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 static int add_bound_rdev(struct md_rdev *rdev)
2380 struct mddev *mddev = rdev->mddev;
2383 if (!mddev->pers->hot_remove_disk) {
2384 /* If there is hot_add_disk but no hot_remove_disk
2385 * then added disks for geometry changes,
2386 * and should be added immediately.
2388 super_types[mddev->major_version].
2389 validate_super(mddev, rdev);
2390 err = mddev->pers->hot_add_disk(mddev, rdev);
2392 unbind_rdev_from_array(rdev);
2397 sysfs_notify_dirent_safe(rdev->sysfs_state);
2399 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2400 if (mddev->degraded)
2401 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
2402 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2403 md_new_event(mddev);
2404 md_wakeup_thread(mddev->thread);
2408 /* words written to sysfs files may, or may not, be \n terminated.
2409 * We want to accept with case. For this we use cmd_match.
2411 static int cmd_match(const char *cmd, const char *str)
2413 /* See if cmd, written into a sysfs file, matches
2414 * str. They must either be the same, or cmd can
2415 * have a trailing newline
2417 while (*cmd && *str && *cmd == *str) {
2428 struct rdev_sysfs_entry {
2429 struct attribute attr;
2430 ssize_t (*show)(struct md_rdev *, char *);
2431 ssize_t (*store)(struct md_rdev *, const char *, size_t);
2435 state_show(struct md_rdev *rdev, char *page)
2439 unsigned long flags = ACCESS_ONCE(rdev->flags);
2441 if (test_bit(Faulty, &flags) ||
2442 rdev->badblocks.unacked_exist) {
2443 len+= sprintf(page+len, "%sfaulty",sep);
2446 if (test_bit(In_sync, &flags)) {
2447 len += sprintf(page+len, "%sin_sync",sep);
2450 if (test_bit(WriteMostly, &flags)) {
2451 len += sprintf(page+len, "%swrite_mostly",sep);
2454 if (test_bit(Blocked, &flags) ||
2455 (rdev->badblocks.unacked_exist
2456 && !test_bit(Faulty, &flags))) {
2457 len += sprintf(page+len, "%sblocked", sep);
2460 if (!test_bit(Faulty, &flags) &&
2461 !test_bit(In_sync, &flags)) {
2462 len += sprintf(page+len, "%sspare", sep);
2465 if (test_bit(WriteErrorSeen, &flags)) {
2466 len += sprintf(page+len, "%swrite_error", sep);
2469 if (test_bit(WantReplacement, &flags)) {
2470 len += sprintf(page+len, "%swant_replacement", sep);
2473 if (test_bit(Replacement, &flags)) {
2474 len += sprintf(page+len, "%sreplacement", sep);
2478 return len+sprintf(page+len, "\n");
2482 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2485 * faulty - simulates an error
2486 * remove - disconnects the device
2487 * writemostly - sets write_mostly
2488 * -writemostly - clears write_mostly
2489 * blocked - sets the Blocked flags
2490 * -blocked - clears the Blocked and possibly simulates an error
2491 * insync - sets Insync providing device isn't active
2492 * -insync - clear Insync for a device with a slot assigned,
2493 * so that it gets rebuilt based on bitmap
2494 * write_error - sets WriteErrorSeen
2495 * -write_error - clears WriteErrorSeen
2498 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2499 md_error(rdev->mddev, rdev);
2500 if (test_bit(Faulty, &rdev->flags))
2504 } else if (cmd_match(buf, "remove")) {
2505 if (rdev->raid_disk >= 0)
2508 struct mddev *mddev = rdev->mddev;
2509 if (mddev_is_clustered(mddev))
2510 md_cluster_ops->remove_disk(mddev, rdev);
2511 md_kick_rdev_from_array(rdev);
2512 if (mddev_is_clustered(mddev))
2513 md_cluster_ops->metadata_update_start(mddev);
2515 md_update_sb(mddev, 1);
2516 md_new_event(mddev);
2517 if (mddev_is_clustered(mddev))
2518 md_cluster_ops->metadata_update_finish(mddev);
2521 } else if (cmd_match(buf, "writemostly")) {
2522 set_bit(WriteMostly, &rdev->flags);
2524 } else if (cmd_match(buf, "-writemostly")) {
2525 clear_bit(WriteMostly, &rdev->flags);
2527 } else if (cmd_match(buf, "blocked")) {
2528 set_bit(Blocked, &rdev->flags);
2530 } else if (cmd_match(buf, "-blocked")) {
2531 if (!test_bit(Faulty, &rdev->flags) &&
2532 rdev->badblocks.unacked_exist) {
2533 /* metadata handler doesn't understand badblocks,
2534 * so we need to fail the device
2536 md_error(rdev->mddev, rdev);
2538 clear_bit(Blocked, &rdev->flags);
2539 clear_bit(BlockedBadBlocks, &rdev->flags);
2540 wake_up(&rdev->blocked_wait);
2541 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2542 md_wakeup_thread(rdev->mddev->thread);
2545 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2546 set_bit(In_sync, &rdev->flags);
2548 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
2549 if (rdev->mddev->pers == NULL) {
2550 clear_bit(In_sync, &rdev->flags);
2551 rdev->saved_raid_disk = rdev->raid_disk;
2552 rdev->raid_disk = -1;
2555 } else if (cmd_match(buf, "write_error")) {
2556 set_bit(WriteErrorSeen, &rdev->flags);
2558 } else if (cmd_match(buf, "-write_error")) {
2559 clear_bit(WriteErrorSeen, &rdev->flags);
2561 } else if (cmd_match(buf, "want_replacement")) {
2562 /* Any non-spare device that is not a replacement can
2563 * become want_replacement at any time, but we then need to
2564 * check if recovery is needed.
2566 if (rdev->raid_disk >= 0 &&
2567 !test_bit(Replacement, &rdev->flags))
2568 set_bit(WantReplacement, &rdev->flags);
2569 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2570 md_wakeup_thread(rdev->mddev->thread);
2572 } else if (cmd_match(buf, "-want_replacement")) {
2573 /* Clearing 'want_replacement' is always allowed.
2574 * Once replacements starts it is too late though.
2577 clear_bit(WantReplacement, &rdev->flags);
2578 } else if (cmd_match(buf, "replacement")) {
2579 /* Can only set a device as a replacement when array has not
2580 * yet been started. Once running, replacement is automatic
2581 * from spares, or by assigning 'slot'.
2583 if (rdev->mddev->pers)
2586 set_bit(Replacement, &rdev->flags);
2589 } else if (cmd_match(buf, "-replacement")) {
2590 /* Similarly, can only clear Replacement before start */
2591 if (rdev->mddev->pers)
2594 clear_bit(Replacement, &rdev->flags);
2597 } else if (cmd_match(buf, "re-add")) {
2598 if (test_bit(Faulty, &rdev->flags) && (rdev->raid_disk == -1)) {
2599 clear_bit(Faulty, &rdev->flags);
2600 err = add_bound_rdev(rdev);
2605 sysfs_notify_dirent_safe(rdev->sysfs_state);
2606 return err ? err : len;
2608 static struct rdev_sysfs_entry rdev_state =
2609 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2612 errors_show(struct md_rdev *rdev, char *page)
2614 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2618 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2621 unsigned long n = simple_strtoul(buf, &e, 10);
2622 if (*buf && (*e == 0 || *e == '\n')) {
2623 atomic_set(&rdev->corrected_errors, n);
2628 static struct rdev_sysfs_entry rdev_errors =
2629 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2632 slot_show(struct md_rdev *rdev, char *page)
2634 if (rdev->raid_disk < 0)
2635 return sprintf(page, "none\n");
2637 return sprintf(page, "%d\n", rdev->raid_disk);
2641 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2645 int slot = simple_strtoul(buf, &e, 10);
2646 if (strncmp(buf, "none", 4)==0)
2648 else if (e==buf || (*e && *e!= '\n'))
2650 if (rdev->mddev->pers && slot == -1) {
2651 /* Setting 'slot' on an active array requires also
2652 * updating the 'rd%d' link, and communicating
2653 * with the personality with ->hot_*_disk.
2654 * For now we only support removing
2655 * failed/spare devices. This normally happens automatically,
2656 * but not when the metadata is externally managed.
2658 if (rdev->raid_disk == -1)
2660 /* personality does all needed checks */
2661 if (rdev->mddev->pers->hot_remove_disk == NULL)
2663 clear_bit(Blocked, &rdev->flags);
2664 remove_and_add_spares(rdev->mddev, rdev);
2665 if (rdev->raid_disk >= 0)
2667 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2668 md_wakeup_thread(rdev->mddev->thread);
2669 } else if (rdev->mddev->pers) {
2670 /* Activating a spare .. or possibly reactivating
2671 * if we ever get bitmaps working here.
2674 if (rdev->raid_disk != -1)
2677 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2680 if (rdev->mddev->pers->hot_add_disk == NULL)
2683 if (slot >= rdev->mddev->raid_disks &&
2684 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2687 rdev->raid_disk = slot;
2688 if (test_bit(In_sync, &rdev->flags))
2689 rdev->saved_raid_disk = slot;
2691 rdev->saved_raid_disk = -1;
2692 clear_bit(In_sync, &rdev->flags);
2693 clear_bit(Bitmap_sync, &rdev->flags);
2694 err = rdev->mddev->pers->
2695 hot_add_disk(rdev->mddev, rdev);
2697 rdev->raid_disk = -1;
2700 sysfs_notify_dirent_safe(rdev->sysfs_state);
2701 if (sysfs_link_rdev(rdev->mddev, rdev))
2702 /* failure here is OK */;
2703 /* don't wakeup anyone, leave that to userspace. */
2705 if (slot >= rdev->mddev->raid_disks &&
2706 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2708 rdev->raid_disk = slot;
2709 /* assume it is working */
2710 clear_bit(Faulty, &rdev->flags);
2711 clear_bit(WriteMostly, &rdev->flags);
2712 set_bit(In_sync, &rdev->flags);
2713 sysfs_notify_dirent_safe(rdev->sysfs_state);
2718 static struct rdev_sysfs_entry rdev_slot =
2719 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2722 offset_show(struct md_rdev *rdev, char *page)
2724 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2728 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2730 unsigned long long offset;
2731 if (kstrtoull(buf, 10, &offset) < 0)
2733 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2735 if (rdev->sectors && rdev->mddev->external)
2736 /* Must set offset before size, so overlap checks
2739 rdev->data_offset = offset;
2740 rdev->new_data_offset = offset;
2744 static struct rdev_sysfs_entry rdev_offset =
2745 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2747 static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2749 return sprintf(page, "%llu\n",
2750 (unsigned long long)rdev->new_data_offset);
2753 static ssize_t new_offset_store(struct md_rdev *rdev,
2754 const char *buf, size_t len)
2756 unsigned long long new_offset;
2757 struct mddev *mddev = rdev->mddev;
2759 if (kstrtoull(buf, 10, &new_offset) < 0)
2762 if (mddev->sync_thread ||
2763 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
2765 if (new_offset == rdev->data_offset)
2766 /* reset is always permitted */
2768 else if (new_offset > rdev->data_offset) {
2769 /* must not push array size beyond rdev_sectors */
2770 if (new_offset - rdev->data_offset
2771 + mddev->dev_sectors > rdev->sectors)
2774 /* Metadata worries about other space details. */
2776 /* decreasing the offset is inconsistent with a backwards
2779 if (new_offset < rdev->data_offset &&
2780 mddev->reshape_backwards)
2782 /* Increasing offset is inconsistent with forwards
2783 * reshape. reshape_direction should be set to
2784 * 'backwards' first.
2786 if (new_offset > rdev->data_offset &&
2787 !mddev->reshape_backwards)
2790 if (mddev->pers && mddev->persistent &&
2791 !super_types[mddev->major_version]
2792 .allow_new_offset(rdev, new_offset))
2794 rdev->new_data_offset = new_offset;
2795 if (new_offset > rdev->data_offset)
2796 mddev->reshape_backwards = 1;
2797 else if (new_offset < rdev->data_offset)
2798 mddev->reshape_backwards = 0;
2802 static struct rdev_sysfs_entry rdev_new_offset =
2803 __ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2806 rdev_size_show(struct md_rdev *rdev, char *page)
2808 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2811 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2813 /* check if two start/length pairs overlap */
2821 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2823 unsigned long long blocks;
2826 if (kstrtoull(buf, 10, &blocks) < 0)
2829 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2830 return -EINVAL; /* sector conversion overflow */
2833 if (new != blocks * 2)
2834 return -EINVAL; /* unsigned long long to sector_t overflow */
2841 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2843 struct mddev *my_mddev = rdev->mddev;
2844 sector_t oldsectors = rdev->sectors;
2847 if (strict_blocks_to_sectors(buf, §ors) < 0)
2849 if (rdev->data_offset != rdev->new_data_offset)
2850 return -EINVAL; /* too confusing */
2851 if (my_mddev->pers && rdev->raid_disk >= 0) {
2852 if (my_mddev->persistent) {
2853 sectors = super_types[my_mddev->major_version].
2854 rdev_size_change(rdev, sectors);
2857 } else if (!sectors)
2858 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2860 if (!my_mddev->pers->resize)
2861 /* Cannot change size for RAID0 or Linear etc */
2864 if (sectors < my_mddev->dev_sectors)
2865 return -EINVAL; /* component must fit device */
2867 rdev->sectors = sectors;
2868 if (sectors > oldsectors && my_mddev->external) {
2869 /* Need to check that all other rdevs with the same
2870 * ->bdev do not overlap. 'rcu' is sufficient to walk
2871 * the rdev lists safely.
2872 * This check does not provide a hard guarantee, it
2873 * just helps avoid dangerous mistakes.
2875 struct mddev *mddev;
2877 struct list_head *tmp;
2880 for_each_mddev(mddev, tmp) {
2881 struct md_rdev *rdev2;
2883 rdev_for_each(rdev2, mddev)
2884 if (rdev->bdev == rdev2->bdev &&
2886 overlaps(rdev->data_offset, rdev->sectors,
2899 /* Someone else could have slipped in a size
2900 * change here, but doing so is just silly.
2901 * We put oldsectors back because we *know* it is
2902 * safe, and trust userspace not to race with
2905 rdev->sectors = oldsectors;
2912 static struct rdev_sysfs_entry rdev_size =
2913 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2915 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2917 unsigned long long recovery_start = rdev->recovery_offset;
2919 if (test_bit(In_sync, &rdev->flags) ||
2920 recovery_start == MaxSector)
2921 return sprintf(page, "none\n");
2923 return sprintf(page, "%llu\n", recovery_start);
2926 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2928 unsigned long long recovery_start;
2930 if (cmd_match(buf, "none"))
2931 recovery_start = MaxSector;
2932 else if (kstrtoull(buf, 10, &recovery_start))
2935 if (rdev->mddev->pers &&
2936 rdev->raid_disk >= 0)
2939 rdev->recovery_offset = recovery_start;
2940 if (recovery_start == MaxSector)
2941 set_bit(In_sync, &rdev->flags);
2943 clear_bit(In_sync, &rdev->flags);
2947 static struct rdev_sysfs_entry rdev_recovery_start =
2948 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2951 badblocks_show(struct badblocks *bb, char *page, int unack);
2953 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2955 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2957 return badblocks_show(&rdev->badblocks, page, 0);
2959 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2961 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2962 /* Maybe that ack was all we needed */
2963 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2964 wake_up(&rdev->blocked_wait);
2967 static struct rdev_sysfs_entry rdev_bad_blocks =
2968 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2970 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2972 return badblocks_show(&rdev->badblocks, page, 1);
2974 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2976 return badblocks_store(&rdev->badblocks, page, len, 1);
2978 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2979 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2981 static struct attribute *rdev_default_attrs[] = {
2986 &rdev_new_offset.attr,
2988 &rdev_recovery_start.attr,
2989 &rdev_bad_blocks.attr,
2990 &rdev_unack_bad_blocks.attr,
2994 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2996 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2997 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3003 return entry->show(rdev, page);
3007 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3008 const char *page, size_t length)
3010 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3011 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3013 struct mddev *mddev = rdev->mddev;
3017 if (!capable(CAP_SYS_ADMIN))
3019 rv = mddev ? mddev_lock(mddev): -EBUSY;
3021 if (rdev->mddev == NULL)
3024 rv = entry->store(rdev, page, length);
3025 mddev_unlock(mddev);
3030 static void rdev_free(struct kobject *ko)
3032 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3035 static const struct sysfs_ops rdev_sysfs_ops = {
3036 .show = rdev_attr_show,
3037 .store = rdev_attr_store,
3039 static struct kobj_type rdev_ktype = {
3040 .release = rdev_free,
3041 .sysfs_ops = &rdev_sysfs_ops,
3042 .default_attrs = rdev_default_attrs,
3045 int md_rdev_init(struct md_rdev *rdev)
3048 rdev->saved_raid_disk = -1;
3049 rdev->raid_disk = -1;
3051 rdev->data_offset = 0;
3052 rdev->new_data_offset = 0;
3053 rdev->sb_events = 0;
3054 rdev->last_read_error.tv_sec = 0;
3055 rdev->last_read_error.tv_nsec = 0;
3056 rdev->sb_loaded = 0;
3057 rdev->bb_page = NULL;
3058 atomic_set(&rdev->nr_pending, 0);
3059 atomic_set(&rdev->read_errors, 0);
3060 atomic_set(&rdev->corrected_errors, 0);
3062 INIT_LIST_HEAD(&rdev->same_set);
3063 init_waitqueue_head(&rdev->blocked_wait);
3065 /* Add space to store bad block list.
3066 * This reserves the space even on arrays where it cannot
3067 * be used - I wonder if that matters
3069 rdev->badblocks.count = 0;
3070 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
3071 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3072 seqlock_init(&rdev->badblocks.lock);
3073 if (rdev->badblocks.page == NULL)
3078 EXPORT_SYMBOL_GPL(md_rdev_init);
3080 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3082 * mark the device faulty if:
3084 * - the device is nonexistent (zero size)
3085 * - the device has no valid superblock
3087 * a faulty rdev _never_ has rdev->sb set.
3089 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3091 char b[BDEVNAME_SIZE];
3093 struct md_rdev *rdev;
3096 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3098 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3099 return ERR_PTR(-ENOMEM);
3102 err = md_rdev_init(rdev);
3105 err = alloc_disk_sb(rdev);
3109 err = lock_rdev(rdev, newdev, super_format == -2);
3113 kobject_init(&rdev->kobj, &rdev_ktype);
3115 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3118 "md: %s has zero or unknown size, marking faulty!\n",
3119 bdevname(rdev->bdev,b));
3124 if (super_format >= 0) {
3125 err = super_types[super_format].
3126 load_super(rdev, NULL, super_minor);
3127 if (err == -EINVAL) {
3129 "md: %s does not have a valid v%d.%d "
3130 "superblock, not importing!\n",
3131 bdevname(rdev->bdev,b),
3132 super_format, super_minor);
3137 "md: could not read %s's sb, not importing!\n",
3138 bdevname(rdev->bdev,b));
3148 md_rdev_clear(rdev);
3150 return ERR_PTR(err);
3154 * Check a full RAID array for plausibility
3157 static void analyze_sbs(struct mddev *mddev)
3160 struct md_rdev *rdev, *freshest, *tmp;
3161 char b[BDEVNAME_SIZE];
3164 rdev_for_each_safe(rdev, tmp, mddev)
3165 switch (super_types[mddev->major_version].
3166 load_super(rdev, freshest, mddev->minor_version)) {
3174 "md: fatal superblock inconsistency in %s"
3175 " -- removing from array\n",
3176 bdevname(rdev->bdev,b));
3177 md_kick_rdev_from_array(rdev);
3180 super_types[mddev->major_version].
3181 validate_super(mddev, freshest);
3184 rdev_for_each_safe(rdev, tmp, mddev) {
3185 if (mddev->max_disks &&
3186 (rdev->desc_nr >= mddev->max_disks ||
3187 i > mddev->max_disks)) {
3189 "md: %s: %s: only %d devices permitted\n",
3190 mdname(mddev), bdevname(rdev->bdev, b),
3192 md_kick_rdev_from_array(rdev);
3195 if (rdev != freshest) {
3196 if (super_types[mddev->major_version].
3197 validate_super(mddev, rdev)) {
3198 printk(KERN_WARNING "md: kicking non-fresh %s"
3200 bdevname(rdev->bdev,b));
3201 md_kick_rdev_from_array(rdev);
3204 /* No device should have a Candidate flag
3205 * when reading devices
3207 if (test_bit(Candidate, &rdev->flags)) {
3208 pr_info("md: kicking Cluster Candidate %s from array!\n",
3209 bdevname(rdev->bdev, b));
3210 md_kick_rdev_from_array(rdev);
3213 if (mddev->level == LEVEL_MULTIPATH) {
3214 rdev->desc_nr = i++;
3215 rdev->raid_disk = rdev->desc_nr;
3216 set_bit(In_sync, &rdev->flags);
3217 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3218 rdev->raid_disk = -1;
3219 clear_bit(In_sync, &rdev->flags);
3224 /* Read a fixed-point number.
3225 * Numbers in sysfs attributes should be in "standard" units where
3226 * possible, so time should be in seconds.
3227 * However we internally use a a much smaller unit such as
3228 * milliseconds or jiffies.
3229 * This function takes a decimal number with a possible fractional
3230 * component, and produces an integer which is the result of
3231 * multiplying that number by 10^'scale'.
3232 * all without any floating-point arithmetic.
3234 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3236 unsigned long result = 0;
3238 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3241 else if (decimals < scale) {
3244 result = result * 10 + value;
3256 while (decimals < scale) {
3264 static void md_safemode_timeout(unsigned long data);
3267 safe_delay_show(struct mddev *mddev, char *page)
3269 int msec = (mddev->safemode_delay*1000)/HZ;
3270 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3273 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3277 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3280 mddev->safemode_delay = 0;
3282 unsigned long old_delay = mddev->safemode_delay;
3283 unsigned long new_delay = (msec*HZ)/1000;
3287 mddev->safemode_delay = new_delay;
3288 if (new_delay < old_delay || old_delay == 0)
3289 mod_timer(&mddev->safemode_timer, jiffies+1);
3293 static struct md_sysfs_entry md_safe_delay =
3294 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3297 level_show(struct mddev *mddev, char *page)
3299 struct md_personality *p;
3301 spin_lock(&mddev->lock);
3304 ret = sprintf(page, "%s\n", p->name);
3305 else if (mddev->clevel[0])
3306 ret = sprintf(page, "%s\n", mddev->clevel);
3307 else if (mddev->level != LEVEL_NONE)
3308 ret = sprintf(page, "%d\n", mddev->level);
3311 spin_unlock(&mddev->lock);
3316 level_store(struct mddev *mddev, const char *buf, size_t len)
3321 struct md_personality *pers, *oldpers;
3323 void *priv, *oldpriv;
3324 struct md_rdev *rdev;
3326 if (slen == 0 || slen >= sizeof(clevel))
3329 rv = mddev_lock(mddev);
3333 if (mddev->pers == NULL) {
3334 strncpy(mddev->clevel, buf, slen);
3335 if (mddev->clevel[slen-1] == '\n')
3337 mddev->clevel[slen] = 0;
3338 mddev->level = LEVEL_NONE;
3346 /* request to change the personality. Need to ensure:
3347 * - array is not engaged in resync/recovery/reshape
3348 * - old personality can be suspended
3349 * - new personality will access other array.
3353 if (mddev->sync_thread ||
3354 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3355 mddev->reshape_position != MaxSector ||
3356 mddev->sysfs_active)
3360 if (!mddev->pers->quiesce) {
3361 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3362 mdname(mddev), mddev->pers->name);
3366 /* Now find the new personality */
3367 strncpy(clevel, buf, slen);
3368 if (clevel[slen-1] == '\n')
3371 if (kstrtol(clevel, 10, &level))
3374 if (request_module("md-%s", clevel) != 0)
3375 request_module("md-level-%s", clevel);
3376 spin_lock(&pers_lock);
3377 pers = find_pers(level, clevel);
3378 if (!pers || !try_module_get(pers->owner)) {
3379 spin_unlock(&pers_lock);
3380 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3384 spin_unlock(&pers_lock);
3386 if (pers == mddev->pers) {
3387 /* Nothing to do! */
3388 module_put(pers->owner);
3392 if (!pers->takeover) {
3393 module_put(pers->owner);
3394 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3395 mdname(mddev), clevel);
3400 rdev_for_each(rdev, mddev)
3401 rdev->new_raid_disk = rdev->raid_disk;
3403 /* ->takeover must set new_* and/or delta_disks
3404 * if it succeeds, and may set them when it fails.
3406 priv = pers->takeover(mddev);
3408 mddev->new_level = mddev->level;
3409 mddev->new_layout = mddev->layout;
3410 mddev->new_chunk_sectors = mddev->chunk_sectors;
3411 mddev->raid_disks -= mddev->delta_disks;
3412 mddev->delta_disks = 0;
3413 mddev->reshape_backwards = 0;
3414 module_put(pers->owner);
3415 printk(KERN_WARNING "md: %s: %s would not accept array\n",
3416 mdname(mddev), clevel);
3421 /* Looks like we have a winner */
3422 mddev_suspend(mddev);
3423 mddev_detach(mddev);
3425 spin_lock(&mddev->lock);
3426 oldpers = mddev->pers;
3427 oldpriv = mddev->private;
3429 mddev->private = priv;
3430 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3431 mddev->level = mddev->new_level;
3432 mddev->layout = mddev->new_layout;
3433 mddev->chunk_sectors = mddev->new_chunk_sectors;
3434 mddev->delta_disks = 0;
3435 mddev->reshape_backwards = 0;
3436 mddev->degraded = 0;
3437 spin_unlock(&mddev->lock);
3439 if (oldpers->sync_request == NULL &&
3441 /* We are converting from a no-redundancy array
3442 * to a redundancy array and metadata is managed
3443 * externally so we need to be sure that writes
3444 * won't block due to a need to transition
3446 * until external management is started.
3449 mddev->safemode_delay = 0;
3450 mddev->safemode = 0;
3453 oldpers->free(mddev, oldpriv);
3455 if (oldpers->sync_request == NULL &&
3456 pers->sync_request != NULL) {
3457 /* need to add the md_redundancy_group */
3458 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3460 "md: cannot register extra attributes for %s\n",
3462 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3464 if (oldpers->sync_request != NULL &&
3465 pers->sync_request == NULL) {
3466 /* need to remove the md_redundancy_group */
3467 if (mddev->to_remove == NULL)
3468 mddev->to_remove = &md_redundancy_group;
3471 rdev_for_each(rdev, mddev) {
3472 if (rdev->raid_disk < 0)
3474 if (rdev->new_raid_disk >= mddev->raid_disks)
3475 rdev->new_raid_disk = -1;
3476 if (rdev->new_raid_disk == rdev->raid_disk)
3478 sysfs_unlink_rdev(mddev, rdev);
3480 rdev_for_each(rdev, mddev) {
3481 if (rdev->raid_disk < 0)
3483 if (rdev->new_raid_disk == rdev->raid_disk)
3485 rdev->raid_disk = rdev->new_raid_disk;
3486 if (rdev->raid_disk < 0)
3487 clear_bit(In_sync, &rdev->flags);
3489 if (sysfs_link_rdev(mddev, rdev))
3490 printk(KERN_WARNING "md: cannot register rd%d"
3491 " for %s after level change\n",
3492 rdev->raid_disk, mdname(mddev));
3496 if (pers->sync_request == NULL) {
3497 /* this is now an array without redundancy, so
3498 * it must always be in_sync
3501 del_timer_sync(&mddev->safemode_timer);
3503 blk_set_stacking_limits(&mddev->queue->limits);
3505 set_bit(MD_CHANGE_DEVS, &mddev->flags);
3506 mddev_resume(mddev);
3508 md_update_sb(mddev, 1);
3509 sysfs_notify(&mddev->kobj, NULL, "level");
3510 md_new_event(mddev);
3513 mddev_unlock(mddev);
3517 static struct md_sysfs_entry md_level =
3518 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3521 layout_show(struct mddev *mddev, char *page)
3523 /* just a number, not meaningful for all levels */
3524 if (mddev->reshape_position != MaxSector &&
3525 mddev->layout != mddev->new_layout)
3526 return sprintf(page, "%d (%d)\n",
3527 mddev->new_layout, mddev->layout);
3528 return sprintf(page, "%d\n", mddev->layout);
3532 layout_store(struct mddev *mddev, const char *buf, size_t len)
3535 unsigned long n = simple_strtoul(buf, &e, 10);
3538 if (!*buf || (*e && *e != '\n'))
3540 err = mddev_lock(mddev);
3545 if (mddev->pers->check_reshape == NULL)
3550 mddev->new_layout = n;
3551 err = mddev->pers->check_reshape(mddev);
3553 mddev->new_layout = mddev->layout;
3556 mddev->new_layout = n;
3557 if (mddev->reshape_position == MaxSector)
3560 mddev_unlock(mddev);
3563 static struct md_sysfs_entry md_layout =
3564 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3567 raid_disks_show(struct mddev *mddev, char *page)
3569 if (mddev->raid_disks == 0)
3571 if (mddev->reshape_position != MaxSector &&
3572 mddev->delta_disks != 0)
3573 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3574 mddev->raid_disks - mddev->delta_disks);
3575 return sprintf(page, "%d\n", mddev->raid_disks);
3578 static int update_raid_disks(struct mddev *mddev, int raid_disks);
3581 raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
3585 unsigned long n = simple_strtoul(buf, &e, 10);
3587 if (!*buf || (*e && *e != '\n'))
3590 err = mddev_lock(mddev);
3594 err = update_raid_disks(mddev, n);
3595 else if (mddev->reshape_position != MaxSector) {
3596 struct md_rdev *rdev;
3597 int olddisks = mddev->raid_disks - mddev->delta_disks;
3600 rdev_for_each(rdev, mddev) {
3602 rdev->data_offset < rdev->new_data_offset)
3605 rdev->data_offset > rdev->new_data_offset)
3609 mddev->delta_disks = n - olddisks;
3610 mddev->raid_disks = n;
3611 mddev->reshape_backwards = (mddev->delta_disks < 0);
3613 mddev->raid_disks = n;
3615 mddev_unlock(mddev);
3616 return err ? err : len;
3618 static struct md_sysfs_entry md_raid_disks =
3619 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3622 chunk_size_show(struct mddev *mddev, char *page)
3624 if (mddev->reshape_position != MaxSector &&
3625 mddev->chunk_sectors != mddev->new_chunk_sectors)
3626 return sprintf(page, "%d (%d)\n",
3627 mddev->new_chunk_sectors << 9,
3628 mddev->chunk_sectors << 9);
3629 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3633 chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
3637 unsigned long n = simple_strtoul(buf, &e, 10);
3639 if (!*buf || (*e && *e != '\n'))
3642 err = mddev_lock(mddev);
3646 if (mddev->pers->check_reshape == NULL)
3651 mddev->new_chunk_sectors = n >> 9;
3652 err = mddev->pers->check_reshape(mddev);
3654 mddev->new_chunk_sectors = mddev->chunk_sectors;
3657 mddev->new_chunk_sectors = n >> 9;
3658 if (mddev->reshape_position == MaxSector)
3659 mddev->chunk_sectors = n >> 9;
3661 mddev_unlock(mddev);
3664 static struct md_sysfs_entry md_chunk_size =
3665 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3668 resync_start_show(struct mddev *mddev, char *page)
3670 if (mddev->recovery_cp == MaxSector)
3671 return sprintf(page, "none\n");
3672 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3676 resync_start_store(struct mddev *mddev, const char *buf, size_t len)
3680 unsigned long long n = simple_strtoull(buf, &e, 10);
3682 err = mddev_lock(mddev);
3685 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3687 else if (cmd_match(buf, "none"))
3689 else if (!*buf || (*e && *e != '\n'))
3693 mddev->recovery_cp = n;
3695 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3697 mddev_unlock(mddev);
3700 static struct md_sysfs_entry md_resync_start =
3701 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3704 * The array state can be:
3707 * No devices, no size, no level
3708 * Equivalent to STOP_ARRAY ioctl
3710 * May have some settings, but array is not active
3711 * all IO results in error
3712 * When written, doesn't tear down array, but just stops it
3713 * suspended (not supported yet)
3714 * All IO requests will block. The array can be reconfigured.
3715 * Writing this, if accepted, will block until array is quiescent
3717 * no resync can happen. no superblocks get written.
3718 * write requests fail
3720 * like readonly, but behaves like 'clean' on a write request.
3722 * clean - no pending writes, but otherwise active.
3723 * When written to inactive array, starts without resync
3724 * If a write request arrives then
3725 * if metadata is known, mark 'dirty' and switch to 'active'.
3726 * if not known, block and switch to write-pending
3727 * If written to an active array that has pending writes, then fails.
3729 * fully active: IO and resync can be happening.
3730 * When written to inactive array, starts with resync
3733 * clean, but writes are blocked waiting for 'active' to be written.
3736 * like active, but no writes have been seen for a while (100msec).
3739 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3740 write_pending, active_idle, bad_word};
3741 static char *array_states[] = {
3742 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3743 "write-pending", "active-idle", NULL };
3745 static int match_word(const char *word, char **list)
3748 for (n=0; list[n]; n++)
3749 if (cmd_match(word, list[n]))
3755 array_state_show(struct mddev *mddev, char *page)
3757 enum array_state st = inactive;
3770 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3772 else if (mddev->safemode)
3778 if (list_empty(&mddev->disks) &&
3779 mddev->raid_disks == 0 &&
3780 mddev->dev_sectors == 0)
3785 return sprintf(page, "%s\n", array_states[st]);
3788 static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3789 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3790 static int do_md_run(struct mddev *mddev);
3791 static int restart_array(struct mddev *mddev);
3794 array_state_store(struct mddev *mddev, const char *buf, size_t len)
3797 enum array_state st = match_word(buf, array_states);
3799 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3800 /* don't take reconfig_mutex when toggling between
3803 spin_lock(&mddev->lock);
3805 restart_array(mddev);
3806 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3807 wake_up(&mddev->sb_wait);
3809 } else /* st == clean */ {
3810 restart_array(mddev);
3811 if (atomic_read(&mddev->writes_pending) == 0) {
3812 if (mddev->in_sync == 0) {
3814 if (mddev->safemode == 1)
3815 mddev->safemode = 0;
3816 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3822 spin_unlock(&mddev->lock);
3825 err = mddev_lock(mddev);
3833 /* stopping an active array */
3834 err = do_md_stop(mddev, 0, NULL);
3837 /* stopping an active array */
3839 err = do_md_stop(mddev, 2, NULL);
3841 err = 0; /* already inactive */
3844 break; /* not supported yet */
3847 err = md_set_readonly(mddev, NULL);
3850 set_disk_ro(mddev->gendisk, 1);
3851 err = do_md_run(mddev);
3857 err = md_set_readonly(mddev, NULL);
3858 else if (mddev->ro == 1)
3859 err = restart_array(mddev);
3862 set_disk_ro(mddev->gendisk, 0);
3866 err = do_md_run(mddev);
3871 restart_array(mddev);
3872 spin_lock(&mddev->lock);
3873 if (atomic_read(&mddev->writes_pending) == 0) {
3874 if (mddev->in_sync == 0) {
3876 if (mddev->safemode == 1)
3877 mddev->safemode = 0;
3878 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3883 spin_unlock(&mddev->lock);
3889 restart_array(mddev);
3890 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3891 wake_up(&mddev->sb_wait);
3895 set_disk_ro(mddev->gendisk, 0);
3896 err = do_md_run(mddev);
3901 /* these cannot be set */
3906 if (mddev->hold_active == UNTIL_IOCTL)
3907 mddev->hold_active = 0;
3908 sysfs_notify_dirent_safe(mddev->sysfs_state);
3910 mddev_unlock(mddev);
3913 static struct md_sysfs_entry md_array_state =
3914 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3917 max_corrected_read_errors_show(struct mddev *mddev, char *page) {
3918 return sprintf(page, "%d\n",
3919 atomic_read(&mddev->max_corr_read_errors));
3923 max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
3926 unsigned long n = simple_strtoul(buf, &e, 10);
3928 if (*buf && (*e == 0 || *e == '\n')) {
3929 atomic_set(&mddev->max_corr_read_errors, n);
3935 static struct md_sysfs_entry max_corr_read_errors =
3936 __ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3937 max_corrected_read_errors_store);
3940 null_show(struct mddev *mddev, char *page)
3946 new_dev_store(struct mddev *mddev, const char *buf, size_t len)
3948 /* buf must be %d:%d\n? giving major and minor numbers */
3949 /* The new device is added to the array.
3950 * If the array has a persistent superblock, we read the
3951 * superblock to initialise info and check validity.
3952 * Otherwise, only checking done is that in bind_rdev_to_array,
3953 * which mainly checks size.
3956 int major = simple_strtoul(buf, &e, 10);
3959 struct md_rdev *rdev;
3962 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3964 minor = simple_strtoul(e+1, &e, 10);
3965 if (*e && *e != '\n')
3967 dev = MKDEV(major, minor);
3968 if (major != MAJOR(dev) ||
3969 minor != MINOR(dev))
3972 flush_workqueue(md_misc_wq);
3974 err = mddev_lock(mddev);
3977 if (mddev->persistent) {
3978 rdev = md_import_device(dev, mddev->major_version,
3979 mddev->minor_version);
3980 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3981 struct md_rdev *rdev0
3982 = list_entry(mddev->disks.next,
3983 struct md_rdev, same_set);
3984 err = super_types[mddev->major_version]
3985 .load_super(rdev, rdev0, mddev->minor_version);
3989 } else if (mddev->external)
3990 rdev = md_import_device(dev, -2, -1);
3992 rdev = md_import_device(dev, -1, -1);
3995 return PTR_ERR(rdev);
3996 err = bind_rdev_to_array(rdev, mddev);
4000 mddev_unlock(mddev);
4001 return err ? err : len;
4004 static struct md_sysfs_entry md_new_device =
4005 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
4008 bitmap_store(struct mddev *mddev, const char *buf, size_t len)
4011 unsigned long chunk, end_chunk;
4014 err = mddev_lock(mddev);
4019 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4021 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4022 if (buf == end) break;
4023 if (*end == '-') { /* range */
4025 end_chunk = simple_strtoul(buf, &end, 0);
4026 if (buf == end) break;
4028 if (*end && !isspace(*end)) break;
4029 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
4030 buf = skip_spaces(end);
4032 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4034 mddev_unlock(mddev);
4038 static struct md_sysfs_entry md_bitmap =
4039 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4042 size_show(struct mddev *mddev, char *page)
4044 return sprintf(page, "%llu\n",
4045 (unsigned long long)mddev->dev_sectors / 2);
4048 static int update_size(struct mddev *mddev, sector_t num_sectors);
4051 size_store(struct mddev *mddev, const char *buf, size_t len)
4053 /* If array is inactive, we can reduce the component size, but
4054 * not increase it (except from 0).
4055 * If array is active, we can try an on-line resize
4058 int err = strict_blocks_to_sectors(buf, §ors);
4062 err = mddev_lock(mddev);
4066 if (mddev_is_clustered(mddev))
4067 md_cluster_ops->metadata_update_start(mddev);
4068 err = update_size(mddev, sectors);
4069 md_update_sb(mddev, 1);
4070 if (mddev_is_clustered(mddev))
4071 md_cluster_ops->metadata_update_finish(mddev);
4073 if (mddev->dev_sectors == 0 ||
4074 mddev->dev_sectors > sectors)
4075 mddev->dev_sectors = sectors;
4079 mddev_unlock(mddev);
4080 return err ? err : len;
4083 static struct md_sysfs_entry md_size =
4084 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
4086 /* Metadata version.
4088 * 'none' for arrays with no metadata (good luck...)
4089 * 'external' for arrays with externally managed metadata,
4090 * or N.M for internally known formats
4093 metadata_show(struct mddev *mddev, char *page)
4095 if (mddev->persistent)
4096 return sprintf(page, "%d.%d\n",
4097 mddev->major_version, mddev->minor_version);
4098 else if (mddev->external)
4099 return sprintf(page, "external:%s\n", mddev->metadata_type);
4101 return sprintf(page, "none\n");
4105 metadata_store(struct mddev *mddev, const char *buf, size_t len)
4110 /* Changing the details of 'external' metadata is
4111 * always permitted. Otherwise there must be
4112 * no devices attached to the array.
4115 err = mddev_lock(mddev);
4119 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4121 else if (!list_empty(&mddev->disks))
4125 if (cmd_match(buf, "none")) {
4126 mddev->persistent = 0;
4127 mddev->external = 0;
4128 mddev->major_version = 0;
4129 mddev->minor_version = 90;
4132 if (strncmp(buf, "external:", 9) == 0) {
4133 size_t namelen = len-9;
4134 if (namelen >= sizeof(mddev->metadata_type))
4135 namelen = sizeof(mddev->metadata_type)-1;
4136 strncpy(mddev->metadata_type, buf+9, namelen);
4137 mddev->metadata_type[namelen] = 0;
4138 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4139 mddev->metadata_type[--namelen] = 0;
4140 mddev->persistent = 0;
4141 mddev->external = 1;
4142 mddev->major_version = 0;
4143 mddev->minor_version = 90;
4146 major = simple_strtoul(buf, &e, 10);
4148 if (e==buf || *e != '.')
4151 minor = simple_strtoul(buf, &e, 10);
4152 if (e==buf || (*e && *e != '\n') )
4155 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4157 mddev->major_version = major;
4158 mddev->minor_version = minor;
4159 mddev->persistent = 1;
4160 mddev->external = 0;
4163 mddev_unlock(mddev);
4167 static struct md_sysfs_entry md_metadata =
4168 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4171 action_show(struct mddev *mddev, char *page)
4173 char *type = "idle";
4174 unsigned long recovery = mddev->recovery;
4175 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
4177 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4178 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4179 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
4181 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4182 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
4184 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
4188 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
4191 return sprintf(page, "%s\n", type);
4195 action_store(struct mddev *mddev, const char *page, size_t len)
4197 if (!mddev->pers || !mddev->pers->sync_request)
4200 if (cmd_match(page, "frozen"))
4201 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4203 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4205 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4206 flush_workqueue(md_misc_wq);
4207 if (mddev->sync_thread) {
4208 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4209 if (mddev_lock(mddev) == 0) {
4210 md_reap_sync_thread(mddev);
4211 mddev_unlock(mddev);
4214 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4215 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4217 else if (cmd_match(page, "resync"))
4218 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4219 else if (cmd_match(page, "recover")) {
4220 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4221 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4222 } else if (cmd_match(page, "reshape")) {
4224 if (mddev->pers->start_reshape == NULL)
4226 err = mddev_lock(mddev);
4228 err = mddev->pers->start_reshape(mddev);
4229 mddev_unlock(mddev);
4233 sysfs_notify(&mddev->kobj, NULL, "degraded");
4235 if (cmd_match(page, "check"))
4236 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4237 else if (!cmd_match(page, "repair"))
4239 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4240 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4242 if (mddev->ro == 2) {
4243 /* A write to sync_action is enough to justify
4244 * canceling read-auto mode
4247 md_wakeup_thread(mddev->sync_thread);
4249 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4250 md_wakeup_thread(mddev->thread);
4251 sysfs_notify_dirent_safe(mddev->sysfs_action);
4255 static struct md_sysfs_entry md_scan_mode =
4256 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4259 last_sync_action_show(struct mddev *mddev, char *page)
4261 return sprintf(page, "%s\n", mddev->last_sync_action);
4264 static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4267 mismatch_cnt_show(struct mddev *mddev, char *page)
4269 return sprintf(page, "%llu\n",
4270 (unsigned long long)
4271 atomic64_read(&mddev->resync_mismatches));
4274 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4277 sync_min_show(struct mddev *mddev, char *page)
4279 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4280 mddev->sync_speed_min ? "local": "system");
4284 sync_min_store(struct mddev *mddev, const char *buf, size_t len)
4288 if (strncmp(buf, "system", 6)==0) {
4289 mddev->sync_speed_min = 0;
4292 min = simple_strtoul(buf, &e, 10);
4293 if (buf == e || (*e && *e != '\n') || min <= 0)
4295 mddev->sync_speed_min = min;
4299 static struct md_sysfs_entry md_sync_min =
4300 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4303 sync_max_show(struct mddev *mddev, char *page)
4305 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4306 mddev->sync_speed_max ? "local": "system");
4310 sync_max_store(struct mddev *mddev, const char *buf, size_t len)
4314 if (strncmp(buf, "system", 6)==0) {
4315 mddev->sync_speed_max = 0;
4318 max = simple_strtoul(buf, &e, 10);
4319 if (buf == e || (*e && *e != '\n') || max <= 0)
4321 mddev->sync_speed_max = max;
4325 static struct md_sysfs_entry md_sync_max =
4326 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4329 degraded_show(struct mddev *mddev, char *page)
4331 return sprintf(page, "%d\n", mddev->degraded);
4333 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4336 sync_force_parallel_show(struct mddev *mddev, char *page)
4338 return sprintf(page, "%d\n", mddev->parallel_resync);
4342 sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
4346 if (kstrtol(buf, 10, &n))
4349 if (n != 0 && n != 1)
4352 mddev->parallel_resync = n;
4354 if (mddev->sync_thread)
4355 wake_up(&resync_wait);
4360 /* force parallel resync, even with shared block devices */
4361 static struct md_sysfs_entry md_sync_force_parallel =
4362 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4363 sync_force_parallel_show, sync_force_parallel_store);
4366 sync_speed_show(struct mddev *mddev, char *page)
4368 unsigned long resync, dt, db;
4369 if (mddev->curr_resync == 0)
4370 return sprintf(page, "none\n");
4371 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4372 dt = (jiffies - mddev->resync_mark) / HZ;
4374 db = resync - mddev->resync_mark_cnt;
4375 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4378 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4381 sync_completed_show(struct mddev *mddev, char *page)
4383 unsigned long long max_sectors, resync;
4385 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4386 return sprintf(page, "none\n");
4388 if (mddev->curr_resync == 1 ||
4389 mddev->curr_resync == 2)
4390 return sprintf(page, "delayed\n");
4392 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4393 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4394 max_sectors = mddev->resync_max_sectors;
4396 max_sectors = mddev->dev_sectors;
4398 resync = mddev->curr_resync_completed;
4399 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4402 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4405 min_sync_show(struct mddev *mddev, char *page)
4407 return sprintf(page, "%llu\n",
4408 (unsigned long long)mddev->resync_min);
4411 min_sync_store(struct mddev *mddev, const char *buf, size_t len)
4413 unsigned long long min;
4417 if (kstrtoull(buf, 10, &min))
4420 spin_lock(&mddev->lock);
4422 if (min > mddev->resync_max)
4426 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4429 /* Must be a multiple of chunk_size */
4430 chunk = mddev->chunk_sectors;
4432 sector_t temp = min;
4435 if (sector_div(temp, chunk))
4438 mddev->resync_min = min;
4442 spin_unlock(&mddev->lock);
4446 static struct md_sysfs_entry md_min_sync =
4447 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4450 max_sync_show(struct mddev *mddev, char *page)
4452 if (mddev->resync_max == MaxSector)
4453 return sprintf(page, "max\n");
4455 return sprintf(page, "%llu\n",
4456 (unsigned long long)mddev->resync_max);
4459 max_sync_store(struct mddev *mddev, const char *buf, size_t len)
4462 spin_lock(&mddev->lock);
4463 if (strncmp(buf, "max", 3) == 0)
4464 mddev->resync_max = MaxSector;
4466 unsigned long long max;
4470 if (kstrtoull(buf, 10, &max))
4472 if (max < mddev->resync_min)
4476 if (max < mddev->resync_max &&
4478 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4481 /* Must be a multiple of chunk_size */
4482 chunk = mddev->chunk_sectors;
4484 sector_t temp = max;
4487 if (sector_div(temp, chunk))
4490 mddev->resync_max = max;
4492 wake_up(&mddev->recovery_wait);
4495 spin_unlock(&mddev->lock);
4499 static struct md_sysfs_entry md_max_sync =
4500 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4503 suspend_lo_show(struct mddev *mddev, char *page)
4505 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4509 suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
4512 unsigned long long new = simple_strtoull(buf, &e, 10);
4513 unsigned long long old;
4516 if (buf == e || (*e && *e != '\n'))
4519 err = mddev_lock(mddev);
4523 if (mddev->pers == NULL ||
4524 mddev->pers->quiesce == NULL)
4526 old = mddev->suspend_lo;
4527 mddev->suspend_lo = new;
4529 /* Shrinking suspended region */
4530 mddev->pers->quiesce(mddev, 2);
4532 /* Expanding suspended region - need to wait */
4533 mddev->pers->quiesce(mddev, 1);
4534 mddev->pers->quiesce(mddev, 0);
4538 mddev_unlock(mddev);
4541 static struct md_sysfs_entry md_suspend_lo =
4542 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4545 suspend_hi_show(struct mddev *mddev, char *page)
4547 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4551 suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
4554 unsigned long long new = simple_strtoull(buf, &e, 10);
4555 unsigned long long old;
4558 if (buf == e || (*e && *e != '\n'))
4561 err = mddev_lock(mddev);
4565 if (mddev->pers == NULL ||
4566 mddev->pers->quiesce == NULL)
4568 old = mddev->suspend_hi;
4569 mddev->suspend_hi = new;
4571 /* Shrinking suspended region */
4572 mddev->pers->quiesce(mddev, 2);
4574 /* Expanding suspended region - need to wait */
4575 mddev->pers->quiesce(mddev, 1);
4576 mddev->pers->quiesce(mddev, 0);
4580 mddev_unlock(mddev);
4583 static struct md_sysfs_entry md_suspend_hi =
4584 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4587 reshape_position_show(struct mddev *mddev, char *page)
4589 if (mddev->reshape_position != MaxSector)
4590 return sprintf(page, "%llu\n",
4591 (unsigned long long)mddev->reshape_position);
4592 strcpy(page, "none\n");
4597 reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
4599 struct md_rdev *rdev;
4602 unsigned long long new = simple_strtoull(buf, &e, 10);
4604 if (buf == e || (*e && *e != '\n'))
4606 err = mddev_lock(mddev);
4612 mddev->reshape_position = new;
4613 mddev->delta_disks = 0;
4614 mddev->reshape_backwards = 0;
4615 mddev->new_level = mddev->level;
4616 mddev->new_layout = mddev->layout;
4617 mddev->new_chunk_sectors = mddev->chunk_sectors;
4618 rdev_for_each(rdev, mddev)
4619 rdev->new_data_offset = rdev->data_offset;
4622 mddev_unlock(mddev);
4626 static struct md_sysfs_entry md_reshape_position =
4627 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4628 reshape_position_store);
4631 reshape_direction_show(struct mddev *mddev, char *page)
4633 return sprintf(page, "%s\n",
4634 mddev->reshape_backwards ? "backwards" : "forwards");
4638 reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4643 if (cmd_match(buf, "forwards"))
4645 else if (cmd_match(buf, "backwards"))
4649 if (mddev->reshape_backwards == backwards)
4652 err = mddev_lock(mddev);
4655 /* check if we are allowed to change */
4656 if (mddev->delta_disks)
4658 else if (mddev->persistent &&
4659 mddev->major_version == 0)
4662 mddev->reshape_backwards = backwards;
4663 mddev_unlock(mddev);
4667 static struct md_sysfs_entry md_reshape_direction =
4668 __ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4669 reshape_direction_store);
4672 array_size_show(struct mddev *mddev, char *page)
4674 if (mddev->external_size)
4675 return sprintf(page, "%llu\n",
4676 (unsigned long long)mddev->array_sectors/2);
4678 return sprintf(page, "default\n");
4682 array_size_store(struct mddev *mddev, const char *buf, size_t len)
4687 err = mddev_lock(mddev);
4691 if (strncmp(buf, "default", 7) == 0) {
4693 sectors = mddev->pers->size(mddev, 0, 0);
4695 sectors = mddev->array_sectors;
4697 mddev->external_size = 0;
4699 if (strict_blocks_to_sectors(buf, §ors) < 0)
4701 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4704 mddev->external_size = 1;
4708 mddev->array_sectors = sectors;
4710 set_capacity(mddev->gendisk, mddev->array_sectors);
4711 revalidate_disk(mddev->gendisk);
4714 mddev_unlock(mddev);
4718 static struct md_sysfs_entry md_array_size =
4719 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4722 static struct attribute *md_default_attrs[] = {
4725 &md_raid_disks.attr,
4726 &md_chunk_size.attr,
4728 &md_resync_start.attr,
4730 &md_new_device.attr,
4731 &md_safe_delay.attr,
4732 &md_array_state.attr,
4733 &md_reshape_position.attr,
4734 &md_reshape_direction.attr,
4735 &md_array_size.attr,
4736 &max_corr_read_errors.attr,
4740 static struct attribute *md_redundancy_attrs[] = {
4742 &md_last_scan_mode.attr,
4743 &md_mismatches.attr,
4746 &md_sync_speed.attr,
4747 &md_sync_force_parallel.attr,
4748 &md_sync_completed.attr,
4751 &md_suspend_lo.attr,
4752 &md_suspend_hi.attr,
4757 static struct attribute_group md_redundancy_group = {
4759 .attrs = md_redundancy_attrs,
4763 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4765 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4766 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4771 spin_lock(&all_mddevs_lock);
4772 if (list_empty(&mddev->all_mddevs)) {
4773 spin_unlock(&all_mddevs_lock);
4777 spin_unlock(&all_mddevs_lock);
4779 rv = entry->show(mddev, page);
4785 md_attr_store(struct kobject *kobj, struct attribute *attr,
4786 const char *page, size_t length)
4788 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4789 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
4794 if (!capable(CAP_SYS_ADMIN))
4796 spin_lock(&all_mddevs_lock);
4797 if (list_empty(&mddev->all_mddevs)) {
4798 spin_unlock(&all_mddevs_lock);
4802 spin_unlock(&all_mddevs_lock);
4803 rv = entry->store(mddev, page, length);
4808 static void md_free(struct kobject *ko)
4810 struct mddev *mddev = container_of(ko, struct mddev, kobj);
4812 if (mddev->sysfs_state)
4813 sysfs_put(mddev->sysfs_state);
4815 if (mddev->gendisk) {
4816 del_gendisk(mddev->gendisk);
4817 put_disk(mddev->gendisk);
4820 blk_cleanup_queue(mddev->queue);
4825 static const struct sysfs_ops md_sysfs_ops = {
4826 .show = md_attr_show,
4827 .store = md_attr_store,
4829 static struct kobj_type md_ktype = {
4831 .sysfs_ops = &md_sysfs_ops,
4832 .default_attrs = md_default_attrs,
4837 static void mddev_delayed_delete(struct work_struct *ws)
4839 struct mddev *mddev = container_of(ws, struct mddev, del_work);
4841 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4842 kobject_del(&mddev->kobj);
4843 kobject_put(&mddev->kobj);
4846 static int md_alloc(dev_t dev, char *name)
4848 static DEFINE_MUTEX(disks_mutex);
4849 struct mddev *mddev = mddev_find(dev);
4850 struct gendisk *disk;
4859 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4860 shift = partitioned ? MdpMinorShift : 0;
4861 unit = MINOR(mddev->unit) >> shift;
4863 /* wait for any previous instance of this device to be
4864 * completely removed (mddev_delayed_delete).
4866 flush_workqueue(md_misc_wq);
4868 mutex_lock(&disks_mutex);
4874 /* Need to ensure that 'name' is not a duplicate.
4876 struct mddev *mddev2;
4877 spin_lock(&all_mddevs_lock);
4879 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4880 if (mddev2->gendisk &&
4881 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4882 spin_unlock(&all_mddevs_lock);
4885 spin_unlock(&all_mddevs_lock);
4889 mddev->queue = blk_alloc_queue(GFP_KERNEL);
4892 mddev->queue->queuedata = mddev;
4894 blk_queue_make_request(mddev->queue, md_make_request);
4895 blk_set_stacking_limits(&mddev->queue->limits);
4897 disk = alloc_disk(1 << shift);
4899 blk_cleanup_queue(mddev->queue);
4900 mddev->queue = NULL;
4903 disk->major = MAJOR(mddev->unit);
4904 disk->first_minor = unit << shift;
4906 strcpy(disk->disk_name, name);
4907 else if (partitioned)
4908 sprintf(disk->disk_name, "md_d%d", unit);
4910 sprintf(disk->disk_name, "md%d", unit);
4911 disk->fops = &md_fops;
4912 disk->private_data = mddev;
4913 disk->queue = mddev->queue;
4914 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4915 /* Allow extended partitions. This makes the
4916 * 'mdp' device redundant, but we can't really
4919 disk->flags |= GENHD_FL_EXT_DEVT;
4920 mddev->gendisk = disk;
4921 /* As soon as we call add_disk(), another thread could get
4922 * through to md_open, so make sure it doesn't get too far
4924 mutex_lock(&mddev->open_mutex);
4927 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4928 &disk_to_dev(disk)->kobj, "%s", "md");
4930 /* This isn't possible, but as kobject_init_and_add is marked
4931 * __must_check, we must do something with the result
4933 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4937 if (mddev->kobj.sd &&
4938 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4939 printk(KERN_DEBUG "pointless warning\n");
4940 mutex_unlock(&mddev->open_mutex);
4942 mutex_unlock(&disks_mutex);
4943 if (!error && mddev->kobj.sd) {
4944 kobject_uevent(&mddev->kobj, KOBJ_ADD);
4945 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4951 static struct kobject *md_probe(dev_t dev, int *part, void *data)
4953 md_alloc(dev, NULL);
4957 static int add_named_array(const char *val, struct kernel_param *kp)
4959 /* val must be "md_*" where * is not all digits.
4960 * We allocate an array with a large free minor number, and
4961 * set the name to val. val must not already be an active name.
4963 int len = strlen(val);
4964 char buf[DISK_NAME_LEN];
4966 while (len && val[len-1] == '\n')
4968 if (len >= DISK_NAME_LEN)
4970 strlcpy(buf, val, len+1);
4971 if (strncmp(buf, "md_", 3) != 0)
4973 return md_alloc(0, buf);
4976 static void md_safemode_timeout(unsigned long data)
4978 struct mddev *mddev = (struct mddev *) data;
4980 if (!atomic_read(&mddev->writes_pending)) {
4981 mddev->safemode = 1;
4982 if (mddev->external)
4983 sysfs_notify_dirent_safe(mddev->sysfs_state);
4985 md_wakeup_thread(mddev->thread);
4988 static int start_dirty_degraded;
4990 int md_run(struct mddev *mddev)
4993 struct md_rdev *rdev;
4994 struct md_personality *pers;
4996 if (list_empty(&mddev->disks))
4997 /* cannot run an array with no devices.. */
5002 /* Cannot run until previous stop completes properly */
5003 if (mddev->sysfs_active)
5007 * Analyze all RAID superblock(s)
5009 if (!mddev->raid_disks) {
5010 if (!mddev->persistent)
5015 if (mddev->level != LEVEL_NONE)
5016 request_module("md-level-%d", mddev->level);
5017 else if (mddev->clevel[0])
5018 request_module("md-%s", mddev->clevel);
5021 * Drop all container device buffers, from now on
5022 * the only valid external interface is through the md
5025 rdev_for_each(rdev, mddev) {
5026 if (test_bit(Faulty, &rdev->flags))
5028 sync_blockdev(rdev->bdev);
5029 invalidate_bdev(rdev->bdev);
5031 /* perform some consistency tests on the device.
5032 * We don't want the data to overlap the metadata,
5033 * Internal Bitmap issues have been handled elsewhere.
5035 if (rdev->meta_bdev) {
5036 /* Nothing to check */;
5037 } else if (rdev->data_offset < rdev->sb_start) {
5038 if (mddev->dev_sectors &&
5039 rdev->data_offset + mddev->dev_sectors
5041 printk("md: %s: data overlaps metadata\n",
5046 if (rdev->sb_start + rdev->sb_size/512
5047 > rdev->data_offset) {
5048 printk("md: %s: metadata overlaps data\n",
5053 sysfs_notify_dirent_safe(rdev->sysfs_state);
5056 if (mddev->bio_set == NULL)
5057 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
5059 spin_lock(&pers_lock);
5060 pers = find_pers(mddev->level, mddev->clevel);
5061 if (!pers || !try_module_get(pers->owner)) {
5062 spin_unlock(&pers_lock);
5063 if (mddev->level != LEVEL_NONE)
5064 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5067 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5071 spin_unlock(&pers_lock);
5072 if (mddev->level != pers->level) {
5073 mddev->level = pers->level;
5074 mddev->new_level = pers->level;
5076 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
5078 if (mddev->reshape_position != MaxSector &&
5079 pers->start_reshape == NULL) {
5080 /* This personality cannot handle reshaping... */
5081 module_put(pers->owner);
5085 if (pers->sync_request) {
5086 /* Warn if this is a potentially silly
5089 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5090 struct md_rdev *rdev2;
5093 rdev_for_each(rdev, mddev)
5094 rdev_for_each(rdev2, mddev) {
5096 rdev->bdev->bd_contains ==
5097 rdev2->bdev->bd_contains) {
5099 "%s: WARNING: %s appears to be"
5100 " on the same physical disk as"
5103 bdevname(rdev->bdev,b),
5104 bdevname(rdev2->bdev,b2));
5111 "True protection against single-disk"
5112 " failure might be compromised.\n");
5115 mddev->recovery = 0;
5116 /* may be over-ridden by personality */
5117 mddev->resync_max_sectors = mddev->dev_sectors;
5119 mddev->ok_start_degraded = start_dirty_degraded;
5121 if (start_readonly && mddev->ro == 0)
5122 mddev->ro = 2; /* read-only, but switch on first write */
5124 err = pers->run(mddev);
5126 printk(KERN_ERR "md: pers->run() failed ...\n");
5127 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
5128 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5129 " but 'external_size' not in effect?\n", __func__);
5131 "md: invalid array_size %llu > default size %llu\n",
5132 (unsigned long long)mddev->array_sectors / 2,
5133 (unsigned long long)pers->size(mddev, 0, 0) / 2);
5136 if (err == 0 && pers->sync_request &&
5137 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
5138 struct bitmap *bitmap;
5140 bitmap = bitmap_create(mddev, -1);
5141 if (IS_ERR(bitmap)) {
5142 err = PTR_ERR(bitmap);
5143 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5144 mdname(mddev), err);
5146 mddev->bitmap = bitmap;
5150 mddev_detach(mddev);
5151 pers->free(mddev, mddev->private);
5152 module_put(pers->owner);
5153 bitmap_destroy(mddev);
5157 mddev->queue->backing_dev_info.congested_data = mddev;
5158 mddev->queue->backing_dev_info.congested_fn = md_congested;
5159 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
5161 if (pers->sync_request) {
5162 if (mddev->kobj.sd &&
5163 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
5165 "md: cannot register extra attributes for %s\n",
5167 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
5168 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
5171 atomic_set(&mddev->writes_pending,0);
5172 atomic_set(&mddev->max_corr_read_errors,
5173 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
5174 mddev->safemode = 0;
5175 mddev->safemode_timer.function = md_safemode_timeout;
5176 mddev->safemode_timer.data = (unsigned long) mddev;
5177 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
5180 spin_lock(&mddev->lock);
5183 spin_unlock(&mddev->lock);
5184 rdev_for_each(rdev, mddev)
5185 if (rdev->raid_disk >= 0)
5186 if (sysfs_link_rdev(mddev, rdev))
5187 /* failure here is OK */;
5189 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5191 if (mddev->flags & MD_UPDATE_SB_FLAGS)
5192 md_update_sb(mddev, 0);
5194 md_new_event(mddev);
5195 sysfs_notify_dirent_safe(mddev->sysfs_state);
5196 sysfs_notify_dirent_safe(mddev->sysfs_action);
5197 sysfs_notify(&mddev->kobj, NULL, "degraded");
5200 EXPORT_SYMBOL_GPL(md_run);
5202 static int do_md_run(struct mddev *mddev)
5206 err = md_run(mddev);
5209 err = bitmap_load(mddev);
5211 bitmap_destroy(mddev);
5215 md_wakeup_thread(mddev->thread);
5216 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5218 set_capacity(mddev->gendisk, mddev->array_sectors);
5219 revalidate_disk(mddev->gendisk);
5221 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5226 static int restart_array(struct mddev *mddev)
5228 struct gendisk *disk = mddev->gendisk;
5230 /* Complain if it has no devices */
5231 if (list_empty(&mddev->disks))
5237 mddev->safemode = 0;
5239 set_disk_ro(disk, 0);
5240 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5242 /* Kick recovery or resync if necessary */
5243 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5244 md_wakeup_thread(mddev->thread);
5245 md_wakeup_thread(mddev->sync_thread);
5246 sysfs_notify_dirent_safe(mddev->sysfs_state);
5250 static void md_clean(struct mddev *mddev)
5252 mddev->array_sectors = 0;
5253 mddev->external_size = 0;
5254 mddev->dev_sectors = 0;
5255 mddev->raid_disks = 0;
5256 mddev->recovery_cp = 0;
5257 mddev->resync_min = 0;
5258 mddev->resync_max = MaxSector;
5259 mddev->reshape_position = MaxSector;
5260 mddev->external = 0;
5261 mddev->persistent = 0;
5262 mddev->level = LEVEL_NONE;
5263 mddev->clevel[0] = 0;
5266 mddev->metadata_type[0] = 0;
5267 mddev->chunk_sectors = 0;
5268 mddev->ctime = mddev->utime = 0;
5270 mddev->max_disks = 0;
5272 mddev->can_decrease_events = 0;
5273 mddev->delta_disks = 0;
5274 mddev->reshape_backwards = 0;
5275 mddev->new_level = LEVEL_NONE;
5276 mddev->new_layout = 0;
5277 mddev->new_chunk_sectors = 0;
5278 mddev->curr_resync = 0;
5279 atomic64_set(&mddev->resync_mismatches, 0);
5280 mddev->suspend_lo = mddev->suspend_hi = 0;
5281 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5282 mddev->recovery = 0;
5285 mddev->degraded = 0;
5286 mddev->safemode = 0;
5287 mddev->merge_check_needed = 0;
5288 mddev->bitmap_info.offset = 0;
5289 mddev->bitmap_info.default_offset = 0;
5290 mddev->bitmap_info.default_space = 0;
5291 mddev->bitmap_info.chunksize = 0;
5292 mddev->bitmap_info.daemon_sleep = 0;
5293 mddev->bitmap_info.max_write_behind = 0;
5296 static void __md_stop_writes(struct mddev *mddev)
5298 if (mddev_is_clustered(mddev))
5299 md_cluster_ops->metadata_update_start(mddev);
5300 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5301 flush_workqueue(md_misc_wq);
5302 if (mddev->sync_thread) {
5303 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5304 md_reap_sync_thread(mddev);
5307 del_timer_sync(&mddev->safemode_timer);
5309 bitmap_flush(mddev);
5310 md_super_wait(mddev);
5312 if (mddev->ro == 0 &&
5313 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
5314 /* mark array as shutdown cleanly */
5316 md_update_sb(mddev, 1);
5318 if (mddev_is_clustered(mddev))
5319 md_cluster_ops->metadata_update_finish(mddev);
5322 void md_stop_writes(struct mddev *mddev)
5324 mddev_lock_nointr(mddev);
5325 __md_stop_writes(mddev);
5326 mddev_unlock(mddev);
5328 EXPORT_SYMBOL_GPL(md_stop_writes);
5330 static void mddev_detach(struct mddev *mddev)
5332 struct bitmap *bitmap = mddev->bitmap;
5333 /* wait for behind writes to complete */
5334 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5335 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5337 /* need to kick something here to make sure I/O goes? */
5338 wait_event(bitmap->behind_wait,
5339 atomic_read(&bitmap->behind_writes) == 0);
5341 if (mddev->pers && mddev->pers->quiesce) {
5342 mddev->pers->quiesce(mddev, 1);
5343 mddev->pers->quiesce(mddev, 0);
5345 md_unregister_thread(&mddev->thread);
5347 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5350 static void __md_stop(struct mddev *mddev)
5352 struct md_personality *pers = mddev->pers;
5353 mddev_detach(mddev);
5354 spin_lock(&mddev->lock);
5357 spin_unlock(&mddev->lock);
5358 pers->free(mddev, mddev->private);
5359 if (pers->sync_request && mddev->to_remove == NULL)
5360 mddev->to_remove = &md_redundancy_group;
5361 module_put(pers->owner);
5362 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5365 void md_stop(struct mddev *mddev)
5367 /* stop the array and free an attached data structures.
5368 * This is called from dm-raid
5371 bitmap_destroy(mddev);
5373 bioset_free(mddev->bio_set);
5376 EXPORT_SYMBOL_GPL(md_stop);
5378 static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
5383 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5385 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5386 md_wakeup_thread(mddev->thread);
5388 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5389 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5390 if (mddev->sync_thread)
5391 /* Thread might be blocked waiting for metadata update
5392 * which will now never happen */
5393 wake_up_process(mddev->sync_thread->tsk);
5395 mddev_unlock(mddev);
5396 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5398 mddev_lock_nointr(mddev);
5400 mutex_lock(&mddev->open_mutex);
5401 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5402 mddev->sync_thread ||
5403 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5404 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5405 printk("md: %s still in use.\n",mdname(mddev));
5407 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5408 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5409 md_wakeup_thread(mddev->thread);
5415 __md_stop_writes(mddev);
5421 set_disk_ro(mddev->gendisk, 1);
5422 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5423 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5424 md_wakeup_thread(mddev->thread);
5425 sysfs_notify_dirent_safe(mddev->sysfs_state);
5429 mutex_unlock(&mddev->open_mutex);
5434 * 0 - completely stop and dis-assemble array
5435 * 2 - stop but do not disassemble array
5437 static int do_md_stop(struct mddev *mddev, int mode,
5438 struct block_device *bdev)
5440 struct gendisk *disk = mddev->gendisk;
5441 struct md_rdev *rdev;
5444 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5446 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5447 md_wakeup_thread(mddev->thread);
5449 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
5450 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5451 if (mddev->sync_thread)
5452 /* Thread might be blocked waiting for metadata update
5453 * which will now never happen */
5454 wake_up_process(mddev->sync_thread->tsk);
5456 mddev_unlock(mddev);
5457 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5458 !test_bit(MD_RECOVERY_RUNNING,
5459 &mddev->recovery)));
5460 mddev_lock_nointr(mddev);
5462 mutex_lock(&mddev->open_mutex);
5463 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
5464 mddev->sysfs_active ||
5465 mddev->sync_thread ||
5466 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
5467 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
5468 printk("md: %s still in use.\n",mdname(mddev));
5469 mutex_unlock(&mddev->open_mutex);
5471 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5472 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5473 md_wakeup_thread(mddev->thread);
5479 set_disk_ro(disk, 0);
5481 __md_stop_writes(mddev);
5483 mddev->queue->merge_bvec_fn = NULL;
5484 mddev->queue->backing_dev_info.congested_fn = NULL;
5486 /* tell userspace to handle 'inactive' */
5487 sysfs_notify_dirent_safe(mddev->sysfs_state);
5489 rdev_for_each(rdev, mddev)
5490 if (rdev->raid_disk >= 0)
5491 sysfs_unlink_rdev(mddev, rdev);
5493 set_capacity(disk, 0);
5494 mutex_unlock(&mddev->open_mutex);
5496 revalidate_disk(disk);
5501 mutex_unlock(&mddev->open_mutex);
5503 * Free resources if final stop
5506 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5508 bitmap_destroy(mddev);
5509 if (mddev->bitmap_info.file) {
5510 struct file *f = mddev->bitmap_info.file;
5511 spin_lock(&mddev->lock);
5512 mddev->bitmap_info.file = NULL;
5513 spin_unlock(&mddev->lock);
5516 mddev->bitmap_info.offset = 0;
5518 export_array(mddev);
5521 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5522 if (mddev->hold_active == UNTIL_STOP)
5523 mddev->hold_active = 0;
5525 blk_integrity_unregister(disk);
5526 md_new_event(mddev);
5527 sysfs_notify_dirent_safe(mddev->sysfs_state);
5532 static void autorun_array(struct mddev *mddev)
5534 struct md_rdev *rdev;
5537 if (list_empty(&mddev->disks))
5540 printk(KERN_INFO "md: running: ");
5542 rdev_for_each(rdev, mddev) {
5543 char b[BDEVNAME_SIZE];
5544 printk("<%s>", bdevname(rdev->bdev,b));
5548 err = do_md_run(mddev);
5550 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5551 do_md_stop(mddev, 0, NULL);
5556 * lets try to run arrays based on all disks that have arrived
5557 * until now. (those are in pending_raid_disks)
5559 * the method: pick the first pending disk, collect all disks with
5560 * the same UUID, remove all from the pending list and put them into
5561 * the 'same_array' list. Then order this list based on superblock
5562 * update time (freshest comes first), kick out 'old' disks and
5563 * compare superblocks. If everything's fine then run it.
5565 * If "unit" is allocated, then bump its reference count
5567 static void autorun_devices(int part)
5569 struct md_rdev *rdev0, *rdev, *tmp;
5570 struct mddev *mddev;
5571 char b[BDEVNAME_SIZE];
5573 printk(KERN_INFO "md: autorun ...\n");
5574 while (!list_empty(&pending_raid_disks)) {
5577 LIST_HEAD(candidates);
5578 rdev0 = list_entry(pending_raid_disks.next,
5579 struct md_rdev, same_set);
5581 printk(KERN_INFO "md: considering %s ...\n",
5582 bdevname(rdev0->bdev,b));
5583 INIT_LIST_HEAD(&candidates);
5584 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5585 if (super_90_load(rdev, rdev0, 0) >= 0) {
5586 printk(KERN_INFO "md: adding %s ...\n",
5587 bdevname(rdev->bdev,b));
5588 list_move(&rdev->same_set, &candidates);
5591 * now we have a set of devices, with all of them having
5592 * mostly sane superblocks. It's time to allocate the
5596 dev = MKDEV(mdp_major,
5597 rdev0->preferred_minor << MdpMinorShift);
5598 unit = MINOR(dev) >> MdpMinorShift;
5600 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5603 if (rdev0->preferred_minor != unit) {
5604 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5605 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5609 md_probe(dev, NULL, NULL);
5610 mddev = mddev_find(dev);
5611 if (!mddev || !mddev->gendisk) {
5615 "md: cannot allocate memory for md drive.\n");
5618 if (mddev_lock(mddev))
5619 printk(KERN_WARNING "md: %s locked, cannot run\n",
5621 else if (mddev->raid_disks || mddev->major_version
5622 || !list_empty(&mddev->disks)) {
5624 "md: %s already running, cannot run %s\n",
5625 mdname(mddev), bdevname(rdev0->bdev,b));
5626 mddev_unlock(mddev);
5628 printk(KERN_INFO "md: created %s\n", mdname(mddev));
5629 mddev->persistent = 1;
5630 rdev_for_each_list(rdev, tmp, &candidates) {
5631 list_del_init(&rdev->same_set);
5632 if (bind_rdev_to_array(rdev, mddev))
5635 autorun_array(mddev);
5636 mddev_unlock(mddev);
5638 /* on success, candidates will be empty, on error
5641 rdev_for_each_list(rdev, tmp, &candidates) {
5642 list_del_init(&rdev->same_set);
5647 printk(KERN_INFO "md: ... autorun DONE.\n");
5649 #endif /* !MODULE */
5651 static int get_version(void __user *arg)
5655 ver.major = MD_MAJOR_VERSION;
5656 ver.minor = MD_MINOR_VERSION;
5657 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5659 if (copy_to_user(arg, &ver, sizeof(ver)))
5665 static int get_array_info(struct mddev *mddev, void __user *arg)
5667 mdu_array_info_t info;
5668 int nr,working,insync,failed,spare;
5669 struct md_rdev *rdev;
5671 nr = working = insync = failed = spare = 0;
5673 rdev_for_each_rcu(rdev, mddev) {
5675 if (test_bit(Faulty, &rdev->flags))
5679 if (test_bit(In_sync, &rdev->flags))
5687 info.major_version = mddev->major_version;
5688 info.minor_version = mddev->minor_version;
5689 info.patch_version = MD_PATCHLEVEL_VERSION;
5690 info.ctime = mddev->ctime;
5691 info.level = mddev->level;
5692 info.size = mddev->dev_sectors / 2;
5693 if (info.size != mddev->dev_sectors / 2) /* overflow */
5696 info.raid_disks = mddev->raid_disks;
5697 info.md_minor = mddev->md_minor;
5698 info.not_persistent= !mddev->persistent;
5700 info.utime = mddev->utime;
5703 info.state = (1<<MD_SB_CLEAN);
5704 if (mddev->bitmap && mddev->bitmap_info.offset)
5705 info.state |= (1<<MD_SB_BITMAP_PRESENT);
5706 if (mddev_is_clustered(mddev))
5707 info.state |= (1<<MD_SB_CLUSTERED);
5708 info.active_disks = insync;
5709 info.working_disks = working;
5710 info.failed_disks = failed;
5711 info.spare_disks = spare;
5713 info.layout = mddev->layout;
5714 info.chunk_size = mddev->chunk_sectors << 9;
5716 if (copy_to_user(arg, &info, sizeof(info)))
5722 static int get_bitmap_file(struct mddev *mddev, void __user * arg)
5724 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5728 file = kmalloc(sizeof(*file), GFP_NOIO);
5733 spin_lock(&mddev->lock);
5734 /* bitmap disabled, zero the first byte and copy out */
5735 if (!mddev->bitmap_info.file)
5736 file->pathname[0] = '\0';
5737 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5738 file->pathname, sizeof(file->pathname))),
5742 memmove(file->pathname, ptr,
5743 sizeof(file->pathname)-(ptr-file->pathname));
5744 spin_unlock(&mddev->lock);
5747 copy_to_user(arg, file, sizeof(*file)))
5754 static int get_disk_info(struct mddev *mddev, void __user * arg)
5756 mdu_disk_info_t info;
5757 struct md_rdev *rdev;
5759 if (copy_from_user(&info, arg, sizeof(info)))
5763 rdev = md_find_rdev_nr_rcu(mddev, info.number);
5765 info.major = MAJOR(rdev->bdev->bd_dev);
5766 info.minor = MINOR(rdev->bdev->bd_dev);
5767 info.raid_disk = rdev->raid_disk;
5769 if (test_bit(Faulty, &rdev->flags))
5770 info.state |= (1<<MD_DISK_FAULTY);
5771 else if (test_bit(In_sync, &rdev->flags)) {
5772 info.state |= (1<<MD_DISK_ACTIVE);
5773 info.state |= (1<<MD_DISK_SYNC);
5775 if (test_bit(WriteMostly, &rdev->flags))
5776 info.state |= (1<<MD_DISK_WRITEMOSTLY);
5778 info.major = info.minor = 0;
5779 info.raid_disk = -1;
5780 info.state = (1<<MD_DISK_REMOVED);
5784 if (copy_to_user(arg, &info, sizeof(info)))
5790 static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
5792 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5793 struct md_rdev *rdev;
5794 dev_t dev = MKDEV(info->major,info->minor);
5796 if (mddev_is_clustered(mddev) &&
5797 !(info->state & ((1 << MD_DISK_CLUSTER_ADD) | (1 << MD_DISK_CANDIDATE)))) {
5798 pr_err("%s: Cannot add to clustered mddev.\n",
5803 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5806 if (!mddev->raid_disks) {
5808 /* expecting a device which has a superblock */
5809 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5812 "md: md_import_device returned %ld\n",
5814 return PTR_ERR(rdev);
5816 if (!list_empty(&mddev->disks)) {
5817 struct md_rdev *rdev0
5818 = list_entry(mddev->disks.next,
5819 struct md_rdev, same_set);
5820 err = super_types[mddev->major_version]
5821 .load_super(rdev, rdev0, mddev->minor_version);
5824 "md: %s has different UUID to %s\n",
5825 bdevname(rdev->bdev,b),
5826 bdevname(rdev0->bdev,b2));
5831 err = bind_rdev_to_array(rdev, mddev);
5838 * add_new_disk can be used once the array is assembled
5839 * to add "hot spares". They must already have a superblock
5844 if (!mddev->pers->hot_add_disk) {
5846 "%s: personality does not support diskops!\n",
5850 if (mddev->persistent)
5851 rdev = md_import_device(dev, mddev->major_version,
5852 mddev->minor_version);
5854 rdev = md_import_device(dev, -1, -1);
5857 "md: md_import_device returned %ld\n",
5859 return PTR_ERR(rdev);
5861 /* set saved_raid_disk if appropriate */
5862 if (!mddev->persistent) {
5863 if (info->state & (1<<MD_DISK_SYNC) &&
5864 info->raid_disk < mddev->raid_disks) {
5865 rdev->raid_disk = info->raid_disk;
5866 set_bit(In_sync, &rdev->flags);
5867 clear_bit(Bitmap_sync, &rdev->flags);
5869 rdev->raid_disk = -1;
5870 rdev->saved_raid_disk = rdev->raid_disk;
5872 super_types[mddev->major_version].
5873 validate_super(mddev, rdev);
5874 if ((info->state & (1<<MD_DISK_SYNC)) &&
5875 rdev->raid_disk != info->raid_disk) {
5876 /* This was a hot-add request, but events doesn't
5877 * match, so reject it.
5883 clear_bit(In_sync, &rdev->flags); /* just to be sure */
5884 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5885 set_bit(WriteMostly, &rdev->flags);
5887 clear_bit(WriteMostly, &rdev->flags);
5890 * check whether the device shows up in other nodes
5892 if (mddev_is_clustered(mddev)) {
5893 if (info->state & (1 << MD_DISK_CANDIDATE)) {
5894 /* Through --cluster-confirm */
5895 set_bit(Candidate, &rdev->flags);
5896 err = md_cluster_ops->new_disk_ack(mddev, true);
5901 } else if (info->state & (1 << MD_DISK_CLUSTER_ADD)) {
5902 /* --add initiated by this node */
5903 err = md_cluster_ops->add_new_disk_start(mddev, rdev);
5905 md_cluster_ops->add_new_disk_finish(mddev);
5912 rdev->raid_disk = -1;
5913 err = bind_rdev_to_array(rdev, mddev);
5917 err = add_bound_rdev(rdev);
5918 if (mddev_is_clustered(mddev) &&
5919 (info->state & (1 << MD_DISK_CLUSTER_ADD)))
5920 md_cluster_ops->add_new_disk_finish(mddev);
5924 /* otherwise, add_new_disk is only allowed
5925 * for major_version==0 superblocks
5927 if (mddev->major_version != 0) {
5928 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5933 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5935 rdev = md_import_device(dev, -1, 0);
5938 "md: error, md_import_device() returned %ld\n",
5940 return PTR_ERR(rdev);
5942 rdev->desc_nr = info->number;
5943 if (info->raid_disk < mddev->raid_disks)
5944 rdev->raid_disk = info->raid_disk;
5946 rdev->raid_disk = -1;
5948 if (rdev->raid_disk < mddev->raid_disks)
5949 if (info->state & (1<<MD_DISK_SYNC))
5950 set_bit(In_sync, &rdev->flags);
5952 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5953 set_bit(WriteMostly, &rdev->flags);
5955 if (!mddev->persistent) {
5956 printk(KERN_INFO "md: nonpersistent superblock ...\n");
5957 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5959 rdev->sb_start = calc_dev_sboffset(rdev);
5960 rdev->sectors = rdev->sb_start;
5962 err = bind_rdev_to_array(rdev, mddev);
5972 static int hot_remove_disk(struct mddev *mddev, dev_t dev)
5974 char b[BDEVNAME_SIZE];
5975 struct md_rdev *rdev;
5977 rdev = find_rdev(mddev, dev);
5981 if (mddev_is_clustered(mddev))
5982 md_cluster_ops->metadata_update_start(mddev);
5984 clear_bit(Blocked, &rdev->flags);
5985 remove_and_add_spares(mddev, rdev);
5987 if (rdev->raid_disk >= 0)
5990 if (mddev_is_clustered(mddev))
5991 md_cluster_ops->remove_disk(mddev, rdev);
5993 md_kick_rdev_from_array(rdev);
5994 md_update_sb(mddev, 1);
5995 md_new_event(mddev);
5997 if (mddev_is_clustered(mddev))
5998 md_cluster_ops->metadata_update_finish(mddev);
6002 if (mddev_is_clustered(mddev))
6003 md_cluster_ops->metadata_update_cancel(mddev);
6004 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
6005 bdevname(rdev->bdev,b), mdname(mddev));
6009 static int hot_add_disk(struct mddev *mddev, dev_t dev)
6011 char b[BDEVNAME_SIZE];
6013 struct md_rdev *rdev;
6018 if (mddev->major_version != 0) {
6019 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
6020 " version-0 superblocks.\n",
6024 if (!mddev->pers->hot_add_disk) {
6026 "%s: personality does not support diskops!\n",
6031 rdev = md_import_device(dev, -1, 0);
6034 "md: error, md_import_device() returned %ld\n",
6039 if (mddev->persistent)
6040 rdev->sb_start = calc_dev_sboffset(rdev);
6042 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
6044 rdev->sectors = rdev->sb_start;
6046 if (test_bit(Faulty, &rdev->flags)) {
6048 "md: can not hot-add faulty %s disk to %s!\n",
6049 bdevname(rdev->bdev,b), mdname(mddev));
6054 if (mddev_is_clustered(mddev))
6055 md_cluster_ops->metadata_update_start(mddev);
6056 clear_bit(In_sync, &rdev->flags);
6058 rdev->saved_raid_disk = -1;
6059 err = bind_rdev_to_array(rdev, mddev);
6061 goto abort_clustered;
6064 * The rest should better be atomic, we can have disk failures
6065 * noticed in interrupt contexts ...
6068 rdev->raid_disk = -1;
6070 md_update_sb(mddev, 1);
6072 if (mddev_is_clustered(mddev))
6073 md_cluster_ops->metadata_update_finish(mddev);
6075 * Kick recovery, maybe this spare has to be added to the
6076 * array immediately.
6078 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6079 md_wakeup_thread(mddev->thread);
6080 md_new_event(mddev);
6084 if (mddev_is_clustered(mddev))
6085 md_cluster_ops->metadata_update_cancel(mddev);
6091 static int set_bitmap_file(struct mddev *mddev, int fd)
6096 if (!mddev->pers->quiesce || !mddev->thread)
6098 if (mddev->recovery || mddev->sync_thread)
6100 /* we should be able to change the bitmap.. */
6104 struct inode *inode;
6107 if (mddev->bitmap || mddev->bitmap_info.file)
6108 return -EEXIST; /* cannot add when bitmap is present */
6112 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6117 inode = f->f_mapping->host;
6118 if (!S_ISREG(inode->i_mode)) {
6119 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6122 } else if (!(f->f_mode & FMODE_WRITE)) {
6123 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6126 } else if (atomic_read(&inode->i_writecount) != 1) {
6127 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6135 mddev->bitmap_info.file = f;
6136 mddev->bitmap_info.offset = 0; /* file overrides offset */
6137 } else if (mddev->bitmap == NULL)
6138 return -ENOENT; /* cannot remove what isn't there */
6141 mddev->pers->quiesce(mddev, 1);
6143 struct bitmap *bitmap;
6145 bitmap = bitmap_create(mddev, -1);
6146 if (!IS_ERR(bitmap)) {
6147 mddev->bitmap = bitmap;
6148 err = bitmap_load(mddev);
6150 err = PTR_ERR(bitmap);
6152 if (fd < 0 || err) {
6153 bitmap_destroy(mddev);
6154 fd = -1; /* make sure to put the file */
6156 mddev->pers->quiesce(mddev, 0);
6159 struct file *f = mddev->bitmap_info.file;
6161 spin_lock(&mddev->lock);
6162 mddev->bitmap_info.file = NULL;
6163 spin_unlock(&mddev->lock);
6172 * set_array_info is used two different ways
6173 * The original usage is when creating a new array.
6174 * In this usage, raid_disks is > 0 and it together with
6175 * level, size, not_persistent,layout,chunksize determine the
6176 * shape of the array.
6177 * This will always create an array with a type-0.90.0 superblock.
6178 * The newer usage is when assembling an array.
6179 * In this case raid_disks will be 0, and the major_version field is
6180 * use to determine which style super-blocks are to be found on the devices.
6181 * The minor and patch _version numbers are also kept incase the
6182 * super_block handler wishes to interpret them.
6184 static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
6187 if (info->raid_disks == 0) {
6188 /* just setting version number for superblock loading */
6189 if (info->major_version < 0 ||
6190 info->major_version >= ARRAY_SIZE(super_types) ||
6191 super_types[info->major_version].name == NULL) {
6192 /* maybe try to auto-load a module? */
6194 "md: superblock version %d not known\n",
6195 info->major_version);
6198 mddev->major_version = info->major_version;
6199 mddev->minor_version = info->minor_version;
6200 mddev->patch_version = info->patch_version;
6201 mddev->persistent = !info->not_persistent;
6202 /* ensure mddev_put doesn't delete this now that there
6203 * is some minimal configuration.
6205 mddev->ctime = get_seconds();
6208 mddev->major_version = MD_MAJOR_VERSION;
6209 mddev->minor_version = MD_MINOR_VERSION;
6210 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6211 mddev->ctime = get_seconds();
6213 mddev->level = info->level;
6214 mddev->clevel[0] = 0;
6215 mddev->dev_sectors = 2 * (sector_t)info->size;
6216 mddev->raid_disks = info->raid_disks;
6217 /* don't set md_minor, it is determined by which /dev/md* was
6220 if (info->state & (1<<MD_SB_CLEAN))
6221 mddev->recovery_cp = MaxSector;
6223 mddev->recovery_cp = 0;
6224 mddev->persistent = ! info->not_persistent;
6225 mddev->external = 0;
6227 mddev->layout = info->layout;
6228 mddev->chunk_sectors = info->chunk_size >> 9;
6230 mddev->max_disks = MD_SB_DISKS;
6232 if (mddev->persistent)
6234 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6236 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
6237 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
6238 mddev->bitmap_info.offset = 0;
6240 mddev->reshape_position = MaxSector;
6243 * Generate a 128 bit UUID
6245 get_random_bytes(mddev->uuid, 16);
6247 mddev->new_level = mddev->level;
6248 mddev->new_chunk_sectors = mddev->chunk_sectors;
6249 mddev->new_layout = mddev->layout;
6250 mddev->delta_disks = 0;
6251 mddev->reshape_backwards = 0;
6256 void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
6258 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6260 if (mddev->external_size)
6263 mddev->array_sectors = array_sectors;
6265 EXPORT_SYMBOL(md_set_array_sectors);
6267 static int update_size(struct mddev *mddev, sector_t num_sectors)
6269 struct md_rdev *rdev;
6271 int fit = (num_sectors == 0);
6273 if (mddev->pers->resize == NULL)
6275 /* The "num_sectors" is the number of sectors of each device that
6276 * is used. This can only make sense for arrays with redundancy.
6277 * linear and raid0 always use whatever space is available. We can only
6278 * consider changing this number if no resync or reconstruction is
6279 * happening, and if the new size is acceptable. It must fit before the
6280 * sb_start or, if that is <data_offset, it must fit before the size
6281 * of each device. If num_sectors is zero, we find the largest size
6284 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6290 rdev_for_each(rdev, mddev) {
6291 sector_t avail = rdev->sectors;
6293 if (fit && (num_sectors == 0 || num_sectors > avail))
6294 num_sectors = avail;
6295 if (avail < num_sectors)
6298 rv = mddev->pers->resize(mddev, num_sectors);
6300 revalidate_disk(mddev->gendisk);
6304 static int update_raid_disks(struct mddev *mddev, int raid_disks)
6307 struct md_rdev *rdev;
6308 /* change the number of raid disks */
6309 if (mddev->pers->check_reshape == NULL)
6313 if (raid_disks <= 0 ||
6314 (mddev->max_disks && raid_disks >= mddev->max_disks))
6316 if (mddev->sync_thread ||
6317 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6318 mddev->reshape_position != MaxSector)
6321 rdev_for_each(rdev, mddev) {
6322 if (mddev->raid_disks < raid_disks &&
6323 rdev->data_offset < rdev->new_data_offset)
6325 if (mddev->raid_disks > raid_disks &&
6326 rdev->data_offset > rdev->new_data_offset)
6330 mddev->delta_disks = raid_disks - mddev->raid_disks;
6331 if (mddev->delta_disks < 0)
6332 mddev->reshape_backwards = 1;
6333 else if (mddev->delta_disks > 0)
6334 mddev->reshape_backwards = 0;
6336 rv = mddev->pers->check_reshape(mddev);
6338 mddev->delta_disks = 0;
6339 mddev->reshape_backwards = 0;
6345 * update_array_info is used to change the configuration of an
6347 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6348 * fields in the info are checked against the array.
6349 * Any differences that cannot be handled will cause an error.
6350 * Normally, only one change can be managed at a time.
6352 static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
6358 /* calculate expected state,ignoring low bits */
6359 if (mddev->bitmap && mddev->bitmap_info.offset)
6360 state |= (1 << MD_SB_BITMAP_PRESENT);
6362 if (mddev->major_version != info->major_version ||
6363 mddev->minor_version != info->minor_version ||
6364 /* mddev->patch_version != info->patch_version || */
6365 mddev->ctime != info->ctime ||
6366 mddev->level != info->level ||
6367 /* mddev->layout != info->layout || */
6368 !mddev->persistent != info->not_persistent||
6369 mddev->chunk_sectors != info->chunk_size >> 9 ||
6370 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6371 ((state^info->state) & 0xfffffe00)
6374 /* Check there is only one change */
6375 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6377 if (mddev->raid_disks != info->raid_disks)
6379 if (mddev->layout != info->layout)
6381 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6388 if (mddev->layout != info->layout) {
6390 * we don't need to do anything at the md level, the
6391 * personality will take care of it all.
6393 if (mddev->pers->check_reshape == NULL)
6396 mddev->new_layout = info->layout;
6397 rv = mddev->pers->check_reshape(mddev);
6399 mddev->new_layout = mddev->layout;
6403 if (mddev_is_clustered(mddev))
6404 md_cluster_ops->metadata_update_start(mddev);
6405 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6406 rv = update_size(mddev, (sector_t)info->size * 2);
6408 if (mddev->raid_disks != info->raid_disks)
6409 rv = update_raid_disks(mddev, info->raid_disks);
6411 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6412 if (mddev->pers->quiesce == NULL || mddev->thread == NULL) {
6416 if (mddev->recovery || mddev->sync_thread) {
6420 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6421 struct bitmap *bitmap;
6422 /* add the bitmap */
6423 if (mddev->bitmap) {
6427 if (mddev->bitmap_info.default_offset == 0) {
6431 mddev->bitmap_info.offset =
6432 mddev->bitmap_info.default_offset;
6433 mddev->bitmap_info.space =
6434 mddev->bitmap_info.default_space;
6435 mddev->pers->quiesce(mddev, 1);
6436 bitmap = bitmap_create(mddev, -1);
6437 if (!IS_ERR(bitmap)) {
6438 mddev->bitmap = bitmap;
6439 rv = bitmap_load(mddev);
6441 rv = PTR_ERR(bitmap);
6443 bitmap_destroy(mddev);
6444 mddev->pers->quiesce(mddev, 0);
6446 /* remove the bitmap */
6447 if (!mddev->bitmap) {
6451 if (mddev->bitmap->storage.file) {
6455 mddev->pers->quiesce(mddev, 1);
6456 bitmap_destroy(mddev);
6457 mddev->pers->quiesce(mddev, 0);
6458 mddev->bitmap_info.offset = 0;
6461 md_update_sb(mddev, 1);
6462 if (mddev_is_clustered(mddev))
6463 md_cluster_ops->metadata_update_finish(mddev);
6466 if (mddev_is_clustered(mddev))
6467 md_cluster_ops->metadata_update_cancel(mddev);
6471 static int set_disk_faulty(struct mddev *mddev, dev_t dev)
6473 struct md_rdev *rdev;
6476 if (mddev->pers == NULL)
6480 rdev = find_rdev_rcu(mddev, dev);
6484 md_error(mddev, rdev);
6485 if (!test_bit(Faulty, &rdev->flags))
6493 * We have a problem here : there is no easy way to give a CHS
6494 * virtual geometry. We currently pretend that we have a 2 heads
6495 * 4 sectors (with a BIG number of cylinders...). This drives
6496 * dosfs just mad... ;-)
6498 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6500 struct mddev *mddev = bdev->bd_disk->private_data;
6504 geo->cylinders = mddev->array_sectors / 8;
6508 static inline bool md_ioctl_valid(unsigned int cmd)
6513 case GET_ARRAY_INFO:
6514 case GET_BITMAP_FILE:
6517 case HOT_REMOVE_DISK:
6520 case RESTART_ARRAY_RW:
6522 case SET_ARRAY_INFO:
6523 case SET_BITMAP_FILE:
6524 case SET_DISK_FAULTY:
6527 case CLUSTERED_DISK_NACK:
6534 static int md_ioctl(struct block_device *bdev, fmode_t mode,
6535 unsigned int cmd, unsigned long arg)
6538 void __user *argp = (void __user *)arg;
6539 struct mddev *mddev = NULL;
6542 if (!md_ioctl_valid(cmd))
6547 case GET_ARRAY_INFO:
6551 if (!capable(CAP_SYS_ADMIN))
6556 * Commands dealing with the RAID driver but not any
6561 err = get_version(argp);
6567 autostart_arrays(arg);
6574 * Commands creating/starting a new array:
6577 mddev = bdev->bd_disk->private_data;
6584 /* Some actions do not requires the mutex */
6586 case GET_ARRAY_INFO:
6587 if (!mddev->raid_disks && !mddev->external)
6590 err = get_array_info(mddev, argp);
6594 if (!mddev->raid_disks && !mddev->external)
6597 err = get_disk_info(mddev, argp);
6600 case SET_DISK_FAULTY:
6601 err = set_disk_faulty(mddev, new_decode_dev(arg));
6604 case GET_BITMAP_FILE:
6605 err = get_bitmap_file(mddev, argp);
6610 if (cmd == ADD_NEW_DISK)
6611 /* need to ensure md_delayed_delete() has completed */
6612 flush_workqueue(md_misc_wq);
6614 if (cmd == HOT_REMOVE_DISK)
6615 /* need to ensure recovery thread has run */
6616 wait_event_interruptible_timeout(mddev->sb_wait,
6617 !test_bit(MD_RECOVERY_NEEDED,
6619 msecs_to_jiffies(5000));
6620 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6621 /* Need to flush page cache, and ensure no-one else opens
6624 mutex_lock(&mddev->open_mutex);
6625 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
6626 mutex_unlock(&mddev->open_mutex);
6630 set_bit(MD_STILL_CLOSED, &mddev->flags);
6631 mutex_unlock(&mddev->open_mutex);
6632 sync_blockdev(bdev);
6634 err = mddev_lock(mddev);
6637 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6642 if (cmd == SET_ARRAY_INFO) {
6643 mdu_array_info_t info;
6645 memset(&info, 0, sizeof(info));
6646 else if (copy_from_user(&info, argp, sizeof(info))) {
6651 err = update_array_info(mddev, &info);
6653 printk(KERN_WARNING "md: couldn't update"
6654 " array info. %d\n", err);
6659 if (!list_empty(&mddev->disks)) {
6661 "md: array %s already has disks!\n",
6666 if (mddev->raid_disks) {
6668 "md: array %s already initialised!\n",
6673 err = set_array_info(mddev, &info);
6675 printk(KERN_WARNING "md: couldn't set"
6676 " array info. %d\n", err);
6683 * Commands querying/configuring an existing array:
6685 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6686 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6687 if ((!mddev->raid_disks && !mddev->external)
6688 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6689 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6690 && cmd != GET_BITMAP_FILE) {
6696 * Commands even a read-only array can execute:
6699 case RESTART_ARRAY_RW:
6700 err = restart_array(mddev);
6704 err = do_md_stop(mddev, 0, bdev);
6708 err = md_set_readonly(mddev, bdev);
6711 case HOT_REMOVE_DISK:
6712 err = hot_remove_disk(mddev, new_decode_dev(arg));
6716 /* We can support ADD_NEW_DISK on read-only arrays
6717 * on if we are re-adding a preexisting device.
6718 * So require mddev->pers and MD_DISK_SYNC.
6721 mdu_disk_info_t info;
6722 if (copy_from_user(&info, argp, sizeof(info)))
6724 else if (!(info.state & (1<<MD_DISK_SYNC)))
6725 /* Need to clear read-only for this */
6728 err = add_new_disk(mddev, &info);
6734 if (get_user(ro, (int __user *)(arg))) {
6740 /* if the bdev is going readonly the value of mddev->ro
6741 * does not matter, no writes are coming
6746 /* are we are already prepared for writes? */
6750 /* transitioning to readauto need only happen for
6751 * arrays that call md_write_start
6754 err = restart_array(mddev);
6757 set_disk_ro(mddev->gendisk, 0);
6764 * The remaining ioctls are changing the state of the
6765 * superblock, so we do not allow them on read-only arrays.
6767 if (mddev->ro && mddev->pers) {
6768 if (mddev->ro == 2) {
6770 sysfs_notify_dirent_safe(mddev->sysfs_state);
6771 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6772 /* mddev_unlock will wake thread */
6773 /* If a device failed while we were read-only, we
6774 * need to make sure the metadata is updated now.
6776 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6777 mddev_unlock(mddev);
6778 wait_event(mddev->sb_wait,
6779 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6780 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6781 mddev_lock_nointr(mddev);
6792 mdu_disk_info_t info;
6793 if (copy_from_user(&info, argp, sizeof(info)))
6796 err = add_new_disk(mddev, &info);
6800 case CLUSTERED_DISK_NACK:
6801 if (mddev_is_clustered(mddev))
6802 md_cluster_ops->new_disk_ack(mddev, false);
6808 err = hot_add_disk(mddev, new_decode_dev(arg));
6812 err = do_md_run(mddev);
6815 case SET_BITMAP_FILE:
6816 err = set_bitmap_file(mddev, (int)arg);
6825 if (mddev->hold_active == UNTIL_IOCTL &&
6827 mddev->hold_active = 0;
6828 mddev_unlock(mddev);
6832 #ifdef CONFIG_COMPAT
6833 static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6834 unsigned int cmd, unsigned long arg)
6837 case HOT_REMOVE_DISK:
6839 case SET_DISK_FAULTY:
6840 case SET_BITMAP_FILE:
6841 /* These take in integer arg, do not convert */
6844 arg = (unsigned long)compat_ptr(arg);
6848 return md_ioctl(bdev, mode, cmd, arg);
6850 #endif /* CONFIG_COMPAT */
6852 static int md_open(struct block_device *bdev, fmode_t mode)
6855 * Succeed if we can lock the mddev, which confirms that
6856 * it isn't being stopped right now.
6858 struct mddev *mddev = mddev_find(bdev->bd_dev);
6864 if (mddev->gendisk != bdev->bd_disk) {
6865 /* we are racing with mddev_put which is discarding this
6869 /* Wait until bdev->bd_disk is definitely gone */
6870 flush_workqueue(md_misc_wq);
6871 /* Then retry the open from the top */
6872 return -ERESTARTSYS;
6874 BUG_ON(mddev != bdev->bd_disk->private_data);
6876 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6880 atomic_inc(&mddev->openers);
6881 clear_bit(MD_STILL_CLOSED, &mddev->flags);
6882 mutex_unlock(&mddev->open_mutex);
6884 check_disk_change(bdev);
6889 static void md_release(struct gendisk *disk, fmode_t mode)
6891 struct mddev *mddev = disk->private_data;
6894 atomic_dec(&mddev->openers);
6898 static int md_media_changed(struct gendisk *disk)
6900 struct mddev *mddev = disk->private_data;
6902 return mddev->changed;
6905 static int md_revalidate(struct gendisk *disk)
6907 struct mddev *mddev = disk->private_data;
6912 static const struct block_device_operations md_fops =
6914 .owner = THIS_MODULE,
6916 .release = md_release,
6918 #ifdef CONFIG_COMPAT
6919 .compat_ioctl = md_compat_ioctl,
6921 .getgeo = md_getgeo,
6922 .media_changed = md_media_changed,
6923 .revalidate_disk= md_revalidate,
6926 static int md_thread(void *arg)
6928 struct md_thread *thread = arg;
6931 * md_thread is a 'system-thread', it's priority should be very
6932 * high. We avoid resource deadlocks individually in each
6933 * raid personality. (RAID5 does preallocation) We also use RR and
6934 * the very same RT priority as kswapd, thus we will never get
6935 * into a priority inversion deadlock.
6937 * we definitely have to have equal or higher priority than
6938 * bdflush, otherwise bdflush will deadlock if there are too
6939 * many dirty RAID5 blocks.
6942 allow_signal(SIGKILL);
6943 while (!kthread_should_stop()) {
6945 /* We need to wait INTERRUPTIBLE so that
6946 * we don't add to the load-average.
6947 * That means we need to be sure no signals are
6950 if (signal_pending(current))
6951 flush_signals(current);
6953 wait_event_interruptible_timeout
6955 test_bit(THREAD_WAKEUP, &thread->flags)
6956 || kthread_should_stop(),
6959 clear_bit(THREAD_WAKEUP, &thread->flags);
6960 if (!kthread_should_stop())
6961 thread->run(thread);
6967 void md_wakeup_thread(struct md_thread *thread)
6970 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
6971 set_bit(THREAD_WAKEUP, &thread->flags);
6972 wake_up(&thread->wqueue);
6975 EXPORT_SYMBOL(md_wakeup_thread);
6977 struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6978 struct mddev *mddev, const char *name)
6980 struct md_thread *thread;
6982 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
6986 init_waitqueue_head(&thread->wqueue);
6989 thread->mddev = mddev;
6990 thread->timeout = MAX_SCHEDULE_TIMEOUT;
6991 thread->tsk = kthread_run(md_thread, thread,
6993 mdname(thread->mddev),
6995 if (IS_ERR(thread->tsk)) {
7001 EXPORT_SYMBOL(md_register_thread);
7003 void md_unregister_thread(struct md_thread **threadp)
7005 struct md_thread *thread = *threadp;
7008 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
7009 /* Locking ensures that mddev_unlock does not wake_up a
7010 * non-existent thread
7012 spin_lock(&pers_lock);
7014 spin_unlock(&pers_lock);
7016 kthread_stop(thread->tsk);
7019 EXPORT_SYMBOL(md_unregister_thread);
7021 void md_error(struct mddev *mddev, struct md_rdev *rdev)
7023 if (!rdev || test_bit(Faulty, &rdev->flags))
7026 if (!mddev->pers || !mddev->pers->error_handler)
7028 mddev->pers->error_handler(mddev,rdev);
7029 if (mddev->degraded)
7030 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7031 sysfs_notify_dirent_safe(rdev->sysfs_state);
7032 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7033 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7034 md_wakeup_thread(mddev->thread);
7035 if (mddev->event_work.func)
7036 queue_work(md_misc_wq, &mddev->event_work);
7037 md_new_event_inintr(mddev);
7039 EXPORT_SYMBOL(md_error);
7041 /* seq_file implementation /proc/mdstat */
7043 static void status_unused(struct seq_file *seq)
7046 struct md_rdev *rdev;
7048 seq_printf(seq, "unused devices: ");
7050 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
7051 char b[BDEVNAME_SIZE];
7053 seq_printf(seq, "%s ",
7054 bdevname(rdev->bdev,b));
7057 seq_printf(seq, "<none>");
7059 seq_printf(seq, "\n");
7062 static void status_resync(struct seq_file *seq, struct mddev *mddev)
7064 sector_t max_sectors, resync, res;
7065 unsigned long dt, db;
7068 unsigned int per_milli;
7070 if (mddev->curr_resync <= 3)
7073 resync = mddev->curr_resync
7074 - atomic_read(&mddev->recovery_active);
7076 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
7077 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7078 max_sectors = mddev->resync_max_sectors;
7080 max_sectors = mddev->dev_sectors;
7082 WARN_ON(max_sectors == 0);
7083 /* Pick 'scale' such that (resync>>scale)*1000 will fit
7084 * in a sector_t, and (max_sectors>>scale) will fit in a
7085 * u32, as those are the requirements for sector_div.
7086 * Thus 'scale' must be at least 10
7089 if (sizeof(sector_t) > sizeof(unsigned long)) {
7090 while ( max_sectors/2 > (1ULL<<(scale+32)))
7093 res = (resync>>scale)*1000;
7094 sector_div(res, (u32)((max_sectors>>scale)+1));
7098 int i, x = per_milli/50, y = 20-x;
7099 seq_printf(seq, "[");
7100 for (i = 0; i < x; i++)
7101 seq_printf(seq, "=");
7102 seq_printf(seq, ">");
7103 for (i = 0; i < y; i++)
7104 seq_printf(seq, ".");
7105 seq_printf(seq, "] ");
7107 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
7108 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
7110 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
7112 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
7113 "resync" : "recovery"))),
7114 per_milli/10, per_milli % 10,
7115 (unsigned long long) resync/2,
7116 (unsigned long long) max_sectors/2);
7119 * dt: time from mark until now
7120 * db: blocks written from mark until now
7121 * rt: remaining time
7123 * rt is a sector_t, so could be 32bit or 64bit.
7124 * So we divide before multiply in case it is 32bit and close
7126 * We scale the divisor (db) by 32 to avoid losing precision
7127 * near the end of resync when the number of remaining sectors
7129 * We then divide rt by 32 after multiplying by db to compensate.
7130 * The '+1' avoids division by zero if db is very small.
7132 dt = ((jiffies - mddev->resync_mark) / HZ);
7134 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
7135 - mddev->resync_mark_cnt;
7137 rt = max_sectors - resync; /* number of remaining sectors */
7138 sector_div(rt, db/32+1);
7142 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7143 ((unsigned long)rt % 60)/6);
7145 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
7148 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7150 struct list_head *tmp;
7152 struct mddev *mddev;
7160 spin_lock(&all_mddevs_lock);
7161 list_for_each(tmp,&all_mddevs)
7163 mddev = list_entry(tmp, struct mddev, all_mddevs);
7165 spin_unlock(&all_mddevs_lock);
7168 spin_unlock(&all_mddevs_lock);
7170 return (void*)2;/* tail */
7174 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7176 struct list_head *tmp;
7177 struct mddev *next_mddev, *mddev = v;
7183 spin_lock(&all_mddevs_lock);
7185 tmp = all_mddevs.next;
7187 tmp = mddev->all_mddevs.next;
7188 if (tmp != &all_mddevs)
7189 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
7191 next_mddev = (void*)2;
7194 spin_unlock(&all_mddevs_lock);
7202 static void md_seq_stop(struct seq_file *seq, void *v)
7204 struct mddev *mddev = v;
7206 if (mddev && v != (void*)1 && v != (void*)2)
7210 static int md_seq_show(struct seq_file *seq, void *v)
7212 struct mddev *mddev = v;
7214 struct md_rdev *rdev;
7216 if (v == (void*)1) {
7217 struct md_personality *pers;
7218 seq_printf(seq, "Personalities : ");
7219 spin_lock(&pers_lock);
7220 list_for_each_entry(pers, &pers_list, list)
7221 seq_printf(seq, "[%s] ", pers->name);
7223 spin_unlock(&pers_lock);
7224 seq_printf(seq, "\n");
7225 seq->poll_event = atomic_read(&md_event_count);
7228 if (v == (void*)2) {
7233 spin_lock(&mddev->lock);
7234 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7235 seq_printf(seq, "%s : %sactive", mdname(mddev),
7236 mddev->pers ? "" : "in");
7239 seq_printf(seq, " (read-only)");
7241 seq_printf(seq, " (auto-read-only)");
7242 seq_printf(seq, " %s", mddev->pers->name);
7247 rdev_for_each_rcu(rdev, mddev) {
7248 char b[BDEVNAME_SIZE];
7249 seq_printf(seq, " %s[%d]",
7250 bdevname(rdev->bdev,b), rdev->desc_nr);
7251 if (test_bit(WriteMostly, &rdev->flags))
7252 seq_printf(seq, "(W)");
7253 if (test_bit(Faulty, &rdev->flags)) {
7254 seq_printf(seq, "(F)");
7257 if (rdev->raid_disk < 0)
7258 seq_printf(seq, "(S)"); /* spare */
7259 if (test_bit(Replacement, &rdev->flags))
7260 seq_printf(seq, "(R)");
7261 sectors += rdev->sectors;
7265 if (!list_empty(&mddev->disks)) {
7267 seq_printf(seq, "\n %llu blocks",
7268 (unsigned long long)
7269 mddev->array_sectors / 2);
7271 seq_printf(seq, "\n %llu blocks",
7272 (unsigned long long)sectors / 2);
7274 if (mddev->persistent) {
7275 if (mddev->major_version != 0 ||
7276 mddev->minor_version != 90) {
7277 seq_printf(seq," super %d.%d",
7278 mddev->major_version,
7279 mddev->minor_version);
7281 } else if (mddev->external)
7282 seq_printf(seq, " super external:%s",
7283 mddev->metadata_type);
7285 seq_printf(seq, " super non-persistent");
7288 mddev->pers->status(seq, mddev);
7289 seq_printf(seq, "\n ");
7290 if (mddev->pers->sync_request) {
7291 if (mddev->curr_resync > 2) {
7292 status_resync(seq, mddev);
7293 seq_printf(seq, "\n ");
7294 } else if (mddev->curr_resync >= 1)
7295 seq_printf(seq, "\tresync=DELAYED\n ");
7296 else if (mddev->recovery_cp < MaxSector)
7297 seq_printf(seq, "\tresync=PENDING\n ");
7300 seq_printf(seq, "\n ");
7302 bitmap_status(seq, mddev->bitmap);
7304 seq_printf(seq, "\n");
7306 spin_unlock(&mddev->lock);
7311 static const struct seq_operations md_seq_ops = {
7312 .start = md_seq_start,
7313 .next = md_seq_next,
7314 .stop = md_seq_stop,
7315 .show = md_seq_show,
7318 static int md_seq_open(struct inode *inode, struct file *file)
7320 struct seq_file *seq;
7323 error = seq_open(file, &md_seq_ops);
7327 seq = file->private_data;
7328 seq->poll_event = atomic_read(&md_event_count);
7332 static int md_unloading;
7333 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7335 struct seq_file *seq = filp->private_data;
7339 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
7340 poll_wait(filp, &md_event_waiters, wait);
7342 /* always allow read */
7343 mask = POLLIN | POLLRDNORM;
7345 if (seq->poll_event != atomic_read(&md_event_count))
7346 mask |= POLLERR | POLLPRI;
7350 static const struct file_operations md_seq_fops = {
7351 .owner = THIS_MODULE,
7352 .open = md_seq_open,
7354 .llseek = seq_lseek,
7355 .release = seq_release_private,
7356 .poll = mdstat_poll,
7359 int register_md_personality(struct md_personality *p)
7361 printk(KERN_INFO "md: %s personality registered for level %d\n",
7363 spin_lock(&pers_lock);
7364 list_add_tail(&p->list, &pers_list);
7365 spin_unlock(&pers_lock);
7368 EXPORT_SYMBOL(register_md_personality);
7370 int unregister_md_personality(struct md_personality *p)
7372 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
7373 spin_lock(&pers_lock);
7374 list_del_init(&p->list);
7375 spin_unlock(&pers_lock);
7378 EXPORT_SYMBOL(unregister_md_personality);
7380 int register_md_cluster_operations(struct md_cluster_operations *ops, struct module *module)
7382 if (md_cluster_ops != NULL)
7384 spin_lock(&pers_lock);
7385 md_cluster_ops = ops;
7386 md_cluster_mod = module;
7387 spin_unlock(&pers_lock);
7390 EXPORT_SYMBOL(register_md_cluster_operations);
7392 int unregister_md_cluster_operations(void)
7394 spin_lock(&pers_lock);
7395 md_cluster_ops = NULL;
7396 spin_unlock(&pers_lock);
7399 EXPORT_SYMBOL(unregister_md_cluster_operations);
7401 int md_setup_cluster(struct mddev *mddev, int nodes)
7405 err = request_module("md-cluster");
7407 pr_err("md-cluster module not found.\n");
7411 spin_lock(&pers_lock);
7412 if (!md_cluster_ops || !try_module_get(md_cluster_mod)) {
7413 spin_unlock(&pers_lock);
7416 spin_unlock(&pers_lock);
7418 return md_cluster_ops->join(mddev, nodes);
7421 void md_cluster_stop(struct mddev *mddev)
7423 if (!md_cluster_ops)
7425 md_cluster_ops->leave(mddev);
7426 module_put(md_cluster_mod);
7429 static int is_mddev_idle(struct mddev *mddev, int init)
7431 struct md_rdev *rdev;
7437 rdev_for_each_rcu(rdev, mddev) {
7438 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
7439 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7440 (int)part_stat_read(&disk->part0, sectors[1]) -
7441 atomic_read(&disk->sync_io);
7442 /* sync IO will cause sync_io to increase before the disk_stats
7443 * as sync_io is counted when a request starts, and
7444 * disk_stats is counted when it completes.
7445 * So resync activity will cause curr_events to be smaller than
7446 * when there was no such activity.
7447 * non-sync IO will cause disk_stat to increase without
7448 * increasing sync_io so curr_events will (eventually)
7449 * be larger than it was before. Once it becomes
7450 * substantially larger, the test below will cause
7451 * the array to appear non-idle, and resync will slow
7453 * If there is a lot of outstanding resync activity when
7454 * we set last_event to curr_events, then all that activity
7455 * completing might cause the array to appear non-idle
7456 * and resync will be slowed down even though there might
7457 * not have been non-resync activity. This will only
7458 * happen once though. 'last_events' will soon reflect
7459 * the state where there is little or no outstanding
7460 * resync requests, and further resync activity will
7461 * always make curr_events less than last_events.
7464 if (init || curr_events - rdev->last_events > 64) {
7465 rdev->last_events = curr_events;
7473 void md_done_sync(struct mddev *mddev, int blocks, int ok)
7475 /* another "blocks" (512byte) blocks have been synced */
7476 atomic_sub(blocks, &mddev->recovery_active);
7477 wake_up(&mddev->recovery_wait);
7479 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7480 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
7481 md_wakeup_thread(mddev->thread);
7482 // stop recovery, signal do_sync ....
7485 EXPORT_SYMBOL(md_done_sync);
7487 /* md_write_start(mddev, bi)
7488 * If we need to update some array metadata (e.g. 'active' flag
7489 * in superblock) before writing, schedule a superblock update
7490 * and wait for it to complete.
7492 void md_write_start(struct mddev *mddev, struct bio *bi)
7495 if (bio_data_dir(bi) != WRITE)
7498 BUG_ON(mddev->ro == 1);
7499 if (mddev->ro == 2) {
7500 /* need to switch to read/write */
7502 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7503 md_wakeup_thread(mddev->thread);
7504 md_wakeup_thread(mddev->sync_thread);
7507 atomic_inc(&mddev->writes_pending);
7508 if (mddev->safemode == 1)
7509 mddev->safemode = 0;
7510 if (mddev->in_sync) {
7511 spin_lock(&mddev->lock);
7512 if (mddev->in_sync) {
7514 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7515 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7516 md_wakeup_thread(mddev->thread);
7519 spin_unlock(&mddev->lock);
7522 sysfs_notify_dirent_safe(mddev->sysfs_state);
7523 wait_event(mddev->sb_wait,
7524 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
7526 EXPORT_SYMBOL(md_write_start);
7528 void md_write_end(struct mddev *mddev)
7530 if (atomic_dec_and_test(&mddev->writes_pending)) {
7531 if (mddev->safemode == 2)
7532 md_wakeup_thread(mddev->thread);
7533 else if (mddev->safemode_delay)
7534 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7537 EXPORT_SYMBOL(md_write_end);
7539 /* md_allow_write(mddev)
7540 * Calling this ensures that the array is marked 'active' so that writes
7541 * may proceed without blocking. It is important to call this before
7542 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7543 * Must be called with mddev_lock held.
7545 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7546 * is dropped, so return -EAGAIN after notifying userspace.
7548 int md_allow_write(struct mddev *mddev)
7554 if (!mddev->pers->sync_request)
7557 spin_lock(&mddev->lock);
7558 if (mddev->in_sync) {
7560 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7561 set_bit(MD_CHANGE_PENDING, &mddev->flags);
7562 if (mddev->safemode_delay &&
7563 mddev->safemode == 0)
7564 mddev->safemode = 1;
7565 spin_unlock(&mddev->lock);
7566 if (mddev_is_clustered(mddev))
7567 md_cluster_ops->metadata_update_start(mddev);
7568 md_update_sb(mddev, 0);
7569 if (mddev_is_clustered(mddev))
7570 md_cluster_ops->metadata_update_finish(mddev);
7571 sysfs_notify_dirent_safe(mddev->sysfs_state);
7573 spin_unlock(&mddev->lock);
7575 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
7580 EXPORT_SYMBOL_GPL(md_allow_write);
7582 #define SYNC_MARKS 10
7583 #define SYNC_MARK_STEP (3*HZ)
7584 #define UPDATE_FREQUENCY (5*60*HZ)
7585 void md_do_sync(struct md_thread *thread)
7587 struct mddev *mddev = thread->mddev;
7588 struct mddev *mddev2;
7589 unsigned int currspeed = 0,
7591 sector_t max_sectors,j, io_sectors, recovery_done;
7592 unsigned long mark[SYNC_MARKS];
7593 unsigned long update_time;
7594 sector_t mark_cnt[SYNC_MARKS];
7596 struct list_head *tmp;
7597 sector_t last_check;
7599 struct md_rdev *rdev;
7600 char *desc, *action = NULL;
7601 struct blk_plug plug;
7603 /* just incase thread restarts... */
7604 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7606 if (mddev->ro) {/* never try to sync a read-only array */
7607 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7611 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7612 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
7613 desc = "data-check";
7615 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7616 desc = "requested-resync";
7620 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7625 mddev->last_sync_action = action ?: desc;
7627 /* we overload curr_resync somewhat here.
7628 * 0 == not engaged in resync at all
7629 * 2 == checking that there is no conflict with another sync
7630 * 1 == like 2, but have yielded to allow conflicting resync to
7632 * other == active in resync - this many blocks
7634 * Before starting a resync we must have set curr_resync to
7635 * 2, and then checked that every "conflicting" array has curr_resync
7636 * less than ours. When we find one that is the same or higher
7637 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7638 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7639 * This will mean we have to start checking from the beginning again.
7644 mddev->curr_resync = 2;
7647 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7649 for_each_mddev(mddev2, tmp) {
7650 if (mddev2 == mddev)
7652 if (!mddev->parallel_resync
7653 && mddev2->curr_resync
7654 && match_mddev_units(mddev, mddev2)) {
7656 if (mddev < mddev2 && mddev->curr_resync == 2) {
7657 /* arbitrarily yield */
7658 mddev->curr_resync = 1;
7659 wake_up(&resync_wait);
7661 if (mddev > mddev2 && mddev->curr_resync == 1)
7662 /* no need to wait here, we can wait the next
7663 * time 'round when curr_resync == 2
7666 /* We need to wait 'interruptible' so as not to
7667 * contribute to the load average, and not to
7668 * be caught by 'softlockup'
7670 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7671 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7672 mddev2->curr_resync >= mddev->curr_resync) {
7673 printk(KERN_INFO "md: delaying %s of %s"
7674 " until %s has finished (they"
7675 " share one or more physical units)\n",
7676 desc, mdname(mddev), mdname(mddev2));
7678 if (signal_pending(current))
7679 flush_signals(current);
7681 finish_wait(&resync_wait, &wq);
7684 finish_wait(&resync_wait, &wq);
7687 } while (mddev->curr_resync < 2);
7690 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7691 /* resync follows the size requested by the personality,
7692 * which defaults to physical size, but can be virtual size
7694 max_sectors = mddev->resync_max_sectors;
7695 atomic64_set(&mddev->resync_mismatches, 0);
7696 /* we don't use the checkpoint if there's a bitmap */
7697 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7698 j = mddev->resync_min;
7699 else if (!mddev->bitmap)
7700 j = mddev->recovery_cp;
7702 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7703 max_sectors = mddev->resync_max_sectors;
7705 /* recovery follows the physical size of devices */
7706 max_sectors = mddev->dev_sectors;
7709 rdev_for_each_rcu(rdev, mddev)
7710 if (rdev->raid_disk >= 0 &&
7711 !test_bit(Faulty, &rdev->flags) &&
7712 !test_bit(In_sync, &rdev->flags) &&
7713 rdev->recovery_offset < j)
7714 j = rdev->recovery_offset;
7717 /* If there is a bitmap, we need to make sure all
7718 * writes that started before we added a spare
7719 * complete before we start doing a recovery.
7720 * Otherwise the write might complete and (via
7721 * bitmap_endwrite) set a bit in the bitmap after the
7722 * recovery has checked that bit and skipped that
7725 if (mddev->bitmap) {
7726 mddev->pers->quiesce(mddev, 1);
7727 mddev->pers->quiesce(mddev, 0);
7731 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7732 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7733 " %d KB/sec/disk.\n", speed_min(mddev));
7734 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7735 "(but not more than %d KB/sec) for %s.\n",
7736 speed_max(mddev), desc);
7738 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7741 for (m = 0; m < SYNC_MARKS; m++) {
7743 mark_cnt[m] = io_sectors;
7746 mddev->resync_mark = mark[last_mark];
7747 mddev->resync_mark_cnt = mark_cnt[last_mark];
7750 * Tune reconstruction:
7752 window = 32*(PAGE_SIZE/512);
7753 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7754 window/2, (unsigned long long)max_sectors/2);
7756 atomic_set(&mddev->recovery_active, 0);
7761 "md: resuming %s of %s from checkpoint.\n",
7762 desc, mdname(mddev));
7763 mddev->curr_resync = j;
7765 mddev->curr_resync = 3; /* no longer delayed */
7766 mddev->curr_resync_completed = j;
7767 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7768 md_new_event(mddev);
7769 update_time = jiffies;
7771 if (mddev_is_clustered(mddev))
7772 md_cluster_ops->resync_start(mddev, j, max_sectors);
7774 blk_start_plug(&plug);
7775 while (j < max_sectors) {
7780 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7781 ((mddev->curr_resync > mddev->curr_resync_completed &&
7782 (mddev->curr_resync - mddev->curr_resync_completed)
7783 > (max_sectors >> 4)) ||
7784 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
7785 (j - mddev->curr_resync_completed)*2
7786 >= mddev->resync_max - mddev->curr_resync_completed
7788 /* time to update curr_resync_completed */
7789 wait_event(mddev->recovery_wait,
7790 atomic_read(&mddev->recovery_active) == 0);
7791 mddev->curr_resync_completed = j;
7792 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7793 j > mddev->recovery_cp)
7794 mddev->recovery_cp = j;
7795 update_time = jiffies;
7796 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7797 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7800 while (j >= mddev->resync_max &&
7801 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7802 /* As this condition is controlled by user-space,
7803 * we can block indefinitely, so use '_interruptible'
7804 * to avoid triggering warnings.
7806 flush_signals(current); /* just in case */
7807 wait_event_interruptible(mddev->recovery_wait,
7808 mddev->resync_max > j
7809 || test_bit(MD_RECOVERY_INTR,
7813 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7816 sectors = mddev->pers->sync_request(mddev, j, &skipped,
7817 currspeed < speed_min(mddev));
7819 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7823 if (!skipped) { /* actual IO requested */
7824 io_sectors += sectors;
7825 atomic_add(sectors, &mddev->recovery_active);
7828 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7833 mddev->curr_resync = j;
7834 if (mddev_is_clustered(mddev))
7835 md_cluster_ops->resync_info_update(mddev, j, max_sectors);
7836 mddev->curr_mark_cnt = io_sectors;
7837 if (last_check == 0)
7838 /* this is the earliest that rebuild will be
7839 * visible in /proc/mdstat
7841 md_new_event(mddev);
7843 if (last_check + window > io_sectors || j == max_sectors)
7846 last_check = io_sectors;
7848 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7850 int next = (last_mark+1) % SYNC_MARKS;
7852 mddev->resync_mark = mark[next];
7853 mddev->resync_mark_cnt = mark_cnt[next];
7854 mark[next] = jiffies;
7855 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7859 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7863 * this loop exits only if either when we are slower than
7864 * the 'hard' speed limit, or the system was IO-idle for
7866 * the system might be non-idle CPU-wise, but we only care
7867 * about not overloading the IO subsystem. (things like an
7868 * e2fsck being done on the RAID array should execute fast)
7872 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7873 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
7874 /((jiffies-mddev->resync_mark)/HZ +1) +1;
7876 if (currspeed > speed_min(mddev)) {
7877 if ((currspeed > speed_max(mddev)) ||
7878 !is_mddev_idle(mddev, 0)) {
7884 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7885 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7886 ? "interrupted" : "done");
7888 * this also signals 'finished resyncing' to md_stop
7890 blk_finish_plug(&plug);
7891 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7893 /* tell personality that we are finished */
7894 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7896 if (mddev_is_clustered(mddev))
7897 md_cluster_ops->resync_finish(mddev);
7899 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7900 mddev->curr_resync > 2) {
7901 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7902 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7903 if (mddev->curr_resync >= mddev->recovery_cp) {
7905 "md: checkpointing %s of %s.\n",
7906 desc, mdname(mddev));
7907 if (test_bit(MD_RECOVERY_ERROR,
7909 mddev->recovery_cp =
7910 mddev->curr_resync_completed;
7912 mddev->recovery_cp =
7916 mddev->recovery_cp = MaxSector;
7918 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7919 mddev->curr_resync = MaxSector;
7921 rdev_for_each_rcu(rdev, mddev)
7922 if (rdev->raid_disk >= 0 &&
7923 mddev->delta_disks >= 0 &&
7924 !test_bit(Faulty, &rdev->flags) &&
7925 !test_bit(In_sync, &rdev->flags) &&
7926 rdev->recovery_offset < mddev->curr_resync)
7927 rdev->recovery_offset = mddev->curr_resync;
7932 set_bit(MD_CHANGE_DEVS, &mddev->flags);
7934 spin_lock(&mddev->lock);
7935 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7936 /* We completed so min/max setting can be forgotten if used. */
7937 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7938 mddev->resync_min = 0;
7939 mddev->resync_max = MaxSector;
7940 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7941 mddev->resync_min = mddev->curr_resync_completed;
7942 mddev->curr_resync = 0;
7943 spin_unlock(&mddev->lock);
7945 wake_up(&resync_wait);
7946 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7947 md_wakeup_thread(mddev->thread);
7950 EXPORT_SYMBOL_GPL(md_do_sync);
7952 static int remove_and_add_spares(struct mddev *mddev,
7953 struct md_rdev *this)
7955 struct md_rdev *rdev;
7959 rdev_for_each(rdev, mddev)
7960 if ((this == NULL || rdev == this) &&
7961 rdev->raid_disk >= 0 &&
7962 !test_bit(Blocked, &rdev->flags) &&
7963 (test_bit(Faulty, &rdev->flags) ||
7964 ! test_bit(In_sync, &rdev->flags)) &&
7965 atomic_read(&rdev->nr_pending)==0) {
7966 if (mddev->pers->hot_remove_disk(
7967 mddev, rdev) == 0) {
7968 sysfs_unlink_rdev(mddev, rdev);
7969 rdev->raid_disk = -1;
7973 if (removed && mddev->kobj.sd)
7974 sysfs_notify(&mddev->kobj, NULL, "degraded");
7979 rdev_for_each(rdev, mddev) {
7980 if (rdev->raid_disk >= 0 &&
7981 !test_bit(In_sync, &rdev->flags) &&
7982 !test_bit(Faulty, &rdev->flags))
7984 if (rdev->raid_disk >= 0)
7986 if (test_bit(Faulty, &rdev->flags))
7989 ! (rdev->saved_raid_disk >= 0 &&
7990 !test_bit(Bitmap_sync, &rdev->flags)))
7993 if (rdev->saved_raid_disk < 0)
7994 rdev->recovery_offset = 0;
7996 hot_add_disk(mddev, rdev) == 0) {
7997 if (sysfs_link_rdev(mddev, rdev))
7998 /* failure here is OK */;
8000 md_new_event(mddev);
8001 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8006 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8010 static void md_start_sync(struct work_struct *ws)
8012 struct mddev *mddev = container_of(ws, struct mddev, del_work);
8014 mddev->sync_thread = md_register_thread(md_do_sync,
8017 if (!mddev->sync_thread) {
8018 printk(KERN_ERR "%s: could not start resync"
8021 /* leave the spares where they are, it shouldn't hurt */
8022 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8023 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8024 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8025 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8026 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8027 wake_up(&resync_wait);
8028 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8030 if (mddev->sysfs_action)
8031 sysfs_notify_dirent_safe(mddev->sysfs_action);
8033 md_wakeup_thread(mddev->sync_thread);
8034 sysfs_notify_dirent_safe(mddev->sysfs_action);
8035 md_new_event(mddev);
8039 * This routine is regularly called by all per-raid-array threads to
8040 * deal with generic issues like resync and super-block update.
8041 * Raid personalities that don't have a thread (linear/raid0) do not
8042 * need this as they never do any recovery or update the superblock.
8044 * It does not do any resync itself, but rather "forks" off other threads
8045 * to do that as needed.
8046 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
8047 * "->recovery" and create a thread at ->sync_thread.
8048 * When the thread finishes it sets MD_RECOVERY_DONE
8049 * and wakeups up this thread which will reap the thread and finish up.
8050 * This thread also removes any faulty devices (with nr_pending == 0).
8052 * The overall approach is:
8053 * 1/ if the superblock needs updating, update it.
8054 * 2/ If a recovery thread is running, don't do anything else.
8055 * 3/ If recovery has finished, clean up, possibly marking spares active.
8056 * 4/ If there are any faulty devices, remove them.
8057 * 5/ If array is degraded, try to add spares devices
8058 * 6/ If array has spares or is not in-sync, start a resync thread.
8060 void md_check_recovery(struct mddev *mddev)
8062 if (mddev->suspended)
8066 bitmap_daemon_work(mddev);
8068 if (signal_pending(current)) {
8069 if (mddev->pers->sync_request && !mddev->external) {
8070 printk(KERN_INFO "md: %s in immediate safe mode\n",
8072 mddev->safemode = 2;
8074 flush_signals(current);
8077 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
8080 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
8081 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8082 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
8083 (mddev->external == 0 && mddev->safemode == 1) ||
8084 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
8085 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
8089 if (mddev_trylock(mddev)) {
8093 /* On a read-only array we can:
8094 * - remove failed devices
8095 * - add already-in_sync devices if the array itself
8097 * As we only add devices that are already in-sync,
8098 * we can activate the spares immediately.
8100 remove_and_add_spares(mddev, NULL);
8101 /* There is no thread, but we need to call
8102 * ->spare_active and clear saved_raid_disk
8104 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
8105 md_reap_sync_thread(mddev);
8106 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8110 if (!mddev->external) {
8112 spin_lock(&mddev->lock);
8113 if (mddev->safemode &&
8114 !atomic_read(&mddev->writes_pending) &&
8116 mddev->recovery_cp == MaxSector) {
8119 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
8121 if (mddev->safemode == 1)
8122 mddev->safemode = 0;
8123 spin_unlock(&mddev->lock);
8125 sysfs_notify_dirent_safe(mddev->sysfs_state);
8128 if (mddev->flags & MD_UPDATE_SB_FLAGS) {
8129 if (mddev_is_clustered(mddev))
8130 md_cluster_ops->metadata_update_start(mddev);
8131 md_update_sb(mddev, 0);
8132 if (mddev_is_clustered(mddev))
8133 md_cluster_ops->metadata_update_finish(mddev);
8136 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
8137 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
8138 /* resync/recovery still happening */
8139 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8142 if (mddev->sync_thread) {
8143 md_reap_sync_thread(mddev);
8146 /* Set RUNNING before clearing NEEDED to avoid
8147 * any transients in the value of "sync_action".
8149 mddev->curr_resync_completed = 0;
8150 spin_lock(&mddev->lock);
8151 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8152 spin_unlock(&mddev->lock);
8153 /* Clear some bits that don't mean anything, but
8156 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
8157 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
8159 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
8160 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
8162 /* no recovery is running.
8163 * remove any failed drives, then
8164 * add spares if possible.
8165 * Spares are also removed and re-added, to allow
8166 * the personality to fail the re-add.
8169 if (mddev->reshape_position != MaxSector) {
8170 if (mddev->pers->check_reshape == NULL ||
8171 mddev->pers->check_reshape(mddev) != 0)
8172 /* Cannot proceed */
8174 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8175 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8176 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
8177 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8178 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8179 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8180 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8181 } else if (mddev->recovery_cp < MaxSector) {
8182 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8183 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
8184 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
8185 /* nothing to be done ... */
8188 if (mddev->pers->sync_request) {
8190 /* We are adding a device or devices to an array
8191 * which has the bitmap stored on all devices.
8192 * So make sure all bitmap pages get written
8194 bitmap_write_all(mddev->bitmap);
8196 INIT_WORK(&mddev->del_work, md_start_sync);
8197 queue_work(md_misc_wq, &mddev->del_work);
8201 if (!mddev->sync_thread) {
8202 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8203 wake_up(&resync_wait);
8204 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8206 if (mddev->sysfs_action)
8207 sysfs_notify_dirent_safe(mddev->sysfs_action);
8210 wake_up(&mddev->sb_wait);
8211 mddev_unlock(mddev);
8214 EXPORT_SYMBOL(md_check_recovery);
8216 void md_reap_sync_thread(struct mddev *mddev)
8218 struct md_rdev *rdev;
8220 /* resync has finished, collect result */
8221 md_unregister_thread(&mddev->sync_thread);
8222 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8223 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8225 /* activate any spares */
8226 if (mddev->pers->spare_active(mddev)) {
8227 sysfs_notify(&mddev->kobj, NULL,
8229 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8232 if (mddev_is_clustered(mddev))
8233 md_cluster_ops->metadata_update_start(mddev);
8234 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8235 mddev->pers->finish_reshape)
8236 mddev->pers->finish_reshape(mddev);
8238 /* If array is no-longer degraded, then any saved_raid_disk
8239 * information must be scrapped.
8241 if (!mddev->degraded)
8242 rdev_for_each(rdev, mddev)
8243 rdev->saved_raid_disk = -1;
8245 md_update_sb(mddev, 1);
8246 if (mddev_is_clustered(mddev))
8247 md_cluster_ops->metadata_update_finish(mddev);
8248 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8249 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8250 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8251 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8252 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
8253 wake_up(&resync_wait);
8254 /* flag recovery needed just to double check */
8255 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8256 sysfs_notify_dirent_safe(mddev->sysfs_action);
8257 md_new_event(mddev);
8258 if (mddev->event_work.func)
8259 queue_work(md_misc_wq, &mddev->event_work);
8261 EXPORT_SYMBOL(md_reap_sync_thread);
8263 void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
8265 sysfs_notify_dirent_safe(rdev->sysfs_state);
8266 wait_event_timeout(rdev->blocked_wait,
8267 !test_bit(Blocked, &rdev->flags) &&
8268 !test_bit(BlockedBadBlocks, &rdev->flags),
8269 msecs_to_jiffies(5000));
8270 rdev_dec_pending(rdev, mddev);
8272 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8274 void md_finish_reshape(struct mddev *mddev)
8276 /* called be personality module when reshape completes. */
8277 struct md_rdev *rdev;
8279 rdev_for_each(rdev, mddev) {
8280 if (rdev->data_offset > rdev->new_data_offset)
8281 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8283 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8284 rdev->data_offset = rdev->new_data_offset;
8287 EXPORT_SYMBOL(md_finish_reshape);
8289 /* Bad block management.
8290 * We can record which blocks on each device are 'bad' and so just
8291 * fail those blocks, or that stripe, rather than the whole device.
8292 * Entries in the bad-block table are 64bits wide. This comprises:
8293 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8294 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8295 * A 'shift' can be set so that larger blocks are tracked and
8296 * consequently larger devices can be covered.
8297 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8299 * Locking of the bad-block table uses a seqlock so md_is_badblock
8300 * might need to retry if it is very unlucky.
8301 * We will sometimes want to check for bad blocks in a bi_end_io function,
8302 * so we use the write_seqlock_irq variant.
8304 * When looking for a bad block we specify a range and want to
8305 * know if any block in the range is bad. So we binary-search
8306 * to the last range that starts at-or-before the given endpoint,
8307 * (or "before the sector after the target range")
8308 * then see if it ends after the given start.
8310 * 0 if there are no known bad blocks in the range
8311 * 1 if there are known bad block which are all acknowledged
8312 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8313 * plus the start/length of the first bad section we overlap.
8315 int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8316 sector_t *first_bad, int *bad_sectors)
8322 sector_t target = s + sectors;
8325 if (bb->shift > 0) {
8326 /* round the start down, and the end up */
8328 target += (1<<bb->shift) - 1;
8329 target >>= bb->shift;
8330 sectors = target - s;
8332 /* 'target' is now the first block after the bad range */
8335 seq = read_seqbegin(&bb->lock);
8340 /* Binary search between lo and hi for 'target'
8341 * i.e. for the last range that starts before 'target'
8343 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8344 * are known not to be the last range before target.
8345 * VARIANT: hi-lo is the number of possible
8346 * ranges, and decreases until it reaches 1
8348 while (hi - lo > 1) {
8349 int mid = (lo + hi) / 2;
8350 sector_t a = BB_OFFSET(p[mid]);
8352 /* This could still be the one, earlier ranges
8356 /* This and later ranges are definitely out. */
8359 /* 'lo' might be the last that started before target, but 'hi' isn't */
8361 /* need to check all range that end after 's' to see if
8362 * any are unacknowledged.
8365 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8366 if (BB_OFFSET(p[lo]) < target) {
8367 /* starts before the end, and finishes after
8368 * the start, so they must overlap
8370 if (rv != -1 && BB_ACK(p[lo]))
8374 *first_bad = BB_OFFSET(p[lo]);
8375 *bad_sectors = BB_LEN(p[lo]);
8381 if (read_seqretry(&bb->lock, seq))
8386 EXPORT_SYMBOL_GPL(md_is_badblock);
8389 * Add a range of bad blocks to the table.
8390 * This might extend the table, or might contract it
8391 * if two adjacent ranges can be merged.
8392 * We binary-search to find the 'insertion' point, then
8393 * decide how best to handle it.
8395 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8401 unsigned long flags;
8404 /* badblocks are disabled */
8408 /* round the start down, and the end up */
8409 sector_t next = s + sectors;
8411 next += (1<<bb->shift) - 1;
8416 write_seqlock_irqsave(&bb->lock, flags);
8421 /* Find the last range that starts at-or-before 's' */
8422 while (hi - lo > 1) {
8423 int mid = (lo + hi) / 2;
8424 sector_t a = BB_OFFSET(p[mid]);
8430 if (hi > lo && BB_OFFSET(p[lo]) > s)
8434 /* we found a range that might merge with the start
8437 sector_t a = BB_OFFSET(p[lo]);
8438 sector_t e = a + BB_LEN(p[lo]);
8439 int ack = BB_ACK(p[lo]);
8441 /* Yes, we can merge with a previous range */
8442 if (s == a && s + sectors >= e)
8443 /* new range covers old */
8446 ack = ack && acknowledged;
8448 if (e < s + sectors)
8450 if (e - a <= BB_MAX_LEN) {
8451 p[lo] = BB_MAKE(a, e-a, ack);
8454 /* does not all fit in one range,
8455 * make p[lo] maximal
8457 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8458 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8464 if (sectors && hi < bb->count) {
8465 /* 'hi' points to the first range that starts after 's'.
8466 * Maybe we can merge with the start of that range */
8467 sector_t a = BB_OFFSET(p[hi]);
8468 sector_t e = a + BB_LEN(p[hi]);
8469 int ack = BB_ACK(p[hi]);
8470 if (a <= s + sectors) {
8471 /* merging is possible */
8472 if (e <= s + sectors) {
8477 ack = ack && acknowledged;
8480 if (e - a <= BB_MAX_LEN) {
8481 p[hi] = BB_MAKE(a, e-a, ack);
8484 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8492 if (sectors == 0 && hi < bb->count) {
8493 /* we might be able to combine lo and hi */
8494 /* Note: 's' is at the end of 'lo' */
8495 sector_t a = BB_OFFSET(p[hi]);
8496 int lolen = BB_LEN(p[lo]);
8497 int hilen = BB_LEN(p[hi]);
8498 int newlen = lolen + hilen - (s - a);
8499 if (s >= a && newlen < BB_MAX_LEN) {
8500 /* yes, we can combine them */
8501 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8502 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8503 memmove(p + hi, p + hi + 1,
8504 (bb->count - hi - 1) * 8);
8509 /* didn't merge (it all).
8510 * Need to add a range just before 'hi' */
8511 if (bb->count >= MD_MAX_BADBLOCKS) {
8512 /* No room for more */
8516 int this_sectors = sectors;
8517 memmove(p + hi + 1, p + hi,
8518 (bb->count - hi) * 8);
8521 if (this_sectors > BB_MAX_LEN)
8522 this_sectors = BB_MAX_LEN;
8523 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8524 sectors -= this_sectors;
8531 bb->unacked_exist = 1;
8532 write_sequnlock_irqrestore(&bb->lock, flags);
8537 int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8542 s += rdev->new_data_offset;
8544 s += rdev->data_offset;
8545 rv = md_set_badblocks(&rdev->badblocks,
8548 /* Make sure they get written out promptly */
8549 sysfs_notify_dirent_safe(rdev->sysfs_state);
8550 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8551 md_wakeup_thread(rdev->mddev->thread);
8555 EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8558 * Remove a range of bad blocks from the table.
8559 * This may involve extending the table if we spilt a region,
8560 * but it must not fail. So if the table becomes full, we just
8561 * drop the remove request.
8563 static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8567 sector_t target = s + sectors;
8570 if (bb->shift > 0) {
8571 /* When clearing we round the start up and the end down.
8572 * This should not matter as the shift should align with
8573 * the block size and no rounding should ever be needed.
8574 * However it is better the think a block is bad when it
8575 * isn't than to think a block is not bad when it is.
8577 s += (1<<bb->shift) - 1;
8579 target >>= bb->shift;
8580 sectors = target - s;
8583 write_seqlock_irq(&bb->lock);
8588 /* Find the last range that starts before 'target' */
8589 while (hi - lo > 1) {
8590 int mid = (lo + hi) / 2;
8591 sector_t a = BB_OFFSET(p[mid]);
8598 /* p[lo] is the last range that could overlap the
8599 * current range. Earlier ranges could also overlap,
8600 * but only this one can overlap the end of the range.
8602 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8603 /* Partial overlap, leave the tail of this range */
8604 int ack = BB_ACK(p[lo]);
8605 sector_t a = BB_OFFSET(p[lo]);
8606 sector_t end = a + BB_LEN(p[lo]);
8609 /* we need to split this range */
8610 if (bb->count >= MD_MAX_BADBLOCKS) {
8614 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8616 p[lo] = BB_MAKE(a, s-a, ack);
8619 p[lo] = BB_MAKE(target, end - target, ack);
8620 /* there is no longer an overlap */
8625 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8626 /* This range does overlap */
8627 if (BB_OFFSET(p[lo]) < s) {
8628 /* Keep the early parts of this range. */
8629 int ack = BB_ACK(p[lo]);
8630 sector_t start = BB_OFFSET(p[lo]);
8631 p[lo] = BB_MAKE(start, s - start, ack);
8632 /* now low doesn't overlap, so.. */
8637 /* 'lo' is strictly before, 'hi' is strictly after,
8638 * anything between needs to be discarded
8641 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8642 bb->count -= (hi - lo - 1);
8648 write_sequnlock_irq(&bb->lock);
8652 int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8656 s += rdev->new_data_offset;
8658 s += rdev->data_offset;
8659 return md_clear_badblocks(&rdev->badblocks,
8662 EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8665 * Acknowledge all bad blocks in a list.
8666 * This only succeeds if ->changed is clear. It is used by
8667 * in-kernel metadata updates
8669 void md_ack_all_badblocks(struct badblocks *bb)
8671 if (bb->page == NULL || bb->changed)
8672 /* no point even trying */
8674 write_seqlock_irq(&bb->lock);
8676 if (bb->changed == 0 && bb->unacked_exist) {
8679 for (i = 0; i < bb->count ; i++) {
8680 if (!BB_ACK(p[i])) {
8681 sector_t start = BB_OFFSET(p[i]);
8682 int len = BB_LEN(p[i]);
8683 p[i] = BB_MAKE(start, len, 1);
8686 bb->unacked_exist = 0;
8688 write_sequnlock_irq(&bb->lock);
8690 EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8692 /* sysfs access to bad-blocks list.
8693 * We present two files.
8694 * 'bad-blocks' lists sector numbers and lengths of ranges that
8695 * are recorded as bad. The list is truncated to fit within
8696 * the one-page limit of sysfs.
8697 * Writing "sector length" to this file adds an acknowledged
8699 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8700 * been acknowledged. Writing to this file adds bad blocks
8701 * without acknowledging them. This is largely for testing.
8705 badblocks_show(struct badblocks *bb, char *page, int unack)
8716 seq = read_seqbegin(&bb->lock);
8721 while (len < PAGE_SIZE && i < bb->count) {
8722 sector_t s = BB_OFFSET(p[i]);
8723 unsigned int length = BB_LEN(p[i]);
8724 int ack = BB_ACK(p[i]);
8730 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8731 (unsigned long long)s << bb->shift,
8732 length << bb->shift);
8734 if (unack && len == 0)
8735 bb->unacked_exist = 0;
8737 if (read_seqretry(&bb->lock, seq))
8746 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8748 unsigned long long sector;
8752 /* Allow clearing via sysfs *only* for testing/debugging.
8753 * Normally only a successful write may clear a badblock
8756 if (page[0] == '-') {
8760 #endif /* DO_DEBUG */
8762 switch (sscanf(page, "%llu %d%c", §or, &length, &newline)) {
8764 if (newline != '\n')
8776 md_clear_badblocks(bb, sector, length);
8779 #endif /* DO_DEBUG */
8780 if (md_set_badblocks(bb, sector, length, !unack))
8786 static int md_notify_reboot(struct notifier_block *this,
8787 unsigned long code, void *x)
8789 struct list_head *tmp;
8790 struct mddev *mddev;
8793 for_each_mddev(mddev, tmp) {
8794 if (mddev_trylock(mddev)) {
8796 __md_stop_writes(mddev);
8797 if (mddev->persistent)
8798 mddev->safemode = 2;
8799 mddev_unlock(mddev);
8804 * certain more exotic SCSI devices are known to be
8805 * volatile wrt too early system reboots. While the
8806 * right place to handle this issue is the given
8807 * driver, we do want to have a safe RAID driver ...
8815 static struct notifier_block md_notifier = {
8816 .notifier_call = md_notify_reboot,
8818 .priority = INT_MAX, /* before any real devices */
8821 static void md_geninit(void)
8823 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8825 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8828 static int __init md_init(void)
8832 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8836 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8840 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8843 if ((ret = register_blkdev(0, "mdp")) < 0)
8847 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
8848 md_probe, NULL, NULL);
8849 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8850 md_probe, NULL, NULL);
8852 register_reboot_notifier(&md_notifier);
8853 raid_table_header = register_sysctl_table(raid_root_table);
8859 unregister_blkdev(MD_MAJOR, "md");
8861 destroy_workqueue(md_misc_wq);
8863 destroy_workqueue(md_wq);
8868 void md_reload_sb(struct mddev *mddev)
8870 struct md_rdev *rdev, *tmp;
8872 rdev_for_each_safe(rdev, tmp, mddev) {
8873 rdev->sb_loaded = 0;
8874 ClearPageUptodate(rdev->sb_page);
8876 mddev->raid_disks = 0;
8878 rdev_for_each_safe(rdev, tmp, mddev) {
8879 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
8880 /* since we don't write to faulty devices, we figure out if the
8881 * disk is faulty by comparing events
8883 if (mddev->events > sb->events)
8884 set_bit(Faulty, &rdev->flags);
8888 EXPORT_SYMBOL(md_reload_sb);
8893 * Searches all registered partitions for autorun RAID arrays
8897 static LIST_HEAD(all_detected_devices);
8898 struct detected_devices_node {
8899 struct list_head list;
8903 void md_autodetect_dev(dev_t dev)
8905 struct detected_devices_node *node_detected_dev;
8907 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8908 if (node_detected_dev) {
8909 node_detected_dev->dev = dev;
8910 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8912 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8913 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8917 static void autostart_arrays(int part)
8919 struct md_rdev *rdev;
8920 struct detected_devices_node *node_detected_dev;
8922 int i_scanned, i_passed;
8927 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8929 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8931 node_detected_dev = list_entry(all_detected_devices.next,
8932 struct detected_devices_node, list);
8933 list_del(&node_detected_dev->list);
8934 dev = node_detected_dev->dev;
8935 kfree(node_detected_dev);
8936 rdev = md_import_device(dev,0, 90);
8940 if (test_bit(Faulty, &rdev->flags))
8943 set_bit(AutoDetected, &rdev->flags);
8944 list_add(&rdev->same_set, &pending_raid_disks);
8948 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8949 i_scanned, i_passed);
8951 autorun_devices(part);
8954 #endif /* !MODULE */
8956 static __exit void md_exit(void)
8958 struct mddev *mddev;
8959 struct list_head *tmp;
8962 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
8963 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8965 unregister_blkdev(MD_MAJOR,"md");
8966 unregister_blkdev(mdp_major, "mdp");
8967 unregister_reboot_notifier(&md_notifier);
8968 unregister_sysctl_table(raid_table_header);
8970 /* We cannot unload the modules while some process is
8971 * waiting for us in select() or poll() - wake them up
8974 while (waitqueue_active(&md_event_waiters)) {
8975 /* not safe to leave yet */
8976 wake_up(&md_event_waiters);
8980 remove_proc_entry("mdstat", NULL);
8982 for_each_mddev(mddev, tmp) {
8983 export_array(mddev);
8984 mddev->hold_active = 0;
8986 destroy_workqueue(md_misc_wq);
8987 destroy_workqueue(md_wq);
8990 subsys_initcall(md_init);
8991 module_exit(md_exit)
8993 static int get_ro(char *buffer, struct kernel_param *kp)
8995 return sprintf(buffer, "%d", start_readonly);
8997 static int set_ro(const char *val, struct kernel_param *kp)
9000 int num = simple_strtoul(val, &e, 10);
9001 if (*val && (*e == '\0' || *e == '\n')) {
9002 start_readonly = num;
9008 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
9009 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
9010 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
9012 MODULE_LICENSE("GPL");
9013 MODULE_DESCRIPTION("MD RAID framework");
9015 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);