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 /* bio could be mergeable after passing to underlayer */
297 bio->bi_rw &= ~REQ_NOMERGE;
298 mddev->pers->make_request(mddev, bio);
300 cpu = part_stat_lock();
301 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
302 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
305 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
306 wake_up(&mddev->sb_wait);
308 return BLK_QC_T_NONE;
311 /* mddev_suspend makes sure no new requests are submitted
312 * to the device, and that any requests that have been submitted
313 * are completely handled.
314 * Once mddev_detach() is called and completes, the module will be
317 void mddev_suspend(struct mddev *mddev)
319 if (mddev->suspended++)
322 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
323 mddev->pers->quiesce(mddev, 1);
325 del_timer_sync(&mddev->safemode_timer);
327 EXPORT_SYMBOL_GPL(mddev_suspend);
329 void mddev_resume(struct mddev *mddev)
331 if (--mddev->suspended)
333 wake_up(&mddev->sb_wait);
334 mddev->pers->quiesce(mddev, 0);
336 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
337 md_wakeup_thread(mddev->thread);
338 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
340 EXPORT_SYMBOL_GPL(mddev_resume);
342 int mddev_congested(struct mddev *mddev, int bits)
344 struct md_personality *pers = mddev->pers;
348 if (mddev->suspended)
350 else if (pers && pers->congested)
351 ret = pers->congested(mddev, bits);
355 EXPORT_SYMBOL_GPL(mddev_congested);
356 static int md_congested(void *data, int bits)
358 struct mddev *mddev = data;
359 return mddev_congested(mddev, bits);
363 * Generic flush handling for md
366 static void md_end_flush(struct bio *bio)
368 struct md_rdev *rdev = bio->bi_private;
369 struct mddev *mddev = rdev->mddev;
371 rdev_dec_pending(rdev, mddev);
373 if (atomic_dec_and_test(&mddev->flush_pending)) {
374 /* The pre-request flush has finished */
375 queue_work(md_wq, &mddev->flush_work);
380 static void md_submit_flush_data(struct work_struct *ws);
382 static void submit_flushes(struct work_struct *ws)
384 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
385 struct md_rdev *rdev;
387 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
388 atomic_set(&mddev->flush_pending, 1);
390 rdev_for_each_rcu(rdev, mddev)
391 if (rdev->raid_disk >= 0 &&
392 !test_bit(Faulty, &rdev->flags)) {
393 /* Take two references, one is dropped
394 * when request finishes, one after
395 * we reclaim rcu_read_lock
398 atomic_inc(&rdev->nr_pending);
399 atomic_inc(&rdev->nr_pending);
401 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
402 bi->bi_end_io = md_end_flush;
403 bi->bi_private = rdev;
404 bi->bi_bdev = rdev->bdev;
405 atomic_inc(&mddev->flush_pending);
406 submit_bio(WRITE_FLUSH, bi);
408 rdev_dec_pending(rdev, mddev);
411 if (atomic_dec_and_test(&mddev->flush_pending))
412 queue_work(md_wq, &mddev->flush_work);
415 static void md_submit_flush_data(struct work_struct *ws)
417 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
418 struct bio *bio = mddev->flush_bio;
420 if (bio->bi_iter.bi_size == 0)
421 /* an empty barrier - all done */
424 bio->bi_rw &= ~REQ_FLUSH;
425 mddev->pers->make_request(mddev, bio);
428 mddev->flush_bio = NULL;
429 wake_up(&mddev->sb_wait);
432 void md_flush_request(struct mddev *mddev, struct bio *bio)
434 spin_lock_irq(&mddev->lock);
435 wait_event_lock_irq(mddev->sb_wait,
438 mddev->flush_bio = bio;
439 spin_unlock_irq(&mddev->lock);
441 INIT_WORK(&mddev->flush_work, submit_flushes);
442 queue_work(md_wq, &mddev->flush_work);
444 EXPORT_SYMBOL(md_flush_request);
446 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
448 struct mddev *mddev = cb->data;
449 md_wakeup_thread(mddev->thread);
452 EXPORT_SYMBOL(md_unplug);
454 static inline struct mddev *mddev_get(struct mddev *mddev)
456 atomic_inc(&mddev->active);
460 static void mddev_delayed_delete(struct work_struct *ws);
462 static void mddev_put(struct mddev *mddev)
464 struct bio_set *bs = NULL;
466 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
468 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
469 mddev->ctime == 0 && !mddev->hold_active) {
470 /* Array is not configured at all, and not held active,
472 list_del_init(&mddev->all_mddevs);
474 mddev->bio_set = NULL;
475 if (mddev->gendisk) {
476 /* We did a probe so need to clean up. Call
477 * queue_work inside the spinlock so that
478 * flush_workqueue() after mddev_find will
479 * succeed in waiting for the work to be done.
481 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
482 queue_work(md_misc_wq, &mddev->del_work);
486 spin_unlock(&all_mddevs_lock);
491 static void md_safemode_timeout(unsigned long data);
493 void mddev_init(struct mddev *mddev)
495 mutex_init(&mddev->open_mutex);
496 mutex_init(&mddev->reconfig_mutex);
497 mutex_init(&mddev->bitmap_info.mutex);
498 INIT_LIST_HEAD(&mddev->disks);
499 INIT_LIST_HEAD(&mddev->all_mddevs);
500 setup_timer(&mddev->safemode_timer, md_safemode_timeout,
501 (unsigned long) mddev);
502 atomic_set(&mddev->active, 1);
503 atomic_set(&mddev->openers, 0);
504 atomic_set(&mddev->active_io, 0);
505 spin_lock_init(&mddev->lock);
506 atomic_set(&mddev->flush_pending, 0);
507 init_waitqueue_head(&mddev->sb_wait);
508 init_waitqueue_head(&mddev->recovery_wait);
509 mddev->reshape_position = MaxSector;
510 mddev->reshape_backwards = 0;
511 mddev->last_sync_action = "none";
512 mddev->resync_min = 0;
513 mddev->resync_max = MaxSector;
514 mddev->level = LEVEL_NONE;
516 EXPORT_SYMBOL_GPL(mddev_init);
518 static struct mddev *mddev_find(dev_t unit)
520 struct mddev *mddev, *new = NULL;
522 if (unit && MAJOR(unit) != MD_MAJOR)
523 unit &= ~((1<<MdpMinorShift)-1);
526 spin_lock(&all_mddevs_lock);
529 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
530 if (mddev->unit == unit) {
532 spin_unlock(&all_mddevs_lock);
538 list_add(&new->all_mddevs, &all_mddevs);
539 spin_unlock(&all_mddevs_lock);
540 new->hold_active = UNTIL_IOCTL;
544 /* find an unused unit number */
545 static int next_minor = 512;
546 int start = next_minor;
550 dev = MKDEV(MD_MAJOR, next_minor);
552 if (next_minor > MINORMASK)
554 if (next_minor == start) {
555 /* Oh dear, all in use. */
556 spin_unlock(&all_mddevs_lock);
562 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
563 if (mddev->unit == dev) {
569 new->md_minor = MINOR(dev);
570 new->hold_active = UNTIL_STOP;
571 list_add(&new->all_mddevs, &all_mddevs);
572 spin_unlock(&all_mddevs_lock);
575 spin_unlock(&all_mddevs_lock);
577 new = kzalloc(sizeof(*new), GFP_KERNEL);
582 if (MAJOR(unit) == MD_MAJOR)
583 new->md_minor = MINOR(unit);
585 new->md_minor = MINOR(unit) >> MdpMinorShift;
592 static struct attribute_group md_redundancy_group;
594 void mddev_unlock(struct mddev *mddev)
596 if (mddev->to_remove) {
597 /* These cannot be removed under reconfig_mutex as
598 * an access to the files will try to take reconfig_mutex
599 * while holding the file unremovable, which leads to
601 * So hold set sysfs_active while the remove in happeing,
602 * and anything else which might set ->to_remove or my
603 * otherwise change the sysfs namespace will fail with
604 * -EBUSY if sysfs_active is still set.
605 * We set sysfs_active under reconfig_mutex and elsewhere
606 * test it under the same mutex to ensure its correct value
609 struct attribute_group *to_remove = mddev->to_remove;
610 mddev->to_remove = NULL;
611 mddev->sysfs_active = 1;
612 mutex_unlock(&mddev->reconfig_mutex);
614 if (mddev->kobj.sd) {
615 if (to_remove != &md_redundancy_group)
616 sysfs_remove_group(&mddev->kobj, to_remove);
617 if (mddev->pers == NULL ||
618 mddev->pers->sync_request == NULL) {
619 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
620 if (mddev->sysfs_action)
621 sysfs_put(mddev->sysfs_action);
622 mddev->sysfs_action = NULL;
625 mddev->sysfs_active = 0;
627 mutex_unlock(&mddev->reconfig_mutex);
629 /* As we've dropped the mutex we need a spinlock to
630 * make sure the thread doesn't disappear
632 spin_lock(&pers_lock);
633 md_wakeup_thread(mddev->thread);
634 spin_unlock(&pers_lock);
636 EXPORT_SYMBOL_GPL(mddev_unlock);
638 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
640 struct md_rdev *rdev;
642 rdev_for_each_rcu(rdev, mddev)
643 if (rdev->desc_nr == nr)
648 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
650 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
652 struct md_rdev *rdev;
654 rdev_for_each(rdev, mddev)
655 if (rdev->bdev->bd_dev == dev)
661 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
663 struct md_rdev *rdev;
665 rdev_for_each_rcu(rdev, mddev)
666 if (rdev->bdev->bd_dev == dev)
672 static struct md_personality *find_pers(int level, char *clevel)
674 struct md_personality *pers;
675 list_for_each_entry(pers, &pers_list, list) {
676 if (level != LEVEL_NONE && pers->level == level)
678 if (strcmp(pers->name, clevel)==0)
684 /* return the offset of the super block in 512byte sectors */
685 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
687 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
688 return MD_NEW_SIZE_SECTORS(num_sectors);
691 static int alloc_disk_sb(struct md_rdev *rdev)
693 rdev->sb_page = alloc_page(GFP_KERNEL);
694 if (!rdev->sb_page) {
695 printk(KERN_ALERT "md: out of memory.\n");
702 void md_rdev_clear(struct md_rdev *rdev)
705 put_page(rdev->sb_page);
707 rdev->sb_page = NULL;
712 put_page(rdev->bb_page);
713 rdev->bb_page = NULL;
715 kfree(rdev->badblocks.page);
716 rdev->badblocks.page = NULL;
718 EXPORT_SYMBOL_GPL(md_rdev_clear);
720 static void super_written(struct bio *bio)
722 struct md_rdev *rdev = bio->bi_private;
723 struct mddev *mddev = rdev->mddev;
726 printk("md: super_written gets error=%d\n", bio->bi_error);
727 md_error(mddev, rdev);
730 if (atomic_dec_and_test(&mddev->pending_writes))
731 wake_up(&mddev->sb_wait);
735 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
736 sector_t sector, int size, struct page *page)
738 /* write first size bytes of page to sector of rdev
739 * Increment mddev->pending_writes before returning
740 * and decrement it on completion, waking up sb_wait
741 * if zero is reached.
742 * If an error occurred, call md_error
744 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
746 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
747 bio->bi_iter.bi_sector = sector;
748 bio_add_page(bio, page, size, 0);
749 bio->bi_private = rdev;
750 bio->bi_end_io = super_written;
752 atomic_inc(&mddev->pending_writes);
753 submit_bio(WRITE_FLUSH_FUA, bio);
756 void md_super_wait(struct mddev *mddev)
758 /* wait for all superblock writes that were scheduled to complete */
759 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
762 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
763 struct page *page, int rw, bool metadata_op)
765 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
768 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
769 rdev->meta_bdev : rdev->bdev;
771 bio->bi_iter.bi_sector = sector + rdev->sb_start;
772 else if (rdev->mddev->reshape_position != MaxSector &&
773 (rdev->mddev->reshape_backwards ==
774 (sector >= rdev->mddev->reshape_position)))
775 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
777 bio->bi_iter.bi_sector = sector + rdev->data_offset;
778 bio_add_page(bio, page, size, 0);
779 submit_bio_wait(rw, bio);
781 ret = !bio->bi_error;
785 EXPORT_SYMBOL_GPL(sync_page_io);
787 static int read_disk_sb(struct md_rdev *rdev, int size)
789 char b[BDEVNAME_SIZE];
794 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
800 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
801 bdevname(rdev->bdev,b));
805 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
807 return sb1->set_uuid0 == sb2->set_uuid0 &&
808 sb1->set_uuid1 == sb2->set_uuid1 &&
809 sb1->set_uuid2 == sb2->set_uuid2 &&
810 sb1->set_uuid3 == sb2->set_uuid3;
813 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
816 mdp_super_t *tmp1, *tmp2;
818 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
819 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
821 if (!tmp1 || !tmp2) {
823 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
831 * nr_disks is not constant
836 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
843 static u32 md_csum_fold(u32 csum)
845 csum = (csum & 0xffff) + (csum >> 16);
846 return (csum & 0xffff) + (csum >> 16);
849 static unsigned int calc_sb_csum(mdp_super_t *sb)
852 u32 *sb32 = (u32*)sb;
854 unsigned int disk_csum, csum;
856 disk_csum = sb->sb_csum;
859 for (i = 0; i < MD_SB_BYTES/4 ; i++)
861 csum = (newcsum & 0xffffffff) + (newcsum>>32);
864 /* This used to use csum_partial, which was wrong for several
865 * reasons including that different results are returned on
866 * different architectures. It isn't critical that we get exactly
867 * the same return value as before (we always csum_fold before
868 * testing, and that removes any differences). However as we
869 * know that csum_partial always returned a 16bit value on
870 * alphas, do a fold to maximise conformity to previous behaviour.
872 sb->sb_csum = md_csum_fold(disk_csum);
874 sb->sb_csum = disk_csum;
880 * Handle superblock details.
881 * We want to be able to handle multiple superblock formats
882 * so we have a common interface to them all, and an array of
883 * different handlers.
884 * We rely on user-space to write the initial superblock, and support
885 * reading and updating of superblocks.
886 * Interface methods are:
887 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
888 * loads and validates a superblock on dev.
889 * if refdev != NULL, compare superblocks on both devices
891 * 0 - dev has a superblock that is compatible with refdev
892 * 1 - dev has a superblock that is compatible and newer than refdev
893 * so dev should be used as the refdev in future
894 * -EINVAL superblock incompatible or invalid
895 * -othererror e.g. -EIO
897 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
898 * Verify that dev is acceptable into mddev.
899 * The first time, mddev->raid_disks will be 0, and data from
900 * dev should be merged in. Subsequent calls check that dev
901 * is new enough. Return 0 or -EINVAL
903 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
904 * Update the superblock for rdev with data in mddev
905 * This does not write to disc.
911 struct module *owner;
912 int (*load_super)(struct md_rdev *rdev,
913 struct md_rdev *refdev,
915 int (*validate_super)(struct mddev *mddev,
916 struct md_rdev *rdev);
917 void (*sync_super)(struct mddev *mddev,
918 struct md_rdev *rdev);
919 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
920 sector_t num_sectors);
921 int (*allow_new_offset)(struct md_rdev *rdev,
922 unsigned long long new_offset);
926 * Check that the given mddev has no bitmap.
928 * This function is called from the run method of all personalities that do not
929 * support bitmaps. It prints an error message and returns non-zero if mddev
930 * has a bitmap. Otherwise, it returns 0.
933 int md_check_no_bitmap(struct mddev *mddev)
935 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
937 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
938 mdname(mddev), mddev->pers->name);
941 EXPORT_SYMBOL(md_check_no_bitmap);
944 * load_super for 0.90.0
946 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
948 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
953 * Calculate the position of the superblock (512byte sectors),
954 * it's at the end of the disk.
956 * It also happens to be a multiple of 4Kb.
958 rdev->sb_start = calc_dev_sboffset(rdev);
960 ret = read_disk_sb(rdev, MD_SB_BYTES);
965 bdevname(rdev->bdev, b);
966 sb = page_address(rdev->sb_page);
968 if (sb->md_magic != MD_SB_MAGIC) {
969 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
974 if (sb->major_version != 0 ||
975 sb->minor_version < 90 ||
976 sb->minor_version > 91) {
977 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
978 sb->major_version, sb->minor_version,
983 if (sb->raid_disks <= 0)
986 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
987 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
992 rdev->preferred_minor = sb->md_minor;
993 rdev->data_offset = 0;
994 rdev->new_data_offset = 0;
995 rdev->sb_size = MD_SB_BYTES;
996 rdev->badblocks.shift = -1;
998 if (sb->level == LEVEL_MULTIPATH)
1001 rdev->desc_nr = sb->this_disk.number;
1007 mdp_super_t *refsb = page_address(refdev->sb_page);
1008 if (!uuid_equal(refsb, sb)) {
1009 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1010 b, bdevname(refdev->bdev,b2));
1013 if (!sb_equal(refsb, sb)) {
1014 printk(KERN_WARNING "md: %s has same UUID"
1015 " but different superblock to %s\n",
1016 b, bdevname(refdev->bdev, b2));
1020 ev2 = md_event(refsb);
1026 rdev->sectors = rdev->sb_start;
1027 /* Limit to 4TB as metadata cannot record more than that.
1028 * (not needed for Linear and RAID0 as metadata doesn't
1031 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1032 rdev->sectors = (2ULL << 32) - 2;
1034 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1035 /* "this cannot possibly happen" ... */
1043 * validate_super for 0.90.0
1045 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1048 mdp_super_t *sb = page_address(rdev->sb_page);
1049 __u64 ev1 = md_event(sb);
1051 rdev->raid_disk = -1;
1052 clear_bit(Faulty, &rdev->flags);
1053 clear_bit(In_sync, &rdev->flags);
1054 clear_bit(Bitmap_sync, &rdev->flags);
1055 clear_bit(WriteMostly, &rdev->flags);
1057 if (mddev->raid_disks == 0) {
1058 mddev->major_version = 0;
1059 mddev->minor_version = sb->minor_version;
1060 mddev->patch_version = sb->patch_version;
1061 mddev->external = 0;
1062 mddev->chunk_sectors = sb->chunk_size >> 9;
1063 mddev->ctime = sb->ctime;
1064 mddev->utime = sb->utime;
1065 mddev->level = sb->level;
1066 mddev->clevel[0] = 0;
1067 mddev->layout = sb->layout;
1068 mddev->raid_disks = sb->raid_disks;
1069 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1070 mddev->events = ev1;
1071 mddev->bitmap_info.offset = 0;
1072 mddev->bitmap_info.space = 0;
1073 /* bitmap can use 60 K after the 4K superblocks */
1074 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1075 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1076 mddev->reshape_backwards = 0;
1078 if (mddev->minor_version >= 91) {
1079 mddev->reshape_position = sb->reshape_position;
1080 mddev->delta_disks = sb->delta_disks;
1081 mddev->new_level = sb->new_level;
1082 mddev->new_layout = sb->new_layout;
1083 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1084 if (mddev->delta_disks < 0)
1085 mddev->reshape_backwards = 1;
1087 mddev->reshape_position = MaxSector;
1088 mddev->delta_disks = 0;
1089 mddev->new_level = mddev->level;
1090 mddev->new_layout = mddev->layout;
1091 mddev->new_chunk_sectors = mddev->chunk_sectors;
1094 if (sb->state & (1<<MD_SB_CLEAN))
1095 mddev->recovery_cp = MaxSector;
1097 if (sb->events_hi == sb->cp_events_hi &&
1098 sb->events_lo == sb->cp_events_lo) {
1099 mddev->recovery_cp = sb->recovery_cp;
1101 mddev->recovery_cp = 0;
1104 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1105 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1106 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1107 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1109 mddev->max_disks = MD_SB_DISKS;
1111 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1112 mddev->bitmap_info.file == NULL) {
1113 mddev->bitmap_info.offset =
1114 mddev->bitmap_info.default_offset;
1115 mddev->bitmap_info.space =
1116 mddev->bitmap_info.default_space;
1119 } else if (mddev->pers == NULL) {
1120 /* Insist on good event counter while assembling, except
1121 * for spares (which don't need an event count) */
1123 if (sb->disks[rdev->desc_nr].state & (
1124 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1125 if (ev1 < mddev->events)
1127 } else if (mddev->bitmap) {
1128 /* if adding to array with a bitmap, then we can accept an
1129 * older device ... but not too old.
1131 if (ev1 < mddev->bitmap->events_cleared)
1133 if (ev1 < mddev->events)
1134 set_bit(Bitmap_sync, &rdev->flags);
1136 if (ev1 < mddev->events)
1137 /* just a hot-add of a new device, leave raid_disk at -1 */
1141 if (mddev->level != LEVEL_MULTIPATH) {
1142 desc = sb->disks + rdev->desc_nr;
1144 if (desc->state & (1<<MD_DISK_FAULTY))
1145 set_bit(Faulty, &rdev->flags);
1146 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1147 desc->raid_disk < mddev->raid_disks */) {
1148 set_bit(In_sync, &rdev->flags);
1149 rdev->raid_disk = desc->raid_disk;
1150 rdev->saved_raid_disk = desc->raid_disk;
1151 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1152 /* active but not in sync implies recovery up to
1153 * reshape position. We don't know exactly where
1154 * that is, so set to zero for now */
1155 if (mddev->minor_version >= 91) {
1156 rdev->recovery_offset = 0;
1157 rdev->raid_disk = desc->raid_disk;
1160 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1161 set_bit(WriteMostly, &rdev->flags);
1162 } else /* MULTIPATH are always insync */
1163 set_bit(In_sync, &rdev->flags);
1168 * sync_super for 0.90.0
1170 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1173 struct md_rdev *rdev2;
1174 int next_spare = mddev->raid_disks;
1176 /* make rdev->sb match mddev data..
1179 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1180 * 3/ any empty disks < next_spare become removed
1182 * disks[0] gets initialised to REMOVED because
1183 * we cannot be sure from other fields if it has
1184 * been initialised or not.
1187 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1189 rdev->sb_size = MD_SB_BYTES;
1191 sb = page_address(rdev->sb_page);
1193 memset(sb, 0, sizeof(*sb));
1195 sb->md_magic = MD_SB_MAGIC;
1196 sb->major_version = mddev->major_version;
1197 sb->patch_version = mddev->patch_version;
1198 sb->gvalid_words = 0; /* ignored */
1199 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1200 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1201 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1202 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1204 sb->ctime = mddev->ctime;
1205 sb->level = mddev->level;
1206 sb->size = mddev->dev_sectors / 2;
1207 sb->raid_disks = mddev->raid_disks;
1208 sb->md_minor = mddev->md_minor;
1209 sb->not_persistent = 0;
1210 sb->utime = mddev->utime;
1212 sb->events_hi = (mddev->events>>32);
1213 sb->events_lo = (u32)mddev->events;
1215 if (mddev->reshape_position == MaxSector)
1216 sb->minor_version = 90;
1218 sb->minor_version = 91;
1219 sb->reshape_position = mddev->reshape_position;
1220 sb->new_level = mddev->new_level;
1221 sb->delta_disks = mddev->delta_disks;
1222 sb->new_layout = mddev->new_layout;
1223 sb->new_chunk = mddev->new_chunk_sectors << 9;
1225 mddev->minor_version = sb->minor_version;
1228 sb->recovery_cp = mddev->recovery_cp;
1229 sb->cp_events_hi = (mddev->events>>32);
1230 sb->cp_events_lo = (u32)mddev->events;
1231 if (mddev->recovery_cp == MaxSector)
1232 sb->state = (1<< MD_SB_CLEAN);
1234 sb->recovery_cp = 0;
1236 sb->layout = mddev->layout;
1237 sb->chunk_size = mddev->chunk_sectors << 9;
1239 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1240 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1242 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1243 rdev_for_each(rdev2, mddev) {
1246 int is_active = test_bit(In_sync, &rdev2->flags);
1248 if (rdev2->raid_disk >= 0 &&
1249 sb->minor_version >= 91)
1250 /* we have nowhere to store the recovery_offset,
1251 * but if it is not below the reshape_position,
1252 * we can piggy-back on that.
1255 if (rdev2->raid_disk < 0 ||
1256 test_bit(Faulty, &rdev2->flags))
1259 desc_nr = rdev2->raid_disk;
1261 desc_nr = next_spare++;
1262 rdev2->desc_nr = desc_nr;
1263 d = &sb->disks[rdev2->desc_nr];
1265 d->number = rdev2->desc_nr;
1266 d->major = MAJOR(rdev2->bdev->bd_dev);
1267 d->minor = MINOR(rdev2->bdev->bd_dev);
1269 d->raid_disk = rdev2->raid_disk;
1271 d->raid_disk = rdev2->desc_nr; /* compatibility */
1272 if (test_bit(Faulty, &rdev2->flags))
1273 d->state = (1<<MD_DISK_FAULTY);
1274 else if (is_active) {
1275 d->state = (1<<MD_DISK_ACTIVE);
1276 if (test_bit(In_sync, &rdev2->flags))
1277 d->state |= (1<<MD_DISK_SYNC);
1285 if (test_bit(WriteMostly, &rdev2->flags))
1286 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1288 /* now set the "removed" and "faulty" bits on any missing devices */
1289 for (i=0 ; i < mddev->raid_disks ; i++) {
1290 mdp_disk_t *d = &sb->disks[i];
1291 if (d->state == 0 && d->number == 0) {
1294 d->state = (1<<MD_DISK_REMOVED);
1295 d->state |= (1<<MD_DISK_FAULTY);
1299 sb->nr_disks = nr_disks;
1300 sb->active_disks = active;
1301 sb->working_disks = working;
1302 sb->failed_disks = failed;
1303 sb->spare_disks = spare;
1305 sb->this_disk = sb->disks[rdev->desc_nr];
1306 sb->sb_csum = calc_sb_csum(sb);
1310 * rdev_size_change for 0.90.0
1312 static unsigned long long
1313 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1315 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1316 return 0; /* component must fit device */
1317 if (rdev->mddev->bitmap_info.offset)
1318 return 0; /* can't move bitmap */
1319 rdev->sb_start = calc_dev_sboffset(rdev);
1320 if (!num_sectors || num_sectors > rdev->sb_start)
1321 num_sectors = rdev->sb_start;
1322 /* Limit to 4TB as metadata cannot record more than that.
1323 * 4TB == 2^32 KB, or 2*2^32 sectors.
1325 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1326 num_sectors = (2ULL << 32) - 2;
1327 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1329 md_super_wait(rdev->mddev);
1334 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1336 /* non-zero offset changes not possible with v0.90 */
1337 return new_offset == 0;
1341 * version 1 superblock
1344 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1348 unsigned long long newcsum;
1349 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1350 __le32 *isuper = (__le32*)sb;
1352 disk_csum = sb->sb_csum;
1355 for (; size >= 4; size -= 4)
1356 newcsum += le32_to_cpu(*isuper++);
1359 newcsum += le16_to_cpu(*(__le16*) isuper);
1361 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1362 sb->sb_csum = disk_csum;
1363 return cpu_to_le32(csum);
1366 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1368 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1370 struct mdp_superblock_1 *sb;
1374 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1378 * Calculate the position of the superblock in 512byte sectors.
1379 * It is always aligned to a 4K boundary and
1380 * depeding on minor_version, it can be:
1381 * 0: At least 8K, but less than 12K, from end of device
1382 * 1: At start of device
1383 * 2: 4K from start of device.
1385 switch(minor_version) {
1387 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1389 sb_start &= ~(sector_t)(4*2-1);
1400 rdev->sb_start = sb_start;
1402 /* superblock is rarely larger than 1K, but it can be larger,
1403 * and it is safe to read 4k, so we do that
1405 ret = read_disk_sb(rdev, 4096);
1406 if (ret) return ret;
1408 sb = page_address(rdev->sb_page);
1410 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1411 sb->major_version != cpu_to_le32(1) ||
1412 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1413 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1414 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1417 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1418 printk("md: invalid superblock checksum on %s\n",
1419 bdevname(rdev->bdev,b));
1422 if (le64_to_cpu(sb->data_size) < 10) {
1423 printk("md: data_size too small on %s\n",
1424 bdevname(rdev->bdev,b));
1429 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1430 /* Some padding is non-zero, might be a new feature */
1433 rdev->preferred_minor = 0xffff;
1434 rdev->data_offset = le64_to_cpu(sb->data_offset);
1435 rdev->new_data_offset = rdev->data_offset;
1436 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1437 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1438 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1439 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1441 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1442 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1443 if (rdev->sb_size & bmask)
1444 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1447 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1450 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1453 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1456 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1458 if (!rdev->bb_page) {
1459 rdev->bb_page = alloc_page(GFP_KERNEL);
1463 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1464 rdev->badblocks.count == 0) {
1465 /* need to load the bad block list.
1466 * Currently we limit it to one page.
1472 int sectors = le16_to_cpu(sb->bblog_size);
1473 if (sectors > (PAGE_SIZE / 512))
1475 offset = le32_to_cpu(sb->bblog_offset);
1478 bb_sector = (long long)offset;
1479 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1480 rdev->bb_page, READ, true))
1482 bbp = (u64 *)page_address(rdev->bb_page);
1483 rdev->badblocks.shift = sb->bblog_shift;
1484 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1485 u64 bb = le64_to_cpu(*bbp);
1486 int count = bb & (0x3ff);
1487 u64 sector = bb >> 10;
1488 sector <<= sb->bblog_shift;
1489 count <<= sb->bblog_shift;
1492 if (md_set_badblocks(&rdev->badblocks,
1493 sector, count, 1) == 0)
1496 } else if (sb->bblog_offset != 0)
1497 rdev->badblocks.shift = 0;
1503 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1505 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1506 sb->level != refsb->level ||
1507 sb->layout != refsb->layout ||
1508 sb->chunksize != refsb->chunksize) {
1509 printk(KERN_WARNING "md: %s has strangely different"
1510 " superblock to %s\n",
1511 bdevname(rdev->bdev,b),
1512 bdevname(refdev->bdev,b2));
1515 ev1 = le64_to_cpu(sb->events);
1516 ev2 = le64_to_cpu(refsb->events);
1523 if (minor_version) {
1524 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1525 sectors -= rdev->data_offset;
1527 sectors = rdev->sb_start;
1528 if (sectors < le64_to_cpu(sb->data_size))
1530 rdev->sectors = le64_to_cpu(sb->data_size);
1534 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1536 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1537 __u64 ev1 = le64_to_cpu(sb->events);
1539 rdev->raid_disk = -1;
1540 clear_bit(Faulty, &rdev->flags);
1541 clear_bit(In_sync, &rdev->flags);
1542 clear_bit(Bitmap_sync, &rdev->flags);
1543 clear_bit(WriteMostly, &rdev->flags);
1545 if (mddev->raid_disks == 0) {
1546 mddev->major_version = 1;
1547 mddev->patch_version = 0;
1548 mddev->external = 0;
1549 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1550 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1551 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1552 mddev->level = le32_to_cpu(sb->level);
1553 mddev->clevel[0] = 0;
1554 mddev->layout = le32_to_cpu(sb->layout);
1555 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1556 mddev->dev_sectors = le64_to_cpu(sb->size);
1557 mddev->events = ev1;
1558 mddev->bitmap_info.offset = 0;
1559 mddev->bitmap_info.space = 0;
1560 /* Default location for bitmap is 1K after superblock
1561 * using 3K - total of 4K
1563 mddev->bitmap_info.default_offset = 1024 >> 9;
1564 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1565 mddev->reshape_backwards = 0;
1567 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1568 memcpy(mddev->uuid, sb->set_uuid, 16);
1570 mddev->max_disks = (4096-256)/2;
1572 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1573 mddev->bitmap_info.file == NULL) {
1574 mddev->bitmap_info.offset =
1575 (__s32)le32_to_cpu(sb->bitmap_offset);
1576 /* Metadata doesn't record how much space is available.
1577 * For 1.0, we assume we can use up to the superblock
1578 * if before, else to 4K beyond superblock.
1579 * For others, assume no change is possible.
1581 if (mddev->minor_version > 0)
1582 mddev->bitmap_info.space = 0;
1583 else if (mddev->bitmap_info.offset > 0)
1584 mddev->bitmap_info.space =
1585 8 - mddev->bitmap_info.offset;
1587 mddev->bitmap_info.space =
1588 -mddev->bitmap_info.offset;
1591 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1592 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1593 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1594 mddev->new_level = le32_to_cpu(sb->new_level);
1595 mddev->new_layout = le32_to_cpu(sb->new_layout);
1596 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1597 if (mddev->delta_disks < 0 ||
1598 (mddev->delta_disks == 0 &&
1599 (le32_to_cpu(sb->feature_map)
1600 & MD_FEATURE_RESHAPE_BACKWARDS)))
1601 mddev->reshape_backwards = 1;
1603 mddev->reshape_position = MaxSector;
1604 mddev->delta_disks = 0;
1605 mddev->new_level = mddev->level;
1606 mddev->new_layout = mddev->layout;
1607 mddev->new_chunk_sectors = mddev->chunk_sectors;
1610 } else if (mddev->pers == NULL) {
1611 /* Insist of good event counter while assembling, except for
1612 * spares (which don't need an event count) */
1614 if (rdev->desc_nr >= 0 &&
1615 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1616 (le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < MD_DISK_ROLE_MAX ||
1617 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) == MD_DISK_ROLE_JOURNAL))
1618 if (ev1 < mddev->events)
1620 } else if (mddev->bitmap) {
1621 /* If adding to array with a bitmap, then we can accept an
1622 * older device, but not too old.
1624 if (ev1 < mddev->bitmap->events_cleared)
1626 if (ev1 < mddev->events)
1627 set_bit(Bitmap_sync, &rdev->flags);
1629 if (ev1 < mddev->events)
1630 /* just a hot-add of a new device, leave raid_disk at -1 */
1633 if (mddev->level != LEVEL_MULTIPATH) {
1635 if (rdev->desc_nr < 0 ||
1636 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1637 role = MD_DISK_ROLE_SPARE;
1640 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1642 case MD_DISK_ROLE_SPARE: /* spare */
1644 case MD_DISK_ROLE_FAULTY: /* faulty */
1645 set_bit(Faulty, &rdev->flags);
1647 case MD_DISK_ROLE_JOURNAL: /* journal device */
1648 if (!(le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)) {
1649 /* journal device without journal feature */
1651 "md: journal device provided without journal feature, ignoring the device\n");
1654 set_bit(Journal, &rdev->flags);
1655 rdev->journal_tail = le64_to_cpu(sb->journal_tail);
1656 if (mddev->recovery_cp == MaxSector)
1657 set_bit(MD_JOURNAL_CLEAN, &mddev->flags);
1658 rdev->raid_disk = 0;
1661 rdev->saved_raid_disk = role;
1662 if ((le32_to_cpu(sb->feature_map) &
1663 MD_FEATURE_RECOVERY_OFFSET)) {
1664 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1665 if (!(le32_to_cpu(sb->feature_map) &
1666 MD_FEATURE_RECOVERY_BITMAP))
1667 rdev->saved_raid_disk = -1;
1669 set_bit(In_sync, &rdev->flags);
1670 rdev->raid_disk = role;
1673 if (sb->devflags & WriteMostly1)
1674 set_bit(WriteMostly, &rdev->flags);
1675 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1676 set_bit(Replacement, &rdev->flags);
1677 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_JOURNAL)
1678 set_bit(MD_HAS_JOURNAL, &mddev->flags);
1679 } else /* MULTIPATH are always insync */
1680 set_bit(In_sync, &rdev->flags);
1685 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1687 struct mdp_superblock_1 *sb;
1688 struct md_rdev *rdev2;
1690 /* make rdev->sb match mddev and rdev data. */
1692 sb = page_address(rdev->sb_page);
1694 sb->feature_map = 0;
1696 sb->recovery_offset = cpu_to_le64(0);
1697 memset(sb->pad3, 0, sizeof(sb->pad3));
1699 sb->utime = cpu_to_le64((__u64)mddev->utime);
1700 sb->events = cpu_to_le64(mddev->events);
1702 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1703 else if (test_bit(MD_JOURNAL_CLEAN, &mddev->flags))
1704 sb->resync_offset = cpu_to_le64(MaxSector);
1706 sb->resync_offset = cpu_to_le64(0);
1708 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1710 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1711 sb->size = cpu_to_le64(mddev->dev_sectors);
1712 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1713 sb->level = cpu_to_le32(mddev->level);
1714 sb->layout = cpu_to_le32(mddev->layout);
1716 if (test_bit(WriteMostly, &rdev->flags))
1717 sb->devflags |= WriteMostly1;
1719 sb->devflags &= ~WriteMostly1;
1720 sb->data_offset = cpu_to_le64(rdev->data_offset);
1721 sb->data_size = cpu_to_le64(rdev->sectors);
1723 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1724 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1725 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1728 if (rdev->raid_disk >= 0 && !test_bit(Journal, &rdev->flags) &&
1729 !test_bit(In_sync, &rdev->flags)) {
1731 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1732 sb->recovery_offset =
1733 cpu_to_le64(rdev->recovery_offset);
1734 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1736 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1738 /* Note: recovery_offset and journal_tail share space */
1739 if (test_bit(Journal, &rdev->flags))
1740 sb->journal_tail = cpu_to_le64(rdev->journal_tail);
1741 if (test_bit(Replacement, &rdev->flags))
1743 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1745 if (mddev->reshape_position != MaxSector) {
1746 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1747 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1748 sb->new_layout = cpu_to_le32(mddev->new_layout);
1749 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1750 sb->new_level = cpu_to_le32(mddev->new_level);
1751 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1752 if (mddev->delta_disks == 0 &&
1753 mddev->reshape_backwards)
1755 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1756 if (rdev->new_data_offset != rdev->data_offset) {
1758 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1759 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1760 - rdev->data_offset));
1764 if (mddev_is_clustered(mddev))
1765 sb->feature_map |= cpu_to_le32(MD_FEATURE_CLUSTERED);
1767 if (rdev->badblocks.count == 0)
1768 /* Nothing to do for bad blocks*/ ;
1769 else if (sb->bblog_offset == 0)
1770 /* Cannot record bad blocks on this device */
1771 md_error(mddev, rdev);
1773 struct badblocks *bb = &rdev->badblocks;
1774 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1776 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1781 seq = read_seqbegin(&bb->lock);
1783 memset(bbp, 0xff, PAGE_SIZE);
1785 for (i = 0 ; i < bb->count ; i++) {
1786 u64 internal_bb = p[i];
1787 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1788 | BB_LEN(internal_bb));
1789 bbp[i] = cpu_to_le64(store_bb);
1792 if (read_seqretry(&bb->lock, seq))
1795 bb->sector = (rdev->sb_start +
1796 (int)le32_to_cpu(sb->bblog_offset));
1797 bb->size = le16_to_cpu(sb->bblog_size);
1802 rdev_for_each(rdev2, mddev)
1803 if (rdev2->desc_nr+1 > max_dev)
1804 max_dev = rdev2->desc_nr+1;
1806 if (max_dev > le32_to_cpu(sb->max_dev)) {
1808 sb->max_dev = cpu_to_le32(max_dev);
1809 rdev->sb_size = max_dev * 2 + 256;
1810 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1811 if (rdev->sb_size & bmask)
1812 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1814 max_dev = le32_to_cpu(sb->max_dev);
1816 for (i=0; i<max_dev;i++)
1817 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1819 if (test_bit(MD_HAS_JOURNAL, &mddev->flags))
1820 sb->feature_map |= cpu_to_le32(MD_FEATURE_JOURNAL);
1822 rdev_for_each(rdev2, mddev) {
1824 if (test_bit(Faulty, &rdev2->flags))
1825 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_FAULTY);
1826 else if (test_bit(In_sync, &rdev2->flags))
1827 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1828 else if (test_bit(Journal, &rdev2->flags))
1829 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_JOURNAL);
1830 else if (rdev2->raid_disk >= 0)
1831 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1833 sb->dev_roles[i] = cpu_to_le16(MD_DISK_ROLE_SPARE);
1836 sb->sb_csum = calc_sb_1_csum(sb);
1839 static unsigned long long
1840 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1842 struct mdp_superblock_1 *sb;
1843 sector_t max_sectors;
1844 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1845 return 0; /* component must fit device */
1846 if (rdev->data_offset != rdev->new_data_offset)
1847 return 0; /* too confusing */
1848 if (rdev->sb_start < rdev->data_offset) {
1849 /* minor versions 1 and 2; superblock before data */
1850 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1851 max_sectors -= rdev->data_offset;
1852 if (!num_sectors || num_sectors > max_sectors)
1853 num_sectors = max_sectors;
1854 } else if (rdev->mddev->bitmap_info.offset) {
1855 /* minor version 0 with bitmap we can't move */
1858 /* minor version 0; superblock after data */
1860 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1861 sb_start &= ~(sector_t)(4*2 - 1);
1862 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1863 if (!num_sectors || num_sectors > max_sectors)
1864 num_sectors = max_sectors;
1865 rdev->sb_start = sb_start;
1867 sb = page_address(rdev->sb_page);
1868 sb->data_size = cpu_to_le64(num_sectors);
1869 sb->super_offset = rdev->sb_start;
1870 sb->sb_csum = calc_sb_1_csum(sb);
1871 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1873 md_super_wait(rdev->mddev);
1879 super_1_allow_new_offset(struct md_rdev *rdev,
1880 unsigned long long new_offset)
1882 /* All necessary checks on new >= old have been done */
1883 struct bitmap *bitmap;
1884 if (new_offset >= rdev->data_offset)
1887 /* with 1.0 metadata, there is no metadata to tread on
1888 * so we can always move back */
1889 if (rdev->mddev->minor_version == 0)
1892 /* otherwise we must be sure not to step on
1893 * any metadata, so stay:
1894 * 36K beyond start of superblock
1895 * beyond end of badblocks
1896 * beyond write-intent bitmap
1898 if (rdev->sb_start + (32+4)*2 > new_offset)
1900 bitmap = rdev->mddev->bitmap;
1901 if (bitmap && !rdev->mddev->bitmap_info.file &&
1902 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1903 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1905 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1911 static struct super_type super_types[] = {
1914 .owner = THIS_MODULE,
1915 .load_super = super_90_load,
1916 .validate_super = super_90_validate,
1917 .sync_super = super_90_sync,
1918 .rdev_size_change = super_90_rdev_size_change,
1919 .allow_new_offset = super_90_allow_new_offset,
1923 .owner = THIS_MODULE,
1924 .load_super = super_1_load,
1925 .validate_super = super_1_validate,
1926 .sync_super = super_1_sync,
1927 .rdev_size_change = super_1_rdev_size_change,
1928 .allow_new_offset = super_1_allow_new_offset,
1932 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1934 if (mddev->sync_super) {
1935 mddev->sync_super(mddev, rdev);
1939 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1941 super_types[mddev->major_version].sync_super(mddev, rdev);
1944 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1946 struct md_rdev *rdev, *rdev2;
1949 rdev_for_each_rcu(rdev, mddev1) {
1950 if (test_bit(Faulty, &rdev->flags) ||
1951 test_bit(Journal, &rdev->flags) ||
1952 rdev->raid_disk == -1)
1954 rdev_for_each_rcu(rdev2, mddev2) {
1955 if (test_bit(Faulty, &rdev2->flags) ||
1956 test_bit(Journal, &rdev2->flags) ||
1957 rdev2->raid_disk == -1)
1959 if (rdev->bdev->bd_contains ==
1960 rdev2->bdev->bd_contains) {
1970 static LIST_HEAD(pending_raid_disks);
1973 * Try to register data integrity profile for an mddev
1975 * This is called when an array is started and after a disk has been kicked
1976 * from the array. It only succeeds if all working and active component devices
1977 * are integrity capable with matching profiles.
1979 int md_integrity_register(struct mddev *mddev)
1981 struct md_rdev *rdev, *reference = NULL;
1983 if (list_empty(&mddev->disks))
1984 return 0; /* nothing to do */
1985 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1986 return 0; /* shouldn't register, or already is */
1987 rdev_for_each(rdev, mddev) {
1988 /* skip spares and non-functional disks */
1989 if (test_bit(Faulty, &rdev->flags))
1991 if (rdev->raid_disk < 0)
1994 /* Use the first rdev as the reference */
1998 /* does this rdev's profile match the reference profile? */
1999 if (blk_integrity_compare(reference->bdev->bd_disk,
2000 rdev->bdev->bd_disk) < 0)
2003 if (!reference || !bdev_get_integrity(reference->bdev))
2006 * All component devices are integrity capable and have matching
2007 * profiles, register the common profile for the md device.
2009 blk_integrity_register(mddev->gendisk,
2010 bdev_get_integrity(reference->bdev));
2012 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2013 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2014 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2020 EXPORT_SYMBOL(md_integrity_register);
2023 * Attempt to add an rdev, but only if it is consistent with the current
2026 int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
2028 struct blk_integrity *bi_rdev;
2029 struct blk_integrity *bi_mddev;
2030 char name[BDEVNAME_SIZE];
2032 if (!mddev->gendisk)
2035 bi_rdev = bdev_get_integrity(rdev->bdev);
2036 bi_mddev = blk_get_integrity(mddev->gendisk);
2038 if (!bi_mddev) /* nothing to do */
2041 if (blk_integrity_compare(mddev->gendisk, rdev->bdev->bd_disk) != 0) {
2042 printk(KERN_NOTICE "%s: incompatible integrity profile for %s\n",
2043 mdname(mddev), bdevname(rdev->bdev, name));
2049 EXPORT_SYMBOL(md_integrity_add_rdev);
2051 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2053 char b[BDEVNAME_SIZE];
2057 /* prevent duplicates */
2058 if (find_rdev(mddev, rdev->bdev->bd_dev))
2061 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2062 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2063 rdev->sectors < mddev->dev_sectors)) {
2065 /* Cannot change size, so fail
2066 * If mddev->level <= 0, then we don't care
2067 * about aligning sizes (e.g. linear)
2069 if (mddev->level > 0)
2072 mddev->dev_sectors = rdev->sectors;
2075 /* Verify rdev->desc_nr is unique.
2076 * If it is -1, assign a free number, else
2077 * check number is not in use
2080 if (rdev->desc_nr < 0) {
2083 choice = mddev->raid_disks;
2084 while (md_find_rdev_nr_rcu(mddev, choice))
2086 rdev->desc_nr = choice;
2088 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2094 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2095 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2096 mdname(mddev), mddev->max_disks);
2099 bdevname(rdev->bdev,b);
2100 strreplace(b, '/', '!');
2102 rdev->mddev = mddev;
2103 printk(KERN_INFO "md: bind<%s>\n", b);
2105 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2108 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2109 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2110 /* failure here is OK */;
2111 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2113 list_add_rcu(&rdev->same_set, &mddev->disks);
2114 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2116 /* May as well allow recovery to be retried once */
2117 mddev->recovery_disabled++;
2122 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2127 static void md_delayed_delete(struct work_struct *ws)
2129 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2130 kobject_del(&rdev->kobj);
2131 kobject_put(&rdev->kobj);
2134 static void unbind_rdev_from_array(struct md_rdev *rdev)
2136 char b[BDEVNAME_SIZE];
2138 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2139 list_del_rcu(&rdev->same_set);
2140 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2142 sysfs_remove_link(&rdev->kobj, "block");
2143 sysfs_put(rdev->sysfs_state);
2144 rdev->sysfs_state = NULL;
2145 rdev->badblocks.count = 0;
2146 /* We need to delay this, otherwise we can deadlock when
2147 * writing to 'remove' to "dev/state". We also need
2148 * to delay it due to rcu usage.
2151 INIT_WORK(&rdev->del_work, md_delayed_delete);
2152 kobject_get(&rdev->kobj);
2153 queue_work(md_misc_wq, &rdev->del_work);
2157 * prevent the device from being mounted, repartitioned or
2158 * otherwise reused by a RAID array (or any other kernel
2159 * subsystem), by bd_claiming the device.
2161 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2164 struct block_device *bdev;
2165 char b[BDEVNAME_SIZE];
2167 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2168 shared ? (struct md_rdev *)lock_rdev : rdev);
2170 printk(KERN_ERR "md: could not open %s.\n",
2171 __bdevname(dev, b));
2172 return PTR_ERR(bdev);
2178 static void unlock_rdev(struct md_rdev *rdev)
2180 struct block_device *bdev = rdev->bdev;
2182 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2185 void md_autodetect_dev(dev_t dev);
2187 static void export_rdev(struct md_rdev *rdev)
2189 char b[BDEVNAME_SIZE];
2191 printk(KERN_INFO "md: export_rdev(%s)\n",
2192 bdevname(rdev->bdev,b));
2193 md_rdev_clear(rdev);
2195 if (test_bit(AutoDetected, &rdev->flags))
2196 md_autodetect_dev(rdev->bdev->bd_dev);
2199 kobject_put(&rdev->kobj);
2202 void md_kick_rdev_from_array(struct md_rdev *rdev)
2204 unbind_rdev_from_array(rdev);
2207 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2209 static void export_array(struct mddev *mddev)
2211 struct md_rdev *rdev;
2213 while (!list_empty(&mddev->disks)) {
2214 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2216 md_kick_rdev_from_array(rdev);
2218 mddev->raid_disks = 0;
2219 mddev->major_version = 0;
2222 static void sync_sbs(struct mddev *mddev, int nospares)
2224 /* Update each superblock (in-memory image), but
2225 * if we are allowed to, skip spares which already
2226 * have the right event counter, or have one earlier
2227 * (which would mean they aren't being marked as dirty
2228 * with the rest of the array)
2230 struct md_rdev *rdev;
2231 rdev_for_each(rdev, mddev) {
2232 if (rdev->sb_events == mddev->events ||
2234 rdev->raid_disk < 0 &&
2235 rdev->sb_events+1 == mddev->events)) {
2236 /* Don't update this superblock */
2237 rdev->sb_loaded = 2;
2239 sync_super(mddev, rdev);
2240 rdev->sb_loaded = 1;
2245 static bool does_sb_need_changing(struct mddev *mddev)
2247 struct md_rdev *rdev;
2248 struct mdp_superblock_1 *sb;
2251 /* Find a good rdev */
2252 rdev_for_each(rdev, mddev)
2253 if ((rdev->raid_disk >= 0) && !test_bit(Faulty, &rdev->flags))
2256 /* No good device found. */
2260 sb = page_address(rdev->sb_page);
2261 /* Check if a device has become faulty or a spare become active */
2262 rdev_for_each(rdev, mddev) {
2263 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
2264 /* Device activated? */
2265 if (role == 0xffff && rdev->raid_disk >=0 &&
2266 !test_bit(Faulty, &rdev->flags))
2268 /* Device turned faulty? */
2269 if (test_bit(Faulty, &rdev->flags) && (role < 0xfffd))
2273 /* Check if any mddev parameters have changed */
2274 if ((mddev->dev_sectors != le64_to_cpu(sb->size)) ||
2275 (mddev->reshape_position != le64_to_cpu(sb->reshape_position)) ||
2276 (mddev->layout != le64_to_cpu(sb->layout)) ||
2277 (mddev->raid_disks != le32_to_cpu(sb->raid_disks)) ||
2278 (mddev->chunk_sectors != le32_to_cpu(sb->chunksize)))
2284 void md_update_sb(struct mddev *mddev, int force_change)
2286 struct md_rdev *rdev;
2289 int any_badblocks_changed = 0;
2294 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2298 if (mddev_is_clustered(mddev)) {
2299 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2301 ret = md_cluster_ops->metadata_update_start(mddev);
2302 /* Has someone else has updated the sb */
2303 if (!does_sb_need_changing(mddev)) {
2305 md_cluster_ops->metadata_update_cancel(mddev);
2306 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2311 /* First make sure individual recovery_offsets are correct */
2312 rdev_for_each(rdev, mddev) {
2313 if (rdev->raid_disk >= 0 &&
2314 mddev->delta_disks >= 0 &&
2315 !test_bit(Journal, &rdev->flags) &&
2316 !test_bit(In_sync, &rdev->flags) &&
2317 mddev->curr_resync_completed > rdev->recovery_offset)
2318 rdev->recovery_offset = mddev->curr_resync_completed;
2321 if (!mddev->persistent) {
2322 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2323 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2324 if (!mddev->external) {
2325 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2326 rdev_for_each(rdev, mddev) {
2327 if (rdev->badblocks.changed) {
2328 rdev->badblocks.changed = 0;
2329 md_ack_all_badblocks(&rdev->badblocks);
2330 md_error(mddev, rdev);
2332 clear_bit(Blocked, &rdev->flags);
2333 clear_bit(BlockedBadBlocks, &rdev->flags);
2334 wake_up(&rdev->blocked_wait);
2337 wake_up(&mddev->sb_wait);
2341 spin_lock(&mddev->lock);
2343 mddev->utime = get_seconds();
2345 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2347 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2348 /* just a clean<-> dirty transition, possibly leave spares alone,
2349 * though if events isn't the right even/odd, we will have to do
2355 if (mddev->degraded)
2356 /* If the array is degraded, then skipping spares is both
2357 * dangerous and fairly pointless.
2358 * Dangerous because a device that was removed from the array
2359 * might have a event_count that still looks up-to-date,
2360 * so it can be re-added without a resync.
2361 * Pointless because if there are any spares to skip,
2362 * then a recovery will happen and soon that array won't
2363 * be degraded any more and the spare can go back to sleep then.
2367 sync_req = mddev->in_sync;
2369 /* If this is just a dirty<->clean transition, and the array is clean
2370 * and 'events' is odd, we can roll back to the previous clean state */
2372 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2373 && mddev->can_decrease_events
2374 && mddev->events != 1) {
2376 mddev->can_decrease_events = 0;
2378 /* otherwise we have to go forward and ... */
2380 mddev->can_decrease_events = nospares;
2384 * This 64-bit counter should never wrap.
2385 * Either we are in around ~1 trillion A.C., assuming
2386 * 1 reboot per second, or we have a bug...
2388 WARN_ON(mddev->events == 0);
2390 rdev_for_each(rdev, mddev) {
2391 if (rdev->badblocks.changed)
2392 any_badblocks_changed++;
2393 if (test_bit(Faulty, &rdev->flags))
2394 set_bit(FaultRecorded, &rdev->flags);
2397 sync_sbs(mddev, nospares);
2398 spin_unlock(&mddev->lock);
2400 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2401 mdname(mddev), mddev->in_sync);
2403 bitmap_update_sb(mddev->bitmap);
2404 rdev_for_each(rdev, mddev) {
2405 char b[BDEVNAME_SIZE];
2407 if (rdev->sb_loaded != 1)
2408 continue; /* no noise on spare devices */
2410 if (!test_bit(Faulty, &rdev->flags)) {
2411 md_super_write(mddev,rdev,
2412 rdev->sb_start, rdev->sb_size,
2414 pr_debug("md: (write) %s's sb offset: %llu\n",
2415 bdevname(rdev->bdev, b),
2416 (unsigned long long)rdev->sb_start);
2417 rdev->sb_events = mddev->events;
2418 if (rdev->badblocks.size) {
2419 md_super_write(mddev, rdev,
2420 rdev->badblocks.sector,
2421 rdev->badblocks.size << 9,
2423 rdev->badblocks.size = 0;
2427 pr_debug("md: %s (skipping faulty)\n",
2428 bdevname(rdev->bdev, b));
2430 if (mddev->level == LEVEL_MULTIPATH)
2431 /* only need to write one superblock... */
2434 md_super_wait(mddev);
2435 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2437 spin_lock(&mddev->lock);
2438 if (mddev->in_sync != sync_req ||
2439 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2440 /* have to write it out again */
2441 spin_unlock(&mddev->lock);
2444 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2445 spin_unlock(&mddev->lock);
2446 wake_up(&mddev->sb_wait);
2447 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2448 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2450 rdev_for_each(rdev, mddev) {
2451 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2452 clear_bit(Blocked, &rdev->flags);
2454 if (any_badblocks_changed)
2455 md_ack_all_badblocks(&rdev->badblocks);
2456 clear_bit(BlockedBadBlocks, &rdev->flags);
2457 wake_up(&rdev->blocked_wait);
2460 if (mddev_is_clustered(mddev) && ret == 0)
2461 md_cluster_ops->metadata_update_finish(mddev);
2463 EXPORT_SYMBOL(md_update_sb);
2465 static int add_bound_rdev(struct md_rdev *rdev)
2467 struct mddev *mddev = rdev->mddev;
2470 if (!mddev->pers->hot_remove_disk) {
2471 /* If there is hot_add_disk but no hot_remove_disk
2472 * then added disks for geometry changes,
2473 * and should be added immediately.
2475 super_types[mddev->major_version].
2476 validate_super(mddev, rdev);
2477 err = mddev->pers->hot_add_disk(mddev, rdev);
2479 unbind_rdev_from_array(rdev);
2484 sysfs_notify_dirent_safe(rdev->sysfs_state);
2486 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2487 if (mddev->degraded)
2488 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2489 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2490 md_new_event(mddev);
2491 md_wakeup_thread(mddev->thread);
2495 /* words written to sysfs files may, or may not, be \n terminated.
2496 * We want to accept with case. For this we use cmd_match.
2498 static int cmd_match(const char *cmd, const char *str)
2500 /* See if cmd, written into a sysfs file, matches
2501 * str. They must either be the same, or cmd can
2502 * have a trailing newline
2504 while (*cmd && *str && *cmd == *str) {
2515 struct rdev_sysfs_entry {
2516 struct attribute attr;
2517 ssize_t (*show)(struct md_rdev *, char *);
2518 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2522 state_show(struct md_rdev *rdev, char *page)
2526 unsigned long flags = ACCESS_ONCE(rdev->flags);
2528 if (test_bit(Faulty, &flags) ||
2529 rdev->badblocks.unacked_exist) {
2530 len+= sprintf(page+len, "%sfaulty",sep);
2533 if (test_bit(In_sync, &flags)) {
2534 len += sprintf(page+len, "%sin_sync",sep);
2537 if (test_bit(Journal, &flags)) {
2538 len += sprintf(page+len, "%sjournal",sep);
2541 if (test_bit(WriteMostly, &flags)) {
2542 len += sprintf(page+len, "%swrite_mostly",sep);
2545 if (test_bit(Blocked, &flags) ||
2546 (rdev->badblocks.unacked_exist
2547 && !test_bit(Faulty, &flags))) {
2548 len += sprintf(page+len, "%sblocked", sep);
2551 if (!test_bit(Faulty, &flags) &&
2552 !test_bit(Journal, &flags) &&
2553 !test_bit(In_sync, &flags)) {
2554 len += sprintf(page+len, "%sspare", sep);
2557 if (test_bit(WriteErrorSeen, &flags)) {
2558 len += sprintf(page+len, "%swrite_error", sep);
2561 if (test_bit(WantReplacement, &flags)) {
2562 len += sprintf(page+len, "%swant_replacement", sep);
2565 if (test_bit(Replacement, &flags)) {
2566 len += sprintf(page+len, "%sreplacement", sep);
2570 return len+sprintf(page+len, "\n");
2574 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2577 * faulty - simulates an error
2578 * remove - disconnects the device
2579 * writemostly - sets write_mostly
2580 * -writemostly - clears write_mostly
2581 * blocked - sets the Blocked flags
2582 * -blocked - clears the Blocked and possibly simulates an error
2583 * insync - sets Insync providing device isn't active
2584 * -insync - clear Insync for a device with a slot assigned,
2585 * so that it gets rebuilt based on bitmap
2586 * write_error - sets WriteErrorSeen
2587 * -write_error - clears WriteErrorSeen
2590 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2591 md_error(rdev->mddev, rdev);
2592 if (test_bit(Faulty, &rdev->flags))
2596 } else if (cmd_match(buf, "remove")) {
2597 if (rdev->raid_disk >= 0)
2600 struct mddev *mddev = rdev->mddev;
2602 if (mddev_is_clustered(mddev))
2603 err = md_cluster_ops->remove_disk(mddev, rdev);
2606 md_kick_rdev_from_array(rdev);
2608 md_update_sb(mddev, 1);
2609 md_new_event(mddev);
2612 } else if (cmd_match(buf, "writemostly")) {
2613 set_bit(WriteMostly, &rdev->flags);
2615 } else if (cmd_match(buf, "-writemostly")) {
2616 clear_bit(WriteMostly, &rdev->flags);
2618 } else if (cmd_match(buf, "blocked")) {
2619 set_bit(Blocked, &rdev->flags);
2621 } else if (cmd_match(buf, "-blocked")) {
2622 if (!test_bit(Faulty, &rdev->flags) &&
2623 rdev->badblocks.unacked_exist) {
2624 /* metadata handler doesn't understand badblocks,
2625 * so we need to fail the device
2627 md_error(rdev->mddev, rdev);
2629 clear_bit(Blocked, &rdev->flags);
2630 clear_bit(BlockedBadBlocks, &rdev->flags);
2631 wake_up(&rdev->blocked_wait);
2632 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2633 md_wakeup_thread(rdev->mddev->thread);
2636 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2637 set_bit(In_sync, &rdev->flags);
2639 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0 &&
2640 !test_bit(Journal, &rdev->flags)) {
2641 if (rdev->mddev->pers == NULL) {
2642 clear_bit(In_sync, &rdev->flags);
2643 rdev->saved_raid_disk = rdev->raid_disk;
2644 rdev->raid_disk = -1;
2647 } else if (cmd_match(buf, "write_error")) {
2648 set_bit(WriteErrorSeen, &rdev->flags);
2650 } else if (cmd_match(buf, "-write_error")) {
2651 clear_bit(WriteErrorSeen, &rdev->flags);
2653 } else if (cmd_match(buf, "want_replacement")) {
2654 /* Any non-spare device that is not a replacement can
2655 * become want_replacement at any time, but we then need to
2656 * check if recovery is needed.
2658 if (rdev->raid_disk >= 0 &&
2659 !test_bit(Journal, &rdev->flags) &&
2660 !test_bit(Replacement, &rdev->flags))
2661 set_bit(WantReplacement, &rdev->flags);
2662 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2663 md_wakeup_thread(rdev->mddev->thread);
2665 } else if (cmd_match(buf, "-want_replacement")) {
2666 /* Clearing 'want_replacement' is always allowed.
2667 * Once replacements starts it is too late though.
2670 clear_bit(WantReplacement, &rdev->flags);
2671 } else if (cmd_match(buf, "replacement")) {
2672 /* Can only set a device as a replacement when array has not
2673 * yet been started. Once running, replacement is automatic
2674 * from spares, or by assigning 'slot'.
2676 if (rdev->mddev->pers)
2679 set_bit(Replacement, &rdev->flags);
2682 } else if (cmd_match(buf, "-replacement")) {
2683 /* Similarly, can only clear Replacement before start */
2684 if (rdev->mddev->pers)
2687 clear_bit(Replacement, &rdev->flags);
2690 } else if (cmd_match(buf, "re-add")) {
2691 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2692 /* clear_bit is performed _after_ all the devices
2693 * have their local Faulty bit cleared. If any writes
2694 * happen in the meantime in the local node, they
2695 * will land in the local bitmap, which will be synced
2696 * by this node eventually
2698 if (!mddev_is_clustered(rdev->mddev) ||
2699 (err = md_cluster_ops->gather_bitmaps(rdev)) == 0) {
2700 clear_bit(Faulty, &rdev->flags);
2701 err = add_bound_rdev(rdev);
2707 sysfs_notify_dirent_safe(rdev->sysfs_state);
2708 return err ? err : len;
2710 static struct rdev_sysfs_entry rdev_state =
2711 __ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
2714 errors_show(struct md_rdev *rdev, char *page)
2716 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2720 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2725 rv = kstrtouint(buf, 10, &n);
2728 atomic_set(&rdev->corrected_errors, n);
2731 static struct rdev_sysfs_entry rdev_errors =
2732 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2735 slot_show(struct md_rdev *rdev, char *page)
2737 if (test_bit(Journal, &rdev->flags))
2738 return sprintf(page, "journal\n");
2739 else if (rdev->raid_disk < 0)
2740 return sprintf(page, "none\n");
2742 return sprintf(page, "%d\n", rdev->raid_disk);
2746 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2751 if (test_bit(Journal, &rdev->flags))
2753 if (strncmp(buf, "none", 4)==0)
2756 err = kstrtouint(buf, 10, (unsigned int *)&slot);
2760 if (rdev->mddev->pers && slot == -1) {
2761 /* Setting 'slot' on an active array requires also
2762 * updating the 'rd%d' link, and communicating
2763 * with the personality with ->hot_*_disk.
2764 * For now we only support removing
2765 * failed/spare devices. This normally happens automatically,
2766 * but not when the metadata is externally managed.
2768 if (rdev->raid_disk == -1)
2770 /* personality does all needed checks */
2771 if (rdev->mddev->pers->hot_remove_disk == NULL)
2773 clear_bit(Blocked, &rdev->flags);
2774 remove_and_add_spares(rdev->mddev, rdev);
2775 if (rdev->raid_disk >= 0)
2777 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2778 md_wakeup_thread(rdev->mddev->thread);
2779 } else if (rdev->mddev->pers) {
2780 /* Activating a spare .. or possibly reactivating
2781 * if we ever get bitmaps working here.
2785 if (rdev->raid_disk != -1)
2788 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2791 if (rdev->mddev->pers->hot_add_disk == NULL)
2794 if (slot >= rdev->mddev->raid_disks &&
2795 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2798 rdev->raid_disk = slot;
2799 if (test_bit(In_sync, &rdev->flags))
2800 rdev->saved_raid_disk = slot;
2802 rdev->saved_raid_disk = -1;
2803 clear_bit(In_sync, &rdev->flags);
2804 clear_bit(Bitmap_sync, &rdev->flags);
2805 err = rdev->mddev->pers->
2806 hot_add_disk(rdev->mddev, rdev);
2808 rdev->raid_disk = -1;
2811 sysfs_notify_dirent_safe(rdev->sysfs_state);
2812 if (sysfs_link_rdev(rdev->mddev, rdev))
2813 /* failure here is OK */;
2814 /* don't wakeup anyone, leave that to userspace. */
2816 if (slot >= rdev->mddev->raid_disks &&
2817 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2819 rdev->raid_disk = slot;
2820 /* assume it is working */
2821 clear_bit(Faulty, &rdev->flags);
2822 clear_bit(WriteMostly, &rdev->flags);
2823 set_bit(In_sync, &rdev->flags);
2824 sysfs_notify_dirent_safe(rdev->sysfs_state);
2829 static struct rdev_sysfs_entry rdev_slot =
2830 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2833 offset_show(struct md_rdev *rdev, char *page)
2835 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2839 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2841 unsigned long long offset;
2842 if (kstrtoull(buf, 10, &offset) < 0)
2844 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2846 if (rdev->sectors && rdev->mddev->external)
2847 /* Must set offset before size, so overlap checks
2850 rdev->data_offset = offset;
2851 rdev->new_data_offset = offset;
2855 static struct rdev_sysfs_entry rdev_offset =
2856 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2858 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2860 return sprintf(page, "%llu\n",
2861 (unsigned long long)rdev->new_data_offset);
2864 static ssize_t new_offset_store(struct md_rdev *rdev,
2865 const char *buf, size_t len)
2867 unsigned long long new_offset;
2868 struct mddev *mddev = rdev->mddev;
2870 if (kstrtoull(buf, 10, &new_offset) < 0)
2873 if (mddev->sync_thread ||
2874 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2876 if (new_offset == rdev->data_offset)
2877 /* reset is always permitted */
2879 else if (new_offset > rdev->data_offset) {
2880 /* must not push array size beyond rdev_sectors */
2881 if (new_offset - rdev->data_offset
2882 + mddev->dev_sectors > rdev->sectors)
2885 /* Metadata worries about other space details. */
2887 /* decreasing the offset is inconsistent with a backwards
2890 if (new_offset < rdev->data_offset &&
2891 mddev->reshape_backwards)
2893 /* Increasing offset is inconsistent with forwards
2894 * reshape. reshape_direction should be set to
2895 * 'backwards' first.
2897 if (new_offset > rdev->data_offset &&
2898 !mddev->reshape_backwards)
2901 if (mddev->pers && mddev->persistent &&
2902 !super_types[mddev->major_version]
2903 .allow_new_offset(rdev, new_offset))
2905 rdev->new_data_offset = new_offset;
2906 if (new_offset > rdev->data_offset)
2907 mddev->reshape_backwards = 1;
2908 else if (new_offset < rdev->data_offset)
2909 mddev->reshape_backwards = 0;
2913 static struct rdev_sysfs_entry rdev_new_offset =
2914 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2917 rdev_size_show(struct md_rdev *rdev, char *page)
2919 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2922 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2924 /* check if two start/length pairs overlap */
2932 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2934 unsigned long long blocks;
2937 if (kstrtoull(buf, 10, &blocks) < 0)
2940 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2941 return -EINVAL; /* sector conversion overflow */
2944 if (new != blocks * 2)
2945 return -EINVAL; /* unsigned long long to sector_t overflow */
2952 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2954 struct mddev *my_mddev = rdev->mddev;
2955 sector_t oldsectors = rdev->sectors;
2958 if (test_bit(Journal, &rdev->flags))
2960 if (strict_blocks_to_sectors(buf, §ors) < 0)
2962 if (rdev->data_offset != rdev->new_data_offset)
2963 return -EINVAL; /* too confusing */
2964 if (my_mddev->pers && rdev->raid_disk >= 0) {
2965 if (my_mddev->persistent) {
2966 sectors = super_types[my_mddev->major_version].
2967 rdev_size_change(rdev, sectors);
2970 } else if (!sectors)
2971 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2973 if (!my_mddev->pers->resize)
2974 /* Cannot change size for RAID0 or Linear etc */
2977 if (sectors < my_mddev->dev_sectors)
2978 return -EINVAL; /* component must fit device */
2980 rdev->sectors = sectors;
2981 if (sectors > oldsectors && my_mddev->external) {
2982 /* Need to check that all other rdevs with the same
2983 * ->bdev do not overlap. 'rcu' is sufficient to walk
2984 * the rdev lists safely.
2985 * This check does not provide a hard guarantee, it
2986 * just helps avoid dangerous mistakes.
2988 struct mddev *mddev;
2990 struct list_head *tmp;
2993 for_each_mddev(mddev, tmp) {
2994 struct md_rdev *rdev2;
2996 rdev_for_each(rdev2, mddev)
2997 if (rdev->bdev == rdev2->bdev &&
2999 overlaps(rdev->data_offset, rdev->sectors,
3012 /* Someone else could have slipped in a size
3013 * change here, but doing so is just silly.
3014 * We put oldsectors back because we *know* it is
3015 * safe, and trust userspace not to race with
3018 rdev->sectors = oldsectors;
3025 static struct rdev_sysfs_entry rdev_size =
3026 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
3028 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
3030 unsigned long long recovery_start = rdev->recovery_offset;
3032 if (test_bit(In_sync, &rdev->flags) ||
3033 recovery_start == MaxSector)
3034 return sprintf(page, "none\n");
3036 return sprintf(page, "%llu\n", recovery_start);
3039 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
3041 unsigned long long recovery_start;
3043 if (cmd_match(buf, "none"))
3044 recovery_start = MaxSector;
3045 else if (kstrtoull(buf, 10, &recovery_start))
3048 if (rdev->mddev->pers &&
3049 rdev->raid_disk >= 0)
3052 rdev->recovery_offset = recovery_start;
3053 if (recovery_start == MaxSector)
3054 set_bit(In_sync, &rdev->flags);
3056 clear_bit(In_sync, &rdev->flags);
3060 static struct rdev_sysfs_entry rdev_recovery_start =
3061 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3064 badblocks_show(struct badblocks *bb, char *page, int unack);
3066 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3068 static ssize_t bb_show(struct md_rdev *rdev, char *page)
3070 return badblocks_show(&rdev->badblocks, page, 0);
3072 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
3074 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3075 /* Maybe that ack was all we needed */
3076 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3077 wake_up(&rdev->blocked_wait);
3080 static struct rdev_sysfs_entry rdev_bad_blocks =
3081 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3083 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
3085 return badblocks_show(&rdev->badblocks, page, 1);
3087 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
3089 return badblocks_store(&rdev->badblocks, page, len, 1);
3091 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3092 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3094 static struct attribute *rdev_default_attrs[] = {
3099 &rdev_new_offset.attr,
3101 &rdev_recovery_start.attr,
3102 &rdev_bad_blocks.attr,
3103 &rdev_unack_bad_blocks.attr,
3107 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3109 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3110 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3116 return entry->show(rdev, page);
3120 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3121 const char *page, size_t length)
3123 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3124 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3126 struct mddev *mddev = rdev->mddev;
3130 if (!capable(CAP_SYS_ADMIN))
3132 rv = mddev ? mddev_lock(mddev): -EBUSY;
3134 if (rdev->mddev == NULL)
3137 rv = entry->store(rdev, page, length);
3138 mddev_unlock(mddev);
3143 static void rdev_free(struct kobject *ko)
3145 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3148 static const struct sysfs_ops rdev_sysfs_ops = {
3149 .show = rdev_attr_show,
3150 .store = rdev_attr_store,
3152 static struct kobj_type rdev_ktype = {
3153 .release = rdev_free,
3154 .sysfs_ops = &rdev_sysfs_ops,
3155 .default_attrs = rdev_default_attrs,
3158 int md_rdev_init(struct md_rdev *rdev)
3161 rdev->saved_raid_disk = -1;
3162 rdev->raid_disk = -1;
3164 rdev->data_offset = 0;
3165 rdev->new_data_offset = 0;
3166 rdev->sb_events = 0;
3167 rdev->last_read_error.tv_sec = 0;
3168 rdev->last_read_error.tv_nsec = 0;
3169 rdev->sb_loaded = 0;
3170 rdev->bb_page = NULL;
3171 atomic_set(&rdev->nr_pending, 0);
3172 atomic_set(&rdev->read_errors, 0);
3173 atomic_set(&rdev->corrected_errors, 0);
3175 INIT_LIST_HEAD(&rdev->same_set);
3176 init_waitqueue_head(&rdev->blocked_wait);
3178 /* Add space to store bad block list.
3179 * This reserves the space even on arrays where it cannot
3180 * be used - I wonder if that matters
3182 rdev->badblocks.count = 0;
3183 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3184 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3185 seqlock_init(&rdev->badblocks.lock);
3186 if (rdev->badblocks.page == NULL)
3191 EXPORT_SYMBOL_GPL(md_rdev_init);
3193 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3195 * mark the device faulty if:
3197 * - the device is nonexistent (zero size)
3198 * - the device has no valid superblock
3200 * a faulty rdev _never_ has rdev->sb set.
3202 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3204 char b[BDEVNAME_SIZE];
3206 struct md_rdev *rdev;
3209 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3211 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3212 return ERR_PTR(-ENOMEM);
3215 err = md_rdev_init(rdev);
3218 err = alloc_disk_sb(rdev);
3222 err = lock_rdev(rdev, newdev, super_format == -2);
3226 kobject_init(&rdev->kobj, &rdev_ktype);
3228 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3231 "md: %s has zero or unknown size, marking faulty!\n",
3232 bdevname(rdev->bdev,b));
3237 if (super_format >= 0) {
3238 err = super_types[super_format].
3239 load_super(rdev, NULL, super_minor);
3240 if (err == -EINVAL) {
3242 "md: %s does not have a valid v%d.%d "
3243 "superblock, not importing!\n",
3244 bdevname(rdev->bdev,b),
3245 super_format, super_minor);
3250 "md: could not read %s's sb, not importing!\n",
3251 bdevname(rdev->bdev,b));
3261 md_rdev_clear(rdev);
3263 return ERR_PTR(err);
3267 * Check a full RAID array for plausibility
3270 static void analyze_sbs(struct mddev *mddev)
3273 struct md_rdev *rdev, *freshest, *tmp;
3274 char b[BDEVNAME_SIZE];
3277 rdev_for_each_safe(rdev, tmp, mddev)
3278 switch (super_types[mddev->major_version].
3279 load_super(rdev, freshest, mddev->minor_version)) {
3287 "md: fatal superblock inconsistency in %s"
3288 " -- removing from array\n",
3289 bdevname(rdev->bdev,b));
3290 md_kick_rdev_from_array(rdev);
3293 super_types[mddev->major_version].
3294 validate_super(mddev, freshest);
3297 rdev_for_each_safe(rdev, tmp, mddev) {
3298 if (mddev->max_disks &&
3299 (rdev->desc_nr >= mddev->max_disks ||
3300 i > mddev->max_disks)) {
3302 "md: %s: %s: only %d devices permitted\n",
3303 mdname(mddev), bdevname(rdev->bdev, b),
3305 md_kick_rdev_from_array(rdev);
3308 if (rdev != freshest) {
3309 if (super_types[mddev->major_version].
3310 validate_super(mddev, rdev)) {
3311 printk(KERN_WARNING "md: kicking non-fresh %s"
3313 bdevname(rdev->bdev,b));
3314 md_kick_rdev_from_array(rdev);
3318 if (mddev->level == LEVEL_MULTIPATH) {
3319 rdev->desc_nr = i++;
3320 rdev->raid_disk = rdev->desc_nr;
3321 set_bit(In_sync, &rdev->flags);
3322 } else if (rdev->raid_disk >=
3323 (mddev->raid_disks - min(0, mddev->delta_disks)) &&
3324 !test_bit(Journal, &rdev->flags)) {
3325 rdev->raid_disk = -1;
3326 clear_bit(In_sync, &rdev->flags);
3331 /* Read a fixed-point number.
3332 * Numbers in sysfs attributes should be in "standard" units where
3333 * possible, so time should be in seconds.
3334 * However we internally use a a much smaller unit such as
3335 * milliseconds or jiffies.
3336 * This function takes a decimal number with a possible fractional
3337 * component, and produces an integer which is the result of
3338 * multiplying that number by 10^'scale'.
3339 * all without any floating-point arithmetic.
3341 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3343 unsigned long result = 0;
3345 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3348 else if (decimals < scale) {
3351 result = result * 10 + value;
3363 while (decimals < scale) {
3372 safe_delay_show(struct mddev *mddev, char *page)
3374 int msec = (mddev->safemode_delay*1000)/HZ;
3375 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3378 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3382 if (mddev_is_clustered(mddev)) {
3383 pr_info("md: Safemode is disabled for clustered mode\n");
3387 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3390 mddev->safemode_delay = 0;
3392 unsigned long old_delay = mddev->safemode_delay;
3393 unsigned long new_delay = (msec*HZ)/1000;
3397 mddev->safemode_delay = new_delay;
3398 if (new_delay < old_delay || old_delay == 0)
3399 mod_timer(&mddev->safemode_timer, jiffies+1);
3403 static struct md_sysfs_entry md_safe_delay =
3404 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3407 level_show(struct mddev *mddev, char *page)
3409 struct md_personality *p;
3411 spin_lock(&mddev->lock);
3414 ret = sprintf(page, "%s\n", p->name);
3415 else if (mddev->clevel[0])
3416 ret = sprintf(page, "%s\n", mddev->clevel);
3417 else if (mddev->level != LEVEL_NONE)
3418 ret = sprintf(page, "%d\n", mddev->level);
3421 spin_unlock(&mddev->lock);
3426 level_store(struct mddev *mddev, const char *buf, size_t len)
3431 struct md_personality *pers, *oldpers;
3433 void *priv, *oldpriv;
3434 struct md_rdev *rdev;
3436 if (slen == 0 || slen >= sizeof(clevel))
3439 rv = mddev_lock(mddev);
3443 if (mddev->pers == NULL) {
3444 strncpy(mddev->clevel, buf, slen);
3445 if (mddev->clevel[slen-1] == '\n')
3447 mddev->clevel[slen] = 0;
3448 mddev->level = LEVEL_NONE;
3456 /* request to change the personality. Need to ensure:
3457 * - array is not engaged in resync/recovery/reshape
3458 * - old personality can be suspended
3459 * - new personality will access other array.
3463 if (mddev->sync_thread ||
3464 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3465 mddev->reshape_position != MaxSector ||
3466 mddev->sysfs_active)
3470 if (!mddev->pers->quiesce) {
3471 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3472 mdname(mddev), mddev->pers->name);
3476 /* Now find the new personality */
3477 strncpy(clevel, buf, slen);
3478 if (clevel[slen-1] == '\n')
3481 if (kstrtol(clevel, 10, &level))
3484 if (request_module("md-%s", clevel) != 0)
3485 request_module("md-level-%s", clevel);
3486 spin_lock(&pers_lock);
3487 pers = find_pers(level, clevel);
3488 if (!pers || !try_module_get(pers->owner)) {
3489 spin_unlock(&pers_lock);
3490 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3494 spin_unlock(&pers_lock);
3496 if (pers == mddev->pers) {
3497 /* Nothing to do! */
3498 module_put(pers->owner);
3502 if (!pers->takeover) {
3503 module_put(pers->owner);
3504 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3505 mdname(mddev), clevel);
3510 rdev_for_each(rdev, mddev)
3511 rdev->new_raid_disk = rdev->raid_disk;
3513 /* ->takeover must set new_* and/or delta_disks
3514 * if it succeeds, and may set them when it fails.
3516 priv = pers->takeover(mddev);
3518 mddev->new_level = mddev->level;
3519 mddev->new_layout = mddev->layout;
3520 mddev->new_chunk_sectors = mddev->chunk_sectors;
3521 mddev->raid_disks -= mddev->delta_disks;
3522 mddev->delta_disks = 0;
3523 mddev->reshape_backwards = 0;
3524 module_put(pers->owner);
3525 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3526 mdname(mddev), clevel);
3531 /* Looks like we have a winner */
3532 mddev_suspend(mddev);
3533 mddev_detach(mddev);
3535 spin_lock(&mddev->lock);
3536 oldpers = mddev->pers;
3537 oldpriv = mddev->private;
3539 mddev->private = priv;
3540 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3541 mddev->level = mddev->new_level;
3542 mddev->layout = mddev->new_layout;
3543 mddev->chunk_sectors = mddev->new_chunk_sectors;
3544 mddev->delta_disks = 0;
3545 mddev->reshape_backwards = 0;
3546 mddev->degraded = 0;
3547 spin_unlock(&mddev->lock);
3549 if (oldpers->sync_request == NULL &&
3551 /* We are converting from a no-redundancy array
3552 * to a redundancy array and metadata is managed
3553 * externally so we need to be sure that writes
3554 * won't block due to a need to transition
3556 * until external management is started.
3559 mddev->safemode_delay = 0;
3560 mddev->safemode = 0;
3563 oldpers->free(mddev, oldpriv);
3565 if (oldpers->sync_request == NULL &&
3566 pers->sync_request != NULL) {
3567 /* need to add the md_redundancy_group */
3568 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3570 "md: cannot register extra attributes for %s\n",
3572 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3574 if (oldpers->sync_request != NULL &&
3575 pers->sync_request == NULL) {
3576 /* need to remove the md_redundancy_group */
3577 if (mddev->to_remove == NULL)
3578 mddev->to_remove = &md_redundancy_group;
3581 rdev_for_each(rdev, mddev) {
3582 if (rdev->raid_disk < 0)
3584 if (rdev->new_raid_disk >= mddev->raid_disks)
3585 rdev->new_raid_disk = -1;
3586 if (rdev->new_raid_disk == rdev->raid_disk)
3588 sysfs_unlink_rdev(mddev, rdev);
3590 rdev_for_each(rdev, mddev) {
3591 if (rdev->raid_disk < 0)
3593 if (rdev->new_raid_disk == rdev->raid_disk)
3595 rdev->raid_disk = rdev->new_raid_disk;
3596 if (rdev->raid_disk < 0)
3597 clear_bit(In_sync, &rdev->flags);
3599 if (sysfs_link_rdev(mddev, rdev))
3600 printk(KERN_WARNING "md: cannot register rd%d"
3601 " for %s after level change\n",
3602 rdev->raid_disk, mdname(mddev));
3606 if (pers->sync_request == NULL) {
3607 /* this is now an array without redundancy, so
3608 * it must always be in_sync
3611 del_timer_sync(&mddev->safemode_timer);
3613 blk_set_stacking_limits(&mddev->queue->limits);
3615 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3616 mddev_resume(mddev);
3618 md_update_sb(mddev, 1);
3619 sysfs_notify(&mddev->kobj, NULL, "level");
3620 md_new_event(mddev);
3623 mddev_unlock(mddev);
3627 static struct md_sysfs_entry md_level =
3628 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3631 layout_show(struct mddev *mddev, char *page)
3633 /* just a number, not meaningful for all levels */
3634 if (mddev->reshape_position != MaxSector &&
3635 mddev->layout != mddev->new_layout)
3636 return sprintf(page, "%d (%d)\n",
3637 mddev->new_layout, mddev->layout);
3638 return sprintf(page, "%d\n", mddev->layout);
3642 layout_store(struct mddev *mddev, const char *buf, size_t len)
3647 err = kstrtouint(buf, 10, &n);
3650 err = mddev_lock(mddev);
3655 if (mddev->pers->check_reshape == NULL)
3660 mddev->new_layout = n;
3661 err = mddev->pers->check_reshape(mddev);
3663 mddev->new_layout = mddev->layout;
3666 mddev->new_layout = n;
3667 if (mddev->reshape_position == MaxSector)
3670 mddev_unlock(mddev);
3673 static struct md_sysfs_entry md_layout =
3674 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3677 raid_disks_show(struct mddev *mddev, char *page)
3679 if (mddev->raid_disks == 0)
3681 if (mddev->reshape_position != MaxSector &&
3682 mddev->delta_disks != 0)
3683 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3684 mddev->raid_disks - mddev->delta_disks);
3685 return sprintf(page, "%d\n", mddev->raid_disks);
3688 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3691 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3696 err = kstrtouint(buf, 10, &n);
3700 err = mddev_lock(mddev);
3704 err = update_raid_disks(mddev, n);
3705 else if (mddev->reshape_position != MaxSector) {
3706 struct md_rdev *rdev;
3707 int olddisks = mddev->raid_disks - mddev->delta_disks;
3710 rdev_for_each(rdev, mddev) {
3712 rdev->data_offset < rdev->new_data_offset)
3715 rdev->data_offset > rdev->new_data_offset)
3719 mddev->delta_disks = n - olddisks;
3720 mddev->raid_disks = n;
3721 mddev->reshape_backwards = (mddev->delta_disks < 0);
3723 mddev->raid_disks = n;
3725 mddev_unlock(mddev);
3726 return err ? err : len;
3728 static struct md_sysfs_entry md_raid_disks =
3729 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3732 chunk_size_show(struct mddev *mddev, char *page)
3734 if (mddev->reshape_position != MaxSector &&
3735 mddev->chunk_sectors != mddev->new_chunk_sectors)
3736 return sprintf(page, "%d (%d)\n",
3737 mddev->new_chunk_sectors << 9,
3738 mddev->chunk_sectors << 9);
3739 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3743 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3748 err = kstrtoul(buf, 10, &n);
3752 err = mddev_lock(mddev);
3756 if (mddev->pers->check_reshape == NULL)
3761 mddev->new_chunk_sectors = n >> 9;
3762 err = mddev->pers->check_reshape(mddev);
3764 mddev->new_chunk_sectors = mddev->chunk_sectors;
3767 mddev->new_chunk_sectors = n >> 9;
3768 if (mddev->reshape_position == MaxSector)
3769 mddev->chunk_sectors = n >> 9;
3771 mddev_unlock(mddev);
3774 static struct md_sysfs_entry md_chunk_size =
3775 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3778 resync_start_show(struct mddev *mddev, char *page)
3780 if (mddev->recovery_cp == MaxSector)
3781 return sprintf(page, "none\n");
3782 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3786 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3788 unsigned long long n;
3791 if (cmd_match(buf, "none"))
3794 err = kstrtoull(buf, 10, &n);
3797 if (n != (sector_t)n)
3801 err = mddev_lock(mddev);
3804 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3808 mddev->recovery_cp = n;
3810 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3812 mddev_unlock(mddev);
3815 static struct md_sysfs_entry md_resync_start =
3816 __ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3817 resync_start_show, resync_start_store);
3820 * The array state can be:
3823 * No devices, no size, no level
3824 * Equivalent to STOP_ARRAY ioctl
3826 * May have some settings, but array is not active
3827 * all IO results in error
3828 * When written, doesn't tear down array, but just stops it
3829 * suspended (not supported yet)
3830 * All IO requests will block. The array can be reconfigured.
3831 * Writing this, if accepted, will block until array is quiescent
3833 * no resync can happen. no superblocks get written.
3834 * write requests fail
3836 * like readonly, but behaves like 'clean' on a write request.
3838 * clean - no pending writes, but otherwise active.
3839 * When written to inactive array, starts without resync
3840 * If a write request arrives then
3841 * if metadata is known, mark 'dirty' and switch to 'active'.
3842 * if not known, block and switch to write-pending
3843 * If written to an active array that has pending writes, then fails.
3845 * fully active: IO and resync can be happening.
3846 * When written to inactive array, starts with resync
3849 * clean, but writes are blocked waiting for 'active' to be written.
3852 * like active, but no writes have been seen for a while (100msec).
3855 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3856 write_pending, active_idle, bad_word};
3857 static char *array_states[] = {
3858 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3859 "write-pending", "active-idle", NULL };
3861 static int match_word(const char *word, char **list)
3864 for (n=0; list[n]; n++)
3865 if (cmd_match(word, list[n]))
3871 array_state_show(struct mddev *mddev, char *page)
3873 enum array_state st = inactive;
3886 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3888 else if (mddev->safemode)
3894 if (list_empty(&mddev->disks) &&
3895 mddev->raid_disks == 0 &&
3896 mddev->dev_sectors == 0)
3901 return sprintf(page, "%s\n", array_states[st]);
3904 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3905 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3906 static int do_md_run(struct mddev *mddev);
3907 static int restart_array(struct mddev *mddev);
3910 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3913 enum array_state st = match_word(buf, array_states);
3915 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3916 /* don't take reconfig_mutex when toggling between
3919 spin_lock(&mddev->lock);
3921 restart_array(mddev);
3922 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3923 wake_up(&mddev->sb_wait);
3925 } else /* st == clean */ {
3926 restart_array(mddev);
3927 if (atomic_read(&mddev->writes_pending) == 0) {
3928 if (mddev->in_sync == 0) {
3930 if (mddev->safemode == 1)
3931 mddev->safemode = 0;
3932 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3938 spin_unlock(&mddev->lock);
3941 err = mddev_lock(mddev);
3949 /* stopping an active array */
3950 err = do_md_stop(mddev, 0, NULL);
3953 /* stopping an active array */
3955 err = do_md_stop(mddev, 2, NULL);
3957 err = 0; /* already inactive */
3960 break; /* not supported yet */
3963 err = md_set_readonly(mddev, NULL);
3966 set_disk_ro(mddev->gendisk, 1);
3967 err = do_md_run(mddev);
3973 err = md_set_readonly(mddev, NULL);
3974 else if (mddev->ro == 1)
3975 err = restart_array(mddev);
3978 set_disk_ro(mddev->gendisk, 0);
3982 err = do_md_run(mddev);
3987 err = restart_array(mddev);
3990 spin_lock(&mddev->lock);
3991 if (atomic_read(&mddev->writes_pending) == 0) {
3992 if (mddev->in_sync == 0) {
3994 if (mddev->safemode == 1)
3995 mddev->safemode = 0;
3996 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
4001 spin_unlock(&mddev->lock);
4007 err = restart_array(mddev);
4010 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
4011 wake_up(&mddev->sb_wait);
4015 set_disk_ro(mddev->gendisk, 0);
4016 err = do_md_run(mddev);
4021 /* these cannot be set */
4026 if (mddev->hold_active == UNTIL_IOCTL)
4027 mddev->hold_active = 0;
4028 sysfs_notify_dirent_safe(mddev->sysfs_state);
4030 mddev_unlock(mddev);
4033 static struct md_sysfs_entry md_array_state =
4034 __ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
4037 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
4038 return sprintf(page, "%d\n",
4039 atomic_read(&mddev->max_corr_read_errors));
4043 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
4048 rv = kstrtouint(buf, 10, &n);
4051 atomic_set(&mddev->max_corr_read_errors, n);
4055 static struct md_sysfs_entry max_corr_read_errors =
4056 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
4057 max_corrected_read_errors_store);
4060 null_show(struct mddev *mddev, char *page)
4066 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
4068 /* buf must be %d:%d\n? giving major and minor numbers */
4069 /* The new device is added to the array.
4070 * If the array has a persistent superblock, we read the
4071 * superblock to initialise info and check validity.
4072 * Otherwise, only checking done is that in bind_rdev_to_array,
4073 * which mainly checks size.
4076 int major = simple_strtoul(buf, &e, 10);
4079 struct md_rdev *rdev;
4082 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4084 minor = simple_strtoul(e+1, &e, 10);
4085 if (*e && *e != '\n')
4087 dev = MKDEV(major, minor);
4088 if (major != MAJOR(dev) ||
4089 minor != MINOR(dev))
4092 flush_workqueue(md_misc_wq);
4094 err = mddev_lock(mddev);
4097 if (mddev->persistent) {
4098 rdev = md_import_device(dev, mddev->major_version,
4099 mddev->minor_version);
4100 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
4101 struct md_rdev *rdev0
4102 = list_entry(mddev->disks.next,
4103 struct md_rdev, same_set);
4104 err = super_types[mddev->major_version]
4105 .load_super(rdev, rdev0, mddev->minor_version);
4109 } else if (mddev->external)
4110 rdev = md_import_device(dev, -2, -1);
4112 rdev = md_import_device(dev, -1, -1);
4115 mddev_unlock(mddev);
4116 return PTR_ERR(rdev);
4118 err = bind_rdev_to_array(rdev, mddev);
4122 mddev_unlock(mddev);
4123 return err ? err : len;
4126 static struct md_sysfs_entry md_new_device =
4127 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4130 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4133 unsigned long chunk, end_chunk;
4136 err = mddev_lock(mddev);
4141 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4143 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4144 if (buf == end) break;
4145 if (*end == '-') { /* range */
4147 end_chunk = simple_strtoul(buf, &end, 0);
4148 if (buf == end) break;
4150 if (*end && !isspace(*end)) break;
4151 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4152 buf = skip_spaces(end);
4154 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4156 mddev_unlock(mddev);
4160 static struct md_sysfs_entry md_bitmap =
4161 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4164 size_show(struct mddev *mddev, char *page)
4166 return sprintf(page, "%llu\n",
4167 (unsigned long long)mddev->dev_sectors / 2);
4170 static int update_size(struct mddev *mddev, sector_t num_sectors);
4173 size_store(struct mddev *mddev, const char *buf, size_t len)
4175 /* If array is inactive, we can reduce the component size, but
4176 * not increase it (except from 0).
4177 * If array is active, we can try an on-line resize
4180 int err = strict_blocks_to_sectors(buf, §ors);
4184 err = mddev_lock(mddev);
4188 err = update_size(mddev, sectors);
4189 md_update_sb(mddev, 1);
4191 if (mddev->dev_sectors == 0 ||
4192 mddev->dev_sectors > sectors)
4193 mddev->dev_sectors = sectors;
4197 mddev_unlock(mddev);
4198 return err ? err : len;
4201 static struct md_sysfs_entry md_size =
4202 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4204 /* Metadata version.
4206 * 'none' for arrays with no metadata (good luck...)
4207 * 'external' for arrays with externally managed metadata,
4208 * or N.M for internally known formats
4211 metadata_show(struct mddev *mddev, char *page)
4213 if (mddev->persistent)
4214 return sprintf(page, "%d.%d\n",
4215 mddev->major_version, mddev->minor_version);
4216 else if (mddev->external)
4217 return sprintf(page, "external:%s\n", mddev->metadata_type);
4219 return sprintf(page, "none\n");
4223 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4228 /* Changing the details of 'external' metadata is
4229 * always permitted. Otherwise there must be
4230 * no devices attached to the array.
4233 err = mddev_lock(mddev);
4237 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4239 else if (!list_empty(&mddev->disks))
4243 if (cmd_match(buf, "none")) {
4244 mddev->persistent = 0;
4245 mddev->external = 0;
4246 mddev->major_version = 0;
4247 mddev->minor_version = 90;
4250 if (strncmp(buf, "external:", 9) == 0) {
4251 size_t namelen = len-9;
4252 if (namelen >= sizeof(mddev->metadata_type))
4253 namelen = sizeof(mddev->metadata_type)-1;
4254 strncpy(mddev->metadata_type, buf+9, namelen);
4255 mddev->metadata_type[namelen] = 0;
4256 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4257 mddev->metadata_type[--namelen] = 0;
4258 mddev->persistent = 0;
4259 mddev->external = 1;
4260 mddev->major_version = 0;
4261 mddev->minor_version = 90;
4264 major = simple_strtoul(buf, &e, 10);
4266 if (e==buf || *e != '.')
4269 minor = simple_strtoul(buf, &e, 10);
4270 if (e==buf || (*e && *e != '\n') )
4273 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4275 mddev->major_version = major;
4276 mddev->minor_version = minor;
4277 mddev->persistent = 1;
4278 mddev->external = 0;
4281 mddev_unlock(mddev);
4285 static struct md_sysfs_entry md_metadata =
4286 __ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4289 action_show(struct mddev *mddev, char *page)
4291 char *type = "idle";
4292 unsigned long recovery = mddev->recovery;
4293 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4295 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4296 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4297 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4299 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4300 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4302 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4306 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4308 else if (mddev->reshape_position != MaxSector)
4311 return sprintf(page, "%s\n", type);
4315 action_store(struct mddev *mddev, const char *page, size_t len)
4317 if (!mddev->pers || !mddev->pers->sync_request)
4321 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4322 if (cmd_match(page, "frozen"))
4323 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4325 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4326 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4327 mddev_lock(mddev) == 0) {
4328 flush_workqueue(md_misc_wq);
4329 if (mddev->sync_thread) {
4330 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4331 md_reap_sync_thread(mddev);
4333 mddev_unlock(mddev);
4335 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4337 else if (cmd_match(page, "resync"))
4338 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4339 else if (cmd_match(page, "recover")) {
4340 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4341 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4342 } else if (cmd_match(page, "reshape")) {
4344 if (mddev->pers->start_reshape == NULL)
4346 err = mddev_lock(mddev);
4348 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4351 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4352 err = mddev->pers->start_reshape(mddev);
4354 mddev_unlock(mddev);
4358 sysfs_notify(&mddev->kobj, NULL, "degraded");
4360 if (cmd_match(page, "check"))
4361 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4362 else if (!cmd_match(page, "repair"))
4364 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4365 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4366 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4368 if (mddev->ro == 2) {
4369 /* A write to sync_action is enough to justify
4370 * canceling read-auto mode
4373 md_wakeup_thread(mddev->sync_thread);
4375 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4376 md_wakeup_thread(mddev->thread);
4377 sysfs_notify_dirent_safe(mddev->sysfs_action);
4381 static struct md_sysfs_entry md_scan_mode =
4382 __ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4385 last_sync_action_show(struct mddev *mddev, char *page)
4387 return sprintf(page, "%s\n", mddev->last_sync_action);
4390 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4393 mismatch_cnt_show(struct mddev *mddev, char *page)
4395 return sprintf(page, "%llu\n",
4396 (unsigned long long)
4397 atomic64_read(&mddev->resync_mismatches));
4400 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4403 sync_min_show(struct mddev *mddev, char *page)
4405 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4406 mddev->sync_speed_min ? "local": "system");
4410 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4415 if (strncmp(buf, "system", 6)==0) {
4418 rv = kstrtouint(buf, 10, &min);
4424 mddev->sync_speed_min = min;
4428 static struct md_sysfs_entry md_sync_min =
4429 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4432 sync_max_show(struct mddev *mddev, char *page)
4434 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4435 mddev->sync_speed_max ? "local": "system");
4439 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4444 if (strncmp(buf, "system", 6)==0) {
4447 rv = kstrtouint(buf, 10, &max);
4453 mddev->sync_speed_max = max;
4457 static struct md_sysfs_entry md_sync_max =
4458 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4461 degraded_show(struct mddev *mddev, char *page)
4463 return sprintf(page, "%d\n", mddev->degraded);
4465 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4468 sync_force_parallel_show(struct mddev *mddev, char *page)
4470 return sprintf(page, "%d\n", mddev->parallel_resync);
4474 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4478 if (kstrtol(buf, 10, &n))
4481 if (n != 0 && n != 1)
4484 mddev->parallel_resync = n;
4486 if (mddev->sync_thread)
4487 wake_up(&resync_wait);
4492 /* force parallel resync, even with shared block devices */
4493 static struct md_sysfs_entry md_sync_force_parallel =
4494 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4495 sync_force_parallel_show, sync_force_parallel_store);
4498 sync_speed_show(struct mddev *mddev, char *page)
4500 unsigned long resync, dt, db;
4501 if (mddev->curr_resync == 0)
4502 return sprintf(page, "none\n");
4503 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4504 dt = (jiffies - mddev->resync_mark) / HZ;
4506 db = resync - mddev->resync_mark_cnt;
4507 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4510 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4513 sync_completed_show(struct mddev *mddev, char *page)
4515 unsigned long long max_sectors, resync;
4517 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4518 return sprintf(page, "none\n");
4520 if (mddev->curr_resync == 1 ||
4521 mddev->curr_resync == 2)
4522 return sprintf(page, "delayed\n");
4524 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4525 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4526 max_sectors = mddev->resync_max_sectors;
4528 max_sectors = mddev->dev_sectors;
4530 resync = mddev->curr_resync_completed;
4531 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4534 static struct md_sysfs_entry md_sync_completed =
4535 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
4538 min_sync_show(struct mddev *mddev, char *page)
4540 return sprintf(page, "%llu\n",
4541 (unsigned long long)mddev->resync_min);
4544 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4546 unsigned long long min;
4549 if (kstrtoull(buf, 10, &min))
4552 spin_lock(&mddev->lock);
4554 if (min > mddev->resync_max)
4558 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4561 /* Round down to multiple of 4K for safety */
4562 mddev->resync_min = round_down(min, 8);
4566 spin_unlock(&mddev->lock);
4570 static struct md_sysfs_entry md_min_sync =
4571 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4574 max_sync_show(struct mddev *mddev, char *page)
4576 if (mddev->resync_max == MaxSector)
4577 return sprintf(page, "max\n");
4579 return sprintf(page, "%llu\n",
4580 (unsigned long long)mddev->resync_max);
4583 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4586 spin_lock(&mddev->lock);
4587 if (strncmp(buf, "max", 3) == 0)
4588 mddev->resync_max = MaxSector;
4590 unsigned long long max;
4594 if (kstrtoull(buf, 10, &max))
4596 if (max < mddev->resync_min)
4600 if (max < mddev->resync_max &&
4602 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4605 /* Must be a multiple of chunk_size */
4606 chunk = mddev->chunk_sectors;
4608 sector_t temp = max;
4611 if (sector_div(temp, chunk))
4614 mddev->resync_max = max;
4616 wake_up(&mddev->recovery_wait);
4619 spin_unlock(&mddev->lock);
4623 static struct md_sysfs_entry md_max_sync =
4624 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4627 suspend_lo_show(struct mddev *mddev, char *page)
4629 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4633 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4635 unsigned long long old, new;
4638 err = kstrtoull(buf, 10, &new);
4641 if (new != (sector_t)new)
4644 err = mddev_lock(mddev);
4648 if (mddev->pers == NULL ||
4649 mddev->pers->quiesce == NULL)
4651 old = mddev->suspend_lo;
4652 mddev->suspend_lo = new;
4654 /* Shrinking suspended region */
4655 mddev->pers->quiesce(mddev, 2);
4657 /* Expanding suspended region - need to wait */
4658 mddev->pers->quiesce(mddev, 1);
4659 mddev->pers->quiesce(mddev, 0);
4663 mddev_unlock(mddev);
4666 static struct md_sysfs_entry md_suspend_lo =
4667 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4670 suspend_hi_show(struct mddev *mddev, char *page)
4672 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4676 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4678 unsigned long long old, new;
4681 err = kstrtoull(buf, 10, &new);
4684 if (new != (sector_t)new)
4687 err = mddev_lock(mddev);
4691 if (mddev->pers == NULL ||
4692 mddev->pers->quiesce == NULL)
4694 old = mddev->suspend_hi;
4695 mddev->suspend_hi = new;
4697 /* Shrinking suspended region */
4698 mddev->pers->quiesce(mddev, 2);
4700 /* Expanding suspended region - need to wait */
4701 mddev->pers->quiesce(mddev, 1);
4702 mddev->pers->quiesce(mddev, 0);
4706 mddev_unlock(mddev);
4709 static struct md_sysfs_entry md_suspend_hi =
4710 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4713 reshape_position_show(struct mddev *mddev, char *page)
4715 if (mddev->reshape_position != MaxSector)
4716 return sprintf(page, "%llu\n",
4717 (unsigned long long)mddev->reshape_position);
4718 strcpy(page, "none\n");
4723 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4725 struct md_rdev *rdev;
4726 unsigned long long new;
4729 err = kstrtoull(buf, 10, &new);
4732 if (new != (sector_t)new)
4734 err = mddev_lock(mddev);
4740 mddev->reshape_position = new;
4741 mddev->delta_disks = 0;
4742 mddev->reshape_backwards = 0;
4743 mddev->new_level = mddev->level;
4744 mddev->new_layout = mddev->layout;
4745 mddev->new_chunk_sectors = mddev->chunk_sectors;
4746 rdev_for_each(rdev, mddev)
4747 rdev->new_data_offset = rdev->data_offset;
4750 mddev_unlock(mddev);
4754 static struct md_sysfs_entry md_reshape_position =
4755 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4756 reshape_position_store);
4759 reshape_direction_show(struct mddev *mddev, char *page)
4761 return sprintf(page, "%s\n",
4762 mddev->reshape_backwards ? "backwards" : "forwards");
4766 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4771 if (cmd_match(buf, "forwards"))
4773 else if (cmd_match(buf, "backwards"))
4777 if (mddev->reshape_backwards == backwards)
4780 err = mddev_lock(mddev);
4783 /* check if we are allowed to change */
4784 if (mddev->delta_disks)
4786 else if (mddev->persistent &&
4787 mddev->major_version == 0)
4790 mddev->reshape_backwards = backwards;
4791 mddev_unlock(mddev);
4795 static struct md_sysfs_entry md_reshape_direction =
4796 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4797 reshape_direction_store);
4800 array_size_show(struct mddev *mddev, char *page)
4802 if (mddev->external_size)
4803 return sprintf(page, "%llu\n",
4804 (unsigned long long)mddev->array_sectors/2);
4806 return sprintf(page, "default\n");
4810 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4815 err = mddev_lock(mddev);
4819 if (strncmp(buf, "default", 7) == 0) {
4821 sectors = mddev->pers->size(mddev, 0, 0);
4823 sectors = mddev->array_sectors;
4825 mddev->external_size = 0;
4827 if (strict_blocks_to_sectors(buf, §ors) < 0)
4829 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4832 mddev->external_size = 1;
4836 mddev->array_sectors = sectors;
4838 set_capacity(mddev->gendisk, mddev->array_sectors);
4839 revalidate_disk(mddev->gendisk);
4842 mddev_unlock(mddev);
4846 static struct md_sysfs_entry md_array_size =
4847 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4850 static struct attribute *md_default_attrs[] = {
4853 &md_raid_disks.attr,
4854 &md_chunk_size.attr,
4856 &md_resync_start.attr,
4858 &md_new_device.attr,
4859 &md_safe_delay.attr,
4860 &md_array_state.attr,
4861 &md_reshape_position.attr,
4862 &md_reshape_direction.attr,
4863 &md_array_size.attr,
4864 &max_corr_read_errors.attr,
4868 static struct attribute *md_redundancy_attrs[] = {
4870 &md_last_scan_mode.attr,
4871 &md_mismatches.attr,
4874 &md_sync_speed.attr,
4875 &md_sync_force_parallel.attr,
4876 &md_sync_completed.attr,
4879 &md_suspend_lo.attr,
4880 &md_suspend_hi.attr,
4885 static struct attribute_group md_redundancy_group = {
4887 .attrs = md_redundancy_attrs,
4891 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4893 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4894 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4899 spin_lock(&all_mddevs_lock);
4900 if (list_empty(&mddev->all_mddevs)) {
4901 spin_unlock(&all_mddevs_lock);
4905 spin_unlock(&all_mddevs_lock);
4907 rv = entry->show(mddev, page);
4913 md_attr_store(struct kobject *kobj, struct attribute *attr,
4914 const char *page, size_t length)
4916 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4917 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4922 if (!capable(CAP_SYS_ADMIN))
4924 spin_lock(&all_mddevs_lock);
4925 if (list_empty(&mddev->all_mddevs)) {
4926 spin_unlock(&all_mddevs_lock);
4930 spin_unlock(&all_mddevs_lock);
4931 rv = entry->store(mddev, page, length);
4936 static void md_free(struct kobject *ko)
4938 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4940 if (mddev->sysfs_state)
4941 sysfs_put(mddev->sysfs_state);
4944 blk_cleanup_queue(mddev->queue);
4945 if (mddev->gendisk) {
4946 del_gendisk(mddev->gendisk);
4947 put_disk(mddev->gendisk);
4953 static const struct sysfs_ops md_sysfs_ops = {
4954 .show = md_attr_show,
4955 .store = md_attr_store,
4957 static struct kobj_type md_ktype = {
4959 .sysfs_ops = &md_sysfs_ops,
4960 .default_attrs = md_default_attrs,
4965 static void mddev_delayed_delete(struct work_struct *ws)
4967 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4969 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4970 kobject_del(&mddev->kobj);
4971 kobject_put(&mddev->kobj);
4974 static int md_alloc(dev_t dev, char *name)
4976 static DEFINE_MUTEX(disks_mutex);
4977 struct mddev *mddev = mddev_find(dev);
4978 struct gendisk *disk;
4987 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4988 shift = partitioned ? MdpMinorShift : 0;
4989 unit = MINOR(mddev->unit) >> shift;
4991 /* wait for any previous instance of this device to be
4992 * completely removed (mddev_delayed_delete).
4994 flush_workqueue(md_misc_wq);
4996 mutex_lock(&disks_mutex);
5002 /* Need to ensure that 'name' is not a duplicate.
5004 struct mddev *mddev2;
5005 spin_lock(&all_mddevs_lock);
5007 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
5008 if (mddev2->gendisk &&
5009 strcmp(mddev2->gendisk->disk_name, name) == 0) {
5010 spin_unlock(&all_mddevs_lock);
5013 spin_unlock(&all_mddevs_lock);
5017 mddev->queue = blk_alloc_queue(GFP_KERNEL);
5020 mddev->queue->queuedata = mddev;
5022 blk_queue_make_request(mddev->queue, md_make_request);
5023 blk_set_stacking_limits(&mddev->queue->limits);
5025 disk = alloc_disk(1 << shift);
5027 blk_cleanup_queue(mddev->queue);
5028 mddev->queue = NULL;
5031 disk->major = MAJOR(mddev->unit);
5032 disk->first_minor = unit << shift;
5034 strcpy(disk->disk_name, name);
5035 else if (partitioned)
5036 sprintf(disk->disk_name, "md_d%d", unit);
5038 sprintf(disk->disk_name, "md%d", unit);
5039 disk->fops = &md_fops;
5040 disk->private_data = mddev;
5041 disk->queue = mddev->queue;
5042 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
5043 /* Allow extended partitions. This makes the
5044 * 'mdp' device redundant, but we can't really
5047 disk->flags |= GENHD_FL_EXT_DEVT;
5048 mddev->gendisk = disk;
5049 /* As soon as we call add_disk(), another thread could get
5050 * through to md_open, so make sure it doesn't get too far
5052 mutex_lock(&mddev->open_mutex);
5055 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
5056 &disk_to_dev(disk)->kobj, "%s", "md");
5058 /* This isn't possible, but as kobject_init_and_add is marked
5059 * __must_check, we must do something with the result
5061 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
5065 if (mddev->kobj.sd &&
5066 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
5067 printk(KERN_DEBUG "pointless warning\n");
5068 mutex_unlock(&mddev->open_mutex);
5070 mutex_unlock(&disks_mutex);
5071 if (!error && mddev->kobj.sd) {
5072 kobject_uevent(&mddev->kobj, KOBJ_ADD);
5073 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
5079 static struct kobject *md_probe(dev_t dev, int *part, void *data)
5081 md_alloc(dev, NULL);
5085 static int add_named_array(const char *val, struct kernel_param *kp)
5087 /* val must be "md_*" where * is not all digits.
5088 * We allocate an array with a large free minor number, and
5089 * set the name to val. val must not already be an active name.
5091 int len = strlen(val);
5092 char buf[DISK_NAME_LEN];
5094 while (len && val[len-1] == '\n')
5096 if (len >= DISK_NAME_LEN)
5098 strlcpy(buf, val, len+1);
5099 if (strncmp(buf, "md_", 3) != 0)
5101 return md_alloc(0, buf);
5104 static void md_safemode_timeout(unsigned long data)
5106 struct mddev *mddev = (struct mddev *) data;
5108 if (!atomic_read(&mddev->writes_pending)) {
5109 mddev->safemode = 1;
5110 if (mddev->external)
5111 sysfs_notify_dirent_safe(mddev->sysfs_state);
5113 md_wakeup_thread(mddev->thread);
5116 static int start_dirty_degraded;
5118 int md_run(struct mddev *mddev)
5121 struct md_rdev *rdev;
5122 struct md_personality *pers;
5124 if (list_empty(&mddev->disks))
5125 /* cannot run an array with no devices.. */
5130 /* Cannot run until previous stop completes properly */
5131 if (mddev->sysfs_active)
5135 * Analyze all RAID superblock(s)
5137 if (!mddev->raid_disks) {
5138 if (!mddev->persistent)
5143 if (mddev->level != LEVEL_NONE)
5144 request_module("md-level-%d", mddev->level);
5145 else if (mddev->clevel[0])
5146 request_module("md-%s", mddev->clevel);
5149 * Drop all container device buffers, from now on
5150 * the only valid external interface is through the md
5153 rdev_for_each(rdev, mddev) {
5154 if (test_bit(Faulty, &rdev->flags))
5156 sync_blockdev(rdev->bdev);
5157 invalidate_bdev(rdev->bdev);
5159 /* perform some consistency tests on the device.
5160 * We don't want the data to overlap the metadata,
5161 * Internal Bitmap issues have been handled elsewhere.
5163 if (rdev->meta_bdev) {
5164 /* Nothing to check */;
5165 } else if (rdev->data_offset < rdev->sb_start) {
5166 if (mddev->dev_sectors &&
5167 rdev->data_offset + mddev->dev_sectors
5169 printk("md: %s: data overlaps metadata\n",
5174 if (rdev->sb_start + rdev->sb_size/512
5175 > rdev->data_offset) {
5176 printk("md: %s: metadata overlaps data\n",
5181 sysfs_notify_dirent_safe(rdev->sysfs_state);
5184 if (mddev->bio_set == NULL)
5185 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5187 spin_lock(&pers_lock);
5188 pers = find_pers(mddev->level, mddev->clevel);
5189 if (!pers || !try_module_get(pers->owner)) {
5190 spin_unlock(&pers_lock);
5191 if (mddev->level != LEVEL_NONE)
5192 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5195 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5199 spin_unlock(&pers_lock);
5200 if (mddev->level != pers->level) {
5201 mddev->level = pers->level;
5202 mddev->new_level = pers->level;
5204 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5206 if (mddev->reshape_position != MaxSector &&
5207 pers->start_reshape == NULL) {
5208 /* This personality cannot handle reshaping... */
5209 module_put(pers->owner);
5213 if (pers->sync_request) {
5214 /* Warn if this is a potentially silly
5217 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5218 struct md_rdev *rdev2;
5221 rdev_for_each(rdev, mddev)
5222 rdev_for_each(rdev2, mddev) {
5224 rdev->bdev->bd_contains ==
5225 rdev2->bdev->bd_contains) {
5227 "%s: WARNING: %s appears to be"
5228 " on the same physical disk as"
5231 bdevname(rdev->bdev,b),
5232 bdevname(rdev2->bdev,b2));
5239 "True protection against single-disk"
5240 " failure might be compromised.\n");
5243 mddev->recovery = 0;
5244 /* may be over-ridden by personality */
5245 mddev->resync_max_sectors = mddev->dev_sectors;
5247 mddev->ok_start_degraded = start_dirty_degraded;
5249 if (start_readonly && mddev->ro == 0)
5250 mddev->ro = 2; /* read-only, but switch on first write */
5252 err = pers->run(mddev);
5254 printk(KERN_ERR "md: pers->run() failed ...\n");
5255 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5256 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5257 " but 'external_size' not in effect?\n", __func__);
5259 "md: invalid array_size %llu > default size %llu\n",
5260 (unsigned long long)mddev->array_sectors / 2,
5261 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5264 if (err == 0 && pers->sync_request &&
5265 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5266 struct bitmap *bitmap;
5268 bitmap = bitmap_create(mddev, -1);
5269 if (IS_ERR(bitmap)) {
5270 err = PTR_ERR(bitmap);
5271 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5272 mdname(mddev), err);
5274 mddev->bitmap = bitmap;
5278 mddev_detach(mddev);
5280 pers->free(mddev, mddev->private);
5281 mddev->private = NULL;
5282 module_put(pers->owner);
5283 bitmap_destroy(mddev);
5287 mddev->queue->backing_dev_info.congested_data = mddev;
5288 mddev->queue->backing_dev_info.congested_fn = md_congested;
5290 if (pers->sync_request) {
5291 if (mddev->kobj.sd &&
5292 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5294 "md: cannot register extra attributes for %s\n",
5296 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5297 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5300 atomic_set(&mddev->writes_pending,0);
5301 atomic_set(&mddev->max_corr_read_errors,
5302 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5303 mddev->safemode = 0;
5304 if (mddev_is_clustered(mddev))
5305 mddev->safemode_delay = 0;
5307 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5310 spin_lock(&mddev->lock);
5313 spin_unlock(&mddev->lock);
5314 rdev_for_each(rdev, mddev)
5315 if (rdev->raid_disk >= 0)
5316 if (sysfs_link_rdev(mddev, rdev))
5317 /* failure here is OK */;
5319 if (mddev->degraded && !mddev->ro)
5320 /* This ensures that recovering status is reported immediately
5321 * via sysfs - until a lack of spares is confirmed.
5323 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5324 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5326 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5327 md_update_sb(mddev, 0);
5329 md_new_event(mddev);
5330 sysfs_notify_dirent_safe(mddev->sysfs_state);
5331 sysfs_notify_dirent_safe(mddev->sysfs_action);
5332 sysfs_notify(&mddev->kobj, NULL, "degraded");
5335 EXPORT_SYMBOL_GPL(md_run);
5337 static int do_md_run(struct mddev *mddev)
5341 err = md_run(mddev);
5344 err = bitmap_load(mddev);
5346 bitmap_destroy(mddev);
5350 if (mddev_is_clustered(mddev))
5351 md_allow_write(mddev);
5353 md_wakeup_thread(mddev->thread);
5354 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5356 set_capacity(mddev->gendisk, mddev->array_sectors);
5357 revalidate_disk(mddev->gendisk);
5359 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5364 static int restart_array(struct mddev *mddev)
5366 struct gendisk *disk = mddev->gendisk;
5368 /* Complain if it has no devices */
5369 if (list_empty(&mddev->disks))
5375 if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) {
5376 struct md_rdev *rdev;
5377 bool has_journal = false;
5380 rdev_for_each_rcu(rdev, mddev) {
5381 if (test_bit(Journal, &rdev->flags) &&
5382 !test_bit(Faulty, &rdev->flags)) {
5389 /* Don't restart rw with journal missing/faulty */
5394 mddev->safemode = 0;
5396 set_disk_ro(disk, 0);
5397 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5399 /* Kick recovery or resync if necessary */
5400 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5401 md_wakeup_thread(mddev->thread);
5402 md_wakeup_thread(mddev->sync_thread);
5403 sysfs_notify_dirent_safe(mddev->sysfs_state);
5407 static void md_clean(struct mddev *mddev)
5409 mddev->array_sectors = 0;
5410 mddev->external_size = 0;
5411 mddev->dev_sectors = 0;
5412 mddev->raid_disks = 0;
5413 mddev->recovery_cp = 0;
5414 mddev->resync_min = 0;
5415 mddev->resync_max = MaxSector;
5416 mddev->reshape_position = MaxSector;
5417 mddev->external = 0;
5418 mddev->persistent = 0;
5419 mddev->level = LEVEL_NONE;
5420 mddev->clevel[0] = 0;
5423 mddev->metadata_type[0] = 0;
5424 mddev->chunk_sectors = 0;
5425 mddev->ctime = mddev->utime = 0;
5427 mddev->max_disks = 0;
5429 mddev->can_decrease_events = 0;
5430 mddev->delta_disks = 0;
5431 mddev->reshape_backwards = 0;
5432 mddev->new_level = LEVEL_NONE;
5433 mddev->new_layout = 0;
5434 mddev->new_chunk_sectors = 0;
5435 mddev->curr_resync = 0;
5436 atomic64_set(&mddev->resync_mismatches, 0);
5437 mddev->suspend_lo = mddev->suspend_hi = 0;
5438 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5439 mddev->recovery = 0;
5442 mddev->degraded = 0;
5443 mddev->safemode = 0;
5444 mddev->private = NULL;
5445 mddev->bitmap_info.offset = 0;
5446 mddev->bitmap_info.default_offset = 0;
5447 mddev->bitmap_info.default_space = 0;
5448 mddev->bitmap_info.chunksize = 0;
5449 mddev->bitmap_info.daemon_sleep = 0;
5450 mddev->bitmap_info.max_write_behind = 0;
5453 static void __md_stop_writes(struct mddev *mddev)
5455 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5456 flush_workqueue(md_misc_wq);
5457 if (mddev->sync_thread) {
5458 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5459 md_reap_sync_thread(mddev);
5462 del_timer_sync(&mddev->safemode_timer);
5464 bitmap_flush(mddev);
5465 md_super_wait(mddev);
5467 if (mddev->ro == 0 &&
5468 ((!mddev->in_sync && !mddev_is_clustered(mddev)) ||
5469 (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5470 /* mark array as shutdown cleanly */
5471 if (!mddev_is_clustered(mddev))
5473 md_update_sb(mddev, 1);
5477 void md_stop_writes(struct mddev *mddev)
5479 mddev_lock_nointr(mddev);
5480 __md_stop_writes(mddev);
5481 mddev_unlock(mddev);
5483 EXPORT_SYMBOL_GPL(md_stop_writes);
5485 static void mddev_detach(struct mddev *mddev)
5487 struct bitmap *bitmap = mddev->bitmap;
5488 /* wait for behind writes to complete */
5489 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5490 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5492 /* need to kick something here to make sure I/O goes? */
5493 wait_event(bitmap->behind_wait,
5494 atomic_read(&bitmap->behind_writes) == 0);
5496 if (mddev->pers && mddev->pers->quiesce) {
5497 mddev->pers->quiesce(mddev, 1);
5498 mddev->pers->quiesce(mddev, 0);
5500 md_unregister_thread(&mddev->thread);
5502 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5505 static void __md_stop(struct mddev *mddev)
5507 struct md_personality *pers = mddev->pers;
5508 mddev_detach(mddev);
5509 /* Ensure ->event_work is done */
5510 flush_workqueue(md_misc_wq);
5511 spin_lock(&mddev->lock);
5514 spin_unlock(&mddev->lock);
5515 pers->free(mddev, mddev->private);
5516 mddev->private = NULL;
5517 if (pers->sync_request && mddev->to_remove == NULL)
5518 mddev->to_remove = &md_redundancy_group;
5519 module_put(pers->owner);
5520 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5523 void md_stop(struct mddev *mddev)
5525 /* stop the array and free an attached data structures.
5526 * This is called from dm-raid
5529 bitmap_destroy(mddev);
5531 bioset_free(mddev->bio_set);
5534 EXPORT_SYMBOL_GPL(md_stop);
5536 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5541 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5543 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5544 md_wakeup_thread(mddev->thread);
5546 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5547 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5548 if (mddev->sync_thread)
5549 /* Thread might be blocked waiting for metadata update
5550 * which will now never happen */
5551 wake_up_process(mddev->sync_thread->tsk);
5553 if (mddev->external && test_bit(MD_CHANGE_PENDING, &mddev->flags))
5555 mddev_unlock(mddev);
5556 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5558 wait_event(mddev->sb_wait,
5559 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5560 mddev_lock_nointr(mddev);
5562 mutex_lock(&mddev->open_mutex);
5563 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5564 mddev->sync_thread ||
5565 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5566 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5567 printk("md: %s still in use.\n",mdname(mddev));
5569 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5571 md_wakeup_thread(mddev->thread);
5577 __md_stop_writes(mddev);
5583 set_disk_ro(mddev->gendisk, 1);
5584 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5585 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5586 md_wakeup_thread(mddev->thread);
5587 sysfs_notify_dirent_safe(mddev->sysfs_state);
5591 mutex_unlock(&mddev->open_mutex);
5596 * 0 - completely stop and dis-assemble array
5597 * 2 - stop but do not disassemble array
5599 static int do_md_stop(struct mddev *mddev, int mode,
5600 struct block_device *bdev)
5602 struct gendisk *disk = mddev->gendisk;
5603 struct md_rdev *rdev;
5606 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5608 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5609 md_wakeup_thread(mddev->thread);
5611 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5612 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5613 if (mddev->sync_thread)
5614 /* Thread might be blocked waiting for metadata update
5615 * which will now never happen */
5616 wake_up_process(mddev->sync_thread->tsk);
5618 mddev_unlock(mddev);
5619 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5620 !test_bit(MD_RECOVERY_RUNNING,
5621 &mddev->recovery)));
5622 mddev_lock_nointr(mddev);
5624 mutex_lock(&mddev->open_mutex);
5625 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5626 mddev->sysfs_active ||
5627 mddev->sync_thread ||
5628 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5629 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5630 printk("md: %s still in use.\n",mdname(mddev));
5631 mutex_unlock(&mddev->open_mutex);
5633 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5634 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5635 md_wakeup_thread(mddev->thread);
5641 set_disk_ro(disk, 0);
5643 __md_stop_writes(mddev);
5645 mddev->queue->backing_dev_info.congested_fn = NULL;
5647 /* tell userspace to handle 'inactive' */
5648 sysfs_notify_dirent_safe(mddev->sysfs_state);
5650 rdev_for_each(rdev, mddev)
5651 if (rdev->raid_disk >= 0)
5652 sysfs_unlink_rdev(mddev, rdev);
5654 set_capacity(disk, 0);
5655 mutex_unlock(&mddev->open_mutex);
5657 revalidate_disk(disk);
5662 mutex_unlock(&mddev->open_mutex);
5664 * Free resources if final stop
5667 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5669 bitmap_destroy(mddev);
5670 if (mddev->bitmap_info.file) {
5671 struct file *f = mddev->bitmap_info.file;
5672 spin_lock(&mddev->lock);
5673 mddev->bitmap_info.file = NULL;
5674 spin_unlock(&mddev->lock);
5677 mddev->bitmap_info.offset = 0;
5679 export_array(mddev);
5682 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5683 if (mddev->hold_active == UNTIL_STOP)
5684 mddev->hold_active = 0;
5686 md_new_event(mddev);
5687 sysfs_notify_dirent_safe(mddev->sysfs_state);
5692 static void autorun_array(struct mddev *mddev)
5694 struct md_rdev *rdev;
5697 if (list_empty(&mddev->disks))
5700 printk(KERN_INFO "md: running: ");
5702 rdev_for_each(rdev, mddev) {
5703 char b[BDEVNAME_SIZE];
5704 printk("<%s>", bdevname(rdev->bdev,b));
5708 err = do_md_run(mddev);
5710 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5711 do_md_stop(mddev, 0, NULL);
5716 * lets try to run arrays based on all disks that have arrived
5717 * until now. (those are in pending_raid_disks)
5719 * the method: pick the first pending disk, collect all disks with
5720 * the same UUID, remove all from the pending list and put them into
5721 * the 'same_array' list. Then order this list based on superblock
5722 * update time (freshest comes first), kick out 'old' disks and
5723 * compare superblocks. If everything's fine then run it.
5725 * If "unit" is allocated, then bump its reference count
5727 static void autorun_devices(int part)
5729 struct md_rdev *rdev0, *rdev, *tmp;
5730 struct mddev *mddev;
5731 char b[BDEVNAME_SIZE];
5733 printk(KERN_INFO "md: autorun ...\n");
5734 while (!list_empty(&pending_raid_disks)) {
5737 LIST_HEAD(candidates);
5738 rdev0 = list_entry(pending_raid_disks.next,
5739 struct md_rdev, same_set);
5741 printk(KERN_INFO "md: considering %s ...\n",
5742 bdevname(rdev0->bdev,b));
5743 INIT_LIST_HEAD(&candidates);
5744 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5745 if (super_90_load(rdev, rdev0, 0) >= 0) {
5746 printk(KERN_INFO "md: adding %s ...\n",
5747 bdevname(rdev->bdev,b));
5748 list_move(&rdev->same_set, &candidates);
5751 * now we have a set of devices, with all of them having
5752 * mostly sane superblocks. It's time to allocate the
5756 dev = MKDEV(mdp_major,
5757 rdev0->preferred_minor << MdpMinorShift);
5758 unit = MINOR(dev) >> MdpMinorShift;
5760 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5763 if (rdev0->preferred_minor != unit) {
5764 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5765 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5769 md_probe(dev, NULL, NULL);
5770 mddev = mddev_find(dev);
5771 if (!mddev || !mddev->gendisk) {
5775 "md: cannot allocate memory for md drive.\n");
5778 if (mddev_lock(mddev))
5779 printk(KERN_WARNING "md: %s locked, cannot run\n",
5781 else if (mddev->raid_disks || mddev->major_version
5782 || !list_empty(&mddev->disks)) {
5784 "md: %s already running, cannot run %s\n",
5785 mdname(mddev), bdevname(rdev0->bdev,b));
5786 mddev_unlock(mddev);
5788 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5789 mddev->persistent = 1;
5790 rdev_for_each_list(rdev, tmp, &candidates) {
5791 list_del_init(&rdev->same_set);
5792 if (bind_rdev_to_array(rdev, mddev))
5795 autorun_array(mddev);
5796 mddev_unlock(mddev);
5798 /* on success, candidates will be empty, on error
5801 rdev_for_each_list(rdev, tmp, &candidates) {
5802 list_del_init(&rdev->same_set);
5807 printk(KERN_INFO "md: ... autorun DONE.\n");
5809 #endif /* !MODULE */
5811 static int get_version(void __user *arg)
5815 ver.major = MD_MAJOR_VERSION;
5816 ver.minor = MD_MINOR_VERSION;
5817 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5819 if (copy_to_user(arg, &ver, sizeof(ver)))
5825 static int get_array_info(struct mddev *mddev, void __user *arg)
5827 mdu_array_info_t info;
5828 int nr,working,insync,failed,spare;
5829 struct md_rdev *rdev;
5831 nr = working = insync = failed = spare = 0;
5833 rdev_for_each_rcu(rdev, mddev) {
5835 if (test_bit(Faulty, &rdev->flags))
5839 if (test_bit(In_sync, &rdev->flags))
5847 info.major_version = mddev->major_version;
5848 info.minor_version = mddev->minor_version;
5849 info.patch_version = MD_PATCHLEVEL_VERSION;
5850 info.ctime = mddev->ctime;
5851 info.level = mddev->level;
5852 info.size = mddev->dev_sectors / 2;
5853 if (info.size != mddev->dev_sectors / 2) /* overflow */
5856 info.raid_disks = mddev->raid_disks;
5857 info.md_minor = mddev->md_minor;
5858 info.not_persistent= !mddev->persistent;
5860 info.utime = mddev->utime;
5863 info.state = (1<<MD_SB_CLEAN);
5864 if (mddev->bitmap && mddev->bitmap_info.offset)
5865 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5866 if (mddev_is_clustered(mddev))
5867 info.state |= (1<<MD_SB_CLUSTERED);
5868 info.active_disks = insync;
5869 info.working_disks = working;
5870 info.failed_disks = failed;
5871 info.spare_disks = spare;
5873 info.layout = mddev->layout;
5874 info.chunk_size = mddev->chunk_sectors << 9;
5876 if (copy_to_user(arg, &info, sizeof(info)))
5882 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5884 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5888 file = kzalloc(sizeof(*file), GFP_NOIO);
5893 spin_lock(&mddev->lock);
5894 /* bitmap enabled */
5895 if (mddev->bitmap_info.file) {
5896 ptr = file_path(mddev->bitmap_info.file, file->pathname,
5897 sizeof(file->pathname));
5901 memmove(file->pathname, ptr,
5902 sizeof(file->pathname)-(ptr-file->pathname));
5904 spin_unlock(&mddev->lock);
5907 copy_to_user(arg, file, sizeof(*file)))
5914 static int get_disk_info(struct mddev *mddev, void __user * arg)
5916 mdu_disk_info_t info;
5917 struct md_rdev *rdev;
5919 if (copy_from_user(&info, arg, sizeof(info)))
5923 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5925 info.major = MAJOR(rdev->bdev->bd_dev);
5926 info.minor = MINOR(rdev->bdev->bd_dev);
5927 info.raid_disk = rdev->raid_disk;
5929 if (test_bit(Faulty, &rdev->flags))
5930 info.state |= (1<<MD_DISK_FAULTY);
5931 else if (test_bit(In_sync, &rdev->flags)) {
5932 info.state |= (1<<MD_DISK_ACTIVE);
5933 info.state |= (1<<MD_DISK_SYNC);
5935 if (test_bit(Journal, &rdev->flags))
5936 info.state |= (1<<MD_DISK_JOURNAL);
5937 if (test_bit(WriteMostly, &rdev->flags))
5938 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5940 info.major = info.minor = 0;
5941 info.raid_disk = -1;
5942 info.state = (1<<MD_DISK_REMOVED);
5946 if (copy_to_user(arg, &info, sizeof(info)))
5952 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5954 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5955 struct md_rdev *rdev;
5956 dev_t dev = MKDEV(info->major,info->minor);
5958 if (mddev_is_clustered(mddev) &&
5959 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5960 pr_err("%s: Cannot add to clustered mddev.\n",
5965 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5968 if (!mddev->raid_disks) {
5970 /* expecting a device which has a superblock */
5971 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5974 "md: md_import_device returned %ld\n",
5976 return PTR_ERR(rdev);
5978 if (!list_empty(&mddev->disks)) {
5979 struct md_rdev *rdev0
5980 = list_entry(mddev->disks.next,
5981 struct md_rdev, same_set);
5982 err = super_types[mddev->major_version]
5983 .load_super(rdev, rdev0, mddev->minor_version);
5986 "md: %s has different UUID to %s\n",
5987 bdevname(rdev->bdev,b),
5988 bdevname(rdev0->bdev,b2));
5993 err = bind_rdev_to_array(rdev, mddev);
6000 * add_new_disk can be used once the array is assembled
6001 * to add "hot spares". They must already have a superblock
6006 if (!mddev->pers->hot_add_disk) {
6008 "%s: personality does not support diskops!\n",
6012 if (mddev->persistent)
6013 rdev = md_import_device(dev, mddev->major_version,
6014 mddev->minor_version);
6016 rdev = md_import_device(dev, -1, -1);
6019 "md: md_import_device returned %ld\n",
6021 return PTR_ERR(rdev);
6023 /* set saved_raid_disk if appropriate */
6024 if (!mddev->persistent) {
6025 if (info->state & (1<<MD_DISK_SYNC) &&
6026 info->raid_disk < mddev->raid_disks) {
6027 rdev->raid_disk = info->raid_disk;
6028 set_bit(In_sync, &rdev->flags);
6029 clear_bit(Bitmap_sync, &rdev->flags);
6031 rdev->raid_disk = -1;
6032 rdev->saved_raid_disk = rdev->raid_disk;
6034 super_types[mddev->major_version].
6035 validate_super(mddev, rdev);
6036 if ((info->state & (1<<MD_DISK_SYNC)) &&
6037 rdev->raid_disk != info->raid_disk) {
6038 /* This was a hot-add request, but events doesn't
6039 * match, so reject it.
6045 clear_bit(In_sync, &rdev->flags); /* just to be sure */
6046 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6047 set_bit(WriteMostly, &rdev->flags);
6049 clear_bit(WriteMostly, &rdev->flags);
6051 if (info->state & (1<<MD_DISK_JOURNAL))
6052 set_bit(Journal, &rdev->flags);
6054 * check whether the device shows up in other nodes
6056 if (mddev_is_clustered(mddev)) {
6057 if (info->state & (1 << MD_DISK_CANDIDATE))
6058 set_bit(Candidate, &rdev->flags);
6059 else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
6060 /* --add initiated by this node */
6061 err = md_cluster_ops->add_new_disk(mddev, rdev);
6069 rdev->raid_disk = -1;
6070 err = bind_rdev_to_array(rdev, mddev);
6075 if (mddev_is_clustered(mddev)) {
6076 if (info->state & (1 << MD_DISK_CANDIDATE))
6077 md_cluster_ops->new_disk_ack(mddev, (err == 0));
6080 md_cluster_ops->add_new_disk_cancel(mddev);
6082 err = add_bound_rdev(rdev);
6086 err = add_bound_rdev(rdev);
6091 /* otherwise, add_new_disk is only allowed
6092 * for major_version==0 superblocks
6094 if (mddev->major_version != 0) {
6095 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
6100 if (!(info->state & (1<<MD_DISK_FAULTY))) {
6102 rdev = md_import_device(dev, -1, 0);
6105 "md: error, md_import_device() returned %ld\n",
6107 return PTR_ERR(rdev);
6109 rdev->desc_nr = info->number;
6110 if (info->raid_disk < mddev->raid_disks)
6111 rdev->raid_disk = info->raid_disk;
6113 rdev->raid_disk = -1;
6115 if (rdev->raid_disk < mddev->raid_disks)
6116 if (info->state & (1<<MD_DISK_SYNC))
6117 set_bit(In_sync, &rdev->flags);
6119 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
6120 set_bit(WriteMostly, &rdev->flags);
6122 if (!mddev->persistent) {
6123 printk(KERN_INFO "md: nonpersistent superblock ...\n");
6124 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6126 rdev->sb_start = calc_dev_sboffset(rdev);
6127 rdev->sectors = rdev->sb_start;
6129 err = bind_rdev_to_array(rdev, mddev);
6139 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
6141 char b[BDEVNAME_SIZE];
6142 struct md_rdev *rdev;
6145 rdev = find_rdev(mddev, dev);
6149 if (mddev_is_clustered(mddev))
6150 ret = md_cluster_ops->metadata_update_start(mddev);
6152 if (rdev->raid_disk < 0)
6155 clear_bit(Blocked, &rdev->flags);
6156 remove_and_add_spares(mddev, rdev);
6158 if (rdev->raid_disk >= 0)
6162 if (mddev_is_clustered(mddev) && ret == 0)
6163 md_cluster_ops->remove_disk(mddev, rdev);
6165 md_kick_rdev_from_array(rdev);
6166 md_update_sb(mddev, 1);
6167 md_new_event(mddev);
6171 if (mddev_is_clustered(mddev) && ret == 0)
6172 md_cluster_ops->metadata_update_cancel(mddev);
6174 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6175 bdevname(rdev->bdev,b), mdname(mddev));
6179 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6181 char b[BDEVNAME_SIZE];
6183 struct md_rdev *rdev;
6188 if (mddev->major_version != 0) {
6189 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6190 " version-0 superblocks.\n",
6194 if (!mddev->pers->hot_add_disk) {
6196 "%s: personality does not support diskops!\n",
6201 rdev = md_import_device(dev, -1, 0);
6204 "md: error, md_import_device() returned %ld\n",
6209 if (mddev->persistent)
6210 rdev->sb_start = calc_dev_sboffset(rdev);
6212 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6214 rdev->sectors = rdev->sb_start;
6216 if (test_bit(Faulty, &rdev->flags)) {
6218 "md: can not hot-add faulty %s disk to %s!\n",
6219 bdevname(rdev->bdev,b), mdname(mddev));
6224 clear_bit(In_sync, &rdev->flags);
6226 rdev->saved_raid_disk = -1;
6227 err = bind_rdev_to_array(rdev, mddev);
6232 * The rest should better be atomic, we can have disk failures
6233 * noticed in interrupt contexts ...
6236 rdev->raid_disk = -1;
6238 md_update_sb(mddev, 1);
6240 * Kick recovery, maybe this spare has to be added to the
6241 * array immediately.
6243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6244 md_wakeup_thread(mddev->thread);
6245 md_new_event(mddev);
6253 static int set_bitmap_file(struct mddev *mddev, int fd)
6258 if (!mddev->pers->quiesce || !mddev->thread)
6260 if (mddev->recovery || mddev->sync_thread)
6262 /* we should be able to change the bitmap.. */
6266 struct inode *inode;
6269 if (mddev->bitmap || mddev->bitmap_info.file)
6270 return -EEXIST; /* cannot add when bitmap is present */
6274 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6279 inode = f->f_mapping->host;
6280 if (!S_ISREG(inode->i_mode)) {
6281 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6284 } else if (!(f->f_mode & FMODE_WRITE)) {
6285 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6288 } else if (atomic_read(&inode->i_writecount) != 1) {
6289 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6297 mddev->bitmap_info.file = f;
6298 mddev->bitmap_info.offset = 0; /* file overrides offset */
6299 } else if (mddev->bitmap == NULL)
6300 return -ENOENT; /* cannot remove what isn't there */
6303 mddev->pers->quiesce(mddev, 1);
6305 struct bitmap *bitmap;
6307 bitmap = bitmap_create(mddev, -1);
6308 if (!IS_ERR(bitmap)) {
6309 mddev->bitmap = bitmap;
6310 err = bitmap_load(mddev);
6312 err = PTR_ERR(bitmap);
6314 if (fd < 0 || err) {
6315 bitmap_destroy(mddev);
6316 fd = -1; /* make sure to put the file */
6318 mddev->pers->quiesce(mddev, 0);
6321 struct file *f = mddev->bitmap_info.file;
6323 spin_lock(&mddev->lock);
6324 mddev->bitmap_info.file = NULL;
6325 spin_unlock(&mddev->lock);
6334 * set_array_info is used two different ways
6335 * The original usage is when creating a new array.
6336 * In this usage, raid_disks is > 0 and it together with
6337 * level, size, not_persistent,layout,chunksize determine the
6338 * shape of the array.
6339 * This will always create an array with a type-0.90.0 superblock.
6340 * The newer usage is when assembling an array.
6341 * In this case raid_disks will be 0, and the major_version field is
6342 * use to determine which style super-blocks are to be found on the devices.
6343 * The minor and patch _version numbers are also kept incase the
6344 * super_block handler wishes to interpret them.
6346 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6349 if (info->raid_disks == 0) {
6350 /* just setting version number for superblock loading */
6351 if (info->major_version < 0 ||
6352 info->major_version >= ARRAY_SIZE(super_types) ||
6353 super_types[info->major_version].name == NULL) {
6354 /* maybe try to auto-load a module? */
6356 "md: superblock version %d not known\n",
6357 info->major_version);
6360 mddev->major_version = info->major_version;
6361 mddev->minor_version = info->minor_version;
6362 mddev->patch_version = info->patch_version;
6363 mddev->persistent = !info->not_persistent;
6364 /* ensure mddev_put doesn't delete this now that there
6365 * is some minimal configuration.
6367 mddev->ctime = get_seconds();
6370 mddev->major_version = MD_MAJOR_VERSION;
6371 mddev->minor_version = MD_MINOR_VERSION;
6372 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6373 mddev->ctime = get_seconds();
6375 mddev->level = info->level;
6376 mddev->clevel[0] = 0;
6377 mddev->dev_sectors = 2 * (sector_t)info->size;
6378 mddev->raid_disks = info->raid_disks;
6379 /* don't set md_minor, it is determined by which /dev/md* was
6382 if (info->state & (1<<MD_SB_CLEAN))
6383 mddev->recovery_cp = MaxSector;
6385 mddev->recovery_cp = 0;
6386 mddev->persistent = ! info->not_persistent;
6387 mddev->external = 0;
6389 mddev->layout = info->layout;
6390 mddev->chunk_sectors = info->chunk_size >> 9;
6392 mddev->max_disks = MD_SB_DISKS;
6394 if (mddev->persistent)
6396 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6398 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6399 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6400 mddev->bitmap_info.offset = 0;
6402 mddev->reshape_position = MaxSector;
6405 * Generate a 128 bit UUID
6407 get_random_bytes(mddev->uuid, 16);
6409 mddev->new_level = mddev->level;
6410 mddev->new_chunk_sectors = mddev->chunk_sectors;
6411 mddev->new_layout = mddev->layout;
6412 mddev->delta_disks = 0;
6413 mddev->reshape_backwards = 0;
6418 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6420 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6422 if (mddev->external_size)
6425 mddev->array_sectors = array_sectors;
6427 EXPORT_SYMBOL(md_set_array_sectors);
6429 static int update_size(struct mddev *mddev, sector_t num_sectors)
6431 struct md_rdev *rdev;
6433 int fit = (num_sectors == 0);
6435 if (mddev->pers->resize == NULL)
6437 /* The "num_sectors" is the number of sectors of each device that
6438 * is used. This can only make sense for arrays with redundancy.
6439 * linear and raid0 always use whatever space is available. We can only
6440 * consider changing this number if no resync or reconstruction is
6441 * happening, and if the new size is acceptable. It must fit before the
6442 * sb_start or, if that is <data_offset, it must fit before the size
6443 * of each device. If num_sectors is zero, we find the largest size
6446 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6452 rdev_for_each(rdev, mddev) {
6453 sector_t avail = rdev->sectors;
6455 if (fit && (num_sectors == 0 || num_sectors > avail))
6456 num_sectors = avail;
6457 if (avail < num_sectors)
6460 rv = mddev->pers->resize(mddev, num_sectors);
6462 revalidate_disk(mddev->gendisk);
6466 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6469 struct md_rdev *rdev;
6470 /* change the number of raid disks */
6471 if (mddev->pers->check_reshape == NULL)
6475 if (raid_disks <= 0 ||
6476 (mddev->max_disks && raid_disks >= mddev->max_disks))
6478 if (mddev->sync_thread ||
6479 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6480 mddev->reshape_position != MaxSector)
6483 rdev_for_each(rdev, mddev) {
6484 if (mddev->raid_disks < raid_disks &&
6485 rdev->data_offset < rdev->new_data_offset)
6487 if (mddev->raid_disks > raid_disks &&
6488 rdev->data_offset > rdev->new_data_offset)
6492 mddev->delta_disks = raid_disks - mddev->raid_disks;
6493 if (mddev->delta_disks < 0)
6494 mddev->reshape_backwards = 1;
6495 else if (mddev->delta_disks > 0)
6496 mddev->reshape_backwards = 0;
6498 rv = mddev->pers->check_reshape(mddev);
6500 mddev->delta_disks = 0;
6501 mddev->reshape_backwards = 0;
6507 * update_array_info is used to change the configuration of an
6509 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6510 * fields in the info are checked against the array.
6511 * Any differences that cannot be handled will cause an error.
6512 * Normally, only one change can be managed at a time.
6514 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6520 /* calculate expected state,ignoring low bits */
6521 if (mddev->bitmap && mddev->bitmap_info.offset)
6522 state |= (1 << MD_SB_BITMAP_PRESENT);
6524 if (mddev->major_version != info->major_version ||
6525 mddev->minor_version != info->minor_version ||
6526 /* mddev->patch_version != info->patch_version || */
6527 mddev->ctime != info->ctime ||
6528 mddev->level != info->level ||
6529 /* mddev->layout != info->layout || */
6530 mddev->persistent != !info->not_persistent ||
6531 mddev->chunk_sectors != info->chunk_size >> 9 ||
6532 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6533 ((state^info->state) & 0xfffffe00)
6536 /* Check there is only one change */
6537 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6539 if (mddev->raid_disks != info->raid_disks)
6541 if (mddev->layout != info->layout)
6543 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6550 if (mddev->layout != info->layout) {
6552 * we don't need to do anything at the md level, the
6553 * personality will take care of it all.
6555 if (mddev->pers->check_reshape == NULL)
6558 mddev->new_layout = info->layout;
6559 rv = mddev->pers->check_reshape(mddev);
6561 mddev->new_layout = mddev->layout;
6565 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6566 rv = update_size(mddev, (sector_t)info->size * 2);
6568 if (mddev->raid_disks != info->raid_disks)
6569 rv = update_raid_disks(mddev, info->raid_disks);
6571 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6572 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6576 if (mddev->recovery || mddev->sync_thread) {
6580 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6581 struct bitmap *bitmap;
6582 /* add the bitmap */
6583 if (mddev->bitmap) {
6587 if (mddev->bitmap_info.default_offset == 0) {
6591 mddev->bitmap_info.offset =
6592 mddev->bitmap_info.default_offset;
6593 mddev->bitmap_info.space =
6594 mddev->bitmap_info.default_space;
6595 mddev->pers->quiesce(mddev, 1);
6596 bitmap = bitmap_create(mddev, -1);
6597 if (!IS_ERR(bitmap)) {
6598 mddev->bitmap = bitmap;
6599 rv = bitmap_load(mddev);
6601 rv = PTR_ERR(bitmap);
6603 bitmap_destroy(mddev);
6604 mddev->pers->quiesce(mddev, 0);
6606 /* remove the bitmap */
6607 if (!mddev->bitmap) {
6611 if (mddev->bitmap->storage.file) {
6615 mddev->pers->quiesce(mddev, 1);
6616 bitmap_destroy(mddev);
6617 mddev->pers->quiesce(mddev, 0);
6618 mddev->bitmap_info.offset = 0;
6621 md_update_sb(mddev, 1);
6627 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6629 struct md_rdev *rdev;
6632 if (mddev->pers == NULL)
6636 rdev = find_rdev_rcu(mddev, dev);
6640 md_error(mddev, rdev);
6641 if (!test_bit(Faulty, &rdev->flags))
6649 * We have a problem here : there is no easy way to give a CHS
6650 * virtual geometry. We currently pretend that we have a 2 heads
6651 * 4 sectors (with a BIG number of cylinders...). This drives
6652 * dosfs just mad... ;-)
6654 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6656 struct mddev *mddev = bdev->bd_disk->private_data;
6660 geo->cylinders = mddev->array_sectors / 8;
6664 static inline bool md_ioctl_valid(unsigned int cmd)
6669 case GET_ARRAY_INFO:
6670 case GET_BITMAP_FILE:
6673 case HOT_REMOVE_DISK:
6676 case RESTART_ARRAY_RW:
6678 case SET_ARRAY_INFO:
6679 case SET_BITMAP_FILE:
6680 case SET_DISK_FAULTY:
6683 case CLUSTERED_DISK_NACK:
6690 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6691 unsigned int cmd, unsigned long arg)
6694 void __user *argp = (void __user *)arg;
6695 struct mddev *mddev = NULL;
6698 if (!md_ioctl_valid(cmd))
6703 case GET_ARRAY_INFO:
6707 if (!capable(CAP_SYS_ADMIN))
6712 * Commands dealing with the RAID driver but not any
6717 err = get_version(argp);
6723 autostart_arrays(arg);
6730 * Commands creating/starting a new array:
6733 mddev = bdev->bd_disk->private_data;
6740 /* Some actions do not requires the mutex */
6742 case GET_ARRAY_INFO:
6743 if (!mddev->raid_disks && !mddev->external)
6746 err = get_array_info(mddev, argp);
6750 if (!mddev->raid_disks && !mddev->external)
6753 err = get_disk_info(mddev, argp);
6756 case SET_DISK_FAULTY:
6757 err = set_disk_faulty(mddev, new_decode_dev(arg));
6760 case GET_BITMAP_FILE:
6761 err = get_bitmap_file(mddev, argp);
6766 if (cmd == ADD_NEW_DISK)
6767 /* need to ensure md_delayed_delete() has completed */
6768 flush_workqueue(md_misc_wq);
6770 if (cmd == HOT_REMOVE_DISK)
6771 /* need to ensure recovery thread has run */
6772 wait_event_interruptible_timeout(mddev->sb_wait,
6773 !test_bit(MD_RECOVERY_NEEDED,
6775 msecs_to_jiffies(5000));
6776 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6777 /* Need to flush page cache, and ensure no-one else opens
6780 mutex_lock(&mddev->open_mutex);
6781 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6782 mutex_unlock(&mddev->open_mutex);
6786 set_bit(MD_STILL_CLOSED, &mddev->flags);
6787 mutex_unlock(&mddev->open_mutex);
6788 sync_blockdev(bdev);
6790 err = mddev_lock(mddev);
6793 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6798 if (cmd == SET_ARRAY_INFO) {
6799 mdu_array_info_t info;
6801 memset(&info, 0, sizeof(info));
6802 else if (copy_from_user(&info, argp, sizeof(info))) {
6807 err = update_array_info(mddev, &info);
6809 printk(KERN_WARNING "md: couldn't update"
6810 " array info. %d\n", err);
6815 if (!list_empty(&mddev->disks)) {
6817 "md: array %s already has disks!\n",
6822 if (mddev->raid_disks) {
6824 "md: array %s already initialised!\n",
6829 err = set_array_info(mddev, &info);
6831 printk(KERN_WARNING "md: couldn't set"
6832 " array info. %d\n", err);
6839 * Commands querying/configuring an existing array:
6841 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6842 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6843 if ((!mddev->raid_disks && !mddev->external)
6844 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6845 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6846 && cmd != GET_BITMAP_FILE) {
6852 * Commands even a read-only array can execute:
6855 case RESTART_ARRAY_RW:
6856 err = restart_array(mddev);
6860 err = do_md_stop(mddev, 0, bdev);
6864 err = md_set_readonly(mddev, bdev);
6867 case HOT_REMOVE_DISK:
6868 err = hot_remove_disk(mddev, new_decode_dev(arg));
6872 /* We can support ADD_NEW_DISK on read-only arrays
6873 * on if we are re-adding a preexisting device.
6874 * So require mddev->pers and MD_DISK_SYNC.
6877 mdu_disk_info_t info;
6878 if (copy_from_user(&info, argp, sizeof(info)))
6880 else if (!(info.state & (1<<MD_DISK_SYNC)))
6881 /* Need to clear read-only for this */
6884 err = add_new_disk(mddev, &info);
6890 if (get_user(ro, (int __user *)(arg))) {
6896 /* if the bdev is going readonly the value of mddev->ro
6897 * does not matter, no writes are coming
6902 /* are we are already prepared for writes? */
6906 /* transitioning to readauto need only happen for
6907 * arrays that call md_write_start
6910 err = restart_array(mddev);
6913 set_disk_ro(mddev->gendisk, 0);
6920 * The remaining ioctls are changing the state of the
6921 * superblock, so we do not allow them on read-only arrays.
6923 if (mddev->ro && mddev->pers) {
6924 if (mddev->ro == 2) {
6926 sysfs_notify_dirent_safe(mddev->sysfs_state);
6927 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6928 /* mddev_unlock will wake thread */
6929 /* If a device failed while we were read-only, we
6930 * need to make sure the metadata is updated now.
6932 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6933 mddev_unlock(mddev);
6934 wait_event(mddev->sb_wait,
6935 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6936 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6937 mddev_lock_nointr(mddev);
6948 mdu_disk_info_t info;
6949 if (copy_from_user(&info, argp, sizeof(info)))
6952 err = add_new_disk(mddev, &info);
6956 case CLUSTERED_DISK_NACK:
6957 if (mddev_is_clustered(mddev))
6958 md_cluster_ops->new_disk_ack(mddev, false);
6964 err = hot_add_disk(mddev, new_decode_dev(arg));
6968 err = do_md_run(mddev);
6971 case SET_BITMAP_FILE:
6972 err = set_bitmap_file(mddev, (int)arg);
6981 if (mddev->hold_active == UNTIL_IOCTL &&
6983 mddev->hold_active = 0;
6984 mddev_unlock(mddev);
6988 #ifdef CONFIG_COMPAT
6989 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6990 unsigned int cmd, unsigned long arg)
6993 case HOT_REMOVE_DISK:
6995 case SET_DISK_FAULTY:
6996 case SET_BITMAP_FILE:
6997 /* These take in integer arg, do not convert */
7000 arg = (unsigned long)compat_ptr(arg);
7004 return md_ioctl(bdev, mode, cmd, arg);
7006 #endif /* CONFIG_COMPAT */
7008 static int md_open(struct block_device *bdev, fmode_t mode)
7011 * Succeed if we can lock the mddev, which confirms that
7012 * it isn't being stopped right now.
7014 struct mddev *mddev = mddev_find(bdev->bd_dev);
7020 if (mddev->gendisk != bdev->bd_disk) {
7021 /* we are racing with mddev_put which is discarding this
7025 /* Wait until bdev->bd_disk is definitely gone */
7026 flush_workqueue(md_misc_wq);
7027 /* Then retry the open from the top */
7028 return -ERESTARTSYS;
7030 BUG_ON(mddev != bdev->bd_disk->private_data);
7032 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
7036 atomic_inc(&mddev->openers);
7037 clear_bit(MD_STILL_CLOSED, &mddev->flags);
7038 mutex_unlock(&mddev->open_mutex);
7040 check_disk_change(bdev);
7045 static void md_release(struct gendisk *disk, fmode_t mode)
7047 struct mddev *mddev = disk->private_data;
7050 atomic_dec(&mddev->openers);
7054 static int md_media_changed(struct gendisk *disk)
7056 struct mddev *mddev = disk->private_data;
7058 return mddev->changed;
7061 static int md_revalidate(struct gendisk *disk)
7063 struct mddev *mddev = disk->private_data;
7068 static const struct block_device_operations md_fops =
7070 .owner = THIS_MODULE,
7072 .release = md_release,
7074 #ifdef CONFIG_COMPAT
7075 .compat_ioctl = md_compat_ioctl,
7077 .getgeo = md_getgeo,
7078 .media_changed = md_media_changed,
7079 .revalidate_disk= md_revalidate,
7082 static int md_thread(void *arg)
7084 struct md_thread *thread = arg;
7087 * md_thread is a 'system-thread', it's priority should be very
7088 * high. We avoid resource deadlocks individually in each
7089 * raid personality. (RAID5 does preallocation) We also use RR and
7090 * the very same RT priority as kswapd, thus we will never get
7091 * into a priority inversion deadlock.
7093 * we definitely have to have equal or higher priority than
7094 * bdflush, otherwise bdflush will deadlock if there are too
7095 * many dirty RAID5 blocks.
7098 allow_signal(SIGKILL);
7099 while (!kthread_should_stop()) {
7101 /* We need to wait INTERRUPTIBLE so that
7102 * we don't add to the load-average.
7103 * That means we need to be sure no signals are
7106 if (signal_pending(current))
7107 flush_signals(current);
7109 wait_event_interruptible_timeout
7111 test_bit(THREAD_WAKEUP, &thread->flags)
7112 || kthread_should_stop(),
7115 clear_bit(THREAD_WAKEUP, &thread->flags);
7116 if (!kthread_should_stop())
7117 thread->run(thread);
7123 void md_wakeup_thread(struct md_thread *thread)
7126 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
7127 set_bit(THREAD_WAKEUP, &thread->flags);
7128 wake_up(&thread->wqueue);
7131 EXPORT_SYMBOL(md_wakeup_thread);
7133 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
7134 struct mddev *mddev, const char *name)
7136 struct md_thread *thread;
7138 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
7142 init_waitqueue_head(&thread->wqueue);
7145 thread->mddev = mddev;
7146 thread->timeout = MAX_SCHEDULE_TIMEOUT;
7147 thread->tsk = kthread_run(md_thread, thread,
7149 mdname(thread->mddev),
7151 if (IS_ERR(thread->tsk)) {
7157 EXPORT_SYMBOL(md_register_thread);
7159 void md_unregister_thread(struct md_thread **threadp)
7161 struct md_thread *thread = *threadp;
7164 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7165 /* Locking ensures that mddev_unlock does not wake_up a
7166 * non-existent thread
7168 spin_lock(&pers_lock);
7170 spin_unlock(&pers_lock);
7172 kthread_stop(thread->tsk);
7175 EXPORT_SYMBOL(md_unregister_thread);
7177 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7179 if (!rdev || test_bit(Faulty, &rdev->flags))
7182 if (!mddev->pers || !mddev->pers->error_handler)
7184 mddev->pers->error_handler(mddev,rdev);
7185 if (mddev->degraded)
7186 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7187 sysfs_notify_dirent_safe(rdev->sysfs_state);
7188 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7189 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7190 md_wakeup_thread(mddev->thread);
7191 if (mddev->event_work.func)
7192 queue_work(md_misc_wq, &mddev->event_work);
7193 md_new_event_inintr(mddev);
7195 EXPORT_SYMBOL(md_error);
7197 /* seq_file implementation /proc/mdstat */
7199 static void status_unused(struct seq_file *seq)
7202 struct md_rdev *rdev;
7204 seq_printf(seq, "unused devices: ");
7206 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7207 char b[BDEVNAME_SIZE];
7209 seq_printf(seq, "%s ",
7210 bdevname(rdev->bdev,b));
7213 seq_printf(seq, "<none>");
7215 seq_printf(seq, "\n");
7218 static int status_resync(struct seq_file *seq, struct mddev *mddev)
7220 sector_t max_sectors, resync, res;
7221 unsigned long dt, db;
7224 unsigned int per_milli;
7226 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7227 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7228 max_sectors = mddev->resync_max_sectors;
7230 max_sectors = mddev->dev_sectors;
7232 resync = mddev->curr_resync;
7234 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7235 /* Still cleaning up */
7236 resync = max_sectors;
7238 resync -= atomic_read(&mddev->recovery_active);
7241 if (mddev->recovery_cp < MaxSector) {
7242 seq_printf(seq, "\tresync=PENDING");
7248 seq_printf(seq, "\tresync=DELAYED");
7252 WARN_ON(max_sectors == 0);
7253 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7254 * in a sector_t, and (max_sectors>>scale) will fit in a
7255 * u32, as those are the requirements for sector_div.
7256 * Thus 'scale' must be at least 10
7259 if (sizeof(sector_t) > sizeof(unsigned long)) {
7260 while ( max_sectors/2 > (1ULL<<(scale+32)))
7263 res = (resync>>scale)*1000;
7264 sector_div(res, (u32)((max_sectors>>scale)+1));
7268 int i, x = per_milli/50, y = 20-x;
7269 seq_printf(seq, "[");
7270 for (i = 0; i < x; i++)
7271 seq_printf(seq, "=");
7272 seq_printf(seq, ">");
7273 for (i = 0; i < y; i++)
7274 seq_printf(seq, ".");
7275 seq_printf(seq, "] ");
7277 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7278 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7280 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7282 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7283 "resync" : "recovery"))),
7284 per_milli/10, per_milli % 10,
7285 (unsigned long long) resync/2,
7286 (unsigned long long) max_sectors/2);
7289 * dt: time from mark until now
7290 * db: blocks written from mark until now
7291 * rt: remaining time
7293 * rt is a sector_t, so could be 32bit or 64bit.
7294 * So we divide before multiply in case it is 32bit and close
7296 * We scale the divisor (db) by 32 to avoid losing precision
7297 * near the end of resync when the number of remaining sectors
7299 * We then divide rt by 32 after multiplying by db to compensate.
7300 * The '+1' avoids division by zero if db is very small.
7302 dt = ((jiffies - mddev->resync_mark) / HZ);
7304 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7305 - mddev->resync_mark_cnt;
7307 rt = max_sectors - resync; /* number of remaining sectors */
7308 sector_div(rt, db/32+1);
7312 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7313 ((unsigned long)rt % 60)/6);
7315 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7319 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7321 struct list_head *tmp;
7323 struct mddev *mddev;
7331 spin_lock(&all_mddevs_lock);
7332 list_for_each(tmp,&all_mddevs)
7334 mddev = list_entry(tmp, struct mddev, all_mddevs);
7336 spin_unlock(&all_mddevs_lock);
7339 spin_unlock(&all_mddevs_lock);
7341 return (void*)2;/* tail */
7345 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7347 struct list_head *tmp;
7348 struct mddev *next_mddev, *mddev = v;
7354 spin_lock(&all_mddevs_lock);
7356 tmp = all_mddevs.next;
7358 tmp = mddev->all_mddevs.next;
7359 if (tmp != &all_mddevs)
7360 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7362 next_mddev = (void*)2;
7365 spin_unlock(&all_mddevs_lock);
7373 static void md_seq_stop(struct seq_file *seq, void *v)
7375 struct mddev *mddev = v;
7377 if (mddev && v != (void*)1 && v != (void*)2)
7381 static int md_seq_show(struct seq_file *seq, void *v)
7383 struct mddev *mddev = v;
7385 struct md_rdev *rdev;
7387 if (v == (void*)1) {
7388 struct md_personality *pers;
7389 seq_printf(seq, "Personalities : ");
7390 spin_lock(&pers_lock);
7391 list_for_each_entry(pers, &pers_list, list)
7392 seq_printf(seq, "[%s] ", pers->name);
7394 spin_unlock(&pers_lock);
7395 seq_printf(seq, "\n");
7396 seq->poll_event = atomic_read(&md_event_count);
7399 if (v == (void*)2) {
7404 spin_lock(&mddev->lock);
7405 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7406 seq_printf(seq, "%s : %sactive", mdname(mddev),
7407 mddev->pers ? "" : "in");
7410 seq_printf(seq, " (read-only)");
7412 seq_printf(seq, " (auto-read-only)");
7413 seq_printf(seq, " %s", mddev->pers->name);
7418 rdev_for_each_rcu(rdev, mddev) {
7419 char b[BDEVNAME_SIZE];
7420 seq_printf(seq, " %s[%d]",
7421 bdevname(rdev->bdev,b), rdev->desc_nr);
7422 if (test_bit(WriteMostly, &rdev->flags))
7423 seq_printf(seq, "(W)");
7424 if (test_bit(Journal, &rdev->flags))
7425 seq_printf(seq, "(J)");
7426 if (test_bit(Faulty, &rdev->flags)) {
7427 seq_printf(seq, "(F)");
7430 if (rdev->raid_disk < 0)
7431 seq_printf(seq, "(S)"); /* spare */
7432 if (test_bit(Replacement, &rdev->flags))
7433 seq_printf(seq, "(R)");
7434 sectors += rdev->sectors;
7438 if (!list_empty(&mddev->disks)) {
7440 seq_printf(seq, "\n %llu blocks",
7441 (unsigned long long)
7442 mddev->array_sectors / 2);
7444 seq_printf(seq, "\n %llu blocks",
7445 (unsigned long long)sectors / 2);
7447 if (mddev->persistent) {
7448 if (mddev->major_version != 0 ||
7449 mddev->minor_version != 90) {
7450 seq_printf(seq," super %d.%d",
7451 mddev->major_version,
7452 mddev->minor_version);
7454 } else if (mddev->external)
7455 seq_printf(seq, " super external:%s",
7456 mddev->metadata_type);
7458 seq_printf(seq, " super non-persistent");
7461 mddev->pers->status(seq, mddev);
7462 seq_printf(seq, "\n ");
7463 if (mddev->pers->sync_request) {
7464 if (status_resync(seq, mddev))
7465 seq_printf(seq, "\n ");
7468 seq_printf(seq, "\n ");
7470 bitmap_status(seq, mddev->bitmap);
7472 seq_printf(seq, "\n");
7474 spin_unlock(&mddev->lock);
7479 static const struct seq_operations md_seq_ops = {
7480 .start = md_seq_start,
7481 .next = md_seq_next,
7482 .stop = md_seq_stop,
7483 .show = md_seq_show,
7486 static int md_seq_open(struct inode *inode, struct file *file)
7488 struct seq_file *seq;
7491 error = seq_open(file, &md_seq_ops);
7495 seq = file->private_data;
7496 seq->poll_event = atomic_read(&md_event_count);
7500 static int md_unloading;
7501 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7503 struct seq_file *seq = filp->private_data;
7507 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7508 poll_wait(filp, &md_event_waiters, wait);
7510 /* always allow read */
7511 mask = POLLIN | POLLRDNORM;
7513 if (seq->poll_event != atomic_read(&md_event_count))
7514 mask |= POLLERR | POLLPRI;
7518 static const struct file_operations md_seq_fops = {
7519 .owner = THIS_MODULE,
7520 .open = md_seq_open,
7522 .llseek = seq_lseek,
7523 .release = seq_release_private,
7524 .poll = mdstat_poll,
7527 int register_md_personality(struct md_personality *p)
7529 printk(KERN_INFO "md: %s personality registered for level %d\n",
7531 spin_lock(&pers_lock);
7532 list_add_tail(&p->list, &pers_list);
7533 spin_unlock(&pers_lock);
7536 EXPORT_SYMBOL(register_md_personality);
7538 int unregister_md_personality(struct md_personality *p)
7540 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7541 spin_lock(&pers_lock);
7542 list_del_init(&p->list);
7543 spin_unlock(&pers_lock);
7546 EXPORT_SYMBOL(unregister_md_personality);
7548 int register_md_cluster_operations(struct md_cluster_operations *ops,
7549 struct module *module)
7552 spin_lock(&pers_lock);
7553 if (md_cluster_ops != NULL)
7556 md_cluster_ops = ops;
7557 md_cluster_mod = module;
7559 spin_unlock(&pers_lock);
7562 EXPORT_SYMBOL(register_md_cluster_operations);
7564 int unregister_md_cluster_operations(void)
7566 spin_lock(&pers_lock);
7567 md_cluster_ops = NULL;
7568 spin_unlock(&pers_lock);
7571 EXPORT_SYMBOL(unregister_md_cluster_operations);
7573 int md_setup_cluster(struct mddev *mddev, int nodes)
7575 if (!md_cluster_ops)
7576 request_module("md-cluster");
7577 spin_lock(&pers_lock);
7578 /* ensure module won't be unloaded */
7579 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7580 pr_err("can't find md-cluster module or get it's reference.\n");
7581 spin_unlock(&pers_lock);
7584 spin_unlock(&pers_lock);
7586 return md_cluster_ops->join(mddev, nodes);
7589 void md_cluster_stop(struct mddev *mddev)
7591 if (!md_cluster_ops)
7593 md_cluster_ops->leave(mddev);
7594 module_put(md_cluster_mod);
7597 static int is_mddev_idle(struct mddev *mddev, int init)
7599 struct md_rdev *rdev;
7605 rdev_for_each_rcu(rdev, mddev) {
7606 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7607 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7608 (int)part_stat_read(&disk->part0, sectors[1]) -
7609 atomic_read(&disk->sync_io);
7610 /* sync IO will cause sync_io to increase before the disk_stats
7611 * as sync_io is counted when a request starts, and
7612 * disk_stats is counted when it completes.
7613 * So resync activity will cause curr_events to be smaller than
7614 * when there was no such activity.
7615 * non-sync IO will cause disk_stat to increase without
7616 * increasing sync_io so curr_events will (eventually)
7617 * be larger than it was before. Once it becomes
7618 * substantially larger, the test below will cause
7619 * the array to appear non-idle, and resync will slow
7621 * If there is a lot of outstanding resync activity when
7622 * we set last_event to curr_events, then all that activity
7623 * completing might cause the array to appear non-idle
7624 * and resync will be slowed down even though there might
7625 * not have been non-resync activity. This will only
7626 * happen once though. 'last_events' will soon reflect
7627 * the state where there is little or no outstanding
7628 * resync requests, and further resync activity will
7629 * always make curr_events less than last_events.
7632 if (init || curr_events - rdev->last_events > 64) {
7633 rdev->last_events = curr_events;
7641 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7643 /* another "blocks" (512byte) blocks have been synced */
7644 atomic_sub(blocks, &mddev->recovery_active);
7645 wake_up(&mddev->recovery_wait);
7647 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7648 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7649 md_wakeup_thread(mddev->thread);
7650 // stop recovery, signal do_sync ....
7653 EXPORT_SYMBOL(md_done_sync);
7655 /* md_write_start(mddev, bi)
7656 * If we need to update some array metadata (e.g. 'active' flag
7657 * in superblock) before writing, schedule a superblock update
7658 * and wait for it to complete.
7660 void md_write_start(struct mddev *mddev, struct bio *bi)
7663 if (bio_data_dir(bi) != WRITE)
7666 BUG_ON(mddev->ro == 1);
7667 if (mddev->ro == 2) {
7668 /* need to switch to read/write */
7670 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7671 md_wakeup_thread(mddev->thread);
7672 md_wakeup_thread(mddev->sync_thread);
7675 atomic_inc(&mddev->writes_pending);
7676 if (mddev->safemode == 1)
7677 mddev->safemode = 0;
7678 if (mddev->in_sync) {
7679 spin_lock(&mddev->lock);
7680 if (mddev->in_sync) {
7682 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7683 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7684 md_wakeup_thread(mddev->thread);
7687 spin_unlock(&mddev->lock);
7690 sysfs_notify_dirent_safe(mddev->sysfs_state);
7691 wait_event(mddev->sb_wait,
7692 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7694 EXPORT_SYMBOL(md_write_start);
7696 void md_write_end(struct mddev *mddev)
7698 if (atomic_dec_and_test(&mddev->writes_pending)) {
7699 if (mddev->safemode == 2)
7700 md_wakeup_thread(mddev->thread);
7701 else if (mddev->safemode_delay)
7702 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7705 EXPORT_SYMBOL(md_write_end);
7707 /* md_allow_write(mddev)
7708 * Calling this ensures that the array is marked 'active' so that writes
7709 * may proceed without blocking. It is important to call this before
7710 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7711 * Must be called with mddev_lock held.
7713 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7714 * is dropped, so return -EAGAIN after notifying userspace.
7716 int md_allow_write(struct mddev *mddev)
7722 if (!mddev->pers->sync_request)
7725 spin_lock(&mddev->lock);
7726 if (mddev->in_sync) {
7728 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7729 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7730 if (mddev->safemode_delay &&
7731 mddev->safemode == 0)
7732 mddev->safemode = 1;
7733 spin_unlock(&mddev->lock);
7734 md_update_sb(mddev, 0);
7735 sysfs_notify_dirent_safe(mddev->sysfs_state);
7737 spin_unlock(&mddev->lock);
7739 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7744 EXPORT_SYMBOL_GPL(md_allow_write);
7746 #define SYNC_MARKS 10
7747 #define SYNC_MARK_STEP (3*HZ)
7748 #define UPDATE_FREQUENCY (5*60*HZ)
7749 void md_do_sync(struct md_thread *thread)
7751 struct mddev *mddev = thread->mddev;
7752 struct mddev *mddev2;
7753 unsigned int currspeed = 0,
7755 sector_t max_sectors,j, io_sectors, recovery_done;
7756 unsigned long mark[SYNC_MARKS];
7757 unsigned long update_time;
7758 sector_t mark_cnt[SYNC_MARKS];
7760 struct list_head *tmp;
7761 sector_t last_check;
7763 struct md_rdev *rdev;
7764 char *desc, *action = NULL;
7765 struct blk_plug plug;
7766 bool cluster_resync_finished = false;
7768 /* just incase thread restarts... */
7769 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7771 if (mddev->ro) {/* never try to sync a read-only array */
7772 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7776 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7777 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7778 desc = "data-check";
7780 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7781 desc = "requested-resync";
7785 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7790 mddev->last_sync_action = action ?: desc;
7792 /* we overload curr_resync somewhat here.
7793 * 0 == not engaged in resync at all
7794 * 2 == checking that there is no conflict with another sync
7795 * 1 == like 2, but have yielded to allow conflicting resync to
7797 * other == active in resync - this many blocks
7799 * Before starting a resync we must have set curr_resync to
7800 * 2, and then checked that every "conflicting" array has curr_resync
7801 * less than ours. When we find one that is the same or higher
7802 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7803 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7804 * This will mean we have to start checking from the beginning again.
7809 mddev->curr_resync = 2;
7812 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7814 for_each_mddev(mddev2, tmp) {
7815 if (mddev2 == mddev)
7817 if (!mddev->parallel_resync
7818 && mddev2->curr_resync
7819 && match_mddev_units(mddev, mddev2)) {
7821 if (mddev < mddev2 && mddev->curr_resync == 2) {
7822 /* arbitrarily yield */
7823 mddev->curr_resync = 1;
7824 wake_up(&resync_wait);
7826 if (mddev > mddev2 && mddev->curr_resync == 1)
7827 /* no need to wait here, we can wait the next
7828 * time 'round when curr_resync == 2
7831 /* We need to wait 'interruptible' so as not to
7832 * contribute to the load average, and not to
7833 * be caught by 'softlockup'
7835 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7836 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7837 mddev2->curr_resync >= mddev->curr_resync) {
7838 printk(KERN_INFO "md: delaying %s of %s"
7839 " until %s has finished (they"
7840 " share one or more physical units)\n",
7841 desc, mdname(mddev), mdname(mddev2));
7843 if (signal_pending(current))
7844 flush_signals(current);
7846 finish_wait(&resync_wait, &wq);
7849 finish_wait(&resync_wait, &wq);
7852 } while (mddev->curr_resync < 2);
7855 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7856 /* resync follows the size requested by the personality,
7857 * which defaults to physical size, but can be virtual size
7859 max_sectors = mddev->resync_max_sectors;
7860 atomic64_set(&mddev->resync_mismatches, 0);
7861 /* we don't use the checkpoint if there's a bitmap */
7862 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7863 j = mddev->resync_min;
7864 else if (!mddev->bitmap)
7865 j = mddev->recovery_cp;
7867 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7868 max_sectors = mddev->resync_max_sectors;
7870 /* recovery follows the physical size of devices */
7871 max_sectors = mddev->dev_sectors;
7874 rdev_for_each_rcu(rdev, mddev)
7875 if (rdev->raid_disk >= 0 &&
7876 !test_bit(Journal, &rdev->flags) &&
7877 !test_bit(Faulty, &rdev->flags) &&
7878 !test_bit(In_sync, &rdev->flags) &&
7879 rdev->recovery_offset < j)
7880 j = rdev->recovery_offset;
7883 /* If there is a bitmap, we need to make sure all
7884 * writes that started before we added a spare
7885 * complete before we start doing a recovery.
7886 * Otherwise the write might complete and (via
7887 * bitmap_endwrite) set a bit in the bitmap after the
7888 * recovery has checked that bit and skipped that
7891 if (mddev->bitmap) {
7892 mddev->pers->quiesce(mddev, 1);
7893 mddev->pers->quiesce(mddev, 0);
7897 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7898 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7899 " %d KB/sec/disk.\n", speed_min(mddev));
7900 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7901 "(but not more than %d KB/sec) for %s.\n",
7902 speed_max(mddev), desc);
7904 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7907 for (m = 0; m < SYNC_MARKS; m++) {
7909 mark_cnt[m] = io_sectors;
7912 mddev->resync_mark = mark[last_mark];
7913 mddev->resync_mark_cnt = mark_cnt[last_mark];
7916 * Tune reconstruction:
7918 window = 32*(PAGE_SIZE/512);
7919 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7920 window/2, (unsigned long long)max_sectors/2);
7922 atomic_set(&mddev->recovery_active, 0);
7927 "md: resuming %s of %s from checkpoint.\n",
7928 desc, mdname(mddev));
7929 mddev->curr_resync = j;
7931 mddev->curr_resync = 3; /* no longer delayed */
7932 mddev->curr_resync_completed = j;
7933 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7934 md_new_event(mddev);
7935 update_time = jiffies;
7937 blk_start_plug(&plug);
7938 while (j < max_sectors) {
7943 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7944 ((mddev->curr_resync > mddev->curr_resync_completed &&
7945 (mddev->curr_resync - mddev->curr_resync_completed)
7946 > (max_sectors >> 4)) ||
7947 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7948 (j - mddev->curr_resync_completed)*2
7949 >= mddev->resync_max - mddev->curr_resync_completed ||
7950 mddev->curr_resync_completed > mddev->resync_max
7952 /* time to update curr_resync_completed */
7953 wait_event(mddev->recovery_wait,
7954 atomic_read(&mddev->recovery_active) == 0);
7955 mddev->curr_resync_completed = j;
7956 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7957 j > mddev->recovery_cp)
7958 mddev->recovery_cp = j;
7959 update_time = jiffies;
7960 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7961 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7964 while (j >= mddev->resync_max &&
7965 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7966 /* As this condition is controlled by user-space,
7967 * we can block indefinitely, so use '_interruptible'
7968 * to avoid triggering warnings.
7970 flush_signals(current); /* just in case */
7971 wait_event_interruptible(mddev->recovery_wait,
7972 mddev->resync_max > j
7973 || test_bit(MD_RECOVERY_INTR,
7977 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7980 sectors = mddev->pers->sync_request(mddev, j, &skipped);
7982 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7986 if (!skipped) { /* actual IO requested */
7987 io_sectors += sectors;
7988 atomic_add(sectors, &mddev->recovery_active);
7991 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7995 if (j > max_sectors)
7996 /* when skipping, extra large numbers can be returned. */
7999 mddev->curr_resync = j;
8000 mddev->curr_mark_cnt = io_sectors;
8001 if (last_check == 0)
8002 /* this is the earliest that rebuild will be
8003 * visible in /proc/mdstat
8005 md_new_event(mddev);
8007 if (last_check + window > io_sectors || j == max_sectors)
8010 last_check = io_sectors;
8012 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
8014 int next = (last_mark+1) % SYNC_MARKS;
8016 mddev->resync_mark = mark[next];
8017 mddev->resync_mark_cnt = mark_cnt[next];
8018 mark[next] = jiffies;
8019 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
8023 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8027 * this loop exits only if either when we are slower than
8028 * the 'hard' speed limit, or the system was IO-idle for
8030 * the system might be non-idle CPU-wise, but we only care
8031 * about not overloading the IO subsystem. (things like an
8032 * e2fsck being done on the RAID array should execute fast)
8036 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
8037 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
8038 /((jiffies-mddev->resync_mark)/HZ +1) +1;
8040 if (currspeed > speed_min(mddev)) {
8041 if (currspeed > speed_max(mddev)) {
8045 if (!is_mddev_idle(mddev, 0)) {
8047 * Give other IO more of a chance.
8048 * The faster the devices, the less we wait.
8050 wait_event(mddev->recovery_wait,
8051 !atomic_read(&mddev->recovery_active));
8055 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
8056 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
8057 ? "interrupted" : "done");
8059 * this also signals 'finished resyncing' to md_stop
8061 blk_finish_plug(&plug);
8062 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
8064 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8065 !test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8066 mddev->curr_resync > 2) {
8067 mddev->curr_resync_completed = mddev->curr_resync;
8068 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
8070 /* tell personality and other nodes that we are finished */
8071 if (mddev_is_clustered(mddev)) {
8072 md_cluster_ops->resync_finish(mddev);
8073 cluster_resync_finished = true;
8075 mddev->pers->sync_request(mddev, max_sectors, &skipped);
8077 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
8078 mddev->curr_resync > 2) {
8079 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
8080 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8081 if (mddev->curr_resync >= mddev->recovery_cp) {
8083 "md: checkpointing %s of %s.\n",
8084 desc, mdname(mddev));
8085 if (test_bit(MD_RECOVERY_ERROR,
8087 mddev->recovery_cp =
8088 mddev->curr_resync_completed;
8090 mddev->recovery_cp =
8094 mddev->recovery_cp = MaxSector;
8096 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
8097 mddev->curr_resync = MaxSector;
8099 rdev_for_each_rcu(rdev, mddev)
8100 if (rdev->raid_disk >= 0 &&
8101 mddev->delta_disks >= 0 &&
8102 !test_bit(Journal, &rdev->flags) &&
8103 !test_bit(Faulty, &rdev->flags) &&
8104 !test_bit(In_sync, &rdev->flags) &&
8105 rdev->recovery_offset < mddev->curr_resync)
8106 rdev->recovery_offset = mddev->curr_resync;
8111 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8113 if (mddev_is_clustered(mddev) &&
8114 test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8115 !cluster_resync_finished)
8116 md_cluster_ops->resync_finish(mddev);
8118 spin_lock(&mddev->lock);
8119 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
8120 /* We completed so min/max setting can be forgotten if used. */
8121 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8122 mddev->resync_min = 0;
8123 mddev->resync_max = MaxSector;
8124 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
8125 mddev->resync_min = mddev->curr_resync_completed;
8126 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
8127 mddev->curr_resync = 0;
8128 spin_unlock(&mddev->lock);
8130 wake_up(&resync_wait);
8131 md_wakeup_thread(mddev->thread);
8134 EXPORT_SYMBOL_GPL(md_do_sync);
8136 static int remove_and_add_spares(struct mddev *mddev,
8137 struct md_rdev *this)
8139 struct md_rdev *rdev;
8143 rdev_for_each(rdev, mddev)
8144 if ((this == NULL || rdev == this) &&
8145 rdev->raid_disk >= 0 &&
8146 !test_bit(Blocked, &rdev->flags) &&
8147 (test_bit(Faulty, &rdev->flags) ||
8148 (!test_bit(In_sync, &rdev->flags) &&
8149 !test_bit(Journal, &rdev->flags))) &&
8150 atomic_read(&rdev->nr_pending)==0) {
8151 if (mddev->pers->hot_remove_disk(
8152 mddev, rdev) == 0) {
8153 sysfs_unlink_rdev(mddev, rdev);
8154 rdev->raid_disk = -1;
8158 if (removed && mddev->kobj.sd)
8159 sysfs_notify(&mddev->kobj, NULL, "degraded");
8161 if (this && removed)
8164 rdev_for_each(rdev, mddev) {
8165 if (this && this != rdev)
8167 if (test_bit(Candidate, &rdev->flags))
8169 if (rdev->raid_disk >= 0 &&
8170 !test_bit(In_sync, &rdev->flags) &&
8171 !test_bit(Journal, &rdev->flags) &&
8172 !test_bit(Faulty, &rdev->flags))
8174 if (rdev->raid_disk >= 0)
8176 if (test_bit(Faulty, &rdev->flags))
8178 if (test_bit(Journal, &rdev->flags))
8181 ! (rdev->saved_raid_disk >= 0 &&
8182 !test_bit(Bitmap_sync, &rdev->flags)))
8185 rdev->recovery_offset = 0;
8187 hot_add_disk(mddev, rdev) == 0) {
8188 if (sysfs_link_rdev(mddev, rdev))
8189 /* failure here is OK */;
8191 md_new_event(mddev);
8192 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8197 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8201 static void md_start_sync(struct work_struct *ws)
8203 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8206 if (mddev_is_clustered(mddev)) {
8207 ret = md_cluster_ops->resync_start(mddev);
8209 mddev->sync_thread = NULL;
8214 mddev->sync_thread = md_register_thread(md_do_sync,
8218 if (!mddev->sync_thread) {
8219 if (!(mddev_is_clustered(mddev) && ret == -EAGAIN))
8220 printk(KERN_ERR "%s: could not start resync"
8223 /* leave the spares where they are, it shouldn't hurt */
8224 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8225 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8226 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8227 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8228 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8229 wake_up(&resync_wait);
8230 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8232 if (mddev->sysfs_action)
8233 sysfs_notify_dirent_safe(mddev->sysfs_action);
8235 md_wakeup_thread(mddev->sync_thread);
8236 sysfs_notify_dirent_safe(mddev->sysfs_action);
8237 md_new_event(mddev);
8241 * This routine is regularly called by all per-raid-array threads to
8242 * deal with generic issues like resync and super-block update.
8243 * Raid personalities that don't have a thread (linear/raid0) do not
8244 * need this as they never do any recovery or update the superblock.
8246 * It does not do any resync itself, but rather "forks" off other threads
8247 * to do that as needed.
8248 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8249 * "->recovery" and create a thread at ->sync_thread.
8250 * When the thread finishes it sets MD_RECOVERY_DONE
8251 * and wakeups up this thread which will reap the thread and finish up.
8252 * This thread also removes any faulty devices (with nr_pending == 0).
8254 * The overall approach is:
8255 * 1/ if the superblock needs updating, update it.
8256 * 2/ If a recovery thread is running, don't do anything else.
8257 * 3/ If recovery has finished, clean up, possibly marking spares active.
8258 * 4/ If there are any faulty devices, remove them.
8259 * 5/ If array is degraded, try to add spares devices
8260 * 6/ If array has spares or is not in-sync, start a resync thread.
8262 void md_check_recovery(struct mddev *mddev)
8264 if (mddev->suspended)
8268 bitmap_daemon_work(mddev);
8270 if (signal_pending(current)) {
8271 if (mddev->pers->sync_request && !mddev->external) {
8272 printk(KERN_INFO "md: %s in immediate safe mode\n",
8274 mddev->safemode = 2;
8276 flush_signals(current);
8279 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8282 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8283 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8284 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8285 (mddev->external == 0 && mddev->safemode == 1) ||
8286 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8287 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8291 if (mddev_trylock(mddev)) {
8295 struct md_rdev *rdev;
8296 if (!mddev->external && mddev->in_sync)
8297 /* 'Blocked' flag not needed as failed devices
8298 * will be recorded if array switched to read/write.
8299 * Leaving it set will prevent the device
8300 * from being removed.
8302 rdev_for_each(rdev, mddev)
8303 clear_bit(Blocked, &rdev->flags);
8304 /* On a read-only array we can:
8305 * - remove failed devices
8306 * - add already-in_sync devices if the array itself
8308 * As we only add devices that are already in-sync,
8309 * we can activate the spares immediately.
8311 remove_and_add_spares(mddev, NULL);
8312 /* There is no thread, but we need to call
8313 * ->spare_active and clear saved_raid_disk
8315 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8316 md_reap_sync_thread(mddev);
8317 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8318 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8319 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
8323 if (!mddev->external) {
8325 spin_lock(&mddev->lock);
8326 if (mddev->safemode &&
8327 !atomic_read(&mddev->writes_pending) &&
8329 mddev->recovery_cp == MaxSector) {
8332 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8334 if (mddev->safemode == 1)
8335 mddev->safemode = 0;
8336 spin_unlock(&mddev->lock);
8338 sysfs_notify_dirent_safe(mddev->sysfs_state);
8341 if (mddev->flags & MD_UPDATE_SB_FLAGS)
8342 md_update_sb(mddev, 0);
8344 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8345 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8346 /* resync/recovery still happening */
8347 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8350 if (mddev->sync_thread) {
8351 md_reap_sync_thread(mddev);
8354 /* Set RUNNING before clearing NEEDED to avoid
8355 * any transients in the value of "sync_action".
8357 mddev->curr_resync_completed = 0;
8358 spin_lock(&mddev->lock);
8359 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8360 spin_unlock(&mddev->lock);
8361 /* Clear some bits that don't mean anything, but
8364 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8365 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8367 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8368 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8370 /* no recovery is running.
8371 * remove any failed drives, then
8372 * add spares if possible.
8373 * Spares are also removed and re-added, to allow
8374 * the personality to fail the re-add.
8377 if (mddev->reshape_position != MaxSector) {
8378 if (mddev->pers->check_reshape == NULL ||
8379 mddev->pers->check_reshape(mddev) != 0)
8380 /* Cannot proceed */
8382 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8383 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8384 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8385 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8386 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8387 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8388 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8389 } else if (mddev->recovery_cp < MaxSector) {
8390 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8391 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8392 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8393 /* nothing to be done ... */
8396 if (mddev->pers->sync_request) {
8398 /* We are adding a device or devices to an array
8399 * which has the bitmap stored on all devices.
8400 * So make sure all bitmap pages get written
8402 bitmap_write_all(mddev->bitmap);
8404 INIT_WORK(&mddev->del_work, md_start_sync);
8405 queue_work(md_misc_wq, &mddev->del_work);
8409 if (!mddev->sync_thread) {
8410 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8411 wake_up(&resync_wait);
8412 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8414 if (mddev->sysfs_action)
8415 sysfs_notify_dirent_safe(mddev->sysfs_action);
8418 wake_up(&mddev->sb_wait);
8419 mddev_unlock(mddev);
8422 EXPORT_SYMBOL(md_check_recovery);
8424 void md_reap_sync_thread(struct mddev *mddev)
8426 struct md_rdev *rdev;
8428 /* resync has finished, collect result */
8429 md_unregister_thread(&mddev->sync_thread);
8430 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8431 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8433 /* activate any spares */
8434 if (mddev->pers->spare_active(mddev)) {
8435 sysfs_notify(&mddev->kobj, NULL,
8437 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8440 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8441 mddev->pers->finish_reshape)
8442 mddev->pers->finish_reshape(mddev);
8444 /* If array is no-longer degraded, then any saved_raid_disk
8445 * information must be scrapped.
8447 if (!mddev->degraded)
8448 rdev_for_each(rdev, mddev)
8449 rdev->saved_raid_disk = -1;
8451 md_update_sb(mddev, 1);
8452 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8453 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8454 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8455 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8456 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8457 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8458 wake_up(&resync_wait);
8459 /* flag recovery needed just to double check */
8460 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8461 sysfs_notify_dirent_safe(mddev->sysfs_action);
8462 md_new_event(mddev);
8463 if (mddev->event_work.func)
8464 queue_work(md_misc_wq, &mddev->event_work);
8466 EXPORT_SYMBOL(md_reap_sync_thread);
8468 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8470 sysfs_notify_dirent_safe(rdev->sysfs_state);
8471 wait_event_timeout(rdev->blocked_wait,
8472 !test_bit(Blocked, &rdev->flags) &&
8473 !test_bit(BlockedBadBlocks, &rdev->flags),
8474 msecs_to_jiffies(5000));
8475 rdev_dec_pending(rdev, mddev);
8477 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8479 void md_finish_reshape(struct mddev *mddev)
8481 /* called be personality module when reshape completes. */
8482 struct md_rdev *rdev;
8484 rdev_for_each(rdev, mddev) {
8485 if (rdev->data_offset > rdev->new_data_offset)
8486 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8488 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8489 rdev->data_offset = rdev->new_data_offset;
8492 EXPORT_SYMBOL(md_finish_reshape);
8494 /* Bad block management.
8495 * We can record which blocks on each device are 'bad' and so just
8496 * fail those blocks, or that stripe, rather than the whole device.
8497 * Entries in the bad-block table are 64bits wide. This comprises:
8498 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8499 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8500 * A 'shift' can be set so that larger blocks are tracked and
8501 * consequently larger devices can be covered.
8502 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8504 * Locking of the bad-block table uses a seqlock so md_is_badblock
8505 * might need to retry if it is very unlucky.
8506 * We will sometimes want to check for bad blocks in a bi_end_io function,
8507 * so we use the write_seqlock_irq variant.
8509 * When looking for a bad block we specify a range and want to
8510 * know if any block in the range is bad. So we binary-search
8511 * to the last range that starts at-or-before the given endpoint,
8512 * (or "before the sector after the target range")
8513 * then see if it ends after the given start.
8515 * 0 if there are no known bad blocks in the range
8516 * 1 if there are known bad block which are all acknowledged
8517 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8518 * plus the start/length of the first bad section we overlap.
8520 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8521 sector_t *first_bad, int *bad_sectors)
8527 sector_t target = s + sectors;
8530 if (bb->shift > 0) {
8531 /* round the start down, and the end up */
8533 target += (1<<bb->shift) - 1;
8534 target >>= bb->shift;
8535 sectors = target - s;
8537 /* 'target' is now the first block after the bad range */
8540 seq = read_seqbegin(&bb->lock);
8545 /* Binary search between lo and hi for 'target'
8546 * i.e. for the last range that starts before 'target'
8548 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8549 * are known not to be the last range before target.
8550 * VARIANT: hi-lo is the number of possible
8551 * ranges, and decreases until it reaches 1
8553 while (hi - lo > 1) {
8554 int mid = (lo + hi) / 2;
8555 sector_t a = BB_OFFSET(p[mid]);
8557 /* This could still be the one, earlier ranges
8561 /* This and later ranges are definitely out. */
8564 /* 'lo' might be the last that started before target, but 'hi' isn't */
8566 /* need to check all range that end after 's' to see if
8567 * any are unacknowledged.
8570 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8571 if (BB_OFFSET(p[lo]) < target) {
8572 /* starts before the end, and finishes after
8573 * the start, so they must overlap
8575 if (rv != -1 && BB_ACK(p[lo]))
8579 *first_bad = BB_OFFSET(p[lo]);
8580 *bad_sectors = BB_LEN(p[lo]);
8586 if (read_seqretry(&bb->lock, seq))
8591 EXPORT_SYMBOL_GPL(md_is_badblock);
8594 * Add a range of bad blocks to the table.
8595 * This might extend the table, or might contract it
8596 * if two adjacent ranges can be merged.
8597 * We binary-search to find the 'insertion' point, then
8598 * decide how best to handle it.
8600 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8606 unsigned long flags;
8609 /* badblocks are disabled */
8613 /* round the start down, and the end up */
8614 sector_t next = s + sectors;
8616 next += (1<<bb->shift) - 1;
8621 write_seqlock_irqsave(&bb->lock, flags);
8626 /* Find the last range that starts at-or-before 's' */
8627 while (hi - lo > 1) {
8628 int mid = (lo + hi) / 2;
8629 sector_t a = BB_OFFSET(p[mid]);
8635 if (hi > lo && BB_OFFSET(p[lo]) > s)
8639 /* we found a range that might merge with the start
8642 sector_t a = BB_OFFSET(p[lo]);
8643 sector_t e = a + BB_LEN(p[lo]);
8644 int ack = BB_ACK(p[lo]);
8646 /* Yes, we can merge with a previous range */
8647 if (s == a && s + sectors >= e)
8648 /* new range covers old */
8651 ack = ack && acknowledged;
8653 if (e < s + sectors)
8655 if (e - a <= BB_MAX_LEN) {
8656 p[lo] = BB_MAKE(a, e-a, ack);
8659 /* does not all fit in one range,
8660 * make p[lo] maximal
8662 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8663 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8669 if (sectors && hi < bb->count) {
8670 /* 'hi' points to the first range that starts after 's'.
8671 * Maybe we can merge with the start of that range */
8672 sector_t a = BB_OFFSET(p[hi]);
8673 sector_t e = a + BB_LEN(p[hi]);
8674 int ack = BB_ACK(p[hi]);
8675 if (a <= s + sectors) {
8676 /* merging is possible */
8677 if (e <= s + sectors) {
8682 ack = ack && acknowledged;
8685 if (e - a <= BB_MAX_LEN) {
8686 p[hi] = BB_MAKE(a, e-a, ack);
8689 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8697 if (sectors == 0 && hi < bb->count) {
8698 /* we might be able to combine lo and hi */
8699 /* Note: 's' is at the end of 'lo' */
8700 sector_t a = BB_OFFSET(p[hi]);
8701 int lolen = BB_LEN(p[lo]);
8702 int hilen = BB_LEN(p[hi]);
8703 int newlen = lolen + hilen - (s - a);
8704 if (s >= a && newlen < BB_MAX_LEN) {
8705 /* yes, we can combine them */
8706 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8707 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8708 memmove(p + hi, p + hi + 1,
8709 (bb->count - hi - 1) * 8);
8714 /* didn't merge (it all).
8715 * Need to add a range just before 'hi' */
8716 if (bb->count >= MD_MAX_BADBLOCKS) {
8717 /* No room for more */
8721 int this_sectors = sectors;
8722 memmove(p + hi + 1, p + hi,
8723 (bb->count - hi) * 8);
8726 if (this_sectors > BB_MAX_LEN)
8727 this_sectors = BB_MAX_LEN;
8728 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8729 sectors -= this_sectors;
8736 bb->unacked_exist = 1;
8737 write_sequnlock_irqrestore(&bb->lock, flags);
8742 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8747 s += rdev->new_data_offset;
8749 s += rdev->data_offset;
8750 rv = md_set_badblocks(&rdev->badblocks,
8753 /* Make sure they get written out promptly */
8754 sysfs_notify_dirent_safe(rdev->sysfs_state);
8755 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8756 set_bit(MD_CHANGE_PENDING, &rdev->mddev->flags);
8757 md_wakeup_thread(rdev->mddev->thread);
8761 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8764 * Remove a range of bad blocks from the table.
8765 * This may involve extending the table if we spilt a region,
8766 * but it must not fail. So if the table becomes full, we just
8767 * drop the remove request.
8769 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8773 sector_t target = s + sectors;
8776 if (bb->shift > 0) {
8777 /* When clearing we round the start up and the end down.
8778 * This should not matter as the shift should align with
8779 * the block size and no rounding should ever be needed.
8780 * However it is better the think a block is bad when it
8781 * isn't than to think a block is not bad when it is.
8783 s += (1<<bb->shift) - 1;
8785 target >>= bb->shift;
8786 sectors = target - s;
8789 write_seqlock_irq(&bb->lock);
8794 /* Find the last range that starts before 'target' */
8795 while (hi - lo > 1) {
8796 int mid = (lo + hi) / 2;
8797 sector_t a = BB_OFFSET(p[mid]);
8804 /* p[lo] is the last range that could overlap the
8805 * current range. Earlier ranges could also overlap,
8806 * but only this one can overlap the end of the range.
8808 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8809 /* Partial overlap, leave the tail of this range */
8810 int ack = BB_ACK(p[lo]);
8811 sector_t a = BB_OFFSET(p[lo]);
8812 sector_t end = a + BB_LEN(p[lo]);
8815 /* we need to split this range */
8816 if (bb->count >= MD_MAX_BADBLOCKS) {
8820 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8822 p[lo] = BB_MAKE(a, s-a, ack);
8825 p[lo] = BB_MAKE(target, end - target, ack);
8826 /* there is no longer an overlap */
8831 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8832 /* This range does overlap */
8833 if (BB_OFFSET(p[lo]) < s) {
8834 /* Keep the early parts of this range. */
8835 int ack = BB_ACK(p[lo]);
8836 sector_t start = BB_OFFSET(p[lo]);
8837 p[lo] = BB_MAKE(start, s - start, ack);
8838 /* now low doesn't overlap, so.. */
8843 /* 'lo' is strictly before, 'hi' is strictly after,
8844 * anything between needs to be discarded
8847 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8848 bb->count -= (hi - lo - 1);
8854 write_sequnlock_irq(&bb->lock);
8858 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8862 s += rdev->new_data_offset;
8864 s += rdev->data_offset;
8865 return md_clear_badblocks(&rdev->badblocks,
8868 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8871 * Acknowledge all bad blocks in a list.
8872 * This only succeeds if ->changed is clear. It is used by
8873 * in-kernel metadata updates
8875 void md_ack_all_badblocks(struct badblocks *bb)
8877 if (bb->page == NULL || bb->changed)
8878 /* no point even trying */
8880 write_seqlock_irq(&bb->lock);
8882 if (bb->changed == 0 && bb->unacked_exist) {
8885 for (i = 0; i < bb->count ; i++) {
8886 if (!BB_ACK(p[i])) {
8887 sector_t start = BB_OFFSET(p[i]);
8888 int len = BB_LEN(p[i]);
8889 p[i] = BB_MAKE(start, len, 1);
8892 bb->unacked_exist = 0;
8894 write_sequnlock_irq(&bb->lock);
8896 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8898 /* sysfs access to bad-blocks list.
8899 * We present two files.
8900 * 'bad-blocks' lists sector numbers and lengths of ranges that
8901 * are recorded as bad. The list is truncated to fit within
8902 * the one-page limit of sysfs.
8903 * Writing "sector length" to this file adds an acknowledged
8905 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8906 * been acknowledged. Writing to this file adds bad blocks
8907 * without acknowledging them. This is largely for testing.
8911 badblocks_show(struct badblocks *bb, char *page, int unack)
8922 seq = read_seqbegin(&bb->lock);
8927 while (len < PAGE_SIZE && i < bb->count) {
8928 sector_t s = BB_OFFSET(p[i]);
8929 unsigned int length = BB_LEN(p[i]);
8930 int ack = BB_ACK(p[i]);
8936 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8937 (unsigned long long)s << bb->shift,
8938 length << bb->shift);
8940 if (unack && len == 0)
8941 bb->unacked_exist = 0;
8943 if (read_seqretry(&bb->lock, seq))
8952 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8954 unsigned long long sector;
8958 /* Allow clearing via sysfs *only* for testing/debugging.
8959 * Normally only a successful write may clear a badblock
8962 if (page[0] == '-') {
8966 #endif /* DO_DEBUG */
8968 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8970 if (newline != '\n')
8982 md_clear_badblocks(bb, sector, length);
8985 #endif /* DO_DEBUG */
8986 if (md_set_badblocks(bb, sector, length, !unack))
8992 static int md_notify_reboot(struct notifier_block *this,
8993 unsigned long code, void *x)
8995 struct list_head *tmp;
8996 struct mddev *mddev;
8999 for_each_mddev(mddev, tmp) {
9000 if (mddev_trylock(mddev)) {
9002 __md_stop_writes(mddev);
9003 if (mddev->persistent)
9004 mddev->safemode = 2;
9005 mddev_unlock(mddev);
9010 * certain more exotic SCSI devices are known to be
9011 * volatile wrt too early system reboots. While the
9012 * right place to handle this issue is the given
9013 * driver, we do want to have a safe RAID driver ...
9021 static struct notifier_block md_notifier = {
9022 .notifier_call = md_notify_reboot,
9024 .priority = INT_MAX, /* before any real devices */
9027 static void md_geninit(void)
9029 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
9031 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
9034 static int __init md_init(void)
9038 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
9042 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
9046 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
9049 if ((ret = register_blkdev(0, "mdp")) < 0)
9053 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
9054 md_probe, NULL, NULL);
9055 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
9056 md_probe, NULL, NULL);
9058 register_reboot_notifier(&md_notifier);
9059 raid_table_header = register_sysctl_table(raid_root_table);
9065 unregister_blkdev(MD_MAJOR, "md");
9067 destroy_workqueue(md_misc_wq);
9069 destroy_workqueue(md_wq);
9074 static void check_sb_changes(struct mddev *mddev, struct md_rdev *rdev)
9076 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
9077 struct md_rdev *rdev2;
9079 char b[BDEVNAME_SIZE];
9081 /* Check for change of roles in the active devices */
9082 rdev_for_each(rdev2, mddev) {
9083 if (test_bit(Faulty, &rdev2->flags))
9086 /* Check if the roles changed */
9087 role = le16_to_cpu(sb->dev_roles[rdev2->desc_nr]);
9089 if (test_bit(Candidate, &rdev2->flags)) {
9090 if (role == 0xfffe) {
9091 pr_info("md: Removing Candidate device %s because add failed\n", bdevname(rdev2->bdev,b));
9092 md_kick_rdev_from_array(rdev2);
9096 clear_bit(Candidate, &rdev2->flags);
9099 if (role != rdev2->raid_disk) {
9101 if (rdev2->raid_disk == -1 && role != 0xffff) {
9102 rdev2->saved_raid_disk = role;
9103 ret = remove_and_add_spares(mddev, rdev2);
9104 pr_info("Activated spare: %s\n",
9105 bdevname(rdev2->bdev,b));
9109 * We just want to do the minimum to mark the disk
9110 * as faulty. The recovery is performed by the
9111 * one who initiated the error.
9113 if ((role == 0xfffe) || (role == 0xfffd)) {
9114 md_error(mddev, rdev2);
9115 clear_bit(Blocked, &rdev2->flags);
9120 if (mddev->raid_disks != le32_to_cpu(sb->raid_disks))
9121 update_raid_disks(mddev, le32_to_cpu(sb->raid_disks));
9123 /* Finally set the event to be up to date */
9124 mddev->events = le64_to_cpu(sb->events);
9127 static int read_rdev(struct mddev *mddev, struct md_rdev *rdev)
9130 struct page *swapout = rdev->sb_page;
9131 struct mdp_superblock_1 *sb;
9133 /* Store the sb page of the rdev in the swapout temporary
9134 * variable in case we err in the future
9136 rdev->sb_page = NULL;
9137 alloc_disk_sb(rdev);
9138 ClearPageUptodate(rdev->sb_page);
9139 rdev->sb_loaded = 0;
9140 err = super_types[mddev->major_version].load_super(rdev, NULL, mddev->minor_version);
9143 pr_warn("%s: %d Could not reload rdev(%d) err: %d. Restoring old values\n",
9144 __func__, __LINE__, rdev->desc_nr, err);
9145 put_page(rdev->sb_page);
9146 rdev->sb_page = swapout;
9147 rdev->sb_loaded = 1;
9151 sb = page_address(rdev->sb_page);
9152 /* Read the offset unconditionally, even if MD_FEATURE_RECOVERY_OFFSET
9156 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RECOVERY_OFFSET))
9157 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
9159 /* The other node finished recovery, call spare_active to set
9160 * device In_sync and mddev->degraded
9162 if (rdev->recovery_offset == MaxSector &&
9163 !test_bit(In_sync, &rdev->flags) &&
9164 mddev->pers->spare_active(mddev))
9165 sysfs_notify(&mddev->kobj, NULL, "degraded");
9171 void md_reload_sb(struct mddev *mddev, int nr)
9173 struct md_rdev *rdev;
9177 rdev_for_each_rcu(rdev, mddev) {
9178 if (rdev->desc_nr == nr)
9182 if (!rdev || rdev->desc_nr != nr) {
9183 pr_warn("%s: %d Could not find rdev with nr %d\n", __func__, __LINE__, nr);
9187 err = read_rdev(mddev, rdev);
9191 check_sb_changes(mddev, rdev);
9193 /* Read all rdev's to update recovery_offset */
9194 rdev_for_each_rcu(rdev, mddev)
9195 read_rdev(mddev, rdev);
9197 EXPORT_SYMBOL(md_reload_sb);
9202 * Searches all registered partitions for autorun RAID arrays
9206 static LIST_HEAD(all_detected_devices);
9207 struct detected_devices_node {
9208 struct list_head list;
9212 void md_autodetect_dev(dev_t dev)
9214 struct detected_devices_node *node_detected_dev;
9216 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
9217 if (node_detected_dev) {
9218 node_detected_dev->dev = dev;
9219 list_add_tail(&node_detected_dev->list, &all_detected_devices);
9221 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
9222 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
9226 static void autostart_arrays(int part)
9228 struct md_rdev *rdev;
9229 struct detected_devices_node *node_detected_dev;
9231 int i_scanned, i_passed;
9236 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
9238 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
9240 node_detected_dev = list_entry(all_detected_devices.next,
9241 struct detected_devices_node, list);
9242 list_del(&node_detected_dev->list);
9243 dev = node_detected_dev->dev;
9244 kfree(node_detected_dev);
9245 rdev = md_import_device(dev,0, 90);
9249 if (test_bit(Faulty, &rdev->flags))
9252 set_bit(AutoDetected, &rdev->flags);
9253 list_add(&rdev->same_set, &pending_raid_disks);
9257 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
9258 i_scanned, i_passed);
9260 autorun_devices(part);
9263 #endif /* !MODULE */
9265 static __exit void md_exit(void)
9267 struct mddev *mddev;
9268 struct list_head *tmp;
9271 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
9272 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
9274 unregister_blkdev(MD_MAJOR,"md");
9275 unregister_blkdev(mdp_major, "mdp");
9276 unregister_reboot_notifier(&md_notifier);
9277 unregister_sysctl_table(raid_table_header);
9279 /* We cannot unload the modules while some process is
9280 * waiting for us in select() or poll() - wake them up
9283 while (waitqueue_active(&md_event_waiters)) {
9284 /* not safe to leave yet */
9285 wake_up(&md_event_waiters);
9289 remove_proc_entry("mdstat", NULL);
9291 for_each_mddev(mddev, tmp) {
9292 export_array(mddev);
9293 mddev->hold_active = 0;
9295 destroy_workqueue(md_misc_wq);
9296 destroy_workqueue(md_wq);
9299 subsys_initcall(md_init);
9300 module_exit(md_exit)
9302 static int get_ro(char *buffer, struct kernel_param *kp)
9304 return sprintf(buffer, "%d", start_readonly);
9306 static int set_ro(const char *val, struct kernel_param *kp)
9308 return kstrtouint(val, 10, (unsigned int *)&start_readonly);
9311 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9312 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9313 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9315 MODULE_LICENSE("GPL");
9316 MODULE_DESCRIPTION("MD RAID framework");
9318 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);