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);
2020 /* Disable data integrity if non-capable/non-matching disk is being added */
2021 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2023 struct blk_integrity *bi_rdev;
2024 struct blk_integrity *bi_mddev;
2026 if (!mddev->gendisk)
2029 bi_rdev = bdev_get_integrity(rdev->bdev);
2030 bi_mddev = blk_get_integrity(mddev->gendisk);
2032 if (!bi_mddev) /* nothing to do */
2034 if (rdev->raid_disk < 0) /* skip spares */
2036 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2037 rdev->bdev->bd_disk) >= 0)
2039 WARN_ON_ONCE(!mddev->suspended);
2040 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2041 blk_integrity_unregister(mddev->gendisk);
2043 EXPORT_SYMBOL(md_integrity_add_rdev);
2045 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2047 char b[BDEVNAME_SIZE];
2051 /* prevent duplicates */
2052 if (find_rdev(mddev, rdev->bdev->bd_dev))
2055 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2056 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2057 rdev->sectors < mddev->dev_sectors)) {
2059 /* Cannot change size, so fail
2060 * If mddev->level <= 0, then we don't care
2061 * about aligning sizes (e.g. linear)
2063 if (mddev->level > 0)
2066 mddev->dev_sectors = rdev->sectors;
2069 /* Verify rdev->desc_nr is unique.
2070 * If it is -1, assign a free number, else
2071 * check number is not in use
2074 if (rdev->desc_nr < 0) {
2077 choice = mddev->raid_disks;
2078 while (md_find_rdev_nr_rcu(mddev, choice))
2080 rdev->desc_nr = choice;
2082 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2088 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2089 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2090 mdname(mddev), mddev->max_disks);
2093 bdevname(rdev->bdev,b);
2094 strreplace(b, '/', '!');
2096 rdev->mddev = mddev;
2097 printk(KERN_INFO "md: bind<%s>\n", b);
2099 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2102 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2103 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2104 /* failure here is OK */;
2105 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2107 list_add_rcu(&rdev->same_set, &mddev->disks);
2108 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2110 /* May as well allow recovery to be retried once */
2111 mddev->recovery_disabled++;
2116 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2121 static void md_delayed_delete(struct work_struct *ws)
2123 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2124 kobject_del(&rdev->kobj);
2125 kobject_put(&rdev->kobj);
2128 static void unbind_rdev_from_array(struct md_rdev *rdev)
2130 char b[BDEVNAME_SIZE];
2132 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2133 list_del_rcu(&rdev->same_set);
2134 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2136 sysfs_remove_link(&rdev->kobj, "block");
2137 sysfs_put(rdev->sysfs_state);
2138 rdev->sysfs_state = NULL;
2139 rdev->badblocks.count = 0;
2140 /* We need to delay this, otherwise we can deadlock when
2141 * writing to 'remove' to "dev/state". We also need
2142 * to delay it due to rcu usage.
2145 INIT_WORK(&rdev->del_work, md_delayed_delete);
2146 kobject_get(&rdev->kobj);
2147 queue_work(md_misc_wq, &rdev->del_work);
2151 * prevent the device from being mounted, repartitioned or
2152 * otherwise reused by a RAID array (or any other kernel
2153 * subsystem), by bd_claiming the device.
2155 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2158 struct block_device *bdev;
2159 char b[BDEVNAME_SIZE];
2161 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2162 shared ? (struct md_rdev *)lock_rdev : rdev);
2164 printk(KERN_ERR "md: could not open %s.\n",
2165 __bdevname(dev, b));
2166 return PTR_ERR(bdev);
2172 static void unlock_rdev(struct md_rdev *rdev)
2174 struct block_device *bdev = rdev->bdev;
2176 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2179 void md_autodetect_dev(dev_t dev);
2181 static void export_rdev(struct md_rdev *rdev)
2183 char b[BDEVNAME_SIZE];
2185 printk(KERN_INFO "md: export_rdev(%s)\n",
2186 bdevname(rdev->bdev,b));
2187 md_rdev_clear(rdev);
2189 if (test_bit(AutoDetected, &rdev->flags))
2190 md_autodetect_dev(rdev->bdev->bd_dev);
2193 kobject_put(&rdev->kobj);
2196 void md_kick_rdev_from_array(struct md_rdev *rdev)
2198 unbind_rdev_from_array(rdev);
2201 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2203 static void export_array(struct mddev *mddev)
2205 struct md_rdev *rdev;
2207 while (!list_empty(&mddev->disks)) {
2208 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2210 md_kick_rdev_from_array(rdev);
2212 mddev->raid_disks = 0;
2213 mddev->major_version = 0;
2216 static void sync_sbs(struct mddev *mddev, int nospares)
2218 /* Update each superblock (in-memory image), but
2219 * if we are allowed to, skip spares which already
2220 * have the right event counter, or have one earlier
2221 * (which would mean they aren't being marked as dirty
2222 * with the rest of the array)
2224 struct md_rdev *rdev;
2225 rdev_for_each(rdev, mddev) {
2226 if (rdev->sb_events == mddev->events ||
2228 rdev->raid_disk < 0 &&
2229 rdev->sb_events+1 == mddev->events)) {
2230 /* Don't update this superblock */
2231 rdev->sb_loaded = 2;
2233 sync_super(mddev, rdev);
2234 rdev->sb_loaded = 1;
2239 static bool does_sb_need_changing(struct mddev *mddev)
2241 struct md_rdev *rdev;
2242 struct mdp_superblock_1 *sb;
2245 /* Find a good rdev */
2246 rdev_for_each(rdev, mddev)
2247 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2250 /* No good device found. */
2254 sb = page_address(rdev->sb_page);
2255 /* Check if a device has become faulty or a spare become active */
2256 rdev_for_each(rdev, mddev) {
2257 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2258 /* Device activated? */
2259 if (role == 0xffff && rdev->raid_disk >=0 &&
2260 !test_bit(Faulty, &rdev->flags))
2262 /* Device turned faulty? */
2263 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2267 /* Check if any mddev parameters have changed */
2268 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2269 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2270 (mddev->layout != le64_to_cpu(sb->layout)) ||
2271 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2272 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2278 void md_update_sb(struct mddev *mddev, int force_change)
2280 struct md_rdev *rdev;
2283 int any_badblocks_changed = 0;
2288 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2292 if (mddev_is_clustered(mddev)) {
2293 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2295 ret = md_cluster_ops->metadata_update_start(mddev);
2296 /* Has someone else has updated the sb */
2297 if (!does_sb_need_changing(mddev)) {
2299 md_cluster_ops->metadata_update_cancel(mddev);
2300 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2305 /* First make sure individual recovery_offsets are correct */
2306 rdev_for_each(rdev, mddev) {
2307 if (rdev->raid_disk >= 0 &&
2308 mddev->delta_disks >= 0 &&
2309 !test_bit(Journal, &rdev->flags) &&
2310 !test_bit(In_sync, &rdev->flags) &&
2311 mddev->curr_resync_completed > rdev->recovery_offset)
2312 rdev->recovery_offset = mddev->curr_resync_completed;
2315 if (!mddev->persistent) {
2316 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2317 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2318 if (!mddev->external) {
2319 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2320 rdev_for_each(rdev, mddev) {
2321 if (rdev->badblocks.changed) {
2322 rdev->badblocks.changed = 0;
2323 md_ack_all_badblocks(&rdev->badblocks);
2324 md_error(mddev, rdev);
2326 clear_bit(Blocked, &rdev->flags);
2327 clear_bit(BlockedBadBlocks, &rdev->flags);
2328 wake_up(&rdev->blocked_wait);
2331 wake_up(&mddev->sb_wait);
2335 spin_lock(&mddev->lock);
2337 mddev->utime = get_seconds();
2339 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2341 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2342 /* just a clean<-> dirty transition, possibly leave spares alone,
2343 * though if events isn't the right even/odd, we will have to do
2349 if (mddev->degraded)
2350 /* If the array is degraded, then skipping spares is both
2351 * dangerous and fairly pointless.
2352 * Dangerous because a device that was removed from the array
2353 * might have a event_count that still looks up-to-date,
2354 * so it can be re-added without a resync.
2355 * Pointless because if there are any spares to skip,
2356 * then a recovery will happen and soon that array won't
2357 * be degraded any more and the spare can go back to sleep then.
2361 sync_req = mddev->in_sync;
2363 /* If this is just a dirty<->clean transition, and the array is clean
2364 * and 'events' is odd, we can roll back to the previous clean state */
2366 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2367 && mddev->can_decrease_events
2368 && mddev->events != 1) {
2370 mddev->can_decrease_events = 0;
2372 /* otherwise we have to go forward and ... */
2374 mddev->can_decrease_events = nospares;
2378 * This 64-bit counter should never wrap.
2379 * Either we are in around ~1 trillion A.C., assuming
2380 * 1 reboot per second, or we have a bug...
2382 WARN_ON(mddev->events == 0);
2384 rdev_for_each(rdev, mddev) {
2385 if (rdev->badblocks.changed)
2386 any_badblocks_changed++;
2387 if (test_bit(Faulty, &rdev->flags))
2388 set_bit(FaultRecorded, &rdev->flags);
2391 sync_sbs(mddev, nospares);
2392 spin_unlock(&mddev->lock);
2394 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2395 mdname(mddev), mddev->in_sync);
2397 bitmap_update_sb(mddev->bitmap);
2398 rdev_for_each(rdev, mddev) {
2399 char b[BDEVNAME_SIZE];
2401 if (rdev->sb_loaded != 1)
2402 continue; /* no noise on spare devices */
2404 if (!test_bit(Faulty, &rdev->flags)) {
2405 md_super_write(mddev,rdev,
2406 rdev->sb_start, rdev->sb_size,
2408 pr_debug("md: (write) %s's sb offset: %llu\n",
2409 bdevname(rdev->bdev, b),
2410 (unsigned long long)rdev->sb_start);
2411 rdev->sb_events = mddev->events;
2412 if (rdev->badblocks.size) {
2413 md_super_write(mddev, rdev,
2414 rdev->badblocks.sector,
2415 rdev->badblocks.size << 9,
2417 rdev->badblocks.size = 0;
2421 pr_debug("md: %s (skipping faulty)\n",
2422 bdevname(rdev->bdev, b));
2424 if (mddev->level == LEVEL_MULTIPATH)
2425 /* only need to write one superblock... */
2428 md_super_wait(mddev);
2429 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2431 spin_lock(&mddev->lock);
2432 if (mddev->in_sync != sync_req ||
2433 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2434 /* have to write it out again */
2435 spin_unlock(&mddev->lock);
2438 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2439 spin_unlock(&mddev->lock);
2440 wake_up(&mddev->sb_wait);
2441 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2442 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2444 rdev_for_each(rdev, mddev) {
2445 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2446 clear_bit(Blocked, &rdev->flags);
2448 if (any_badblocks_changed)
2449 md_ack_all_badblocks(&rdev->badblocks);
2450 clear_bit(BlockedBadBlocks, &rdev->flags);
2451 wake_up(&rdev->blocked_wait);
2454 if (mddev_is_clustered(mddev) && ret == 0)
2455 md_cluster_ops->metadata_update_finish(mddev);
2457 EXPORT_SYMBOL(md_update_sb);
2459 static int add_bound_rdev(struct md_rdev *rdev)
2461 struct mddev *mddev = rdev->mddev;
2464 if (!mddev->pers->hot_remove_disk) {
2465 /* If there is hot_add_disk but no hot_remove_disk
2466 * then added disks for geometry changes,
2467 * and should be added immediately.
2469 super_types[mddev->major_version].
2470 validate_super(mddev, rdev);
2471 err = mddev->pers->hot_add_disk(mddev, rdev);
2473 unbind_rdev_from_array(rdev);
2478 sysfs_notify_dirent_safe(rdev->sysfs_state);
2480 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2481 if (mddev->degraded)
2482 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2483 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2484 md_new_event(mddev);
2485 md_wakeup_thread(mddev->thread);
2489 /* words written to sysfs files may, or may not, be \n terminated.
2490 * We want to accept with case. For this we use cmd_match.
2492 static int cmd_match(const char *cmd, const char *str)
2494 /* See if cmd, written into a sysfs file, matches
2495 * str. They must either be the same, or cmd can
2496 * have a trailing newline
2498 while (*cmd && *str && *cmd == *str) {
2509 struct rdev_sysfs_entry {
2510 struct attribute attr;
2511 ssize_t (*show)(struct md_rdev *, char *);
2512 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2516 state_show(struct md_rdev *rdev, char *page)
2520 unsigned long flags = ACCESS_ONCE(rdev->flags);
2522 if (test_bit(Faulty, &flags) ||
2523 rdev->badblocks.unacked_exist) {
2524 len+= sprintf(page+len, "%sfaulty",sep);
2527 if (test_bit(In_sync, &flags)) {
2528 len += sprintf(page+len, "%sin_sync",sep);
2531 if (test_bit(Journal, &flags)) {
2532 len += sprintf(page+len, "%sjournal",sep);
2535 if (test_bit(WriteMostly, &flags)) {
2536 len += sprintf(page+len, "%swrite_mostly",sep);
2539 if (test_bit(Blocked, &flags) ||
2540 (rdev->badblocks.unacked_exist
2541 && !test_bit(Faulty, &flags))) {
2542 len += sprintf(page+len, "%sblocked", sep);
2545 if (!test_bit(Faulty, &flags) &&
2546 !test_bit(Journal, &flags) &&
2547 !test_bit(In_sync, &flags)) {
2548 len += sprintf(page+len, "%sspare", sep);
2551 if (test_bit(WriteErrorSeen, &flags)) {
2552 len += sprintf(page+len, "%swrite_error", sep);
2555 if (test_bit(WantReplacement, &flags)) {
2556 len += sprintf(page+len, "%swant_replacement", sep);
2559 if (test_bit(Replacement, &flags)) {
2560 len += sprintf(page+len, "%sreplacement", sep);
2564 return len+sprintf(page+len, "\n");
2568 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2571 * faulty - simulates an error
2572 * remove - disconnects the device
2573 * writemostly - sets write_mostly
2574 * -writemostly - clears write_mostly
2575 * blocked - sets the Blocked flags
2576 * -blocked - clears the Blocked and possibly simulates an error
2577 * insync - sets Insync providing device isn't active
2578 * -insync - clear Insync for a device with a slot assigned,
2579 * so that it gets rebuilt based on bitmap
2580 * write_error - sets WriteErrorSeen
2581 * -write_error - clears WriteErrorSeen
2584 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2585 md_error(rdev->mddev, rdev);
2586 if (test_bit(Faulty, &rdev->flags))
2590 } else if (cmd_match(buf, "remove")) {
2591 if (rdev->raid_disk >= 0)
2594 struct mddev *mddev = rdev->mddev;
2596 if (mddev_is_clustered(mddev))
2597 err = md_cluster_ops->remove_disk(mddev, rdev);
2600 md_kick_rdev_from_array(rdev);
2602 md_update_sb(mddev, 1);
2603 md_new_event(mddev);
2606 } else if (cmd_match(buf, "writemostly")) {
2607 set_bit(WriteMostly, &rdev->flags);
2609 } else if (cmd_match(buf, "-writemostly")) {
2610 clear_bit(WriteMostly, &rdev->flags);
2612 } else if (cmd_match(buf, "blocked")) {
2613 set_bit(Blocked, &rdev->flags);
2615 } else if (cmd_match(buf, "-blocked")) {
2616 if (!test_bit(Faulty, &rdev->flags) &&
2617 rdev->badblocks.unacked_exist) {
2618 /* metadata handler doesn't understand badblocks,
2619 * so we need to fail the device
2621 md_error(rdev->mddev, rdev);
2623 clear_bit(Blocked, &rdev->flags);
2624 clear_bit(BlockedBadBlocks, &rdev->flags);
2625 wake_up(&rdev->blocked_wait);
2626 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2627 md_wakeup_thread(rdev->mddev->thread);
2630 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2631 set_bit(In_sync, &rdev->flags);
2633 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2634 !test_bit(Journal, &rdev->flags)) {
2635 if (rdev->mddev->pers == NULL) {
2636 clear_bit(In_sync, &rdev->flags);
2637 rdev->saved_raid_disk = rdev->raid_disk;
2638 rdev->raid_disk = -1;
2641 } else if (cmd_match(buf, "write_error")) {
2642 set_bit(WriteErrorSeen, &rdev->flags);
2644 } else if (cmd_match(buf, "-write_error")) {
2645 clear_bit(WriteErrorSeen, &rdev->flags);
2647 } else if (cmd_match(buf, "want_replacement")) {
2648 /* Any non-spare device that is not a replacement can
2649 * become want_replacement at any time, but we then need to
2650 * check if recovery is needed.
2652 if (rdev->raid_disk >= 0 &&
2653 !test_bit(Journal, &rdev->flags) &&
2654 !test_bit(Replacement, &rdev->flags))
2655 set_bit(WantReplacement, &rdev->flags);
2656 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2657 md_wakeup_thread(rdev->mddev->thread);
2659 } else if (cmd_match(buf, "-want_replacement")) {
2660 /* Clearing 'want_replacement' is always allowed.
2661 * Once replacements starts it is too late though.
2664 clear_bit(WantReplacement, &rdev->flags);
2665 } else if (cmd_match(buf, "replacement")) {
2666 /* Can only set a device as a replacement when array has not
2667 * yet been started. Once running, replacement is automatic
2668 * from spares, or by assigning 'slot'.
2670 if (rdev->mddev->pers)
2673 set_bit(Replacement, &rdev->flags);
2676 } else if (cmd_match(buf, "-replacement")) {
2677 /* Similarly, can only clear Replacement before start */
2678 if (rdev->mddev->pers)
2681 clear_bit(Replacement, &rdev->flags);
2684 } else if (cmd_match(buf, "re-add")) {
2685 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2686 /* clear_bit is performed _after_ all the devices
2687 * have their local Faulty bit cleared. If any writes
2688 * happen in the meantime in the local node, they
2689 * will land in the local bitmap, which will be synced
2690 * by this node eventually
2692 if (!mddev_is_clustered(rdev->mddev) ||
2693 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2694 clear_bit(Faulty, &rdev->flags);
2695 err = add_bound_rdev(rdev);
2701 sysfs_notify_dirent_safe(rdev->sysfs_state);
2702 return err ? err : len;
2704 static struct rdev_sysfs_entry rdev_state =
2705 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2708 errors_show(struct md_rdev *rdev, char *page)
2710 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2714 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2719 rv = kstrtouint(buf, 10, &n);
2722 atomic_set(&rdev->corrected_errors, n);
2725 static struct rdev_sysfs_entry rdev_errors =
2726 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2729 slot_show(struct md_rdev *rdev, char *page)
2731 if (test_bit(Journal, &rdev->flags))
2732 return sprintf(page, "journal\n");
2733 else if (rdev->raid_disk < 0)
2734 return sprintf(page, "none\n");
2736 return sprintf(page, "%d\n", rdev->raid_disk);
2740 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2745 if (test_bit(Journal, &rdev->flags))
2747 if (strncmp(buf, "none", 4)==0)
2750 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2754 if (rdev->mddev->pers && slot == -1) {
2755 /* Setting 'slot' on an active array requires also
2756 * updating the 'rd%d' link, and communicating
2757 * with the personality with ->hot_*_disk.
2758 * For now we only support removing
2759 * failed/spare devices. This normally happens automatically,
2760 * but not when the metadata is externally managed.
2762 if (rdev->raid_disk == -1)
2764 /* personality does all needed checks */
2765 if (rdev->mddev->pers->hot_remove_disk == NULL)
2767 clear_bit(Blocked, &rdev->flags);
2768 remove_and_add_spares(rdev->mddev, rdev);
2769 if (rdev->raid_disk >= 0)
2771 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2772 md_wakeup_thread(rdev->mddev->thread);
2773 } else if (rdev->mddev->pers) {
2774 /* Activating a spare .. or possibly reactivating
2775 * if we ever get bitmaps working here.
2779 if (rdev->raid_disk != -1)
2782 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2785 if (rdev->mddev->pers->hot_add_disk == NULL)
2788 if (slot >= rdev->mddev->raid_disks &&
2789 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2792 rdev->raid_disk = slot;
2793 if (test_bit(In_sync, &rdev->flags))
2794 rdev->saved_raid_disk = slot;
2796 rdev->saved_raid_disk = -1;
2797 clear_bit(In_sync, &rdev->flags);
2798 clear_bit(Bitmap_sync, &rdev->flags);
2799 err = rdev->mddev->pers->
2800 hot_add_disk(rdev->mddev, rdev);
2802 rdev->raid_disk = -1;
2805 sysfs_notify_dirent_safe(rdev->sysfs_state);
2806 if (sysfs_link_rdev(rdev->mddev, rdev))
2807 /* failure here is OK */;
2808 /* don't wakeup anyone, leave that to userspace. */
2810 if (slot >= rdev->mddev->raid_disks &&
2811 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2813 rdev->raid_disk = slot;
2814 /* assume it is working */
2815 clear_bit(Faulty, &rdev->flags);
2816 clear_bit(WriteMostly, &rdev->flags);
2817 set_bit(In_sync, &rdev->flags);
2818 sysfs_notify_dirent_safe(rdev->sysfs_state);
2823 static struct rdev_sysfs_entry rdev_slot =
2824 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2827 offset_show(struct md_rdev *rdev, char *page)
2829 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2833 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2835 unsigned long long offset;
2836 if (kstrtoull(buf, 10, &offset) < 0)
2838 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2840 if (rdev->sectors && rdev->mddev->external)
2841 /* Must set offset before size, so overlap checks
2844 rdev->data_offset = offset;
2845 rdev->new_data_offset = offset;
2849 static struct rdev_sysfs_entry rdev_offset =
2850 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2852 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2854 return sprintf(page, "%llu\n",
2855 (unsigned long long)rdev->new_data_offset);
2858 static ssize_t new_offset_store(struct md_rdev *rdev,
2859 const char *buf, size_t len)
2861 unsigned long long new_offset;
2862 struct mddev *mddev = rdev->mddev;
2864 if (kstrtoull(buf, 10, &new_offset) < 0)
2867 if (mddev->sync_thread ||
2868 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2870 if (new_offset == rdev->data_offset)
2871 /* reset is always permitted */
2873 else if (new_offset > rdev->data_offset) {
2874 /* must not push array size beyond rdev_sectors */
2875 if (new_offset - rdev->data_offset
2876 + mddev->dev_sectors > rdev->sectors)
2879 /* Metadata worries about other space details. */
2881 /* decreasing the offset is inconsistent with a backwards
2884 if (new_offset < rdev->data_offset &&
2885 mddev->reshape_backwards)
2887 /* Increasing offset is inconsistent with forwards
2888 * reshape. reshape_direction should be set to
2889 * 'backwards' first.
2891 if (new_offset > rdev->data_offset &&
2892 !mddev->reshape_backwards)
2895 if (mddev->pers && mddev->persistent &&
2896 !super_types[mddev->major_version]
2897 .allow_new_offset(rdev, new_offset))
2899 rdev->new_data_offset = new_offset;
2900 if (new_offset > rdev->data_offset)
2901 mddev->reshape_backwards = 1;
2902 else if (new_offset < rdev->data_offset)
2903 mddev->reshape_backwards = 0;
2907 static struct rdev_sysfs_entry rdev_new_offset =
2908 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2911 rdev_size_show(struct md_rdev *rdev, char *page)
2913 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2916 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2918 /* check if two start/length pairs overlap */
2926 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2928 unsigned long long blocks;
2931 if (kstrtoull(buf, 10, &blocks) < 0)
2934 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2935 return -EINVAL; /* sector conversion overflow */
2938 if (new != blocks * 2)
2939 return -EINVAL; /* unsigned long long to sector_t overflow */
2946 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2948 struct mddev *my_mddev = rdev->mddev;
2949 sector_t oldsectors = rdev->sectors;
2952 if (test_bit(Journal, &rdev->flags))
2954 if (strict_blocks_to_sectors(buf, §ors) < 0)
2956 if (rdev->data_offset != rdev->new_data_offset)
2957 return -EINVAL; /* too confusing */
2958 if (my_mddev->pers && rdev->raid_disk >= 0) {
2959 if (my_mddev->persistent) {
2960 sectors = super_types[my_mddev->major_version].
2961 rdev_size_change(rdev, sectors);
2964 } else if (!sectors)
2965 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2967 if (!my_mddev->pers->resize)
2968 /* Cannot change size for RAID0 or Linear etc */
2971 if (sectors < my_mddev->dev_sectors)
2972 return -EINVAL; /* component must fit device */
2974 rdev->sectors = sectors;
2975 if (sectors > oldsectors && my_mddev->external) {
2976 /* Need to check that all other rdevs with the same
2977 * ->bdev do not overlap. 'rcu' is sufficient to walk
2978 * the rdev lists safely.
2979 * This check does not provide a hard guarantee, it
2980 * just helps avoid dangerous mistakes.
2982 struct mddev *mddev;
2984 struct list_head *tmp;
2987 for_each_mddev(mddev, tmp) {
2988 struct md_rdev *rdev2;
2990 rdev_for_each(rdev2, mddev)
2991 if (rdev->bdev == rdev2->bdev &&
2993 overlaps(rdev->data_offset, rdev->sectors,
3006 /* Someone else could have slipped in a size
3007 * change here, but doing so is just silly.
3008 * We put oldsectors back because we *know* it is
3009 * safe, and trust userspace not to race with
3012 rdev->sectors = oldsectors;
3019 static struct rdev_sysfs_entry rdev_size =
3020 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3022 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3024 unsigned long long recovery_start = rdev->recovery_offset;
3026 if (test_bit(In_sync, &rdev->flags) ||
3027 recovery_start == MaxSector)
3028 return sprintf(page, "none\n");
3030 return sprintf(page, "%llu\n", recovery_start);
3033 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3035 unsigned long long recovery_start;
3037 if (cmd_match(buf, "none"))
3038 recovery_start = MaxSector;
3039 else if (kstrtoull(buf, 10, &recovery_start))
3042 if (rdev->mddev->pers &&
3043 rdev->raid_disk >= 0)
3046 rdev->recovery_offset = recovery_start;
3047 if (recovery_start == MaxSector)
3048 set_bit(In_sync, &rdev->flags);
3050 clear_bit(In_sync, &rdev->flags);
3054 static struct rdev_sysfs_entry rdev_recovery_start =
3055 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3058 badblocks_show(struct badblocks *bb, char *page, int unack);
3060 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3062 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3064 return badblocks_show(&rdev->badblocks, page, 0);
3066 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3068 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3069 /* Maybe that ack was all we needed */
3070 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3071 wake_up(&rdev->blocked_wait);
3074 static struct rdev_sysfs_entry rdev_bad_blocks =
3075 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3077 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3079 return badblocks_show(&rdev->badblocks, page, 1);
3081 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3083 return badblocks_store(&rdev->badblocks, page, len, 1);
3085 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3086 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3088 static struct attribute *rdev_default_attrs[] = {
3093 &rdev_new_offset.attr,
3095 &rdev_recovery_start.attr,
3096 &rdev_bad_blocks.attr,
3097 &rdev_unack_bad_blocks.attr,
3101 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3103 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3104 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3110 return entry->show(rdev, page);
3114 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3115 const char *page, size_t length)
3117 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3118 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3120 struct mddev *mddev = rdev->mddev;
3124 if (!capable(CAP_SYS_ADMIN))
3126 rv = mddev ? mddev_lock(mddev): -EBUSY;
3128 if (rdev->mddev == NULL)
3131 rv = entry->store(rdev, page, length);
3132 mddev_unlock(mddev);
3137 static void rdev_free(struct kobject *ko)
3139 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3142 static const struct sysfs_ops rdev_sysfs_ops = {
3143 .show = rdev_attr_show,
3144 .store = rdev_attr_store,
3146 static struct kobj_type rdev_ktype = {
3147 .release = rdev_free,
3148 .sysfs_ops = &rdev_sysfs_ops,
3149 .default_attrs = rdev_default_attrs,
3152 int md_rdev_init(struct md_rdev *rdev)
3155 rdev->saved_raid_disk = -1;
3156 rdev->raid_disk = -1;
3158 rdev->data_offset = 0;
3159 rdev->new_data_offset = 0;
3160 rdev->sb_events = 0;
3161 rdev->last_read_error.tv_sec = 0;
3162 rdev->last_read_error.tv_nsec = 0;
3163 rdev->sb_loaded = 0;
3164 rdev->bb_page = NULL;
3165 atomic_set(&rdev->nr_pending, 0);
3166 atomic_set(&rdev->read_errors, 0);
3167 atomic_set(&rdev->corrected_errors, 0);
3169 INIT_LIST_HEAD(&rdev->same_set);
3170 init_waitqueue_head(&rdev->blocked_wait);
3172 /* Add space to store bad block list.
3173 * This reserves the space even on arrays where it cannot
3174 * be used - I wonder if that matters
3176 rdev->badblocks.count = 0;
3177 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3178 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3179 seqlock_init(&rdev->badblocks.lock);
3180 if (rdev->badblocks.page == NULL)
3185 EXPORT_SYMBOL_GPL(md_rdev_init);
3187 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3189 * mark the device faulty if:
3191 * - the device is nonexistent (zero size)
3192 * - the device has no valid superblock
3194 * a faulty rdev _never_ has rdev->sb set.
3196 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3198 char b[BDEVNAME_SIZE];
3200 struct md_rdev *rdev;
3203 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3205 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3206 return ERR_PTR(-ENOMEM);
3209 err = md_rdev_init(rdev);
3212 err = alloc_disk_sb(rdev);
3216 err = lock_rdev(rdev, newdev, super_format == -2);
3220 kobject_init(&rdev->kobj, &rdev_ktype);
3222 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3225 "md: %s has zero or unknown size, marking faulty!\n",
3226 bdevname(rdev->bdev,b));
3231 if (super_format >= 0) {
3232 err = super_types[super_format].
3233 load_super(rdev, NULL, super_minor);
3234 if (err == -EINVAL) {
3236 "md: %s does not have a valid v%d.%d "
3237 "superblock, not importing!\n",
3238 bdevname(rdev->bdev,b),
3239 super_format, super_minor);
3244 "md: could not read %s's sb, not importing!\n",
3245 bdevname(rdev->bdev,b));
3255 md_rdev_clear(rdev);
3257 return ERR_PTR(err);
3261 * Check a full RAID array for plausibility
3264 static void analyze_sbs(struct mddev *mddev)
3267 struct md_rdev *rdev, *freshest, *tmp;
3268 char b[BDEVNAME_SIZE];
3271 rdev_for_each_safe(rdev, tmp, mddev)
3272 switch (super_types[mddev->major_version].
3273 load_super(rdev, freshest, mddev->minor_version)) {
3281 "md: fatal superblock inconsistency in %s"
3282 " -- removing from array\n",
3283 bdevname(rdev->bdev,b));
3284 md_kick_rdev_from_array(rdev);
3287 super_types[mddev->major_version].
3288 validate_super(mddev, freshest);
3291 rdev_for_each_safe(rdev, tmp, mddev) {
3292 if (mddev->max_disks &&
3293 (rdev->desc_nr >= mddev->max_disks ||
3294 i > mddev->max_disks)) {
3296 "md: %s: %s: only %d devices permitted\n",
3297 mdname(mddev), bdevname(rdev->bdev, b),
3299 md_kick_rdev_from_array(rdev);
3302 if (rdev != freshest) {
3303 if (super_types[mddev->major_version].
3304 validate_super(mddev, rdev)) {
3305 printk(KERN_WARNING "md: kicking non-fresh %s"
3307 bdevname(rdev->bdev,b));
3308 md_kick_rdev_from_array(rdev);
3312 if (mddev->level == LEVEL_MULTIPATH) {
3313 rdev->desc_nr = i++;
3314 rdev->raid_disk = rdev->desc_nr;
3315 set_bit(In_sync, &rdev->flags);
3316 } else if (rdev->raid_disk >=
3317 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3318 !test_bit(Journal, &rdev->flags)) {
3319 rdev->raid_disk = -1;
3320 clear_bit(In_sync, &rdev->flags);
3325 /* Read a fixed-point number.
3326 * Numbers in sysfs attributes should be in "standard" units where
3327 * possible, so time should be in seconds.
3328 * However we internally use a a much smaller unit such as
3329 * milliseconds or jiffies.
3330 * This function takes a decimal number with a possible fractional
3331 * component, and produces an integer which is the result of
3332 * multiplying that number by 10^'scale'.
3333 * all without any floating-point arithmetic.
3335 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3337 unsigned long result = 0;
3339 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3342 else if (decimals < scale) {
3345 result = result * 10 + value;
3357 while (decimals < scale) {
3366 safe_delay_show(struct mddev *mddev, char *page)
3368 int msec = (mddev->safemode_delay*1000)/HZ;
3369 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3372 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3376 if (mddev_is_clustered(mddev)) {
3377 pr_info("md: Safemode is disabled for clustered mode\n");
3381 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3384 mddev->safemode_delay = 0;
3386 unsigned long old_delay = mddev->safemode_delay;
3387 unsigned long new_delay = (msec*HZ)/1000;
3391 mddev->safemode_delay = new_delay;
3392 if (new_delay < old_delay || old_delay == 0)
3393 mod_timer(&mddev->safemode_timer, jiffies+1);
3397 static struct md_sysfs_entry md_safe_delay =
3398 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3401 level_show(struct mddev *mddev, char *page)
3403 struct md_personality *p;
3405 spin_lock(&mddev->lock);
3408 ret = sprintf(page, "%s\n", p->name);
3409 else if (mddev->clevel[0])
3410 ret = sprintf(page, "%s\n", mddev->clevel);
3411 else if (mddev->level != LEVEL_NONE)
3412 ret = sprintf(page, "%d\n", mddev->level);
3415 spin_unlock(&mddev->lock);
3420 level_store(struct mddev *mddev, const char *buf, size_t len)
3425 struct md_personality *pers, *oldpers;
3427 void *priv, *oldpriv;
3428 struct md_rdev *rdev;
3430 if (slen == 0 || slen >= sizeof(clevel))
3433 rv = mddev_lock(mddev);
3437 if (mddev->pers == NULL) {
3438 strncpy(mddev->clevel, buf, slen);
3439 if (mddev->clevel[slen-1] == '\n')
3441 mddev->clevel[slen] = 0;
3442 mddev->level = LEVEL_NONE;
3450 /* request to change the personality. Need to ensure:
3451 * - array is not engaged in resync/recovery/reshape
3452 * - old personality can be suspended
3453 * - new personality will access other array.
3457 if (mddev->sync_thread ||
3458 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3459 mddev->reshape_position != MaxSector ||
3460 mddev->sysfs_active)
3464 if (!mddev->pers->quiesce) {
3465 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3466 mdname(mddev), mddev->pers->name);
3470 /* Now find the new personality */
3471 strncpy(clevel, buf, slen);
3472 if (clevel[slen-1] == '\n')
3475 if (kstrtol(clevel, 10, &level))
3478 if (request_module("md-%s", clevel) != 0)
3479 request_module("md-level-%s", clevel);
3480 spin_lock(&pers_lock);
3481 pers = find_pers(level, clevel);
3482 if (!pers || !try_module_get(pers->owner)) {
3483 spin_unlock(&pers_lock);
3484 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3488 spin_unlock(&pers_lock);
3490 if (pers == mddev->pers) {
3491 /* Nothing to do! */
3492 module_put(pers->owner);
3496 if (!pers->takeover) {
3497 module_put(pers->owner);
3498 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3499 mdname(mddev), clevel);
3504 rdev_for_each(rdev, mddev)
3505 rdev->new_raid_disk = rdev->raid_disk;
3507 /* ->takeover must set new_* and/or delta_disks
3508 * if it succeeds, and may set them when it fails.
3510 priv = pers->takeover(mddev);
3512 mddev->new_level = mddev->level;
3513 mddev->new_layout = mddev->layout;
3514 mddev->new_chunk_sectors = mddev->chunk_sectors;
3515 mddev->raid_disks -= mddev->delta_disks;
3516 mddev->delta_disks = 0;
3517 mddev->reshape_backwards = 0;
3518 module_put(pers->owner);
3519 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3520 mdname(mddev), clevel);
3525 /* Looks like we have a winner */
3526 mddev_suspend(mddev);
3527 mddev_detach(mddev);
3529 spin_lock(&mddev->lock);
3530 oldpers = mddev->pers;
3531 oldpriv = mddev->private;
3533 mddev->private = priv;
3534 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3535 mddev->level = mddev->new_level;
3536 mddev->layout = mddev->new_layout;
3537 mddev->chunk_sectors = mddev->new_chunk_sectors;
3538 mddev->delta_disks = 0;
3539 mddev->reshape_backwards = 0;
3540 mddev->degraded = 0;
3541 spin_unlock(&mddev->lock);
3543 if (oldpers->sync_request == NULL &&
3545 /* We are converting from a no-redundancy array
3546 * to a redundancy array and metadata is managed
3547 * externally so we need to be sure that writes
3548 * won't block due to a need to transition
3550 * until external management is started.
3553 mddev->safemode_delay = 0;
3554 mddev->safemode = 0;
3557 oldpers->free(mddev, oldpriv);
3559 if (oldpers->sync_request == NULL &&
3560 pers->sync_request != NULL) {
3561 /* need to add the md_redundancy_group */
3562 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3564 "md: cannot register extra attributes for %s\n",
3566 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3568 if (oldpers->sync_request != NULL &&
3569 pers->sync_request == NULL) {
3570 /* need to remove the md_redundancy_group */
3571 if (mddev->to_remove == NULL)
3572 mddev->to_remove = &md_redundancy_group;
3575 rdev_for_each(rdev, mddev) {
3576 if (rdev->raid_disk < 0)
3578 if (rdev->new_raid_disk >= mddev->raid_disks)
3579 rdev->new_raid_disk = -1;
3580 if (rdev->new_raid_disk == rdev->raid_disk)
3582 sysfs_unlink_rdev(mddev, rdev);
3584 rdev_for_each(rdev, mddev) {
3585 if (rdev->raid_disk < 0)
3587 if (rdev->new_raid_disk == rdev->raid_disk)
3589 rdev->raid_disk = rdev->new_raid_disk;
3590 if (rdev->raid_disk < 0)
3591 clear_bit(In_sync, &rdev->flags);
3593 if (sysfs_link_rdev(mddev, rdev))
3594 printk(KERN_WARNING "md: cannot register rd%d"
3595 " for %s after level change\n",
3596 rdev->raid_disk, mdname(mddev));
3600 if (pers->sync_request == NULL) {
3601 /* this is now an array without redundancy, so
3602 * it must always be in_sync
3605 del_timer_sync(&mddev->safemode_timer);
3607 blk_set_stacking_limits(&mddev->queue->limits);
3609 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3610 mddev_resume(mddev);
3612 md_update_sb(mddev, 1);
3613 sysfs_notify(&mddev->kobj, NULL, "level");
3614 md_new_event(mddev);
3617 mddev_unlock(mddev);
3621 static struct md_sysfs_entry md_level =
3622 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3625 layout_show(struct mddev *mddev, char *page)
3627 /* just a number, not meaningful for all levels */
3628 if (mddev->reshape_position != MaxSector &&
3629 mddev->layout != mddev->new_layout)
3630 return sprintf(page, "%d (%d)\n",
3631 mddev->new_layout, mddev->layout);
3632 return sprintf(page, "%d\n", mddev->layout);
3636 layout_store(struct mddev *mddev, const char *buf, size_t len)
3641 err = kstrtouint(buf, 10, &n);
3644 err = mddev_lock(mddev);
3649 if (mddev->pers->check_reshape == NULL)
3654 mddev->new_layout = n;
3655 err = mddev->pers->check_reshape(mddev);
3657 mddev->new_layout = mddev->layout;
3660 mddev->new_layout = n;
3661 if (mddev->reshape_position == MaxSector)
3664 mddev_unlock(mddev);
3667 static struct md_sysfs_entry md_layout =
3668 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3671 raid_disks_show(struct mddev *mddev, char *page)
3673 if (mddev->raid_disks == 0)
3675 if (mddev->reshape_position != MaxSector &&
3676 mddev->delta_disks != 0)
3677 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3678 mddev->raid_disks - mddev->delta_disks);
3679 return sprintf(page, "%d\n", mddev->raid_disks);
3682 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3685 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3690 err = kstrtouint(buf, 10, &n);
3694 err = mddev_lock(mddev);
3698 err = update_raid_disks(mddev, n);
3699 else if (mddev->reshape_position != MaxSector) {
3700 struct md_rdev *rdev;
3701 int olddisks = mddev->raid_disks - mddev->delta_disks;
3704 rdev_for_each(rdev, mddev) {
3706 rdev->data_offset < rdev->new_data_offset)
3709 rdev->data_offset > rdev->new_data_offset)
3713 mddev->delta_disks = n - olddisks;
3714 mddev->raid_disks = n;
3715 mddev->reshape_backwards = (mddev->delta_disks < 0);
3717 mddev->raid_disks = n;
3719 mddev_unlock(mddev);
3720 return err ? err : len;
3722 static struct md_sysfs_entry md_raid_disks =
3723 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3726 chunk_size_show(struct mddev *mddev, char *page)
3728 if (mddev->reshape_position != MaxSector &&
3729 mddev->chunk_sectors != mddev->new_chunk_sectors)
3730 return sprintf(page, "%d (%d)\n",
3731 mddev->new_chunk_sectors << 9,
3732 mddev->chunk_sectors << 9);
3733 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3737 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3742 err = kstrtoul(buf, 10, &n);
3746 err = mddev_lock(mddev);
3750 if (mddev->pers->check_reshape == NULL)
3755 mddev->new_chunk_sectors = n >> 9;
3756 err = mddev->pers->check_reshape(mddev);
3758 mddev->new_chunk_sectors = mddev->chunk_sectors;
3761 mddev->new_chunk_sectors = n >> 9;
3762 if (mddev->reshape_position == MaxSector)
3763 mddev->chunk_sectors = n >> 9;
3765 mddev_unlock(mddev);
3768 static struct md_sysfs_entry md_chunk_size =
3769 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3772 resync_start_show(struct mddev *mddev, char *page)
3774 if (mddev->recovery_cp == MaxSector)
3775 return sprintf(page, "none\n");
3776 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3780 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3782 unsigned long long n;
3785 if (cmd_match(buf, "none"))
3788 err = kstrtoull(buf, 10, &n);
3791 if (n != (sector_t)n)
3795 err = mddev_lock(mddev);
3798 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3802 mddev->recovery_cp = n;
3804 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3806 mddev_unlock(mddev);
3809 static struct md_sysfs_entry md_resync_start =
3810 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3811 resync_start_show, resync_start_store);
3814 * The array state can be:
3817 * No devices, no size, no level
3818 * Equivalent to STOP_ARRAY ioctl
3820 * May have some settings, but array is not active
3821 * all IO results in error
3822 * When written, doesn't tear down array, but just stops it
3823 * suspended (not supported yet)
3824 * All IO requests will block. The array can be reconfigured.
3825 * Writing this, if accepted, will block until array is quiescent
3827 * no resync can happen. no superblocks get written.
3828 * write requests fail
3830 * like readonly, but behaves like 'clean' on a write request.
3832 * clean - no pending writes, but otherwise active.
3833 * When written to inactive array, starts without resync
3834 * If a write request arrives then
3835 * if metadata is known, mark 'dirty' and switch to 'active'.
3836 * if not known, block and switch to write-pending
3837 * If written to an active array that has pending writes, then fails.
3839 * fully active: IO and resync can be happening.
3840 * When written to inactive array, starts with resync
3843 * clean, but writes are blocked waiting for 'active' to be written.
3846 * like active, but no writes have been seen for a while (100msec).
3849 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3850 write_pending, active_idle, bad_word};
3851 static char *array_states[] = {
3852 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3853 "write-pending", "active-idle", NULL };
3855 static int match_word(const char *word, char **list)
3858 for (n=0; list[n]; n++)
3859 if (cmd_match(word, list[n]))
3865 array_state_show(struct mddev *mddev, char *page)
3867 enum array_state st = inactive;
3880 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3882 else if (mddev->safemode)
3888 if (list_empty(&mddev->disks) &&
3889 mddev->raid_disks == 0 &&
3890 mddev->dev_sectors == 0)
3895 return sprintf(page, "%s\n", array_states[st]);
3898 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3899 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3900 static int do_md_run(struct mddev *mddev);
3901 static int restart_array(struct mddev *mddev);
3904 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3907 enum array_state st = match_word(buf, array_states);
3909 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3910 /* don't take reconfig_mutex when toggling between
3913 spin_lock(&mddev->lock);
3915 restart_array(mddev);
3916 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3917 wake_up(&mddev->sb_wait);
3919 } else /* st == clean */ {
3920 restart_array(mddev);
3921 if (atomic_read(&mddev->writes_pending) == 0) {
3922 if (mddev->in_sync == 0) {
3924 if (mddev->safemode == 1)
3925 mddev->safemode = 0;
3926 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3932 spin_unlock(&mddev->lock);
3935 err = mddev_lock(mddev);
3943 /* stopping an active array */
3944 err = do_md_stop(mddev, 0, NULL);
3947 /* stopping an active array */
3949 err = do_md_stop(mddev, 2, NULL);
3951 err = 0; /* already inactive */
3954 break; /* not supported yet */
3957 err = md_set_readonly(mddev, NULL);
3960 set_disk_ro(mddev->gendisk, 1);
3961 err = do_md_run(mddev);
3967 err = md_set_readonly(mddev, NULL);
3968 else if (mddev->ro == 1)
3969 err = restart_array(mddev);
3972 set_disk_ro(mddev->gendisk, 0);
3976 err = do_md_run(mddev);
3981 err = restart_array(mddev);
3984 spin_lock(&mddev->lock);
3985 if (atomic_read(&mddev->writes_pending) == 0) {
3986 if (mddev->in_sync == 0) {
3988 if (mddev->safemode == 1)
3989 mddev->safemode = 0;
3990 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3995 spin_unlock(&mddev->lock);
4001 err = restart_array(mddev);
4004 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4005 wake_up(&mddev->sb_wait);
4009 set_disk_ro(mddev->gendisk, 0);
4010 err = do_md_run(mddev);
4015 /* these cannot be set */
4020 if (mddev->hold_active == UNTIL_IOCTL)
4021 mddev->hold_active = 0;
4022 sysfs_notify_dirent_safe(mddev->sysfs_state);
4024 mddev_unlock(mddev);
4027 static struct md_sysfs_entry md_array_state =
4028 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4031 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4032 return sprintf(page, "%d\n",
4033 atomic_read(&mddev->max_corr_read_errors));
4037 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4042 rv = kstrtouint(buf, 10, &n);
4045 atomic_set(&mddev->max_corr_read_errors, n);
4049 static struct md_sysfs_entry max_corr_read_errors =
4050 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4051 max_corrected_read_errors_store);
4054 null_show(struct mddev *mddev, char *page)
4060 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4062 /* buf must be %d:%d\n? giving major and minor numbers */
4063 /* The new device is added to the array.
4064 * If the array has a persistent superblock, we read the
4065 * superblock to initialise info and check validity.
4066 * Otherwise, only checking done is that in bind_rdev_to_array,
4067 * which mainly checks size.
4070 int major = simple_strtoul(buf, &e, 10);
4073 struct md_rdev *rdev;
4076 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4078 minor = simple_strtoul(e+1, &e, 10);
4079 if (*e && *e != '\n')
4081 dev = MKDEV(major, minor);
4082 if (major != MAJOR(dev) ||
4083 minor != MINOR(dev))
4086 flush_workqueue(md_misc_wq);
4088 err = mddev_lock(mddev);
4091 if (mddev->persistent) {
4092 rdev = md_import_device(dev, mddev->major_version,
4093 mddev->minor_version);
4094 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4095 struct md_rdev *rdev0
4096 = list_entry(mddev->disks.next,
4097 struct md_rdev, same_set);
4098 err = super_types[mddev->major_version]
4099 .load_super(rdev, rdev0, mddev->minor_version);
4103 } else if (mddev->external)
4104 rdev = md_import_device(dev, -2, -1);
4106 rdev = md_import_device(dev, -1, -1);
4109 mddev_unlock(mddev);
4110 return PTR_ERR(rdev);
4112 err = bind_rdev_to_array(rdev, mddev);
4116 mddev_unlock(mddev);
4117 return err ? err : len;
4120 static struct md_sysfs_entry md_new_device =
4121 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4124 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4127 unsigned long chunk, end_chunk;
4130 err = mddev_lock(mddev);
4135 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4137 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4138 if (buf == end) break;
4139 if (*end == '-') { /* range */
4141 end_chunk = simple_strtoul(buf, &end, 0);
4142 if (buf == end) break;
4144 if (*end && !isspace(*end)) break;
4145 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4146 buf = skip_spaces(end);
4148 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4150 mddev_unlock(mddev);
4154 static struct md_sysfs_entry md_bitmap =
4155 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4158 size_show(struct mddev *mddev, char *page)
4160 return sprintf(page, "%llu\n",
4161 (unsigned long long)mddev->dev_sectors / 2);
4164 static int update_size(struct mddev *mddev, sector_t num_sectors);
4167 size_store(struct mddev *mddev, const char *buf, size_t len)
4169 /* If array is inactive, we can reduce the component size, but
4170 * not increase it (except from 0).
4171 * If array is active, we can try an on-line resize
4174 int err = strict_blocks_to_sectors(buf, §ors);
4178 err = mddev_lock(mddev);
4182 err = update_size(mddev, sectors);
4183 md_update_sb(mddev, 1);
4185 if (mddev->dev_sectors == 0 ||
4186 mddev->dev_sectors > sectors)
4187 mddev->dev_sectors = sectors;
4191 mddev_unlock(mddev);
4192 return err ? err : len;
4195 static struct md_sysfs_entry md_size =
4196 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4198 /* Metadata version.
4200 * 'none' for arrays with no metadata (good luck...)
4201 * 'external' for arrays with externally managed metadata,
4202 * or N.M for internally known formats
4205 metadata_show(struct mddev *mddev, char *page)
4207 if (mddev->persistent)
4208 return sprintf(page, "%d.%d\n",
4209 mddev->major_version, mddev->minor_version);
4210 else if (mddev->external)
4211 return sprintf(page, "external:%s\n", mddev->metadata_type);
4213 return sprintf(page, "none\n");
4217 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4222 /* Changing the details of 'external' metadata is
4223 * always permitted. Otherwise there must be
4224 * no devices attached to the array.
4227 err = mddev_lock(mddev);
4231 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4233 else if (!list_empty(&mddev->disks))
4237 if (cmd_match(buf, "none")) {
4238 mddev->persistent = 0;
4239 mddev->external = 0;
4240 mddev->major_version = 0;
4241 mddev->minor_version = 90;
4244 if (strncmp(buf, "external:", 9) == 0) {
4245 size_t namelen = len-9;
4246 if (namelen >= sizeof(mddev->metadata_type))
4247 namelen = sizeof(mddev->metadata_type)-1;
4248 strncpy(mddev->metadata_type, buf+9, namelen);
4249 mddev->metadata_type[namelen] = 0;
4250 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4251 mddev->metadata_type[--namelen] = 0;
4252 mddev->persistent = 0;
4253 mddev->external = 1;
4254 mddev->major_version = 0;
4255 mddev->minor_version = 90;
4258 major = simple_strtoul(buf, &e, 10);
4260 if (e==buf || *e != '.')
4263 minor = simple_strtoul(buf, &e, 10);
4264 if (e==buf || (*e && *e != '\n') )
4267 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4269 mddev->major_version = major;
4270 mddev->minor_version = minor;
4271 mddev->persistent = 1;
4272 mddev->external = 0;
4275 mddev_unlock(mddev);
4279 static struct md_sysfs_entry md_metadata =
4280 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4283 action_show(struct mddev *mddev, char *page)
4285 char *type = "idle";
4286 unsigned long recovery = mddev->recovery;
4287 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4289 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4290 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4291 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4293 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4294 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4296 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4300 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4302 else if (mddev->reshape_position != MaxSector)
4305 return sprintf(page, "%s\n", type);
4309 action_store(struct mddev *mddev, const char *page, size_t len)
4311 if (!mddev->pers || !mddev->pers->sync_request)
4315 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4316 if (cmd_match(page, "frozen"))
4317 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4319 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4320 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4321 mddev_lock(mddev) == 0) {
4322 flush_workqueue(md_misc_wq);
4323 if (mddev->sync_thread) {
4324 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4325 md_reap_sync_thread(mddev);
4327 mddev_unlock(mddev);
4329 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4330 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4332 else if (cmd_match(page, "resync"))
4333 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4334 else if (cmd_match(page, "recover")) {
4335 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4336 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4337 } else if (cmd_match(page, "reshape")) {
4339 if (mddev->pers->start_reshape == NULL)
4341 err = mddev_lock(mddev);
4343 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4344 err = mddev->pers->start_reshape(mddev);
4345 mddev_unlock(mddev);
4349 sysfs_notify(&mddev->kobj, NULL, "degraded");
4351 if (cmd_match(page, "check"))
4352 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4353 else if (!cmd_match(page, "repair"))
4355 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4356 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4357 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4359 if (mddev->ro == 2) {
4360 /* A write to sync_action is enough to justify
4361 * canceling read-auto mode
4364 md_wakeup_thread(mddev->sync_thread);
4366 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4367 md_wakeup_thread(mddev->thread);
4368 sysfs_notify_dirent_safe(mddev->sysfs_action);
4372 static struct md_sysfs_entry md_scan_mode =
4373 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4376 last_sync_action_show(struct mddev *mddev, char *page)
4378 return sprintf(page, "%s\n", mddev->last_sync_action);
4381 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4384 mismatch_cnt_show(struct mddev *mddev, char *page)
4386 return sprintf(page, "%llu\n",
4387 (unsigned long long)
4388 atomic64_read(&mddev->resync_mismatches));
4391 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4394 sync_min_show(struct mddev *mddev, char *page)
4396 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4397 mddev->sync_speed_min ? "local": "system");
4401 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4406 if (strncmp(buf, "system", 6)==0) {
4409 rv = kstrtouint(buf, 10, &min);
4415 mddev->sync_speed_min = min;
4419 static struct md_sysfs_entry md_sync_min =
4420 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4423 sync_max_show(struct mddev *mddev, char *page)
4425 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4426 mddev->sync_speed_max ? "local": "system");
4430 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4435 if (strncmp(buf, "system", 6)==0) {
4438 rv = kstrtouint(buf, 10, &max);
4444 mddev->sync_speed_max = max;
4448 static struct md_sysfs_entry md_sync_max =
4449 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4452 degraded_show(struct mddev *mddev, char *page)
4454 return sprintf(page, "%d\n", mddev->degraded);
4456 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4459 sync_force_parallel_show(struct mddev *mddev, char *page)
4461 return sprintf(page, "%d\n", mddev->parallel_resync);
4465 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4469 if (kstrtol(buf, 10, &n))
4472 if (n != 0 && n != 1)
4475 mddev->parallel_resync = n;
4477 if (mddev->sync_thread)
4478 wake_up(&resync_wait);
4483 /* force parallel resync, even with shared block devices */
4484 static struct md_sysfs_entry md_sync_force_parallel =
4485 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4486 sync_force_parallel_show, sync_force_parallel_store);
4489 sync_speed_show(struct mddev *mddev, char *page)
4491 unsigned long resync, dt, db;
4492 if (mddev->curr_resync == 0)
4493 return sprintf(page, "none\n");
4494 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4495 dt = (jiffies - mddev->resync_mark) / HZ;
4497 db = resync - mddev->resync_mark_cnt;
4498 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4501 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4504 sync_completed_show(struct mddev *mddev, char *page)
4506 unsigned long long max_sectors, resync;
4508 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4509 return sprintf(page, "none\n");
4511 if (mddev->curr_resync == 1 ||
4512 mddev->curr_resync == 2)
4513 return sprintf(page, "delayed\n");
4515 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4516 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4517 max_sectors = mddev->resync_max_sectors;
4519 max_sectors = mddev->dev_sectors;
4521 resync = mddev->curr_resync_completed;
4522 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4525 static struct md_sysfs_entry md_sync_completed =
4526 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4529 min_sync_show(struct mddev *mddev, char *page)
4531 return sprintf(page, "%llu\n",
4532 (unsigned long long)mddev->resync_min);
4535 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4537 unsigned long long min;
4540 if (kstrtoull(buf, 10, &min))
4543 spin_lock(&mddev->lock);
4545 if (min > mddev->resync_max)
4549 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4552 /* Round down to multiple of 4K for safety */
4553 mddev->resync_min = round_down(min, 8);
4557 spin_unlock(&mddev->lock);
4561 static struct md_sysfs_entry md_min_sync =
4562 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4565 max_sync_show(struct mddev *mddev, char *page)
4567 if (mddev->resync_max == MaxSector)
4568 return sprintf(page, "max\n");
4570 return sprintf(page, "%llu\n",
4571 (unsigned long long)mddev->resync_max);
4574 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4577 spin_lock(&mddev->lock);
4578 if (strncmp(buf, "max", 3) == 0)
4579 mddev->resync_max = MaxSector;
4581 unsigned long long max;
4585 if (kstrtoull(buf, 10, &max))
4587 if (max < mddev->resync_min)
4591 if (max < mddev->resync_max &&
4593 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4596 /* Must be a multiple of chunk_size */
4597 chunk = mddev->chunk_sectors;
4599 sector_t temp = max;
4602 if (sector_div(temp, chunk))
4605 mddev->resync_max = max;
4607 wake_up(&mddev->recovery_wait);
4610 spin_unlock(&mddev->lock);
4614 static struct md_sysfs_entry md_max_sync =
4615 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4618 suspend_lo_show(struct mddev *mddev, char *page)
4620 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4624 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4626 unsigned long long old, new;
4629 err = kstrtoull(buf, 10, &new);
4632 if (new != (sector_t)new)
4635 err = mddev_lock(mddev);
4639 if (mddev->pers == NULL ||
4640 mddev->pers->quiesce == NULL)
4642 old = mddev->suspend_lo;
4643 mddev->suspend_lo = new;
4645 /* Shrinking suspended region */
4646 mddev->pers->quiesce(mddev, 2);
4648 /* Expanding suspended region - need to wait */
4649 mddev->pers->quiesce(mddev, 1);
4650 mddev->pers->quiesce(mddev, 0);
4654 mddev_unlock(mddev);
4657 static struct md_sysfs_entry md_suspend_lo =
4658 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4661 suspend_hi_show(struct mddev *mddev, char *page)
4663 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4667 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4669 unsigned long long old, new;
4672 err = kstrtoull(buf, 10, &new);
4675 if (new != (sector_t)new)
4678 err = mddev_lock(mddev);
4682 if (mddev->pers == NULL ||
4683 mddev->pers->quiesce == NULL)
4685 old = mddev->suspend_hi;
4686 mddev->suspend_hi = new;
4688 /* Shrinking suspended region */
4689 mddev->pers->quiesce(mddev, 2);
4691 /* Expanding suspended region - need to wait */
4692 mddev->pers->quiesce(mddev, 1);
4693 mddev->pers->quiesce(mddev, 0);
4697 mddev_unlock(mddev);
4700 static struct md_sysfs_entry md_suspend_hi =
4701 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4704 reshape_position_show(struct mddev *mddev, char *page)
4706 if (mddev->reshape_position != MaxSector)
4707 return sprintf(page, "%llu\n",
4708 (unsigned long long)mddev->reshape_position);
4709 strcpy(page, "none\n");
4714 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4716 struct md_rdev *rdev;
4717 unsigned long long new;
4720 err = kstrtoull(buf, 10, &new);
4723 if (new != (sector_t)new)
4725 err = mddev_lock(mddev);
4731 mddev->reshape_position = new;
4732 mddev->delta_disks = 0;
4733 mddev->reshape_backwards = 0;
4734 mddev->new_level = mddev->level;
4735 mddev->new_layout = mddev->layout;
4736 mddev->new_chunk_sectors = mddev->chunk_sectors;
4737 rdev_for_each(rdev, mddev)
4738 rdev->new_data_offset = rdev->data_offset;
4741 mddev_unlock(mddev);
4745 static struct md_sysfs_entry md_reshape_position =
4746 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4747 reshape_position_store);
4750 reshape_direction_show(struct mddev *mddev, char *page)
4752 return sprintf(page, "%s\n",
4753 mddev->reshape_backwards ? "backwards" : "forwards");
4757 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4762 if (cmd_match(buf, "forwards"))
4764 else if (cmd_match(buf, "backwards"))
4768 if (mddev->reshape_backwards == backwards)
4771 err = mddev_lock(mddev);
4774 /* check if we are allowed to change */
4775 if (mddev->delta_disks)
4777 else if (mddev->persistent &&
4778 mddev->major_version == 0)
4781 mddev->reshape_backwards = backwards;
4782 mddev_unlock(mddev);
4786 static struct md_sysfs_entry md_reshape_direction =
4787 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4788 reshape_direction_store);
4791 array_size_show(struct mddev *mddev, char *page)
4793 if (mddev->external_size)
4794 return sprintf(page, "%llu\n",
4795 (unsigned long long)mddev->array_sectors/2);
4797 return sprintf(page, "default\n");
4801 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4806 err = mddev_lock(mddev);
4810 if (strncmp(buf, "default", 7) == 0) {
4812 sectors = mddev->pers->size(mddev, 0, 0);
4814 sectors = mddev->array_sectors;
4816 mddev->external_size = 0;
4818 if (strict_blocks_to_sectors(buf, §ors) < 0)
4820 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4823 mddev->external_size = 1;
4827 mddev->array_sectors = sectors;
4829 set_capacity(mddev->gendisk, mddev->array_sectors);
4830 revalidate_disk(mddev->gendisk);
4833 mddev_unlock(mddev);
4837 static struct md_sysfs_entry md_array_size =
4838 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4841 static struct attribute *md_default_attrs[] = {
4844 &md_raid_disks.attr,
4845 &md_chunk_size.attr,
4847 &md_resync_start.attr,
4849 &md_new_device.attr,
4850 &md_safe_delay.attr,
4851 &md_array_state.attr,
4852 &md_reshape_position.attr,
4853 &md_reshape_direction.attr,
4854 &md_array_size.attr,
4855 &max_corr_read_errors.attr,
4859 static struct attribute *md_redundancy_attrs[] = {
4861 &md_last_scan_mode.attr,
4862 &md_mismatches.attr,
4865 &md_sync_speed.attr,
4866 &md_sync_force_parallel.attr,
4867 &md_sync_completed.attr,
4870 &md_suspend_lo.attr,
4871 &md_suspend_hi.attr,
4876 static struct attribute_group md_redundancy_group = {
4878 .attrs = md_redundancy_attrs,
4882 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4884 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4885 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4890 spin_lock(&all_mddevs_lock);
4891 if (list_empty(&mddev->all_mddevs)) {
4892 spin_unlock(&all_mddevs_lock);
4896 spin_unlock(&all_mddevs_lock);
4898 rv = entry->show(mddev, page);
4904 md_attr_store(struct kobject *kobj, struct attribute *attr,
4905 const char *page, size_t length)
4907 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4908 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4913 if (!capable(CAP_SYS_ADMIN))
4915 spin_lock(&all_mddevs_lock);
4916 if (list_empty(&mddev->all_mddevs)) {
4917 spin_unlock(&all_mddevs_lock);
4921 spin_unlock(&all_mddevs_lock);
4922 rv = entry->store(mddev, page, length);
4927 static void md_free(struct kobject *ko)
4929 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4931 if (mddev->sysfs_state)
4932 sysfs_put(mddev->sysfs_state);
4935 blk_cleanup_queue(mddev->queue);
4936 if (mddev->gendisk) {
4937 del_gendisk(mddev->gendisk);
4938 put_disk(mddev->gendisk);
4944 static const struct sysfs_ops md_sysfs_ops = {
4945 .show = md_attr_show,
4946 .store = md_attr_store,
4948 static struct kobj_type md_ktype = {
4950 .sysfs_ops = &md_sysfs_ops,
4951 .default_attrs = md_default_attrs,
4956 static void mddev_delayed_delete(struct work_struct *ws)
4958 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4960 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4961 kobject_del(&mddev->kobj);
4962 kobject_put(&mddev->kobj);
4965 static int md_alloc(dev_t dev, char *name)
4967 static DEFINE_MUTEX(disks_mutex);
4968 struct mddev *mddev = mddev_find(dev);
4969 struct gendisk *disk;
4978 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4979 shift = partitioned ? MdpMinorShift : 0;
4980 unit = MINOR(mddev->unit) >> shift;
4982 /* wait for any previous instance of this device to be
4983 * completely removed (mddev_delayed_delete).
4985 flush_workqueue(md_misc_wq);
4987 mutex_lock(&disks_mutex);
4993 /* Need to ensure that 'name' is not a duplicate.
4995 struct mddev *mddev2;
4996 spin_lock(&all_mddevs_lock);
4998 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4999 if (mddev2->gendisk &&
5000 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5001 spin_unlock(&all_mddevs_lock);
5004 spin_unlock(&all_mddevs_lock);
5008 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5011 mddev->queue->queuedata = mddev;
5013 blk_queue_make_request(mddev->queue, md_make_request);
5014 blk_set_stacking_limits(&mddev->queue->limits);
5016 disk = alloc_disk(1 << shift);
5018 blk_cleanup_queue(mddev->queue);
5019 mddev->queue = NULL;
5022 disk->major = MAJOR(mddev->unit);
5023 disk->first_minor = unit << shift;
5025 strcpy(disk->disk_name, name);
5026 else if (partitioned)
5027 sprintf(disk->disk_name, "md_d%d", unit);
5029 sprintf(disk->disk_name, "md%d", unit);
5030 disk->fops = &md_fops;
5031 disk->private_data = mddev;
5032 disk->queue = mddev->queue;
5033 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5034 /* Allow extended partitions. This makes the
5035 * 'mdp' device redundant, but we can't really
5038 disk->flags |= GENHD_FL_EXT_DEVT;
5039 mddev->gendisk = disk;
5040 /* As soon as we call add_disk(), another thread could get
5041 * through to md_open, so make sure it doesn't get too far
5043 mutex_lock(&mddev->open_mutex);
5046 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5047 &disk_to_dev(disk)->kobj, "%s", "md");
5049 /* This isn't possible, but as kobject_init_and_add is marked
5050 * __must_check, we must do something with the result
5052 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5056 if (mddev->kobj.sd &&
5057 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5058 printk(KERN_DEBUG "pointless warning\n");
5059 mutex_unlock(&mddev->open_mutex);
5061 mutex_unlock(&disks_mutex);
5062 if (!error && mddev->kobj.sd) {
5063 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5064 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5070 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5072 md_alloc(dev, NULL);
5076 static int add_named_array(const char *val, struct kernel_param *kp)
5078 /* val must be "md_*" where * is not all digits.
5079 * We allocate an array with a large free minor number, and
5080 * set the name to val. val must not already be an active name.
5082 int len = strlen(val);
5083 char buf[DISK_NAME_LEN];
5085 while (len && val[len-1] == '\n')
5087 if (len >= DISK_NAME_LEN)
5089 strlcpy(buf, val, len+1);
5090 if (strncmp(buf, "md_", 3) != 0)
5092 return md_alloc(0, buf);
5095 static void md_safemode_timeout(unsigned long data)
5097 struct mddev *mddev = (struct mddev *) data;
5099 if (!atomic_read(&mddev->writes_pending)) {
5100 mddev->safemode = 1;
5101 if (mddev->external)
5102 sysfs_notify_dirent_safe(mddev->sysfs_state);
5104 md_wakeup_thread(mddev->thread);
5107 static int start_dirty_degraded;
5109 int md_run(struct mddev *mddev)
5112 struct md_rdev *rdev;
5113 struct md_personality *pers;
5115 if (list_empty(&mddev->disks))
5116 /* cannot run an array with no devices.. */
5121 /* Cannot run until previous stop completes properly */
5122 if (mddev->sysfs_active)
5126 * Analyze all RAID superblock(s)
5128 if (!mddev->raid_disks) {
5129 if (!mddev->persistent)
5134 if (mddev->level != LEVEL_NONE)
5135 request_module("md-level-%d", mddev->level);
5136 else if (mddev->clevel[0])
5137 request_module("md-%s", mddev->clevel);
5140 * Drop all container device buffers, from now on
5141 * the only valid external interface is through the md
5144 rdev_for_each(rdev, mddev) {
5145 if (test_bit(Faulty, &rdev->flags))
5147 sync_blockdev(rdev->bdev);
5148 invalidate_bdev(rdev->bdev);
5150 /* perform some consistency tests on the device.
5151 * We don't want the data to overlap the metadata,
5152 * Internal Bitmap issues have been handled elsewhere.
5154 if (rdev->meta_bdev) {
5155 /* Nothing to check */;
5156 } else if (rdev->data_offset < rdev->sb_start) {
5157 if (mddev->dev_sectors &&
5158 rdev->data_offset + mddev->dev_sectors
5160 printk("md: %s: data overlaps metadata\n",
5165 if (rdev->sb_start + rdev->sb_size/512
5166 > rdev->data_offset) {
5167 printk("md: %s: metadata overlaps data\n",
5172 sysfs_notify_dirent_safe(rdev->sysfs_state);
5175 if (mddev->bio_set == NULL)
5176 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5178 spin_lock(&pers_lock);
5179 pers = find_pers(mddev->level, mddev->clevel);
5180 if (!pers || !try_module_get(pers->owner)) {
5181 spin_unlock(&pers_lock);
5182 if (mddev->level != LEVEL_NONE)
5183 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5186 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5190 spin_unlock(&pers_lock);
5191 if (mddev->level != pers->level) {
5192 mddev->level = pers->level;
5193 mddev->new_level = pers->level;
5195 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5197 if (mddev->reshape_position != MaxSector &&
5198 pers->start_reshape == NULL) {
5199 /* This personality cannot handle reshaping... */
5200 module_put(pers->owner);
5204 if (pers->sync_request) {
5205 /* Warn if this is a potentially silly
5208 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5209 struct md_rdev *rdev2;
5212 rdev_for_each(rdev, mddev)
5213 rdev_for_each(rdev2, mddev) {
5215 rdev->bdev->bd_contains ==
5216 rdev2->bdev->bd_contains) {
5218 "%s: WARNING: %s appears to be"
5219 " on the same physical disk as"
5222 bdevname(rdev->bdev,b),
5223 bdevname(rdev2->bdev,b2));
5230 "True protection against single-disk"
5231 " failure might be compromised.\n");
5234 mddev->recovery = 0;
5235 /* may be over-ridden by personality */
5236 mddev->resync_max_sectors = mddev->dev_sectors;
5238 mddev->ok_start_degraded = start_dirty_degraded;
5240 if (start_readonly && mddev->ro == 0)
5241 mddev->ro = 2; /* read-only, but switch on first write */
5243 err = pers->run(mddev);
5245 printk(KERN_ERR "md: pers->run() failed ...\n");
5246 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5247 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5248 " but 'external_size' not in effect?\n", __func__);
5250 "md: invalid array_size %llu > default size %llu\n",
5251 (unsigned long long)mddev->array_sectors / 2,
5252 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5255 if (err == 0 && pers->sync_request &&
5256 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5257 struct bitmap *bitmap;
5259 bitmap = bitmap_create(mddev, -1);
5260 if (IS_ERR(bitmap)) {
5261 err = PTR_ERR(bitmap);
5262 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5263 mdname(mddev), err);
5265 mddev->bitmap = bitmap;
5269 mddev_detach(mddev);
5271 pers->free(mddev, mddev->private);
5272 mddev->private = NULL;
5273 module_put(pers->owner);
5274 bitmap_destroy(mddev);
5278 mddev->queue->backing_dev_info.congested_data = mddev;
5279 mddev->queue->backing_dev_info.congested_fn = md_congested;
5281 if (pers->sync_request) {
5282 if (mddev->kobj.sd &&
5283 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5285 "md: cannot register extra attributes for %s\n",
5287 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5288 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5291 atomic_set(&mddev->writes_pending,0);
5292 atomic_set(&mddev->max_corr_read_errors,
5293 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5294 mddev->safemode = 0;
5295 if (mddev_is_clustered(mddev))
5296 mddev->safemode_delay = 0;
5298 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5301 spin_lock(&mddev->lock);
5304 spin_unlock(&mddev->lock);
5305 rdev_for_each(rdev, mddev)
5306 if (rdev->raid_disk >= 0)
5307 if (sysfs_link_rdev(mddev, rdev))
5308 /* failure here is OK */;
5310 if (mddev->degraded && !mddev->ro)
5311 /* This ensures that recovering status is reported immediately
5312 * via sysfs - until a lack of spares is confirmed.
5314 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5315 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5317 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5318 md_update_sb(mddev, 0);
5320 md_new_event(mddev);
5321 sysfs_notify_dirent_safe(mddev->sysfs_state);
5322 sysfs_notify_dirent_safe(mddev->sysfs_action);
5323 sysfs_notify(&mddev->kobj, NULL, "degraded");
5326 EXPORT_SYMBOL_GPL(md_run);
5328 static int do_md_run(struct mddev *mddev)
5332 err = md_run(mddev);
5335 err = bitmap_load(mddev);
5337 bitmap_destroy(mddev);
5341 if (mddev_is_clustered(mddev))
5342 md_allow_write(mddev);
5344 md_wakeup_thread(mddev->thread);
5345 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5347 set_capacity(mddev->gendisk, mddev->array_sectors);
5348 revalidate_disk(mddev->gendisk);
5350 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5355 static int restart_array(struct mddev *mddev)
5357 struct gendisk *disk = mddev->gendisk;
5359 /* Complain if it has no devices */
5360 if (list_empty(&mddev->disks))
5366 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5367 struct md_rdev *rdev;
5368 bool has_journal = false;
5371 rdev_for_each_rcu(rdev, mddev) {
5372 if (test_bit(Journal, &rdev->flags) &&
5373 !test_bit(Faulty, &rdev->flags)) {
5380 /* Don't restart rw with journal missing/faulty */
5385 mddev->safemode = 0;
5387 set_disk_ro(disk, 0);
5388 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5390 /* Kick recovery or resync if necessary */
5391 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5392 md_wakeup_thread(mddev->thread);
5393 md_wakeup_thread(mddev->sync_thread);
5394 sysfs_notify_dirent_safe(mddev->sysfs_state);
5398 static void md_clean(struct mddev *mddev)
5400 mddev->array_sectors = 0;
5401 mddev->external_size = 0;
5402 mddev->dev_sectors = 0;
5403 mddev->raid_disks = 0;
5404 mddev->recovery_cp = 0;
5405 mddev->resync_min = 0;
5406 mddev->resync_max = MaxSector;
5407 mddev->reshape_position = MaxSector;
5408 mddev->external = 0;
5409 mddev->persistent = 0;
5410 mddev->level = LEVEL_NONE;
5411 mddev->clevel[0] = 0;
5414 mddev->metadata_type[0] = 0;
5415 mddev->chunk_sectors = 0;
5416 mddev->ctime = mddev->utime = 0;
5418 mddev->max_disks = 0;
5420 mddev->can_decrease_events = 0;
5421 mddev->delta_disks = 0;
5422 mddev->reshape_backwards = 0;
5423 mddev->new_level = LEVEL_NONE;
5424 mddev->new_layout = 0;
5425 mddev->new_chunk_sectors = 0;
5426 mddev->curr_resync = 0;
5427 atomic64_set(&mddev->resync_mismatches, 0);
5428 mddev->suspend_lo = mddev->suspend_hi = 0;
5429 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5430 mddev->recovery = 0;
5433 mddev->degraded = 0;
5434 mddev->safemode = 0;
5435 mddev->private = NULL;
5436 mddev->bitmap_info.offset = 0;
5437 mddev->bitmap_info.default_offset = 0;
5438 mddev->bitmap_info.default_space = 0;
5439 mddev->bitmap_info.chunksize = 0;
5440 mddev->bitmap_info.daemon_sleep = 0;
5441 mddev->bitmap_info.max_write_behind = 0;
5444 static void __md_stop_writes(struct mddev *mddev)
5446 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5447 flush_workqueue(md_misc_wq);
5448 if (mddev->sync_thread) {
5449 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5450 md_reap_sync_thread(mddev);
5453 del_timer_sync(&mddev->safemode_timer);
5455 bitmap_flush(mddev);
5456 md_super_wait(mddev);
5458 if (mddev->ro == 0 &&
5459 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5460 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5461 /* mark array as shutdown cleanly */
5462 if (!mddev_is_clustered(mddev))
5464 md_update_sb(mddev, 1);
5468 void md_stop_writes(struct mddev *mddev)
5470 mddev_lock_nointr(mddev);
5471 __md_stop_writes(mddev);
5472 mddev_unlock(mddev);
5474 EXPORT_SYMBOL_GPL(md_stop_writes);
5476 static void mddev_detach(struct mddev *mddev)
5478 struct bitmap *bitmap = mddev->bitmap;
5479 /* wait for behind writes to complete */
5480 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5481 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5483 /* need to kick something here to make sure I/O goes? */
5484 wait_event(bitmap->behind_wait,
5485 atomic_read(&bitmap->behind_writes) == 0);
5487 if (mddev->pers && mddev->pers->quiesce) {
5488 mddev->pers->quiesce(mddev, 1);
5489 mddev->pers->quiesce(mddev, 0);
5491 md_unregister_thread(&mddev->thread);
5493 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5496 static void __md_stop(struct mddev *mddev)
5498 struct md_personality *pers = mddev->pers;
5499 mddev_detach(mddev);
5500 /* Ensure ->event_work is done */
5501 flush_workqueue(md_misc_wq);
5502 spin_lock(&mddev->lock);
5505 spin_unlock(&mddev->lock);
5506 pers->free(mddev, mddev->private);
5507 mddev->private = NULL;
5508 if (pers->sync_request && mddev->to_remove == NULL)
5509 mddev->to_remove = &md_redundancy_group;
5510 module_put(pers->owner);
5511 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5514 void md_stop(struct mddev *mddev)
5516 /* stop the array and free an attached data structures.
5517 * This is called from dm-raid
5520 bitmap_destroy(mddev);
5522 bioset_free(mddev->bio_set);
5525 EXPORT_SYMBOL_GPL(md_stop);
5527 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5532 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5534 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5535 md_wakeup_thread(mddev->thread);
5537 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5538 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5539 if (mddev->sync_thread)
5540 /* Thread might be blocked waiting for metadata update
5541 * which will now never happen */
5542 wake_up_process(mddev->sync_thread->tsk);
5544 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5546 mddev_unlock(mddev);
5547 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5549 wait_event(mddev->sb_wait,
5550 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5551 mddev_lock_nointr(mddev);
5553 mutex_lock(&mddev->open_mutex);
5554 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5555 mddev->sync_thread ||
5556 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5557 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5558 printk("md: %s still in use.\n",mdname(mddev));
5560 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5561 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5562 md_wakeup_thread(mddev->thread);
5568 __md_stop_writes(mddev);
5574 set_disk_ro(mddev->gendisk, 1);
5575 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5576 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5577 md_wakeup_thread(mddev->thread);
5578 sysfs_notify_dirent_safe(mddev->sysfs_state);
5582 mutex_unlock(&mddev->open_mutex);
5587 * 0 - completely stop and dis-assemble array
5588 * 2 - stop but do not disassemble array
5590 static int do_md_stop(struct mddev *mddev, int mode,
5591 struct block_device *bdev)
5593 struct gendisk *disk = mddev->gendisk;
5594 struct md_rdev *rdev;
5597 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5599 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5600 md_wakeup_thread(mddev->thread);
5602 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5603 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5604 if (mddev->sync_thread)
5605 /* Thread might be blocked waiting for metadata update
5606 * which will now never happen */
5607 wake_up_process(mddev->sync_thread->tsk);
5609 mddev_unlock(mddev);
5610 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5611 !test_bit(MD_RECOVERY_RUNNING,
5612 &mddev->recovery)));
5613 mddev_lock_nointr(mddev);
5615 mutex_lock(&mddev->open_mutex);
5616 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5617 mddev->sysfs_active ||
5618 mddev->sync_thread ||
5619 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5620 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5621 printk("md: %s still in use.\n",mdname(mddev));
5622 mutex_unlock(&mddev->open_mutex);
5624 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5625 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5626 md_wakeup_thread(mddev->thread);
5632 set_disk_ro(disk, 0);
5634 __md_stop_writes(mddev);
5636 mddev->queue->backing_dev_info.congested_fn = NULL;
5638 /* tell userspace to handle 'inactive' */
5639 sysfs_notify_dirent_safe(mddev->sysfs_state);
5641 rdev_for_each(rdev, mddev)
5642 if (rdev->raid_disk >= 0)
5643 sysfs_unlink_rdev(mddev, rdev);
5645 set_capacity(disk, 0);
5646 mutex_unlock(&mddev->open_mutex);
5648 revalidate_disk(disk);
5653 mutex_unlock(&mddev->open_mutex);
5655 * Free resources if final stop
5658 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5660 bitmap_destroy(mddev);
5661 if (mddev->bitmap_info.file) {
5662 struct file *f = mddev->bitmap_info.file;
5663 spin_lock(&mddev->lock);
5664 mddev->bitmap_info.file = NULL;
5665 spin_unlock(&mddev->lock);
5668 mddev->bitmap_info.offset = 0;
5670 export_array(mddev);
5673 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5674 if (mddev->hold_active == UNTIL_STOP)
5675 mddev->hold_active = 0;
5677 md_new_event(mddev);
5678 sysfs_notify_dirent_safe(mddev->sysfs_state);
5683 static void autorun_array(struct mddev *mddev)
5685 struct md_rdev *rdev;
5688 if (list_empty(&mddev->disks))
5691 printk(KERN_INFO "md: running: ");
5693 rdev_for_each(rdev, mddev) {
5694 char b[BDEVNAME_SIZE];
5695 printk("<%s>", bdevname(rdev->bdev,b));
5699 err = do_md_run(mddev);
5701 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5702 do_md_stop(mddev, 0, NULL);
5707 * lets try to run arrays based on all disks that have arrived
5708 * until now. (those are in pending_raid_disks)
5710 * the method: pick the first pending disk, collect all disks with
5711 * the same UUID, remove all from the pending list and put them into
5712 * the 'same_array' list. Then order this list based on superblock
5713 * update time (freshest comes first), kick out 'old' disks and
5714 * compare superblocks. If everything's fine then run it.
5716 * If "unit" is allocated, then bump its reference count
5718 static void autorun_devices(int part)
5720 struct md_rdev *rdev0, *rdev, *tmp;
5721 struct mddev *mddev;
5722 char b[BDEVNAME_SIZE];
5724 printk(KERN_INFO "md: autorun ...\n");
5725 while (!list_empty(&pending_raid_disks)) {
5728 LIST_HEAD(candidates);
5729 rdev0 = list_entry(pending_raid_disks.next,
5730 struct md_rdev, same_set);
5732 printk(KERN_INFO "md: considering %s ...\n",
5733 bdevname(rdev0->bdev,b));
5734 INIT_LIST_HEAD(&candidates);
5735 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5736 if (super_90_load(rdev, rdev0, 0) >= 0) {
5737 printk(KERN_INFO "md: adding %s ...\n",
5738 bdevname(rdev->bdev,b));
5739 list_move(&rdev->same_set, &candidates);
5742 * now we have a set of devices, with all of them having
5743 * mostly sane superblocks. It's time to allocate the
5747 dev = MKDEV(mdp_major,
5748 rdev0->preferred_minor << MdpMinorShift);
5749 unit = MINOR(dev) >> MdpMinorShift;
5751 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5754 if (rdev0->preferred_minor != unit) {
5755 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5756 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5760 md_probe(dev, NULL, NULL);
5761 mddev = mddev_find(dev);
5762 if (!mddev || !mddev->gendisk) {
5766 "md: cannot allocate memory for md drive.\n");
5769 if (mddev_lock(mddev))
5770 printk(KERN_WARNING "md: %s locked, cannot run\n",
5772 else if (mddev->raid_disks || mddev->major_version
5773 || !list_empty(&mddev->disks)) {
5775 "md: %s already running, cannot run %s\n",
5776 mdname(mddev), bdevname(rdev0->bdev,b));
5777 mddev_unlock(mddev);
5779 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5780 mddev->persistent = 1;
5781 rdev_for_each_list(rdev, tmp, &candidates) {
5782 list_del_init(&rdev->same_set);
5783 if (bind_rdev_to_array(rdev, mddev))
5786 autorun_array(mddev);
5787 mddev_unlock(mddev);
5789 /* on success, candidates will be empty, on error
5792 rdev_for_each_list(rdev, tmp, &candidates) {
5793 list_del_init(&rdev->same_set);
5798 printk(KERN_INFO "md: ... autorun DONE.\n");
5800 #endif /* !MODULE */
5802 static int get_version(void __user *arg)
5806 ver.major = MD_MAJOR_VERSION;
5807 ver.minor = MD_MINOR_VERSION;
5808 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5810 if (copy_to_user(arg, &ver, sizeof(ver)))
5816 static int get_array_info(struct mddev *mddev, void __user *arg)
5818 mdu_array_info_t info;
5819 int nr,working,insync,failed,spare;
5820 struct md_rdev *rdev;
5822 nr = working = insync = failed = spare = 0;
5824 rdev_for_each_rcu(rdev, mddev) {
5826 if (test_bit(Faulty, &rdev->flags))
5830 if (test_bit(In_sync, &rdev->flags))
5838 info.major_version = mddev->major_version;
5839 info.minor_version = mddev->minor_version;
5840 info.patch_version = MD_PATCHLEVEL_VERSION;
5841 info.ctime = mddev->ctime;
5842 info.level = mddev->level;
5843 info.size = mddev->dev_sectors / 2;
5844 if (info.size != mddev->dev_sectors / 2) /* overflow */
5847 info.raid_disks = mddev->raid_disks;
5848 info.md_minor = mddev->md_minor;
5849 info.not_persistent= !mddev->persistent;
5851 info.utime = mddev->utime;
5854 info.state = (1<<MD_SB_CLEAN);
5855 if (mddev->bitmap && mddev->bitmap_info.offset)
5856 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5857 if (mddev_is_clustered(mddev))
5858 info.state |= (1<<MD_SB_CLUSTERED);
5859 info.active_disks = insync;
5860 info.working_disks = working;
5861 info.failed_disks = failed;
5862 info.spare_disks = spare;
5864 info.layout = mddev->layout;
5865 info.chunk_size = mddev->chunk_sectors << 9;
5867 if (copy_to_user(arg, &info, sizeof(info)))
5873 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5875 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5879 file = kzalloc(sizeof(*file), GFP_NOIO);
5884 spin_lock(&mddev->lock);
5885 /* bitmap enabled */
5886 if (mddev->bitmap_info.file) {
5887 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5888 sizeof(file->pathname));
5892 memmove(file->pathname, ptr,
5893 sizeof(file->pathname)-(ptr-file->pathname));
5895 spin_unlock(&mddev->lock);
5898 copy_to_user(arg, file, sizeof(*file)))
5905 static int get_disk_info(struct mddev *mddev, void __user * arg)
5907 mdu_disk_info_t info;
5908 struct md_rdev *rdev;
5910 if (copy_from_user(&info, arg, sizeof(info)))
5914 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5916 info.major = MAJOR(rdev->bdev->bd_dev);
5917 info.minor = MINOR(rdev->bdev->bd_dev);
5918 info.raid_disk = rdev->raid_disk;
5920 if (test_bit(Faulty, &rdev->flags))
5921 info.state |= (1<<MD_DISK_FAULTY);
5922 else if (test_bit(In_sync, &rdev->flags)) {
5923 info.state |= (1<<MD_DISK_ACTIVE);
5924 info.state |= (1<<MD_DISK_SYNC);
5926 if (test_bit(Journal, &rdev->flags))
5927 info.state |= (1<<MD_DISK_JOURNAL);
5928 if (test_bit(WriteMostly, &rdev->flags))
5929 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5931 info.major = info.minor = 0;
5932 info.raid_disk = -1;
5933 info.state = (1<<MD_DISK_REMOVED);
5937 if (copy_to_user(arg, &info, sizeof(info)))
5943 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5945 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5946 struct md_rdev *rdev;
5947 dev_t dev = MKDEV(info->major,info->minor);
5949 if (mddev_is_clustered(mddev) &&
5950 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5951 pr_err("%s: Cannot add to clustered mddev.\n",
5956 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5959 if (!mddev->raid_disks) {
5961 /* expecting a device which has a superblock */
5962 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5965 "md: md_import_device returned %ld\n",
5967 return PTR_ERR(rdev);
5969 if (!list_empty(&mddev->disks)) {
5970 struct md_rdev *rdev0
5971 = list_entry(mddev->disks.next,
5972 struct md_rdev, same_set);
5973 err = super_types[mddev->major_version]
5974 .load_super(rdev, rdev0, mddev->minor_version);
5977 "md: %s has different UUID to %s\n",
5978 bdevname(rdev->bdev,b),
5979 bdevname(rdev0->bdev,b2));
5984 err = bind_rdev_to_array(rdev, mddev);
5991 * add_new_disk can be used once the array is assembled
5992 * to add "hot spares". They must already have a superblock
5997 if (!mddev->pers->hot_add_disk) {
5999 "%s: personality does not support diskops!\n",
6003 if (mddev->persistent)
6004 rdev = md_import_device(dev, mddev->major_version,
6005 mddev->minor_version);
6007 rdev = md_import_device(dev, -1, -1);
6010 "md: md_import_device returned %ld\n",
6012 return PTR_ERR(rdev);
6014 /* set saved_raid_disk if appropriate */
6015 if (!mddev->persistent) {
6016 if (info->state & (1<<MD_DISK_SYNC) &&
6017 info->raid_disk < mddev->raid_disks) {
6018 rdev->raid_disk = info->raid_disk;
6019 set_bit(In_sync, &rdev->flags);
6020 clear_bit(Bitmap_sync, &rdev->flags);
6022 rdev->raid_disk = -1;
6023 rdev->saved_raid_disk = rdev->raid_disk;
6025 super_types[mddev->major_version].
6026 validate_super(mddev, rdev);
6027 if ((info->state & (1<<MD_DISK_SYNC)) &&
6028 rdev->raid_disk != info->raid_disk) {
6029 /* This was a hot-add request, but events doesn't
6030 * match, so reject it.
6036 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6037 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6038 set_bit(WriteMostly, &rdev->flags);
6040 clear_bit(WriteMostly, &rdev->flags);
6042 if (info->state & (1<<MD_DISK_JOURNAL))
6043 set_bit(Journal, &rdev->flags);
6045 * check whether the device shows up in other nodes
6047 if (mddev_is_clustered(mddev)) {
6048 if (info->state & (1 << MD_DISK_CANDIDATE))
6049 set_bit(Candidate, &rdev->flags);
6050 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6051 /* --add initiated by this node */
6052 err = md_cluster_ops->add_new_disk(mddev, rdev);
6060 rdev->raid_disk = -1;
6061 err = bind_rdev_to_array(rdev, mddev);
6066 if (mddev_is_clustered(mddev)) {
6067 if (info->state & (1 << MD_DISK_CANDIDATE))
6068 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6071 md_cluster_ops->add_new_disk_cancel(mddev);
6073 err = add_bound_rdev(rdev);
6077 err = add_bound_rdev(rdev);
6082 /* otherwise, add_new_disk is only allowed
6083 * for major_version==0 superblocks
6085 if (mddev->major_version != 0) {
6086 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6091 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6093 rdev = md_import_device(dev, -1, 0);
6096 "md: error, md_import_device() returned %ld\n",
6098 return PTR_ERR(rdev);
6100 rdev->desc_nr = info->number;
6101 if (info->raid_disk < mddev->raid_disks)
6102 rdev->raid_disk = info->raid_disk;
6104 rdev->raid_disk = -1;
6106 if (rdev->raid_disk < mddev->raid_disks)
6107 if (info->state & (1<<MD_DISK_SYNC))
6108 set_bit(In_sync, &rdev->flags);
6110 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6111 set_bit(WriteMostly, &rdev->flags);
6113 if (!mddev->persistent) {
6114 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6115 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6117 rdev->sb_start = calc_dev_sboffset(rdev);
6118 rdev->sectors = rdev->sb_start;
6120 err = bind_rdev_to_array(rdev, mddev);
6130 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6132 char b[BDEVNAME_SIZE];
6133 struct md_rdev *rdev;
6136 rdev = find_rdev(mddev, dev);
6140 if (mddev_is_clustered(mddev))
6141 ret = md_cluster_ops->metadata_update_start(mddev);
6143 if (rdev->raid_disk < 0)
6146 clear_bit(Blocked, &rdev->flags);
6147 remove_and_add_spares(mddev, rdev);
6149 if (rdev->raid_disk >= 0)
6153 if (mddev_is_clustered(mddev) && ret == 0)
6154 md_cluster_ops->remove_disk(mddev, rdev);
6156 md_kick_rdev_from_array(rdev);
6157 md_update_sb(mddev, 1);
6158 md_new_event(mddev);
6162 if (mddev_is_clustered(mddev) && ret == 0)
6163 md_cluster_ops->metadata_update_cancel(mddev);
6165 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6166 bdevname(rdev->bdev,b), mdname(mddev));
6170 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6172 char b[BDEVNAME_SIZE];
6174 struct md_rdev *rdev;
6179 if (mddev->major_version != 0) {
6180 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6181 " version-0 superblocks.\n",
6185 if (!mddev->pers->hot_add_disk) {
6187 "%s: personality does not support diskops!\n",
6192 rdev = md_import_device(dev, -1, 0);
6195 "md: error, md_import_device() returned %ld\n",
6200 if (mddev->persistent)
6201 rdev->sb_start = calc_dev_sboffset(rdev);
6203 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6205 rdev->sectors = rdev->sb_start;
6207 if (test_bit(Faulty, &rdev->flags)) {
6209 "md: can not hot-add faulty %s disk to %s!\n",
6210 bdevname(rdev->bdev,b), mdname(mddev));
6215 clear_bit(In_sync, &rdev->flags);
6217 rdev->saved_raid_disk = -1;
6218 err = bind_rdev_to_array(rdev, mddev);
6223 * The rest should better be atomic, we can have disk failures
6224 * noticed in interrupt contexts ...
6227 rdev->raid_disk = -1;
6229 md_update_sb(mddev, 1);
6231 * Kick recovery, maybe this spare has to be added to the
6232 * array immediately.
6234 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6235 md_wakeup_thread(mddev->thread);
6236 md_new_event(mddev);
6244 static int set_bitmap_file(struct mddev *mddev, int fd)
6249 if (!mddev->pers->quiesce || !mddev->thread)
6251 if (mddev->recovery || mddev->sync_thread)
6253 /* we should be able to change the bitmap.. */
6257 struct inode *inode;
6260 if (mddev->bitmap || mddev->bitmap_info.file)
6261 return -EEXIST; /* cannot add when bitmap is present */
6265 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6270 inode = f->f_mapping->host;
6271 if (!S_ISREG(inode->i_mode)) {
6272 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6275 } else if (!(f->f_mode & FMODE_WRITE)) {
6276 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6279 } else if (atomic_read(&inode->i_writecount) != 1) {
6280 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6288 mddev->bitmap_info.file = f;
6289 mddev->bitmap_info.offset = 0; /* file overrides offset */
6290 } else if (mddev->bitmap == NULL)
6291 return -ENOENT; /* cannot remove what isn't there */
6294 mddev->pers->quiesce(mddev, 1);
6296 struct bitmap *bitmap;
6298 bitmap = bitmap_create(mddev, -1);
6299 if (!IS_ERR(bitmap)) {
6300 mddev->bitmap = bitmap;
6301 err = bitmap_load(mddev);
6303 err = PTR_ERR(bitmap);
6305 if (fd < 0 || err) {
6306 bitmap_destroy(mddev);
6307 fd = -1; /* make sure to put the file */
6309 mddev->pers->quiesce(mddev, 0);
6312 struct file *f = mddev->bitmap_info.file;
6314 spin_lock(&mddev->lock);
6315 mddev->bitmap_info.file = NULL;
6316 spin_unlock(&mddev->lock);
6325 * set_array_info is used two different ways
6326 * The original usage is when creating a new array.
6327 * In this usage, raid_disks is > 0 and it together with
6328 * level, size, not_persistent,layout,chunksize determine the
6329 * shape of the array.
6330 * This will always create an array with a type-0.90.0 superblock.
6331 * The newer usage is when assembling an array.
6332 * In this case raid_disks will be 0, and the major_version field is
6333 * use to determine which style super-blocks are to be found on the devices.
6334 * The minor and patch _version numbers are also kept incase the
6335 * super_block handler wishes to interpret them.
6337 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6340 if (info->raid_disks == 0) {
6341 /* just setting version number for superblock loading */
6342 if (info->major_version < 0 ||
6343 info->major_version >= ARRAY_SIZE(super_types) ||
6344 super_types[info->major_version].name == NULL) {
6345 /* maybe try to auto-load a module? */
6347 "md: superblock version %d not known\n",
6348 info->major_version);
6351 mddev->major_version = info->major_version;
6352 mddev->minor_version = info->minor_version;
6353 mddev->patch_version = info->patch_version;
6354 mddev->persistent = !info->not_persistent;
6355 /* ensure mddev_put doesn't delete this now that there
6356 * is some minimal configuration.
6358 mddev->ctime = get_seconds();
6361 mddev->major_version = MD_MAJOR_VERSION;
6362 mddev->minor_version = MD_MINOR_VERSION;
6363 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6364 mddev->ctime = get_seconds();
6366 mddev->level = info->level;
6367 mddev->clevel[0] = 0;
6368 mddev->dev_sectors = 2 * (sector_t)info->size;
6369 mddev->raid_disks = info->raid_disks;
6370 /* don't set md_minor, it is determined by which /dev/md* was
6373 if (info->state & (1<<MD_SB_CLEAN))
6374 mddev->recovery_cp = MaxSector;
6376 mddev->recovery_cp = 0;
6377 mddev->persistent = ! info->not_persistent;
6378 mddev->external = 0;
6380 mddev->layout = info->layout;
6381 mddev->chunk_sectors = info->chunk_size >> 9;
6383 mddev->max_disks = MD_SB_DISKS;
6385 if (mddev->persistent)
6387 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6389 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6390 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6391 mddev->bitmap_info.offset = 0;
6393 mddev->reshape_position = MaxSector;
6396 * Generate a 128 bit UUID
6398 get_random_bytes(mddev->uuid, 16);
6400 mddev->new_level = mddev->level;
6401 mddev->new_chunk_sectors = mddev->chunk_sectors;
6402 mddev->new_layout = mddev->layout;
6403 mddev->delta_disks = 0;
6404 mddev->reshape_backwards = 0;
6409 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6411 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6413 if (mddev->external_size)
6416 mddev->array_sectors = array_sectors;
6418 EXPORT_SYMBOL(md_set_array_sectors);
6420 static int update_size(struct mddev *mddev, sector_t num_sectors)
6422 struct md_rdev *rdev;
6424 int fit = (num_sectors == 0);
6426 if (mddev->pers->resize == NULL)
6428 /* The "num_sectors" is the number of sectors of each device that
6429 * is used. This can only make sense for arrays with redundancy.
6430 * linear and raid0 always use whatever space is available. We can only
6431 * consider changing this number if no resync or reconstruction is
6432 * happening, and if the new size is acceptable. It must fit before the
6433 * sb_start or, if that is <data_offset, it must fit before the size
6434 * of each device. If num_sectors is zero, we find the largest size
6437 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6443 rdev_for_each(rdev, mddev) {
6444 sector_t avail = rdev->sectors;
6446 if (fit && (num_sectors == 0 || num_sectors > avail))
6447 num_sectors = avail;
6448 if (avail < num_sectors)
6451 rv = mddev->pers->resize(mddev, num_sectors);
6453 revalidate_disk(mddev->gendisk);
6457 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6460 struct md_rdev *rdev;
6461 /* change the number of raid disks */
6462 if (mddev->pers->check_reshape == NULL)
6466 if (raid_disks <= 0 ||
6467 (mddev->max_disks && raid_disks >= mddev->max_disks))
6469 if (mddev->sync_thread ||
6470 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6471 mddev->reshape_position != MaxSector)
6474 rdev_for_each(rdev, mddev) {
6475 if (mddev->raid_disks < raid_disks &&
6476 rdev->data_offset < rdev->new_data_offset)
6478 if (mddev->raid_disks > raid_disks &&
6479 rdev->data_offset > rdev->new_data_offset)
6483 mddev->delta_disks = raid_disks - mddev->raid_disks;
6484 if (mddev->delta_disks < 0)
6485 mddev->reshape_backwards = 1;
6486 else if (mddev->delta_disks > 0)
6487 mddev->reshape_backwards = 0;
6489 rv = mddev->pers->check_reshape(mddev);
6491 mddev->delta_disks = 0;
6492 mddev->reshape_backwards = 0;
6498 * update_array_info is used to change the configuration of an
6500 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6501 * fields in the info are checked against the array.
6502 * Any differences that cannot be handled will cause an error.
6503 * Normally, only one change can be managed at a time.
6505 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6511 /* calculate expected state,ignoring low bits */
6512 if (mddev->bitmap && mddev->bitmap_info.offset)
6513 state |= (1 << MD_SB_BITMAP_PRESENT);
6515 if (mddev->major_version != info->major_version ||
6516 mddev->minor_version != info->minor_version ||
6517 /* mddev->patch_version != info->patch_version || */
6518 mddev->ctime != info->ctime ||
6519 mddev->level != info->level ||
6520 /* mddev->layout != info->layout || */
6521 mddev->persistent != !info->not_persistent ||
6522 mddev->chunk_sectors != info->chunk_size >> 9 ||
6523 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6524 ((state^info->state) & 0xfffffe00)
6527 /* Check there is only one change */
6528 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6530 if (mddev->raid_disks != info->raid_disks)
6532 if (mddev->layout != info->layout)
6534 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6541 if (mddev->layout != info->layout) {
6543 * we don't need to do anything at the md level, the
6544 * personality will take care of it all.
6546 if (mddev->pers->check_reshape == NULL)
6549 mddev->new_layout = info->layout;
6550 rv = mddev->pers->check_reshape(mddev);
6552 mddev->new_layout = mddev->layout;
6556 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6557 rv = update_size(mddev, (sector_t)info->size * 2);
6559 if (mddev->raid_disks != info->raid_disks)
6560 rv = update_raid_disks(mddev, info->raid_disks);
6562 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6563 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6567 if (mddev->recovery || mddev->sync_thread) {
6571 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6572 struct bitmap *bitmap;
6573 /* add the bitmap */
6574 if (mddev->bitmap) {
6578 if (mddev->bitmap_info.default_offset == 0) {
6582 mddev->bitmap_info.offset =
6583 mddev->bitmap_info.default_offset;
6584 mddev->bitmap_info.space =
6585 mddev->bitmap_info.default_space;
6586 mddev->pers->quiesce(mddev, 1);
6587 bitmap = bitmap_create(mddev, -1);
6588 if (!IS_ERR(bitmap)) {
6589 mddev->bitmap = bitmap;
6590 rv = bitmap_load(mddev);
6592 rv = PTR_ERR(bitmap);
6594 bitmap_destroy(mddev);
6595 mddev->pers->quiesce(mddev, 0);
6597 /* remove the bitmap */
6598 if (!mddev->bitmap) {
6602 if (mddev->bitmap->storage.file) {
6606 mddev->pers->quiesce(mddev, 1);
6607 bitmap_destroy(mddev);
6608 mddev->pers->quiesce(mddev, 0);
6609 mddev->bitmap_info.offset = 0;
6612 md_update_sb(mddev, 1);
6618 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6620 struct md_rdev *rdev;
6623 if (mddev->pers == NULL)
6627 rdev = find_rdev_rcu(mddev, dev);
6631 md_error(mddev, rdev);
6632 if (!test_bit(Faulty, &rdev->flags))
6640 * We have a problem here : there is no easy way to give a CHS
6641 * virtual geometry. We currently pretend that we have a 2 heads
6642 * 4 sectors (with a BIG number of cylinders...). This drives
6643 * dosfs just mad... ;-)
6645 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6647 struct mddev *mddev = bdev->bd_disk->private_data;
6651 geo->cylinders = mddev->array_sectors / 8;
6655 static inline bool md_ioctl_valid(unsigned int cmd)
6660 case GET_ARRAY_INFO:
6661 case GET_BITMAP_FILE:
6664 case HOT_REMOVE_DISK:
6667 case RESTART_ARRAY_RW:
6669 case SET_ARRAY_INFO:
6670 case SET_BITMAP_FILE:
6671 case SET_DISK_FAULTY:
6674 case CLUSTERED_DISK_NACK:
6681 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6682 unsigned int cmd, unsigned long arg)
6685 void __user *argp = (void __user *)arg;
6686 struct mddev *mddev = NULL;
6689 if (!md_ioctl_valid(cmd))
6694 case GET_ARRAY_INFO:
6698 if (!capable(CAP_SYS_ADMIN))
6703 * Commands dealing with the RAID driver but not any
6708 err = get_version(argp);
6714 autostart_arrays(arg);
6721 * Commands creating/starting a new array:
6724 mddev = bdev->bd_disk->private_data;
6731 /* Some actions do not requires the mutex */
6733 case GET_ARRAY_INFO:
6734 if (!mddev->raid_disks && !mddev->external)
6737 err = get_array_info(mddev, argp);
6741 if (!mddev->raid_disks && !mddev->external)
6744 err = get_disk_info(mddev, argp);
6747 case SET_DISK_FAULTY:
6748 err = set_disk_faulty(mddev, new_decode_dev(arg));
6751 case GET_BITMAP_FILE:
6752 err = get_bitmap_file(mddev, argp);
6757 if (cmd == ADD_NEW_DISK)
6758 /* need to ensure md_delayed_delete() has completed */
6759 flush_workqueue(md_misc_wq);
6761 if (cmd == HOT_REMOVE_DISK)
6762 /* need to ensure recovery thread has run */
6763 wait_event_interruptible_timeout(mddev->sb_wait,
6764 !test_bit(MD_RECOVERY_NEEDED,
6766 msecs_to_jiffies(5000));
6767 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6768 /* Need to flush page cache, and ensure no-one else opens
6771 mutex_lock(&mddev->open_mutex);
6772 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6773 mutex_unlock(&mddev->open_mutex);
6777 set_bit(MD_STILL_CLOSED, &mddev->flags);
6778 mutex_unlock(&mddev->open_mutex);
6779 sync_blockdev(bdev);
6781 err = mddev_lock(mddev);
6784 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6789 if (cmd == SET_ARRAY_INFO) {
6790 mdu_array_info_t info;
6792 memset(&info, 0, sizeof(info));
6793 else if (copy_from_user(&info, argp, sizeof(info))) {
6798 err = update_array_info(mddev, &info);
6800 printk(KERN_WARNING "md: couldn't update"
6801 " array info. %d\n", err);
6806 if (!list_empty(&mddev->disks)) {
6808 "md: array %s already has disks!\n",
6813 if (mddev->raid_disks) {
6815 "md: array %s already initialised!\n",
6820 err = set_array_info(mddev, &info);
6822 printk(KERN_WARNING "md: couldn't set"
6823 " array info. %d\n", err);
6830 * Commands querying/configuring an existing array:
6832 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6833 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6834 if ((!mddev->raid_disks && !mddev->external)
6835 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6836 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6837 && cmd != GET_BITMAP_FILE) {
6843 * Commands even a read-only array can execute:
6846 case RESTART_ARRAY_RW:
6847 err = restart_array(mddev);
6851 err = do_md_stop(mddev, 0, bdev);
6855 err = md_set_readonly(mddev, bdev);
6858 case HOT_REMOVE_DISK:
6859 err = hot_remove_disk(mddev, new_decode_dev(arg));
6863 /* We can support ADD_NEW_DISK on read-only arrays
6864 * on if we are re-adding a preexisting device.
6865 * So require mddev->pers and MD_DISK_SYNC.
6868 mdu_disk_info_t info;
6869 if (copy_from_user(&info, argp, sizeof(info)))
6871 else if (!(info.state & (1<<MD_DISK_SYNC)))
6872 /* Need to clear read-only for this */
6875 err = add_new_disk(mddev, &info);
6881 if (get_user(ro, (int __user *)(arg))) {
6887 /* if the bdev is going readonly the value of mddev->ro
6888 * does not matter, no writes are coming
6893 /* are we are already prepared for writes? */
6897 /* transitioning to readauto need only happen for
6898 * arrays that call md_write_start
6901 err = restart_array(mddev);
6904 set_disk_ro(mddev->gendisk, 0);
6911 * The remaining ioctls are changing the state of the
6912 * superblock, so we do not allow them on read-only arrays.
6914 if (mddev->ro && mddev->pers) {
6915 if (mddev->ro == 2) {
6917 sysfs_notify_dirent_safe(mddev->sysfs_state);
6918 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6919 /* mddev_unlock will wake thread */
6920 /* If a device failed while we were read-only, we
6921 * need to make sure the metadata is updated now.
6923 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6924 mddev_unlock(mddev);
6925 wait_event(mddev->sb_wait,
6926 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6927 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6928 mddev_lock_nointr(mddev);
6939 mdu_disk_info_t info;
6940 if (copy_from_user(&info, argp, sizeof(info)))
6943 err = add_new_disk(mddev, &info);
6947 case CLUSTERED_DISK_NACK:
6948 if (mddev_is_clustered(mddev))
6949 md_cluster_ops->new_disk_ack(mddev, false);
6955 err = hot_add_disk(mddev, new_decode_dev(arg));
6959 err = do_md_run(mddev);
6962 case SET_BITMAP_FILE:
6963 err = set_bitmap_file(mddev, (int)arg);
6972 if (mddev->hold_active == UNTIL_IOCTL &&
6974 mddev->hold_active = 0;
6975 mddev_unlock(mddev);
6979 #ifdef CONFIG_COMPAT
6980 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6981 unsigned int cmd, unsigned long arg)
6984 case HOT_REMOVE_DISK:
6986 case SET_DISK_FAULTY:
6987 case SET_BITMAP_FILE:
6988 /* These take in integer arg, do not convert */
6991 arg = (unsigned long)compat_ptr(arg);
6995 return md_ioctl(bdev, mode, cmd, arg);
6997 #endif /* CONFIG_COMPAT */
6999 static int md_open(struct block_device *bdev, fmode_t mode)
7002 * Succeed if we can lock the mddev, which confirms that
7003 * it isn't being stopped right now.
7005 struct mddev *mddev = mddev_find(bdev->bd_dev);
7011 if (mddev->gendisk != bdev->bd_disk) {
7012 /* we are racing with mddev_put which is discarding this
7016 /* Wait until bdev->bd_disk is definitely gone */
7017 flush_workqueue(md_misc_wq);
7018 /* Then retry the open from the top */
7019 return -ERESTARTSYS;
7021 BUG_ON(mddev != bdev->bd_disk->private_data);
7023 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7027 atomic_inc(&mddev->openers);
7028 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7029 mutex_unlock(&mddev->open_mutex);
7031 check_disk_change(bdev);
7036 static void md_release(struct gendisk *disk, fmode_t mode)
7038 struct mddev *mddev = disk->private_data;
7041 atomic_dec(&mddev->openers);
7045 static int md_media_changed(struct gendisk *disk)
7047 struct mddev *mddev = disk->private_data;
7049 return mddev->changed;
7052 static int md_revalidate(struct gendisk *disk)
7054 struct mddev *mddev = disk->private_data;
7059 static const struct block_device_operations md_fops =
7061 .owner = THIS_MODULE,
7063 .release = md_release,
7065 #ifdef CONFIG_COMPAT
7066 .compat_ioctl = md_compat_ioctl,
7068 .getgeo = md_getgeo,
7069 .media_changed = md_media_changed,
7070 .revalidate_disk= md_revalidate,
7073 static int md_thread(void *arg)
7075 struct md_thread *thread = arg;
7078 * md_thread is a 'system-thread', it's priority should be very
7079 * high. We avoid resource deadlocks individually in each
7080 * raid personality. (RAID5 does preallocation) We also use RR and
7081 * the very same RT priority as kswapd, thus we will never get
7082 * into a priority inversion deadlock.
7084 * we definitely have to have equal or higher priority than
7085 * bdflush, otherwise bdflush will deadlock if there are too
7086 * many dirty RAID5 blocks.
7089 allow_signal(SIGKILL);
7090 while (!kthread_should_stop()) {
7092 /* We need to wait INTERRUPTIBLE so that
7093 * we don't add to the load-average.
7094 * That means we need to be sure no signals are
7097 if (signal_pending(current))
7098 flush_signals(current);
7100 wait_event_interruptible_timeout
7102 test_bit(THREAD_WAKEUP, &thread->flags)
7103 || kthread_should_stop(),
7106 clear_bit(THREAD_WAKEUP, &thread->flags);
7107 if (!kthread_should_stop())
7108 thread->run(thread);
7114 void md_wakeup_thread(struct md_thread *thread)
7117 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7118 set_bit(THREAD_WAKEUP, &thread->flags);
7119 wake_up(&thread->wqueue);
7122 EXPORT_SYMBOL(md_wakeup_thread);
7124 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7125 struct mddev *mddev, const char *name)
7127 struct md_thread *thread;
7129 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7133 init_waitqueue_head(&thread->wqueue);
7136 thread->mddev = mddev;
7137 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7138 thread->tsk = kthread_run(md_thread, thread,
7140 mdname(thread->mddev),
7142 if (IS_ERR(thread->tsk)) {
7148 EXPORT_SYMBOL(md_register_thread);
7150 void md_unregister_thread(struct md_thread **threadp)
7152 struct md_thread *thread = *threadp;
7155 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7156 /* Locking ensures that mddev_unlock does not wake_up a
7157 * non-existent thread
7159 spin_lock(&pers_lock);
7161 spin_unlock(&pers_lock);
7163 kthread_stop(thread->tsk);
7166 EXPORT_SYMBOL(md_unregister_thread);
7168 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7170 if (!rdev || test_bit(Faulty, &rdev->flags))
7173 if (!mddev->pers || !mddev->pers->error_handler)
7175 mddev->pers->error_handler(mddev,rdev);
7176 if (mddev->degraded)
7177 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7178 sysfs_notify_dirent_safe(rdev->sysfs_state);
7179 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7180 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7181 md_wakeup_thread(mddev->thread);
7182 if (mddev->event_work.func)
7183 queue_work(md_misc_wq, &mddev->event_work);
7184 md_new_event_inintr(mddev);
7186 EXPORT_SYMBOL(md_error);
7188 /* seq_file implementation /proc/mdstat */
7190 static void status_unused(struct seq_file *seq)
7193 struct md_rdev *rdev;
7195 seq_printf(seq, "unused devices: ");
7197 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7198 char b[BDEVNAME_SIZE];
7200 seq_printf(seq, "%s ",
7201 bdevname(rdev->bdev,b));
7204 seq_printf(seq, "<none>");
7206 seq_printf(seq, "\n");
7209 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7211 sector_t max_sectors, resync, res;
7212 unsigned long dt, db;
7215 unsigned int per_milli;
7217 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7218 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7219 max_sectors = mddev->resync_max_sectors;
7221 max_sectors = mddev->dev_sectors;
7223 resync = mddev->curr_resync;
7225 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7226 /* Still cleaning up */
7227 resync = max_sectors;
7229 resync -= atomic_read(&mddev->recovery_active);
7232 if (mddev->recovery_cp < MaxSector) {
7233 seq_printf(seq, "\tresync=PENDING");
7239 seq_printf(seq, "\tresync=DELAYED");
7243 WARN_ON(max_sectors == 0);
7244 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7245 * in a sector_t, and (max_sectors>>scale) will fit in a
7246 * u32, as those are the requirements for sector_div.
7247 * Thus 'scale' must be at least 10
7250 if (sizeof(sector_t) > sizeof(unsigned long)) {
7251 while ( max_sectors/2 > (1ULL<<(scale+32)))
7254 res = (resync>>scale)*1000;
7255 sector_div(res, (u32)((max_sectors>>scale)+1));
7259 int i, x = per_milli/50, y = 20-x;
7260 seq_printf(seq, "[");
7261 for (i = 0; i < x; i++)
7262 seq_printf(seq, "=");
7263 seq_printf(seq, ">");
7264 for (i = 0; i < y; i++)
7265 seq_printf(seq, ".");
7266 seq_printf(seq, "] ");
7268 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7269 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7271 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7273 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7274 "resync" : "recovery"))),
7275 per_milli/10, per_milli % 10,
7276 (unsigned long long) resync/2,
7277 (unsigned long long) max_sectors/2);
7280 * dt: time from mark until now
7281 * db: blocks written from mark until now
7282 * rt: remaining time
7284 * rt is a sector_t, so could be 32bit or 64bit.
7285 * So we divide before multiply in case it is 32bit and close
7287 * We scale the divisor (db) by 32 to avoid losing precision
7288 * near the end of resync when the number of remaining sectors
7290 * We then divide rt by 32 after multiplying by db to compensate.
7291 * The '+1' avoids division by zero if db is very small.
7293 dt = ((jiffies - mddev->resync_mark) / HZ);
7295 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7296 - mddev->resync_mark_cnt;
7298 rt = max_sectors - resync; /* number of remaining sectors */
7299 sector_div(rt, db/32+1);
7303 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7304 ((unsigned long)rt % 60)/6);
7306 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7310 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7312 struct list_head *tmp;
7314 struct mddev *mddev;
7322 spin_lock(&all_mddevs_lock);
7323 list_for_each(tmp,&all_mddevs)
7325 mddev = list_entry(tmp, struct mddev, all_mddevs);
7327 spin_unlock(&all_mddevs_lock);
7330 spin_unlock(&all_mddevs_lock);
7332 return (void*)2;/* tail */
7336 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7338 struct list_head *tmp;
7339 struct mddev *next_mddev, *mddev = v;
7345 spin_lock(&all_mddevs_lock);
7347 tmp = all_mddevs.next;
7349 tmp = mddev->all_mddevs.next;
7350 if (tmp != &all_mddevs)
7351 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7353 next_mddev = (void*)2;
7356 spin_unlock(&all_mddevs_lock);
7364 static void md_seq_stop(struct seq_file *seq, void *v)
7366 struct mddev *mddev = v;
7368 if (mddev && v != (void*)1 && v != (void*)2)
7372 static int md_seq_show(struct seq_file *seq, void *v)
7374 struct mddev *mddev = v;
7376 struct md_rdev *rdev;
7378 if (v == (void*)1) {
7379 struct md_personality *pers;
7380 seq_printf(seq, "Personalities : ");
7381 spin_lock(&pers_lock);
7382 list_for_each_entry(pers, &pers_list, list)
7383 seq_printf(seq, "[%s] ", pers->name);
7385 spin_unlock(&pers_lock);
7386 seq_printf(seq, "\n");
7387 seq->poll_event = atomic_read(&md_event_count);
7390 if (v == (void*)2) {
7395 spin_lock(&mddev->lock);
7396 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7397 seq_printf(seq, "%s : %sactive", mdname(mddev),
7398 mddev->pers ? "" : "in");
7401 seq_printf(seq, " (read-only)");
7403 seq_printf(seq, " (auto-read-only)");
7404 seq_printf(seq, " %s", mddev->pers->name);
7409 rdev_for_each_rcu(rdev, mddev) {
7410 char b[BDEVNAME_SIZE];
7411 seq_printf(seq, " %s[%d]",
7412 bdevname(rdev->bdev,b), rdev->desc_nr);
7413 if (test_bit(WriteMostly, &rdev->flags))
7414 seq_printf(seq, "(W)");
7415 if (test_bit(Journal, &rdev->flags))
7416 seq_printf(seq, "(J)");
7417 if (test_bit(Faulty, &rdev->flags)) {
7418 seq_printf(seq, "(F)");
7421 if (rdev->raid_disk < 0)
7422 seq_printf(seq, "(S)"); /* spare */
7423 if (test_bit(Replacement, &rdev->flags))
7424 seq_printf(seq, "(R)");
7425 sectors += rdev->sectors;
7429 if (!list_empty(&mddev->disks)) {
7431 seq_printf(seq, "\n %llu blocks",
7432 (unsigned long long)
7433 mddev->array_sectors / 2);
7435 seq_printf(seq, "\n %llu blocks",
7436 (unsigned long long)sectors / 2);
7438 if (mddev->persistent) {
7439 if (mddev->major_version != 0 ||
7440 mddev->minor_version != 90) {
7441 seq_printf(seq," super %d.%d",
7442 mddev->major_version,
7443 mddev->minor_version);
7445 } else if (mddev->external)
7446 seq_printf(seq, " super external:%s",
7447 mddev->metadata_type);
7449 seq_printf(seq, " super non-persistent");
7452 mddev->pers->status(seq, mddev);
7453 seq_printf(seq, "\n ");
7454 if (mddev->pers->sync_request) {
7455 if (status_resync(seq, mddev))
7456 seq_printf(seq, "\n ");
7459 seq_printf(seq, "\n ");
7461 bitmap_status(seq, mddev->bitmap);
7463 seq_printf(seq, "\n");
7465 spin_unlock(&mddev->lock);
7470 static const struct seq_operations md_seq_ops = {
7471 .start = md_seq_start,
7472 .next = md_seq_next,
7473 .stop = md_seq_stop,
7474 .show = md_seq_show,
7477 static int md_seq_open(struct inode *inode, struct file *file)
7479 struct seq_file *seq;
7482 error = seq_open(file, &md_seq_ops);
7486 seq = file->private_data;
7487 seq->poll_event = atomic_read(&md_event_count);
7491 static int md_unloading;
7492 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7494 struct seq_file *seq = filp->private_data;
7498 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7499 poll_wait(filp, &md_event_waiters, wait);
7501 /* always allow read */
7502 mask = POLLIN | POLLRDNORM;
7504 if (seq->poll_event != atomic_read(&md_event_count))
7505 mask |= POLLERR | POLLPRI;
7509 static const struct file_operations md_seq_fops = {
7510 .owner = THIS_MODULE,
7511 .open = md_seq_open,
7513 .llseek = seq_lseek,
7514 .release = seq_release_private,
7515 .poll = mdstat_poll,
7518 int register_md_personality(struct md_personality *p)
7520 printk(KERN_INFO "md: %s personality registered for level %d\n",
7522 spin_lock(&pers_lock);
7523 list_add_tail(&p->list, &pers_list);
7524 spin_unlock(&pers_lock);
7527 EXPORT_SYMBOL(register_md_personality);
7529 int unregister_md_personality(struct md_personality *p)
7531 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7532 spin_lock(&pers_lock);
7533 list_del_init(&p->list);
7534 spin_unlock(&pers_lock);
7537 EXPORT_SYMBOL(unregister_md_personality);
7539 int register_md_cluster_operations(struct md_cluster_operations *ops,
7540 struct module *module)
7543 spin_lock(&pers_lock);
7544 if (md_cluster_ops != NULL)
7547 md_cluster_ops = ops;
7548 md_cluster_mod = module;
7550 spin_unlock(&pers_lock);
7553 EXPORT_SYMBOL(register_md_cluster_operations);
7555 int unregister_md_cluster_operations(void)
7557 spin_lock(&pers_lock);
7558 md_cluster_ops = NULL;
7559 spin_unlock(&pers_lock);
7562 EXPORT_SYMBOL(unregister_md_cluster_operations);
7564 int md_setup_cluster(struct mddev *mddev, int nodes)
7568 err = request_module("md-cluster");
7570 pr_err("md-cluster module not found.\n");
7574 spin_lock(&pers_lock);
7575 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7576 spin_unlock(&pers_lock);
7579 spin_unlock(&pers_lock);
7581 return md_cluster_ops->join(mddev, nodes);
7584 void md_cluster_stop(struct mddev *mddev)
7586 if (!md_cluster_ops)
7588 md_cluster_ops->leave(mddev);
7589 module_put(md_cluster_mod);
7592 static int is_mddev_idle(struct mddev *mddev, int init)
7594 struct md_rdev *rdev;
7600 rdev_for_each_rcu(rdev, mddev) {
7601 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7602 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7603 (int)part_stat_read(&disk->part0, sectors[1]) -
7604 atomic_read(&disk->sync_io);
7605 /* sync IO will cause sync_io to increase before the disk_stats
7606 * as sync_io is counted when a request starts, and
7607 * disk_stats is counted when it completes.
7608 * So resync activity will cause curr_events to be smaller than
7609 * when there was no such activity.
7610 * non-sync IO will cause disk_stat to increase without
7611 * increasing sync_io so curr_events will (eventually)
7612 * be larger than it was before. Once it becomes
7613 * substantially larger, the test below will cause
7614 * the array to appear non-idle, and resync will slow
7616 * If there is a lot of outstanding resync activity when
7617 * we set last_event to curr_events, then all that activity
7618 * completing might cause the array to appear non-idle
7619 * and resync will be slowed down even though there might
7620 * not have been non-resync activity. This will only
7621 * happen once though. 'last_events' will soon reflect
7622 * the state where there is little or no outstanding
7623 * resync requests, and further resync activity will
7624 * always make curr_events less than last_events.
7627 if (init || curr_events - rdev->last_events > 64) {
7628 rdev->last_events = curr_events;
7636 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7638 /* another "blocks" (512byte) blocks have been synced */
7639 atomic_sub(blocks, &mddev->recovery_active);
7640 wake_up(&mddev->recovery_wait);
7642 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7643 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7644 md_wakeup_thread(mddev->thread);
7645 // stop recovery, signal do_sync ....
7648 EXPORT_SYMBOL(md_done_sync);
7650 /* md_write_start(mddev, bi)
7651 * If we need to update some array metadata (e.g. 'active' flag
7652 * in superblock) before writing, schedule a superblock update
7653 * and wait for it to complete.
7655 void md_write_start(struct mddev *mddev, struct bio *bi)
7658 if (bio_data_dir(bi) != WRITE)
7661 BUG_ON(mddev->ro == 1);
7662 if (mddev->ro == 2) {
7663 /* need to switch to read/write */
7665 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7666 md_wakeup_thread(mddev->thread);
7667 md_wakeup_thread(mddev->sync_thread);
7670 atomic_inc(&mddev->writes_pending);
7671 if (mddev->safemode == 1)
7672 mddev->safemode = 0;
7673 if (mddev->in_sync) {
7674 spin_lock(&mddev->lock);
7675 if (mddev->in_sync) {
7677 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7678 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7679 md_wakeup_thread(mddev->thread);
7682 spin_unlock(&mddev->lock);
7685 sysfs_notify_dirent_safe(mddev->sysfs_state);
7686 wait_event(mddev->sb_wait,
7687 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7689 EXPORT_SYMBOL(md_write_start);
7691 void md_write_end(struct mddev *mddev)
7693 if (atomic_dec_and_test(&mddev->writes_pending)) {
7694 if (mddev->safemode == 2)
7695 md_wakeup_thread(mddev->thread);
7696 else if (mddev->safemode_delay)
7697 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7700 EXPORT_SYMBOL(md_write_end);
7702 /* md_allow_write(mddev)
7703 * Calling this ensures that the array is marked 'active' so that writes
7704 * may proceed without blocking. It is important to call this before
7705 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7706 * Must be called with mddev_lock held.
7708 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7709 * is dropped, so return -EAGAIN after notifying userspace.
7711 int md_allow_write(struct mddev *mddev)
7717 if (!mddev->pers->sync_request)
7720 spin_lock(&mddev->lock);
7721 if (mddev->in_sync) {
7723 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7724 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7725 if (mddev->safemode_delay &&
7726 mddev->safemode == 0)
7727 mddev->safemode = 1;
7728 spin_unlock(&mddev->lock);
7729 md_update_sb(mddev, 0);
7730 sysfs_notify_dirent_safe(mddev->sysfs_state);
7732 spin_unlock(&mddev->lock);
7734 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7739 EXPORT_SYMBOL_GPL(md_allow_write);
7741 #define SYNC_MARKS 10
7742 #define SYNC_MARK_STEP (3*HZ)
7743 #define UPDATE_FREQUENCY (5*60*HZ)
7744 void md_do_sync(struct md_thread *thread)
7746 struct mddev *mddev = thread->mddev;
7747 struct mddev *mddev2;
7748 unsigned int currspeed = 0,
7750 sector_t max_sectors,j, io_sectors, recovery_done;
7751 unsigned long mark[SYNC_MARKS];
7752 unsigned long update_time;
7753 sector_t mark_cnt[SYNC_MARKS];
7755 struct list_head *tmp;
7756 sector_t last_check;
7758 struct md_rdev *rdev;
7759 char *desc, *action = NULL;
7760 struct blk_plug plug;
7761 bool cluster_resync_finished = false;
7763 /* just incase thread restarts... */
7764 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7766 if (mddev->ro) {/* never try to sync a read-only array */
7767 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7771 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7772 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7773 desc = "data-check";
7775 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7776 desc = "requested-resync";
7780 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7785 mddev->last_sync_action = action ?: desc;
7787 /* we overload curr_resync somewhat here.
7788 * 0 == not engaged in resync at all
7789 * 2 == checking that there is no conflict with another sync
7790 * 1 == like 2, but have yielded to allow conflicting resync to
7792 * other == active in resync - this many blocks
7794 * Before starting a resync we must have set curr_resync to
7795 * 2, and then checked that every "conflicting" array has curr_resync
7796 * less than ours. When we find one that is the same or higher
7797 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7798 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7799 * This will mean we have to start checking from the beginning again.
7804 mddev->curr_resync = 2;
7807 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7809 for_each_mddev(mddev2, tmp) {
7810 if (mddev2 == mddev)
7812 if (!mddev->parallel_resync
7813 && mddev2->curr_resync
7814 && match_mddev_units(mddev, mddev2)) {
7816 if (mddev < mddev2 && mddev->curr_resync == 2) {
7817 /* arbitrarily yield */
7818 mddev->curr_resync = 1;
7819 wake_up(&resync_wait);
7821 if (mddev > mddev2 && mddev->curr_resync == 1)
7822 /* no need to wait here, we can wait the next
7823 * time 'round when curr_resync == 2
7826 /* We need to wait 'interruptible' so as not to
7827 * contribute to the load average, and not to
7828 * be caught by 'softlockup'
7830 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7831 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7832 mddev2->curr_resync >= mddev->curr_resync) {
7833 printk(KERN_INFO "md: delaying %s of %s"
7834 " until %s has finished (they"
7835 " share one or more physical units)\n",
7836 desc, mdname(mddev), mdname(mddev2));
7838 if (signal_pending(current))
7839 flush_signals(current);
7841 finish_wait(&resync_wait, &wq);
7844 finish_wait(&resync_wait, &wq);
7847 } while (mddev->curr_resync < 2);
7850 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7851 /* resync follows the size requested by the personality,
7852 * which defaults to physical size, but can be virtual size
7854 max_sectors = mddev->resync_max_sectors;
7855 atomic64_set(&mddev->resync_mismatches, 0);
7856 /* we don't use the checkpoint if there's a bitmap */
7857 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7858 j = mddev->resync_min;
7859 else if (!mddev->bitmap)
7860 j = mddev->recovery_cp;
7862 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7863 max_sectors = mddev->resync_max_sectors;
7865 /* recovery follows the physical size of devices */
7866 max_sectors = mddev->dev_sectors;
7869 rdev_for_each_rcu(rdev, mddev)
7870 if (rdev->raid_disk >= 0 &&
7871 !test_bit(Journal, &rdev->flags) &&
7872 !test_bit(Faulty, &rdev->flags) &&
7873 !test_bit(In_sync, &rdev->flags) &&
7874 rdev->recovery_offset < j)
7875 j = rdev->recovery_offset;
7878 /* If there is a bitmap, we need to make sure all
7879 * writes that started before we added a spare
7880 * complete before we start doing a recovery.
7881 * Otherwise the write might complete and (via
7882 * bitmap_endwrite) set a bit in the bitmap after the
7883 * recovery has checked that bit and skipped that
7886 if (mddev->bitmap) {
7887 mddev->pers->quiesce(mddev, 1);
7888 mddev->pers->quiesce(mddev, 0);
7892 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7893 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7894 " %d KB/sec/disk.\n", speed_min(mddev));
7895 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7896 "(but not more than %d KB/sec) for %s.\n",
7897 speed_max(mddev), desc);
7899 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7902 for (m = 0; m < SYNC_MARKS; m++) {
7904 mark_cnt[m] = io_sectors;
7907 mddev->resync_mark = mark[last_mark];
7908 mddev->resync_mark_cnt = mark_cnt[last_mark];
7911 * Tune reconstruction:
7913 window = 32*(PAGE_SIZE/512);
7914 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7915 window/2, (unsigned long long)max_sectors/2);
7917 atomic_set(&mddev->recovery_active, 0);
7922 "md: resuming %s of %s from checkpoint.\n",
7923 desc, mdname(mddev));
7924 mddev->curr_resync = j;
7926 mddev->curr_resync = 3; /* no longer delayed */
7927 mddev->curr_resync_completed = j;
7928 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7929 md_new_event(mddev);
7930 update_time = jiffies;
7932 blk_start_plug(&plug);
7933 while (j < max_sectors) {
7938 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7939 ((mddev->curr_resync > mddev->curr_resync_completed &&
7940 (mddev->curr_resync - mddev->curr_resync_completed)
7941 > (max_sectors >> 4)) ||
7942 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7943 (j - mddev->curr_resync_completed)*2
7944 >= mddev->resync_max - mddev->curr_resync_completed ||
7945 mddev->curr_resync_completed > mddev->resync_max
7947 /* time to update curr_resync_completed */
7948 wait_event(mddev->recovery_wait,
7949 atomic_read(&mddev->recovery_active) == 0);
7950 mddev->curr_resync_completed = j;
7951 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7952 j > mddev->recovery_cp)
7953 mddev->recovery_cp = j;
7954 update_time = jiffies;
7955 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7956 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7959 while (j >= mddev->resync_max &&
7960 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7961 /* As this condition is controlled by user-space,
7962 * we can block indefinitely, so use '_interruptible'
7963 * to avoid triggering warnings.
7965 flush_signals(current); /* just in case */
7966 wait_event_interruptible(mddev->recovery_wait,
7967 mddev->resync_max > j
7968 || test_bit(MD_RECOVERY_INTR,
7972 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7975 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7977 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7981 if (!skipped) { /* actual IO requested */
7982 io_sectors += sectors;
7983 atomic_add(sectors, &mddev->recovery_active);
7986 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7990 if (j > max_sectors)
7991 /* when skipping, extra large numbers can be returned. */
7994 mddev->curr_resync = j;
7995 mddev->curr_mark_cnt = io_sectors;
7996 if (last_check == 0)
7997 /* this is the earliest that rebuild will be
7998 * visible in /proc/mdstat
8000 md_new_event(mddev);
8002 if (last_check + window > io_sectors || j == max_sectors)
8005 last_check = io_sectors;
8007 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8009 int next = (last_mark+1) % SYNC_MARKS;
8011 mddev->resync_mark = mark[next];
8012 mddev->resync_mark_cnt = mark_cnt[next];
8013 mark[next] = jiffies;
8014 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8018 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8022 * this loop exits only if either when we are slower than
8023 * the 'hard' speed limit, or the system was IO-idle for
8025 * the system might be non-idle CPU-wise, but we only care
8026 * about not overloading the IO subsystem. (things like an
8027 * e2fsck being done on the RAID array should execute fast)
8031 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8032 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8033 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8035 if (currspeed > speed_min(mddev)) {
8036 if (currspeed > speed_max(mddev)) {
8040 if (!is_mddev_idle(mddev, 0)) {
8042 * Give other IO more of a chance.
8043 * The faster the devices, the less we wait.
8045 wait_event(mddev->recovery_wait,
8046 !atomic_read(&mddev->recovery_active));
8050 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8051 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8052 ? "interrupted" : "done");
8054 * this also signals 'finished resyncing' to md_stop
8056 blk_finish_plug(&plug);
8057 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8059 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8060 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8061 mddev->curr_resync > 2) {
8062 mddev->curr_resync_completed = mddev->curr_resync;
8063 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8065 /* tell personality and other nodes that we are finished */
8066 if (mddev_is_clustered(mddev)) {
8067 md_cluster_ops->resync_finish(mddev);
8068 cluster_resync_finished = true;
8070 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8072 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8073 mddev->curr_resync > 2) {
8074 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8075 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8076 if (mddev->curr_resync >= mddev->recovery_cp) {
8078 "md: checkpointing %s of %s.\n",
8079 desc, mdname(mddev));
8080 if (test_bit(MD_RECOVERY_ERROR,
8082 mddev->recovery_cp =
8083 mddev->curr_resync_completed;
8085 mddev->recovery_cp =
8089 mddev->recovery_cp = MaxSector;
8091 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8092 mddev->curr_resync = MaxSector;
8094 rdev_for_each_rcu(rdev, mddev)
8095 if (rdev->raid_disk >= 0 &&
8096 mddev->delta_disks >= 0 &&
8097 !test_bit(Journal, &rdev->flags) &&
8098 !test_bit(Faulty, &rdev->flags) &&
8099 !test_bit(In_sync, &rdev->flags) &&
8100 rdev->recovery_offset < mddev->curr_resync)
8101 rdev->recovery_offset = mddev->curr_resync;
8106 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8108 if (mddev_is_clustered(mddev) &&
8109 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8110 !cluster_resync_finished)
8111 md_cluster_ops->resync_finish(mddev);
8113 spin_lock(&mddev->lock);
8114 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8115 /* We completed so min/max setting can be forgotten if used. */
8116 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8117 mddev->resync_min = 0;
8118 mddev->resync_max = MaxSector;
8119 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8120 mddev->resync_min = mddev->curr_resync_completed;
8121 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8122 mddev->curr_resync = 0;
8123 spin_unlock(&mddev->lock);
8125 wake_up(&resync_wait);
8126 md_wakeup_thread(mddev->thread);
8129 EXPORT_SYMBOL_GPL(md_do_sync);
8131 static int remove_and_add_spares(struct mddev *mddev,
8132 struct md_rdev *this)
8134 struct md_rdev *rdev;
8138 rdev_for_each(rdev, mddev)
8139 if ((this == NULL || rdev == this) &&
8140 rdev->raid_disk >= 0 &&
8141 !test_bit(Blocked, &rdev->flags) &&
8142 (test_bit(Faulty, &rdev->flags) ||
8143 (!test_bit(In_sync, &rdev->flags) &&
8144 !test_bit(Journal, &rdev->flags))) &&
8145 atomic_read(&rdev->nr_pending)==0) {
8146 if (mddev->pers->hot_remove_disk(
8147 mddev, rdev) == 0) {
8148 sysfs_unlink_rdev(mddev, rdev);
8149 rdev->raid_disk = -1;
8153 if (removed && mddev->kobj.sd)
8154 sysfs_notify(&mddev->kobj, NULL, "degraded");
8156 if (this && removed)
8159 rdev_for_each(rdev, mddev) {
8160 if (this && this != rdev)
8162 if (test_bit(Candidate, &rdev->flags))
8164 if (rdev->raid_disk >= 0 &&
8165 !test_bit(In_sync, &rdev->flags) &&
8166 !test_bit(Journal, &rdev->flags) &&
8167 !test_bit(Faulty, &rdev->flags))
8169 if (rdev->raid_disk >= 0)
8171 if (test_bit(Faulty, &rdev->flags))
8173 if (test_bit(Journal, &rdev->flags))
8176 ! (rdev->saved_raid_disk >= 0 &&
8177 !test_bit(Bitmap_sync, &rdev->flags)))
8180 rdev->recovery_offset = 0;
8182 hot_add_disk(mddev, rdev) == 0) {
8183 if (sysfs_link_rdev(mddev, rdev))
8184 /* failure here is OK */;
8186 md_new_event(mddev);
8187 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8192 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8196 static void md_start_sync(struct work_struct *ws)
8198 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8201 if (mddev_is_clustered(mddev)) {
8202 ret = md_cluster_ops->resync_start(mddev);
8204 mddev->sync_thread = NULL;
8209 mddev->sync_thread = md_register_thread(md_do_sync,
8213 if (!mddev->sync_thread) {
8214 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8215 printk(KERN_ERR "%s: could not start resync"
8218 /* leave the spares where they are, it shouldn't hurt */
8219 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8220 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8221 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8222 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8223 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8224 wake_up(&resync_wait);
8225 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8227 if (mddev->sysfs_action)
8228 sysfs_notify_dirent_safe(mddev->sysfs_action);
8230 md_wakeup_thread(mddev->sync_thread);
8231 sysfs_notify_dirent_safe(mddev->sysfs_action);
8232 md_new_event(mddev);
8236 * This routine is regularly called by all per-raid-array threads to
8237 * deal with generic issues like resync and super-block update.
8238 * Raid personalities that don't have a thread (linear/raid0) do not
8239 * need this as they never do any recovery or update the superblock.
8241 * It does not do any resync itself, but rather "forks" off other threads
8242 * to do that as needed.
8243 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8244 * "->recovery" and create a thread at ->sync_thread.
8245 * When the thread finishes it sets MD_RECOVERY_DONE
8246 * and wakeups up this thread which will reap the thread and finish up.
8247 * This thread also removes any faulty devices (with nr_pending == 0).
8249 * The overall approach is:
8250 * 1/ if the superblock needs updating, update it.
8251 * 2/ If a recovery thread is running, don't do anything else.
8252 * 3/ If recovery has finished, clean up, possibly marking spares active.
8253 * 4/ If there are any faulty devices, remove them.
8254 * 5/ If array is degraded, try to add spares devices
8255 * 6/ If array has spares or is not in-sync, start a resync thread.
8257 void md_check_recovery(struct mddev *mddev)
8259 if (mddev->suspended)
8263 bitmap_daemon_work(mddev);
8265 if (signal_pending(current)) {
8266 if (mddev->pers->sync_request && !mddev->external) {
8267 printk(KERN_INFO "md: %s in immediate safe mode\n",
8269 mddev->safemode = 2;
8271 flush_signals(current);
8274 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8277 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8278 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8279 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8280 (mddev->external == 0 && mddev->safemode == 1) ||
8281 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8282 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8286 if (mddev_trylock(mddev)) {
8290 struct md_rdev *rdev;
8291 if (!mddev->external && mddev->in_sync)
8292 /* 'Blocked' flag not needed as failed devices
8293 * will be recorded if array switched to read/write.
8294 * Leaving it set will prevent the device
8295 * from being removed.
8297 rdev_for_each(rdev, mddev)
8298 clear_bit(Blocked, &rdev->flags);
8299 /* On a read-only array we can:
8300 * - remove failed devices
8301 * - add already-in_sync devices if the array itself
8303 * As we only add devices that are already in-sync,
8304 * we can activate the spares immediately.
8306 remove_and_add_spares(mddev, NULL);
8307 /* There is no thread, but we need to call
8308 * ->spare_active and clear saved_raid_disk
8310 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8311 md_reap_sync_thread(mddev);
8312 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8313 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8314 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8318 if (!mddev->external) {
8320 spin_lock(&mddev->lock);
8321 if (mddev->safemode &&
8322 !atomic_read(&mddev->writes_pending) &&
8324 mddev->recovery_cp == MaxSector) {
8327 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8329 if (mddev->safemode == 1)
8330 mddev->safemode = 0;
8331 spin_unlock(&mddev->lock);
8333 sysfs_notify_dirent_safe(mddev->sysfs_state);
8336 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8337 md_update_sb(mddev, 0);
8339 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8340 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8341 /* resync/recovery still happening */
8342 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8345 if (mddev->sync_thread) {
8346 md_reap_sync_thread(mddev);
8349 /* Set RUNNING before clearing NEEDED to avoid
8350 * any transients in the value of "sync_action".
8352 mddev->curr_resync_completed = 0;
8353 spin_lock(&mddev->lock);
8354 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8355 spin_unlock(&mddev->lock);
8356 /* Clear some bits that don't mean anything, but
8359 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8360 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8362 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8363 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8365 /* no recovery is running.
8366 * remove any failed drives, then
8367 * add spares if possible.
8368 * Spares are also removed and re-added, to allow
8369 * the personality to fail the re-add.
8372 if (mddev->reshape_position != MaxSector) {
8373 if (mddev->pers->check_reshape == NULL ||
8374 mddev->pers->check_reshape(mddev) != 0)
8375 /* Cannot proceed */
8377 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8378 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8379 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8380 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8381 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8382 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8383 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8384 } else if (mddev->recovery_cp < MaxSector) {
8385 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8386 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8387 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8388 /* nothing to be done ... */
8391 if (mddev->pers->sync_request) {
8393 /* We are adding a device or devices to an array
8394 * which has the bitmap stored on all devices.
8395 * So make sure all bitmap pages get written
8397 bitmap_write_all(mddev->bitmap);
8399 INIT_WORK(&mddev->del_work, md_start_sync);
8400 queue_work(md_misc_wq, &mddev->del_work);
8404 if (!mddev->sync_thread) {
8405 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8406 wake_up(&resync_wait);
8407 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8409 if (mddev->sysfs_action)
8410 sysfs_notify_dirent_safe(mddev->sysfs_action);
8413 wake_up(&mddev->sb_wait);
8414 mddev_unlock(mddev);
8417 EXPORT_SYMBOL(md_check_recovery);
8419 void md_reap_sync_thread(struct mddev *mddev)
8421 struct md_rdev *rdev;
8423 /* resync has finished, collect result */
8424 md_unregister_thread(&mddev->sync_thread);
8425 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8426 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8428 /* activate any spares */
8429 if (mddev->pers->spare_active(mddev)) {
8430 sysfs_notify(&mddev->kobj, NULL,
8432 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8435 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8436 mddev->pers->finish_reshape)
8437 mddev->pers->finish_reshape(mddev);
8439 /* If array is no-longer degraded, then any saved_raid_disk
8440 * information must be scrapped.
8442 if (!mddev->degraded)
8443 rdev_for_each(rdev, mddev)
8444 rdev->saved_raid_disk = -1;
8446 md_update_sb(mddev, 1);
8447 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8448 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8449 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8450 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8451 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8452 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8453 wake_up(&resync_wait);
8454 /* flag recovery needed just to double check */
8455 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8456 sysfs_notify_dirent_safe(mddev->sysfs_action);
8457 md_new_event(mddev);
8458 if (mddev->event_work.func)
8459 queue_work(md_misc_wq, &mddev->event_work);
8461 EXPORT_SYMBOL(md_reap_sync_thread);
8463 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8465 sysfs_notify_dirent_safe(rdev->sysfs_state);
8466 wait_event_timeout(rdev->blocked_wait,
8467 !test_bit(Blocked, &rdev->flags) &&
8468 !test_bit(BlockedBadBlocks, &rdev->flags),
8469 msecs_to_jiffies(5000));
8470 rdev_dec_pending(rdev, mddev);
8472 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8474 void md_finish_reshape(struct mddev *mddev)
8476 /* called be personality module when reshape completes. */
8477 struct md_rdev *rdev;
8479 rdev_for_each(rdev, mddev) {
8480 if (rdev->data_offset > rdev->new_data_offset)
8481 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8483 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8484 rdev->data_offset = rdev->new_data_offset;
8487 EXPORT_SYMBOL(md_finish_reshape);
8489 /* Bad block management.
8490 * We can record which blocks on each device are 'bad' and so just
8491 * fail those blocks, or that stripe, rather than the whole device.
8492 * Entries in the bad-block table are 64bits wide. This comprises:
8493 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8494 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8495 * A 'shift' can be set so that larger blocks are tracked and
8496 * consequently larger devices can be covered.
8497 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8499 * Locking of the bad-block table uses a seqlock so md_is_badblock
8500 * might need to retry if it is very unlucky.
8501 * We will sometimes want to check for bad blocks in a bi_end_io function,
8502 * so we use the write_seqlock_irq variant.
8504 * When looking for a bad block we specify a range and want to
8505 * know if any block in the range is bad. So we binary-search
8506 * to the last range that starts at-or-before the given endpoint,
8507 * (or "before the sector after the target range")
8508 * then see if it ends after the given start.
8510 * 0 if there are no known bad blocks in the range
8511 * 1 if there are known bad block which are all acknowledged
8512 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8513 * plus the start/length of the first bad section we overlap.
8515 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8516 sector_t *first_bad, int *bad_sectors)
8522 sector_t target = s + sectors;
8525 if (bb->shift > 0) {
8526 /* round the start down, and the end up */
8528 target += (1<<bb->shift) - 1;
8529 target >>= bb->shift;
8530 sectors = target - s;
8532 /* 'target' is now the first block after the bad range */
8535 seq = read_seqbegin(&bb->lock);
8540 /* Binary search between lo and hi for 'target'
8541 * i.e. for the last range that starts before 'target'
8543 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8544 * are known not to be the last range before target.
8545 * VARIANT: hi-lo is the number of possible
8546 * ranges, and decreases until it reaches 1
8548 while (hi - lo > 1) {
8549 int mid = (lo + hi) / 2;
8550 sector_t a = BB_OFFSET(p[mid]);
8552 /* This could still be the one, earlier ranges
8556 /* This and later ranges are definitely out. */
8559 /* 'lo' might be the last that started before target, but 'hi' isn't */
8561 /* need to check all range that end after 's' to see if
8562 * any are unacknowledged.
8565 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8566 if (BB_OFFSET(p[lo]) < target) {
8567 /* starts before the end, and finishes after
8568 * the start, so they must overlap
8570 if (rv != -1 && BB_ACK(p[lo]))
8574 *first_bad = BB_OFFSET(p[lo]);
8575 *bad_sectors = BB_LEN(p[lo]);
8581 if (read_seqretry(&bb->lock, seq))
8586 EXPORT_SYMBOL_GPL(md_is_badblock);
8589 * Add a range of bad blocks to the table.
8590 * This might extend the table, or might contract it
8591 * if two adjacent ranges can be merged.
8592 * We binary-search to find the 'insertion' point, then
8593 * decide how best to handle it.
8595 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8601 unsigned long flags;
8604 /* badblocks are disabled */
8608 /* round the start down, and the end up */
8609 sector_t next = s + sectors;
8611 next += (1<<bb->shift) - 1;
8616 write_seqlock_irqsave(&bb->lock, flags);
8621 /* Find the last range that starts at-or-before 's' */
8622 while (hi - lo > 1) {
8623 int mid = (lo + hi) / 2;
8624 sector_t a = BB_OFFSET(p[mid]);
8630 if (hi > lo && BB_OFFSET(p[lo]) > s)
8634 /* we found a range that might merge with the start
8637 sector_t a = BB_OFFSET(p[lo]);
8638 sector_t e = a + BB_LEN(p[lo]);
8639 int ack = BB_ACK(p[lo]);
8641 /* Yes, we can merge with a previous range */
8642 if (s == a && s + sectors >= e)
8643 /* new range covers old */
8646 ack = ack && acknowledged;
8648 if (e < s + sectors)
8650 if (e - a <= BB_MAX_LEN) {
8651 p[lo] = BB_MAKE(a, e-a, ack);
8654 /* does not all fit in one range,
8655 * make p[lo] maximal
8657 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8658 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8664 if (sectors && hi < bb->count) {
8665 /* 'hi' points to the first range that starts after 's'.
8666 * Maybe we can merge with the start of that range */
8667 sector_t a = BB_OFFSET(p[hi]);
8668 sector_t e = a + BB_LEN(p[hi]);
8669 int ack = BB_ACK(p[hi]);
8670 if (a <= s + sectors) {
8671 /* merging is possible */
8672 if (e <= s + sectors) {
8677 ack = ack && acknowledged;
8680 if (e - a <= BB_MAX_LEN) {
8681 p[hi] = BB_MAKE(a, e-a, ack);
8684 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8692 if (sectors == 0 && hi < bb->count) {
8693 /* we might be able to combine lo and hi */
8694 /* Note: 's' is at the end of 'lo' */
8695 sector_t a = BB_OFFSET(p[hi]);
8696 int lolen = BB_LEN(p[lo]);
8697 int hilen = BB_LEN(p[hi]);
8698 int newlen = lolen + hilen - (s - a);
8699 if (s >= a && newlen < BB_MAX_LEN) {
8700 /* yes, we can combine them */
8701 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8702 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8703 memmove(p + hi, p + hi + 1,
8704 (bb->count - hi - 1) * 8);
8709 /* didn't merge (it all).
8710 * Need to add a range just before 'hi' */
8711 if (bb->count >= MD_MAX_BADBLOCKS) {
8712 /* No room for more */
8716 int this_sectors = sectors;
8717 memmove(p + hi + 1, p + hi,
8718 (bb->count - hi) * 8);
8721 if (this_sectors > BB_MAX_LEN)
8722 this_sectors = BB_MAX_LEN;
8723 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8724 sectors -= this_sectors;
8731 bb->unacked_exist = 1;
8732 write_sequnlock_irqrestore(&bb->lock, flags);
8737 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8742 s += rdev->new_data_offset;
8744 s += rdev->data_offset;
8745 rv = md_set_badblocks(&rdev->badblocks,
8748 /* Make sure they get written out promptly */
8749 sysfs_notify_dirent_safe(rdev->sysfs_state);
8750 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8751 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8752 md_wakeup_thread(rdev->mddev->thread);
8756 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8759 * Remove a range of bad blocks from the table.
8760 * This may involve extending the table if we spilt a region,
8761 * but it must not fail. So if the table becomes full, we just
8762 * drop the remove request.
8764 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8768 sector_t target = s + sectors;
8771 if (bb->shift > 0) {
8772 /* When clearing we round the start up and the end down.
8773 * This should not matter as the shift should align with
8774 * the block size and no rounding should ever be needed.
8775 * However it is better the think a block is bad when it
8776 * isn't than to think a block is not bad when it is.
8778 s += (1<<bb->shift) - 1;
8780 target >>= bb->shift;
8781 sectors = target - s;
8784 write_seqlock_irq(&bb->lock);
8789 /* Find the last range that starts before 'target' */
8790 while (hi - lo > 1) {
8791 int mid = (lo + hi) / 2;
8792 sector_t a = BB_OFFSET(p[mid]);
8799 /* p[lo] is the last range that could overlap the
8800 * current range. Earlier ranges could also overlap,
8801 * but only this one can overlap the end of the range.
8803 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8804 /* Partial overlap, leave the tail of this range */
8805 int ack = BB_ACK(p[lo]);
8806 sector_t a = BB_OFFSET(p[lo]);
8807 sector_t end = a + BB_LEN(p[lo]);
8810 /* we need to split this range */
8811 if (bb->count >= MD_MAX_BADBLOCKS) {
8815 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8817 p[lo] = BB_MAKE(a, s-a, ack);
8820 p[lo] = BB_MAKE(target, end - target, ack);
8821 /* there is no longer an overlap */
8826 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8827 /* This range does overlap */
8828 if (BB_OFFSET(p[lo]) < s) {
8829 /* Keep the early parts of this range. */
8830 int ack = BB_ACK(p[lo]);
8831 sector_t start = BB_OFFSET(p[lo]);
8832 p[lo] = BB_MAKE(start, s - start, ack);
8833 /* now low doesn't overlap, so.. */
8838 /* 'lo' is strictly before, 'hi' is strictly after,
8839 * anything between needs to be discarded
8842 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8843 bb->count -= (hi - lo - 1);
8849 write_sequnlock_irq(&bb->lock);
8853 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8857 s += rdev->new_data_offset;
8859 s += rdev->data_offset;
8860 return md_clear_badblocks(&rdev->badblocks,
8863 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8866 * Acknowledge all bad blocks in a list.
8867 * This only succeeds if ->changed is clear. It is used by
8868 * in-kernel metadata updates
8870 void md_ack_all_badblocks(struct badblocks *bb)
8872 if (bb->page == NULL || bb->changed)
8873 /* no point even trying */
8875 write_seqlock_irq(&bb->lock);
8877 if (bb->changed == 0 && bb->unacked_exist) {
8880 for (i = 0; i < bb->count ; i++) {
8881 if (!BB_ACK(p[i])) {
8882 sector_t start = BB_OFFSET(p[i]);
8883 int len = BB_LEN(p[i]);
8884 p[i] = BB_MAKE(start, len, 1);
8887 bb->unacked_exist = 0;
8889 write_sequnlock_irq(&bb->lock);
8891 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8893 /* sysfs access to bad-blocks list.
8894 * We present two files.
8895 * 'bad-blocks' lists sector numbers and lengths of ranges that
8896 * are recorded as bad. The list is truncated to fit within
8897 * the one-page limit of sysfs.
8898 * Writing "sector length" to this file adds an acknowledged
8900 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8901 * been acknowledged. Writing to this file adds bad blocks
8902 * without acknowledging them. This is largely for testing.
8906 badblocks_show(struct badblocks *bb, char *page, int unack)
8917 seq = read_seqbegin(&bb->lock);
8922 while (len < PAGE_SIZE && i < bb->count) {
8923 sector_t s = BB_OFFSET(p[i]);
8924 unsigned int length = BB_LEN(p[i]);
8925 int ack = BB_ACK(p[i]);
8931 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8932 (unsigned long long)s << bb->shift,
8933 length << bb->shift);
8935 if (unack && len == 0)
8936 bb->unacked_exist = 0;
8938 if (read_seqretry(&bb->lock, seq))
8947 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8949 unsigned long long sector;
8953 /* Allow clearing via sysfs *only* for testing/debugging.
8954 * Normally only a successful write may clear a badblock
8957 if (page[0] == '-') {
8961 #endif /* DO_DEBUG */
8963 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8965 if (newline != '\n')
8977 md_clear_badblocks(bb, sector, length);
8980 #endif /* DO_DEBUG */
8981 if (md_set_badblocks(bb, sector, length, !unack))
8987 static int md_notify_reboot(struct notifier_block *this,
8988 unsigned long code, void *x)
8990 struct list_head *tmp;
8991 struct mddev *mddev;
8994 for_each_mddev(mddev, tmp) {
8995 if (mddev_trylock(mddev)) {
8997 __md_stop_writes(mddev);
8998 if (mddev->persistent)
8999 mddev->safemode = 2;
9000 mddev_unlock(mddev);
9005 * certain more exotic SCSI devices are known to be
9006 * volatile wrt too early system reboots. While the
9007 * right place to handle this issue is the given
9008 * driver, we do want to have a safe RAID driver ...
9016 static struct notifier_block md_notifier = {
9017 .notifier_call = md_notify_reboot,
9019 .priority = INT_MAX, /* before any real devices */
9022 static void md_geninit(void)
9024 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9026 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9029 static int __init md_init(void)
9033 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9037 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9041 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9044 if ((ret = register_blkdev(0, "mdp")) < 0)
9048 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9049 md_probe, NULL, NULL);
9050 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9051 md_probe, NULL, NULL);
9053 register_reboot_notifier(&md_notifier);
9054 raid_table_header = register_sysctl_table(raid_root_table);
9060 unregister_blkdev(MD_MAJOR, "md");
9062 destroy_workqueue(md_misc_wq);
9064 destroy_workqueue(md_wq);
9069 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9071 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9072 struct md_rdev *rdev2;
9074 char b[BDEVNAME_SIZE];
9076 /* Check for change of roles in the active devices */
9077 rdev_for_each(rdev2, mddev) {
9078 if (test_bit(Faulty, &rdev2->flags))
9081 /* Check if the roles changed */
9082 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9084 if (test_bit(Candidate, &rdev2->flags)) {
9085 if (role == 0xfffe) {
9086 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9087 md_kick_rdev_from_array(rdev2);
9091 clear_bit(Candidate, &rdev2->flags);
9094 if (role != rdev2->raid_disk) {
9096 if (rdev2->raid_disk == -1 && role != 0xffff) {
9097 rdev2->saved_raid_disk = role;
9098 ret = remove_and_add_spares(mddev, rdev2);
9099 pr_info("Activated spare: %s\n",
9100 bdevname(rdev2->bdev,b));
9104 * We just want to do the minimum to mark the disk
9105 * as faulty. The recovery is performed by the
9106 * one who initiated the error.
9108 if ((role == 0xfffe) || (role == 0xfffd)) {
9109 md_error(mddev, rdev2);
9110 clear_bit(Blocked, &rdev2->flags);
9115 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9116 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9118 /* Finally set the event to be up to date */
9119 mddev->events = le64_to_cpu(sb->events);
9122 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9125 struct page *swapout = rdev->sb_page;
9126 struct mdp_superblock_1 *sb;
9128 /* Store the sb page of the rdev in the swapout temporary
9129 * variable in case we err in the future
9131 rdev->sb_page = NULL;
9132 alloc_disk_sb(rdev);
9133 ClearPageUptodate(rdev->sb_page);
9134 rdev->sb_loaded = 0;
9135 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9138 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9139 __func__, __LINE__, rdev->desc_nr, err);
9140 put_page(rdev->sb_page);
9141 rdev->sb_page = swapout;
9142 rdev->sb_loaded = 1;
9146 sb = page_address(rdev->sb_page);
9147 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9151 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9152 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9154 /* The other node finished recovery, call spare_active to set
9155 * device In_sync and mddev->degraded
9157 if (rdev->recovery_offset == MaxSector &&
9158 !test_bit(In_sync, &rdev->flags) &&
9159 mddev->pers->spare_active(mddev))
9160 sysfs_notify(&mddev->kobj, NULL, "degraded");
9166 void md_reload_sb(struct mddev *mddev, int nr)
9168 struct md_rdev *rdev;
9172 rdev_for_each_rcu(rdev, mddev) {
9173 if (rdev->desc_nr == nr)
9177 if (!rdev || rdev->desc_nr != nr) {
9178 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9182 err = read_rdev(mddev, rdev);
9186 check_sb_changes(mddev, rdev);
9188 /* Read all rdev's to update recovery_offset */
9189 rdev_for_each_rcu(rdev, mddev)
9190 read_rdev(mddev, rdev);
9192 EXPORT_SYMBOL(md_reload_sb);
9197 * Searches all registered partitions for autorun RAID arrays
9201 static LIST_HEAD(all_detected_devices);
9202 struct detected_devices_node {
9203 struct list_head list;
9207 void md_autodetect_dev(dev_t dev)
9209 struct detected_devices_node *node_detected_dev;
9211 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9212 if (node_detected_dev) {
9213 node_detected_dev->dev = dev;
9214 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9216 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9217 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9221 static void autostart_arrays(int part)
9223 struct md_rdev *rdev;
9224 struct detected_devices_node *node_detected_dev;
9226 int i_scanned, i_passed;
9231 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
9233 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9235 node_detected_dev = list_entry(all_detected_devices.next,
9236 struct detected_devices_node, list);
9237 list_del(&node_detected_dev->list);
9238 dev = node_detected_dev->dev;
9239 kfree(node_detected_dev);
9240 rdev = md_import_device(dev,0, 90);
9244 if (test_bit(Faulty, &rdev->flags))
9247 set_bit(AutoDetected, &rdev->flags);
9248 list_add(&rdev->same_set, &pending_raid_disks);
9252 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9253 i_scanned, i_passed);
9255 autorun_devices(part);
9258 #endif /* !MODULE */
9260 static __exit void md_exit(void)
9262 struct mddev *mddev;
9263 struct list_head *tmp;
9266 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9267 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9269 unregister_blkdev(MD_MAJOR,"md");
9270 unregister_blkdev(mdp_major, "mdp");
9271 unregister_reboot_notifier(&md_notifier);
9272 unregister_sysctl_table(raid_table_header);
9274 /* We cannot unload the modules while some process is
9275 * waiting for us in select() or poll() - wake them up
9278 while (waitqueue_active(&md_event_waiters)) {
9279 /* not safe to leave yet */
9280 wake_up(&md_event_waiters);
9284 remove_proc_entry("mdstat", NULL);
9286 for_each_mddev(mddev, tmp) {
9287 export_array(mddev);
9288 mddev->hold_active = 0;
9290 destroy_workqueue(md_misc_wq);
9291 destroy_workqueue(md_wq);
9294 subsys_initcall(md_init);
9295 module_exit(md_exit)
9297 static int get_ro(char *buffer, struct kernel_param *kp)
9299 return sprintf(buffer, "%d", start_readonly);
9301 static int set_ro(const char *val, struct kernel_param *kp)
9303 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9306 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9307 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9308 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9310 MODULE_LICENSE("GPL");
9311 MODULE_DESCRIPTION("MD RAID framework");
9313 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);