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/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part);
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
75 static void md_print_devices(void);
77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min = 1000;
95 static int sysctl_speed_limit_max = 200000;
96 static inline int speed_min(mddev_t *mddev)
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
102 static inline int speed_max(mddev_t *mddev)
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
108 static struct ctl_table_header *raid_table_header;
110 static ctl_table raid_table[] = {
112 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
113 .procname = "speed_limit_min",
114 .data = &sysctl_speed_limit_min,
115 .maxlen = sizeof(int),
116 .mode = S_IRUGO|S_IWUSR,
117 .proc_handler = &proc_dointvec,
120 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = &proc_dointvec,
130 static ctl_table raid_dir_table[] = {
132 .ctl_name = DEV_RAID,
135 .mode = S_IRUGO|S_IXUGO,
141 static ctl_table raid_root_table[] = {
147 .child = raid_dir_table,
152 static struct block_device_operations md_fops;
154 static int start_readonly;
157 * We have a system wide 'event count' that is incremented
158 * on any 'interesting' event, and readers of /proc/mdstat
159 * can use 'poll' or 'select' to find out when the event
163 * start array, stop array, error, add device, remove device,
164 * start build, activate spare
166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
167 static atomic_t md_event_count;
168 void md_new_event(mddev_t *mddev)
170 atomic_inc(&md_event_count);
171 wake_up(&md_event_waiters);
173 EXPORT_SYMBOL_GPL(md_new_event);
175 /* Alternate version that can be called from interrupts
176 * when calling sysfs_notify isn't needed.
178 static void md_new_event_inintr(mddev_t *mddev)
180 atomic_inc(&md_event_count);
181 wake_up(&md_event_waiters);
185 * Enables to iterate over all existing md arrays
186 * all_mddevs_lock protects this list.
188 static LIST_HEAD(all_mddevs);
189 static DEFINE_SPINLOCK(all_mddevs_lock);
193 * iterates through all used mddevs in the system.
194 * We take care to grab the all_mddevs_lock whenever navigating
195 * the list, and to always hold a refcount when unlocked.
196 * Any code which breaks out of this loop while own
197 * a reference to the current mddev and must mddev_put it.
199 #define for_each_mddev(mddev,tmp) \
201 for (({ spin_lock(&all_mddevs_lock); \
202 tmp = all_mddevs.next; \
204 ({ if (tmp != &all_mddevs) \
205 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
206 spin_unlock(&all_mddevs_lock); \
207 if (mddev) mddev_put(mddev); \
208 mddev = list_entry(tmp, mddev_t, all_mddevs); \
209 tmp != &all_mddevs;}); \
210 ({ spin_lock(&all_mddevs_lock); \
215 static int md_fail_request (struct request_queue *q, struct bio *bio)
221 static inline mddev_t *mddev_get(mddev_t *mddev)
223 atomic_inc(&mddev->active);
227 static void mddev_put(mddev_t *mddev)
229 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
231 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
232 list_del(&mddev->all_mddevs);
233 spin_unlock(&all_mddevs_lock);
234 blk_cleanup_queue(mddev->queue);
235 kobject_put(&mddev->kobj);
237 spin_unlock(&all_mddevs_lock);
240 static mddev_t * mddev_find(dev_t unit)
242 mddev_t *mddev, *new = NULL;
245 spin_lock(&all_mddevs_lock);
246 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
247 if (mddev->unit == unit) {
249 spin_unlock(&all_mddevs_lock);
255 list_add(&new->all_mddevs, &all_mddevs);
256 spin_unlock(&all_mddevs_lock);
259 spin_unlock(&all_mddevs_lock);
261 new = kzalloc(sizeof(*new), GFP_KERNEL);
266 if (MAJOR(unit) == MD_MAJOR)
267 new->md_minor = MINOR(unit);
269 new->md_minor = MINOR(unit) >> MdpMinorShift;
271 mutex_init(&new->reconfig_mutex);
272 INIT_LIST_HEAD(&new->disks);
273 INIT_LIST_HEAD(&new->all_mddevs);
274 init_timer(&new->safemode_timer);
275 atomic_set(&new->active, 1);
276 spin_lock_init(&new->write_lock);
277 init_waitqueue_head(&new->sb_wait);
278 init_waitqueue_head(&new->recovery_wait);
279 new->reshape_position = MaxSector;
281 new->resync_max = MaxSector;
282 new->level = LEVEL_NONE;
284 new->queue = blk_alloc_queue(GFP_KERNEL);
289 /* Can be unlocked because the queue is new: no concurrency */
290 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
292 blk_queue_make_request(new->queue, md_fail_request);
297 static inline int mddev_lock(mddev_t * mddev)
299 return mutex_lock_interruptible(&mddev->reconfig_mutex);
302 static inline int mddev_trylock(mddev_t * mddev)
304 return mutex_trylock(&mddev->reconfig_mutex);
307 static inline void mddev_unlock(mddev_t * mddev)
309 mutex_unlock(&mddev->reconfig_mutex);
311 md_wakeup_thread(mddev->thread);
314 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
317 struct list_head *tmp;
319 rdev_for_each(rdev, tmp, mddev) {
320 if (rdev->desc_nr == nr)
326 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
328 struct list_head *tmp;
331 rdev_for_each(rdev, tmp, mddev) {
332 if (rdev->bdev->bd_dev == dev)
338 static struct mdk_personality *find_pers(int level, char *clevel)
340 struct mdk_personality *pers;
341 list_for_each_entry(pers, &pers_list, list) {
342 if (level != LEVEL_NONE && pers->level == level)
344 if (strcmp(pers->name, clevel)==0)
350 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
352 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
353 return MD_NEW_SIZE_BLOCKS(size);
356 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
360 size = rdev->sb_offset;
363 size &= ~((sector_t)chunk_size/1024 - 1);
367 static int alloc_disk_sb(mdk_rdev_t * rdev)
372 rdev->sb_page = alloc_page(GFP_KERNEL);
373 if (!rdev->sb_page) {
374 printk(KERN_ALERT "md: out of memory.\n");
381 static void free_disk_sb(mdk_rdev_t * rdev)
384 put_page(rdev->sb_page);
386 rdev->sb_page = NULL;
393 static void super_written(struct bio *bio, int error)
395 mdk_rdev_t *rdev = bio->bi_private;
396 mddev_t *mddev = rdev->mddev;
398 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
399 printk("md: super_written gets error=%d, uptodate=%d\n",
400 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
401 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
402 md_error(mddev, rdev);
405 if (atomic_dec_and_test(&mddev->pending_writes))
406 wake_up(&mddev->sb_wait);
410 static void super_written_barrier(struct bio *bio, int error)
412 struct bio *bio2 = bio->bi_private;
413 mdk_rdev_t *rdev = bio2->bi_private;
414 mddev_t *mddev = rdev->mddev;
416 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
417 error == -EOPNOTSUPP) {
419 /* barriers don't appear to be supported :-( */
420 set_bit(BarriersNotsupp, &rdev->flags);
421 mddev->barriers_work = 0;
422 spin_lock_irqsave(&mddev->write_lock, flags);
423 bio2->bi_next = mddev->biolist;
424 mddev->biolist = bio2;
425 spin_unlock_irqrestore(&mddev->write_lock, flags);
426 wake_up(&mddev->sb_wait);
430 bio->bi_private = rdev;
431 super_written(bio, error);
435 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
436 sector_t sector, int size, struct page *page)
438 /* write first size bytes of page to sector of rdev
439 * Increment mddev->pending_writes before returning
440 * and decrement it on completion, waking up sb_wait
441 * if zero is reached.
442 * If an error occurred, call md_error
444 * As we might need to resubmit the request if BIO_RW_BARRIER
445 * causes ENOTSUPP, we allocate a spare bio...
447 struct bio *bio = bio_alloc(GFP_NOIO, 1);
448 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
450 bio->bi_bdev = rdev->bdev;
451 bio->bi_sector = sector;
452 bio_add_page(bio, page, size, 0);
453 bio->bi_private = rdev;
454 bio->bi_end_io = super_written;
457 atomic_inc(&mddev->pending_writes);
458 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
460 rw |= (1<<BIO_RW_BARRIER);
461 rbio = bio_clone(bio, GFP_NOIO);
462 rbio->bi_private = bio;
463 rbio->bi_end_io = super_written_barrier;
464 submit_bio(rw, rbio);
469 void md_super_wait(mddev_t *mddev)
471 /* wait for all superblock writes that were scheduled to complete.
472 * if any had to be retried (due to BARRIER problems), retry them
476 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
477 if (atomic_read(&mddev->pending_writes)==0)
479 while (mddev->biolist) {
481 spin_lock_irq(&mddev->write_lock);
482 bio = mddev->biolist;
483 mddev->biolist = bio->bi_next ;
485 spin_unlock_irq(&mddev->write_lock);
486 submit_bio(bio->bi_rw, bio);
490 finish_wait(&mddev->sb_wait, &wq);
493 static void bi_complete(struct bio *bio, int error)
495 complete((struct completion*)bio->bi_private);
498 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
499 struct page *page, int rw)
501 struct bio *bio = bio_alloc(GFP_NOIO, 1);
502 struct completion event;
505 rw |= (1 << BIO_RW_SYNC);
508 bio->bi_sector = sector;
509 bio_add_page(bio, page, size, 0);
510 init_completion(&event);
511 bio->bi_private = &event;
512 bio->bi_end_io = bi_complete;
514 wait_for_completion(&event);
516 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
520 EXPORT_SYMBOL_GPL(sync_page_io);
522 static int read_disk_sb(mdk_rdev_t * rdev, int size)
524 char b[BDEVNAME_SIZE];
525 if (!rdev->sb_page) {
533 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
539 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
540 bdevname(rdev->bdev,b));
544 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
546 return sb1->set_uuid0 == sb2->set_uuid0 &&
547 sb1->set_uuid1 == sb2->set_uuid1 &&
548 sb1->set_uuid2 == sb2->set_uuid2 &&
549 sb1->set_uuid3 == sb2->set_uuid3;
552 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
555 mdp_super_t *tmp1, *tmp2;
557 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
558 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
560 if (!tmp1 || !tmp2) {
562 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
570 * nr_disks is not constant
575 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
587 static u32 md_csum_fold(u32 csum)
589 csum = (csum & 0xffff) + (csum >> 16);
590 return (csum & 0xffff) + (csum >> 16);
593 static unsigned int calc_sb_csum(mdp_super_t * sb)
596 u32 *sb32 = (u32*)sb;
598 unsigned int disk_csum, csum;
600 disk_csum = sb->sb_csum;
603 for (i = 0; i < MD_SB_BYTES/4 ; i++)
605 csum = (newcsum & 0xffffffff) + (newcsum>>32);
609 /* This used to use csum_partial, which was wrong for several
610 * reasons including that different results are returned on
611 * different architectures. It isn't critical that we get exactly
612 * the same return value as before (we always csum_fold before
613 * testing, and that removes any differences). However as we
614 * know that csum_partial always returned a 16bit value on
615 * alphas, do a fold to maximise conformity to previous behaviour.
617 sb->sb_csum = md_csum_fold(disk_csum);
619 sb->sb_csum = disk_csum;
626 * Handle superblock details.
627 * We want to be able to handle multiple superblock formats
628 * so we have a common interface to them all, and an array of
629 * different handlers.
630 * We rely on user-space to write the initial superblock, and support
631 * reading and updating of superblocks.
632 * Interface methods are:
633 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
634 * loads and validates a superblock on dev.
635 * if refdev != NULL, compare superblocks on both devices
637 * 0 - dev has a superblock that is compatible with refdev
638 * 1 - dev has a superblock that is compatible and newer than refdev
639 * so dev should be used as the refdev in future
640 * -EINVAL superblock incompatible or invalid
641 * -othererror e.g. -EIO
643 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
644 * Verify that dev is acceptable into mddev.
645 * The first time, mddev->raid_disks will be 0, and data from
646 * dev should be merged in. Subsequent calls check that dev
647 * is new enough. Return 0 or -EINVAL
649 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
650 * Update the superblock for rdev with data in mddev
651 * This does not write to disc.
657 struct module *owner;
658 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
660 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
661 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
662 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
663 unsigned long long size);
667 * load_super for 0.90.0
669 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
671 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
677 * Calculate the position of the superblock,
678 * it's at the end of the disk.
680 * It also happens to be a multiple of 4Kb.
682 sb_offset = calc_dev_sboffset(rdev->bdev);
683 rdev->sb_offset = sb_offset;
685 ret = read_disk_sb(rdev, MD_SB_BYTES);
690 bdevname(rdev->bdev, b);
691 sb = (mdp_super_t*)page_address(rdev->sb_page);
693 if (sb->md_magic != MD_SB_MAGIC) {
694 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
699 if (sb->major_version != 0 ||
700 sb->minor_version < 90 ||
701 sb->minor_version > 91) {
702 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
703 sb->major_version, sb->minor_version,
708 if (sb->raid_disks <= 0)
711 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
712 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
717 rdev->preferred_minor = sb->md_minor;
718 rdev->data_offset = 0;
719 rdev->sb_size = MD_SB_BYTES;
721 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
722 if (sb->level != 1 && sb->level != 4
723 && sb->level != 5 && sb->level != 6
724 && sb->level != 10) {
725 /* FIXME use a better test */
727 "md: bitmaps not supported for this level.\n");
732 if (sb->level == LEVEL_MULTIPATH)
735 rdev->desc_nr = sb->this_disk.number;
741 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
742 if (!uuid_equal(refsb, sb)) {
743 printk(KERN_WARNING "md: %s has different UUID to %s\n",
744 b, bdevname(refdev->bdev,b2));
747 if (!sb_equal(refsb, sb)) {
748 printk(KERN_WARNING "md: %s has same UUID"
749 " but different superblock to %s\n",
750 b, bdevname(refdev->bdev, b2));
754 ev2 = md_event(refsb);
760 rdev->size = calc_dev_size(rdev, sb->chunk_size);
762 if (rdev->size < sb->size && sb->level > 1)
763 /* "this cannot possibly happen" ... */
771 * validate_super for 0.90.0
773 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
776 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
777 __u64 ev1 = md_event(sb);
779 rdev->raid_disk = -1;
780 clear_bit(Faulty, &rdev->flags);
781 clear_bit(In_sync, &rdev->flags);
782 clear_bit(WriteMostly, &rdev->flags);
783 clear_bit(BarriersNotsupp, &rdev->flags);
785 if (mddev->raid_disks == 0) {
786 mddev->major_version = 0;
787 mddev->minor_version = sb->minor_version;
788 mddev->patch_version = sb->patch_version;
790 mddev->chunk_size = sb->chunk_size;
791 mddev->ctime = sb->ctime;
792 mddev->utime = sb->utime;
793 mddev->level = sb->level;
794 mddev->clevel[0] = 0;
795 mddev->layout = sb->layout;
796 mddev->raid_disks = sb->raid_disks;
797 mddev->size = sb->size;
799 mddev->bitmap_offset = 0;
800 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
802 if (mddev->minor_version >= 91) {
803 mddev->reshape_position = sb->reshape_position;
804 mddev->delta_disks = sb->delta_disks;
805 mddev->new_level = sb->new_level;
806 mddev->new_layout = sb->new_layout;
807 mddev->new_chunk = sb->new_chunk;
809 mddev->reshape_position = MaxSector;
810 mddev->delta_disks = 0;
811 mddev->new_level = mddev->level;
812 mddev->new_layout = mddev->layout;
813 mddev->new_chunk = mddev->chunk_size;
816 if (sb->state & (1<<MD_SB_CLEAN))
817 mddev->recovery_cp = MaxSector;
819 if (sb->events_hi == sb->cp_events_hi &&
820 sb->events_lo == sb->cp_events_lo) {
821 mddev->recovery_cp = sb->recovery_cp;
823 mddev->recovery_cp = 0;
826 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
827 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
828 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
829 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
831 mddev->max_disks = MD_SB_DISKS;
833 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
834 mddev->bitmap_file == NULL)
835 mddev->bitmap_offset = mddev->default_bitmap_offset;
837 } else if (mddev->pers == NULL) {
838 /* Insist on good event counter while assembling */
840 if (ev1 < mddev->events)
842 } else if (mddev->bitmap) {
843 /* if adding to array with a bitmap, then we can accept an
844 * older device ... but not too old.
846 if (ev1 < mddev->bitmap->events_cleared)
849 if (ev1 < mddev->events)
850 /* just a hot-add of a new device, leave raid_disk at -1 */
854 if (mddev->level != LEVEL_MULTIPATH) {
855 desc = sb->disks + rdev->desc_nr;
857 if (desc->state & (1<<MD_DISK_FAULTY))
858 set_bit(Faulty, &rdev->flags);
859 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
860 desc->raid_disk < mddev->raid_disks */) {
861 set_bit(In_sync, &rdev->flags);
862 rdev->raid_disk = desc->raid_disk;
864 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
865 set_bit(WriteMostly, &rdev->flags);
866 } else /* MULTIPATH are always insync */
867 set_bit(In_sync, &rdev->flags);
872 * sync_super for 0.90.0
874 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
877 struct list_head *tmp;
879 int next_spare = mddev->raid_disks;
882 /* make rdev->sb match mddev data..
885 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
886 * 3/ any empty disks < next_spare become removed
888 * disks[0] gets initialised to REMOVED because
889 * we cannot be sure from other fields if it has
890 * been initialised or not.
893 int active=0, working=0,failed=0,spare=0,nr_disks=0;
895 rdev->sb_size = MD_SB_BYTES;
897 sb = (mdp_super_t*)page_address(rdev->sb_page);
899 memset(sb, 0, sizeof(*sb));
901 sb->md_magic = MD_SB_MAGIC;
902 sb->major_version = mddev->major_version;
903 sb->patch_version = mddev->patch_version;
904 sb->gvalid_words = 0; /* ignored */
905 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
906 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
907 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
908 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
910 sb->ctime = mddev->ctime;
911 sb->level = mddev->level;
912 sb->size = mddev->size;
913 sb->raid_disks = mddev->raid_disks;
914 sb->md_minor = mddev->md_minor;
915 sb->not_persistent = 0;
916 sb->utime = mddev->utime;
918 sb->events_hi = (mddev->events>>32);
919 sb->events_lo = (u32)mddev->events;
921 if (mddev->reshape_position == MaxSector)
922 sb->minor_version = 90;
924 sb->minor_version = 91;
925 sb->reshape_position = mddev->reshape_position;
926 sb->new_level = mddev->new_level;
927 sb->delta_disks = mddev->delta_disks;
928 sb->new_layout = mddev->new_layout;
929 sb->new_chunk = mddev->new_chunk;
931 mddev->minor_version = sb->minor_version;
934 sb->recovery_cp = mddev->recovery_cp;
935 sb->cp_events_hi = (mddev->events>>32);
936 sb->cp_events_lo = (u32)mddev->events;
937 if (mddev->recovery_cp == MaxSector)
938 sb->state = (1<< MD_SB_CLEAN);
942 sb->layout = mddev->layout;
943 sb->chunk_size = mddev->chunk_size;
945 if (mddev->bitmap && mddev->bitmap_file == NULL)
946 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
948 sb->disks[0].state = (1<<MD_DISK_REMOVED);
949 rdev_for_each(rdev2, tmp, mddev) {
952 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
953 && !test_bit(Faulty, &rdev2->flags))
954 desc_nr = rdev2->raid_disk;
956 desc_nr = next_spare++;
957 rdev2->desc_nr = desc_nr;
958 d = &sb->disks[rdev2->desc_nr];
960 d->number = rdev2->desc_nr;
961 d->major = MAJOR(rdev2->bdev->bd_dev);
962 d->minor = MINOR(rdev2->bdev->bd_dev);
963 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
964 && !test_bit(Faulty, &rdev2->flags))
965 d->raid_disk = rdev2->raid_disk;
967 d->raid_disk = rdev2->desc_nr; /* compatibility */
968 if (test_bit(Faulty, &rdev2->flags))
969 d->state = (1<<MD_DISK_FAULTY);
970 else if (test_bit(In_sync, &rdev2->flags)) {
971 d->state = (1<<MD_DISK_ACTIVE);
972 d->state |= (1<<MD_DISK_SYNC);
980 if (test_bit(WriteMostly, &rdev2->flags))
981 d->state |= (1<<MD_DISK_WRITEMOSTLY);
983 /* now set the "removed" and "faulty" bits on any missing devices */
984 for (i=0 ; i < mddev->raid_disks ; i++) {
985 mdp_disk_t *d = &sb->disks[i];
986 if (d->state == 0 && d->number == 0) {
989 d->state = (1<<MD_DISK_REMOVED);
990 d->state |= (1<<MD_DISK_FAULTY);
994 sb->nr_disks = nr_disks;
995 sb->active_disks = active;
996 sb->working_disks = working;
997 sb->failed_disks = failed;
998 sb->spare_disks = spare;
1000 sb->this_disk = sb->disks[rdev->desc_nr];
1001 sb->sb_csum = calc_sb_csum(sb);
1005 * rdev_size_change for 0.90.0
1007 static unsigned long long
1008 super_90_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1010 if (size && size < rdev->mddev->size)
1011 return 0; /* component must fit device */
1012 size *= 2; /* convert to sectors */
1013 if (rdev->mddev->bitmap_offset)
1014 return 0; /* can't move bitmap */
1015 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
1016 if (!size || size > rdev->sb_offset*2)
1017 size = rdev->sb_offset*2;
1018 md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1020 md_super_wait(rdev->mddev);
1021 return size/2; /* kB for sysfs */
1026 * version 1 superblock
1029 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1033 unsigned long long newcsum;
1034 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1035 __le32 *isuper = (__le32*)sb;
1038 disk_csum = sb->sb_csum;
1041 for (i=0; size>=4; size -= 4 )
1042 newcsum += le32_to_cpu(*isuper++);
1045 newcsum += le16_to_cpu(*(__le16*) isuper);
1047 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1048 sb->sb_csum = disk_csum;
1049 return cpu_to_le32(csum);
1052 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1054 struct mdp_superblock_1 *sb;
1057 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1061 * Calculate the position of the superblock.
1062 * It is always aligned to a 4K boundary and
1063 * depeding on minor_version, it can be:
1064 * 0: At least 8K, but less than 12K, from end of device
1065 * 1: At start of device
1066 * 2: 4K from start of device.
1068 switch(minor_version) {
1070 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1072 sb_offset &= ~(sector_t)(4*2-1);
1073 /* convert from sectors to K */
1085 rdev->sb_offset = sb_offset;
1087 /* superblock is rarely larger than 1K, but it can be larger,
1088 * and it is safe to read 4k, so we do that
1090 ret = read_disk_sb(rdev, 4096);
1091 if (ret) return ret;
1094 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1096 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1097 sb->major_version != cpu_to_le32(1) ||
1098 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1099 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1100 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1103 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1104 printk("md: invalid superblock checksum on %s\n",
1105 bdevname(rdev->bdev,b));
1108 if (le64_to_cpu(sb->data_size) < 10) {
1109 printk("md: data_size too small on %s\n",
1110 bdevname(rdev->bdev,b));
1113 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1114 if (sb->level != cpu_to_le32(1) &&
1115 sb->level != cpu_to_le32(4) &&
1116 sb->level != cpu_to_le32(5) &&
1117 sb->level != cpu_to_le32(6) &&
1118 sb->level != cpu_to_le32(10)) {
1120 "md: bitmaps not supported for this level.\n");
1125 rdev->preferred_minor = 0xffff;
1126 rdev->data_offset = le64_to_cpu(sb->data_offset);
1127 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1129 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1130 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1131 if (rdev->sb_size & bmask)
1132 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1135 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1138 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1141 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1147 struct mdp_superblock_1 *refsb =
1148 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1150 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1151 sb->level != refsb->level ||
1152 sb->layout != refsb->layout ||
1153 sb->chunksize != refsb->chunksize) {
1154 printk(KERN_WARNING "md: %s has strangely different"
1155 " superblock to %s\n",
1156 bdevname(rdev->bdev,b),
1157 bdevname(refdev->bdev,b2));
1160 ev1 = le64_to_cpu(sb->events);
1161 ev2 = le64_to_cpu(refsb->events);
1169 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1171 rdev->size = rdev->sb_offset;
1172 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1174 rdev->size = le64_to_cpu(sb->data_size)/2;
1175 if (le32_to_cpu(sb->chunksize))
1176 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1178 if (le64_to_cpu(sb->size) > rdev->size*2)
1183 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1185 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1186 __u64 ev1 = le64_to_cpu(sb->events);
1188 rdev->raid_disk = -1;
1189 clear_bit(Faulty, &rdev->flags);
1190 clear_bit(In_sync, &rdev->flags);
1191 clear_bit(WriteMostly, &rdev->flags);
1192 clear_bit(BarriersNotsupp, &rdev->flags);
1194 if (mddev->raid_disks == 0) {
1195 mddev->major_version = 1;
1196 mddev->patch_version = 0;
1197 mddev->external = 0;
1198 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1199 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1200 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1201 mddev->level = le32_to_cpu(sb->level);
1202 mddev->clevel[0] = 0;
1203 mddev->layout = le32_to_cpu(sb->layout);
1204 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1205 mddev->size = le64_to_cpu(sb->size)/2;
1206 mddev->events = ev1;
1207 mddev->bitmap_offset = 0;
1208 mddev->default_bitmap_offset = 1024 >> 9;
1210 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1211 memcpy(mddev->uuid, sb->set_uuid, 16);
1213 mddev->max_disks = (4096-256)/2;
1215 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1216 mddev->bitmap_file == NULL )
1217 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1219 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1220 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1221 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1222 mddev->new_level = le32_to_cpu(sb->new_level);
1223 mddev->new_layout = le32_to_cpu(sb->new_layout);
1224 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1226 mddev->reshape_position = MaxSector;
1227 mddev->delta_disks = 0;
1228 mddev->new_level = mddev->level;
1229 mddev->new_layout = mddev->layout;
1230 mddev->new_chunk = mddev->chunk_size;
1233 } else if (mddev->pers == NULL) {
1234 /* Insist of good event counter while assembling */
1236 if (ev1 < mddev->events)
1238 } else if (mddev->bitmap) {
1239 /* If adding to array with a bitmap, then we can accept an
1240 * older device, but not too old.
1242 if (ev1 < mddev->bitmap->events_cleared)
1245 if (ev1 < mddev->events)
1246 /* just a hot-add of a new device, leave raid_disk at -1 */
1249 if (mddev->level != LEVEL_MULTIPATH) {
1251 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1253 case 0xffff: /* spare */
1255 case 0xfffe: /* faulty */
1256 set_bit(Faulty, &rdev->flags);
1259 if ((le32_to_cpu(sb->feature_map) &
1260 MD_FEATURE_RECOVERY_OFFSET))
1261 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1263 set_bit(In_sync, &rdev->flags);
1264 rdev->raid_disk = role;
1267 if (sb->devflags & WriteMostly1)
1268 set_bit(WriteMostly, &rdev->flags);
1269 } else /* MULTIPATH are always insync */
1270 set_bit(In_sync, &rdev->flags);
1275 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1277 struct mdp_superblock_1 *sb;
1278 struct list_head *tmp;
1281 /* make rdev->sb match mddev and rdev data. */
1283 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1285 sb->feature_map = 0;
1287 sb->recovery_offset = cpu_to_le64(0);
1288 memset(sb->pad1, 0, sizeof(sb->pad1));
1289 memset(sb->pad2, 0, sizeof(sb->pad2));
1290 memset(sb->pad3, 0, sizeof(sb->pad3));
1292 sb->utime = cpu_to_le64((__u64)mddev->utime);
1293 sb->events = cpu_to_le64(mddev->events);
1295 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1297 sb->resync_offset = cpu_to_le64(0);
1299 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1301 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1302 sb->size = cpu_to_le64(mddev->size<<1);
1304 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1305 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1306 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1309 if (rdev->raid_disk >= 0 &&
1310 !test_bit(In_sync, &rdev->flags) &&
1311 rdev->recovery_offset > 0) {
1312 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1313 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1316 if (mddev->reshape_position != MaxSector) {
1317 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1318 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1319 sb->new_layout = cpu_to_le32(mddev->new_layout);
1320 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1321 sb->new_level = cpu_to_le32(mddev->new_level);
1322 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1326 rdev_for_each(rdev2, tmp, mddev)
1327 if (rdev2->desc_nr+1 > max_dev)
1328 max_dev = rdev2->desc_nr+1;
1330 if (max_dev > le32_to_cpu(sb->max_dev))
1331 sb->max_dev = cpu_to_le32(max_dev);
1332 for (i=0; i<max_dev;i++)
1333 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1335 rdev_for_each(rdev2, tmp, mddev) {
1337 if (test_bit(Faulty, &rdev2->flags))
1338 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1339 else if (test_bit(In_sync, &rdev2->flags))
1340 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1341 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1342 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1344 sb->dev_roles[i] = cpu_to_le16(0xffff);
1347 sb->sb_csum = calc_sb_1_csum(sb);
1350 static unsigned long long
1351 super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1353 struct mdp_superblock_1 *sb;
1354 unsigned long long max_size;
1355 if (size && size < rdev->mddev->size)
1356 return 0; /* component must fit device */
1357 size *= 2; /* convert to sectors */
1358 if (rdev->sb_offset < rdev->data_offset/2) {
1359 /* minor versions 1 and 2; superblock before data */
1360 max_size = (rdev->bdev->bd_inode->i_size >> 9);
1361 max_size -= rdev->data_offset;
1362 if (!size || size > max_size)
1364 } else if (rdev->mddev->bitmap_offset) {
1365 /* minor version 0 with bitmap we can't move */
1368 /* minor version 0; superblock after data */
1370 sb_offset = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1371 sb_offset &= ~(sector_t)(4*2 - 1);
1372 max_size = rdev->size*2 + sb_offset - rdev->sb_offset*2;
1373 if (!size || size > max_size)
1375 rdev->sb_offset = sb_offset/2;
1377 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1378 sb->data_size = cpu_to_le64(size);
1379 sb->super_offset = rdev->sb_offset*2;
1380 sb->sb_csum = calc_sb_1_csum(sb);
1381 md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1383 md_super_wait(rdev->mddev);
1384 return size/2; /* kB for sysfs */
1387 static struct super_type super_types[] = {
1390 .owner = THIS_MODULE,
1391 .load_super = super_90_load,
1392 .validate_super = super_90_validate,
1393 .sync_super = super_90_sync,
1394 .rdev_size_change = super_90_rdev_size_change,
1398 .owner = THIS_MODULE,
1399 .load_super = super_1_load,
1400 .validate_super = super_1_validate,
1401 .sync_super = super_1_sync,
1402 .rdev_size_change = super_1_rdev_size_change,
1406 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1408 struct list_head *tmp, *tmp2;
1409 mdk_rdev_t *rdev, *rdev2;
1411 rdev_for_each(rdev, tmp, mddev1)
1412 rdev_for_each(rdev2, tmp2, mddev2)
1413 if (rdev->bdev->bd_contains ==
1414 rdev2->bdev->bd_contains)
1420 static LIST_HEAD(pending_raid_disks);
1422 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1424 char b[BDEVNAME_SIZE];
1434 /* prevent duplicates */
1435 if (find_rdev(mddev, rdev->bdev->bd_dev))
1438 /* make sure rdev->size exceeds mddev->size */
1439 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1441 /* Cannot change size, so fail
1442 * If mddev->level <= 0, then we don't care
1443 * about aligning sizes (e.g. linear)
1445 if (mddev->level > 0)
1448 mddev->size = rdev->size;
1451 /* Verify rdev->desc_nr is unique.
1452 * If it is -1, assign a free number, else
1453 * check number is not in use
1455 if (rdev->desc_nr < 0) {
1457 if (mddev->pers) choice = mddev->raid_disks;
1458 while (find_rdev_nr(mddev, choice))
1460 rdev->desc_nr = choice;
1462 if (find_rdev_nr(mddev, rdev->desc_nr))
1465 bdevname(rdev->bdev,b);
1466 while ( (s=strchr(b, '/')) != NULL)
1469 rdev->mddev = mddev;
1470 printk(KERN_INFO "md: bind<%s>\n", b);
1472 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1475 if (rdev->bdev->bd_part)
1476 ko = &rdev->bdev->bd_part->dev.kobj;
1478 ko = &rdev->bdev->bd_disk->dev.kobj;
1479 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1480 kobject_del(&rdev->kobj);
1483 list_add(&rdev->same_set, &mddev->disks);
1484 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1488 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1493 static void md_delayed_delete(struct work_struct *ws)
1495 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1496 kobject_del(&rdev->kobj);
1497 kobject_put(&rdev->kobj);
1500 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1502 char b[BDEVNAME_SIZE];
1507 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1508 list_del_init(&rdev->same_set);
1509 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1511 sysfs_remove_link(&rdev->kobj, "block");
1513 /* We need to delay this, otherwise we can deadlock when
1514 * writing to 'remove' to "dev/state"
1516 INIT_WORK(&rdev->del_work, md_delayed_delete);
1517 kobject_get(&rdev->kobj);
1518 schedule_work(&rdev->del_work);
1522 * prevent the device from being mounted, repartitioned or
1523 * otherwise reused by a RAID array (or any other kernel
1524 * subsystem), by bd_claiming the device.
1526 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1529 struct block_device *bdev;
1530 char b[BDEVNAME_SIZE];
1532 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1534 printk(KERN_ERR "md: could not open %s.\n",
1535 __bdevname(dev, b));
1536 return PTR_ERR(bdev);
1538 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1540 printk(KERN_ERR "md: could not bd_claim %s.\n",
1546 set_bit(AllReserved, &rdev->flags);
1551 static void unlock_rdev(mdk_rdev_t *rdev)
1553 struct block_device *bdev = rdev->bdev;
1561 void md_autodetect_dev(dev_t dev);
1563 static void export_rdev(mdk_rdev_t * rdev)
1565 char b[BDEVNAME_SIZE];
1566 printk(KERN_INFO "md: export_rdev(%s)\n",
1567 bdevname(rdev->bdev,b));
1571 list_del_init(&rdev->same_set);
1573 if (test_bit(AutoDetected, &rdev->flags))
1574 md_autodetect_dev(rdev->bdev->bd_dev);
1577 kobject_put(&rdev->kobj);
1580 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1582 unbind_rdev_from_array(rdev);
1586 static void export_array(mddev_t *mddev)
1588 struct list_head *tmp;
1591 rdev_for_each(rdev, tmp, mddev) {
1596 kick_rdev_from_array(rdev);
1598 if (!list_empty(&mddev->disks))
1600 mddev->raid_disks = 0;
1601 mddev->major_version = 0;
1604 static void print_desc(mdp_disk_t *desc)
1606 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1607 desc->major,desc->minor,desc->raid_disk,desc->state);
1610 static void print_sb(mdp_super_t *sb)
1615 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1616 sb->major_version, sb->minor_version, sb->patch_version,
1617 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1619 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1620 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1621 sb->md_minor, sb->layout, sb->chunk_size);
1622 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1623 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1624 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1625 sb->failed_disks, sb->spare_disks,
1626 sb->sb_csum, (unsigned long)sb->events_lo);
1629 for (i = 0; i < MD_SB_DISKS; i++) {
1632 desc = sb->disks + i;
1633 if (desc->number || desc->major || desc->minor ||
1634 desc->raid_disk || (desc->state && (desc->state != 4))) {
1635 printk(" D %2d: ", i);
1639 printk(KERN_INFO "md: THIS: ");
1640 print_desc(&sb->this_disk);
1644 static void print_rdev(mdk_rdev_t *rdev)
1646 char b[BDEVNAME_SIZE];
1647 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1648 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1649 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1651 if (rdev->sb_loaded) {
1652 printk(KERN_INFO "md: rdev superblock:\n");
1653 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1655 printk(KERN_INFO "md: no rdev superblock!\n");
1658 static void md_print_devices(void)
1660 struct list_head *tmp, *tmp2;
1663 char b[BDEVNAME_SIZE];
1666 printk("md: **********************************\n");
1667 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1668 printk("md: **********************************\n");
1669 for_each_mddev(mddev, tmp) {
1672 bitmap_print_sb(mddev->bitmap);
1674 printk("%s: ", mdname(mddev));
1675 rdev_for_each(rdev, tmp2, mddev)
1676 printk("<%s>", bdevname(rdev->bdev,b));
1679 rdev_for_each(rdev, tmp2, mddev)
1682 printk("md: **********************************\n");
1687 static void sync_sbs(mddev_t * mddev, int nospares)
1689 /* Update each superblock (in-memory image), but
1690 * if we are allowed to, skip spares which already
1691 * have the right event counter, or have one earlier
1692 * (which would mean they aren't being marked as dirty
1693 * with the rest of the array)
1696 struct list_head *tmp;
1698 rdev_for_each(rdev, tmp, mddev) {
1699 if (rdev->sb_events == mddev->events ||
1701 rdev->raid_disk < 0 &&
1702 (rdev->sb_events&1)==0 &&
1703 rdev->sb_events+1 == mddev->events)) {
1704 /* Don't update this superblock */
1705 rdev->sb_loaded = 2;
1707 super_types[mddev->major_version].
1708 sync_super(mddev, rdev);
1709 rdev->sb_loaded = 1;
1714 static void md_update_sb(mddev_t * mddev, int force_change)
1716 struct list_head *tmp;
1721 if (mddev->external)
1724 spin_lock_irq(&mddev->write_lock);
1726 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1727 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1729 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1730 /* just a clean<-> dirty transition, possibly leave spares alone,
1731 * though if events isn't the right even/odd, we will have to do
1737 if (mddev->degraded)
1738 /* If the array is degraded, then skipping spares is both
1739 * dangerous and fairly pointless.
1740 * Dangerous because a device that was removed from the array
1741 * might have a event_count that still looks up-to-date,
1742 * so it can be re-added without a resync.
1743 * Pointless because if there are any spares to skip,
1744 * then a recovery will happen and soon that array won't
1745 * be degraded any more and the spare can go back to sleep then.
1749 sync_req = mddev->in_sync;
1750 mddev->utime = get_seconds();
1752 /* If this is just a dirty<->clean transition, and the array is clean
1753 * and 'events' is odd, we can roll back to the previous clean state */
1755 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1756 && (mddev->events & 1)
1757 && mddev->events != 1)
1760 /* otherwise we have to go forward and ... */
1762 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1763 /* .. if the array isn't clean, insist on an odd 'events' */
1764 if ((mddev->events&1)==0) {
1769 /* otherwise insist on an even 'events' (for clean states) */
1770 if ((mddev->events&1)) {
1777 if (!mddev->events) {
1779 * oops, this 64-bit counter should never wrap.
1780 * Either we are in around ~1 trillion A.C., assuming
1781 * 1 reboot per second, or we have a bug:
1788 * do not write anything to disk if using
1789 * nonpersistent superblocks
1791 if (!mddev->persistent) {
1792 if (!mddev->external)
1793 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1795 spin_unlock_irq(&mddev->write_lock);
1796 wake_up(&mddev->sb_wait);
1799 sync_sbs(mddev, nospares);
1800 spin_unlock_irq(&mddev->write_lock);
1803 "md: updating %s RAID superblock on device (in sync %d)\n",
1804 mdname(mddev),mddev->in_sync);
1806 bitmap_update_sb(mddev->bitmap);
1807 rdev_for_each(rdev, tmp, mddev) {
1808 char b[BDEVNAME_SIZE];
1809 dprintk(KERN_INFO "md: ");
1810 if (rdev->sb_loaded != 1)
1811 continue; /* no noise on spare devices */
1812 if (test_bit(Faulty, &rdev->flags))
1813 dprintk("(skipping faulty ");
1815 dprintk("%s ", bdevname(rdev->bdev,b));
1816 if (!test_bit(Faulty, &rdev->flags)) {
1817 md_super_write(mddev,rdev,
1818 rdev->sb_offset<<1, rdev->sb_size,
1820 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1821 bdevname(rdev->bdev,b),
1822 (unsigned long long)rdev->sb_offset);
1823 rdev->sb_events = mddev->events;
1827 if (mddev->level == LEVEL_MULTIPATH)
1828 /* only need to write one superblock... */
1831 md_super_wait(mddev);
1832 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1834 spin_lock_irq(&mddev->write_lock);
1835 if (mddev->in_sync != sync_req ||
1836 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1837 /* have to write it out again */
1838 spin_unlock_irq(&mddev->write_lock);
1841 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1842 spin_unlock_irq(&mddev->write_lock);
1843 wake_up(&mddev->sb_wait);
1847 /* words written to sysfs files may, or may not, be \n terminated.
1848 * We want to accept with case. For this we use cmd_match.
1850 static int cmd_match(const char *cmd, const char *str)
1852 /* See if cmd, written into a sysfs file, matches
1853 * str. They must either be the same, or cmd can
1854 * have a trailing newline
1856 while (*cmd && *str && *cmd == *str) {
1867 struct rdev_sysfs_entry {
1868 struct attribute attr;
1869 ssize_t (*show)(mdk_rdev_t *, char *);
1870 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1874 state_show(mdk_rdev_t *rdev, char *page)
1879 if (test_bit(Faulty, &rdev->flags)) {
1880 len+= sprintf(page+len, "%sfaulty",sep);
1883 if (test_bit(In_sync, &rdev->flags)) {
1884 len += sprintf(page+len, "%sin_sync",sep);
1887 if (test_bit(WriteMostly, &rdev->flags)) {
1888 len += sprintf(page+len, "%swrite_mostly",sep);
1891 if (test_bit(Blocked, &rdev->flags)) {
1892 len += sprintf(page+len, "%sblocked", sep);
1895 if (!test_bit(Faulty, &rdev->flags) &&
1896 !test_bit(In_sync, &rdev->flags)) {
1897 len += sprintf(page+len, "%sspare", sep);
1900 return len+sprintf(page+len, "\n");
1904 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1907 * faulty - simulates and error
1908 * remove - disconnects the device
1909 * writemostly - sets write_mostly
1910 * -writemostly - clears write_mostly
1911 * blocked - sets the Blocked flag
1912 * -blocked - clears the Blocked flag
1915 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1916 md_error(rdev->mddev, rdev);
1918 } else if (cmd_match(buf, "remove")) {
1919 if (rdev->raid_disk >= 0)
1922 mddev_t *mddev = rdev->mddev;
1923 kick_rdev_from_array(rdev);
1925 md_update_sb(mddev, 1);
1926 md_new_event(mddev);
1929 } else if (cmd_match(buf, "writemostly")) {
1930 set_bit(WriteMostly, &rdev->flags);
1932 } else if (cmd_match(buf, "-writemostly")) {
1933 clear_bit(WriteMostly, &rdev->flags);
1935 } else if (cmd_match(buf, "blocked")) {
1936 set_bit(Blocked, &rdev->flags);
1938 } else if (cmd_match(buf, "-blocked")) {
1939 clear_bit(Blocked, &rdev->flags);
1940 wake_up(&rdev->blocked_wait);
1941 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1942 md_wakeup_thread(rdev->mddev->thread);
1947 sysfs_notify(&rdev->kobj, NULL, "state");
1948 return err ? err : len;
1950 static struct rdev_sysfs_entry rdev_state =
1951 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1954 errors_show(mdk_rdev_t *rdev, char *page)
1956 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1960 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1963 unsigned long n = simple_strtoul(buf, &e, 10);
1964 if (*buf && (*e == 0 || *e == '\n')) {
1965 atomic_set(&rdev->corrected_errors, n);
1970 static struct rdev_sysfs_entry rdev_errors =
1971 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1974 slot_show(mdk_rdev_t *rdev, char *page)
1976 if (rdev->raid_disk < 0)
1977 return sprintf(page, "none\n");
1979 return sprintf(page, "%d\n", rdev->raid_disk);
1983 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1988 int slot = simple_strtoul(buf, &e, 10);
1989 if (strncmp(buf, "none", 4)==0)
1991 else if (e==buf || (*e && *e!= '\n'))
1993 if (rdev->mddev->pers && slot == -1) {
1994 /* Setting 'slot' on an active array requires also
1995 * updating the 'rd%d' link, and communicating
1996 * with the personality with ->hot_*_disk.
1997 * For now we only support removing
1998 * failed/spare devices. This normally happens automatically,
1999 * but not when the metadata is externally managed.
2001 if (rdev->raid_disk == -1)
2003 /* personality does all needed checks */
2004 if (rdev->mddev->pers->hot_add_disk == NULL)
2006 err = rdev->mddev->pers->
2007 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2010 sprintf(nm, "rd%d", rdev->raid_disk);
2011 sysfs_remove_link(&rdev->mddev->kobj, nm);
2012 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2013 md_wakeup_thread(rdev->mddev->thread);
2014 } else if (rdev->mddev->pers) {
2016 struct list_head *tmp;
2017 /* Activating a spare .. or possibly reactivating
2018 * if we every get bitmaps working here.
2021 if (rdev->raid_disk != -1)
2024 if (rdev->mddev->pers->hot_add_disk == NULL)
2027 rdev_for_each(rdev2, tmp, rdev->mddev)
2028 if (rdev2->raid_disk == slot)
2031 rdev->raid_disk = slot;
2032 if (test_bit(In_sync, &rdev->flags))
2033 rdev->saved_raid_disk = slot;
2035 rdev->saved_raid_disk = -1;
2036 err = rdev->mddev->pers->
2037 hot_add_disk(rdev->mddev, rdev);
2039 rdev->raid_disk = -1;
2042 sysfs_notify(&rdev->kobj, NULL, "state");
2043 sprintf(nm, "rd%d", rdev->raid_disk);
2044 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2046 "md: cannot register "
2048 nm, mdname(rdev->mddev));
2050 /* don't wakeup anyone, leave that to userspace. */
2052 if (slot >= rdev->mddev->raid_disks)
2054 rdev->raid_disk = slot;
2055 /* assume it is working */
2056 clear_bit(Faulty, &rdev->flags);
2057 clear_bit(WriteMostly, &rdev->flags);
2058 set_bit(In_sync, &rdev->flags);
2059 sysfs_notify(&rdev->kobj, NULL, "state");
2065 static struct rdev_sysfs_entry rdev_slot =
2066 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2069 offset_show(mdk_rdev_t *rdev, char *page)
2071 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2075 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2078 unsigned long long offset = simple_strtoull(buf, &e, 10);
2079 if (e==buf || (*e && *e != '\n'))
2081 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2083 if (rdev->size && rdev->mddev->external)
2084 /* Must set offset before size, so overlap checks
2087 rdev->data_offset = offset;
2091 static struct rdev_sysfs_entry rdev_offset =
2092 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2095 rdev_size_show(mdk_rdev_t *rdev, char *page)
2097 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2100 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2102 /* check if two start/length pairs overlap */
2111 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2114 unsigned long long size = simple_strtoull(buf, &e, 10);
2115 unsigned long long oldsize = rdev->size;
2116 mddev_t *my_mddev = rdev->mddev;
2118 if (e==buf || (*e && *e != '\n'))
2120 if (my_mddev->pers && rdev->raid_disk >= 0) {
2121 if (rdev->mddev->persistent) {
2122 size = super_types[rdev->mddev->major_version].
2123 rdev_size_change(rdev, size);
2127 size = (rdev->bdev->bd_inode->i_size >> 10);
2128 size -= rdev->data_offset/2;
2130 if (size < rdev->mddev->size)
2131 return -EINVAL; /* component must fit device */
2135 if (size > oldsize && rdev->mddev->external) {
2136 /* need to check that all other rdevs with the same ->bdev
2137 * do not overlap. We need to unlock the mddev to avoid
2138 * a deadlock. We have already changed rdev->size, and if
2139 * we have to change it back, we will have the lock again.
2143 struct list_head *tmp, *tmp2;
2145 mddev_unlock(my_mddev);
2146 for_each_mddev(mddev, tmp) {
2150 rdev_for_each(rdev2, tmp2, mddev)
2151 if (test_bit(AllReserved, &rdev2->flags) ||
2152 (rdev->bdev == rdev2->bdev &&
2154 overlaps(rdev->data_offset, rdev->size,
2155 rdev2->data_offset, rdev2->size))) {
2159 mddev_unlock(mddev);
2165 mddev_lock(my_mddev);
2167 /* Someone else could have slipped in a size
2168 * change here, but doing so is just silly.
2169 * We put oldsize back because we *know* it is
2170 * safe, and trust userspace not to race with
2173 rdev->size = oldsize;
2177 if (size < my_mddev->size || my_mddev->size == 0)
2178 my_mddev->size = size;
2182 static struct rdev_sysfs_entry rdev_size =
2183 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2185 static struct attribute *rdev_default_attrs[] = {
2194 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2196 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2197 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2198 mddev_t *mddev = rdev->mddev;
2204 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2206 if (rdev->mddev == NULL)
2209 rv = entry->show(rdev, page);
2210 mddev_unlock(mddev);
2216 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2217 const char *page, size_t length)
2219 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2220 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2222 mddev_t *mddev = rdev->mddev;
2226 if (!capable(CAP_SYS_ADMIN))
2228 rv = mddev ? mddev_lock(mddev): -EBUSY;
2230 if (rdev->mddev == NULL)
2233 rv = entry->store(rdev, page, length);
2234 mddev_unlock(mddev);
2239 static void rdev_free(struct kobject *ko)
2241 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2244 static struct sysfs_ops rdev_sysfs_ops = {
2245 .show = rdev_attr_show,
2246 .store = rdev_attr_store,
2248 static struct kobj_type rdev_ktype = {
2249 .release = rdev_free,
2250 .sysfs_ops = &rdev_sysfs_ops,
2251 .default_attrs = rdev_default_attrs,
2255 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2257 * mark the device faulty if:
2259 * - the device is nonexistent (zero size)
2260 * - the device has no valid superblock
2262 * a faulty rdev _never_ has rdev->sb set.
2264 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2266 char b[BDEVNAME_SIZE];
2271 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2273 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2274 return ERR_PTR(-ENOMEM);
2277 if ((err = alloc_disk_sb(rdev)))
2280 err = lock_rdev(rdev, newdev, super_format == -2);
2284 kobject_init(&rdev->kobj, &rdev_ktype);
2287 rdev->saved_raid_disk = -1;
2288 rdev->raid_disk = -1;
2290 rdev->data_offset = 0;
2291 rdev->sb_events = 0;
2292 atomic_set(&rdev->nr_pending, 0);
2293 atomic_set(&rdev->read_errors, 0);
2294 atomic_set(&rdev->corrected_errors, 0);
2296 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2299 "md: %s has zero or unknown size, marking faulty!\n",
2300 bdevname(rdev->bdev,b));
2305 if (super_format >= 0) {
2306 err = super_types[super_format].
2307 load_super(rdev, NULL, super_minor);
2308 if (err == -EINVAL) {
2310 "md: %s does not have a valid v%d.%d "
2311 "superblock, not importing!\n",
2312 bdevname(rdev->bdev,b),
2313 super_format, super_minor);
2318 "md: could not read %s's sb, not importing!\n",
2319 bdevname(rdev->bdev,b));
2324 INIT_LIST_HEAD(&rdev->same_set);
2325 init_waitqueue_head(&rdev->blocked_wait);
2330 if (rdev->sb_page) {
2336 return ERR_PTR(err);
2340 * Check a full RAID array for plausibility
2344 static void analyze_sbs(mddev_t * mddev)
2347 struct list_head *tmp;
2348 mdk_rdev_t *rdev, *freshest;
2349 char b[BDEVNAME_SIZE];
2352 rdev_for_each(rdev, tmp, mddev)
2353 switch (super_types[mddev->major_version].
2354 load_super(rdev, freshest, mddev->minor_version)) {
2362 "md: fatal superblock inconsistency in %s"
2363 " -- removing from array\n",
2364 bdevname(rdev->bdev,b));
2365 kick_rdev_from_array(rdev);
2369 super_types[mddev->major_version].
2370 validate_super(mddev, freshest);
2373 rdev_for_each(rdev, tmp, mddev) {
2374 if (rdev != freshest)
2375 if (super_types[mddev->major_version].
2376 validate_super(mddev, rdev)) {
2377 printk(KERN_WARNING "md: kicking non-fresh %s"
2379 bdevname(rdev->bdev,b));
2380 kick_rdev_from_array(rdev);
2383 if (mddev->level == LEVEL_MULTIPATH) {
2384 rdev->desc_nr = i++;
2385 rdev->raid_disk = rdev->desc_nr;
2386 set_bit(In_sync, &rdev->flags);
2387 } else if (rdev->raid_disk >= mddev->raid_disks) {
2388 rdev->raid_disk = -1;
2389 clear_bit(In_sync, &rdev->flags);
2395 if (mddev->recovery_cp != MaxSector &&
2397 printk(KERN_ERR "md: %s: raid array is not clean"
2398 " -- starting background reconstruction\n",
2404 safe_delay_show(mddev_t *mddev, char *page)
2406 int msec = (mddev->safemode_delay*1000)/HZ;
2407 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2410 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2418 /* remove a period, and count digits after it */
2419 if (len >= sizeof(buf))
2421 strlcpy(buf, cbuf, len);
2423 for (i=0; i<len; i++) {
2425 if (isdigit(buf[i])) {
2430 } else if (buf[i] == '.') {
2435 msec = simple_strtoul(buf, &e, 10);
2436 if (e == buf || (*e && *e != '\n'))
2438 msec = (msec * 1000) / scale;
2440 mddev->safemode_delay = 0;
2442 mddev->safemode_delay = (msec*HZ)/1000;
2443 if (mddev->safemode_delay == 0)
2444 mddev->safemode_delay = 1;
2448 static struct md_sysfs_entry md_safe_delay =
2449 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2452 level_show(mddev_t *mddev, char *page)
2454 struct mdk_personality *p = mddev->pers;
2456 return sprintf(page, "%s\n", p->name);
2457 else if (mddev->clevel[0])
2458 return sprintf(page, "%s\n", mddev->clevel);
2459 else if (mddev->level != LEVEL_NONE)
2460 return sprintf(page, "%d\n", mddev->level);
2466 level_store(mddev_t *mddev, const char *buf, size_t len)
2473 if (len >= sizeof(mddev->clevel))
2475 strncpy(mddev->clevel, buf, len);
2476 if (mddev->clevel[len-1] == '\n')
2478 mddev->clevel[len] = 0;
2479 mddev->level = LEVEL_NONE;
2483 static struct md_sysfs_entry md_level =
2484 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2488 layout_show(mddev_t *mddev, char *page)
2490 /* just a number, not meaningful for all levels */
2491 if (mddev->reshape_position != MaxSector &&
2492 mddev->layout != mddev->new_layout)
2493 return sprintf(page, "%d (%d)\n",
2494 mddev->new_layout, mddev->layout);
2495 return sprintf(page, "%d\n", mddev->layout);
2499 layout_store(mddev_t *mddev, const char *buf, size_t len)
2502 unsigned long n = simple_strtoul(buf, &e, 10);
2504 if (!*buf || (*e && *e != '\n'))
2509 if (mddev->reshape_position != MaxSector)
2510 mddev->new_layout = n;
2515 static struct md_sysfs_entry md_layout =
2516 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2520 raid_disks_show(mddev_t *mddev, char *page)
2522 if (mddev->raid_disks == 0)
2524 if (mddev->reshape_position != MaxSector &&
2525 mddev->delta_disks != 0)
2526 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2527 mddev->raid_disks - mddev->delta_disks);
2528 return sprintf(page, "%d\n", mddev->raid_disks);
2531 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2534 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2538 unsigned long n = simple_strtoul(buf, &e, 10);
2540 if (!*buf || (*e && *e != '\n'))
2544 rv = update_raid_disks(mddev, n);
2545 else if (mddev->reshape_position != MaxSector) {
2546 int olddisks = mddev->raid_disks - mddev->delta_disks;
2547 mddev->delta_disks = n - olddisks;
2548 mddev->raid_disks = n;
2550 mddev->raid_disks = n;
2551 return rv ? rv : len;
2553 static struct md_sysfs_entry md_raid_disks =
2554 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2557 chunk_size_show(mddev_t *mddev, char *page)
2559 if (mddev->reshape_position != MaxSector &&
2560 mddev->chunk_size != mddev->new_chunk)
2561 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2563 return sprintf(page, "%d\n", mddev->chunk_size);
2567 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2569 /* can only set chunk_size if array is not yet active */
2571 unsigned long n = simple_strtoul(buf, &e, 10);
2573 if (!*buf || (*e && *e != '\n'))
2578 else if (mddev->reshape_position != MaxSector)
2579 mddev->new_chunk = n;
2581 mddev->chunk_size = n;
2584 static struct md_sysfs_entry md_chunk_size =
2585 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2588 resync_start_show(mddev_t *mddev, char *page)
2590 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2594 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2597 unsigned long long n = simple_strtoull(buf, &e, 10);
2601 if (!*buf || (*e && *e != '\n'))
2604 mddev->recovery_cp = n;
2607 static struct md_sysfs_entry md_resync_start =
2608 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2611 * The array state can be:
2614 * No devices, no size, no level
2615 * Equivalent to STOP_ARRAY ioctl
2617 * May have some settings, but array is not active
2618 * all IO results in error
2619 * When written, doesn't tear down array, but just stops it
2620 * suspended (not supported yet)
2621 * All IO requests will block. The array can be reconfigured.
2622 * Writing this, if accepted, will block until array is quiescent
2624 * no resync can happen. no superblocks get written.
2625 * write requests fail
2627 * like readonly, but behaves like 'clean' on a write request.
2629 * clean - no pending writes, but otherwise active.
2630 * When written to inactive array, starts without resync
2631 * If a write request arrives then
2632 * if metadata is known, mark 'dirty' and switch to 'active'.
2633 * if not known, block and switch to write-pending
2634 * If written to an active array that has pending writes, then fails.
2636 * fully active: IO and resync can be happening.
2637 * When written to inactive array, starts with resync
2640 * clean, but writes are blocked waiting for 'active' to be written.
2643 * like active, but no writes have been seen for a while (100msec).
2646 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2647 write_pending, active_idle, bad_word};
2648 static char *array_states[] = {
2649 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2650 "write-pending", "active-idle", NULL };
2652 static int match_word(const char *word, char **list)
2655 for (n=0; list[n]; n++)
2656 if (cmd_match(word, list[n]))
2662 array_state_show(mddev_t *mddev, char *page)
2664 enum array_state st = inactive;
2677 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2679 else if (mddev->safemode)
2685 if (list_empty(&mddev->disks) &&
2686 mddev->raid_disks == 0 &&
2692 return sprintf(page, "%s\n", array_states[st]);
2695 static int do_md_stop(mddev_t * mddev, int ro);
2696 static int do_md_run(mddev_t * mddev);
2697 static int restart_array(mddev_t *mddev);
2700 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2703 enum array_state st = match_word(buf, array_states);
2708 /* stopping an active array */
2709 if (atomic_read(&mddev->active) > 1)
2711 err = do_md_stop(mddev, 0);
2714 /* stopping an active array */
2716 if (atomic_read(&mddev->active) > 1)
2718 err = do_md_stop(mddev, 2);
2720 err = 0; /* already inactive */
2723 break; /* not supported yet */
2726 err = do_md_stop(mddev, 1);
2729 set_disk_ro(mddev->gendisk, 1);
2730 err = do_md_run(mddev);
2736 err = do_md_stop(mddev, 1);
2738 err = restart_array(mddev);
2741 set_disk_ro(mddev->gendisk, 0);
2745 err = do_md_run(mddev);
2750 restart_array(mddev);
2751 spin_lock_irq(&mddev->write_lock);
2752 if (atomic_read(&mddev->writes_pending) == 0) {
2753 if (mddev->in_sync == 0) {
2755 if (mddev->safemode == 1)
2756 mddev->safemode = 0;
2757 if (mddev->persistent)
2758 set_bit(MD_CHANGE_CLEAN,
2764 spin_unlock_irq(&mddev->write_lock);
2767 mddev->recovery_cp = MaxSector;
2768 err = do_md_run(mddev);
2773 restart_array(mddev);
2774 if (mddev->external)
2775 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2776 wake_up(&mddev->sb_wait);
2780 set_disk_ro(mddev->gendisk, 0);
2781 err = do_md_run(mddev);
2786 /* these cannot be set */
2792 sysfs_notify(&mddev->kobj, NULL, "array_state");
2796 static struct md_sysfs_entry md_array_state =
2797 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2800 null_show(mddev_t *mddev, char *page)
2806 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2808 /* buf must be %d:%d\n? giving major and minor numbers */
2809 /* The new device is added to the array.
2810 * If the array has a persistent superblock, we read the
2811 * superblock to initialise info and check validity.
2812 * Otherwise, only checking done is that in bind_rdev_to_array,
2813 * which mainly checks size.
2816 int major = simple_strtoul(buf, &e, 10);
2822 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2824 minor = simple_strtoul(e+1, &e, 10);
2825 if (*e && *e != '\n')
2827 dev = MKDEV(major, minor);
2828 if (major != MAJOR(dev) ||
2829 minor != MINOR(dev))
2833 if (mddev->persistent) {
2834 rdev = md_import_device(dev, mddev->major_version,
2835 mddev->minor_version);
2836 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2837 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2838 mdk_rdev_t, same_set);
2839 err = super_types[mddev->major_version]
2840 .load_super(rdev, rdev0, mddev->minor_version);
2844 } else if (mddev->external)
2845 rdev = md_import_device(dev, -2, -1);
2847 rdev = md_import_device(dev, -1, -1);
2850 return PTR_ERR(rdev);
2851 err = bind_rdev_to_array(rdev, mddev);
2855 return err ? err : len;
2858 static struct md_sysfs_entry md_new_device =
2859 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2862 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2865 unsigned long chunk, end_chunk;
2869 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2871 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2872 if (buf == end) break;
2873 if (*end == '-') { /* range */
2875 end_chunk = simple_strtoul(buf, &end, 0);
2876 if (buf == end) break;
2878 if (*end && !isspace(*end)) break;
2879 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2881 while (isspace(*buf)) buf++;
2883 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2888 static struct md_sysfs_entry md_bitmap =
2889 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2892 size_show(mddev_t *mddev, char *page)
2894 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2897 static int update_size(mddev_t *mddev, unsigned long size);
2900 size_store(mddev_t *mddev, const char *buf, size_t len)
2902 /* If array is inactive, we can reduce the component size, but
2903 * not increase it (except from 0).
2904 * If array is active, we can try an on-line resize
2908 unsigned long long size = simple_strtoull(buf, &e, 10);
2909 if (!*buf || *buf == '\n' ||
2914 err = update_size(mddev, size);
2915 md_update_sb(mddev, 1);
2917 if (mddev->size == 0 ||
2923 return err ? err : len;
2926 static struct md_sysfs_entry md_size =
2927 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2932 * 'none' for arrays with no metadata (good luck...)
2933 * 'external' for arrays with externally managed metadata,
2934 * or N.M for internally known formats
2937 metadata_show(mddev_t *mddev, char *page)
2939 if (mddev->persistent)
2940 return sprintf(page, "%d.%d\n",
2941 mddev->major_version, mddev->minor_version);
2942 else if (mddev->external)
2943 return sprintf(page, "external:%s\n", mddev->metadata_type);
2945 return sprintf(page, "none\n");
2949 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2953 if (!list_empty(&mddev->disks))
2956 if (cmd_match(buf, "none")) {
2957 mddev->persistent = 0;
2958 mddev->external = 0;
2959 mddev->major_version = 0;
2960 mddev->minor_version = 90;
2963 if (strncmp(buf, "external:", 9) == 0) {
2964 size_t namelen = len-9;
2965 if (namelen >= sizeof(mddev->metadata_type))
2966 namelen = sizeof(mddev->metadata_type)-1;
2967 strncpy(mddev->metadata_type, buf+9, namelen);
2968 mddev->metadata_type[namelen] = 0;
2969 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2970 mddev->metadata_type[--namelen] = 0;
2971 mddev->persistent = 0;
2972 mddev->external = 1;
2973 mddev->major_version = 0;
2974 mddev->minor_version = 90;
2977 major = simple_strtoul(buf, &e, 10);
2978 if (e==buf || *e != '.')
2981 minor = simple_strtoul(buf, &e, 10);
2982 if (e==buf || (*e && *e != '\n') )
2984 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2986 mddev->major_version = major;
2987 mddev->minor_version = minor;
2988 mddev->persistent = 1;
2989 mddev->external = 0;
2993 static struct md_sysfs_entry md_metadata =
2994 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2997 action_show(mddev_t *mddev, char *page)
2999 char *type = "idle";
3000 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3001 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3002 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3004 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3005 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3007 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3011 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3014 return sprintf(page, "%s\n", type);
3018 action_store(mddev_t *mddev, const char *page, size_t len)
3020 if (!mddev->pers || !mddev->pers->sync_request)
3023 if (cmd_match(page, "idle")) {
3024 if (mddev->sync_thread) {
3025 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3026 md_unregister_thread(mddev->sync_thread);
3027 mddev->sync_thread = NULL;
3028 mddev->recovery = 0;
3030 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3031 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3033 else if (cmd_match(page, "resync"))
3034 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3035 else if (cmd_match(page, "recover")) {
3036 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3037 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3038 } else if (cmd_match(page, "reshape")) {
3040 if (mddev->pers->start_reshape == NULL)
3042 err = mddev->pers->start_reshape(mddev);
3045 sysfs_notify(&mddev->kobj, NULL, "degraded");
3047 if (cmd_match(page, "check"))
3048 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3049 else if (!cmd_match(page, "repair"))
3051 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3052 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3054 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3055 md_wakeup_thread(mddev->thread);
3056 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3061 mismatch_cnt_show(mddev_t *mddev, char *page)
3063 return sprintf(page, "%llu\n",
3064 (unsigned long long) mddev->resync_mismatches);
3067 static struct md_sysfs_entry md_scan_mode =
3068 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3071 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3074 sync_min_show(mddev_t *mddev, char *page)
3076 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3077 mddev->sync_speed_min ? "local": "system");
3081 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3085 if (strncmp(buf, "system", 6)==0) {
3086 mddev->sync_speed_min = 0;
3089 min = simple_strtoul(buf, &e, 10);
3090 if (buf == e || (*e && *e != '\n') || min <= 0)
3092 mddev->sync_speed_min = min;
3096 static struct md_sysfs_entry md_sync_min =
3097 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3100 sync_max_show(mddev_t *mddev, char *page)
3102 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3103 mddev->sync_speed_max ? "local": "system");
3107 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3111 if (strncmp(buf, "system", 6)==0) {
3112 mddev->sync_speed_max = 0;
3115 max = simple_strtoul(buf, &e, 10);
3116 if (buf == e || (*e && *e != '\n') || max <= 0)
3118 mddev->sync_speed_max = max;
3122 static struct md_sysfs_entry md_sync_max =
3123 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3126 degraded_show(mddev_t *mddev, char *page)
3128 return sprintf(page, "%d\n", mddev->degraded);
3130 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3133 sync_force_parallel_show(mddev_t *mddev, char *page)
3135 return sprintf(page, "%d\n", mddev->parallel_resync);
3139 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3143 if (strict_strtol(buf, 10, &n))
3146 if (n != 0 && n != 1)
3149 mddev->parallel_resync = n;
3151 if (mddev->sync_thread)
3152 wake_up(&resync_wait);
3157 /* force parallel resync, even with shared block devices */
3158 static struct md_sysfs_entry md_sync_force_parallel =
3159 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3160 sync_force_parallel_show, sync_force_parallel_store);
3163 sync_speed_show(mddev_t *mddev, char *page)
3165 unsigned long resync, dt, db;
3166 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3167 dt = (jiffies - mddev->resync_mark) / HZ;
3169 db = resync - mddev->resync_mark_cnt;
3170 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3173 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3176 sync_completed_show(mddev_t *mddev, char *page)
3178 unsigned long max_blocks, resync;
3180 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3181 max_blocks = mddev->resync_max_sectors;
3183 max_blocks = mddev->size << 1;
3185 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3186 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3189 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3192 min_sync_show(mddev_t *mddev, char *page)
3194 return sprintf(page, "%llu\n",
3195 (unsigned long long)mddev->resync_min);
3198 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3200 unsigned long long min;
3201 if (strict_strtoull(buf, 10, &min))
3203 if (min > mddev->resync_max)
3205 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3208 /* Must be a multiple of chunk_size */
3209 if (mddev->chunk_size) {
3210 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3213 mddev->resync_min = min;
3218 static struct md_sysfs_entry md_min_sync =
3219 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3222 max_sync_show(mddev_t *mddev, char *page)
3224 if (mddev->resync_max == MaxSector)
3225 return sprintf(page, "max\n");
3227 return sprintf(page, "%llu\n",
3228 (unsigned long long)mddev->resync_max);
3231 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3233 if (strncmp(buf, "max", 3) == 0)
3234 mddev->resync_max = MaxSector;
3236 unsigned long long max;
3237 if (strict_strtoull(buf, 10, &max))
3239 if (max < mddev->resync_min)
3241 if (max < mddev->resync_max &&
3242 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3245 /* Must be a multiple of chunk_size */
3246 if (mddev->chunk_size) {
3247 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3250 mddev->resync_max = max;
3252 wake_up(&mddev->recovery_wait);
3256 static struct md_sysfs_entry md_max_sync =
3257 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3260 suspend_lo_show(mddev_t *mddev, char *page)
3262 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3266 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3269 unsigned long long new = simple_strtoull(buf, &e, 10);
3271 if (mddev->pers->quiesce == NULL)
3273 if (buf == e || (*e && *e != '\n'))
3275 if (new >= mddev->suspend_hi ||
3276 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3277 mddev->suspend_lo = new;
3278 mddev->pers->quiesce(mddev, 2);
3283 static struct md_sysfs_entry md_suspend_lo =
3284 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3288 suspend_hi_show(mddev_t *mddev, char *page)
3290 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3294 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3297 unsigned long long new = simple_strtoull(buf, &e, 10);
3299 if (mddev->pers->quiesce == NULL)
3301 if (buf == e || (*e && *e != '\n'))
3303 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3304 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3305 mddev->suspend_hi = new;
3306 mddev->pers->quiesce(mddev, 1);
3307 mddev->pers->quiesce(mddev, 0);
3312 static struct md_sysfs_entry md_suspend_hi =
3313 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3316 reshape_position_show(mddev_t *mddev, char *page)
3318 if (mddev->reshape_position != MaxSector)
3319 return sprintf(page, "%llu\n",
3320 (unsigned long long)mddev->reshape_position);
3321 strcpy(page, "none\n");
3326 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3329 unsigned long long new = simple_strtoull(buf, &e, 10);
3332 if (buf == e || (*e && *e != '\n'))
3334 mddev->reshape_position = new;
3335 mddev->delta_disks = 0;
3336 mddev->new_level = mddev->level;
3337 mddev->new_layout = mddev->layout;
3338 mddev->new_chunk = mddev->chunk_size;
3342 static struct md_sysfs_entry md_reshape_position =
3343 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3344 reshape_position_store);
3347 static struct attribute *md_default_attrs[] = {
3350 &md_raid_disks.attr,
3351 &md_chunk_size.attr,
3353 &md_resync_start.attr,
3355 &md_new_device.attr,
3356 &md_safe_delay.attr,
3357 &md_array_state.attr,
3358 &md_reshape_position.attr,
3362 static struct attribute *md_redundancy_attrs[] = {
3364 &md_mismatches.attr,
3367 &md_sync_speed.attr,
3368 &md_sync_force_parallel.attr,
3369 &md_sync_completed.attr,
3372 &md_suspend_lo.attr,
3373 &md_suspend_hi.attr,
3378 static struct attribute_group md_redundancy_group = {
3380 .attrs = md_redundancy_attrs,
3385 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3387 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3388 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3393 rv = mddev_lock(mddev);
3395 rv = entry->show(mddev, page);
3396 mddev_unlock(mddev);
3402 md_attr_store(struct kobject *kobj, struct attribute *attr,
3403 const char *page, size_t length)
3405 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3406 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3411 if (!capable(CAP_SYS_ADMIN))
3413 rv = mddev_lock(mddev);
3415 rv = entry->store(mddev, page, length);
3416 mddev_unlock(mddev);
3421 static void md_free(struct kobject *ko)
3423 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3427 static struct sysfs_ops md_sysfs_ops = {
3428 .show = md_attr_show,
3429 .store = md_attr_store,
3431 static struct kobj_type md_ktype = {
3433 .sysfs_ops = &md_sysfs_ops,
3434 .default_attrs = md_default_attrs,
3439 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3441 static DEFINE_MUTEX(disks_mutex);
3442 mddev_t *mddev = mddev_find(dev);
3443 struct gendisk *disk;
3444 int partitioned = (MAJOR(dev) != MD_MAJOR);
3445 int shift = partitioned ? MdpMinorShift : 0;
3446 int unit = MINOR(dev) >> shift;
3452 mutex_lock(&disks_mutex);
3453 if (mddev->gendisk) {
3454 mutex_unlock(&disks_mutex);
3458 disk = alloc_disk(1 << shift);
3460 mutex_unlock(&disks_mutex);
3464 disk->major = MAJOR(dev);
3465 disk->first_minor = unit << shift;
3467 sprintf(disk->disk_name, "md_d%d", unit);
3469 sprintf(disk->disk_name, "md%d", unit);
3470 disk->fops = &md_fops;
3471 disk->private_data = mddev;
3472 disk->queue = mddev->queue;
3474 mddev->gendisk = disk;
3475 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3477 mutex_unlock(&disks_mutex);
3479 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3482 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3486 static void md_safemode_timeout(unsigned long data)
3488 mddev_t *mddev = (mddev_t *) data;
3490 if (!atomic_read(&mddev->writes_pending)) {
3491 mddev->safemode = 1;
3492 if (mddev->external)
3493 sysfs_notify(&mddev->kobj, NULL, "array_state");
3495 md_wakeup_thread(mddev->thread);
3498 static int start_dirty_degraded;
3500 static int do_md_run(mddev_t * mddev)
3504 struct list_head *tmp;
3506 struct gendisk *disk;
3507 struct mdk_personality *pers;
3508 char b[BDEVNAME_SIZE];
3510 if (list_empty(&mddev->disks))
3511 /* cannot run an array with no devices.. */
3518 * Analyze all RAID superblock(s)
3520 if (!mddev->raid_disks) {
3521 if (!mddev->persistent)
3526 chunk_size = mddev->chunk_size;
3529 if (chunk_size > MAX_CHUNK_SIZE) {
3530 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3531 chunk_size, MAX_CHUNK_SIZE);
3535 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3537 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3538 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3541 if (chunk_size < PAGE_SIZE) {
3542 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3543 chunk_size, PAGE_SIZE);
3547 /* devices must have minimum size of one chunk */
3548 rdev_for_each(rdev, tmp, mddev) {
3549 if (test_bit(Faulty, &rdev->flags))
3551 if (rdev->size < chunk_size / 1024) {
3553 "md: Dev %s smaller than chunk_size:"
3555 bdevname(rdev->bdev,b),
3556 (unsigned long long)rdev->size,
3564 if (mddev->level != LEVEL_NONE)
3565 request_module("md-level-%d", mddev->level);
3566 else if (mddev->clevel[0])
3567 request_module("md-%s", mddev->clevel);
3571 * Drop all container device buffers, from now on
3572 * the only valid external interface is through the md
3575 rdev_for_each(rdev, tmp, mddev) {
3576 if (test_bit(Faulty, &rdev->flags))
3578 sync_blockdev(rdev->bdev);
3579 invalidate_bdev(rdev->bdev);
3581 /* perform some consistency tests on the device.
3582 * We don't want the data to overlap the metadata,
3583 * Internal Bitmap issues has handled elsewhere.
3585 if (rdev->data_offset < rdev->sb_offset) {
3587 rdev->data_offset + mddev->size*2
3588 > rdev->sb_offset*2) {
3589 printk("md: %s: data overlaps metadata\n",
3594 if (rdev->sb_offset*2 + rdev->sb_size/512
3595 > rdev->data_offset) {
3596 printk("md: %s: metadata overlaps data\n",
3601 sysfs_notify(&rdev->kobj, NULL, "state");
3604 md_probe(mddev->unit, NULL, NULL);
3605 disk = mddev->gendisk;
3609 spin_lock(&pers_lock);
3610 pers = find_pers(mddev->level, mddev->clevel);
3611 if (!pers || !try_module_get(pers->owner)) {
3612 spin_unlock(&pers_lock);
3613 if (mddev->level != LEVEL_NONE)
3614 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3617 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3622 spin_unlock(&pers_lock);
3623 mddev->level = pers->level;
3624 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3626 if (mddev->reshape_position != MaxSector &&
3627 pers->start_reshape == NULL) {
3628 /* This personality cannot handle reshaping... */
3630 module_put(pers->owner);
3634 if (pers->sync_request) {
3635 /* Warn if this is a potentially silly
3638 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3640 struct list_head *tmp2;
3642 rdev_for_each(rdev, tmp, mddev) {
3643 rdev_for_each(rdev2, tmp2, mddev) {
3645 rdev->bdev->bd_contains ==
3646 rdev2->bdev->bd_contains) {
3648 "%s: WARNING: %s appears to be"
3649 " on the same physical disk as"
3652 bdevname(rdev->bdev,b),
3653 bdevname(rdev2->bdev,b2));
3660 "True protection against single-disk"
3661 " failure might be compromised.\n");
3664 mddev->recovery = 0;
3665 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3666 mddev->barriers_work = 1;
3667 mddev->ok_start_degraded = start_dirty_degraded;
3670 mddev->ro = 2; /* read-only, but switch on first write */
3672 err = mddev->pers->run(mddev);
3674 printk(KERN_ERR "md: pers->run() failed ...\n");
3675 else if (mddev->pers->sync_request) {
3676 err = bitmap_create(mddev);
3678 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3679 mdname(mddev), err);
3680 mddev->pers->stop(mddev);
3684 module_put(mddev->pers->owner);
3686 bitmap_destroy(mddev);
3689 if (mddev->pers->sync_request) {
3690 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3692 "md: cannot register extra attributes for %s\n",
3694 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3697 atomic_set(&mddev->writes_pending,0);
3698 mddev->safemode = 0;
3699 mddev->safemode_timer.function = md_safemode_timeout;
3700 mddev->safemode_timer.data = (unsigned long) mddev;
3701 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3704 rdev_for_each(rdev, tmp, mddev)
3705 if (rdev->raid_disk >= 0) {
3707 sprintf(nm, "rd%d", rdev->raid_disk);
3708 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3709 printk("md: cannot register %s for %s\n",
3713 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3716 md_update_sb(mddev, 0);
3718 set_capacity(disk, mddev->array_size<<1);
3720 /* If we call blk_queue_make_request here, it will
3721 * re-initialise max_sectors etc which may have been
3722 * refined inside -> run. So just set the bits we need to set.
3723 * Most initialisation happended when we called
3724 * blk_queue_make_request(..., md_fail_request)
3727 mddev->queue->queuedata = mddev;
3728 mddev->queue->make_request_fn = mddev->pers->make_request;
3730 /* If there is a partially-recovered drive we need to
3731 * start recovery here. If we leave it to md_check_recovery,
3732 * it will remove the drives and not do the right thing
3734 if (mddev->degraded && !mddev->sync_thread) {
3735 struct list_head *rtmp;
3737 rdev_for_each(rdev, rtmp, mddev)
3738 if (rdev->raid_disk >= 0 &&
3739 !test_bit(In_sync, &rdev->flags) &&
3740 !test_bit(Faulty, &rdev->flags))
3741 /* complete an interrupted recovery */
3743 if (spares && mddev->pers->sync_request) {
3744 mddev->recovery = 0;
3745 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3746 mddev->sync_thread = md_register_thread(md_do_sync,
3749 if (!mddev->sync_thread) {
3750 printk(KERN_ERR "%s: could not start resync"
3753 /* leave the spares where they are, it shouldn't hurt */
3754 mddev->recovery = 0;
3758 md_wakeup_thread(mddev->thread);
3759 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3762 md_new_event(mddev);
3763 sysfs_notify(&mddev->kobj, NULL, "array_state");
3764 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3765 sysfs_notify(&mddev->kobj, NULL, "degraded");
3766 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3770 static int restart_array(mddev_t *mddev)
3772 struct gendisk *disk = mddev->gendisk;
3776 * Complain if it has no devices
3779 if (list_empty(&mddev->disks))
3787 mddev->safemode = 0;
3789 set_disk_ro(disk, 0);
3791 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3794 * Kick recovery or resync if necessary
3796 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3797 md_wakeup_thread(mddev->thread);
3798 md_wakeup_thread(mddev->sync_thread);
3800 sysfs_notify(&mddev->kobj, NULL, "array_state");
3809 /* similar to deny_write_access, but accounts for our holding a reference
3810 * to the file ourselves */
3811 static int deny_bitmap_write_access(struct file * file)
3813 struct inode *inode = file->f_mapping->host;
3815 spin_lock(&inode->i_lock);
3816 if (atomic_read(&inode->i_writecount) > 1) {
3817 spin_unlock(&inode->i_lock);
3820 atomic_set(&inode->i_writecount, -1);
3821 spin_unlock(&inode->i_lock);
3826 static void restore_bitmap_write_access(struct file *file)
3828 struct inode *inode = file->f_mapping->host;
3830 spin_lock(&inode->i_lock);
3831 atomic_set(&inode->i_writecount, 1);
3832 spin_unlock(&inode->i_lock);
3836 * 0 - completely stop and dis-assemble array
3837 * 1 - switch to readonly
3838 * 2 - stop but do not disassemble array
3840 static int do_md_stop(mddev_t * mddev, int mode)
3843 struct gendisk *disk = mddev->gendisk;
3846 if (atomic_read(&mddev->active)>2) {
3847 printk("md: %s still in use.\n",mdname(mddev));
3851 if (mddev->sync_thread) {
3852 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3853 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3854 md_unregister_thread(mddev->sync_thread);
3855 mddev->sync_thread = NULL;
3858 del_timer_sync(&mddev->safemode_timer);
3860 invalidate_partition(disk, 0);
3863 case 1: /* readonly */
3869 case 0: /* disassemble */
3871 bitmap_flush(mddev);
3872 md_super_wait(mddev);
3874 set_disk_ro(disk, 0);
3875 blk_queue_make_request(mddev->queue, md_fail_request);
3876 mddev->pers->stop(mddev);
3877 mddev->queue->merge_bvec_fn = NULL;
3878 mddev->queue->unplug_fn = NULL;
3879 mddev->queue->backing_dev_info.congested_fn = NULL;
3880 if (mddev->pers->sync_request)
3881 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3883 module_put(mddev->pers->owner);
3885 /* tell userspace to handle 'inactive' */
3886 sysfs_notify(&mddev->kobj, NULL, "array_state");
3888 set_capacity(disk, 0);
3894 if (!mddev->in_sync || mddev->flags) {
3895 /* mark array as shutdown cleanly */
3897 md_update_sb(mddev, 1);
3900 set_disk_ro(disk, 1);
3901 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3905 * Free resources if final stop
3909 struct list_head *tmp;
3911 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3913 bitmap_destroy(mddev);
3914 if (mddev->bitmap_file) {
3915 restore_bitmap_write_access(mddev->bitmap_file);
3916 fput(mddev->bitmap_file);
3917 mddev->bitmap_file = NULL;
3919 mddev->bitmap_offset = 0;
3921 rdev_for_each(rdev, tmp, mddev)
3922 if (rdev->raid_disk >= 0) {
3924 sprintf(nm, "rd%d", rdev->raid_disk);
3925 sysfs_remove_link(&mddev->kobj, nm);
3928 /* make sure all md_delayed_delete calls have finished */
3929 flush_scheduled_work();
3931 export_array(mddev);
3933 mddev->array_size = 0;
3935 mddev->raid_disks = 0;
3936 mddev->recovery_cp = 0;
3937 mddev->resync_min = 0;
3938 mddev->resync_max = MaxSector;
3939 mddev->reshape_position = MaxSector;
3940 mddev->external = 0;
3941 mddev->persistent = 0;
3942 mddev->level = LEVEL_NONE;
3943 mddev->clevel[0] = 0;
3946 mddev->metadata_type[0] = 0;
3947 mddev->chunk_size = 0;
3948 mddev->ctime = mddev->utime = 0;
3950 mddev->max_disks = 0;
3952 mddev->delta_disks = 0;
3953 mddev->new_level = LEVEL_NONE;
3954 mddev->new_layout = 0;
3955 mddev->new_chunk = 0;
3956 mddev->curr_resync = 0;
3957 mddev->resync_mismatches = 0;
3958 mddev->suspend_lo = mddev->suspend_hi = 0;
3959 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3960 mddev->recovery = 0;
3963 mddev->degraded = 0;
3964 mddev->barriers_work = 0;
3965 mddev->safemode = 0;
3967 } else if (mddev->pers)
3968 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3971 md_new_event(mddev);
3972 sysfs_notify(&mddev->kobj, NULL, "array_state");
3978 static void autorun_array(mddev_t *mddev)
3981 struct list_head *tmp;
3984 if (list_empty(&mddev->disks))
3987 printk(KERN_INFO "md: running: ");
3989 rdev_for_each(rdev, tmp, mddev) {
3990 char b[BDEVNAME_SIZE];
3991 printk("<%s>", bdevname(rdev->bdev,b));
3995 err = do_md_run (mddev);
3997 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3998 do_md_stop (mddev, 0);
4003 * lets try to run arrays based on all disks that have arrived
4004 * until now. (those are in pending_raid_disks)
4006 * the method: pick the first pending disk, collect all disks with
4007 * the same UUID, remove all from the pending list and put them into
4008 * the 'same_array' list. Then order this list based on superblock
4009 * update time (freshest comes first), kick out 'old' disks and
4010 * compare superblocks. If everything's fine then run it.
4012 * If "unit" is allocated, then bump its reference count
4014 static void autorun_devices(int part)
4016 struct list_head *tmp;
4017 mdk_rdev_t *rdev0, *rdev;
4019 char b[BDEVNAME_SIZE];
4021 printk(KERN_INFO "md: autorun ...\n");
4022 while (!list_empty(&pending_raid_disks)) {
4025 LIST_HEAD(candidates);
4026 rdev0 = list_entry(pending_raid_disks.next,
4027 mdk_rdev_t, same_set);
4029 printk(KERN_INFO "md: considering %s ...\n",
4030 bdevname(rdev0->bdev,b));
4031 INIT_LIST_HEAD(&candidates);
4032 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4033 if (super_90_load(rdev, rdev0, 0) >= 0) {
4034 printk(KERN_INFO "md: adding %s ...\n",
4035 bdevname(rdev->bdev,b));
4036 list_move(&rdev->same_set, &candidates);
4039 * now we have a set of devices, with all of them having
4040 * mostly sane superblocks. It's time to allocate the
4044 dev = MKDEV(mdp_major,
4045 rdev0->preferred_minor << MdpMinorShift);
4046 unit = MINOR(dev) >> MdpMinorShift;
4048 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4051 if (rdev0->preferred_minor != unit) {
4052 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4053 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4057 md_probe(dev, NULL, NULL);
4058 mddev = mddev_find(dev);
4059 if (!mddev || !mddev->gendisk) {
4063 "md: cannot allocate memory for md drive.\n");
4066 if (mddev_lock(mddev))
4067 printk(KERN_WARNING "md: %s locked, cannot run\n",
4069 else if (mddev->raid_disks || mddev->major_version
4070 || !list_empty(&mddev->disks)) {
4072 "md: %s already running, cannot run %s\n",
4073 mdname(mddev), bdevname(rdev0->bdev,b));
4074 mddev_unlock(mddev);
4076 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4077 mddev->persistent = 1;
4078 rdev_for_each_list(rdev, tmp, candidates) {
4079 list_del_init(&rdev->same_set);
4080 if (bind_rdev_to_array(rdev, mddev))
4083 autorun_array(mddev);
4084 mddev_unlock(mddev);
4086 /* on success, candidates will be empty, on error
4089 rdev_for_each_list(rdev, tmp, candidates)
4093 printk(KERN_INFO "md: ... autorun DONE.\n");
4095 #endif /* !MODULE */
4097 static int get_version(void __user * arg)
4101 ver.major = MD_MAJOR_VERSION;
4102 ver.minor = MD_MINOR_VERSION;
4103 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4105 if (copy_to_user(arg, &ver, sizeof(ver)))
4111 static int get_array_info(mddev_t * mddev, void __user * arg)
4113 mdu_array_info_t info;
4114 int nr,working,active,failed,spare;
4116 struct list_head *tmp;
4118 nr=working=active=failed=spare=0;
4119 rdev_for_each(rdev, tmp, mddev) {
4121 if (test_bit(Faulty, &rdev->flags))
4125 if (test_bit(In_sync, &rdev->flags))
4132 info.major_version = mddev->major_version;
4133 info.minor_version = mddev->minor_version;
4134 info.patch_version = MD_PATCHLEVEL_VERSION;
4135 info.ctime = mddev->ctime;
4136 info.level = mddev->level;
4137 info.size = mddev->size;
4138 if (info.size != mddev->size) /* overflow */
4141 info.raid_disks = mddev->raid_disks;
4142 info.md_minor = mddev->md_minor;
4143 info.not_persistent= !mddev->persistent;
4145 info.utime = mddev->utime;
4148 info.state = (1<<MD_SB_CLEAN);
4149 if (mddev->bitmap && mddev->bitmap_offset)
4150 info.state = (1<<MD_SB_BITMAP_PRESENT);
4151 info.active_disks = active;
4152 info.working_disks = working;
4153 info.failed_disks = failed;
4154 info.spare_disks = spare;
4156 info.layout = mddev->layout;
4157 info.chunk_size = mddev->chunk_size;
4159 if (copy_to_user(arg, &info, sizeof(info)))
4165 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4167 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4168 char *ptr, *buf = NULL;
4171 if (md_allow_write(mddev))
4172 file = kmalloc(sizeof(*file), GFP_NOIO);
4174 file = kmalloc(sizeof(*file), GFP_KERNEL);
4179 /* bitmap disabled, zero the first byte and copy out */
4180 if (!mddev->bitmap || !mddev->bitmap->file) {
4181 file->pathname[0] = '\0';
4185 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4189 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4193 strcpy(file->pathname, ptr);
4197 if (copy_to_user(arg, file, sizeof(*file)))
4205 static int get_disk_info(mddev_t * mddev, void __user * arg)
4207 mdu_disk_info_t info;
4211 if (copy_from_user(&info, arg, sizeof(info)))
4216 rdev = find_rdev_nr(mddev, nr);
4218 info.major = MAJOR(rdev->bdev->bd_dev);
4219 info.minor = MINOR(rdev->bdev->bd_dev);
4220 info.raid_disk = rdev->raid_disk;
4222 if (test_bit(Faulty, &rdev->flags))
4223 info.state |= (1<<MD_DISK_FAULTY);
4224 else if (test_bit(In_sync, &rdev->flags)) {
4225 info.state |= (1<<MD_DISK_ACTIVE);
4226 info.state |= (1<<MD_DISK_SYNC);
4228 if (test_bit(WriteMostly, &rdev->flags))
4229 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4231 info.major = info.minor = 0;
4232 info.raid_disk = -1;
4233 info.state = (1<<MD_DISK_REMOVED);
4236 if (copy_to_user(arg, &info, sizeof(info)))
4242 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4244 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4246 dev_t dev = MKDEV(info->major,info->minor);
4248 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4251 if (!mddev->raid_disks) {
4253 /* expecting a device which has a superblock */
4254 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4257 "md: md_import_device returned %ld\n",
4259 return PTR_ERR(rdev);
4261 if (!list_empty(&mddev->disks)) {
4262 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4263 mdk_rdev_t, same_set);
4264 int err = super_types[mddev->major_version]
4265 .load_super(rdev, rdev0, mddev->minor_version);
4268 "md: %s has different UUID to %s\n",
4269 bdevname(rdev->bdev,b),
4270 bdevname(rdev0->bdev,b2));
4275 err = bind_rdev_to_array(rdev, mddev);
4282 * add_new_disk can be used once the array is assembled
4283 * to add "hot spares". They must already have a superblock
4288 if (!mddev->pers->hot_add_disk) {
4290 "%s: personality does not support diskops!\n",
4294 if (mddev->persistent)
4295 rdev = md_import_device(dev, mddev->major_version,
4296 mddev->minor_version);
4298 rdev = md_import_device(dev, -1, -1);
4301 "md: md_import_device returned %ld\n",
4303 return PTR_ERR(rdev);
4305 /* set save_raid_disk if appropriate */
4306 if (!mddev->persistent) {
4307 if (info->state & (1<<MD_DISK_SYNC) &&
4308 info->raid_disk < mddev->raid_disks)
4309 rdev->raid_disk = info->raid_disk;
4311 rdev->raid_disk = -1;
4313 super_types[mddev->major_version].
4314 validate_super(mddev, rdev);
4315 rdev->saved_raid_disk = rdev->raid_disk;
4317 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4318 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4319 set_bit(WriteMostly, &rdev->flags);
4321 rdev->raid_disk = -1;
4322 err = bind_rdev_to_array(rdev, mddev);
4323 if (!err && !mddev->pers->hot_remove_disk) {
4324 /* If there is hot_add_disk but no hot_remove_disk
4325 * then added disks for geometry changes,
4326 * and should be added immediately.
4328 super_types[mddev->major_version].
4329 validate_super(mddev, rdev);
4330 err = mddev->pers->hot_add_disk(mddev, rdev);
4332 unbind_rdev_from_array(rdev);
4337 sysfs_notify(&rdev->kobj, NULL, "state");
4339 md_update_sb(mddev, 1);
4340 if (mddev->degraded)
4341 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4342 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4343 md_wakeup_thread(mddev->thread);
4347 /* otherwise, add_new_disk is only allowed
4348 * for major_version==0 superblocks
4350 if (mddev->major_version != 0) {
4351 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4356 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4358 rdev = md_import_device (dev, -1, 0);
4361 "md: error, md_import_device() returned %ld\n",
4363 return PTR_ERR(rdev);
4365 rdev->desc_nr = info->number;
4366 if (info->raid_disk < mddev->raid_disks)
4367 rdev->raid_disk = info->raid_disk;
4369 rdev->raid_disk = -1;
4371 if (rdev->raid_disk < mddev->raid_disks)
4372 if (info->state & (1<<MD_DISK_SYNC))
4373 set_bit(In_sync, &rdev->flags);
4375 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4376 set_bit(WriteMostly, &rdev->flags);
4378 if (!mddev->persistent) {
4379 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4380 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4382 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4383 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4385 err = bind_rdev_to_array(rdev, mddev);
4395 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4397 char b[BDEVNAME_SIZE];
4400 rdev = find_rdev(mddev, dev);
4404 if (rdev->raid_disk >= 0)
4407 kick_rdev_from_array(rdev);
4408 md_update_sb(mddev, 1);
4409 md_new_event(mddev);
4413 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4414 bdevname(rdev->bdev,b), mdname(mddev));
4418 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4420 char b[BDEVNAME_SIZE];
4428 if (mddev->major_version != 0) {
4429 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4430 " version-0 superblocks.\n",
4434 if (!mddev->pers->hot_add_disk) {
4436 "%s: personality does not support diskops!\n",
4441 rdev = md_import_device (dev, -1, 0);
4444 "md: error, md_import_device() returned %ld\n",
4449 if (mddev->persistent)
4450 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4453 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4455 size = calc_dev_size(rdev, mddev->chunk_size);
4458 if (test_bit(Faulty, &rdev->flags)) {
4460 "md: can not hot-add faulty %s disk to %s!\n",
4461 bdevname(rdev->bdev,b), mdname(mddev));
4465 clear_bit(In_sync, &rdev->flags);
4467 rdev->saved_raid_disk = -1;
4468 err = bind_rdev_to_array(rdev, mddev);
4473 * The rest should better be atomic, we can have disk failures
4474 * noticed in interrupt contexts ...
4477 if (rdev->desc_nr == mddev->max_disks) {
4478 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4481 goto abort_unbind_export;
4484 rdev->raid_disk = -1;
4486 md_update_sb(mddev, 1);
4489 * Kick recovery, maybe this spare has to be added to the
4490 * array immediately.
4492 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4493 md_wakeup_thread(mddev->thread);
4494 md_new_event(mddev);
4497 abort_unbind_export:
4498 unbind_rdev_from_array(rdev);
4505 static int set_bitmap_file(mddev_t *mddev, int fd)
4510 if (!mddev->pers->quiesce)
4512 if (mddev->recovery || mddev->sync_thread)
4514 /* we should be able to change the bitmap.. */
4520 return -EEXIST; /* cannot add when bitmap is present */
4521 mddev->bitmap_file = fget(fd);
4523 if (mddev->bitmap_file == NULL) {
4524 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4529 err = deny_bitmap_write_access(mddev->bitmap_file);
4531 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4533 fput(mddev->bitmap_file);
4534 mddev->bitmap_file = NULL;
4537 mddev->bitmap_offset = 0; /* file overrides offset */
4538 } else if (mddev->bitmap == NULL)
4539 return -ENOENT; /* cannot remove what isn't there */
4542 mddev->pers->quiesce(mddev, 1);
4544 err = bitmap_create(mddev);
4545 if (fd < 0 || err) {
4546 bitmap_destroy(mddev);
4547 fd = -1; /* make sure to put the file */
4549 mddev->pers->quiesce(mddev, 0);
4552 if (mddev->bitmap_file) {
4553 restore_bitmap_write_access(mddev->bitmap_file);
4554 fput(mddev->bitmap_file);
4556 mddev->bitmap_file = NULL;
4563 * set_array_info is used two different ways
4564 * The original usage is when creating a new array.
4565 * In this usage, raid_disks is > 0 and it together with
4566 * level, size, not_persistent,layout,chunksize determine the
4567 * shape of the array.
4568 * This will always create an array with a type-0.90.0 superblock.
4569 * The newer usage is when assembling an array.
4570 * In this case raid_disks will be 0, and the major_version field is
4571 * use to determine which style super-blocks are to be found on the devices.
4572 * The minor and patch _version numbers are also kept incase the
4573 * super_block handler wishes to interpret them.
4575 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4578 if (info->raid_disks == 0) {
4579 /* just setting version number for superblock loading */
4580 if (info->major_version < 0 ||
4581 info->major_version >= ARRAY_SIZE(super_types) ||
4582 super_types[info->major_version].name == NULL) {
4583 /* maybe try to auto-load a module? */
4585 "md: superblock version %d not known\n",
4586 info->major_version);
4589 mddev->major_version = info->major_version;
4590 mddev->minor_version = info->minor_version;
4591 mddev->patch_version = info->patch_version;
4592 mddev->persistent = !info->not_persistent;
4595 mddev->major_version = MD_MAJOR_VERSION;
4596 mddev->minor_version = MD_MINOR_VERSION;
4597 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4598 mddev->ctime = get_seconds();
4600 mddev->level = info->level;
4601 mddev->clevel[0] = 0;
4602 mddev->size = info->size;
4603 mddev->raid_disks = info->raid_disks;
4604 /* don't set md_minor, it is determined by which /dev/md* was
4607 if (info->state & (1<<MD_SB_CLEAN))
4608 mddev->recovery_cp = MaxSector;
4610 mddev->recovery_cp = 0;
4611 mddev->persistent = ! info->not_persistent;
4612 mddev->external = 0;
4614 mddev->layout = info->layout;
4615 mddev->chunk_size = info->chunk_size;
4617 mddev->max_disks = MD_SB_DISKS;
4619 if (mddev->persistent)
4621 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4623 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4624 mddev->bitmap_offset = 0;
4626 mddev->reshape_position = MaxSector;
4629 * Generate a 128 bit UUID
4631 get_random_bytes(mddev->uuid, 16);
4633 mddev->new_level = mddev->level;
4634 mddev->new_chunk = mddev->chunk_size;
4635 mddev->new_layout = mddev->layout;
4636 mddev->delta_disks = 0;
4641 static int update_size(mddev_t *mddev, unsigned long size)
4645 struct list_head *tmp;
4646 int fit = (size == 0);
4648 if (mddev->pers->resize == NULL)
4650 /* The "size" is the amount of each device that is used.
4651 * This can only make sense for arrays with redundancy.
4652 * linear and raid0 always use whatever space is available
4653 * We can only consider changing the size if no resync
4654 * or reconstruction is happening, and if the new size
4655 * is acceptable. It must fit before the sb_offset or,
4656 * if that is <data_offset, it must fit before the
4657 * size of each device.
4658 * If size is zero, we find the largest size that fits.
4660 if (mddev->sync_thread)
4662 rdev_for_each(rdev, tmp, mddev) {
4664 avail = rdev->size * 2;
4666 if (fit && (size == 0 || size > avail/2))
4668 if (avail < ((sector_t)size << 1))
4671 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4673 struct block_device *bdev;
4675 bdev = bdget_disk(mddev->gendisk, 0);
4677 mutex_lock(&bdev->bd_inode->i_mutex);
4678 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4679 mutex_unlock(&bdev->bd_inode->i_mutex);
4686 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4689 /* change the number of raid disks */
4690 if (mddev->pers->check_reshape == NULL)
4692 if (raid_disks <= 0 ||
4693 raid_disks >= mddev->max_disks)
4695 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4697 mddev->delta_disks = raid_disks - mddev->raid_disks;
4699 rv = mddev->pers->check_reshape(mddev);
4705 * update_array_info is used to change the configuration of an
4707 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4708 * fields in the info are checked against the array.
4709 * Any differences that cannot be handled will cause an error.
4710 * Normally, only one change can be managed at a time.
4712 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4718 /* calculate expected state,ignoring low bits */
4719 if (mddev->bitmap && mddev->bitmap_offset)
4720 state |= (1 << MD_SB_BITMAP_PRESENT);
4722 if (mddev->major_version != info->major_version ||
4723 mddev->minor_version != info->minor_version ||
4724 /* mddev->patch_version != info->patch_version || */
4725 mddev->ctime != info->ctime ||
4726 mddev->level != info->level ||
4727 /* mddev->layout != info->layout || */
4728 !mddev->persistent != info->not_persistent||
4729 mddev->chunk_size != info->chunk_size ||
4730 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4731 ((state^info->state) & 0xfffffe00)
4734 /* Check there is only one change */
4735 if (info->size >= 0 && mddev->size != info->size) cnt++;
4736 if (mddev->raid_disks != info->raid_disks) cnt++;
4737 if (mddev->layout != info->layout) cnt++;
4738 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4739 if (cnt == 0) return 0;
4740 if (cnt > 1) return -EINVAL;
4742 if (mddev->layout != info->layout) {
4744 * we don't need to do anything at the md level, the
4745 * personality will take care of it all.
4747 if (mddev->pers->reconfig == NULL)
4750 return mddev->pers->reconfig(mddev, info->layout, -1);
4752 if (info->size >= 0 && mddev->size != info->size)
4753 rv = update_size(mddev, info->size);
4755 if (mddev->raid_disks != info->raid_disks)
4756 rv = update_raid_disks(mddev, info->raid_disks);
4758 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4759 if (mddev->pers->quiesce == NULL)
4761 if (mddev->recovery || mddev->sync_thread)
4763 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4764 /* add the bitmap */
4767 if (mddev->default_bitmap_offset == 0)
4769 mddev->bitmap_offset = mddev->default_bitmap_offset;
4770 mddev->pers->quiesce(mddev, 1);
4771 rv = bitmap_create(mddev);
4773 bitmap_destroy(mddev);
4774 mddev->pers->quiesce(mddev, 0);
4776 /* remove the bitmap */
4779 if (mddev->bitmap->file)
4781 mddev->pers->quiesce(mddev, 1);
4782 bitmap_destroy(mddev);
4783 mddev->pers->quiesce(mddev, 0);
4784 mddev->bitmap_offset = 0;
4787 md_update_sb(mddev, 1);
4791 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4795 if (mddev->pers == NULL)
4798 rdev = find_rdev(mddev, dev);
4802 md_error(mddev, rdev);
4807 * We have a problem here : there is no easy way to give a CHS
4808 * virtual geometry. We currently pretend that we have a 2 heads
4809 * 4 sectors (with a BIG number of cylinders...). This drives
4810 * dosfs just mad... ;-)
4812 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4814 mddev_t *mddev = bdev->bd_disk->private_data;
4818 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4822 static int md_ioctl(struct inode *inode, struct file *file,
4823 unsigned int cmd, unsigned long arg)
4826 void __user *argp = (void __user *)arg;
4827 mddev_t *mddev = NULL;
4829 if (!capable(CAP_SYS_ADMIN))
4833 * Commands dealing with the RAID driver but not any
4839 err = get_version(argp);
4842 case PRINT_RAID_DEBUG:
4850 autostart_arrays(arg);
4857 * Commands creating/starting a new array:
4860 mddev = inode->i_bdev->bd_disk->private_data;
4867 err = mddev_lock(mddev);
4870 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4877 case SET_ARRAY_INFO:
4879 mdu_array_info_t info;
4881 memset(&info, 0, sizeof(info));
4882 else if (copy_from_user(&info, argp, sizeof(info))) {
4887 err = update_array_info(mddev, &info);
4889 printk(KERN_WARNING "md: couldn't update"
4890 " array info. %d\n", err);
4895 if (!list_empty(&mddev->disks)) {
4897 "md: array %s already has disks!\n",
4902 if (mddev->raid_disks) {
4904 "md: array %s already initialised!\n",
4909 err = set_array_info(mddev, &info);
4911 printk(KERN_WARNING "md: couldn't set"
4912 " array info. %d\n", err);
4922 * Commands querying/configuring an existing array:
4924 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4925 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4926 if ((!mddev->raid_disks && !mddev->external)
4927 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4928 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4929 && cmd != GET_BITMAP_FILE) {
4935 * Commands even a read-only array can execute:
4939 case GET_ARRAY_INFO:
4940 err = get_array_info(mddev, argp);
4943 case GET_BITMAP_FILE:
4944 err = get_bitmap_file(mddev, argp);
4948 err = get_disk_info(mddev, argp);
4951 case RESTART_ARRAY_RW:
4952 err = restart_array(mddev);
4956 err = do_md_stop (mddev, 0);
4960 err = do_md_stop (mddev, 1);
4966 * The remaining ioctls are changing the state of the
4967 * superblock, so we do not allow them on read-only arrays.
4968 * However non-MD ioctls (e.g. get-size) will still come through
4969 * here and hit the 'default' below, so only disallow
4970 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4972 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4973 if (mddev->ro == 2) {
4975 sysfs_notify(&mddev->kobj, NULL, "array_state");
4976 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4977 md_wakeup_thread(mddev->thread);
4988 mdu_disk_info_t info;
4989 if (copy_from_user(&info, argp, sizeof(info)))
4992 err = add_new_disk(mddev, &info);
4996 case HOT_REMOVE_DISK:
4997 err = hot_remove_disk(mddev, new_decode_dev(arg));
5001 err = hot_add_disk(mddev, new_decode_dev(arg));
5004 case SET_DISK_FAULTY:
5005 err = set_disk_faulty(mddev, new_decode_dev(arg));
5009 err = do_md_run (mddev);
5012 case SET_BITMAP_FILE:
5013 err = set_bitmap_file(mddev, (int)arg);
5023 mddev_unlock(mddev);
5033 static int md_open(struct inode *inode, struct file *file)
5036 * Succeed if we can lock the mddev, which confirms that
5037 * it isn't being stopped right now.
5039 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5042 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5047 mddev_unlock(mddev);
5049 check_disk_change(inode->i_bdev);
5054 static int md_release(struct inode *inode, struct file * file)
5056 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5064 static int md_media_changed(struct gendisk *disk)
5066 mddev_t *mddev = disk->private_data;
5068 return mddev->changed;
5071 static int md_revalidate(struct gendisk *disk)
5073 mddev_t *mddev = disk->private_data;
5078 static struct block_device_operations md_fops =
5080 .owner = THIS_MODULE,
5082 .release = md_release,
5084 .getgeo = md_getgeo,
5085 .media_changed = md_media_changed,
5086 .revalidate_disk= md_revalidate,
5089 static int md_thread(void * arg)
5091 mdk_thread_t *thread = arg;
5094 * md_thread is a 'system-thread', it's priority should be very
5095 * high. We avoid resource deadlocks individually in each
5096 * raid personality. (RAID5 does preallocation) We also use RR and
5097 * the very same RT priority as kswapd, thus we will never get
5098 * into a priority inversion deadlock.
5100 * we definitely have to have equal or higher priority than
5101 * bdflush, otherwise bdflush will deadlock if there are too
5102 * many dirty RAID5 blocks.
5105 allow_signal(SIGKILL);
5106 while (!kthread_should_stop()) {
5108 /* We need to wait INTERRUPTIBLE so that
5109 * we don't add to the load-average.
5110 * That means we need to be sure no signals are
5113 if (signal_pending(current))
5114 flush_signals(current);
5116 wait_event_interruptible_timeout
5118 test_bit(THREAD_WAKEUP, &thread->flags)
5119 || kthread_should_stop(),
5122 clear_bit(THREAD_WAKEUP, &thread->flags);
5124 thread->run(thread->mddev);
5130 void md_wakeup_thread(mdk_thread_t *thread)
5133 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5134 set_bit(THREAD_WAKEUP, &thread->flags);
5135 wake_up(&thread->wqueue);
5139 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5142 mdk_thread_t *thread;
5144 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5148 init_waitqueue_head(&thread->wqueue);
5151 thread->mddev = mddev;
5152 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5153 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5154 if (IS_ERR(thread->tsk)) {
5161 void md_unregister_thread(mdk_thread_t *thread)
5163 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5165 kthread_stop(thread->tsk);
5169 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5176 if (!rdev || test_bit(Faulty, &rdev->flags))
5179 if (mddev->external)
5180 set_bit(Blocked, &rdev->flags);
5182 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5184 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5185 __builtin_return_address(0),__builtin_return_address(1),
5186 __builtin_return_address(2),__builtin_return_address(3));
5190 if (!mddev->pers->error_handler)
5192 mddev->pers->error_handler(mddev,rdev);
5193 if (mddev->degraded)
5194 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5195 set_bit(StateChanged, &rdev->flags);
5196 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5197 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5198 md_wakeup_thread(mddev->thread);
5199 md_new_event_inintr(mddev);
5202 /* seq_file implementation /proc/mdstat */
5204 static void status_unused(struct seq_file *seq)
5208 struct list_head *tmp;
5210 seq_printf(seq, "unused devices: ");
5212 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5213 char b[BDEVNAME_SIZE];
5215 seq_printf(seq, "%s ",
5216 bdevname(rdev->bdev,b));
5219 seq_printf(seq, "<none>");
5221 seq_printf(seq, "\n");
5225 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5227 sector_t max_blocks, resync, res;
5228 unsigned long dt, db, rt;
5230 unsigned int per_milli;
5232 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5234 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5235 max_blocks = mddev->resync_max_sectors >> 1;
5237 max_blocks = mddev->size;
5240 * Should not happen.
5246 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5247 * in a sector_t, and (max_blocks>>scale) will fit in a
5248 * u32, as those are the requirements for sector_div.
5249 * Thus 'scale' must be at least 10
5252 if (sizeof(sector_t) > sizeof(unsigned long)) {
5253 while ( max_blocks/2 > (1ULL<<(scale+32)))
5256 res = (resync>>scale)*1000;
5257 sector_div(res, (u32)((max_blocks>>scale)+1));
5261 int i, x = per_milli/50, y = 20-x;
5262 seq_printf(seq, "[");
5263 for (i = 0; i < x; i++)
5264 seq_printf(seq, "=");
5265 seq_printf(seq, ">");
5266 for (i = 0; i < y; i++)
5267 seq_printf(seq, ".");
5268 seq_printf(seq, "] ");
5270 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5271 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5273 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5275 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5276 "resync" : "recovery"))),
5277 per_milli/10, per_milli % 10,
5278 (unsigned long long) resync,
5279 (unsigned long long) max_blocks);
5282 * We do not want to overflow, so the order of operands and
5283 * the * 100 / 100 trick are important. We do a +1 to be
5284 * safe against division by zero. We only estimate anyway.
5286 * dt: time from mark until now
5287 * db: blocks written from mark until now
5288 * rt: remaining time
5290 dt = ((jiffies - mddev->resync_mark) / HZ);
5292 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5293 - mddev->resync_mark_cnt;
5294 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5296 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5298 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5301 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5303 struct list_head *tmp;
5313 spin_lock(&all_mddevs_lock);
5314 list_for_each(tmp,&all_mddevs)
5316 mddev = list_entry(tmp, mddev_t, all_mddevs);
5318 spin_unlock(&all_mddevs_lock);
5321 spin_unlock(&all_mddevs_lock);
5323 return (void*)2;/* tail */
5327 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5329 struct list_head *tmp;
5330 mddev_t *next_mddev, *mddev = v;
5336 spin_lock(&all_mddevs_lock);
5338 tmp = all_mddevs.next;
5340 tmp = mddev->all_mddevs.next;
5341 if (tmp != &all_mddevs)
5342 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5344 next_mddev = (void*)2;
5347 spin_unlock(&all_mddevs_lock);
5355 static void md_seq_stop(struct seq_file *seq, void *v)
5359 if (mddev && v != (void*)1 && v != (void*)2)
5363 struct mdstat_info {
5367 static int md_seq_show(struct seq_file *seq, void *v)
5371 struct list_head *tmp2;
5373 struct mdstat_info *mi = seq->private;
5374 struct bitmap *bitmap;
5376 if (v == (void*)1) {
5377 struct mdk_personality *pers;
5378 seq_printf(seq, "Personalities : ");
5379 spin_lock(&pers_lock);
5380 list_for_each_entry(pers, &pers_list, list)
5381 seq_printf(seq, "[%s] ", pers->name);
5383 spin_unlock(&pers_lock);
5384 seq_printf(seq, "\n");
5385 mi->event = atomic_read(&md_event_count);
5388 if (v == (void*)2) {
5393 if (mddev_lock(mddev) < 0)
5396 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5397 seq_printf(seq, "%s : %sactive", mdname(mddev),
5398 mddev->pers ? "" : "in");
5401 seq_printf(seq, " (read-only)");
5403 seq_printf(seq, " (auto-read-only)");
5404 seq_printf(seq, " %s", mddev->pers->name);
5408 rdev_for_each(rdev, tmp2, mddev) {
5409 char b[BDEVNAME_SIZE];
5410 seq_printf(seq, " %s[%d]",
5411 bdevname(rdev->bdev,b), rdev->desc_nr);
5412 if (test_bit(WriteMostly, &rdev->flags))
5413 seq_printf(seq, "(W)");
5414 if (test_bit(Faulty, &rdev->flags)) {
5415 seq_printf(seq, "(F)");
5417 } else if (rdev->raid_disk < 0)
5418 seq_printf(seq, "(S)"); /* spare */
5422 if (!list_empty(&mddev->disks)) {
5424 seq_printf(seq, "\n %llu blocks",
5425 (unsigned long long)mddev->array_size);
5427 seq_printf(seq, "\n %llu blocks",
5428 (unsigned long long)size);
5430 if (mddev->persistent) {
5431 if (mddev->major_version != 0 ||
5432 mddev->minor_version != 90) {
5433 seq_printf(seq," super %d.%d",
5434 mddev->major_version,
5435 mddev->minor_version);
5437 } else if (mddev->external)
5438 seq_printf(seq, " super external:%s",
5439 mddev->metadata_type);
5441 seq_printf(seq, " super non-persistent");
5444 mddev->pers->status (seq, mddev);
5445 seq_printf(seq, "\n ");
5446 if (mddev->pers->sync_request) {
5447 if (mddev->curr_resync > 2) {
5448 status_resync (seq, mddev);
5449 seq_printf(seq, "\n ");
5450 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5451 seq_printf(seq, "\tresync=DELAYED\n ");
5452 else if (mddev->recovery_cp < MaxSector)
5453 seq_printf(seq, "\tresync=PENDING\n ");
5456 seq_printf(seq, "\n ");
5458 if ((bitmap = mddev->bitmap)) {
5459 unsigned long chunk_kb;
5460 unsigned long flags;
5461 spin_lock_irqsave(&bitmap->lock, flags);
5462 chunk_kb = bitmap->chunksize >> 10;
5463 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5465 bitmap->pages - bitmap->missing_pages,
5467 (bitmap->pages - bitmap->missing_pages)
5468 << (PAGE_SHIFT - 10),
5469 chunk_kb ? chunk_kb : bitmap->chunksize,
5470 chunk_kb ? "KB" : "B");
5472 seq_printf(seq, ", file: ");
5473 seq_path(seq, &bitmap->file->f_path, " \t\n");
5476 seq_printf(seq, "\n");
5477 spin_unlock_irqrestore(&bitmap->lock, flags);
5480 seq_printf(seq, "\n");
5482 mddev_unlock(mddev);
5487 static struct seq_operations md_seq_ops = {
5488 .start = md_seq_start,
5489 .next = md_seq_next,
5490 .stop = md_seq_stop,
5491 .show = md_seq_show,
5494 static int md_seq_open(struct inode *inode, struct file *file)
5497 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5501 error = seq_open(file, &md_seq_ops);
5505 struct seq_file *p = file->private_data;
5507 mi->event = atomic_read(&md_event_count);
5512 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5514 struct seq_file *m = filp->private_data;
5515 struct mdstat_info *mi = m->private;
5518 poll_wait(filp, &md_event_waiters, wait);
5520 /* always allow read */
5521 mask = POLLIN | POLLRDNORM;
5523 if (mi->event != atomic_read(&md_event_count))
5524 mask |= POLLERR | POLLPRI;
5528 static const struct file_operations md_seq_fops = {
5529 .owner = THIS_MODULE,
5530 .open = md_seq_open,
5532 .llseek = seq_lseek,
5533 .release = seq_release_private,
5534 .poll = mdstat_poll,
5537 int register_md_personality(struct mdk_personality *p)
5539 spin_lock(&pers_lock);
5540 list_add_tail(&p->list, &pers_list);
5541 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5542 spin_unlock(&pers_lock);
5546 int unregister_md_personality(struct mdk_personality *p)
5548 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5549 spin_lock(&pers_lock);
5550 list_del_init(&p->list);
5551 spin_unlock(&pers_lock);
5555 static int is_mddev_idle(mddev_t *mddev)
5558 struct list_head *tmp;
5563 rdev_for_each(rdev, tmp, mddev) {
5564 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5565 curr_events = disk_stat_read(disk, sectors[0]) +
5566 disk_stat_read(disk, sectors[1]) -
5567 atomic_read(&disk->sync_io);
5568 /* sync IO will cause sync_io to increase before the disk_stats
5569 * as sync_io is counted when a request starts, and
5570 * disk_stats is counted when it completes.
5571 * So resync activity will cause curr_events to be smaller than
5572 * when there was no such activity.
5573 * non-sync IO will cause disk_stat to increase without
5574 * increasing sync_io so curr_events will (eventually)
5575 * be larger than it was before. Once it becomes
5576 * substantially larger, the test below will cause
5577 * the array to appear non-idle, and resync will slow
5579 * If there is a lot of outstanding resync activity when
5580 * we set last_event to curr_events, then all that activity
5581 * completing might cause the array to appear non-idle
5582 * and resync will be slowed down even though there might
5583 * not have been non-resync activity. This will only
5584 * happen once though. 'last_events' will soon reflect
5585 * the state where there is little or no outstanding
5586 * resync requests, and further resync activity will
5587 * always make curr_events less than last_events.
5590 if (curr_events - rdev->last_events > 4096) {
5591 rdev->last_events = curr_events;
5598 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5600 /* another "blocks" (512byte) blocks have been synced */
5601 atomic_sub(blocks, &mddev->recovery_active);
5602 wake_up(&mddev->recovery_wait);
5604 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5605 md_wakeup_thread(mddev->thread);
5606 // stop recovery, signal do_sync ....
5611 /* md_write_start(mddev, bi)
5612 * If we need to update some array metadata (e.g. 'active' flag
5613 * in superblock) before writing, schedule a superblock update
5614 * and wait for it to complete.
5616 void md_write_start(mddev_t *mddev, struct bio *bi)
5619 if (bio_data_dir(bi) != WRITE)
5622 BUG_ON(mddev->ro == 1);
5623 if (mddev->ro == 2) {
5624 /* need to switch to read/write */
5626 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5627 md_wakeup_thread(mddev->thread);
5628 md_wakeup_thread(mddev->sync_thread);
5631 atomic_inc(&mddev->writes_pending);
5632 if (mddev->safemode == 1)
5633 mddev->safemode = 0;
5634 if (mddev->in_sync) {
5635 spin_lock_irq(&mddev->write_lock);
5636 if (mddev->in_sync) {
5638 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5639 md_wakeup_thread(mddev->thread);
5642 spin_unlock_irq(&mddev->write_lock);
5645 sysfs_notify(&mddev->kobj, NULL, "array_state");
5646 wait_event(mddev->sb_wait,
5647 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5648 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5651 void md_write_end(mddev_t *mddev)
5653 if (atomic_dec_and_test(&mddev->writes_pending)) {
5654 if (mddev->safemode == 2)
5655 md_wakeup_thread(mddev->thread);
5656 else if (mddev->safemode_delay)
5657 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5661 /* md_allow_write(mddev)
5662 * Calling this ensures that the array is marked 'active' so that writes
5663 * may proceed without blocking. It is important to call this before
5664 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5665 * Must be called with mddev_lock held.
5667 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5668 * is dropped, so return -EAGAIN after notifying userspace.
5670 int md_allow_write(mddev_t *mddev)
5676 if (!mddev->pers->sync_request)
5679 spin_lock_irq(&mddev->write_lock);
5680 if (mddev->in_sync) {
5682 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5683 if (mddev->safemode_delay &&
5684 mddev->safemode == 0)
5685 mddev->safemode = 1;
5686 spin_unlock_irq(&mddev->write_lock);
5687 md_update_sb(mddev, 0);
5688 sysfs_notify(&mddev->kobj, NULL, "array_state");
5690 spin_unlock_irq(&mddev->write_lock);
5692 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5697 EXPORT_SYMBOL_GPL(md_allow_write);
5699 #define SYNC_MARKS 10
5700 #define SYNC_MARK_STEP (3*HZ)
5701 void md_do_sync(mddev_t *mddev)
5704 unsigned int currspeed = 0,
5706 sector_t max_sectors,j, io_sectors;
5707 unsigned long mark[SYNC_MARKS];
5708 sector_t mark_cnt[SYNC_MARKS];
5710 struct list_head *tmp;
5711 sector_t last_check;
5713 struct list_head *rtmp;
5717 /* just incase thread restarts... */
5718 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5720 if (mddev->ro) /* never try to sync a read-only array */
5723 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5724 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5725 desc = "data-check";
5726 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5727 desc = "requested-resync";
5730 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5735 /* we overload curr_resync somewhat here.
5736 * 0 == not engaged in resync at all
5737 * 2 == checking that there is no conflict with another sync
5738 * 1 == like 2, but have yielded to allow conflicting resync to
5740 * other == active in resync - this many blocks
5742 * Before starting a resync we must have set curr_resync to
5743 * 2, and then checked that every "conflicting" array has curr_resync
5744 * less than ours. When we find one that is the same or higher
5745 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5746 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5747 * This will mean we have to start checking from the beginning again.
5752 mddev->curr_resync = 2;
5755 if (kthread_should_stop()) {
5756 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5759 for_each_mddev(mddev2, tmp) {
5760 if (mddev2 == mddev)
5762 if (!mddev->parallel_resync
5763 && mddev2->curr_resync
5764 && match_mddev_units(mddev, mddev2)) {
5766 if (mddev < mddev2 && mddev->curr_resync == 2) {
5767 /* arbitrarily yield */
5768 mddev->curr_resync = 1;
5769 wake_up(&resync_wait);
5771 if (mddev > mddev2 && mddev->curr_resync == 1)
5772 /* no need to wait here, we can wait the next
5773 * time 'round when curr_resync == 2
5776 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5777 if (!kthread_should_stop() &&
5778 mddev2->curr_resync >= mddev->curr_resync) {
5779 printk(KERN_INFO "md: delaying %s of %s"
5780 " until %s has finished (they"
5781 " share one or more physical units)\n",
5782 desc, mdname(mddev), mdname(mddev2));
5785 finish_wait(&resync_wait, &wq);
5788 finish_wait(&resync_wait, &wq);
5791 } while (mddev->curr_resync < 2);
5794 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5795 /* resync follows the size requested by the personality,
5796 * which defaults to physical size, but can be virtual size
5798 max_sectors = mddev->resync_max_sectors;
5799 mddev->resync_mismatches = 0;
5800 /* we don't use the checkpoint if there's a bitmap */
5801 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5802 j = mddev->resync_min;
5803 else if (!mddev->bitmap)
5804 j = mddev->recovery_cp;
5806 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5807 max_sectors = mddev->size << 1;
5809 /* recovery follows the physical size of devices */
5810 max_sectors = mddev->size << 1;
5812 rdev_for_each(rdev, rtmp, mddev)
5813 if (rdev->raid_disk >= 0 &&
5814 !test_bit(Faulty, &rdev->flags) &&
5815 !test_bit(In_sync, &rdev->flags) &&
5816 rdev->recovery_offset < j)
5817 j = rdev->recovery_offset;
5820 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5821 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5822 " %d KB/sec/disk.\n", speed_min(mddev));
5823 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5824 "(but not more than %d KB/sec) for %s.\n",
5825 speed_max(mddev), desc);
5827 is_mddev_idle(mddev); /* this also initializes IO event counters */
5830 for (m = 0; m < SYNC_MARKS; m++) {
5832 mark_cnt[m] = io_sectors;
5835 mddev->resync_mark = mark[last_mark];
5836 mddev->resync_mark_cnt = mark_cnt[last_mark];
5839 * Tune reconstruction:
5841 window = 32*(PAGE_SIZE/512);
5842 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5843 window/2,(unsigned long long) max_sectors/2);
5845 atomic_set(&mddev->recovery_active, 0);
5850 "md: resuming %s of %s from checkpoint.\n",
5851 desc, mdname(mddev));
5852 mddev->curr_resync = j;
5855 while (j < max_sectors) {
5859 if (j >= mddev->resync_max) {
5860 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5861 wait_event(mddev->recovery_wait,
5862 mddev->resync_max > j
5863 || kthread_should_stop());
5865 if (kthread_should_stop())
5867 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5868 currspeed < speed_min(mddev));
5870 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5874 if (!skipped) { /* actual IO requested */
5875 io_sectors += sectors;
5876 atomic_add(sectors, &mddev->recovery_active);
5880 if (j>1) mddev->curr_resync = j;
5881 mddev->curr_mark_cnt = io_sectors;
5882 if (last_check == 0)
5883 /* this is the earliers that rebuilt will be
5884 * visible in /proc/mdstat
5886 md_new_event(mddev);
5888 if (last_check + window > io_sectors || j == max_sectors)
5891 last_check = io_sectors;
5893 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5897 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5899 int next = (last_mark+1) % SYNC_MARKS;
5901 mddev->resync_mark = mark[next];
5902 mddev->resync_mark_cnt = mark_cnt[next];
5903 mark[next] = jiffies;
5904 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5909 if (kthread_should_stop())
5914 * this loop exits only if either when we are slower than
5915 * the 'hard' speed limit, or the system was IO-idle for
5917 * the system might be non-idle CPU-wise, but we only care
5918 * about not overloading the IO subsystem. (things like an
5919 * e2fsck being done on the RAID array should execute fast)
5921 blk_unplug(mddev->queue);
5924 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5925 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5927 if (currspeed > speed_min(mddev)) {
5928 if ((currspeed > speed_max(mddev)) ||
5929 !is_mddev_idle(mddev)) {
5935 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5937 * this also signals 'finished resyncing' to md_stop
5940 blk_unplug(mddev->queue);
5942 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5944 /* tell personality that we are finished */
5945 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5947 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5948 mddev->curr_resync > 2) {
5949 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5950 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5951 if (mddev->curr_resync >= mddev->recovery_cp) {
5953 "md: checkpointing %s of %s.\n",
5954 desc, mdname(mddev));
5955 mddev->recovery_cp = mddev->curr_resync;
5958 mddev->recovery_cp = MaxSector;
5960 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5961 mddev->curr_resync = MaxSector;
5962 rdev_for_each(rdev, rtmp, mddev)
5963 if (rdev->raid_disk >= 0 &&
5964 !test_bit(Faulty, &rdev->flags) &&
5965 !test_bit(In_sync, &rdev->flags) &&
5966 rdev->recovery_offset < mddev->curr_resync)
5967 rdev->recovery_offset = mddev->curr_resync;
5970 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5973 mddev->curr_resync = 0;
5974 mddev->resync_min = 0;
5975 mddev->resync_max = MaxSector;
5976 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5977 wake_up(&resync_wait);
5978 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5979 md_wakeup_thread(mddev->thread);
5984 * got a signal, exit.
5987 "md: md_do_sync() got signal ... exiting\n");
5988 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5992 EXPORT_SYMBOL_GPL(md_do_sync);
5995 static int remove_and_add_spares(mddev_t *mddev)
5998 struct list_head *rtmp;
6001 rdev_for_each(rdev, rtmp, mddev)
6002 if (rdev->raid_disk >= 0 &&
6003 !test_bit(Blocked, &rdev->flags) &&
6004 (test_bit(Faulty, &rdev->flags) ||
6005 ! test_bit(In_sync, &rdev->flags)) &&
6006 atomic_read(&rdev->nr_pending)==0) {
6007 if (mddev->pers->hot_remove_disk(
6008 mddev, rdev->raid_disk)==0) {
6010 sprintf(nm,"rd%d", rdev->raid_disk);
6011 sysfs_remove_link(&mddev->kobj, nm);
6012 rdev->raid_disk = -1;
6016 if (mddev->degraded) {
6017 rdev_for_each(rdev, rtmp, mddev) {
6018 if (rdev->raid_disk >= 0 &&
6019 !test_bit(In_sync, &rdev->flags))
6021 if (rdev->raid_disk < 0
6022 && !test_bit(Faulty, &rdev->flags)) {
6023 rdev->recovery_offset = 0;
6025 hot_add_disk(mddev, rdev) == 0) {
6027 sprintf(nm, "rd%d", rdev->raid_disk);
6028 if (sysfs_create_link(&mddev->kobj,
6031 "md: cannot register "
6035 md_new_event(mddev);
6044 * This routine is regularly called by all per-raid-array threads to
6045 * deal with generic issues like resync and super-block update.
6046 * Raid personalities that don't have a thread (linear/raid0) do not
6047 * need this as they never do any recovery or update the superblock.
6049 * It does not do any resync itself, but rather "forks" off other threads
6050 * to do that as needed.
6051 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6052 * "->recovery" and create a thread at ->sync_thread.
6053 * When the thread finishes it sets MD_RECOVERY_DONE
6054 * and wakeups up this thread which will reap the thread and finish up.
6055 * This thread also removes any faulty devices (with nr_pending == 0).
6057 * The overall approach is:
6058 * 1/ if the superblock needs updating, update it.
6059 * 2/ If a recovery thread is running, don't do anything else.
6060 * 3/ If recovery has finished, clean up, possibly marking spares active.
6061 * 4/ If there are any faulty devices, remove them.
6062 * 5/ If array is degraded, try to add spares devices
6063 * 6/ If array has spares or is not in-sync, start a resync thread.
6065 void md_check_recovery(mddev_t *mddev)
6068 struct list_head *rtmp;
6072 bitmap_daemon_work(mddev->bitmap);
6077 if (signal_pending(current)) {
6078 if (mddev->pers->sync_request && !mddev->external) {
6079 printk(KERN_INFO "md: %s in immediate safe mode\n",
6081 mddev->safemode = 2;
6083 flush_signals(current);
6087 (mddev->flags && !mddev->external) ||
6088 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6089 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6090 (mddev->external == 0 && mddev->safemode == 1) ||
6091 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6092 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6096 if (mddev_trylock(mddev)) {
6099 if (!mddev->external) {
6101 spin_lock_irq(&mddev->write_lock);
6102 if (mddev->safemode &&
6103 !atomic_read(&mddev->writes_pending) &&
6105 mddev->recovery_cp == MaxSector) {
6108 if (mddev->persistent)
6109 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6111 if (mddev->safemode == 1)
6112 mddev->safemode = 0;
6113 spin_unlock_irq(&mddev->write_lock);
6115 sysfs_notify(&mddev->kobj, NULL, "array_state");
6119 md_update_sb(mddev, 0);
6121 rdev_for_each(rdev, rtmp, mddev)
6122 if (test_and_clear_bit(StateChanged, &rdev->flags))
6123 sysfs_notify(&rdev->kobj, NULL, "state");
6126 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6127 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6128 /* resync/recovery still happening */
6129 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6132 if (mddev->sync_thread) {
6133 /* resync has finished, collect result */
6134 md_unregister_thread(mddev->sync_thread);
6135 mddev->sync_thread = NULL;
6136 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6138 /* activate any spares */
6139 if (mddev->pers->spare_active(mddev))
6140 sysfs_notify(&mddev->kobj, NULL,
6143 md_update_sb(mddev, 1);
6145 /* if array is no-longer degraded, then any saved_raid_disk
6146 * information must be scrapped
6148 if (!mddev->degraded)
6149 rdev_for_each(rdev, rtmp, mddev)
6150 rdev->saved_raid_disk = -1;
6152 mddev->recovery = 0;
6153 /* flag recovery needed just to double check */
6154 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6155 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6156 md_new_event(mddev);
6159 /* Set RUNNING before clearing NEEDED to avoid
6160 * any transients in the value of "sync_action".
6162 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6163 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6164 /* Clear some bits that don't mean anything, but
6167 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6168 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6170 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6172 /* no recovery is running.
6173 * remove any failed drives, then
6174 * add spares if possible.
6175 * Spare are also removed and re-added, to allow
6176 * the personality to fail the re-add.
6179 if (mddev->reshape_position != MaxSector) {
6180 if (mddev->pers->check_reshape(mddev) != 0)
6181 /* Cannot proceed */
6183 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6184 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6185 } else if ((spares = remove_and_add_spares(mddev))) {
6186 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6187 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6188 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6189 } else if (mddev->recovery_cp < MaxSector) {
6190 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6191 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6192 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6193 /* nothing to be done ... */
6196 if (mddev->pers->sync_request) {
6197 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6198 /* We are adding a device or devices to an array
6199 * which has the bitmap stored on all devices.
6200 * So make sure all bitmap pages get written
6202 bitmap_write_all(mddev->bitmap);
6204 mddev->sync_thread = md_register_thread(md_do_sync,
6207 if (!mddev->sync_thread) {
6208 printk(KERN_ERR "%s: could not start resync"
6211 /* leave the spares where they are, it shouldn't hurt */
6212 mddev->recovery = 0;
6214 md_wakeup_thread(mddev->sync_thread);
6215 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6216 md_new_event(mddev);
6219 if (!mddev->sync_thread) {
6220 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6221 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6223 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6225 mddev_unlock(mddev);
6229 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6231 sysfs_notify(&rdev->kobj, NULL, "state");
6232 wait_event_timeout(rdev->blocked_wait,
6233 !test_bit(Blocked, &rdev->flags),
6234 msecs_to_jiffies(5000));
6235 rdev_dec_pending(rdev, mddev);
6237 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6239 static int md_notify_reboot(struct notifier_block *this,
6240 unsigned long code, void *x)
6242 struct list_head *tmp;
6245 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6247 printk(KERN_INFO "md: stopping all md devices.\n");
6249 for_each_mddev(mddev, tmp)
6250 if (mddev_trylock(mddev)) {
6251 do_md_stop (mddev, 1);
6252 mddev_unlock(mddev);
6255 * certain more exotic SCSI devices are known to be
6256 * volatile wrt too early system reboots. While the
6257 * right place to handle this issue is the given
6258 * driver, we do want to have a safe RAID driver ...
6265 static struct notifier_block md_notifier = {
6266 .notifier_call = md_notify_reboot,
6268 .priority = INT_MAX, /* before any real devices */
6271 static void md_geninit(void)
6273 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6275 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6278 static int __init md_init(void)
6280 if (register_blkdev(MAJOR_NR, "md"))
6282 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6283 unregister_blkdev(MAJOR_NR, "md");
6286 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6287 md_probe, NULL, NULL);
6288 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6289 md_probe, NULL, NULL);
6291 register_reboot_notifier(&md_notifier);
6292 raid_table_header = register_sysctl_table(raid_root_table);
6302 * Searches all registered partitions for autorun RAID arrays
6306 static LIST_HEAD(all_detected_devices);
6307 struct detected_devices_node {
6308 struct list_head list;
6312 void md_autodetect_dev(dev_t dev)
6314 struct detected_devices_node *node_detected_dev;
6316 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6317 if (node_detected_dev) {
6318 node_detected_dev->dev = dev;
6319 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6321 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6322 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6327 static void autostart_arrays(int part)
6330 struct detected_devices_node *node_detected_dev;
6332 int i_scanned, i_passed;
6337 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6339 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6341 node_detected_dev = list_entry(all_detected_devices.next,
6342 struct detected_devices_node, list);
6343 list_del(&node_detected_dev->list);
6344 dev = node_detected_dev->dev;
6345 kfree(node_detected_dev);
6346 rdev = md_import_device(dev,0, 90);
6350 if (test_bit(Faulty, &rdev->flags)) {
6354 set_bit(AutoDetected, &rdev->flags);
6355 list_add(&rdev->same_set, &pending_raid_disks);
6359 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6360 i_scanned, i_passed);
6362 autorun_devices(part);
6365 #endif /* !MODULE */
6367 static __exit void md_exit(void)
6370 struct list_head *tmp;
6372 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6373 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6375 unregister_blkdev(MAJOR_NR,"md");
6376 unregister_blkdev(mdp_major, "mdp");
6377 unregister_reboot_notifier(&md_notifier);
6378 unregister_sysctl_table(raid_table_header);
6379 remove_proc_entry("mdstat", NULL);
6380 for_each_mddev(mddev, tmp) {
6381 struct gendisk *disk = mddev->gendisk;
6384 export_array(mddev);
6387 mddev->gendisk = NULL;
6392 subsys_initcall(md_init);
6393 module_exit(md_exit)
6395 static int get_ro(char *buffer, struct kernel_param *kp)
6397 return sprintf(buffer, "%d", start_readonly);
6399 static int set_ro(const char *val, struct kernel_param *kp)
6402 int num = simple_strtoul(val, &e, 10);
6403 if (*val && (*e == '\0' || *e == '\n')) {
6404 start_readonly = num;
6410 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6411 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6414 EXPORT_SYMBOL(register_md_personality);
6415 EXPORT_SYMBOL(unregister_md_personality);
6416 EXPORT_SYMBOL(md_error);
6417 EXPORT_SYMBOL(md_done_sync);
6418 EXPORT_SYMBOL(md_write_start);
6419 EXPORT_SYMBOL(md_write_end);
6420 EXPORT_SYMBOL(md_register_thread);
6421 EXPORT_SYMBOL(md_unregister_thread);
6422 EXPORT_SYMBOL(md_wakeup_thread);
6423 EXPORT_SYMBOL(md_check_recovery);
6424 MODULE_LICENSE("GPL");
6426 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);