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 gets a ref on the parent, so
809 * we have to drop the one we already have
811 kobject_add(&rdev2->kobj);
812 kobject_put(rdev->kobj.parent);
813 sysfs_create_link(&rdev2->kobj,
814 &rdev2->bdev->bd_disk->kobj,
816 if (rdev2->raid_disk >= 0) {
818 sprintf(nm, "rd%d", rdev2->raid_disk);
819 sysfs_create_link(&mddev->kobj,
823 /* now set the "removed" and "faulty" bits on any missing devices */
824 for (i=0 ; i < mddev->raid_disks ; i++) {
825 mdp_disk_t *d = &sb->disks[i];
826 if (d->state == 0 && d->number == 0) {
829 d->state = (1<<MD_DISK_REMOVED);
830 d->state |= (1<<MD_DISK_FAULTY);
834 sb->nr_disks = nr_disks;
835 sb->active_disks = active;
836 sb->working_disks = working;
837 sb->failed_disks = failed;
838 sb->spare_disks = spare;
840 sb->this_disk = sb->disks[rdev->desc_nr];
841 sb->sb_csum = calc_sb_csum(sb);
845 * version 1 superblock
848 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
850 unsigned int disk_csum, csum;
851 unsigned long long newcsum;
852 int size = 256 + le32_to_cpu(sb->max_dev)*2;
853 unsigned int *isuper = (unsigned int*)sb;
856 disk_csum = sb->sb_csum;
859 for (i=0; size>=4; size -= 4 )
860 newcsum += le32_to_cpu(*isuper++);
863 newcsum += le16_to_cpu(*(unsigned short*) isuper);
865 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
866 sb->sb_csum = disk_csum;
867 return cpu_to_le32(csum);
870 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
872 struct mdp_superblock_1 *sb;
875 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
879 * Calculate the position of the superblock.
880 * It is always aligned to a 4K boundary and
881 * depeding on minor_version, it can be:
882 * 0: At least 8K, but less than 12K, from end of device
883 * 1: At start of device
884 * 2: 4K from start of device.
886 switch(minor_version) {
888 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
890 sb_offset &= ~(sector_t)(4*2-1);
891 /* convert from sectors to K */
903 rdev->sb_offset = sb_offset;
905 /* superblock is rarely larger than 1K, but it can be larger,
906 * and it is safe to read 4k, so we do that
908 ret = read_disk_sb(rdev, 4096);
912 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
914 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
915 sb->major_version != cpu_to_le32(1) ||
916 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
917 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
918 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
921 if (calc_sb_1_csum(sb) != sb->sb_csum) {
922 printk("md: invalid superblock checksum on %s\n",
923 bdevname(rdev->bdev,b));
926 if (le64_to_cpu(sb->data_size) < 10) {
927 printk("md: data_size too small on %s\n",
928 bdevname(rdev->bdev,b));
931 rdev->preferred_minor = 0xffff;
932 rdev->data_offset = le64_to_cpu(sb->data_offset);
934 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
935 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
936 if (rdev->sb_size & bmask)
937 rdev-> sb_size = (rdev->sb_size | bmask)+1;
943 struct mdp_superblock_1 *refsb =
944 (struct mdp_superblock_1*)page_address(refdev->sb_page);
946 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
947 sb->level != refsb->level ||
948 sb->layout != refsb->layout ||
949 sb->chunksize != refsb->chunksize) {
950 printk(KERN_WARNING "md: %s has strangely different"
951 " superblock to %s\n",
952 bdevname(rdev->bdev,b),
953 bdevname(refdev->bdev,b2));
956 ev1 = le64_to_cpu(sb->events);
957 ev2 = le64_to_cpu(refsb->events);
963 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
965 rdev->size = rdev->sb_offset;
966 if (rdev->size < le64_to_cpu(sb->data_size)/2)
968 rdev->size = le64_to_cpu(sb->data_size)/2;
969 if (le32_to_cpu(sb->chunksize))
970 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
974 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
976 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
978 rdev->raid_disk = -1;
980 if (mddev->raid_disks == 0) {
981 mddev->major_version = 1;
982 mddev->patch_version = 0;
983 mddev->persistent = 1;
984 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
985 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
986 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
987 mddev->level = le32_to_cpu(sb->level);
988 mddev->layout = le32_to_cpu(sb->layout);
989 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
990 mddev->size = le64_to_cpu(sb->size)/2;
991 mddev->events = le64_to_cpu(sb->events);
992 mddev->bitmap_offset = 0;
993 mddev->default_bitmap_offset = 0;
994 mddev->default_bitmap_offset = 1024;
996 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
997 memcpy(mddev->uuid, sb->set_uuid, 16);
999 mddev->max_disks = (4096-256)/2;
1001 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1002 mddev->bitmap_file == NULL ) {
1003 if (mddev->level != 1) {
1004 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1007 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1009 } else if (mddev->pers == NULL) {
1010 /* Insist of good event counter while assembling */
1011 __u64 ev1 = le64_to_cpu(sb->events);
1013 if (ev1 < mddev->events)
1015 } else if (mddev->bitmap) {
1016 /* If adding to array with a bitmap, then we can accept an
1017 * older device, but not too old.
1019 __u64 ev1 = le64_to_cpu(sb->events);
1020 if (ev1 < mddev->bitmap->events_cleared)
1022 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1025 if (mddev->level != LEVEL_MULTIPATH) {
1027 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1028 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1030 case 0xffff: /* spare */
1033 case 0xfffe: /* faulty */
1039 rdev->raid_disk = role;
1043 if (sb->devflags & WriteMostly1)
1044 set_bit(WriteMostly, &rdev->flags);
1045 } else /* MULTIPATH are always insync */
1051 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1053 struct mdp_superblock_1 *sb;
1054 struct list_head *tmp;
1057 /* make rdev->sb match mddev and rdev data. */
1059 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1061 sb->feature_map = 0;
1063 memset(sb->pad1, 0, sizeof(sb->pad1));
1064 memset(sb->pad2, 0, sizeof(sb->pad2));
1065 memset(sb->pad3, 0, sizeof(sb->pad3));
1067 sb->utime = cpu_to_le64((__u64)mddev->utime);
1068 sb->events = cpu_to_le64(mddev->events);
1070 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1072 sb->resync_offset = cpu_to_le64(0);
1074 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1075 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1076 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1080 ITERATE_RDEV(mddev,rdev2,tmp)
1081 if (rdev2->desc_nr+1 > max_dev)
1082 max_dev = rdev2->desc_nr+1;
1084 sb->max_dev = cpu_to_le32(max_dev);
1085 for (i=0; i<max_dev;i++)
1086 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1088 ITERATE_RDEV(mddev,rdev2,tmp) {
1091 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1092 else if (rdev2->in_sync)
1093 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1095 sb->dev_roles[i] = cpu_to_le16(0xffff);
1098 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1099 sb->sb_csum = calc_sb_1_csum(sb);
1103 static struct super_type super_types[] = {
1106 .owner = THIS_MODULE,
1107 .load_super = super_90_load,
1108 .validate_super = super_90_validate,
1109 .sync_super = super_90_sync,
1113 .owner = THIS_MODULE,
1114 .load_super = super_1_load,
1115 .validate_super = super_1_validate,
1116 .sync_super = super_1_sync,
1120 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1122 struct list_head *tmp;
1125 ITERATE_RDEV(mddev,rdev,tmp)
1126 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1132 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1134 struct list_head *tmp;
1137 ITERATE_RDEV(mddev1,rdev,tmp)
1138 if (match_dev_unit(mddev2, rdev))
1144 static LIST_HEAD(pending_raid_disks);
1146 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1148 mdk_rdev_t *same_pdev;
1149 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1155 same_pdev = match_dev_unit(mddev, rdev);
1158 "%s: WARNING: %s appears to be on the same physical"
1159 " disk as %s. True\n protection against single-disk"
1160 " failure might be compromised.\n",
1161 mdname(mddev), bdevname(rdev->bdev,b),
1162 bdevname(same_pdev->bdev,b2));
1164 /* Verify rdev->desc_nr is unique.
1165 * If it is -1, assign a free number, else
1166 * check number is not in use
1168 if (rdev->desc_nr < 0) {
1170 if (mddev->pers) choice = mddev->raid_disks;
1171 while (find_rdev_nr(mddev, choice))
1173 rdev->desc_nr = choice;
1175 if (find_rdev_nr(mddev, rdev->desc_nr))
1179 list_add(&rdev->same_set, &mddev->disks);
1180 rdev->mddev = mddev;
1181 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1183 rdev->kobj.k_name = NULL;
1184 snprintf(rdev->kobj.name, KOBJ_NAME_LEN, "dev%d", rdev->desc_nr);
1185 rdev->kobj.parent = &mddev->kobj;
1186 kobject_add(&rdev->kobj);
1188 sysfs_create_link(&rdev->kobj, &rdev->bdev->bd_disk->kobj, "block");
1192 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1194 char b[BDEVNAME_SIZE];
1199 list_del_init(&rdev->same_set);
1200 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1202 sysfs_remove_link(&rdev->kobj, "block");
1203 kobject_del(&rdev->kobj);
1207 * prevent the device from being mounted, repartitioned or
1208 * otherwise reused by a RAID array (or any other kernel
1209 * subsystem), by bd_claiming the device.
1211 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1214 struct block_device *bdev;
1215 char b[BDEVNAME_SIZE];
1217 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1219 printk(KERN_ERR "md: could not open %s.\n",
1220 __bdevname(dev, b));
1221 return PTR_ERR(bdev);
1223 err = bd_claim(bdev, rdev);
1225 printk(KERN_ERR "md: could not bd_claim %s.\n",
1234 static void unlock_rdev(mdk_rdev_t *rdev)
1236 struct block_device *bdev = rdev->bdev;
1244 void md_autodetect_dev(dev_t dev);
1246 static void export_rdev(mdk_rdev_t * rdev)
1248 char b[BDEVNAME_SIZE];
1249 printk(KERN_INFO "md: export_rdev(%s)\n",
1250 bdevname(rdev->bdev,b));
1254 list_del_init(&rdev->same_set);
1256 md_autodetect_dev(rdev->bdev->bd_dev);
1259 kobject_put(&rdev->kobj);
1262 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1264 unbind_rdev_from_array(rdev);
1268 static void export_array(mddev_t *mddev)
1270 struct list_head *tmp;
1273 ITERATE_RDEV(mddev,rdev,tmp) {
1278 kick_rdev_from_array(rdev);
1280 if (!list_empty(&mddev->disks))
1282 mddev->raid_disks = 0;
1283 mddev->major_version = 0;
1286 static void print_desc(mdp_disk_t *desc)
1288 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1289 desc->major,desc->minor,desc->raid_disk,desc->state);
1292 static void print_sb(mdp_super_t *sb)
1297 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1298 sb->major_version, sb->minor_version, sb->patch_version,
1299 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1301 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1302 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1303 sb->md_minor, sb->layout, sb->chunk_size);
1304 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1305 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1306 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1307 sb->failed_disks, sb->spare_disks,
1308 sb->sb_csum, (unsigned long)sb->events_lo);
1311 for (i = 0; i < MD_SB_DISKS; i++) {
1314 desc = sb->disks + i;
1315 if (desc->number || desc->major || desc->minor ||
1316 desc->raid_disk || (desc->state && (desc->state != 4))) {
1317 printk(" D %2d: ", i);
1321 printk(KERN_INFO "md: THIS: ");
1322 print_desc(&sb->this_disk);
1326 static void print_rdev(mdk_rdev_t *rdev)
1328 char b[BDEVNAME_SIZE];
1329 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1330 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1331 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1332 if (rdev->sb_loaded) {
1333 printk(KERN_INFO "md: rdev superblock:\n");
1334 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1336 printk(KERN_INFO "md: no rdev superblock!\n");
1339 void md_print_devices(void)
1341 struct list_head *tmp, *tmp2;
1344 char b[BDEVNAME_SIZE];
1347 printk("md: **********************************\n");
1348 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1349 printk("md: **********************************\n");
1350 ITERATE_MDDEV(mddev,tmp) {
1353 bitmap_print_sb(mddev->bitmap);
1355 printk("%s: ", mdname(mddev));
1356 ITERATE_RDEV(mddev,rdev,tmp2)
1357 printk("<%s>", bdevname(rdev->bdev,b));
1360 ITERATE_RDEV(mddev,rdev,tmp2)
1363 printk("md: **********************************\n");
1368 static void sync_sbs(mddev_t * mddev)
1371 struct list_head *tmp;
1373 ITERATE_RDEV(mddev,rdev,tmp) {
1374 super_types[mddev->major_version].
1375 sync_super(mddev, rdev);
1376 rdev->sb_loaded = 1;
1380 static void md_update_sb(mddev_t * mddev)
1383 struct list_head *tmp;
1388 spin_lock(&mddev->write_lock);
1389 sync_req = mddev->in_sync;
1390 mddev->utime = get_seconds();
1393 if (!mddev->events) {
1395 * oops, this 64-bit counter should never wrap.
1396 * Either we are in around ~1 trillion A.C., assuming
1397 * 1 reboot per second, or we have a bug:
1402 mddev->sb_dirty = 2;
1406 * do not write anything to disk if using
1407 * nonpersistent superblocks
1409 if (!mddev->persistent) {
1410 mddev->sb_dirty = 0;
1411 spin_unlock(&mddev->write_lock);
1412 wake_up(&mddev->sb_wait);
1415 spin_unlock(&mddev->write_lock);
1418 "md: updating %s RAID superblock on device (in sync %d)\n",
1419 mdname(mddev),mddev->in_sync);
1421 err = bitmap_update_sb(mddev->bitmap);
1422 ITERATE_RDEV(mddev,rdev,tmp) {
1423 char b[BDEVNAME_SIZE];
1424 dprintk(KERN_INFO "md: ");
1426 dprintk("(skipping faulty ");
1428 dprintk("%s ", bdevname(rdev->bdev,b));
1429 if (!rdev->faulty) {
1430 md_super_write(mddev,rdev,
1431 rdev->sb_offset<<1, rdev->sb_size,
1433 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1434 bdevname(rdev->bdev,b),
1435 (unsigned long long)rdev->sb_offset);
1439 if (mddev->level == LEVEL_MULTIPATH)
1440 /* only need to write one superblock... */
1443 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1444 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1446 spin_lock(&mddev->write_lock);
1447 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1448 /* have to write it out again */
1449 spin_unlock(&mddev->write_lock);
1452 mddev->sb_dirty = 0;
1453 spin_unlock(&mddev->write_lock);
1454 wake_up(&mddev->sb_wait);
1458 struct rdev_sysfs_entry {
1459 struct attribute attr;
1460 ssize_t (*show)(mdk_rdev_t *, char *);
1461 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1465 rdev_show_state(mdk_rdev_t *rdev, char *page)
1471 len+= sprintf(page+len, "%sfaulty",sep);
1474 if (rdev->in_sync) {
1475 len += sprintf(page+len, "%sin_sync",sep);
1478 if (!rdev->faulty && !rdev->in_sync) {
1479 len += sprintf(page+len, "%sspare", sep);
1482 return len+sprintf(page+len, "\n");
1485 static struct rdev_sysfs_entry rdev_state = {
1486 .attr = {.name = "state", .mode = S_IRUGO },
1487 .show = rdev_show_state,
1491 rdev_show_super(mdk_rdev_t *rdev, char *page)
1493 if (rdev->sb_loaded && rdev->sb_size) {
1494 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1495 return rdev->sb_size;
1499 static struct rdev_sysfs_entry rdev_super = {
1500 .attr = {.name = "super", .mode = S_IRUGO },
1501 .show = rdev_show_super,
1503 static struct attribute *rdev_default_attrs[] = {
1509 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1511 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1512 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1516 return entry->show(rdev, page);
1520 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1521 const char *page, size_t length)
1523 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1524 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1528 return entry->store(rdev, page, length);
1531 static void rdev_free(struct kobject *ko)
1533 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1536 static struct sysfs_ops rdev_sysfs_ops = {
1537 .show = rdev_attr_show,
1538 .store = rdev_attr_store,
1540 static struct kobj_type rdev_ktype = {
1541 .release = rdev_free,
1542 .sysfs_ops = &rdev_sysfs_ops,
1543 .default_attrs = rdev_default_attrs,
1547 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1549 * mark the device faulty if:
1551 * - the device is nonexistent (zero size)
1552 * - the device has no valid superblock
1554 * a faulty rdev _never_ has rdev->sb set.
1556 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1558 char b[BDEVNAME_SIZE];
1563 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1565 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1566 return ERR_PTR(-ENOMEM);
1568 memset(rdev, 0, sizeof(*rdev));
1570 if ((err = alloc_disk_sb(rdev)))
1573 err = lock_rdev(rdev, newdev);
1577 rdev->kobj.parent = NULL;
1578 rdev->kobj.ktype = &rdev_ktype;
1579 kobject_init(&rdev->kobj);
1584 rdev->data_offset = 0;
1585 atomic_set(&rdev->nr_pending, 0);
1587 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1590 "md: %s has zero or unknown size, marking faulty!\n",
1591 bdevname(rdev->bdev,b));
1596 if (super_format >= 0) {
1597 err = super_types[super_format].
1598 load_super(rdev, NULL, super_minor);
1599 if (err == -EINVAL) {
1601 "md: %s has invalid sb, not importing!\n",
1602 bdevname(rdev->bdev,b));
1607 "md: could not read %s's sb, not importing!\n",
1608 bdevname(rdev->bdev,b));
1612 INIT_LIST_HEAD(&rdev->same_set);
1617 if (rdev->sb_page) {
1623 return ERR_PTR(err);
1627 * Check a full RAID array for plausibility
1631 static void analyze_sbs(mddev_t * mddev)
1634 struct list_head *tmp;
1635 mdk_rdev_t *rdev, *freshest;
1636 char b[BDEVNAME_SIZE];
1639 ITERATE_RDEV(mddev,rdev,tmp)
1640 switch (super_types[mddev->major_version].
1641 load_super(rdev, freshest, mddev->minor_version)) {
1649 "md: fatal superblock inconsistency in %s"
1650 " -- removing from array\n",
1651 bdevname(rdev->bdev,b));
1652 kick_rdev_from_array(rdev);
1656 super_types[mddev->major_version].
1657 validate_super(mddev, freshest);
1660 ITERATE_RDEV(mddev,rdev,tmp) {
1661 if (rdev != freshest)
1662 if (super_types[mddev->major_version].
1663 validate_super(mddev, rdev)) {
1664 printk(KERN_WARNING "md: kicking non-fresh %s"
1666 bdevname(rdev->bdev,b));
1667 kick_rdev_from_array(rdev);
1670 if (mddev->level == LEVEL_MULTIPATH) {
1671 rdev->desc_nr = i++;
1672 rdev->raid_disk = rdev->desc_nr;
1679 if (mddev->recovery_cp != MaxSector &&
1681 printk(KERN_ERR "md: %s: raid array is not clean"
1682 " -- starting background reconstruction\n",
1687 struct md_sysfs_entry {
1688 struct attribute attr;
1689 ssize_t (*show)(mddev_t *, char *);
1690 ssize_t (*store)(mddev_t *, const char *, size_t);
1694 md_show_level(mddev_t *mddev, char *page)
1696 mdk_personality_t *p = mddev->pers;
1699 if (mddev->level >= 0)
1700 return sprintf(page, "RAID-%d\n", mddev->level);
1702 return sprintf(page, "%s\n", p->name);
1705 static struct md_sysfs_entry md_level = {
1706 .attr = {.name = "level", .mode = S_IRUGO },
1707 .show = md_show_level,
1711 md_show_rdisks(mddev_t *mddev, char *page)
1713 return sprintf(page, "%d\n", mddev->raid_disks);
1716 static struct md_sysfs_entry md_raid_disks = {
1717 .attr = {.name = "raid_disks", .mode = S_IRUGO },
1718 .show = md_show_rdisks,
1722 md_show_scan(mddev_t *mddev, char *page)
1724 char *type = "none";
1725 if (mddev->recovery &
1726 ((1<<MD_RECOVERY_RUNNING) || (1<<MD_RECOVERY_NEEDED))) {
1727 if (mddev->recovery & (1<<MD_RECOVERY_SYNC)) {
1728 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1730 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1737 return sprintf(page, "%s\n", type);
1741 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1744 if (mddev->recovery &
1745 ((1<<MD_RECOVERY_RUNNING) || (1<<MD_RECOVERY_NEEDED)))
1747 down(&mddev->reconfig_sem);
1748 if (mddev->pers && mddev->pers->sync_request)
1750 up(&mddev->reconfig_sem);
1754 if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1755 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1756 else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1758 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1759 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1761 md_wakeup_thread(mddev->thread);
1766 md_show_mismatch(mddev_t *mddev, char *page)
1768 return sprintf(page, "%llu\n",
1769 (unsigned long long) mddev->resync_mismatches);
1772 static struct md_sysfs_entry md_scan_mode = {
1773 .attr = {.name = "scan_mode", .mode = S_IRUGO|S_IWUSR },
1774 .show = md_show_scan,
1775 .store = md_store_scan,
1778 static struct md_sysfs_entry md_mismatches = {
1779 .attr = {.name = "mismatch_cnt", .mode = S_IRUGO },
1780 .show = md_show_mismatch,
1783 static struct attribute *md_default_attrs[] = {
1785 &md_raid_disks.attr,
1787 &md_mismatches.attr,
1792 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1794 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1795 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1799 return entry->show(mddev, page);
1803 md_attr_store(struct kobject *kobj, struct attribute *attr,
1804 const char *page, size_t length)
1806 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1807 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1811 return entry->store(mddev, page, length);
1814 static void md_free(struct kobject *ko)
1816 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1820 static struct sysfs_ops md_sysfs_ops = {
1821 .show = md_attr_show,
1822 .store = md_attr_store,
1824 static struct kobj_type md_ktype = {
1826 .sysfs_ops = &md_sysfs_ops,
1827 .default_attrs = md_default_attrs,
1832 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1834 static DECLARE_MUTEX(disks_sem);
1835 mddev_t *mddev = mddev_find(dev);
1836 struct gendisk *disk;
1837 int partitioned = (MAJOR(dev) != MD_MAJOR);
1838 int shift = partitioned ? MdpMinorShift : 0;
1839 int unit = MINOR(dev) >> shift;
1845 if (mddev->gendisk) {
1850 disk = alloc_disk(1 << shift);
1856 disk->major = MAJOR(dev);
1857 disk->first_minor = unit << shift;
1859 sprintf(disk->disk_name, "md_d%d", unit);
1860 sprintf(disk->devfs_name, "md/d%d", unit);
1862 sprintf(disk->disk_name, "md%d", unit);
1863 sprintf(disk->devfs_name, "md/%d", unit);
1865 disk->fops = &md_fops;
1866 disk->private_data = mddev;
1867 disk->queue = mddev->queue;
1869 mddev->gendisk = disk;
1871 mddev->kobj.parent = &disk->kobj;
1872 mddev->kobj.k_name = NULL;
1873 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1874 mddev->kobj.ktype = &md_ktype;
1875 kobject_register(&mddev->kobj);
1879 void md_wakeup_thread(mdk_thread_t *thread);
1881 static void md_safemode_timeout(unsigned long data)
1883 mddev_t *mddev = (mddev_t *) data;
1885 mddev->safemode = 1;
1886 md_wakeup_thread(mddev->thread);
1890 static int do_md_run(mddev_t * mddev)
1894 struct list_head *tmp;
1896 struct gendisk *disk;
1897 char b[BDEVNAME_SIZE];
1899 if (list_empty(&mddev->disks))
1900 /* cannot run an array with no devices.. */
1907 * Analyze all RAID superblock(s)
1909 if (!mddev->raid_disks)
1912 chunk_size = mddev->chunk_size;
1913 pnum = level_to_pers(mddev->level);
1915 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1918 * 'default chunksize' in the old md code used to
1919 * be PAGE_SIZE, baaad.
1920 * we abort here to be on the safe side. We don't
1921 * want to continue the bad practice.
1924 "no chunksize specified, see 'man raidtab'\n");
1927 if (chunk_size > MAX_CHUNK_SIZE) {
1928 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1929 chunk_size, MAX_CHUNK_SIZE);
1933 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1935 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1936 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1939 if (chunk_size < PAGE_SIZE) {
1940 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1941 chunk_size, PAGE_SIZE);
1945 /* devices must have minimum size of one chunk */
1946 ITERATE_RDEV(mddev,rdev,tmp) {
1949 if (rdev->size < chunk_size / 1024) {
1951 "md: Dev %s smaller than chunk_size:"
1953 bdevname(rdev->bdev,b),
1954 (unsigned long long)rdev->size,
1964 request_module("md-personality-%d", pnum);
1969 * Drop all container device buffers, from now on
1970 * the only valid external interface is through the md
1972 * Also find largest hardsector size
1974 ITERATE_RDEV(mddev,rdev,tmp) {
1977 sync_blockdev(rdev->bdev);
1978 invalidate_bdev(rdev->bdev, 0);
1981 md_probe(mddev->unit, NULL, NULL);
1982 disk = mddev->gendisk;
1986 spin_lock(&pers_lock);
1987 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1988 spin_unlock(&pers_lock);
1989 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1994 mddev->pers = pers[pnum];
1995 spin_unlock(&pers_lock);
1997 mddev->recovery = 0;
1998 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2000 /* before we start the array running, initialise the bitmap */
2001 err = bitmap_create(mddev);
2003 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2004 mdname(mddev), err);
2006 err = mddev->pers->run(mddev);
2008 printk(KERN_ERR "md: pers->run() failed ...\n");
2009 module_put(mddev->pers->owner);
2011 bitmap_destroy(mddev);
2014 atomic_set(&mddev->writes_pending,0);
2015 mddev->safemode = 0;
2016 mddev->safemode_timer.function = md_safemode_timeout;
2017 mddev->safemode_timer.data = (unsigned long) mddev;
2018 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2021 ITERATE_RDEV(mddev,rdev,tmp)
2022 if (rdev->raid_disk >= 0) {
2024 sprintf(nm, "rd%d", rdev->raid_disk);
2025 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2028 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2029 md_wakeup_thread(mddev->thread);
2031 if (mddev->sb_dirty)
2032 md_update_sb(mddev);
2034 set_capacity(disk, mddev->array_size<<1);
2036 /* If we call blk_queue_make_request here, it will
2037 * re-initialise max_sectors etc which may have been
2038 * refined inside -> run. So just set the bits we need to set.
2039 * Most initialisation happended when we called
2040 * blk_queue_make_request(..., md_fail_request)
2043 mddev->queue->queuedata = mddev;
2044 mddev->queue->make_request_fn = mddev->pers->make_request;
2050 static int restart_array(mddev_t *mddev)
2052 struct gendisk *disk = mddev->gendisk;
2056 * Complain if it has no devices
2059 if (list_empty(&mddev->disks))
2067 mddev->safemode = 0;
2069 set_disk_ro(disk, 0);
2071 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2074 * Kick recovery or resync if necessary
2076 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2077 md_wakeup_thread(mddev->thread);
2080 printk(KERN_ERR "md: %s has no personality assigned.\n",
2089 static int do_md_stop(mddev_t * mddev, int ro)
2092 struct gendisk *disk = mddev->gendisk;
2095 if (atomic_read(&mddev->active)>2) {
2096 printk("md: %s still in use.\n",mdname(mddev));
2100 if (mddev->sync_thread) {
2101 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2102 md_unregister_thread(mddev->sync_thread);
2103 mddev->sync_thread = NULL;
2106 del_timer_sync(&mddev->safemode_timer);
2108 invalidate_partition(disk, 0);
2116 bitmap_flush(mddev);
2117 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
2119 set_disk_ro(disk, 0);
2120 blk_queue_make_request(mddev->queue, md_fail_request);
2121 mddev->pers->stop(mddev);
2122 module_put(mddev->pers->owner);
2127 if (!mddev->in_sync) {
2128 /* mark array as shutdown cleanly */
2130 md_update_sb(mddev);
2133 set_disk_ro(disk, 1);
2136 bitmap_destroy(mddev);
2137 if (mddev->bitmap_file) {
2138 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2139 fput(mddev->bitmap_file);
2140 mddev->bitmap_file = NULL;
2142 mddev->bitmap_offset = 0;
2145 * Free resources if final stop
2149 struct list_head *tmp;
2150 struct gendisk *disk;
2151 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2153 ITERATE_RDEV(mddev,rdev,tmp)
2154 if (rdev->raid_disk >= 0) {
2156 sprintf(nm, "rd%d", rdev->raid_disk);
2157 sysfs_remove_link(&mddev->kobj, nm);
2160 export_array(mddev);
2162 mddev->array_size = 0;
2163 disk = mddev->gendisk;
2165 set_capacity(disk, 0);
2168 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2175 static void autorun_array(mddev_t *mddev)
2178 struct list_head *tmp;
2181 if (list_empty(&mddev->disks))
2184 printk(KERN_INFO "md: running: ");
2186 ITERATE_RDEV(mddev,rdev,tmp) {
2187 char b[BDEVNAME_SIZE];
2188 printk("<%s>", bdevname(rdev->bdev,b));
2192 err = do_md_run (mddev);
2194 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2195 do_md_stop (mddev, 0);
2200 * lets try to run arrays based on all disks that have arrived
2201 * until now. (those are in pending_raid_disks)
2203 * the method: pick the first pending disk, collect all disks with
2204 * the same UUID, remove all from the pending list and put them into
2205 * the 'same_array' list. Then order this list based on superblock
2206 * update time (freshest comes first), kick out 'old' disks and
2207 * compare superblocks. If everything's fine then run it.
2209 * If "unit" is allocated, then bump its reference count
2211 static void autorun_devices(int part)
2213 struct list_head candidates;
2214 struct list_head *tmp;
2215 mdk_rdev_t *rdev0, *rdev;
2217 char b[BDEVNAME_SIZE];
2219 printk(KERN_INFO "md: autorun ...\n");
2220 while (!list_empty(&pending_raid_disks)) {
2222 rdev0 = list_entry(pending_raid_disks.next,
2223 mdk_rdev_t, same_set);
2225 printk(KERN_INFO "md: considering %s ...\n",
2226 bdevname(rdev0->bdev,b));
2227 INIT_LIST_HEAD(&candidates);
2228 ITERATE_RDEV_PENDING(rdev,tmp)
2229 if (super_90_load(rdev, rdev0, 0) >= 0) {
2230 printk(KERN_INFO "md: adding %s ...\n",
2231 bdevname(rdev->bdev,b));
2232 list_move(&rdev->same_set, &candidates);
2235 * now we have a set of devices, with all of them having
2236 * mostly sane superblocks. It's time to allocate the
2239 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2240 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2241 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2245 dev = MKDEV(mdp_major,
2246 rdev0->preferred_minor << MdpMinorShift);
2248 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2250 md_probe(dev, NULL, NULL);
2251 mddev = mddev_find(dev);
2254 "md: cannot allocate memory for md drive.\n");
2257 if (mddev_lock(mddev))
2258 printk(KERN_WARNING "md: %s locked, cannot run\n",
2260 else if (mddev->raid_disks || mddev->major_version
2261 || !list_empty(&mddev->disks)) {
2263 "md: %s already running, cannot run %s\n",
2264 mdname(mddev), bdevname(rdev0->bdev,b));
2265 mddev_unlock(mddev);
2267 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2268 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2269 list_del_init(&rdev->same_set);
2270 if (bind_rdev_to_array(rdev, mddev))
2273 autorun_array(mddev);
2274 mddev_unlock(mddev);
2276 /* on success, candidates will be empty, on error
2279 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2283 printk(KERN_INFO "md: ... autorun DONE.\n");
2287 * import RAID devices based on one partition
2288 * if possible, the array gets run as well.
2291 static int autostart_array(dev_t startdev)
2293 char b[BDEVNAME_SIZE];
2294 int err = -EINVAL, i;
2295 mdp_super_t *sb = NULL;
2296 mdk_rdev_t *start_rdev = NULL, *rdev;
2298 start_rdev = md_import_device(startdev, 0, 0);
2299 if (IS_ERR(start_rdev))
2303 /* NOTE: this can only work for 0.90.0 superblocks */
2304 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2305 if (sb->major_version != 0 ||
2306 sb->minor_version != 90 ) {
2307 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2308 export_rdev(start_rdev);
2312 if (start_rdev->faulty) {
2314 "md: can not autostart based on faulty %s!\n",
2315 bdevname(start_rdev->bdev,b));
2316 export_rdev(start_rdev);
2319 list_add(&start_rdev->same_set, &pending_raid_disks);
2321 for (i = 0; i < MD_SB_DISKS; i++) {
2322 mdp_disk_t *desc = sb->disks + i;
2323 dev_t dev = MKDEV(desc->major, desc->minor);
2327 if (dev == startdev)
2329 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2331 rdev = md_import_device(dev, 0, 0);
2335 list_add(&rdev->same_set, &pending_raid_disks);
2339 * possibly return codes
2347 static int get_version(void __user * arg)
2351 ver.major = MD_MAJOR_VERSION;
2352 ver.minor = MD_MINOR_VERSION;
2353 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2355 if (copy_to_user(arg, &ver, sizeof(ver)))
2361 static int get_array_info(mddev_t * mddev, void __user * arg)
2363 mdu_array_info_t info;
2364 int nr,working,active,failed,spare;
2366 struct list_head *tmp;
2368 nr=working=active=failed=spare=0;
2369 ITERATE_RDEV(mddev,rdev,tmp) {
2382 info.major_version = mddev->major_version;
2383 info.minor_version = mddev->minor_version;
2384 info.patch_version = MD_PATCHLEVEL_VERSION;
2385 info.ctime = mddev->ctime;
2386 info.level = mddev->level;
2387 info.size = mddev->size;
2389 info.raid_disks = mddev->raid_disks;
2390 info.md_minor = mddev->md_minor;
2391 info.not_persistent= !mddev->persistent;
2393 info.utime = mddev->utime;
2396 info.state = (1<<MD_SB_CLEAN);
2397 if (mddev->bitmap && mddev->bitmap_offset)
2398 info.state = (1<<MD_SB_BITMAP_PRESENT);
2399 info.active_disks = active;
2400 info.working_disks = working;
2401 info.failed_disks = failed;
2402 info.spare_disks = spare;
2404 info.layout = mddev->layout;
2405 info.chunk_size = mddev->chunk_size;
2407 if (copy_to_user(arg, &info, sizeof(info)))
2413 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2415 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2416 char *ptr, *buf = NULL;
2419 file = kmalloc(sizeof(*file), GFP_KERNEL);
2423 /* bitmap disabled, zero the first byte and copy out */
2424 if (!mddev->bitmap || !mddev->bitmap->file) {
2425 file->pathname[0] = '\0';
2429 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2433 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2437 strcpy(file->pathname, ptr);
2441 if (copy_to_user(arg, file, sizeof(*file)))
2449 static int get_disk_info(mddev_t * mddev, void __user * arg)
2451 mdu_disk_info_t info;
2455 if (copy_from_user(&info, arg, sizeof(info)))
2460 rdev = find_rdev_nr(mddev, nr);
2462 info.major = MAJOR(rdev->bdev->bd_dev);
2463 info.minor = MINOR(rdev->bdev->bd_dev);
2464 info.raid_disk = rdev->raid_disk;
2467 info.state |= (1<<MD_DISK_FAULTY);
2468 else if (rdev->in_sync) {
2469 info.state |= (1<<MD_DISK_ACTIVE);
2470 info.state |= (1<<MD_DISK_SYNC);
2472 if (test_bit(WriteMostly, &rdev->flags))
2473 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2475 info.major = info.minor = 0;
2476 info.raid_disk = -1;
2477 info.state = (1<<MD_DISK_REMOVED);
2480 if (copy_to_user(arg, &info, sizeof(info)))
2486 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2488 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2490 dev_t dev = MKDEV(info->major,info->minor);
2492 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2495 if (!mddev->raid_disks) {
2497 /* expecting a device which has a superblock */
2498 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2501 "md: md_import_device returned %ld\n",
2503 return PTR_ERR(rdev);
2505 if (!list_empty(&mddev->disks)) {
2506 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2507 mdk_rdev_t, same_set);
2508 int err = super_types[mddev->major_version]
2509 .load_super(rdev, rdev0, mddev->minor_version);
2512 "md: %s has different UUID to %s\n",
2513 bdevname(rdev->bdev,b),
2514 bdevname(rdev0->bdev,b2));
2519 err = bind_rdev_to_array(rdev, mddev);
2526 * add_new_disk can be used once the array is assembled
2527 * to add "hot spares". They must already have a superblock
2532 if (!mddev->pers->hot_add_disk) {
2534 "%s: personality does not support diskops!\n",
2538 if (mddev->persistent)
2539 rdev = md_import_device(dev, mddev->major_version,
2540 mddev->minor_version);
2542 rdev = md_import_device(dev, -1, -1);
2545 "md: md_import_device returned %ld\n",
2547 return PTR_ERR(rdev);
2549 /* set save_raid_disk if appropriate */
2550 if (!mddev->persistent) {
2551 if (info->state & (1<<MD_DISK_SYNC) &&
2552 info->raid_disk < mddev->raid_disks)
2553 rdev->raid_disk = info->raid_disk;
2555 rdev->raid_disk = -1;
2557 super_types[mddev->major_version].
2558 validate_super(mddev, rdev);
2559 rdev->saved_raid_disk = rdev->raid_disk;
2561 rdev->in_sync = 0; /* just to be sure */
2562 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2563 set_bit(WriteMostly, &rdev->flags);
2565 rdev->raid_disk = -1;
2566 err = bind_rdev_to_array(rdev, mddev);
2570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2571 md_wakeup_thread(mddev->thread);
2575 /* otherwise, add_new_disk is only allowed
2576 * for major_version==0 superblocks
2578 if (mddev->major_version != 0) {
2579 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2584 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2586 rdev = md_import_device (dev, -1, 0);
2589 "md: error, md_import_device() returned %ld\n",
2591 return PTR_ERR(rdev);
2593 rdev->desc_nr = info->number;
2594 if (info->raid_disk < mddev->raid_disks)
2595 rdev->raid_disk = info->raid_disk;
2597 rdev->raid_disk = -1;
2600 if (rdev->raid_disk < mddev->raid_disks)
2601 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2605 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2606 set_bit(WriteMostly, &rdev->flags);
2608 err = bind_rdev_to_array(rdev, mddev);
2614 if (!mddev->persistent) {
2615 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2616 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2618 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2619 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2621 if (!mddev->size || (mddev->size > rdev->size))
2622 mddev->size = rdev->size;
2628 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2630 char b[BDEVNAME_SIZE];
2636 rdev = find_rdev(mddev, dev);
2640 if (rdev->raid_disk >= 0)
2643 kick_rdev_from_array(rdev);
2644 md_update_sb(mddev);
2648 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2649 bdevname(rdev->bdev,b), mdname(mddev));
2653 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2655 char b[BDEVNAME_SIZE];
2663 if (mddev->major_version != 0) {
2664 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2665 " version-0 superblocks.\n",
2669 if (!mddev->pers->hot_add_disk) {
2671 "%s: personality does not support diskops!\n",
2676 rdev = md_import_device (dev, -1, 0);
2679 "md: error, md_import_device() returned %ld\n",
2684 if (mddev->persistent)
2685 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2688 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2690 size = calc_dev_size(rdev, mddev->chunk_size);
2693 if (size < mddev->size) {
2695 "%s: disk size %llu blocks < array size %llu\n",
2696 mdname(mddev), (unsigned long long)size,
2697 (unsigned long long)mddev->size);
2704 "md: can not hot-add faulty %s disk to %s!\n",
2705 bdevname(rdev->bdev,b), mdname(mddev));
2711 bind_rdev_to_array(rdev, mddev);
2714 * The rest should better be atomic, we can have disk failures
2715 * noticed in interrupt contexts ...
2718 if (rdev->desc_nr == mddev->max_disks) {
2719 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2722 goto abort_unbind_export;
2725 rdev->raid_disk = -1;
2727 md_update_sb(mddev);
2730 * Kick recovery, maybe this spare has to be added to the
2731 * array immediately.
2733 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2734 md_wakeup_thread(mddev->thread);
2738 abort_unbind_export:
2739 unbind_rdev_from_array(rdev);
2746 /* similar to deny_write_access, but accounts for our holding a reference
2747 * to the file ourselves */
2748 static int deny_bitmap_write_access(struct file * file)
2750 struct inode *inode = file->f_mapping->host;
2752 spin_lock(&inode->i_lock);
2753 if (atomic_read(&inode->i_writecount) > 1) {
2754 spin_unlock(&inode->i_lock);
2757 atomic_set(&inode->i_writecount, -1);
2758 spin_unlock(&inode->i_lock);
2763 static int set_bitmap_file(mddev_t *mddev, int fd)
2768 if (!mddev->pers->quiesce)
2770 if (mddev->recovery || mddev->sync_thread)
2772 /* we should be able to change the bitmap.. */
2778 return -EEXIST; /* cannot add when bitmap is present */
2779 mddev->bitmap_file = fget(fd);
2781 if (mddev->bitmap_file == NULL) {
2782 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2787 err = deny_bitmap_write_access(mddev->bitmap_file);
2789 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2791 fput(mddev->bitmap_file);
2792 mddev->bitmap_file = NULL;
2795 mddev->bitmap_offset = 0; /* file overrides offset */
2796 } else if (mddev->bitmap == NULL)
2797 return -ENOENT; /* cannot remove what isn't there */
2800 mddev->pers->quiesce(mddev, 1);
2802 err = bitmap_create(mddev);
2804 bitmap_destroy(mddev);
2805 mddev->pers->quiesce(mddev, 0);
2806 } else if (fd < 0) {
2807 if (mddev->bitmap_file)
2808 fput(mddev->bitmap_file);
2809 mddev->bitmap_file = NULL;
2816 * set_array_info is used two different ways
2817 * The original usage is when creating a new array.
2818 * In this usage, raid_disks is > 0 and it together with
2819 * level, size, not_persistent,layout,chunksize determine the
2820 * shape of the array.
2821 * This will always create an array with a type-0.90.0 superblock.
2822 * The newer usage is when assembling an array.
2823 * In this case raid_disks will be 0, and the major_version field is
2824 * use to determine which style super-blocks are to be found on the devices.
2825 * The minor and patch _version numbers are also kept incase the
2826 * super_block handler wishes to interpret them.
2828 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2831 if (info->raid_disks == 0) {
2832 /* just setting version number for superblock loading */
2833 if (info->major_version < 0 ||
2834 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2835 super_types[info->major_version].name == NULL) {
2836 /* maybe try to auto-load a module? */
2838 "md: superblock version %d not known\n",
2839 info->major_version);
2842 mddev->major_version = info->major_version;
2843 mddev->minor_version = info->minor_version;
2844 mddev->patch_version = info->patch_version;
2847 mddev->major_version = MD_MAJOR_VERSION;
2848 mddev->minor_version = MD_MINOR_VERSION;
2849 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2850 mddev->ctime = get_seconds();
2852 mddev->level = info->level;
2853 mddev->size = info->size;
2854 mddev->raid_disks = info->raid_disks;
2855 /* don't set md_minor, it is determined by which /dev/md* was
2858 if (info->state & (1<<MD_SB_CLEAN))
2859 mddev->recovery_cp = MaxSector;
2861 mddev->recovery_cp = 0;
2862 mddev->persistent = ! info->not_persistent;
2864 mddev->layout = info->layout;
2865 mddev->chunk_size = info->chunk_size;
2867 mddev->max_disks = MD_SB_DISKS;
2869 mddev->sb_dirty = 1;
2872 * Generate a 128 bit UUID
2874 get_random_bytes(mddev->uuid, 16);
2880 * update_array_info is used to change the configuration of an
2882 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2883 * fields in the info are checked against the array.
2884 * Any differences that cannot be handled will cause an error.
2885 * Normally, only one change can be managed at a time.
2887 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2893 /* calculate expected state,ignoring low bits */
2894 if (mddev->bitmap && mddev->bitmap_offset)
2895 state |= (1 << MD_SB_BITMAP_PRESENT);
2897 if (mddev->major_version != info->major_version ||
2898 mddev->minor_version != info->minor_version ||
2899 /* mddev->patch_version != info->patch_version || */
2900 mddev->ctime != info->ctime ||
2901 mddev->level != info->level ||
2902 /* mddev->layout != info->layout || */
2903 !mddev->persistent != info->not_persistent||
2904 mddev->chunk_size != info->chunk_size ||
2905 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2906 ((state^info->state) & 0xfffffe00)
2909 /* Check there is only one change */
2910 if (mddev->size != info->size) cnt++;
2911 if (mddev->raid_disks != info->raid_disks) cnt++;
2912 if (mddev->layout != info->layout) cnt++;
2913 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2914 if (cnt == 0) return 0;
2915 if (cnt > 1) return -EINVAL;
2917 if (mddev->layout != info->layout) {
2919 * we don't need to do anything at the md level, the
2920 * personality will take care of it all.
2922 if (mddev->pers->reconfig == NULL)
2925 return mddev->pers->reconfig(mddev, info->layout, -1);
2927 if (mddev->size != info->size) {
2929 struct list_head *tmp;
2930 if (mddev->pers->resize == NULL)
2932 /* The "size" is the amount of each device that is used.
2933 * This can only make sense for arrays with redundancy.
2934 * linear and raid0 always use whatever space is available
2935 * We can only consider changing the size if no resync
2936 * or reconstruction is happening, and if the new size
2937 * is acceptable. It must fit before the sb_offset or,
2938 * if that is <data_offset, it must fit before the
2939 * size of each device.
2940 * If size is zero, we find the largest size that fits.
2942 if (mddev->sync_thread)
2944 ITERATE_RDEV(mddev,rdev,tmp) {
2946 int fit = (info->size == 0);
2947 if (rdev->sb_offset > rdev->data_offset)
2948 avail = (rdev->sb_offset*2) - rdev->data_offset;
2950 avail = get_capacity(rdev->bdev->bd_disk)
2951 - rdev->data_offset;
2952 if (fit && (info->size == 0 || info->size > avail/2))
2953 info->size = avail/2;
2954 if (avail < ((sector_t)info->size << 1))
2957 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2959 struct block_device *bdev;
2961 bdev = bdget_disk(mddev->gendisk, 0);
2963 down(&bdev->bd_inode->i_sem);
2964 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2965 up(&bdev->bd_inode->i_sem);
2970 if (mddev->raid_disks != info->raid_disks) {
2971 /* change the number of raid disks */
2972 if (mddev->pers->reshape == NULL)
2974 if (info->raid_disks <= 0 ||
2975 info->raid_disks >= mddev->max_disks)
2977 if (mddev->sync_thread)
2979 rv = mddev->pers->reshape(mddev, info->raid_disks);
2981 struct block_device *bdev;
2983 bdev = bdget_disk(mddev->gendisk, 0);
2985 down(&bdev->bd_inode->i_sem);
2986 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2987 up(&bdev->bd_inode->i_sem);
2992 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2993 if (mddev->pers->quiesce == NULL)
2995 if (mddev->recovery || mddev->sync_thread)
2997 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2998 /* add the bitmap */
3001 if (mddev->default_bitmap_offset == 0)
3003 mddev->bitmap_offset = mddev->default_bitmap_offset;
3004 mddev->pers->quiesce(mddev, 1);
3005 rv = bitmap_create(mddev);
3007 bitmap_destroy(mddev);
3008 mddev->pers->quiesce(mddev, 0);
3010 /* remove the bitmap */
3013 if (mddev->bitmap->file)
3015 mddev->pers->quiesce(mddev, 1);
3016 bitmap_destroy(mddev);
3017 mddev->pers->quiesce(mddev, 0);
3018 mddev->bitmap_offset = 0;
3021 md_update_sb(mddev);
3025 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3029 if (mddev->pers == NULL)
3032 rdev = find_rdev(mddev, dev);
3036 md_error(mddev, rdev);
3040 static int md_ioctl(struct inode *inode, struct file *file,
3041 unsigned int cmd, unsigned long arg)
3044 void __user *argp = (void __user *)arg;
3045 struct hd_geometry __user *loc = argp;
3046 mddev_t *mddev = NULL;
3048 if (!capable(CAP_SYS_ADMIN))
3052 * Commands dealing with the RAID driver but not any
3058 err = get_version(argp);
3061 case PRINT_RAID_DEBUG:
3069 autostart_arrays(arg);
3076 * Commands creating/starting a new array:
3079 mddev = inode->i_bdev->bd_disk->private_data;
3087 if (cmd == START_ARRAY) {
3088 /* START_ARRAY doesn't need to lock the array as autostart_array
3089 * does the locking, and it could even be a different array
3094 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3095 "This will not be supported beyond 2.6\n",
3096 current->comm, current->pid);
3099 err = autostart_array(new_decode_dev(arg));
3101 printk(KERN_WARNING "md: autostart failed!\n");
3107 err = mddev_lock(mddev);
3110 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3117 case SET_ARRAY_INFO:
3119 mdu_array_info_t info;
3121 memset(&info, 0, sizeof(info));
3122 else if (copy_from_user(&info, argp, sizeof(info))) {
3127 err = update_array_info(mddev, &info);
3129 printk(KERN_WARNING "md: couldn't update"
3130 " array info. %d\n", err);
3135 if (!list_empty(&mddev->disks)) {
3137 "md: array %s already has disks!\n",
3142 if (mddev->raid_disks) {
3144 "md: array %s already initialised!\n",
3149 err = set_array_info(mddev, &info);
3151 printk(KERN_WARNING "md: couldn't set"
3152 " array info. %d\n", err);
3162 * Commands querying/configuring an existing array:
3164 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3165 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3166 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3167 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3173 * Commands even a read-only array can execute:
3177 case GET_ARRAY_INFO:
3178 err = get_array_info(mddev, argp);
3181 case GET_BITMAP_FILE:
3182 err = get_bitmap_file(mddev, argp);
3186 err = get_disk_info(mddev, argp);
3189 case RESTART_ARRAY_RW:
3190 err = restart_array(mddev);
3194 err = do_md_stop (mddev, 0);
3198 err = do_md_stop (mddev, 1);
3202 * We have a problem here : there is no easy way to give a CHS
3203 * virtual geometry. We currently pretend that we have a 2 heads
3204 * 4 sectors (with a BIG number of cylinders...). This drives
3205 * dosfs just mad... ;-)
3212 err = put_user (2, (char __user *) &loc->heads);
3215 err = put_user (4, (char __user *) &loc->sectors);
3218 err = put_user(get_capacity(mddev->gendisk)/8,
3219 (short __user *) &loc->cylinders);
3222 err = put_user (get_start_sect(inode->i_bdev),
3223 (long __user *) &loc->start);
3228 * The remaining ioctls are changing the state of the
3229 * superblock, so we do not allow read-only arrays
3241 mdu_disk_info_t info;
3242 if (copy_from_user(&info, argp, sizeof(info)))
3245 err = add_new_disk(mddev, &info);
3249 case HOT_REMOVE_DISK:
3250 err = hot_remove_disk(mddev, new_decode_dev(arg));
3254 err = hot_add_disk(mddev, new_decode_dev(arg));
3257 case SET_DISK_FAULTY:
3258 err = set_disk_faulty(mddev, new_decode_dev(arg));
3262 err = do_md_run (mddev);
3265 case SET_BITMAP_FILE:
3266 err = set_bitmap_file(mddev, (int)arg);
3270 if (_IOC_TYPE(cmd) == MD_MAJOR)
3271 printk(KERN_WARNING "md: %s(pid %d) used"
3272 " obsolete MD ioctl, upgrade your"
3273 " software to use new ictls.\n",
3274 current->comm, current->pid);
3281 mddev_unlock(mddev);
3291 static int md_open(struct inode *inode, struct file *file)
3294 * Succeed if we can lock the mddev, which confirms that
3295 * it isn't being stopped right now.
3297 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3300 if ((err = mddev_lock(mddev)))
3305 mddev_unlock(mddev);
3307 check_disk_change(inode->i_bdev);
3312 static int md_release(struct inode *inode, struct file * file)
3314 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3323 static int md_media_changed(struct gendisk *disk)
3325 mddev_t *mddev = disk->private_data;
3327 return mddev->changed;
3330 static int md_revalidate(struct gendisk *disk)
3332 mddev_t *mddev = disk->private_data;
3337 static struct block_device_operations md_fops =
3339 .owner = THIS_MODULE,
3341 .release = md_release,
3343 .media_changed = md_media_changed,
3344 .revalidate_disk= md_revalidate,
3347 static int md_thread(void * arg)
3349 mdk_thread_t *thread = arg;
3352 * md_thread is a 'system-thread', it's priority should be very
3353 * high. We avoid resource deadlocks individually in each
3354 * raid personality. (RAID5 does preallocation) We also use RR and
3355 * the very same RT priority as kswapd, thus we will never get
3356 * into a priority inversion deadlock.
3358 * we definitely have to have equal or higher priority than
3359 * bdflush, otherwise bdflush will deadlock if there are too
3360 * many dirty RAID5 blocks.
3363 allow_signal(SIGKILL);
3364 complete(thread->event);
3365 while (!kthread_should_stop()) {
3366 void (*run)(mddev_t *);
3368 wait_event_interruptible_timeout(thread->wqueue,
3369 test_bit(THREAD_WAKEUP, &thread->flags)
3370 || kthread_should_stop(),
3374 clear_bit(THREAD_WAKEUP, &thread->flags);
3384 void md_wakeup_thread(mdk_thread_t *thread)
3387 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3388 set_bit(THREAD_WAKEUP, &thread->flags);
3389 wake_up(&thread->wqueue);
3393 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3396 mdk_thread_t *thread;
3397 struct completion event;
3399 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3403 memset(thread, 0, sizeof(mdk_thread_t));
3404 init_waitqueue_head(&thread->wqueue);
3406 init_completion(&event);
3407 thread->event = &event;
3409 thread->mddev = mddev;
3410 thread->name = name;
3411 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3412 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3413 if (IS_ERR(thread->tsk)) {
3417 wait_for_completion(&event);
3421 void md_unregister_thread(mdk_thread_t *thread)
3423 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3425 kthread_stop(thread->tsk);
3429 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3436 if (!rdev || rdev->faulty)
3439 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3441 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3442 __builtin_return_address(0),__builtin_return_address(1),
3443 __builtin_return_address(2),__builtin_return_address(3));
3445 if (!mddev->pers->error_handler)
3447 mddev->pers->error_handler(mddev,rdev);
3448 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3449 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3450 md_wakeup_thread(mddev->thread);
3453 /* seq_file implementation /proc/mdstat */
3455 static void status_unused(struct seq_file *seq)
3459 struct list_head *tmp;
3461 seq_printf(seq, "unused devices: ");
3463 ITERATE_RDEV_PENDING(rdev,tmp) {
3464 char b[BDEVNAME_SIZE];
3466 seq_printf(seq, "%s ",
3467 bdevname(rdev->bdev,b));
3470 seq_printf(seq, "<none>");
3472 seq_printf(seq, "\n");
3476 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3478 unsigned long max_blocks, resync, res, dt, db, rt;
3480 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3482 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3483 max_blocks = mddev->resync_max_sectors >> 1;
3485 max_blocks = mddev->size;
3488 * Should not happen.
3494 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3496 int i, x = res/50, y = 20-x;
3497 seq_printf(seq, "[");
3498 for (i = 0; i < x; i++)
3499 seq_printf(seq, "=");
3500 seq_printf(seq, ">");
3501 for (i = 0; i < y; i++)
3502 seq_printf(seq, ".");
3503 seq_printf(seq, "] ");
3505 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3506 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3507 "resync" : "recovery"),
3508 res/10, res % 10, resync, max_blocks);
3511 * We do not want to overflow, so the order of operands and
3512 * the * 100 / 100 trick are important. We do a +1 to be
3513 * safe against division by zero. We only estimate anyway.
3515 * dt: time from mark until now
3516 * db: blocks written from mark until now
3517 * rt: remaining time
3519 dt = ((jiffies - mddev->resync_mark) / HZ);
3521 db = resync - (mddev->resync_mark_cnt/2);
3522 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3524 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3526 seq_printf(seq, " speed=%ldK/sec", db/dt);
3529 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3531 struct list_head *tmp;
3541 spin_lock(&all_mddevs_lock);
3542 list_for_each(tmp,&all_mddevs)
3544 mddev = list_entry(tmp, mddev_t, all_mddevs);
3546 spin_unlock(&all_mddevs_lock);
3549 spin_unlock(&all_mddevs_lock);
3551 return (void*)2;/* tail */
3555 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3557 struct list_head *tmp;
3558 mddev_t *next_mddev, *mddev = v;
3564 spin_lock(&all_mddevs_lock);
3566 tmp = all_mddevs.next;
3568 tmp = mddev->all_mddevs.next;
3569 if (tmp != &all_mddevs)
3570 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3572 next_mddev = (void*)2;
3575 spin_unlock(&all_mddevs_lock);
3583 static void md_seq_stop(struct seq_file *seq, void *v)
3587 if (mddev && v != (void*)1 && v != (void*)2)
3591 static int md_seq_show(struct seq_file *seq, void *v)
3595 struct list_head *tmp2;
3598 struct bitmap *bitmap;
3600 if (v == (void*)1) {
3601 seq_printf(seq, "Personalities : ");
3602 spin_lock(&pers_lock);
3603 for (i = 0; i < MAX_PERSONALITY; i++)
3605 seq_printf(seq, "[%s] ", pers[i]->name);
3607 spin_unlock(&pers_lock);
3608 seq_printf(seq, "\n");
3611 if (v == (void*)2) {
3616 if (mddev_lock(mddev)!=0)
3618 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3619 seq_printf(seq, "%s : %sactive", mdname(mddev),
3620 mddev->pers ? "" : "in");
3623 seq_printf(seq, " (read-only)");
3624 seq_printf(seq, " %s", mddev->pers->name);
3628 ITERATE_RDEV(mddev,rdev,tmp2) {
3629 char b[BDEVNAME_SIZE];
3630 seq_printf(seq, " %s[%d]",
3631 bdevname(rdev->bdev,b), rdev->desc_nr);
3632 if (test_bit(WriteMostly, &rdev->flags))
3633 seq_printf(seq, "(W)");
3635 seq_printf(seq, "(F)");
3637 } else if (rdev->raid_disk < 0)
3638 seq_printf(seq, "(S)"); /* spare */
3642 if (!list_empty(&mddev->disks)) {
3644 seq_printf(seq, "\n %llu blocks",
3645 (unsigned long long)mddev->array_size);
3647 seq_printf(seq, "\n %llu blocks",
3648 (unsigned long long)size);
3650 if (mddev->persistent) {
3651 if (mddev->major_version != 0 ||
3652 mddev->minor_version != 90) {
3653 seq_printf(seq," super %d.%d",
3654 mddev->major_version,
3655 mddev->minor_version);
3658 seq_printf(seq, " super non-persistent");
3661 mddev->pers->status (seq, mddev);
3662 seq_printf(seq, "\n ");
3663 if (mddev->curr_resync > 2) {
3664 status_resync (seq, mddev);
3665 seq_printf(seq, "\n ");
3666 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3667 seq_printf(seq, " resync=DELAYED\n ");
3669 seq_printf(seq, "\n ");
3671 if ((bitmap = mddev->bitmap)) {
3672 unsigned long chunk_kb;
3673 unsigned long flags;
3674 spin_lock_irqsave(&bitmap->lock, flags);
3675 chunk_kb = bitmap->chunksize >> 10;
3676 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3678 bitmap->pages - bitmap->missing_pages,
3680 (bitmap->pages - bitmap->missing_pages)
3681 << (PAGE_SHIFT - 10),
3682 chunk_kb ? chunk_kb : bitmap->chunksize,
3683 chunk_kb ? "KB" : "B");
3685 seq_printf(seq, ", file: ");
3686 seq_path(seq, bitmap->file->f_vfsmnt,
3687 bitmap->file->f_dentry," \t\n");
3690 seq_printf(seq, "\n");
3691 spin_unlock_irqrestore(&bitmap->lock, flags);
3694 seq_printf(seq, "\n");
3696 mddev_unlock(mddev);
3701 static struct seq_operations md_seq_ops = {
3702 .start = md_seq_start,
3703 .next = md_seq_next,
3704 .stop = md_seq_stop,
3705 .show = md_seq_show,
3708 static int md_seq_open(struct inode *inode, struct file *file)
3712 error = seq_open(file, &md_seq_ops);
3716 static struct file_operations md_seq_fops = {
3717 .open = md_seq_open,
3719 .llseek = seq_lseek,
3720 .release = seq_release,
3723 int register_md_personality(int pnum, mdk_personality_t *p)
3725 if (pnum >= MAX_PERSONALITY) {
3727 "md: tried to install personality %s as nr %d, but max is %lu\n",
3728 p->name, pnum, MAX_PERSONALITY-1);
3732 spin_lock(&pers_lock);
3734 spin_unlock(&pers_lock);
3739 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3740 spin_unlock(&pers_lock);
3744 int unregister_md_personality(int pnum)
3746 if (pnum >= MAX_PERSONALITY)
3749 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3750 spin_lock(&pers_lock);
3752 spin_unlock(&pers_lock);
3756 static int is_mddev_idle(mddev_t *mddev)
3759 struct list_head *tmp;
3761 unsigned long curr_events;
3764 ITERATE_RDEV(mddev,rdev,tmp) {
3765 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3766 curr_events = disk_stat_read(disk, sectors[0]) +
3767 disk_stat_read(disk, sectors[1]) -
3768 atomic_read(&disk->sync_io);
3769 /* Allow some slack between valud of curr_events and last_events,
3770 * as there are some uninteresting races.
3771 * Note: the following is an unsigned comparison.
3773 if ((curr_events - rdev->last_events + 32) > 64) {
3774 rdev->last_events = curr_events;
3781 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3783 /* another "blocks" (512byte) blocks have been synced */
3784 atomic_sub(blocks, &mddev->recovery_active);
3785 wake_up(&mddev->recovery_wait);
3787 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3788 md_wakeup_thread(mddev->thread);
3789 // stop recovery, signal do_sync ....
3794 /* md_write_start(mddev, bi)
3795 * If we need to update some array metadata (e.g. 'active' flag
3796 * in superblock) before writing, schedule a superblock update
3797 * and wait for it to complete.
3799 void md_write_start(mddev_t *mddev, struct bio *bi)
3801 if (bio_data_dir(bi) != WRITE)
3804 atomic_inc(&mddev->writes_pending);
3805 if (mddev->in_sync) {
3806 spin_lock(&mddev->write_lock);
3807 if (mddev->in_sync) {
3809 mddev->sb_dirty = 1;
3810 md_wakeup_thread(mddev->thread);
3812 spin_unlock(&mddev->write_lock);
3814 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3817 void md_write_end(mddev_t *mddev)
3819 if (atomic_dec_and_test(&mddev->writes_pending)) {
3820 if (mddev->safemode == 2)
3821 md_wakeup_thread(mddev->thread);
3823 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3827 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3829 #define SYNC_MARKS 10
3830 #define SYNC_MARK_STEP (3*HZ)
3831 static void md_do_sync(mddev_t *mddev)
3834 unsigned int currspeed = 0,
3836 sector_t max_sectors,j, io_sectors;
3837 unsigned long mark[SYNC_MARKS];
3838 sector_t mark_cnt[SYNC_MARKS];
3840 struct list_head *tmp;
3841 sector_t last_check;
3844 /* just incase thread restarts... */
3845 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3848 /* we overload curr_resync somewhat here.
3849 * 0 == not engaged in resync at all
3850 * 2 == checking that there is no conflict with another sync
3851 * 1 == like 2, but have yielded to allow conflicting resync to
3853 * other == active in resync - this many blocks
3855 * Before starting a resync we must have set curr_resync to
3856 * 2, and then checked that every "conflicting" array has curr_resync
3857 * less than ours. When we find one that is the same or higher
3858 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3859 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3860 * This will mean we have to start checking from the beginning again.
3865 mddev->curr_resync = 2;
3868 if (signal_pending(current) ||
3869 kthread_should_stop()) {
3870 flush_signals(current);
3871 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3874 ITERATE_MDDEV(mddev2,tmp) {
3875 if (mddev2 == mddev)
3877 if (mddev2->curr_resync &&
3878 match_mddev_units(mddev,mddev2)) {
3880 if (mddev < mddev2 && mddev->curr_resync == 2) {
3881 /* arbitrarily yield */
3882 mddev->curr_resync = 1;
3883 wake_up(&resync_wait);
3885 if (mddev > mddev2 && mddev->curr_resync == 1)
3886 /* no need to wait here, we can wait the next
3887 * time 'round when curr_resync == 2
3890 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3891 if (!signal_pending(current) &&
3892 !kthread_should_stop() &&
3893 mddev2->curr_resync >= mddev->curr_resync) {
3894 printk(KERN_INFO "md: delaying resync of %s"
3895 " until %s has finished resync (they"
3896 " share one or more physical units)\n",
3897 mdname(mddev), mdname(mddev2));
3900 finish_wait(&resync_wait, &wq);
3903 finish_wait(&resync_wait, &wq);
3906 } while (mddev->curr_resync < 2);
3908 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3909 /* resync follows the size requested by the personality,
3910 * which defaults to physical size, but can be virtual size
3912 max_sectors = mddev->resync_max_sectors;
3913 mddev->resync_mismatches = 0;
3915 /* recovery follows the physical size of devices */
3916 max_sectors = mddev->size << 1;
3918 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3919 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3920 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3921 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3922 "(but not more than %d KB/sec) for reconstruction.\n",
3923 sysctl_speed_limit_max);
3925 is_mddev_idle(mddev); /* this also initializes IO event counters */
3926 /* we don't use the checkpoint if there's a bitmap */
3927 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3928 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3929 j = mddev->recovery_cp;
3933 for (m = 0; m < SYNC_MARKS; m++) {
3935 mark_cnt[m] = io_sectors;
3938 mddev->resync_mark = mark[last_mark];
3939 mddev->resync_mark_cnt = mark_cnt[last_mark];
3942 * Tune reconstruction:
3944 window = 32*(PAGE_SIZE/512);
3945 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3946 window/2,(unsigned long long) max_sectors/2);
3948 atomic_set(&mddev->recovery_active, 0);
3949 init_waitqueue_head(&mddev->recovery_wait);
3954 "md: resuming recovery of %s from checkpoint.\n",
3956 mddev->curr_resync = j;
3959 while (j < max_sectors) {
3963 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3964 currspeed < sysctl_speed_limit_min);
3966 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3970 if (!skipped) { /* actual IO requested */
3971 io_sectors += sectors;
3972 atomic_add(sectors, &mddev->recovery_active);
3976 if (j>1) mddev->curr_resync = j;
3979 if (last_check + window > io_sectors || j == max_sectors)
3982 last_check = io_sectors;
3984 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3985 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3989 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3991 int next = (last_mark+1) % SYNC_MARKS;
3993 mddev->resync_mark = mark[next];
3994 mddev->resync_mark_cnt = mark_cnt[next];
3995 mark[next] = jiffies;
3996 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4001 if (signal_pending(current) || kthread_should_stop()) {
4003 * got a signal, exit.
4006 "md: md_do_sync() got signal ... exiting\n");
4007 flush_signals(current);
4008 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4013 * this loop exits only if either when we are slower than
4014 * the 'hard' speed limit, or the system was IO-idle for
4016 * the system might be non-idle CPU-wise, but we only care
4017 * about not overloading the IO subsystem. (things like an
4018 * e2fsck being done on the RAID array should execute fast)
4020 mddev->queue->unplug_fn(mddev->queue);
4023 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4024 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4026 if (currspeed > sysctl_speed_limit_min) {
4027 if ((currspeed > sysctl_speed_limit_max) ||
4028 !is_mddev_idle(mddev)) {
4029 msleep_interruptible(250);
4034 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4036 * this also signals 'finished resyncing' to md_stop
4039 mddev->queue->unplug_fn(mddev->queue);
4041 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4043 /* tell personality that we are finished */
4044 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4046 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4047 mddev->curr_resync > 2 &&
4048 mddev->curr_resync >= mddev->recovery_cp) {
4049 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4051 "md: checkpointing recovery of %s.\n",
4053 mddev->recovery_cp = mddev->curr_resync;
4055 mddev->recovery_cp = MaxSector;
4059 mddev->curr_resync = 0;
4060 wake_up(&resync_wait);
4061 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4062 md_wakeup_thread(mddev->thread);
4067 * This routine is regularly called by all per-raid-array threads to
4068 * deal with generic issues like resync and super-block update.
4069 * Raid personalities that don't have a thread (linear/raid0) do not
4070 * need this as they never do any recovery or update the superblock.
4072 * It does not do any resync itself, but rather "forks" off other threads
4073 * to do that as needed.
4074 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4075 * "->recovery" and create a thread at ->sync_thread.
4076 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4077 * and wakeups up this thread which will reap the thread and finish up.
4078 * This thread also removes any faulty devices (with nr_pending == 0).
4080 * The overall approach is:
4081 * 1/ if the superblock needs updating, update it.
4082 * 2/ If a recovery thread is running, don't do anything else.
4083 * 3/ If recovery has finished, clean up, possibly marking spares active.
4084 * 4/ If there are any faulty devices, remove them.
4085 * 5/ If array is degraded, try to add spares devices
4086 * 6/ If array has spares or is not in-sync, start a resync thread.
4088 void md_check_recovery(mddev_t *mddev)
4091 struct list_head *rtmp;
4095 bitmap_daemon_work(mddev->bitmap);
4100 if (signal_pending(current)) {
4101 if (mddev->pers->sync_request) {
4102 printk(KERN_INFO "md: %s in immediate safe mode\n",
4104 mddev->safemode = 2;
4106 flush_signals(current);
4111 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4112 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4113 (mddev->safemode == 1) ||
4114 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4115 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4119 if (mddev_trylock(mddev)==0) {
4122 spin_lock(&mddev->write_lock);
4123 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4124 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4126 mddev->sb_dirty = 1;
4128 if (mddev->safemode == 1)
4129 mddev->safemode = 0;
4130 spin_unlock(&mddev->write_lock);
4132 if (mddev->sb_dirty)
4133 md_update_sb(mddev);
4136 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4137 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4138 /* resync/recovery still happening */
4139 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4142 if (mddev->sync_thread) {
4143 /* resync has finished, collect result */
4144 md_unregister_thread(mddev->sync_thread);
4145 mddev->sync_thread = NULL;
4146 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4147 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4149 /* activate any spares */
4150 mddev->pers->spare_active(mddev);
4152 md_update_sb(mddev);
4154 /* if array is no-longer degraded, then any saved_raid_disk
4155 * information must be scrapped
4157 if (!mddev->degraded)
4158 ITERATE_RDEV(mddev,rdev,rtmp)
4159 rdev->saved_raid_disk = -1;
4161 mddev->recovery = 0;
4162 /* flag recovery needed just to double check */
4163 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4166 /* Clear some bits that don't mean anything, but
4169 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4170 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4171 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4172 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4174 /* no recovery is running.
4175 * remove any failed drives, then
4176 * add spares if possible.
4177 * Spare are also removed and re-added, to allow
4178 * the personality to fail the re-add.
4180 ITERATE_RDEV(mddev,rdev,rtmp)
4181 if (rdev->raid_disk >= 0 &&
4182 (rdev->faulty || ! rdev->in_sync) &&
4183 atomic_read(&rdev->nr_pending)==0) {
4184 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4186 sprintf(nm,"rd%d", rdev->raid_disk);
4187 sysfs_remove_link(&mddev->kobj, nm);
4188 rdev->raid_disk = -1;
4192 if (mddev->degraded) {
4193 ITERATE_RDEV(mddev,rdev,rtmp)
4194 if (rdev->raid_disk < 0
4196 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4198 sprintf(nm, "rd%d", rdev->raid_disk);
4199 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4207 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4208 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4209 } else if (mddev->recovery_cp < MaxSector) {
4210 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4211 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4212 /* nothing to be done ... */
4215 if (mddev->pers->sync_request) {
4216 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4217 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4218 /* We are adding a device or devices to an array
4219 * which has the bitmap stored on all devices.
4220 * So make sure all bitmap pages get written
4222 bitmap_write_all(mddev->bitmap);
4224 mddev->sync_thread = md_register_thread(md_do_sync,
4227 if (!mddev->sync_thread) {
4228 printk(KERN_ERR "%s: could not start resync"
4231 /* leave the spares where they are, it shouldn't hurt */
4232 mddev->recovery = 0;
4234 md_wakeup_thread(mddev->sync_thread);
4238 mddev_unlock(mddev);
4242 static int md_notify_reboot(struct notifier_block *this,
4243 unsigned long code, void *x)
4245 struct list_head *tmp;
4248 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4250 printk(KERN_INFO "md: stopping all md devices.\n");
4252 ITERATE_MDDEV(mddev,tmp)
4253 if (mddev_trylock(mddev)==0)
4254 do_md_stop (mddev, 1);
4256 * certain more exotic SCSI devices are known to be
4257 * volatile wrt too early system reboots. While the
4258 * right place to handle this issue is the given
4259 * driver, we do want to have a safe RAID driver ...
4266 static struct notifier_block md_notifier = {
4267 .notifier_call = md_notify_reboot,
4269 .priority = INT_MAX, /* before any real devices */
4272 static void md_geninit(void)
4274 struct proc_dir_entry *p;
4276 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4278 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4280 p->proc_fops = &md_seq_fops;
4283 static int __init md_init(void)
4287 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4288 " MD_SB_DISKS=%d\n",
4289 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4290 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4291 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4294 if (register_blkdev(MAJOR_NR, "md"))
4296 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4297 unregister_blkdev(MAJOR_NR, "md");
4301 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4302 md_probe, NULL, NULL);
4303 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4304 md_probe, NULL, NULL);
4306 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4307 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4308 S_IFBLK|S_IRUSR|S_IWUSR,
4311 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4312 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4313 S_IFBLK|S_IRUSR|S_IWUSR,
4317 register_reboot_notifier(&md_notifier);
4318 raid_table_header = register_sysctl_table(raid_root_table, 1);
4328 * Searches all registered partitions for autorun RAID arrays
4331 static dev_t detected_devices[128];
4334 void md_autodetect_dev(dev_t dev)
4336 if (dev_cnt >= 0 && dev_cnt < 127)
4337 detected_devices[dev_cnt++] = dev;
4341 static void autostart_arrays(int part)
4346 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4348 for (i = 0; i < dev_cnt; i++) {
4349 dev_t dev = detected_devices[i];
4351 rdev = md_import_device(dev,0, 0);
4359 list_add(&rdev->same_set, &pending_raid_disks);
4363 autorun_devices(part);
4368 static __exit void md_exit(void)
4371 struct list_head *tmp;
4373 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4374 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4375 for (i=0; i < MAX_MD_DEVS; i++)
4376 devfs_remove("md/%d", i);
4377 for (i=0; i < MAX_MD_DEVS; i++)
4378 devfs_remove("md/d%d", i);
4382 unregister_blkdev(MAJOR_NR,"md");
4383 unregister_blkdev(mdp_major, "mdp");
4384 unregister_reboot_notifier(&md_notifier);
4385 unregister_sysctl_table(raid_table_header);
4386 remove_proc_entry("mdstat", NULL);
4387 ITERATE_MDDEV(mddev,tmp) {
4388 struct gendisk *disk = mddev->gendisk;
4391 export_array(mddev);
4394 mddev->gendisk = NULL;
4399 module_init(md_init)
4400 module_exit(md_exit)
4402 EXPORT_SYMBOL(register_md_personality);
4403 EXPORT_SYMBOL(unregister_md_personality);
4404 EXPORT_SYMBOL(md_error);
4405 EXPORT_SYMBOL(md_done_sync);
4406 EXPORT_SYMBOL(md_write_start);
4407 EXPORT_SYMBOL(md_write_end);
4408 EXPORT_SYMBOL(md_register_thread);
4409 EXPORT_SYMBOL(md_unregister_thread);
4410 EXPORT_SYMBOL(md_wakeup_thread);
4411 EXPORT_SYMBOL(md_print_devices);
4412 EXPORT_SYMBOL(md_check_recovery);
4413 MODULE_LICENSE("GPL");
4415 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);