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/config.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/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
46 #include <linux/init.h>
48 #include <linux/file.h>
51 #include <linux/kmod.h>
54 #include <asm/unaligned.h>
56 #define MAJOR_NR MD_MAJOR
59 /* 63 partitions with the alternate major number (mdp) */
60 #define MdpMinorShift 6
63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
67 static void autostart_arrays (int part);
70 static mdk_personality_t *pers[MAX_PERSONALITY];
71 static DEFINE_SPINLOCK(pers_lock);
74 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
75 * is 1000 KB/sec, so the extra system load does not show up that much.
76 * Increase it if you want to have more _guaranteed_ speed. Note that
77 * the RAID driver will use the maximum available bandwidth if the IO
78 * subsystem is idle. There is also an 'absolute maximum' reconstruction
79 * speed limit - in case reconstruction slows down your system despite
82 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
85 static int sysctl_speed_limit_min = 1000;
86 static int sysctl_speed_limit_max = 200000;
88 static struct ctl_table_header *raid_table_header;
90 static ctl_table raid_table[] = {
92 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
93 .procname = "speed_limit_min",
94 .data = &sysctl_speed_limit_min,
95 .maxlen = sizeof(int),
97 .proc_handler = &proc_dointvec,
100 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
101 .procname = "speed_limit_max",
102 .data = &sysctl_speed_limit_max,
103 .maxlen = sizeof(int),
105 .proc_handler = &proc_dointvec,
110 static ctl_table raid_dir_table[] = {
112 .ctl_name = DEV_RAID,
121 static ctl_table raid_root_table[] = {
127 .child = raid_dir_table,
132 static struct block_device_operations md_fops;
134 static int start_readonly;
137 * Enables to iterate over all existing md arrays
138 * all_mddevs_lock protects this list.
140 static LIST_HEAD(all_mddevs);
141 static DEFINE_SPINLOCK(all_mddevs_lock);
145 * iterates through all used mddevs in the system.
146 * We take care to grab the all_mddevs_lock whenever navigating
147 * the list, and to always hold a refcount when unlocked.
148 * Any code which breaks out of this loop while own
149 * a reference to the current mddev and must mddev_put it.
151 #define ITERATE_MDDEV(mddev,tmp) \
153 for (({ spin_lock(&all_mddevs_lock); \
154 tmp = all_mddevs.next; \
156 ({ if (tmp != &all_mddevs) \
157 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
158 spin_unlock(&all_mddevs_lock); \
159 if (mddev) mddev_put(mddev); \
160 mddev = list_entry(tmp, mddev_t, all_mddevs); \
161 tmp != &all_mddevs;}); \
162 ({ spin_lock(&all_mddevs_lock); \
167 static int md_fail_request (request_queue_t *q, struct bio *bio)
169 bio_io_error(bio, bio->bi_size);
173 static inline mddev_t *mddev_get(mddev_t *mddev)
175 atomic_inc(&mddev->active);
179 static void mddev_put(mddev_t *mddev)
181 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
183 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
184 list_del(&mddev->all_mddevs);
185 blk_put_queue(mddev->queue);
186 kobject_unregister(&mddev->kobj);
188 spin_unlock(&all_mddevs_lock);
191 static mddev_t * mddev_find(dev_t unit)
193 mddev_t *mddev, *new = NULL;
196 spin_lock(&all_mddevs_lock);
197 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
198 if (mddev->unit == unit) {
200 spin_unlock(&all_mddevs_lock);
206 list_add(&new->all_mddevs, &all_mddevs);
207 spin_unlock(&all_mddevs_lock);
210 spin_unlock(&all_mddevs_lock);
212 new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
216 memset(new, 0, sizeof(*new));
219 if (MAJOR(unit) == MD_MAJOR)
220 new->md_minor = MINOR(unit);
222 new->md_minor = MINOR(unit) >> MdpMinorShift;
224 init_MUTEX(&new->reconfig_sem);
225 INIT_LIST_HEAD(&new->disks);
226 INIT_LIST_HEAD(&new->all_mddevs);
227 init_timer(&new->safemode_timer);
228 atomic_set(&new->active, 1);
229 spin_lock_init(&new->write_lock);
230 init_waitqueue_head(&new->sb_wait);
232 new->queue = blk_alloc_queue(GFP_KERNEL);
238 blk_queue_make_request(new->queue, md_fail_request);
243 static inline int mddev_lock(mddev_t * mddev)
245 return down_interruptible(&mddev->reconfig_sem);
248 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
250 down(&mddev->reconfig_sem);
253 static inline int mddev_trylock(mddev_t * mddev)
255 return down_trylock(&mddev->reconfig_sem);
258 static inline void mddev_unlock(mddev_t * mddev)
260 up(&mddev->reconfig_sem);
262 md_wakeup_thread(mddev->thread);
265 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
268 struct list_head *tmp;
270 ITERATE_RDEV(mddev,rdev,tmp) {
271 if (rdev->desc_nr == nr)
277 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
279 struct list_head *tmp;
282 ITERATE_RDEV(mddev,rdev,tmp) {
283 if (rdev->bdev->bd_dev == dev)
289 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
291 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
292 return MD_NEW_SIZE_BLOCKS(size);
295 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
299 size = rdev->sb_offset;
302 size &= ~((sector_t)chunk_size/1024 - 1);
306 static int alloc_disk_sb(mdk_rdev_t * rdev)
311 rdev->sb_page = alloc_page(GFP_KERNEL);
312 if (!rdev->sb_page) {
313 printk(KERN_ALERT "md: out of memory.\n");
320 static void free_disk_sb(mdk_rdev_t * rdev)
323 page_cache_release(rdev->sb_page);
325 rdev->sb_page = NULL;
332 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
334 mdk_rdev_t *rdev = bio->bi_private;
335 mddev_t *mddev = rdev->mddev;
339 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
340 md_error(mddev, rdev);
342 if (atomic_dec_and_test(&mddev->pending_writes))
343 wake_up(&mddev->sb_wait);
348 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
350 struct bio *bio2 = bio->bi_private;
351 mdk_rdev_t *rdev = bio2->bi_private;
352 mddev_t *mddev = rdev->mddev;
356 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
357 error == -EOPNOTSUPP) {
359 /* barriers don't appear to be supported :-( */
360 set_bit(BarriersNotsupp, &rdev->flags);
361 mddev->barriers_work = 0;
362 spin_lock_irqsave(&mddev->write_lock, flags);
363 bio2->bi_next = mddev->biolist;
364 mddev->biolist = bio2;
365 spin_unlock_irqrestore(&mddev->write_lock, flags);
366 wake_up(&mddev->sb_wait);
371 bio->bi_private = rdev;
372 return super_written(bio, bytes_done, error);
375 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
376 sector_t sector, int size, struct page *page)
378 /* write first size bytes of page to sector of rdev
379 * Increment mddev->pending_writes before returning
380 * and decrement it on completion, waking up sb_wait
381 * if zero is reached.
382 * If an error occurred, call md_error
384 * As we might need to resubmit the request if BIO_RW_BARRIER
385 * causes ENOTSUPP, we allocate a spare bio...
387 struct bio *bio = bio_alloc(GFP_NOIO, 1);
388 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
390 bio->bi_bdev = rdev->bdev;
391 bio->bi_sector = sector;
392 bio_add_page(bio, page, size, 0);
393 bio->bi_private = rdev;
394 bio->bi_end_io = super_written;
397 atomic_inc(&mddev->pending_writes);
398 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
400 rw |= (1<<BIO_RW_BARRIER);
401 rbio = bio_clone(bio, GFP_NOIO);
402 rbio->bi_private = bio;
403 rbio->bi_end_io = super_written_barrier;
404 submit_bio(rw, rbio);
409 void md_super_wait(mddev_t *mddev)
411 /* wait for all superblock writes that were scheduled to complete.
412 * if any had to be retried (due to BARRIER problems), retry them
416 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
417 if (atomic_read(&mddev->pending_writes)==0)
419 while (mddev->biolist) {
421 spin_lock_irq(&mddev->write_lock);
422 bio = mddev->biolist;
423 mddev->biolist = bio->bi_next ;
425 spin_unlock_irq(&mddev->write_lock);
426 submit_bio(bio->bi_rw, bio);
430 finish_wait(&mddev->sb_wait, &wq);
433 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
438 complete((struct completion*)bio->bi_private);
442 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
443 struct page *page, int rw)
445 struct bio *bio = bio_alloc(GFP_NOIO, 1);
446 struct completion event;
449 rw |= (1 << BIO_RW_SYNC);
452 bio->bi_sector = sector;
453 bio_add_page(bio, page, size, 0);
454 init_completion(&event);
455 bio->bi_private = &event;
456 bio->bi_end_io = bi_complete;
458 wait_for_completion(&event);
460 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
465 static int read_disk_sb(mdk_rdev_t * rdev, int size)
467 char b[BDEVNAME_SIZE];
468 if (!rdev->sb_page) {
476 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
482 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
483 bdevname(rdev->bdev,b));
487 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
489 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
490 (sb1->set_uuid1 == sb2->set_uuid1) &&
491 (sb1->set_uuid2 == sb2->set_uuid2) &&
492 (sb1->set_uuid3 == sb2->set_uuid3))
500 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
503 mdp_super_t *tmp1, *tmp2;
505 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
506 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
508 if (!tmp1 || !tmp2) {
510 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
518 * nr_disks is not constant
523 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
534 static unsigned int calc_sb_csum(mdp_super_t * sb)
536 unsigned int disk_csum, csum;
538 disk_csum = sb->sb_csum;
540 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
541 sb->sb_csum = disk_csum;
547 * Handle superblock details.
548 * We want to be able to handle multiple superblock formats
549 * so we have a common interface to them all, and an array of
550 * different handlers.
551 * We rely on user-space to write the initial superblock, and support
552 * reading and updating of superblocks.
553 * Interface methods are:
554 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
555 * loads and validates a superblock on dev.
556 * if refdev != NULL, compare superblocks on both devices
558 * 0 - dev has a superblock that is compatible with refdev
559 * 1 - dev has a superblock that is compatible and newer than refdev
560 * so dev should be used as the refdev in future
561 * -EINVAL superblock incompatible or invalid
562 * -othererror e.g. -EIO
564 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
565 * Verify that dev is acceptable into mddev.
566 * The first time, mddev->raid_disks will be 0, and data from
567 * dev should be merged in. Subsequent calls check that dev
568 * is new enough. Return 0 or -EINVAL
570 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
571 * Update the superblock for rdev with data in mddev
572 * This does not write to disc.
578 struct module *owner;
579 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
580 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
581 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
585 * load_super for 0.90.0
587 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
589 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
595 * Calculate the position of the superblock,
596 * it's at the end of the disk.
598 * It also happens to be a multiple of 4Kb.
600 sb_offset = calc_dev_sboffset(rdev->bdev);
601 rdev->sb_offset = sb_offset;
603 ret = read_disk_sb(rdev, MD_SB_BYTES);
608 bdevname(rdev->bdev, b);
609 sb = (mdp_super_t*)page_address(rdev->sb_page);
611 if (sb->md_magic != MD_SB_MAGIC) {
612 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
617 if (sb->major_version != 0 ||
618 sb->minor_version != 90) {
619 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
620 sb->major_version, sb->minor_version,
625 if (sb->raid_disks <= 0)
628 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
629 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
634 rdev->preferred_minor = sb->md_minor;
635 rdev->data_offset = 0;
636 rdev->sb_size = MD_SB_BYTES;
638 if (sb->level == LEVEL_MULTIPATH)
641 rdev->desc_nr = sb->this_disk.number;
647 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
648 if (!uuid_equal(refsb, sb)) {
649 printk(KERN_WARNING "md: %s has different UUID to %s\n",
650 b, bdevname(refdev->bdev,b2));
653 if (!sb_equal(refsb, sb)) {
654 printk(KERN_WARNING "md: %s has same UUID"
655 " but different superblock to %s\n",
656 b, bdevname(refdev->bdev, b2));
660 ev2 = md_event(refsb);
666 rdev->size = calc_dev_size(rdev, sb->chunk_size);
673 * validate_super for 0.90.0
675 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
678 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
680 rdev->raid_disk = -1;
682 if (mddev->raid_disks == 0) {
683 mddev->major_version = 0;
684 mddev->minor_version = sb->minor_version;
685 mddev->patch_version = sb->patch_version;
686 mddev->persistent = ! sb->not_persistent;
687 mddev->chunk_size = sb->chunk_size;
688 mddev->ctime = sb->ctime;
689 mddev->utime = sb->utime;
690 mddev->level = sb->level;
691 mddev->layout = sb->layout;
692 mddev->raid_disks = sb->raid_disks;
693 mddev->size = sb->size;
694 mddev->events = md_event(sb);
695 mddev->bitmap_offset = 0;
696 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
698 if (sb->state & (1<<MD_SB_CLEAN))
699 mddev->recovery_cp = MaxSector;
701 if (sb->events_hi == sb->cp_events_hi &&
702 sb->events_lo == sb->cp_events_lo) {
703 mddev->recovery_cp = sb->recovery_cp;
705 mddev->recovery_cp = 0;
708 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
709 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
710 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
711 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
713 mddev->max_disks = MD_SB_DISKS;
715 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
716 mddev->bitmap_file == NULL) {
717 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
718 && mddev->level != 10) {
719 /* FIXME use a better test */
720 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
723 mddev->bitmap_offset = mddev->default_bitmap_offset;
726 } else if (mddev->pers == NULL) {
727 /* Insist on good event counter while assembling */
728 __u64 ev1 = md_event(sb);
730 if (ev1 < mddev->events)
732 } else if (mddev->bitmap) {
733 /* if adding to array with a bitmap, then we can accept an
734 * older device ... but not too old.
736 __u64 ev1 = md_event(sb);
737 if (ev1 < mddev->bitmap->events_cleared)
739 } else /* just a hot-add of a new device, leave raid_disk at -1 */
742 if (mddev->level != LEVEL_MULTIPATH) {
743 desc = sb->disks + rdev->desc_nr;
745 if (desc->state & (1<<MD_DISK_FAULTY))
746 set_bit(Faulty, &rdev->flags);
747 else if (desc->state & (1<<MD_DISK_SYNC) &&
748 desc->raid_disk < mddev->raid_disks) {
749 set_bit(In_sync, &rdev->flags);
750 rdev->raid_disk = desc->raid_disk;
752 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
753 set_bit(WriteMostly, &rdev->flags);
754 } else /* MULTIPATH are always insync */
755 set_bit(In_sync, &rdev->flags);
760 * sync_super for 0.90.0
762 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
765 struct list_head *tmp;
767 int next_spare = mddev->raid_disks;
770 /* make rdev->sb match mddev data..
773 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
774 * 3/ any empty disks < next_spare become removed
776 * disks[0] gets initialised to REMOVED because
777 * we cannot be sure from other fields if it has
778 * been initialised or not.
781 int active=0, working=0,failed=0,spare=0,nr_disks=0;
783 rdev->sb_size = MD_SB_BYTES;
785 sb = (mdp_super_t*)page_address(rdev->sb_page);
787 memset(sb, 0, sizeof(*sb));
789 sb->md_magic = MD_SB_MAGIC;
790 sb->major_version = mddev->major_version;
791 sb->minor_version = mddev->minor_version;
792 sb->patch_version = mddev->patch_version;
793 sb->gvalid_words = 0; /* ignored */
794 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
795 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
796 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
797 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
799 sb->ctime = mddev->ctime;
800 sb->level = mddev->level;
801 sb->size = mddev->size;
802 sb->raid_disks = mddev->raid_disks;
803 sb->md_minor = mddev->md_minor;
804 sb->not_persistent = !mddev->persistent;
805 sb->utime = mddev->utime;
807 sb->events_hi = (mddev->events>>32);
808 sb->events_lo = (u32)mddev->events;
812 sb->recovery_cp = mddev->recovery_cp;
813 sb->cp_events_hi = (mddev->events>>32);
814 sb->cp_events_lo = (u32)mddev->events;
815 if (mddev->recovery_cp == MaxSector)
816 sb->state = (1<< MD_SB_CLEAN);
820 sb->layout = mddev->layout;
821 sb->chunk_size = mddev->chunk_size;
823 if (mddev->bitmap && mddev->bitmap_file == NULL)
824 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
826 sb->disks[0].state = (1<<MD_DISK_REMOVED);
827 ITERATE_RDEV(mddev,rdev2,tmp) {
830 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
831 && !test_bit(Faulty, &rdev2->flags))
832 desc_nr = rdev2->raid_disk;
834 desc_nr = next_spare++;
835 rdev2->desc_nr = desc_nr;
836 d = &sb->disks[rdev2->desc_nr];
838 d->number = rdev2->desc_nr;
839 d->major = MAJOR(rdev2->bdev->bd_dev);
840 d->minor = MINOR(rdev2->bdev->bd_dev);
841 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
842 && !test_bit(Faulty, &rdev2->flags))
843 d->raid_disk = rdev2->raid_disk;
845 d->raid_disk = rdev2->desc_nr; /* compatibility */
846 if (test_bit(Faulty, &rdev2->flags)) {
847 d->state = (1<<MD_DISK_FAULTY);
849 } else if (test_bit(In_sync, &rdev2->flags)) {
850 d->state = (1<<MD_DISK_ACTIVE);
851 d->state |= (1<<MD_DISK_SYNC);
859 if (test_bit(WriteMostly, &rdev2->flags))
860 d->state |= (1<<MD_DISK_WRITEMOSTLY);
862 /* now set the "removed" and "faulty" bits on any missing devices */
863 for (i=0 ; i < mddev->raid_disks ; i++) {
864 mdp_disk_t *d = &sb->disks[i];
865 if (d->state == 0 && d->number == 0) {
868 d->state = (1<<MD_DISK_REMOVED);
869 d->state |= (1<<MD_DISK_FAULTY);
873 sb->nr_disks = nr_disks;
874 sb->active_disks = active;
875 sb->working_disks = working;
876 sb->failed_disks = failed;
877 sb->spare_disks = spare;
879 sb->this_disk = sb->disks[rdev->desc_nr];
880 sb->sb_csum = calc_sb_csum(sb);
884 * version 1 superblock
887 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
889 unsigned int disk_csum, csum;
890 unsigned long long newcsum;
891 int size = 256 + le32_to_cpu(sb->max_dev)*2;
892 unsigned int *isuper = (unsigned int*)sb;
895 disk_csum = sb->sb_csum;
898 for (i=0; size>=4; size -= 4 )
899 newcsum += le32_to_cpu(*isuper++);
902 newcsum += le16_to_cpu(*(unsigned short*) isuper);
904 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
905 sb->sb_csum = disk_csum;
906 return cpu_to_le32(csum);
909 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
911 struct mdp_superblock_1 *sb;
914 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
918 * Calculate the position of the superblock.
919 * It is always aligned to a 4K boundary and
920 * depeding on minor_version, it can be:
921 * 0: At least 8K, but less than 12K, from end of device
922 * 1: At start of device
923 * 2: 4K from start of device.
925 switch(minor_version) {
927 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
929 sb_offset &= ~(sector_t)(4*2-1);
930 /* convert from sectors to K */
942 rdev->sb_offset = sb_offset;
944 /* superblock is rarely larger than 1K, but it can be larger,
945 * and it is safe to read 4k, so we do that
947 ret = read_disk_sb(rdev, 4096);
951 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
953 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
954 sb->major_version != cpu_to_le32(1) ||
955 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
956 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
957 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
960 if (calc_sb_1_csum(sb) != sb->sb_csum) {
961 printk("md: invalid superblock checksum on %s\n",
962 bdevname(rdev->bdev,b));
965 if (le64_to_cpu(sb->data_size) < 10) {
966 printk("md: data_size too small on %s\n",
967 bdevname(rdev->bdev,b));
970 rdev->preferred_minor = 0xffff;
971 rdev->data_offset = le64_to_cpu(sb->data_offset);
973 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
974 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
975 if (rdev->sb_size & bmask)
976 rdev-> sb_size = (rdev->sb_size | bmask)+1;
982 struct mdp_superblock_1 *refsb =
983 (struct mdp_superblock_1*)page_address(refdev->sb_page);
985 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
986 sb->level != refsb->level ||
987 sb->layout != refsb->layout ||
988 sb->chunksize != refsb->chunksize) {
989 printk(KERN_WARNING "md: %s has strangely different"
990 " superblock to %s\n",
991 bdevname(rdev->bdev,b),
992 bdevname(refdev->bdev,b2));
995 ev1 = le64_to_cpu(sb->events);
996 ev2 = le64_to_cpu(refsb->events);
1002 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1004 rdev->size = rdev->sb_offset;
1005 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1007 rdev->size = le64_to_cpu(sb->data_size)/2;
1008 if (le32_to_cpu(sb->chunksize))
1009 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1013 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1015 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1017 rdev->raid_disk = -1;
1019 if (mddev->raid_disks == 0) {
1020 mddev->major_version = 1;
1021 mddev->patch_version = 0;
1022 mddev->persistent = 1;
1023 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1024 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1025 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1026 mddev->level = le32_to_cpu(sb->level);
1027 mddev->layout = le32_to_cpu(sb->layout);
1028 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1029 mddev->size = le64_to_cpu(sb->size)/2;
1030 mddev->events = le64_to_cpu(sb->events);
1031 mddev->bitmap_offset = 0;
1032 mddev->default_bitmap_offset = 1024;
1034 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1035 memcpy(mddev->uuid, sb->set_uuid, 16);
1037 mddev->max_disks = (4096-256)/2;
1039 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1040 mddev->bitmap_file == NULL ) {
1041 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1042 && mddev->level != 10) {
1043 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1046 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1048 } else if (mddev->pers == NULL) {
1049 /* Insist of good event counter while assembling */
1050 __u64 ev1 = le64_to_cpu(sb->events);
1052 if (ev1 < mddev->events)
1054 } else if (mddev->bitmap) {
1055 /* If adding to array with a bitmap, then we can accept an
1056 * older device, but not too old.
1058 __u64 ev1 = le64_to_cpu(sb->events);
1059 if (ev1 < mddev->bitmap->events_cleared)
1061 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1064 if (mddev->level != LEVEL_MULTIPATH) {
1066 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1067 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1069 case 0xffff: /* spare */
1071 case 0xfffe: /* faulty */
1072 set_bit(Faulty, &rdev->flags);
1075 set_bit(In_sync, &rdev->flags);
1076 rdev->raid_disk = role;
1079 if (sb->devflags & WriteMostly1)
1080 set_bit(WriteMostly, &rdev->flags);
1081 } else /* MULTIPATH are always insync */
1082 set_bit(In_sync, &rdev->flags);
1087 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1089 struct mdp_superblock_1 *sb;
1090 struct list_head *tmp;
1093 /* make rdev->sb match mddev and rdev data. */
1095 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1097 sb->feature_map = 0;
1099 memset(sb->pad1, 0, sizeof(sb->pad1));
1100 memset(sb->pad2, 0, sizeof(sb->pad2));
1101 memset(sb->pad3, 0, sizeof(sb->pad3));
1103 sb->utime = cpu_to_le64((__u64)mddev->utime);
1104 sb->events = cpu_to_le64(mddev->events);
1106 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1108 sb->resync_offset = cpu_to_le64(0);
1110 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1111 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1112 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1116 ITERATE_RDEV(mddev,rdev2,tmp)
1117 if (rdev2->desc_nr+1 > max_dev)
1118 max_dev = rdev2->desc_nr+1;
1120 sb->max_dev = cpu_to_le32(max_dev);
1121 for (i=0; i<max_dev;i++)
1122 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1124 ITERATE_RDEV(mddev,rdev2,tmp) {
1126 if (test_bit(Faulty, &rdev2->flags))
1127 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1128 else if (test_bit(In_sync, &rdev2->flags))
1129 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1131 sb->dev_roles[i] = cpu_to_le16(0xffff);
1134 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1135 sb->sb_csum = calc_sb_1_csum(sb);
1139 static struct super_type super_types[] = {
1142 .owner = THIS_MODULE,
1143 .load_super = super_90_load,
1144 .validate_super = super_90_validate,
1145 .sync_super = super_90_sync,
1149 .owner = THIS_MODULE,
1150 .load_super = super_1_load,
1151 .validate_super = super_1_validate,
1152 .sync_super = super_1_sync,
1156 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1158 struct list_head *tmp;
1161 ITERATE_RDEV(mddev,rdev,tmp)
1162 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1168 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1170 struct list_head *tmp;
1173 ITERATE_RDEV(mddev1,rdev,tmp)
1174 if (match_dev_unit(mddev2, rdev))
1180 static LIST_HEAD(pending_raid_disks);
1182 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1184 mdk_rdev_t *same_pdev;
1185 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1192 same_pdev = match_dev_unit(mddev, rdev);
1195 "%s: WARNING: %s appears to be on the same physical"
1196 " disk as %s. True\n protection against single-disk"
1197 " failure might be compromised.\n",
1198 mdname(mddev), bdevname(rdev->bdev,b),
1199 bdevname(same_pdev->bdev,b2));
1201 /* Verify rdev->desc_nr is unique.
1202 * If it is -1, assign a free number, else
1203 * check number is not in use
1205 if (rdev->desc_nr < 0) {
1207 if (mddev->pers) choice = mddev->raid_disks;
1208 while (find_rdev_nr(mddev, choice))
1210 rdev->desc_nr = choice;
1212 if (find_rdev_nr(mddev, rdev->desc_nr))
1215 bdevname(rdev->bdev,b);
1216 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1219 list_add(&rdev->same_set, &mddev->disks);
1220 rdev->mddev = mddev;
1221 printk(KERN_INFO "md: bind<%s>\n", b);
1223 rdev->kobj.parent = &mddev->kobj;
1224 kobject_add(&rdev->kobj);
1226 if (rdev->bdev->bd_part)
1227 ko = &rdev->bdev->bd_part->kobj;
1229 ko = &rdev->bdev->bd_disk->kobj;
1230 sysfs_create_link(&rdev->kobj, ko, "block");
1234 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1236 char b[BDEVNAME_SIZE];
1241 list_del_init(&rdev->same_set);
1242 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1244 sysfs_remove_link(&rdev->kobj, "block");
1245 kobject_del(&rdev->kobj);
1249 * prevent the device from being mounted, repartitioned or
1250 * otherwise reused by a RAID array (or any other kernel
1251 * subsystem), by bd_claiming the device.
1253 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1256 struct block_device *bdev;
1257 char b[BDEVNAME_SIZE];
1259 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1261 printk(KERN_ERR "md: could not open %s.\n",
1262 __bdevname(dev, b));
1263 return PTR_ERR(bdev);
1265 err = bd_claim(bdev, rdev);
1267 printk(KERN_ERR "md: could not bd_claim %s.\n",
1276 static void unlock_rdev(mdk_rdev_t *rdev)
1278 struct block_device *bdev = rdev->bdev;
1286 void md_autodetect_dev(dev_t dev);
1288 static void export_rdev(mdk_rdev_t * rdev)
1290 char b[BDEVNAME_SIZE];
1291 printk(KERN_INFO "md: export_rdev(%s)\n",
1292 bdevname(rdev->bdev,b));
1296 list_del_init(&rdev->same_set);
1298 md_autodetect_dev(rdev->bdev->bd_dev);
1301 kobject_put(&rdev->kobj);
1304 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1306 unbind_rdev_from_array(rdev);
1310 static void export_array(mddev_t *mddev)
1312 struct list_head *tmp;
1315 ITERATE_RDEV(mddev,rdev,tmp) {
1320 kick_rdev_from_array(rdev);
1322 if (!list_empty(&mddev->disks))
1324 mddev->raid_disks = 0;
1325 mddev->major_version = 0;
1328 static void print_desc(mdp_disk_t *desc)
1330 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1331 desc->major,desc->minor,desc->raid_disk,desc->state);
1334 static void print_sb(mdp_super_t *sb)
1339 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1340 sb->major_version, sb->minor_version, sb->patch_version,
1341 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1343 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1344 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1345 sb->md_minor, sb->layout, sb->chunk_size);
1346 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1347 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1348 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1349 sb->failed_disks, sb->spare_disks,
1350 sb->sb_csum, (unsigned long)sb->events_lo);
1353 for (i = 0; i < MD_SB_DISKS; i++) {
1356 desc = sb->disks + i;
1357 if (desc->number || desc->major || desc->minor ||
1358 desc->raid_disk || (desc->state && (desc->state != 4))) {
1359 printk(" D %2d: ", i);
1363 printk(KERN_INFO "md: THIS: ");
1364 print_desc(&sb->this_disk);
1368 static void print_rdev(mdk_rdev_t *rdev)
1370 char b[BDEVNAME_SIZE];
1371 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1372 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1373 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1375 if (rdev->sb_loaded) {
1376 printk(KERN_INFO "md: rdev superblock:\n");
1377 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1379 printk(KERN_INFO "md: no rdev superblock!\n");
1382 void md_print_devices(void)
1384 struct list_head *tmp, *tmp2;
1387 char b[BDEVNAME_SIZE];
1390 printk("md: **********************************\n");
1391 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1392 printk("md: **********************************\n");
1393 ITERATE_MDDEV(mddev,tmp) {
1396 bitmap_print_sb(mddev->bitmap);
1398 printk("%s: ", mdname(mddev));
1399 ITERATE_RDEV(mddev,rdev,tmp2)
1400 printk("<%s>", bdevname(rdev->bdev,b));
1403 ITERATE_RDEV(mddev,rdev,tmp2)
1406 printk("md: **********************************\n");
1411 static void sync_sbs(mddev_t * mddev)
1414 struct list_head *tmp;
1416 ITERATE_RDEV(mddev,rdev,tmp) {
1417 super_types[mddev->major_version].
1418 sync_super(mddev, rdev);
1419 rdev->sb_loaded = 1;
1423 static void md_update_sb(mddev_t * mddev)
1426 struct list_head *tmp;
1431 spin_lock_irq(&mddev->write_lock);
1432 sync_req = mddev->in_sync;
1433 mddev->utime = get_seconds();
1436 if (!mddev->events) {
1438 * oops, this 64-bit counter should never wrap.
1439 * Either we are in around ~1 trillion A.C., assuming
1440 * 1 reboot per second, or we have a bug:
1445 mddev->sb_dirty = 2;
1449 * do not write anything to disk if using
1450 * nonpersistent superblocks
1452 if (!mddev->persistent) {
1453 mddev->sb_dirty = 0;
1454 spin_unlock_irq(&mddev->write_lock);
1455 wake_up(&mddev->sb_wait);
1458 spin_unlock_irq(&mddev->write_lock);
1461 "md: updating %s RAID superblock on device (in sync %d)\n",
1462 mdname(mddev),mddev->in_sync);
1464 err = bitmap_update_sb(mddev->bitmap);
1465 ITERATE_RDEV(mddev,rdev,tmp) {
1466 char b[BDEVNAME_SIZE];
1467 dprintk(KERN_INFO "md: ");
1468 if (test_bit(Faulty, &rdev->flags))
1469 dprintk("(skipping faulty ");
1471 dprintk("%s ", bdevname(rdev->bdev,b));
1472 if (!test_bit(Faulty, &rdev->flags)) {
1473 md_super_write(mddev,rdev,
1474 rdev->sb_offset<<1, rdev->sb_size,
1476 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1477 bdevname(rdev->bdev,b),
1478 (unsigned long long)rdev->sb_offset);
1482 if (mddev->level == LEVEL_MULTIPATH)
1483 /* only need to write one superblock... */
1486 md_super_wait(mddev);
1487 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1489 spin_lock_irq(&mddev->write_lock);
1490 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1491 /* have to write it out again */
1492 spin_unlock_irq(&mddev->write_lock);
1495 mddev->sb_dirty = 0;
1496 spin_unlock_irq(&mddev->write_lock);
1497 wake_up(&mddev->sb_wait);
1501 struct rdev_sysfs_entry {
1502 struct attribute attr;
1503 ssize_t (*show)(mdk_rdev_t *, char *);
1504 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1508 state_show(mdk_rdev_t *rdev, char *page)
1513 if (test_bit(Faulty, &rdev->flags)) {
1514 len+= sprintf(page+len, "%sfaulty",sep);
1517 if (test_bit(In_sync, &rdev->flags)) {
1518 len += sprintf(page+len, "%sin_sync",sep);
1521 if (!test_bit(Faulty, &rdev->flags) &&
1522 !test_bit(In_sync, &rdev->flags)) {
1523 len += sprintf(page+len, "%sspare", sep);
1526 return len+sprintf(page+len, "\n");
1529 static struct rdev_sysfs_entry
1530 rdev_state = __ATTR_RO(state);
1533 super_show(mdk_rdev_t *rdev, char *page)
1535 if (rdev->sb_loaded && rdev->sb_size) {
1536 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1537 return rdev->sb_size;
1541 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1543 static struct attribute *rdev_default_attrs[] = {
1549 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1551 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1552 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1556 return entry->show(rdev, page);
1560 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1561 const char *page, size_t length)
1563 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1564 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1568 return entry->store(rdev, page, length);
1571 static void rdev_free(struct kobject *ko)
1573 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1576 static struct sysfs_ops rdev_sysfs_ops = {
1577 .show = rdev_attr_show,
1578 .store = rdev_attr_store,
1580 static struct kobj_type rdev_ktype = {
1581 .release = rdev_free,
1582 .sysfs_ops = &rdev_sysfs_ops,
1583 .default_attrs = rdev_default_attrs,
1587 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1589 * mark the device faulty if:
1591 * - the device is nonexistent (zero size)
1592 * - the device has no valid superblock
1594 * a faulty rdev _never_ has rdev->sb set.
1596 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1598 char b[BDEVNAME_SIZE];
1603 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1605 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1606 return ERR_PTR(-ENOMEM);
1608 memset(rdev, 0, sizeof(*rdev));
1610 if ((err = alloc_disk_sb(rdev)))
1613 err = lock_rdev(rdev, newdev);
1617 rdev->kobj.parent = NULL;
1618 rdev->kobj.ktype = &rdev_ktype;
1619 kobject_init(&rdev->kobj);
1623 rdev->data_offset = 0;
1624 atomic_set(&rdev->nr_pending, 0);
1625 atomic_set(&rdev->read_errors, 0);
1627 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1630 "md: %s has zero or unknown size, marking faulty!\n",
1631 bdevname(rdev->bdev,b));
1636 if (super_format >= 0) {
1637 err = super_types[super_format].
1638 load_super(rdev, NULL, super_minor);
1639 if (err == -EINVAL) {
1641 "md: %s has invalid sb, not importing!\n",
1642 bdevname(rdev->bdev,b));
1647 "md: could not read %s's sb, not importing!\n",
1648 bdevname(rdev->bdev,b));
1652 INIT_LIST_HEAD(&rdev->same_set);
1657 if (rdev->sb_page) {
1663 return ERR_PTR(err);
1667 * Check a full RAID array for plausibility
1671 static void analyze_sbs(mddev_t * mddev)
1674 struct list_head *tmp;
1675 mdk_rdev_t *rdev, *freshest;
1676 char b[BDEVNAME_SIZE];
1679 ITERATE_RDEV(mddev,rdev,tmp)
1680 switch (super_types[mddev->major_version].
1681 load_super(rdev, freshest, mddev->minor_version)) {
1689 "md: fatal superblock inconsistency in %s"
1690 " -- removing from array\n",
1691 bdevname(rdev->bdev,b));
1692 kick_rdev_from_array(rdev);
1696 super_types[mddev->major_version].
1697 validate_super(mddev, freshest);
1700 ITERATE_RDEV(mddev,rdev,tmp) {
1701 if (rdev != freshest)
1702 if (super_types[mddev->major_version].
1703 validate_super(mddev, rdev)) {
1704 printk(KERN_WARNING "md: kicking non-fresh %s"
1706 bdevname(rdev->bdev,b));
1707 kick_rdev_from_array(rdev);
1710 if (mddev->level == LEVEL_MULTIPATH) {
1711 rdev->desc_nr = i++;
1712 rdev->raid_disk = rdev->desc_nr;
1713 set_bit(In_sync, &rdev->flags);
1719 if (mddev->recovery_cp != MaxSector &&
1721 printk(KERN_ERR "md: %s: raid array is not clean"
1722 " -- starting background reconstruction\n",
1728 level_show(mddev_t *mddev, char *page)
1730 mdk_personality_t *p = mddev->pers;
1731 if (p == NULL && mddev->raid_disks == 0)
1733 if (mddev->level >= 0)
1734 return sprintf(page, "raid%d\n", mddev->level);
1736 return sprintf(page, "%s\n", p->name);
1739 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1742 raid_disks_show(mddev_t *mddev, char *page)
1744 if (mddev->raid_disks == 0)
1746 return sprintf(page, "%d\n", mddev->raid_disks);
1749 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1752 action_show(mddev_t *mddev, char *page)
1754 char *type = "idle";
1755 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1756 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1757 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1758 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1760 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1767 return sprintf(page, "%s\n", type);
1771 action_store(mddev_t *mddev, const char *page, size_t len)
1773 if (!mddev->pers || !mddev->pers->sync_request)
1776 if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1777 if (mddev->sync_thread) {
1778 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1779 md_unregister_thread(mddev->sync_thread);
1780 mddev->sync_thread = NULL;
1781 mddev->recovery = 0;
1786 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1787 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1789 if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1790 strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1791 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1793 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1794 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1795 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1797 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1798 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1799 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1801 md_wakeup_thread(mddev->thread);
1806 mismatch_cnt_show(mddev_t *mddev, char *page)
1808 return sprintf(page, "%llu\n",
1809 (unsigned long long) mddev->resync_mismatches);
1812 static struct md_sysfs_entry
1813 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1816 static struct md_sysfs_entry
1817 md_mismatches = __ATTR_RO(mismatch_cnt);
1819 static struct attribute *md_default_attrs[] = {
1821 &md_raid_disks.attr,
1825 static struct attribute *md_redundancy_attrs[] = {
1827 &md_mismatches.attr,
1830 static struct attribute_group md_redundancy_group = {
1832 .attrs = md_redundancy_attrs,
1837 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1839 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1840 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1846 rv = entry->show(mddev, page);
1847 mddev_unlock(mddev);
1852 md_attr_store(struct kobject *kobj, struct attribute *attr,
1853 const char *page, size_t length)
1855 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1856 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1862 rv = entry->store(mddev, page, length);
1863 mddev_unlock(mddev);
1867 static void md_free(struct kobject *ko)
1869 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1873 static struct sysfs_ops md_sysfs_ops = {
1874 .show = md_attr_show,
1875 .store = md_attr_store,
1877 static struct kobj_type md_ktype = {
1879 .sysfs_ops = &md_sysfs_ops,
1880 .default_attrs = md_default_attrs,
1885 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1887 static DECLARE_MUTEX(disks_sem);
1888 mddev_t *mddev = mddev_find(dev);
1889 struct gendisk *disk;
1890 int partitioned = (MAJOR(dev) != MD_MAJOR);
1891 int shift = partitioned ? MdpMinorShift : 0;
1892 int unit = MINOR(dev) >> shift;
1898 if (mddev->gendisk) {
1903 disk = alloc_disk(1 << shift);
1909 disk->major = MAJOR(dev);
1910 disk->first_minor = unit << shift;
1912 sprintf(disk->disk_name, "md_d%d", unit);
1913 sprintf(disk->devfs_name, "md/d%d", unit);
1915 sprintf(disk->disk_name, "md%d", unit);
1916 sprintf(disk->devfs_name, "md/%d", unit);
1918 disk->fops = &md_fops;
1919 disk->private_data = mddev;
1920 disk->queue = mddev->queue;
1922 mddev->gendisk = disk;
1924 mddev->kobj.parent = &disk->kobj;
1925 mddev->kobj.k_name = NULL;
1926 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1927 mddev->kobj.ktype = &md_ktype;
1928 kobject_register(&mddev->kobj);
1932 void md_wakeup_thread(mdk_thread_t *thread);
1934 static void md_safemode_timeout(unsigned long data)
1936 mddev_t *mddev = (mddev_t *) data;
1938 mddev->safemode = 1;
1939 md_wakeup_thread(mddev->thread);
1942 static int start_dirty_degraded;
1944 static int do_md_run(mddev_t * mddev)
1948 struct list_head *tmp;
1950 struct gendisk *disk;
1951 char b[BDEVNAME_SIZE];
1953 if (list_empty(&mddev->disks))
1954 /* cannot run an array with no devices.. */
1961 * Analyze all RAID superblock(s)
1963 if (!mddev->raid_disks)
1966 chunk_size = mddev->chunk_size;
1967 pnum = level_to_pers(mddev->level);
1969 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1972 * 'default chunksize' in the old md code used to
1973 * be PAGE_SIZE, baaad.
1974 * we abort here to be on the safe side. We don't
1975 * want to continue the bad practice.
1978 "no chunksize specified, see 'man raidtab'\n");
1981 if (chunk_size > MAX_CHUNK_SIZE) {
1982 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1983 chunk_size, MAX_CHUNK_SIZE);
1987 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1989 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1990 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1993 if (chunk_size < PAGE_SIZE) {
1994 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1995 chunk_size, PAGE_SIZE);
1999 /* devices must have minimum size of one chunk */
2000 ITERATE_RDEV(mddev,rdev,tmp) {
2001 if (test_bit(Faulty, &rdev->flags))
2003 if (rdev->size < chunk_size / 1024) {
2005 "md: Dev %s smaller than chunk_size:"
2007 bdevname(rdev->bdev,b),
2008 (unsigned long long)rdev->size,
2018 request_module("md-personality-%d", pnum);
2023 * Drop all container device buffers, from now on
2024 * the only valid external interface is through the md
2026 * Also find largest hardsector size
2028 ITERATE_RDEV(mddev,rdev,tmp) {
2029 if (test_bit(Faulty, &rdev->flags))
2031 sync_blockdev(rdev->bdev);
2032 invalidate_bdev(rdev->bdev, 0);
2035 md_probe(mddev->unit, NULL, NULL);
2036 disk = mddev->gendisk;
2040 spin_lock(&pers_lock);
2041 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2042 spin_unlock(&pers_lock);
2043 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2048 mddev->pers = pers[pnum];
2049 spin_unlock(&pers_lock);
2051 mddev->recovery = 0;
2052 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2053 mddev->barriers_work = 1;
2054 mddev->ok_start_degraded = start_dirty_degraded;
2057 mddev->ro = 2; /* read-only, but switch on first write */
2059 err = mddev->pers->run(mddev);
2060 if (!err && mddev->pers->sync_request) {
2061 err = bitmap_create(mddev);
2063 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2064 mdname(mddev), err);
2065 mddev->pers->stop(mddev);
2069 printk(KERN_ERR "md: pers->run() failed ...\n");
2070 module_put(mddev->pers->owner);
2072 bitmap_destroy(mddev);
2075 if (mddev->pers->sync_request)
2076 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2077 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2080 atomic_set(&mddev->writes_pending,0);
2081 mddev->safemode = 0;
2082 mddev->safemode_timer.function = md_safemode_timeout;
2083 mddev->safemode_timer.data = (unsigned long) mddev;
2084 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2087 ITERATE_RDEV(mddev,rdev,tmp)
2088 if (rdev->raid_disk >= 0) {
2090 sprintf(nm, "rd%d", rdev->raid_disk);
2091 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2094 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2095 md_wakeup_thread(mddev->thread);
2097 if (mddev->sb_dirty)
2098 md_update_sb(mddev);
2100 set_capacity(disk, mddev->array_size<<1);
2102 /* If we call blk_queue_make_request here, it will
2103 * re-initialise max_sectors etc which may have been
2104 * refined inside -> run. So just set the bits we need to set.
2105 * Most initialisation happended when we called
2106 * blk_queue_make_request(..., md_fail_request)
2109 mddev->queue->queuedata = mddev;
2110 mddev->queue->make_request_fn = mddev->pers->make_request;
2116 static int restart_array(mddev_t *mddev)
2118 struct gendisk *disk = mddev->gendisk;
2122 * Complain if it has no devices
2125 if (list_empty(&mddev->disks))
2133 mddev->safemode = 0;
2135 set_disk_ro(disk, 0);
2137 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2140 * Kick recovery or resync if necessary
2142 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2143 md_wakeup_thread(mddev->thread);
2146 printk(KERN_ERR "md: %s has no personality assigned.\n",
2155 static int do_md_stop(mddev_t * mddev, int ro)
2158 struct gendisk *disk = mddev->gendisk;
2161 if (atomic_read(&mddev->active)>2) {
2162 printk("md: %s still in use.\n",mdname(mddev));
2166 if (mddev->sync_thread) {
2167 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2168 md_unregister_thread(mddev->sync_thread);
2169 mddev->sync_thread = NULL;
2172 del_timer_sync(&mddev->safemode_timer);
2174 invalidate_partition(disk, 0);
2182 bitmap_flush(mddev);
2183 md_super_wait(mddev);
2185 set_disk_ro(disk, 0);
2186 blk_queue_make_request(mddev->queue, md_fail_request);
2187 mddev->pers->stop(mddev);
2188 if (mddev->pers->sync_request)
2189 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2191 module_put(mddev->pers->owner);
2196 if (!mddev->in_sync) {
2197 /* mark array as shutdown cleanly */
2199 md_update_sb(mddev);
2202 set_disk_ro(disk, 1);
2205 bitmap_destroy(mddev);
2206 if (mddev->bitmap_file) {
2207 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2208 fput(mddev->bitmap_file);
2209 mddev->bitmap_file = NULL;
2211 mddev->bitmap_offset = 0;
2214 * Free resources if final stop
2218 struct list_head *tmp;
2219 struct gendisk *disk;
2220 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2222 ITERATE_RDEV(mddev,rdev,tmp)
2223 if (rdev->raid_disk >= 0) {
2225 sprintf(nm, "rd%d", rdev->raid_disk);
2226 sysfs_remove_link(&mddev->kobj, nm);
2229 export_array(mddev);
2231 mddev->array_size = 0;
2232 disk = mddev->gendisk;
2234 set_capacity(disk, 0);
2237 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2244 static void autorun_array(mddev_t *mddev)
2247 struct list_head *tmp;
2250 if (list_empty(&mddev->disks))
2253 printk(KERN_INFO "md: running: ");
2255 ITERATE_RDEV(mddev,rdev,tmp) {
2256 char b[BDEVNAME_SIZE];
2257 printk("<%s>", bdevname(rdev->bdev,b));
2261 err = do_md_run (mddev);
2263 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2264 do_md_stop (mddev, 0);
2269 * lets try to run arrays based on all disks that have arrived
2270 * until now. (those are in pending_raid_disks)
2272 * the method: pick the first pending disk, collect all disks with
2273 * the same UUID, remove all from the pending list and put them into
2274 * the 'same_array' list. Then order this list based on superblock
2275 * update time (freshest comes first), kick out 'old' disks and
2276 * compare superblocks. If everything's fine then run it.
2278 * If "unit" is allocated, then bump its reference count
2280 static void autorun_devices(int part)
2282 struct list_head candidates;
2283 struct list_head *tmp;
2284 mdk_rdev_t *rdev0, *rdev;
2286 char b[BDEVNAME_SIZE];
2288 printk(KERN_INFO "md: autorun ...\n");
2289 while (!list_empty(&pending_raid_disks)) {
2291 rdev0 = list_entry(pending_raid_disks.next,
2292 mdk_rdev_t, same_set);
2294 printk(KERN_INFO "md: considering %s ...\n",
2295 bdevname(rdev0->bdev,b));
2296 INIT_LIST_HEAD(&candidates);
2297 ITERATE_RDEV_PENDING(rdev,tmp)
2298 if (super_90_load(rdev, rdev0, 0) >= 0) {
2299 printk(KERN_INFO "md: adding %s ...\n",
2300 bdevname(rdev->bdev,b));
2301 list_move(&rdev->same_set, &candidates);
2304 * now we have a set of devices, with all of them having
2305 * mostly sane superblocks. It's time to allocate the
2308 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2309 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2310 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2314 dev = MKDEV(mdp_major,
2315 rdev0->preferred_minor << MdpMinorShift);
2317 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2319 md_probe(dev, NULL, NULL);
2320 mddev = mddev_find(dev);
2323 "md: cannot allocate memory for md drive.\n");
2326 if (mddev_lock(mddev))
2327 printk(KERN_WARNING "md: %s locked, cannot run\n",
2329 else if (mddev->raid_disks || mddev->major_version
2330 || !list_empty(&mddev->disks)) {
2332 "md: %s already running, cannot run %s\n",
2333 mdname(mddev), bdevname(rdev0->bdev,b));
2334 mddev_unlock(mddev);
2336 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2337 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2338 list_del_init(&rdev->same_set);
2339 if (bind_rdev_to_array(rdev, mddev))
2342 autorun_array(mddev);
2343 mddev_unlock(mddev);
2345 /* on success, candidates will be empty, on error
2348 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2352 printk(KERN_INFO "md: ... autorun DONE.\n");
2356 * import RAID devices based on one partition
2357 * if possible, the array gets run as well.
2360 static int autostart_array(dev_t startdev)
2362 char b[BDEVNAME_SIZE];
2363 int err = -EINVAL, i;
2364 mdp_super_t *sb = NULL;
2365 mdk_rdev_t *start_rdev = NULL, *rdev;
2367 start_rdev = md_import_device(startdev, 0, 0);
2368 if (IS_ERR(start_rdev))
2372 /* NOTE: this can only work for 0.90.0 superblocks */
2373 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2374 if (sb->major_version != 0 ||
2375 sb->minor_version != 90 ) {
2376 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2377 export_rdev(start_rdev);
2381 if (test_bit(Faulty, &start_rdev->flags)) {
2383 "md: can not autostart based on faulty %s!\n",
2384 bdevname(start_rdev->bdev,b));
2385 export_rdev(start_rdev);
2388 list_add(&start_rdev->same_set, &pending_raid_disks);
2390 for (i = 0; i < MD_SB_DISKS; i++) {
2391 mdp_disk_t *desc = sb->disks + i;
2392 dev_t dev = MKDEV(desc->major, desc->minor);
2396 if (dev == startdev)
2398 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2400 rdev = md_import_device(dev, 0, 0);
2404 list_add(&rdev->same_set, &pending_raid_disks);
2408 * possibly return codes
2416 static int get_version(void __user * arg)
2420 ver.major = MD_MAJOR_VERSION;
2421 ver.minor = MD_MINOR_VERSION;
2422 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2424 if (copy_to_user(arg, &ver, sizeof(ver)))
2430 static int get_array_info(mddev_t * mddev, void __user * arg)
2432 mdu_array_info_t info;
2433 int nr,working,active,failed,spare;
2435 struct list_head *tmp;
2437 nr=working=active=failed=spare=0;
2438 ITERATE_RDEV(mddev,rdev,tmp) {
2440 if (test_bit(Faulty, &rdev->flags))
2444 if (test_bit(In_sync, &rdev->flags))
2451 info.major_version = mddev->major_version;
2452 info.minor_version = mddev->minor_version;
2453 info.patch_version = MD_PATCHLEVEL_VERSION;
2454 info.ctime = mddev->ctime;
2455 info.level = mddev->level;
2456 info.size = mddev->size;
2458 info.raid_disks = mddev->raid_disks;
2459 info.md_minor = mddev->md_minor;
2460 info.not_persistent= !mddev->persistent;
2462 info.utime = mddev->utime;
2465 info.state = (1<<MD_SB_CLEAN);
2466 if (mddev->bitmap && mddev->bitmap_offset)
2467 info.state = (1<<MD_SB_BITMAP_PRESENT);
2468 info.active_disks = active;
2469 info.working_disks = working;
2470 info.failed_disks = failed;
2471 info.spare_disks = spare;
2473 info.layout = mddev->layout;
2474 info.chunk_size = mddev->chunk_size;
2476 if (copy_to_user(arg, &info, sizeof(info)))
2482 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2484 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2485 char *ptr, *buf = NULL;
2488 file = kmalloc(sizeof(*file), GFP_KERNEL);
2492 /* bitmap disabled, zero the first byte and copy out */
2493 if (!mddev->bitmap || !mddev->bitmap->file) {
2494 file->pathname[0] = '\0';
2498 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2502 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2506 strcpy(file->pathname, ptr);
2510 if (copy_to_user(arg, file, sizeof(*file)))
2518 static int get_disk_info(mddev_t * mddev, void __user * arg)
2520 mdu_disk_info_t info;
2524 if (copy_from_user(&info, arg, sizeof(info)))
2529 rdev = find_rdev_nr(mddev, nr);
2531 info.major = MAJOR(rdev->bdev->bd_dev);
2532 info.minor = MINOR(rdev->bdev->bd_dev);
2533 info.raid_disk = rdev->raid_disk;
2535 if (test_bit(Faulty, &rdev->flags))
2536 info.state |= (1<<MD_DISK_FAULTY);
2537 else if (test_bit(In_sync, &rdev->flags)) {
2538 info.state |= (1<<MD_DISK_ACTIVE);
2539 info.state |= (1<<MD_DISK_SYNC);
2541 if (test_bit(WriteMostly, &rdev->flags))
2542 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2544 info.major = info.minor = 0;
2545 info.raid_disk = -1;
2546 info.state = (1<<MD_DISK_REMOVED);
2549 if (copy_to_user(arg, &info, sizeof(info)))
2555 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2557 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2559 dev_t dev = MKDEV(info->major,info->minor);
2561 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2564 if (!mddev->raid_disks) {
2566 /* expecting a device which has a superblock */
2567 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2570 "md: md_import_device returned %ld\n",
2572 return PTR_ERR(rdev);
2574 if (!list_empty(&mddev->disks)) {
2575 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2576 mdk_rdev_t, same_set);
2577 int err = super_types[mddev->major_version]
2578 .load_super(rdev, rdev0, mddev->minor_version);
2581 "md: %s has different UUID to %s\n",
2582 bdevname(rdev->bdev,b),
2583 bdevname(rdev0->bdev,b2));
2588 err = bind_rdev_to_array(rdev, mddev);
2595 * add_new_disk can be used once the array is assembled
2596 * to add "hot spares". They must already have a superblock
2601 if (!mddev->pers->hot_add_disk) {
2603 "%s: personality does not support diskops!\n",
2607 if (mddev->persistent)
2608 rdev = md_import_device(dev, mddev->major_version,
2609 mddev->minor_version);
2611 rdev = md_import_device(dev, -1, -1);
2614 "md: md_import_device returned %ld\n",
2616 return PTR_ERR(rdev);
2618 /* set save_raid_disk if appropriate */
2619 if (!mddev->persistent) {
2620 if (info->state & (1<<MD_DISK_SYNC) &&
2621 info->raid_disk < mddev->raid_disks)
2622 rdev->raid_disk = info->raid_disk;
2624 rdev->raid_disk = -1;
2626 super_types[mddev->major_version].
2627 validate_super(mddev, rdev);
2628 rdev->saved_raid_disk = rdev->raid_disk;
2630 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2631 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2632 set_bit(WriteMostly, &rdev->flags);
2634 rdev->raid_disk = -1;
2635 err = bind_rdev_to_array(rdev, mddev);
2639 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2640 md_wakeup_thread(mddev->thread);
2644 /* otherwise, add_new_disk is only allowed
2645 * for major_version==0 superblocks
2647 if (mddev->major_version != 0) {
2648 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2653 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2655 rdev = md_import_device (dev, -1, 0);
2658 "md: error, md_import_device() returned %ld\n",
2660 return PTR_ERR(rdev);
2662 rdev->desc_nr = info->number;
2663 if (info->raid_disk < mddev->raid_disks)
2664 rdev->raid_disk = info->raid_disk;
2666 rdev->raid_disk = -1;
2670 if (rdev->raid_disk < mddev->raid_disks)
2671 if (info->state & (1<<MD_DISK_SYNC))
2672 set_bit(In_sync, &rdev->flags);
2674 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2675 set_bit(WriteMostly, &rdev->flags);
2677 err = bind_rdev_to_array(rdev, mddev);
2683 if (!mddev->persistent) {
2684 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2685 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2687 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2688 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2690 if (!mddev->size || (mddev->size > rdev->size))
2691 mddev->size = rdev->size;
2697 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2699 char b[BDEVNAME_SIZE];
2705 rdev = find_rdev(mddev, dev);
2709 if (rdev->raid_disk >= 0)
2712 kick_rdev_from_array(rdev);
2713 md_update_sb(mddev);
2717 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2718 bdevname(rdev->bdev,b), mdname(mddev));
2722 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2724 char b[BDEVNAME_SIZE];
2732 if (mddev->major_version != 0) {
2733 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2734 " version-0 superblocks.\n",
2738 if (!mddev->pers->hot_add_disk) {
2740 "%s: personality does not support diskops!\n",
2745 rdev = md_import_device (dev, -1, 0);
2748 "md: error, md_import_device() returned %ld\n",
2753 if (mddev->persistent)
2754 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2757 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2759 size = calc_dev_size(rdev, mddev->chunk_size);
2762 if (size < mddev->size) {
2764 "%s: disk size %llu blocks < array size %llu\n",
2765 mdname(mddev), (unsigned long long)size,
2766 (unsigned long long)mddev->size);
2771 if (test_bit(Faulty, &rdev->flags)) {
2773 "md: can not hot-add faulty %s disk to %s!\n",
2774 bdevname(rdev->bdev,b), mdname(mddev));
2778 clear_bit(In_sync, &rdev->flags);
2780 bind_rdev_to_array(rdev, mddev);
2783 * The rest should better be atomic, we can have disk failures
2784 * noticed in interrupt contexts ...
2787 if (rdev->desc_nr == mddev->max_disks) {
2788 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2791 goto abort_unbind_export;
2794 rdev->raid_disk = -1;
2796 md_update_sb(mddev);
2799 * Kick recovery, maybe this spare has to be added to the
2800 * array immediately.
2802 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2803 md_wakeup_thread(mddev->thread);
2807 abort_unbind_export:
2808 unbind_rdev_from_array(rdev);
2815 /* similar to deny_write_access, but accounts for our holding a reference
2816 * to the file ourselves */
2817 static int deny_bitmap_write_access(struct file * file)
2819 struct inode *inode = file->f_mapping->host;
2821 spin_lock(&inode->i_lock);
2822 if (atomic_read(&inode->i_writecount) > 1) {
2823 spin_unlock(&inode->i_lock);
2826 atomic_set(&inode->i_writecount, -1);
2827 spin_unlock(&inode->i_lock);
2832 static int set_bitmap_file(mddev_t *mddev, int fd)
2837 if (!mddev->pers->quiesce)
2839 if (mddev->recovery || mddev->sync_thread)
2841 /* we should be able to change the bitmap.. */
2847 return -EEXIST; /* cannot add when bitmap is present */
2848 mddev->bitmap_file = fget(fd);
2850 if (mddev->bitmap_file == NULL) {
2851 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2856 err = deny_bitmap_write_access(mddev->bitmap_file);
2858 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2860 fput(mddev->bitmap_file);
2861 mddev->bitmap_file = NULL;
2864 mddev->bitmap_offset = 0; /* file overrides offset */
2865 } else if (mddev->bitmap == NULL)
2866 return -ENOENT; /* cannot remove what isn't there */
2869 mddev->pers->quiesce(mddev, 1);
2871 err = bitmap_create(mddev);
2873 bitmap_destroy(mddev);
2874 mddev->pers->quiesce(mddev, 0);
2875 } else if (fd < 0) {
2876 if (mddev->bitmap_file)
2877 fput(mddev->bitmap_file);
2878 mddev->bitmap_file = NULL;
2885 * set_array_info is used two different ways
2886 * The original usage is when creating a new array.
2887 * In this usage, raid_disks is > 0 and it together with
2888 * level, size, not_persistent,layout,chunksize determine the
2889 * shape of the array.
2890 * This will always create an array with a type-0.90.0 superblock.
2891 * The newer usage is when assembling an array.
2892 * In this case raid_disks will be 0, and the major_version field is
2893 * use to determine which style super-blocks are to be found on the devices.
2894 * The minor and patch _version numbers are also kept incase the
2895 * super_block handler wishes to interpret them.
2897 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2900 if (info->raid_disks == 0) {
2901 /* just setting version number for superblock loading */
2902 if (info->major_version < 0 ||
2903 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2904 super_types[info->major_version].name == NULL) {
2905 /* maybe try to auto-load a module? */
2907 "md: superblock version %d not known\n",
2908 info->major_version);
2911 mddev->major_version = info->major_version;
2912 mddev->minor_version = info->minor_version;
2913 mddev->patch_version = info->patch_version;
2916 mddev->major_version = MD_MAJOR_VERSION;
2917 mddev->minor_version = MD_MINOR_VERSION;
2918 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2919 mddev->ctime = get_seconds();
2921 mddev->level = info->level;
2922 mddev->size = info->size;
2923 mddev->raid_disks = info->raid_disks;
2924 /* don't set md_minor, it is determined by which /dev/md* was
2927 if (info->state & (1<<MD_SB_CLEAN))
2928 mddev->recovery_cp = MaxSector;
2930 mddev->recovery_cp = 0;
2931 mddev->persistent = ! info->not_persistent;
2933 mddev->layout = info->layout;
2934 mddev->chunk_size = info->chunk_size;
2936 mddev->max_disks = MD_SB_DISKS;
2938 mddev->sb_dirty = 1;
2940 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2941 mddev->bitmap_offset = 0;
2944 * Generate a 128 bit UUID
2946 get_random_bytes(mddev->uuid, 16);
2952 * update_array_info is used to change the configuration of an
2954 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2955 * fields in the info are checked against the array.
2956 * Any differences that cannot be handled will cause an error.
2957 * Normally, only one change can be managed at a time.
2959 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2965 /* calculate expected state,ignoring low bits */
2966 if (mddev->bitmap && mddev->bitmap_offset)
2967 state |= (1 << MD_SB_BITMAP_PRESENT);
2969 if (mddev->major_version != info->major_version ||
2970 mddev->minor_version != info->minor_version ||
2971 /* mddev->patch_version != info->patch_version || */
2972 mddev->ctime != info->ctime ||
2973 mddev->level != info->level ||
2974 /* mddev->layout != info->layout || */
2975 !mddev->persistent != info->not_persistent||
2976 mddev->chunk_size != info->chunk_size ||
2977 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2978 ((state^info->state) & 0xfffffe00)
2981 /* Check there is only one change */
2982 if (mddev->size != info->size) cnt++;
2983 if (mddev->raid_disks != info->raid_disks) cnt++;
2984 if (mddev->layout != info->layout) cnt++;
2985 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2986 if (cnt == 0) return 0;
2987 if (cnt > 1) return -EINVAL;
2989 if (mddev->layout != info->layout) {
2991 * we don't need to do anything at the md level, the
2992 * personality will take care of it all.
2994 if (mddev->pers->reconfig == NULL)
2997 return mddev->pers->reconfig(mddev, info->layout, -1);
2999 if (mddev->size != info->size) {
3001 struct list_head *tmp;
3002 if (mddev->pers->resize == NULL)
3004 /* The "size" is the amount of each device that is used.
3005 * This can only make sense for arrays with redundancy.
3006 * linear and raid0 always use whatever space is available
3007 * We can only consider changing the size if no resync
3008 * or reconstruction is happening, and if the new size
3009 * is acceptable. It must fit before the sb_offset or,
3010 * if that is <data_offset, it must fit before the
3011 * size of each device.
3012 * If size is zero, we find the largest size that fits.
3014 if (mddev->sync_thread)
3016 ITERATE_RDEV(mddev,rdev,tmp) {
3018 int fit = (info->size == 0);
3019 if (rdev->sb_offset > rdev->data_offset)
3020 avail = (rdev->sb_offset*2) - rdev->data_offset;
3022 avail = get_capacity(rdev->bdev->bd_disk)
3023 - rdev->data_offset;
3024 if (fit && (info->size == 0 || info->size > avail/2))
3025 info->size = avail/2;
3026 if (avail < ((sector_t)info->size << 1))
3029 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3031 struct block_device *bdev;
3033 bdev = bdget_disk(mddev->gendisk, 0);
3035 down(&bdev->bd_inode->i_sem);
3036 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3037 up(&bdev->bd_inode->i_sem);
3042 if (mddev->raid_disks != info->raid_disks) {
3043 /* change the number of raid disks */
3044 if (mddev->pers->reshape == NULL)
3046 if (info->raid_disks <= 0 ||
3047 info->raid_disks >= mddev->max_disks)
3049 if (mddev->sync_thread)
3051 rv = mddev->pers->reshape(mddev, info->raid_disks);
3053 struct block_device *bdev;
3055 bdev = bdget_disk(mddev->gendisk, 0);
3057 down(&bdev->bd_inode->i_sem);
3058 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3059 up(&bdev->bd_inode->i_sem);
3064 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3065 if (mddev->pers->quiesce == NULL)
3067 if (mddev->recovery || mddev->sync_thread)
3069 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3070 /* add the bitmap */
3073 if (mddev->default_bitmap_offset == 0)
3075 mddev->bitmap_offset = mddev->default_bitmap_offset;
3076 mddev->pers->quiesce(mddev, 1);
3077 rv = bitmap_create(mddev);
3079 bitmap_destroy(mddev);
3080 mddev->pers->quiesce(mddev, 0);
3082 /* remove the bitmap */
3085 if (mddev->bitmap->file)
3087 mddev->pers->quiesce(mddev, 1);
3088 bitmap_destroy(mddev);
3089 mddev->pers->quiesce(mddev, 0);
3090 mddev->bitmap_offset = 0;
3093 md_update_sb(mddev);
3097 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3101 if (mddev->pers == NULL)
3104 rdev = find_rdev(mddev, dev);
3108 md_error(mddev, rdev);
3112 static int md_ioctl(struct inode *inode, struct file *file,
3113 unsigned int cmd, unsigned long arg)
3116 void __user *argp = (void __user *)arg;
3117 struct hd_geometry __user *loc = argp;
3118 mddev_t *mddev = NULL;
3120 if (!capable(CAP_SYS_ADMIN))
3124 * Commands dealing with the RAID driver but not any
3130 err = get_version(argp);
3133 case PRINT_RAID_DEBUG:
3141 autostart_arrays(arg);
3148 * Commands creating/starting a new array:
3151 mddev = inode->i_bdev->bd_disk->private_data;
3159 if (cmd == START_ARRAY) {
3160 /* START_ARRAY doesn't need to lock the array as autostart_array
3161 * does the locking, and it could even be a different array
3166 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3167 "This will not be supported beyond July 2006\n",
3168 current->comm, current->pid);
3171 err = autostart_array(new_decode_dev(arg));
3173 printk(KERN_WARNING "md: autostart failed!\n");
3179 err = mddev_lock(mddev);
3182 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3189 case SET_ARRAY_INFO:
3191 mdu_array_info_t info;
3193 memset(&info, 0, sizeof(info));
3194 else if (copy_from_user(&info, argp, sizeof(info))) {
3199 err = update_array_info(mddev, &info);
3201 printk(KERN_WARNING "md: couldn't update"
3202 " array info. %d\n", err);
3207 if (!list_empty(&mddev->disks)) {
3209 "md: array %s already has disks!\n",
3214 if (mddev->raid_disks) {
3216 "md: array %s already initialised!\n",
3221 err = set_array_info(mddev, &info);
3223 printk(KERN_WARNING "md: couldn't set"
3224 " array info. %d\n", err);
3234 * Commands querying/configuring an existing array:
3236 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3237 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3238 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3239 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3245 * Commands even a read-only array can execute:
3249 case GET_ARRAY_INFO:
3250 err = get_array_info(mddev, argp);
3253 case GET_BITMAP_FILE:
3254 err = get_bitmap_file(mddev, argp);
3258 err = get_disk_info(mddev, argp);
3261 case RESTART_ARRAY_RW:
3262 err = restart_array(mddev);
3266 err = do_md_stop (mddev, 0);
3270 err = do_md_stop (mddev, 1);
3274 * We have a problem here : there is no easy way to give a CHS
3275 * virtual geometry. We currently pretend that we have a 2 heads
3276 * 4 sectors (with a BIG number of cylinders...). This drives
3277 * dosfs just mad... ;-)
3284 err = put_user (2, (char __user *) &loc->heads);
3287 err = put_user (4, (char __user *) &loc->sectors);
3290 err = put_user(get_capacity(mddev->gendisk)/8,
3291 (short __user *) &loc->cylinders);
3294 err = put_user (get_start_sect(inode->i_bdev),
3295 (long __user *) &loc->start);
3300 * The remaining ioctls are changing the state of the
3301 * superblock, so we do not allow them on read-only arrays.
3302 * However non-MD ioctls (e.g. get-size) will still come through
3303 * here and hit the 'default' below, so only disallow
3304 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3306 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3307 mddev->ro && mddev->pers) {
3308 if (mddev->ro == 2) {
3310 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3311 md_wakeup_thread(mddev->thread);
3323 mdu_disk_info_t info;
3324 if (copy_from_user(&info, argp, sizeof(info)))
3327 err = add_new_disk(mddev, &info);
3331 case HOT_REMOVE_DISK:
3332 err = hot_remove_disk(mddev, new_decode_dev(arg));
3336 err = hot_add_disk(mddev, new_decode_dev(arg));
3339 case SET_DISK_FAULTY:
3340 err = set_disk_faulty(mddev, new_decode_dev(arg));
3344 err = do_md_run (mddev);
3347 case SET_BITMAP_FILE:
3348 err = set_bitmap_file(mddev, (int)arg);
3352 if (_IOC_TYPE(cmd) == MD_MAJOR)
3353 printk(KERN_WARNING "md: %s(pid %d) used"
3354 " obsolete MD ioctl, upgrade your"
3355 " software to use new ictls.\n",
3356 current->comm, current->pid);
3363 mddev_unlock(mddev);
3373 static int md_open(struct inode *inode, struct file *file)
3376 * Succeed if we can lock the mddev, which confirms that
3377 * it isn't being stopped right now.
3379 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3382 if ((err = mddev_lock(mddev)))
3387 mddev_unlock(mddev);
3389 check_disk_change(inode->i_bdev);
3394 static int md_release(struct inode *inode, struct file * file)
3396 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3405 static int md_media_changed(struct gendisk *disk)
3407 mddev_t *mddev = disk->private_data;
3409 return mddev->changed;
3412 static int md_revalidate(struct gendisk *disk)
3414 mddev_t *mddev = disk->private_data;
3419 static struct block_device_operations md_fops =
3421 .owner = THIS_MODULE,
3423 .release = md_release,
3425 .media_changed = md_media_changed,
3426 .revalidate_disk= md_revalidate,
3429 static int md_thread(void * arg)
3431 mdk_thread_t *thread = arg;
3434 * md_thread is a 'system-thread', it's priority should be very
3435 * high. We avoid resource deadlocks individually in each
3436 * raid personality. (RAID5 does preallocation) We also use RR and
3437 * the very same RT priority as kswapd, thus we will never get
3438 * into a priority inversion deadlock.
3440 * we definitely have to have equal or higher priority than
3441 * bdflush, otherwise bdflush will deadlock if there are too
3442 * many dirty RAID5 blocks.
3445 allow_signal(SIGKILL);
3446 while (!kthread_should_stop()) {
3448 /* We need to wait INTERRUPTIBLE so that
3449 * we don't add to the load-average.
3450 * That means we need to be sure no signals are
3453 if (signal_pending(current))
3454 flush_signals(current);
3456 wait_event_interruptible_timeout
3458 test_bit(THREAD_WAKEUP, &thread->flags)
3459 || kthread_should_stop(),
3463 clear_bit(THREAD_WAKEUP, &thread->flags);
3465 thread->run(thread->mddev);
3471 void md_wakeup_thread(mdk_thread_t *thread)
3474 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3475 set_bit(THREAD_WAKEUP, &thread->flags);
3476 wake_up(&thread->wqueue);
3480 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3483 mdk_thread_t *thread;
3485 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3489 memset(thread, 0, sizeof(mdk_thread_t));
3490 init_waitqueue_head(&thread->wqueue);
3493 thread->mddev = mddev;
3494 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3495 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3496 if (IS_ERR(thread->tsk)) {
3503 void md_unregister_thread(mdk_thread_t *thread)
3505 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3507 kthread_stop(thread->tsk);
3511 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3518 if (!rdev || test_bit(Faulty, &rdev->flags))
3521 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3523 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3524 __builtin_return_address(0),__builtin_return_address(1),
3525 __builtin_return_address(2),__builtin_return_address(3));
3527 if (!mddev->pers->error_handler)
3529 mddev->pers->error_handler(mddev,rdev);
3530 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3531 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3532 md_wakeup_thread(mddev->thread);
3535 /* seq_file implementation /proc/mdstat */
3537 static void status_unused(struct seq_file *seq)
3541 struct list_head *tmp;
3543 seq_printf(seq, "unused devices: ");
3545 ITERATE_RDEV_PENDING(rdev,tmp) {
3546 char b[BDEVNAME_SIZE];
3548 seq_printf(seq, "%s ",
3549 bdevname(rdev->bdev,b));
3552 seq_printf(seq, "<none>");
3554 seq_printf(seq, "\n");
3558 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3560 unsigned long max_blocks, resync, res, dt, db, rt;
3562 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3564 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3565 max_blocks = mddev->resync_max_sectors >> 1;
3567 max_blocks = mddev->size;
3570 * Should not happen.
3576 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3578 int i, x = res/50, y = 20-x;
3579 seq_printf(seq, "[");
3580 for (i = 0; i < x; i++)
3581 seq_printf(seq, "=");
3582 seq_printf(seq, ">");
3583 for (i = 0; i < y; i++)
3584 seq_printf(seq, ".");
3585 seq_printf(seq, "] ");
3587 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3588 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3589 "resync" : "recovery"),
3590 res/10, res % 10, resync, max_blocks);
3593 * We do not want to overflow, so the order of operands and
3594 * the * 100 / 100 trick are important. We do a +1 to be
3595 * safe against division by zero. We only estimate anyway.
3597 * dt: time from mark until now
3598 * db: blocks written from mark until now
3599 * rt: remaining time
3601 dt = ((jiffies - mddev->resync_mark) / HZ);
3603 db = resync - (mddev->resync_mark_cnt/2);
3604 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3606 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3608 seq_printf(seq, " speed=%ldK/sec", db/dt);
3611 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3613 struct list_head *tmp;
3623 spin_lock(&all_mddevs_lock);
3624 list_for_each(tmp,&all_mddevs)
3626 mddev = list_entry(tmp, mddev_t, all_mddevs);
3628 spin_unlock(&all_mddevs_lock);
3631 spin_unlock(&all_mddevs_lock);
3633 return (void*)2;/* tail */
3637 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3639 struct list_head *tmp;
3640 mddev_t *next_mddev, *mddev = v;
3646 spin_lock(&all_mddevs_lock);
3648 tmp = all_mddevs.next;
3650 tmp = mddev->all_mddevs.next;
3651 if (tmp != &all_mddevs)
3652 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3654 next_mddev = (void*)2;
3657 spin_unlock(&all_mddevs_lock);
3665 static void md_seq_stop(struct seq_file *seq, void *v)
3669 if (mddev && v != (void*)1 && v != (void*)2)
3673 static int md_seq_show(struct seq_file *seq, void *v)
3677 struct list_head *tmp2;
3680 struct bitmap *bitmap;
3682 if (v == (void*)1) {
3683 seq_printf(seq, "Personalities : ");
3684 spin_lock(&pers_lock);
3685 for (i = 0; i < MAX_PERSONALITY; i++)
3687 seq_printf(seq, "[%s] ", pers[i]->name);
3689 spin_unlock(&pers_lock);
3690 seq_printf(seq, "\n");
3693 if (v == (void*)2) {
3698 if (mddev_lock(mddev)!=0)
3700 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3701 seq_printf(seq, "%s : %sactive", mdname(mddev),
3702 mddev->pers ? "" : "in");
3705 seq_printf(seq, " (read-only)");
3707 seq_printf(seq, "(auto-read-only)");
3708 seq_printf(seq, " %s", mddev->pers->name);
3712 ITERATE_RDEV(mddev,rdev,tmp2) {
3713 char b[BDEVNAME_SIZE];
3714 seq_printf(seq, " %s[%d]",
3715 bdevname(rdev->bdev,b), rdev->desc_nr);
3716 if (test_bit(WriteMostly, &rdev->flags))
3717 seq_printf(seq, "(W)");
3718 if (test_bit(Faulty, &rdev->flags)) {
3719 seq_printf(seq, "(F)");
3721 } else if (rdev->raid_disk < 0)
3722 seq_printf(seq, "(S)"); /* spare */
3726 if (!list_empty(&mddev->disks)) {
3728 seq_printf(seq, "\n %llu blocks",
3729 (unsigned long long)mddev->array_size);
3731 seq_printf(seq, "\n %llu blocks",
3732 (unsigned long long)size);
3734 if (mddev->persistent) {
3735 if (mddev->major_version != 0 ||
3736 mddev->minor_version != 90) {
3737 seq_printf(seq," super %d.%d",
3738 mddev->major_version,
3739 mddev->minor_version);
3742 seq_printf(seq, " super non-persistent");
3745 mddev->pers->status (seq, mddev);
3746 seq_printf(seq, "\n ");
3747 if (mddev->pers->sync_request) {
3748 if (mddev->curr_resync > 2) {
3749 status_resync (seq, mddev);
3750 seq_printf(seq, "\n ");
3751 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3752 seq_printf(seq, "\tresync=DELAYED\n ");
3753 else if (mddev->recovery_cp < MaxSector)
3754 seq_printf(seq, "\tresync=PENDING\n ");
3757 seq_printf(seq, "\n ");
3759 if ((bitmap = mddev->bitmap)) {
3760 unsigned long chunk_kb;
3761 unsigned long flags;
3762 spin_lock_irqsave(&bitmap->lock, flags);
3763 chunk_kb = bitmap->chunksize >> 10;
3764 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3766 bitmap->pages - bitmap->missing_pages,
3768 (bitmap->pages - bitmap->missing_pages)
3769 << (PAGE_SHIFT - 10),
3770 chunk_kb ? chunk_kb : bitmap->chunksize,
3771 chunk_kb ? "KB" : "B");
3773 seq_printf(seq, ", file: ");
3774 seq_path(seq, bitmap->file->f_vfsmnt,
3775 bitmap->file->f_dentry," \t\n");
3778 seq_printf(seq, "\n");
3779 spin_unlock_irqrestore(&bitmap->lock, flags);
3782 seq_printf(seq, "\n");
3784 mddev_unlock(mddev);
3789 static struct seq_operations md_seq_ops = {
3790 .start = md_seq_start,
3791 .next = md_seq_next,
3792 .stop = md_seq_stop,
3793 .show = md_seq_show,
3796 static int md_seq_open(struct inode *inode, struct file *file)
3800 error = seq_open(file, &md_seq_ops);
3804 static struct file_operations md_seq_fops = {
3805 .open = md_seq_open,
3807 .llseek = seq_lseek,
3808 .release = seq_release,
3811 int register_md_personality(int pnum, mdk_personality_t *p)
3813 if (pnum >= MAX_PERSONALITY) {
3815 "md: tried to install personality %s as nr %d, but max is %lu\n",
3816 p->name, pnum, MAX_PERSONALITY-1);
3820 spin_lock(&pers_lock);
3822 spin_unlock(&pers_lock);
3827 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3828 spin_unlock(&pers_lock);
3832 int unregister_md_personality(int pnum)
3834 if (pnum >= MAX_PERSONALITY)
3837 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3838 spin_lock(&pers_lock);
3840 spin_unlock(&pers_lock);
3844 static int is_mddev_idle(mddev_t *mddev)
3847 struct list_head *tmp;
3849 unsigned long curr_events;
3852 ITERATE_RDEV(mddev,rdev,tmp) {
3853 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3854 curr_events = disk_stat_read(disk, sectors[0]) +
3855 disk_stat_read(disk, sectors[1]) -
3856 atomic_read(&disk->sync_io);
3857 /* The difference between curr_events and last_events
3858 * will be affected by any new non-sync IO (making
3859 * curr_events bigger) and any difference in the amount of
3860 * in-flight syncio (making current_events bigger or smaller)
3861 * The amount in-flight is currently limited to
3862 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3863 * which is at most 4096 sectors.
3864 * These numbers are fairly fragile and should be made
3865 * more robust, probably by enforcing the
3866 * 'window size' that md_do_sync sort-of uses.
3868 * Note: the following is an unsigned comparison.
3870 if ((curr_events - rdev->last_events + 4096) > 8192) {
3871 rdev->last_events = curr_events;
3878 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3880 /* another "blocks" (512byte) blocks have been synced */
3881 atomic_sub(blocks, &mddev->recovery_active);
3882 wake_up(&mddev->recovery_wait);
3884 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3885 md_wakeup_thread(mddev->thread);
3886 // stop recovery, signal do_sync ....
3891 /* md_write_start(mddev, bi)
3892 * If we need to update some array metadata (e.g. 'active' flag
3893 * in superblock) before writing, schedule a superblock update
3894 * and wait for it to complete.
3896 void md_write_start(mddev_t *mddev, struct bio *bi)
3898 if (bio_data_dir(bi) != WRITE)
3901 BUG_ON(mddev->ro == 1);
3902 if (mddev->ro == 2) {
3903 /* need to switch to read/write */
3905 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3906 md_wakeup_thread(mddev->thread);
3908 atomic_inc(&mddev->writes_pending);
3909 if (mddev->in_sync) {
3910 spin_lock_irq(&mddev->write_lock);
3911 if (mddev->in_sync) {
3913 mddev->sb_dirty = 1;
3914 md_wakeup_thread(mddev->thread);
3916 spin_unlock_irq(&mddev->write_lock);
3918 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3921 void md_write_end(mddev_t *mddev)
3923 if (atomic_dec_and_test(&mddev->writes_pending)) {
3924 if (mddev->safemode == 2)
3925 md_wakeup_thread(mddev->thread);
3927 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3931 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3933 #define SYNC_MARKS 10
3934 #define SYNC_MARK_STEP (3*HZ)
3935 static void md_do_sync(mddev_t *mddev)
3938 unsigned int currspeed = 0,
3940 sector_t max_sectors,j, io_sectors;
3941 unsigned long mark[SYNC_MARKS];
3942 sector_t mark_cnt[SYNC_MARKS];
3944 struct list_head *tmp;
3945 sector_t last_check;
3948 /* just incase thread restarts... */
3949 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3952 /* we overload curr_resync somewhat here.
3953 * 0 == not engaged in resync at all
3954 * 2 == checking that there is no conflict with another sync
3955 * 1 == like 2, but have yielded to allow conflicting resync to
3957 * other == active in resync - this many blocks
3959 * Before starting a resync we must have set curr_resync to
3960 * 2, and then checked that every "conflicting" array has curr_resync
3961 * less than ours. When we find one that is the same or higher
3962 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3963 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3964 * This will mean we have to start checking from the beginning again.
3969 mddev->curr_resync = 2;
3972 if (kthread_should_stop()) {
3973 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3976 ITERATE_MDDEV(mddev2,tmp) {
3977 if (mddev2 == mddev)
3979 if (mddev2->curr_resync &&
3980 match_mddev_units(mddev,mddev2)) {
3982 if (mddev < mddev2 && mddev->curr_resync == 2) {
3983 /* arbitrarily yield */
3984 mddev->curr_resync = 1;
3985 wake_up(&resync_wait);
3987 if (mddev > mddev2 && mddev->curr_resync == 1)
3988 /* no need to wait here, we can wait the next
3989 * time 'round when curr_resync == 2
3992 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
3993 if (!kthread_should_stop() &&
3994 mddev2->curr_resync >= mddev->curr_resync) {
3995 printk(KERN_INFO "md: delaying resync of %s"
3996 " until %s has finished resync (they"
3997 " share one or more physical units)\n",
3998 mdname(mddev), mdname(mddev2));
4001 finish_wait(&resync_wait, &wq);
4004 finish_wait(&resync_wait, &wq);
4007 } while (mddev->curr_resync < 2);
4009 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4010 /* resync follows the size requested by the personality,
4011 * which defaults to physical size, but can be virtual size
4013 max_sectors = mddev->resync_max_sectors;
4014 mddev->resync_mismatches = 0;
4016 /* recovery follows the physical size of devices */
4017 max_sectors = mddev->size << 1;
4019 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4020 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4021 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4022 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4023 "(but not more than %d KB/sec) for reconstruction.\n",
4024 sysctl_speed_limit_max);
4026 is_mddev_idle(mddev); /* this also initializes IO event counters */
4027 /* we don't use the checkpoint if there's a bitmap */
4028 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4029 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4030 j = mddev->recovery_cp;
4034 for (m = 0; m < SYNC_MARKS; m++) {
4036 mark_cnt[m] = io_sectors;
4039 mddev->resync_mark = mark[last_mark];
4040 mddev->resync_mark_cnt = mark_cnt[last_mark];
4043 * Tune reconstruction:
4045 window = 32*(PAGE_SIZE/512);
4046 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4047 window/2,(unsigned long long) max_sectors/2);
4049 atomic_set(&mddev->recovery_active, 0);
4050 init_waitqueue_head(&mddev->recovery_wait);
4055 "md: resuming recovery of %s from checkpoint.\n",
4057 mddev->curr_resync = j;
4060 while (j < max_sectors) {
4064 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4065 currspeed < sysctl_speed_limit_min);
4067 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4071 if (!skipped) { /* actual IO requested */
4072 io_sectors += sectors;
4073 atomic_add(sectors, &mddev->recovery_active);
4077 if (j>1) mddev->curr_resync = j;
4080 if (last_check + window > io_sectors || j == max_sectors)
4083 last_check = io_sectors;
4085 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4086 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4090 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4092 int next = (last_mark+1) % SYNC_MARKS;
4094 mddev->resync_mark = mark[next];
4095 mddev->resync_mark_cnt = mark_cnt[next];
4096 mark[next] = jiffies;
4097 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4102 if (kthread_should_stop()) {
4104 * got a signal, exit.
4107 "md: md_do_sync() got signal ... exiting\n");
4108 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4113 * this loop exits only if either when we are slower than
4114 * the 'hard' speed limit, or the system was IO-idle for
4116 * the system might be non-idle CPU-wise, but we only care
4117 * about not overloading the IO subsystem. (things like an
4118 * e2fsck being done on the RAID array should execute fast)
4120 mddev->queue->unplug_fn(mddev->queue);
4123 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4124 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4126 if (currspeed > sysctl_speed_limit_min) {
4127 if ((currspeed > sysctl_speed_limit_max) ||
4128 !is_mddev_idle(mddev)) {
4134 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4136 * this also signals 'finished resyncing' to md_stop
4139 mddev->queue->unplug_fn(mddev->queue);
4141 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4143 /* tell personality that we are finished */
4144 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4146 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4147 mddev->curr_resync > 2 &&
4148 mddev->curr_resync >= mddev->recovery_cp) {
4149 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4151 "md: checkpointing recovery of %s.\n",
4153 mddev->recovery_cp = mddev->curr_resync;
4155 mddev->recovery_cp = MaxSector;
4159 mddev->curr_resync = 0;
4160 wake_up(&resync_wait);
4161 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4162 md_wakeup_thread(mddev->thread);
4167 * This routine is regularly called by all per-raid-array threads to
4168 * deal with generic issues like resync and super-block update.
4169 * Raid personalities that don't have a thread (linear/raid0) do not
4170 * need this as they never do any recovery or update the superblock.
4172 * It does not do any resync itself, but rather "forks" off other threads
4173 * to do that as needed.
4174 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4175 * "->recovery" and create a thread at ->sync_thread.
4176 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4177 * and wakeups up this thread which will reap the thread and finish up.
4178 * This thread also removes any faulty devices (with nr_pending == 0).
4180 * The overall approach is:
4181 * 1/ if the superblock needs updating, update it.
4182 * 2/ If a recovery thread is running, don't do anything else.
4183 * 3/ If recovery has finished, clean up, possibly marking spares active.
4184 * 4/ If there are any faulty devices, remove them.
4185 * 5/ If array is degraded, try to add spares devices
4186 * 6/ If array has spares or is not in-sync, start a resync thread.
4188 void md_check_recovery(mddev_t *mddev)
4191 struct list_head *rtmp;
4195 bitmap_daemon_work(mddev->bitmap);
4200 if (signal_pending(current)) {
4201 if (mddev->pers->sync_request) {
4202 printk(KERN_INFO "md: %s in immediate safe mode\n",
4204 mddev->safemode = 2;
4206 flush_signals(current);
4211 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4212 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4213 (mddev->safemode == 1) ||
4214 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4215 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4219 if (mddev_trylock(mddev)==0) {
4222 spin_lock_irq(&mddev->write_lock);
4223 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4224 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4226 mddev->sb_dirty = 1;
4228 if (mddev->safemode == 1)
4229 mddev->safemode = 0;
4230 spin_unlock_irq(&mddev->write_lock);
4232 if (mddev->sb_dirty)
4233 md_update_sb(mddev);
4236 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4237 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4238 /* resync/recovery still happening */
4239 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4242 if (mddev->sync_thread) {
4243 /* resync has finished, collect result */
4244 md_unregister_thread(mddev->sync_thread);
4245 mddev->sync_thread = NULL;
4246 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4247 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4249 /* activate any spares */
4250 mddev->pers->spare_active(mddev);
4252 md_update_sb(mddev);
4254 /* if array is no-longer degraded, then any saved_raid_disk
4255 * information must be scrapped
4257 if (!mddev->degraded)
4258 ITERATE_RDEV(mddev,rdev,rtmp)
4259 rdev->saved_raid_disk = -1;
4261 mddev->recovery = 0;
4262 /* flag recovery needed just to double check */
4263 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4266 /* Clear some bits that don't mean anything, but
4269 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4270 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4271 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4272 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4274 /* no recovery is running.
4275 * remove any failed drives, then
4276 * add spares if possible.
4277 * Spare are also removed and re-added, to allow
4278 * the personality to fail the re-add.
4280 ITERATE_RDEV(mddev,rdev,rtmp)
4281 if (rdev->raid_disk >= 0 &&
4282 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4283 atomic_read(&rdev->nr_pending)==0) {
4284 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4286 sprintf(nm,"rd%d", rdev->raid_disk);
4287 sysfs_remove_link(&mddev->kobj, nm);
4288 rdev->raid_disk = -1;
4292 if (mddev->degraded) {
4293 ITERATE_RDEV(mddev,rdev,rtmp)
4294 if (rdev->raid_disk < 0
4295 && !test_bit(Faulty, &rdev->flags)) {
4296 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4298 sprintf(nm, "rd%d", rdev->raid_disk);
4299 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4307 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4308 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4309 } else if (mddev->recovery_cp < MaxSector) {
4310 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4311 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4312 /* nothing to be done ... */
4315 if (mddev->pers->sync_request) {
4316 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4317 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4318 /* We are adding a device or devices to an array
4319 * which has the bitmap stored on all devices.
4320 * So make sure all bitmap pages get written
4322 bitmap_write_all(mddev->bitmap);
4324 mddev->sync_thread = md_register_thread(md_do_sync,
4327 if (!mddev->sync_thread) {
4328 printk(KERN_ERR "%s: could not start resync"
4331 /* leave the spares where they are, it shouldn't hurt */
4332 mddev->recovery = 0;
4334 md_wakeup_thread(mddev->sync_thread);
4338 mddev_unlock(mddev);
4342 static int md_notify_reboot(struct notifier_block *this,
4343 unsigned long code, void *x)
4345 struct list_head *tmp;
4348 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4350 printk(KERN_INFO "md: stopping all md devices.\n");
4352 ITERATE_MDDEV(mddev,tmp)
4353 if (mddev_trylock(mddev)==0)
4354 do_md_stop (mddev, 1);
4356 * certain more exotic SCSI devices are known to be
4357 * volatile wrt too early system reboots. While the
4358 * right place to handle this issue is the given
4359 * driver, we do want to have a safe RAID driver ...
4366 static struct notifier_block md_notifier = {
4367 .notifier_call = md_notify_reboot,
4369 .priority = INT_MAX, /* before any real devices */
4372 static void md_geninit(void)
4374 struct proc_dir_entry *p;
4376 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4378 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4380 p->proc_fops = &md_seq_fops;
4383 static int __init md_init(void)
4387 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4388 " MD_SB_DISKS=%d\n",
4389 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4390 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4391 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4394 if (register_blkdev(MAJOR_NR, "md"))
4396 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4397 unregister_blkdev(MAJOR_NR, "md");
4401 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4402 md_probe, NULL, NULL);
4403 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4404 md_probe, NULL, NULL);
4406 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4407 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4408 S_IFBLK|S_IRUSR|S_IWUSR,
4411 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4412 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4413 S_IFBLK|S_IRUSR|S_IWUSR,
4417 register_reboot_notifier(&md_notifier);
4418 raid_table_header = register_sysctl_table(raid_root_table, 1);
4428 * Searches all registered partitions for autorun RAID arrays
4431 static dev_t detected_devices[128];
4434 void md_autodetect_dev(dev_t dev)
4436 if (dev_cnt >= 0 && dev_cnt < 127)
4437 detected_devices[dev_cnt++] = dev;
4441 static void autostart_arrays(int part)
4446 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4448 for (i = 0; i < dev_cnt; i++) {
4449 dev_t dev = detected_devices[i];
4451 rdev = md_import_device(dev,0, 0);
4455 if (test_bit(Faulty, &rdev->flags)) {
4459 list_add(&rdev->same_set, &pending_raid_disks);
4463 autorun_devices(part);
4468 static __exit void md_exit(void)
4471 struct list_head *tmp;
4473 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4474 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4475 for (i=0; i < MAX_MD_DEVS; i++)
4476 devfs_remove("md/%d", i);
4477 for (i=0; i < MAX_MD_DEVS; i++)
4478 devfs_remove("md/d%d", i);
4482 unregister_blkdev(MAJOR_NR,"md");
4483 unregister_blkdev(mdp_major, "mdp");
4484 unregister_reboot_notifier(&md_notifier);
4485 unregister_sysctl_table(raid_table_header);
4486 remove_proc_entry("mdstat", NULL);
4487 ITERATE_MDDEV(mddev,tmp) {
4488 struct gendisk *disk = mddev->gendisk;
4491 export_array(mddev);
4494 mddev->gendisk = NULL;
4499 module_init(md_init)
4500 module_exit(md_exit)
4502 static int get_ro(char *buffer, struct kernel_param *kp)
4504 return sprintf(buffer, "%d", start_readonly);
4506 static int set_ro(const char *val, struct kernel_param *kp)
4509 int num = simple_strtoul(val, &e, 10);
4510 if (*val && (*e == '\0' || *e == '\n')) {
4511 start_readonly = num;
4517 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4518 module_param(start_dirty_degraded, int, 0644);
4521 EXPORT_SYMBOL(register_md_personality);
4522 EXPORT_SYMBOL(unregister_md_personality);
4523 EXPORT_SYMBOL(md_error);
4524 EXPORT_SYMBOL(md_done_sync);
4525 EXPORT_SYMBOL(md_write_start);
4526 EXPORT_SYMBOL(md_write_end);
4527 EXPORT_SYMBOL(md_register_thread);
4528 EXPORT_SYMBOL(md_unregister_thread);
4529 EXPORT_SYMBOL(md_wakeup_thread);
4530 EXPORT_SYMBOL(md_print_devices);
4531 EXPORT_SYMBOL(md_check_recovery);
4532 MODULE_LICENSE("GPL");
4534 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);