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 void 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;
259 if (mddev == NULL || mddev->pers == NULL
264 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
265 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
268 smp_rmb(); /* Ensure implications of 'active' are visible */
270 if (mddev->suspended) {
273 prepare_to_wait(&mddev->sb_wait, &__wait,
274 TASK_UNINTERRUPTIBLE);
275 if (!mddev->suspended)
281 finish_wait(&mddev->sb_wait, &__wait);
283 atomic_inc(&mddev->active_io);
287 * save the sectors now since our bio can
288 * go away inside make_request
290 sectors = bio_sectors(bio);
291 mddev->pers->make_request(mddev, bio);
293 generic_start_io_acct(rw, sectors, &mddev->gendisk->part0);
295 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
296 wake_up(&mddev->sb_wait);
299 /* mddev_suspend makes sure no new requests are submitted
300 * to the device, and that any requests that have been submitted
301 * are completely handled.
302 * Once mddev_detach() is called and completes, the module will be
305 void mddev_suspend(struct mddev *mddev)
307 BUG_ON(mddev->suspended);
308 mddev->suspended = 1;
310 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
311 mddev->pers->quiesce(mddev, 1);
313 del_timer_sync(&mddev->safemode_timer);
315 EXPORT_SYMBOL_GPL(mddev_suspend);
317 void mddev_resume(struct mddev *mddev)
319 mddev->suspended = 0;
320 wake_up(&mddev->sb_wait);
321 mddev->pers->quiesce(mddev, 0);
323 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
324 md_wakeup_thread(mddev->thread);
325 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
327 EXPORT_SYMBOL_GPL(mddev_resume);
329 int mddev_congested(struct mddev *mddev, int bits)
331 struct md_personality *pers = mddev->pers;
335 if (mddev->suspended)
337 else if (pers && pers->congested)
338 ret = pers->congested(mddev, bits);
342 EXPORT_SYMBOL_GPL(mddev_congested);
343 static int md_congested(void *data, int bits)
345 struct mddev *mddev = data;
346 return mddev_congested(mddev, bits);
349 static int md_mergeable_bvec(struct request_queue *q,
350 struct bvec_merge_data *bvm,
351 struct bio_vec *biovec)
353 struct mddev *mddev = q->queuedata;
356 if (mddev->suspended) {
357 /* Must always allow one vec */
358 if (bvm->bi_size == 0)
359 ret = biovec->bv_len;
363 struct md_personality *pers = mddev->pers;
364 if (pers && pers->mergeable_bvec)
365 ret = pers->mergeable_bvec(mddev, bvm, biovec);
367 ret = biovec->bv_len;
373 * Generic flush handling for md
376 static void md_end_flush(struct bio *bio, int err)
378 struct md_rdev *rdev = bio->bi_private;
379 struct mddev *mddev = rdev->mddev;
381 rdev_dec_pending(rdev, mddev);
383 if (atomic_dec_and_test(&mddev->flush_pending)) {
384 /* The pre-request flush has finished */
385 queue_work(md_wq, &mddev->flush_work);
390 static void md_submit_flush_data(struct work_struct *ws);
392 static void submit_flushes(struct work_struct *ws)
394 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
395 struct md_rdev *rdev;
397 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
398 atomic_set(&mddev->flush_pending, 1);
400 rdev_for_each_rcu(rdev, mddev)
401 if (rdev->raid_disk >= 0 &&
402 !test_bit(Faulty, &rdev->flags)) {
403 /* Take two references, one is dropped
404 * when request finishes, one after
405 * we reclaim rcu_read_lock
408 atomic_inc(&rdev->nr_pending);
409 atomic_inc(&rdev->nr_pending);
411 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
412 bi->bi_end_io = md_end_flush;
413 bi->bi_private = rdev;
414 bi->bi_bdev = rdev->bdev;
415 atomic_inc(&mddev->flush_pending);
416 submit_bio(WRITE_FLUSH, bi);
418 rdev_dec_pending(rdev, mddev);
421 if (atomic_dec_and_test(&mddev->flush_pending))
422 queue_work(md_wq, &mddev->flush_work);
425 static void md_submit_flush_data(struct work_struct *ws)
427 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
428 struct bio *bio = mddev->flush_bio;
430 if (bio->bi_iter.bi_size == 0)
431 /* an empty barrier - all done */
434 bio->bi_rw &= ~REQ_FLUSH;
435 mddev->pers->make_request(mddev, bio);
438 mddev->flush_bio = NULL;
439 wake_up(&mddev->sb_wait);
442 void md_flush_request(struct mddev *mddev, struct bio *bio)
444 spin_lock_irq(&mddev->lock);
445 wait_event_lock_irq(mddev->sb_wait,
448 mddev->flush_bio = bio;
449 spin_unlock_irq(&mddev->lock);
451 INIT_WORK(&mddev->flush_work, submit_flushes);
452 queue_work(md_wq, &mddev->flush_work);
454 EXPORT_SYMBOL(md_flush_request);
456 void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
458 struct mddev *mddev = cb->data;
459 md_wakeup_thread(mddev->thread);
462 EXPORT_SYMBOL(md_unplug);
464 static inline struct mddev *mddev_get(struct mddev *mddev)
466 atomic_inc(&mddev->active);
470 static void mddev_delayed_delete(struct work_struct *ws);
472 static void mddev_put(struct mddev *mddev)
474 struct bio_set *bs = NULL;
476 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
478 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
479 mddev->ctime == 0 && !mddev->hold_active) {
480 /* Array is not configured at all, and not held active,
482 list_del_init(&mddev->all_mddevs);
484 mddev->bio_set = NULL;
485 if (mddev->gendisk) {
486 /* We did a probe so need to clean up. Call
487 * queue_work inside the spinlock so that
488 * flush_workqueue() after mddev_find will
489 * succeed in waiting for the work to be done.
491 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
492 queue_work(md_misc_wq, &mddev->del_work);
496 spin_unlock(&all_mddevs_lock);
501 void mddev_init(struct mddev *mddev)
503 mutex_init(&mddev->open_mutex);
504 mutex_init(&mddev->reconfig_mutex);
505 mutex_init(&mddev->bitmap_info.mutex);
506 INIT_LIST_HEAD(&mddev->disks);
507 INIT_LIST_HEAD(&mddev->all_mddevs);
508 init_timer(&mddev->safemode_timer);
509 atomic_set(&mddev->active, 1);
510 atomic_set(&mddev->openers, 0);
511 atomic_set(&mddev->active_io, 0);
512 spin_lock_init(&mddev->lock);
513 atomic_set(&mddev->flush_pending, 0);
514 init_waitqueue_head(&mddev->sb_wait);
515 init_waitqueue_head(&mddev->recovery_wait);
516 mddev->reshape_position = MaxSector;
517 mddev->reshape_backwards = 0;
518 mddev->last_sync_action = "none";
519 mddev->resync_min = 0;
520 mddev->resync_max = MaxSector;
521 mddev->level = LEVEL_NONE;
523 EXPORT_SYMBOL_GPL(mddev_init);
525 static struct mddev *mddev_find(dev_t unit)
527 struct mddev *mddev, *new = NULL;
529 if (unit && MAJOR(unit) != MD_MAJOR)
530 unit &= ~((1<<MdpMinorShift)-1);
533 spin_lock(&all_mddevs_lock);
536 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
537 if (mddev->unit == unit) {
539 spin_unlock(&all_mddevs_lock);
545 list_add(&new->all_mddevs, &all_mddevs);
546 spin_unlock(&all_mddevs_lock);
547 new->hold_active = UNTIL_IOCTL;
551 /* find an unused unit number */
552 static int next_minor = 512;
553 int start = next_minor;
557 dev = MKDEV(MD_MAJOR, next_minor);
559 if (next_minor > MINORMASK)
561 if (next_minor == start) {
562 /* Oh dear, all in use. */
563 spin_unlock(&all_mddevs_lock);
569 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
570 if (mddev->unit == dev) {
576 new->md_minor = MINOR(dev);
577 new->hold_active = UNTIL_STOP;
578 list_add(&new->all_mddevs, &all_mddevs);
579 spin_unlock(&all_mddevs_lock);
582 spin_unlock(&all_mddevs_lock);
584 new = kzalloc(sizeof(*new), GFP_KERNEL);
589 if (MAJOR(unit) == MD_MAJOR)
590 new->md_minor = MINOR(unit);
592 new->md_minor = MINOR(unit) >> MdpMinorShift;
599 static struct attribute_group md_redundancy_group;
601 void mddev_unlock(struct mddev *mddev)
603 if (mddev->to_remove) {
604 /* These cannot be removed under reconfig_mutex as
605 * an access to the files will try to take reconfig_mutex
606 * while holding the file unremovable, which leads to
608 * So hold set sysfs_active while the remove in happeing,
609 * and anything else which might set ->to_remove or my
610 * otherwise change the sysfs namespace will fail with
611 * -EBUSY if sysfs_active is still set.
612 * We set sysfs_active under reconfig_mutex and elsewhere
613 * test it under the same mutex to ensure its correct value
616 struct attribute_group *to_remove = mddev->to_remove;
617 mddev->to_remove = NULL;
618 mddev->sysfs_active = 1;
619 mutex_unlock(&mddev->reconfig_mutex);
621 if (mddev->kobj.sd) {
622 if (to_remove != &md_redundancy_group)
623 sysfs_remove_group(&mddev->kobj, to_remove);
624 if (mddev->pers == NULL ||
625 mddev->pers->sync_request == NULL) {
626 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
627 if (mddev->sysfs_action)
628 sysfs_put(mddev->sysfs_action);
629 mddev->sysfs_action = NULL;
632 mddev->sysfs_active = 0;
634 mutex_unlock(&mddev->reconfig_mutex);
636 /* As we've dropped the mutex we need a spinlock to
637 * make sure the thread doesn't disappear
639 spin_lock(&pers_lock);
640 md_wakeup_thread(mddev->thread);
641 spin_unlock(&pers_lock);
643 EXPORT_SYMBOL_GPL(mddev_unlock);
645 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr)
647 struct md_rdev *rdev;
649 rdev_for_each_rcu(rdev, mddev)
650 if (rdev->desc_nr == nr)
655 EXPORT_SYMBOL_GPL(md_find_rdev_nr_rcu);
657 static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
659 struct md_rdev *rdev;
661 rdev_for_each(rdev, mddev)
662 if (rdev->bdev->bd_dev == dev)
668 static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
670 struct md_rdev *rdev;
672 rdev_for_each_rcu(rdev, mddev)
673 if (rdev->bdev->bd_dev == dev)
679 static struct md_personality *find_pers(int level, char *clevel)
681 struct md_personality *pers;
682 list_for_each_entry(pers, &pers_list, list) {
683 if (level != LEVEL_NONE && pers->level == level)
685 if (strcmp(pers->name, clevel)==0)
691 /* return the offset of the super block in 512byte sectors */
692 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
694 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
695 return MD_NEW_SIZE_SECTORS(num_sectors);
698 static int alloc_disk_sb(struct md_rdev *rdev)
700 rdev->sb_page = alloc_page(GFP_KERNEL);
701 if (!rdev->sb_page) {
702 printk(KERN_ALERT "md: out of memory.\n");
709 void md_rdev_clear(struct md_rdev *rdev)
712 put_page(rdev->sb_page);
714 rdev->sb_page = NULL;
719 put_page(rdev->bb_page);
720 rdev->bb_page = NULL;
722 kfree(rdev->badblocks.page);
723 rdev->badblocks.page = NULL;
725 EXPORT_SYMBOL_GPL(md_rdev_clear);
727 static void super_written(struct bio *bio, int error)
729 struct md_rdev *rdev = bio->bi_private;
730 struct mddev *mddev = rdev->mddev;
732 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
733 printk("md: super_written gets error=%d, uptodate=%d\n",
734 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
735 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
736 md_error(mddev, rdev);
739 if (atomic_dec_and_test(&mddev->pending_writes))
740 wake_up(&mddev->sb_wait);
744 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
745 sector_t sector, int size, struct page *page)
747 /* write first size bytes of page to sector of rdev
748 * Increment mddev->pending_writes before returning
749 * and decrement it on completion, waking up sb_wait
750 * if zero is reached.
751 * If an error occurred, call md_error
753 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
755 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
756 bio->bi_iter.bi_sector = sector;
757 bio_add_page(bio, page, size, 0);
758 bio->bi_private = rdev;
759 bio->bi_end_io = super_written;
761 atomic_inc(&mddev->pending_writes);
762 submit_bio(WRITE_FLUSH_FUA, bio);
765 void md_super_wait(struct mddev *mddev)
767 /* wait for all superblock writes that were scheduled to complete */
768 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
771 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
772 struct page *page, int rw, bool metadata_op)
774 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
777 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
778 rdev->meta_bdev : rdev->bdev;
780 bio->bi_iter.bi_sector = sector + rdev->sb_start;
781 else if (rdev->mddev->reshape_position != MaxSector &&
782 (rdev->mddev->reshape_backwards ==
783 (sector >= rdev->mddev->reshape_position)))
784 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
786 bio->bi_iter.bi_sector = sector + rdev->data_offset;
787 bio_add_page(bio, page, size, 0);
788 submit_bio_wait(rw, bio);
790 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
794 EXPORT_SYMBOL_GPL(sync_page_io);
796 static int read_disk_sb(struct md_rdev *rdev, int size)
798 char b[BDEVNAME_SIZE];
803 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
809 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
810 bdevname(rdev->bdev,b));
814 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
816 return sb1->set_uuid0 == sb2->set_uuid0 &&
817 sb1->set_uuid1 == sb2->set_uuid1 &&
818 sb1->set_uuid2 == sb2->set_uuid2 &&
819 sb1->set_uuid3 == sb2->set_uuid3;
822 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
825 mdp_super_t *tmp1, *tmp2;
827 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
828 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
830 if (!tmp1 || !tmp2) {
832 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
840 * nr_disks is not constant
845 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
852 static u32 md_csum_fold(u32 csum)
854 csum = (csum & 0xffff) + (csum >> 16);
855 return (csum & 0xffff) + (csum >> 16);
858 static unsigned int calc_sb_csum(mdp_super_t *sb)
861 u32 *sb32 = (u32*)sb;
863 unsigned int disk_csum, csum;
865 disk_csum = sb->sb_csum;
868 for (i = 0; i < MD_SB_BYTES/4 ; i++)
870 csum = (newcsum & 0xffffffff) + (newcsum>>32);
873 /* This used to use csum_partial, which was wrong for several
874 * reasons including that different results are returned on
875 * different architectures. It isn't critical that we get exactly
876 * the same return value as before (we always csum_fold before
877 * testing, and that removes any differences). However as we
878 * know that csum_partial always returned a 16bit value on
879 * alphas, do a fold to maximise conformity to previous behaviour.
881 sb->sb_csum = md_csum_fold(disk_csum);
883 sb->sb_csum = disk_csum;
889 * Handle superblock details.
890 * We want to be able to handle multiple superblock formats
891 * so we have a common interface to them all, and an array of
892 * different handlers.
893 * We rely on user-space to write the initial superblock, and support
894 * reading and updating of superblocks.
895 * Interface methods are:
896 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
897 * loads and validates a superblock on dev.
898 * if refdev != NULL, compare superblocks on both devices
900 * 0 - dev has a superblock that is compatible with refdev
901 * 1 - dev has a superblock that is compatible and newer than refdev
902 * so dev should be used as the refdev in future
903 * -EINVAL superblock incompatible or invalid
904 * -othererror e.g. -EIO
906 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
907 * Verify that dev is acceptable into mddev.
908 * The first time, mddev->raid_disks will be 0, and data from
909 * dev should be merged in. Subsequent calls check that dev
910 * is new enough. Return 0 or -EINVAL
912 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
913 * Update the superblock for rdev with data in mddev
914 * This does not write to disc.
920 struct module *owner;
921 int (*load_super)(struct md_rdev *rdev,
922 struct md_rdev *refdev,
924 int (*validate_super)(struct mddev *mddev,
925 struct md_rdev *rdev);
926 void (*sync_super)(struct mddev *mddev,
927 struct md_rdev *rdev);
928 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
929 sector_t num_sectors);
930 int (*allow_new_offset)(struct md_rdev *rdev,
931 unsigned long long new_offset);
935 * Check that the given mddev has no bitmap.
937 * This function is called from the run method of all personalities that do not
938 * support bitmaps. It prints an error message and returns non-zero if mddev
939 * has a bitmap. Otherwise, it returns 0.
942 int md_check_no_bitmap(struct mddev *mddev)
944 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
946 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
947 mdname(mddev), mddev->pers->name);
950 EXPORT_SYMBOL(md_check_no_bitmap);
953 * load_super for 0.90.0
955 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
957 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
962 * Calculate the position of the superblock (512byte sectors),
963 * it's at the end of the disk.
965 * It also happens to be a multiple of 4Kb.
967 rdev->sb_start = calc_dev_sboffset(rdev);
969 ret = read_disk_sb(rdev, MD_SB_BYTES);
974 bdevname(rdev->bdev, b);
975 sb = page_address(rdev->sb_page);
977 if (sb->md_magic != MD_SB_MAGIC) {
978 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
983 if (sb->major_version != 0 ||
984 sb->minor_version < 90 ||
985 sb->minor_version > 91) {
986 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
987 sb->major_version, sb->minor_version,
992 if (sb->raid_disks <= 0)
995 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
996 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1001 rdev->preferred_minor = sb->md_minor;
1002 rdev->data_offset = 0;
1003 rdev->new_data_offset = 0;
1004 rdev->sb_size = MD_SB_BYTES;
1005 rdev->badblocks.shift = -1;
1007 if (sb->level == LEVEL_MULTIPATH)
1010 rdev->desc_nr = sb->this_disk.number;
1016 mdp_super_t *refsb = page_address(refdev->sb_page);
1017 if (!uuid_equal(refsb, sb)) {
1018 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1019 b, bdevname(refdev->bdev,b2));
1022 if (!sb_equal(refsb, sb)) {
1023 printk(KERN_WARNING "md: %s has same UUID"
1024 " but different superblock to %s\n",
1025 b, bdevname(refdev->bdev, b2));
1029 ev2 = md_event(refsb);
1035 rdev->sectors = rdev->sb_start;
1036 /* Limit to 4TB as metadata cannot record more than that.
1037 * (not needed for Linear and RAID0 as metadata doesn't
1040 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
1041 rdev->sectors = (2ULL << 32) - 2;
1043 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1044 /* "this cannot possibly happen" ... */
1052 * validate_super for 0.90.0
1054 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1057 mdp_super_t *sb = page_address(rdev->sb_page);
1058 __u64 ev1 = md_event(sb);
1060 rdev->raid_disk = -1;
1061 clear_bit(Faulty, &rdev->flags);
1062 clear_bit(In_sync, &rdev->flags);
1063 clear_bit(Bitmap_sync, &rdev->flags);
1064 clear_bit(WriteMostly, &rdev->flags);
1066 if (mddev->raid_disks == 0) {
1067 mddev->major_version = 0;
1068 mddev->minor_version = sb->minor_version;
1069 mddev->patch_version = sb->patch_version;
1070 mddev->external = 0;
1071 mddev->chunk_sectors = sb->chunk_size >> 9;
1072 mddev->ctime = sb->ctime;
1073 mddev->utime = sb->utime;
1074 mddev->level = sb->level;
1075 mddev->clevel[0] = 0;
1076 mddev->layout = sb->layout;
1077 mddev->raid_disks = sb->raid_disks;
1078 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1079 mddev->events = ev1;
1080 mddev->bitmap_info.offset = 0;
1081 mddev->bitmap_info.space = 0;
1082 /* bitmap can use 60 K after the 4K superblocks */
1083 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1084 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
1085 mddev->reshape_backwards = 0;
1087 if (mddev->minor_version >= 91) {
1088 mddev->reshape_position = sb->reshape_position;
1089 mddev->delta_disks = sb->delta_disks;
1090 mddev->new_level = sb->new_level;
1091 mddev->new_layout = sb->new_layout;
1092 mddev->new_chunk_sectors = sb->new_chunk >> 9;
1093 if (mddev->delta_disks < 0)
1094 mddev->reshape_backwards = 1;
1096 mddev->reshape_position = MaxSector;
1097 mddev->delta_disks = 0;
1098 mddev->new_level = mddev->level;
1099 mddev->new_layout = mddev->layout;
1100 mddev->new_chunk_sectors = mddev->chunk_sectors;
1103 if (sb->state & (1<<MD_SB_CLEAN))
1104 mddev->recovery_cp = MaxSector;
1106 if (sb->events_hi == sb->cp_events_hi &&
1107 sb->events_lo == sb->cp_events_lo) {
1108 mddev->recovery_cp = sb->recovery_cp;
1110 mddev->recovery_cp = 0;
1113 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1114 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1115 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1116 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1118 mddev->max_disks = MD_SB_DISKS;
1120 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1121 mddev->bitmap_info.file == NULL) {
1122 mddev->bitmap_info.offset =
1123 mddev->bitmap_info.default_offset;
1124 mddev->bitmap_info.space =
1125 mddev->bitmap_info.default_space;
1128 } else if (mddev->pers == NULL) {
1129 /* Insist on good event counter while assembling, except
1130 * for spares (which don't need an event count) */
1132 if (sb->disks[rdev->desc_nr].state & (
1133 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1134 if (ev1 < mddev->events)
1136 } else if (mddev->bitmap) {
1137 /* if adding to array with a bitmap, then we can accept an
1138 * older device ... but not too old.
1140 if (ev1 < mddev->bitmap->events_cleared)
1142 if (ev1 < mddev->events)
1143 set_bit(Bitmap_sync, &rdev->flags);
1145 if (ev1 < mddev->events)
1146 /* just a hot-add of a new device, leave raid_disk at -1 */
1150 if (mddev->level != LEVEL_MULTIPATH) {
1151 desc = sb->disks + rdev->desc_nr;
1153 if (desc->state & (1<<MD_DISK_FAULTY))
1154 set_bit(Faulty, &rdev->flags);
1155 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1156 desc->raid_disk < mddev->raid_disks */) {
1157 set_bit(In_sync, &rdev->flags);
1158 rdev->raid_disk = desc->raid_disk;
1159 rdev->saved_raid_disk = desc->raid_disk;
1160 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1161 /* active but not in sync implies recovery up to
1162 * reshape position. We don't know exactly where
1163 * that is, so set to zero for now */
1164 if (mddev->minor_version >= 91) {
1165 rdev->recovery_offset = 0;
1166 rdev->raid_disk = desc->raid_disk;
1169 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1170 set_bit(WriteMostly, &rdev->flags);
1171 } else /* MULTIPATH are always insync */
1172 set_bit(In_sync, &rdev->flags);
1177 * sync_super for 0.90.0
1179 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1182 struct md_rdev *rdev2;
1183 int next_spare = mddev->raid_disks;
1185 /* make rdev->sb match mddev data..
1188 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1189 * 3/ any empty disks < next_spare become removed
1191 * disks[0] gets initialised to REMOVED because
1192 * we cannot be sure from other fields if it has
1193 * been initialised or not.
1196 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1198 rdev->sb_size = MD_SB_BYTES;
1200 sb = page_address(rdev->sb_page);
1202 memset(sb, 0, sizeof(*sb));
1204 sb->md_magic = MD_SB_MAGIC;
1205 sb->major_version = mddev->major_version;
1206 sb->patch_version = mddev->patch_version;
1207 sb->gvalid_words = 0; /* ignored */
1208 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1209 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1210 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1211 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1213 sb->ctime = mddev->ctime;
1214 sb->level = mddev->level;
1215 sb->size = mddev->dev_sectors / 2;
1216 sb->raid_disks = mddev->raid_disks;
1217 sb->md_minor = mddev->md_minor;
1218 sb->not_persistent = 0;
1219 sb->utime = mddev->utime;
1221 sb->events_hi = (mddev->events>>32);
1222 sb->events_lo = (u32)mddev->events;
1224 if (mddev->reshape_position == MaxSector)
1225 sb->minor_version = 90;
1227 sb->minor_version = 91;
1228 sb->reshape_position = mddev->reshape_position;
1229 sb->new_level = mddev->new_level;
1230 sb->delta_disks = mddev->delta_disks;
1231 sb->new_layout = mddev->new_layout;
1232 sb->new_chunk = mddev->new_chunk_sectors << 9;
1234 mddev->minor_version = sb->minor_version;
1237 sb->recovery_cp = mddev->recovery_cp;
1238 sb->cp_events_hi = (mddev->events>>32);
1239 sb->cp_events_lo = (u32)mddev->events;
1240 if (mddev->recovery_cp == MaxSector)
1241 sb->state = (1<< MD_SB_CLEAN);
1243 sb->recovery_cp = 0;
1245 sb->layout = mddev->layout;
1246 sb->chunk_size = mddev->chunk_sectors << 9;
1248 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1249 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1251 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1252 rdev_for_each(rdev2, mddev) {
1255 int is_active = test_bit(In_sync, &rdev2->flags);
1257 if (rdev2->raid_disk >= 0 &&
1258 sb->minor_version >= 91)
1259 /* we have nowhere to store the recovery_offset,
1260 * but if it is not below the reshape_position,
1261 * we can piggy-back on that.
1264 if (rdev2->raid_disk < 0 ||
1265 test_bit(Faulty, &rdev2->flags))
1268 desc_nr = rdev2->raid_disk;
1270 desc_nr = next_spare++;
1271 rdev2->desc_nr = desc_nr;
1272 d = &sb->disks[rdev2->desc_nr];
1274 d->number = rdev2->desc_nr;
1275 d->major = MAJOR(rdev2->bdev->bd_dev);
1276 d->minor = MINOR(rdev2->bdev->bd_dev);
1278 d->raid_disk = rdev2->raid_disk;
1280 d->raid_disk = rdev2->desc_nr; /* compatibility */
1281 if (test_bit(Faulty, &rdev2->flags))
1282 d->state = (1<<MD_DISK_FAULTY);
1283 else if (is_active) {
1284 d->state = (1<<MD_DISK_ACTIVE);
1285 if (test_bit(In_sync, &rdev2->flags))
1286 d->state |= (1<<MD_DISK_SYNC);
1294 if (test_bit(WriteMostly, &rdev2->flags))
1295 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1297 /* now set the "removed" and "faulty" bits on any missing devices */
1298 for (i=0 ; i < mddev->raid_disks ; i++) {
1299 mdp_disk_t *d = &sb->disks[i];
1300 if (d->state == 0 && d->number == 0) {
1303 d->state = (1<<MD_DISK_REMOVED);
1304 d->state |= (1<<MD_DISK_FAULTY);
1308 sb->nr_disks = nr_disks;
1309 sb->active_disks = active;
1310 sb->working_disks = working;
1311 sb->failed_disks = failed;
1312 sb->spare_disks = spare;
1314 sb->this_disk = sb->disks[rdev->desc_nr];
1315 sb->sb_csum = calc_sb_csum(sb);
1319 * rdev_size_change for 0.90.0
1321 static unsigned long long
1322 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1324 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1325 return 0; /* component must fit device */
1326 if (rdev->mddev->bitmap_info.offset)
1327 return 0; /* can't move bitmap */
1328 rdev->sb_start = calc_dev_sboffset(rdev);
1329 if (!num_sectors || num_sectors > rdev->sb_start)
1330 num_sectors = rdev->sb_start;
1331 /* Limit to 4TB as metadata cannot record more than that.
1332 * 4TB == 2^32 KB, or 2*2^32 sectors.
1334 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
1335 num_sectors = (2ULL << 32) - 2;
1336 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1338 md_super_wait(rdev->mddev);
1343 super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1345 /* non-zero offset changes not possible with v0.90 */
1346 return new_offset == 0;
1350 * version 1 superblock
1353 static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
1357 unsigned long long newcsum;
1358 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1359 __le32 *isuper = (__le32*)sb;
1361 disk_csum = sb->sb_csum;
1364 for (; size >= 4; size -= 4)
1365 newcsum += le32_to_cpu(*isuper++);
1368 newcsum += le16_to_cpu(*(__le16*) isuper);
1370 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1371 sb->sb_csum = disk_csum;
1372 return cpu_to_le32(csum);
1375 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1377 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1379 struct mdp_superblock_1 *sb;
1383 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1387 * Calculate the position of the superblock in 512byte sectors.
1388 * It is always aligned to a 4K boundary and
1389 * depeding on minor_version, it can be:
1390 * 0: At least 8K, but less than 12K, from end of device
1391 * 1: At start of device
1392 * 2: 4K from start of device.
1394 switch(minor_version) {
1396 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1398 sb_start &= ~(sector_t)(4*2-1);
1409 rdev->sb_start = sb_start;
1411 /* superblock is rarely larger than 1K, but it can be larger,
1412 * and it is safe to read 4k, so we do that
1414 ret = read_disk_sb(rdev, 4096);
1415 if (ret) return ret;
1417 sb = page_address(rdev->sb_page);
1419 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1420 sb->major_version != cpu_to_le32(1) ||
1421 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1422 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1423 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1426 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1427 printk("md: invalid superblock checksum on %s\n",
1428 bdevname(rdev->bdev,b));
1431 if (le64_to_cpu(sb->data_size) < 10) {
1432 printk("md: data_size too small on %s\n",
1433 bdevname(rdev->bdev,b));
1438 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1439 /* Some padding is non-zero, might be a new feature */
1442 rdev->preferred_minor = 0xffff;
1443 rdev->data_offset = le64_to_cpu(sb->data_offset);
1444 rdev->new_data_offset = rdev->data_offset;
1445 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1446 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1447 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
1448 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1450 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1451 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1452 if (rdev->sb_size & bmask)
1453 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1456 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1459 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1462 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1465 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1467 if (!rdev->bb_page) {
1468 rdev->bb_page = alloc_page(GFP_KERNEL);
1472 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1473 rdev->badblocks.count == 0) {
1474 /* need to load the bad block list.
1475 * Currently we limit it to one page.
1481 int sectors = le16_to_cpu(sb->bblog_size);
1482 if (sectors > (PAGE_SIZE / 512))
1484 offset = le32_to_cpu(sb->bblog_offset);
1487 bb_sector = (long long)offset;
1488 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1489 rdev->bb_page, READ, true))
1491 bbp = (u64 *)page_address(rdev->bb_page);
1492 rdev->badblocks.shift = sb->bblog_shift;
1493 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1494 u64 bb = le64_to_cpu(*bbp);
1495 int count = bb & (0x3ff);
1496 u64 sector = bb >> 10;
1497 sector <<= sb->bblog_shift;
1498 count <<= sb->bblog_shift;
1501 if (md_set_badblocks(&rdev->badblocks,
1502 sector, count, 1) == 0)
1505 } else if (sb->bblog_offset != 0)
1506 rdev->badblocks.shift = 0;
1512 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1514 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1515 sb->level != refsb->level ||
1516 sb->layout != refsb->layout ||
1517 sb->chunksize != refsb->chunksize) {
1518 printk(KERN_WARNING "md: %s has strangely different"
1519 " superblock to %s\n",
1520 bdevname(rdev->bdev,b),
1521 bdevname(refdev->bdev,b2));
1524 ev1 = le64_to_cpu(sb->events);
1525 ev2 = le64_to_cpu(refsb->events);
1532 if (minor_version) {
1533 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1534 sectors -= rdev->data_offset;
1536 sectors = rdev->sb_start;
1537 if (sectors < le64_to_cpu(sb->data_size))
1539 rdev->sectors = le64_to_cpu(sb->data_size);
1543 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1545 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1546 __u64 ev1 = le64_to_cpu(sb->events);
1548 rdev->raid_disk = -1;
1549 clear_bit(Faulty, &rdev->flags);
1550 clear_bit(In_sync, &rdev->flags);
1551 clear_bit(Bitmap_sync, &rdev->flags);
1552 clear_bit(WriteMostly, &rdev->flags);
1554 if (mddev->raid_disks == 0) {
1555 mddev->major_version = 1;
1556 mddev->patch_version = 0;
1557 mddev->external = 0;
1558 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1559 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1560 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1561 mddev->level = le32_to_cpu(sb->level);
1562 mddev->clevel[0] = 0;
1563 mddev->layout = le32_to_cpu(sb->layout);
1564 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1565 mddev->dev_sectors = le64_to_cpu(sb->size);
1566 mddev->events = ev1;
1567 mddev->bitmap_info.offset = 0;
1568 mddev->bitmap_info.space = 0;
1569 /* Default location for bitmap is 1K after superblock
1570 * using 3K - total of 4K
1572 mddev->bitmap_info.default_offset = 1024 >> 9;
1573 mddev->bitmap_info.default_space = (4096-1024) >> 9;
1574 mddev->reshape_backwards = 0;
1576 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1577 memcpy(mddev->uuid, sb->set_uuid, 16);
1579 mddev->max_disks = (4096-256)/2;
1581 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1582 mddev->bitmap_info.file == NULL) {
1583 mddev->bitmap_info.offset =
1584 (__s32)le32_to_cpu(sb->bitmap_offset);
1585 /* Metadata doesn't record how much space is available.
1586 * For 1.0, we assume we can use up to the superblock
1587 * if before, else to 4K beyond superblock.
1588 * For others, assume no change is possible.
1590 if (mddev->minor_version > 0)
1591 mddev->bitmap_info.space = 0;
1592 else if (mddev->bitmap_info.offset > 0)
1593 mddev->bitmap_info.space =
1594 8 - mddev->bitmap_info.offset;
1596 mddev->bitmap_info.space =
1597 -mddev->bitmap_info.offset;
1600 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1601 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1602 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1603 mddev->new_level = le32_to_cpu(sb->new_level);
1604 mddev->new_layout = le32_to_cpu(sb->new_layout);
1605 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1606 if (mddev->delta_disks < 0 ||
1607 (mddev->delta_disks == 0 &&
1608 (le32_to_cpu(sb->feature_map)
1609 & MD_FEATURE_RESHAPE_BACKWARDS)))
1610 mddev->reshape_backwards = 1;
1612 mddev->reshape_position = MaxSector;
1613 mddev->delta_disks = 0;
1614 mddev->new_level = mddev->level;
1615 mddev->new_layout = mddev->layout;
1616 mddev->new_chunk_sectors = mddev->chunk_sectors;
1619 } else if (mddev->pers == NULL) {
1620 /* Insist of good event counter while assembling, except for
1621 * spares (which don't need an event count) */
1623 if (rdev->desc_nr >= 0 &&
1624 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1625 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1626 if (ev1 < mddev->events)
1628 } else if (mddev->bitmap) {
1629 /* If adding to array with a bitmap, then we can accept an
1630 * older device, but not too old.
1632 if (ev1 < mddev->bitmap->events_cleared)
1634 if (ev1 < mddev->events)
1635 set_bit(Bitmap_sync, &rdev->flags);
1637 if (ev1 < mddev->events)
1638 /* just a hot-add of a new device, leave raid_disk at -1 */
1641 if (mddev->level != LEVEL_MULTIPATH) {
1643 if (rdev->desc_nr < 0 ||
1644 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1648 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1650 case 0xffff: /* spare */
1652 case 0xfffe: /* faulty */
1653 set_bit(Faulty, &rdev->flags);
1656 rdev->saved_raid_disk = role;
1657 if ((le32_to_cpu(sb->feature_map) &
1658 MD_FEATURE_RECOVERY_OFFSET)) {
1659 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1660 if (!(le32_to_cpu(sb->feature_map) &
1661 MD_FEATURE_RECOVERY_BITMAP))
1662 rdev->saved_raid_disk = -1;
1664 set_bit(In_sync, &rdev->flags);
1665 rdev->raid_disk = role;
1668 if (sb->devflags & WriteMostly1)
1669 set_bit(WriteMostly, &rdev->flags);
1670 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1671 set_bit(Replacement, &rdev->flags);
1672 } else /* MULTIPATH are always insync */
1673 set_bit(In_sync, &rdev->flags);
1678 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1680 struct mdp_superblock_1 *sb;
1681 struct md_rdev *rdev2;
1683 /* make rdev->sb match mddev and rdev data. */
1685 sb = page_address(rdev->sb_page);
1687 sb->feature_map = 0;
1689 sb->recovery_offset = cpu_to_le64(0);
1690 memset(sb->pad3, 0, sizeof(sb->pad3));
1692 sb->utime = cpu_to_le64((__u64)mddev->utime);
1693 sb->events = cpu_to_le64(mddev->events);
1695 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1697 sb->resync_offset = cpu_to_le64(0);
1699 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1701 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1702 sb->size = cpu_to_le64(mddev->dev_sectors);
1703 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1704 sb->level = cpu_to_le32(mddev->level);
1705 sb->layout = cpu_to_le32(mddev->layout);
1707 if (test_bit(WriteMostly, &rdev->flags))
1708 sb->devflags |= WriteMostly1;
1710 sb->devflags &= ~WriteMostly1;
1711 sb->data_offset = cpu_to_le64(rdev->data_offset);
1712 sb->data_size = cpu_to_le64(rdev->sectors);
1714 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1715 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1716 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1719 if (rdev->raid_disk >= 0 &&
1720 !test_bit(In_sync, &rdev->flags)) {
1722 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1723 sb->recovery_offset =
1724 cpu_to_le64(rdev->recovery_offset);
1725 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1727 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
1729 if (test_bit(Replacement, &rdev->flags))
1731 cpu_to_le32(MD_FEATURE_REPLACEMENT);
1733 if (mddev->reshape_position != MaxSector) {
1734 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1735 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1736 sb->new_layout = cpu_to_le32(mddev->new_layout);
1737 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1738 sb->new_level = cpu_to_le32(mddev->new_level);
1739 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1740 if (mddev->delta_disks == 0 &&
1741 mddev->reshape_backwards)
1743 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
1744 if (rdev->new_data_offset != rdev->data_offset) {
1746 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1747 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1748 - rdev->data_offset));
1752 if (rdev->badblocks.count == 0)
1753 /* Nothing to do for bad blocks*/ ;
1754 else if (sb->bblog_offset == 0)
1755 /* Cannot record bad blocks on this device */
1756 md_error(mddev, rdev);
1758 struct badblocks *bb = &rdev->badblocks;
1759 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1761 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1766 seq = read_seqbegin(&bb->lock);
1768 memset(bbp, 0xff, PAGE_SIZE);
1770 for (i = 0 ; i < bb->count ; i++) {
1771 u64 internal_bb = p[i];
1772 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1773 | BB_LEN(internal_bb));
1774 bbp[i] = cpu_to_le64(store_bb);
1777 if (read_seqretry(&bb->lock, seq))
1780 bb->sector = (rdev->sb_start +
1781 (int)le32_to_cpu(sb->bblog_offset));
1782 bb->size = le16_to_cpu(sb->bblog_size);
1787 rdev_for_each(rdev2, mddev)
1788 if (rdev2->desc_nr+1 > max_dev)
1789 max_dev = rdev2->desc_nr+1;
1791 if (max_dev > le32_to_cpu(sb->max_dev)) {
1793 sb->max_dev = cpu_to_le32(max_dev);
1794 rdev->sb_size = max_dev * 2 + 256;
1795 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1796 if (rdev->sb_size & bmask)
1797 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1799 max_dev = le32_to_cpu(sb->max_dev);
1801 for (i=0; i<max_dev;i++)
1802 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1804 rdev_for_each(rdev2, mddev) {
1806 if (test_bit(Faulty, &rdev2->flags))
1807 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1808 else if (test_bit(In_sync, &rdev2->flags))
1809 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1810 else if (rdev2->raid_disk >= 0)
1811 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1813 sb->dev_roles[i] = cpu_to_le16(0xffff);
1816 sb->sb_csum = calc_sb_1_csum(sb);
1819 static unsigned long long
1820 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1822 struct mdp_superblock_1 *sb;
1823 sector_t max_sectors;
1824 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1825 return 0; /* component must fit device */
1826 if (rdev->data_offset != rdev->new_data_offset)
1827 return 0; /* too confusing */
1828 if (rdev->sb_start < rdev->data_offset) {
1829 /* minor versions 1 and 2; superblock before data */
1830 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1831 max_sectors -= rdev->data_offset;
1832 if (!num_sectors || num_sectors > max_sectors)
1833 num_sectors = max_sectors;
1834 } else if (rdev->mddev->bitmap_info.offset) {
1835 /* minor version 0 with bitmap we can't move */
1838 /* minor version 0; superblock after data */
1840 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1841 sb_start &= ~(sector_t)(4*2 - 1);
1842 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1843 if (!num_sectors || num_sectors > max_sectors)
1844 num_sectors = max_sectors;
1845 rdev->sb_start = sb_start;
1847 sb = page_address(rdev->sb_page);
1848 sb->data_size = cpu_to_le64(num_sectors);
1849 sb->super_offset = rdev->sb_start;
1850 sb->sb_csum = calc_sb_1_csum(sb);
1851 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1853 md_super_wait(rdev->mddev);
1859 super_1_allow_new_offset(struct md_rdev *rdev,
1860 unsigned long long new_offset)
1862 /* All necessary checks on new >= old have been done */
1863 struct bitmap *bitmap;
1864 if (new_offset >= rdev->data_offset)
1867 /* with 1.0 metadata, there is no metadata to tread on
1868 * so we can always move back */
1869 if (rdev->mddev->minor_version == 0)
1872 /* otherwise we must be sure not to step on
1873 * any metadata, so stay:
1874 * 36K beyond start of superblock
1875 * beyond end of badblocks
1876 * beyond write-intent bitmap
1878 if (rdev->sb_start + (32+4)*2 > new_offset)
1880 bitmap = rdev->mddev->bitmap;
1881 if (bitmap && !rdev->mddev->bitmap_info.file &&
1882 rdev->sb_start + rdev->mddev->bitmap_info.offset +
1883 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
1885 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1891 static struct super_type super_types[] = {
1894 .owner = THIS_MODULE,
1895 .load_super = super_90_load,
1896 .validate_super = super_90_validate,
1897 .sync_super = super_90_sync,
1898 .rdev_size_change = super_90_rdev_size_change,
1899 .allow_new_offset = super_90_allow_new_offset,
1903 .owner = THIS_MODULE,
1904 .load_super = super_1_load,
1905 .validate_super = super_1_validate,
1906 .sync_super = super_1_sync,
1907 .rdev_size_change = super_1_rdev_size_change,
1908 .allow_new_offset = super_1_allow_new_offset,
1912 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1914 if (mddev->sync_super) {
1915 mddev->sync_super(mddev, rdev);
1919 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1921 super_types[mddev->major_version].sync_super(mddev, rdev);
1924 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1926 struct md_rdev *rdev, *rdev2;
1929 rdev_for_each_rcu(rdev, mddev1)
1930 rdev_for_each_rcu(rdev2, mddev2)
1931 if (rdev->bdev->bd_contains ==
1932 rdev2->bdev->bd_contains) {
1940 static LIST_HEAD(pending_raid_disks);
1943 * Try to register data integrity profile for an mddev
1945 * This is called when an array is started and after a disk has been kicked
1946 * from the array. It only succeeds if all working and active component devices
1947 * are integrity capable with matching profiles.
1949 int md_integrity_register(struct mddev *mddev)
1951 struct md_rdev *rdev, *reference = NULL;
1953 if (list_empty(&mddev->disks))
1954 return 0; /* nothing to do */
1955 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1956 return 0; /* shouldn't register, or already is */
1957 rdev_for_each(rdev, mddev) {
1958 /* skip spares and non-functional disks */
1959 if (test_bit(Faulty, &rdev->flags))
1961 if (rdev->raid_disk < 0)
1964 /* Use the first rdev as the reference */
1968 /* does this rdev's profile match the reference profile? */
1969 if (blk_integrity_compare(reference->bdev->bd_disk,
1970 rdev->bdev->bd_disk) < 0)
1973 if (!reference || !bdev_get_integrity(reference->bdev))
1976 * All component devices are integrity capable and have matching
1977 * profiles, register the common profile for the md device.
1979 if (blk_integrity_register(mddev->gendisk,
1980 bdev_get_integrity(reference->bdev)) != 0) {
1981 printk(KERN_ERR "md: failed to register integrity for %s\n",
1985 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1986 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1987 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1993 EXPORT_SYMBOL(md_integrity_register);
1995 /* Disable data integrity if non-capable/non-matching disk is being added */
1996 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1998 struct blk_integrity *bi_rdev;
1999 struct blk_integrity *bi_mddev;
2001 if (!mddev->gendisk)
2004 bi_rdev = bdev_get_integrity(rdev->bdev);
2005 bi_mddev = blk_get_integrity(mddev->gendisk);
2007 if (!bi_mddev) /* nothing to do */
2009 if (rdev->raid_disk < 0) /* skip spares */
2011 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2012 rdev->bdev->bd_disk) >= 0)
2014 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2015 blk_integrity_unregister(mddev->gendisk);
2017 EXPORT_SYMBOL(md_integrity_add_rdev);
2019 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
2021 char b[BDEVNAME_SIZE];
2026 /* prevent duplicates */
2027 if (find_rdev(mddev, rdev->bdev->bd_dev))
2030 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2031 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2032 rdev->sectors < mddev->dev_sectors)) {
2034 /* Cannot change size, so fail
2035 * If mddev->level <= 0, then we don't care
2036 * about aligning sizes (e.g. linear)
2038 if (mddev->level > 0)
2041 mddev->dev_sectors = rdev->sectors;
2044 /* Verify rdev->desc_nr is unique.
2045 * If it is -1, assign a free number, else
2046 * check number is not in use
2049 if (rdev->desc_nr < 0) {
2052 choice = mddev->raid_disks;
2053 while (md_find_rdev_nr_rcu(mddev, choice))
2055 rdev->desc_nr = choice;
2057 if (md_find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2063 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2064 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2065 mdname(mddev), mddev->max_disks);
2068 bdevname(rdev->bdev,b);
2069 while ( (s=strchr(b, '/')) != NULL)
2072 rdev->mddev = mddev;
2073 printk(KERN_INFO "md: bind<%s>\n", b);
2075 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2078 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2079 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2080 /* failure here is OK */;
2081 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2083 list_add_rcu(&rdev->same_set, &mddev->disks);
2084 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2086 /* May as well allow recovery to be retried once */
2087 mddev->recovery_disabled++;
2092 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2097 static void md_delayed_delete(struct work_struct *ws)
2099 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2100 kobject_del(&rdev->kobj);
2101 kobject_put(&rdev->kobj);
2104 static void unbind_rdev_from_array(struct md_rdev *rdev)
2106 char b[BDEVNAME_SIZE];
2108 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2109 list_del_rcu(&rdev->same_set);
2110 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2112 sysfs_remove_link(&rdev->kobj, "block");
2113 sysfs_put(rdev->sysfs_state);
2114 rdev->sysfs_state = NULL;
2115 rdev->badblocks.count = 0;
2116 /* We need to delay this, otherwise we can deadlock when
2117 * writing to 'remove' to "dev/state". We also need
2118 * to delay it due to rcu usage.
2121 INIT_WORK(&rdev->del_work, md_delayed_delete);
2122 kobject_get(&rdev->kobj);
2123 queue_work(md_misc_wq, &rdev->del_work);
2127 * prevent the device from being mounted, repartitioned or
2128 * otherwise reused by a RAID array (or any other kernel
2129 * subsystem), by bd_claiming the device.
2131 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2134 struct block_device *bdev;
2135 char b[BDEVNAME_SIZE];
2137 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2138 shared ? (struct md_rdev *)lock_rdev : rdev);
2140 printk(KERN_ERR "md: could not open %s.\n",
2141 __bdevname(dev, b));
2142 return PTR_ERR(bdev);
2148 static void unlock_rdev(struct md_rdev *rdev)
2150 struct block_device *bdev = rdev->bdev;
2152 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2155 void md_autodetect_dev(dev_t dev);
2157 static void export_rdev(struct md_rdev *rdev)
2159 char b[BDEVNAME_SIZE];
2161 printk(KERN_INFO "md: export_rdev(%s)\n",
2162 bdevname(rdev->bdev,b));
2163 md_rdev_clear(rdev);
2165 if (test_bit(AutoDetected, &rdev->flags))
2166 md_autodetect_dev(rdev->bdev->bd_dev);
2169 kobject_put(&rdev->kobj);
2172 void md_kick_rdev_from_array(struct md_rdev *rdev)
2174 unbind_rdev_from_array(rdev);
2177 EXPORT_SYMBOL_GPL(md_kick_rdev_from_array);
2179 static void export_array(struct mddev *mddev)
2181 struct md_rdev *rdev;
2183 while (!list_empty(&mddev->disks)) {
2184 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2186 md_kick_rdev_from_array(rdev);
2188 mddev->raid_disks = 0;
2189 mddev->major_version = 0;
2192 static void sync_sbs(struct mddev *mddev, int nospares)
2194 /* Update each superblock (in-memory image), but
2195 * if we are allowed to, skip spares which already
2196 * have the right event counter, or have one earlier
2197 * (which would mean they aren't being marked as dirty
2198 * with the rest of the array)
2200 struct md_rdev *rdev;
2201 rdev_for_each(rdev, mddev) {
2202 if (rdev->sb_events == mddev->events ||
2204 rdev->raid_disk < 0 &&
2205 rdev->sb_events+1 == mddev->events)) {
2206 /* Don't update this superblock */
2207 rdev->sb_loaded = 2;
2209 sync_super(mddev, rdev);
2210 rdev->sb_loaded = 1;
2215 void md_update_sb(struct mddev *mddev, int force_change)
2217 struct md_rdev *rdev;
2220 int any_badblocks_changed = 0;
2224 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2228 /* First make sure individual recovery_offsets are correct */
2229 rdev_for_each(rdev, mddev) {
2230 if (rdev->raid_disk >= 0 &&
2231 mddev->delta_disks >= 0 &&
2232 !test_bit(In_sync, &rdev->flags) &&
2233 mddev->curr_resync_completed > rdev->recovery_offset)
2234 rdev->recovery_offset = mddev->curr_resync_completed;
2237 if (!mddev->persistent) {
2238 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2239 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2240 if (!mddev->external) {
2241 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2242 rdev_for_each(rdev, mddev) {
2243 if (rdev->badblocks.changed) {
2244 rdev->badblocks.changed = 0;
2245 md_ack_all_badblocks(&rdev->badblocks);
2246 md_error(mddev, rdev);
2248 clear_bit(Blocked, &rdev->flags);
2249 clear_bit(BlockedBadBlocks, &rdev->flags);
2250 wake_up(&rdev->blocked_wait);
2253 wake_up(&mddev->sb_wait);
2257 spin_lock(&mddev->lock);
2259 mddev->utime = get_seconds();
2261 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2263 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2264 /* just a clean<-> dirty transition, possibly leave spares alone,
2265 * though if events isn't the right even/odd, we will have to do
2271 if (mddev->degraded)
2272 /* If the array is degraded, then skipping spares is both
2273 * dangerous and fairly pointless.
2274 * Dangerous because a device that was removed from the array
2275 * might have a event_count that still looks up-to-date,
2276 * so it can be re-added without a resync.
2277 * Pointless because if there are any spares to skip,
2278 * then a recovery will happen and soon that array won't
2279 * be degraded any more and the spare can go back to sleep then.
2283 sync_req = mddev->in_sync;
2285 /* If this is just a dirty<->clean transition, and the array is clean
2286 * and 'events' is odd, we can roll back to the previous clean state */
2288 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2289 && mddev->can_decrease_events
2290 && mddev->events != 1) {
2292 mddev->can_decrease_events = 0;
2294 /* otherwise we have to go forward and ... */
2296 mddev->can_decrease_events = nospares;
2300 * This 64-bit counter should never wrap.
2301 * Either we are in around ~1 trillion A.C., assuming
2302 * 1 reboot per second, or we have a bug...
2304 WARN_ON(mddev->events == 0);
2306 rdev_for_each(rdev, mddev) {
2307 if (rdev->badblocks.changed)
2308 any_badblocks_changed++;
2309 if (test_bit(Faulty, &rdev->flags))
2310 set_bit(FaultRecorded, &rdev->flags);
2313 sync_sbs(mddev, nospares);
2314 spin_unlock(&mddev->lock);
2316 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2317 mdname(mddev), mddev->in_sync);
2319 bitmap_update_sb(mddev->bitmap);
2320 rdev_for_each(rdev, mddev) {
2321 char b[BDEVNAME_SIZE];
2323 if (rdev->sb_loaded != 1)
2324 continue; /* no noise on spare devices */
2326 if (!test_bit(Faulty, &rdev->flags)) {
2327 md_super_write(mddev,rdev,
2328 rdev->sb_start, rdev->sb_size,
2330 pr_debug("md: (write) %s's sb offset: %llu\n",
2331 bdevname(rdev->bdev, b),
2332 (unsigned long long)rdev->sb_start);
2333 rdev->sb_events = mddev->events;
2334 if (rdev->badblocks.size) {
2335 md_super_write(mddev, rdev,
2336 rdev->badblocks.sector,
2337 rdev->badblocks.size << 9,
2339 rdev->badblocks.size = 0;
2343 pr_debug("md: %s (skipping faulty)\n",
2344 bdevname(rdev->bdev, b));
2346 if (mddev->level == LEVEL_MULTIPATH)
2347 /* only need to write one superblock... */
2350 md_super_wait(mddev);
2351 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2353 spin_lock(&mddev->lock);
2354 if (mddev->in_sync != sync_req ||
2355 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2356 /* have to write it out again */
2357 spin_unlock(&mddev->lock);
2360 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2361 spin_unlock(&mddev->lock);
2362 wake_up(&mddev->sb_wait);
2363 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2364 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2366 rdev_for_each(rdev, mddev) {
2367 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2368 clear_bit(Blocked, &rdev->flags);
2370 if (any_badblocks_changed)
2371 md_ack_all_badblocks(&rdev->badblocks);
2372 clear_bit(BlockedBadBlocks, &rdev->flags);
2373 wake_up(&rdev->blocked_wait);
2376 EXPORT_SYMBOL(md_update_sb);
2378 /* words written to sysfs files may, or may not, be \n terminated.
2379 * We want to accept with case. For this we use cmd_match.
2381 static int cmd_match(const char *cmd, const char *str)
2383 /* See if cmd, written into a sysfs file, matches
2384 * str. They must either be the same, or cmd can
2385 * have a trailing newline
2387 while (*cmd && *str && *cmd == *str) {
2398 struct rdev_sysfs_entry {
2399 struct attribute attr;
2400 ssize_t (*show)(struct md_rdev *, char *);
2401 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2405 state_show(struct md_rdev *rdev, char *page)
2409 unsigned long flags = ACCESS_ONCE(rdev->flags);
2411 if (test_bit(Faulty, &flags) ||
2412 rdev->badblocks.unacked_exist) {
2413 len+= sprintf(page+len, "%sfaulty",sep);
2416 if (test_bit(In_sync, &flags)) {
2417 len += sprintf(page+len, "%sin_sync",sep);
2420 if (test_bit(WriteMostly, &flags)) {
2421 len += sprintf(page+len, "%swrite_mostly",sep);
2424 if (test_bit(Blocked, &flags) ||
2425 (rdev->badblocks.unacked_exist
2426 && !test_bit(Faulty, &flags))) {
2427 len += sprintf(page+len, "%sblocked", sep);
2430 if (!test_bit(Faulty, &flags) &&
2431 !test_bit(In_sync, &flags)) {
2432 len += sprintf(page+len, "%sspare", sep);
2435 if (test_bit(WriteErrorSeen, &flags)) {
2436 len += sprintf(page+len, "%swrite_error", sep);
2439 if (test_bit(WantReplacement, &flags)) {
2440 len += sprintf(page+len, "%swant_replacement", sep);
2443 if (test_bit(Replacement, &flags)) {
2444 len += sprintf(page+len, "%sreplacement", sep);
2448 return len+sprintf(page+len, "\n");
2452 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2455 * faulty - simulates an error
2456 * remove - disconnects the device
2457 * writemostly - sets write_mostly
2458 * -writemostly - clears write_mostly
2459 * blocked - sets the Blocked flags
2460 * -blocked - clears the Blocked and possibly simulates an error
2461 * insync - sets Insync providing device isn't active
2462 * -insync - clear Insync for a device with a slot assigned,
2463 * so that it gets rebuilt based on bitmap
2464 * write_error - sets WriteErrorSeen
2465 * -write_error - clears WriteErrorSeen
2468 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2469 md_error(rdev->mddev, rdev);
2470 if (test_bit(Faulty, &rdev->flags))
2474 } else if (cmd_match(buf, "remove")) {
2475 if (rdev->raid_disk >= 0)
2478 struct mddev *mddev = rdev->mddev;
2479 if (mddev_is_clustered(mddev))
2480 md_cluster_ops->metadata_update_start(mddev);
2481 md_kick_rdev_from_array(rdev);
2483 md_update_sb(mddev, 1);
2484 md_new_event(mddev);
2485 if (mddev_is_clustered(mddev))
2486 md_cluster_ops->metadata_update_finish(mddev);
2489 } else if (cmd_match(buf, "writemostly")) {
2490 set_bit(WriteMostly, &rdev->flags);
2492 } else if (cmd_match(buf, "-writemostly")) {
2493 clear_bit(WriteMostly, &rdev->flags);
2495 } else if (cmd_match(buf, "blocked")) {
2496 set_bit(Blocked, &rdev->flags);
2498 } else if (cmd_match(buf, "-blocked")) {
2499 if (!test_bit(Faulty, &rdev->flags) &&
2500 rdev->badblocks.unacked_exist) {
2501 /* metadata handler doesn't understand badblocks,
2502 * so we need to fail the device
2504 md_error(rdev->mddev, rdev);
2506 clear_bit(Blocked, &rdev->flags);
2507 clear_bit(BlockedBadBlocks, &rdev->flags);
2508 wake_up(&rdev->blocked_wait);
2509 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2510 md_wakeup_thread(rdev->mddev->thread);
2513 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2514 set_bit(In_sync, &rdev->flags);
2516 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2517 if (rdev->mddev->pers == NULL) {
2518 clear_bit(In_sync, &rdev->flags);
2519 rdev->saved_raid_disk = rdev->raid_disk;
2520 rdev->raid_disk = -1;
2523 } else if (cmd_match(buf, "write_error")) {
2524 set_bit(WriteErrorSeen, &rdev->flags);
2526 } else if (cmd_match(buf, "-write_error")) {
2527 clear_bit(WriteErrorSeen, &rdev->flags);
2529 } else if (cmd_match(buf, "want_replacement")) {
2530 /* Any non-spare device that is not a replacement can
2531 * become want_replacement at any time, but we then need to
2532 * check if recovery is needed.
2534 if (rdev->raid_disk >= 0 &&
2535 !test_bit(Replacement, &rdev->flags))
2536 set_bit(WantReplacement, &rdev->flags);
2537 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2538 md_wakeup_thread(rdev->mddev->thread);
2540 } else if (cmd_match(buf, "-want_replacement")) {
2541 /* Clearing 'want_replacement' is always allowed.
2542 * Once replacements starts it is too late though.
2545 clear_bit(WantReplacement, &rdev->flags);
2546 } else if (cmd_match(buf, "replacement")) {
2547 /* Can only set a device as a replacement when array has not
2548 * yet been started. Once running, replacement is automatic
2549 * from spares, or by assigning 'slot'.
2551 if (rdev->mddev->pers)
2554 set_bit(Replacement, &rdev->flags);
2557 } else if (cmd_match(buf, "-replacement")) {
2558 /* Similarly, can only clear Replacement before start */
2559 if (rdev->mddev->pers)
2562 clear_bit(Replacement, &rdev->flags);
2567 sysfs_notify_dirent_safe(rdev->sysfs_state);
2568 return err ? err : len;
2570 static struct rdev_sysfs_entry rdev_state =
2571 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2574 errors_show(struct md_rdev *rdev, char *page)
2576 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2580 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2583 unsigned long n = simple_strtoul(buf, &e, 10);
2584 if (*buf && (*e == 0 || *e == '\n')) {
2585 atomic_set(&rdev->corrected_errors, n);
2590 static struct rdev_sysfs_entry rdev_errors =
2591 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2594 slot_show(struct md_rdev *rdev, char *page)
2596 if (rdev->raid_disk < 0)
2597 return sprintf(page, "none\n");
2599 return sprintf(page, "%d\n", rdev->raid_disk);
2603 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2607 int slot = simple_strtoul(buf, &e, 10);
2608 if (strncmp(buf, "none", 4)==0)
2610 else if (e==buf || (*e && *e!= '\n'))
2612 if (rdev->mddev->pers && slot == -1) {
2613 /* Setting 'slot' on an active array requires also
2614 * updating the 'rd%d' link, and communicating
2615 * with the personality with ->hot_*_disk.
2616 * For now we only support removing
2617 * failed/spare devices. This normally happens automatically,
2618 * but not when the metadata is externally managed.
2620 if (rdev->raid_disk == -1)
2622 /* personality does all needed checks */
2623 if (rdev->mddev->pers->hot_remove_disk == NULL)
2625 clear_bit(Blocked, &rdev->flags);
2626 remove_and_add_spares(rdev->mddev, rdev);
2627 if (rdev->raid_disk >= 0)
2629 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2630 md_wakeup_thread(rdev->mddev->thread);
2631 } else if (rdev->mddev->pers) {
2632 /* Activating a spare .. or possibly reactivating
2633 * if we ever get bitmaps working here.
2636 if (rdev->raid_disk != -1)
2639 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2642 if (rdev->mddev->pers->hot_add_disk == NULL)
2645 if (slot >= rdev->mddev->raid_disks &&
2646 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2649 rdev->raid_disk = slot;
2650 if (test_bit(In_sync, &rdev->flags))
2651 rdev->saved_raid_disk = slot;
2653 rdev->saved_raid_disk = -1;
2654 clear_bit(In_sync, &rdev->flags);
2655 clear_bit(Bitmap_sync, &rdev->flags);
2656 err = rdev->mddev->pers->
2657 hot_add_disk(rdev->mddev, rdev);
2659 rdev->raid_disk = -1;
2662 sysfs_notify_dirent_safe(rdev->sysfs_state);
2663 if (sysfs_link_rdev(rdev->mddev, rdev))
2664 /* failure here is OK */;
2665 /* don't wakeup anyone, leave that to userspace. */
2667 if (slot >= rdev->mddev->raid_disks &&
2668 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2670 rdev->raid_disk = slot;
2671 /* assume it is working */
2672 clear_bit(Faulty, &rdev->flags);
2673 clear_bit(WriteMostly, &rdev->flags);
2674 set_bit(In_sync, &rdev->flags);
2675 sysfs_notify_dirent_safe(rdev->sysfs_state);
2680 static struct rdev_sysfs_entry rdev_slot =
2681 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2684 offset_show(struct md_rdev *rdev, char *page)
2686 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2690 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2692 unsigned long long offset;
2693 if (kstrtoull(buf, 10, &offset) < 0)
2695 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2697 if (rdev->sectors && rdev->mddev->external)
2698 /* Must set offset before size, so overlap checks
2701 rdev->data_offset = offset;
2702 rdev->new_data_offset = offset;
2706 static struct rdev_sysfs_entry rdev_offset =
2707 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2709 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2711 return sprintf(page, "%llu\n",
2712 (unsigned long long)rdev->new_data_offset);
2715 static ssize_t new_offset_store(struct md_rdev *rdev,
2716 const char *buf, size_t len)
2718 unsigned long long new_offset;
2719 struct mddev *mddev = rdev->mddev;
2721 if (kstrtoull(buf, 10, &new_offset) < 0)
2724 if (mddev->sync_thread ||
2725 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2727 if (new_offset == rdev->data_offset)
2728 /* reset is always permitted */
2730 else if (new_offset > rdev->data_offset) {
2731 /* must not push array size beyond rdev_sectors */
2732 if (new_offset - rdev->data_offset
2733 + mddev->dev_sectors > rdev->sectors)
2736 /* Metadata worries about other space details. */
2738 /* decreasing the offset is inconsistent with a backwards
2741 if (new_offset < rdev->data_offset &&
2742 mddev->reshape_backwards)
2744 /* Increasing offset is inconsistent with forwards
2745 * reshape. reshape_direction should be set to
2746 * 'backwards' first.
2748 if (new_offset > rdev->data_offset &&
2749 !mddev->reshape_backwards)
2752 if (mddev->pers && mddev->persistent &&
2753 !super_types[mddev->major_version]
2754 .allow_new_offset(rdev, new_offset))
2756 rdev->new_data_offset = new_offset;
2757 if (new_offset > rdev->data_offset)
2758 mddev->reshape_backwards = 1;
2759 else if (new_offset < rdev->data_offset)
2760 mddev->reshape_backwards = 0;
2764 static struct rdev_sysfs_entry rdev_new_offset =
2765 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2768 rdev_size_show(struct md_rdev *rdev, char *page)
2770 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2773 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2775 /* check if two start/length pairs overlap */
2783 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2785 unsigned long long blocks;
2788 if (kstrtoull(buf, 10, &blocks) < 0)
2791 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2792 return -EINVAL; /* sector conversion overflow */
2795 if (new != blocks * 2)
2796 return -EINVAL; /* unsigned long long to sector_t overflow */
2803 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2805 struct mddev *my_mddev = rdev->mddev;
2806 sector_t oldsectors = rdev->sectors;
2809 if (strict_blocks_to_sectors(buf, §ors) < 0)
2811 if (rdev->data_offset != rdev->new_data_offset)
2812 return -EINVAL; /* too confusing */
2813 if (my_mddev->pers && rdev->raid_disk >= 0) {
2814 if (my_mddev->persistent) {
2815 sectors = super_types[my_mddev->major_version].
2816 rdev_size_change(rdev, sectors);
2819 } else if (!sectors)
2820 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2822 if (!my_mddev->pers->resize)
2823 /* Cannot change size for RAID0 or Linear etc */
2826 if (sectors < my_mddev->dev_sectors)
2827 return -EINVAL; /* component must fit device */
2829 rdev->sectors = sectors;
2830 if (sectors > oldsectors && my_mddev->external) {
2831 /* Need to check that all other rdevs with the same
2832 * ->bdev do not overlap. 'rcu' is sufficient to walk
2833 * the rdev lists safely.
2834 * This check does not provide a hard guarantee, it
2835 * just helps avoid dangerous mistakes.
2837 struct mddev *mddev;
2839 struct list_head *tmp;
2842 for_each_mddev(mddev, tmp) {
2843 struct md_rdev *rdev2;
2845 rdev_for_each(rdev2, mddev)
2846 if (rdev->bdev == rdev2->bdev &&
2848 overlaps(rdev->data_offset, rdev->sectors,
2861 /* Someone else could have slipped in a size
2862 * change here, but doing so is just silly.
2863 * We put oldsectors back because we *know* it is
2864 * safe, and trust userspace not to race with
2867 rdev->sectors = oldsectors;
2874 static struct rdev_sysfs_entry rdev_size =
2875 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2877 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2879 unsigned long long recovery_start = rdev->recovery_offset;
2881 if (test_bit(In_sync, &rdev->flags) ||
2882 recovery_start == MaxSector)
2883 return sprintf(page, "none\n");
2885 return sprintf(page, "%llu\n", recovery_start);
2888 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2890 unsigned long long recovery_start;
2892 if (cmd_match(buf, "none"))
2893 recovery_start = MaxSector;
2894 else if (kstrtoull(buf, 10, &recovery_start))
2897 if (rdev->mddev->pers &&
2898 rdev->raid_disk >= 0)
2901 rdev->recovery_offset = recovery_start;
2902 if (recovery_start == MaxSector)
2903 set_bit(In_sync, &rdev->flags);
2905 clear_bit(In_sync, &rdev->flags);
2909 static struct rdev_sysfs_entry rdev_recovery_start =
2910 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2913 badblocks_show(struct badblocks *bb, char *page, int unack);
2915 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2917 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2919 return badblocks_show(&rdev->badblocks, page, 0);
2921 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2923 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2924 /* Maybe that ack was all we needed */
2925 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2926 wake_up(&rdev->blocked_wait);
2929 static struct rdev_sysfs_entry rdev_bad_blocks =
2930 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2932 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2934 return badblocks_show(&rdev->badblocks, page, 1);
2936 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2938 return badblocks_store(&rdev->badblocks, page, len, 1);
2940 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2941 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2943 static struct attribute *rdev_default_attrs[] = {
2948 &rdev_new_offset.attr,
2950 &rdev_recovery_start.attr,
2951 &rdev_bad_blocks.attr,
2952 &rdev_unack_bad_blocks.attr,
2956 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2958 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2959 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2965 return entry->show(rdev, page);
2969 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2970 const char *page, size_t length)
2972 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2973 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
2975 struct mddev *mddev = rdev->mddev;
2979 if (!capable(CAP_SYS_ADMIN))
2981 rv = mddev ? mddev_lock(mddev): -EBUSY;
2983 if (rdev->mddev == NULL)
2986 rv = entry->store(rdev, page, length);
2987 mddev_unlock(mddev);
2992 static void rdev_free(struct kobject *ko)
2994 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
2997 static const struct sysfs_ops rdev_sysfs_ops = {
2998 .show = rdev_attr_show,
2999 .store = rdev_attr_store,
3001 static struct kobj_type rdev_ktype = {
3002 .release = rdev_free,
3003 .sysfs_ops = &rdev_sysfs_ops,
3004 .default_attrs = rdev_default_attrs,
3007 int md_rdev_init(struct md_rdev *rdev)
3010 rdev->saved_raid_disk = -1;
3011 rdev->raid_disk = -1;
3013 rdev->data_offset = 0;
3014 rdev->new_data_offset = 0;
3015 rdev->sb_events = 0;
3016 rdev->last_read_error.tv_sec = 0;
3017 rdev->last_read_error.tv_nsec = 0;
3018 rdev->sb_loaded = 0;
3019 rdev->bb_page = NULL;
3020 atomic_set(&rdev->nr_pending, 0);
3021 atomic_set(&rdev->read_errors, 0);
3022 atomic_set(&rdev->corrected_errors, 0);
3024 INIT_LIST_HEAD(&rdev->same_set);
3025 init_waitqueue_head(&rdev->blocked_wait);
3027 /* Add space to store bad block list.
3028 * This reserves the space even on arrays where it cannot
3029 * be used - I wonder if that matters
3031 rdev->badblocks.count = 0;
3032 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3033 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3034 seqlock_init(&rdev->badblocks.lock);
3035 if (rdev->badblocks.page == NULL)
3040 EXPORT_SYMBOL_GPL(md_rdev_init);
3042 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3044 * mark the device faulty if:
3046 * - the device is nonexistent (zero size)
3047 * - the device has no valid superblock
3049 * a faulty rdev _never_ has rdev->sb set.
3051 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3053 char b[BDEVNAME_SIZE];
3055 struct md_rdev *rdev;
3058 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3060 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3061 return ERR_PTR(-ENOMEM);
3064 err = md_rdev_init(rdev);
3067 err = alloc_disk_sb(rdev);
3071 err = lock_rdev(rdev, newdev, super_format == -2);
3075 kobject_init(&rdev->kobj, &rdev_ktype);
3077 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3080 "md: %s has zero or unknown size, marking faulty!\n",
3081 bdevname(rdev->bdev,b));
3086 if (super_format >= 0) {
3087 err = super_types[super_format].
3088 load_super(rdev, NULL, super_minor);
3089 if (err == -EINVAL) {
3091 "md: %s does not have a valid v%d.%d "
3092 "superblock, not importing!\n",
3093 bdevname(rdev->bdev,b),
3094 super_format, super_minor);
3099 "md: could not read %s's sb, not importing!\n",
3100 bdevname(rdev->bdev,b));
3110 md_rdev_clear(rdev);
3112 return ERR_PTR(err);
3116 * Check a full RAID array for plausibility
3119 static void analyze_sbs(struct mddev *mddev)
3122 struct md_rdev *rdev, *freshest, *tmp;
3123 char b[BDEVNAME_SIZE];
3126 rdev_for_each_safe(rdev, tmp, mddev)
3127 switch (super_types[mddev->major_version].
3128 load_super(rdev, freshest, mddev->minor_version)) {
3136 "md: fatal superblock inconsistency in %s"
3137 " -- removing from array\n",
3138 bdevname(rdev->bdev,b));
3139 md_kick_rdev_from_array(rdev);
3142 super_types[mddev->major_version].
3143 validate_super(mddev, freshest);
3146 rdev_for_each_safe(rdev, tmp, mddev) {
3147 if (mddev->max_disks &&
3148 (rdev->desc_nr >= mddev->max_disks ||
3149 i > mddev->max_disks)) {
3151 "md: %s: %s: only %d devices permitted\n",
3152 mdname(mddev), bdevname(rdev->bdev, b),
3154 md_kick_rdev_from_array(rdev);
3157 if (rdev != freshest) {
3158 if (super_types[mddev->major_version].
3159 validate_super(mddev, rdev)) {
3160 printk(KERN_WARNING "md: kicking non-fresh %s"
3162 bdevname(rdev->bdev,b));
3163 md_kick_rdev_from_array(rdev);
3166 /* No device should have a Candidate flag
3167 * when reading devices
3169 if (test_bit(Candidate, &rdev->flags)) {
3170 pr_info("md: kicking Cluster Candidate %s from array!\n",
3171 bdevname(rdev->bdev, b));
3172 md_kick_rdev_from_array(rdev);
3175 if (mddev->level == LEVEL_MULTIPATH) {
3176 rdev->desc_nr = i++;
3177 rdev->raid_disk = rdev->desc_nr;
3178 set_bit(In_sync, &rdev->flags);
3179 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3180 rdev->raid_disk = -1;
3181 clear_bit(In_sync, &rdev->flags);
3186 /* Read a fixed-point number.
3187 * Numbers in sysfs attributes should be in "standard" units where
3188 * possible, so time should be in seconds.
3189 * However we internally use a a much smaller unit such as
3190 * milliseconds or jiffies.
3191 * This function takes a decimal number with a possible fractional
3192 * component, and produces an integer which is the result of
3193 * multiplying that number by 10^'scale'.
3194 * all without any floating-point arithmetic.
3196 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3198 unsigned long result = 0;
3200 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3203 else if (decimals < scale) {
3206 result = result * 10 + value;
3218 while (decimals < scale) {
3226 static void md_safemode_timeout(unsigned long data);
3229 safe_delay_show(struct mddev *mddev, char *page)
3231 int msec = (mddev->safemode_delay*1000)/HZ;
3232 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3235 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3239 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3242 mddev->safemode_delay = 0;
3244 unsigned long old_delay = mddev->safemode_delay;
3245 unsigned long new_delay = (msec*HZ)/1000;
3249 mddev->safemode_delay = new_delay;
3250 if (new_delay < old_delay || old_delay == 0)
3251 mod_timer(&mddev->safemode_timer, jiffies+1);
3255 static struct md_sysfs_entry md_safe_delay =
3256 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3259 level_show(struct mddev *mddev, char *page)
3261 struct md_personality *p;
3263 spin_lock(&mddev->lock);
3266 ret = sprintf(page, "%s\n", p->name);
3267 else if (mddev->clevel[0])
3268 ret = sprintf(page, "%s\n", mddev->clevel);
3269 else if (mddev->level != LEVEL_NONE)
3270 ret = sprintf(page, "%d\n", mddev->level);
3273 spin_unlock(&mddev->lock);
3278 level_store(struct mddev *mddev, const char *buf, size_t len)
3283 struct md_personality *pers, *oldpers;
3285 void *priv, *oldpriv;
3286 struct md_rdev *rdev;
3288 if (slen == 0 || slen >= sizeof(clevel))
3291 rv = mddev_lock(mddev);
3295 if (mddev->pers == NULL) {
3296 strncpy(mddev->clevel, buf, slen);
3297 if (mddev->clevel[slen-1] == '\n')
3299 mddev->clevel[slen] = 0;
3300 mddev->level = LEVEL_NONE;
3308 /* request to change the personality. Need to ensure:
3309 * - array is not engaged in resync/recovery/reshape
3310 * - old personality can be suspended
3311 * - new personality will access other array.
3315 if (mddev->sync_thread ||
3316 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3317 mddev->reshape_position != MaxSector ||
3318 mddev->sysfs_active)
3322 if (!mddev->pers->quiesce) {
3323 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3324 mdname(mddev), mddev->pers->name);
3328 /* Now find the new personality */
3329 strncpy(clevel, buf, slen);
3330 if (clevel[slen-1] == '\n')
3333 if (kstrtol(clevel, 10, &level))
3336 if (request_module("md-%s", clevel) != 0)
3337 request_module("md-level-%s", clevel);
3338 spin_lock(&pers_lock);
3339 pers = find_pers(level, clevel);
3340 if (!pers || !try_module_get(pers->owner)) {
3341 spin_unlock(&pers_lock);
3342 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3346 spin_unlock(&pers_lock);
3348 if (pers == mddev->pers) {
3349 /* Nothing to do! */
3350 module_put(pers->owner);
3354 if (!pers->takeover) {
3355 module_put(pers->owner);
3356 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3357 mdname(mddev), clevel);
3362 rdev_for_each(rdev, mddev)
3363 rdev->new_raid_disk = rdev->raid_disk;
3365 /* ->takeover must set new_* and/or delta_disks
3366 * if it succeeds, and may set them when it fails.
3368 priv = pers->takeover(mddev);
3370 mddev->new_level = mddev->level;
3371 mddev->new_layout = mddev->layout;
3372 mddev->new_chunk_sectors = mddev->chunk_sectors;
3373 mddev->raid_disks -= mddev->delta_disks;
3374 mddev->delta_disks = 0;
3375 mddev->reshape_backwards = 0;
3376 module_put(pers->owner);
3377 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3378 mdname(mddev), clevel);
3383 /* Looks like we have a winner */
3384 mddev_suspend(mddev);
3385 mddev_detach(mddev);
3387 spin_lock(&mddev->lock);
3388 oldpers = mddev->pers;
3389 oldpriv = mddev->private;
3391 mddev->private = priv;
3392 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3393 mddev->level = mddev->new_level;
3394 mddev->layout = mddev->new_layout;
3395 mddev->chunk_sectors = mddev->new_chunk_sectors;
3396 mddev->delta_disks = 0;
3397 mddev->reshape_backwards = 0;
3398 mddev->degraded = 0;
3399 spin_unlock(&mddev->lock);
3401 if (oldpers->sync_request == NULL &&
3403 /* We are converting from a no-redundancy array
3404 * to a redundancy array and metadata is managed
3405 * externally so we need to be sure that writes
3406 * won't block due to a need to transition
3408 * until external management is started.
3411 mddev->safemode_delay = 0;
3412 mddev->safemode = 0;
3415 oldpers->free(mddev, oldpriv);
3417 if (oldpers->sync_request == NULL &&
3418 pers->sync_request != NULL) {
3419 /* need to add the md_redundancy_group */
3420 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3422 "md: cannot register extra attributes for %s\n",
3424 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3426 if (oldpers->sync_request != NULL &&
3427 pers->sync_request == NULL) {
3428 /* need to remove the md_redundancy_group */
3429 if (mddev->to_remove == NULL)
3430 mddev->to_remove = &md_redundancy_group;
3433 rdev_for_each(rdev, mddev) {
3434 if (rdev->raid_disk < 0)
3436 if (rdev->new_raid_disk >= mddev->raid_disks)
3437 rdev->new_raid_disk = -1;
3438 if (rdev->new_raid_disk == rdev->raid_disk)
3440 sysfs_unlink_rdev(mddev, rdev);
3442 rdev_for_each(rdev, mddev) {
3443 if (rdev->raid_disk < 0)
3445 if (rdev->new_raid_disk == rdev->raid_disk)
3447 rdev->raid_disk = rdev->new_raid_disk;
3448 if (rdev->raid_disk < 0)
3449 clear_bit(In_sync, &rdev->flags);
3451 if (sysfs_link_rdev(mddev, rdev))
3452 printk(KERN_WARNING "md: cannot register rd%d"
3453 " for %s after level change\n",
3454 rdev->raid_disk, mdname(mddev));
3458 if (pers->sync_request == NULL) {
3459 /* this is now an array without redundancy, so
3460 * it must always be in_sync
3463 del_timer_sync(&mddev->safemode_timer);
3465 blk_set_stacking_limits(&mddev->queue->limits);
3467 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3468 mddev_resume(mddev);
3470 md_update_sb(mddev, 1);
3471 sysfs_notify(&mddev->kobj, NULL, "level");
3472 md_new_event(mddev);
3475 mddev_unlock(mddev);
3479 static struct md_sysfs_entry md_level =
3480 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3483 layout_show(struct mddev *mddev, char *page)
3485 /* just a number, not meaningful for all levels */
3486 if (mddev->reshape_position != MaxSector &&
3487 mddev->layout != mddev->new_layout)
3488 return sprintf(page, "%d (%d)\n",
3489 mddev->new_layout, mddev->layout);
3490 return sprintf(page, "%d\n", mddev->layout);
3494 layout_store(struct mddev *mddev, const char *buf, size_t len)
3497 unsigned long n = simple_strtoul(buf, &e, 10);
3500 if (!*buf || (*e && *e != '\n'))
3502 err = mddev_lock(mddev);
3507 if (mddev->pers->check_reshape == NULL)
3512 mddev->new_layout = n;
3513 err = mddev->pers->check_reshape(mddev);
3515 mddev->new_layout = mddev->layout;
3518 mddev->new_layout = n;
3519 if (mddev->reshape_position == MaxSector)
3522 mddev_unlock(mddev);
3525 static struct md_sysfs_entry md_layout =
3526 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3529 raid_disks_show(struct mddev *mddev, char *page)
3531 if (mddev->raid_disks == 0)
3533 if (mddev->reshape_position != MaxSector &&
3534 mddev->delta_disks != 0)
3535 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3536 mddev->raid_disks - mddev->delta_disks);
3537 return sprintf(page, "%d\n", mddev->raid_disks);
3540 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3543 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3547 unsigned long n = simple_strtoul(buf, &e, 10);
3549 if (!*buf || (*e && *e != '\n'))
3552 err = mddev_lock(mddev);
3556 err = update_raid_disks(mddev, n);
3557 else if (mddev->reshape_position != MaxSector) {
3558 struct md_rdev *rdev;
3559 int olddisks = mddev->raid_disks - mddev->delta_disks;
3562 rdev_for_each(rdev, mddev) {
3564 rdev->data_offset < rdev->new_data_offset)
3567 rdev->data_offset > rdev->new_data_offset)
3571 mddev->delta_disks = n - olddisks;
3572 mddev->raid_disks = n;
3573 mddev->reshape_backwards = (mddev->delta_disks < 0);
3575 mddev->raid_disks = n;
3577 mddev_unlock(mddev);
3578 return err ? err : len;
3580 static struct md_sysfs_entry md_raid_disks =
3581 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3584 chunk_size_show(struct mddev *mddev, char *page)
3586 if (mddev->reshape_position != MaxSector &&
3587 mddev->chunk_sectors != mddev->new_chunk_sectors)
3588 return sprintf(page, "%d (%d)\n",
3589 mddev->new_chunk_sectors << 9,
3590 mddev->chunk_sectors << 9);
3591 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3595 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3599 unsigned long n = simple_strtoul(buf, &e, 10);
3601 if (!*buf || (*e && *e != '\n'))
3604 err = mddev_lock(mddev);
3608 if (mddev->pers->check_reshape == NULL)
3613 mddev->new_chunk_sectors = n >> 9;
3614 err = mddev->pers->check_reshape(mddev);
3616 mddev->new_chunk_sectors = mddev->chunk_sectors;
3619 mddev->new_chunk_sectors = n >> 9;
3620 if (mddev->reshape_position == MaxSector)
3621 mddev->chunk_sectors = n >> 9;
3623 mddev_unlock(mddev);
3626 static struct md_sysfs_entry md_chunk_size =
3627 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3630 resync_start_show(struct mddev *mddev, char *page)
3632 if (mddev->recovery_cp == MaxSector)
3633 return sprintf(page, "none\n");
3634 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3638 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3642 unsigned long long n = simple_strtoull(buf, &e, 10);
3644 err = mddev_lock(mddev);
3647 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3649 else if (cmd_match(buf, "none"))
3651 else if (!*buf || (*e && *e != '\n'))
3655 mddev->recovery_cp = n;
3657 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3659 mddev_unlock(mddev);
3662 static struct md_sysfs_entry md_resync_start =
3663 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3666 * The array state can be:
3669 * No devices, no size, no level
3670 * Equivalent to STOP_ARRAY ioctl
3672 * May have some settings, but array is not active
3673 * all IO results in error
3674 * When written, doesn't tear down array, but just stops it
3675 * suspended (not supported yet)
3676 * All IO requests will block. The array can be reconfigured.
3677 * Writing this, if accepted, will block until array is quiescent
3679 * no resync can happen. no superblocks get written.
3680 * write requests fail
3682 * like readonly, but behaves like 'clean' on a write request.
3684 * clean - no pending writes, but otherwise active.
3685 * When written to inactive array, starts without resync
3686 * If a write request arrives then
3687 * if metadata is known, mark 'dirty' and switch to 'active'.
3688 * if not known, block and switch to write-pending
3689 * If written to an active array that has pending writes, then fails.
3691 * fully active: IO and resync can be happening.
3692 * When written to inactive array, starts with resync
3695 * clean, but writes are blocked waiting for 'active' to be written.
3698 * like active, but no writes have been seen for a while (100msec).
3701 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3702 write_pending, active_idle, bad_word};
3703 static char *array_states[] = {
3704 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3705 "write-pending", "active-idle", NULL };
3707 static int match_word(const char *word, char **list)
3710 for (n=0; list[n]; n++)
3711 if (cmd_match(word, list[n]))
3717 array_state_show(struct mddev *mddev, char *page)
3719 enum array_state st = inactive;
3732 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3734 else if (mddev->safemode)
3740 if (list_empty(&mddev->disks) &&
3741 mddev->raid_disks == 0 &&
3742 mddev->dev_sectors == 0)
3747 return sprintf(page, "%s\n", array_states[st]);
3750 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3751 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3752 static int do_md_run(struct mddev *mddev);
3753 static int restart_array(struct mddev *mddev);
3756 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3759 enum array_state st = match_word(buf, array_states);
3761 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3762 /* don't take reconfig_mutex when toggling between
3765 spin_lock(&mddev->lock);
3767 restart_array(mddev);
3768 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3769 wake_up(&mddev->sb_wait);
3771 } else /* st == clean */ {
3772 restart_array(mddev);
3773 if (atomic_read(&mddev->writes_pending) == 0) {
3774 if (mddev->in_sync == 0) {
3776 if (mddev->safemode == 1)
3777 mddev->safemode = 0;
3778 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3784 spin_unlock(&mddev->lock);
3787 err = mddev_lock(mddev);
3795 /* stopping an active array */
3796 err = do_md_stop(mddev, 0, NULL);
3799 /* stopping an active array */
3801 err = do_md_stop(mddev, 2, NULL);
3803 err = 0; /* already inactive */
3806 break; /* not supported yet */
3809 err = md_set_readonly(mddev, NULL);
3812 set_disk_ro(mddev->gendisk, 1);
3813 err = do_md_run(mddev);
3819 err = md_set_readonly(mddev, NULL);
3820 else if (mddev->ro == 1)
3821 err = restart_array(mddev);
3824 set_disk_ro(mddev->gendisk, 0);
3828 err = do_md_run(mddev);
3833 restart_array(mddev);
3834 spin_lock(&mddev->lock);
3835 if (atomic_read(&mddev->writes_pending) == 0) {
3836 if (mddev->in_sync == 0) {
3838 if (mddev->safemode == 1)
3839 mddev->safemode = 0;
3840 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3845 spin_unlock(&mddev->lock);
3851 restart_array(mddev);
3852 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3853 wake_up(&mddev->sb_wait);
3857 set_disk_ro(mddev->gendisk, 0);
3858 err = do_md_run(mddev);
3863 /* these cannot be set */
3868 if (mddev->hold_active == UNTIL_IOCTL)
3869 mddev->hold_active = 0;
3870 sysfs_notify_dirent_safe(mddev->sysfs_state);
3872 mddev_unlock(mddev);
3875 static struct md_sysfs_entry md_array_state =
3876 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3879 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3880 return sprintf(page, "%d\n",
3881 atomic_read(&mddev->max_corr_read_errors));
3885 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3888 unsigned long n = simple_strtoul(buf, &e, 10);
3890 if (*buf && (*e == 0 || *e == '\n')) {
3891 atomic_set(&mddev->max_corr_read_errors, n);
3897 static struct md_sysfs_entry max_corr_read_errors =
3898 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3899 max_corrected_read_errors_store);
3902 null_show(struct mddev *mddev, char *page)
3908 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3910 /* buf must be %d:%d\n? giving major and minor numbers */
3911 /* The new device is added to the array.
3912 * If the array has a persistent superblock, we read the
3913 * superblock to initialise info and check validity.
3914 * Otherwise, only checking done is that in bind_rdev_to_array,
3915 * which mainly checks size.
3918 int major = simple_strtoul(buf, &e, 10);
3921 struct md_rdev *rdev;
3924 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3926 minor = simple_strtoul(e+1, &e, 10);
3927 if (*e && *e != '\n')
3929 dev = MKDEV(major, minor);
3930 if (major != MAJOR(dev) ||
3931 minor != MINOR(dev))
3934 flush_workqueue(md_misc_wq);
3936 err = mddev_lock(mddev);
3939 if (mddev->persistent) {
3940 rdev = md_import_device(dev, mddev->major_version,
3941 mddev->minor_version);
3942 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3943 struct md_rdev *rdev0
3944 = list_entry(mddev->disks.next,
3945 struct md_rdev, same_set);
3946 err = super_types[mddev->major_version]
3947 .load_super(rdev, rdev0, mddev->minor_version);
3951 } else if (mddev->external)
3952 rdev = md_import_device(dev, -2, -1);
3954 rdev = md_import_device(dev, -1, -1);
3957 return PTR_ERR(rdev);
3958 err = bind_rdev_to_array(rdev, mddev);
3962 mddev_unlock(mddev);
3963 return err ? err : len;
3966 static struct md_sysfs_entry md_new_device =
3967 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3970 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
3973 unsigned long chunk, end_chunk;
3976 err = mddev_lock(mddev);
3981 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3983 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3984 if (buf == end) break;
3985 if (*end == '-') { /* range */
3987 end_chunk = simple_strtoul(buf, &end, 0);
3988 if (buf == end) break;
3990 if (*end && !isspace(*end)) break;
3991 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3992 buf = skip_spaces(end);
3994 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3996 mddev_unlock(mddev);
4000 static struct md_sysfs_entry md_bitmap =
4001 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4004 size_show(struct mddev *mddev, char *page)
4006 return sprintf(page, "%llu\n",
4007 (unsigned long long)mddev->dev_sectors / 2);
4010 static int update_size(struct mddev *mddev, sector_t num_sectors);
4013 size_store(struct mddev *mddev, const char *buf, size_t len)
4015 /* If array is inactive, we can reduce the component size, but
4016 * not increase it (except from 0).
4017 * If array is active, we can try an on-line resize
4020 int err = strict_blocks_to_sectors(buf, §ors);
4024 err = mddev_lock(mddev);
4028 if (mddev_is_clustered(mddev))
4029 md_cluster_ops->metadata_update_start(mddev);
4030 err = update_size(mddev, sectors);
4031 md_update_sb(mddev, 1);
4032 if (mddev_is_clustered(mddev))
4033 md_cluster_ops->metadata_update_finish(mddev);
4035 if (mddev->dev_sectors == 0 ||
4036 mddev->dev_sectors > sectors)
4037 mddev->dev_sectors = sectors;
4041 mddev_unlock(mddev);
4042 return err ? err : len;
4045 static struct md_sysfs_entry md_size =
4046 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4048 /* Metadata version.
4050 * 'none' for arrays with no metadata (good luck...)
4051 * 'external' for arrays with externally managed metadata,
4052 * or N.M for internally known formats
4055 metadata_show(struct mddev *mddev, char *page)
4057 if (mddev->persistent)
4058 return sprintf(page, "%d.%d\n",
4059 mddev->major_version, mddev->minor_version);
4060 else if (mddev->external)
4061 return sprintf(page, "external:%s\n", mddev->metadata_type);
4063 return sprintf(page, "none\n");
4067 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4072 /* Changing the details of 'external' metadata is
4073 * always permitted. Otherwise there must be
4074 * no devices attached to the array.
4077 err = mddev_lock(mddev);
4081 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4083 else if (!list_empty(&mddev->disks))
4087 if (cmd_match(buf, "none")) {
4088 mddev->persistent = 0;
4089 mddev->external = 0;
4090 mddev->major_version = 0;
4091 mddev->minor_version = 90;
4094 if (strncmp(buf, "external:", 9) == 0) {
4095 size_t namelen = len-9;
4096 if (namelen >= sizeof(mddev->metadata_type))
4097 namelen = sizeof(mddev->metadata_type)-1;
4098 strncpy(mddev->metadata_type, buf+9, namelen);
4099 mddev->metadata_type[namelen] = 0;
4100 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4101 mddev->metadata_type[--namelen] = 0;
4102 mddev->persistent = 0;
4103 mddev->external = 1;
4104 mddev->major_version = 0;
4105 mddev->minor_version = 90;
4108 major = simple_strtoul(buf, &e, 10);
4110 if (e==buf || *e != '.')
4113 minor = simple_strtoul(buf, &e, 10);
4114 if (e==buf || (*e && *e != '\n') )
4117 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4119 mddev->major_version = major;
4120 mddev->minor_version = minor;
4121 mddev->persistent = 1;
4122 mddev->external = 0;
4125 mddev_unlock(mddev);
4129 static struct md_sysfs_entry md_metadata =
4130 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4133 action_show(struct mddev *mddev, char *page)
4135 char *type = "idle";
4136 unsigned long recovery = mddev->recovery;
4137 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4139 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4140 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4141 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4143 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4144 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4146 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4150 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4153 return sprintf(page, "%s\n", type);
4157 action_store(struct mddev *mddev, const char *page, size_t len)
4159 if (!mddev->pers || !mddev->pers->sync_request)
4162 if (cmd_match(page, "frozen"))
4163 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4165 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4167 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4168 flush_workqueue(md_misc_wq);
4169 if (mddev->sync_thread) {
4170 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4171 if (mddev_lock(mddev) == 0) {
4172 md_reap_sync_thread(mddev);
4173 mddev_unlock(mddev);
4176 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4177 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4179 else if (cmd_match(page, "resync"))
4180 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4181 else if (cmd_match(page, "recover")) {
4182 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4183 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4184 } else if (cmd_match(page, "reshape")) {
4186 if (mddev->pers->start_reshape == NULL)
4188 err = mddev_lock(mddev);
4190 err = mddev->pers->start_reshape(mddev);
4191 mddev_unlock(mddev);
4195 sysfs_notify(&mddev->kobj, NULL, "degraded");
4197 if (cmd_match(page, "check"))
4198 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4199 else if (!cmd_match(page, "repair"))
4201 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4202 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4204 if (mddev->ro == 2) {
4205 /* A write to sync_action is enough to justify
4206 * canceling read-auto mode
4209 md_wakeup_thread(mddev->sync_thread);
4211 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4212 md_wakeup_thread(mddev->thread);
4213 sysfs_notify_dirent_safe(mddev->sysfs_action);
4217 static struct md_sysfs_entry md_scan_mode =
4218 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4221 last_sync_action_show(struct mddev *mddev, char *page)
4223 return sprintf(page, "%s\n", mddev->last_sync_action);
4226 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4229 mismatch_cnt_show(struct mddev *mddev, char *page)
4231 return sprintf(page, "%llu\n",
4232 (unsigned long long)
4233 atomic64_read(&mddev->resync_mismatches));
4236 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4239 sync_min_show(struct mddev *mddev, char *page)
4241 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4242 mddev->sync_speed_min ? "local": "system");
4246 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4250 if (strncmp(buf, "system", 6)==0) {
4251 mddev->sync_speed_min = 0;
4254 min = simple_strtoul(buf, &e, 10);
4255 if (buf == e || (*e && *e != '\n') || min <= 0)
4257 mddev->sync_speed_min = min;
4261 static struct md_sysfs_entry md_sync_min =
4262 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4265 sync_max_show(struct mddev *mddev, char *page)
4267 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4268 mddev->sync_speed_max ? "local": "system");
4272 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4276 if (strncmp(buf, "system", 6)==0) {
4277 mddev->sync_speed_max = 0;
4280 max = simple_strtoul(buf, &e, 10);
4281 if (buf == e || (*e && *e != '\n') || max <= 0)
4283 mddev->sync_speed_max = max;
4287 static struct md_sysfs_entry md_sync_max =
4288 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4291 degraded_show(struct mddev *mddev, char *page)
4293 return sprintf(page, "%d\n", mddev->degraded);
4295 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4298 sync_force_parallel_show(struct mddev *mddev, char *page)
4300 return sprintf(page, "%d\n", mddev->parallel_resync);
4304 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4308 if (kstrtol(buf, 10, &n))
4311 if (n != 0 && n != 1)
4314 mddev->parallel_resync = n;
4316 if (mddev->sync_thread)
4317 wake_up(&resync_wait);
4322 /* force parallel resync, even with shared block devices */
4323 static struct md_sysfs_entry md_sync_force_parallel =
4324 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4325 sync_force_parallel_show, sync_force_parallel_store);
4328 sync_speed_show(struct mddev *mddev, char *page)
4330 unsigned long resync, dt, db;
4331 if (mddev->curr_resync == 0)
4332 return sprintf(page, "none\n");
4333 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4334 dt = (jiffies - mddev->resync_mark) / HZ;
4336 db = resync - mddev->resync_mark_cnt;
4337 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4340 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4343 sync_completed_show(struct mddev *mddev, char *page)
4345 unsigned long long max_sectors, resync;
4347 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4348 return sprintf(page, "none\n");
4350 if (mddev->curr_resync == 1 ||
4351 mddev->curr_resync == 2)
4352 return sprintf(page, "delayed\n");
4354 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4355 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4356 max_sectors = mddev->resync_max_sectors;
4358 max_sectors = mddev->dev_sectors;
4360 resync = mddev->curr_resync_completed;
4361 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4364 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4367 min_sync_show(struct mddev *mddev, char *page)
4369 return sprintf(page, "%llu\n",
4370 (unsigned long long)mddev->resync_min);
4373 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4375 unsigned long long min;
4379 if (kstrtoull(buf, 10, &min))
4382 spin_lock(&mddev->lock);
4384 if (min > mddev->resync_max)
4388 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4391 /* Must be a multiple of chunk_size */
4392 chunk = mddev->chunk_sectors;
4394 sector_t temp = min;
4397 if (sector_div(temp, chunk))
4400 mddev->resync_min = min;
4404 spin_unlock(&mddev->lock);
4408 static struct md_sysfs_entry md_min_sync =
4409 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4412 max_sync_show(struct mddev *mddev, char *page)
4414 if (mddev->resync_max == MaxSector)
4415 return sprintf(page, "max\n");
4417 return sprintf(page, "%llu\n",
4418 (unsigned long long)mddev->resync_max);
4421 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4424 spin_lock(&mddev->lock);
4425 if (strncmp(buf, "max", 3) == 0)
4426 mddev->resync_max = MaxSector;
4428 unsigned long long max;
4432 if (kstrtoull(buf, 10, &max))
4434 if (max < mddev->resync_min)
4438 if (max < mddev->resync_max &&
4440 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4443 /* Must be a multiple of chunk_size */
4444 chunk = mddev->chunk_sectors;
4446 sector_t temp = max;
4449 if (sector_div(temp, chunk))
4452 mddev->resync_max = max;
4454 wake_up(&mddev->recovery_wait);
4457 spin_unlock(&mddev->lock);
4461 static struct md_sysfs_entry md_max_sync =
4462 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4465 suspend_lo_show(struct mddev *mddev, char *page)
4467 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4471 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4474 unsigned long long new = simple_strtoull(buf, &e, 10);
4475 unsigned long long old;
4478 if (buf == e || (*e && *e != '\n'))
4481 err = mddev_lock(mddev);
4485 if (mddev->pers == NULL ||
4486 mddev->pers->quiesce == NULL)
4488 old = mddev->suspend_lo;
4489 mddev->suspend_lo = new;
4491 /* Shrinking suspended region */
4492 mddev->pers->quiesce(mddev, 2);
4494 /* Expanding suspended region - need to wait */
4495 mddev->pers->quiesce(mddev, 1);
4496 mddev->pers->quiesce(mddev, 0);
4500 mddev_unlock(mddev);
4503 static struct md_sysfs_entry md_suspend_lo =
4504 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4507 suspend_hi_show(struct mddev *mddev, char *page)
4509 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4513 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4516 unsigned long long new = simple_strtoull(buf, &e, 10);
4517 unsigned long long old;
4520 if (buf == e || (*e && *e != '\n'))
4523 err = mddev_lock(mddev);
4527 if (mddev->pers == NULL ||
4528 mddev->pers->quiesce == NULL)
4530 old = mddev->suspend_hi;
4531 mddev->suspend_hi = new;
4533 /* Shrinking suspended region */
4534 mddev->pers->quiesce(mddev, 2);
4536 /* Expanding suspended region - need to wait */
4537 mddev->pers->quiesce(mddev, 1);
4538 mddev->pers->quiesce(mddev, 0);
4542 mddev_unlock(mddev);
4545 static struct md_sysfs_entry md_suspend_hi =
4546 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4549 reshape_position_show(struct mddev *mddev, char *page)
4551 if (mddev->reshape_position != MaxSector)
4552 return sprintf(page, "%llu\n",
4553 (unsigned long long)mddev->reshape_position);
4554 strcpy(page, "none\n");
4559 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4561 struct md_rdev *rdev;
4564 unsigned long long new = simple_strtoull(buf, &e, 10);
4566 if (buf == e || (*e && *e != '\n'))
4568 err = mddev_lock(mddev);
4574 mddev->reshape_position = new;
4575 mddev->delta_disks = 0;
4576 mddev->reshape_backwards = 0;
4577 mddev->new_level = mddev->level;
4578 mddev->new_layout = mddev->layout;
4579 mddev->new_chunk_sectors = mddev->chunk_sectors;
4580 rdev_for_each(rdev, mddev)
4581 rdev->new_data_offset = rdev->data_offset;
4584 mddev_unlock(mddev);
4588 static struct md_sysfs_entry md_reshape_position =
4589 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4590 reshape_position_store);
4593 reshape_direction_show(struct mddev *mddev, char *page)
4595 return sprintf(page, "%s\n",
4596 mddev->reshape_backwards ? "backwards" : "forwards");
4600 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4605 if (cmd_match(buf, "forwards"))
4607 else if (cmd_match(buf, "backwards"))
4611 if (mddev->reshape_backwards == backwards)
4614 err = mddev_lock(mddev);
4617 /* check if we are allowed to change */
4618 if (mddev->delta_disks)
4620 else if (mddev->persistent &&
4621 mddev->major_version == 0)
4624 mddev->reshape_backwards = backwards;
4625 mddev_unlock(mddev);
4629 static struct md_sysfs_entry md_reshape_direction =
4630 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4631 reshape_direction_store);
4634 array_size_show(struct mddev *mddev, char *page)
4636 if (mddev->external_size)
4637 return sprintf(page, "%llu\n",
4638 (unsigned long long)mddev->array_sectors/2);
4640 return sprintf(page, "default\n");
4644 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4649 err = mddev_lock(mddev);
4653 if (strncmp(buf, "default", 7) == 0) {
4655 sectors = mddev->pers->size(mddev, 0, 0);
4657 sectors = mddev->array_sectors;
4659 mddev->external_size = 0;
4661 if (strict_blocks_to_sectors(buf, §ors) < 0)
4663 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4666 mddev->external_size = 1;
4670 mddev->array_sectors = sectors;
4672 set_capacity(mddev->gendisk, mddev->array_sectors);
4673 revalidate_disk(mddev->gendisk);
4676 mddev_unlock(mddev);
4680 static struct md_sysfs_entry md_array_size =
4681 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4684 static struct attribute *md_default_attrs[] = {
4687 &md_raid_disks.attr,
4688 &md_chunk_size.attr,
4690 &md_resync_start.attr,
4692 &md_new_device.attr,
4693 &md_safe_delay.attr,
4694 &md_array_state.attr,
4695 &md_reshape_position.attr,
4696 &md_reshape_direction.attr,
4697 &md_array_size.attr,
4698 &max_corr_read_errors.attr,
4702 static struct attribute *md_redundancy_attrs[] = {
4704 &md_last_scan_mode.attr,
4705 &md_mismatches.attr,
4708 &md_sync_speed.attr,
4709 &md_sync_force_parallel.attr,
4710 &md_sync_completed.attr,
4713 &md_suspend_lo.attr,
4714 &md_suspend_hi.attr,
4719 static struct attribute_group md_redundancy_group = {
4721 .attrs = md_redundancy_attrs,
4725 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4727 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4728 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4733 spin_lock(&all_mddevs_lock);
4734 if (list_empty(&mddev->all_mddevs)) {
4735 spin_unlock(&all_mddevs_lock);
4739 spin_unlock(&all_mddevs_lock);
4741 rv = entry->show(mddev, page);
4747 md_attr_store(struct kobject *kobj, struct attribute *attr,
4748 const char *page, size_t length)
4750 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4751 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4756 if (!capable(CAP_SYS_ADMIN))
4758 spin_lock(&all_mddevs_lock);
4759 if (list_empty(&mddev->all_mddevs)) {
4760 spin_unlock(&all_mddevs_lock);
4764 spin_unlock(&all_mddevs_lock);
4765 rv = entry->store(mddev, page, length);
4770 static void md_free(struct kobject *ko)
4772 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4774 if (mddev->sysfs_state)
4775 sysfs_put(mddev->sysfs_state);
4777 if (mddev->gendisk) {
4778 del_gendisk(mddev->gendisk);
4779 put_disk(mddev->gendisk);
4782 blk_cleanup_queue(mddev->queue);
4787 static const struct sysfs_ops md_sysfs_ops = {
4788 .show = md_attr_show,
4789 .store = md_attr_store,
4791 static struct kobj_type md_ktype = {
4793 .sysfs_ops = &md_sysfs_ops,
4794 .default_attrs = md_default_attrs,
4799 static void mddev_delayed_delete(struct work_struct *ws)
4801 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4803 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4804 kobject_del(&mddev->kobj);
4805 kobject_put(&mddev->kobj);
4808 static int md_alloc(dev_t dev, char *name)
4810 static DEFINE_MUTEX(disks_mutex);
4811 struct mddev *mddev = mddev_find(dev);
4812 struct gendisk *disk;
4821 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4822 shift = partitioned ? MdpMinorShift : 0;
4823 unit = MINOR(mddev->unit) >> shift;
4825 /* wait for any previous instance of this device to be
4826 * completely removed (mddev_delayed_delete).
4828 flush_workqueue(md_misc_wq);
4830 mutex_lock(&disks_mutex);
4836 /* Need to ensure that 'name' is not a duplicate.
4838 struct mddev *mddev2;
4839 spin_lock(&all_mddevs_lock);
4841 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4842 if (mddev2->gendisk &&
4843 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4844 spin_unlock(&all_mddevs_lock);
4847 spin_unlock(&all_mddevs_lock);
4851 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4854 mddev->queue->queuedata = mddev;
4856 blk_queue_make_request(mddev->queue, md_make_request);
4857 blk_set_stacking_limits(&mddev->queue->limits);
4859 disk = alloc_disk(1 << shift);
4861 blk_cleanup_queue(mddev->queue);
4862 mddev->queue = NULL;
4865 disk->major = MAJOR(mddev->unit);
4866 disk->first_minor = unit << shift;
4868 strcpy(disk->disk_name, name);
4869 else if (partitioned)
4870 sprintf(disk->disk_name, "md_d%d", unit);
4872 sprintf(disk->disk_name, "md%d", unit);
4873 disk->fops = &md_fops;
4874 disk->private_data = mddev;
4875 disk->queue = mddev->queue;
4876 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4877 /* Allow extended partitions. This makes the
4878 * 'mdp' device redundant, but we can't really
4881 disk->flags |= GENHD_FL_EXT_DEVT;
4882 mddev->gendisk = disk;
4883 /* As soon as we call add_disk(), another thread could get
4884 * through to md_open, so make sure it doesn't get too far
4886 mutex_lock(&mddev->open_mutex);
4889 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4890 &disk_to_dev(disk)->kobj, "%s", "md");
4892 /* This isn't possible, but as kobject_init_and_add is marked
4893 * __must_check, we must do something with the result
4895 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4899 if (mddev->kobj.sd &&
4900 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4901 printk(KERN_DEBUG "pointless warning\n");
4902 mutex_unlock(&mddev->open_mutex);
4904 mutex_unlock(&disks_mutex);
4905 if (!error && mddev->kobj.sd) {
4906 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4907 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4913 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4915 md_alloc(dev, NULL);
4919 static int add_named_array(const char *val, struct kernel_param *kp)
4921 /* val must be "md_*" where * is not all digits.
4922 * We allocate an array with a large free minor number, and
4923 * set the name to val. val must not already be an active name.
4925 int len = strlen(val);
4926 char buf[DISK_NAME_LEN];
4928 while (len && val[len-1] == '\n')
4930 if (len >= DISK_NAME_LEN)
4932 strlcpy(buf, val, len+1);
4933 if (strncmp(buf, "md_", 3) != 0)
4935 return md_alloc(0, buf);
4938 static void md_safemode_timeout(unsigned long data)
4940 struct mddev *mddev = (struct mddev *) data;
4942 if (!atomic_read(&mddev->writes_pending)) {
4943 mddev->safemode = 1;
4944 if (mddev->external)
4945 sysfs_notify_dirent_safe(mddev->sysfs_state);
4947 md_wakeup_thread(mddev->thread);
4950 static int start_dirty_degraded;
4952 int md_run(struct mddev *mddev)
4955 struct md_rdev *rdev;
4956 struct md_personality *pers;
4958 if (list_empty(&mddev->disks))
4959 /* cannot run an array with no devices.. */
4964 /* Cannot run until previous stop completes properly */
4965 if (mddev->sysfs_active)
4969 * Analyze all RAID superblock(s)
4971 if (!mddev->raid_disks) {
4972 if (!mddev->persistent)
4977 if (mddev->level != LEVEL_NONE)
4978 request_module("md-level-%d", mddev->level);
4979 else if (mddev->clevel[0])
4980 request_module("md-%s", mddev->clevel);
4983 * Drop all container device buffers, from now on
4984 * the only valid external interface is through the md
4987 rdev_for_each(rdev, mddev) {
4988 if (test_bit(Faulty, &rdev->flags))
4990 sync_blockdev(rdev->bdev);
4991 invalidate_bdev(rdev->bdev);
4993 /* perform some consistency tests on the device.
4994 * We don't want the data to overlap the metadata,
4995 * Internal Bitmap issues have been handled elsewhere.
4997 if (rdev->meta_bdev) {
4998 /* Nothing to check */;
4999 } else if (rdev->data_offset < rdev->sb_start) {
5000 if (mddev->dev_sectors &&
5001 rdev->data_offset + mddev->dev_sectors
5003 printk("md: %s: data overlaps metadata\n",
5008 if (rdev->sb_start + rdev->sb_size/512
5009 > rdev->data_offset) {
5010 printk("md: %s: metadata overlaps data\n",
5015 sysfs_notify_dirent_safe(rdev->sysfs_state);
5018 if (mddev->bio_set == NULL)
5019 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5021 spin_lock(&pers_lock);
5022 pers = find_pers(mddev->level, mddev->clevel);
5023 if (!pers || !try_module_get(pers->owner)) {
5024 spin_unlock(&pers_lock);
5025 if (mddev->level != LEVEL_NONE)
5026 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5029 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5033 spin_unlock(&pers_lock);
5034 if (mddev->level != pers->level) {
5035 mddev->level = pers->level;
5036 mddev->new_level = pers->level;
5038 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5040 if (mddev->reshape_position != MaxSector &&
5041 pers->start_reshape == NULL) {
5042 /* This personality cannot handle reshaping... */
5043 module_put(pers->owner);
5047 if (pers->sync_request) {
5048 /* Warn if this is a potentially silly
5051 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5052 struct md_rdev *rdev2;
5055 rdev_for_each(rdev, mddev)
5056 rdev_for_each(rdev2, mddev) {
5058 rdev->bdev->bd_contains ==
5059 rdev2->bdev->bd_contains) {
5061 "%s: WARNING: %s appears to be"
5062 " on the same physical disk as"
5065 bdevname(rdev->bdev,b),
5066 bdevname(rdev2->bdev,b2));
5073 "True protection against single-disk"
5074 " failure might be compromised.\n");
5077 mddev->recovery = 0;
5078 /* may be over-ridden by personality */
5079 mddev->resync_max_sectors = mddev->dev_sectors;
5081 mddev->ok_start_degraded = start_dirty_degraded;
5083 if (start_readonly && mddev->ro == 0)
5084 mddev->ro = 2; /* read-only, but switch on first write */
5086 err = pers->run(mddev);
5088 printk(KERN_ERR "md: pers->run() failed ...\n");
5089 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5090 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5091 " but 'external_size' not in effect?\n", __func__);
5093 "md: invalid array_size %llu > default size %llu\n",
5094 (unsigned long long)mddev->array_sectors / 2,
5095 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5098 if (err == 0 && pers->sync_request &&
5099 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5100 struct bitmap *bitmap;
5102 bitmap = bitmap_create(mddev, -1);
5103 if (IS_ERR(bitmap)) {
5104 err = PTR_ERR(bitmap);
5105 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5106 mdname(mddev), err);
5108 mddev->bitmap = bitmap;
5112 mddev_detach(mddev);
5113 pers->free(mddev, mddev->private);
5114 module_put(pers->owner);
5115 bitmap_destroy(mddev);
5119 mddev->queue->backing_dev_info.congested_data = mddev;
5120 mddev->queue->backing_dev_info.congested_fn = md_congested;
5121 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5123 if (pers->sync_request) {
5124 if (mddev->kobj.sd &&
5125 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5127 "md: cannot register extra attributes for %s\n",
5129 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5130 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5133 atomic_set(&mddev->writes_pending,0);
5134 atomic_set(&mddev->max_corr_read_errors,
5135 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5136 mddev->safemode = 0;
5137 mddev->safemode_timer.function = md_safemode_timeout;
5138 mddev->safemode_timer.data = (unsigned long) mddev;
5139 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5142 spin_lock(&mddev->lock);
5145 spin_unlock(&mddev->lock);
5146 rdev_for_each(rdev, mddev)
5147 if (rdev->raid_disk >= 0)
5148 if (sysfs_link_rdev(mddev, rdev))
5149 /* failure here is OK */;
5151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5153 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5154 md_update_sb(mddev, 0);
5156 md_new_event(mddev);
5157 sysfs_notify_dirent_safe(mddev->sysfs_state);
5158 sysfs_notify_dirent_safe(mddev->sysfs_action);
5159 sysfs_notify(&mddev->kobj, NULL, "degraded");
5162 EXPORT_SYMBOL_GPL(md_run);
5164 static int do_md_run(struct mddev *mddev)
5168 err = md_run(mddev);
5171 err = bitmap_load(mddev);
5173 bitmap_destroy(mddev);
5177 md_wakeup_thread(mddev->thread);
5178 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5180 set_capacity(mddev->gendisk, mddev->array_sectors);
5181 revalidate_disk(mddev->gendisk);
5183 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5188 static int restart_array(struct mddev *mddev)
5190 struct gendisk *disk = mddev->gendisk;
5192 /* Complain if it has no devices */
5193 if (list_empty(&mddev->disks))
5199 mddev->safemode = 0;
5201 set_disk_ro(disk, 0);
5202 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5204 /* Kick recovery or resync if necessary */
5205 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5206 md_wakeup_thread(mddev->thread);
5207 md_wakeup_thread(mddev->sync_thread);
5208 sysfs_notify_dirent_safe(mddev->sysfs_state);
5212 static void md_clean(struct mddev *mddev)
5214 mddev->array_sectors = 0;
5215 mddev->external_size = 0;
5216 mddev->dev_sectors = 0;
5217 mddev->raid_disks = 0;
5218 mddev->recovery_cp = 0;
5219 mddev->resync_min = 0;
5220 mddev->resync_max = MaxSector;
5221 mddev->reshape_position = MaxSector;
5222 mddev->external = 0;
5223 mddev->persistent = 0;
5224 mddev->level = LEVEL_NONE;
5225 mddev->clevel[0] = 0;
5228 mddev->metadata_type[0] = 0;
5229 mddev->chunk_sectors = 0;
5230 mddev->ctime = mddev->utime = 0;
5232 mddev->max_disks = 0;
5234 mddev->can_decrease_events = 0;
5235 mddev->delta_disks = 0;
5236 mddev->reshape_backwards = 0;
5237 mddev->new_level = LEVEL_NONE;
5238 mddev->new_layout = 0;
5239 mddev->new_chunk_sectors = 0;
5240 mddev->curr_resync = 0;
5241 atomic64_set(&mddev->resync_mismatches, 0);
5242 mddev->suspend_lo = mddev->suspend_hi = 0;
5243 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5244 mddev->recovery = 0;
5247 mddev->degraded = 0;
5248 mddev->safemode = 0;
5249 mddev->merge_check_needed = 0;
5250 mddev->bitmap_info.offset = 0;
5251 mddev->bitmap_info.default_offset = 0;
5252 mddev->bitmap_info.default_space = 0;
5253 mddev->bitmap_info.chunksize = 0;
5254 mddev->bitmap_info.daemon_sleep = 0;
5255 mddev->bitmap_info.max_write_behind = 0;
5258 static void __md_stop_writes(struct mddev *mddev)
5260 if (mddev_is_clustered(mddev))
5261 md_cluster_ops->metadata_update_start(mddev);
5262 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5263 flush_workqueue(md_misc_wq);
5264 if (mddev->sync_thread) {
5265 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5266 md_reap_sync_thread(mddev);
5269 del_timer_sync(&mddev->safemode_timer);
5271 bitmap_flush(mddev);
5272 md_super_wait(mddev);
5274 if (mddev->ro == 0 &&
5275 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5276 /* mark array as shutdown cleanly */
5278 md_update_sb(mddev, 1);
5280 if (mddev_is_clustered(mddev))
5281 md_cluster_ops->metadata_update_finish(mddev);
5284 void md_stop_writes(struct mddev *mddev)
5286 mddev_lock_nointr(mddev);
5287 __md_stop_writes(mddev);
5288 mddev_unlock(mddev);
5290 EXPORT_SYMBOL_GPL(md_stop_writes);
5292 static void mddev_detach(struct mddev *mddev)
5294 struct bitmap *bitmap = mddev->bitmap;
5295 /* wait for behind writes to complete */
5296 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5297 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5299 /* need to kick something here to make sure I/O goes? */
5300 wait_event(bitmap->behind_wait,
5301 atomic_read(&bitmap->behind_writes) == 0);
5303 if (mddev->pers && mddev->pers->quiesce) {
5304 mddev->pers->quiesce(mddev, 1);
5305 mddev->pers->quiesce(mddev, 0);
5307 md_unregister_thread(&mddev->thread);
5309 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5312 static void __md_stop(struct mddev *mddev)
5314 struct md_personality *pers = mddev->pers;
5315 mddev_detach(mddev);
5316 spin_lock(&mddev->lock);
5319 spin_unlock(&mddev->lock);
5320 pers->free(mddev, mddev->private);
5321 if (pers->sync_request && mddev->to_remove == NULL)
5322 mddev->to_remove = &md_redundancy_group;
5323 module_put(pers->owner);
5324 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5327 void md_stop(struct mddev *mddev)
5329 /* stop the array and free an attached data structures.
5330 * This is called from dm-raid
5333 bitmap_destroy(mddev);
5335 bioset_free(mddev->bio_set);
5338 EXPORT_SYMBOL_GPL(md_stop);
5340 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5345 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5347 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5348 md_wakeup_thread(mddev->thread);
5350 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5351 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5352 if (mddev->sync_thread)
5353 /* Thread might be blocked waiting for metadata update
5354 * which will now never happen */
5355 wake_up_process(mddev->sync_thread->tsk);
5357 mddev_unlock(mddev);
5358 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5360 mddev_lock_nointr(mddev);
5362 mutex_lock(&mddev->open_mutex);
5363 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5364 mddev->sync_thread ||
5365 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5366 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5367 printk("md: %s still in use.\n",mdname(mddev));
5369 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5370 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5371 md_wakeup_thread(mddev->thread);
5377 __md_stop_writes(mddev);
5383 set_disk_ro(mddev->gendisk, 1);
5384 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5385 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5386 md_wakeup_thread(mddev->thread);
5387 sysfs_notify_dirent_safe(mddev->sysfs_state);
5391 mutex_unlock(&mddev->open_mutex);
5396 * 0 - completely stop and dis-assemble array
5397 * 2 - stop but do not disassemble array
5399 static int do_md_stop(struct mddev *mddev, int mode,
5400 struct block_device *bdev)
5402 struct gendisk *disk = mddev->gendisk;
5403 struct md_rdev *rdev;
5406 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5408 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5409 md_wakeup_thread(mddev->thread);
5411 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5412 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5413 if (mddev->sync_thread)
5414 /* Thread might be blocked waiting for metadata update
5415 * which will now never happen */
5416 wake_up_process(mddev->sync_thread->tsk);
5418 mddev_unlock(mddev);
5419 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5420 !test_bit(MD_RECOVERY_RUNNING,
5421 &mddev->recovery)));
5422 mddev_lock_nointr(mddev);
5424 mutex_lock(&mddev->open_mutex);
5425 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5426 mddev->sysfs_active ||
5427 mddev->sync_thread ||
5428 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5429 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5430 printk("md: %s still in use.\n",mdname(mddev));
5431 mutex_unlock(&mddev->open_mutex);
5433 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5434 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5435 md_wakeup_thread(mddev->thread);
5441 set_disk_ro(disk, 0);
5443 __md_stop_writes(mddev);
5445 mddev->queue->merge_bvec_fn = NULL;
5446 mddev->queue->backing_dev_info.congested_fn = NULL;
5448 /* tell userspace to handle 'inactive' */
5449 sysfs_notify_dirent_safe(mddev->sysfs_state);
5451 rdev_for_each(rdev, mddev)
5452 if (rdev->raid_disk >= 0)
5453 sysfs_unlink_rdev(mddev, rdev);
5455 set_capacity(disk, 0);
5456 mutex_unlock(&mddev->open_mutex);
5458 revalidate_disk(disk);
5463 mutex_unlock(&mddev->open_mutex);
5465 * Free resources if final stop
5468 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5470 bitmap_destroy(mddev);
5471 if (mddev->bitmap_info.file) {
5472 struct file *f = mddev->bitmap_info.file;
5473 spin_lock(&mddev->lock);
5474 mddev->bitmap_info.file = NULL;
5475 spin_unlock(&mddev->lock);
5478 mddev->bitmap_info.offset = 0;
5480 export_array(mddev);
5483 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5484 if (mddev->hold_active == UNTIL_STOP)
5485 mddev->hold_active = 0;
5487 blk_integrity_unregister(disk);
5488 md_new_event(mddev);
5489 sysfs_notify_dirent_safe(mddev->sysfs_state);
5494 static void autorun_array(struct mddev *mddev)
5496 struct md_rdev *rdev;
5499 if (list_empty(&mddev->disks))
5502 printk(KERN_INFO "md: running: ");
5504 rdev_for_each(rdev, mddev) {
5505 char b[BDEVNAME_SIZE];
5506 printk("<%s>", bdevname(rdev->bdev,b));
5510 err = do_md_run(mddev);
5512 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5513 do_md_stop(mddev, 0, NULL);
5518 * lets try to run arrays based on all disks that have arrived
5519 * until now. (those are in pending_raid_disks)
5521 * the method: pick the first pending disk, collect all disks with
5522 * the same UUID, remove all from the pending list and put them into
5523 * the 'same_array' list. Then order this list based on superblock
5524 * update time (freshest comes first), kick out 'old' disks and
5525 * compare superblocks. If everything's fine then run it.
5527 * If "unit" is allocated, then bump its reference count
5529 static void autorun_devices(int part)
5531 struct md_rdev *rdev0, *rdev, *tmp;
5532 struct mddev *mddev;
5533 char b[BDEVNAME_SIZE];
5535 printk(KERN_INFO "md: autorun ...\n");
5536 while (!list_empty(&pending_raid_disks)) {
5539 LIST_HEAD(candidates);
5540 rdev0 = list_entry(pending_raid_disks.next,
5541 struct md_rdev, same_set);
5543 printk(KERN_INFO "md: considering %s ...\n",
5544 bdevname(rdev0->bdev,b));
5545 INIT_LIST_HEAD(&candidates);
5546 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5547 if (super_90_load(rdev, rdev0, 0) >= 0) {
5548 printk(KERN_INFO "md: adding %s ...\n",
5549 bdevname(rdev->bdev,b));
5550 list_move(&rdev->same_set, &candidates);
5553 * now we have a set of devices, with all of them having
5554 * mostly sane superblocks. It's time to allocate the
5558 dev = MKDEV(mdp_major,
5559 rdev0->preferred_minor << MdpMinorShift);
5560 unit = MINOR(dev) >> MdpMinorShift;
5562 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5565 if (rdev0->preferred_minor != unit) {
5566 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5567 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5571 md_probe(dev, NULL, NULL);
5572 mddev = mddev_find(dev);
5573 if (!mddev || !mddev->gendisk) {
5577 "md: cannot allocate memory for md drive.\n");
5580 if (mddev_lock(mddev))
5581 printk(KERN_WARNING "md: %s locked, cannot run\n",
5583 else if (mddev->raid_disks || mddev->major_version
5584 || !list_empty(&mddev->disks)) {
5586 "md: %s already running, cannot run %s\n",
5587 mdname(mddev), bdevname(rdev0->bdev,b));
5588 mddev_unlock(mddev);
5590 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5591 mddev->persistent = 1;
5592 rdev_for_each_list(rdev, tmp, &candidates) {
5593 list_del_init(&rdev->same_set);
5594 if (bind_rdev_to_array(rdev, mddev))
5597 autorun_array(mddev);
5598 mddev_unlock(mddev);
5600 /* on success, candidates will be empty, on error
5603 rdev_for_each_list(rdev, tmp, &candidates) {
5604 list_del_init(&rdev->same_set);
5609 printk(KERN_INFO "md: ... autorun DONE.\n");
5611 #endif /* !MODULE */
5613 static int get_version(void __user *arg)
5617 ver.major = MD_MAJOR_VERSION;
5618 ver.minor = MD_MINOR_VERSION;
5619 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5621 if (copy_to_user(arg, &ver, sizeof(ver)))
5627 static int get_array_info(struct mddev *mddev, void __user *arg)
5629 mdu_array_info_t info;
5630 int nr,working,insync,failed,spare;
5631 struct md_rdev *rdev;
5633 nr = working = insync = failed = spare = 0;
5635 rdev_for_each_rcu(rdev, mddev) {
5637 if (test_bit(Faulty, &rdev->flags))
5641 if (test_bit(In_sync, &rdev->flags))
5649 info.major_version = mddev->major_version;
5650 info.minor_version = mddev->minor_version;
5651 info.patch_version = MD_PATCHLEVEL_VERSION;
5652 info.ctime = mddev->ctime;
5653 info.level = mddev->level;
5654 info.size = mddev->dev_sectors / 2;
5655 if (info.size != mddev->dev_sectors / 2) /* overflow */
5658 info.raid_disks = mddev->raid_disks;
5659 info.md_minor = mddev->md_minor;
5660 info.not_persistent= !mddev->persistent;
5662 info.utime = mddev->utime;
5665 info.state = (1<<MD_SB_CLEAN);
5666 if (mddev->bitmap && mddev->bitmap_info.offset)
5667 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5668 if (mddev_is_clustered(mddev))
5669 info.state |= (1<<MD_SB_CLUSTERED);
5670 info.active_disks = insync;
5671 info.working_disks = working;
5672 info.failed_disks = failed;
5673 info.spare_disks = spare;
5675 info.layout = mddev->layout;
5676 info.chunk_size = mddev->chunk_sectors << 9;
5678 if (copy_to_user(arg, &info, sizeof(info)))
5684 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5686 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5690 file = kmalloc(sizeof(*file), GFP_NOIO);
5695 spin_lock(&mddev->lock);
5696 /* bitmap disabled, zero the first byte and copy out */
5697 if (!mddev->bitmap_info.file)
5698 file->pathname[0] = '\0';
5699 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5700 file->pathname, sizeof(file->pathname))),
5704 memmove(file->pathname, ptr,
5705 sizeof(file->pathname)-(ptr-file->pathname));
5706 spin_unlock(&mddev->lock);
5709 copy_to_user(arg, file, sizeof(*file)))
5716 static int get_disk_info(struct mddev *mddev, void __user * arg)
5718 mdu_disk_info_t info;
5719 struct md_rdev *rdev;
5721 if (copy_from_user(&info, arg, sizeof(info)))
5725 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5727 info.major = MAJOR(rdev->bdev->bd_dev);
5728 info.minor = MINOR(rdev->bdev->bd_dev);
5729 info.raid_disk = rdev->raid_disk;
5731 if (test_bit(Faulty, &rdev->flags))
5732 info.state |= (1<<MD_DISK_FAULTY);
5733 else if (test_bit(In_sync, &rdev->flags)) {
5734 info.state |= (1<<MD_DISK_ACTIVE);
5735 info.state |= (1<<MD_DISK_SYNC);
5737 if (test_bit(WriteMostly, &rdev->flags))
5738 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5740 info.major = info.minor = 0;
5741 info.raid_disk = -1;
5742 info.state = (1<<MD_DISK_REMOVED);
5746 if (copy_to_user(arg, &info, sizeof(info)))
5752 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5754 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5755 struct md_rdev *rdev;
5756 dev_t dev = MKDEV(info->major,info->minor);
5758 if (mddev_is_clustered(mddev) &&
5759 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5760 pr_err("%s: Cannot add to clustered mddev.\n",
5765 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5768 if (!mddev->raid_disks) {
5770 /* expecting a device which has a superblock */
5771 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5774 "md: md_import_device returned %ld\n",
5776 return PTR_ERR(rdev);
5778 if (!list_empty(&mddev->disks)) {
5779 struct md_rdev *rdev0
5780 = list_entry(mddev->disks.next,
5781 struct md_rdev, same_set);
5782 err = super_types[mddev->major_version]
5783 .load_super(rdev, rdev0, mddev->minor_version);
5786 "md: %s has different UUID to %s\n",
5787 bdevname(rdev->bdev,b),
5788 bdevname(rdev0->bdev,b2));
5793 err = bind_rdev_to_array(rdev, mddev);
5800 * add_new_disk can be used once the array is assembled
5801 * to add "hot spares". They must already have a superblock
5806 if (!mddev->pers->hot_add_disk) {
5808 "%s: personality does not support diskops!\n",
5812 if (mddev->persistent)
5813 rdev = md_import_device(dev, mddev->major_version,
5814 mddev->minor_version);
5816 rdev = md_import_device(dev, -1, -1);
5819 "md: md_import_device returned %ld\n",
5821 return PTR_ERR(rdev);
5823 /* set saved_raid_disk if appropriate */
5824 if (!mddev->persistent) {
5825 if (info->state & (1<<MD_DISK_SYNC) &&
5826 info->raid_disk < mddev->raid_disks) {
5827 rdev->raid_disk = info->raid_disk;
5828 set_bit(In_sync, &rdev->flags);
5829 clear_bit(Bitmap_sync, &rdev->flags);
5831 rdev->raid_disk = -1;
5832 rdev->saved_raid_disk = rdev->raid_disk;
5834 super_types[mddev->major_version].
5835 validate_super(mddev, rdev);
5836 if ((info->state & (1<<MD_DISK_SYNC)) &&
5837 rdev->raid_disk != info->raid_disk) {
5838 /* This was a hot-add request, but events doesn't
5839 * match, so reject it.
5845 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5846 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5847 set_bit(WriteMostly, &rdev->flags);
5849 clear_bit(WriteMostly, &rdev->flags);
5852 * check whether the device shows up in other nodes
5854 if (mddev_is_clustered(mddev)) {
5855 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5856 /* Through --cluster-confirm */
5857 set_bit(Candidate, &rdev->flags);
5858 err = md_cluster_ops->new_disk_ack(mddev, true);
5863 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5864 /* --add initiated by this node */
5865 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5867 md_cluster_ops->add_new_disk_finish(mddev);
5874 rdev->raid_disk = -1;
5875 err = bind_rdev_to_array(rdev, mddev);
5876 if (!err && !mddev->pers->hot_remove_disk) {
5877 /* If there is hot_add_disk but no hot_remove_disk
5878 * then added disks for geometry changes,
5879 * and should be added immediately.
5881 super_types[mddev->major_version].
5882 validate_super(mddev, rdev);
5883 err = mddev->pers->hot_add_disk(mddev, rdev);
5885 unbind_rdev_from_array(rdev);
5890 sysfs_notify_dirent_safe(rdev->sysfs_state);
5892 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5893 if (mddev->degraded)
5894 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5895 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5897 md_new_event(mddev);
5898 md_wakeup_thread(mddev->thread);
5899 if (mddev_is_clustered(mddev) &&
5900 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5901 md_cluster_ops->add_new_disk_finish(mddev);
5905 /* otherwise, add_new_disk is only allowed
5906 * for major_version==0 superblocks
5908 if (mddev->major_version != 0) {
5909 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5914 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5916 rdev = md_import_device(dev, -1, 0);
5919 "md: error, md_import_device() returned %ld\n",
5921 return PTR_ERR(rdev);
5923 rdev->desc_nr = info->number;
5924 if (info->raid_disk < mddev->raid_disks)
5925 rdev->raid_disk = info->raid_disk;
5927 rdev->raid_disk = -1;
5929 if (rdev->raid_disk < mddev->raid_disks)
5930 if (info->state & (1<<MD_DISK_SYNC))
5931 set_bit(In_sync, &rdev->flags);
5933 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5934 set_bit(WriteMostly, &rdev->flags);
5936 if (!mddev->persistent) {
5937 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5938 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5940 rdev->sb_start = calc_dev_sboffset(rdev);
5941 rdev->sectors = rdev->sb_start;
5943 err = bind_rdev_to_array(rdev, mddev);
5953 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5955 char b[BDEVNAME_SIZE];
5956 struct md_rdev *rdev;
5958 rdev = find_rdev(mddev, dev);
5962 if (mddev_is_clustered(mddev))
5963 md_cluster_ops->metadata_update_start(mddev);
5965 clear_bit(Blocked, &rdev->flags);
5966 remove_and_add_spares(mddev, rdev);
5968 if (rdev->raid_disk >= 0)
5971 md_kick_rdev_from_array(rdev);
5972 md_update_sb(mddev, 1);
5973 md_new_event(mddev);
5975 if (mddev_is_clustered(mddev))
5976 md_cluster_ops->metadata_update_finish(mddev);
5980 if (mddev_is_clustered(mddev))
5981 md_cluster_ops->metadata_update_cancel(mddev);
5982 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5983 bdevname(rdev->bdev,b), mdname(mddev));
5987 static int hot_add_disk(struct mddev *mddev, dev_t dev)
5989 char b[BDEVNAME_SIZE];
5991 struct md_rdev *rdev;
5996 if (mddev->major_version != 0) {
5997 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5998 " version-0 superblocks.\n",
6002 if (!mddev->pers->hot_add_disk) {
6004 "%s: personality does not support diskops!\n",
6009 rdev = md_import_device(dev, -1, 0);
6012 "md: error, md_import_device() returned %ld\n",
6017 if (mddev->persistent)
6018 rdev->sb_start = calc_dev_sboffset(rdev);
6020 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6022 rdev->sectors = rdev->sb_start;
6024 if (test_bit(Faulty, &rdev->flags)) {
6026 "md: can not hot-add faulty %s disk to %s!\n",
6027 bdevname(rdev->bdev,b), mdname(mddev));
6032 if (mddev_is_clustered(mddev))
6033 md_cluster_ops->metadata_update_start(mddev);
6034 clear_bit(In_sync, &rdev->flags);
6036 rdev->saved_raid_disk = -1;
6037 err = bind_rdev_to_array(rdev, mddev);
6039 goto abort_clustered;
6042 * The rest should better be atomic, we can have disk failures
6043 * noticed in interrupt contexts ...
6046 rdev->raid_disk = -1;
6048 md_update_sb(mddev, 1);
6050 if (mddev_is_clustered(mddev))
6051 md_cluster_ops->metadata_update_finish(mddev);
6053 * Kick recovery, maybe this spare has to be added to the
6054 * array immediately.
6056 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6057 md_wakeup_thread(mddev->thread);
6058 md_new_event(mddev);
6062 if (mddev_is_clustered(mddev))
6063 md_cluster_ops->metadata_update_cancel(mddev);
6069 static int set_bitmap_file(struct mddev *mddev, int fd)
6074 if (!mddev->pers->quiesce || !mddev->thread)
6076 if (mddev->recovery || mddev->sync_thread)
6078 /* we should be able to change the bitmap.. */
6082 struct inode *inode;
6085 if (mddev->bitmap || mddev->bitmap_info.file)
6086 return -EEXIST; /* cannot add when bitmap is present */
6090 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6095 inode = f->f_mapping->host;
6096 if (!S_ISREG(inode->i_mode)) {
6097 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6100 } else if (!(f->f_mode & FMODE_WRITE)) {
6101 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6104 } else if (atomic_read(&inode->i_writecount) != 1) {
6105 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6113 mddev->bitmap_info.file = f;
6114 mddev->bitmap_info.offset = 0; /* file overrides offset */
6115 } else if (mddev->bitmap == NULL)
6116 return -ENOENT; /* cannot remove what isn't there */
6119 mddev->pers->quiesce(mddev, 1);
6121 struct bitmap *bitmap;
6123 bitmap = bitmap_create(mddev, -1);
6124 if (!IS_ERR(bitmap)) {
6125 mddev->bitmap = bitmap;
6126 err = bitmap_load(mddev);
6128 err = PTR_ERR(bitmap);
6130 if (fd < 0 || err) {
6131 bitmap_destroy(mddev);
6132 fd = -1; /* make sure to put the file */
6134 mddev->pers->quiesce(mddev, 0);
6137 struct file *f = mddev->bitmap_info.file;
6139 spin_lock(&mddev->lock);
6140 mddev->bitmap_info.file = NULL;
6141 spin_unlock(&mddev->lock);
6150 * set_array_info is used two different ways
6151 * The original usage is when creating a new array.
6152 * In this usage, raid_disks is > 0 and it together with
6153 * level, size, not_persistent,layout,chunksize determine the
6154 * shape of the array.
6155 * This will always create an array with a type-0.90.0 superblock.
6156 * The newer usage is when assembling an array.
6157 * In this case raid_disks will be 0, and the major_version field is
6158 * use to determine which style super-blocks are to be found on the devices.
6159 * The minor and patch _version numbers are also kept incase the
6160 * super_block handler wishes to interpret them.
6162 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6165 if (info->raid_disks == 0) {
6166 /* just setting version number for superblock loading */
6167 if (info->major_version < 0 ||
6168 info->major_version >= ARRAY_SIZE(super_types) ||
6169 super_types[info->major_version].name == NULL) {
6170 /* maybe try to auto-load a module? */
6172 "md: superblock version %d not known\n",
6173 info->major_version);
6176 mddev->major_version = info->major_version;
6177 mddev->minor_version = info->minor_version;
6178 mddev->patch_version = info->patch_version;
6179 mddev->persistent = !info->not_persistent;
6180 /* ensure mddev_put doesn't delete this now that there
6181 * is some minimal configuration.
6183 mddev->ctime = get_seconds();
6186 mddev->major_version = MD_MAJOR_VERSION;
6187 mddev->minor_version = MD_MINOR_VERSION;
6188 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6189 mddev->ctime = get_seconds();
6191 mddev->level = info->level;
6192 mddev->clevel[0] = 0;
6193 mddev->dev_sectors = 2 * (sector_t)info->size;
6194 mddev->raid_disks = info->raid_disks;
6195 /* don't set md_minor, it is determined by which /dev/md* was
6198 if (info->state & (1<<MD_SB_CLEAN))
6199 mddev->recovery_cp = MaxSector;
6201 mddev->recovery_cp = 0;
6202 mddev->persistent = ! info->not_persistent;
6203 mddev->external = 0;
6205 mddev->layout = info->layout;
6206 mddev->chunk_sectors = info->chunk_size >> 9;
6208 mddev->max_disks = MD_SB_DISKS;
6210 if (mddev->persistent)
6212 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6214 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6215 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6216 mddev->bitmap_info.offset = 0;
6218 mddev->reshape_position = MaxSector;
6221 * Generate a 128 bit UUID
6223 get_random_bytes(mddev->uuid, 16);
6225 mddev->new_level = mddev->level;
6226 mddev->new_chunk_sectors = mddev->chunk_sectors;
6227 mddev->new_layout = mddev->layout;
6228 mddev->delta_disks = 0;
6229 mddev->reshape_backwards = 0;
6234 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6236 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6238 if (mddev->external_size)
6241 mddev->array_sectors = array_sectors;
6243 EXPORT_SYMBOL(md_set_array_sectors);
6245 static int update_size(struct mddev *mddev, sector_t num_sectors)
6247 struct md_rdev *rdev;
6249 int fit = (num_sectors == 0);
6251 if (mddev->pers->resize == NULL)
6253 /* The "num_sectors" is the number of sectors of each device that
6254 * is used. This can only make sense for arrays with redundancy.
6255 * linear and raid0 always use whatever space is available. We can only
6256 * consider changing this number if no resync or reconstruction is
6257 * happening, and if the new size is acceptable. It must fit before the
6258 * sb_start or, if that is <data_offset, it must fit before the size
6259 * of each device. If num_sectors is zero, we find the largest size
6262 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6268 rdev_for_each(rdev, mddev) {
6269 sector_t avail = rdev->sectors;
6271 if (fit && (num_sectors == 0 || num_sectors > avail))
6272 num_sectors = avail;
6273 if (avail < num_sectors)
6276 rv = mddev->pers->resize(mddev, num_sectors);
6278 revalidate_disk(mddev->gendisk);
6282 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6285 struct md_rdev *rdev;
6286 /* change the number of raid disks */
6287 if (mddev->pers->check_reshape == NULL)
6291 if (raid_disks <= 0 ||
6292 (mddev->max_disks && raid_disks >= mddev->max_disks))
6294 if (mddev->sync_thread ||
6295 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6296 mddev->reshape_position != MaxSector)
6299 rdev_for_each(rdev, mddev) {
6300 if (mddev->raid_disks < raid_disks &&
6301 rdev->data_offset < rdev->new_data_offset)
6303 if (mddev->raid_disks > raid_disks &&
6304 rdev->data_offset > rdev->new_data_offset)
6308 mddev->delta_disks = raid_disks - mddev->raid_disks;
6309 if (mddev->delta_disks < 0)
6310 mddev->reshape_backwards = 1;
6311 else if (mddev->delta_disks > 0)
6312 mddev->reshape_backwards = 0;
6314 rv = mddev->pers->check_reshape(mddev);
6316 mddev->delta_disks = 0;
6317 mddev->reshape_backwards = 0;
6323 * update_array_info is used to change the configuration of an
6325 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6326 * fields in the info are checked against the array.
6327 * Any differences that cannot be handled will cause an error.
6328 * Normally, only one change can be managed at a time.
6330 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6336 /* calculate expected state,ignoring low bits */
6337 if (mddev->bitmap && mddev->bitmap_info.offset)
6338 state |= (1 << MD_SB_BITMAP_PRESENT);
6340 if (mddev->major_version != info->major_version ||
6341 mddev->minor_version != info->minor_version ||
6342 /* mddev->patch_version != info->patch_version || */
6343 mddev->ctime != info->ctime ||
6344 mddev->level != info->level ||
6345 /* mddev->layout != info->layout || */
6346 !mddev->persistent != info->not_persistent||
6347 mddev->chunk_sectors != info->chunk_size >> 9 ||
6348 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6349 ((state^info->state) & 0xfffffe00)
6352 /* Check there is only one change */
6353 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6355 if (mddev->raid_disks != info->raid_disks)
6357 if (mddev->layout != info->layout)
6359 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6366 if (mddev->layout != info->layout) {
6368 * we don't need to do anything at the md level, the
6369 * personality will take care of it all.
6371 if (mddev->pers->check_reshape == NULL)
6374 mddev->new_layout = info->layout;
6375 rv = mddev->pers->check_reshape(mddev);
6377 mddev->new_layout = mddev->layout;
6381 if (mddev_is_clustered(mddev))
6382 md_cluster_ops->metadata_update_start(mddev);
6383 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6384 rv = update_size(mddev, (sector_t)info->size * 2);
6386 if (mddev->raid_disks != info->raid_disks)
6387 rv = update_raid_disks(mddev, info->raid_disks);
6389 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6390 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6394 if (mddev->recovery || mddev->sync_thread) {
6398 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6399 struct bitmap *bitmap;
6400 /* add the bitmap */
6401 if (mddev->bitmap) {
6405 if (mddev->bitmap_info.default_offset == 0) {
6409 mddev->bitmap_info.offset =
6410 mddev->bitmap_info.default_offset;
6411 mddev->bitmap_info.space =
6412 mddev->bitmap_info.default_space;
6413 mddev->pers->quiesce(mddev, 1);
6414 bitmap = bitmap_create(mddev, -1);
6415 if (!IS_ERR(bitmap)) {
6416 mddev->bitmap = bitmap;
6417 rv = bitmap_load(mddev);
6419 rv = PTR_ERR(bitmap);
6421 bitmap_destroy(mddev);
6422 mddev->pers->quiesce(mddev, 0);
6424 /* remove the bitmap */
6425 if (!mddev->bitmap) {
6429 if (mddev->bitmap->storage.file) {
6433 mddev->pers->quiesce(mddev, 1);
6434 bitmap_destroy(mddev);
6435 mddev->pers->quiesce(mddev, 0);
6436 mddev->bitmap_info.offset = 0;
6439 md_update_sb(mddev, 1);
6440 if (mddev_is_clustered(mddev))
6441 md_cluster_ops->metadata_update_finish(mddev);
6444 if (mddev_is_clustered(mddev))
6445 md_cluster_ops->metadata_update_cancel(mddev);
6449 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6451 struct md_rdev *rdev;
6454 if (mddev->pers == NULL)
6458 rdev = find_rdev_rcu(mddev, dev);
6462 md_error(mddev, rdev);
6463 if (!test_bit(Faulty, &rdev->flags))
6471 * We have a problem here : there is no easy way to give a CHS
6472 * virtual geometry. We currently pretend that we have a 2 heads
6473 * 4 sectors (with a BIG number of cylinders...). This drives
6474 * dosfs just mad... ;-)
6476 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6478 struct mddev *mddev = bdev->bd_disk->private_data;
6482 geo->cylinders = mddev->array_sectors / 8;
6486 static inline bool md_ioctl_valid(unsigned int cmd)
6491 case GET_ARRAY_INFO:
6492 case GET_BITMAP_FILE:
6495 case HOT_REMOVE_DISK:
6498 case RESTART_ARRAY_RW:
6500 case SET_ARRAY_INFO:
6501 case SET_BITMAP_FILE:
6502 case SET_DISK_FAULTY:
6505 case CLUSTERED_DISK_NACK:
6512 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6513 unsigned int cmd, unsigned long arg)
6516 void __user *argp = (void __user *)arg;
6517 struct mddev *mddev = NULL;
6520 if (!md_ioctl_valid(cmd))
6525 case GET_ARRAY_INFO:
6529 if (!capable(CAP_SYS_ADMIN))
6534 * Commands dealing with the RAID driver but not any
6539 err = get_version(argp);
6545 autostart_arrays(arg);
6552 * Commands creating/starting a new array:
6555 mddev = bdev->bd_disk->private_data;
6562 /* Some actions do not requires the mutex */
6564 case GET_ARRAY_INFO:
6565 if (!mddev->raid_disks && !mddev->external)
6568 err = get_array_info(mddev, argp);
6572 if (!mddev->raid_disks && !mddev->external)
6575 err = get_disk_info(mddev, argp);
6578 case SET_DISK_FAULTY:
6579 err = set_disk_faulty(mddev, new_decode_dev(arg));
6582 case GET_BITMAP_FILE:
6583 err = get_bitmap_file(mddev, argp);
6588 if (cmd == ADD_NEW_DISK)
6589 /* need to ensure md_delayed_delete() has completed */
6590 flush_workqueue(md_misc_wq);
6592 if (cmd == HOT_REMOVE_DISK)
6593 /* need to ensure recovery thread has run */
6594 wait_event_interruptible_timeout(mddev->sb_wait,
6595 !test_bit(MD_RECOVERY_NEEDED,
6597 msecs_to_jiffies(5000));
6598 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6599 /* Need to flush page cache, and ensure no-one else opens
6602 mutex_lock(&mddev->open_mutex);
6603 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6604 mutex_unlock(&mddev->open_mutex);
6608 set_bit(MD_STILL_CLOSED, &mddev->flags);
6609 mutex_unlock(&mddev->open_mutex);
6610 sync_blockdev(bdev);
6612 err = mddev_lock(mddev);
6615 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6620 if (cmd == SET_ARRAY_INFO) {
6621 mdu_array_info_t info;
6623 memset(&info, 0, sizeof(info));
6624 else if (copy_from_user(&info, argp, sizeof(info))) {
6629 err = update_array_info(mddev, &info);
6631 printk(KERN_WARNING "md: couldn't update"
6632 " array info. %d\n", err);
6637 if (!list_empty(&mddev->disks)) {
6639 "md: array %s already has disks!\n",
6644 if (mddev->raid_disks) {
6646 "md: array %s already initialised!\n",
6651 err = set_array_info(mddev, &info);
6653 printk(KERN_WARNING "md: couldn't set"
6654 " array info. %d\n", err);
6661 * Commands querying/configuring an existing array:
6663 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6664 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6665 if ((!mddev->raid_disks && !mddev->external)
6666 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6667 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6668 && cmd != GET_BITMAP_FILE) {
6674 * Commands even a read-only array can execute:
6677 case RESTART_ARRAY_RW:
6678 err = restart_array(mddev);
6682 err = do_md_stop(mddev, 0, bdev);
6686 err = md_set_readonly(mddev, bdev);
6689 case HOT_REMOVE_DISK:
6690 err = hot_remove_disk(mddev, new_decode_dev(arg));
6694 /* We can support ADD_NEW_DISK on read-only arrays
6695 * on if we are re-adding a preexisting device.
6696 * So require mddev->pers and MD_DISK_SYNC.
6699 mdu_disk_info_t info;
6700 if (copy_from_user(&info, argp, sizeof(info)))
6702 else if (!(info.state & (1<<MD_DISK_SYNC)))
6703 /* Need to clear read-only for this */
6706 err = add_new_disk(mddev, &info);
6712 if (get_user(ro, (int __user *)(arg))) {
6718 /* if the bdev is going readonly the value of mddev->ro
6719 * does not matter, no writes are coming
6724 /* are we are already prepared for writes? */
6728 /* transitioning to readauto need only happen for
6729 * arrays that call md_write_start
6732 err = restart_array(mddev);
6735 set_disk_ro(mddev->gendisk, 0);
6742 * The remaining ioctls are changing the state of the
6743 * superblock, so we do not allow them on read-only arrays.
6745 if (mddev->ro && mddev->pers) {
6746 if (mddev->ro == 2) {
6748 sysfs_notify_dirent_safe(mddev->sysfs_state);
6749 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6750 /* mddev_unlock will wake thread */
6751 /* If a device failed while we were read-only, we
6752 * need to make sure the metadata is updated now.
6754 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6755 mddev_unlock(mddev);
6756 wait_event(mddev->sb_wait,
6757 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6758 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6759 mddev_lock_nointr(mddev);
6770 mdu_disk_info_t info;
6771 if (copy_from_user(&info, argp, sizeof(info)))
6774 err = add_new_disk(mddev, &info);
6778 case CLUSTERED_DISK_NACK:
6779 if (mddev_is_clustered(mddev))
6780 md_cluster_ops->new_disk_ack(mddev, false);
6786 err = hot_add_disk(mddev, new_decode_dev(arg));
6790 err = do_md_run(mddev);
6793 case SET_BITMAP_FILE:
6794 err = set_bitmap_file(mddev, (int)arg);
6803 if (mddev->hold_active == UNTIL_IOCTL &&
6805 mddev->hold_active = 0;
6806 mddev_unlock(mddev);
6810 #ifdef CONFIG_COMPAT
6811 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6812 unsigned int cmd, unsigned long arg)
6815 case HOT_REMOVE_DISK:
6817 case SET_DISK_FAULTY:
6818 case SET_BITMAP_FILE:
6819 /* These take in integer arg, do not convert */
6822 arg = (unsigned long)compat_ptr(arg);
6826 return md_ioctl(bdev, mode, cmd, arg);
6828 #endif /* CONFIG_COMPAT */
6830 static int md_open(struct block_device *bdev, fmode_t mode)
6833 * Succeed if we can lock the mddev, which confirms that
6834 * it isn't being stopped right now.
6836 struct mddev *mddev = mddev_find(bdev->bd_dev);
6842 if (mddev->gendisk != bdev->bd_disk) {
6843 /* we are racing with mddev_put which is discarding this
6847 /* Wait until bdev->bd_disk is definitely gone */
6848 flush_workqueue(md_misc_wq);
6849 /* Then retry the open from the top */
6850 return -ERESTARTSYS;
6852 BUG_ON(mddev != bdev->bd_disk->private_data);
6854 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6858 atomic_inc(&mddev->openers);
6859 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6860 mutex_unlock(&mddev->open_mutex);
6862 check_disk_change(bdev);
6867 static void md_release(struct gendisk *disk, fmode_t mode)
6869 struct mddev *mddev = disk->private_data;
6872 atomic_dec(&mddev->openers);
6876 static int md_media_changed(struct gendisk *disk)
6878 struct mddev *mddev = disk->private_data;
6880 return mddev->changed;
6883 static int md_revalidate(struct gendisk *disk)
6885 struct mddev *mddev = disk->private_data;
6890 static const struct block_device_operations md_fops =
6892 .owner = THIS_MODULE,
6894 .release = md_release,
6896 #ifdef CONFIG_COMPAT
6897 .compat_ioctl = md_compat_ioctl,
6899 .getgeo = md_getgeo,
6900 .media_changed = md_media_changed,
6901 .revalidate_disk= md_revalidate,
6904 static int md_thread(void *arg)
6906 struct md_thread *thread = arg;
6909 * md_thread is a 'system-thread', it's priority should be very
6910 * high. We avoid resource deadlocks individually in each
6911 * raid personality. (RAID5 does preallocation) We also use RR and
6912 * the very same RT priority as kswapd, thus we will never get
6913 * into a priority inversion deadlock.
6915 * we definitely have to have equal or higher priority than
6916 * bdflush, otherwise bdflush will deadlock if there are too
6917 * many dirty RAID5 blocks.
6920 allow_signal(SIGKILL);
6921 while (!kthread_should_stop()) {
6923 /* We need to wait INTERRUPTIBLE so that
6924 * we don't add to the load-average.
6925 * That means we need to be sure no signals are
6928 if (signal_pending(current))
6929 flush_signals(current);
6931 wait_event_interruptible_timeout
6933 test_bit(THREAD_WAKEUP, &thread->flags)
6934 || kthread_should_stop(),
6937 clear_bit(THREAD_WAKEUP, &thread->flags);
6938 if (!kthread_should_stop())
6939 thread->run(thread);
6945 void md_wakeup_thread(struct md_thread *thread)
6948 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6949 set_bit(THREAD_WAKEUP, &thread->flags);
6950 wake_up(&thread->wqueue);
6953 EXPORT_SYMBOL(md_wakeup_thread);
6955 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6956 struct mddev *mddev, const char *name)
6958 struct md_thread *thread;
6960 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6964 init_waitqueue_head(&thread->wqueue);
6967 thread->mddev = mddev;
6968 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6969 thread->tsk = kthread_run(md_thread, thread,
6971 mdname(thread->mddev),
6973 if (IS_ERR(thread->tsk)) {
6979 EXPORT_SYMBOL(md_register_thread);
6981 void md_unregister_thread(struct md_thread **threadp)
6983 struct md_thread *thread = *threadp;
6986 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6987 /* Locking ensures that mddev_unlock does not wake_up a
6988 * non-existent thread
6990 spin_lock(&pers_lock);
6992 spin_unlock(&pers_lock);
6994 kthread_stop(thread->tsk);
6997 EXPORT_SYMBOL(md_unregister_thread);
6999 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7001 if (!rdev || test_bit(Faulty, &rdev->flags))
7004 if (!mddev->pers || !mddev->pers->error_handler)
7006 mddev->pers->error_handler(mddev,rdev);
7007 if (mddev->degraded)
7008 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7009 sysfs_notify_dirent_safe(rdev->sysfs_state);
7010 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7011 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7012 md_wakeup_thread(mddev->thread);
7013 if (mddev->event_work.func)
7014 queue_work(md_misc_wq, &mddev->event_work);
7015 md_new_event_inintr(mddev);
7017 EXPORT_SYMBOL(md_error);
7019 /* seq_file implementation /proc/mdstat */
7021 static void status_unused(struct seq_file *seq)
7024 struct md_rdev *rdev;
7026 seq_printf(seq, "unused devices: ");
7028 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7029 char b[BDEVNAME_SIZE];
7031 seq_printf(seq, "%s ",
7032 bdevname(rdev->bdev,b));
7035 seq_printf(seq, "<none>");
7037 seq_printf(seq, "\n");
7040 static void status_resync(struct seq_file *seq, struct mddev *mddev)
7042 sector_t max_sectors, resync, res;
7043 unsigned long dt, db;
7046 unsigned int per_milli;
7048 if (mddev->curr_resync <= 3)
7051 resync = mddev->curr_resync
7052 - atomic_read(&mddev->recovery_active);
7054 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7055 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7056 max_sectors = mddev->resync_max_sectors;
7058 max_sectors = mddev->dev_sectors;
7060 WARN_ON(max_sectors == 0);
7061 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7062 * in a sector_t, and (max_sectors>>scale) will fit in a
7063 * u32, as those are the requirements for sector_div.
7064 * Thus 'scale' must be at least 10
7067 if (sizeof(sector_t) > sizeof(unsigned long)) {
7068 while ( max_sectors/2 > (1ULL<<(scale+32)))
7071 res = (resync>>scale)*1000;
7072 sector_div(res, (u32)((max_sectors>>scale)+1));
7076 int i, x = per_milli/50, y = 20-x;
7077 seq_printf(seq, "[");
7078 for (i = 0; i < x; i++)
7079 seq_printf(seq, "=");
7080 seq_printf(seq, ">");
7081 for (i = 0; i < y; i++)
7082 seq_printf(seq, ".");
7083 seq_printf(seq, "] ");
7085 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7086 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7088 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7090 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7091 "resync" : "recovery"))),
7092 per_milli/10, per_milli % 10,
7093 (unsigned long long) resync/2,
7094 (unsigned long long) max_sectors/2);
7097 * dt: time from mark until now
7098 * db: blocks written from mark until now
7099 * rt: remaining time
7101 * rt is a sector_t, so could be 32bit or 64bit.
7102 * So we divide before multiply in case it is 32bit and close
7104 * We scale the divisor (db) by 32 to avoid losing precision
7105 * near the end of resync when the number of remaining sectors
7107 * We then divide rt by 32 after multiplying by db to compensate.
7108 * The '+1' avoids division by zero if db is very small.
7110 dt = ((jiffies - mddev->resync_mark) / HZ);
7112 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7113 - mddev->resync_mark_cnt;
7115 rt = max_sectors - resync; /* number of remaining sectors */
7116 sector_div(rt, db/32+1);
7120 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7121 ((unsigned long)rt % 60)/6);
7123 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7126 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7128 struct list_head *tmp;
7130 struct mddev *mddev;
7138 spin_lock(&all_mddevs_lock);
7139 list_for_each(tmp,&all_mddevs)
7141 mddev = list_entry(tmp, struct mddev, all_mddevs);
7143 spin_unlock(&all_mddevs_lock);
7146 spin_unlock(&all_mddevs_lock);
7148 return (void*)2;/* tail */
7152 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7154 struct list_head *tmp;
7155 struct mddev *next_mddev, *mddev = v;
7161 spin_lock(&all_mddevs_lock);
7163 tmp = all_mddevs.next;
7165 tmp = mddev->all_mddevs.next;
7166 if (tmp != &all_mddevs)
7167 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7169 next_mddev = (void*)2;
7172 spin_unlock(&all_mddevs_lock);
7180 static void md_seq_stop(struct seq_file *seq, void *v)
7182 struct mddev *mddev = v;
7184 if (mddev && v != (void*)1 && v != (void*)2)
7188 static int md_seq_show(struct seq_file *seq, void *v)
7190 struct mddev *mddev = v;
7192 struct md_rdev *rdev;
7194 if (v == (void*)1) {
7195 struct md_personality *pers;
7196 seq_printf(seq, "Personalities : ");
7197 spin_lock(&pers_lock);
7198 list_for_each_entry(pers, &pers_list, list)
7199 seq_printf(seq, "[%s] ", pers->name);
7201 spin_unlock(&pers_lock);
7202 seq_printf(seq, "\n");
7203 seq->poll_event = atomic_read(&md_event_count);
7206 if (v == (void*)2) {
7211 spin_lock(&mddev->lock);
7212 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7213 seq_printf(seq, "%s : %sactive", mdname(mddev),
7214 mddev->pers ? "" : "in");
7217 seq_printf(seq, " (read-only)");
7219 seq_printf(seq, " (auto-read-only)");
7220 seq_printf(seq, " %s", mddev->pers->name);
7225 rdev_for_each_rcu(rdev, mddev) {
7226 char b[BDEVNAME_SIZE];
7227 seq_printf(seq, " %s[%d]",
7228 bdevname(rdev->bdev,b), rdev->desc_nr);
7229 if (test_bit(WriteMostly, &rdev->flags))
7230 seq_printf(seq, "(W)");
7231 if (test_bit(Faulty, &rdev->flags)) {
7232 seq_printf(seq, "(F)");
7235 if (rdev->raid_disk < 0)
7236 seq_printf(seq, "(S)"); /* spare */
7237 if (test_bit(Replacement, &rdev->flags))
7238 seq_printf(seq, "(R)");
7239 sectors += rdev->sectors;
7243 if (!list_empty(&mddev->disks)) {
7245 seq_printf(seq, "\n %llu blocks",
7246 (unsigned long long)
7247 mddev->array_sectors / 2);
7249 seq_printf(seq, "\n %llu blocks",
7250 (unsigned long long)sectors / 2);
7252 if (mddev->persistent) {
7253 if (mddev->major_version != 0 ||
7254 mddev->minor_version != 90) {
7255 seq_printf(seq," super %d.%d",
7256 mddev->major_version,
7257 mddev->minor_version);
7259 } else if (mddev->external)
7260 seq_printf(seq, " super external:%s",
7261 mddev->metadata_type);
7263 seq_printf(seq, " super non-persistent");
7266 mddev->pers->status(seq, mddev);
7267 seq_printf(seq, "\n ");
7268 if (mddev->pers->sync_request) {
7269 if (mddev->curr_resync > 2) {
7270 status_resync(seq, mddev);
7271 seq_printf(seq, "\n ");
7272 } else if (mddev->curr_resync >= 1)
7273 seq_printf(seq, "\tresync=DELAYED\n ");
7274 else if (mddev->recovery_cp < MaxSector)
7275 seq_printf(seq, "\tresync=PENDING\n ");
7278 seq_printf(seq, "\n ");
7280 bitmap_status(seq, mddev->bitmap);
7282 seq_printf(seq, "\n");
7284 spin_unlock(&mddev->lock);
7289 static const struct seq_operations md_seq_ops = {
7290 .start = md_seq_start,
7291 .next = md_seq_next,
7292 .stop = md_seq_stop,
7293 .show = md_seq_show,
7296 static int md_seq_open(struct inode *inode, struct file *file)
7298 struct seq_file *seq;
7301 error = seq_open(file, &md_seq_ops);
7305 seq = file->private_data;
7306 seq->poll_event = atomic_read(&md_event_count);
7310 static int md_unloading;
7311 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7313 struct seq_file *seq = filp->private_data;
7317 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7318 poll_wait(filp, &md_event_waiters, wait);
7320 /* always allow read */
7321 mask = POLLIN | POLLRDNORM;
7323 if (seq->poll_event != atomic_read(&md_event_count))
7324 mask |= POLLERR | POLLPRI;
7328 static const struct file_operations md_seq_fops = {
7329 .owner = THIS_MODULE,
7330 .open = md_seq_open,
7332 .llseek = seq_lseek,
7333 .release = seq_release_private,
7334 .poll = mdstat_poll,
7337 int register_md_personality(struct md_personality *p)
7339 printk(KERN_INFO "md: %s personality registered for level %d\n",
7341 spin_lock(&pers_lock);
7342 list_add_tail(&p->list, &pers_list);
7343 spin_unlock(&pers_lock);
7346 EXPORT_SYMBOL(register_md_personality);
7348 int unregister_md_personality(struct md_personality *p)
7350 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7351 spin_lock(&pers_lock);
7352 list_del_init(&p->list);
7353 spin_unlock(&pers_lock);
7356 EXPORT_SYMBOL(unregister_md_personality);
7358 int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7360 if (md_cluster_ops != NULL)
7362 spin_lock(&pers_lock);
7363 md_cluster_ops = ops;
7364 md_cluster_mod = module;
7365 spin_unlock(&pers_lock);
7368 EXPORT_SYMBOL(register_md_cluster_operations);
7370 int unregister_md_cluster_operations(void)
7372 spin_lock(&pers_lock);
7373 md_cluster_ops = NULL;
7374 spin_unlock(&pers_lock);
7377 EXPORT_SYMBOL(unregister_md_cluster_operations);
7379 int md_setup_cluster(struct mddev *mddev, int nodes)
7383 err = request_module("md-cluster");
7385 pr_err("md-cluster module not found.\n");
7389 spin_lock(&pers_lock);
7390 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7391 spin_unlock(&pers_lock);
7394 spin_unlock(&pers_lock);
7396 return md_cluster_ops->join(mddev, nodes);
7399 void md_cluster_stop(struct mddev *mddev)
7401 if (!md_cluster_ops)
7403 md_cluster_ops->leave(mddev);
7404 module_put(md_cluster_mod);
7407 static int is_mddev_idle(struct mddev *mddev, int init)
7409 struct md_rdev *rdev;
7415 rdev_for_each_rcu(rdev, mddev) {
7416 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7417 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7418 (int)part_stat_read(&disk->part0, sectors[1]) -
7419 atomic_read(&disk->sync_io);
7420 /* sync IO will cause sync_io to increase before the disk_stats
7421 * as sync_io is counted when a request starts, and
7422 * disk_stats is counted when it completes.
7423 * So resync activity will cause curr_events to be smaller than
7424 * when there was no such activity.
7425 * non-sync IO will cause disk_stat to increase without
7426 * increasing sync_io so curr_events will (eventually)
7427 * be larger than it was before. Once it becomes
7428 * substantially larger, the test below will cause
7429 * the array to appear non-idle, and resync will slow
7431 * If there is a lot of outstanding resync activity when
7432 * we set last_event to curr_events, then all that activity
7433 * completing might cause the array to appear non-idle
7434 * and resync will be slowed down even though there might
7435 * not have been non-resync activity. This will only
7436 * happen once though. 'last_events' will soon reflect
7437 * the state where there is little or no outstanding
7438 * resync requests, and further resync activity will
7439 * always make curr_events less than last_events.
7442 if (init || curr_events - rdev->last_events > 64) {
7443 rdev->last_events = curr_events;
7451 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7453 /* another "blocks" (512byte) blocks have been synced */
7454 atomic_sub(blocks, &mddev->recovery_active);
7455 wake_up(&mddev->recovery_wait);
7457 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7458 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7459 md_wakeup_thread(mddev->thread);
7460 // stop recovery, signal do_sync ....
7463 EXPORT_SYMBOL(md_done_sync);
7465 /* md_write_start(mddev, bi)
7466 * If we need to update some array metadata (e.g. 'active' flag
7467 * in superblock) before writing, schedule a superblock update
7468 * and wait for it to complete.
7470 void md_write_start(struct mddev *mddev, struct bio *bi)
7473 if (bio_data_dir(bi) != WRITE)
7476 BUG_ON(mddev->ro == 1);
7477 if (mddev->ro == 2) {
7478 /* need to switch to read/write */
7480 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7481 md_wakeup_thread(mddev->thread);
7482 md_wakeup_thread(mddev->sync_thread);
7485 atomic_inc(&mddev->writes_pending);
7486 if (mddev->safemode == 1)
7487 mddev->safemode = 0;
7488 if (mddev->in_sync) {
7489 spin_lock(&mddev->lock);
7490 if (mddev->in_sync) {
7492 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7493 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7494 md_wakeup_thread(mddev->thread);
7497 spin_unlock(&mddev->lock);
7500 sysfs_notify_dirent_safe(mddev->sysfs_state);
7501 wait_event(mddev->sb_wait,
7502 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7504 EXPORT_SYMBOL(md_write_start);
7506 void md_write_end(struct mddev *mddev)
7508 if (atomic_dec_and_test(&mddev->writes_pending)) {
7509 if (mddev->safemode == 2)
7510 md_wakeup_thread(mddev->thread);
7511 else if (mddev->safemode_delay)
7512 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7515 EXPORT_SYMBOL(md_write_end);
7517 /* md_allow_write(mddev)
7518 * Calling this ensures that the array is marked 'active' so that writes
7519 * may proceed without blocking. It is important to call this before
7520 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7521 * Must be called with mddev_lock held.
7523 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7524 * is dropped, so return -EAGAIN after notifying userspace.
7526 int md_allow_write(struct mddev *mddev)
7532 if (!mddev->pers->sync_request)
7535 spin_lock(&mddev->lock);
7536 if (mddev->in_sync) {
7538 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7539 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7540 if (mddev->safemode_delay &&
7541 mddev->safemode == 0)
7542 mddev->safemode = 1;
7543 spin_unlock(&mddev->lock);
7544 if (mddev_is_clustered(mddev))
7545 md_cluster_ops->metadata_update_start(mddev);
7546 md_update_sb(mddev, 0);
7547 if (mddev_is_clustered(mddev))
7548 md_cluster_ops->metadata_update_finish(mddev);
7549 sysfs_notify_dirent_safe(mddev->sysfs_state);
7551 spin_unlock(&mddev->lock);
7553 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7558 EXPORT_SYMBOL_GPL(md_allow_write);
7560 #define SYNC_MARKS 10
7561 #define SYNC_MARK_STEP (3*HZ)
7562 #define UPDATE_FREQUENCY (5*60*HZ)
7563 void md_do_sync(struct md_thread *thread)
7565 struct mddev *mddev = thread->mddev;
7566 struct mddev *mddev2;
7567 unsigned int currspeed = 0,
7569 sector_t max_sectors,j, io_sectors, recovery_done;
7570 unsigned long mark[SYNC_MARKS];
7571 unsigned long update_time;
7572 sector_t mark_cnt[SYNC_MARKS];
7574 struct list_head *tmp;
7575 sector_t last_check;
7577 struct md_rdev *rdev;
7578 char *desc, *action = NULL;
7579 struct blk_plug plug;
7581 /* just incase thread restarts... */
7582 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7584 if (mddev->ro) {/* never try to sync a read-only array */
7585 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7589 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7590 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7591 desc = "data-check";
7593 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7594 desc = "requested-resync";
7598 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7603 mddev->last_sync_action = action ?: desc;
7605 /* we overload curr_resync somewhat here.
7606 * 0 == not engaged in resync at all
7607 * 2 == checking that there is no conflict with another sync
7608 * 1 == like 2, but have yielded to allow conflicting resync to
7610 * other == active in resync - this many blocks
7612 * Before starting a resync we must have set curr_resync to
7613 * 2, and then checked that every "conflicting" array has curr_resync
7614 * less than ours. When we find one that is the same or higher
7615 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7616 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7617 * This will mean we have to start checking from the beginning again.
7622 mddev->curr_resync = 2;
7625 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7627 for_each_mddev(mddev2, tmp) {
7628 if (mddev2 == mddev)
7630 if (!mddev->parallel_resync
7631 && mddev2->curr_resync
7632 && match_mddev_units(mddev, mddev2)) {
7634 if (mddev < mddev2 && mddev->curr_resync == 2) {
7635 /* arbitrarily yield */
7636 mddev->curr_resync = 1;
7637 wake_up(&resync_wait);
7639 if (mddev > mddev2 && mddev->curr_resync == 1)
7640 /* no need to wait here, we can wait the next
7641 * time 'round when curr_resync == 2
7644 /* We need to wait 'interruptible' so as not to
7645 * contribute to the load average, and not to
7646 * be caught by 'softlockup'
7648 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7649 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7650 mddev2->curr_resync >= mddev->curr_resync) {
7651 printk(KERN_INFO "md: delaying %s of %s"
7652 " until %s has finished (they"
7653 " share one or more physical units)\n",
7654 desc, mdname(mddev), mdname(mddev2));
7656 if (signal_pending(current))
7657 flush_signals(current);
7659 finish_wait(&resync_wait, &wq);
7662 finish_wait(&resync_wait, &wq);
7665 } while (mddev->curr_resync < 2);
7668 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7669 /* resync follows the size requested by the personality,
7670 * which defaults to physical size, but can be virtual size
7672 max_sectors = mddev->resync_max_sectors;
7673 atomic64_set(&mddev->resync_mismatches, 0);
7674 /* we don't use the checkpoint if there's a bitmap */
7675 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7676 j = mddev->resync_min;
7677 else if (!mddev->bitmap)
7678 j = mddev->recovery_cp;
7680 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7681 max_sectors = mddev->resync_max_sectors;
7683 /* recovery follows the physical size of devices */
7684 max_sectors = mddev->dev_sectors;
7687 rdev_for_each_rcu(rdev, mddev)
7688 if (rdev->raid_disk >= 0 &&
7689 !test_bit(Faulty, &rdev->flags) &&
7690 !test_bit(In_sync, &rdev->flags) &&
7691 rdev->recovery_offset < j)
7692 j = rdev->recovery_offset;
7695 /* If there is a bitmap, we need to make sure all
7696 * writes that started before we added a spare
7697 * complete before we start doing a recovery.
7698 * Otherwise the write might complete and (via
7699 * bitmap_endwrite) set a bit in the bitmap after the
7700 * recovery has checked that bit and skipped that
7703 if (mddev->bitmap) {
7704 mddev->pers->quiesce(mddev, 1);
7705 mddev->pers->quiesce(mddev, 0);
7709 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7710 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7711 " %d KB/sec/disk.\n", speed_min(mddev));
7712 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7713 "(but not more than %d KB/sec) for %s.\n",
7714 speed_max(mddev), desc);
7716 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7719 for (m = 0; m < SYNC_MARKS; m++) {
7721 mark_cnt[m] = io_sectors;
7724 mddev->resync_mark = mark[last_mark];
7725 mddev->resync_mark_cnt = mark_cnt[last_mark];
7728 * Tune reconstruction:
7730 window = 32*(PAGE_SIZE/512);
7731 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7732 window/2, (unsigned long long)max_sectors/2);
7734 atomic_set(&mddev->recovery_active, 0);
7739 "md: resuming %s of %s from checkpoint.\n",
7740 desc, mdname(mddev));
7741 mddev->curr_resync = j;
7743 mddev->curr_resync = 3; /* no longer delayed */
7744 mddev->curr_resync_completed = j;
7745 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7746 md_new_event(mddev);
7747 update_time = jiffies;
7749 if (mddev_is_clustered(mddev))
7750 md_cluster_ops->resync_start(mddev, j, max_sectors);
7752 blk_start_plug(&plug);
7753 while (j < max_sectors) {
7758 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7759 ((mddev->curr_resync > mddev->curr_resync_completed &&
7760 (mddev->curr_resync - mddev->curr_resync_completed)
7761 > (max_sectors >> 4)) ||
7762 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7763 (j - mddev->curr_resync_completed)*2
7764 >= mddev->resync_max - mddev->curr_resync_completed
7766 /* time to update curr_resync_completed */
7767 wait_event(mddev->recovery_wait,
7768 atomic_read(&mddev->recovery_active) == 0);
7769 mddev->curr_resync_completed = j;
7770 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7771 j > mddev->recovery_cp)
7772 mddev->recovery_cp = j;
7773 update_time = jiffies;
7774 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7775 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7778 while (j >= mddev->resync_max &&
7779 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7780 /* As this condition is controlled by user-space,
7781 * we can block indefinitely, so use '_interruptible'
7782 * to avoid triggering warnings.
7784 flush_signals(current); /* just in case */
7785 wait_event_interruptible(mddev->recovery_wait,
7786 mddev->resync_max > j
7787 || test_bit(MD_RECOVERY_INTR,
7791 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7794 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7795 currspeed < speed_min(mddev));
7797 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7801 if (!skipped) { /* actual IO requested */
7802 io_sectors += sectors;
7803 atomic_add(sectors, &mddev->recovery_active);
7806 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7811 mddev->curr_resync = j;
7812 if (mddev_is_clustered(mddev))
7813 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7814 mddev->curr_mark_cnt = io_sectors;
7815 if (last_check == 0)
7816 /* this is the earliest that rebuild will be
7817 * visible in /proc/mdstat
7819 md_new_event(mddev);
7821 if (last_check + window > io_sectors || j == max_sectors)
7824 last_check = io_sectors;
7826 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7828 int next = (last_mark+1) % SYNC_MARKS;
7830 mddev->resync_mark = mark[next];
7831 mddev->resync_mark_cnt = mark_cnt[next];
7832 mark[next] = jiffies;
7833 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7837 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7841 * this loop exits only if either when we are slower than
7842 * the 'hard' speed limit, or the system was IO-idle for
7844 * the system might be non-idle CPU-wise, but we only care
7845 * about not overloading the IO subsystem. (things like an
7846 * e2fsck being done on the RAID array should execute fast)
7850 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7851 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7852 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7854 if (currspeed > speed_min(mddev)) {
7855 if ((currspeed > speed_max(mddev)) ||
7856 !is_mddev_idle(mddev, 0)) {
7862 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7863 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7864 ? "interrupted" : "done");
7866 * this also signals 'finished resyncing' to md_stop
7868 blk_finish_plug(&plug);
7869 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7871 /* tell personality that we are finished */
7872 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7874 if (mddev_is_clustered(mddev))
7875 md_cluster_ops->resync_finish(mddev);
7877 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7878 mddev->curr_resync > 2) {
7879 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7880 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7881 if (mddev->curr_resync >= mddev->recovery_cp) {
7883 "md: checkpointing %s of %s.\n",
7884 desc, mdname(mddev));
7885 if (test_bit(MD_RECOVERY_ERROR,
7887 mddev->recovery_cp =
7888 mddev->curr_resync_completed;
7890 mddev->recovery_cp =
7894 mddev->recovery_cp = MaxSector;
7896 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7897 mddev->curr_resync = MaxSector;
7899 rdev_for_each_rcu(rdev, mddev)
7900 if (rdev->raid_disk >= 0 &&
7901 mddev->delta_disks >= 0 &&
7902 !test_bit(Faulty, &rdev->flags) &&
7903 !test_bit(In_sync, &rdev->flags) &&
7904 rdev->recovery_offset < mddev->curr_resync)
7905 rdev->recovery_offset = mddev->curr_resync;
7910 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7912 spin_lock(&mddev->lock);
7913 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7914 /* We completed so min/max setting can be forgotten if used. */
7915 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7916 mddev->resync_min = 0;
7917 mddev->resync_max = MaxSector;
7918 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7919 mddev->resync_min = mddev->curr_resync_completed;
7920 mddev->curr_resync = 0;
7921 spin_unlock(&mddev->lock);
7923 wake_up(&resync_wait);
7924 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7925 md_wakeup_thread(mddev->thread);
7928 EXPORT_SYMBOL_GPL(md_do_sync);
7930 static int remove_and_add_spares(struct mddev *mddev,
7931 struct md_rdev *this)
7933 struct md_rdev *rdev;
7937 rdev_for_each(rdev, mddev)
7938 if ((this == NULL || rdev == this) &&
7939 rdev->raid_disk >= 0 &&
7940 !test_bit(Blocked, &rdev->flags) &&
7941 (test_bit(Faulty, &rdev->flags) ||
7942 ! test_bit(In_sync, &rdev->flags)) &&
7943 atomic_read(&rdev->nr_pending)==0) {
7944 if (mddev->pers->hot_remove_disk(
7945 mddev, rdev) == 0) {
7946 sysfs_unlink_rdev(mddev, rdev);
7947 rdev->raid_disk = -1;
7951 if (removed && mddev->kobj.sd)
7952 sysfs_notify(&mddev->kobj, NULL, "degraded");
7957 rdev_for_each(rdev, mddev) {
7958 if (rdev->raid_disk >= 0 &&
7959 !test_bit(In_sync, &rdev->flags) &&
7960 !test_bit(Faulty, &rdev->flags))
7962 if (rdev->raid_disk >= 0)
7964 if (test_bit(Faulty, &rdev->flags))
7967 ! (rdev->saved_raid_disk >= 0 &&
7968 !test_bit(Bitmap_sync, &rdev->flags)))
7971 if (rdev->saved_raid_disk < 0)
7972 rdev->recovery_offset = 0;
7974 hot_add_disk(mddev, rdev) == 0) {
7975 if (sysfs_link_rdev(mddev, rdev))
7976 /* failure here is OK */;
7978 md_new_event(mddev);
7979 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7984 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7988 static void md_start_sync(struct work_struct *ws)
7990 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7992 mddev->sync_thread = md_register_thread(md_do_sync,
7995 if (!mddev->sync_thread) {
7996 printk(KERN_ERR "%s: could not start resync"
7999 /* leave the spares where they are, it shouldn't hurt */
8000 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8001 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8002 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8003 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8004 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8005 wake_up(&resync_wait);
8006 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8008 if (mddev->sysfs_action)
8009 sysfs_notify_dirent_safe(mddev->sysfs_action);
8011 md_wakeup_thread(mddev->sync_thread);
8012 sysfs_notify_dirent_safe(mddev->sysfs_action);
8013 md_new_event(mddev);
8017 * This routine is regularly called by all per-raid-array threads to
8018 * deal with generic issues like resync and super-block update.
8019 * Raid personalities that don't have a thread (linear/raid0) do not
8020 * need this as they never do any recovery or update the superblock.
8022 * It does not do any resync itself, but rather "forks" off other threads
8023 * to do that as needed.
8024 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8025 * "->recovery" and create a thread at ->sync_thread.
8026 * When the thread finishes it sets MD_RECOVERY_DONE
8027 * and wakeups up this thread which will reap the thread and finish up.
8028 * This thread also removes any faulty devices (with nr_pending == 0).
8030 * The overall approach is:
8031 * 1/ if the superblock needs updating, update it.
8032 * 2/ If a recovery thread is running, don't do anything else.
8033 * 3/ If recovery has finished, clean up, possibly marking spares active.
8034 * 4/ If there are any faulty devices, remove them.
8035 * 5/ If array is degraded, try to add spares devices
8036 * 6/ If array has spares or is not in-sync, start a resync thread.
8038 void md_check_recovery(struct mddev *mddev)
8040 if (mddev->suspended)
8044 bitmap_daemon_work(mddev);
8046 if (signal_pending(current)) {
8047 if (mddev->pers->sync_request && !mddev->external) {
8048 printk(KERN_INFO "md: %s in immediate safe mode\n",
8050 mddev->safemode = 2;
8052 flush_signals(current);
8055 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8058 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8059 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8060 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8061 (mddev->external == 0 && mddev->safemode == 1) ||
8062 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8063 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8067 if (mddev_trylock(mddev)) {
8071 /* On a read-only array we can:
8072 * - remove failed devices
8073 * - add already-in_sync devices if the array itself
8075 * As we only add devices that are already in-sync,
8076 * we can activate the spares immediately.
8078 remove_and_add_spares(mddev, NULL);
8079 /* There is no thread, but we need to call
8080 * ->spare_active and clear saved_raid_disk
8082 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8083 md_reap_sync_thread(mddev);
8084 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8088 if (!mddev->external) {
8090 spin_lock(&mddev->lock);
8091 if (mddev->safemode &&
8092 !atomic_read(&mddev->writes_pending) &&
8094 mddev->recovery_cp == MaxSector) {
8097 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8099 if (mddev->safemode == 1)
8100 mddev->safemode = 0;
8101 spin_unlock(&mddev->lock);
8103 sysfs_notify_dirent_safe(mddev->sysfs_state);
8106 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8107 if (mddev_is_clustered(mddev))
8108 md_cluster_ops->metadata_update_start(mddev);
8109 md_update_sb(mddev, 0);
8110 if (mddev_is_clustered(mddev))
8111 md_cluster_ops->metadata_update_finish(mddev);
8114 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8115 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8116 /* resync/recovery still happening */
8117 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8120 if (mddev->sync_thread) {
8121 md_reap_sync_thread(mddev);
8124 /* Set RUNNING before clearing NEEDED to avoid
8125 * any transients in the value of "sync_action".
8127 mddev->curr_resync_completed = 0;
8128 spin_lock(&mddev->lock);
8129 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8130 spin_unlock(&mddev->lock);
8131 /* Clear some bits that don't mean anything, but
8134 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8135 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8137 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8138 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8140 /* no recovery is running.
8141 * remove any failed drives, then
8142 * add spares if possible.
8143 * Spares are also removed and re-added, to allow
8144 * the personality to fail the re-add.
8147 if (mddev->reshape_position != MaxSector) {
8148 if (mddev->pers->check_reshape == NULL ||
8149 mddev->pers->check_reshape(mddev) != 0)
8150 /* Cannot proceed */
8152 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8153 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8154 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8155 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8156 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8157 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8158 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8159 } else if (mddev->recovery_cp < MaxSector) {
8160 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8161 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8162 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8163 /* nothing to be done ... */
8166 if (mddev->pers->sync_request) {
8168 /* We are adding a device or devices to an array
8169 * which has the bitmap stored on all devices.
8170 * So make sure all bitmap pages get written
8172 bitmap_write_all(mddev->bitmap);
8174 INIT_WORK(&mddev->del_work, md_start_sync);
8175 queue_work(md_misc_wq, &mddev->del_work);
8179 if (!mddev->sync_thread) {
8180 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8181 wake_up(&resync_wait);
8182 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8184 if (mddev->sysfs_action)
8185 sysfs_notify_dirent_safe(mddev->sysfs_action);
8188 wake_up(&mddev->sb_wait);
8189 mddev_unlock(mddev);
8192 EXPORT_SYMBOL(md_check_recovery);
8194 void md_reap_sync_thread(struct mddev *mddev)
8196 struct md_rdev *rdev;
8198 /* resync has finished, collect result */
8199 md_unregister_thread(&mddev->sync_thread);
8200 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8201 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8203 /* activate any spares */
8204 if (mddev->pers->spare_active(mddev)) {
8205 sysfs_notify(&mddev->kobj, NULL,
8207 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8210 if (mddev_is_clustered(mddev))
8211 md_cluster_ops->metadata_update_start(mddev);
8212 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8213 mddev->pers->finish_reshape)
8214 mddev->pers->finish_reshape(mddev);
8216 /* If array is no-longer degraded, then any saved_raid_disk
8217 * information must be scrapped.
8219 if (!mddev->degraded)
8220 rdev_for_each(rdev, mddev)
8221 rdev->saved_raid_disk = -1;
8223 md_update_sb(mddev, 1);
8224 if (mddev_is_clustered(mddev))
8225 md_cluster_ops->metadata_update_finish(mddev);
8226 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8227 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8228 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8229 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8230 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8231 wake_up(&resync_wait);
8232 /* flag recovery needed just to double check */
8233 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8234 sysfs_notify_dirent_safe(mddev->sysfs_action);
8235 md_new_event(mddev);
8236 if (mddev->event_work.func)
8237 queue_work(md_misc_wq, &mddev->event_work);
8239 EXPORT_SYMBOL(md_reap_sync_thread);
8241 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8243 sysfs_notify_dirent_safe(rdev->sysfs_state);
8244 wait_event_timeout(rdev->blocked_wait,
8245 !test_bit(Blocked, &rdev->flags) &&
8246 !test_bit(BlockedBadBlocks, &rdev->flags),
8247 msecs_to_jiffies(5000));
8248 rdev_dec_pending(rdev, mddev);
8250 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8252 void md_finish_reshape(struct mddev *mddev)
8254 /* called be personality module when reshape completes. */
8255 struct md_rdev *rdev;
8257 rdev_for_each(rdev, mddev) {
8258 if (rdev->data_offset > rdev->new_data_offset)
8259 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8261 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8262 rdev->data_offset = rdev->new_data_offset;
8265 EXPORT_SYMBOL(md_finish_reshape);
8267 /* Bad block management.
8268 * We can record which blocks on each device are 'bad' and so just
8269 * fail those blocks, or that stripe, rather than the whole device.
8270 * Entries in the bad-block table are 64bits wide. This comprises:
8271 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8272 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8273 * A 'shift' can be set so that larger blocks are tracked and
8274 * consequently larger devices can be covered.
8275 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8277 * Locking of the bad-block table uses a seqlock so md_is_badblock
8278 * might need to retry if it is very unlucky.
8279 * We will sometimes want to check for bad blocks in a bi_end_io function,
8280 * so we use the write_seqlock_irq variant.
8282 * When looking for a bad block we specify a range and want to
8283 * know if any block in the range is bad. So we binary-search
8284 * to the last range that starts at-or-before the given endpoint,
8285 * (or "before the sector after the target range")
8286 * then see if it ends after the given start.
8288 * 0 if there are no known bad blocks in the range
8289 * 1 if there are known bad block which are all acknowledged
8290 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8291 * plus the start/length of the first bad section we overlap.
8293 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8294 sector_t *first_bad, int *bad_sectors)
8300 sector_t target = s + sectors;
8303 if (bb->shift > 0) {
8304 /* round the start down, and the end up */
8306 target += (1<<bb->shift) - 1;
8307 target >>= bb->shift;
8308 sectors = target - s;
8310 /* 'target' is now the first block after the bad range */
8313 seq = read_seqbegin(&bb->lock);
8318 /* Binary search between lo and hi for 'target'
8319 * i.e. for the last range that starts before 'target'
8321 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8322 * are known not to be the last range before target.
8323 * VARIANT: hi-lo is the number of possible
8324 * ranges, and decreases until it reaches 1
8326 while (hi - lo > 1) {
8327 int mid = (lo + hi) / 2;
8328 sector_t a = BB_OFFSET(p[mid]);
8330 /* This could still be the one, earlier ranges
8334 /* This and later ranges are definitely out. */
8337 /* 'lo' might be the last that started before target, but 'hi' isn't */
8339 /* need to check all range that end after 's' to see if
8340 * any are unacknowledged.
8343 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8344 if (BB_OFFSET(p[lo]) < target) {
8345 /* starts before the end, and finishes after
8346 * the start, so they must overlap
8348 if (rv != -1 && BB_ACK(p[lo]))
8352 *first_bad = BB_OFFSET(p[lo]);
8353 *bad_sectors = BB_LEN(p[lo]);
8359 if (read_seqretry(&bb->lock, seq))
8364 EXPORT_SYMBOL_GPL(md_is_badblock);
8367 * Add a range of bad blocks to the table.
8368 * This might extend the table, or might contract it
8369 * if two adjacent ranges can be merged.
8370 * We binary-search to find the 'insertion' point, then
8371 * decide how best to handle it.
8373 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8379 unsigned long flags;
8382 /* badblocks are disabled */
8386 /* round the start down, and the end up */
8387 sector_t next = s + sectors;
8389 next += (1<<bb->shift) - 1;
8394 write_seqlock_irqsave(&bb->lock, flags);
8399 /* Find the last range that starts at-or-before 's' */
8400 while (hi - lo > 1) {
8401 int mid = (lo + hi) / 2;
8402 sector_t a = BB_OFFSET(p[mid]);
8408 if (hi > lo && BB_OFFSET(p[lo]) > s)
8412 /* we found a range that might merge with the start
8415 sector_t a = BB_OFFSET(p[lo]);
8416 sector_t e = a + BB_LEN(p[lo]);
8417 int ack = BB_ACK(p[lo]);
8419 /* Yes, we can merge with a previous range */
8420 if (s == a && s + sectors >= e)
8421 /* new range covers old */
8424 ack = ack && acknowledged;
8426 if (e < s + sectors)
8428 if (e - a <= BB_MAX_LEN) {
8429 p[lo] = BB_MAKE(a, e-a, ack);
8432 /* does not all fit in one range,
8433 * make p[lo] maximal
8435 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8436 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8442 if (sectors && hi < bb->count) {
8443 /* 'hi' points to the first range that starts after 's'.
8444 * Maybe we can merge with the start of that range */
8445 sector_t a = BB_OFFSET(p[hi]);
8446 sector_t e = a + BB_LEN(p[hi]);
8447 int ack = BB_ACK(p[hi]);
8448 if (a <= s + sectors) {
8449 /* merging is possible */
8450 if (e <= s + sectors) {
8455 ack = ack && acknowledged;
8458 if (e - a <= BB_MAX_LEN) {
8459 p[hi] = BB_MAKE(a, e-a, ack);
8462 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8470 if (sectors == 0 && hi < bb->count) {
8471 /* we might be able to combine lo and hi */
8472 /* Note: 's' is at the end of 'lo' */
8473 sector_t a = BB_OFFSET(p[hi]);
8474 int lolen = BB_LEN(p[lo]);
8475 int hilen = BB_LEN(p[hi]);
8476 int newlen = lolen + hilen - (s - a);
8477 if (s >= a && newlen < BB_MAX_LEN) {
8478 /* yes, we can combine them */
8479 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8480 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8481 memmove(p + hi, p + hi + 1,
8482 (bb->count - hi - 1) * 8);
8487 /* didn't merge (it all).
8488 * Need to add a range just before 'hi' */
8489 if (bb->count >= MD_MAX_BADBLOCKS) {
8490 /* No room for more */
8494 int this_sectors = sectors;
8495 memmove(p + hi + 1, p + hi,
8496 (bb->count - hi) * 8);
8499 if (this_sectors > BB_MAX_LEN)
8500 this_sectors = BB_MAX_LEN;
8501 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8502 sectors -= this_sectors;
8509 bb->unacked_exist = 1;
8510 write_sequnlock_irqrestore(&bb->lock, flags);
8515 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8520 s += rdev->new_data_offset;
8522 s += rdev->data_offset;
8523 rv = md_set_badblocks(&rdev->badblocks,
8526 /* Make sure they get written out promptly */
8527 sysfs_notify_dirent_safe(rdev->sysfs_state);
8528 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8529 md_wakeup_thread(rdev->mddev->thread);
8533 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8536 * Remove a range of bad blocks from the table.
8537 * This may involve extending the table if we spilt a region,
8538 * but it must not fail. So if the table becomes full, we just
8539 * drop the remove request.
8541 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8545 sector_t target = s + sectors;
8548 if (bb->shift > 0) {
8549 /* When clearing we round the start up and the end down.
8550 * This should not matter as the shift should align with
8551 * the block size and no rounding should ever be needed.
8552 * However it is better the think a block is bad when it
8553 * isn't than to think a block is not bad when it is.
8555 s += (1<<bb->shift) - 1;
8557 target >>= bb->shift;
8558 sectors = target - s;
8561 write_seqlock_irq(&bb->lock);
8566 /* Find the last range that starts before 'target' */
8567 while (hi - lo > 1) {
8568 int mid = (lo + hi) / 2;
8569 sector_t a = BB_OFFSET(p[mid]);
8576 /* p[lo] is the last range that could overlap the
8577 * current range. Earlier ranges could also overlap,
8578 * but only this one can overlap the end of the range.
8580 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8581 /* Partial overlap, leave the tail of this range */
8582 int ack = BB_ACK(p[lo]);
8583 sector_t a = BB_OFFSET(p[lo]);
8584 sector_t end = a + BB_LEN(p[lo]);
8587 /* we need to split this range */
8588 if (bb->count >= MD_MAX_BADBLOCKS) {
8592 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8594 p[lo] = BB_MAKE(a, s-a, ack);
8597 p[lo] = BB_MAKE(target, end - target, ack);
8598 /* there is no longer an overlap */
8603 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8604 /* This range does overlap */
8605 if (BB_OFFSET(p[lo]) < s) {
8606 /* Keep the early parts of this range. */
8607 int ack = BB_ACK(p[lo]);
8608 sector_t start = BB_OFFSET(p[lo]);
8609 p[lo] = BB_MAKE(start, s - start, ack);
8610 /* now low doesn't overlap, so.. */
8615 /* 'lo' is strictly before, 'hi' is strictly after,
8616 * anything between needs to be discarded
8619 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8620 bb->count -= (hi - lo - 1);
8626 write_sequnlock_irq(&bb->lock);
8630 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8634 s += rdev->new_data_offset;
8636 s += rdev->data_offset;
8637 return md_clear_badblocks(&rdev->badblocks,
8640 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8643 * Acknowledge all bad blocks in a list.
8644 * This only succeeds if ->changed is clear. It is used by
8645 * in-kernel metadata updates
8647 void md_ack_all_badblocks(struct badblocks *bb)
8649 if (bb->page == NULL || bb->changed)
8650 /* no point even trying */
8652 write_seqlock_irq(&bb->lock);
8654 if (bb->changed == 0 && bb->unacked_exist) {
8657 for (i = 0; i < bb->count ; i++) {
8658 if (!BB_ACK(p[i])) {
8659 sector_t start = BB_OFFSET(p[i]);
8660 int len = BB_LEN(p[i]);
8661 p[i] = BB_MAKE(start, len, 1);
8664 bb->unacked_exist = 0;
8666 write_sequnlock_irq(&bb->lock);
8668 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8670 /* sysfs access to bad-blocks list.
8671 * We present two files.
8672 * 'bad-blocks' lists sector numbers and lengths of ranges that
8673 * are recorded as bad. The list is truncated to fit within
8674 * the one-page limit of sysfs.
8675 * Writing "sector length" to this file adds an acknowledged
8677 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8678 * been acknowledged. Writing to this file adds bad blocks
8679 * without acknowledging them. This is largely for testing.
8683 badblocks_show(struct badblocks *bb, char *page, int unack)
8694 seq = read_seqbegin(&bb->lock);
8699 while (len < PAGE_SIZE && i < bb->count) {
8700 sector_t s = BB_OFFSET(p[i]);
8701 unsigned int length = BB_LEN(p[i]);
8702 int ack = BB_ACK(p[i]);
8708 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8709 (unsigned long long)s << bb->shift,
8710 length << bb->shift);
8712 if (unack && len == 0)
8713 bb->unacked_exist = 0;
8715 if (read_seqretry(&bb->lock, seq))
8724 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8726 unsigned long long sector;
8730 /* Allow clearing via sysfs *only* for testing/debugging.
8731 * Normally only a successful write may clear a badblock
8734 if (page[0] == '-') {
8738 #endif /* DO_DEBUG */
8740 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8742 if (newline != '\n')
8754 md_clear_badblocks(bb, sector, length);
8757 #endif /* DO_DEBUG */
8758 if (md_set_badblocks(bb, sector, length, !unack))
8764 static int md_notify_reboot(struct notifier_block *this,
8765 unsigned long code, void *x)
8767 struct list_head *tmp;
8768 struct mddev *mddev;
8771 for_each_mddev(mddev, tmp) {
8772 if (mddev_trylock(mddev)) {
8774 __md_stop_writes(mddev);
8775 if (mddev->persistent)
8776 mddev->safemode = 2;
8777 mddev_unlock(mddev);
8782 * certain more exotic SCSI devices are known to be
8783 * volatile wrt too early system reboots. While the
8784 * right place to handle this issue is the given
8785 * driver, we do want to have a safe RAID driver ...
8793 static struct notifier_block md_notifier = {
8794 .notifier_call = md_notify_reboot,
8796 .priority = INT_MAX, /* before any real devices */
8799 static void md_geninit(void)
8801 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8803 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8806 static int __init md_init(void)
8810 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8814 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8818 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8821 if ((ret = register_blkdev(0, "mdp")) < 0)
8825 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8826 md_probe, NULL, NULL);
8827 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8828 md_probe, NULL, NULL);
8830 register_reboot_notifier(&md_notifier);
8831 raid_table_header = register_sysctl_table(raid_root_table);
8837 unregister_blkdev(MD_MAJOR, "md");
8839 destroy_workqueue(md_misc_wq);
8841 destroy_workqueue(md_wq);
8846 void md_reload_sb(struct mddev *mddev)
8848 struct md_rdev *rdev, *tmp;
8850 rdev_for_each_safe(rdev, tmp, mddev) {
8851 rdev->sb_loaded = 0;
8852 ClearPageUptodate(rdev->sb_page);
8854 mddev->raid_disks = 0;
8856 rdev_for_each_safe(rdev, tmp, mddev) {
8857 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8858 /* since we don't write to faulty devices, we figure out if the
8859 * disk is faulty by comparing events
8861 if (mddev->events > sb->events)
8862 set_bit(Faulty, &rdev->flags);
8866 EXPORT_SYMBOL(md_reload_sb);
8871 * Searches all registered partitions for autorun RAID arrays
8875 static LIST_HEAD(all_detected_devices);
8876 struct detected_devices_node {
8877 struct list_head list;
8881 void md_autodetect_dev(dev_t dev)
8883 struct detected_devices_node *node_detected_dev;
8885 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8886 if (node_detected_dev) {
8887 node_detected_dev->dev = dev;
8888 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8890 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8891 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8895 static void autostart_arrays(int part)
8897 struct md_rdev *rdev;
8898 struct detected_devices_node *node_detected_dev;
8900 int i_scanned, i_passed;
8905 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8907 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8909 node_detected_dev = list_entry(all_detected_devices.next,
8910 struct detected_devices_node, list);
8911 list_del(&node_detected_dev->list);
8912 dev = node_detected_dev->dev;
8913 kfree(node_detected_dev);
8914 rdev = md_import_device(dev,0, 90);
8918 if (test_bit(Faulty, &rdev->flags))
8921 set_bit(AutoDetected, &rdev->flags);
8922 list_add(&rdev->same_set, &pending_raid_disks);
8926 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8927 i_scanned, i_passed);
8929 autorun_devices(part);
8932 #endif /* !MODULE */
8934 static __exit void md_exit(void)
8936 struct mddev *mddev;
8937 struct list_head *tmp;
8940 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8941 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8943 unregister_blkdev(MD_MAJOR,"md");
8944 unregister_blkdev(mdp_major, "mdp");
8945 unregister_reboot_notifier(&md_notifier);
8946 unregister_sysctl_table(raid_table_header);
8948 /* We cannot unload the modules while some process is
8949 * waiting for us in select() or poll() - wake them up
8952 while (waitqueue_active(&md_event_waiters)) {
8953 /* not safe to leave yet */
8954 wake_up(&md_event_waiters);
8958 remove_proc_entry("mdstat", NULL);
8960 for_each_mddev(mddev, tmp) {
8961 export_array(mddev);
8962 mddev->hold_active = 0;
8964 destroy_workqueue(md_misc_wq);
8965 destroy_workqueue(md_wq);
8968 subsys_initcall(md_init);
8969 module_exit(md_exit)
8971 static int get_ro(char *buffer, struct kernel_param *kp)
8973 return sprintf(buffer, "%d", start_readonly);
8975 static int set_ro(const char *val, struct kernel_param *kp)
8978 int num = simple_strtoul(val, &e, 10);
8979 if (*val && (*e == '\0' || *e == '\n')) {
8980 start_readonly = num;
8986 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8987 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8988 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8990 MODULE_LICENSE("GPL");
8991 MODULE_DESCRIPTION("MD RAID framework");
8993 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);