983 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1984 return 0; /* shouldn't register, or already is */
1985 rdev_for_each(rdev, mddev) {
1986 /* skip spares and non-functional disks */
1987 if (test_bit(Faulty, &rdev->flags))
1989 if (rdev->raid_disk < 0)
1992 /* Use the first rdev as the reference */
1996 /* does this rdev's profile match the reference profile? */
1997 if (blk_integrity_compare(reference->bdev->bd_disk,
1998 rdev->bdev->bd_disk) < 0)
2001 if (!reference || !bdev_get_integrity(reference->bdev))
2004 * All component devices are integrity capable and have matching
2005 * profiles, register the common profile for the md device.
2007 blk_integrity_register(mddev->gendisk,
2008 bdev_get_integrity(reference->bdev));
2010 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2011 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2012 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2018 EXPORT_SYMBOL(md_integrity_register);
2021 * Attempt to add an rdev, but only if it is consistent with the current
2024 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2026 struct blk_integrity *bi_rdev;
2027 struct blk_integrity *bi_mddev;
2028 char name[BDEVNAME_SIZE];
2030 if (!mddev->gendisk)
2033 bi_rdev = bdev_get_integrity(rdev->bdev);
2034 bi_mddev = blk_get_integrity(mddev->gendisk);
2036 if (!bi_mddev) /* nothing to do */
2039 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2040 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2041 mdname(mddev), bdevname(rdev->bdev, name));
2047 EXPORT_SYMBOL(md_integrity_add_rdev);
2049 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2051 char b[BDEVNAME_SIZE];
2055 /* prevent duplicates */
2056 if (find_rdev(mddev, rdev->bdev->bd_dev))
2059 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2060 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2061 rdev->sectors < mddev->dev_sectors)) {
2063 /* Cannot change size, so fail
2064 * If mddev->level <= 0, then we don't care
2065 * about aligning sizes (e.g. linear)
2067 if (mddev->level > 0)
2070 mddev->dev_sectors = rdev->sectors;
2073 /* Verify rdev->desc_nr is unique.
2074 * If it is -1, assign a free number, else
2075 * check number is not in use
2078 if (rdev->desc_nr < 0) {
2081 choice = mddev->raid_disks;
2082 while (md_find_rdev_nr_rcu(mddev, choice))
2084 rdev->desc_nr = choice;
2086 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2092 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2093 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2094 mdname(mddev), mddev->max_disks);
2097 bdevname(rdev->bdev,b);
2098 strreplace(b, '/', '!');
2100 rdev->mddev = mddev;
2101 printk(KERN_INFO "md: bind<%s>\n", b);
2103 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2106 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2107 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2108 /* failure here is OK */;
2109 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2111 list_add_rcu(&rdev->same_set, &mddev->disks);
2112 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2114 /* May as well allow recovery to be retried once */
2115 mddev->recovery_disabled++;
2120 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2125 static void md_delayed_delete(struct work_struct *ws)
2127 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2128 kobject_del(&rdev->kobj);
2129 kobject_put(&rdev->kobj);
2132 static void unbind_rdev_from_array(struct md_rdev *rdev)
2134 char b[BDEVNAME_SIZE];
2136 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2137 list_del_rcu(&rdev->same_set);
2138 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2140 sysfs_remove_link(&rdev->kobj, "block");
2141 sysfs_put(rdev->sysfs_state);
2142 rdev->sysfs_state = NULL;
2143 rdev->badblocks.count = 0;
2144 /* We need to delay this, otherwise we can deadlock when
2145 * writing to 'remove' to "dev/state". We also need
2146 * to delay it due to rcu usage.
2149 INIT_WORK(&rdev->del_work, md_delayed_delete);
2150 kobject_get(&rdev->kobj);
2151 queue_work(md_misc_wq, &rdev->del_work);
2155 * prevent the device from being mounted, repartitioned or
2156 * otherwise reused by a RAID array (or any other kernel
2157 * subsystem), by bd_claiming the device.
2159 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2162 struct block_device *bdev;
2163 char b[BDEVNAME_SIZE];
2165 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2166 shared ? (struct md_rdev *)lock_rdev : rdev);
2168 printk(KERN_ERR "md: could not open %s.\n",
2169 __bdevname(dev, b));
2170 return PTR_ERR(bdev);
2176 static void unlock_rdev(struct md_rdev *rdev)
2178 struct block_device *bdev = rdev->bdev;
2180 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2183 void md_autodetect_dev(dev_t dev);
2185 static void export_rdev(struct md_rdev *rdev)
2187 char b[BDEVNAME_SIZE];
2189 printk(KERN_INFO "md: export_rdev(%s)\n",
2190 bdevname(rdev->bdev,b));
2191 md_rdev_clear(rdev);
2193 if (test_bit(AutoDetected, &rdev->flags))
2194 md_autodetect_dev(rdev->bdev->bd_dev);
2197 kobject_put(&rdev->kobj);
2200 void md_kick_rdev_from_array(struct md_rdev *rdev)
2202 unbind_rdev_from_array(rdev);
2205 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2207 static void export_array(struct mddev *mddev)
2209 struct md_rdev *rdev;
2211 while (!list_empty(&mddev->disks)) {
2212 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2214 md_kick_rdev_from_array(rdev);
2216 mddev->raid_disks = 0;
2217 mddev->major_version = 0;
2220 static void sync_sbs(struct mddev *mddev, int nospares)
2222 /* Update each superblock (in-memory image), but
2223 * if we are allowed to, skip spares which already
2224 * have the right event counter, or have one earlier
2225 * (which would mean they aren't being marked as dirty
2226 * with the rest of the array)
2228 struct md_rdev *rdev;
2229 rdev_for_each(rdev, mddev) {
2230 if (rdev->sb_events == mddev->events ||
2232 rdev->raid_disk < 0 &&
2233 rdev->sb_events+1 == mddev->events)) {
2234 /* Don't update this superblock */
2235 rdev->sb_loaded = 2;
2237 sync_super(mddev, rdev);
2238 rdev->sb_loaded = 1;
2243 static bool does_sb_need_changing(struct mddev *mddev)
2245 struct md_rdev *rdev;
2246 struct mdp_superblock_1 *sb;
2249 /* Find a good rdev */
2250 rdev_for_each(rdev, mddev)
2251 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2254 /* No good device found. */
2258 sb = page_address(rdev->sb_page);
2259 /* Check if a device has become faulty or a spare become active */
2260 rdev_for_each(rdev, mddev) {
2261 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2262 /* Device activated? */
2263 if (role == 0xffff && rdev->raid_disk >=0 &&
2264 !test_bit(Faulty, &rdev->flags))
2266 /* Device turned faulty? */
2267 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2271 /* Check if any mddev parameters have changed */
2272 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2273 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2274 (mddev->layout != le64_to_cpu(sb->layout)) ||
2275 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2276 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2282 void md_update_sb(struct mddev *mddev, int force_change)
2284 struct md_rdev *rdev;
2287 int any_badblocks_changed = 0;
2292 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2296 if (mddev_is_clustered(mddev)) {
2297 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2299 ret = md_cluster_ops->metadata_update_start(mddev);
2300 /* Has someone else has updated the sb */
2301 if (!does_sb_need_changing(mddev)) {
2303 md_cluster_ops->metadata_update_cancel(mddev);
2304 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2309 /* First make sure individual recovery_offsets are correct */
2310 rdev_for_each(rdev, mddev) {
2311 if (rdev->raid_disk >= 0 &&
2312 mddev->delta_disks >= 0 &&
2313 !test_bit(Journal, &rdev->flags) &&
2314 !test_bit(In_sync, &rdev->flags) &&
2315 mddev->curr_resync_completed > rdev->recovery_offset)
2316 rdev->recovery_offset = mddev->curr_resync_completed;
2319 if (!mddev->persistent) {
2320 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2321 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2322 if (!mddev->external) {
2323 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2324 rdev_for_each(rdev, mddev) {
2325 if (rdev->badblocks.changed) {
2326 rdev->badblocks.changed = 0;
2327 md_ack_all_badblocks(&rdev->badblocks);
2328 md_error(mddev, rdev);
2330 clear_bit(Blocked, &rdev->flags);
2331 clear_bit(BlockedBadBlocks, &rdev->flags);
2332 wake_up(&rdev->blocked_wait);
2335 wake_up(&mddev->sb_wait);
2339 spin_lock(&mddev->lock);
2341 mddev->utime = get_seconds();
2343 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2345 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2346 /* just a clean<-> dirty transition, possibly leave spares alone,
2347 * though if events isn't the right even/odd, we will have to do
2353 if (mddev->degraded)
2354 /* If the array is degraded, then skipping spares is both
2355 * dangerous and fairly pointless.
2356 * Dangerous because a device that was removed from the array
2357 * might have a event_count that still looks up-to-date,
2358 * so it can be re-added without a resync.
2359 * Pointless because if there are any spares to skip,
2360 * then a recovery will happen and soon that array won't
2361 * be degraded any more and the spare can go back to sleep then.
2365 sync_req = mddev->in_sync;
2367 /* If this is just a dirty<->clean transition, and the array is clean
2368 * and 'events' is odd, we can roll back to the previous clean state */
2370 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2371 && mddev->can_decrease_events
2372 && mddev->events != 1) {
2374 mddev->can_decrease_events = 0;
2376 /* otherwise we have to go forward and ... */
2378 mddev->can_decrease_events = nospares;
2382 * This 64-bit counter should never wrap.
2383 * Either we are in around ~1 trillion A.C., assuming
2384 * 1 reboot per second, or we have a bug...
2386 WARN_ON(mddev->events == 0);
2388 rdev_for_each(rdev, mddev) {
2389 if (rdev->badblocks.changed)
2390 any_badblocks_changed++;
2391 if (test_bit(Faulty, &rdev->flags))
2392 set_bit(FaultRecorded, &rdev->flags);
2395 sync_sbs(mddev, nospares);
2396 spin_unlock(&mddev->lock);
2398 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2399 mdname(mddev), mddev->in_sync);
2401 bitmap_update_sb(mddev->bitmap);
2402 rdev_for_each(rdev, mddev) {
2403 char b[BDEVNAME_SIZE];
2405 if (rdev->sb_loaded != 1)
2406 continue; /* no noise on spare devices */
2408 if (!test_bit(Faulty, &rdev->flags)) {
2409 md_super_write(mddev,rdev,
2410 rdev->sb_start, rdev->sb_size,
2412 pr_debug("md: (write) %s's sb offset: %llu\n",
2413 bdevname(rdev->bdev, b),
2414 (unsigned long long)rdev->sb_start);
2415 rdev->sb_events = mddev->events;
2416 if (rdev->badblocks.size) {
2417 md_super_write(mddev, rdev,
2418 rdev->badblocks.sector,
2419 rdev->badblocks.size << 9,
2421 rdev->badblocks.size = 0;
2425 pr_debug("md: %s (skipping faulty)\n",
2426 bdevname(rdev->bdev, b));
2428 if (mddev->level == LEVEL_MULTIPATH)
2429 /* only need to write one superblock... */
2432 md_super_wait(mddev);
2433 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2435 spin_lock(&mddev->lock);
2436 if (mddev->in_sync != sync_req ||
2437 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2438 /* have to write it out again */
2439 spin_unlock(&mddev->lock);
2442 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2443 spin_unlock(&mddev->lock);
2444 wake_up(&mddev->sb_wait);
2445 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2446 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2448 rdev_for_each(rdev, mddev) {
2449 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2450 clear_bit(Blocked, &rdev->flags);
2452 if (any_badblocks_changed)
2453 md_ack_all_badblocks(&rdev->badblocks);
2454 clear_bit(BlockedBadBlocks, &rdev->flags);
2455 wake_up(&rdev->blocked_wait);
2458 if (mddev_is_clustered(mddev) && ret == 0)
2459 md_cluster_ops->metadata_update_finish(mddev);
2461 EXPORT_SYMBOL(md_update_sb);
2463 static int add_bound_rdev(struct md_rdev *rdev)
2465 struct mddev *mddev = rdev->mddev;
2468 if (!mddev->pers->hot_remove_disk) {
2469 /* If there is hot_add_disk but no hot_remove_disk
2470 * then added disks for geometry changes,
2471 * and should be added immediately.
2473 super_types[mddev->major_version].
2474 validate_super(mddev, rdev);
2475 err = mddev->pers->hot_add_disk(mddev, rdev);
2477 unbind_rdev_from_array(rdev);
2482 sysfs_notify_dirent_safe(rdev->sysfs_state);
2484 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2485 if (mddev->degraded)
2486 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2487 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2488 md_new_event(mddev);
2489 md_wakeup_thread(mddev->thread);
2493 /* words written to sysfs files may, or may not, be \n terminated.
2494 * We want to accept with case. For this we use cmd_match.
2496 static int cmd_match(const char *cmd, const char *str)
2498 /* See if cmd, written into a sysfs file, matches
2499 * str. They must either be the same, or cmd can
2500 * have a trailing newline
2502 while (*cmd && *str && *cmd == *str) {
2513 struct rdev_sysfs_entry {
2514 struct attribute attr;
2515 ssize_t (*show)(struct md_rdev *, char *);
2516 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2520 state_show(struct md_rdev *rdev, char *page)
2524 unsigned long flags = ACCESS_ONCE(rdev->flags);
2526 if (test_bit(Faulty, &flags) ||
2527 rdev->badblocks.unacked_exist) {
2528 len+= sprintf(page+len, "%sfaulty",sep);
2531 if (test_bit(In_sync, &flags)) {
2532 len += sprintf(page+len, "%sin_sync",sep);
2535 if (test_bit(Journal, &flags)) {
2536 len += sprintf(page+len, "%sjournal",sep);
2539 if (test_bit(WriteMostly, &flags)) {
2540 len += sprintf(page+len, "%swrite_mostly",sep);
2543 if (test_bit(Blocked, &flags) ||
2544 (rdev->badblocks.unacked_exist
2545 && !test_bit(Faulty, &flags))) {
2546 len += sprintf(page+len, "%sblocked", sep);
2549 if (!test_bit(Faulty, &flags) &&
2550 !test_bit(Journal, &flags) &&
2551 !test_bit(In_sync, &flags)) {
2552 len += sprintf(page+len, "%sspare", sep);
2555 if (test_bit(WriteErrorSeen, &flags)) {
2556 len += sprintf(page+len, "%swrite_error", sep);
2559 if (test_bit(WantReplacement, &flags)) {
2560 len += sprintf(page+len, "%swant_replacement", sep);
2563 if (test_bit(Replacement, &flags)) {
2564 len += sprintf(page+len, "%sreplacement", sep);
2568 return len+sprintf(page+len, "\n");
2572 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2575 * faulty - simulates an error
2576 * remove - disconnects the device
2577 * writemostly - sets write_mostly
2578 * -writemostly - clears write_mostly
2579 * blocked - sets the Blocked flags
2580 * -blocked - clears the Blocked and possibly simulates an error
2581 * insync - sets Insync providing device isn't active
2582 * -insync - clear Insync for a device with a slot assigned,
2583 * so that it gets rebuilt based on bitmap
2584 * write_error - sets WriteErrorSeen
2585 * -write_error - clears WriteErrorSeen
2588 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2589 md_error(rdev->mddev, rdev);
2590 if (test_bit(Faulty, &rdev->flags))
2594 } else if (cmd_match(buf, "remove")) {
2595 if (rdev->raid_disk >= 0)
2598 struct mddev *mddev = rdev->mddev;
2600 if (mddev_is_clustered(mddev))
2601 err = md_cluster_ops->remove_disk(mddev, rdev);
2604 md_kick_rdev_from_array(rdev);
2606 md_update_sb(mddev, 1);
2607 md_new_event(mddev);
2610 } else if (cmd_match(buf, "writemostly")) {
2611 set_bit(WriteMostly, &rdev->flags);
2613 } else if (cmd_match(buf, "-writemostly")) {
2614 clear_bit(WriteMostly, &rdev->flags);
2616 } else if (cmd_match(buf, "blocked")) {
2617 set_bit(Blocked, &rdev->flags);
2619 } else if (cmd_match(buf, "-blocked")) {
2620 if (!test_bit(Faulty, &rdev->flags) &&
2621 rdev->badblocks.unacked_exist) {
2622 /* metadata handler doesn't understand badblocks,
2623 * so we need to fail the device
2625 md_error(rdev->mddev, rdev);
2627 clear_bit(Blocked, &rdev->flags);
2628 clear_bit(BlockedBadBlocks, &rdev->flags);
2629 wake_up(&rdev->blocked_wait);
2630 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2631 md_wakeup_thread(rdev->mddev->thread);
2634 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2635 set_bit(In_sync, &rdev->flags);
2637 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2638 !test_bit(Journal, &rdev->flags)) {
2639 if (rdev->mddev->pers == NULL) {
2640 clear_bit(In_sync, &rdev->flags);
2641 rdev->saved_raid_disk = rdev->raid_disk;
2642 rdev->raid_disk = -1;
2645 } else if (cmd_match(buf, "write_error")) {
2646 set_bit(WriteErrorSeen, &rdev->flags);
2648 } else if (cmd_match(buf, "-write_error")) {
2649 clear_bit(WriteErrorSeen, &rdev->flags);
2651 } else if (cmd_match(buf, "want_replacement")) {
2652 /* Any non-spare device that is not a replacement can
2653 * become want_replacement at any time, but we then need to
2654 * check if recovery is needed.
2656 if (rdev->raid_disk >= 0 &&
2657 !test_bit(Journal, &rdev->flags) &&
2658 !test_bit(Replacement, &rdev->flags))
2659 set_bit(WantReplacement, &rdev->flags);
2660 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2661 md_wakeup_thread(rdev->mddev->thread);
2663 } else if (cmd_match(buf, "-want_replacement")) {
2664 /* Clearing 'want_replacement' is always allowed.
2665 * Once replacements starts it is too late though.
2668 clear_bit(WantReplacement, &rdev->flags);
2669 } else if (cmd_match(buf, "replacement")) {
2670 /* Can only set a device as a replacement when array has not
2671 * yet been started. Once running, replacement is automatic
2672 * from spares, or by assigning 'slot'.
2674 if (rdev->mddev->pers)
2677 set_bit(Replacement, &rdev->flags);
2680 } else if (cmd_match(buf, "-replacement")) {
2681 /* Similarly, can only clear Replacement before start */
2682 if (rdev->mddev->pers)
2685 clear_bit(Replacement, &rdev->flags);
2688 } else if (cmd_match(buf, "re-add")) {
2689 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2690 /* clear_bit is performed _after_ all the devices
2691 * have their local Faulty bit cleared. If any writes
2692 * happen in the meantime in the local node, they
2693 * will land in the local bitmap, which will be synced
2694 * by this node eventually
2696 if (!mddev_is_clustered(rdev->mddev) ||
2697 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2698 clear_bit(Faulty, &rdev->flags);
2699 err = add_bound_rdev(rdev);
2705 sysfs_notify_dirent_safe(rdev->sysfs_state);
2706 return err ? err : len;
2708 static struct rdev_sysfs_entry rdev_state =
2709 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2712 errors_show(struct md_rdev *rdev, char *page)
2714 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2718 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2723 rv = kstrtouint(buf, 10, &n);
2726 atomic_set(&rdev->corrected_errors, n);
2729 static struct rdev_sysfs_entry rdev_errors =
2730 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2733 slot_show(struct md_rdev *rdev, char *page)
2735 if (test_bit(Journal, &rdev->flags))
2736 return sprintf(page, "journal\n");
2737 else if (rdev->raid_disk < 0)
2738 return sprintf(page, "none\n");
2740 return sprintf(page, "%d\n", rdev->raid_disk);
2744 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2749 if (test_bit(Journal, &rdev->flags))
2751 if (strncmp(buf, "none", 4)==0)
2754 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2758 if (rdev->mddev->pers && slot == -1) {
2759 /* Setting 'slot' on an active array requires also
2760 * updating the 'rd%d' link, and communicating
2761 * with the personality with ->hot_*_disk.
2762 * For now we only support removing
2763 * failed/spare devices. This normally happens automatically,
2764 * but not when the metadata is externally managed.
2766 if (rdev->raid_disk == -1)
2768 /* personality does all needed checks */
2769 if (rdev->mddev->pers->hot_remove_disk == NULL)
2771 clear_bit(Blocked, &rdev->flags);
2772 remove_and_add_spares(rdev->mddev, rdev);
2773 if (rdev->raid_disk >= 0)
2775 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2776 md_wakeup_thread(rdev->mddev->thread);
2777 } else if (rdev->mddev->pers) {
2778 /* Activating a spare .. or possibly reactivating
2779 * if we ever get bitmaps working here.
2783 if (rdev->raid_disk != -1)
2786 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2789 if (rdev->mddev->pers->hot_add_disk == NULL)
2792 if (slot >= rdev->mddev->raid_disks &&
2793 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2796 rdev->raid_disk = slot;
2797 if (test_bit(In_sync, &rdev->flags))
2798 rdev->saved_raid_disk = slot;
2800 rdev->saved_raid_disk = -1;
2801 clear_bit(In_sync, &rdev->flags);
2802 clear_bit(Bitmap_sync, &rdev->flags);
2803 err = rdev->mddev->pers->
2804 hot_add_disk(rdev->mddev, rdev);
2806 rdev->raid_disk = -1;
2809 sysfs_notify_dirent_safe(rdev->sysfs_state);
2810 if (sysfs_link_rdev(rdev->mddev, rdev))
2811 /* failure here is OK */;
2812 /* don't wakeup anyone, leave that to userspace. */
2814 if (slot >= rdev->mddev->raid_disks &&
2815 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2817 rdev->raid_disk = slot;
2818 /* assume it is working */
2819 clear_bit(Faulty, &rdev->flags);
2820 clear_bit(WriteMostly, &rdev->flags);
2821 set_bit(In_sync, &rdev->flags);
2822 sysfs_notify_dirent_safe(rdev->sysfs_state);
2827 static struct rdev_sysfs_entry rdev_slot =
2828 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2831 offset_show(struct md_rdev *rdev, char *page)
2833 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2837 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2839 unsigned long long offset;
2840 if (kstrtoull(buf, 10, &offset) < 0)
2842 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2844 if (rdev->sectors && rdev->mddev->external)
2845 /* Must set offset before size, so overlap checks
2848 rdev->data_offset = offset;
2849 rdev->new_data_offset = offset;
2853 static struct rdev_sysfs_entry rdev_offset =
2854 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2856 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2858 return sprintf(page, "%llu\n",
2859 (unsigned long long)rdev->new_data_offset);
2862 static ssize_t new_offset_store(struct md_rdev *rdev,
2863 const char *buf, size_t len)
2865 unsigned long long new_offset;
2866 struct mddev *mddev = rdev->mddev;
2868 if (kstrtoull(buf, 10, &new_offset) < 0)
2871 if (mddev->sync_thread ||
2872 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2874 if (new_offset == rdev->data_offset)
2875 /* reset is always permitted */
2877 else if (new_offset > rdev->data_offset) {
2878 /* must not push array size beyond rdev_sectors */
2879 if (new_offset - rdev->data_offset
2880 + mddev->dev_sectors > rdev->sectors)
2883 /* Metadata worries about other space details. */
2885 /* decreasing the offset is inconsistent with a backwards
2888 if (new_offset < rdev->data_offset &&
2889 mddev->reshape_backwards)
2891 /* Increasing offset is inconsistent with forwards
2892 * reshape. reshape_direction should be set to
2893 * 'backwards' first.
2895 if (new_offset > rdev->data_offset &&
2896 !mddev->reshape_backwards)
2899 if (mddev->pers && mddev->persistent &&
2900 !super_types[mddev->major_version]
2901 .allow_new_offset(rdev, new_offset))
2903 rdev->new_data_offset = new_offset;
2904 if (new_offset > rdev->data_offset)
2905 mddev->reshape_backwards = 1;
2906 else if (new_offset < rdev->data_offset)
2907 mddev->reshape_backwards = 0;
2911 static struct rdev_sysfs_entry rdev_new_offset =
2912 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2915 rdev_size_show(struct md_rdev *rdev, char *page)
2917 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2920 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2922 /* check if two start/length pairs overlap */
2930 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2932 unsigned long long blocks;
2935 if (kstrtoull(buf, 10, &blocks) < 0)
2938 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2939 return -EINVAL; /* sector conversion overflow */
2942 if (new != blocks * 2)
2943 return -EINVAL; /* unsigned long long to sector_t overflow */
2950 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2952 struct mddev *my_mddev = rdev->mddev;
2953 sector_t oldsectors = rdev->sectors;
2956 if (test_bit(Journal, &rdev->flags))
2958 if (strict_blocks_to_sectors(buf, §ors) < 0)
2960 if (rdev->data_offset != rdev->new_data_offset)
2961 return -EINVAL; /* too confusing */
2962 if (my_mddev->pers && rdev->raid_disk >= 0) {
2963 if (my_mddev->persistent) {
2964 sectors = super_types[my_mddev->major_version].
2965 rdev_size_change(rdev, sectors);
2968 } else if (!sectors)
2969 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2971 if (!my_mddev->pers->resize)
2972 /* Cannot change size for RAID0 or Linear etc */
2975 if (sectors < my_mddev->dev_sectors)
2976 return -EINVAL; /* component must fit device */
2978 rdev->sectors = sectors;
2979 if (sectors > oldsectors && my_mddev->external) {
2980 /* Need to check that all other rdevs with the same
2981 * ->bdev do not overlap. 'rcu' is sufficient to walk
2982 * the rdev lists safely.
2983 * This check does not provide a hard guarantee, it
2984 * just helps avoid dangerous mistakes.
2986 struct mddev *mddev;
2988 struct list_head *tmp;
2991 for_each_mddev(mddev, tmp) {
2992 struct md_rdev *rdev2;
2994 rdev_for_each(rdev2, mddev)
2995 if (rdev->bdev == rdev2->bdev &&
2997 overlaps(rdev->data_offset, rdev->sectors,
3010 /* Someone else could have slipped in a size
3011 * change here, but doing so is just silly.
3012 * We put oldsectors back because we *know* it is
3013 * safe, and trust userspace not to race with
3016 rdev->sectors = oldsectors;
3023 static struct rdev_sysfs_entry rdev_size =
3024 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3026 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3028 unsigned long long recovery_start = rdev->recovery_offset;
3030 if (test_bit(In_sync, &rdev->flags) ||
3031 recovery_start == MaxSector)
3032 return sprintf(page, "none\n");
3034 return sprintf(page, "%llu\n", recovery_start);
3037 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3039 unsigned long long recovery_start;
3041 if (cmd_match(buf, "none"))
3042 recovery_start = MaxSector;
3043 else if (kstrtoull(buf, 10, &recovery_start))
3046 if (rdev->mddev->pers &&
3047 rdev->raid_disk >= 0)
3050 rdev->recovery_offset = recovery_start;
3051 if (recovery_start == MaxSector)
3052 set_bit(In_sync, &rdev->flags);
3054 clear_bit(In_sync, &rdev->flags);
3058 static struct rdev_sysfs_entry rdev_recovery_start =
3059 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3062 badblocks_show(struct badblocks *bb, char *page, int unack);
3064 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3066 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3068 return badblocks_show(&rdev->badblocks, page, 0);
3070 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3072 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3073 /* Maybe that ack was all we needed */
3074 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3075 wake_up(&rdev->blocked_wait);
3078 static struct rdev_sysfs_entry rdev_bad_blocks =
3079 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3081 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3083 return badblocks_show(&rdev->badblocks, page, 1);
3085 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3087 return badblocks_store(&rdev->badblocks, page, len, 1);
3089 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3090 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3092 static struct attribute *rdev_default_attrs[] = {
3097 &rdev_new_offset.attr,
3099 &rdev_recovery_start.attr,
3100 &rdev_bad_blocks.attr,
3101 &rdev_unack_bad_blocks.attr,
3105 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3107 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3108 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3114 return entry->show(rdev, page);
3118 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3119 const char *page, size_t length)
3121 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3122 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3124 struct mddev *mddev = rdev->mddev;
3128 if (!capable(CAP_SYS_ADMIN))
3130 rv = mddev ? mddev_lock(mddev): -EBUSY;
3132 if (rdev->mddev == NULL)
3135 rv = entry->store(rdev, page, length);
3136 mddev_unlock(mddev);
3141 static void rdev_free(struct kobject *ko)
3143 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3146 static const struct sysfs_ops rdev_sysfs_ops = {
3147 .show = rdev_attr_show,
3148 .store = rdev_attr_store,
3150 static struct kobj_type rdev_ktype = {
3151 .release = rdev_free,
3152 .sysfs_ops = &rdev_sysfs_ops,
3153 .default_attrs = rdev_default_attrs,
3156 int md_rdev_init(struct md_rdev *rdev)
3159 rdev->saved_raid_disk = -1;
3160 rdev->raid_disk = -1;
3162 rdev->data_offset = 0;
3163 rdev->new_data_offset = 0;
3164 rdev->sb_events = 0;
3165 rdev->last_read_error.tv_sec = 0;
3166 rdev->last_read_error.tv_nsec = 0;
3167 rdev->sb_loaded = 0;
3168 rdev->bb_page = NULL;
3169 atomic_set(&rdev->nr_pending, 0);
3170 atomic_set(&rdev->read_errors, 0);
3171 atomic_set(&rdev->corrected_errors, 0);
3173 INIT_LIST_HEAD(&rdev->same_set);
3174 init_waitqueue_head(&rdev->blocked_wait);
3176 /* Add space to store bad block list.
3177 * This reserves the space even on arrays where it cannot
3178 * be used - I wonder if that matters
3180 rdev->badblocks.count = 0;
3181 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3182 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3183 seqlock_init(&rdev->badblocks.lock);
3184 if (rdev->badblocks.page == NULL)
3189 EXPORT_SYMBOL_GPL(md_rdev_init);
3191 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3193 * mark the device faulty if:
3195 * - the device is nonexistent (zero size)
3196 * - the device has no valid superblock
3198 * a faulty rdev _never_ has rdev->sb set.
3200 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3202 char b[BDEVNAME_SIZE];
3204 struct md_rdev *rdev;
3207 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3209 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3210 return ERR_PTR(-ENOMEM);
3213 err = md_rdev_init(rdev);
3216 err = alloc_disk_sb(rdev);
3220 err = lock_rdev(rdev, newdev, super_format == -2);
3224 kobject_init(&rdev->kobj, &rdev_ktype);
3226 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3229 "md: %s has zero or unknown size, marking faulty!\n",
3230 bdevname(rdev->bdev,b));
3235 if (super_format >= 0) {
3236 err = super_types[super_format].
3237 load_super(rdev, NULL, super_minor);
3238 if (err == -EINVAL) {
3240 "md: %s does not have a valid v%d.%d "
3241 "superblock, not importing!\n",
3242 bdevname(rdev->bdev,b),
3243 super_format, super_minor);
3248 "md: could not read %s's sb, not importing!\n",
3249 bdevname(rdev->bdev,b));
3259 md_rdev_clear(rdev);
3261 return ERR_PTR(err);
3265 * Check a full RAID array for plausibility
3268 static void analyze_sbs(struct mddev *mddev)
3271 struct md_rdev *rdev, *freshest, *tmp;
3272 char b[BDEVNAME_SIZE];
3275 rdev_for_each_safe(rdev, tmp, mddev)
3276 switch (super_types[mddev->major_version].
3277 load_super(rdev, freshest, mddev->minor_version)) {
3285 "md: fatal superblock inconsistency in %s"
3286 " -- removing from array\n",
3287 bdevname(rdev->bdev,b));
3288 md_kick_rdev_from_array(rdev);
3291 super_types[mddev->major_version].
3292 validate_super(mddev, freshest);
3295 rdev_for_each_safe(rdev, tmp, mddev) {
3296 if (mddev->max_disks &&
3297 (rdev->desc_nr >= mddev->max_disks ||
3298 i > mddev->max_disks)) {
3300 "md: %s: %s: only %d devices permitted\n",
3301 mdname(mddev), bdevname(rdev->bdev, b),
3303 md_kick_rdev_from_array(rdev);
3306 if (rdev != freshest) {
3307 if (super_types[mddev->major_version].
3308 validate_super(mddev, rdev)) {
3309 printk(KERN_WARNING "md: kicking non-fresh %s"
3311 bdevname(rdev->bdev,b));
3312 md_kick_rdev_from_array(rdev);
3316 if (mddev->level == LEVEL_MULTIPATH) {
3317 rdev->desc_nr = i++;
3318 rdev->raid_disk = rdev->desc_nr;
3319 set_bit(In_sync, &rdev->flags);
3320 } else if (rdev->raid_disk >=
3321 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3322 !test_bit(Journal, &rdev->flags)) {
3323 rdev->raid_disk = -1;
3324 clear_bit(In_sync, &rdev->flags);
3329 /* Read a fixed-point number.
3330 * Numbers in sysfs attributes should be in "standard" units where
3331 * possible, so time should be in seconds.
3332 * However we internally use a a much smaller unit such as
3333 * milliseconds or jiffies.
3334 * This function takes a decimal number with a possible fractional
3335 * component, and produces an integer which is the result of
3336 * multiplying that number by 10^'scale'.
3337 * all without any floating-point arithmetic.
3339 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3341 unsigned long result = 0;
3343 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3346 else if (decimals < scale) {
3349 result = result * 10 + value;
3361 while (decimals < scale) {
3370 safe_delay_show(struct mddev *mddev, char *page)
3372 int msec = (mddev->safemode_delay*1000)/HZ;
3373 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3376 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3380 if (mddev_is_clustered(mddev)) {
3381 pr_info("md: Safemode is disabled for clustered mode\n");
3385 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3388 mddev->safemode_delay = 0;
3390 unsigned long old_delay = mddev->safemode_delay;
3391 unsigned long new_delay = (msec*HZ)/1000;
3395 mddev->safemode_delay = new_delay;
3396 if (new_delay < old_delay || old_delay == 0)
3397 mod_timer(&mddev->safemode_timer, jiffies+1);
3401 static struct md_sysfs_entry md_safe_delay =
3402 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3405 level_show(struct mddev *mddev, char *page)
3407 struct md_personality *p;
3409 spin_lock(&mddev->lock);
3412 ret = sprintf(page, "%s\n", p->name);
3413 else if (mddev->clevel[0])
3414 ret = sprintf(page, "%s\n", mddev->clevel);
3415 else if (mddev->level != LEVEL_NONE)
3416 ret = sprintf(page, "%d\n", mddev->level);
3419 spin_unlock(&mddev->lock);
3424 level_store(struct mddev *mddev, const char *buf, size_t len)
3429 struct md_personality *pers, *oldpers;
3431 void *priv, *oldpriv;
3432 struct md_rdev *rdev;
3434 if (slen == 0 || slen >= sizeof(clevel))
3437 rv = mddev_lock(mddev);
3441 if (mddev->pers == NULL) {
3442 strncpy(mddev->clevel, buf, slen);
3443 if (mddev->clevel[slen-1] == '\n')
3445 mddev->clevel[slen] = 0;
3446 mddev->level = LEVEL_NONE;
3454 /* request to change the personality. Need to ensure:
3455 * - array is not engaged in resync/recovery/reshape
3456 * - old personality can be suspended
3457 * - new personality will access other array.
3461 if (mddev->sync_thread ||
3462 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3463 mddev->reshape_position != MaxSector ||
3464 mddev->sysfs_active)
3468 if (!mddev->pers->quiesce) {
3469 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3470 mdname(mddev), mddev->pers->name);
3474 /* Now find the new personality */
3475 strncpy(clevel, buf, slen);
3476 if (clevel[slen-1] == '\n')
3479 if (kstrtol(clevel, 10, &level))
3482 if (request_module("md-%s", clevel) != 0)
3483 request_module("md-level-%s", clevel);
3484 spin_lock(&pers_lock);
3485 pers = find_pers(level, clevel);
3486 if (!pers || !try_module_get(pers->owner)) {
3487 spin_unlock(&pers_lock);
3488 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3492 spin_unlock(&pers_lock);
3494 if (pers == mddev->pers) {
3495 /* Nothing to do! */
3496 module_put(pers->owner);
3500 if (!pers->takeover) {
3501 module_put(pers->owner);
3502 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3503 mdname(mddev), clevel);
3508 rdev_for_each(rdev, mddev)
3509 rdev->new_raid_disk = rdev->raid_disk;
3511 /* ->takeover must set new_* and/or delta_disks
3512 * if it succeeds, and may set them when it fails.
3514 priv = pers->takeover(mddev);
3516 mddev->new_level = mddev->level;
3517 mddev->new_layout = mddev->layout;
3518 mddev->new_chunk_sectors = mddev->chunk_sectors;
3519 mddev->raid_disks -= mddev->delta_disks;
3520 mddev->delta_disks = 0;
3521 mddev->reshape_backwards = 0;
3522 module_put(pers->owner);
3523 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3524 mdname(mddev), clevel);
3529 /* Looks like we have a winner */
3530 mddev_suspend(mddev);
3531 mddev_detach(mddev);
3533 spin_lock(&mddev->lock);
3534 oldpers = mddev->pers;
3535 oldpriv = mddev->private;
3537 mddev->private = priv;
3538 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3539 mddev->level = mddev->new_level;
3540 mddev->layout = mddev->new_layout;
3541 mddev->chunk_sectors = mddev->new_chunk_sectors;
3542 mddev->delta_disks = 0;
3543 mddev->reshape_backwards = 0;
3544 mddev->degraded = 0;
3545 spin_unlock(&mddev->lock);
3547 if (oldpers->sync_request == NULL &&
3549 /* We are converting from a no-redundancy array
3550 * to a redundancy array and metadata is managed
3551 * externally so we need to be sure that writes
3552 * won't block due to a need to transition
3554 * until external management is started.
3557 mddev->safemode_delay = 0;
3558 mddev->safemode = 0;
3561 oldpers->free(mddev, oldpriv);
3563 if (oldpers->sync_request == NULL &&
3564 pers->sync_request != NULL) {
3565 /* need to add the md_redundancy_group */
3566 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3568 "md: cannot register extra attributes for %s\n",
3570 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3572 if (oldpers->sync_request != NULL &&
3573 pers->sync_request == NULL) {
3574 /* need to remove the md_redundancy_group */
3575 if (mddev->to_remove == NULL)
3576 mddev->to_remove = &md_redundancy_group;
3579 rdev_for_each(rdev, mddev) {
3580 if (rdev->raid_disk < 0)
3582 if (rdev->new_raid_disk >= mddev->raid_disks)
3583 rdev->new_raid_disk = -1;
3584 if (rdev->new_raid_disk == rdev->raid_disk)
3586 sysfs_unlink_rdev(mddev, rdev);
3588 rdev_for_each(rdev, mddev) {
3589 if (rdev->raid_disk < 0)
3591 if (rdev->new_raid_disk == rdev->raid_disk)
3593 rdev->raid_disk = rdev->new_raid_disk;
3594 if (rdev->raid_disk < 0)
3595 clear_bit(In_sync, &rdev->flags);
3597 if (sysfs_link_rdev(mddev, rdev))
3598 printk(KERN_WARNING "md: cannot register rd%d"
3599 " for %s after level change\n",
3600 rdev->raid_disk, mdname(mddev));
3604 if (pers->sync_request == NULL) {
3605 /* this is now an array without redundancy, so
3606 * it must always be in_sync
3609 del_timer_sync(&mddev->safemode_timer);
3611 blk_set_stacking_limits(&mddev->queue->limits);
3613 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3614 mddev_resume(mddev);
3616 md_update_sb(mddev, 1);
3617 sysfs_notify(&mddev->kobj, NULL, "level");
3618 md_new_event(mddev);
3621 mddev_unlock(mddev);
3625 static struct md_sysfs_entry md_level =
3626 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3629 layout_show(struct mddev *mddev, char *page)
3631 /* just a number, not meaningful for all levels */
3632 if (mddev->reshape_position != MaxSector &&
3633 mddev->layout != mddev->new_layout)
3634 return sprintf(page, "%d (%d)\n",
3635 mddev->new_layout, mddev->layout);
3636 return sprintf(page, "%d\n", mddev->layout);
3640 layout_store(struct mddev *mddev, const char *buf, size_t len)
3645 err = kstrtouint(buf, 10, &n);
3648 err = mddev_lock(mddev);
3653 if (mddev->pers->check_reshape == NULL)
3658 mddev->new_layout = n;
3659 err = mddev->pers->check_reshape(mddev);
3661 mddev->new_layout = mddev->layout;
3664 mddev->new_layout = n;
3665 if (mddev->reshape_position == MaxSector)
3668 mddev_unlock(mddev);
3671 static struct md_sysfs_entry md_layout =
3672 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3675 raid_disks_show(struct mddev *mddev, char *page)
3677 if (mddev->raid_disks == 0)
3679 if (mddev->reshape_position != MaxSector &&
3680 mddev->delta_disks != 0)
3681 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3682 mddev->raid_disks - mddev->delta_disks);
3683 return sprintf(page, "%d\n", mddev->raid_disks);
3686 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3689 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3694 err = kstrtouint(buf, 10, &n);
3698 err = mddev_lock(mddev);
3702 err = update_raid_disks(mddev, n);
3703 else if (mddev->reshape_position != MaxSector) {
3704 struct md_rdev *rdev;
3705 int olddisks = mddev->raid_disks - mddev->delta_disks;
3708 rdev_for_each(rdev, mddev) {
3710 rdev->data_offset < rdev->new_data_offset)
3713 rdev->data_offset > rdev->new_data_offset)
3717 mddev->delta_disks = n - olddisks;
3718 mddev->raid_disks = n;
3719 mddev->reshape_backwards = (mddev->delta_disks < 0);
3721 mddev->raid_disks = n;
3723 mddev_unlock(mddev);
3724 return err ? err : len;
3726 static struct md_sysfs_entry md_raid_disks =
3727 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3730 chunk_size_show(struct mddev *mddev, char *page)
3732 if (mddev->reshape_position != MaxSector &&
3733 mddev->chunk_sectors != mddev->new_chunk_sectors)
3734 return sprintf(page, "%d (%d)\n",
3735 mddev->new_chunk_sectors << 9,
3736 mddev->chunk_sectors << 9);
3737 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3741 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3746 err = kstrtoul(buf, 10, &n);
3750 err = mddev_lock(mddev);
3754 if (mddev->pers->check_reshape == NULL)
3759 mddev->new_chunk_sectors = n >> 9;
3760 err = mddev->pers->check_reshape(mddev);
3762 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;
3769 mddev_unlock(mddev);
3772 static struct md_sysfs_entry md_chunk_size =
3773 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3776 resync_start_show(struct mddev *mddev, char *page)
3778 if (mddev->recovery_cp == MaxSector)
3779 return sprintf(page, "none\n");
3780 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3784 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3786 unsigned long long n;
3789 if (cmd_match(buf, "none"))
3792 err = kstrtoull(buf, 10, &n);
3795 if (n != (sector_t)n)
3799 err = mddev_lock(mddev);
3802 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3806 mddev->recovery_cp = n;
3808 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3810 mddev_unlock(mddev);
3813 static struct md_sysfs_entry md_resync_start =
3814 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3815 resync_start_show, resync_start_store);
3818 * The array state can be:
3821 * No devices, no size, no level
3822 * Equivalent to STOP_ARRAY ioctl
3824 * May have some settings, but array is not active
3825 * all IO results in error
3826 * When written, doesn't tear down array, but just stops it
3827 * suspended (not supported yet)
3828 * All IO requests will block. The array can be reconfigured.
3829 * Writing this, if accepted, will block until array is quiescent
3831 * no resync can happen. no superblocks get written.
3832 * write requests fail
3834 * like readonly, but behaves like 'clean' on a write request.
3836 * clean - no pending writes, but otherwise active.
3837 * When written to inactive array, starts without resync
3838 * If a write request arrives then
3839 * if metadata is known, mark 'dirty' and switch to 'active'.
3840 * if not known, block and switch to write-pending
3841 * If written to an active array that has pending writes, then fails.
3843 * fully active: IO and resync can be happening.
3844 * When written to inactive array, starts with resync
3847 * clean, but writes are blocked waiting for 'active' to be written.
3850 * like active, but no writes have been seen for a while (100msec).
3853 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3854 write_pending, active_idle, bad_word};
3855 static char *array_states[] = {
3856 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3857 "write-pending", "active-idle", NULL };
3859 static int match_word(const char *word, char **list)
3862 for (n=0; list[n]; n++)
3863 if (cmd_match(word, list[n]))
3869 array_state_show(struct mddev *mddev, char *page)
3871 enum array_state st = inactive;
3884 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3886 else if (mddev->safemode)
3892 if (list_empty(&mddev->disks) &&
3893 mddev->raid_disks == 0 &&
3894 mddev->dev_sectors == 0)
3899 return sprintf(page, "%s\n", array_states[st]);
3902 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3903 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3904 static int do_md_run(struct mddev *mddev);
3905 static int restart_array(struct mddev *mddev);
3908 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3911 enum array_state st = match_word(buf, array_states);
3913 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3914 /* don't take reconfig_mutex when toggling between
3917 spin_lock(&mddev->lock);
3919 restart_array(mddev);
3920 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3921 wake_up(&mddev->sb_wait);
3923 } else /* st == clean */ {
3924 restart_array(mddev);
3925 if (atomic_read(&mddev->writes_pending) == 0) {
3926 if (mddev->in_sync == 0) {
3928 if (mddev->safemode == 1)
3929 mddev->safemode = 0;
3930 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3936 spin_unlock(&mddev->lock);
3939 err = mddev_lock(mddev);
3947 /* stopping an active array */
3948 err = do_md_stop(mddev, 0, NULL);
3951 /* stopping an active array */
3953 err = do_md_stop(mddev, 2, NULL);
3955 err = 0; /* already inactive */
3958 break; /* not supported yet */
3961 err = md_set_readonly(mddev, NULL);
3964 set_disk_ro(mddev->gendisk, 1);
3965 err = do_md_run(mddev);
3971 err = md_set_readonly(mddev, NULL);
3972 else if (mddev->ro == 1)
3973 err = restart_array(mddev);
3976 set_disk_ro(mddev->gendisk, 0);
3980 err = do_md_run(mddev);
3985 err = restart_array(mddev);
3988 spin_lock(&mddev->lock);
3989 if (atomic_read(&mddev->writes_pending) == 0) {
3990 if (mddev->in_sync == 0) {
3992 if (mddev->safemode == 1)
3993 mddev->safemode = 0;
3994 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3999 spin_unlock(&mddev->lock);
4005 err = restart_array(mddev);
4008 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4009 wake_up(&mddev->sb_wait);
4013 set_disk_ro(mddev->gendisk, 0);
4014 err = do_md_run(mddev);
4019 /* these cannot be set */
4024 if (mddev->hold_active == UNTIL_IOCTL)
4025 mddev->hold_active = 0;
4026 sysfs_notify_dirent_safe(mddev->sysfs_state);
4028 mddev_unlock(mddev);
4031 static struct md_sysfs_entry md_array_state =
4032 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4035 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4036 return sprintf(page, "%d\n",
4037 atomic_read(&mddev->max_corr_read_errors));
4041 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4046 rv = kstrtouint(buf, 10, &n);
4049 atomic_set(&mddev->max_corr_read_errors, n);
4053 static struct md_sysfs_entry max_corr_read_errors =
4054 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4055 max_corrected_read_errors_store);
4058 null_show(struct mddev *mddev, char *page)
4064 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4066 /* buf must be %d:%d\n? giving major and minor numbers */
4067 /* The new device is added to the array.
4068 * If the array has a persistent superblock, we read the
4069 * superblock to initialise info and check validity.
4070 * Otherwise, only checking done is that in bind_rdev_to_array,
4071 * which mainly checks size.
4074 int major = simple_strtoul(buf, &e, 10);
4077 struct md_rdev *rdev;
4080 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4082 minor = simple_strtoul(e+1, &e, 10);
4083 if (*e && *e != '\n')
4085 dev = MKDEV(major, minor);
4086 if (major != MAJOR(dev) ||
4087 minor != MINOR(dev))
4090 flush_workqueue(md_misc_wq);
4092 err = mddev_lock(mddev);
4095 if (mddev->persistent) {
4096 rdev = md_import_device(dev, mddev->major_version,
4097 mddev->minor_version);
4098 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4099 struct md_rdev *rdev0
4100 = list_entry(mddev->disks.next,
4101 struct md_rdev, same_set);
4102 err = super_types[mddev->major_version]
4103 .load_super(rdev, rdev0, mddev->minor_version);
4107 } else if (mddev->external)
4108 rdev = md_import_device(dev, -2, -1);
4110 rdev = md_import_device(dev, -1, -1);
4113 mddev_unlock(mddev);
4114 return PTR_ERR(rdev);
4116 err = bind_rdev_to_array(rdev, mddev);
4120 mddev_unlock(mddev);
4121 return err ? err : len;
4124 static struct md_sysfs_entry md_new_device =
4125 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4128 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4131 unsigned long chunk, end_chunk;
4134 err = mddev_lock(mddev);
4139 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4141 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4142 if (buf == end) break;
4143 if (*end == '-') { /* range */
4145 end_chunk = simple_strtoul(buf, &end, 0);
4146 if (buf == end) break;
4148 if (*end && !isspace(*end)) break;
4149 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4150 buf = skip_spaces(end);
4152 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4154 mddev_unlock(mddev);
4158 static struct md_sysfs_entry md_bitmap =
4159 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4162 size_show(struct mddev *mddev, char *page)
4164 return sprintf(page, "%llu\n",
4165 (unsigned long long)mddev->dev_sectors / 2);
4168 static int update_size(struct mddev *mddev, sector_t num_sectors);
4171 size_store(struct mddev *mddev, const char *buf, size_t len)
4173 /* If array is inactive, we can reduce the component size, but
4174 * not increase it (except from 0).
4175 * If array is active, we can try an on-line resize
4178 int err = strict_blocks_to_sectors(buf, §ors);
4182 err = mddev_lock(mddev);
4186 err = update_size(mddev, sectors);
4187 md_update_sb(mddev, 1);
4189 if (mddev->dev_sectors == 0 ||
4190 mddev->dev_sectors > sectors)
4191 mddev->dev_sectors = sectors;
4195 mddev_unlock(mddev);
4196 return err ? err : len;
4199 static struct md_sysfs_entry md_size =
4200 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4202 /* Metadata version.
4204 * 'none' for arrays with no metadata (good luck...)
4205 * 'external' for arrays with externally managed metadata,
4206 * or N.M for internally known formats
4209 metadata_show(struct mddev *mddev, char *page)
4211 if (mddev->persistent)
4212 return sprintf(page, "%d.%d\n",
4213 mddev->major_version, mddev->minor_version);
4214 else if (mddev->external)
4215 return sprintf(page, "external:%s\n", mddev->metadata_type);
4217 return sprintf(page, "none\n");
4221 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4226 /* Changing the details of 'external' metadata is
4227 * always permitted. Otherwise there must be
4228 * no devices attached to the array.
4231 err = mddev_lock(mddev);
4235 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4237 else if (!list_empty(&mddev->disks))
4241 if (cmd_match(buf, "none")) {
4242 mddev->persistent = 0;
4243 mddev->external = 0;
4244 mddev->major_version = 0;
4245 mddev->minor_version = 90;
4248 if (strncmp(buf, "external:", 9) == 0) {
4249 size_t namelen = len-9;
4250 if (namelen >= sizeof(mddev->metadata_type))
4251 namelen = sizeof(mddev->metadata_type)-1;
4252 strncpy(mddev->metadata_type, buf+9, namelen);
4253 mddev->metadata_type[namelen] = 0;
4254 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4255 mddev->metadata_type[--namelen] = 0;
4256 mddev->persistent = 0;
4257 mddev->external = 1;
4258 mddev->major_version = 0;
4259 mddev->minor_version = 90;
4262 major = simple_strtoul(buf, &e, 10);
4264 if (e==buf || *e != '.')
4267 minor = simple_strtoul(buf, &e, 10);
4268 if (e==buf || (*e && *e != '\n') )
4271 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4273 mddev->major_version = major;
4274 mddev->minor_version = minor;
4275 mddev->persistent = 1;
4276 mddev->external = 0;
4279 mddev_unlock(mddev);
4283 static struct md_sysfs_entry md_metadata =
4284 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4287 action_show(struct mddev *mddev, char *page)
4289 char *type = "idle";
4290 unsigned long recovery = mddev->recovery;
4291 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4293 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4294 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4295 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4297 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4298 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4300 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4304 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4306 else if (mddev->reshape_position != MaxSector)
4309 return sprintf(page, "%s\n", type);
4313 action_store(struct mddev *mddev, const char *page, size_t len)
4315 if (!mddev->pers || !mddev->pers->sync_request)
4319 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4320 if (cmd_match(page, "frozen"))
4321 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4323 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4324 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4325 mddev_lock(mddev) == 0) {
4326 flush_workqueue(md_misc_wq);
4327 if (mddev->sync_thread) {
4328 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4329 md_reap_sync_thread(mddev);
4331 mddev_unlock(mddev);
4333 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4335 else if (cmd_match(page, "resync"))
4336 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4337 else if (cmd_match(page, "recover")) {
4338 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4339 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4340 } else if (cmd_match(page, "reshape")) {
4342 if (mddev->pers->start_reshape == NULL)
4344 err = mddev_lock(mddev);
4346 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4349 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4350 err = mddev->pers->start_reshape(mddev);
4352 mddev_unlock(mddev);
4356 sysfs_notify(&mddev->kobj, NULL, "degraded");
4358 if (cmd_match(page, "check"))
4359 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4360 else if (!cmd_match(page, "repair"))
4362 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4363 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4364 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4366 if (mddev->ro == 2) {
4367 /* A write to sync_action is enough to justify
4368 * canceling read-auto mode
4371 md_wakeup_thread(mddev->sync_thread);
4373 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4374 md_wakeup_thread(mddev->thread);
4375 sysfs_notify_dirent_safe(mddev->sysfs_action);
4379 static struct md_sysfs_entry md_scan_mode =
4380 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4383 last_sync_action_show(struct mddev *mddev, char *page)
4385 return sprintf(page, "%s\n", mddev->last_sync_action);
4388 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4391 mismatch_cnt_show(struct mddev *mddev, char *page)
4393 return sprintf(page, "%llu\n",
4394 (unsigned long long)
4395 atomic64_read(&mddev->resync_mismatches));
4398 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4401 sync_min_show(struct mddev *mddev, char *page)
4403 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4404 mddev->sync_speed_min ? "local": "system");
4408 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4413 if (strncmp(buf, "system", 6)==0) {
4416 rv = kstrtouint(buf, 10, &min);
4422 mddev->sync_speed_min = min;
4426 static struct md_sysfs_entry md_sync_min =
4427 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4430 sync_max_show(struct mddev *mddev, char *page)
4432 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4433 mddev->sync_speed_max ? "local": "system");
4437 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4442 if (strncmp(buf, "system", 6)==0) {
4445 rv = kstrtouint(buf, 10, &max);
4451 mddev->sync_speed_max = max;
4455 static struct md_sysfs_entry md_sync_max =
4456 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4459 degraded_show(struct mddev *mddev, char *page)
4461 return sprintf(page, "%d\n", mddev->degraded);
4463 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4466 sync_force_parallel_show(struct mddev *mddev, char *page)
4468 return sprintf(page, "%d\n", mddev->parallel_resync);
4472 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4476 if (kstrtol(buf, 10, &n))
4479 if (n != 0 && n != 1)
4482 mddev->parallel_resync = n;
4484 if (mddev->sync_thread)
4485 wake_up(&resync_wait);
4490 /* force parallel resync, even with shared block devices */
4491 static struct md_sysfs_entry md_sync_force_parallel =
4492 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4493 sync_force_parallel_show, sync_force_parallel_store);
4496 sync_speed_show(struct mddev *mddev, char *page)
4498 unsigned long resync, dt, db;
4499 if (mddev->curr_resync == 0)
4500 return sprintf(page, "none\n");
4501 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4502 dt = (jiffies - mddev->resync_mark) / HZ;
4504 db = resync - mddev->resync_mark_cnt;
4505 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4508 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4511 sync_completed_show(struct mddev *mddev, char *page)
4513 unsigned long long max_sectors, resync;
4515 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4516 return sprintf(page, "none\n");
4518 if (mddev->curr_resync == 1 ||
4519 mddev->curr_resync == 2)
4520 return sprintf(page, "delayed\n");
4522 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4523 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4524 max_sectors = mddev->resync_max_sectors;
4526 max_sectors = mddev->dev_sectors;
4528 resync = mddev->curr_resync_completed;
4529 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4532 static struct md_sysfs_entry md_sync_completed =
4533 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4536 min_sync_show(struct mddev *mddev, char *page)
4538 return sprintf(page, "%llu\n",
4539 (unsigned long long)mddev->resync_min);
4542 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4544 unsigned long long min;
4547 if (kstrtoull(buf, 10, &min))
4550 spin_lock(&mddev->lock);
4552 if (min > mddev->resync_max)
4556 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4559 /* Round down to multiple of 4K for safety */
4560 mddev->resync_min = round_down(min, 8);
4564 spin_unlock(&mddev->lock);
4568 static struct md_sysfs_entry md_min_sync =
4569 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4572 max_sync_show(struct mddev *mddev, char *page)
4574 if (mddev->resync_max == MaxSector)
4575 return sprintf(page, "max\n");
4577 return sprintf(page, "%llu\n",
4578 (unsigned long long)mddev->resync_max);
4581 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4584 spin_lock(&mddev->lock);
4585 if (strncmp(buf, "max", 3) == 0)
4586 mddev->resync_max = MaxSector;
4588 unsigned long long max;
4592 if (kstrtoull(buf, 10, &max))
4594 if (max < mddev->resync_min)
4598 if (max < mddev->resync_max &&
4600 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4603 /* Must be a multiple of chunk_size */
4604 chunk = mddev->chunk_sectors;
4606 sector_t temp = max;
4609 if (sector_div(temp, chunk))
4612 mddev->resync_max = max;
4614 wake_up(&mddev->recovery_wait);
4617 spin_unlock(&mddev->lock);
4621 static struct md_sysfs_entry md_max_sync =
4622 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4625 suspend_lo_show(struct mddev *mddev, char *page)
4627 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4631 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4633 unsigned long long old, new;
4636 err = kstrtoull(buf, 10, &new);
4639 if (new != (sector_t)new)
4642 err = mddev_lock(mddev);
4646 if (mddev->pers == NULL ||
4647 mddev->pers->quiesce == NULL)
4649 old = mddev->suspend_lo;
4650 mddev->suspend_lo = new;
4652 /* Shrinking suspended region */
4653 mddev->pers->quiesce(mddev, 2);
4655 /* Expanding suspended region - need to wait */
4656 mddev->pers->quiesce(mddev, 1);
4657 mddev->pers->quiesce(mddev, 0);
4661 mddev_unlock(mddev);
4664 static struct md_sysfs_entry md_suspend_lo =
4665 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4668 suspend_hi_show(struct mddev *mddev, char *page)
4670 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4674 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4676 unsigned long long old, new;
4679 err = kstrtoull(buf, 10, &new);
4682 if (new != (sector_t)new)
4685 err = mddev_lock(mddev);
4689 if (mddev->pers == NULL ||
4690 mddev->pers->quiesce == NULL)
4692 old = mddev->suspend_hi;
4693 mddev->suspend_hi = new;
4695 /* Shrinking suspended region */
4696 mddev->pers->quiesce(mddev, 2);
4698 /* Expanding suspended region - need to wait */
4699 mddev->pers->quiesce(mddev, 1);
4700 mddev->pers->quiesce(mddev, 0);
4704 mddev_unlock(mddev);
4707 static struct md_sysfs_entry md_suspend_hi =
4708 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4711 reshape_position_show(struct mddev *mddev, char *page)
4713 if (mddev->reshape_position != MaxSector)
4714 return sprintf(page, "%llu\n",
4715 (unsigned long long)mddev->reshape_position);
4716 strcpy(page, "none\n");
4721 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4723 struct md_rdev *rdev;
4724 unsigned long long new;
4727 err = kstrtoull(buf, 10, &new);
4730 if (new != (sector_t)new)
4732 err = mddev_lock(mddev);
4738 mddev->reshape_position = new;
4739 mddev->delta_disks = 0;
4740 mddev->reshape_backwards = 0;
4741 mddev->new_level = mddev->level;
4742 mddev->new_layout = mddev->layout;
4743 mddev->new_chunk_sectors = mddev->chunk_sectors;
4744 rdev_for_each(rdev, mddev)
4745 rdev->new_data_offset = rdev->data_offset;
4748 mddev_unlock(mddev);
4752 static struct md_sysfs_entry md_reshape_position =
4753 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4754 reshape_position_store);
4757 reshape_direction_show(struct mddev *mddev, char *page)
4759 return sprintf(page, "%s\n",
4760 mddev->reshape_backwards ? "backwards" : "forwards");
4764 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4769 if (cmd_match(buf, "forwards"))
4771 else if (cmd_match(buf, "backwards"))
4775 if (mddev->reshape_backwards == backwards)
4778 err = mddev_lock(mddev);
4781 /* check if we are allowed to change */
4782 if (mddev->delta_disks)
4784 else if (mddev->persistent &&
4785 mddev->major_version == 0)
4788 mddev->reshape_backwards = backwards;
4789 mddev_unlock(mddev);
4793 static struct md_sysfs_entry md_reshape_direction =
4794 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4795 reshape_direction_store);
4798 array_size_show(struct mddev *mddev, char *page)
4800 if (mddev->external_size)
4801 return sprintf(page, "%llu\n",
4802 (unsigned long long)mddev->array_sectors/2);
4804 return sprintf(page, "default\n");
4808 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4813 err = mddev_lock(mddev);
4817 if (strncmp(buf, "default", 7) == 0) {
4819 sectors = mddev->pers->size(mddev, 0, 0);
4821 sectors = mddev->array_sectors;
4823 mddev->external_size = 0;
4825 if (strict_blocks_to_sectors(buf, §ors) < 0)
4827 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4830 mddev->external_size = 1;
4834 mddev->array_sectors = sectors;
4836 set_capacity(mddev->gendisk, mddev->array_sectors);
4837 revalidate_disk(mddev->gendisk);
4840 mddev_unlock(mddev);
4844 static struct md_sysfs_entry md_array_size =
4845 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4848 static struct attribute *md_default_attrs[] = {
4851 &md_raid_disks.attr,
4852 &md_chunk_size.attr,
4854 &md_resync_start.attr,
4856 &md_new_device.attr,
4857 &md_safe_delay.attr,
4858 &md_array_state.attr,
4859 &md_reshape_position.attr,
4860 &md_reshape_direction.attr,
4861 &md_array_size.attr,
4862 &max_corr_read_errors.attr,
4866 static struct attribute *md_redundancy_attrs[] = {
4868 &md_last_scan_mode.attr,
4869 &md_mismatches.attr,
4872 &md_sync_speed.attr,
4873 &md_sync_force_parallel.attr,
4874 &md_sync_completed.attr,
4877 &md_suspend_lo.attr,
4878 &md_suspend_hi.attr,
4883 static struct attribute_group md_redundancy_group = {
4885 .attrs = md_redundancy_attrs,
4889 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4891 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4892 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4897 spin_lock(&all_mddevs_lock);
4898 if (list_empty(&mddev->all_mddevs)) {
4899 spin_unlock(&all_mddevs_lock);
4903 spin_unlock(&all_mddevs_lock);
4905 rv = entry->show(mddev, page);
4911 md_attr_store(struct kobject *kobj, struct attribute *attr,
4912 const char *page, size_t length)
4914 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4915 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4920 if (!capable(CAP_SYS_ADMIN))
4922 spin_lock(&all_mddevs_lock);
4923 if (list_empty(&mddev->all_mddevs)) {
4924 spin_unlock(&all_mddevs_lock);
4928 spin_unlock(&all_mddevs_lock);
4929 rv = entry->store(mddev, page, length);
4934 static void md_free(struct kobject *ko)
4936 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4938 if (mddev->sysfs_state)
4939 sysfs_put(mddev->sysfs_state);
4942 blk_cleanup_queue(mddev->queue);
4943 if (mddev->gendisk) {
4944 del_gendisk(mddev->gendisk);
4945 put_disk(mddev->gendisk);
4951 static const struct sysfs_ops md_sysfs_ops = {
4952 .show = md_attr_show,
4953 .store = md_attr_store,
4955 static struct kobj_type md_ktype = {
4957 .sysfs_ops = &md_sysfs_ops,
4958 .default_attrs = md_default_attrs,
4963 static void mddev_delayed_delete(struct work_struct *ws)
4965 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4967 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4968 kobject_del(&mddev->kobj);
4969 kobject_put(&mddev->kobj);
4972 static int md_alloc(dev_t dev, char *name)
4974 static DEFINE_MUTEX(disks_mutex);
4975 struct mddev *mddev = mddev_find(dev);
4976 struct gendisk *disk;
4985 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4986 shift = partitioned ? MdpMinorShift : 0;
4987 unit = MINOR(mddev->unit) >> shift;
4989 /* wait for any previous instance of this device to be
4990 * completely removed (mddev_delayed_delete).
4992 flush_workqueue(md_misc_wq);
4994 mutex_lock(&disks_mutex);
5000 /* Need to ensure that 'name' is not a duplicate.
5002 struct mddev *mddev2;
5003 spin_lock(&all_mddevs_lock);
5005 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5006 if (mddev2->gendisk &&
5007 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5008 spin_unlock(&all_mddevs_lock);
5011 spin_unlock(&all_mddevs_lock);
5015 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5018 mddev->queue->queuedata = mddev;
5020 blk_queue_make_request(mddev->queue, md_make_request);
5021 blk_set_stacking_limits(&mddev->queue->limits);
5023 disk = alloc_disk(1 << shift);
5025 blk_cleanup_queue(mddev->queue);
5026 mddev->queue = NULL;
5029 disk->major = MAJOR(mddev->unit);
5030 disk->first_minor = unit << shift;
5032 strcpy(disk->disk_name, name);
5033 else if (partitioned)
5034 sprintf(disk->disk_name, "md_d%d", unit);
5036 sprintf(disk->disk_name, "md%d", unit);
5037 disk->fops = &md_fops;
5038 disk->private_data = mddev;
5039 disk->queue = mddev->queue;
5040 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5041 /* Allow extended partitions. This makes the
5042 * 'mdp' device redundant, but we can't really
5045 disk->flags |= GENHD_FL_EXT_DEVT;
5046 mddev->gendisk = disk;
5047 /* As soon as we call add_disk(), another thread could get
5048 * through to md_open, so make sure it doesn't get too far
5050 mutex_lock(&mddev->open_mutex);
5053 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5054 &disk_to_dev(disk)->kobj, "%s", "md");
5056 /* This isn't possible, but as kobject_init_and_add is marked
5057 * __must_check, we must do something with the result
5059 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5063 if (mddev->kobj.sd &&
5064 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5065 printk(KERN_DEBUG "pointless warning\n");
5066 mutex_unlock(&mddev->open_mutex);
5068 mutex_unlock(&disks_mutex);
5069 if (!error && mddev->kobj.sd) {
5070 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5071 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5077 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5079 md_alloc(dev, NULL);
5083 static int add_named_array(const char *val, struct kernel_param *kp)
5085 /* val must be "md_*" where * is not all digits.
5086 * We allocate an array with a large free minor number, and
5087 * set the name to val. val must not already be an active name.
5089 int len = strlen(val);
5090 char buf[DISK_NAME_LEN];
5092 while (len && val[len-1] == '\n')
5094 if (len >= DISK_NAME_LEN)
5096 strlcpy(buf, val, len+1);
5097 if (strncmp(buf, "md_", 3) != 0)
5099 return md_alloc(0, buf);
5102 static void md_safemode_timeout(unsigned long data)
5104 struct mddev *mddev = (struct mddev *) data;
5106 if (!atomic_read(&mddev->writes_pending)) {
5107 mddev->safemode = 1;
5108 if (mddev->external)
5109 sysfs_notify_dirent_safe(mddev->sysfs_state);
5111 md_wakeup_thread(mddev->thread);
5114 static int start_dirty_degraded;
5116 int md_run(struct mddev *mddev)
5119 struct md_rdev *rdev;
5120 struct md_personality *pers;
5122 if (list_empty(&mddev->disks))
5123 /* cannot run an array with no devices.. */
5128 /* Cannot run until previous stop completes properly */
5129 if (mddev->sysfs_active)
5133 * Analyze all RAID superblock(s)
5135 if (!mddev->raid_disks) {
5136 if (!mddev->persistent)
5141 if (mddev->level != LEVEL_NONE)
5142 request_module("md-level-%d", mddev->level);
5143 else if (mddev->clevel[0])
5144 request_module("md-%s", mddev->clevel);
5147 * Drop all container device buffers, from now on
5148 * the only valid external interface is through the md
5151 rdev_for_each(rdev, mddev) {
5152 if (test_bit(Faulty, &rdev->flags))
5154 sync_blockdev(rdev->bdev);
5155 invalidate_bdev(rdev->bdev);
5157 /* perform some consistency tests on the device.
5158 * We don't want the data to overlap the metadata,
5159 * Internal Bitmap issues have been handled elsewhere.
5161 if (rdev->meta_bdev) {
5162 /* Nothing to check */;
5163 } else if (rdev->data_offset < rdev->sb_start) {
5164 if (mddev->dev_sectors &&
5165 rdev->data_offset + mddev->dev_sectors
5167 printk("md: %s: data overlaps metadata\n",
5172 if (rdev->sb_start + rdev->sb_size/512
5173 > rdev->data_offset) {
5174 printk("md: %s: metadata overlaps data\n",
5179 sysfs_notify_dirent_safe(rdev->sysfs_state);
5182 if (mddev->bio_set == NULL)
5183 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5185 spin_lock(&pers_lock);
5186 pers = find_pers(mddev->level, mddev->clevel);
5187 if (!pers || !try_module_get(pers->owner)) {
5188 spin_unlock(&pers_lock);
5189 if (mddev->level != LEVEL_NONE)
5190 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5193 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5197 spin_unlock(&pers_lock);
5198 if (mddev->level != pers->level) {
5199 mddev->level = pers->level;
5200 mddev->new_level = pers->level;
5202 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5204 if (mddev->reshape_position != MaxSector &&
5205 pers->start_reshape == NULL) {
5206 /* This personality cannot handle reshaping... */
5207 module_put(pers->owner);
5211 if (pers->sync_request) {
5212 /* Warn if this is a potentially silly
5215 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5216 struct md_rdev *rdev2;
5219 rdev_for_each(rdev, mddev)
5220 rdev_for_each(rdev2, mddev) {
5222 rdev->bdev->bd_contains ==
5223 rdev2->bdev->bd_contains) {
5225 "%s: WARNING: %s appears to be"
5226 " on the same physical disk as"
5229 bdevname(rdev->bdev,b),
5230 bdevname(rdev2->bdev,b2));
5237 "True protection against single-disk"
5238 " failure might be compromised.\n");
5241 mddev->recovery = 0;
5242 /* may be over-ridden by personality */
5243 mddev->resync_max_sectors = mddev->dev_sectors;
5245 mddev->ok_start_degraded = start_dirty_degraded;
5247 if (start_readonly && mddev->ro == 0)
5248 mddev->ro = 2; /* read-only, but switch on first write */
5250 err = pers->run(mddev);
5252 printk(KERN_ERR "md: pers->run() failed ...\n");
5253 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5254 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5255 " but 'external_size' not in effect?\n", __func__);
5257 "md: invalid array_size %llu > default size %llu\n",
5258 (unsigned long long)mddev->array_sectors / 2,
5259 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5262 if (err == 0 && pers->sync_request &&
5263 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5264 struct bitmap *bitmap;
5266 bitmap = bitmap_create(mddev, -1);
5267 if (IS_ERR(bitmap)) {
5268 err = PTR_ERR(bitmap);
5269 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5270 mdname(mddev), err);
5272 mddev->bitmap = bitmap;
5276 mddev_detach(mddev);
5278 pers->free(mddev, mddev->private);
5279 mddev->private = NULL;
5280 module_put(pers->owner);
5281 bitmap_destroy(mddev);
5285 mddev->queue->backing_dev_info.congested_data = mddev;
5286 mddev->queue->backing_dev_info.congested_fn = md_congested;
5288 if (pers->sync_request) {
5289 if (mddev->kobj.sd &&
5290 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5292 "md: cannot register extra attributes for %s\n",
5294 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5295 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5298 atomic_set(&mddev->writes_pending,0);
5299 atomic_set(&mddev->max_corr_read_errors,
5300 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5301 mddev->safemode = 0;
5302 if (mddev_is_clustered(mddev))
5303 mddev->safemode_delay = 0;
5305 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5308 spin_lock(&mddev->lock);
5311 spin_unlock(&mddev->lock);
5312 rdev_for_each(rdev, mddev)
5313 if (rdev->raid_disk >= 0)
5314 if (sysfs_link_rdev(mddev, rdev))
5315 /* failure here is OK */;
5317 if (mddev->degraded && !mddev->ro)
5318 /* This ensures that recovering status is reported immediately
5319 * via sysfs - until a lack of spares is confirmed.
5321 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5322 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5324 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5325 md_update_sb(mddev, 0);
5327 md_new_event(mddev);
5328 sysfs_notify_dirent_safe(mddev->sysfs_state);
5329 sysfs_notify_dirent_safe(mddev->sysfs_action);
5330 sysfs_notify(&mddev->kobj, NULL, "degraded");
5333 EXPORT_SYMBOL_GPL(md_run);
5335 static int do_md_run(struct mddev *mddev)
5339 err = md_run(mddev);
5342 err = bitmap_load(mddev);
5344 bitmap_destroy(mddev);
5348 if (mddev_is_clustered(mddev))
5349 md_allow_write(mddev);
5351 md_wakeup_thread(mddev->thread);
5352 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5354 set_capacity(mddev->gendisk, mddev->array_sectors);
5355 revalidate_disk(mddev->gendisk);
5357 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5362 static int restart_array(struct mddev *mddev)
5364 struct gendisk *disk = mddev->gendisk;
5366 /* Complain if it has no devices */
5367 if (list_empty(&mddev->disks))
5373 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5374 struct md_rdev *rdev;
5375 bool has_journal = false;
5378 rdev_for_each_rcu(rdev, mddev) {
5379 if (test_bit(Journal, &rdev->flags) &&
5380 !test_bit(Faulty, &rdev->flags)) {
5387 /* Don't restart rw with journal missing/faulty */
5392 mddev->safemode = 0;
5394 set_disk_ro(disk, 0);
5395 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5397 /* Kick recovery or resync if necessary */
5398 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5399 md_wakeup_thread(mddev->thread);
5400 md_wakeup_thread(mddev->sync_thread);
5401 sysfs_notify_dirent_safe(mddev->sysfs_state);
5405 static void md_clean(struct mddev *mddev)
5407 mddev->array_sectors = 0;
5408 mddev->external_size = 0;
5409 mddev->dev_sectors = 0;
5410 mddev->raid_disks = 0;
5411 mddev->recovery_cp = 0;
5412 mddev->resync_min = 0;
5413 mddev->resync_max = MaxSector;
5414 mddev->reshape_position = MaxSector;
5415 mddev->external = 0;
5416 mddev->persistent = 0;
5417 mddev->level = LEVEL_NONE;
5418 mddev->clevel[0] = 0;
5421 mddev->metadata_type[0] = 0;
5422 mddev->chunk_sectors = 0;
5423 mddev->ctime = mddev->utime = 0;
5425 mddev->max_disks = 0;
5427 mddev->can_decrease_events = 0;
5428 mddev->delta_disks = 0;
5429 mddev->reshape_backwards = 0;
5430 mddev->new_level = LEVEL_NONE;
5431 mddev->new_layout = 0;
5432 mddev->new_chunk_sectors = 0;
5433 mddev->curr_resync = 0;
5434 atomic64_set(&mddev->resync_mismatches, 0);
5435 mddev->suspend_lo = mddev->suspend_hi = 0;
5436 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5437 mddev->recovery = 0;
5440 mddev->degraded = 0;
5441 mddev->safemode = 0;
5442 mddev->private = NULL;
5443 mddev->bitmap_info.offset = 0;
5444 mddev->bitmap_info.default_offset = 0;
5445 mddev->bitmap_info.default_space = 0;
5446 mddev->bitmap_info.chunksize = 0;
5447 mddev->bitmap_info.daemon_sleep = 0;
5448 mddev->bitmap_info.max_write_behind = 0;
5451 static void __md_stop_writes(struct mddev *mddev)
5453 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5454 flush_workqueue(md_misc_wq);
5455 if (mddev->sync_thread) {
5456 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5457 md_reap_sync_thread(mddev);
5460 del_timer_sync(&mddev->safemode_timer);
5462 bitmap_flush(mddev);
5463 md_super_wait(mddev);
5465 if (mddev->ro == 0 &&
5466 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5467 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5468 /* mark array as shutdown cleanly */
5469 if (!mddev_is_clustered(mddev))
5471 md_update_sb(mddev, 1);
5475 void md_stop_writes(struct mddev *mddev)
5477 mddev_lock_nointr(mddev);
5478 __md_stop_writes(mddev);
5479 mddev_unlock(mddev);
5481 EXPORT_SYMBOL_GPL(md_stop_writes);
5483 static void mddev_detach(struct mddev *mddev)
5485 struct bitmap *bitmap = mddev->bitmap;
5486 /* wait for behind writes to complete */
5487 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5488 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5490 /* need to kick something here to make sure I/O goes? */
5491 wait_event(bitmap->behind_wait,
5492 atomic_read(&bitmap->behind_writes) == 0);
5494 if (mddev->pers && mddev->pers->quiesce) {
5495 mddev->pers->quiesce(mddev, 1);
5496 mddev->pers->quiesce(mddev, 0);
5498 md_unregister_thread(&mddev->thread);
5500 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5503 static void __md_stop(struct mddev *mddev)
5505 struct md_personality *pers = mddev->pers;
5506 mddev_detach(mddev);
5507 /* Ensure ->event_work is done */
5508 flush_workqueue(md_misc_wq);
5509 spin_lock(&mddev->lock);
5512 spin_unlock(&mddev->lock);
5513 pers->free(mddev, mddev->private);
5514 mddev->private = NULL;
5515 if (pers->sync_request && mddev->to_remove == NULL)
5516 mddev->to_remove = &md_redundancy_group;
5517 module_put(pers->owner);
5518 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5521 void md_stop(struct mddev *mddev)
5523 /* stop the array and free an attached data structures.
5524 * This is called from dm-raid
5527 bitmap_destroy(mddev);
5529 bioset_free(mddev->bio_set);
5532 EXPORT_SYMBOL_GPL(md_stop);
5534 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5539 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5541 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5542 md_wakeup_thread(mddev->thread);
5544 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5545 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5546 if (mddev->sync_thread)
5547 /* Thread might be blocked waiting for metadata update
5548 * which will now never happen */
5549 wake_up_process(mddev->sync_thread->tsk);
5551 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5553 mddev_unlock(mddev);
5554 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5556 wait_event(mddev->sb_wait,
5557 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5558 mddev_lock_nointr(mddev);
5560 mutex_lock(&mddev->open_mutex);
5561 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5562 mddev->sync_thread ||
5563 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5564 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5565 printk("md: %s still in use.\n",mdname(mddev));
5567 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5568 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5569 md_wakeup_thread(mddev->thread);
5575 __md_stop_writes(mddev);
5581 set_disk_ro(mddev->gendisk, 1);
5582 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5583 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5584 md_wakeup_thread(mddev->thread);
5585 sysfs_notify_dirent_safe(mddev->sysfs_state);
5589 mutex_unlock(&mddev->open_mutex);
5594 * 0 - completely stop and dis-assemble array
5595 * 2 - stop but do not disassemble array
5597 static int do_md_stop(struct mddev *mddev, int mode,
5598 struct block_device *bdev)
5600 struct gendisk *disk = mddev->gendisk;
5601 struct md_rdev *rdev;
5604 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5606 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5607 md_wakeup_thread(mddev->thread);
5609 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5610 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5611 if (mddev->sync_thread)
5612 /* Thread might be blocked waiting for metadata update
5613 * which will now never happen */
5614 wake_up_process(mddev->sync_thread->tsk);
5616 mddev_unlock(mddev);
5617 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5618 !test_bit(MD_RECOVERY_RUNNING,
5619 &mddev->recovery)));
5620 mddev_lock_nointr(mddev);
5622 mutex_lock(&mddev->open_mutex);
5623 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5624 mddev->sysfs_active ||
5625 mddev->sync_thread ||
5626 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5627 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5628 printk("md: %s still in use.\n",mdname(mddev));
5629 mutex_unlock(&mddev->open_mutex);
5631 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5632 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5633 md_wakeup_thread(mddev->thread);
5639 set_disk_ro(disk, 0);
5641 __md_stop_writes(mddev);
5643 mddev->queue->backing_dev_info.congested_fn = NULL;
5645 /* tell userspace to handle 'inactive' */
5646 sysfs_notify_dirent_safe(mddev->sysfs_state);
5648 rdev_for_each(rdev, mddev)
5649 if (rdev->raid_disk >= 0)
5650 sysfs_unlink_rdev(mddev, rdev);
5652 set_capacity(disk, 0);
5653 mutex_unlock(&mddev->open_mutex);
5655 revalidate_disk(disk);
5660 mutex_unlock(&mddev->open_mutex);
5662 * Free resources if final stop
5665 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5667 bitmap_destroy(mddev);
5668 if (mddev->bitmap_info.file) {
5669 struct file *f = mddev->bitmap_info.file;
5670 spin_lock(&mddev->lock);
5671 mddev->bitmap_info.file = NULL;
5672 spin_unlock(&mddev->lock);
5675 mddev->bitmap_info.offset = 0;
5677 export_array(mddev);
5680 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5681 if (mddev->hold_active == UNTIL_STOP)
5682 mddev->hold_active = 0;
5684 md_new_event(mddev);
5685 sysfs_notify_dirent_safe(mddev->sysfs_state);
5690 static void autorun_array(struct mddev *mddev)
5692 struct md_rdev *rdev;
5695 if (list_empty(&mddev->disks))
5698 printk(KERN_INFO "md: running: ");
5700 rdev_for_each(rdev, mddev) {
5701 char b[BDEVNAME_SIZE];
5702 printk("<%s>", bdevname(rdev->bdev,b));
5706 err = do_md_run(mddev);
5708 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5709 do_md_stop(mddev, 0, NULL);
5714 * lets try to run arrays based on all disks that have arrived
5715 * until now. (those are in pending_raid_disks)
5717 * the method: pick the first pending disk, collect all disks with
5718 * the same UUID, remove all from the pending list and put them into
5719 * the 'same_array' list. Then order this list based on superblock
5720 * update time (freshest comes first), kick out 'old' disks and
5721 * compare superblocks. If everything's fine then run it.
5723 * If "unit" is allocated, then bump its reference count
5725 static void autorun_devices(int part)
5727 struct md_rdev *rdev0, *rdev, *tmp;
5728 struct mddev *mddev;
5729 char b[BDEVNAME_SIZE];
5731 printk(KERN_INFO "md: autorun ...\n");
5732 while (!list_empty(&pending_raid_disks)) {
5735 LIST_HEAD(candidates);
5736 rdev0 = list_entry(pending_raid_disks.next,
5737 struct md_rdev, same_set);
5739 printk(KERN_INFO "md: considering %s ...\n",
5740 bdevname(rdev0->bdev,b));
5741 INIT_LIST_HEAD(&candidates);
5742 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5743 if (super_90_load(rdev, rdev0, 0) >= 0) {
5744 printk(KERN_INFO "md: adding %s ...\n",
5745 bdevname(rdev->bdev,b));
5746 list_move(&rdev->same_set, &candidates);
5749 * now we have a set of devices, with all of them having
5750 * mostly sane superblocks. It's time to allocate the
5754 dev = MKDEV(mdp_major,
5755 rdev0->preferred_minor << MdpMinorShift);
5756 unit = MINOR(dev) >> MdpMinorShift;
5758 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5761 if (rdev0->preferred_minor != unit) {
5762 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5763 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5767 md_probe(dev, NULL, NULL);
5768 mddev = mddev_find(dev);
5769 if (!mddev || !mddev->gendisk) {
5773 "md: cannot allocate memory for md drive.\n");
5776 if (mddev_lock(mddev))
5777 printk(KERN_WARNING "md: %s locked, cannot run\n",
5779 else if (mddev->raid_disks || mddev->major_version
5780 || !list_empty(&mddev->disks)) {
5782 "md: %s already running, cannot run %s\n",
5783 mdname(mddev), bdevname(rdev0->bdev,b));
5784 mddev_unlock(mddev);
5786 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5787 mddev->persistent = 1;
5788 rdev_for_each_list(rdev, tmp, &candidates) {
5789 list_del_init(&rdev->same_set);
5790 if (bind_rdev_to_array(rdev, mddev))
5793 autorun_array(mddev);
5794 mddev_unlock(mddev);
5796 /* on success, candidates will be empty, on error
5799 rdev_for_each_list(rdev, tmp, &candidates) {
5800 list_del_init(&rdev->same_set);
5805 printk(KERN_INFO "md: ... autorun DONE.\n");
5807 #endif /* !MODULE */
5809 static int get_version(void __user *arg)
5813 ver.major = MD_MAJOR_VERSION;
5814 ver.minor = MD_MINOR_VERSION;
5815 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5817 if (copy_to_user(arg, &ver, sizeof(ver)))
5823 static int get_array_info(struct mddev *mddev, void __user *arg)
5825 mdu_array_info_t info;
5826 int nr,working,insync,failed,spare;
5827 struct md_rdev *rdev;
5829 nr = working = insync = failed = spare = 0;
5831 rdev_for_each_rcu(rdev, mddev) {
5833 if (test_bit(Faulty, &rdev->flags))
5837 if (test_bit(In_sync, &rdev->flags))
5845 info.major_version = mddev->major_version;
5846 info.minor_version = mddev->minor_version;
5847 info.patch_version = MD_PATCHLEVEL_VERSION;
5848 info.ctime = mddev->ctime;
5849 info.level = mddev->level;
5850 info.size = mddev->dev_sectors / 2;
5851 if (info.size != mddev->dev_sectors / 2) /* overflow */
5854 info.raid_disks = mddev->raid_disks;
5855 info.md_minor = mddev->md_minor;
5856 info.not_persistent= !mddev->persistent;
5858 info.utime = mddev->utime;
5861 info.state = (1<<MD_SB_CLEAN);
5862 if (mddev->bitmap && mddev->bitmap_info.offset)
5863 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5864 if (mddev_is_clustered(mddev))
5865 info.state |= (1<<MD_SB_CLUSTERED);
5866 info.active_disks = insync;
5867 info.working_disks = working;
5868 info.failed_disks = failed;
5869 info.spare_disks = spare;
5871 info.layout = mddev->layout;
5872 info.chunk_size = mddev->chunk_sectors << 9;
5874 if (copy_to_user(arg, &info, sizeof(info)))
5880 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5882 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5886 file = kzalloc(sizeof(*file), GFP_NOIO);
5891 spin_lock(&mddev->lock);
5892 /* bitmap enabled */
5893 if (mddev->bitmap_info.file) {
5894 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5895 sizeof(file->pathname));
5899 memmove(file->pathname, ptr,
5900 sizeof(file->pathname)-(ptr-file->pathname));
5902 spin_unlock(&mddev->lock);
5905 copy_to_user(arg, file, sizeof(*file)))
5912 static int get_disk_info(struct mddev *mddev, void __user * arg)
5914 mdu_disk_info_t info;
5915 struct md_rdev *rdev;
5917 if (copy_from_user(&info, arg, sizeof(info)))
5921 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5923 info.major = MAJOR(rdev->bdev->bd_dev);
5924 info.minor = MINOR(rdev->bdev->bd_dev);
5925 info.raid_disk = rdev->raid_disk;
5927 if (test_bit(Faulty, &rdev->flags))
5928 info.state |= (1<<MD_DISK_FAULTY);
5929 else if (test_bit(In_sync, &rdev->flags)) {
5930 info.state |= (1<<MD_DISK_ACTIVE);
5931 info.state |= (1<<MD_DISK_SYNC);
5933 if (test_bit(Journal, &rdev->flags))
5934 info.state |= (1<<MD_DISK_JOURNAL);
5935 if (test_bit(WriteMostly, &rdev->flags))
5936 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5938 info.major = info.minor = 0;
5939 info.raid_disk = -1;
5940 info.state = (1<<MD_DISK_REMOVED);
5944 if (copy_to_user(arg, &info, sizeof(info)))
5950 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5952 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5953 struct md_rdev *rdev;
5954 dev_t dev = MKDEV(info->major,info->minor);
5956 if (mddev_is_clustered(mddev) &&
5957 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5958 pr_err("%s: Cannot add to clustered mddev.\n",
5963 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5966 if (!mddev->raid_disks) {
5968 /* expecting a device which has a superblock */
5969 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5972 "md: md_import_device returned %ld\n",
5974 return PTR_ERR(rdev);
5976 if (!list_empty(&mddev->disks)) {
5977 struct md_rdev *rdev0
5978 = list_entry(mddev->disks.next,
5979 struct md_rdev, same_set);
5980 err = super_types[mddev->major_version]
5981 .load_super(rdev, rdev0, mddev->minor_version);
5984 "md: %s has different UUID to %s\n",
5985 bdevname(rdev->bdev,b),
5986 bdevname(rdev0->bdev,b2));
5991 err = bind_rdev_to_array(rdev, mddev);
5998 * add_new_disk can be used once the array is assembled
5999 * to add "hot spares". They must already have a superblock
6004 if (!mddev->pers->hot_add_disk) {
6006 "%s: personality does not support diskops!\n",
6010 if (mddev->persistent)
6011 rdev = md_import_device(dev, mddev->major_version,
6012 mddev->minor_version);
6014 rdev = md_import_device(dev, -1, -1);
6017 "md: md_import_device returned %ld\n",
6019 return PTR_ERR(rdev);
6021 /* set saved_raid_disk if appropriate */
6022 if (!mddev->persistent) {
6023 if (info->state & (1<<MD_DISK_SYNC) &&
6024 info->raid_disk < mddev->raid_disks) {
6025 rdev->raid_disk = info->raid_disk;
6026 set_bit(In_sync, &rdev->flags);
6027 clear_bit(Bitmap_sync, &rdev->flags);
6029 rdev->raid_disk = -1;
6030 rdev->saved_raid_disk = rdev->raid_disk;
6032 super_types[mddev->major_version].
6033 validate_super(mddev, rdev);
6034 if ((info->state & (1<<MD_DISK_SYNC)) &&
6035 rdev->raid_disk != info->raid_disk) {
6036 /* This was a hot-add request, but events doesn't
6037 * match, so reject it.
6043 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6044 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6045 set_bit(WriteMostly, &rdev->flags);
6047 clear_bit(WriteMostly, &rdev->flags);
6049 if (info->state & (1<<MD_DISK_JOURNAL))
6050 set_bit(Journal, &rdev->flags);
6052 * check whether the device shows up in other nodes
6054 if (mddev_is_clustered(mddev)) {
6055 if (info->state & (1 << MD_DISK_CANDIDATE))
6056 set_bit(Candidate, &rdev->flags);
6057 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6058 /* --add initiated by this node */
6059 err = md_cluster_ops->add_new_disk(mddev, rdev);
6067 rdev->raid_disk = -1;
6068 err = bind_rdev_to_array(rdev, mddev);
6073 if (mddev_is_clustered(mddev)) {
6074 if (info->state & (1 << MD_DISK_CANDIDATE))
6075 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6078 md_cluster_ops->add_new_disk_cancel(mddev);
6080 err = add_bound_rdev(rdev);
6084 err = add_bound_rdev(rdev);
6089 /* otherwise, add_new_disk is only allowed
6090 * for major_version==0 superblocks
6092 if (mddev->major_version != 0) {
6093 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6098 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6100 rdev = md_import_device(dev, -1, 0);
6103 "md: error, md_import_device() returned %ld\n",
6105 return PTR_ERR(rdev);
6107 rdev->desc_nr = info->number;
6108 if (info->raid_disk < mddev->raid_disks)
6109 rdev->raid_disk = info->raid_disk;
6111 rdev->raid_disk = -1;
6113 if (rdev->raid_disk < mddev->raid_disks)
6114 if (info->state & (1<<MD_DISK_SYNC))
6115 set_bit(In_sync, &rdev->flags);
6117 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6118 set_bit(WriteMostly, &rdev->flags);
6120 if (!mddev->persistent) {
6121 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6122 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6124 rdev->sb_start = calc_dev_sboffset(rdev);
6125 rdev->sectors = rdev->sb_start;
6127 err = bind_rdev_to_array(rdev, mddev);
6137 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6139 char b[BDEVNAME_SIZE];
6140 struct md_rdev *rdev;
6143 rdev = find_rdev(mddev, dev);
6147 if (mddev_is_clustered(mddev))
6148 ret = md_cluster_ops->metadata_update_start(mddev);
6150 if (rdev->raid_disk < 0)
6153 clear_bit(Blocked, &rdev->flags);
6154 remove_and_add_spares(mddev, rdev);
6156 if (rdev->raid_disk >= 0)
6160 if (mddev_is_clustered(mddev) && ret == 0)
6161 md_cluster_ops->remove_disk(mddev, rdev);
6163 md_kick_rdev_from_array(rdev);
6164 md_update_sb(mddev, 1);
6165 md_new_event(mddev);
6169 if (mddev_is_clustered(mddev) && ret == 0)
6170 md_cluster_ops->metadata_update_cancel(mddev);
6172 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6173 bdevname(rdev->bdev,b), mdname(mddev));
6177 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6179 char b[BDEVNAME_SIZE];
6181 struct md_rdev *rdev;
6186 if (mddev->major_version != 0) {
6187 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6188 " version-0 superblocks.\n",
6192 if (!mddev->pers->hot_add_disk) {
6194 "%s: personality does not support diskops!\n",
6199 rdev = md_import_device(dev, -1, 0);
6202 "md: error, md_import_device() returned %ld\n",
6207 if (mddev->persistent)
6208 rdev->sb_start = calc_dev_sboffset(rdev);
6210 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6212 rdev->sectors = rdev->sb_start;
6214 if (test_bit(Faulty, &rdev->flags)) {
6216 "md: can not hot-add faulty %s disk to %s!\n",
6217 bdevname(rdev->bdev,b), mdname(mddev));
6222 clear_bit(In_sync, &rdev->flags);
6224 rdev->saved_raid_disk = -1;
6225 err = bind_rdev_to_array(rdev, mddev);
6230 * The rest should better be atomic, we can have disk failures
6231 * noticed in interrupt contexts ...
6234 rdev->raid_disk = -1;
6236 md_update_sb(mddev, 1);
6238 * Kick recovery, maybe this spare has to be added to the
6239 * array immediately.
6241 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6242 md_wakeup_thread(mddev->thread);
6243 md_new_event(mddev);
6251 static int set_bitmap_file(struct mddev *mddev, int fd)
6256 if (!mddev->pers->quiesce || !mddev->thread)
6258 if (mddev->recovery || mddev->sync_thread)
6260 /* we should be able to change the bitmap.. */
6264 struct inode *inode;
6267 if (mddev->bitmap || mddev->bitmap_info.file)
6268 return -EEXIST; /* cannot add when bitmap is present */
6272 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6277 inode = f->f_mapping->host;
6278 if (!S_ISREG(inode->i_mode)) {
6279 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6282 } else if (!(f->f_mode & FMODE_WRITE)) {
6283 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6286 } else if (atomic_read(&inode->i_writecount) != 1) {
6287 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6295 mddev->bitmap_info.file = f;
6296 mddev->bitmap_info.offset = 0; /* file overrides offset */
6297 } else if (mddev->bitmap == NULL)
6298 return -ENOENT; /* cannot remove what isn't there */
6301 mddev->pers->quiesce(mddev, 1);
6303 struct bitmap *bitmap;
6305 bitmap = bitmap_create(mddev, -1);
6306 if (!IS_ERR(bitmap)) {
6307 mddev->bitmap = bitmap;
6308 err = bitmap_load(mddev);
6310 err = PTR_ERR(bitmap);
6312 if (fd < 0 || err) {
6313 bitmap_destroy(mddev);
6314 fd = -1; /* make sure to put the file */
6316 mddev->pers->quiesce(mddev, 0);
6319 struct file *f = mddev->bitmap_info.file;
6321 spin_lock(&mddev->lock);
6322 mddev->bitmap_info.file = NULL;
6323 spin_unlock(&mddev->lock);
6332 * set_array_info is used two different ways
6333 * The original usage is when creating a new array.
6334 * In this usage, raid_disks is > 0 and it together with
6335 * level, size, not_persistent,layout,chunksize determine the
6336 * shape of the array.
6337 * This will always create an array with a type-0.90.0 superblock.
6338 * The newer usage is when assembling an array.
6339 * In this case raid_disks will be 0, and the major_version field is
6340 * use to determine which style super-blocks are to be found on the devices.
6341 * The minor and patch _version numbers are also kept incase the
6342 * super_block handler wishes to interpret them.
6344 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6347 if (info->raid_disks == 0) {
6348 /* just setting version number for superblock loading */
6349 if (info->major_version < 0 ||
6350 info->major_version >= ARRAY_SIZE(super_types) ||
6351 super_types[info->major_version].name == NULL) {
6352 /* maybe try to auto-load a module? */
6354 "md: superblock version %d not known\n",
6355 info->major_version);
6358 mddev->major_version = info->major_version;
6359 mddev->minor_version = info->minor_version;
6360 mddev->patch_version = info->patch_version;
6361 mddev->persistent = !info->not_persistent;
6362 /* ensure mddev_put doesn't delete this now that there
6363 * is some minimal configuration.
6365 mddev->ctime = get_seconds();
6368 mddev->major_version = MD_MAJOR_VERSION;
6369 mddev->minor_version = MD_MINOR_VERSION;
6370 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6371 mddev->ctime = get_seconds();
6373 mddev->level = info->level;
6374 mddev->clevel[0] = 0;
6375 mddev->dev_sectors = 2 * (sector_t)info->size;
6376 mddev->raid_disks = info->raid_disks;
6377 /* don't set md_minor, it is determined by which /dev/md* was
6380 if (info->state & (1<<MD_SB_CLEAN))
6381 mddev->recovery_cp = MaxSector;
6383 mddev->recovery_cp = 0;
6384 mddev->persistent = ! info->not_persistent;
6385 mddev->external = 0;
6387 mddev->layout = info->layout;
6388 mddev->chunk_sectors = info->chunk_size >> 9;
6390 mddev->max_disks = MD_SB_DISKS;
6392 if (mddev->persistent)
6394 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6396 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6397 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6398 mddev->bitmap_info.offset = 0;
6400 mddev->reshape_position = MaxSector;
6403 * Generate a 128 bit UUID
6405 get_random_bytes(mddev->uuid, 16);
6407 mddev->new_level = mddev->level;
6408 mddev->new_chunk_sectors = mddev->chunk_sectors;
6409 mddev->new_layout = mddev->layout;
6410 mddev->delta_disks = 0;
6411 mddev->reshape_backwards = 0;
6416 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6418 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6420 if (mddev->external_size)
6423 mddev->array_sectors = array_sectors;
6425 EXPORT_SYMBOL(md_set_array_sectors);
6427 static int update_size(struct mddev *mddev, sector_t num_sectors)
6429 struct md_rdev *rdev;
6431 int fit = (num_sectors == 0);
6433 if (mddev->pers->resize == NULL)
6435 /* The "num_sectors" is the number of sectors of each device that
6436 * is used. This can only make sense for arrays with redundancy.
6437 * linear and raid0 always use whatever space is available. We can only
6438 * consider changing this number if no resync or reconstruction is
6439 * happening, and if the new size is acceptable. It must fit before the
6440 * sb_start or, if that is <data_offset, it must fit before the size
6441 * of each device. If num_sectors is zero, we find the largest size
6444 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6450 rdev_for_each(rdev, mddev) {
6451 sector_t avail = rdev->sectors;
6453 if (fit && (num_sectors == 0 || num_sectors > avail))
6454 num_sectors = avail;
6455 if (avail < num_sectors)
6458 rv = mddev->pers->resize(mddev, num_sectors);
6460 revalidate_disk(mddev->gendisk);
6464 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6467 struct md_rdev *rdev;
6468 /* change the number of raid disks */
6469 if (mddev->pers->check_reshape == NULL)
6473 if (raid_disks <= 0 ||
6474 (mddev->max_disks && raid_disks >= mddev->max_disks))
6476 if (mddev->sync_thread ||
6477 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6478 mddev->reshape_position != MaxSector)
6481 rdev_for_each(rdev, mddev) {
6482 if (mddev->raid_disks < raid_disks &&
6483 rdev->data_offset < rdev->new_data_offset)
6485 if (mddev->raid_disks > raid_disks &&
6486 rdev->data_offset > rdev->new_data_offset)
6490 mddev->delta_disks = raid_disks - mddev->raid_disks;
6491 if (mddev->delta_disks < 0)
6492 mddev->reshape_backwards = 1;
6493 else if (mddev->delta_disks > 0)
6494 mddev->reshape_backwards = 0;
6496 rv = mddev->pers->check_reshape(mddev);
6498 mddev->delta_disks = 0;
6499 mddev->reshape_backwards = 0;
6505 * update_array_info is used to change the configuration of an
6507 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6508 * fields in the info are checked against the array.
6509 * Any differences that cannot be handled will cause an error.
6510 * Normally, only one change can be managed at a time.
6512 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6518 /* calculate expected state,ignoring low bits */
6519 if (mddev->bitmap && mddev->bitmap_info.offset)
6520 state |= (1 << MD_SB_BITMAP_PRESENT);
6522 if (mddev->major_version != info->major_version ||
6523 mddev->minor_version != info->minor_version ||
6524 /* mddev->patch_version != info->patch_version || */
6525 mddev->ctime != info->ctime ||
6526 mddev->level != info->level ||
6527 /* mddev->layout != info->layout || */
6528 mddev->persistent != !info->not_persistent ||
6529 mddev->chunk_sectors != info->chunk_size >> 9 ||
6530 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6531 ((state^info->state) & 0xfffffe00)
6534 /* Check there is only one change */
6535 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6537 if (mddev->raid_disks != info->raid_disks)
6539 if (mddev->layout != info->layout)
6541 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6548 if (mddev->layout != info->layout) {
6550 * we don't need to do anything at the md level, the
6551 * personality will take care of it all.
6553 if (mddev->pers->check_reshape == NULL)
6556 mddev->new_layout = info->layout;
6557 rv = mddev->pers->check_reshape(mddev);
6559 mddev->new_layout = mddev->layout;
6563 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6564 rv = update_size(mddev, (sector_t)info->size * 2);
6566 if (mddev->raid_disks != info->raid_disks)
6567 rv = update_raid_disks(mddev, info->raid_disks);
6569 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6570 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6574 if (mddev->recovery || mddev->sync_thread) {
6578 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6579 struct bitmap *bitmap;
6580 /* add the bitmap */
6581 if (mddev->bitmap) {
6585 if (mddev->bitmap_info.default_offset == 0) {
6589 mddev->bitmap_info.offset =
6590 mddev->bitmap_info.default_offset;
6591 mddev->bitmap_info.space =
6592 mddev->bitmap_info.default_space;
6593 mddev->pers->quiesce(mddev, 1);
6594 bitmap = bitmap_create(mddev, -1);
6595 if (!IS_ERR(bitmap)) {
6596 mddev->bitmap = bitmap;
6597 rv = bitmap_load(mddev);
6599 rv = PTR_ERR(bitmap);
6601 bitmap_destroy(mddev);
6602 mddev->pers->quiesce(mddev, 0);
6604 /* remove the bitmap */
6605 if (!mddev->bitmap) {
6609 if (mddev->bitmap->storage.file) {
6613 mddev->pers->quiesce(mddev, 1);
6614 bitmap_destroy(mddev);
6615 mddev->pers->quiesce(mddev, 0);
6616 mddev->bitmap_info.offset = 0;
6619 md_update_sb(mddev, 1);
6625 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6627 struct md_rdev *rdev;
6630 if (mddev->pers == NULL)
6634 rdev = find_rdev_rcu(mddev, dev);
6638 md_error(mddev, rdev);
6639 if (!test_bit(Faulty, &rdev->flags))
6647 * We have a problem here : there is no easy way to give a CHS
6648 * virtual geometry. We currently pretend that we have a 2 heads
6649 * 4 sectors (with a BIG number of cylinders...). This drives
6650 * dosfs just mad... ;-)
6652 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6654 struct mddev *mddev = bdev->bd_disk->private_data;
6658 geo->cylinders = mddev->array_sectors / 8;
6662 static inline bool md_ioctl_valid(unsigned int cmd)
6667 case GET_ARRAY_INFO:
6668 case GET_BITMAP_FILE:
6671 case HOT_REMOVE_DISK:
6674 case RESTART_ARRAY_RW:
6676 case SET_ARRAY_INFO:
6677 case SET_BITMAP_FILE:
6678 case SET_DISK_FAULTY:
6681 case CLUSTERED_DISK_NACK:
6688 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6689 unsigned int cmd, unsigned long arg)
6692 void __user *argp = (void __user *)arg;
6693 struct mddev *mddev = NULL;
6696 if (!md_ioctl_valid(cmd))
6701 case GET_ARRAY_INFO:
6705 if (!capable(CAP_SYS_ADMIN))
6710 * Commands dealing with the RAID driver but not any
6715 err = get_version(argp);
6721 autostart_arrays(arg);
6728 * Commands creating/starting a new array:
6731 mddev = bdev->bd_disk->private_data;
6738 /* Some actions do not requires the mutex */
6740 case GET_ARRAY_INFO:
6741 if (!mddev->raid_disks && !mddev->external)
6744 err = get_array_info(mddev, argp);
6748 if (!mddev->raid_disks && !mddev->external)
6751 err = get_disk_info(mddev, argp);
6754 case SET_DISK_FAULTY:
6755 err = set_disk_faulty(mddev, new_decode_dev(arg));
6758 case GET_BITMAP_FILE:
6759 err = get_bitmap_file(mddev, argp);
6764 if (cmd == ADD_NEW_DISK)
6765 /* need to ensure md_delayed_delete() has completed */
6766 flush_workqueue(md_misc_wq);
6768 if (cmd == HOT_REMOVE_DISK)
6769 /* need to ensure recovery thread has run */
6770 wait_event_interruptible_timeout(mddev->sb_wait,
6771 !test_bit(MD_RECOVERY_NEEDED,
6773 msecs_to_jiffies(5000));
6774 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6775 /* Need to flush page cache, and ensure no-one else opens
6778 mutex_lock(&mddev->open_mutex);
6779 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6780 mutex_unlock(&mddev->open_mutex);
6784 set_bit(MD_STILL_CLOSED, &mddev->flags);
6785 mutex_unlock(&mddev->open_mutex);
6786 sync_blockdev(bdev);
6788 err = mddev_lock(mddev);
6791 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6796 if (cmd == SET_ARRAY_INFO) {
6797 mdu_array_info_t info;
6799 memset(&info, 0, sizeof(info));
6800 else if (copy_from_user(&info, argp, sizeof(info))) {
6805 err = update_array_info(mddev, &info);
6807 printk(KERN_WARNING "md: couldn't update"
6808 " array info. %d\n", err);
6813 if (!list_empty(&mddev->disks)) {
6815 "md: array %s already has disks!\n",
6820 if (mddev->raid_disks) {
6822 "md: array %s already initialised!\n",
6827 err = set_array_info(mddev, &info);
6829 printk(KERN_WARNING "md: couldn't set"
6830 " array info. %d\n", err);
6837 * Commands querying/configuring an existing array:
6839 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6840 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6841 if ((!mddev->raid_disks && !mddev->external)
6842 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6843 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6844 && cmd != GET_BITMAP_FILE) {
6850 * Commands even a read-only array can execute:
6853 case RESTART_ARRAY_RW:
6854 err = restart_array(mddev);
6858 err = do_md_stop(mddev, 0, bdev);
6862 err = md_set_readonly(mddev, bdev);
6865 case HOT_REMOVE_DISK:
6866 err = hot_remove_disk(mddev, new_decode_dev(arg));
6870 /* We can support ADD_NEW_DISK on read-only arrays
6871 * on if we are re-adding a preexisting device.
6872 * So require mddev->pers and MD_DISK_SYNC.
6875 mdu_disk_info_t info;
6876 if (copy_from_user(&info, argp, sizeof(info)))
6878 else if (!(info.state & (1<<MD_DISK_SYNC)))
6879 /* Need to clear read-only for this */
6882 err = add_new_disk(mddev, &info);
6888 if (get_user(ro, (int __user *)(arg))) {
6894 /* if the bdev is going readonly the value of mddev->ro
6895 * does not matter, no writes are coming
6900 /* are we are already prepared for writes? */
6904 /* transitioning to readauto need only happen for
6905 * arrays that call md_write_start
6908 err = restart_array(mddev);
6911 set_disk_ro(mddev->gendisk, 0);
6918 * The remaining ioctls are changing the state of the
6919 * superblock, so we do not allow them on read-only arrays.
6921 if (mddev->ro && mddev->pers) {
6922 if (mddev->ro == 2) {
6924 sysfs_notify_dirent_safe(mddev->sysfs_state);
6925 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6926 /* mddev_unlock will wake thread */
6927 /* If a device failed while we were read-only, we
6928 * need to make sure the metadata is updated now.
6930 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6931 mddev_unlock(mddev);
6932 wait_event(mddev->sb_wait,
6933 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6934 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6935 mddev_lock_nointr(mddev);
6946 mdu_disk_info_t info;
6947 if (copy_from_user(&info, argp, sizeof(info)))
6950 err = add_new_disk(mddev, &info);
6954 case CLUSTERED_DISK_NACK:
6955 if (mddev_is_clustered(mddev))
6956 md_cluster_ops->new_disk_ack(mddev, false);
6962 err = hot_add_disk(mddev, new_decode_dev(arg));
6966 err = do_md_run(mddev);
6969 case SET_BITMAP_FILE:
6970 err = set_bitmap_file(mddev, (int)arg);
6979 if (mddev->hold_active == UNTIL_IOCTL &&
6981 mddev->hold_active = 0;
6982 mddev_unlock(mddev);
6986 #ifdef CONFIG_COMPAT
6987 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6988 unsigned int cmd, unsigned long arg)
6991 case HOT_REMOVE_DISK:
6993 case SET_DISK_FAULTY:
6994 case SET_BITMAP_FILE:
6995 /* These take in integer arg, do not convert */
6998 arg = (unsigned long)compat_ptr(arg);
7002 return md_ioctl(bdev, mode, cmd, arg);
7004 #endif /* CONFIG_COMPAT */
7006 static int md_open(struct block_device *bdev, fmode_t mode)
7009 * Succeed if we can lock the mddev, which confirms that
7010 * it isn't being stopped right now.
7012 struct mddev *mddev = mddev_find(bdev->bd_dev);
7018 if (mddev->gendisk != bdev->bd_disk) {
7019 /* we are racing with mddev_put which is discarding this
7023 /* Wait until bdev->bd_disk is definitely gone */
7024 flush_workqueue(md_misc_wq);
7025 /* Then retry the open from the top */
7026 return -ERESTARTSYS;
7028 BUG_ON(mddev != bdev->bd_disk->private_data);
7030 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7034 atomic_inc(&mddev->openers);
7035 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7036 mutex_unlock(&mddev->open_mutex);
7038 check_disk_change(bdev);
7043 static void md_release(struct gendisk *disk, fmode_t mode)
7045 struct mddev *mddev = disk->private_data;
7048 atomic_dec(&mddev->openers);
7052 static int md_media_changed(struct gendisk *disk)
7054 struct mddev *mddev = disk->private_data;
7056 return mddev->changed;
7059 static int md_revalidate(struct gendisk *disk)
7061 struct mddev *mddev = disk->private_data;
7066 static const struct block_device_operations md_fops =
7068 .owner = THIS_MODULE,
7070 .release = md_release,
7072 #ifdef CONFIG_COMPAT
7073 .compat_ioctl = md_compat_ioctl,
7075 .getgeo = md_getgeo,
7076 .media_changed = md_media_changed,
7077 .revalidate_disk= md_revalidate,
7080 static int md_thread(void *arg)
7082 struct md_thread *thread = arg;
7085 * md_thread is a 'system-thread', it's priority should be very
7086 * high. We avoid resource deadlocks individually in each
7087 * raid personality. (RAID5 does preallocation) We also use RR and
7088 * the very same RT priority as kswapd, thus we will never get
7089 * into a priority inversion deadlock.
7091 * we definitely have to have equal or higher priority than
7092 * bdflush, otherwise bdflush will deadlock if there are too
7093 * many dirty RAID5 blocks.
7096 allow_signal(SIGKILL);
7097 while (!kthread_should_stop()) {
7099 /* We need to wait INTERRUPTIBLE so that
7100 * we don't add to the load-average.
7101 * That means we need to be sure no signals are
7104 if (signal_pending(current))
7105 flush_signals(current);
7107 wait_event_interruptible_timeout
7109 test_bit(THREAD_WAKEUP, &thread->flags)
7110 || kthread_should_stop(),
7113 clear_bit(THREAD_WAKEUP, &thread->flags);
7114 if (!kthread_should_stop())
7115 thread->run(thread);
7121 void md_wakeup_thread(struct md_thread *thread)
7124 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7125 set_bit(THREAD_WAKEUP, &thread->flags);
7126 wake_up(&thread->wqueue);
7129 EXPORT_SYMBOL(md_wakeup_thread);
7131 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7132 struct mddev *mddev, const char *name)
7134 struct md_thread *thread;
7136 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7140 init_waitqueue_head(&thread->wqueue);
7143 thread->mddev = mddev;
7144 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7145 thread->tsk = kthread_run(md_thread, thread,
7147 mdname(thread->mddev),
7149 if (IS_ERR(thread->tsk)) {
7155 EXPORT_SYMBOL(md_register_thread);
7157 void md_unregister_thread(struct md_thread **threadp)
7159 struct md_thread *thread = *threadp;
7162 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7163 /* Locking ensures that mddev_unlock does not wake_up a
7164 * non-existent thread
7166 spin_lock(&pers_lock);
7168 spin_unlock(&pers_lock);
7170 kthread_stop(thread->tsk);
7173 EXPORT_SYMBOL(md_unregister_thread);
7175 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7177 if (!rdev || test_bit(Faulty, &rdev->flags))
7180 if (!mddev->pers || !mddev->pers->error_handler)
7182 mddev->pers->error_handler(mddev,rdev);
7183 if (mddev->degraded)
7184 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7185 sysfs_notify_dirent_safe(rdev->sysfs_state);
7186 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7187 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7188 md_wakeup_thread(mddev->thread);
7189 if (mddev->event_work.func)
7190 queue_work(md_misc_wq, &mddev->event_work);
7191 md_new_event_inintr(mddev);
7193 EXPORT_SYMBOL(md_error);
7195 /* seq_file implementation /proc/mdstat */
7197 static void status_unused(struct seq_file *seq)
7200 struct md_rdev *rdev;
7202 seq_printf(seq, "unused devices: ");
7204 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7205 char b[BDEVNAME_SIZE];
7207 seq_printf(seq, "%s ",
7208 bdevname(rdev->bdev,b));
7211 seq_printf(seq, "<none>");
7213 seq_printf(seq, "\n");
7216 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7218 sector_t max_sectors, resync, res;
7219 unsigned long dt, db;
7222 unsigned int per_milli;
7224 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7225 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7226 max_sectors = mddev->resync_max_sectors;
7228 max_sectors = mddev->dev_sectors;
7230 resync = mddev->curr_resync;
7232 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7233 /* Still cleaning up */
7234 resync = max_sectors;
7236 resync -= atomic_read(&mddev->recovery_active);
7239 if (mddev->recovery_cp < MaxSector) {
7240 seq_printf(seq, "\tresync=PENDING");
7246 seq_printf(seq, "\tresync=DELAYED");
7250 WARN_ON(max_sectors == 0);
7251 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7252 * in a sector_t, and (max_sectors>>scale) will fit in a
7253 * u32, as those are the requirements for sector_div.
7254 * Thus 'scale' must be at least 10
7257 if (sizeof(sector_t) > sizeof(unsigned long)) {
7258 while ( max_sectors/2 > (1ULL<<(scale+32)))
7261 res = (resync>>scale)*1000;
7262 sector_div(res, (u32)((max_sectors>>scale)+1));
7266 int i, x = per_milli/50, y = 20-x;
7267 seq_printf(seq, "[");
7268 for (i = 0; i < x; i++)
7269 seq_printf(seq, "=");
7270 seq_printf(seq, ">");
7271 for (i = 0; i < y; i++)
7272 seq_printf(seq, ".");
7273 seq_printf(seq, "] ");
7275 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7276 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7278 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7280 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7281 "resync" : "recovery"))),
7282 per_milli/10, per_milli % 10,
7283 (unsigned long long) resync/2,
7284 (unsigned long long) max_sectors/2);
7287 * dt: time from mark until now
7288 * db: blocks written from mark until now
7289 * rt: remaining time
7291 * rt is a sector_t, so could be 32bit or 64bit.
7292 * So we divide before multiply in case it is 32bit and close
7294 * We scale the divisor (db) by 32 to avoid losing precision
7295 * near the end of resync when the number of remaining sectors
7297 * We then divide rt by 32 after multiplying by db to compensate.
7298 * The '+1' avoids division by zero if db is very small.
7300 dt = ((jiffies - mddev->resync_mark) / HZ);
7302 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7303 - mddev->resync_mark_cnt;
7305 rt = max_sectors - resync; /* number of remaining sectors */
7306 sector_div(rt, db/32+1);
7310 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7311 ((unsigned long)rt % 60)/6);
7313 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7317 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7319 struct list_head *tmp;
7321 struct mddev *mddev;
7329 spin_lock(&all_mddevs_lock);
7330 list_for_each(tmp,&all_mddevs)
7332 mddev = list_entry(tmp, struct mddev, all_mddevs);
7334 spin_unlock(&all_mddevs_lock);
7337 spin_unlock(&all_mddevs_lock);
7339 return (void*)2;/* tail */
7343 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7345 struct list_head *tmp;
7346 struct mddev *next_mddev, *mddev = v;
7352 spin_lock(&all_mddevs_lock);
7354 tmp = all_mddevs.next;
7356 tmp = mddev->all_mddevs.next;
7357 if (tmp != &all_mddevs)
7358 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7360 next_mddev = (void*)2;
7363 spin_unlock(&all_mddevs_lock);
7371 static void md_seq_stop(struct seq_file *seq, void *v)
7373 struct mddev *mddev = v;
7375 if (mddev && v != (void*)1 && v != (void*)2)
7379 static int md_seq_show(struct seq_file *seq, void *v)
7381 struct mddev *mddev = v;
7383 struct md_rdev *rdev;
7385 if (v == (void*)1) {
7386 struct md_personality *pers;
7387 seq_printf(seq, "Personalities : ");
7388 spin_lock(&pers_lock);
7389 list_for_each_entry(pers, &pers_list, list)
7390 seq_printf(seq, "[%s] ", pers->name);
7392 spin_unlock(&pers_lock);
7393 seq_printf(seq, "\n");
7394 seq->poll_event = atomic_read(&md_event_count);
7397 if (v == (void*)2) {
7402 spin_lock(&mddev->lock);
7403 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7404 seq_printf(seq, "%s : %sactive", mdname(mddev),
7405 mddev->pers ? "" : "in");
7408 seq_printf(seq, " (read-only)");
7410 seq_printf(seq, " (auto-read-only)");
7411 seq_printf(seq, " %s", mddev->pers->name);
7416 rdev_for_each_rcu(rdev, mddev) {
7417 char b[BDEVNAME_SIZE];
7418 seq_printf(seq, " %s[%d]",
7419 bdevname(rdev->bdev,b), rdev->desc_nr);
7420 if (test_bit(WriteMostly, &rdev->flags))
7421 seq_printf(seq, "(W)");
7422 if (test_bit(Journal, &rdev->flags))
7423 seq_printf(seq, "(J)");
7424 if (test_bit(Faulty, &rdev->flags)) {
7425 seq_printf(seq, "(F)");
7428 if (rdev->raid_disk < 0)
7429 seq_printf(seq, "(S)"); /* spare */
7430 if (test_bit(Replacement, &rdev->flags))
7431 seq_printf(seq, "(R)");
7432 sectors += rdev->sectors;
7436 if (!list_empty(&mddev->disks)) {
7438 seq_printf(seq, "\n %llu blocks",
7439 (unsigned long long)
7440 mddev->array_sectors / 2);
7442 seq_printf(seq, "\n %llu blocks",
7443 (unsigned long long)sectors / 2);
7445 if (mddev->persistent) {
7446 if (mddev->major_version != 0 ||
7447 mddev->minor_version != 90) {
7448 seq_printf(seq," super %d.%d",
7449 mddev->major_version,
7450 mddev->minor_version);
7452 } else if (mddev->external)
7453 seq_printf(seq, " super external:%s",
7454 mddev->metadata_type);
7456 seq_printf(seq, " super non-persistent");
7459 mddev->pers->status(seq, mddev);
7460 seq_printf(seq, "\n ");
7461 if (mddev->pers->sync_request) {
7462 if (status_resync(seq, mddev))
7463 seq_printf(seq, "\n ");
7466 seq_printf(seq, "\n ");
7468 bitmap_status(seq, mddev->bitmap);
7470 seq_printf(seq, "\n");
7472 spin_unlock(&mddev->lock);
7477 static const struct seq_operations md_seq_ops = {
7478 .start = md_seq_start,
7479 .next = md_seq_next,
7480 .stop = md_seq_stop,
7481 .show = md_seq_show,
7484 static int md_seq_open(struct inode *inode, struct file *file)
7486 struct seq_file *seq;
7489 error = seq_open(file, &md_seq_ops);
7493 seq = file->private_data;
7494 seq->poll_event = atomic_read(&md_event_count);
7498 static int md_unloading;
7499 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7501 struct seq_file *seq = filp->private_data;
7505 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7506 poll_wait(filp, &md_event_waiters, wait);
7508 /* always allow read */
7509 mask = POLLIN | POLLRDNORM;
7511 if (seq->poll_event != atomic_read(&md_event_count))
7512 mask |= POLLERR | POLLPRI;
7516 static const struct file_operations md_seq_fops = {
7517 .owner = THIS_MODULE,
7518 .open = md_seq_open,
7520 .llseek = seq_lseek,
7521 .release = seq_release_private,
7522 .poll = mdstat_poll,
7525 int register_md_personality(struct md_personality *p)
7527 printk(KERN_INFO "md: %s personality registered for level %d\n",
7529 spin_lock(&pers_lock);
7530 list_add_tail(&p->list, &pers_list);
7531 spin_unlock(&pers_lock);
7534 EXPORT_SYMBOL(register_md_personality);
7536 int unregister_md_personality(struct md_personality *p)
7538 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7539 spin_lock(&pers_lock);
7540 list_del_init(&p->list);
7541 spin_unlock(&pers_lock);
7544 EXPORT_SYMBOL(unregister_md_personality);
7546 int register_md_cluster_operations(struct md_cluster_operations *ops,
7547 struct module *module)
7550 spin_lock(&pers_lock);
7551 if (md_cluster_ops != NULL)
7554 md_cluster_ops = ops;
7555 md_cluster_mod = module;
7557 spin_unlock(&pers_lock);
7560 EXPORT_SYMBOL(register_md_cluster_operations);
7562 int unregister_md_cluster_operations(void)
7564 spin_lock(&pers_lock);
7565 md_cluster_ops = NULL;
7566 spin_unlock(&pers_lock);
7569 EXPORT_SYMBOL(unregister_md_cluster_operations);
7571 int md_setup_cluster(struct mddev *mddev, int nodes)
7575 err = request_module("md-cluster");
7577 pr_err("md-cluster module not found.\n");
7581 spin_lock(&pers_lock);
7582 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7583 spin_unlock(&pers_lock);
7586 spin_unlock(&pers_lock);
7588 return md_cluster_ops->join(mddev, nodes);
7591 void md_cluster_stop(struct mddev *mddev)
7593 if (!md_cluster_ops)
7595 md_cluster_ops->leave(mddev);
7596 module_put(md_cluster_mod);
7599 static int is_mddev_idle(struct mddev *mddev, int init)
7601 struct md_rdev *rdev;
7607 rdev_for_each_rcu(rdev, mddev) {
7608 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7609 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7610 (int)part_stat_read(&disk->part0, sectors[1]) -
7611 atomic_read(&disk->sync_io);
7612 /* sync IO will cause sync_io to increase before the disk_stats
7613 * as sync_io is counted when a request starts, and
7614 * disk_stats is counted when it completes.
7615 * So resync activity will cause curr_events to be smaller than
7616 * when there was no such activity.
7617 * non-sync IO will cause disk_stat to increase without
7618 * increasing sync_io so curr_events will (eventually)
7619 * be larger than it was before. Once it becomes
7620 * substantially larger, the test below will cause
7621 * the array to appear non-idle, and resync will slow
7623 * If there is a lot of outstanding resync activity when
7624 * we set last_event to curr_events, then all that activity
7625 * completing might cause the array to appear non-idle
7626 * and resync will be slowed down even though there might
7627 * not have been non-resync activity. This will only
7628 * happen once though. 'last_events' will soon reflect
7629 * the state where there is little or no outstanding
7630 * resync requests, and further resync activity will
7631 * always make curr_events less than last_events.
7634 if (init || curr_events - rdev->last_events > 64) {
7635 rdev->last_events = curr_events;
7643 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7645 /* another "blocks" (512byte) blocks have been synced */
7646 atomic_sub(blocks, &mddev->recovery_active);
7647 wake_up(&mddev->recovery_wait);
7649 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7650 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7651 md_wakeup_thread(mddev->thread);
7652 // stop recovery, signal do_sync ....
7655 EXPORT_SYMBOL(md_done_sync);
7657 /* md_write_start(mddev, bi)
7658 * If we need to update some array metadata (e.g. 'active' flag
7659 * in superblock) before writing, schedule a superblock update
7660 * and wait for it to complete.
7662 void md_write_start(struct mddev *mddev, struct bio *bi)
7665 if (bio_data_dir(bi) != WRITE)
7668 BUG_ON(mddev->ro == 1);
7669 if (mddev->ro == 2) {
7670 /* need to switch to read/write */
7672 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7673 md_wakeup_thread(mddev->thread);
7674 md_wakeup_thread(mddev->sync_thread);
7677 atomic_inc(&mddev->writes_pending);
7678 if (mddev->safemode == 1)
7679 mddev->safemode = 0;
7680 if (mddev->in_sync) {
7681 spin_lock(&mddev->lock);
7682 if (mddev->in_sync) {
7684 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7685 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7686 md_wakeup_thread(mddev->thread);
7689 spin_unlock(&mddev->lock);
7692 sysfs_notify_dirent_safe(mddev->sysfs_state);
7693 wait_event(mddev->sb_wait,
7694 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7696 EXPORT_SYMBOL(md_write_start);
7698 void md_write_end(struct mddev *mddev)
7700 if (atomic_dec_and_test(&mddev->writes_pending)) {
7701 if (mddev->safemode == 2)
7702 md_wakeup_thread(mddev->thread);
7703 else if (mddev->safemode_delay)
7704 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7707 EXPORT_SYMBOL(md_write_end);
7709 /* md_allow_write(mddev)
7710 * Calling this ensures that the array is marked 'active' so that writes
7711 * may proceed without blocking. It is important to call this before
7712 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7713 * Must be called with mddev_lock held.
7715 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7716 * is dropped, so return -EAGAIN after notifying userspace.
7718 int md_allow_write(struct mddev *mddev)
7724 if (!mddev->pers->sync_request)
7727 spin_lock(&mddev->lock);
7728 if (mddev->in_sync) {
7730 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7731 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7732 if (mddev->safemode_delay &&
7733 mddev->safemode == 0)
7734 mddev->safemode = 1;
7735 spin_unlock(&mddev->lock);
7736 md_update_sb(mddev, 0);
7737 sysfs_notify_dirent_safe(mddev->sysfs_state);
7739 spin_unlock(&mddev->lock);
7741 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7746 EXPORT_SYMBOL_GPL(md_allow_write);
7748 #define SYNC_MARKS 10
7749 #define SYNC_MARK_STEP (3*HZ)
7750 #define UPDATE_FREQUENCY (5*60*HZ)
7751 void md_do_sync(struct md_thread *thread)
7753 struct mddev *mddev = thread->mddev;
7754 struct mddev *mddev2;
7755 unsigned int currspeed = 0,
7757 sector_t max_sectors,j, io_sectors, recovery_done;
7758 unsigned long mark[SYNC_MARKS];
7759 unsigned long update_time;
7760 sector_t mark_cnt[SYNC_MARKS];
7762 struct list_head *tmp;
7763 sector_t last_check;
7765 struct md_rdev *rdev;
7766 char *desc, *action = NULL;
7767 struct blk_plug plug;
7768 bool cluster_resync_finished = false;
7770 /* just incase thread restarts... */
7771 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7773 if (mddev->ro) {/* never try to sync a read-only array */
7774 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7778 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7779 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7780 desc = "data-check";
7782 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7783 desc = "requested-resync";
7787 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7792 mddev->last_sync_action = action ?: desc;
7794 /* we overload curr_resync somewhat here.
7795 * 0 == not engaged in resync at all
7796 * 2 == checking that there is no conflict with another sync
7797 * 1 == like 2, but have yielded to allow conflicting resync to
7799 * other == active in resync - this many blocks
7801 * Before starting a resync we must have set curr_resync to
7802 * 2, and then checked that every "conflicting" array has curr_resync
7803 * less than ours. When we find one that is the same or higher
7804 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7805 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7806 * This will mean we have to start checking from the beginning again.
7811 mddev->curr_resync = 2;
7814 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7816 for_each_mddev(mddev2, tmp) {
7817 if (mddev2 == mddev)
7819 if (!mddev->parallel_resync
7820 && mddev2->curr_resync
7821 && match_mddev_units(mddev, mddev2)) {
7823 if (mddev < mddev2 && mddev->curr_resync == 2) {
7824 /* arbitrarily yield */
7825 mddev->curr_resync = 1;
7826 wake_up(&resync_wait);
7828 if (mddev > mddev2 && mddev->curr_resync == 1)
7829 /* no need to wait here, we can wait the next
7830 * time 'round when curr_resync == 2
7833 /* We need to wait 'interruptible' so as not to
7834 * contribute to the load average, and not to
7835 * be caught by 'softlockup'
7837 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7838 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7839 mddev2->curr_resync >= mddev->curr_resync) {
7840 printk(KERN_INFO "md: delaying %s of %s"
7841 " until %s has finished (they"
7842 " share one or more physical units)\n",
7843 desc, mdname(mddev), mdname(mddev2));
7845 if (signal_pending(current))
7846 flush_signals(current);
7848 finish_wait(&resync_wait, &wq);
7851 finish_wait(&resync_wait, &wq);
7854 } while (mddev->curr_resync < 2);
7857 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7858 /* resync follows the size requested by the personality,
7859 * which defaults to physical size, but can be virtual size
7861 max_sectors = mddev->resync_max_sectors;
7862 atomic64_set(&mddev->resync_mismatches, 0);
7863 /* we don't use the checkpoint if there's a bitmap */
7864 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7865 j = mddev->resync_min;
7866 else if (!mddev->bitmap)
7867 j = mddev->recovery_cp;
7869 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7870 max_sectors = mddev->resync_max_sectors;
7872 /* recovery follows the physical size of devices */
7873 max_sectors = mddev->dev_sectors;
7876 rdev_for_each_rcu(rdev, mddev)
7877 if (rdev->raid_disk >= 0 &&
7878 !test_bit(Journal, &rdev->flags) &&
7879 !test_bit(Faulty, &rdev->flags) &&
7880 !test_bit(In_sync, &rdev->flags) &&
7881 rdev->recovery_offset < j)
7882 j = rdev->recovery_offset;
7885 /* If there is a bitmap, we need to make sure all
7886 * writes that started before we added a spare
7887 * complete before we start doing a recovery.
7888 * Otherwise the write might complete and (via
7889 * bitmap_endwrite) set a bit in the bitmap after the
7890 * recovery has checked that bit and skipped that
7893 if (mddev->bitmap) {
7894 mddev->pers->quiesce(mddev, 1);
7895 mddev->pers->quiesce(mddev, 0);
7899 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7900 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7901 " %d KB/sec/disk.\n", speed_min(mddev));
7902 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7903 "(but not more than %d KB/sec) for %s.\n",
7904 speed_max(mddev), desc);
7906 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7909 for (m = 0; m < SYNC_MARKS; m++) {
7911 mark_cnt[m] = io_sectors;
7914 mddev->resync_mark = mark[last_mark];
7915 mddev->resync_mark_cnt = mark_cnt[last_mark];
7918 * Tune reconstruction:
7920 window = 32*(PAGE_SIZE/512);
7921 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7922 window/2, (unsigned long long)max_sectors/2);
7924 atomic_set(&mddev->recovery_active, 0);
7929 "md: resuming %s of %s from checkpoint.\n",
7930 desc, mdname(mddev));
7931 mddev->curr_resync = j;
7933 mddev->curr_resync = 3; /* no longer delayed */
7934 mddev->curr_resync_completed = j;
7935 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7936 md_new_event(mddev);
7937 update_time = jiffies;
7939 blk_start_plug(&plug);
7940 while (j < max_sectors) {
7945 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7946 ((mddev->curr_resync > mddev->curr_resync_completed &&
7947 (mddev->curr_resync - mddev->curr_resync_completed)
7948 > (max_sectors >> 4)) ||
7949 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7950 (j - mddev->curr_resync_completed)*2
7951 >= mddev->resync_max - mddev->curr_resync_completed ||
7952 mddev->curr_resync_completed > mddev->resync_max
7954 /* time to update curr_resync_completed */
7955 wait_event(mddev->recovery_wait,
7956 atomic_read(&mddev->recovery_active) == 0);
7957 mddev->curr_resync_completed = j;
7958 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7959 j > mddev->recovery_cp)
7960 mddev->recovery_cp = j;
7961 update_time = jiffies;
7962 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7963 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7966 while (j >= mddev->resync_max &&
7967 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7968 /* As this condition is controlled by user-space,
7969 * we can block indefinitely, so use '_interruptible'
7970 * to avoid triggering warnings.
7972 flush_signals(current); /* just in case */
7973 wait_event_interruptible(mddev->recovery_wait,
7974 mddev->resync_max > j
7975 || test_bit(MD_RECOVERY_INTR,
7979 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7982 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7984 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7988 if (!skipped) { /* actual IO requested */
7989 io_sectors += sectors;
7990 atomic_add(sectors, &mddev->recovery_active);
7993 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7997 if (j > max_sectors)
7998 /* when skipping, extra large numbers can be returned. */
8001 mddev->curr_resync = j;
8002 mddev->curr_mark_cnt = io_sectors;
8003 if (last_check == 0)
8004 /* this is the earliest that rebuild will be
8005 * visible in /proc/mdstat
8007 md_new_event(mddev);
8009 if (last_check + window > io_sectors || j == max_sectors)
8012 last_check = io_sectors;
8014 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8016 int next = (last_mark+1) % SYNC_MARKS;
8018 mddev->resync_mark = mark[next];
8019 mddev->resync_mark_cnt = mark_cnt[next];
8020 mark[next] = jiffies;
8021 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8025 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8029 * this loop exits only if either when we are slower than
8030 * the 'hard' speed limit, or the system was IO-idle for
8032 * the system might be non-idle CPU-wise, but we only care
8033 * about not overloading the IO subsystem. (things like an
8034 * e2fsck being done on the RAID array should execute fast)
8038 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8039 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8040 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8042 if (currspeed > speed_min(mddev)) {
8043 if (currspeed > speed_max(mddev)) {
8047 if (!is_mddev_idle(mddev, 0)) {
8049 * Give other IO more of a chance.
8050 * The faster the devices, the less we wait.
8052 wait_event(mddev->recovery_wait,
8053 !atomic_read(&mddev->recovery_active));
8057 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8058 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8059 ? "interrupted" : "done");
8061 * this also signals 'finished resyncing' to md_stop
8063 blk_finish_plug(&plug);
8064 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8066 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8067 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8068 mddev->curr_resync > 2) {
8069 mddev->curr_resync_completed = mddev->curr_resync;
8070 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8072 /* tell personality and other nodes that we are finished */
8073 if (mddev_is_clustered(mddev)) {
8074 md_cluster_ops->resync_finish(mddev);
8075 cluster_resync_finished = true;
8077 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8079 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8080 mddev->curr_resync > 2) {
8081 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8082 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8083 if (mddev->curr_resync >= mddev->recovery_cp) {
8085 "md: checkpointing %s of %s.\n",
8086 desc, mdname(mddev));
8087 if (test_bit(MD_RECOVERY_ERROR,
8089 mddev->recovery_cp =
8090 mddev->curr_resync_completed;
8092 mddev->recovery_cp =
8096 mddev->recovery_cp = MaxSector;
8098 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8099 mddev->curr_resync = MaxSector;
8101 rdev_for_each_rcu(rdev, mddev)
8102 if (rdev->raid_disk >= 0 &&
8103 mddev->delta_disks >= 0 &&
8104 !test_bit(Journal, &rdev->flags) &&
8105 !test_bit(Faulty, &rdev->flags) &&
8106 !test_bit(In_sync, &rdev->flags) &&
8107 rdev->recovery_offset < mddev->curr_resync)
8108 rdev->recovery_offset = mddev->curr_resync;
8113 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8115 if (mddev_is_clustered(mddev) &&
8116 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8117 !cluster_resync_finished)
8118 md_cluster_ops->resync_finish(mddev);
8120 spin_lock(&mddev->lock);
8121 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8122 /* We completed so min/max setting can be forgotten if used. */
8123 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8124 mddev->resync_min = 0;
8125 mddev->resync_max = MaxSector;
8126 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8127 mddev->resync_min = mddev->curr_resync_completed;
8128 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8129 mddev->curr_resync = 0;
8130 spin_unlock(&mddev->lock);
8132 wake_up(&resync_wait);
8133 md_wakeup_thread(mddev->thread);
8136 EXPORT_SYMBOL_GPL(md_do_sync);
8138 static int remove_and_add_spares(struct mddev *mddev,
8139 struct md_rdev *this)
8141 struct md_rdev *rdev;
8145 rdev_for_each(rdev, mddev)
8146 if ((this == NULL || rdev == this) &&
8147 rdev->raid_disk >= 0 &&
8148 !test_bit(Blocked, &rdev->flags) &&
8149 (test_bit(Faulty, &rdev->flags) ||
8150 (!test_bit(In_sync, &rdev->flags) &&
8151 !test_bit(Journal, &rdev->flags))) &&
8152 atomic_read(&rdev->nr_pending)==0) {
8153 if (mddev->pers->hot_remove_disk(
8154 mddev, rdev) == 0) {
8155 sysfs_unlink_rdev(mddev, rdev);
8156 rdev->raid_disk = -1;
8160 if (removed && mddev->kobj.sd)
8161 sysfs_notify(&mddev->kobj, NULL, "degraded");
8163 if (this && removed)
8166 rdev_for_each(rdev, mddev) {
8167 if (this && this != rdev)
8169 if (test_bit(Candidate, &rdev->flags))
8171 if (rdev->raid_disk >= 0 &&
8172 !test_bit(In_sync, &rdev->flags) &&
8173 !test_bit(Journal, &rdev->flags) &&
8174 !test_bit(Faulty, &rdev->flags))
8176 if (rdev->raid_disk >= 0)
8178 if (test_bit(Faulty, &rdev->flags))
8180 if (test_bit(Journal, &rdev->flags))
8183 ! (rdev->saved_raid_disk >= 0 &&
8184 !test_bit(Bitmap_sync, &rdev->flags)))
8187 rdev->recovery_offset = 0;
8189 hot_add_disk(mddev, rdev) == 0) {
8190 if (sysfs_link_rdev(mddev, rdev))
8191 /* failure here is OK */;
8193 md_new_event(mddev);
8194 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8199 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8203 static void md_start_sync(struct work_struct *ws)
8205 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8208 if (mddev_is_clustered(mddev)) {
8209 ret = md_cluster_ops->resync_start(mddev);
8211 mddev->sync_thread = NULL;
8216 mddev->sync_thread = md_register_thread(md_do_sync,
8220 if (!mddev->sync_thread) {
8221 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8222 printk(KERN_ERR "%s: could not start resync"
8225 /* leave the spares where they are, it shouldn't hurt */
8226 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8227 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8228 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8229 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8230 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8231 wake_up(&resync_wait);
8232 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8234 if (mddev->sysfs_action)
8235 sysfs_notify_dirent_safe(mddev->sysfs_action);
8237 md_wakeup_thread(mddev->sync_thread);
8238 sysfs_notify_dirent_safe(mddev->sysfs_action);
8239 md_new_event(mddev);
8243 * This routine is regularly called by all per-raid-array threads to
8244 * deal with generic issues like resync and super-block update.
8245 * Raid personalities that don't have a thread (linear/raid0) do not
8246 * need this as they never do any recovery or update the superblock.
8248 * It does not do any resync itself, but rather "forks" off other threads
8249 * to do that as needed.
8250 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8251 * "->recovery" and create a thread at ->sync_thread.
8252 * When the thread finishes it sets MD_RECOVERY_DONE
8253 * and wakeups up this thread which will reap the thread and finish up.
8254 * This thread also removes any faulty devices (with nr_pending == 0).
8256 * The overall approach is:
8257 * 1/ if the superblock needs updating, update it.
8258 * 2/ If a recovery thread is running, don't do anything else.
8259 * 3/ If recovery has finished, clean up, possibly marking spares active.
8260 * 4/ If there are any faulty devices, remove them.
8261 * 5/ If array is degraded, try to add spares devices
8262 * 6/ If array has spares or is not in-sync, start a resync thread.
8264 void md_check_recovery(struct mddev *mddev)
8266 if (mddev->suspended)
8270 bitmap_daemon_work(mddev);
8272 if (signal_pending(current)) {
8273 if (mddev->pers->sync_request && !mddev->external) {
8274 printk(KERN_INFO "md: %s in immediate safe mode\n",
8276 mddev->safemode = 2;
8278 flush_signals(current);
8281 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8284 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8285 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8286 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8287 (mddev->external == 0 && mddev->safemode == 1) ||
8288 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8289 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8293 if (mddev_trylock(mddev)) {
8297 struct md_rdev *rdev;
8298 if (!mddev->external && mddev->in_sync)
8299 /* 'Blocked' flag not needed as failed devices
8300 * will be recorded if array switched to read/write.
8301 * Leaving it set will prevent the device
8302 * from being removed.
8304 rdev_for_each(rdev, mddev)
8305 clear_bit(Blocked, &rdev->flags);
8306 /* On a read-only array we can:
8307 * - remove failed devices
8308 * - add already-in_sync devices if the array itself
8310 * As we only add devices that are already in-sync,
8311 * we can activate the spares immediately.
8313 remove_and_add_spares(mddev, NULL);
8314 /* There is no thread, but we need to call
8315 * ->spare_active and clear saved_raid_disk
8317 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8318 md_reap_sync_thread(mddev);
8319 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8320 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8321 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8325 if (!mddev->external) {
8327 spin_lock(&mddev->lock);
8328 if (mddev->safemode &&
8329 !atomic_read(&mddev->writes_pending) &&
8331 mddev->recovery_cp == MaxSector) {
8334 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8336 if (mddev->safemode == 1)
8337 mddev->safemode = 0;
8338 spin_unlock(&mddev->lock);
8340 sysfs_notify_dirent_safe(mddev->sysfs_state);
8343 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8344 md_update_sb(mddev, 0);
8346 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8347 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8348 /* resync/recovery still happening */
8349 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8352 if (mddev->sync_thread) {
8353 md_reap_sync_thread(mddev);
8356 /* Set RUNNING before clearing NEEDED to avoid
8357 * any transients in the value of "sync_action".
8359 mddev->curr_resync_completed = 0;
8360 spin_lock(&mddev->lock);
8361 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8362 spin_unlock(&mddev->lock);
8363 /* Clear some bits that don't mean anything, but
8366 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8367 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8369 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8370 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8372 /* no recovery is running.
8373 * remove any failed drives, then
8374 * add spares if possible.
8375 * Spares are also removed and re-added, to allow
8376 * the personality to fail the re-add.
8379 if (mddev->reshape_position != MaxSector) {
8380 if (mddev->pers->check_reshape == NULL ||
8381 mddev->pers->check_reshape(mddev) != 0)
8382 /* Cannot proceed */
8384 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8385 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8386 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8387 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8388 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8389 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8390 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8391 } else if (mddev->recovery_cp < MaxSector) {
8392 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8393 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8394 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8395 /* nothing to be done ... */
8398 if (mddev->pers->sync_request) {
8400 /* We are adding a device or devices to an array
8401 * which has the bitmap stored on all devices.
8402 * So make sure all bitmap pages get written
8404 bitmap_write_all(mddev->bitmap);
8406 INIT_WORK(&mddev->del_work, md_start_sync);
8407 queue_work(md_misc_wq, &mddev->del_work);
8411 if (!mddev->sync_thread) {
8412 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8413 wake_up(&resync_wait);
8414 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8416 if (mddev->sysfs_action)
8417 sysfs_notify_dirent_safe(mddev->sysfs_action);
8420 wake_up(&mddev->sb_wait);
8421 mddev_unlock(mddev);
8424 EXPORT_SYMBOL(md_check_recovery);
8426 void md_reap_sync_thread(struct mddev *mddev)
8428 struct md_rdev *rdev;
8430 /* resync has finished, collect result */
8431 md_unregister_thread(&mddev->sync_thread);
8432 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8433 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8435 /* activate any spares */
8436 if (mddev->pers->spare_active(mddev)) {
8437 sysfs_notify(&mddev->kobj, NULL,
8439 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8442 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8443 mddev->pers->finish_reshape)
8444 mddev->pers->finish_reshape(mddev);
8446 /* If array is no-longer degraded, then any saved_raid_disk
8447 * information must be scrapped.
8449 if (!mddev->degraded)
8450 rdev_for_each(rdev, mddev)
8451 rdev->saved_raid_disk = -1;
8453 md_update_sb(mddev, 1);
8454 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8455 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8456 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8457 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8458 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8459 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8460 wake_up(&resync_wait);
8461 /* flag recovery needed just to double check */
8462 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8463 sysfs_notify_dirent_safe(mddev->sysfs_action);
8464 md_new_event(mddev);
8465 if (mddev->event_work.func)
8466 queue_work(md_misc_wq, &mddev->event_work);
8468 EXPORT_SYMBOL(md_reap_sync_thread);
8470 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8472 sysfs_notify_dirent_safe(rdev->sysfs_state);
8473 wait_event_timeout(rdev->blocked_wait,
8474 !test_bit(Blocked, &rdev->flags) &&
8475 !test_bit(BlockedBadBlocks, &rdev->flags),
8476 msecs_to_jiffies(5000));
8477 rdev_dec_pending(rdev, mddev);
8479 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8481 void md_finish_reshape(struct mddev *mddev)
8483 /* called be personality module when reshape completes. */
8484 struct md_rdev *rdev;
8486 rdev_for_each(rdev, mddev) {
8487 if (rdev->data_offset > rdev->new_data_offset)
8488 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8490 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8491 rdev->data_offset = rdev->new_data_offset;
8494 EXPORT_SYMBOL(md_finish_reshape);
8496 /* Bad block management.
8497 * We can record which blocks on each device are 'bad' and so just
8498 * fail those blocks, or that stripe, rather than the whole device.
8499 * Entries in the bad-block table are 64bits wide. This comprises:
8500 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8501 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8502 * A 'shift' can be set so that larger blocks are tracked and
8503 * consequently larger devices can be covered.
8504 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8506 * Locking of the bad-block table uses a seqlock so md_is_badblock
8507 * might need to retry if it is very unlucky.
8508 * We will sometimes want to check for bad blocks in a bi_end_io function,
8509 * so we use the write_seqlock_irq variant.
8511 * When looking for a bad block we specify a range and want to
8512 * know if any block in the range is bad. So we binary-search
8513 * to the last range that starts at-or-before the given endpoint,
8514 * (or "before the sector after the target range")
8515 * then see if it ends after the given start.
8517 * 0 if there are no known bad blocks in the range
8518 * 1 if there are known bad block which are all acknowledged
8519 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8520 * plus the start/length of the first bad section we overlap.
8522 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8523 sector_t *first_bad, int *bad_sectors)
8529 sector_t target = s + sectors;
8532 if (bb->shift > 0) {
8533 /* round the start down, and the end up */
8535 target += (1<<bb->shift) - 1;
8536 target >>= bb->shift;
8537 sectors = target - s;
8539 /* 'target' is now the first block after the bad range */
8542 seq = read_seqbegin(&bb->lock);
8547 /* Binary search between lo and hi for 'target'
8548 * i.e. for the last range that starts before 'target'
8550 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8551 * are known not to be the last range before target.
8552 * VARIANT: hi-lo is the number of possible
8553 * ranges, and decreases until it reaches 1
8555 while (hi - lo > 1) {
8556 int mid = (lo + hi) / 2;
8557 sector_t a = BB_OFFSET(p[mid]);
8559 /* This could still be the one, earlier ranges
8563 /* This and later ranges are definitely out. */
8566 /* 'lo' might be the last that started before target, but 'hi' isn't */
8568 /* need to check all range that end after 's' to see if
8569 * any are unacknowledged.
8572 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8573 if (BB_OFFSET(p[lo]) < target) {
8574 /* starts before the end, and finishes after
8575 * the start, so they must overlap
8577 if (rv != -1 && BB_ACK(p[lo]))
8581 *first_bad = BB_OFFSET(p[lo]);
8582 *bad_sectors = BB_LEN(p[lo]);
8588 if (read_seqretry(&bb->lock, seq))
8593 EXPORT_SYMBOL_GPL(md_is_badblock);
8596 * Add a range of bad blocks to the table.
8597 * This might extend the table, or might contract it
8598 * if two adjacent ranges can be merged.
8599 * We binary-search to find the 'insertion' point, then
8600 * decide how best to handle it.
8602 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8608 unsigned long flags;
8611 /* badblocks are disabled */
8615 /* round the start down, and the end up */
8616 sector_t next = s + sectors;
8618 next += (1<<bb->shift) - 1;
8623 write_seqlock_irqsave(&bb->lock, flags);
8628 /* Find the last range that starts at-or-before 's' */
8629 while (hi - lo > 1) {
8630 int mid = (lo + hi) / 2;
8631 sector_t a = BB_OFFSET(p[mid]);
8637 if (hi > lo && BB_OFFSET(p[lo]) > s)
8641 /* we found a range that might merge with the start
8644 sector_t a = BB_OFFSET(p[lo]);
8645 sector_t e = a + BB_LEN(p[lo]);
8646 int ack = BB_ACK(p[lo]);
8648 /* Yes, we can merge with a previous range */
8649 if (s == a && s + sectors >= e)
8650 /* new range covers old */
8653 ack = ack && acknowledged;
8655 if (e < s + sectors)
8657 if (e - a <= BB_MAX_LEN) {
8658 p[lo] = BB_MAKE(a, e-a, ack);
8661 /* does not all fit in one range,
8662 * make p[lo] maximal
8664 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8665 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8671 if (sectors && hi < bb->count) {
8672 /* 'hi' points to the first range that starts after 's'.
8673 * Maybe we can merge with the start of that range */
8674 sector_t a = BB_OFFSET(p[hi]);
8675 sector_t e = a + BB_LEN(p[hi]);
8676 int ack = BB_ACK(p[hi]);
8677 if (a <= s + sectors) {
8678 /* merging is possible */
8679 if (e <= s + sectors) {
8684 ack = ack && acknowledged;
8687 if (e - a <= BB_MAX_LEN) {
8688 p[hi] = BB_MAKE(a, e-a, ack);
8691 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8699 if (sectors == 0 && hi < bb->count) {
8700 /* we might be able to combine lo and hi */
8701 /* Note: 's' is at the end of 'lo' */
8702 sector_t a = BB_OFFSET(p[hi]);
8703 int lolen = BB_LEN(p[lo]);
8704 int hilen = BB_LEN(p[hi]);
8705 int newlen = lolen + hilen - (s - a);
8706 if (s >= a && newlen < BB_MAX_LEN) {
8707 /* yes, we can combine them */
8708 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8709 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8710 memmove(p + hi, p + hi + 1,
8711 (bb->count - hi - 1) * 8);
8716 /* didn't merge (it all).
8717 * Need to add a range just before 'hi' */
8718 if (bb->count >= MD_MAX_BADBLOCKS) {
8719 /* No room for more */
8723 int this_sectors = sectors;
8724 memmove(p + hi + 1, p + hi,
8725 (bb->count - hi) * 8);
8728 if (this_sectors > BB_MAX_LEN)
8729 this_sectors = BB_MAX_LEN;
8730 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8731 sectors -= this_sectors;
8738 bb->unacked_exist = 1;
8739 write_sequnlock_irqrestore(&bb->lock, flags);
8744 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8749 s += rdev->new_data_offset;
8751 s += rdev->data_offset;
8752 rv = md_set_badblocks(&rdev->badblocks,
8755 /* Make sure they get written out promptly */
8756 sysfs_notify_dirent_safe(rdev->sysfs_state);
8757 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8758 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8759 md_wakeup_thread(rdev->mddev->thread);
8763 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8766 * Remove a range of bad blocks from the table.
8767 * This may involve extending the table if we spilt a region,
8768 * but it must not fail. So if the table becomes full, we just
8769 * drop the remove request.
8771 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8775 sector_t target = s + sectors;
8778 if (bb->shift > 0) {
8779 /* When clearing we round the start up and the end down.
8780 * This should not matter as the shift should align with
8781 * the block size and no rounding should ever be needed.
8782 * However it is better the think a block is bad when it
8783 * isn't than to think a block is not bad when it is.
8785 s += (1<<bb->shift) - 1;
8787 target >>= bb->shift;
8788 sectors = target - s;
8791 write_seqlock_irq(&bb->lock);
8796 /* Find the last range that starts before 'target' */
8797 while (hi - lo > 1) {
8798 int mid = (lo + hi) / 2;
8799 sector_t a = BB_OFFSET(p[mid]);
8806 /* p[lo] is the last range that could overlap the
8807 * current range. Earlier ranges could also overlap,
8808 * but only this one can overlap the end of the range.
8810 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8811 /* Partial overlap, leave the tail of this range */
8812 int ack = BB_ACK(p[lo]);
8813 sector_t a = BB_OFFSET(p[lo]);
8814 sector_t end = a + BB_LEN(p[lo]);
8817 /* we need to split this range */
8818 if (bb->count >= MD_MAX_BADBLOCKS) {
8822 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8824 p[lo] = BB_MAKE(a, s-a, ack);
8827 p[lo] = BB_MAKE(target, end - target, ack);
8828 /* there is no longer an overlap */
8833 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8834 /* This range does overlap */
8835 if (BB_OFFSET(p[lo]) < s) {
8836 /* Keep the early parts of this range. */
8837 int ack = BB_ACK(p[lo]);
8838 sector_t start = BB_OFFSET(p[lo]);
8839 p[lo] = BB_MAKE(start, s - start, ack);
8840 /* now low doesn't overlap, so.. */
8845 /* 'lo' is strictly before, 'hi' is strictly after,
8846 * anything between needs to be discarded
8849 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8850 bb->count -= (hi - lo - 1);
8856 write_sequnlock_irq(&bb->lock);
8860 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8864 s += rdev->new_data_offset;
8866 s += rdev->data_offset;
8867 return md_clear_badblocks(&rdev->badblocks,
8870 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8873 * Acknowledge all bad blocks in a list.
8874 * This only succeeds if ->changed is clear. It is used by
8875 * in-kernel metadata updates
8877 void md_ack_all_badblocks(struct badblocks *bb)
8879 if (bb->page == NULL || bb->changed)
8880 /* no point even trying */
8882 write_seqlock_irq(&bb->lock);
8884 if (bb->changed == 0 && bb->unacked_exist) {
8887 for (i = 0; i < bb->count ; i++) {
8888 if (!BB_ACK(p[i])) {
8889 sector_t start = BB_OFFSET(p[i]);
8890 int len = BB_LEN(p[i]);
8891 p[i] = BB_MAKE(start, len, 1);
8894 bb->unacked_exist = 0;
8896 write_sequnlock_irq(&bb->lock);
8898 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8900 /* sysfs access to bad-blocks list.
8901 * We present two files.
8902 * 'bad-blocks' lists sector numbers and lengths of ranges that
8903 * are recorded as bad. The list is truncated to fit within
8904 * the one-page limit of sysfs.
8905 * Writing "sector length" to this file adds an acknowledged
8907 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8908 * been acknowledged. Writing to this file adds bad blocks
8909 * without acknowledging them. This is largely for testing.
8913 badblocks_show(struct badblocks *bb, char *page, int unack)
8924 seq = read_seqbegin(&bb->lock);
8929 while (len < PAGE_SIZE && i < bb->count) {
8930 sector_t s = BB_OFFSET(p[i]);
8931 unsigned int length = BB_LEN(p[i]);
8932 int ack = BB_ACK(p[i]);
8938 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8939 (unsigned long long)s << bb->shift,
8940 length << bb->shift);
8942 if (unack && len == 0)
8943 bb->unacked_exist = 0;
8945 if (read_seqretry(&bb->lock, seq))
8954 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8956 unsigned long long sector;
8960 /* Allow clearing via sysfs *only* for testing/debugging.
8961 * Normally only a successful write may clear a badblock
8964 if (page[0] == '-') {
8968 #endif /* DO_DEBUG */
8970 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8972 if (newline != '\n')
8984 md_clear_badblocks(bb, sector, length);
8987 #endif /* DO_DEBUG */
8988 if (md_set_badblocks(bb, sector, length, !unack))
8994 static int md_notify_reboot(struct notifier_block *this,
8995 unsigned long code, void *x)
8997 struct list_head *tmp;
8998 struct mddev *mddev;
9001 for_each_mddev(mddev, tmp) {
9002 if (mddev_trylock(mddev)) {
9004 __md_stop_writes(mddev);
9005 if (mddev->persistent)
9006 mddev->safemode = 2;
9007 mddev_unlock(mddev);
9012 * certain more exotic SCSI devices are known to be
9013 * volatile wrt too early system reboots. While the
9014 * right place to handle this issue is the given
9015 * driver, we do want to have a safe RAID driver ...
9023 static struct notifier_block md_notifier = {
9024 .notifier_call = md_notify_reboot,
9026 .priority = INT_MAX, /* before any real devices */
9029 static void md_geninit(void)
9031 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9033 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9036 static int __init md_init(void)
9040 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9044 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9048 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9051 if ((ret = register_blkdev(0, "mdp")) < 0)
9055 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9056 md_probe, NULL, NULL);
9057 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9058 md_probe, NULL, NULL);
9060 register_reboot_notifier(&md_notifier);
9061 raid_table_header = register_sysctl_table(raid_root_table);
9067 unregister_blkdev(MD_MAJOR, "md");
9069 destroy_workqueue(md_misc_wq);
9071 destroy_workqueue(md_wq);
9076 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9078 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9079 struct md_rdev *rdev2;
9081 char b[BDEVNAME_SIZE];
9083 /* Check for change of roles in the active devices */
9084 rdev_for_each(rdev2, mddev) {
9085 if (test_bit(Faulty, &rdev2->flags))
9088 /* Check if the roles changed */
9089 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9091 if (test_bit(Candidate, &rdev2->flags)) {
9092 if (role == 0xfffe) {
9093 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9094 md_kick_rdev_from_array(rdev2);
9098 clear_bit(Candidate, &rdev2->flags);
9101 if (role != rdev2->raid_disk) {
9103 if (rdev2->raid_disk == -1 && role != 0xffff) {
9104 rdev2->saved_raid_disk = role;
9105 ret = remove_and_add_spares(mddev, rdev2);
9106 pr_info("Activated spare: %s\n",
9107 bdevname(rdev2->bdev,b));
9111 * We just want to do the minimum to mark the disk
9112 * as faulty. The recovery is performed by the
9113 * one who initiated the error.
9115 if ((role == 0xfffe) || (role == 0xfffd)) {
9116 md_error(mddev, rdev2);
9117 clear_bit(Blocked, &rdev2->flags);
9122 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9123 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9125 /* Finally set the event to be up to date */
9126 mddev->events = le64_to_cpu(sb->events);
9129 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9132 struct page *swapout = rdev->sb_page;
9133 struct mdp_superblock_1 *sb;
9135 /* Store the sb page of the rdev in the swapout temporary
9136 * variable in case we err in the future
9138 rdev->sb_page = NULL;
9139 alloc_disk_sb(rdev);
9140 ClearPageUptodate(rdev->sb_page);
9141 rdev->sb_loaded = 0;
9142 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9145 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9146 __func__, __LINE__, rdev->desc_nr, err);
9147 put_page(rdev->sb_page);
9148 rdev->sb_page = swapout;
9149 rdev->sb_loaded = 1;
9153 sb = page_address(rdev->sb_page);
9154 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9158 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9159 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9161 /* The other node finished recovery, call spare_active to set
9162 * device In_sync and mddev->degraded
9164 if (rdev->recovery_offset == MaxSector &&
9165 !test_bit(In_sync, &rdev->flags) &&
9166 mddev->pers->spare_active(mddev))
9167 sysfs_notify(&mddev->kobj, NULL, "degraded");
9173 void md_reload_sb(struct mddev *mddev, int nr)
9175 struct md_rdev *rdev;
9179 rdev_for_each_rcu(rdev, mddev) {
9180 if (rdev->desc_nr == nr)
9184 if (!rdev || rdev->desc_nr != nr) {
9185 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9189 err = read_rdev(mddev, rdev);
9193 check_sb_changes(mddev, rdev);
9195 /* Read all rdev's to update recovery_offset */
9196 rdev_for_each_rcu(rdev, mddev)
9197 read_rdev(mddev, rdev);
9199 EXPORT_SYMBOL(md_reload_sb);
9204 * Searches all registered partitions for autorun RAID arrays
9208 static LIST_HEAD(all_detected_devices);
9209 struct detected_devices_node {
9210 struct list_head list;
9214 void md_autodetect_dev(dev_t dev)
9216 struct detected_devices_node *node_detected_dev;
9218 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9219 if (node_detected_dev) {
9220 node_detected_dev->dev = dev;
9221 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9223 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9224 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9228 static void autostart_arrays(int part)
9230 struct md_rdev *rdev;
9231 struct detected_devices_node *node_detected_dev;
9233 int i_scanned, i_passed;
9238 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
9240 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9242 node_detected_dev = list_entry(all_detected_devices.next,
9243 struct detected_devices_node, list);
9244 list_del(&node_detected_dev->list);
9245 dev = node_detected_dev->dev;
9246 kfree(node_detected_dev);
9247 rdev = md_import_device(dev,0, 90);
9251 if (test_bit(Faulty, &rdev->flags))
9254 set_bit(AutoDetected, &rdev->flags);
9255 list_add(&rdev->same_set, &pending_raid_disks);
9259 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9260 i_scanned, i_passed);
9262 autorun_devices(part);
9265 #endif /* !MODULE */
9267 static __exit void md_exit(void)
9269 struct mddev *mddev;
9270 struct list_head *tmp;
9273 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9274 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9276 unregister_blkdev(MD_MAJOR,"md");
9277 unregister_blkdev(mdp_major, "mdp");
9278 unregister_reboot_notifier(&md_notifier);
9279 unregister_sysctl_table(raid_table_header);
9281 /* We cannot unload the modules while some process is
9282 * waiting for us in select() or poll() - wake them up
9285 while (waitqueue_active(&md_event_waiters)) {
9286 /* not safe to leave yet */
9287 wake_up(&md_event_waiters);
9291 remove_proc_entry("mdstat", NULL);
9293 for_each_mddev(mddev, tmp) {
9294 export_array(mddev);
9295 mddev->hold_active = 0;
9297 destroy_workqueue(md_misc_wq);
9298 destroy_workqueue(md_wq);
9301 subsys_initcall(md_init);
9302 module_exit(md_exit)
9304 static int get_ro(char *buffer, struct kernel_param *kp)
9306 return sprintf(buffer, "%d", start_readonly);
9308 static int set_ro(const char *val, struct kernel_param *kp)
9310 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9313 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9314 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9315 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9317 MODULE_LICENSE("GPL");
9318 MODULE_DESCRIPTION("MD RAID framework");
9320 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);