2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
46 #include <linux/init.h>
48 #include <linux/file.h>
51 #include <linux/kmod.h>
54 #include <asm/unaligned.h>
56 #define MAJOR_NR MD_MAJOR
59 /* 63 partitions with the alternate major number (mdp) */
60 #define MdpMinorShift 6
63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
67 static void autostart_arrays (int part);
70 static mdk_personality_t *pers[MAX_PERSONALITY];
71 static DEFINE_SPINLOCK(pers_lock);
74 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
75 * is 1000 KB/sec, so the extra system load does not show up that much.
76 * Increase it if you want to have more _guaranteed_ speed. Note that
77 * the RAID driver will use the maximum available bandwidth if the IO
78 * subsystem is idle. There is also an 'absolute maximum' reconstruction
79 * speed limit - in case reconstruction slows down your system despite
82 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
85 static int sysctl_speed_limit_min = 1000;
86 static int sysctl_speed_limit_max = 200000;
88 static struct ctl_table_header *raid_table_header;
90 static ctl_table raid_table[] = {
92 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
93 .procname = "speed_limit_min",
94 .data = &sysctl_speed_limit_min,
95 .maxlen = sizeof(int),
97 .proc_handler = &proc_dointvec,
100 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
101 .procname = "speed_limit_max",
102 .data = &sysctl_speed_limit_max,
103 .maxlen = sizeof(int),
105 .proc_handler = &proc_dointvec,
110 static ctl_table raid_dir_table[] = {
112 .ctl_name = DEV_RAID,
121 static ctl_table raid_root_table[] = {
127 .child = raid_dir_table,
132 static struct block_device_operations md_fops;
135 * Enables to iterate over all existing md arrays
136 * all_mddevs_lock protects this list.
138 static LIST_HEAD(all_mddevs);
139 static DEFINE_SPINLOCK(all_mddevs_lock);
143 * iterates through all used mddevs in the system.
144 * We take care to grab the all_mddevs_lock whenever navigating
145 * the list, and to always hold a refcount when unlocked.
146 * Any code which breaks out of this loop while own
147 * a reference to the current mddev and must mddev_put it.
149 #define ITERATE_MDDEV(mddev,tmp) \
151 for (({ spin_lock(&all_mddevs_lock); \
152 tmp = all_mddevs.next; \
154 ({ if (tmp != &all_mddevs) \
155 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
156 spin_unlock(&all_mddevs_lock); \
157 if (mddev) mddev_put(mddev); \
158 mddev = list_entry(tmp, mddev_t, all_mddevs); \
159 tmp != &all_mddevs;}); \
160 ({ spin_lock(&all_mddevs_lock); \
165 static int md_fail_request (request_queue_t *q, struct bio *bio)
167 bio_io_error(bio, bio->bi_size);
171 static inline mddev_t *mddev_get(mddev_t *mddev)
173 atomic_inc(&mddev->active);
177 static void mddev_put(mddev_t *mddev)
179 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
181 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
182 list_del(&mddev->all_mddevs);
183 blk_put_queue(mddev->queue);
184 kobject_unregister(&mddev->kobj);
186 spin_unlock(&all_mddevs_lock);
189 static mddev_t * mddev_find(dev_t unit)
191 mddev_t *mddev, *new = NULL;
194 spin_lock(&all_mddevs_lock);
195 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
196 if (mddev->unit == unit) {
198 spin_unlock(&all_mddevs_lock);
204 list_add(&new->all_mddevs, &all_mddevs);
205 spin_unlock(&all_mddevs_lock);
208 spin_unlock(&all_mddevs_lock);
210 new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
214 memset(new, 0, sizeof(*new));
217 if (MAJOR(unit) == MD_MAJOR)
218 new->md_minor = MINOR(unit);
220 new->md_minor = MINOR(unit) >> MdpMinorShift;
222 init_MUTEX(&new->reconfig_sem);
223 INIT_LIST_HEAD(&new->disks);
224 INIT_LIST_HEAD(&new->all_mddevs);
225 init_timer(&new->safemode_timer);
226 atomic_set(&new->active, 1);
227 spin_lock_init(&new->write_lock);
228 init_waitqueue_head(&new->sb_wait);
230 new->queue = blk_alloc_queue(GFP_KERNEL);
236 blk_queue_make_request(new->queue, md_fail_request);
241 static inline int mddev_lock(mddev_t * mddev)
243 return down_interruptible(&mddev->reconfig_sem);
246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
248 down(&mddev->reconfig_sem);
251 static inline int mddev_trylock(mddev_t * mddev)
253 return down_trylock(&mddev->reconfig_sem);
256 static inline void mddev_unlock(mddev_t * mddev)
258 up(&mddev->reconfig_sem);
260 md_wakeup_thread(mddev->thread);
263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
266 struct list_head *tmp;
268 ITERATE_RDEV(mddev,rdev,tmp) {
269 if (rdev->desc_nr == nr)
275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
277 struct list_head *tmp;
280 ITERATE_RDEV(mddev,rdev,tmp) {
281 if (rdev->bdev->bd_dev == dev)
287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
289 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
290 return MD_NEW_SIZE_BLOCKS(size);
293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
297 size = rdev->sb_offset;
300 size &= ~((sector_t)chunk_size/1024 - 1);
304 static int alloc_disk_sb(mdk_rdev_t * rdev)
309 rdev->sb_page = alloc_page(GFP_KERNEL);
310 if (!rdev->sb_page) {
311 printk(KERN_ALERT "md: out of memory.\n");
318 static void free_disk_sb(mdk_rdev_t * rdev)
321 page_cache_release(rdev->sb_page);
323 rdev->sb_page = NULL;
330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
332 mdk_rdev_t *rdev = bio->bi_private;
336 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
337 md_error(rdev->mddev, rdev);
339 if (atomic_dec_and_test(&rdev->mddev->pending_writes))
340 wake_up(&rdev->mddev->sb_wait);
345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
346 sector_t sector, int size, struct page *page)
348 /* write first size bytes of page to sector of rdev
349 * Increment mddev->pending_writes before returning
350 * and decrement it on completion, waking up sb_wait
351 * if zero is reached.
352 * If an error occurred, call md_error
354 struct bio *bio = bio_alloc(GFP_NOIO, 1);
356 bio->bi_bdev = rdev->bdev;
357 bio->bi_sector = sector;
358 bio_add_page(bio, page, size, 0);
359 bio->bi_private = rdev;
360 bio->bi_end_io = super_written;
361 atomic_inc(&mddev->pending_writes);
362 submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
370 complete((struct completion*)bio->bi_private);
374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
375 struct page *page, int rw)
377 struct bio *bio = bio_alloc(GFP_NOIO, 1);
378 struct completion event;
381 rw |= (1 << BIO_RW_SYNC);
384 bio->bi_sector = sector;
385 bio_add_page(bio, page, size, 0);
386 init_completion(&event);
387 bio->bi_private = &event;
388 bio->bi_end_io = bi_complete;
390 wait_for_completion(&event);
392 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
397 static int read_disk_sb(mdk_rdev_t * rdev, int size)
399 char b[BDEVNAME_SIZE];
400 if (!rdev->sb_page) {
408 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
414 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
415 bdevname(rdev->bdev,b));
419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
421 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
422 (sb1->set_uuid1 == sb2->set_uuid1) &&
423 (sb1->set_uuid2 == sb2->set_uuid2) &&
424 (sb1->set_uuid3 == sb2->set_uuid3))
432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
435 mdp_super_t *tmp1, *tmp2;
437 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
438 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
440 if (!tmp1 || !tmp2) {
442 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
450 * nr_disks is not constant
455 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
466 static unsigned int calc_sb_csum(mdp_super_t * sb)
468 unsigned int disk_csum, csum;
470 disk_csum = sb->sb_csum;
472 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
473 sb->sb_csum = disk_csum;
479 * Handle superblock details.
480 * We want to be able to handle multiple superblock formats
481 * so we have a common interface to them all, and an array of
482 * different handlers.
483 * We rely on user-space to write the initial superblock, and support
484 * reading and updating of superblocks.
485 * Interface methods are:
486 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
487 * loads and validates a superblock on dev.
488 * if refdev != NULL, compare superblocks on both devices
490 * 0 - dev has a superblock that is compatible with refdev
491 * 1 - dev has a superblock that is compatible and newer than refdev
492 * so dev should be used as the refdev in future
493 * -EINVAL superblock incompatible or invalid
494 * -othererror e.g. -EIO
496 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
497 * Verify that dev is acceptable into mddev.
498 * The first time, mddev->raid_disks will be 0, and data from
499 * dev should be merged in. Subsequent calls check that dev
500 * is new enough. Return 0 or -EINVAL
502 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
503 * Update the superblock for rdev with data in mddev
504 * This does not write to disc.
510 struct module *owner;
511 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
512 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
513 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
517 * load_super for 0.90.0
519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
521 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
527 * Calculate the position of the superblock,
528 * it's at the end of the disk.
530 * It also happens to be a multiple of 4Kb.
532 sb_offset = calc_dev_sboffset(rdev->bdev);
533 rdev->sb_offset = sb_offset;
535 ret = read_disk_sb(rdev, MD_SB_BYTES);
540 bdevname(rdev->bdev, b);
541 sb = (mdp_super_t*)page_address(rdev->sb_page);
543 if (sb->md_magic != MD_SB_MAGIC) {
544 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
549 if (sb->major_version != 0 ||
550 sb->minor_version != 90) {
551 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
552 sb->major_version, sb->minor_version,
557 if (sb->raid_disks <= 0)
560 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
561 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
566 rdev->preferred_minor = sb->md_minor;
567 rdev->data_offset = 0;
568 rdev->sb_size = MD_SB_BYTES;
570 if (sb->level == LEVEL_MULTIPATH)
573 rdev->desc_nr = sb->this_disk.number;
579 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
580 if (!uuid_equal(refsb, sb)) {
581 printk(KERN_WARNING "md: %s has different UUID to %s\n",
582 b, bdevname(refdev->bdev,b2));
585 if (!sb_equal(refsb, sb)) {
586 printk(KERN_WARNING "md: %s has same UUID"
587 " but different superblock to %s\n",
588 b, bdevname(refdev->bdev, b2));
592 ev2 = md_event(refsb);
598 rdev->size = calc_dev_size(rdev, sb->chunk_size);
605 * validate_super for 0.90.0
607 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
610 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
612 rdev->raid_disk = -1;
614 if (mddev->raid_disks == 0) {
615 mddev->major_version = 0;
616 mddev->minor_version = sb->minor_version;
617 mddev->patch_version = sb->patch_version;
618 mddev->persistent = ! sb->not_persistent;
619 mddev->chunk_size = sb->chunk_size;
620 mddev->ctime = sb->ctime;
621 mddev->utime = sb->utime;
622 mddev->level = sb->level;
623 mddev->layout = sb->layout;
624 mddev->raid_disks = sb->raid_disks;
625 mddev->size = sb->size;
626 mddev->events = md_event(sb);
627 mddev->bitmap_offset = 0;
628 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
630 if (sb->state & (1<<MD_SB_CLEAN))
631 mddev->recovery_cp = MaxSector;
633 if (sb->events_hi == sb->cp_events_hi &&
634 sb->events_lo == sb->cp_events_lo) {
635 mddev->recovery_cp = sb->recovery_cp;
637 mddev->recovery_cp = 0;
640 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
641 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
642 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
643 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
645 mddev->max_disks = MD_SB_DISKS;
647 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
648 mddev->bitmap_file == NULL) {
649 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
650 /* FIXME use a better test */
651 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
654 mddev->bitmap_offset = mddev->default_bitmap_offset;
657 } else if (mddev->pers == NULL) {
658 /* Insist on good event counter while assembling */
659 __u64 ev1 = md_event(sb);
661 if (ev1 < mddev->events)
663 } else if (mddev->bitmap) {
664 /* if adding to array with a bitmap, then we can accept an
665 * older device ... but not too old.
667 __u64 ev1 = md_event(sb);
668 if (ev1 < mddev->bitmap->events_cleared)
670 } else /* just a hot-add of a new device, leave raid_disk at -1 */
673 if (mddev->level != LEVEL_MULTIPATH) {
676 desc = sb->disks + rdev->desc_nr;
678 if (desc->state & (1<<MD_DISK_FAULTY))
680 else if (desc->state & (1<<MD_DISK_SYNC) &&
681 desc->raid_disk < mddev->raid_disks) {
683 rdev->raid_disk = desc->raid_disk;
685 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
686 set_bit(WriteMostly, &rdev->flags);
687 } else /* MULTIPATH are always insync */
693 * sync_super for 0.90.0
695 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
698 struct list_head *tmp;
700 int next_spare = mddev->raid_disks;
702 /* make rdev->sb match mddev data..
705 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
706 * 3/ any empty disks < next_spare become removed
708 * disks[0] gets initialised to REMOVED because
709 * we cannot be sure from other fields if it has
710 * been initialised or not.
713 int active=0, working=0,failed=0,spare=0,nr_disks=0;
714 unsigned int fixdesc=0;
716 rdev->sb_size = MD_SB_BYTES;
718 sb = (mdp_super_t*)page_address(rdev->sb_page);
720 memset(sb, 0, sizeof(*sb));
722 sb->md_magic = MD_SB_MAGIC;
723 sb->major_version = mddev->major_version;
724 sb->minor_version = mddev->minor_version;
725 sb->patch_version = mddev->patch_version;
726 sb->gvalid_words = 0; /* ignored */
727 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
728 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
729 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
730 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
732 sb->ctime = mddev->ctime;
733 sb->level = mddev->level;
734 sb->size = mddev->size;
735 sb->raid_disks = mddev->raid_disks;
736 sb->md_minor = mddev->md_minor;
737 sb->not_persistent = !mddev->persistent;
738 sb->utime = mddev->utime;
740 sb->events_hi = (mddev->events>>32);
741 sb->events_lo = (u32)mddev->events;
745 sb->recovery_cp = mddev->recovery_cp;
746 sb->cp_events_hi = (mddev->events>>32);
747 sb->cp_events_lo = (u32)mddev->events;
748 if (mddev->recovery_cp == MaxSector)
749 sb->state = (1<< MD_SB_CLEAN);
753 sb->layout = mddev->layout;
754 sb->chunk_size = mddev->chunk_size;
756 if (mddev->bitmap && mddev->bitmap_file == NULL)
757 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
759 sb->disks[0].state = (1<<MD_DISK_REMOVED);
760 ITERATE_RDEV(mddev,rdev2,tmp) {
763 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
764 desc_nr = rdev2->raid_disk;
766 desc_nr = next_spare++;
767 if (desc_nr != rdev2->desc_nr) {
768 fixdesc |= (1 << desc_nr);
769 rdev2->desc_nr = desc_nr;
770 if (rdev2->raid_disk >= 0) {
772 sprintf(nm, "rd%d", rdev2->raid_disk);
773 sysfs_remove_link(&mddev->kobj, nm);
775 sysfs_remove_link(&rdev2->kobj, "block");
776 kobject_del(&rdev2->kobj);
778 d = &sb->disks[rdev2->desc_nr];
780 d->number = rdev2->desc_nr;
781 d->major = MAJOR(rdev2->bdev->bd_dev);
782 d->minor = MINOR(rdev2->bdev->bd_dev);
783 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
784 d->raid_disk = rdev2->raid_disk;
786 d->raid_disk = rdev2->desc_nr; /* compatibility */
788 d->state = (1<<MD_DISK_FAULTY);
790 } else if (rdev2->in_sync) {
791 d->state = (1<<MD_DISK_ACTIVE);
792 d->state |= (1<<MD_DISK_SYNC);
800 if (test_bit(WriteMostly, &rdev2->flags))
801 d->state |= (1<<MD_DISK_WRITEMOSTLY);
804 ITERATE_RDEV(mddev,rdev2,tmp)
805 if (fixdesc & (1<<rdev2->desc_nr)) {
806 snprintf(rdev2->kobj.name, KOBJ_NAME_LEN, "dev%d",
808 kobject_add(&rdev2->kobj);
809 sysfs_create_link(&rdev2->kobj,
810 &rdev2->bdev->bd_disk->kobj,
812 if (rdev2->raid_disk >= 0) {
814 sprintf(nm, "rd%d", rdev2->raid_disk);
815 sysfs_create_link(&mddev->kobj,
819 /* now set the "removed" and "faulty" bits on any missing devices */
820 for (i=0 ; i < mddev->raid_disks ; i++) {
821 mdp_disk_t *d = &sb->disks[i];
822 if (d->state == 0 && d->number == 0) {
825 d->state = (1<<MD_DISK_REMOVED);
826 d->state |= (1<<MD_DISK_FAULTY);
830 sb->nr_disks = nr_disks;
831 sb->active_disks = active;
832 sb->working_disks = working;
833 sb->failed_disks = failed;
834 sb->spare_disks = spare;
836 sb->this_disk = sb->disks[rdev->desc_nr];
837 sb->sb_csum = calc_sb_csum(sb);
841 * version 1 superblock
844 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
846 unsigned int disk_csum, csum;
847 unsigned long long newcsum;
848 int size = 256 + le32_to_cpu(sb->max_dev)*2;
849 unsigned int *isuper = (unsigned int*)sb;
852 disk_csum = sb->sb_csum;
855 for (i=0; size>=4; size -= 4 )
856 newcsum += le32_to_cpu(*isuper++);
859 newcsum += le16_to_cpu(*(unsigned short*) isuper);
861 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
862 sb->sb_csum = disk_csum;
863 return cpu_to_le32(csum);
866 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
868 struct mdp_superblock_1 *sb;
871 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
875 * Calculate the position of the superblock.
876 * It is always aligned to a 4K boundary and
877 * depeding on minor_version, it can be:
878 * 0: At least 8K, but less than 12K, from end of device
879 * 1: At start of device
880 * 2: 4K from start of device.
882 switch(minor_version) {
884 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
886 sb_offset &= ~(sector_t)(4*2-1);
887 /* convert from sectors to K */
899 rdev->sb_offset = sb_offset;
901 /* superblock is rarely larger than 1K, but it can be larger,
902 * and it is safe to read 4k, so we do that
904 ret = read_disk_sb(rdev, 4096);
908 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
910 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
911 sb->major_version != cpu_to_le32(1) ||
912 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
913 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
914 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
917 if (calc_sb_1_csum(sb) != sb->sb_csum) {
918 printk("md: invalid superblock checksum on %s\n",
919 bdevname(rdev->bdev,b));
922 if (le64_to_cpu(sb->data_size) < 10) {
923 printk("md: data_size too small on %s\n",
924 bdevname(rdev->bdev,b));
927 rdev->preferred_minor = 0xffff;
928 rdev->data_offset = le64_to_cpu(sb->data_offset);
930 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
931 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
932 if (rdev->sb_size & bmask)
933 rdev-> sb_size = (rdev->sb_size | bmask)+1;
939 struct mdp_superblock_1 *refsb =
940 (struct mdp_superblock_1*)page_address(refdev->sb_page);
942 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
943 sb->level != refsb->level ||
944 sb->layout != refsb->layout ||
945 sb->chunksize != refsb->chunksize) {
946 printk(KERN_WARNING "md: %s has strangely different"
947 " superblock to %s\n",
948 bdevname(rdev->bdev,b),
949 bdevname(refdev->bdev,b2));
952 ev1 = le64_to_cpu(sb->events);
953 ev2 = le64_to_cpu(refsb->events);
959 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
961 rdev->size = rdev->sb_offset;
962 if (rdev->size < le64_to_cpu(sb->data_size)/2)
964 rdev->size = le64_to_cpu(sb->data_size)/2;
965 if (le32_to_cpu(sb->chunksize))
966 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
970 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
972 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
974 rdev->raid_disk = -1;
976 if (mddev->raid_disks == 0) {
977 mddev->major_version = 1;
978 mddev->patch_version = 0;
979 mddev->persistent = 1;
980 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
981 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
982 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
983 mddev->level = le32_to_cpu(sb->level);
984 mddev->layout = le32_to_cpu(sb->layout);
985 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
986 mddev->size = le64_to_cpu(sb->size)/2;
987 mddev->events = le64_to_cpu(sb->events);
988 mddev->bitmap_offset = 0;
989 mddev->default_bitmap_offset = 0;
990 mddev->default_bitmap_offset = 1024;
992 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
993 memcpy(mddev->uuid, sb->set_uuid, 16);
995 mddev->max_disks = (4096-256)/2;
997 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
998 mddev->bitmap_file == NULL ) {
999 if (mddev->level != 1) {
1000 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1003 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1005 } else if (mddev->pers == NULL) {
1006 /* Insist of good event counter while assembling */
1007 __u64 ev1 = le64_to_cpu(sb->events);
1009 if (ev1 < mddev->events)
1011 } else if (mddev->bitmap) {
1012 /* If adding to array with a bitmap, then we can accept an
1013 * older device, but not too old.
1015 __u64 ev1 = le64_to_cpu(sb->events);
1016 if (ev1 < mddev->bitmap->events_cleared)
1018 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1021 if (mddev->level != LEVEL_MULTIPATH) {
1023 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1024 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1026 case 0xffff: /* spare */
1029 case 0xfffe: /* faulty */
1035 rdev->raid_disk = role;
1039 if (sb->devflags & WriteMostly1)
1040 set_bit(WriteMostly, &rdev->flags);
1041 } else /* MULTIPATH are always insync */
1047 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1049 struct mdp_superblock_1 *sb;
1050 struct list_head *tmp;
1053 /* make rdev->sb match mddev and rdev data. */
1055 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1057 sb->feature_map = 0;
1059 memset(sb->pad1, 0, sizeof(sb->pad1));
1060 memset(sb->pad2, 0, sizeof(sb->pad2));
1061 memset(sb->pad3, 0, sizeof(sb->pad3));
1063 sb->utime = cpu_to_le64((__u64)mddev->utime);
1064 sb->events = cpu_to_le64(mddev->events);
1066 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1068 sb->resync_offset = cpu_to_le64(0);
1070 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1071 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1072 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1076 ITERATE_RDEV(mddev,rdev2,tmp)
1077 if (rdev2->desc_nr+1 > max_dev)
1078 max_dev = rdev2->desc_nr+1;
1080 sb->max_dev = cpu_to_le32(max_dev);
1081 for (i=0; i<max_dev;i++)
1082 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1084 ITERATE_RDEV(mddev,rdev2,tmp) {
1087 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1088 else if (rdev2->in_sync)
1089 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1091 sb->dev_roles[i] = cpu_to_le16(0xffff);
1094 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1095 sb->sb_csum = calc_sb_1_csum(sb);
1099 static struct super_type super_types[] = {
1102 .owner = THIS_MODULE,
1103 .load_super = super_90_load,
1104 .validate_super = super_90_validate,
1105 .sync_super = super_90_sync,
1109 .owner = THIS_MODULE,
1110 .load_super = super_1_load,
1111 .validate_super = super_1_validate,
1112 .sync_super = super_1_sync,
1116 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1118 struct list_head *tmp;
1121 ITERATE_RDEV(mddev,rdev,tmp)
1122 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1128 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1130 struct list_head *tmp;
1133 ITERATE_RDEV(mddev1,rdev,tmp)
1134 if (match_dev_unit(mddev2, rdev))
1140 static LIST_HEAD(pending_raid_disks);
1142 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1144 mdk_rdev_t *same_pdev;
1145 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1151 same_pdev = match_dev_unit(mddev, rdev);
1154 "%s: WARNING: %s appears to be on the same physical"
1155 " disk as %s. True\n protection against single-disk"
1156 " failure might be compromised.\n",
1157 mdname(mddev), bdevname(rdev->bdev,b),
1158 bdevname(same_pdev->bdev,b2));
1160 /* Verify rdev->desc_nr is unique.
1161 * If it is -1, assign a free number, else
1162 * check number is not in use
1164 if (rdev->desc_nr < 0) {
1166 if (mddev->pers) choice = mddev->raid_disks;
1167 while (find_rdev_nr(mddev, choice))
1169 rdev->desc_nr = choice;
1171 if (find_rdev_nr(mddev, rdev->desc_nr))
1175 list_add(&rdev->same_set, &mddev->disks);
1176 rdev->mddev = mddev;
1177 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1179 rdev->kobj.k_name = NULL;
1180 snprintf(rdev->kobj.name, KOBJ_NAME_LEN, "dev%d", rdev->desc_nr);
1181 rdev->kobj.parent = kobject_get(&mddev->kobj);
1182 kobject_add(&rdev->kobj);
1184 sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1188 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1190 char b[BDEVNAME_SIZE];
1195 list_del_init(&rdev->same_set);
1196 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1198 sysfs_remove_link(&rdev->kobj, "block");
1199 kobject_del(&rdev->kobj);
1203 * prevent the device from being mounted, repartitioned or
1204 * otherwise reused by a RAID array (or any other kernel
1205 * subsystem), by bd_claiming the device.
1207 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1210 struct block_device *bdev;
1211 char b[BDEVNAME_SIZE];
1213 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1215 printk(KERN_ERR "md: could not open %s.\n",
1216 __bdevname(dev, b));
1217 return PTR_ERR(bdev);
1219 err = bd_claim(bdev, rdev);
1221 printk(KERN_ERR "md: could not bd_claim %s.\n",
1230 static void unlock_rdev(mdk_rdev_t *rdev)
1232 struct block_device *bdev = rdev->bdev;
1240 void md_autodetect_dev(dev_t dev);
1242 static void export_rdev(mdk_rdev_t * rdev)
1244 char b[BDEVNAME_SIZE];
1245 printk(KERN_INFO "md: export_rdev(%s)\n",
1246 bdevname(rdev->bdev,b));
1250 list_del_init(&rdev->same_set);
1252 md_autodetect_dev(rdev->bdev->bd_dev);
1255 kobject_put(&rdev->kobj);
1258 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1260 unbind_rdev_from_array(rdev);
1264 static void export_array(mddev_t *mddev)
1266 struct list_head *tmp;
1269 ITERATE_RDEV(mddev,rdev,tmp) {
1274 kick_rdev_from_array(rdev);
1276 if (!list_empty(&mddev->disks))
1278 mddev->raid_disks = 0;
1279 mddev->major_version = 0;
1282 static void print_desc(mdp_disk_t *desc)
1284 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1285 desc->major,desc->minor,desc->raid_disk,desc->state);
1288 static void print_sb(mdp_super_t *sb)
1293 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1294 sb->major_version, sb->minor_version, sb->patch_version,
1295 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1297 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1298 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1299 sb->md_minor, sb->layout, sb->chunk_size);
1300 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1301 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1302 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1303 sb->failed_disks, sb->spare_disks,
1304 sb->sb_csum, (unsigned long)sb->events_lo);
1307 for (i = 0; i < MD_SB_DISKS; i++) {
1310 desc = sb->disks + i;
1311 if (desc->number || desc->major || desc->minor ||
1312 desc->raid_disk || (desc->state && (desc->state != 4))) {
1313 printk(" D %2d: ", i);
1317 printk(KERN_INFO "md: THIS: ");
1318 print_desc(&sb->this_disk);
1322 static void print_rdev(mdk_rdev_t *rdev)
1324 char b[BDEVNAME_SIZE];
1325 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1326 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1327 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1328 if (rdev->sb_loaded) {
1329 printk(KERN_INFO "md: rdev superblock:\n");
1330 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1332 printk(KERN_INFO "md: no rdev superblock!\n");
1335 void md_print_devices(void)
1337 struct list_head *tmp, *tmp2;
1340 char b[BDEVNAME_SIZE];
1343 printk("md: **********************************\n");
1344 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1345 printk("md: **********************************\n");
1346 ITERATE_MDDEV(mddev,tmp) {
1349 bitmap_print_sb(mddev->bitmap);
1351 printk("%s: ", mdname(mddev));
1352 ITERATE_RDEV(mddev,rdev,tmp2)
1353 printk("<%s>", bdevname(rdev->bdev,b));
1356 ITERATE_RDEV(mddev,rdev,tmp2)
1359 printk("md: **********************************\n");
1364 static void sync_sbs(mddev_t * mddev)
1367 struct list_head *tmp;
1369 ITERATE_RDEV(mddev,rdev,tmp) {
1370 super_types[mddev->major_version].
1371 sync_super(mddev, rdev);
1372 rdev->sb_loaded = 1;
1376 static void md_update_sb(mddev_t * mddev)
1379 struct list_head *tmp;
1384 spin_lock(&mddev->write_lock);
1385 sync_req = mddev->in_sync;
1386 mddev->utime = get_seconds();
1389 if (!mddev->events) {
1391 * oops, this 64-bit counter should never wrap.
1392 * Either we are in around ~1 trillion A.C., assuming
1393 * 1 reboot per second, or we have a bug:
1398 mddev->sb_dirty = 2;
1402 * do not write anything to disk if using
1403 * nonpersistent superblocks
1405 if (!mddev->persistent) {
1406 mddev->sb_dirty = 0;
1407 spin_unlock(&mddev->write_lock);
1408 wake_up(&mddev->sb_wait);
1411 spin_unlock(&mddev->write_lock);
1414 "md: updating %s RAID superblock on device (in sync %d)\n",
1415 mdname(mddev),mddev->in_sync);
1417 err = bitmap_update_sb(mddev->bitmap);
1418 ITERATE_RDEV(mddev,rdev,tmp) {
1419 char b[BDEVNAME_SIZE];
1420 dprintk(KERN_INFO "md: ");
1422 dprintk("(skipping faulty ");
1424 dprintk("%s ", bdevname(rdev->bdev,b));
1425 if (!rdev->faulty) {
1426 md_super_write(mddev,rdev,
1427 rdev->sb_offset<<1, rdev->sb_size,
1429 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1430 bdevname(rdev->bdev,b),
1431 (unsigned long long)rdev->sb_offset);
1435 if (mddev->level == LEVEL_MULTIPATH)
1436 /* only need to write one superblock... */
1439 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1440 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1442 spin_lock(&mddev->write_lock);
1443 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1444 /* have to write it out again */
1445 spin_unlock(&mddev->write_lock);
1448 mddev->sb_dirty = 0;
1449 spin_unlock(&mddev->write_lock);
1450 wake_up(&mddev->sb_wait);
1454 struct rdev_sysfs_entry {
1455 struct attribute attr;
1456 ssize_t (*show)(mdk_rdev_t *, char *);
1457 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1461 rdev_show_state(mdk_rdev_t *rdev, char *page)
1467 len+= sprintf(page+len, "%sfaulty",sep);
1470 if (rdev->in_sync) {
1471 len += sprintf(page+len, "%sin_sync",sep);
1474 if (!rdev->faulty && !rdev->in_sync) {
1475 len += sprintf(page+len, "%sspare", sep);
1478 return len+sprintf(page+len, "\n");
1481 static struct rdev_sysfs_entry rdev_state = {
1482 .attr = {.name = "state", .mode = S_IRUGO },
1483 .show = rdev_show_state,
1487 rdev_show_super(mdk_rdev_t *rdev, char *page)
1489 if (rdev->sb_loaded && rdev->sb_size) {
1490 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1491 return rdev->sb_size;
1495 static struct rdev_sysfs_entry rdev_super = {
1496 .attr = {.name = "super", .mode = S_IRUGO },
1497 .show = rdev_show_super,
1499 static struct attribute *rdev_default_attrs[] = {
1505 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1507 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1508 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1512 return entry->show(rdev, page);
1516 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1517 const char *page, size_t length)
1519 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1520 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1524 return entry->store(rdev, page, length);
1527 static void rdev_free(struct kobject *ko)
1529 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1532 static struct sysfs_ops rdev_sysfs_ops = {
1533 .show = rdev_attr_show,
1534 .store = rdev_attr_store,
1536 static struct kobj_type rdev_ktype = {
1537 .release = rdev_free,
1538 .sysfs_ops = &rdev_sysfs_ops,
1539 .default_attrs = rdev_default_attrs,
1543 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1545 * mark the device faulty if:
1547 * - the device is nonexistent (zero size)
1548 * - the device has no valid superblock
1550 * a faulty rdev _never_ has rdev->sb set.
1552 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1554 char b[BDEVNAME_SIZE];
1559 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1561 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1562 return ERR_PTR(-ENOMEM);
1564 memset(rdev, 0, sizeof(*rdev));
1566 if ((err = alloc_disk_sb(rdev)))
1569 err = lock_rdev(rdev, newdev);
1573 rdev->kobj.parent = NULL;
1574 rdev->kobj.ktype = &rdev_ktype;
1575 kobject_init(&rdev->kobj);
1580 rdev->data_offset = 0;
1581 atomic_set(&rdev->nr_pending, 0);
1583 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1586 "md: %s has zero or unknown size, marking faulty!\n",
1587 bdevname(rdev->bdev,b));
1592 if (super_format >= 0) {
1593 err = super_types[super_format].
1594 load_super(rdev, NULL, super_minor);
1595 if (err == -EINVAL) {
1597 "md: %s has invalid sb, not importing!\n",
1598 bdevname(rdev->bdev,b));
1603 "md: could not read %s's sb, not importing!\n",
1604 bdevname(rdev->bdev,b));
1608 INIT_LIST_HEAD(&rdev->same_set);
1613 if (rdev->sb_page) {
1619 return ERR_PTR(err);
1623 * Check a full RAID array for plausibility
1627 static void analyze_sbs(mddev_t * mddev)
1630 struct list_head *tmp;
1631 mdk_rdev_t *rdev, *freshest;
1632 char b[BDEVNAME_SIZE];
1635 ITERATE_RDEV(mddev,rdev,tmp)
1636 switch (super_types[mddev->major_version].
1637 load_super(rdev, freshest, mddev->minor_version)) {
1645 "md: fatal superblock inconsistency in %s"
1646 " -- removing from array\n",
1647 bdevname(rdev->bdev,b));
1648 kick_rdev_from_array(rdev);
1652 super_types[mddev->major_version].
1653 validate_super(mddev, freshest);
1656 ITERATE_RDEV(mddev,rdev,tmp) {
1657 if (rdev != freshest)
1658 if (super_types[mddev->major_version].
1659 validate_super(mddev, rdev)) {
1660 printk(KERN_WARNING "md: kicking non-fresh %s"
1662 bdevname(rdev->bdev,b));
1663 kick_rdev_from_array(rdev);
1666 if (mddev->level == LEVEL_MULTIPATH) {
1667 rdev->desc_nr = i++;
1668 rdev->raid_disk = rdev->desc_nr;
1675 if (mddev->recovery_cp != MaxSector &&
1677 printk(KERN_ERR "md: %s: raid array is not clean"
1678 " -- starting background reconstruction\n",
1683 struct md_sysfs_entry {
1684 struct attribute attr;
1685 ssize_t (*show)(mddev_t *, char *);
1686 ssize_t (*store)(mddev_t *, const char *, size_t);
1690 md_show_level(mddev_t *mddev, char *page)
1692 mdk_personality_t *p = mddev->pers;
1695 if (mddev->level >= 0)
1696 return sprintf(page, "RAID-%d\n", mddev->level);
1698 return sprintf(page, "%s\n", p->name);
1701 static struct md_sysfs_entry md_level = {
1702 .attr = {.name = "level", .mode = S_IRUGO },
1703 .show = md_show_level,
1707 md_show_rdisks(mddev_t *mddev, char *page)
1709 return sprintf(page, "%d\n", mddev->raid_disks);
1712 static struct md_sysfs_entry md_raid_disks = {
1713 .attr = {.name = "raid_disks", .mode = S_IRUGO },
1714 .show = md_show_rdisks,
1718 md_show_scan(mddev_t *mddev, char *page)
1720 char *type = "none";
1721 if (mddev->recovery &
1722 ((1<<MD_RECOVERY_RUNNING) || (1<<MD_RECOVERY_NEEDED))) {
1723 if (mddev->recovery & (1<<MD_RECOVERY_SYNC)) {
1724 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1726 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1733 return sprintf(page, "%s\n", type);
1737 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1740 if (mddev->recovery &
1741 ((1<<MD_RECOVERY_RUNNING) || (1<<MD_RECOVERY_NEEDED)))
1743 down(&mddev->reconfig_sem);
1744 if (mddev->pers && mddev->pers->sync_request)
1746 up(&mddev->reconfig_sem);
1750 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1751 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1752 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1754 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1755 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1756 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1757 md_wakeup_thread(mddev->thread);
1762 md_show_mismatch(mddev_t *mddev, char *page)
1764 return sprintf(page, "%llu\n",
1765 (unsigned long long) mddev->resync_mismatches);
1768 static struct md_sysfs_entry md_scan_mode = {
1769 .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1770 .show = md_show_scan,
1771 .store = md_store_scan,
1774 static struct md_sysfs_entry md_mismatches = {
1775 .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1776 .show = md_show_mismatch,
1779 static struct attribute *md_default_attrs[] = {
1781 &md_raid_disks.attr,
1783 &md_mismatches.attr,
1788 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1790 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1791 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1795 return entry->show(mddev, page);
1799 md_attr_store(struct kobject *kobj, struct attribute *attr,
1800 const char *page, size_t length)
1802 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1803 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1807 return entry->store(mddev, page, length);
1810 static void md_free(struct kobject *ko)
1812 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1816 static struct sysfs_ops md_sysfs_ops = {
1817 .show = md_attr_show,
1818 .store = md_attr_store,
1820 static struct kobj_type md_ktype = {
1822 .sysfs_ops = &md_sysfs_ops,
1823 .default_attrs = md_default_attrs,
1828 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1830 static DECLARE_MUTEX(disks_sem);
1831 mddev_t *mddev = mddev_find(dev);
1832 struct gendisk *disk;
1833 int partitioned = (MAJOR(dev) != MD_MAJOR);
1834 int shift = partitioned ? MdpMinorShift : 0;
1835 int unit = MINOR(dev) >> shift;
1841 if (mddev->gendisk) {
1846 disk = alloc_disk(1 << shift);
1852 disk->major = MAJOR(dev);
1853 disk->first_minor = unit << shift;
1855 sprintf(disk->disk_name, "md_d%d", unit);
1856 sprintf(disk->devfs_name, "md/d%d", unit);
1858 sprintf(disk->disk_name, "md%d", unit);
1859 sprintf(disk->devfs_name, "md/%d", unit);
1861 disk->fops = &md_fops;
1862 disk->private_data = mddev;
1863 disk->queue = mddev->queue;
1865 mddev->gendisk = disk;
1867 mddev->kobj.parent = kobject_get(&disk->kobj);
1868 mddev->kobj.k_name = NULL;
1869 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1870 mddev->kobj.ktype = &md_ktype;
1871 kobject_register(&mddev->kobj);
1875 void md_wakeup_thread(mdk_thread_t *thread);
1877 static void md_safemode_timeout(unsigned long data)
1879 mddev_t *mddev = (mddev_t *) data;
1881 mddev->safemode = 1;
1882 md_wakeup_thread(mddev->thread);
1886 static int do_md_run(mddev_t * mddev)
1890 struct list_head *tmp;
1892 struct gendisk *disk;
1893 char b[BDEVNAME_SIZE];
1895 if (list_empty(&mddev->disks))
1896 /* cannot run an array with no devices.. */
1903 * Analyze all RAID superblock(s)
1905 if (!mddev->raid_disks)
1908 chunk_size = mddev->chunk_size;
1909 pnum = level_to_pers(mddev->level);
1911 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1914 * 'default chunksize' in the old md code used to
1915 * be PAGE_SIZE, baaad.
1916 * we abort here to be on the safe side. We don't
1917 * want to continue the bad practice.
1920 "no chunksize specified, see 'man raidtab'\n");
1923 if (chunk_size > MAX_CHUNK_SIZE) {
1924 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1925 chunk_size, MAX_CHUNK_SIZE);
1929 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1931 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1932 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1935 if (chunk_size < PAGE_SIZE) {
1936 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1937 chunk_size, PAGE_SIZE);
1941 /* devices must have minimum size of one chunk */
1942 ITERATE_RDEV(mddev,rdev,tmp) {
1945 if (rdev->size < chunk_size / 1024) {
1947 "md: Dev %s smaller than chunk_size:"
1949 bdevname(rdev->bdev,b),
1950 (unsigned long long)rdev->size,
1960 request_module("md-personality-%d", pnum);
1965 * Drop all container device buffers, from now on
1966 * the only valid external interface is through the md
1968 * Also find largest hardsector size
1970 ITERATE_RDEV(mddev,rdev,tmp) {
1973 sync_blockdev(rdev->bdev);
1974 invalidate_bdev(rdev->bdev, 0);
1977 md_probe(mddev->unit, NULL, NULL);
1978 disk = mddev->gendisk;
1982 spin_lock(&pers_lock);
1983 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1984 spin_unlock(&pers_lock);
1985 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1990 mddev->pers = pers[pnum];
1991 spin_unlock(&pers_lock);
1993 mddev->recovery = 0;
1994 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1996 /* before we start the array running, initialise the bitmap */
1997 err = bitmap_create(mddev);
1999 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2000 mdname(mddev), err);
2002 err = mddev->pers->run(mddev);
2004 printk(KERN_ERR "md: pers->run() failed ...\n");
2005 module_put(mddev->pers->owner);
2007 bitmap_destroy(mddev);
2010 atomic_set(&mddev->writes_pending,0);
2011 mddev->safemode = 0;
2012 mddev->safemode_timer.function = md_safemode_timeout;
2013 mddev->safemode_timer.data = (unsigned long) mddev;
2014 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2017 ITERATE_RDEV(mddev,rdev,tmp)
2018 if (rdev->raid_disk >= 0) {
2020 sprintf(nm, "rd%d", rdev->raid_disk);
2021 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2024 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2025 md_wakeup_thread(mddev->thread);
2027 if (mddev->sb_dirty)
2028 md_update_sb(mddev);
2030 set_capacity(disk, mddev->array_size<<1);
2032 /* If we call blk_queue_make_request here, it will
2033 * re-initialise max_sectors etc which may have been
2034 * refined inside -> run. So just set the bits we need to set.
2035 * Most initialisation happended when we called
2036 * blk_queue_make_request(..., md_fail_request)
2039 mddev->queue->queuedata = mddev;
2040 mddev->queue->make_request_fn = mddev->pers->make_request;
2046 static int restart_array(mddev_t *mddev)
2048 struct gendisk *disk = mddev->gendisk;
2052 * Complain if it has no devices
2055 if (list_empty(&mddev->disks))
2063 mddev->safemode = 0;
2065 set_disk_ro(disk, 0);
2067 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2070 * Kick recovery or resync if necessary
2072 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2073 md_wakeup_thread(mddev->thread);
2076 printk(KERN_ERR "md: %s has no personality assigned.\n",
2085 static int do_md_stop(mddev_t * mddev, int ro)
2088 struct gendisk *disk = mddev->gendisk;
2091 if (atomic_read(&mddev->active)>2) {
2092 printk("md: %s still in use.\n",mdname(mddev));
2096 if (mddev->sync_thread) {
2097 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2098 md_unregister_thread(mddev->sync_thread);
2099 mddev->sync_thread = NULL;
2102 del_timer_sync(&mddev->safemode_timer);
2104 invalidate_partition(disk, 0);
2112 bitmap_flush(mddev);
2113 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2115 set_disk_ro(disk, 0);
2116 blk_queue_make_request(mddev->queue, md_fail_request);
2117 mddev->pers->stop(mddev);
2118 module_put(mddev->pers->owner);
2123 if (!mddev->in_sync) {
2124 /* mark array as shutdown cleanly */
2126 md_update_sb(mddev);
2129 set_disk_ro(disk, 1);
2132 bitmap_destroy(mddev);
2133 if (mddev->bitmap_file) {
2134 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2135 fput(mddev->bitmap_file);
2136 mddev->bitmap_file = NULL;
2138 mddev->bitmap_offset = 0;
2141 * Free resources if final stop
2145 struct list_head *tmp;
2146 struct gendisk *disk;
2147 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2149 ITERATE_RDEV(mddev,rdev,tmp)
2150 if (rdev->raid_disk >= 0) {
2152 sprintf(nm, "rd%d", rdev->raid_disk);
2153 sysfs_remove_link(&mddev->kobj, nm);
2156 export_array(mddev);
2158 mddev->array_size = 0;
2159 disk = mddev->gendisk;
2161 set_capacity(disk, 0);
2164 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2171 static void autorun_array(mddev_t *mddev)
2174 struct list_head *tmp;
2177 if (list_empty(&mddev->disks))
2180 printk(KERN_INFO "md: running: ");
2182 ITERATE_RDEV(mddev,rdev,tmp) {
2183 char b[BDEVNAME_SIZE];
2184 printk("<%s>", bdevname(rdev->bdev,b));
2188 err = do_md_run (mddev);
2190 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2191 do_md_stop (mddev, 0);
2196 * lets try to run arrays based on all disks that have arrived
2197 * until now. (those are in pending_raid_disks)
2199 * the method: pick the first pending disk, collect all disks with
2200 * the same UUID, remove all from the pending list and put them into
2201 * the 'same_array' list. Then order this list based on superblock
2202 * update time (freshest comes first), kick out 'old' disks and
2203 * compare superblocks. If everything's fine then run it.
2205 * If "unit" is allocated, then bump its reference count
2207 static void autorun_devices(int part)
2209 struct list_head candidates;
2210 struct list_head *tmp;
2211 mdk_rdev_t *rdev0, *rdev;
2213 char b[BDEVNAME_SIZE];
2215 printk(KERN_INFO "md: autorun ...\n");
2216 while (!list_empty(&pending_raid_disks)) {
2218 rdev0 = list_entry(pending_raid_disks.next,
2219 mdk_rdev_t, same_set);
2221 printk(KERN_INFO "md: considering %s ...\n",
2222 bdevname(rdev0->bdev,b));
2223 INIT_LIST_HEAD(&candidates);
2224 ITERATE_RDEV_PENDING(rdev,tmp)
2225 if (super_90_load(rdev, rdev0, 0) >= 0) {
2226 printk(KERN_INFO "md: adding %s ...\n",
2227 bdevname(rdev->bdev,b));
2228 list_move(&rdev->same_set, &candidates);
2231 * now we have a set of devices, with all of them having
2232 * mostly sane superblocks. It's time to allocate the
2235 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2236 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2237 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2241 dev = MKDEV(mdp_major,
2242 rdev0->preferred_minor << MdpMinorShift);
2244 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2246 md_probe(dev, NULL, NULL);
2247 mddev = mddev_find(dev);
2250 "md: cannot allocate memory for md drive.\n");
2253 if (mddev_lock(mddev))
2254 printk(KERN_WARNING "md: %s locked, cannot run\n",
2256 else if (mddev->raid_disks || mddev->major_version
2257 || !list_empty(&mddev->disks)) {
2259 "md: %s already running, cannot run %s\n",
2260 mdname(mddev), bdevname(rdev0->bdev,b));
2261 mddev_unlock(mddev);
2263 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2264 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2265 list_del_init(&rdev->same_set);
2266 if (bind_rdev_to_array(rdev, mddev))
2269 autorun_array(mddev);
2270 mddev_unlock(mddev);
2272 /* on success, candidates will be empty, on error
2275 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2279 printk(KERN_INFO "md: ... autorun DONE.\n");
2283 * import RAID devices based on one partition
2284 * if possible, the array gets run as well.
2287 static int autostart_array(dev_t startdev)
2289 char b[BDEVNAME_SIZE];
2290 int err = -EINVAL, i;
2291 mdp_super_t *sb = NULL;
2292 mdk_rdev_t *start_rdev = NULL, *rdev;
2294 start_rdev = md_import_device(startdev, 0, 0);
2295 if (IS_ERR(start_rdev))
2299 /* NOTE: this can only work for 0.90.0 superblocks */
2300 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2301 if (sb->major_version != 0 ||
2302 sb->minor_version != 90 ) {
2303 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2304 export_rdev(start_rdev);
2308 if (start_rdev->faulty) {
2310 "md: can not autostart based on faulty %s!\n",
2311 bdevname(start_rdev->bdev,b));
2312 export_rdev(start_rdev);
2315 list_add(&start_rdev->same_set, &pending_raid_disks);
2317 for (i = 0; i < MD_SB_DISKS; i++) {
2318 mdp_disk_t *desc = sb->disks + i;
2319 dev_t dev = MKDEV(desc->major, desc->minor);
2323 if (dev == startdev)
2325 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2327 rdev = md_import_device(dev, 0, 0);
2331 list_add(&rdev->same_set, &pending_raid_disks);
2335 * possibly return codes
2343 static int get_version(void __user * arg)
2347 ver.major = MD_MAJOR_VERSION;
2348 ver.minor = MD_MINOR_VERSION;
2349 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2351 if (copy_to_user(arg, &ver, sizeof(ver)))
2357 static int get_array_info(mddev_t * mddev, void __user * arg)
2359 mdu_array_info_t info;
2360 int nr,working,active,failed,spare;
2362 struct list_head *tmp;
2364 nr=working=active=failed=spare=0;
2365 ITERATE_RDEV(mddev,rdev,tmp) {
2378 info.major_version = mddev->major_version;
2379 info.minor_version = mddev->minor_version;
2380 info.patch_version = MD_PATCHLEVEL_VERSION;
2381 info.ctime = mddev->ctime;
2382 info.level = mddev->level;
2383 info.size = mddev->size;
2385 info.raid_disks = mddev->raid_disks;
2386 info.md_minor = mddev->md_minor;
2387 info.not_persistent= !mddev->persistent;
2389 info.utime = mddev->utime;
2392 info.state = (1<<MD_SB_CLEAN);
2393 if (mddev->bitmap && mddev->bitmap_offset)
2394 info.state = (1<<MD_SB_BITMAP_PRESENT);
2395 info.active_disks = active;
2396 info.working_disks = working;
2397 info.failed_disks = failed;
2398 info.spare_disks = spare;
2400 info.layout = mddev->layout;
2401 info.chunk_size = mddev->chunk_size;
2403 if (copy_to_user(arg, &info, sizeof(info)))
2409 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2411 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2412 char *ptr, *buf = NULL;
2415 file = kmalloc(sizeof(*file), GFP_KERNEL);
2419 /* bitmap disabled, zero the first byte and copy out */
2420 if (!mddev->bitmap || !mddev->bitmap->file) {
2421 file->pathname[0] = '\0';
2425 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2429 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2433 strcpy(file->pathname, ptr);
2437 if (copy_to_user(arg, file, sizeof(*file)))
2445 static int get_disk_info(mddev_t * mddev, void __user * arg)
2447 mdu_disk_info_t info;
2451 if (copy_from_user(&info, arg, sizeof(info)))
2456 rdev = find_rdev_nr(mddev, nr);
2458 info.major = MAJOR(rdev->bdev->bd_dev);
2459 info.minor = MINOR(rdev->bdev->bd_dev);
2460 info.raid_disk = rdev->raid_disk;
2463 info.state |= (1<<MD_DISK_FAULTY);
2464 else if (rdev->in_sync) {
2465 info.state |= (1<<MD_DISK_ACTIVE);
2466 info.state |= (1<<MD_DISK_SYNC);
2468 if (test_bit(WriteMostly, &rdev->flags))
2469 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2471 info.major = info.minor = 0;
2472 info.raid_disk = -1;
2473 info.state = (1<<MD_DISK_REMOVED);
2476 if (copy_to_user(arg, &info, sizeof(info)))
2482 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2484 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2486 dev_t dev = MKDEV(info->major,info->minor);
2488 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2491 if (!mddev->raid_disks) {
2493 /* expecting a device which has a superblock */
2494 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2497 "md: md_import_device returned %ld\n",
2499 return PTR_ERR(rdev);
2501 if (!list_empty(&mddev->disks)) {
2502 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2503 mdk_rdev_t, same_set);
2504 int err = super_types[mddev->major_version]
2505 .load_super(rdev, rdev0, mddev->minor_version);
2508 "md: %s has different UUID to %s\n",
2509 bdevname(rdev->bdev,b),
2510 bdevname(rdev0->bdev,b2));
2515 err = bind_rdev_to_array(rdev, mddev);
2522 * add_new_disk can be used once the array is assembled
2523 * to add "hot spares". They must already have a superblock
2528 if (!mddev->pers->hot_add_disk) {
2530 "%s: personality does not support diskops!\n",
2534 if (mddev->persistent)
2535 rdev = md_import_device(dev, mddev->major_version,
2536 mddev->minor_version);
2538 rdev = md_import_device(dev, -1, -1);
2541 "md: md_import_device returned %ld\n",
2543 return PTR_ERR(rdev);
2545 /* set save_raid_disk if appropriate */
2546 if (!mddev->persistent) {
2547 if (info->state & (1<<MD_DISK_SYNC) &&
2548 info->raid_disk < mddev->raid_disks)
2549 rdev->raid_disk = info->raid_disk;
2551 rdev->raid_disk = -1;
2553 super_types[mddev->major_version].
2554 validate_super(mddev, rdev);
2555 rdev->saved_raid_disk = rdev->raid_disk;
2557 rdev->in_sync = 0; /* just to be sure */
2558 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2559 set_bit(WriteMostly, &rdev->flags);
2561 rdev->raid_disk = -1;
2562 err = bind_rdev_to_array(rdev, mddev);
2566 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2567 md_wakeup_thread(mddev->thread);
2571 /* otherwise, add_new_disk is only allowed
2572 * for major_version==0 superblocks
2574 if (mddev->major_version != 0) {
2575 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2580 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2582 rdev = md_import_device (dev, -1, 0);
2585 "md: error, md_import_device() returned %ld\n",
2587 return PTR_ERR(rdev);
2589 rdev->desc_nr = info->number;
2590 if (info->raid_disk < mddev->raid_disks)
2591 rdev->raid_disk = info->raid_disk;
2593 rdev->raid_disk = -1;
2596 if (rdev->raid_disk < mddev->raid_disks)
2597 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2601 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2602 set_bit(WriteMostly, &rdev->flags);
2604 err = bind_rdev_to_array(rdev, mddev);
2610 if (!mddev->persistent) {
2611 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2612 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2614 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2615 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2617 if (!mddev->size || (mddev->size > rdev->size))
2618 mddev->size = rdev->size;
2624 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2626 char b[BDEVNAME_SIZE];
2632 rdev = find_rdev(mddev, dev);
2636 if (rdev->raid_disk >= 0)
2639 kick_rdev_from_array(rdev);
2640 md_update_sb(mddev);
2644 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2645 bdevname(rdev->bdev,b), mdname(mddev));
2649 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2651 char b[BDEVNAME_SIZE];
2659 if (mddev->major_version != 0) {
2660 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2661 " version-0 superblocks.\n",
2665 if (!mddev->pers->hot_add_disk) {
2667 "%s: personality does not support diskops!\n",
2672 rdev = md_import_device (dev, -1, 0);
2675 "md: error, md_import_device() returned %ld\n",
2680 if (mddev->persistent)
2681 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2684 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2686 size = calc_dev_size(rdev, mddev->chunk_size);
2689 if (size < mddev->size) {
2691 "%s: disk size %llu blocks < array size %llu\n",
2692 mdname(mddev), (unsigned long long)size,
2693 (unsigned long long)mddev->size);
2700 "md: can not hot-add faulty %s disk to %s!\n",
2701 bdevname(rdev->bdev,b), mdname(mddev));
2707 bind_rdev_to_array(rdev, mddev);
2710 * The rest should better be atomic, we can have disk failures
2711 * noticed in interrupt contexts ...
2714 if (rdev->desc_nr == mddev->max_disks) {
2715 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2718 goto abort_unbind_export;
2721 rdev->raid_disk = -1;
2723 md_update_sb(mddev);
2726 * Kick recovery, maybe this spare has to be added to the
2727 * array immediately.
2729 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2730 md_wakeup_thread(mddev->thread);
2734 abort_unbind_export:
2735 unbind_rdev_from_array(rdev);
2742 /* similar to deny_write_access, but accounts for our holding a reference
2743 * to the file ourselves */
2744 static int deny_bitmap_write_access(struct file * file)
2746 struct inode *inode = file->f_mapping->host;
2748 spin_lock(&inode->i_lock);
2749 if (atomic_read(&inode->i_writecount) > 1) {
2750 spin_unlock(&inode->i_lock);
2753 atomic_set(&inode->i_writecount, -1);
2754 spin_unlock(&inode->i_lock);
2759 static int set_bitmap_file(mddev_t *mddev, int fd)
2764 if (!mddev->pers->quiesce)
2766 if (mddev->recovery || mddev->sync_thread)
2768 /* we should be able to change the bitmap.. */
2774 return -EEXIST; /* cannot add when bitmap is present */
2775 mddev->bitmap_file = fget(fd);
2777 if (mddev->bitmap_file == NULL) {
2778 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2783 err = deny_bitmap_write_access(mddev->bitmap_file);
2785 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2787 fput(mddev->bitmap_file);
2788 mddev->bitmap_file = NULL;
2791 mddev->bitmap_offset = 0; /* file overrides offset */
2792 } else if (mddev->bitmap == NULL)
2793 return -ENOENT; /* cannot remove what isn't there */
2796 mddev->pers->quiesce(mddev, 1);
2798 err = bitmap_create(mddev);
2800 bitmap_destroy(mddev);
2801 mddev->pers->quiesce(mddev, 0);
2802 } else if (fd < 0) {
2803 if (mddev->bitmap_file)
2804 fput(mddev->bitmap_file);
2805 mddev->bitmap_file = NULL;
2812 * set_array_info is used two different ways
2813 * The original usage is when creating a new array.
2814 * In this usage, raid_disks is > 0 and it together with
2815 * level, size, not_persistent,layout,chunksize determine the
2816 * shape of the array.
2817 * This will always create an array with a type-0.90.0 superblock.
2818 * The newer usage is when assembling an array.
2819 * In this case raid_disks will be 0, and the major_version field is
2820 * use to determine which style super-blocks are to be found on the devices.
2821 * The minor and patch _version numbers are also kept incase the
2822 * super_block handler wishes to interpret them.
2824 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2827 if (info->raid_disks == 0) {
2828 /* just setting version number for superblock loading */
2829 if (info->major_version < 0 ||
2830 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2831 super_types[info->major_version].name == NULL) {
2832 /* maybe try to auto-load a module? */
2834 "md: superblock version %d not known\n",
2835 info->major_version);
2838 mddev->major_version = info->major_version;
2839 mddev->minor_version = info->minor_version;
2840 mddev->patch_version = info->patch_version;
2843 mddev->major_version = MD_MAJOR_VERSION;
2844 mddev->minor_version = MD_MINOR_VERSION;
2845 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2846 mddev->ctime = get_seconds();
2848 mddev->level = info->level;
2849 mddev->size = info->size;
2850 mddev->raid_disks = info->raid_disks;
2851 /* don't set md_minor, it is determined by which /dev/md* was
2854 if (info->state & (1<<MD_SB_CLEAN))
2855 mddev->recovery_cp = MaxSector;
2857 mddev->recovery_cp = 0;
2858 mddev->persistent = ! info->not_persistent;
2860 mddev->layout = info->layout;
2861 mddev->chunk_size = info->chunk_size;
2863 mddev->max_disks = MD_SB_DISKS;
2865 mddev->sb_dirty = 1;
2868 * Generate a 128 bit UUID
2870 get_random_bytes(mddev->uuid, 16);
2876 * update_array_info is used to change the configuration of an
2878 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2879 * fields in the info are checked against the array.
2880 * Any differences that cannot be handled will cause an error.
2881 * Normally, only one change can be managed at a time.
2883 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2889 /* calculate expected state,ignoring low bits */
2890 if (mddev->bitmap && mddev->bitmap_offset)
2891 state |= (1 << MD_SB_BITMAP_PRESENT);
2893 if (mddev->major_version != info->major_version ||
2894 mddev->minor_version != info->minor_version ||
2895 /* mddev->patch_version != info->patch_version || */
2896 mddev->ctime != info->ctime ||
2897 mddev->level != info->level ||
2898 /* mddev->layout != info->layout || */
2899 !mddev->persistent != info->not_persistent||
2900 mddev->chunk_size != info->chunk_size ||
2901 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2902 ((state^info->state) & 0xfffffe00)
2905 /* Check there is only one change */
2906 if (mddev->size != info->size) cnt++;
2907 if (mddev->raid_disks != info->raid_disks) cnt++;
2908 if (mddev->layout != info->layout) cnt++;
2909 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2910 if (cnt == 0) return 0;
2911 if (cnt > 1) return -EINVAL;
2913 if (mddev->layout != info->layout) {
2915 * we don't need to do anything at the md level, the
2916 * personality will take care of it all.
2918 if (mddev->pers->reconfig == NULL)
2921 return mddev->pers->reconfig(mddev, info->layout, -1);
2923 if (mddev->size != info->size) {
2925 struct list_head *tmp;
2926 if (mddev->pers->resize == NULL)
2928 /* The "size" is the amount of each device that is used.
2929 * This can only make sense for arrays with redundancy.
2930 * linear and raid0 always use whatever space is available
2931 * We can only consider changing the size if no resync
2932 * or reconstruction is happening, and if the new size
2933 * is acceptable. It must fit before the sb_offset or,
2934 * if that is <data_offset, it must fit before the
2935 * size of each device.
2936 * If size is zero, we find the largest size that fits.
2938 if (mddev->sync_thread)
2940 ITERATE_RDEV(mddev,rdev,tmp) {
2942 int fit = (info->size == 0);
2943 if (rdev->sb_offset > rdev->data_offset)
2944 avail = (rdev->sb_offset*2) - rdev->data_offset;
2946 avail = get_capacity(rdev->bdev->bd_disk)
2947 - rdev->data_offset;
2948 if (fit && (info->size == 0 || info->size > avail/2))
2949 info->size = avail/2;
2950 if (avail < ((sector_t)info->size << 1))
2953 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2955 struct block_device *bdev;
2957 bdev = bdget_disk(mddev->gendisk, 0);
2959 down(&bdev->bd_inode->i_sem);
2960 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2961 up(&bdev->bd_inode->i_sem);
2966 if (mddev->raid_disks != info->raid_disks) {
2967 /* change the number of raid disks */
2968 if (mddev->pers->reshape == NULL)
2970 if (info->raid_disks <= 0 ||
2971 info->raid_disks >= mddev->max_disks)
2973 if (mddev->sync_thread)
2975 rv = mddev->pers->reshape(mddev, info->raid_disks);
2977 struct block_device *bdev;
2979 bdev = bdget_disk(mddev->gendisk, 0);
2981 down(&bdev->bd_inode->i_sem);
2982 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2983 up(&bdev->bd_inode->i_sem);
2988 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2989 if (mddev->pers->quiesce == NULL)
2991 if (mddev->recovery || mddev->sync_thread)
2993 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2994 /* add the bitmap */
2997 if (mddev->default_bitmap_offset == 0)
2999 mddev->bitmap_offset = mddev->default_bitmap_offset;
3000 mddev->pers->quiesce(mddev, 1);
3001 rv = bitmap_create(mddev);
3003 bitmap_destroy(mddev);
3004 mddev->pers->quiesce(mddev, 0);
3006 /* remove the bitmap */
3009 if (mddev->bitmap->file)
3011 mddev->pers->quiesce(mddev, 1);
3012 bitmap_destroy(mddev);
3013 mddev->pers->quiesce(mddev, 0);
3014 mddev->bitmap_offset = 0;
3017 md_update_sb(mddev);
3021 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3025 if (mddev->pers == NULL)
3028 rdev = find_rdev(mddev, dev);
3032 md_error(mddev, rdev);
3036 static int md_ioctl(struct inode *inode, struct file *file,
3037 unsigned int cmd, unsigned long arg)
3040 void __user *argp = (void __user *)arg;
3041 struct hd_geometry __user *loc = argp;
3042 mddev_t *mddev = NULL;
3044 if (!capable(CAP_SYS_ADMIN))
3048 * Commands dealing with the RAID driver but not any
3054 err = get_version(argp);
3057 case PRINT_RAID_DEBUG:
3065 autostart_arrays(arg);
3072 * Commands creating/starting a new array:
3075 mddev = inode->i_bdev->bd_disk->private_data;
3083 if (cmd == START_ARRAY) {
3084 /* START_ARRAY doesn't need to lock the array as autostart_array
3085 * does the locking, and it could even be a different array
3090 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3091 "This will not be supported beyond 2.6\n",
3092 current->comm, current->pid);
3095 err = autostart_array(new_decode_dev(arg));
3097 printk(KERN_WARNING "md: autostart failed!\n");
3103 err = mddev_lock(mddev);
3106 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3113 case SET_ARRAY_INFO:
3115 mdu_array_info_t info;
3117 memset(&info, 0, sizeof(info));
3118 else if (copy_from_user(&info, argp, sizeof(info))) {
3123 err = update_array_info(mddev, &info);
3125 printk(KERN_WARNING "md: couldn't update"
3126 " array info. %d\n", err);
3131 if (!list_empty(&mddev->disks)) {
3133 "md: array %s already has disks!\n",
3138 if (mddev->raid_disks) {
3140 "md: array %s already initialised!\n",
3145 err = set_array_info(mddev, &info);
3147 printk(KERN_WARNING "md: couldn't set"
3148 " array info. %d\n", err);
3158 * Commands querying/configuring an existing array:
3160 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3161 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3162 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3163 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3169 * Commands even a read-only array can execute:
3173 case GET_ARRAY_INFO:
3174 err = get_array_info(mddev, argp);
3177 case GET_BITMAP_FILE:
3178 err = get_bitmap_file(mddev, argp);
3182 err = get_disk_info(mddev, argp);
3185 case RESTART_ARRAY_RW:
3186 err = restart_array(mddev);
3190 err = do_md_stop (mddev, 0);
3194 err = do_md_stop (mddev, 1);
3198 * We have a problem here : there is no easy way to give a CHS
3199 * virtual geometry. We currently pretend that we have a 2 heads
3200 * 4 sectors (with a BIG number of cylinders...). This drives
3201 * dosfs just mad... ;-)
3208 err = put_user (2, (char __user *) &loc->heads);
3211 err = put_user (4, (char __user *) &loc->sectors);
3214 err = put_user(get_capacity(mddev->gendisk)/8,
3215 (short __user *) &loc->cylinders);
3218 err = put_user (get_start_sect(inode->i_bdev),
3219 (long __user *) &loc->start);
3224 * The remaining ioctls are changing the state of the
3225 * superblock, so we do not allow read-only arrays
3237 mdu_disk_info_t info;
3238 if (copy_from_user(&info, argp, sizeof(info)))
3241 err = add_new_disk(mddev, &info);
3245 case HOT_REMOVE_DISK:
3246 err = hot_remove_disk(mddev, new_decode_dev(arg));
3250 err = hot_add_disk(mddev, new_decode_dev(arg));
3253 case SET_DISK_FAULTY:
3254 err = set_disk_faulty(mddev, new_decode_dev(arg));
3258 err = do_md_run (mddev);
3261 case SET_BITMAP_FILE:
3262 err = set_bitmap_file(mddev, (int)arg);
3266 if (_IOC_TYPE(cmd) == MD_MAJOR)
3267 printk(KERN_WARNING "md: %s(pid %d) used"
3268 " obsolete MD ioctl, upgrade your"
3269 " software to use new ictls.\n",
3270 current->comm, current->pid);
3277 mddev_unlock(mddev);
3287 static int md_open(struct inode *inode, struct file *file)
3290 * Succeed if we can lock the mddev, which confirms that
3291 * it isn't being stopped right now.
3293 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3296 if ((err = mddev_lock(mddev)))
3301 mddev_unlock(mddev);
3303 check_disk_change(inode->i_bdev);
3308 static int md_release(struct inode *inode, struct file * file)
3310 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3319 static int md_media_changed(struct gendisk *disk)
3321 mddev_t *mddev = disk->private_data;
3323 return mddev->changed;
3326 static int md_revalidate(struct gendisk *disk)
3328 mddev_t *mddev = disk->private_data;
3333 static struct block_device_operations md_fops =
3335 .owner = THIS_MODULE,
3337 .release = md_release,
3339 .media_changed = md_media_changed,
3340 .revalidate_disk= md_revalidate,
3343 static int md_thread(void * arg)
3345 mdk_thread_t *thread = arg;
3348 * md_thread is a 'system-thread', it's priority should be very
3349 * high. We avoid resource deadlocks individually in each
3350 * raid personality. (RAID5 does preallocation) We also use RR and
3351 * the very same RT priority as kswapd, thus we will never get
3352 * into a priority inversion deadlock.
3354 * we definitely have to have equal or higher priority than
3355 * bdflush, otherwise bdflush will deadlock if there are too
3356 * many dirty RAID5 blocks.
3359 allow_signal(SIGKILL);
3360 complete(thread->event);
3361 while (!kthread_should_stop()) {
3362 void (*run)(mddev_t *);
3364 wait_event_interruptible_timeout(thread->wqueue,
3365 test_bit(THREAD_WAKEUP, &thread->flags)
3366 || kthread_should_stop(),
3370 clear_bit(THREAD_WAKEUP, &thread->flags);
3380 void md_wakeup_thread(mdk_thread_t *thread)
3383 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3384 set_bit(THREAD_WAKEUP, &thread->flags);
3385 wake_up(&thread->wqueue);
3389 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3392 mdk_thread_t *thread;
3393 struct completion event;
3395 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3399 memset(thread, 0, sizeof(mdk_thread_t));
3400 init_waitqueue_head(&thread->wqueue);
3402 init_completion(&event);
3403 thread->event = &event;
3405 thread->mddev = mddev;
3406 thread->name = name;
3407 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3408 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3409 if (IS_ERR(thread->tsk)) {
3413 wait_for_completion(&event);
3417 void md_unregister_thread(mdk_thread_t *thread)
3419 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3421 kthread_stop(thread->tsk);
3425 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3432 if (!rdev || rdev->faulty)
3435 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3437 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3438 __builtin_return_address(0),__builtin_return_address(1),
3439 __builtin_return_address(2),__builtin_return_address(3));
3441 if (!mddev->pers->error_handler)
3443 mddev->pers->error_handler(mddev,rdev);
3444 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3445 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3446 md_wakeup_thread(mddev->thread);
3449 /* seq_file implementation /proc/mdstat */
3451 static void status_unused(struct seq_file *seq)
3455 struct list_head *tmp;
3457 seq_printf(seq, "unused devices: ");
3459 ITERATE_RDEV_PENDING(rdev,tmp) {
3460 char b[BDEVNAME_SIZE];
3462 seq_printf(seq, "%s ",
3463 bdevname(rdev->bdev,b));
3466 seq_printf(seq, "<none>");
3468 seq_printf(seq, "\n");
3472 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3474 unsigned long max_blocks, resync, res, dt, db, rt;
3476 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3478 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3479 max_blocks = mddev->resync_max_sectors >> 1;
3481 max_blocks = mddev->size;
3484 * Should not happen.
3490 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3492 int i, x = res/50, y = 20-x;
3493 seq_printf(seq, "[");
3494 for (i = 0; i < x; i++)
3495 seq_printf(seq, "=");
3496 seq_printf(seq, ">");
3497 for (i = 0; i < y; i++)
3498 seq_printf(seq, ".");
3499 seq_printf(seq, "] ");
3501 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3502 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3503 "resync" : "recovery"),
3504 res/10, res % 10, resync, max_blocks);
3507 * We do not want to overflow, so the order of operands and
3508 * the * 100 / 100 trick are important. We do a +1 to be
3509 * safe against division by zero. We only estimate anyway.
3511 * dt: time from mark until now
3512 * db: blocks written from mark until now
3513 * rt: remaining time
3515 dt = ((jiffies - mddev->resync_mark) / HZ);
3517 db = resync - (mddev->resync_mark_cnt/2);
3518 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3520 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3522 seq_printf(seq, " speed=%ldK/sec", db/dt);
3525 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3527 struct list_head *tmp;
3537 spin_lock(&all_mddevs_lock);
3538 list_for_each(tmp,&all_mddevs)
3540 mddev = list_entry(tmp, mddev_t, all_mddevs);
3542 spin_unlock(&all_mddevs_lock);
3545 spin_unlock(&all_mddevs_lock);
3547 return (void*)2;/* tail */
3551 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3553 struct list_head *tmp;
3554 mddev_t *next_mddev, *mddev = v;
3560 spin_lock(&all_mddevs_lock);
3562 tmp = all_mddevs.next;
3564 tmp = mddev->all_mddevs.next;
3565 if (tmp != &all_mddevs)
3566 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3568 next_mddev = (void*)2;
3571 spin_unlock(&all_mddevs_lock);
3579 static void md_seq_stop(struct seq_file *seq, void *v)
3583 if (mddev && v != (void*)1 && v != (void*)2)
3587 static int md_seq_show(struct seq_file *seq, void *v)
3591 struct list_head *tmp2;
3594 struct bitmap *bitmap;
3596 if (v == (void*)1) {
3597 seq_printf(seq, "Personalities : ");
3598 spin_lock(&pers_lock);
3599 for (i = 0; i < MAX_PERSONALITY; i++)
3601 seq_printf(seq, "[%s] ", pers[i]->name);
3603 spin_unlock(&pers_lock);
3604 seq_printf(seq, "\n");
3607 if (v == (void*)2) {
3612 if (mddev_lock(mddev)!=0)
3614 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3615 seq_printf(seq, "%s : %sactive", mdname(mddev),
3616 mddev->pers ? "" : "in");
3619 seq_printf(seq, " (read-only)");
3620 seq_printf(seq, " %s", mddev->pers->name);
3624 ITERATE_RDEV(mddev,rdev,tmp2) {
3625 char b[BDEVNAME_SIZE];
3626 seq_printf(seq, " %s[%d]",
3627 bdevname(rdev->bdev,b), rdev->desc_nr);
3628 if (test_bit(WriteMostly, &rdev->flags))
3629 seq_printf(seq, "(W)");
3631 seq_printf(seq, "(F)");
3633 } else if (rdev->raid_disk < 0)
3634 seq_printf(seq, "(S)"); /* spare */
3638 if (!list_empty(&mddev->disks)) {
3640 seq_printf(seq, "\n %llu blocks",
3641 (unsigned long long)mddev->array_size);
3643 seq_printf(seq, "\n %llu blocks",
3644 (unsigned long long)size);
3646 if (mddev->persistent) {
3647 if (mddev->major_version != 0 ||
3648 mddev->minor_version != 90) {
3649 seq_printf(seq," super %d.%d",
3650 mddev->major_version,
3651 mddev->minor_version);
3654 seq_printf(seq, " super non-persistent");
3657 mddev->pers->status (seq, mddev);
3658 seq_printf(seq, "\n ");
3659 if (mddev->curr_resync > 2) {
3660 status_resync (seq, mddev);
3661 seq_printf(seq, "\n ");
3662 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3663 seq_printf(seq, " resync=DELAYED\n ");
3665 seq_printf(seq, "\n ");
3667 if ((bitmap = mddev->bitmap)) {
3668 unsigned long chunk_kb;
3669 unsigned long flags;
3670 spin_lock_irqsave(&bitmap->lock, flags);
3671 chunk_kb = bitmap->chunksize >> 10;
3672 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3674 bitmap->pages - bitmap->missing_pages,
3676 (bitmap->pages - bitmap->missing_pages)
3677 << (PAGE_SHIFT - 10),
3678 chunk_kb ? chunk_kb : bitmap->chunksize,
3679 chunk_kb ? "KB" : "B");
3681 seq_printf(seq, ", file: ");
3682 seq_path(seq, bitmap->file->f_vfsmnt,
3683 bitmap->file->f_dentry," \t\n");
3686 seq_printf(seq, "\n");
3687 spin_unlock_irqrestore(&bitmap->lock, flags);
3690 seq_printf(seq, "\n");
3692 mddev_unlock(mddev);
3697 static struct seq_operations md_seq_ops = {
3698 .start = md_seq_start,
3699 .next = md_seq_next,
3700 .stop = md_seq_stop,
3701 .show = md_seq_show,
3704 static int md_seq_open(struct inode *inode, struct file *file)
3708 error = seq_open(file, &md_seq_ops);
3712 static struct file_operations md_seq_fops = {
3713 .open = md_seq_open,
3715 .llseek = seq_lseek,
3716 .release = seq_release,
3719 int register_md_personality(int pnum, mdk_personality_t *p)
3721 if (pnum >= MAX_PERSONALITY) {
3723 "md: tried to install personality %s as nr %d, but max is %lu\n",
3724 p->name, pnum, MAX_PERSONALITY-1);
3728 spin_lock(&pers_lock);
3730 spin_unlock(&pers_lock);
3735 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3736 spin_unlock(&pers_lock);
3740 int unregister_md_personality(int pnum)
3742 if (pnum >= MAX_PERSONALITY)
3745 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3746 spin_lock(&pers_lock);
3748 spin_unlock(&pers_lock);
3752 static int is_mddev_idle(mddev_t *mddev)
3755 struct list_head *tmp;
3757 unsigned long curr_events;
3760 ITERATE_RDEV(mddev,rdev,tmp) {
3761 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3762 curr_events = disk_stat_read(disk, sectors[0]) +
3763 disk_stat_read(disk, sectors[1]) -
3764 atomic_read(&disk->sync_io);
3765 /* Allow some slack between valud of curr_events and last_events,
3766 * as there are some uninteresting races.
3767 * Note: the following is an unsigned comparison.
3769 if ((curr_events - rdev->last_events + 32) > 64) {
3770 rdev->last_events = curr_events;
3777 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3779 /* another "blocks" (512byte) blocks have been synced */
3780 atomic_sub(blocks, &mddev->recovery_active);
3781 wake_up(&mddev->recovery_wait);
3783 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3784 md_wakeup_thread(mddev->thread);
3785 // stop recovery, signal do_sync ....
3790 /* md_write_start(mddev, bi)
3791 * If we need to update some array metadata (e.g. 'active' flag
3792 * in superblock) before writing, schedule a superblock update
3793 * and wait for it to complete.
3795 void md_write_start(mddev_t *mddev, struct bio *bi)
3797 if (bio_data_dir(bi) != WRITE)
3800 atomic_inc(&mddev->writes_pending);
3801 if (mddev->in_sync) {
3802 spin_lock(&mddev->write_lock);
3803 if (mddev->in_sync) {
3805 mddev->sb_dirty = 1;
3806 md_wakeup_thread(mddev->thread);
3808 spin_unlock(&mddev->write_lock);
3810 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3813 void md_write_end(mddev_t *mddev)
3815 if (atomic_dec_and_test(&mddev->writes_pending)) {
3816 if (mddev->safemode == 2)
3817 md_wakeup_thread(mddev->thread);
3819 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3823 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3825 #define SYNC_MARKS 10
3826 #define SYNC_MARK_STEP (3*HZ)
3827 static void md_do_sync(mddev_t *mddev)
3830 unsigned int currspeed = 0,
3832 sector_t max_sectors,j, io_sectors;
3833 unsigned long mark[SYNC_MARKS];
3834 sector_t mark_cnt[SYNC_MARKS];
3836 struct list_head *tmp;
3837 sector_t last_check;
3840 /* just incase thread restarts... */
3841 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3844 /* we overload curr_resync somewhat here.
3845 * 0 == not engaged in resync at all
3846 * 2 == checking that there is no conflict with another sync
3847 * 1 == like 2, but have yielded to allow conflicting resync to
3849 * other == active in resync - this many blocks
3851 * Before starting a resync we must have set curr_resync to
3852 * 2, and then checked that every "conflicting" array has curr_resync
3853 * less than ours. When we find one that is the same or higher
3854 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3855 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3856 * This will mean we have to start checking from the beginning again.
3861 mddev->curr_resync = 2;
3864 if (signal_pending(current) ||
3865 kthread_should_stop()) {
3866 flush_signals(current);
3867 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3870 ITERATE_MDDEV(mddev2,tmp) {
3871 if (mddev2 == mddev)
3873 if (mddev2->curr_resync &&
3874 match_mddev_units(mddev,mddev2)) {
3876 if (mddev < mddev2 && mddev->curr_resync == 2) {
3877 /* arbitrarily yield */
3878 mddev->curr_resync = 1;
3879 wake_up(&resync_wait);
3881 if (mddev > mddev2 && mddev->curr_resync == 1)
3882 /* no need to wait here, we can wait the next
3883 * time 'round when curr_resync == 2
3886 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3887 if (!signal_pending(current) &&
3888 !kthread_should_stop() &&
3889 mddev2->curr_resync >= mddev->curr_resync) {
3890 printk(KERN_INFO "md: delaying resync of %s"
3891 " until %s has finished resync (they"
3892 " share one or more physical units)\n",
3893 mdname(mddev), mdname(mddev2));
3896 finish_wait(&resync_wait, &wq);
3899 finish_wait(&resync_wait, &wq);
3902 } while (mddev->curr_resync < 2);
3904 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3905 /* resync follows the size requested by the personality,
3906 * which defaults to physical size, but can be virtual size
3908 max_sectors = mddev->resync_max_sectors;
3909 mddev->resync_mismatches = 0;
3911 /* recovery follows the physical size of devices */
3912 max_sectors = mddev->size << 1;
3914 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3915 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3916 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3917 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3918 "(but not more than %d KB/sec) for reconstruction.\n",
3919 sysctl_speed_limit_max);
3921 is_mddev_idle(mddev); /* this also initializes IO event counters */
3922 /* we don't use the checkpoint if there's a bitmap */
3923 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3924 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3925 j = mddev->recovery_cp;
3929 for (m = 0; m < SYNC_MARKS; m++) {
3931 mark_cnt[m] = io_sectors;
3934 mddev->resync_mark = mark[last_mark];
3935 mddev->resync_mark_cnt = mark_cnt[last_mark];
3938 * Tune reconstruction:
3940 window = 32*(PAGE_SIZE/512);
3941 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3942 window/2,(unsigned long long) max_sectors/2);
3944 atomic_set(&mddev->recovery_active, 0);
3945 init_waitqueue_head(&mddev->recovery_wait);
3950 "md: resuming recovery of %s from checkpoint.\n",
3952 mddev->curr_resync = j;
3955 while (j < max_sectors) {
3959 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3960 currspeed < sysctl_speed_limit_min);
3962 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3966 if (!skipped) { /* actual IO requested */
3967 io_sectors += sectors;
3968 atomic_add(sectors, &mddev->recovery_active);
3972 if (j>1) mddev->curr_resync = j;
3975 if (last_check + window > io_sectors || j == max_sectors)
3978 last_check = io_sectors;
3980 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3981 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3985 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3987 int next = (last_mark+1) % SYNC_MARKS;
3989 mddev->resync_mark = mark[next];
3990 mddev->resync_mark_cnt = mark_cnt[next];
3991 mark[next] = jiffies;
3992 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3997 if (signal_pending(current) || kthread_should_stop()) {
3999 * got a signal, exit.
4002 "md: md_do_sync() got signal ... exiting\n");
4003 flush_signals(current);
4004 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4009 * this loop exits only if either when we are slower than
4010 * the 'hard' speed limit, or the system was IO-idle for
4012 * the system might be non-idle CPU-wise, but we only care
4013 * about not overloading the IO subsystem. (things like an
4014 * e2fsck being done on the RAID array should execute fast)
4016 mddev->queue->unplug_fn(mddev->queue);
4019 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4020 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4022 if (currspeed > sysctl_speed_limit_min) {
4023 if ((currspeed > sysctl_speed_limit_max) ||
4024 !is_mddev_idle(mddev)) {
4025 msleep_interruptible(250);
4030 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4032 * this also signals 'finished resyncing' to md_stop
4035 mddev->queue->unplug_fn(mddev->queue);
4037 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4039 /* tell personality that we are finished */
4040 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4042 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4043 mddev->curr_resync > 2 &&
4044 mddev->curr_resync >= mddev->recovery_cp) {
4045 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4047 "md: checkpointing recovery of %s.\n",
4049 mddev->recovery_cp = mddev->curr_resync;
4051 mddev->recovery_cp = MaxSector;
4055 mddev->curr_resync = 0;
4056 wake_up(&resync_wait);
4057 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4058 md_wakeup_thread(mddev->thread);
4063 * This routine is regularly called by all per-raid-array threads to
4064 * deal with generic issues like resync and super-block update.
4065 * Raid personalities that don't have a thread (linear/raid0) do not
4066 * need this as they never do any recovery or update the superblock.
4068 * It does not do any resync itself, but rather "forks" off other threads
4069 * to do that as needed.
4070 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4071 * "->recovery" and create a thread at ->sync_thread.
4072 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4073 * and wakeups up this thread which will reap the thread and finish up.
4074 * This thread also removes any faulty devices (with nr_pending == 0).
4076 * The overall approach is:
4077 * 1/ if the superblock needs updating, update it.
4078 * 2/ If a recovery thread is running, don't do anything else.
4079 * 3/ If recovery has finished, clean up, possibly marking spares active.
4080 * 4/ If there are any faulty devices, remove them.
4081 * 5/ If array is degraded, try to add spares devices
4082 * 6/ If array has spares or is not in-sync, start a resync thread.
4084 void md_check_recovery(mddev_t *mddev)
4087 struct list_head *rtmp;
4091 bitmap_daemon_work(mddev->bitmap);
4096 if (signal_pending(current)) {
4097 if (mddev->pers->sync_request) {
4098 printk(KERN_INFO "md: %s in immediate safe mode\n",
4100 mddev->safemode = 2;
4102 flush_signals(current);
4107 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4108 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4109 (mddev->safemode == 1) ||
4110 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4111 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4115 if (mddev_trylock(mddev)==0) {
4118 spin_lock(&mddev->write_lock);
4119 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4120 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4122 mddev->sb_dirty = 1;
4124 if (mddev->safemode == 1)
4125 mddev->safemode = 0;
4126 spin_unlock(&mddev->write_lock);
4128 if (mddev->sb_dirty)
4129 md_update_sb(mddev);
4132 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4133 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4134 /* resync/recovery still happening */
4135 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4138 if (mddev->sync_thread) {
4139 /* resync has finished, collect result */
4140 md_unregister_thread(mddev->sync_thread);
4141 mddev->sync_thread = NULL;
4142 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4143 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4145 /* activate any spares */
4146 mddev->pers->spare_active(mddev);
4148 md_update_sb(mddev);
4150 /* if array is no-longer degraded, then any saved_raid_disk
4151 * information must be scrapped
4153 if (!mddev->degraded)
4154 ITERATE_RDEV(mddev,rdev,rtmp)
4155 rdev->saved_raid_disk = -1;
4157 mddev->recovery = 0;
4158 /* flag recovery needed just to double check */
4159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4162 /* Clear some bits that don't mean anything, but
4165 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4166 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4167 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4168 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4170 /* no recovery is running.
4171 * remove any failed drives, then
4172 * add spares if possible.
4173 * Spare are also removed and re-added, to allow
4174 * the personality to fail the re-add.
4176 ITERATE_RDEV(mddev,rdev,rtmp)
4177 if (rdev->raid_disk >= 0 &&
4178 (rdev->faulty || ! rdev->in_sync) &&
4179 atomic_read(&rdev->nr_pending)==0) {
4180 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4182 sprintf(nm,"rd%d", rdev->raid_disk);
4183 sysfs_remove_link(&mddev->kobj, nm);
4184 rdev->raid_disk = -1;
4188 if (mddev->degraded) {
4189 ITERATE_RDEV(mddev,rdev,rtmp)
4190 if (rdev->raid_disk < 0
4192 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4194 sprintf(nm, "rd%d", rdev->raid_disk);
4195 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4203 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4204 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4205 } else if (mddev->recovery_cp < MaxSector) {
4206 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4207 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4208 /* nothing to be done ... */
4211 if (mddev->pers->sync_request) {
4212 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4213 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4214 /* We are adding a device or devices to an array
4215 * which has the bitmap stored on all devices.
4216 * So make sure all bitmap pages get written
4218 bitmap_write_all(mddev->bitmap);
4220 mddev->sync_thread = md_register_thread(md_do_sync,
4223 if (!mddev->sync_thread) {
4224 printk(KERN_ERR "%s: could not start resync"
4227 /* leave the spares where they are, it shouldn't hurt */
4228 mddev->recovery = 0;
4230 md_wakeup_thread(mddev->sync_thread);
4234 mddev_unlock(mddev);
4238 static int md_notify_reboot(struct notifier_block *this,
4239 unsigned long code, void *x)
4241 struct list_head *tmp;
4244 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4246 printk(KERN_INFO "md: stopping all md devices.\n");
4248 ITERATE_MDDEV(mddev,tmp)
4249 if (mddev_trylock(mddev)==0)
4250 do_md_stop (mddev, 1);
4252 * certain more exotic SCSI devices are known to be
4253 * volatile wrt too early system reboots. While the
4254 * right place to handle this issue is the given
4255 * driver, we do want to have a safe RAID driver ...
4262 static struct notifier_block md_notifier = {
4263 .notifier_call = md_notify_reboot,
4265 .priority = INT_MAX, /* before any real devices */
4268 static void md_geninit(void)
4270 struct proc_dir_entry *p;
4272 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4274 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4276 p->proc_fops = &md_seq_fops;
4279 static int __init md_init(void)
4283 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4284 " MD_SB_DISKS=%d\n",
4285 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4286 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4287 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4290 if (register_blkdev(MAJOR_NR, "md"))
4292 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4293 unregister_blkdev(MAJOR_NR, "md");
4297 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4298 md_probe, NULL, NULL);
4299 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4300 md_probe, NULL, NULL);
4302 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4303 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4304 S_IFBLK|S_IRUSR|S_IWUSR,
4307 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4308 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4309 S_IFBLK|S_IRUSR|S_IWUSR,
4313 register_reboot_notifier(&md_notifier);
4314 raid_table_header = register_sysctl_table(raid_root_table, 1);
4324 * Searches all registered partitions for autorun RAID arrays
4327 static dev_t detected_devices[128];
4330 void md_autodetect_dev(dev_t dev)
4332 if (dev_cnt >= 0 && dev_cnt < 127)
4333 detected_devices[dev_cnt++] = dev;
4337 static void autostart_arrays(int part)
4342 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4344 for (i = 0; i < dev_cnt; i++) {
4345 dev_t dev = detected_devices[i];
4347 rdev = md_import_device(dev,0, 0);
4355 list_add(&rdev->same_set, &pending_raid_disks);
4359 autorun_devices(part);
4364 static __exit void md_exit(void)
4367 struct list_head *tmp;
4369 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4370 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4371 for (i=0; i < MAX_MD_DEVS; i++)
4372 devfs_remove("md/%d", i);
4373 for (i=0; i < MAX_MD_DEVS; i++)
4374 devfs_remove("md/d%d", i);
4378 unregister_blkdev(MAJOR_NR,"md");
4379 unregister_blkdev(mdp_major, "mdp");
4380 unregister_reboot_notifier(&md_notifier);
4381 unregister_sysctl_table(raid_table_header);
4382 remove_proc_entry("mdstat", NULL);
4383 ITERATE_MDDEV(mddev,tmp) {
4384 struct gendisk *disk = mddev->gendisk;
4387 export_array(mddev);
4390 mddev->gendisk = NULL;
4395 module_init(md_init)
4396 module_exit(md_exit)
4398 EXPORT_SYMBOL(register_md_personality);
4399 EXPORT_SYMBOL(unregister_md_personality);
4400 EXPORT_SYMBOL(md_error);
4401 EXPORT_SYMBOL(md_done_sync);
4402 EXPORT_SYMBOL(md_write_start);
4403 EXPORT_SYMBOL(md_write_end);
4404 EXPORT_SYMBOL(md_register_thread);
4405 EXPORT_SYMBOL(md_unregister_thread);
4406 EXPORT_SYMBOL(md_wakeup_thread);
4407 EXPORT_SYMBOL(md_print_devices);
4408 EXPORT_SYMBOL(md_check_recovery);
4409 MODULE_LICENSE("GPL");
4411 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);