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>
45 #include <linux/poll.h>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part);
71 static mdk_personality_t *pers[MAX_PERSONALITY];
72 static DEFINE_SPINLOCK(pers_lock);
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
86 static int sysctl_speed_limit_min = 1000;
87 static int sysctl_speed_limit_max = 200000;
89 static struct ctl_table_header *raid_table_header;
91 static ctl_table raid_table[] = {
93 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
94 .procname = "speed_limit_min",
95 .data = &sysctl_speed_limit_min,
96 .maxlen = sizeof(int),
98 .proc_handler = &proc_dointvec,
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
102 .procname = "speed_limit_max",
103 .data = &sysctl_speed_limit_max,
104 .maxlen = sizeof(int),
106 .proc_handler = &proc_dointvec,
111 static ctl_table raid_dir_table[] = {
113 .ctl_name = DEV_RAID,
122 static ctl_table raid_root_table[] = {
128 .child = raid_dir_table,
133 static struct block_device_operations md_fops;
135 static int start_readonly;
138 * We have a system wide 'event count' that is incremented
139 * on any 'interesting' event, and readers of /proc/mdstat
140 * can use 'poll' or 'select' to find out when the event
144 * start array, stop array, error, add device, remove device,
145 * start build, activate spare
147 DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
148 static atomic_t md_event_count;
149 void md_new_event(mddev_t *mddev)
151 atomic_inc(&md_event_count);
152 wake_up(&md_event_waiters);
156 * Enables to iterate over all existing md arrays
157 * all_mddevs_lock protects this list.
159 static LIST_HEAD(all_mddevs);
160 static DEFINE_SPINLOCK(all_mddevs_lock);
164 * iterates through all used mddevs in the system.
165 * We take care to grab the all_mddevs_lock whenever navigating
166 * the list, and to always hold a refcount when unlocked.
167 * Any code which breaks out of this loop while own
168 * a reference to the current mddev and must mddev_put it.
170 #define ITERATE_MDDEV(mddev,tmp) \
172 for (({ spin_lock(&all_mddevs_lock); \
173 tmp = all_mddevs.next; \
175 ({ if (tmp != &all_mddevs) \
176 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
177 spin_unlock(&all_mddevs_lock); \
178 if (mddev) mddev_put(mddev); \
179 mddev = list_entry(tmp, mddev_t, all_mddevs); \
180 tmp != &all_mddevs;}); \
181 ({ spin_lock(&all_mddevs_lock); \
186 static int md_fail_request (request_queue_t *q, struct bio *bio)
188 bio_io_error(bio, bio->bi_size);
192 static inline mddev_t *mddev_get(mddev_t *mddev)
194 atomic_inc(&mddev->active);
198 static void mddev_put(mddev_t *mddev)
200 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
202 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
203 list_del(&mddev->all_mddevs);
204 blk_put_queue(mddev->queue);
205 kobject_unregister(&mddev->kobj);
207 spin_unlock(&all_mddevs_lock);
210 static mddev_t * mddev_find(dev_t unit)
212 mddev_t *mddev, *new = NULL;
215 spin_lock(&all_mddevs_lock);
216 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
217 if (mddev->unit == unit) {
219 spin_unlock(&all_mddevs_lock);
225 list_add(&new->all_mddevs, &all_mddevs);
226 spin_unlock(&all_mddevs_lock);
229 spin_unlock(&all_mddevs_lock);
231 new = kzalloc(sizeof(*new), GFP_KERNEL);
236 if (MAJOR(unit) == MD_MAJOR)
237 new->md_minor = MINOR(unit);
239 new->md_minor = MINOR(unit) >> MdpMinorShift;
241 init_MUTEX(&new->reconfig_sem);
242 INIT_LIST_HEAD(&new->disks);
243 INIT_LIST_HEAD(&new->all_mddevs);
244 init_timer(&new->safemode_timer);
245 atomic_set(&new->active, 1);
246 spin_lock_init(&new->write_lock);
247 init_waitqueue_head(&new->sb_wait);
249 new->queue = blk_alloc_queue(GFP_KERNEL);
255 blk_queue_make_request(new->queue, md_fail_request);
260 static inline int mddev_lock(mddev_t * mddev)
262 return down_interruptible(&mddev->reconfig_sem);
265 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
267 down(&mddev->reconfig_sem);
270 static inline int mddev_trylock(mddev_t * mddev)
272 return down_trylock(&mddev->reconfig_sem);
275 static inline void mddev_unlock(mddev_t * mddev)
277 up(&mddev->reconfig_sem);
279 md_wakeup_thread(mddev->thread);
282 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
285 struct list_head *tmp;
287 ITERATE_RDEV(mddev,rdev,tmp) {
288 if (rdev->desc_nr == nr)
294 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
296 struct list_head *tmp;
299 ITERATE_RDEV(mddev,rdev,tmp) {
300 if (rdev->bdev->bd_dev == dev)
306 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
308 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
309 return MD_NEW_SIZE_BLOCKS(size);
312 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
316 size = rdev->sb_offset;
319 size &= ~((sector_t)chunk_size/1024 - 1);
323 static int alloc_disk_sb(mdk_rdev_t * rdev)
328 rdev->sb_page = alloc_page(GFP_KERNEL);
329 if (!rdev->sb_page) {
330 printk(KERN_ALERT "md: out of memory.\n");
337 static void free_disk_sb(mdk_rdev_t * rdev)
340 put_page(rdev->sb_page);
342 rdev->sb_page = NULL;
349 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
351 mdk_rdev_t *rdev = bio->bi_private;
352 mddev_t *mddev = rdev->mddev;
356 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
357 md_error(mddev, rdev);
359 if (atomic_dec_and_test(&mddev->pending_writes))
360 wake_up(&mddev->sb_wait);
365 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
367 struct bio *bio2 = bio->bi_private;
368 mdk_rdev_t *rdev = bio2->bi_private;
369 mddev_t *mddev = rdev->mddev;
373 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
374 error == -EOPNOTSUPP) {
376 /* barriers don't appear to be supported :-( */
377 set_bit(BarriersNotsupp, &rdev->flags);
378 mddev->barriers_work = 0;
379 spin_lock_irqsave(&mddev->write_lock, flags);
380 bio2->bi_next = mddev->biolist;
381 mddev->biolist = bio2;
382 spin_unlock_irqrestore(&mddev->write_lock, flags);
383 wake_up(&mddev->sb_wait);
388 bio->bi_private = rdev;
389 return super_written(bio, bytes_done, error);
392 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
393 sector_t sector, int size, struct page *page)
395 /* write first size bytes of page to sector of rdev
396 * Increment mddev->pending_writes before returning
397 * and decrement it on completion, waking up sb_wait
398 * if zero is reached.
399 * If an error occurred, call md_error
401 * As we might need to resubmit the request if BIO_RW_BARRIER
402 * causes ENOTSUPP, we allocate a spare bio...
404 struct bio *bio = bio_alloc(GFP_NOIO, 1);
405 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
407 bio->bi_bdev = rdev->bdev;
408 bio->bi_sector = sector;
409 bio_add_page(bio, page, size, 0);
410 bio->bi_private = rdev;
411 bio->bi_end_io = super_written;
414 atomic_inc(&mddev->pending_writes);
415 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
417 rw |= (1<<BIO_RW_BARRIER);
418 rbio = bio_clone(bio, GFP_NOIO);
419 rbio->bi_private = bio;
420 rbio->bi_end_io = super_written_barrier;
421 submit_bio(rw, rbio);
426 void md_super_wait(mddev_t *mddev)
428 /* wait for all superblock writes that were scheduled to complete.
429 * if any had to be retried (due to BARRIER problems), retry them
433 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
434 if (atomic_read(&mddev->pending_writes)==0)
436 while (mddev->biolist) {
438 spin_lock_irq(&mddev->write_lock);
439 bio = mddev->biolist;
440 mddev->biolist = bio->bi_next ;
442 spin_unlock_irq(&mddev->write_lock);
443 submit_bio(bio->bi_rw, bio);
447 finish_wait(&mddev->sb_wait, &wq);
450 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
455 complete((struct completion*)bio->bi_private);
459 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
460 struct page *page, int rw)
462 struct bio *bio = bio_alloc(GFP_NOIO, 1);
463 struct completion event;
466 rw |= (1 << BIO_RW_SYNC);
469 bio->bi_sector = sector;
470 bio_add_page(bio, page, size, 0);
471 init_completion(&event);
472 bio->bi_private = &event;
473 bio->bi_end_io = bi_complete;
475 wait_for_completion(&event);
477 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
481 EXPORT_SYMBOL_GPL(sync_page_io);
483 static int read_disk_sb(mdk_rdev_t * rdev, int size)
485 char b[BDEVNAME_SIZE];
486 if (!rdev->sb_page) {
494 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
500 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
501 bdevname(rdev->bdev,b));
505 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
507 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
508 (sb1->set_uuid1 == sb2->set_uuid1) &&
509 (sb1->set_uuid2 == sb2->set_uuid2) &&
510 (sb1->set_uuid3 == sb2->set_uuid3))
518 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
521 mdp_super_t *tmp1, *tmp2;
523 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
524 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
526 if (!tmp1 || !tmp2) {
528 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
536 * nr_disks is not constant
541 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
552 static unsigned int calc_sb_csum(mdp_super_t * sb)
554 unsigned int disk_csum, csum;
556 disk_csum = sb->sb_csum;
558 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
559 sb->sb_csum = disk_csum;
565 * Handle superblock details.
566 * We want to be able to handle multiple superblock formats
567 * so we have a common interface to them all, and an array of
568 * different handlers.
569 * We rely on user-space to write the initial superblock, and support
570 * reading and updating of superblocks.
571 * Interface methods are:
572 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
573 * loads and validates a superblock on dev.
574 * if refdev != NULL, compare superblocks on both devices
576 * 0 - dev has a superblock that is compatible with refdev
577 * 1 - dev has a superblock that is compatible and newer than refdev
578 * so dev should be used as the refdev in future
579 * -EINVAL superblock incompatible or invalid
580 * -othererror e.g. -EIO
582 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
583 * Verify that dev is acceptable into mddev.
584 * The first time, mddev->raid_disks will be 0, and data from
585 * dev should be merged in. Subsequent calls check that dev
586 * is new enough. Return 0 or -EINVAL
588 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
589 * Update the superblock for rdev with data in mddev
590 * This does not write to disc.
596 struct module *owner;
597 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
598 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
599 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
603 * load_super for 0.90.0
605 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
607 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
613 * Calculate the position of the superblock,
614 * it's at the end of the disk.
616 * It also happens to be a multiple of 4Kb.
618 sb_offset = calc_dev_sboffset(rdev->bdev);
619 rdev->sb_offset = sb_offset;
621 ret = read_disk_sb(rdev, MD_SB_BYTES);
626 bdevname(rdev->bdev, b);
627 sb = (mdp_super_t*)page_address(rdev->sb_page);
629 if (sb->md_magic != MD_SB_MAGIC) {
630 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
635 if (sb->major_version != 0 ||
636 sb->minor_version != 90) {
637 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
638 sb->major_version, sb->minor_version,
643 if (sb->raid_disks <= 0)
646 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
647 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
652 rdev->preferred_minor = sb->md_minor;
653 rdev->data_offset = 0;
654 rdev->sb_size = MD_SB_BYTES;
656 if (sb->level == LEVEL_MULTIPATH)
659 rdev->desc_nr = sb->this_disk.number;
665 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
666 if (!uuid_equal(refsb, sb)) {
667 printk(KERN_WARNING "md: %s has different UUID to %s\n",
668 b, bdevname(refdev->bdev,b2));
671 if (!sb_equal(refsb, sb)) {
672 printk(KERN_WARNING "md: %s has same UUID"
673 " but different superblock to %s\n",
674 b, bdevname(refdev->bdev, b2));
678 ev2 = md_event(refsb);
684 rdev->size = calc_dev_size(rdev, sb->chunk_size);
691 * validate_super for 0.90.0
693 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
696 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
698 rdev->raid_disk = -1;
700 if (mddev->raid_disks == 0) {
701 mddev->major_version = 0;
702 mddev->minor_version = sb->minor_version;
703 mddev->patch_version = sb->patch_version;
704 mddev->persistent = ! sb->not_persistent;
705 mddev->chunk_size = sb->chunk_size;
706 mddev->ctime = sb->ctime;
707 mddev->utime = sb->utime;
708 mddev->level = sb->level;
709 mddev->layout = sb->layout;
710 mddev->raid_disks = sb->raid_disks;
711 mddev->size = sb->size;
712 mddev->events = md_event(sb);
713 mddev->bitmap_offset = 0;
714 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
716 if (sb->state & (1<<MD_SB_CLEAN))
717 mddev->recovery_cp = MaxSector;
719 if (sb->events_hi == sb->cp_events_hi &&
720 sb->events_lo == sb->cp_events_lo) {
721 mddev->recovery_cp = sb->recovery_cp;
723 mddev->recovery_cp = 0;
726 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
727 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
728 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
729 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
731 mddev->max_disks = MD_SB_DISKS;
733 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
734 mddev->bitmap_file == NULL) {
735 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
736 && mddev->level != 10) {
737 /* FIXME use a better test */
738 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
741 mddev->bitmap_offset = mddev->default_bitmap_offset;
744 } else if (mddev->pers == NULL) {
745 /* Insist on good event counter while assembling */
746 __u64 ev1 = md_event(sb);
748 if (ev1 < mddev->events)
750 } else if (mddev->bitmap) {
751 /* if adding to array with a bitmap, then we can accept an
752 * older device ... but not too old.
754 __u64 ev1 = md_event(sb);
755 if (ev1 < mddev->bitmap->events_cleared)
757 } else /* just a hot-add of a new device, leave raid_disk at -1 */
760 if (mddev->level != LEVEL_MULTIPATH) {
761 desc = sb->disks + rdev->desc_nr;
763 if (desc->state & (1<<MD_DISK_FAULTY))
764 set_bit(Faulty, &rdev->flags);
765 else if (desc->state & (1<<MD_DISK_SYNC) &&
766 desc->raid_disk < mddev->raid_disks) {
767 set_bit(In_sync, &rdev->flags);
768 rdev->raid_disk = desc->raid_disk;
770 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
771 set_bit(WriteMostly, &rdev->flags);
772 } else /* MULTIPATH are always insync */
773 set_bit(In_sync, &rdev->flags);
778 * sync_super for 0.90.0
780 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
783 struct list_head *tmp;
785 int next_spare = mddev->raid_disks;
788 /* make rdev->sb match mddev data..
791 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
792 * 3/ any empty disks < next_spare become removed
794 * disks[0] gets initialised to REMOVED because
795 * we cannot be sure from other fields if it has
796 * been initialised or not.
799 int active=0, working=0,failed=0,spare=0,nr_disks=0;
801 rdev->sb_size = MD_SB_BYTES;
803 sb = (mdp_super_t*)page_address(rdev->sb_page);
805 memset(sb, 0, sizeof(*sb));
807 sb->md_magic = MD_SB_MAGIC;
808 sb->major_version = mddev->major_version;
809 sb->minor_version = mddev->minor_version;
810 sb->patch_version = mddev->patch_version;
811 sb->gvalid_words = 0; /* ignored */
812 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
813 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
814 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
815 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
817 sb->ctime = mddev->ctime;
818 sb->level = mddev->level;
819 sb->size = mddev->size;
820 sb->raid_disks = mddev->raid_disks;
821 sb->md_minor = mddev->md_minor;
822 sb->not_persistent = !mddev->persistent;
823 sb->utime = mddev->utime;
825 sb->events_hi = (mddev->events>>32);
826 sb->events_lo = (u32)mddev->events;
830 sb->recovery_cp = mddev->recovery_cp;
831 sb->cp_events_hi = (mddev->events>>32);
832 sb->cp_events_lo = (u32)mddev->events;
833 if (mddev->recovery_cp == MaxSector)
834 sb->state = (1<< MD_SB_CLEAN);
838 sb->layout = mddev->layout;
839 sb->chunk_size = mddev->chunk_size;
841 if (mddev->bitmap && mddev->bitmap_file == NULL)
842 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
844 sb->disks[0].state = (1<<MD_DISK_REMOVED);
845 ITERATE_RDEV(mddev,rdev2,tmp) {
848 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
849 && !test_bit(Faulty, &rdev2->flags))
850 desc_nr = rdev2->raid_disk;
852 desc_nr = next_spare++;
853 rdev2->desc_nr = desc_nr;
854 d = &sb->disks[rdev2->desc_nr];
856 d->number = rdev2->desc_nr;
857 d->major = MAJOR(rdev2->bdev->bd_dev);
858 d->minor = MINOR(rdev2->bdev->bd_dev);
859 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
860 && !test_bit(Faulty, &rdev2->flags))
861 d->raid_disk = rdev2->raid_disk;
863 d->raid_disk = rdev2->desc_nr; /* compatibility */
864 if (test_bit(Faulty, &rdev2->flags)) {
865 d->state = (1<<MD_DISK_FAULTY);
867 } else if (test_bit(In_sync, &rdev2->flags)) {
868 d->state = (1<<MD_DISK_ACTIVE);
869 d->state |= (1<<MD_DISK_SYNC);
877 if (test_bit(WriteMostly, &rdev2->flags))
878 d->state |= (1<<MD_DISK_WRITEMOSTLY);
880 /* now set the "removed" and "faulty" bits on any missing devices */
881 for (i=0 ; i < mddev->raid_disks ; i++) {
882 mdp_disk_t *d = &sb->disks[i];
883 if (d->state == 0 && d->number == 0) {
886 d->state = (1<<MD_DISK_REMOVED);
887 d->state |= (1<<MD_DISK_FAULTY);
891 sb->nr_disks = nr_disks;
892 sb->active_disks = active;
893 sb->working_disks = working;
894 sb->failed_disks = failed;
895 sb->spare_disks = spare;
897 sb->this_disk = sb->disks[rdev->desc_nr];
898 sb->sb_csum = calc_sb_csum(sb);
902 * version 1 superblock
905 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
907 unsigned int disk_csum, csum;
908 unsigned long long newcsum;
909 int size = 256 + le32_to_cpu(sb->max_dev)*2;
910 unsigned int *isuper = (unsigned int*)sb;
913 disk_csum = sb->sb_csum;
916 for (i=0; size>=4; size -= 4 )
917 newcsum += le32_to_cpu(*isuper++);
920 newcsum += le16_to_cpu(*(unsigned short*) isuper);
922 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
923 sb->sb_csum = disk_csum;
924 return cpu_to_le32(csum);
927 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
929 struct mdp_superblock_1 *sb;
932 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
936 * Calculate the position of the superblock.
937 * It is always aligned to a 4K boundary and
938 * depeding on minor_version, it can be:
939 * 0: At least 8K, but less than 12K, from end of device
940 * 1: At start of device
941 * 2: 4K from start of device.
943 switch(minor_version) {
945 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
947 sb_offset &= ~(sector_t)(4*2-1);
948 /* convert from sectors to K */
960 rdev->sb_offset = sb_offset;
962 /* superblock is rarely larger than 1K, but it can be larger,
963 * and it is safe to read 4k, so we do that
965 ret = read_disk_sb(rdev, 4096);
969 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
971 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
972 sb->major_version != cpu_to_le32(1) ||
973 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
974 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
975 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
978 if (calc_sb_1_csum(sb) != sb->sb_csum) {
979 printk("md: invalid superblock checksum on %s\n",
980 bdevname(rdev->bdev,b));
983 if (le64_to_cpu(sb->data_size) < 10) {
984 printk("md: data_size too small on %s\n",
985 bdevname(rdev->bdev,b));
988 rdev->preferred_minor = 0xffff;
989 rdev->data_offset = le64_to_cpu(sb->data_offset);
991 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
992 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
993 if (rdev->sb_size & bmask)
994 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1000 struct mdp_superblock_1 *refsb =
1001 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1003 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1004 sb->level != refsb->level ||
1005 sb->layout != refsb->layout ||
1006 sb->chunksize != refsb->chunksize) {
1007 printk(KERN_WARNING "md: %s has strangely different"
1008 " superblock to %s\n",
1009 bdevname(rdev->bdev,b),
1010 bdevname(refdev->bdev,b2));
1013 ev1 = le64_to_cpu(sb->events);
1014 ev2 = le64_to_cpu(refsb->events);
1020 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1022 rdev->size = rdev->sb_offset;
1023 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1025 rdev->size = le64_to_cpu(sb->data_size)/2;
1026 if (le32_to_cpu(sb->chunksize))
1027 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1031 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1033 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1035 rdev->raid_disk = -1;
1037 if (mddev->raid_disks == 0) {
1038 mddev->major_version = 1;
1039 mddev->patch_version = 0;
1040 mddev->persistent = 1;
1041 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1042 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1043 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1044 mddev->level = le32_to_cpu(sb->level);
1045 mddev->layout = le32_to_cpu(sb->layout);
1046 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1047 mddev->size = le64_to_cpu(sb->size)/2;
1048 mddev->events = le64_to_cpu(sb->events);
1049 mddev->bitmap_offset = 0;
1050 mddev->default_bitmap_offset = 1024;
1052 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1053 memcpy(mddev->uuid, sb->set_uuid, 16);
1055 mddev->max_disks = (4096-256)/2;
1057 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1058 mddev->bitmap_file == NULL ) {
1059 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1060 && mddev->level != 10) {
1061 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1064 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1066 } else if (mddev->pers == NULL) {
1067 /* Insist of good event counter while assembling */
1068 __u64 ev1 = le64_to_cpu(sb->events);
1070 if (ev1 < mddev->events)
1072 } else if (mddev->bitmap) {
1073 /* If adding to array with a bitmap, then we can accept an
1074 * older device, but not too old.
1076 __u64 ev1 = le64_to_cpu(sb->events);
1077 if (ev1 < mddev->bitmap->events_cleared)
1079 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1082 if (mddev->level != LEVEL_MULTIPATH) {
1084 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1085 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1087 case 0xffff: /* spare */
1089 case 0xfffe: /* faulty */
1090 set_bit(Faulty, &rdev->flags);
1093 set_bit(In_sync, &rdev->flags);
1094 rdev->raid_disk = role;
1097 if (sb->devflags & WriteMostly1)
1098 set_bit(WriteMostly, &rdev->flags);
1099 } else /* MULTIPATH are always insync */
1100 set_bit(In_sync, &rdev->flags);
1105 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1107 struct mdp_superblock_1 *sb;
1108 struct list_head *tmp;
1111 /* make rdev->sb match mddev and rdev data. */
1113 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1115 sb->feature_map = 0;
1117 memset(sb->pad1, 0, sizeof(sb->pad1));
1118 memset(sb->pad2, 0, sizeof(sb->pad2));
1119 memset(sb->pad3, 0, sizeof(sb->pad3));
1121 sb->utime = cpu_to_le64((__u64)mddev->utime);
1122 sb->events = cpu_to_le64(mddev->events);
1124 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1126 sb->resync_offset = cpu_to_le64(0);
1128 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1129 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1130 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1134 ITERATE_RDEV(mddev,rdev2,tmp)
1135 if (rdev2->desc_nr+1 > max_dev)
1136 max_dev = rdev2->desc_nr+1;
1138 sb->max_dev = cpu_to_le32(max_dev);
1139 for (i=0; i<max_dev;i++)
1140 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1142 ITERATE_RDEV(mddev,rdev2,tmp) {
1144 if (test_bit(Faulty, &rdev2->flags))
1145 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1146 else if (test_bit(In_sync, &rdev2->flags))
1147 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1149 sb->dev_roles[i] = cpu_to_le16(0xffff);
1152 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1153 sb->sb_csum = calc_sb_1_csum(sb);
1157 static struct super_type super_types[] = {
1160 .owner = THIS_MODULE,
1161 .load_super = super_90_load,
1162 .validate_super = super_90_validate,
1163 .sync_super = super_90_sync,
1167 .owner = THIS_MODULE,
1168 .load_super = super_1_load,
1169 .validate_super = super_1_validate,
1170 .sync_super = super_1_sync,
1174 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1176 struct list_head *tmp;
1179 ITERATE_RDEV(mddev,rdev,tmp)
1180 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1186 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1188 struct list_head *tmp;
1191 ITERATE_RDEV(mddev1,rdev,tmp)
1192 if (match_dev_unit(mddev2, rdev))
1198 static LIST_HEAD(pending_raid_disks);
1200 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1202 mdk_rdev_t *same_pdev;
1203 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1210 same_pdev = match_dev_unit(mddev, rdev);
1213 "%s: WARNING: %s appears to be on the same physical"
1214 " disk as %s. True\n protection against single-disk"
1215 " failure might be compromised.\n",
1216 mdname(mddev), bdevname(rdev->bdev,b),
1217 bdevname(same_pdev->bdev,b2));
1219 /* Verify rdev->desc_nr is unique.
1220 * If it is -1, assign a free number, else
1221 * check number is not in use
1223 if (rdev->desc_nr < 0) {
1225 if (mddev->pers) choice = mddev->raid_disks;
1226 while (find_rdev_nr(mddev, choice))
1228 rdev->desc_nr = choice;
1230 if (find_rdev_nr(mddev, rdev->desc_nr))
1233 bdevname(rdev->bdev,b);
1234 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1237 list_add(&rdev->same_set, &mddev->disks);
1238 rdev->mddev = mddev;
1239 printk(KERN_INFO "md: bind<%s>\n", b);
1241 rdev->kobj.parent = &mddev->kobj;
1242 kobject_add(&rdev->kobj);
1244 if (rdev->bdev->bd_part)
1245 ko = &rdev->bdev->bd_part->kobj;
1247 ko = &rdev->bdev->bd_disk->kobj;
1248 sysfs_create_link(&rdev->kobj, ko, "block");
1252 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1254 char b[BDEVNAME_SIZE];
1259 list_del_init(&rdev->same_set);
1260 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1262 sysfs_remove_link(&rdev->kobj, "block");
1263 kobject_del(&rdev->kobj);
1267 * prevent the device from being mounted, repartitioned or
1268 * otherwise reused by a RAID array (or any other kernel
1269 * subsystem), by bd_claiming the device.
1271 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1274 struct block_device *bdev;
1275 char b[BDEVNAME_SIZE];
1277 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1279 printk(KERN_ERR "md: could not open %s.\n",
1280 __bdevname(dev, b));
1281 return PTR_ERR(bdev);
1283 err = bd_claim(bdev, rdev);
1285 printk(KERN_ERR "md: could not bd_claim %s.\n",
1294 static void unlock_rdev(mdk_rdev_t *rdev)
1296 struct block_device *bdev = rdev->bdev;
1304 void md_autodetect_dev(dev_t dev);
1306 static void export_rdev(mdk_rdev_t * rdev)
1308 char b[BDEVNAME_SIZE];
1309 printk(KERN_INFO "md: export_rdev(%s)\n",
1310 bdevname(rdev->bdev,b));
1314 list_del_init(&rdev->same_set);
1316 md_autodetect_dev(rdev->bdev->bd_dev);
1319 kobject_put(&rdev->kobj);
1322 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1324 unbind_rdev_from_array(rdev);
1328 static void export_array(mddev_t *mddev)
1330 struct list_head *tmp;
1333 ITERATE_RDEV(mddev,rdev,tmp) {
1338 kick_rdev_from_array(rdev);
1340 if (!list_empty(&mddev->disks))
1342 mddev->raid_disks = 0;
1343 mddev->major_version = 0;
1346 static void print_desc(mdp_disk_t *desc)
1348 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1349 desc->major,desc->minor,desc->raid_disk,desc->state);
1352 static void print_sb(mdp_super_t *sb)
1357 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1358 sb->major_version, sb->minor_version, sb->patch_version,
1359 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1361 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1362 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1363 sb->md_minor, sb->layout, sb->chunk_size);
1364 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1365 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1366 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1367 sb->failed_disks, sb->spare_disks,
1368 sb->sb_csum, (unsigned long)sb->events_lo);
1371 for (i = 0; i < MD_SB_DISKS; i++) {
1374 desc = sb->disks + i;
1375 if (desc->number || desc->major || desc->minor ||
1376 desc->raid_disk || (desc->state && (desc->state != 4))) {
1377 printk(" D %2d: ", i);
1381 printk(KERN_INFO "md: THIS: ");
1382 print_desc(&sb->this_disk);
1386 static void print_rdev(mdk_rdev_t *rdev)
1388 char b[BDEVNAME_SIZE];
1389 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1390 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1391 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1393 if (rdev->sb_loaded) {
1394 printk(KERN_INFO "md: rdev superblock:\n");
1395 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1397 printk(KERN_INFO "md: no rdev superblock!\n");
1400 void md_print_devices(void)
1402 struct list_head *tmp, *tmp2;
1405 char b[BDEVNAME_SIZE];
1408 printk("md: **********************************\n");
1409 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1410 printk("md: **********************************\n");
1411 ITERATE_MDDEV(mddev,tmp) {
1414 bitmap_print_sb(mddev->bitmap);
1416 printk("%s: ", mdname(mddev));
1417 ITERATE_RDEV(mddev,rdev,tmp2)
1418 printk("<%s>", bdevname(rdev->bdev,b));
1421 ITERATE_RDEV(mddev,rdev,tmp2)
1424 printk("md: **********************************\n");
1429 static void sync_sbs(mddev_t * mddev)
1432 struct list_head *tmp;
1434 ITERATE_RDEV(mddev,rdev,tmp) {
1435 super_types[mddev->major_version].
1436 sync_super(mddev, rdev);
1437 rdev->sb_loaded = 1;
1441 static void md_update_sb(mddev_t * mddev)
1444 struct list_head *tmp;
1449 spin_lock_irq(&mddev->write_lock);
1450 sync_req = mddev->in_sync;
1451 mddev->utime = get_seconds();
1454 if (!mddev->events) {
1456 * oops, this 64-bit counter should never wrap.
1457 * Either we are in around ~1 trillion A.C., assuming
1458 * 1 reboot per second, or we have a bug:
1463 mddev->sb_dirty = 2;
1467 * do not write anything to disk if using
1468 * nonpersistent superblocks
1470 if (!mddev->persistent) {
1471 mddev->sb_dirty = 0;
1472 spin_unlock_irq(&mddev->write_lock);
1473 wake_up(&mddev->sb_wait);
1476 spin_unlock_irq(&mddev->write_lock);
1479 "md: updating %s RAID superblock on device (in sync %d)\n",
1480 mdname(mddev),mddev->in_sync);
1482 err = bitmap_update_sb(mddev->bitmap);
1483 ITERATE_RDEV(mddev,rdev,tmp) {
1484 char b[BDEVNAME_SIZE];
1485 dprintk(KERN_INFO "md: ");
1486 if (test_bit(Faulty, &rdev->flags))
1487 dprintk("(skipping faulty ");
1489 dprintk("%s ", bdevname(rdev->bdev,b));
1490 if (!test_bit(Faulty, &rdev->flags)) {
1491 md_super_write(mddev,rdev,
1492 rdev->sb_offset<<1, rdev->sb_size,
1494 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1495 bdevname(rdev->bdev,b),
1496 (unsigned long long)rdev->sb_offset);
1500 if (mddev->level == LEVEL_MULTIPATH)
1501 /* only need to write one superblock... */
1504 md_super_wait(mddev);
1505 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1507 spin_lock_irq(&mddev->write_lock);
1508 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1509 /* have to write it out again */
1510 spin_unlock_irq(&mddev->write_lock);
1513 mddev->sb_dirty = 0;
1514 spin_unlock_irq(&mddev->write_lock);
1515 wake_up(&mddev->sb_wait);
1519 struct rdev_sysfs_entry {
1520 struct attribute attr;
1521 ssize_t (*show)(mdk_rdev_t *, char *);
1522 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1526 state_show(mdk_rdev_t *rdev, char *page)
1531 if (test_bit(Faulty, &rdev->flags)) {
1532 len+= sprintf(page+len, "%sfaulty",sep);
1535 if (test_bit(In_sync, &rdev->flags)) {
1536 len += sprintf(page+len, "%sin_sync",sep);
1539 if (!test_bit(Faulty, &rdev->flags) &&
1540 !test_bit(In_sync, &rdev->flags)) {
1541 len += sprintf(page+len, "%sspare", sep);
1544 return len+sprintf(page+len, "\n");
1547 static struct rdev_sysfs_entry
1548 rdev_state = __ATTR_RO(state);
1551 super_show(mdk_rdev_t *rdev, char *page)
1553 if (rdev->sb_loaded && rdev->sb_size) {
1554 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1555 return rdev->sb_size;
1559 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1561 static struct attribute *rdev_default_attrs[] = {
1567 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1569 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1570 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1574 return entry->show(rdev, page);
1578 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1579 const char *page, size_t length)
1581 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1582 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1586 return entry->store(rdev, page, length);
1589 static void rdev_free(struct kobject *ko)
1591 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1594 static struct sysfs_ops rdev_sysfs_ops = {
1595 .show = rdev_attr_show,
1596 .store = rdev_attr_store,
1598 static struct kobj_type rdev_ktype = {
1599 .release = rdev_free,
1600 .sysfs_ops = &rdev_sysfs_ops,
1601 .default_attrs = rdev_default_attrs,
1605 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1607 * mark the device faulty if:
1609 * - the device is nonexistent (zero size)
1610 * - the device has no valid superblock
1612 * a faulty rdev _never_ has rdev->sb set.
1614 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1616 char b[BDEVNAME_SIZE];
1621 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1623 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1624 return ERR_PTR(-ENOMEM);
1627 if ((err = alloc_disk_sb(rdev)))
1630 err = lock_rdev(rdev, newdev);
1634 rdev->kobj.parent = NULL;
1635 rdev->kobj.ktype = &rdev_ktype;
1636 kobject_init(&rdev->kobj);
1640 rdev->data_offset = 0;
1641 atomic_set(&rdev->nr_pending, 0);
1642 atomic_set(&rdev->read_errors, 0);
1644 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1647 "md: %s has zero or unknown size, marking faulty!\n",
1648 bdevname(rdev->bdev,b));
1653 if (super_format >= 0) {
1654 err = super_types[super_format].
1655 load_super(rdev, NULL, super_minor);
1656 if (err == -EINVAL) {
1658 "md: %s has invalid sb, not importing!\n",
1659 bdevname(rdev->bdev,b));
1664 "md: could not read %s's sb, not importing!\n",
1665 bdevname(rdev->bdev,b));
1669 INIT_LIST_HEAD(&rdev->same_set);
1674 if (rdev->sb_page) {
1680 return ERR_PTR(err);
1684 * Check a full RAID array for plausibility
1688 static void analyze_sbs(mddev_t * mddev)
1691 struct list_head *tmp;
1692 mdk_rdev_t *rdev, *freshest;
1693 char b[BDEVNAME_SIZE];
1696 ITERATE_RDEV(mddev,rdev,tmp)
1697 switch (super_types[mddev->major_version].
1698 load_super(rdev, freshest, mddev->minor_version)) {
1706 "md: fatal superblock inconsistency in %s"
1707 " -- removing from array\n",
1708 bdevname(rdev->bdev,b));
1709 kick_rdev_from_array(rdev);
1713 super_types[mddev->major_version].
1714 validate_super(mddev, freshest);
1717 ITERATE_RDEV(mddev,rdev,tmp) {
1718 if (rdev != freshest)
1719 if (super_types[mddev->major_version].
1720 validate_super(mddev, rdev)) {
1721 printk(KERN_WARNING "md: kicking non-fresh %s"
1723 bdevname(rdev->bdev,b));
1724 kick_rdev_from_array(rdev);
1727 if (mddev->level == LEVEL_MULTIPATH) {
1728 rdev->desc_nr = i++;
1729 rdev->raid_disk = rdev->desc_nr;
1730 set_bit(In_sync, &rdev->flags);
1736 if (mddev->recovery_cp != MaxSector &&
1738 printk(KERN_ERR "md: %s: raid array is not clean"
1739 " -- starting background reconstruction\n",
1745 level_show(mddev_t *mddev, char *page)
1747 mdk_personality_t *p = mddev->pers;
1748 if (p == NULL && mddev->raid_disks == 0)
1750 if (mddev->level >= 0)
1751 return sprintf(page, "raid%d\n", mddev->level);
1753 return sprintf(page, "%s\n", p->name);
1756 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1759 raid_disks_show(mddev_t *mddev, char *page)
1761 if (mddev->raid_disks == 0)
1763 return sprintf(page, "%d\n", mddev->raid_disks);
1766 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1769 action_show(mddev_t *mddev, char *page)
1771 char *type = "idle";
1772 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1773 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1774 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1775 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1777 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1784 return sprintf(page, "%s\n", type);
1788 action_store(mddev_t *mddev, const char *page, size_t len)
1790 if (!mddev->pers || !mddev->pers->sync_request)
1793 if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1794 if (mddev->sync_thread) {
1795 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1796 md_unregister_thread(mddev->sync_thread);
1797 mddev->sync_thread = NULL;
1798 mddev->recovery = 0;
1803 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1804 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1806 if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1807 strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1808 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1810 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1811 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1812 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1814 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1815 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1816 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1818 md_wakeup_thread(mddev->thread);
1823 mismatch_cnt_show(mddev_t *mddev, char *page)
1825 return sprintf(page, "%llu\n",
1826 (unsigned long long) mddev->resync_mismatches);
1829 static struct md_sysfs_entry
1830 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1833 static struct md_sysfs_entry
1834 md_mismatches = __ATTR_RO(mismatch_cnt);
1836 static struct attribute *md_default_attrs[] = {
1838 &md_raid_disks.attr,
1842 static struct attribute *md_redundancy_attrs[] = {
1844 &md_mismatches.attr,
1847 static struct attribute_group md_redundancy_group = {
1849 .attrs = md_redundancy_attrs,
1854 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1856 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1857 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1863 rv = entry->show(mddev, page);
1864 mddev_unlock(mddev);
1869 md_attr_store(struct kobject *kobj, struct attribute *attr,
1870 const char *page, size_t length)
1872 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1873 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1879 rv = entry->store(mddev, page, length);
1880 mddev_unlock(mddev);
1884 static void md_free(struct kobject *ko)
1886 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1890 static struct sysfs_ops md_sysfs_ops = {
1891 .show = md_attr_show,
1892 .store = md_attr_store,
1894 static struct kobj_type md_ktype = {
1896 .sysfs_ops = &md_sysfs_ops,
1897 .default_attrs = md_default_attrs,
1902 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1904 static DECLARE_MUTEX(disks_sem);
1905 mddev_t *mddev = mddev_find(dev);
1906 struct gendisk *disk;
1907 int partitioned = (MAJOR(dev) != MD_MAJOR);
1908 int shift = partitioned ? MdpMinorShift : 0;
1909 int unit = MINOR(dev) >> shift;
1915 if (mddev->gendisk) {
1920 disk = alloc_disk(1 << shift);
1926 disk->major = MAJOR(dev);
1927 disk->first_minor = unit << shift;
1929 sprintf(disk->disk_name, "md_d%d", unit);
1930 sprintf(disk->devfs_name, "md/d%d", unit);
1932 sprintf(disk->disk_name, "md%d", unit);
1933 sprintf(disk->devfs_name, "md/%d", unit);
1935 disk->fops = &md_fops;
1936 disk->private_data = mddev;
1937 disk->queue = mddev->queue;
1939 mddev->gendisk = disk;
1941 mddev->kobj.parent = &disk->kobj;
1942 mddev->kobj.k_name = NULL;
1943 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1944 mddev->kobj.ktype = &md_ktype;
1945 kobject_register(&mddev->kobj);
1949 void md_wakeup_thread(mdk_thread_t *thread);
1951 static void md_safemode_timeout(unsigned long data)
1953 mddev_t *mddev = (mddev_t *) data;
1955 mddev->safemode = 1;
1956 md_wakeup_thread(mddev->thread);
1959 static int start_dirty_degraded;
1961 static int do_md_run(mddev_t * mddev)
1965 struct list_head *tmp;
1967 struct gendisk *disk;
1968 char b[BDEVNAME_SIZE];
1970 if (list_empty(&mddev->disks))
1971 /* cannot run an array with no devices.. */
1978 * Analyze all RAID superblock(s)
1980 if (!mddev->raid_disks)
1983 chunk_size = mddev->chunk_size;
1984 pnum = level_to_pers(mddev->level);
1986 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1989 * 'default chunksize' in the old md code used to
1990 * be PAGE_SIZE, baaad.
1991 * we abort here to be on the safe side. We don't
1992 * want to continue the bad practice.
1995 "no chunksize specified, see 'man raidtab'\n");
1998 if (chunk_size > MAX_CHUNK_SIZE) {
1999 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2000 chunk_size, MAX_CHUNK_SIZE);
2004 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2006 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2007 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2010 if (chunk_size < PAGE_SIZE) {
2011 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2012 chunk_size, PAGE_SIZE);
2016 /* devices must have minimum size of one chunk */
2017 ITERATE_RDEV(mddev,rdev,tmp) {
2018 if (test_bit(Faulty, &rdev->flags))
2020 if (rdev->size < chunk_size / 1024) {
2022 "md: Dev %s smaller than chunk_size:"
2024 bdevname(rdev->bdev,b),
2025 (unsigned long long)rdev->size,
2035 request_module("md-personality-%d", pnum);
2040 * Drop all container device buffers, from now on
2041 * the only valid external interface is through the md
2043 * Also find largest hardsector size
2045 ITERATE_RDEV(mddev,rdev,tmp) {
2046 if (test_bit(Faulty, &rdev->flags))
2048 sync_blockdev(rdev->bdev);
2049 invalidate_bdev(rdev->bdev, 0);
2052 md_probe(mddev->unit, NULL, NULL);
2053 disk = mddev->gendisk;
2057 spin_lock(&pers_lock);
2058 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2059 spin_unlock(&pers_lock);
2060 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2065 mddev->pers = pers[pnum];
2066 spin_unlock(&pers_lock);
2068 mddev->recovery = 0;
2069 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2070 mddev->barriers_work = 1;
2071 mddev->ok_start_degraded = start_dirty_degraded;
2074 mddev->ro = 2; /* read-only, but switch on first write */
2076 err = mddev->pers->run(mddev);
2077 if (!err && mddev->pers->sync_request) {
2078 err = bitmap_create(mddev);
2080 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2081 mdname(mddev), err);
2082 mddev->pers->stop(mddev);
2086 printk(KERN_ERR "md: pers->run() failed ...\n");
2087 module_put(mddev->pers->owner);
2089 bitmap_destroy(mddev);
2092 if (mddev->pers->sync_request)
2093 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2094 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2097 atomic_set(&mddev->writes_pending,0);
2098 mddev->safemode = 0;
2099 mddev->safemode_timer.function = md_safemode_timeout;
2100 mddev->safemode_timer.data = (unsigned long) mddev;
2101 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2104 ITERATE_RDEV(mddev,rdev,tmp)
2105 if (rdev->raid_disk >= 0) {
2107 sprintf(nm, "rd%d", rdev->raid_disk);
2108 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2111 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2112 md_wakeup_thread(mddev->thread);
2114 if (mddev->sb_dirty)
2115 md_update_sb(mddev);
2117 set_capacity(disk, mddev->array_size<<1);
2119 /* If we call blk_queue_make_request here, it will
2120 * re-initialise max_sectors etc which may have been
2121 * refined inside -> run. So just set the bits we need to set.
2122 * Most initialisation happended when we called
2123 * blk_queue_make_request(..., md_fail_request)
2126 mddev->queue->queuedata = mddev;
2127 mddev->queue->make_request_fn = mddev->pers->make_request;
2130 md_new_event(mddev);
2134 static int restart_array(mddev_t *mddev)
2136 struct gendisk *disk = mddev->gendisk;
2140 * Complain if it has no devices
2143 if (list_empty(&mddev->disks))
2151 mddev->safemode = 0;
2153 set_disk_ro(disk, 0);
2155 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2158 * Kick recovery or resync if necessary
2160 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2161 md_wakeup_thread(mddev->thread);
2164 printk(KERN_ERR "md: %s has no personality assigned.\n",
2173 static int do_md_stop(mddev_t * mddev, int ro)
2176 struct gendisk *disk = mddev->gendisk;
2179 if (atomic_read(&mddev->active)>2) {
2180 printk("md: %s still in use.\n",mdname(mddev));
2184 if (mddev->sync_thread) {
2185 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2186 md_unregister_thread(mddev->sync_thread);
2187 mddev->sync_thread = NULL;
2190 del_timer_sync(&mddev->safemode_timer);
2192 invalidate_partition(disk, 0);
2200 bitmap_flush(mddev);
2201 md_super_wait(mddev);
2203 set_disk_ro(disk, 0);
2204 blk_queue_make_request(mddev->queue, md_fail_request);
2205 mddev->pers->stop(mddev);
2206 if (mddev->pers->sync_request)
2207 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2209 module_put(mddev->pers->owner);
2214 if (!mddev->in_sync) {
2215 /* mark array as shutdown cleanly */
2217 md_update_sb(mddev);
2220 set_disk_ro(disk, 1);
2223 bitmap_destroy(mddev);
2224 if (mddev->bitmap_file) {
2225 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2226 fput(mddev->bitmap_file);
2227 mddev->bitmap_file = NULL;
2229 mddev->bitmap_offset = 0;
2232 * Free resources if final stop
2236 struct list_head *tmp;
2237 struct gendisk *disk;
2238 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2240 ITERATE_RDEV(mddev,rdev,tmp)
2241 if (rdev->raid_disk >= 0) {
2243 sprintf(nm, "rd%d", rdev->raid_disk);
2244 sysfs_remove_link(&mddev->kobj, nm);
2247 export_array(mddev);
2249 mddev->array_size = 0;
2250 disk = mddev->gendisk;
2252 set_capacity(disk, 0);
2255 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2258 md_new_event(mddev);
2263 static void autorun_array(mddev_t *mddev)
2266 struct list_head *tmp;
2269 if (list_empty(&mddev->disks))
2272 printk(KERN_INFO "md: running: ");
2274 ITERATE_RDEV(mddev,rdev,tmp) {
2275 char b[BDEVNAME_SIZE];
2276 printk("<%s>", bdevname(rdev->bdev,b));
2280 err = do_md_run (mddev);
2282 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2283 do_md_stop (mddev, 0);
2288 * lets try to run arrays based on all disks that have arrived
2289 * until now. (those are in pending_raid_disks)
2291 * the method: pick the first pending disk, collect all disks with
2292 * the same UUID, remove all from the pending list and put them into
2293 * the 'same_array' list. Then order this list based on superblock
2294 * update time (freshest comes first), kick out 'old' disks and
2295 * compare superblocks. If everything's fine then run it.
2297 * If "unit" is allocated, then bump its reference count
2299 static void autorun_devices(int part)
2301 struct list_head candidates;
2302 struct list_head *tmp;
2303 mdk_rdev_t *rdev0, *rdev;
2305 char b[BDEVNAME_SIZE];
2307 printk(KERN_INFO "md: autorun ...\n");
2308 while (!list_empty(&pending_raid_disks)) {
2310 rdev0 = list_entry(pending_raid_disks.next,
2311 mdk_rdev_t, same_set);
2313 printk(KERN_INFO "md: considering %s ...\n",
2314 bdevname(rdev0->bdev,b));
2315 INIT_LIST_HEAD(&candidates);
2316 ITERATE_RDEV_PENDING(rdev,tmp)
2317 if (super_90_load(rdev, rdev0, 0) >= 0) {
2318 printk(KERN_INFO "md: adding %s ...\n",
2319 bdevname(rdev->bdev,b));
2320 list_move(&rdev->same_set, &candidates);
2323 * now we have a set of devices, with all of them having
2324 * mostly sane superblocks. It's time to allocate the
2327 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2328 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2329 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2333 dev = MKDEV(mdp_major,
2334 rdev0->preferred_minor << MdpMinorShift);
2336 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2338 md_probe(dev, NULL, NULL);
2339 mddev = mddev_find(dev);
2342 "md: cannot allocate memory for md drive.\n");
2345 if (mddev_lock(mddev))
2346 printk(KERN_WARNING "md: %s locked, cannot run\n",
2348 else if (mddev->raid_disks || mddev->major_version
2349 || !list_empty(&mddev->disks)) {
2351 "md: %s already running, cannot run %s\n",
2352 mdname(mddev), bdevname(rdev0->bdev,b));
2353 mddev_unlock(mddev);
2355 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2356 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2357 list_del_init(&rdev->same_set);
2358 if (bind_rdev_to_array(rdev, mddev))
2361 autorun_array(mddev);
2362 mddev_unlock(mddev);
2364 /* on success, candidates will be empty, on error
2367 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2371 printk(KERN_INFO "md: ... autorun DONE.\n");
2375 * import RAID devices based on one partition
2376 * if possible, the array gets run as well.
2379 static int autostart_array(dev_t startdev)
2381 char b[BDEVNAME_SIZE];
2382 int err = -EINVAL, i;
2383 mdp_super_t *sb = NULL;
2384 mdk_rdev_t *start_rdev = NULL, *rdev;
2386 start_rdev = md_import_device(startdev, 0, 0);
2387 if (IS_ERR(start_rdev))
2391 /* NOTE: this can only work for 0.90.0 superblocks */
2392 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2393 if (sb->major_version != 0 ||
2394 sb->minor_version != 90 ) {
2395 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2396 export_rdev(start_rdev);
2400 if (test_bit(Faulty, &start_rdev->flags)) {
2402 "md: can not autostart based on faulty %s!\n",
2403 bdevname(start_rdev->bdev,b));
2404 export_rdev(start_rdev);
2407 list_add(&start_rdev->same_set, &pending_raid_disks);
2409 for (i = 0; i < MD_SB_DISKS; i++) {
2410 mdp_disk_t *desc = sb->disks + i;
2411 dev_t dev = MKDEV(desc->major, desc->minor);
2415 if (dev == startdev)
2417 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2419 rdev = md_import_device(dev, 0, 0);
2423 list_add(&rdev->same_set, &pending_raid_disks);
2427 * possibly return codes
2435 static int get_version(void __user * arg)
2439 ver.major = MD_MAJOR_VERSION;
2440 ver.minor = MD_MINOR_VERSION;
2441 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2443 if (copy_to_user(arg, &ver, sizeof(ver)))
2449 static int get_array_info(mddev_t * mddev, void __user * arg)
2451 mdu_array_info_t info;
2452 int nr,working,active,failed,spare;
2454 struct list_head *tmp;
2456 nr=working=active=failed=spare=0;
2457 ITERATE_RDEV(mddev,rdev,tmp) {
2459 if (test_bit(Faulty, &rdev->flags))
2463 if (test_bit(In_sync, &rdev->flags))
2470 info.major_version = mddev->major_version;
2471 info.minor_version = mddev->minor_version;
2472 info.patch_version = MD_PATCHLEVEL_VERSION;
2473 info.ctime = mddev->ctime;
2474 info.level = mddev->level;
2475 info.size = mddev->size;
2477 info.raid_disks = mddev->raid_disks;
2478 info.md_minor = mddev->md_minor;
2479 info.not_persistent= !mddev->persistent;
2481 info.utime = mddev->utime;
2484 info.state = (1<<MD_SB_CLEAN);
2485 if (mddev->bitmap && mddev->bitmap_offset)
2486 info.state = (1<<MD_SB_BITMAP_PRESENT);
2487 info.active_disks = active;
2488 info.working_disks = working;
2489 info.failed_disks = failed;
2490 info.spare_disks = spare;
2492 info.layout = mddev->layout;
2493 info.chunk_size = mddev->chunk_size;
2495 if (copy_to_user(arg, &info, sizeof(info)))
2501 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2503 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2504 char *ptr, *buf = NULL;
2507 file = kmalloc(sizeof(*file), GFP_KERNEL);
2511 /* bitmap disabled, zero the first byte and copy out */
2512 if (!mddev->bitmap || !mddev->bitmap->file) {
2513 file->pathname[0] = '\0';
2517 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2521 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2525 strcpy(file->pathname, ptr);
2529 if (copy_to_user(arg, file, sizeof(*file)))
2537 static int get_disk_info(mddev_t * mddev, void __user * arg)
2539 mdu_disk_info_t info;
2543 if (copy_from_user(&info, arg, sizeof(info)))
2548 rdev = find_rdev_nr(mddev, nr);
2550 info.major = MAJOR(rdev->bdev->bd_dev);
2551 info.minor = MINOR(rdev->bdev->bd_dev);
2552 info.raid_disk = rdev->raid_disk;
2554 if (test_bit(Faulty, &rdev->flags))
2555 info.state |= (1<<MD_DISK_FAULTY);
2556 else if (test_bit(In_sync, &rdev->flags)) {
2557 info.state |= (1<<MD_DISK_ACTIVE);
2558 info.state |= (1<<MD_DISK_SYNC);
2560 if (test_bit(WriteMostly, &rdev->flags))
2561 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2563 info.major = info.minor = 0;
2564 info.raid_disk = -1;
2565 info.state = (1<<MD_DISK_REMOVED);
2568 if (copy_to_user(arg, &info, sizeof(info)))
2574 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2576 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2578 dev_t dev = MKDEV(info->major,info->minor);
2580 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2583 if (!mddev->raid_disks) {
2585 /* expecting a device which has a superblock */
2586 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2589 "md: md_import_device returned %ld\n",
2591 return PTR_ERR(rdev);
2593 if (!list_empty(&mddev->disks)) {
2594 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2595 mdk_rdev_t, same_set);
2596 int err = super_types[mddev->major_version]
2597 .load_super(rdev, rdev0, mddev->minor_version);
2600 "md: %s has different UUID to %s\n",
2601 bdevname(rdev->bdev,b),
2602 bdevname(rdev0->bdev,b2));
2607 err = bind_rdev_to_array(rdev, mddev);
2614 * add_new_disk can be used once the array is assembled
2615 * to add "hot spares". They must already have a superblock
2620 if (!mddev->pers->hot_add_disk) {
2622 "%s: personality does not support diskops!\n",
2626 if (mddev->persistent)
2627 rdev = md_import_device(dev, mddev->major_version,
2628 mddev->minor_version);
2630 rdev = md_import_device(dev, -1, -1);
2633 "md: md_import_device returned %ld\n",
2635 return PTR_ERR(rdev);
2637 /* set save_raid_disk if appropriate */
2638 if (!mddev->persistent) {
2639 if (info->state & (1<<MD_DISK_SYNC) &&
2640 info->raid_disk < mddev->raid_disks)
2641 rdev->raid_disk = info->raid_disk;
2643 rdev->raid_disk = -1;
2645 super_types[mddev->major_version].
2646 validate_super(mddev, rdev);
2647 rdev->saved_raid_disk = rdev->raid_disk;
2649 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2650 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2651 set_bit(WriteMostly, &rdev->flags);
2653 rdev->raid_disk = -1;
2654 err = bind_rdev_to_array(rdev, mddev);
2658 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2659 md_wakeup_thread(mddev->thread);
2663 /* otherwise, add_new_disk is only allowed
2664 * for major_version==0 superblocks
2666 if (mddev->major_version != 0) {
2667 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2672 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2674 rdev = md_import_device (dev, -1, 0);
2677 "md: error, md_import_device() returned %ld\n",
2679 return PTR_ERR(rdev);
2681 rdev->desc_nr = info->number;
2682 if (info->raid_disk < mddev->raid_disks)
2683 rdev->raid_disk = info->raid_disk;
2685 rdev->raid_disk = -1;
2689 if (rdev->raid_disk < mddev->raid_disks)
2690 if (info->state & (1<<MD_DISK_SYNC))
2691 set_bit(In_sync, &rdev->flags);
2693 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2694 set_bit(WriteMostly, &rdev->flags);
2696 err = bind_rdev_to_array(rdev, mddev);
2702 if (!mddev->persistent) {
2703 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2704 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2706 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2707 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2709 if (!mddev->size || (mddev->size > rdev->size))
2710 mddev->size = rdev->size;
2716 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2718 char b[BDEVNAME_SIZE];
2724 rdev = find_rdev(mddev, dev);
2728 if (rdev->raid_disk >= 0)
2731 kick_rdev_from_array(rdev);
2732 md_update_sb(mddev);
2733 md_new_event(mddev);
2737 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2738 bdevname(rdev->bdev,b), mdname(mddev));
2742 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2744 char b[BDEVNAME_SIZE];
2752 if (mddev->major_version != 0) {
2753 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2754 " version-0 superblocks.\n",
2758 if (!mddev->pers->hot_add_disk) {
2760 "%s: personality does not support diskops!\n",
2765 rdev = md_import_device (dev, -1, 0);
2768 "md: error, md_import_device() returned %ld\n",
2773 if (mddev->persistent)
2774 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2777 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2779 size = calc_dev_size(rdev, mddev->chunk_size);
2782 if (size < mddev->size) {
2784 "%s: disk size %llu blocks < array size %llu\n",
2785 mdname(mddev), (unsigned long long)size,
2786 (unsigned long long)mddev->size);
2791 if (test_bit(Faulty, &rdev->flags)) {
2793 "md: can not hot-add faulty %s disk to %s!\n",
2794 bdevname(rdev->bdev,b), mdname(mddev));
2798 clear_bit(In_sync, &rdev->flags);
2800 bind_rdev_to_array(rdev, mddev);
2803 * The rest should better be atomic, we can have disk failures
2804 * noticed in interrupt contexts ...
2807 if (rdev->desc_nr == mddev->max_disks) {
2808 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2811 goto abort_unbind_export;
2814 rdev->raid_disk = -1;
2816 md_update_sb(mddev);
2819 * Kick recovery, maybe this spare has to be added to the
2820 * array immediately.
2822 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2823 md_wakeup_thread(mddev->thread);
2824 md_new_event(mddev);
2827 abort_unbind_export:
2828 unbind_rdev_from_array(rdev);
2835 /* similar to deny_write_access, but accounts for our holding a reference
2836 * to the file ourselves */
2837 static int deny_bitmap_write_access(struct file * file)
2839 struct inode *inode = file->f_mapping->host;
2841 spin_lock(&inode->i_lock);
2842 if (atomic_read(&inode->i_writecount) > 1) {
2843 spin_unlock(&inode->i_lock);
2846 atomic_set(&inode->i_writecount, -1);
2847 spin_unlock(&inode->i_lock);
2852 static int set_bitmap_file(mddev_t *mddev, int fd)
2857 if (!mddev->pers->quiesce)
2859 if (mddev->recovery || mddev->sync_thread)
2861 /* we should be able to change the bitmap.. */
2867 return -EEXIST; /* cannot add when bitmap is present */
2868 mddev->bitmap_file = fget(fd);
2870 if (mddev->bitmap_file == NULL) {
2871 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2876 err = deny_bitmap_write_access(mddev->bitmap_file);
2878 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2880 fput(mddev->bitmap_file);
2881 mddev->bitmap_file = NULL;
2884 mddev->bitmap_offset = 0; /* file overrides offset */
2885 } else if (mddev->bitmap == NULL)
2886 return -ENOENT; /* cannot remove what isn't there */
2889 mddev->pers->quiesce(mddev, 1);
2891 err = bitmap_create(mddev);
2893 bitmap_destroy(mddev);
2894 mddev->pers->quiesce(mddev, 0);
2895 } else if (fd < 0) {
2896 if (mddev->bitmap_file)
2897 fput(mddev->bitmap_file);
2898 mddev->bitmap_file = NULL;
2905 * set_array_info is used two different ways
2906 * The original usage is when creating a new array.
2907 * In this usage, raid_disks is > 0 and it together with
2908 * level, size, not_persistent,layout,chunksize determine the
2909 * shape of the array.
2910 * This will always create an array with a type-0.90.0 superblock.
2911 * The newer usage is when assembling an array.
2912 * In this case raid_disks will be 0, and the major_version field is
2913 * use to determine which style super-blocks are to be found on the devices.
2914 * The minor and patch _version numbers are also kept incase the
2915 * super_block handler wishes to interpret them.
2917 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2920 if (info->raid_disks == 0) {
2921 /* just setting version number for superblock loading */
2922 if (info->major_version < 0 ||
2923 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2924 super_types[info->major_version].name == NULL) {
2925 /* maybe try to auto-load a module? */
2927 "md: superblock version %d not known\n",
2928 info->major_version);
2931 mddev->major_version = info->major_version;
2932 mddev->minor_version = info->minor_version;
2933 mddev->patch_version = info->patch_version;
2936 mddev->major_version = MD_MAJOR_VERSION;
2937 mddev->minor_version = MD_MINOR_VERSION;
2938 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2939 mddev->ctime = get_seconds();
2941 mddev->level = info->level;
2942 mddev->size = info->size;
2943 mddev->raid_disks = info->raid_disks;
2944 /* don't set md_minor, it is determined by which /dev/md* was
2947 if (info->state & (1<<MD_SB_CLEAN))
2948 mddev->recovery_cp = MaxSector;
2950 mddev->recovery_cp = 0;
2951 mddev->persistent = ! info->not_persistent;
2953 mddev->layout = info->layout;
2954 mddev->chunk_size = info->chunk_size;
2956 mddev->max_disks = MD_SB_DISKS;
2958 mddev->sb_dirty = 1;
2960 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2961 mddev->bitmap_offset = 0;
2964 * Generate a 128 bit UUID
2966 get_random_bytes(mddev->uuid, 16);
2972 * update_array_info is used to change the configuration of an
2974 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2975 * fields in the info are checked against the array.
2976 * Any differences that cannot be handled will cause an error.
2977 * Normally, only one change can be managed at a time.
2979 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2985 /* calculate expected state,ignoring low bits */
2986 if (mddev->bitmap && mddev->bitmap_offset)
2987 state |= (1 << MD_SB_BITMAP_PRESENT);
2989 if (mddev->major_version != info->major_version ||
2990 mddev->minor_version != info->minor_version ||
2991 /* mddev->patch_version != info->patch_version || */
2992 mddev->ctime != info->ctime ||
2993 mddev->level != info->level ||
2994 /* mddev->layout != info->layout || */
2995 !mddev->persistent != info->not_persistent||
2996 mddev->chunk_size != info->chunk_size ||
2997 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2998 ((state^info->state) & 0xfffffe00)
3001 /* Check there is only one change */
3002 if (mddev->size != info->size) cnt++;
3003 if (mddev->raid_disks != info->raid_disks) cnt++;
3004 if (mddev->layout != info->layout) cnt++;
3005 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3006 if (cnt == 0) return 0;
3007 if (cnt > 1) return -EINVAL;
3009 if (mddev->layout != info->layout) {
3011 * we don't need to do anything at the md level, the
3012 * personality will take care of it all.
3014 if (mddev->pers->reconfig == NULL)
3017 return mddev->pers->reconfig(mddev, info->layout, -1);
3019 if (mddev->size != info->size) {
3021 struct list_head *tmp;
3022 if (mddev->pers->resize == NULL)
3024 /* The "size" is the amount of each device that is used.
3025 * This can only make sense for arrays with redundancy.
3026 * linear and raid0 always use whatever space is available
3027 * We can only consider changing the size if no resync
3028 * or reconstruction is happening, and if the new size
3029 * is acceptable. It must fit before the sb_offset or,
3030 * if that is <data_offset, it must fit before the
3031 * size of each device.
3032 * If size is zero, we find the largest size that fits.
3034 if (mddev->sync_thread)
3036 ITERATE_RDEV(mddev,rdev,tmp) {
3038 int fit = (info->size == 0);
3039 if (rdev->sb_offset > rdev->data_offset)
3040 avail = (rdev->sb_offset*2) - rdev->data_offset;
3042 avail = get_capacity(rdev->bdev->bd_disk)
3043 - rdev->data_offset;
3044 if (fit && (info->size == 0 || info->size > avail/2))
3045 info->size = avail/2;
3046 if (avail < ((sector_t)info->size << 1))
3049 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3051 struct block_device *bdev;
3053 bdev = bdget_disk(mddev->gendisk, 0);
3055 down(&bdev->bd_inode->i_sem);
3056 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3057 up(&bdev->bd_inode->i_sem);
3062 if (mddev->raid_disks != info->raid_disks) {
3063 /* change the number of raid disks */
3064 if (mddev->pers->reshape == NULL)
3066 if (info->raid_disks <= 0 ||
3067 info->raid_disks >= mddev->max_disks)
3069 if (mddev->sync_thread)
3071 rv = mddev->pers->reshape(mddev, info->raid_disks);
3073 struct block_device *bdev;
3075 bdev = bdget_disk(mddev->gendisk, 0);
3077 down(&bdev->bd_inode->i_sem);
3078 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3079 up(&bdev->bd_inode->i_sem);
3084 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3085 if (mddev->pers->quiesce == NULL)
3087 if (mddev->recovery || mddev->sync_thread)
3089 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3090 /* add the bitmap */
3093 if (mddev->default_bitmap_offset == 0)
3095 mddev->bitmap_offset = mddev->default_bitmap_offset;
3096 mddev->pers->quiesce(mddev, 1);
3097 rv = bitmap_create(mddev);
3099 bitmap_destroy(mddev);
3100 mddev->pers->quiesce(mddev, 0);
3102 /* remove the bitmap */
3105 if (mddev->bitmap->file)
3107 mddev->pers->quiesce(mddev, 1);
3108 bitmap_destroy(mddev);
3109 mddev->pers->quiesce(mddev, 0);
3110 mddev->bitmap_offset = 0;
3113 md_update_sb(mddev);
3117 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3121 if (mddev->pers == NULL)
3124 rdev = find_rdev(mddev, dev);
3128 md_error(mddev, rdev);
3132 static int md_ioctl(struct inode *inode, struct file *file,
3133 unsigned int cmd, unsigned long arg)
3136 void __user *argp = (void __user *)arg;
3137 struct hd_geometry __user *loc = argp;
3138 mddev_t *mddev = NULL;
3140 if (!capable(CAP_SYS_ADMIN))
3144 * Commands dealing with the RAID driver but not any
3150 err = get_version(argp);
3153 case PRINT_RAID_DEBUG:
3161 autostart_arrays(arg);
3168 * Commands creating/starting a new array:
3171 mddev = inode->i_bdev->bd_disk->private_data;
3179 if (cmd == START_ARRAY) {
3180 /* START_ARRAY doesn't need to lock the array as autostart_array
3181 * does the locking, and it could even be a different array
3186 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3187 "This will not be supported beyond July 2006\n",
3188 current->comm, current->pid);
3191 err = autostart_array(new_decode_dev(arg));
3193 printk(KERN_WARNING "md: autostart failed!\n");
3199 err = mddev_lock(mddev);
3202 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3209 case SET_ARRAY_INFO:
3211 mdu_array_info_t info;
3213 memset(&info, 0, sizeof(info));
3214 else if (copy_from_user(&info, argp, sizeof(info))) {
3219 err = update_array_info(mddev, &info);
3221 printk(KERN_WARNING "md: couldn't update"
3222 " array info. %d\n", err);
3227 if (!list_empty(&mddev->disks)) {
3229 "md: array %s already has disks!\n",
3234 if (mddev->raid_disks) {
3236 "md: array %s already initialised!\n",
3241 err = set_array_info(mddev, &info);
3243 printk(KERN_WARNING "md: couldn't set"
3244 " array info. %d\n", err);
3254 * Commands querying/configuring an existing array:
3256 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3257 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3258 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3259 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3265 * Commands even a read-only array can execute:
3269 case GET_ARRAY_INFO:
3270 err = get_array_info(mddev, argp);
3273 case GET_BITMAP_FILE:
3274 err = get_bitmap_file(mddev, argp);
3278 err = get_disk_info(mddev, argp);
3281 case RESTART_ARRAY_RW:
3282 err = restart_array(mddev);
3286 err = do_md_stop (mddev, 0);
3290 err = do_md_stop (mddev, 1);
3294 * We have a problem here : there is no easy way to give a CHS
3295 * virtual geometry. We currently pretend that we have a 2 heads
3296 * 4 sectors (with a BIG number of cylinders...). This drives
3297 * dosfs just mad... ;-)
3304 err = put_user (2, (char __user *) &loc->heads);
3307 err = put_user (4, (char __user *) &loc->sectors);
3310 err = put_user(get_capacity(mddev->gendisk)/8,
3311 (short __user *) &loc->cylinders);
3314 err = put_user (get_start_sect(inode->i_bdev),
3315 (long __user *) &loc->start);
3320 * The remaining ioctls are changing the state of the
3321 * superblock, so we do not allow them on read-only arrays.
3322 * However non-MD ioctls (e.g. get-size) will still come through
3323 * here and hit the 'default' below, so only disallow
3324 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3326 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3327 mddev->ro && mddev->pers) {
3328 if (mddev->ro == 2) {
3330 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3331 md_wakeup_thread(mddev->thread);
3343 mdu_disk_info_t info;
3344 if (copy_from_user(&info, argp, sizeof(info)))
3347 err = add_new_disk(mddev, &info);
3351 case HOT_REMOVE_DISK:
3352 err = hot_remove_disk(mddev, new_decode_dev(arg));
3356 err = hot_add_disk(mddev, new_decode_dev(arg));
3359 case SET_DISK_FAULTY:
3360 err = set_disk_faulty(mddev, new_decode_dev(arg));
3364 err = do_md_run (mddev);
3367 case SET_BITMAP_FILE:
3368 err = set_bitmap_file(mddev, (int)arg);
3372 if (_IOC_TYPE(cmd) == MD_MAJOR)
3373 printk(KERN_WARNING "md: %s(pid %d) used"
3374 " obsolete MD ioctl, upgrade your"
3375 " software to use new ictls.\n",
3376 current->comm, current->pid);
3383 mddev_unlock(mddev);
3393 static int md_open(struct inode *inode, struct file *file)
3396 * Succeed if we can lock the mddev, which confirms that
3397 * it isn't being stopped right now.
3399 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3402 if ((err = mddev_lock(mddev)))
3407 mddev_unlock(mddev);
3409 check_disk_change(inode->i_bdev);
3414 static int md_release(struct inode *inode, struct file * file)
3416 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3425 static int md_media_changed(struct gendisk *disk)
3427 mddev_t *mddev = disk->private_data;
3429 return mddev->changed;
3432 static int md_revalidate(struct gendisk *disk)
3434 mddev_t *mddev = disk->private_data;
3439 static struct block_device_operations md_fops =
3441 .owner = THIS_MODULE,
3443 .release = md_release,
3445 .media_changed = md_media_changed,
3446 .revalidate_disk= md_revalidate,
3449 static int md_thread(void * arg)
3451 mdk_thread_t *thread = arg;
3454 * md_thread is a 'system-thread', it's priority should be very
3455 * high. We avoid resource deadlocks individually in each
3456 * raid personality. (RAID5 does preallocation) We also use RR and
3457 * the very same RT priority as kswapd, thus we will never get
3458 * into a priority inversion deadlock.
3460 * we definitely have to have equal or higher priority than
3461 * bdflush, otherwise bdflush will deadlock if there are too
3462 * many dirty RAID5 blocks.
3465 allow_signal(SIGKILL);
3466 while (!kthread_should_stop()) {
3468 /* We need to wait INTERRUPTIBLE so that
3469 * we don't add to the load-average.
3470 * That means we need to be sure no signals are
3473 if (signal_pending(current))
3474 flush_signals(current);
3476 wait_event_interruptible_timeout
3478 test_bit(THREAD_WAKEUP, &thread->flags)
3479 || kthread_should_stop(),
3483 clear_bit(THREAD_WAKEUP, &thread->flags);
3485 thread->run(thread->mddev);
3491 void md_wakeup_thread(mdk_thread_t *thread)
3494 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3495 set_bit(THREAD_WAKEUP, &thread->flags);
3496 wake_up(&thread->wqueue);
3500 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3503 mdk_thread_t *thread;
3505 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3509 init_waitqueue_head(&thread->wqueue);
3512 thread->mddev = mddev;
3513 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3514 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3515 if (IS_ERR(thread->tsk)) {
3522 void md_unregister_thread(mdk_thread_t *thread)
3524 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3526 kthread_stop(thread->tsk);
3530 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3537 if (!rdev || test_bit(Faulty, &rdev->flags))
3540 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3542 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3543 __builtin_return_address(0),__builtin_return_address(1),
3544 __builtin_return_address(2),__builtin_return_address(3));
3546 if (!mddev->pers->error_handler)
3548 mddev->pers->error_handler(mddev,rdev);
3549 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3550 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3551 md_wakeup_thread(mddev->thread);
3552 md_new_event(mddev);
3555 /* seq_file implementation /proc/mdstat */
3557 static void status_unused(struct seq_file *seq)
3561 struct list_head *tmp;
3563 seq_printf(seq, "unused devices: ");
3565 ITERATE_RDEV_PENDING(rdev,tmp) {
3566 char b[BDEVNAME_SIZE];
3568 seq_printf(seq, "%s ",
3569 bdevname(rdev->bdev,b));
3572 seq_printf(seq, "<none>");
3574 seq_printf(seq, "\n");
3578 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3580 unsigned long max_blocks, resync, res, dt, db, rt;
3582 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3584 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3585 max_blocks = mddev->resync_max_sectors >> 1;
3587 max_blocks = mddev->size;
3590 * Should not happen.
3596 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3598 int i, x = res/50, y = 20-x;
3599 seq_printf(seq, "[");
3600 for (i = 0; i < x; i++)
3601 seq_printf(seq, "=");
3602 seq_printf(seq, ">");
3603 for (i = 0; i < y; i++)
3604 seq_printf(seq, ".");
3605 seq_printf(seq, "] ");
3607 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3608 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3609 "resync" : "recovery"),
3610 res/10, res % 10, resync, max_blocks);
3613 * We do not want to overflow, so the order of operands and
3614 * the * 100 / 100 trick are important. We do a +1 to be
3615 * safe against division by zero. We only estimate anyway.
3617 * dt: time from mark until now
3618 * db: blocks written from mark until now
3619 * rt: remaining time
3621 dt = ((jiffies - mddev->resync_mark) / HZ);
3623 db = resync - (mddev->resync_mark_cnt/2);
3624 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3626 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3628 seq_printf(seq, " speed=%ldK/sec", db/dt);
3631 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3633 struct list_head *tmp;
3643 spin_lock(&all_mddevs_lock);
3644 list_for_each(tmp,&all_mddevs)
3646 mddev = list_entry(tmp, mddev_t, all_mddevs);
3648 spin_unlock(&all_mddevs_lock);
3651 spin_unlock(&all_mddevs_lock);
3653 return (void*)2;/* tail */
3657 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3659 struct list_head *tmp;
3660 mddev_t *next_mddev, *mddev = v;
3666 spin_lock(&all_mddevs_lock);
3668 tmp = all_mddevs.next;
3670 tmp = mddev->all_mddevs.next;
3671 if (tmp != &all_mddevs)
3672 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3674 next_mddev = (void*)2;
3677 spin_unlock(&all_mddevs_lock);
3685 static void md_seq_stop(struct seq_file *seq, void *v)
3689 if (mddev && v != (void*)1 && v != (void*)2)
3693 struct mdstat_info {
3697 static int md_seq_show(struct seq_file *seq, void *v)
3701 struct list_head *tmp2;
3703 struct mdstat_info *mi = seq->private;
3705 struct bitmap *bitmap;
3707 if (v == (void*)1) {
3708 seq_printf(seq, "Personalities : ");
3709 spin_lock(&pers_lock);
3710 for (i = 0; i < MAX_PERSONALITY; i++)
3712 seq_printf(seq, "[%s] ", pers[i]->name);
3714 spin_unlock(&pers_lock);
3715 seq_printf(seq, "\n");
3716 mi->event = atomic_read(&md_event_count);
3719 if (v == (void*)2) {
3724 if (mddev_lock(mddev)!=0)
3726 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3727 seq_printf(seq, "%s : %sactive", mdname(mddev),
3728 mddev->pers ? "" : "in");
3731 seq_printf(seq, " (read-only)");
3733 seq_printf(seq, "(auto-read-only)");
3734 seq_printf(seq, " %s", mddev->pers->name);
3738 ITERATE_RDEV(mddev,rdev,tmp2) {
3739 char b[BDEVNAME_SIZE];
3740 seq_printf(seq, " %s[%d]",
3741 bdevname(rdev->bdev,b), rdev->desc_nr);
3742 if (test_bit(WriteMostly, &rdev->flags))
3743 seq_printf(seq, "(W)");
3744 if (test_bit(Faulty, &rdev->flags)) {
3745 seq_printf(seq, "(F)");
3747 } else if (rdev->raid_disk < 0)
3748 seq_printf(seq, "(S)"); /* spare */
3752 if (!list_empty(&mddev->disks)) {
3754 seq_printf(seq, "\n %llu blocks",
3755 (unsigned long long)mddev->array_size);
3757 seq_printf(seq, "\n %llu blocks",
3758 (unsigned long long)size);
3760 if (mddev->persistent) {
3761 if (mddev->major_version != 0 ||
3762 mddev->minor_version != 90) {
3763 seq_printf(seq," super %d.%d",
3764 mddev->major_version,
3765 mddev->minor_version);
3768 seq_printf(seq, " super non-persistent");
3771 mddev->pers->status (seq, mddev);
3772 seq_printf(seq, "\n ");
3773 if (mddev->pers->sync_request) {
3774 if (mddev->curr_resync > 2) {
3775 status_resync (seq, mddev);
3776 seq_printf(seq, "\n ");
3777 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3778 seq_printf(seq, "\tresync=DELAYED\n ");
3779 else if (mddev->recovery_cp < MaxSector)
3780 seq_printf(seq, "\tresync=PENDING\n ");
3783 seq_printf(seq, "\n ");
3785 if ((bitmap = mddev->bitmap)) {
3786 unsigned long chunk_kb;
3787 unsigned long flags;
3788 spin_lock_irqsave(&bitmap->lock, flags);
3789 chunk_kb = bitmap->chunksize >> 10;
3790 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3792 bitmap->pages - bitmap->missing_pages,
3794 (bitmap->pages - bitmap->missing_pages)
3795 << (PAGE_SHIFT - 10),
3796 chunk_kb ? chunk_kb : bitmap->chunksize,
3797 chunk_kb ? "KB" : "B");
3799 seq_printf(seq, ", file: ");
3800 seq_path(seq, bitmap->file->f_vfsmnt,
3801 bitmap->file->f_dentry," \t\n");
3804 seq_printf(seq, "\n");
3805 spin_unlock_irqrestore(&bitmap->lock, flags);
3808 seq_printf(seq, "\n");
3810 mddev_unlock(mddev);
3815 static struct seq_operations md_seq_ops = {
3816 .start = md_seq_start,
3817 .next = md_seq_next,
3818 .stop = md_seq_stop,
3819 .show = md_seq_show,
3822 static int md_seq_open(struct inode *inode, struct file *file)
3825 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3829 error = seq_open(file, &md_seq_ops);
3833 struct seq_file *p = file->private_data;
3835 mi->event = atomic_read(&md_event_count);
3840 static int md_seq_release(struct inode *inode, struct file *file)
3842 struct seq_file *m = file->private_data;
3843 struct mdstat_info *mi = m->private;
3846 return seq_release(inode, file);
3849 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
3851 struct seq_file *m = filp->private_data;
3852 struct mdstat_info *mi = m->private;
3855 poll_wait(filp, &md_event_waiters, wait);
3857 /* always allow read */
3858 mask = POLLIN | POLLRDNORM;
3860 if (mi->event != atomic_read(&md_event_count))
3861 mask |= POLLERR | POLLPRI;
3865 static struct file_operations md_seq_fops = {
3866 .open = md_seq_open,
3868 .llseek = seq_lseek,
3869 .release = md_seq_release,
3870 .poll = mdstat_poll,
3873 int register_md_personality(int pnum, mdk_personality_t *p)
3875 if (pnum >= MAX_PERSONALITY) {
3877 "md: tried to install personality %s as nr %d, but max is %lu\n",
3878 p->name, pnum, MAX_PERSONALITY-1);
3882 spin_lock(&pers_lock);
3884 spin_unlock(&pers_lock);
3889 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3890 spin_unlock(&pers_lock);
3894 int unregister_md_personality(int pnum)
3896 if (pnum >= MAX_PERSONALITY)
3899 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3900 spin_lock(&pers_lock);
3902 spin_unlock(&pers_lock);
3906 static int is_mddev_idle(mddev_t *mddev)
3909 struct list_head *tmp;
3911 unsigned long curr_events;
3914 ITERATE_RDEV(mddev,rdev,tmp) {
3915 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3916 curr_events = disk_stat_read(disk, sectors[0]) +
3917 disk_stat_read(disk, sectors[1]) -
3918 atomic_read(&disk->sync_io);
3919 /* The difference between curr_events and last_events
3920 * will be affected by any new non-sync IO (making
3921 * curr_events bigger) and any difference in the amount of
3922 * in-flight syncio (making current_events bigger or smaller)
3923 * The amount in-flight is currently limited to
3924 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3925 * which is at most 4096 sectors.
3926 * These numbers are fairly fragile and should be made
3927 * more robust, probably by enforcing the
3928 * 'window size' that md_do_sync sort-of uses.
3930 * Note: the following is an unsigned comparison.
3932 if ((curr_events - rdev->last_events + 4096) > 8192) {
3933 rdev->last_events = curr_events;
3940 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3942 /* another "blocks" (512byte) blocks have been synced */
3943 atomic_sub(blocks, &mddev->recovery_active);
3944 wake_up(&mddev->recovery_wait);
3946 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3947 md_wakeup_thread(mddev->thread);
3948 // stop recovery, signal do_sync ....
3953 /* md_write_start(mddev, bi)
3954 * If we need to update some array metadata (e.g. 'active' flag
3955 * in superblock) before writing, schedule a superblock update
3956 * and wait for it to complete.
3958 void md_write_start(mddev_t *mddev, struct bio *bi)
3960 if (bio_data_dir(bi) != WRITE)
3963 BUG_ON(mddev->ro == 1);
3964 if (mddev->ro == 2) {
3965 /* need to switch to read/write */
3967 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3968 md_wakeup_thread(mddev->thread);
3970 atomic_inc(&mddev->writes_pending);
3971 if (mddev->in_sync) {
3972 spin_lock_irq(&mddev->write_lock);
3973 if (mddev->in_sync) {
3975 mddev->sb_dirty = 1;
3976 md_wakeup_thread(mddev->thread);
3978 spin_unlock_irq(&mddev->write_lock);
3980 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3983 void md_write_end(mddev_t *mddev)
3985 if (atomic_dec_and_test(&mddev->writes_pending)) {
3986 if (mddev->safemode == 2)
3987 md_wakeup_thread(mddev->thread);
3989 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3993 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3995 #define SYNC_MARKS 10
3996 #define SYNC_MARK_STEP (3*HZ)
3997 static void md_do_sync(mddev_t *mddev)
4000 unsigned int currspeed = 0,
4002 sector_t max_sectors,j, io_sectors;
4003 unsigned long mark[SYNC_MARKS];
4004 sector_t mark_cnt[SYNC_MARKS];
4006 struct list_head *tmp;
4007 sector_t last_check;
4010 /* just incase thread restarts... */
4011 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4014 /* we overload curr_resync somewhat here.
4015 * 0 == not engaged in resync at all
4016 * 2 == checking that there is no conflict with another sync
4017 * 1 == like 2, but have yielded to allow conflicting resync to
4019 * other == active in resync - this many blocks
4021 * Before starting a resync we must have set curr_resync to
4022 * 2, and then checked that every "conflicting" array has curr_resync
4023 * less than ours. When we find one that is the same or higher
4024 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4025 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4026 * This will mean we have to start checking from the beginning again.
4031 mddev->curr_resync = 2;
4034 if (kthread_should_stop()) {
4035 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4038 ITERATE_MDDEV(mddev2,tmp) {
4039 if (mddev2 == mddev)
4041 if (mddev2->curr_resync &&
4042 match_mddev_units(mddev,mddev2)) {
4044 if (mddev < mddev2 && mddev->curr_resync == 2) {
4045 /* arbitrarily yield */
4046 mddev->curr_resync = 1;
4047 wake_up(&resync_wait);
4049 if (mddev > mddev2 && mddev->curr_resync == 1)
4050 /* no need to wait here, we can wait the next
4051 * time 'round when curr_resync == 2
4054 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4055 if (!kthread_should_stop() &&
4056 mddev2->curr_resync >= mddev->curr_resync) {
4057 printk(KERN_INFO "md: delaying resync of %s"
4058 " until %s has finished resync (they"
4059 " share one or more physical units)\n",
4060 mdname(mddev), mdname(mddev2));
4063 finish_wait(&resync_wait, &wq);
4066 finish_wait(&resync_wait, &wq);
4069 } while (mddev->curr_resync < 2);
4071 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4072 /* resync follows the size requested by the personality,
4073 * which defaults to physical size, but can be virtual size
4075 max_sectors = mddev->resync_max_sectors;
4076 mddev->resync_mismatches = 0;
4078 /* recovery follows the physical size of devices */
4079 max_sectors = mddev->size << 1;
4081 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4082 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4083 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4084 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4085 "(but not more than %d KB/sec) for reconstruction.\n",
4086 sysctl_speed_limit_max);
4088 is_mddev_idle(mddev); /* this also initializes IO event counters */
4089 /* we don't use the checkpoint if there's a bitmap */
4090 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4091 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4092 j = mddev->recovery_cp;
4096 for (m = 0; m < SYNC_MARKS; m++) {
4098 mark_cnt[m] = io_sectors;
4101 mddev->resync_mark = mark[last_mark];
4102 mddev->resync_mark_cnt = mark_cnt[last_mark];
4105 * Tune reconstruction:
4107 window = 32*(PAGE_SIZE/512);
4108 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4109 window/2,(unsigned long long) max_sectors/2);
4111 atomic_set(&mddev->recovery_active, 0);
4112 init_waitqueue_head(&mddev->recovery_wait);
4117 "md: resuming recovery of %s from checkpoint.\n",
4119 mddev->curr_resync = j;
4122 while (j < max_sectors) {
4126 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4127 currspeed < sysctl_speed_limit_min);
4129 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4133 if (!skipped) { /* actual IO requested */
4134 io_sectors += sectors;
4135 atomic_add(sectors, &mddev->recovery_active);
4139 if (j>1) mddev->curr_resync = j;
4140 if (last_check == 0)
4141 /* this is the earliers that rebuilt will be
4142 * visible in /proc/mdstat
4144 md_new_event(mddev);
4146 if (last_check + window > io_sectors || j == max_sectors)
4149 last_check = io_sectors;
4151 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4152 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4156 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4158 int next = (last_mark+1) % SYNC_MARKS;
4160 mddev->resync_mark = mark[next];
4161 mddev->resync_mark_cnt = mark_cnt[next];
4162 mark[next] = jiffies;
4163 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4168 if (kthread_should_stop()) {
4170 * got a signal, exit.
4173 "md: md_do_sync() got signal ... exiting\n");
4174 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4179 * this loop exits only if either when we are slower than
4180 * the 'hard' speed limit, or the system was IO-idle for
4182 * the system might be non-idle CPU-wise, but we only care
4183 * about not overloading the IO subsystem. (things like an
4184 * e2fsck being done on the RAID array should execute fast)
4186 mddev->queue->unplug_fn(mddev->queue);
4189 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4190 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4192 if (currspeed > sysctl_speed_limit_min) {
4193 if ((currspeed > sysctl_speed_limit_max) ||
4194 !is_mddev_idle(mddev)) {
4200 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4202 * this also signals 'finished resyncing' to md_stop
4205 mddev->queue->unplug_fn(mddev->queue);
4207 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4209 /* tell personality that we are finished */
4210 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4212 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4213 mddev->curr_resync > 2 &&
4214 mddev->curr_resync >= mddev->recovery_cp) {
4215 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4217 "md: checkpointing recovery of %s.\n",
4219 mddev->recovery_cp = mddev->curr_resync;
4221 mddev->recovery_cp = MaxSector;
4225 mddev->curr_resync = 0;
4226 wake_up(&resync_wait);
4227 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4228 md_wakeup_thread(mddev->thread);
4233 * This routine is regularly called by all per-raid-array threads to
4234 * deal with generic issues like resync and super-block update.
4235 * Raid personalities that don't have a thread (linear/raid0) do not
4236 * need this as they never do any recovery or update the superblock.
4238 * It does not do any resync itself, but rather "forks" off other threads
4239 * to do that as needed.
4240 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4241 * "->recovery" and create a thread at ->sync_thread.
4242 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4243 * and wakeups up this thread which will reap the thread and finish up.
4244 * This thread also removes any faulty devices (with nr_pending == 0).
4246 * The overall approach is:
4247 * 1/ if the superblock needs updating, update it.
4248 * 2/ If a recovery thread is running, don't do anything else.
4249 * 3/ If recovery has finished, clean up, possibly marking spares active.
4250 * 4/ If there are any faulty devices, remove them.
4251 * 5/ If array is degraded, try to add spares devices
4252 * 6/ If array has spares or is not in-sync, start a resync thread.
4254 void md_check_recovery(mddev_t *mddev)
4257 struct list_head *rtmp;
4261 bitmap_daemon_work(mddev->bitmap);
4266 if (signal_pending(current)) {
4267 if (mddev->pers->sync_request) {
4268 printk(KERN_INFO "md: %s in immediate safe mode\n",
4270 mddev->safemode = 2;
4272 flush_signals(current);
4277 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4278 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4279 (mddev->safemode == 1) ||
4280 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4281 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4285 if (mddev_trylock(mddev)==0) {
4288 spin_lock_irq(&mddev->write_lock);
4289 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4290 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4292 mddev->sb_dirty = 1;
4294 if (mddev->safemode == 1)
4295 mddev->safemode = 0;
4296 spin_unlock_irq(&mddev->write_lock);
4298 if (mddev->sb_dirty)
4299 md_update_sb(mddev);
4302 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4303 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4304 /* resync/recovery still happening */
4305 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4308 if (mddev->sync_thread) {
4309 /* resync has finished, collect result */
4310 md_unregister_thread(mddev->sync_thread);
4311 mddev->sync_thread = NULL;
4312 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4313 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4315 /* activate any spares */
4316 mddev->pers->spare_active(mddev);
4318 md_update_sb(mddev);
4320 /* if array is no-longer degraded, then any saved_raid_disk
4321 * information must be scrapped
4323 if (!mddev->degraded)
4324 ITERATE_RDEV(mddev,rdev,rtmp)
4325 rdev->saved_raid_disk = -1;
4327 mddev->recovery = 0;
4328 /* flag recovery needed just to double check */
4329 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4330 md_new_event(mddev);
4333 /* Clear some bits that don't mean anything, but
4336 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4337 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4338 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4339 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4341 /* no recovery is running.
4342 * remove any failed drives, then
4343 * add spares if possible.
4344 * Spare are also removed and re-added, to allow
4345 * the personality to fail the re-add.
4347 ITERATE_RDEV(mddev,rdev,rtmp)
4348 if (rdev->raid_disk >= 0 &&
4349 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4350 atomic_read(&rdev->nr_pending)==0) {
4351 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4353 sprintf(nm,"rd%d", rdev->raid_disk);
4354 sysfs_remove_link(&mddev->kobj, nm);
4355 rdev->raid_disk = -1;
4359 if (mddev->degraded) {
4360 ITERATE_RDEV(mddev,rdev,rtmp)
4361 if (rdev->raid_disk < 0
4362 && !test_bit(Faulty, &rdev->flags)) {
4363 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4365 sprintf(nm, "rd%d", rdev->raid_disk);
4366 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4368 md_new_event(mddev);
4375 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4376 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4377 } else if (mddev->recovery_cp < MaxSector) {
4378 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4379 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4380 /* nothing to be done ... */
4383 if (mddev->pers->sync_request) {
4384 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4385 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4386 /* We are adding a device or devices to an array
4387 * which has the bitmap stored on all devices.
4388 * So make sure all bitmap pages get written
4390 bitmap_write_all(mddev->bitmap);
4392 mddev->sync_thread = md_register_thread(md_do_sync,
4395 if (!mddev->sync_thread) {
4396 printk(KERN_ERR "%s: could not start resync"
4399 /* leave the spares where they are, it shouldn't hurt */
4400 mddev->recovery = 0;
4402 md_wakeup_thread(mddev->sync_thread);
4403 md_new_event(mddev);
4406 mddev_unlock(mddev);
4410 static int md_notify_reboot(struct notifier_block *this,
4411 unsigned long code, void *x)
4413 struct list_head *tmp;
4416 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4418 printk(KERN_INFO "md: stopping all md devices.\n");
4420 ITERATE_MDDEV(mddev,tmp)
4421 if (mddev_trylock(mddev)==0)
4422 do_md_stop (mddev, 1);
4424 * certain more exotic SCSI devices are known to be
4425 * volatile wrt too early system reboots. While the
4426 * right place to handle this issue is the given
4427 * driver, we do want to have a safe RAID driver ...
4434 static struct notifier_block md_notifier = {
4435 .notifier_call = md_notify_reboot,
4437 .priority = INT_MAX, /* before any real devices */
4440 static void md_geninit(void)
4442 struct proc_dir_entry *p;
4444 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4446 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4448 p->proc_fops = &md_seq_fops;
4451 static int __init md_init(void)
4455 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4456 " MD_SB_DISKS=%d\n",
4457 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4458 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4459 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4462 if (register_blkdev(MAJOR_NR, "md"))
4464 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4465 unregister_blkdev(MAJOR_NR, "md");
4469 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4470 md_probe, NULL, NULL);
4471 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4472 md_probe, NULL, NULL);
4474 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4475 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4476 S_IFBLK|S_IRUSR|S_IWUSR,
4479 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4480 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4481 S_IFBLK|S_IRUSR|S_IWUSR,
4485 register_reboot_notifier(&md_notifier);
4486 raid_table_header = register_sysctl_table(raid_root_table, 1);
4496 * Searches all registered partitions for autorun RAID arrays
4499 static dev_t detected_devices[128];
4502 void md_autodetect_dev(dev_t dev)
4504 if (dev_cnt >= 0 && dev_cnt < 127)
4505 detected_devices[dev_cnt++] = dev;
4509 static void autostart_arrays(int part)
4514 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4516 for (i = 0; i < dev_cnt; i++) {
4517 dev_t dev = detected_devices[i];
4519 rdev = md_import_device(dev,0, 0);
4523 if (test_bit(Faulty, &rdev->flags)) {
4527 list_add(&rdev->same_set, &pending_raid_disks);
4531 autorun_devices(part);
4536 static __exit void md_exit(void)
4539 struct list_head *tmp;
4541 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4542 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4543 for (i=0; i < MAX_MD_DEVS; i++)
4544 devfs_remove("md/%d", i);
4545 for (i=0; i < MAX_MD_DEVS; i++)
4546 devfs_remove("md/d%d", i);
4550 unregister_blkdev(MAJOR_NR,"md");
4551 unregister_blkdev(mdp_major, "mdp");
4552 unregister_reboot_notifier(&md_notifier);
4553 unregister_sysctl_table(raid_table_header);
4554 remove_proc_entry("mdstat", NULL);
4555 ITERATE_MDDEV(mddev,tmp) {
4556 struct gendisk *disk = mddev->gendisk;
4559 export_array(mddev);
4562 mddev->gendisk = NULL;
4567 module_init(md_init)
4568 module_exit(md_exit)
4570 static int get_ro(char *buffer, struct kernel_param *kp)
4572 return sprintf(buffer, "%d", start_readonly);
4574 static int set_ro(const char *val, struct kernel_param *kp)
4577 int num = simple_strtoul(val, &e, 10);
4578 if (*val && (*e == '\0' || *e == '\n')) {
4579 start_readonly = num;
4585 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4586 module_param(start_dirty_degraded, int, 0644);
4589 EXPORT_SYMBOL(register_md_personality);
4590 EXPORT_SYMBOL(unregister_md_personality);
4591 EXPORT_SYMBOL(md_error);
4592 EXPORT_SYMBOL(md_done_sync);
4593 EXPORT_SYMBOL(md_write_start);
4594 EXPORT_SYMBOL(md_write_end);
4595 EXPORT_SYMBOL(md_register_thread);
4596 EXPORT_SYMBOL(md_unregister_thread);
4597 EXPORT_SYMBOL(md_wakeup_thread);
4598 EXPORT_SYMBOL(md_print_devices);
4599 EXPORT_SYMBOL(md_check_recovery);
4600 MODULE_LICENSE("GPL");
4602 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);