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 LIST_HEAD(pers_list);
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 struct mdk_personality *find_pers(int level)
308 struct mdk_personality *pers;
309 list_for_each_entry(pers, &pers_list, list)
310 if (pers->level == level)
315 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
317 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
318 return MD_NEW_SIZE_BLOCKS(size);
321 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
325 size = rdev->sb_offset;
328 size &= ~((sector_t)chunk_size/1024 - 1);
332 static int alloc_disk_sb(mdk_rdev_t * rdev)
337 rdev->sb_page = alloc_page(GFP_KERNEL);
338 if (!rdev->sb_page) {
339 printk(KERN_ALERT "md: out of memory.\n");
346 static void free_disk_sb(mdk_rdev_t * rdev)
349 put_page(rdev->sb_page);
351 rdev->sb_page = NULL;
358 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
360 mdk_rdev_t *rdev = bio->bi_private;
361 mddev_t *mddev = rdev->mddev;
365 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
366 md_error(mddev, rdev);
368 if (atomic_dec_and_test(&mddev->pending_writes))
369 wake_up(&mddev->sb_wait);
374 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
376 struct bio *bio2 = bio->bi_private;
377 mdk_rdev_t *rdev = bio2->bi_private;
378 mddev_t *mddev = rdev->mddev;
382 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
383 error == -EOPNOTSUPP) {
385 /* barriers don't appear to be supported :-( */
386 set_bit(BarriersNotsupp, &rdev->flags);
387 mddev->barriers_work = 0;
388 spin_lock_irqsave(&mddev->write_lock, flags);
389 bio2->bi_next = mddev->biolist;
390 mddev->biolist = bio2;
391 spin_unlock_irqrestore(&mddev->write_lock, flags);
392 wake_up(&mddev->sb_wait);
397 bio->bi_private = rdev;
398 return super_written(bio, bytes_done, error);
401 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
402 sector_t sector, int size, struct page *page)
404 /* write first size bytes of page to sector of rdev
405 * Increment mddev->pending_writes before returning
406 * and decrement it on completion, waking up sb_wait
407 * if zero is reached.
408 * If an error occurred, call md_error
410 * As we might need to resubmit the request if BIO_RW_BARRIER
411 * causes ENOTSUPP, we allocate a spare bio...
413 struct bio *bio = bio_alloc(GFP_NOIO, 1);
414 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
416 bio->bi_bdev = rdev->bdev;
417 bio->bi_sector = sector;
418 bio_add_page(bio, page, size, 0);
419 bio->bi_private = rdev;
420 bio->bi_end_io = super_written;
423 atomic_inc(&mddev->pending_writes);
424 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
426 rw |= (1<<BIO_RW_BARRIER);
427 rbio = bio_clone(bio, GFP_NOIO);
428 rbio->bi_private = bio;
429 rbio->bi_end_io = super_written_barrier;
430 submit_bio(rw, rbio);
435 void md_super_wait(mddev_t *mddev)
437 /* wait for all superblock writes that were scheduled to complete.
438 * if any had to be retried (due to BARRIER problems), retry them
442 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
443 if (atomic_read(&mddev->pending_writes)==0)
445 while (mddev->biolist) {
447 spin_lock_irq(&mddev->write_lock);
448 bio = mddev->biolist;
449 mddev->biolist = bio->bi_next ;
451 spin_unlock_irq(&mddev->write_lock);
452 submit_bio(bio->bi_rw, bio);
456 finish_wait(&mddev->sb_wait, &wq);
459 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
464 complete((struct completion*)bio->bi_private);
468 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
469 struct page *page, int rw)
471 struct bio *bio = bio_alloc(GFP_NOIO, 1);
472 struct completion event;
475 rw |= (1 << BIO_RW_SYNC);
478 bio->bi_sector = sector;
479 bio_add_page(bio, page, size, 0);
480 init_completion(&event);
481 bio->bi_private = &event;
482 bio->bi_end_io = bi_complete;
484 wait_for_completion(&event);
486 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
490 EXPORT_SYMBOL_GPL(sync_page_io);
492 static int read_disk_sb(mdk_rdev_t * rdev, int size)
494 char b[BDEVNAME_SIZE];
495 if (!rdev->sb_page) {
503 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
509 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
510 bdevname(rdev->bdev,b));
514 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
516 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
517 (sb1->set_uuid1 == sb2->set_uuid1) &&
518 (sb1->set_uuid2 == sb2->set_uuid2) &&
519 (sb1->set_uuid3 == sb2->set_uuid3))
527 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
530 mdp_super_t *tmp1, *tmp2;
532 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
533 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
535 if (!tmp1 || !tmp2) {
537 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
545 * nr_disks is not constant
550 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
561 static unsigned int calc_sb_csum(mdp_super_t * sb)
563 unsigned int disk_csum, csum;
565 disk_csum = sb->sb_csum;
567 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
568 sb->sb_csum = disk_csum;
574 * Handle superblock details.
575 * We want to be able to handle multiple superblock formats
576 * so we have a common interface to them all, and an array of
577 * different handlers.
578 * We rely on user-space to write the initial superblock, and support
579 * reading and updating of superblocks.
580 * Interface methods are:
581 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
582 * loads and validates a superblock on dev.
583 * if refdev != NULL, compare superblocks on both devices
585 * 0 - dev has a superblock that is compatible with refdev
586 * 1 - dev has a superblock that is compatible and newer than refdev
587 * so dev should be used as the refdev in future
588 * -EINVAL superblock incompatible or invalid
589 * -othererror e.g. -EIO
591 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
592 * Verify that dev is acceptable into mddev.
593 * The first time, mddev->raid_disks will be 0, and data from
594 * dev should be merged in. Subsequent calls check that dev
595 * is new enough. Return 0 or -EINVAL
597 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
598 * Update the superblock for rdev with data in mddev
599 * This does not write to disc.
605 struct module *owner;
606 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
607 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
608 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
612 * load_super for 0.90.0
614 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
616 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
622 * Calculate the position of the superblock,
623 * it's at the end of the disk.
625 * It also happens to be a multiple of 4Kb.
627 sb_offset = calc_dev_sboffset(rdev->bdev);
628 rdev->sb_offset = sb_offset;
630 ret = read_disk_sb(rdev, MD_SB_BYTES);
635 bdevname(rdev->bdev, b);
636 sb = (mdp_super_t*)page_address(rdev->sb_page);
638 if (sb->md_magic != MD_SB_MAGIC) {
639 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
644 if (sb->major_version != 0 ||
645 sb->minor_version != 90) {
646 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
647 sb->major_version, sb->minor_version,
652 if (sb->raid_disks <= 0)
655 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
656 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
661 rdev->preferred_minor = sb->md_minor;
662 rdev->data_offset = 0;
663 rdev->sb_size = MD_SB_BYTES;
665 if (sb->level == LEVEL_MULTIPATH)
668 rdev->desc_nr = sb->this_disk.number;
674 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
675 if (!uuid_equal(refsb, sb)) {
676 printk(KERN_WARNING "md: %s has different UUID to %s\n",
677 b, bdevname(refdev->bdev,b2));
680 if (!sb_equal(refsb, sb)) {
681 printk(KERN_WARNING "md: %s has same UUID"
682 " but different superblock to %s\n",
683 b, bdevname(refdev->bdev, b2));
687 ev2 = md_event(refsb);
693 rdev->size = calc_dev_size(rdev, sb->chunk_size);
700 * validate_super for 0.90.0
702 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
705 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
707 rdev->raid_disk = -1;
709 if (mddev->raid_disks == 0) {
710 mddev->major_version = 0;
711 mddev->minor_version = sb->minor_version;
712 mddev->patch_version = sb->patch_version;
713 mddev->persistent = ! sb->not_persistent;
714 mddev->chunk_size = sb->chunk_size;
715 mddev->ctime = sb->ctime;
716 mddev->utime = sb->utime;
717 mddev->level = sb->level;
718 mddev->layout = sb->layout;
719 mddev->raid_disks = sb->raid_disks;
720 mddev->size = sb->size;
721 mddev->events = md_event(sb);
722 mddev->bitmap_offset = 0;
723 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
725 if (sb->state & (1<<MD_SB_CLEAN))
726 mddev->recovery_cp = MaxSector;
728 if (sb->events_hi == sb->cp_events_hi &&
729 sb->events_lo == sb->cp_events_lo) {
730 mddev->recovery_cp = sb->recovery_cp;
732 mddev->recovery_cp = 0;
735 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
736 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
737 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
738 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
740 mddev->max_disks = MD_SB_DISKS;
742 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
743 mddev->bitmap_file == NULL) {
744 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
745 && mddev->level != 10) {
746 /* FIXME use a better test */
747 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
750 mddev->bitmap_offset = mddev->default_bitmap_offset;
753 } else if (mddev->pers == NULL) {
754 /* Insist on good event counter while assembling */
755 __u64 ev1 = md_event(sb);
757 if (ev1 < mddev->events)
759 } else if (mddev->bitmap) {
760 /* if adding to array with a bitmap, then we can accept an
761 * older device ... but not too old.
763 __u64 ev1 = md_event(sb);
764 if (ev1 < mddev->bitmap->events_cleared)
766 } else /* just a hot-add of a new device, leave raid_disk at -1 */
769 if (mddev->level != LEVEL_MULTIPATH) {
770 desc = sb->disks + rdev->desc_nr;
772 if (desc->state & (1<<MD_DISK_FAULTY))
773 set_bit(Faulty, &rdev->flags);
774 else if (desc->state & (1<<MD_DISK_SYNC) &&
775 desc->raid_disk < mddev->raid_disks) {
776 set_bit(In_sync, &rdev->flags);
777 rdev->raid_disk = desc->raid_disk;
779 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
780 set_bit(WriteMostly, &rdev->flags);
781 } else /* MULTIPATH are always insync */
782 set_bit(In_sync, &rdev->flags);
787 * sync_super for 0.90.0
789 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
792 struct list_head *tmp;
794 int next_spare = mddev->raid_disks;
797 /* make rdev->sb match mddev data..
800 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
801 * 3/ any empty disks < next_spare become removed
803 * disks[0] gets initialised to REMOVED because
804 * we cannot be sure from other fields if it has
805 * been initialised or not.
808 int active=0, working=0,failed=0,spare=0,nr_disks=0;
810 rdev->sb_size = MD_SB_BYTES;
812 sb = (mdp_super_t*)page_address(rdev->sb_page);
814 memset(sb, 0, sizeof(*sb));
816 sb->md_magic = MD_SB_MAGIC;
817 sb->major_version = mddev->major_version;
818 sb->minor_version = mddev->minor_version;
819 sb->patch_version = mddev->patch_version;
820 sb->gvalid_words = 0; /* ignored */
821 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
822 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
823 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
824 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
826 sb->ctime = mddev->ctime;
827 sb->level = mddev->level;
828 sb->size = mddev->size;
829 sb->raid_disks = mddev->raid_disks;
830 sb->md_minor = mddev->md_minor;
831 sb->not_persistent = !mddev->persistent;
832 sb->utime = mddev->utime;
834 sb->events_hi = (mddev->events>>32);
835 sb->events_lo = (u32)mddev->events;
839 sb->recovery_cp = mddev->recovery_cp;
840 sb->cp_events_hi = (mddev->events>>32);
841 sb->cp_events_lo = (u32)mddev->events;
842 if (mddev->recovery_cp == MaxSector)
843 sb->state = (1<< MD_SB_CLEAN);
847 sb->layout = mddev->layout;
848 sb->chunk_size = mddev->chunk_size;
850 if (mddev->bitmap && mddev->bitmap_file == NULL)
851 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
853 sb->disks[0].state = (1<<MD_DISK_REMOVED);
854 ITERATE_RDEV(mddev,rdev2,tmp) {
857 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
858 && !test_bit(Faulty, &rdev2->flags))
859 desc_nr = rdev2->raid_disk;
861 desc_nr = next_spare++;
862 rdev2->desc_nr = desc_nr;
863 d = &sb->disks[rdev2->desc_nr];
865 d->number = rdev2->desc_nr;
866 d->major = MAJOR(rdev2->bdev->bd_dev);
867 d->minor = MINOR(rdev2->bdev->bd_dev);
868 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
869 && !test_bit(Faulty, &rdev2->flags))
870 d->raid_disk = rdev2->raid_disk;
872 d->raid_disk = rdev2->desc_nr; /* compatibility */
873 if (test_bit(Faulty, &rdev2->flags)) {
874 d->state = (1<<MD_DISK_FAULTY);
876 } else if (test_bit(In_sync, &rdev2->flags)) {
877 d->state = (1<<MD_DISK_ACTIVE);
878 d->state |= (1<<MD_DISK_SYNC);
886 if (test_bit(WriteMostly, &rdev2->flags))
887 d->state |= (1<<MD_DISK_WRITEMOSTLY);
889 /* now set the "removed" and "faulty" bits on any missing devices */
890 for (i=0 ; i < mddev->raid_disks ; i++) {
891 mdp_disk_t *d = &sb->disks[i];
892 if (d->state == 0 && d->number == 0) {
895 d->state = (1<<MD_DISK_REMOVED);
896 d->state |= (1<<MD_DISK_FAULTY);
900 sb->nr_disks = nr_disks;
901 sb->active_disks = active;
902 sb->working_disks = working;
903 sb->failed_disks = failed;
904 sb->spare_disks = spare;
906 sb->this_disk = sb->disks[rdev->desc_nr];
907 sb->sb_csum = calc_sb_csum(sb);
911 * version 1 superblock
914 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
916 unsigned int disk_csum, csum;
917 unsigned long long newcsum;
918 int size = 256 + le32_to_cpu(sb->max_dev)*2;
919 unsigned int *isuper = (unsigned int*)sb;
922 disk_csum = sb->sb_csum;
925 for (i=0; size>=4; size -= 4 )
926 newcsum += le32_to_cpu(*isuper++);
929 newcsum += le16_to_cpu(*(unsigned short*) isuper);
931 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
932 sb->sb_csum = disk_csum;
933 return cpu_to_le32(csum);
936 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
938 struct mdp_superblock_1 *sb;
941 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
945 * Calculate the position of the superblock.
946 * It is always aligned to a 4K boundary and
947 * depeding on minor_version, it can be:
948 * 0: At least 8K, but less than 12K, from end of device
949 * 1: At start of device
950 * 2: 4K from start of device.
952 switch(minor_version) {
954 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
956 sb_offset &= ~(sector_t)(4*2-1);
957 /* convert from sectors to K */
969 rdev->sb_offset = sb_offset;
971 /* superblock is rarely larger than 1K, but it can be larger,
972 * and it is safe to read 4k, so we do that
974 ret = read_disk_sb(rdev, 4096);
978 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
980 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
981 sb->major_version != cpu_to_le32(1) ||
982 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
983 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
984 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
987 if (calc_sb_1_csum(sb) != sb->sb_csum) {
988 printk("md: invalid superblock checksum on %s\n",
989 bdevname(rdev->bdev,b));
992 if (le64_to_cpu(sb->data_size) < 10) {
993 printk("md: data_size too small on %s\n",
994 bdevname(rdev->bdev,b));
997 rdev->preferred_minor = 0xffff;
998 rdev->data_offset = le64_to_cpu(sb->data_offset);
1000 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1001 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1002 if (rdev->sb_size & bmask)
1003 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1009 struct mdp_superblock_1 *refsb =
1010 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1012 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1013 sb->level != refsb->level ||
1014 sb->layout != refsb->layout ||
1015 sb->chunksize != refsb->chunksize) {
1016 printk(KERN_WARNING "md: %s has strangely different"
1017 " superblock to %s\n",
1018 bdevname(rdev->bdev,b),
1019 bdevname(refdev->bdev,b2));
1022 ev1 = le64_to_cpu(sb->events);
1023 ev2 = le64_to_cpu(refsb->events);
1029 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1031 rdev->size = rdev->sb_offset;
1032 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1034 rdev->size = le64_to_cpu(sb->data_size)/2;
1035 if (le32_to_cpu(sb->chunksize))
1036 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1040 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1042 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1044 rdev->raid_disk = -1;
1046 if (mddev->raid_disks == 0) {
1047 mddev->major_version = 1;
1048 mddev->patch_version = 0;
1049 mddev->persistent = 1;
1050 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1051 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1052 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1053 mddev->level = le32_to_cpu(sb->level);
1054 mddev->layout = le32_to_cpu(sb->layout);
1055 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1056 mddev->size = le64_to_cpu(sb->size)/2;
1057 mddev->events = le64_to_cpu(sb->events);
1058 mddev->bitmap_offset = 0;
1059 mddev->default_bitmap_offset = 1024;
1061 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1062 memcpy(mddev->uuid, sb->set_uuid, 16);
1064 mddev->max_disks = (4096-256)/2;
1066 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1067 mddev->bitmap_file == NULL ) {
1068 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1069 && mddev->level != 10) {
1070 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1073 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1075 } else if (mddev->pers == NULL) {
1076 /* Insist of good event counter while assembling */
1077 __u64 ev1 = le64_to_cpu(sb->events);
1079 if (ev1 < mddev->events)
1081 } else if (mddev->bitmap) {
1082 /* If adding to array with a bitmap, then we can accept an
1083 * older device, but not too old.
1085 __u64 ev1 = le64_to_cpu(sb->events);
1086 if (ev1 < mddev->bitmap->events_cleared)
1088 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1091 if (mddev->level != LEVEL_MULTIPATH) {
1093 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1094 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1096 case 0xffff: /* spare */
1098 case 0xfffe: /* faulty */
1099 set_bit(Faulty, &rdev->flags);
1102 set_bit(In_sync, &rdev->flags);
1103 rdev->raid_disk = role;
1106 if (sb->devflags & WriteMostly1)
1107 set_bit(WriteMostly, &rdev->flags);
1108 } else /* MULTIPATH are always insync */
1109 set_bit(In_sync, &rdev->flags);
1114 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1116 struct mdp_superblock_1 *sb;
1117 struct list_head *tmp;
1120 /* make rdev->sb match mddev and rdev data. */
1122 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1124 sb->feature_map = 0;
1126 memset(sb->pad1, 0, sizeof(sb->pad1));
1127 memset(sb->pad2, 0, sizeof(sb->pad2));
1128 memset(sb->pad3, 0, sizeof(sb->pad3));
1130 sb->utime = cpu_to_le64((__u64)mddev->utime);
1131 sb->events = cpu_to_le64(mddev->events);
1133 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1135 sb->resync_offset = cpu_to_le64(0);
1137 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1138 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1139 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1143 ITERATE_RDEV(mddev,rdev2,tmp)
1144 if (rdev2->desc_nr+1 > max_dev)
1145 max_dev = rdev2->desc_nr+1;
1147 sb->max_dev = cpu_to_le32(max_dev);
1148 for (i=0; i<max_dev;i++)
1149 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1151 ITERATE_RDEV(mddev,rdev2,tmp) {
1153 if (test_bit(Faulty, &rdev2->flags))
1154 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1155 else if (test_bit(In_sync, &rdev2->flags))
1156 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1158 sb->dev_roles[i] = cpu_to_le16(0xffff);
1161 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1162 sb->sb_csum = calc_sb_1_csum(sb);
1166 static struct super_type super_types[] = {
1169 .owner = THIS_MODULE,
1170 .load_super = super_90_load,
1171 .validate_super = super_90_validate,
1172 .sync_super = super_90_sync,
1176 .owner = THIS_MODULE,
1177 .load_super = super_1_load,
1178 .validate_super = super_1_validate,
1179 .sync_super = super_1_sync,
1183 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1185 struct list_head *tmp;
1188 ITERATE_RDEV(mddev,rdev,tmp)
1189 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1195 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1197 struct list_head *tmp;
1200 ITERATE_RDEV(mddev1,rdev,tmp)
1201 if (match_dev_unit(mddev2, rdev))
1207 static LIST_HEAD(pending_raid_disks);
1209 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1211 mdk_rdev_t *same_pdev;
1212 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1219 same_pdev = match_dev_unit(mddev, rdev);
1222 "%s: WARNING: %s appears to be on the same physical"
1223 " disk as %s. True\n protection against single-disk"
1224 " failure might be compromised.\n",
1225 mdname(mddev), bdevname(rdev->bdev,b),
1226 bdevname(same_pdev->bdev,b2));
1228 /* Verify rdev->desc_nr is unique.
1229 * If it is -1, assign a free number, else
1230 * check number is not in use
1232 if (rdev->desc_nr < 0) {
1234 if (mddev->pers) choice = mddev->raid_disks;
1235 while (find_rdev_nr(mddev, choice))
1237 rdev->desc_nr = choice;
1239 if (find_rdev_nr(mddev, rdev->desc_nr))
1242 bdevname(rdev->bdev,b);
1243 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1246 list_add(&rdev->same_set, &mddev->disks);
1247 rdev->mddev = mddev;
1248 printk(KERN_INFO "md: bind<%s>\n", b);
1250 rdev->kobj.parent = &mddev->kobj;
1251 kobject_add(&rdev->kobj);
1253 if (rdev->bdev->bd_part)
1254 ko = &rdev->bdev->bd_part->kobj;
1256 ko = &rdev->bdev->bd_disk->kobj;
1257 sysfs_create_link(&rdev->kobj, ko, "block");
1261 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1263 char b[BDEVNAME_SIZE];
1268 list_del_init(&rdev->same_set);
1269 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1271 sysfs_remove_link(&rdev->kobj, "block");
1272 kobject_del(&rdev->kobj);
1276 * prevent the device from being mounted, repartitioned or
1277 * otherwise reused by a RAID array (or any other kernel
1278 * subsystem), by bd_claiming the device.
1280 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1283 struct block_device *bdev;
1284 char b[BDEVNAME_SIZE];
1286 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1288 printk(KERN_ERR "md: could not open %s.\n",
1289 __bdevname(dev, b));
1290 return PTR_ERR(bdev);
1292 err = bd_claim(bdev, rdev);
1294 printk(KERN_ERR "md: could not bd_claim %s.\n",
1303 static void unlock_rdev(mdk_rdev_t *rdev)
1305 struct block_device *bdev = rdev->bdev;
1313 void md_autodetect_dev(dev_t dev);
1315 static void export_rdev(mdk_rdev_t * rdev)
1317 char b[BDEVNAME_SIZE];
1318 printk(KERN_INFO "md: export_rdev(%s)\n",
1319 bdevname(rdev->bdev,b));
1323 list_del_init(&rdev->same_set);
1325 md_autodetect_dev(rdev->bdev->bd_dev);
1328 kobject_put(&rdev->kobj);
1331 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1333 unbind_rdev_from_array(rdev);
1337 static void export_array(mddev_t *mddev)
1339 struct list_head *tmp;
1342 ITERATE_RDEV(mddev,rdev,tmp) {
1347 kick_rdev_from_array(rdev);
1349 if (!list_empty(&mddev->disks))
1351 mddev->raid_disks = 0;
1352 mddev->major_version = 0;
1355 static void print_desc(mdp_disk_t *desc)
1357 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1358 desc->major,desc->minor,desc->raid_disk,desc->state);
1361 static void print_sb(mdp_super_t *sb)
1366 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1367 sb->major_version, sb->minor_version, sb->patch_version,
1368 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1370 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1371 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1372 sb->md_minor, sb->layout, sb->chunk_size);
1373 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1374 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1375 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1376 sb->failed_disks, sb->spare_disks,
1377 sb->sb_csum, (unsigned long)sb->events_lo);
1380 for (i = 0; i < MD_SB_DISKS; i++) {
1383 desc = sb->disks + i;
1384 if (desc->number || desc->major || desc->minor ||
1385 desc->raid_disk || (desc->state && (desc->state != 4))) {
1386 printk(" D %2d: ", i);
1390 printk(KERN_INFO "md: THIS: ");
1391 print_desc(&sb->this_disk);
1395 static void print_rdev(mdk_rdev_t *rdev)
1397 char b[BDEVNAME_SIZE];
1398 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1399 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1400 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1402 if (rdev->sb_loaded) {
1403 printk(KERN_INFO "md: rdev superblock:\n");
1404 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1406 printk(KERN_INFO "md: no rdev superblock!\n");
1409 void md_print_devices(void)
1411 struct list_head *tmp, *tmp2;
1414 char b[BDEVNAME_SIZE];
1417 printk("md: **********************************\n");
1418 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1419 printk("md: **********************************\n");
1420 ITERATE_MDDEV(mddev,tmp) {
1423 bitmap_print_sb(mddev->bitmap);
1425 printk("%s: ", mdname(mddev));
1426 ITERATE_RDEV(mddev,rdev,tmp2)
1427 printk("<%s>", bdevname(rdev->bdev,b));
1430 ITERATE_RDEV(mddev,rdev,tmp2)
1433 printk("md: **********************************\n");
1438 static void sync_sbs(mddev_t * mddev)
1441 struct list_head *tmp;
1443 ITERATE_RDEV(mddev,rdev,tmp) {
1444 super_types[mddev->major_version].
1445 sync_super(mddev, rdev);
1446 rdev->sb_loaded = 1;
1450 static void md_update_sb(mddev_t * mddev)
1453 struct list_head *tmp;
1458 spin_lock_irq(&mddev->write_lock);
1459 sync_req = mddev->in_sync;
1460 mddev->utime = get_seconds();
1463 if (!mddev->events) {
1465 * oops, this 64-bit counter should never wrap.
1466 * Either we are in around ~1 trillion A.C., assuming
1467 * 1 reboot per second, or we have a bug:
1472 mddev->sb_dirty = 2;
1476 * do not write anything to disk if using
1477 * nonpersistent superblocks
1479 if (!mddev->persistent) {
1480 mddev->sb_dirty = 0;
1481 spin_unlock_irq(&mddev->write_lock);
1482 wake_up(&mddev->sb_wait);
1485 spin_unlock_irq(&mddev->write_lock);
1488 "md: updating %s RAID superblock on device (in sync %d)\n",
1489 mdname(mddev),mddev->in_sync);
1491 err = bitmap_update_sb(mddev->bitmap);
1492 ITERATE_RDEV(mddev,rdev,tmp) {
1493 char b[BDEVNAME_SIZE];
1494 dprintk(KERN_INFO "md: ");
1495 if (test_bit(Faulty, &rdev->flags))
1496 dprintk("(skipping faulty ");
1498 dprintk("%s ", bdevname(rdev->bdev,b));
1499 if (!test_bit(Faulty, &rdev->flags)) {
1500 md_super_write(mddev,rdev,
1501 rdev->sb_offset<<1, rdev->sb_size,
1503 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1504 bdevname(rdev->bdev,b),
1505 (unsigned long long)rdev->sb_offset);
1509 if (mddev->level == LEVEL_MULTIPATH)
1510 /* only need to write one superblock... */
1513 md_super_wait(mddev);
1514 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1516 spin_lock_irq(&mddev->write_lock);
1517 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1518 /* have to write it out again */
1519 spin_unlock_irq(&mddev->write_lock);
1522 mddev->sb_dirty = 0;
1523 spin_unlock_irq(&mddev->write_lock);
1524 wake_up(&mddev->sb_wait);
1528 struct rdev_sysfs_entry {
1529 struct attribute attr;
1530 ssize_t (*show)(mdk_rdev_t *, char *);
1531 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1535 state_show(mdk_rdev_t *rdev, char *page)
1540 if (test_bit(Faulty, &rdev->flags)) {
1541 len+= sprintf(page+len, "%sfaulty",sep);
1544 if (test_bit(In_sync, &rdev->flags)) {
1545 len += sprintf(page+len, "%sin_sync",sep);
1548 if (!test_bit(Faulty, &rdev->flags) &&
1549 !test_bit(In_sync, &rdev->flags)) {
1550 len += sprintf(page+len, "%sspare", sep);
1553 return len+sprintf(page+len, "\n");
1556 static struct rdev_sysfs_entry
1557 rdev_state = __ATTR_RO(state);
1560 super_show(mdk_rdev_t *rdev, char *page)
1562 if (rdev->sb_loaded && rdev->sb_size) {
1563 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1564 return rdev->sb_size;
1568 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1570 static struct attribute *rdev_default_attrs[] = {
1576 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1578 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1579 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1583 return entry->show(rdev, page);
1587 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1588 const char *page, size_t length)
1590 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1591 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1595 return entry->store(rdev, page, length);
1598 static void rdev_free(struct kobject *ko)
1600 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1603 static struct sysfs_ops rdev_sysfs_ops = {
1604 .show = rdev_attr_show,
1605 .store = rdev_attr_store,
1607 static struct kobj_type rdev_ktype = {
1608 .release = rdev_free,
1609 .sysfs_ops = &rdev_sysfs_ops,
1610 .default_attrs = rdev_default_attrs,
1614 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1616 * mark the device faulty if:
1618 * - the device is nonexistent (zero size)
1619 * - the device has no valid superblock
1621 * a faulty rdev _never_ has rdev->sb set.
1623 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1625 char b[BDEVNAME_SIZE];
1630 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1632 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1633 return ERR_PTR(-ENOMEM);
1636 if ((err = alloc_disk_sb(rdev)))
1639 err = lock_rdev(rdev, newdev);
1643 rdev->kobj.parent = NULL;
1644 rdev->kobj.ktype = &rdev_ktype;
1645 kobject_init(&rdev->kobj);
1649 rdev->data_offset = 0;
1650 atomic_set(&rdev->nr_pending, 0);
1651 atomic_set(&rdev->read_errors, 0);
1653 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1656 "md: %s has zero or unknown size, marking faulty!\n",
1657 bdevname(rdev->bdev,b));
1662 if (super_format >= 0) {
1663 err = super_types[super_format].
1664 load_super(rdev, NULL, super_minor);
1665 if (err == -EINVAL) {
1667 "md: %s has invalid sb, not importing!\n",
1668 bdevname(rdev->bdev,b));
1673 "md: could not read %s's sb, not importing!\n",
1674 bdevname(rdev->bdev,b));
1678 INIT_LIST_HEAD(&rdev->same_set);
1683 if (rdev->sb_page) {
1689 return ERR_PTR(err);
1693 * Check a full RAID array for plausibility
1697 static void analyze_sbs(mddev_t * mddev)
1700 struct list_head *tmp;
1701 mdk_rdev_t *rdev, *freshest;
1702 char b[BDEVNAME_SIZE];
1705 ITERATE_RDEV(mddev,rdev,tmp)
1706 switch (super_types[mddev->major_version].
1707 load_super(rdev, freshest, mddev->minor_version)) {
1715 "md: fatal superblock inconsistency in %s"
1716 " -- removing from array\n",
1717 bdevname(rdev->bdev,b));
1718 kick_rdev_from_array(rdev);
1722 super_types[mddev->major_version].
1723 validate_super(mddev, freshest);
1726 ITERATE_RDEV(mddev,rdev,tmp) {
1727 if (rdev != freshest)
1728 if (super_types[mddev->major_version].
1729 validate_super(mddev, rdev)) {
1730 printk(KERN_WARNING "md: kicking non-fresh %s"
1732 bdevname(rdev->bdev,b));
1733 kick_rdev_from_array(rdev);
1736 if (mddev->level == LEVEL_MULTIPATH) {
1737 rdev->desc_nr = i++;
1738 rdev->raid_disk = rdev->desc_nr;
1739 set_bit(In_sync, &rdev->flags);
1745 if (mddev->recovery_cp != MaxSector &&
1747 printk(KERN_ERR "md: %s: raid array is not clean"
1748 " -- starting background reconstruction\n",
1754 level_show(mddev_t *mddev, char *page)
1756 struct mdk_personality *p = mddev->pers;
1757 if (p == NULL && mddev->raid_disks == 0)
1759 if (mddev->level >= 0)
1760 return sprintf(page, "raid%d\n", mddev->level);
1762 return sprintf(page, "%s\n", p->name);
1765 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1768 raid_disks_show(mddev_t *mddev, char *page)
1770 if (mddev->raid_disks == 0)
1772 return sprintf(page, "%d\n", mddev->raid_disks);
1775 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1778 action_show(mddev_t *mddev, char *page)
1780 char *type = "idle";
1781 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1782 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1783 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1784 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1786 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1793 return sprintf(page, "%s\n", type);
1797 action_store(mddev_t *mddev, const char *page, size_t len)
1799 if (!mddev->pers || !mddev->pers->sync_request)
1802 if (strcmp(page, "idle")==0 || strcmp(page, "idle\n")==0) {
1803 if (mddev->sync_thread) {
1804 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1805 md_unregister_thread(mddev->sync_thread);
1806 mddev->sync_thread = NULL;
1807 mddev->recovery = 0;
1812 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1813 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1815 if (strcmp(page, "resync")==0 || strcmp(page, "resync\n")==0 ||
1816 strcmp(page, "recover")==0 || strcmp(page, "recover\n")==0)
1817 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1819 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1820 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1821 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1823 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1824 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1825 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1827 md_wakeup_thread(mddev->thread);
1832 mismatch_cnt_show(mddev_t *mddev, char *page)
1834 return sprintf(page, "%llu\n",
1835 (unsigned long long) mddev->resync_mismatches);
1838 static struct md_sysfs_entry
1839 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1842 static struct md_sysfs_entry
1843 md_mismatches = __ATTR_RO(mismatch_cnt);
1845 static struct attribute *md_default_attrs[] = {
1847 &md_raid_disks.attr,
1851 static struct attribute *md_redundancy_attrs[] = {
1853 &md_mismatches.attr,
1856 static struct attribute_group md_redundancy_group = {
1858 .attrs = md_redundancy_attrs,
1863 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1865 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1866 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1872 rv = entry->show(mddev, page);
1873 mddev_unlock(mddev);
1878 md_attr_store(struct kobject *kobj, struct attribute *attr,
1879 const char *page, size_t length)
1881 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1882 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1888 rv = entry->store(mddev, page, length);
1889 mddev_unlock(mddev);
1893 static void md_free(struct kobject *ko)
1895 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1899 static struct sysfs_ops md_sysfs_ops = {
1900 .show = md_attr_show,
1901 .store = md_attr_store,
1903 static struct kobj_type md_ktype = {
1905 .sysfs_ops = &md_sysfs_ops,
1906 .default_attrs = md_default_attrs,
1911 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1913 static DECLARE_MUTEX(disks_sem);
1914 mddev_t *mddev = mddev_find(dev);
1915 struct gendisk *disk;
1916 int partitioned = (MAJOR(dev) != MD_MAJOR);
1917 int shift = partitioned ? MdpMinorShift : 0;
1918 int unit = MINOR(dev) >> shift;
1924 if (mddev->gendisk) {
1929 disk = alloc_disk(1 << shift);
1935 disk->major = MAJOR(dev);
1936 disk->first_minor = unit << shift;
1938 sprintf(disk->disk_name, "md_d%d", unit);
1939 sprintf(disk->devfs_name, "md/d%d", unit);
1941 sprintf(disk->disk_name, "md%d", unit);
1942 sprintf(disk->devfs_name, "md/%d", unit);
1944 disk->fops = &md_fops;
1945 disk->private_data = mddev;
1946 disk->queue = mddev->queue;
1948 mddev->gendisk = disk;
1950 mddev->kobj.parent = &disk->kobj;
1951 mddev->kobj.k_name = NULL;
1952 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1953 mddev->kobj.ktype = &md_ktype;
1954 kobject_register(&mddev->kobj);
1958 void md_wakeup_thread(mdk_thread_t *thread);
1960 static void md_safemode_timeout(unsigned long data)
1962 mddev_t *mddev = (mddev_t *) data;
1964 mddev->safemode = 1;
1965 md_wakeup_thread(mddev->thread);
1968 static int start_dirty_degraded;
1970 static int do_md_run(mddev_t * mddev)
1974 struct list_head *tmp;
1976 struct gendisk *disk;
1977 struct mdk_personality *pers;
1978 char b[BDEVNAME_SIZE];
1980 if (list_empty(&mddev->disks))
1981 /* cannot run an array with no devices.. */
1988 * Analyze all RAID superblock(s)
1990 if (!mddev->raid_disks)
1993 chunk_size = mddev->chunk_size;
1996 if (chunk_size > MAX_CHUNK_SIZE) {
1997 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1998 chunk_size, MAX_CHUNK_SIZE);
2002 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2004 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2005 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2008 if (chunk_size < PAGE_SIZE) {
2009 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2010 chunk_size, PAGE_SIZE);
2014 /* devices must have minimum size of one chunk */
2015 ITERATE_RDEV(mddev,rdev,tmp) {
2016 if (test_bit(Faulty, &rdev->flags))
2018 if (rdev->size < chunk_size / 1024) {
2020 "md: Dev %s smaller than chunk_size:"
2022 bdevname(rdev->bdev,b),
2023 (unsigned long long)rdev->size,
2031 request_module("md-level-%d", mddev->level);
2035 * Drop all container device buffers, from now on
2036 * the only valid external interface is through the md
2038 * Also find largest hardsector size
2040 ITERATE_RDEV(mddev,rdev,tmp) {
2041 if (test_bit(Faulty, &rdev->flags))
2043 sync_blockdev(rdev->bdev);
2044 invalidate_bdev(rdev->bdev, 0);
2047 md_probe(mddev->unit, NULL, NULL);
2048 disk = mddev->gendisk;
2052 spin_lock(&pers_lock);
2053 pers = find_pers(mddev->level);
2054 if (!pers || !try_module_get(pers->owner)) {
2055 spin_unlock(&pers_lock);
2056 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2061 spin_unlock(&pers_lock);
2063 mddev->recovery = 0;
2064 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2065 mddev->barriers_work = 1;
2066 mddev->ok_start_degraded = start_dirty_degraded;
2069 mddev->ro = 2; /* read-only, but switch on first write */
2071 err = mddev->pers->run(mddev);
2072 if (!err && mddev->pers->sync_request) {
2073 err = bitmap_create(mddev);
2075 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2076 mdname(mddev), err);
2077 mddev->pers->stop(mddev);
2081 printk(KERN_ERR "md: pers->run() failed ...\n");
2082 module_put(mddev->pers->owner);
2084 bitmap_destroy(mddev);
2087 if (mddev->pers->sync_request)
2088 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2089 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2092 atomic_set(&mddev->writes_pending,0);
2093 mddev->safemode = 0;
2094 mddev->safemode_timer.function = md_safemode_timeout;
2095 mddev->safemode_timer.data = (unsigned long) mddev;
2096 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2099 ITERATE_RDEV(mddev,rdev,tmp)
2100 if (rdev->raid_disk >= 0) {
2102 sprintf(nm, "rd%d", rdev->raid_disk);
2103 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2106 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2107 md_wakeup_thread(mddev->thread);
2109 if (mddev->sb_dirty)
2110 md_update_sb(mddev);
2112 set_capacity(disk, mddev->array_size<<1);
2114 /* If we call blk_queue_make_request here, it will
2115 * re-initialise max_sectors etc which may have been
2116 * refined inside -> run. So just set the bits we need to set.
2117 * Most initialisation happended when we called
2118 * blk_queue_make_request(..., md_fail_request)
2121 mddev->queue->queuedata = mddev;
2122 mddev->queue->make_request_fn = mddev->pers->make_request;
2125 md_new_event(mddev);
2129 static int restart_array(mddev_t *mddev)
2131 struct gendisk *disk = mddev->gendisk;
2135 * Complain if it has no devices
2138 if (list_empty(&mddev->disks))
2146 mddev->safemode = 0;
2148 set_disk_ro(disk, 0);
2150 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2153 * Kick recovery or resync if necessary
2155 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2156 md_wakeup_thread(mddev->thread);
2159 printk(KERN_ERR "md: %s has no personality assigned.\n",
2168 static int do_md_stop(mddev_t * mddev, int ro)
2171 struct gendisk *disk = mddev->gendisk;
2174 if (atomic_read(&mddev->active)>2) {
2175 printk("md: %s still in use.\n",mdname(mddev));
2179 if (mddev->sync_thread) {
2180 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2181 md_unregister_thread(mddev->sync_thread);
2182 mddev->sync_thread = NULL;
2185 del_timer_sync(&mddev->safemode_timer);
2187 invalidate_partition(disk, 0);
2195 bitmap_flush(mddev);
2196 md_super_wait(mddev);
2198 set_disk_ro(disk, 0);
2199 blk_queue_make_request(mddev->queue, md_fail_request);
2200 mddev->pers->stop(mddev);
2201 if (mddev->pers->sync_request)
2202 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2204 module_put(mddev->pers->owner);
2209 if (!mddev->in_sync) {
2210 /* mark array as shutdown cleanly */
2212 md_update_sb(mddev);
2215 set_disk_ro(disk, 1);
2218 bitmap_destroy(mddev);
2219 if (mddev->bitmap_file) {
2220 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2221 fput(mddev->bitmap_file);
2222 mddev->bitmap_file = NULL;
2224 mddev->bitmap_offset = 0;
2227 * Free resources if final stop
2231 struct list_head *tmp;
2232 struct gendisk *disk;
2233 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2235 ITERATE_RDEV(mddev,rdev,tmp)
2236 if (rdev->raid_disk >= 0) {
2238 sprintf(nm, "rd%d", rdev->raid_disk);
2239 sysfs_remove_link(&mddev->kobj, nm);
2242 export_array(mddev);
2244 mddev->array_size = 0;
2245 disk = mddev->gendisk;
2247 set_capacity(disk, 0);
2250 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2253 md_new_event(mddev);
2258 static void autorun_array(mddev_t *mddev)
2261 struct list_head *tmp;
2264 if (list_empty(&mddev->disks))
2267 printk(KERN_INFO "md: running: ");
2269 ITERATE_RDEV(mddev,rdev,tmp) {
2270 char b[BDEVNAME_SIZE];
2271 printk("<%s>", bdevname(rdev->bdev,b));
2275 err = do_md_run (mddev);
2277 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2278 do_md_stop (mddev, 0);
2283 * lets try to run arrays based on all disks that have arrived
2284 * until now. (those are in pending_raid_disks)
2286 * the method: pick the first pending disk, collect all disks with
2287 * the same UUID, remove all from the pending list and put them into
2288 * the 'same_array' list. Then order this list based on superblock
2289 * update time (freshest comes first), kick out 'old' disks and
2290 * compare superblocks. If everything's fine then run it.
2292 * If "unit" is allocated, then bump its reference count
2294 static void autorun_devices(int part)
2296 struct list_head candidates;
2297 struct list_head *tmp;
2298 mdk_rdev_t *rdev0, *rdev;
2300 char b[BDEVNAME_SIZE];
2302 printk(KERN_INFO "md: autorun ...\n");
2303 while (!list_empty(&pending_raid_disks)) {
2305 rdev0 = list_entry(pending_raid_disks.next,
2306 mdk_rdev_t, same_set);
2308 printk(KERN_INFO "md: considering %s ...\n",
2309 bdevname(rdev0->bdev,b));
2310 INIT_LIST_HEAD(&candidates);
2311 ITERATE_RDEV_PENDING(rdev,tmp)
2312 if (super_90_load(rdev, rdev0, 0) >= 0) {
2313 printk(KERN_INFO "md: adding %s ...\n",
2314 bdevname(rdev->bdev,b));
2315 list_move(&rdev->same_set, &candidates);
2318 * now we have a set of devices, with all of them having
2319 * mostly sane superblocks. It's time to allocate the
2322 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2323 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2324 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2328 dev = MKDEV(mdp_major,
2329 rdev0->preferred_minor << MdpMinorShift);
2331 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2333 md_probe(dev, NULL, NULL);
2334 mddev = mddev_find(dev);
2337 "md: cannot allocate memory for md drive.\n");
2340 if (mddev_lock(mddev))
2341 printk(KERN_WARNING "md: %s locked, cannot run\n",
2343 else if (mddev->raid_disks || mddev->major_version
2344 || !list_empty(&mddev->disks)) {
2346 "md: %s already running, cannot run %s\n",
2347 mdname(mddev), bdevname(rdev0->bdev,b));
2348 mddev_unlock(mddev);
2350 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2351 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2352 list_del_init(&rdev->same_set);
2353 if (bind_rdev_to_array(rdev, mddev))
2356 autorun_array(mddev);
2357 mddev_unlock(mddev);
2359 /* on success, candidates will be empty, on error
2362 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2366 printk(KERN_INFO "md: ... autorun DONE.\n");
2370 * import RAID devices based on one partition
2371 * if possible, the array gets run as well.
2374 static int autostart_array(dev_t startdev)
2376 char b[BDEVNAME_SIZE];
2377 int err = -EINVAL, i;
2378 mdp_super_t *sb = NULL;
2379 mdk_rdev_t *start_rdev = NULL, *rdev;
2381 start_rdev = md_import_device(startdev, 0, 0);
2382 if (IS_ERR(start_rdev))
2386 /* NOTE: this can only work for 0.90.0 superblocks */
2387 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2388 if (sb->major_version != 0 ||
2389 sb->minor_version != 90 ) {
2390 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2391 export_rdev(start_rdev);
2395 if (test_bit(Faulty, &start_rdev->flags)) {
2397 "md: can not autostart based on faulty %s!\n",
2398 bdevname(start_rdev->bdev,b));
2399 export_rdev(start_rdev);
2402 list_add(&start_rdev->same_set, &pending_raid_disks);
2404 for (i = 0; i < MD_SB_DISKS; i++) {
2405 mdp_disk_t *desc = sb->disks + i;
2406 dev_t dev = MKDEV(desc->major, desc->minor);
2410 if (dev == startdev)
2412 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2414 rdev = md_import_device(dev, 0, 0);
2418 list_add(&rdev->same_set, &pending_raid_disks);
2422 * possibly return codes
2430 static int get_version(void __user * arg)
2434 ver.major = MD_MAJOR_VERSION;
2435 ver.minor = MD_MINOR_VERSION;
2436 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2438 if (copy_to_user(arg, &ver, sizeof(ver)))
2444 static int get_array_info(mddev_t * mddev, void __user * arg)
2446 mdu_array_info_t info;
2447 int nr,working,active,failed,spare;
2449 struct list_head *tmp;
2451 nr=working=active=failed=spare=0;
2452 ITERATE_RDEV(mddev,rdev,tmp) {
2454 if (test_bit(Faulty, &rdev->flags))
2458 if (test_bit(In_sync, &rdev->flags))
2465 info.major_version = mddev->major_version;
2466 info.minor_version = mddev->minor_version;
2467 info.patch_version = MD_PATCHLEVEL_VERSION;
2468 info.ctime = mddev->ctime;
2469 info.level = mddev->level;
2470 info.size = mddev->size;
2472 info.raid_disks = mddev->raid_disks;
2473 info.md_minor = mddev->md_minor;
2474 info.not_persistent= !mddev->persistent;
2476 info.utime = mddev->utime;
2479 info.state = (1<<MD_SB_CLEAN);
2480 if (mddev->bitmap && mddev->bitmap_offset)
2481 info.state = (1<<MD_SB_BITMAP_PRESENT);
2482 info.active_disks = active;
2483 info.working_disks = working;
2484 info.failed_disks = failed;
2485 info.spare_disks = spare;
2487 info.layout = mddev->layout;
2488 info.chunk_size = mddev->chunk_size;
2490 if (copy_to_user(arg, &info, sizeof(info)))
2496 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2498 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2499 char *ptr, *buf = NULL;
2502 file = kmalloc(sizeof(*file), GFP_KERNEL);
2506 /* bitmap disabled, zero the first byte and copy out */
2507 if (!mddev->bitmap || !mddev->bitmap->file) {
2508 file->pathname[0] = '\0';
2512 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2516 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2520 strcpy(file->pathname, ptr);
2524 if (copy_to_user(arg, file, sizeof(*file)))
2532 static int get_disk_info(mddev_t * mddev, void __user * arg)
2534 mdu_disk_info_t info;
2538 if (copy_from_user(&info, arg, sizeof(info)))
2543 rdev = find_rdev_nr(mddev, nr);
2545 info.major = MAJOR(rdev->bdev->bd_dev);
2546 info.minor = MINOR(rdev->bdev->bd_dev);
2547 info.raid_disk = rdev->raid_disk;
2549 if (test_bit(Faulty, &rdev->flags))
2550 info.state |= (1<<MD_DISK_FAULTY);
2551 else if (test_bit(In_sync, &rdev->flags)) {
2552 info.state |= (1<<MD_DISK_ACTIVE);
2553 info.state |= (1<<MD_DISK_SYNC);
2555 if (test_bit(WriteMostly, &rdev->flags))
2556 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2558 info.major = info.minor = 0;
2559 info.raid_disk = -1;
2560 info.state = (1<<MD_DISK_REMOVED);
2563 if (copy_to_user(arg, &info, sizeof(info)))
2569 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2571 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2573 dev_t dev = MKDEV(info->major,info->minor);
2575 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2578 if (!mddev->raid_disks) {
2580 /* expecting a device which has a superblock */
2581 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2584 "md: md_import_device returned %ld\n",
2586 return PTR_ERR(rdev);
2588 if (!list_empty(&mddev->disks)) {
2589 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2590 mdk_rdev_t, same_set);
2591 int err = super_types[mddev->major_version]
2592 .load_super(rdev, rdev0, mddev->minor_version);
2595 "md: %s has different UUID to %s\n",
2596 bdevname(rdev->bdev,b),
2597 bdevname(rdev0->bdev,b2));
2602 err = bind_rdev_to_array(rdev, mddev);
2609 * add_new_disk can be used once the array is assembled
2610 * to add "hot spares". They must already have a superblock
2615 if (!mddev->pers->hot_add_disk) {
2617 "%s: personality does not support diskops!\n",
2621 if (mddev->persistent)
2622 rdev = md_import_device(dev, mddev->major_version,
2623 mddev->minor_version);
2625 rdev = md_import_device(dev, -1, -1);
2628 "md: md_import_device returned %ld\n",
2630 return PTR_ERR(rdev);
2632 /* set save_raid_disk if appropriate */
2633 if (!mddev->persistent) {
2634 if (info->state & (1<<MD_DISK_SYNC) &&
2635 info->raid_disk < mddev->raid_disks)
2636 rdev->raid_disk = info->raid_disk;
2638 rdev->raid_disk = -1;
2640 super_types[mddev->major_version].
2641 validate_super(mddev, rdev);
2642 rdev->saved_raid_disk = rdev->raid_disk;
2644 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2645 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2646 set_bit(WriteMostly, &rdev->flags);
2648 rdev->raid_disk = -1;
2649 err = bind_rdev_to_array(rdev, mddev);
2653 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2654 md_wakeup_thread(mddev->thread);
2658 /* otherwise, add_new_disk is only allowed
2659 * for major_version==0 superblocks
2661 if (mddev->major_version != 0) {
2662 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2667 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2669 rdev = md_import_device (dev, -1, 0);
2672 "md: error, md_import_device() returned %ld\n",
2674 return PTR_ERR(rdev);
2676 rdev->desc_nr = info->number;
2677 if (info->raid_disk < mddev->raid_disks)
2678 rdev->raid_disk = info->raid_disk;
2680 rdev->raid_disk = -1;
2684 if (rdev->raid_disk < mddev->raid_disks)
2685 if (info->state & (1<<MD_DISK_SYNC))
2686 set_bit(In_sync, &rdev->flags);
2688 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2689 set_bit(WriteMostly, &rdev->flags);
2691 err = bind_rdev_to_array(rdev, mddev);
2697 if (!mddev->persistent) {
2698 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2699 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2701 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2702 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2704 if (!mddev->size || (mddev->size > rdev->size))
2705 mddev->size = rdev->size;
2711 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2713 char b[BDEVNAME_SIZE];
2719 rdev = find_rdev(mddev, dev);
2723 if (rdev->raid_disk >= 0)
2726 kick_rdev_from_array(rdev);
2727 md_update_sb(mddev);
2728 md_new_event(mddev);
2732 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2733 bdevname(rdev->bdev,b), mdname(mddev));
2737 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2739 char b[BDEVNAME_SIZE];
2747 if (mddev->major_version != 0) {
2748 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2749 " version-0 superblocks.\n",
2753 if (!mddev->pers->hot_add_disk) {
2755 "%s: personality does not support diskops!\n",
2760 rdev = md_import_device (dev, -1, 0);
2763 "md: error, md_import_device() returned %ld\n",
2768 if (mddev->persistent)
2769 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2772 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2774 size = calc_dev_size(rdev, mddev->chunk_size);
2777 if (size < mddev->size) {
2779 "%s: disk size %llu blocks < array size %llu\n",
2780 mdname(mddev), (unsigned long long)size,
2781 (unsigned long long)mddev->size);
2786 if (test_bit(Faulty, &rdev->flags)) {
2788 "md: can not hot-add faulty %s disk to %s!\n",
2789 bdevname(rdev->bdev,b), mdname(mddev));
2793 clear_bit(In_sync, &rdev->flags);
2795 bind_rdev_to_array(rdev, mddev);
2798 * The rest should better be atomic, we can have disk failures
2799 * noticed in interrupt contexts ...
2802 if (rdev->desc_nr == mddev->max_disks) {
2803 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2806 goto abort_unbind_export;
2809 rdev->raid_disk = -1;
2811 md_update_sb(mddev);
2814 * Kick recovery, maybe this spare has to be added to the
2815 * array immediately.
2817 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2818 md_wakeup_thread(mddev->thread);
2819 md_new_event(mddev);
2822 abort_unbind_export:
2823 unbind_rdev_from_array(rdev);
2830 /* similar to deny_write_access, but accounts for our holding a reference
2831 * to the file ourselves */
2832 static int deny_bitmap_write_access(struct file * file)
2834 struct inode *inode = file->f_mapping->host;
2836 spin_lock(&inode->i_lock);
2837 if (atomic_read(&inode->i_writecount) > 1) {
2838 spin_unlock(&inode->i_lock);
2841 atomic_set(&inode->i_writecount, -1);
2842 spin_unlock(&inode->i_lock);
2847 static int set_bitmap_file(mddev_t *mddev, int fd)
2852 if (!mddev->pers->quiesce)
2854 if (mddev->recovery || mddev->sync_thread)
2856 /* we should be able to change the bitmap.. */
2862 return -EEXIST; /* cannot add when bitmap is present */
2863 mddev->bitmap_file = fget(fd);
2865 if (mddev->bitmap_file == NULL) {
2866 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2871 err = deny_bitmap_write_access(mddev->bitmap_file);
2873 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2875 fput(mddev->bitmap_file);
2876 mddev->bitmap_file = NULL;
2879 mddev->bitmap_offset = 0; /* file overrides offset */
2880 } else if (mddev->bitmap == NULL)
2881 return -ENOENT; /* cannot remove what isn't there */
2884 mddev->pers->quiesce(mddev, 1);
2886 err = bitmap_create(mddev);
2888 bitmap_destroy(mddev);
2889 mddev->pers->quiesce(mddev, 0);
2890 } else if (fd < 0) {
2891 if (mddev->bitmap_file)
2892 fput(mddev->bitmap_file);
2893 mddev->bitmap_file = NULL;
2900 * set_array_info is used two different ways
2901 * The original usage is when creating a new array.
2902 * In this usage, raid_disks is > 0 and it together with
2903 * level, size, not_persistent,layout,chunksize determine the
2904 * shape of the array.
2905 * This will always create an array with a type-0.90.0 superblock.
2906 * The newer usage is when assembling an array.
2907 * In this case raid_disks will be 0, and the major_version field is
2908 * use to determine which style super-blocks are to be found on the devices.
2909 * The minor and patch _version numbers are also kept incase the
2910 * super_block handler wishes to interpret them.
2912 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2915 if (info->raid_disks == 0) {
2916 /* just setting version number for superblock loading */
2917 if (info->major_version < 0 ||
2918 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2919 super_types[info->major_version].name == NULL) {
2920 /* maybe try to auto-load a module? */
2922 "md: superblock version %d not known\n",
2923 info->major_version);
2926 mddev->major_version = info->major_version;
2927 mddev->minor_version = info->minor_version;
2928 mddev->patch_version = info->patch_version;
2931 mddev->major_version = MD_MAJOR_VERSION;
2932 mddev->minor_version = MD_MINOR_VERSION;
2933 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2934 mddev->ctime = get_seconds();
2936 mddev->level = info->level;
2937 mddev->size = info->size;
2938 mddev->raid_disks = info->raid_disks;
2939 /* don't set md_minor, it is determined by which /dev/md* was
2942 if (info->state & (1<<MD_SB_CLEAN))
2943 mddev->recovery_cp = MaxSector;
2945 mddev->recovery_cp = 0;
2946 mddev->persistent = ! info->not_persistent;
2948 mddev->layout = info->layout;
2949 mddev->chunk_size = info->chunk_size;
2951 mddev->max_disks = MD_SB_DISKS;
2953 mddev->sb_dirty = 1;
2955 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
2956 mddev->bitmap_offset = 0;
2959 * Generate a 128 bit UUID
2961 get_random_bytes(mddev->uuid, 16);
2967 * update_array_info is used to change the configuration of an
2969 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2970 * fields in the info are checked against the array.
2971 * Any differences that cannot be handled will cause an error.
2972 * Normally, only one change can be managed at a time.
2974 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2980 /* calculate expected state,ignoring low bits */
2981 if (mddev->bitmap && mddev->bitmap_offset)
2982 state |= (1 << MD_SB_BITMAP_PRESENT);
2984 if (mddev->major_version != info->major_version ||
2985 mddev->minor_version != info->minor_version ||
2986 /* mddev->patch_version != info->patch_version || */
2987 mddev->ctime != info->ctime ||
2988 mddev->level != info->level ||
2989 /* mddev->layout != info->layout || */
2990 !mddev->persistent != info->not_persistent||
2991 mddev->chunk_size != info->chunk_size ||
2992 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2993 ((state^info->state) & 0xfffffe00)
2996 /* Check there is only one change */
2997 if (mddev->size != info->size) cnt++;
2998 if (mddev->raid_disks != info->raid_disks) cnt++;
2999 if (mddev->layout != info->layout) cnt++;
3000 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3001 if (cnt == 0) return 0;
3002 if (cnt > 1) return -EINVAL;
3004 if (mddev->layout != info->layout) {
3006 * we don't need to do anything at the md level, the
3007 * personality will take care of it all.
3009 if (mddev->pers->reconfig == NULL)
3012 return mddev->pers->reconfig(mddev, info->layout, -1);
3014 if (mddev->size != info->size) {
3016 struct list_head *tmp;
3017 if (mddev->pers->resize == NULL)
3019 /* The "size" is the amount of each device that is used.
3020 * This can only make sense for arrays with redundancy.
3021 * linear and raid0 always use whatever space is available
3022 * We can only consider changing the size if no resync
3023 * or reconstruction is happening, and if the new size
3024 * is acceptable. It must fit before the sb_offset or,
3025 * if that is <data_offset, it must fit before the
3026 * size of each device.
3027 * If size is zero, we find the largest size that fits.
3029 if (mddev->sync_thread)
3031 ITERATE_RDEV(mddev,rdev,tmp) {
3033 int fit = (info->size == 0);
3034 if (rdev->sb_offset > rdev->data_offset)
3035 avail = (rdev->sb_offset*2) - rdev->data_offset;
3037 avail = get_capacity(rdev->bdev->bd_disk)
3038 - rdev->data_offset;
3039 if (fit && (info->size == 0 || info->size > avail/2))
3040 info->size = avail/2;
3041 if (avail < ((sector_t)info->size << 1))
3044 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3046 struct block_device *bdev;
3048 bdev = bdget_disk(mddev->gendisk, 0);
3050 down(&bdev->bd_inode->i_sem);
3051 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3052 up(&bdev->bd_inode->i_sem);
3057 if (mddev->raid_disks != info->raid_disks) {
3058 /* change the number of raid disks */
3059 if (mddev->pers->reshape == NULL)
3061 if (info->raid_disks <= 0 ||
3062 info->raid_disks >= mddev->max_disks)
3064 if (mddev->sync_thread)
3066 rv = mddev->pers->reshape(mddev, info->raid_disks);
3068 struct block_device *bdev;
3070 bdev = bdget_disk(mddev->gendisk, 0);
3072 down(&bdev->bd_inode->i_sem);
3073 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3074 up(&bdev->bd_inode->i_sem);
3079 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3080 if (mddev->pers->quiesce == NULL)
3082 if (mddev->recovery || mddev->sync_thread)
3084 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3085 /* add the bitmap */
3088 if (mddev->default_bitmap_offset == 0)
3090 mddev->bitmap_offset = mddev->default_bitmap_offset;
3091 mddev->pers->quiesce(mddev, 1);
3092 rv = bitmap_create(mddev);
3094 bitmap_destroy(mddev);
3095 mddev->pers->quiesce(mddev, 0);
3097 /* remove the bitmap */
3100 if (mddev->bitmap->file)
3102 mddev->pers->quiesce(mddev, 1);
3103 bitmap_destroy(mddev);
3104 mddev->pers->quiesce(mddev, 0);
3105 mddev->bitmap_offset = 0;
3108 md_update_sb(mddev);
3112 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3116 if (mddev->pers == NULL)
3119 rdev = find_rdev(mddev, dev);
3123 md_error(mddev, rdev);
3127 static int md_ioctl(struct inode *inode, struct file *file,
3128 unsigned int cmd, unsigned long arg)
3131 void __user *argp = (void __user *)arg;
3132 struct hd_geometry __user *loc = argp;
3133 mddev_t *mddev = NULL;
3135 if (!capable(CAP_SYS_ADMIN))
3139 * Commands dealing with the RAID driver but not any
3145 err = get_version(argp);
3148 case PRINT_RAID_DEBUG:
3156 autostart_arrays(arg);
3163 * Commands creating/starting a new array:
3166 mddev = inode->i_bdev->bd_disk->private_data;
3174 if (cmd == START_ARRAY) {
3175 /* START_ARRAY doesn't need to lock the array as autostart_array
3176 * does the locking, and it could even be a different array
3181 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3182 "This will not be supported beyond July 2006\n",
3183 current->comm, current->pid);
3186 err = autostart_array(new_decode_dev(arg));
3188 printk(KERN_WARNING "md: autostart failed!\n");
3194 err = mddev_lock(mddev);
3197 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3204 case SET_ARRAY_INFO:
3206 mdu_array_info_t info;
3208 memset(&info, 0, sizeof(info));
3209 else if (copy_from_user(&info, argp, sizeof(info))) {
3214 err = update_array_info(mddev, &info);
3216 printk(KERN_WARNING "md: couldn't update"
3217 " array info. %d\n", err);
3222 if (!list_empty(&mddev->disks)) {
3224 "md: array %s already has disks!\n",
3229 if (mddev->raid_disks) {
3231 "md: array %s already initialised!\n",
3236 err = set_array_info(mddev, &info);
3238 printk(KERN_WARNING "md: couldn't set"
3239 " array info. %d\n", err);
3249 * Commands querying/configuring an existing array:
3251 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3252 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3253 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3254 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3260 * Commands even a read-only array can execute:
3264 case GET_ARRAY_INFO:
3265 err = get_array_info(mddev, argp);
3268 case GET_BITMAP_FILE:
3269 err = get_bitmap_file(mddev, argp);
3273 err = get_disk_info(mddev, argp);
3276 case RESTART_ARRAY_RW:
3277 err = restart_array(mddev);
3281 err = do_md_stop (mddev, 0);
3285 err = do_md_stop (mddev, 1);
3289 * We have a problem here : there is no easy way to give a CHS
3290 * virtual geometry. We currently pretend that we have a 2 heads
3291 * 4 sectors (with a BIG number of cylinders...). This drives
3292 * dosfs just mad... ;-)
3299 err = put_user (2, (char __user *) &loc->heads);
3302 err = put_user (4, (char __user *) &loc->sectors);
3305 err = put_user(get_capacity(mddev->gendisk)/8,
3306 (short __user *) &loc->cylinders);
3309 err = put_user (get_start_sect(inode->i_bdev),
3310 (long __user *) &loc->start);
3315 * The remaining ioctls are changing the state of the
3316 * superblock, so we do not allow them on read-only arrays.
3317 * However non-MD ioctls (e.g. get-size) will still come through
3318 * here and hit the 'default' below, so only disallow
3319 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3321 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3322 mddev->ro && mddev->pers) {
3323 if (mddev->ro == 2) {
3325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3326 md_wakeup_thread(mddev->thread);
3338 mdu_disk_info_t info;
3339 if (copy_from_user(&info, argp, sizeof(info)))
3342 err = add_new_disk(mddev, &info);
3346 case HOT_REMOVE_DISK:
3347 err = hot_remove_disk(mddev, new_decode_dev(arg));
3351 err = hot_add_disk(mddev, new_decode_dev(arg));
3354 case SET_DISK_FAULTY:
3355 err = set_disk_faulty(mddev, new_decode_dev(arg));
3359 err = do_md_run (mddev);
3362 case SET_BITMAP_FILE:
3363 err = set_bitmap_file(mddev, (int)arg);
3367 if (_IOC_TYPE(cmd) == MD_MAJOR)
3368 printk(KERN_WARNING "md: %s(pid %d) used"
3369 " obsolete MD ioctl, upgrade your"
3370 " software to use new ictls.\n",
3371 current->comm, current->pid);
3378 mddev_unlock(mddev);
3388 static int md_open(struct inode *inode, struct file *file)
3391 * Succeed if we can lock the mddev, which confirms that
3392 * it isn't being stopped right now.
3394 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3397 if ((err = mddev_lock(mddev)))
3402 mddev_unlock(mddev);
3404 check_disk_change(inode->i_bdev);
3409 static int md_release(struct inode *inode, struct file * file)
3411 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3420 static int md_media_changed(struct gendisk *disk)
3422 mddev_t *mddev = disk->private_data;
3424 return mddev->changed;
3427 static int md_revalidate(struct gendisk *disk)
3429 mddev_t *mddev = disk->private_data;
3434 static struct block_device_operations md_fops =
3436 .owner = THIS_MODULE,
3438 .release = md_release,
3440 .media_changed = md_media_changed,
3441 .revalidate_disk= md_revalidate,
3444 static int md_thread(void * arg)
3446 mdk_thread_t *thread = arg;
3449 * md_thread is a 'system-thread', it's priority should be very
3450 * high. We avoid resource deadlocks individually in each
3451 * raid personality. (RAID5 does preallocation) We also use RR and
3452 * the very same RT priority as kswapd, thus we will never get
3453 * into a priority inversion deadlock.
3455 * we definitely have to have equal or higher priority than
3456 * bdflush, otherwise bdflush will deadlock if there are too
3457 * many dirty RAID5 blocks.
3460 allow_signal(SIGKILL);
3461 while (!kthread_should_stop()) {
3463 /* We need to wait INTERRUPTIBLE so that
3464 * we don't add to the load-average.
3465 * That means we need to be sure no signals are
3468 if (signal_pending(current))
3469 flush_signals(current);
3471 wait_event_interruptible_timeout
3473 test_bit(THREAD_WAKEUP, &thread->flags)
3474 || kthread_should_stop(),
3478 clear_bit(THREAD_WAKEUP, &thread->flags);
3480 thread->run(thread->mddev);
3486 void md_wakeup_thread(mdk_thread_t *thread)
3489 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3490 set_bit(THREAD_WAKEUP, &thread->flags);
3491 wake_up(&thread->wqueue);
3495 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3498 mdk_thread_t *thread;
3500 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3504 init_waitqueue_head(&thread->wqueue);
3507 thread->mddev = mddev;
3508 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3509 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3510 if (IS_ERR(thread->tsk)) {
3517 void md_unregister_thread(mdk_thread_t *thread)
3519 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3521 kthread_stop(thread->tsk);
3525 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3532 if (!rdev || test_bit(Faulty, &rdev->flags))
3535 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3537 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3538 __builtin_return_address(0),__builtin_return_address(1),
3539 __builtin_return_address(2),__builtin_return_address(3));
3541 if (!mddev->pers->error_handler)
3543 mddev->pers->error_handler(mddev,rdev);
3544 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3545 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3546 md_wakeup_thread(mddev->thread);
3547 md_new_event(mddev);
3550 /* seq_file implementation /proc/mdstat */
3552 static void status_unused(struct seq_file *seq)
3556 struct list_head *tmp;
3558 seq_printf(seq, "unused devices: ");
3560 ITERATE_RDEV_PENDING(rdev,tmp) {
3561 char b[BDEVNAME_SIZE];
3563 seq_printf(seq, "%s ",
3564 bdevname(rdev->bdev,b));
3567 seq_printf(seq, "<none>");
3569 seq_printf(seq, "\n");
3573 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3575 unsigned long max_blocks, resync, res, dt, db, rt;
3577 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3579 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3580 max_blocks = mddev->resync_max_sectors >> 1;
3582 max_blocks = mddev->size;
3585 * Should not happen.
3591 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3593 int i, x = res/50, y = 20-x;
3594 seq_printf(seq, "[");
3595 for (i = 0; i < x; i++)
3596 seq_printf(seq, "=");
3597 seq_printf(seq, ">");
3598 for (i = 0; i < y; i++)
3599 seq_printf(seq, ".");
3600 seq_printf(seq, "] ");
3602 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3603 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3604 "resync" : "recovery"),
3605 res/10, res % 10, resync, max_blocks);
3608 * We do not want to overflow, so the order of operands and
3609 * the * 100 / 100 trick are important. We do a +1 to be
3610 * safe against division by zero. We only estimate anyway.
3612 * dt: time from mark until now
3613 * db: blocks written from mark until now
3614 * rt: remaining time
3616 dt = ((jiffies - mddev->resync_mark) / HZ);
3618 db = resync - (mddev->resync_mark_cnt/2);
3619 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3621 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3623 seq_printf(seq, " speed=%ldK/sec", db/dt);
3626 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3628 struct list_head *tmp;
3638 spin_lock(&all_mddevs_lock);
3639 list_for_each(tmp,&all_mddevs)
3641 mddev = list_entry(tmp, mddev_t, all_mddevs);
3643 spin_unlock(&all_mddevs_lock);
3646 spin_unlock(&all_mddevs_lock);
3648 return (void*)2;/* tail */
3652 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3654 struct list_head *tmp;
3655 mddev_t *next_mddev, *mddev = v;
3661 spin_lock(&all_mddevs_lock);
3663 tmp = all_mddevs.next;
3665 tmp = mddev->all_mddevs.next;
3666 if (tmp != &all_mddevs)
3667 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3669 next_mddev = (void*)2;
3672 spin_unlock(&all_mddevs_lock);
3680 static void md_seq_stop(struct seq_file *seq, void *v)
3684 if (mddev && v != (void*)1 && v != (void*)2)
3688 struct mdstat_info {
3692 static int md_seq_show(struct seq_file *seq, void *v)
3696 struct list_head *tmp2;
3698 struct mdstat_info *mi = seq->private;
3699 struct bitmap *bitmap;
3701 if (v == (void*)1) {
3702 struct mdk_personality *pers;
3703 seq_printf(seq, "Personalities : ");
3704 spin_lock(&pers_lock);
3705 list_for_each_entry(pers, &pers_list, list)
3706 seq_printf(seq, "[%s] ", pers->name);
3708 spin_unlock(&pers_lock);
3709 seq_printf(seq, "\n");
3710 mi->event = atomic_read(&md_event_count);
3713 if (v == (void*)2) {
3718 if (mddev_lock(mddev)!=0)
3720 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3721 seq_printf(seq, "%s : %sactive", mdname(mddev),
3722 mddev->pers ? "" : "in");
3725 seq_printf(seq, " (read-only)");
3727 seq_printf(seq, "(auto-read-only)");
3728 seq_printf(seq, " %s", mddev->pers->name);
3732 ITERATE_RDEV(mddev,rdev,tmp2) {
3733 char b[BDEVNAME_SIZE];
3734 seq_printf(seq, " %s[%d]",
3735 bdevname(rdev->bdev,b), rdev->desc_nr);
3736 if (test_bit(WriteMostly, &rdev->flags))
3737 seq_printf(seq, "(W)");
3738 if (test_bit(Faulty, &rdev->flags)) {
3739 seq_printf(seq, "(F)");
3741 } else if (rdev->raid_disk < 0)
3742 seq_printf(seq, "(S)"); /* spare */
3746 if (!list_empty(&mddev->disks)) {
3748 seq_printf(seq, "\n %llu blocks",
3749 (unsigned long long)mddev->array_size);
3751 seq_printf(seq, "\n %llu blocks",
3752 (unsigned long long)size);
3754 if (mddev->persistent) {
3755 if (mddev->major_version != 0 ||
3756 mddev->minor_version != 90) {
3757 seq_printf(seq," super %d.%d",
3758 mddev->major_version,
3759 mddev->minor_version);
3762 seq_printf(seq, " super non-persistent");
3765 mddev->pers->status (seq, mddev);
3766 seq_printf(seq, "\n ");
3767 if (mddev->pers->sync_request) {
3768 if (mddev->curr_resync > 2) {
3769 status_resync (seq, mddev);
3770 seq_printf(seq, "\n ");
3771 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3772 seq_printf(seq, "\tresync=DELAYED\n ");
3773 else if (mddev->recovery_cp < MaxSector)
3774 seq_printf(seq, "\tresync=PENDING\n ");
3777 seq_printf(seq, "\n ");
3779 if ((bitmap = mddev->bitmap)) {
3780 unsigned long chunk_kb;
3781 unsigned long flags;
3782 spin_lock_irqsave(&bitmap->lock, flags);
3783 chunk_kb = bitmap->chunksize >> 10;
3784 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3786 bitmap->pages - bitmap->missing_pages,
3788 (bitmap->pages - bitmap->missing_pages)
3789 << (PAGE_SHIFT - 10),
3790 chunk_kb ? chunk_kb : bitmap->chunksize,
3791 chunk_kb ? "KB" : "B");
3793 seq_printf(seq, ", file: ");
3794 seq_path(seq, bitmap->file->f_vfsmnt,
3795 bitmap->file->f_dentry," \t\n");
3798 seq_printf(seq, "\n");
3799 spin_unlock_irqrestore(&bitmap->lock, flags);
3802 seq_printf(seq, "\n");
3804 mddev_unlock(mddev);
3809 static struct seq_operations md_seq_ops = {
3810 .start = md_seq_start,
3811 .next = md_seq_next,
3812 .stop = md_seq_stop,
3813 .show = md_seq_show,
3816 static int md_seq_open(struct inode *inode, struct file *file)
3819 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3823 error = seq_open(file, &md_seq_ops);
3827 struct seq_file *p = file->private_data;
3829 mi->event = atomic_read(&md_event_count);
3834 static int md_seq_release(struct inode *inode, struct file *file)
3836 struct seq_file *m = file->private_data;
3837 struct mdstat_info *mi = m->private;
3840 return seq_release(inode, file);
3843 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
3845 struct seq_file *m = filp->private_data;
3846 struct mdstat_info *mi = m->private;
3849 poll_wait(filp, &md_event_waiters, wait);
3851 /* always allow read */
3852 mask = POLLIN | POLLRDNORM;
3854 if (mi->event != atomic_read(&md_event_count))
3855 mask |= POLLERR | POLLPRI;
3859 static struct file_operations md_seq_fops = {
3860 .open = md_seq_open,
3862 .llseek = seq_lseek,
3863 .release = md_seq_release,
3864 .poll = mdstat_poll,
3867 int register_md_personality(struct mdk_personality *p)
3869 spin_lock(&pers_lock);
3870 list_add_tail(&p->list, &pers_list);
3871 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
3872 spin_unlock(&pers_lock);
3876 int unregister_md_personality(struct mdk_personality *p)
3878 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
3879 spin_lock(&pers_lock);
3880 list_del_init(&p->list);
3881 spin_unlock(&pers_lock);
3885 static int is_mddev_idle(mddev_t *mddev)
3888 struct list_head *tmp;
3890 unsigned long curr_events;
3893 ITERATE_RDEV(mddev,rdev,tmp) {
3894 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3895 curr_events = disk_stat_read(disk, sectors[0]) +
3896 disk_stat_read(disk, sectors[1]) -
3897 atomic_read(&disk->sync_io);
3898 /* The difference between curr_events and last_events
3899 * will be affected by any new non-sync IO (making
3900 * curr_events bigger) and any difference in the amount of
3901 * in-flight syncio (making current_events bigger or smaller)
3902 * The amount in-flight is currently limited to
3903 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
3904 * which is at most 4096 sectors.
3905 * These numbers are fairly fragile and should be made
3906 * more robust, probably by enforcing the
3907 * 'window size' that md_do_sync sort-of uses.
3909 * Note: the following is an unsigned comparison.
3911 if ((curr_events - rdev->last_events + 4096) > 8192) {
3912 rdev->last_events = curr_events;
3919 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3921 /* another "blocks" (512byte) blocks have been synced */
3922 atomic_sub(blocks, &mddev->recovery_active);
3923 wake_up(&mddev->recovery_wait);
3925 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3926 md_wakeup_thread(mddev->thread);
3927 // stop recovery, signal do_sync ....
3932 /* md_write_start(mddev, bi)
3933 * If we need to update some array metadata (e.g. 'active' flag
3934 * in superblock) before writing, schedule a superblock update
3935 * and wait for it to complete.
3937 void md_write_start(mddev_t *mddev, struct bio *bi)
3939 if (bio_data_dir(bi) != WRITE)
3942 BUG_ON(mddev->ro == 1);
3943 if (mddev->ro == 2) {
3944 /* need to switch to read/write */
3946 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3947 md_wakeup_thread(mddev->thread);
3949 atomic_inc(&mddev->writes_pending);
3950 if (mddev->in_sync) {
3951 spin_lock_irq(&mddev->write_lock);
3952 if (mddev->in_sync) {
3954 mddev->sb_dirty = 1;
3955 md_wakeup_thread(mddev->thread);
3957 spin_unlock_irq(&mddev->write_lock);
3959 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3962 void md_write_end(mddev_t *mddev)
3964 if (atomic_dec_and_test(&mddev->writes_pending)) {
3965 if (mddev->safemode == 2)
3966 md_wakeup_thread(mddev->thread);
3968 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3972 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3974 #define SYNC_MARKS 10
3975 #define SYNC_MARK_STEP (3*HZ)
3976 static void md_do_sync(mddev_t *mddev)
3979 unsigned int currspeed = 0,
3981 sector_t max_sectors,j, io_sectors;
3982 unsigned long mark[SYNC_MARKS];
3983 sector_t mark_cnt[SYNC_MARKS];
3985 struct list_head *tmp;
3986 sector_t last_check;
3989 /* just incase thread restarts... */
3990 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3993 /* we overload curr_resync somewhat here.
3994 * 0 == not engaged in resync at all
3995 * 2 == checking that there is no conflict with another sync
3996 * 1 == like 2, but have yielded to allow conflicting resync to
3998 * other == active in resync - this many blocks
4000 * Before starting a resync we must have set curr_resync to
4001 * 2, and then checked that every "conflicting" array has curr_resync
4002 * less than ours. When we find one that is the same or higher
4003 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4004 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4005 * This will mean we have to start checking from the beginning again.
4010 mddev->curr_resync = 2;
4013 if (kthread_should_stop()) {
4014 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4017 ITERATE_MDDEV(mddev2,tmp) {
4018 if (mddev2 == mddev)
4020 if (mddev2->curr_resync &&
4021 match_mddev_units(mddev,mddev2)) {
4023 if (mddev < mddev2 && mddev->curr_resync == 2) {
4024 /* arbitrarily yield */
4025 mddev->curr_resync = 1;
4026 wake_up(&resync_wait);
4028 if (mddev > mddev2 && mddev->curr_resync == 1)
4029 /* no need to wait here, we can wait the next
4030 * time 'round when curr_resync == 2
4033 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4034 if (!kthread_should_stop() &&
4035 mddev2->curr_resync >= mddev->curr_resync) {
4036 printk(KERN_INFO "md: delaying resync of %s"
4037 " until %s has finished resync (they"
4038 " share one or more physical units)\n",
4039 mdname(mddev), mdname(mddev2));
4042 finish_wait(&resync_wait, &wq);
4045 finish_wait(&resync_wait, &wq);
4048 } while (mddev->curr_resync < 2);
4050 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4051 /* resync follows the size requested by the personality,
4052 * which defaults to physical size, but can be virtual size
4054 max_sectors = mddev->resync_max_sectors;
4055 mddev->resync_mismatches = 0;
4057 /* recovery follows the physical size of devices */
4058 max_sectors = mddev->size << 1;
4060 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4061 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4062 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4063 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4064 "(but not more than %d KB/sec) for reconstruction.\n",
4065 sysctl_speed_limit_max);
4067 is_mddev_idle(mddev); /* this also initializes IO event counters */
4068 /* we don't use the checkpoint if there's a bitmap */
4069 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4070 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4071 j = mddev->recovery_cp;
4075 for (m = 0; m < SYNC_MARKS; m++) {
4077 mark_cnt[m] = io_sectors;
4080 mddev->resync_mark = mark[last_mark];
4081 mddev->resync_mark_cnt = mark_cnt[last_mark];
4084 * Tune reconstruction:
4086 window = 32*(PAGE_SIZE/512);
4087 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4088 window/2,(unsigned long long) max_sectors/2);
4090 atomic_set(&mddev->recovery_active, 0);
4091 init_waitqueue_head(&mddev->recovery_wait);
4096 "md: resuming recovery of %s from checkpoint.\n",
4098 mddev->curr_resync = j;
4101 while (j < max_sectors) {
4105 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4106 currspeed < sysctl_speed_limit_min);
4108 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4112 if (!skipped) { /* actual IO requested */
4113 io_sectors += sectors;
4114 atomic_add(sectors, &mddev->recovery_active);
4118 if (j>1) mddev->curr_resync = j;
4119 if (last_check == 0)
4120 /* this is the earliers that rebuilt will be
4121 * visible in /proc/mdstat
4123 md_new_event(mddev);
4125 if (last_check + window > io_sectors || j == max_sectors)
4128 last_check = io_sectors;
4130 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4131 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4135 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4137 int next = (last_mark+1) % SYNC_MARKS;
4139 mddev->resync_mark = mark[next];
4140 mddev->resync_mark_cnt = mark_cnt[next];
4141 mark[next] = jiffies;
4142 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4147 if (kthread_should_stop()) {
4149 * got a signal, exit.
4152 "md: md_do_sync() got signal ... exiting\n");
4153 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4158 * this loop exits only if either when we are slower than
4159 * the 'hard' speed limit, or the system was IO-idle for
4161 * the system might be non-idle CPU-wise, but we only care
4162 * about not overloading the IO subsystem. (things like an
4163 * e2fsck being done on the RAID array should execute fast)
4165 mddev->queue->unplug_fn(mddev->queue);
4168 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4169 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4171 if (currspeed > sysctl_speed_limit_min) {
4172 if ((currspeed > sysctl_speed_limit_max) ||
4173 !is_mddev_idle(mddev)) {
4179 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4181 * this also signals 'finished resyncing' to md_stop
4184 mddev->queue->unplug_fn(mddev->queue);
4186 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4188 /* tell personality that we are finished */
4189 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4191 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4192 mddev->curr_resync > 2 &&
4193 mddev->curr_resync >= mddev->recovery_cp) {
4194 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4196 "md: checkpointing recovery of %s.\n",
4198 mddev->recovery_cp = mddev->curr_resync;
4200 mddev->recovery_cp = MaxSector;
4204 mddev->curr_resync = 0;
4205 wake_up(&resync_wait);
4206 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4207 md_wakeup_thread(mddev->thread);
4212 * This routine is regularly called by all per-raid-array threads to
4213 * deal with generic issues like resync and super-block update.
4214 * Raid personalities that don't have a thread (linear/raid0) do not
4215 * need this as they never do any recovery or update the superblock.
4217 * It does not do any resync itself, but rather "forks" off other threads
4218 * to do that as needed.
4219 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4220 * "->recovery" and create a thread at ->sync_thread.
4221 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4222 * and wakeups up this thread which will reap the thread and finish up.
4223 * This thread also removes any faulty devices (with nr_pending == 0).
4225 * The overall approach is:
4226 * 1/ if the superblock needs updating, update it.
4227 * 2/ If a recovery thread is running, don't do anything else.
4228 * 3/ If recovery has finished, clean up, possibly marking spares active.
4229 * 4/ If there are any faulty devices, remove them.
4230 * 5/ If array is degraded, try to add spares devices
4231 * 6/ If array has spares or is not in-sync, start a resync thread.
4233 void md_check_recovery(mddev_t *mddev)
4236 struct list_head *rtmp;
4240 bitmap_daemon_work(mddev->bitmap);
4245 if (signal_pending(current)) {
4246 if (mddev->pers->sync_request) {
4247 printk(KERN_INFO "md: %s in immediate safe mode\n",
4249 mddev->safemode = 2;
4251 flush_signals(current);
4256 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4257 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4258 (mddev->safemode == 1) ||
4259 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4260 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4264 if (mddev_trylock(mddev)==0) {
4267 spin_lock_irq(&mddev->write_lock);
4268 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4269 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4271 mddev->sb_dirty = 1;
4273 if (mddev->safemode == 1)
4274 mddev->safemode = 0;
4275 spin_unlock_irq(&mddev->write_lock);
4277 if (mddev->sb_dirty)
4278 md_update_sb(mddev);
4281 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4282 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4283 /* resync/recovery still happening */
4284 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4287 if (mddev->sync_thread) {
4288 /* resync has finished, collect result */
4289 md_unregister_thread(mddev->sync_thread);
4290 mddev->sync_thread = NULL;
4291 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4292 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4294 /* activate any spares */
4295 mddev->pers->spare_active(mddev);
4297 md_update_sb(mddev);
4299 /* if array is no-longer degraded, then any saved_raid_disk
4300 * information must be scrapped
4302 if (!mddev->degraded)
4303 ITERATE_RDEV(mddev,rdev,rtmp)
4304 rdev->saved_raid_disk = -1;
4306 mddev->recovery = 0;
4307 /* flag recovery needed just to double check */
4308 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4309 md_new_event(mddev);
4312 /* Clear some bits that don't mean anything, but
4315 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4316 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4317 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4318 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4320 /* no recovery is running.
4321 * remove any failed drives, then
4322 * add spares if possible.
4323 * Spare are also removed and re-added, to allow
4324 * the personality to fail the re-add.
4326 ITERATE_RDEV(mddev,rdev,rtmp)
4327 if (rdev->raid_disk >= 0 &&
4328 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4329 atomic_read(&rdev->nr_pending)==0) {
4330 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4332 sprintf(nm,"rd%d", rdev->raid_disk);
4333 sysfs_remove_link(&mddev->kobj, nm);
4334 rdev->raid_disk = -1;
4338 if (mddev->degraded) {
4339 ITERATE_RDEV(mddev,rdev,rtmp)
4340 if (rdev->raid_disk < 0
4341 && !test_bit(Faulty, &rdev->flags)) {
4342 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4344 sprintf(nm, "rd%d", rdev->raid_disk);
4345 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4347 md_new_event(mddev);
4354 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4355 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4356 } else if (mddev->recovery_cp < MaxSector) {
4357 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4358 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4359 /* nothing to be done ... */
4362 if (mddev->pers->sync_request) {
4363 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4364 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4365 /* We are adding a device or devices to an array
4366 * which has the bitmap stored on all devices.
4367 * So make sure all bitmap pages get written
4369 bitmap_write_all(mddev->bitmap);
4371 mddev->sync_thread = md_register_thread(md_do_sync,
4374 if (!mddev->sync_thread) {
4375 printk(KERN_ERR "%s: could not start resync"
4378 /* leave the spares where they are, it shouldn't hurt */
4379 mddev->recovery = 0;
4381 md_wakeup_thread(mddev->sync_thread);
4382 md_new_event(mddev);
4385 mddev_unlock(mddev);
4389 static int md_notify_reboot(struct notifier_block *this,
4390 unsigned long code, void *x)
4392 struct list_head *tmp;
4395 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4397 printk(KERN_INFO "md: stopping all md devices.\n");
4399 ITERATE_MDDEV(mddev,tmp)
4400 if (mddev_trylock(mddev)==0)
4401 do_md_stop (mddev, 1);
4403 * certain more exotic SCSI devices are known to be
4404 * volatile wrt too early system reboots. While the
4405 * right place to handle this issue is the given
4406 * driver, we do want to have a safe RAID driver ...
4413 static struct notifier_block md_notifier = {
4414 .notifier_call = md_notify_reboot,
4416 .priority = INT_MAX, /* before any real devices */
4419 static void md_geninit(void)
4421 struct proc_dir_entry *p;
4423 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4425 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4427 p->proc_fops = &md_seq_fops;
4430 static int __init md_init(void)
4434 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4435 " MD_SB_DISKS=%d\n",
4436 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4437 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4438 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4441 if (register_blkdev(MAJOR_NR, "md"))
4443 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4444 unregister_blkdev(MAJOR_NR, "md");
4448 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4449 md_probe, NULL, NULL);
4450 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4451 md_probe, NULL, NULL);
4453 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4454 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4455 S_IFBLK|S_IRUSR|S_IWUSR,
4458 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4459 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4460 S_IFBLK|S_IRUSR|S_IWUSR,
4464 register_reboot_notifier(&md_notifier);
4465 raid_table_header = register_sysctl_table(raid_root_table, 1);
4475 * Searches all registered partitions for autorun RAID arrays
4478 static dev_t detected_devices[128];
4481 void md_autodetect_dev(dev_t dev)
4483 if (dev_cnt >= 0 && dev_cnt < 127)
4484 detected_devices[dev_cnt++] = dev;
4488 static void autostart_arrays(int part)
4493 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4495 for (i = 0; i < dev_cnt; i++) {
4496 dev_t dev = detected_devices[i];
4498 rdev = md_import_device(dev,0, 0);
4502 if (test_bit(Faulty, &rdev->flags)) {
4506 list_add(&rdev->same_set, &pending_raid_disks);
4510 autorun_devices(part);
4515 static __exit void md_exit(void)
4518 struct list_head *tmp;
4520 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4521 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4522 for (i=0; i < MAX_MD_DEVS; i++)
4523 devfs_remove("md/%d", i);
4524 for (i=0; i < MAX_MD_DEVS; i++)
4525 devfs_remove("md/d%d", i);
4529 unregister_blkdev(MAJOR_NR,"md");
4530 unregister_blkdev(mdp_major, "mdp");
4531 unregister_reboot_notifier(&md_notifier);
4532 unregister_sysctl_table(raid_table_header);
4533 remove_proc_entry("mdstat", NULL);
4534 ITERATE_MDDEV(mddev,tmp) {
4535 struct gendisk *disk = mddev->gendisk;
4538 export_array(mddev);
4541 mddev->gendisk = NULL;
4546 module_init(md_init)
4547 module_exit(md_exit)
4549 static int get_ro(char *buffer, struct kernel_param *kp)
4551 return sprintf(buffer, "%d", start_readonly);
4553 static int set_ro(const char *val, struct kernel_param *kp)
4556 int num = simple_strtoul(val, &e, 10);
4557 if (*val && (*e == '\0' || *e == '\n')) {
4558 start_readonly = num;
4564 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4565 module_param(start_dirty_degraded, int, 0644);
4568 EXPORT_SYMBOL(register_md_personality);
4569 EXPORT_SYMBOL(unregister_md_personality);
4570 EXPORT_SYMBOL(md_error);
4571 EXPORT_SYMBOL(md_done_sync);
4572 EXPORT_SYMBOL(md_write_start);
4573 EXPORT_SYMBOL(md_write_end);
4574 EXPORT_SYMBOL(md_register_thread);
4575 EXPORT_SYMBOL(md_unregister_thread);
4576 EXPORT_SYMBOL(md_wakeup_thread);
4577 EXPORT_SYMBOL(md_print_devices);
4578 EXPORT_SYMBOL(md_check_recovery);
4579 MODULE_LICENSE("GPL");
4581 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);