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/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/devfs_fs_kernel.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/suspend.h>
45 #include <linux/init.h>
47 #include <linux/file.h>
50 #include <linux/kmod.h>
53 #include <asm/unaligned.h>
55 #define MAJOR_NR MD_MAJOR
58 /* 63 partitions with the alternate major number (mdp) */
59 #define MdpMinorShift 6
62 #define dprintk(x...) ((void)(DEBUG && printk(x)))
66 static void autostart_arrays (int part);
69 static mdk_personality_t *pers[MAX_PERSONALITY];
70 static DEFINE_SPINLOCK(pers_lock);
73 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
74 * is 1000 KB/sec, so the extra system load does not show up that much.
75 * Increase it if you want to have more _guaranteed_ speed. Note that
76 * the RAID driver will use the maximum available bandwith if the IO
77 * subsystem is idle. There is also an 'absolute maximum' reconstruction
78 * speed limit - in case reconstruction slows down your system despite
81 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84 static int sysctl_speed_limit_min = 1000;
85 static int sysctl_speed_limit_max = 200000;
87 static struct ctl_table_header *raid_table_header;
89 static ctl_table raid_table[] = {
91 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
92 .procname = "speed_limit_min",
93 .data = &sysctl_speed_limit_min,
94 .maxlen = sizeof(int),
96 .proc_handler = &proc_dointvec,
99 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
100 .procname = "speed_limit_max",
101 .data = &sysctl_speed_limit_max,
102 .maxlen = sizeof(int),
104 .proc_handler = &proc_dointvec,
109 static ctl_table raid_dir_table[] = {
111 .ctl_name = DEV_RAID,
120 static ctl_table raid_root_table[] = {
126 .child = raid_dir_table,
131 static struct block_device_operations md_fops;
134 * Enables to iterate over all existing md arrays
135 * all_mddevs_lock protects this list.
137 static LIST_HEAD(all_mddevs);
138 static DEFINE_SPINLOCK(all_mddevs_lock);
142 * iterates through all used mddevs in the system.
143 * We take care to grab the all_mddevs_lock whenever navigating
144 * the list, and to always hold a refcount when unlocked.
145 * Any code which breaks out of this loop while own
146 * a reference to the current mddev and must mddev_put it.
148 #define ITERATE_MDDEV(mddev,tmp) \
150 for (({ spin_lock(&all_mddevs_lock); \
151 tmp = all_mddevs.next; \
153 ({ if (tmp != &all_mddevs) \
154 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
155 spin_unlock(&all_mddevs_lock); \
156 if (mddev) mddev_put(mddev); \
157 mddev = list_entry(tmp, mddev_t, all_mddevs); \
158 tmp != &all_mddevs;}); \
159 ({ spin_lock(&all_mddevs_lock); \
164 static int md_fail_request (request_queue_t *q, struct bio *bio)
166 bio_io_error(bio, bio->bi_size);
170 static inline mddev_t *mddev_get(mddev_t *mddev)
172 atomic_inc(&mddev->active);
176 static void mddev_put(mddev_t *mddev)
178 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
180 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
181 list_del(&mddev->all_mddevs);
182 blk_put_queue(mddev->queue);
185 spin_unlock(&all_mddevs_lock);
188 static mddev_t * mddev_find(dev_t unit)
190 mddev_t *mddev, *new = NULL;
193 spin_lock(&all_mddevs_lock);
194 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
195 if (mddev->unit == unit) {
197 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 bio_list_init(&new->write_list);
228 spin_lock_init(&new->write_lock);
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);
261 md_wakeup_thread(mddev->thread);
264 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
267 struct list_head *tmp;
269 ITERATE_RDEV(mddev,rdev,tmp) {
270 if (rdev->desc_nr == nr)
276 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
278 struct list_head *tmp;
281 ITERATE_RDEV(mddev,rdev,tmp) {
282 if (rdev->bdev->bd_dev == dev)
288 inline static sector_t calc_dev_sboffset(struct block_device *bdev)
290 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
291 return MD_NEW_SIZE_BLOCKS(size);
294 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
298 size = rdev->sb_offset;
301 size &= ~((sector_t)chunk_size/1024 - 1);
305 static int alloc_disk_sb(mdk_rdev_t * rdev)
310 rdev->sb_page = alloc_page(GFP_KERNEL);
311 if (!rdev->sb_page) {
312 printk(KERN_ALERT "md: out of memory.\n");
319 static void free_disk_sb(mdk_rdev_t * rdev)
322 page_cache_release(rdev->sb_page);
324 rdev->sb_page = NULL;
331 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
336 complete((struct completion*)bio->bi_private);
340 static int sync_page_io(struct block_device *bdev, sector_t sector, int size,
341 struct page *page, int rw)
343 struct bio *bio = bio_alloc(GFP_NOIO, 1);
344 struct completion event;
347 rw |= (1 << BIO_RW_SYNC);
350 bio->bi_sector = sector;
351 bio_add_page(bio, page, size, 0);
352 init_completion(&event);
353 bio->bi_private = &event;
354 bio->bi_end_io = bi_complete;
356 wait_for_completion(&event);
358 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
363 static int read_disk_sb(mdk_rdev_t * rdev)
365 char b[BDEVNAME_SIZE];
366 if (!rdev->sb_page) {
374 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
380 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
381 bdevname(rdev->bdev,b));
385 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
387 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
388 (sb1->set_uuid1 == sb2->set_uuid1) &&
389 (sb1->set_uuid2 == sb2->set_uuid2) &&
390 (sb1->set_uuid3 == sb2->set_uuid3))
398 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
401 mdp_super_t *tmp1, *tmp2;
403 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
404 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
406 if (!tmp1 || !tmp2) {
408 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
416 * nr_disks is not constant
421 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
435 static unsigned int calc_sb_csum(mdp_super_t * sb)
437 unsigned int disk_csum, csum;
439 disk_csum = sb->sb_csum;
441 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
442 sb->sb_csum = disk_csum;
448 * Handle superblock details.
449 * We want to be able to handle multiple superblock formats
450 * so we have a common interface to them all, and an array of
451 * different handlers.
452 * We rely on user-space to write the initial superblock, and support
453 * reading and updating of superblocks.
454 * Interface methods are:
455 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
456 * loads and validates a superblock on dev.
457 * if refdev != NULL, compare superblocks on both devices
459 * 0 - dev has a superblock that is compatible with refdev
460 * 1 - dev has a superblock that is compatible and newer than refdev
461 * so dev should be used as the refdev in future
462 * -EINVAL superblock incompatible or invalid
463 * -othererror e.g. -EIO
465 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
466 * Verify that dev is acceptable into mddev.
467 * The first time, mddev->raid_disks will be 0, and data from
468 * dev should be merged in. Subsequent calls check that dev
469 * is new enough. Return 0 or -EINVAL
471 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
472 * Update the superblock for rdev with data in mddev
473 * This does not write to disc.
479 struct module *owner;
480 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
481 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
482 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
486 * load_super for 0.90.0
488 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
490 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
496 * Calculate the position of the superblock,
497 * it's at the end of the disk.
499 * It also happens to be a multiple of 4Kb.
501 sb_offset = calc_dev_sboffset(rdev->bdev);
502 rdev->sb_offset = sb_offset;
504 ret = read_disk_sb(rdev);
509 bdevname(rdev->bdev, b);
510 sb = (mdp_super_t*)page_address(rdev->sb_page);
512 if (sb->md_magic != MD_SB_MAGIC) {
513 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
518 if (sb->major_version != 0 ||
519 sb->minor_version != 90) {
520 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
521 sb->major_version, sb->minor_version,
526 if (sb->raid_disks <= 0)
529 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
530 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
535 rdev->preferred_minor = sb->md_minor;
536 rdev->data_offset = 0;
538 if (sb->level == LEVEL_MULTIPATH)
541 rdev->desc_nr = sb->this_disk.number;
547 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
548 if (!uuid_equal(refsb, sb)) {
549 printk(KERN_WARNING "md: %s has different UUID to %s\n",
550 b, bdevname(refdev->bdev,b2));
553 if (!sb_equal(refsb, sb)) {
554 printk(KERN_WARNING "md: %s has same UUID"
555 " but different superblock to %s\n",
556 b, bdevname(refdev->bdev, b2));
560 ev2 = md_event(refsb);
566 rdev->size = calc_dev_size(rdev, sb->chunk_size);
573 * validate_super for 0.90.0
575 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
578 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
580 if (mddev->raid_disks == 0) {
581 mddev->major_version = 0;
582 mddev->minor_version = sb->minor_version;
583 mddev->patch_version = sb->patch_version;
584 mddev->persistent = ! sb->not_persistent;
585 mddev->chunk_size = sb->chunk_size;
586 mddev->ctime = sb->ctime;
587 mddev->utime = sb->utime;
588 mddev->level = sb->level;
589 mddev->layout = sb->layout;
590 mddev->raid_disks = sb->raid_disks;
591 mddev->size = sb->size;
592 mddev->events = md_event(sb);
594 if (sb->state & (1<<MD_SB_CLEAN))
595 mddev->recovery_cp = MaxSector;
597 if (sb->events_hi == sb->cp_events_hi &&
598 sb->events_lo == sb->cp_events_lo) {
599 mddev->recovery_cp = sb->recovery_cp;
601 mddev->recovery_cp = 0;
604 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
605 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
606 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
607 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
609 mddev->max_disks = MD_SB_DISKS;
614 if (ev1 < mddev->events)
617 if (mddev->level != LEVEL_MULTIPATH) {
618 rdev->raid_disk = -1;
619 rdev->in_sync = rdev->faulty = 0;
620 desc = sb->disks + rdev->desc_nr;
622 if (desc->state & (1<<MD_DISK_FAULTY))
624 else if (desc->state & (1<<MD_DISK_SYNC) &&
625 desc->raid_disk < mddev->raid_disks) {
627 rdev->raid_disk = desc->raid_disk;
634 * sync_super for 0.90.0
636 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
639 struct list_head *tmp;
641 int next_spare = mddev->raid_disks;
643 /* make rdev->sb match mddev data..
646 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
647 * 3/ any empty disks < next_spare become removed
649 * disks[0] gets initialised to REMOVED because
650 * we cannot be sure from other fields if it has
651 * been initialised or not.
654 int active=0, working=0,failed=0,spare=0,nr_disks=0;
656 sb = (mdp_super_t*)page_address(rdev->sb_page);
658 memset(sb, 0, sizeof(*sb));
660 sb->md_magic = MD_SB_MAGIC;
661 sb->major_version = mddev->major_version;
662 sb->minor_version = mddev->minor_version;
663 sb->patch_version = mddev->patch_version;
664 sb->gvalid_words = 0; /* ignored */
665 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
666 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
667 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
668 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
670 sb->ctime = mddev->ctime;
671 sb->level = mddev->level;
672 sb->size = mddev->size;
673 sb->raid_disks = mddev->raid_disks;
674 sb->md_minor = mddev->md_minor;
675 sb->not_persistent = !mddev->persistent;
676 sb->utime = mddev->utime;
678 sb->events_hi = (mddev->events>>32);
679 sb->events_lo = (u32)mddev->events;
683 sb->recovery_cp = mddev->recovery_cp;
684 sb->cp_events_hi = (mddev->events>>32);
685 sb->cp_events_lo = (u32)mddev->events;
686 if (mddev->recovery_cp == MaxSector)
687 sb->state = (1<< MD_SB_CLEAN);
691 sb->layout = mddev->layout;
692 sb->chunk_size = mddev->chunk_size;
694 sb->disks[0].state = (1<<MD_DISK_REMOVED);
695 ITERATE_RDEV(mddev,rdev2,tmp) {
697 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
698 rdev2->desc_nr = rdev2->raid_disk;
700 rdev2->desc_nr = next_spare++;
701 d = &sb->disks[rdev2->desc_nr];
703 d->number = rdev2->desc_nr;
704 d->major = MAJOR(rdev2->bdev->bd_dev);
705 d->minor = MINOR(rdev2->bdev->bd_dev);
706 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
707 d->raid_disk = rdev2->raid_disk;
709 d->raid_disk = rdev2->desc_nr; /* compatibility */
711 d->state = (1<<MD_DISK_FAULTY);
713 } else if (rdev2->in_sync) {
714 d->state = (1<<MD_DISK_ACTIVE);
715 d->state |= (1<<MD_DISK_SYNC);
725 /* now set the "removed" and "faulty" bits on any missing devices */
726 for (i=0 ; i < mddev->raid_disks ; i++) {
727 mdp_disk_t *d = &sb->disks[i];
728 if (d->state == 0 && d->number == 0) {
731 d->state = (1<<MD_DISK_REMOVED);
732 d->state |= (1<<MD_DISK_FAULTY);
736 sb->nr_disks = nr_disks;
737 sb->active_disks = active;
738 sb->working_disks = working;
739 sb->failed_disks = failed;
740 sb->spare_disks = spare;
742 sb->this_disk = sb->disks[rdev->desc_nr];
743 sb->sb_csum = calc_sb_csum(sb);
747 * version 1 superblock
750 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
752 unsigned int disk_csum, csum;
753 unsigned long long newcsum;
754 int size = 256 + le32_to_cpu(sb->max_dev)*2;
755 unsigned int *isuper = (unsigned int*)sb;
758 disk_csum = sb->sb_csum;
761 for (i=0; size>=4; size -= 4 )
762 newcsum += le32_to_cpu(*isuper++);
765 newcsum += le16_to_cpu(*(unsigned short*) isuper);
767 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
768 sb->sb_csum = disk_csum;
769 return cpu_to_le32(csum);
772 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
774 struct mdp_superblock_1 *sb;
777 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
780 * Calculate the position of the superblock.
781 * It is always aligned to a 4K boundary and
782 * depeding on minor_version, it can be:
783 * 0: At least 8K, but less than 12K, from end of device
784 * 1: At start of device
785 * 2: 4K from start of device.
787 switch(minor_version) {
789 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
791 sb_offset &= ~(4*2-1);
792 /* convert from sectors to K */
804 rdev->sb_offset = sb_offset;
806 ret = read_disk_sb(rdev);
810 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
812 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
813 sb->major_version != cpu_to_le32(1) ||
814 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
815 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
816 sb->feature_map != 0)
819 if (calc_sb_1_csum(sb) != sb->sb_csum) {
820 printk("md: invalid superblock checksum on %s\n",
821 bdevname(rdev->bdev,b));
824 if (le64_to_cpu(sb->data_size) < 10) {
825 printk("md: data_size too small on %s\n",
826 bdevname(rdev->bdev,b));
829 rdev->preferred_minor = 0xffff;
830 rdev->data_offset = le64_to_cpu(sb->data_offset);
836 struct mdp_superblock_1 *refsb =
837 (struct mdp_superblock_1*)page_address(refdev->sb_page);
839 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
840 sb->level != refsb->level ||
841 sb->layout != refsb->layout ||
842 sb->chunksize != refsb->chunksize) {
843 printk(KERN_WARNING "md: %s has strangely different"
844 " superblock to %s\n",
845 bdevname(rdev->bdev,b),
846 bdevname(refdev->bdev,b2));
849 ev1 = le64_to_cpu(sb->events);
850 ev2 = le64_to_cpu(refsb->events);
856 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
858 rdev->size = rdev->sb_offset;
859 if (rdev->size < le64_to_cpu(sb->data_size)/2)
861 rdev->size = le64_to_cpu(sb->data_size)/2;
862 if (le32_to_cpu(sb->chunksize))
863 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
867 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
869 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
871 if (mddev->raid_disks == 0) {
872 mddev->major_version = 1;
873 mddev->patch_version = 0;
874 mddev->persistent = 1;
875 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
876 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
877 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
878 mddev->level = le32_to_cpu(sb->level);
879 mddev->layout = le32_to_cpu(sb->layout);
880 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
881 mddev->size = le64_to_cpu(sb->size)/2;
882 mddev->events = le64_to_cpu(sb->events);
884 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
885 memcpy(mddev->uuid, sb->set_uuid, 16);
887 mddev->max_disks = (4096-256)/2;
890 ev1 = le64_to_cpu(sb->events);
892 if (ev1 < mddev->events)
896 if (mddev->level != LEVEL_MULTIPATH) {
898 rdev->desc_nr = le32_to_cpu(sb->dev_number);
899 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
901 case 0xffff: /* spare */
904 rdev->raid_disk = -1;
906 case 0xfffe: /* faulty */
909 rdev->raid_disk = -1;
914 rdev->raid_disk = role;
921 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
923 struct mdp_superblock_1 *sb;
924 struct list_head *tmp;
927 /* make rdev->sb match mddev and rdev data. */
929 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
933 memset(sb->pad1, 0, sizeof(sb->pad1));
934 memset(sb->pad2, 0, sizeof(sb->pad2));
935 memset(sb->pad3, 0, sizeof(sb->pad3));
937 sb->utime = cpu_to_le64((__u64)mddev->utime);
938 sb->events = cpu_to_le64(mddev->events);
940 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
942 sb->resync_offset = cpu_to_le64(0);
945 ITERATE_RDEV(mddev,rdev2,tmp)
946 if (rdev2->desc_nr+1 > max_dev)
947 max_dev = rdev2->desc_nr+1;
949 sb->max_dev = cpu_to_le32(max_dev);
950 for (i=0; i<max_dev;i++)
951 sb->dev_roles[i] = cpu_to_le16(0xfffe);
953 ITERATE_RDEV(mddev,rdev2,tmp) {
956 sb->dev_roles[i] = cpu_to_le16(0xfffe);
957 else if (rdev2->in_sync)
958 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
960 sb->dev_roles[i] = cpu_to_le16(0xffff);
963 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
964 sb->sb_csum = calc_sb_1_csum(sb);
968 static struct super_type super_types[] = {
971 .owner = THIS_MODULE,
972 .load_super = super_90_load,
973 .validate_super = super_90_validate,
974 .sync_super = super_90_sync,
978 .owner = THIS_MODULE,
979 .load_super = super_1_load,
980 .validate_super = super_1_validate,
981 .sync_super = super_1_sync,
985 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
987 struct list_head *tmp;
990 ITERATE_RDEV(mddev,rdev,tmp)
991 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
997 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
999 struct list_head *tmp;
1002 ITERATE_RDEV(mddev1,rdev,tmp)
1003 if (match_dev_unit(mddev2, rdev))
1009 static LIST_HEAD(pending_raid_disks);
1011 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1013 mdk_rdev_t *same_pdev;
1014 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1020 same_pdev = match_dev_unit(mddev, rdev);
1023 "%s: WARNING: %s appears to be on the same physical"
1024 " disk as %s. True\n protection against single-disk"
1025 " failure might be compromised.\n",
1026 mdname(mddev), bdevname(rdev->bdev,b),
1027 bdevname(same_pdev->bdev,b2));
1029 /* Verify rdev->desc_nr is unique.
1030 * If it is -1, assign a free number, else
1031 * check number is not in use
1033 if (rdev->desc_nr < 0) {
1035 if (mddev->pers) choice = mddev->raid_disks;
1036 while (find_rdev_nr(mddev, choice))
1038 rdev->desc_nr = choice;
1040 if (find_rdev_nr(mddev, rdev->desc_nr))
1044 list_add(&rdev->same_set, &mddev->disks);
1045 rdev->mddev = mddev;
1046 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1050 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1052 char b[BDEVNAME_SIZE];
1057 list_del_init(&rdev->same_set);
1058 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1063 * prevent the device from being mounted, repartitioned or
1064 * otherwise reused by a RAID array (or any other kernel
1065 * subsystem), by bd_claiming the device.
1067 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1070 struct block_device *bdev;
1071 char b[BDEVNAME_SIZE];
1073 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1075 printk(KERN_ERR "md: could not open %s.\n",
1076 __bdevname(dev, b));
1077 return PTR_ERR(bdev);
1079 err = bd_claim(bdev, rdev);
1081 printk(KERN_ERR "md: could not bd_claim %s.\n",
1090 static void unlock_rdev(mdk_rdev_t *rdev)
1092 struct block_device *bdev = rdev->bdev;
1100 void md_autodetect_dev(dev_t dev);
1102 static void export_rdev(mdk_rdev_t * rdev)
1104 char b[BDEVNAME_SIZE];
1105 printk(KERN_INFO "md: export_rdev(%s)\n",
1106 bdevname(rdev->bdev,b));
1110 list_del_init(&rdev->same_set);
1112 md_autodetect_dev(rdev->bdev->bd_dev);
1118 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1120 unbind_rdev_from_array(rdev);
1124 static void export_array(mddev_t *mddev)
1126 struct list_head *tmp;
1129 ITERATE_RDEV(mddev,rdev,tmp) {
1134 kick_rdev_from_array(rdev);
1136 if (!list_empty(&mddev->disks))
1138 mddev->raid_disks = 0;
1139 mddev->major_version = 0;
1142 static void print_desc(mdp_disk_t *desc)
1144 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1145 desc->major,desc->minor,desc->raid_disk,desc->state);
1148 static void print_sb(mdp_super_t *sb)
1153 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1154 sb->major_version, sb->minor_version, sb->patch_version,
1155 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1157 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1158 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1159 sb->md_minor, sb->layout, sb->chunk_size);
1160 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1161 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1162 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1163 sb->failed_disks, sb->spare_disks,
1164 sb->sb_csum, (unsigned long)sb->events_lo);
1167 for (i = 0; i < MD_SB_DISKS; i++) {
1170 desc = sb->disks + i;
1171 if (desc->number || desc->major || desc->minor ||
1172 desc->raid_disk || (desc->state && (desc->state != 4))) {
1173 printk(" D %2d: ", i);
1177 printk(KERN_INFO "md: THIS: ");
1178 print_desc(&sb->this_disk);
1182 static void print_rdev(mdk_rdev_t *rdev)
1184 char b[BDEVNAME_SIZE];
1185 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1186 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1187 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1188 if (rdev->sb_loaded) {
1189 printk(KERN_INFO "md: rdev superblock:\n");
1190 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1192 printk(KERN_INFO "md: no rdev superblock!\n");
1195 void md_print_devices(void)
1197 struct list_head *tmp, *tmp2;
1200 char b[BDEVNAME_SIZE];
1203 printk("md: **********************************\n");
1204 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1205 printk("md: **********************************\n");
1206 ITERATE_MDDEV(mddev,tmp) {
1209 bitmap_print_sb(mddev->bitmap);
1211 printk("%s: ", mdname(mddev));
1212 ITERATE_RDEV(mddev,rdev,tmp2)
1213 printk("<%s>", bdevname(rdev->bdev,b));
1216 ITERATE_RDEV(mddev,rdev,tmp2)
1219 printk("md: **********************************\n");
1224 static int write_disk_sb(mdk_rdev_t * rdev)
1226 char b[BDEVNAME_SIZE];
1227 if (!rdev->sb_loaded) {
1236 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1237 bdevname(rdev->bdev,b),
1238 (unsigned long long)rdev->sb_offset);
1240 if (sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, WRITE))
1243 printk("md: write_disk_sb failed for device %s\n",
1244 bdevname(rdev->bdev,b));
1248 static void sync_sbs(mddev_t * mddev)
1251 struct list_head *tmp;
1253 ITERATE_RDEV(mddev,rdev,tmp) {
1254 super_types[mddev->major_version].
1255 sync_super(mddev, rdev);
1256 rdev->sb_loaded = 1;
1260 static void md_update_sb(mddev_t * mddev)
1262 int err, count = 100;
1263 struct list_head *tmp;
1268 spin_lock(&mddev->write_lock);
1269 sync_req = mddev->in_sync;
1270 mddev->utime = get_seconds();
1273 if (!mddev->events) {
1275 * oops, this 64-bit counter should never wrap.
1276 * Either we are in around ~1 trillion A.C., assuming
1277 * 1 reboot per second, or we have a bug:
1285 * do not write anything to disk if using
1286 * nonpersistent superblocks
1288 if (!mddev->persistent) {
1289 mddev->sb_dirty = 0;
1290 spin_unlock(&mddev->write_lock);
1293 spin_unlock(&mddev->write_lock);
1296 "md: updating %s RAID superblock on device (in sync %d)\n",
1297 mdname(mddev),mddev->in_sync);
1299 err = bitmap_update_sb(mddev->bitmap);
1300 ITERATE_RDEV(mddev,rdev,tmp) {
1301 char b[BDEVNAME_SIZE];
1302 dprintk(KERN_INFO "md: ");
1304 dprintk("(skipping faulty ");
1306 dprintk("%s ", bdevname(rdev->bdev,b));
1307 if (!rdev->faulty) {
1308 err += write_disk_sb(rdev);
1311 if (!err && mddev->level == LEVEL_MULTIPATH)
1312 /* only need to write one superblock... */
1317 printk(KERN_ERR "md: errors occurred during superblock"
1318 " update, repeating\n");
1322 "md: excessive errors occurred during superblock update, exiting\n");
1324 spin_lock(&mddev->write_lock);
1325 if (mddev->in_sync != sync_req) {
1326 /* have to write it out again */
1327 spin_unlock(&mddev->write_lock);
1330 mddev->sb_dirty = 0;
1331 spin_unlock(&mddev->write_lock);
1336 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1338 * mark the device faulty if:
1340 * - the device is nonexistent (zero size)
1341 * - the device has no valid superblock
1343 * a faulty rdev _never_ has rdev->sb set.
1345 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1347 char b[BDEVNAME_SIZE];
1352 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1354 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1355 return ERR_PTR(-ENOMEM);
1357 memset(rdev, 0, sizeof(*rdev));
1359 if ((err = alloc_disk_sb(rdev)))
1362 err = lock_rdev(rdev, newdev);
1369 rdev->data_offset = 0;
1370 atomic_set(&rdev->nr_pending, 0);
1372 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1375 "md: %s has zero or unknown size, marking faulty!\n",
1376 bdevname(rdev->bdev,b));
1381 if (super_format >= 0) {
1382 err = super_types[super_format].
1383 load_super(rdev, NULL, super_minor);
1384 if (err == -EINVAL) {
1386 "md: %s has invalid sb, not importing!\n",
1387 bdevname(rdev->bdev,b));
1392 "md: could not read %s's sb, not importing!\n",
1393 bdevname(rdev->bdev,b));
1397 INIT_LIST_HEAD(&rdev->same_set);
1402 if (rdev->sb_page) {
1408 return ERR_PTR(err);
1412 * Check a full RAID array for plausibility
1416 static void analyze_sbs(mddev_t * mddev)
1419 struct list_head *tmp;
1420 mdk_rdev_t *rdev, *freshest;
1421 char b[BDEVNAME_SIZE];
1424 ITERATE_RDEV(mddev,rdev,tmp)
1425 switch (super_types[mddev->major_version].
1426 load_super(rdev, freshest, mddev->minor_version)) {
1434 "md: fatal superblock inconsistency in %s"
1435 " -- removing from array\n",
1436 bdevname(rdev->bdev,b));
1437 kick_rdev_from_array(rdev);
1441 super_types[mddev->major_version].
1442 validate_super(mddev, freshest);
1445 ITERATE_RDEV(mddev,rdev,tmp) {
1446 if (rdev != freshest)
1447 if (super_types[mddev->major_version].
1448 validate_super(mddev, rdev)) {
1449 printk(KERN_WARNING "md: kicking non-fresh %s"
1451 bdevname(rdev->bdev,b));
1452 kick_rdev_from_array(rdev);
1455 if (mddev->level == LEVEL_MULTIPATH) {
1456 rdev->desc_nr = i++;
1457 rdev->raid_disk = rdev->desc_nr;
1464 if (mddev->recovery_cp != MaxSector &&
1466 printk(KERN_ERR "md: %s: raid array is not clean"
1467 " -- starting background reconstruction\n",
1474 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1476 static DECLARE_MUTEX(disks_sem);
1477 mddev_t *mddev = mddev_find(dev);
1478 struct gendisk *disk;
1479 int partitioned = (MAJOR(dev) != MD_MAJOR);
1480 int shift = partitioned ? MdpMinorShift : 0;
1481 int unit = MINOR(dev) >> shift;
1487 if (mddev->gendisk) {
1492 disk = alloc_disk(1 << shift);
1498 disk->major = MAJOR(dev);
1499 disk->first_minor = unit << shift;
1501 sprintf(disk->disk_name, "md_d%d", unit);
1502 sprintf(disk->devfs_name, "md/d%d", unit);
1504 sprintf(disk->disk_name, "md%d", unit);
1505 sprintf(disk->devfs_name, "md/%d", unit);
1507 disk->fops = &md_fops;
1508 disk->private_data = mddev;
1509 disk->queue = mddev->queue;
1511 mddev->gendisk = disk;
1516 void md_wakeup_thread(mdk_thread_t *thread);
1518 static void md_safemode_timeout(unsigned long data)
1520 mddev_t *mddev = (mddev_t *) data;
1522 mddev->safemode = 1;
1523 md_wakeup_thread(mddev->thread);
1527 static int do_md_run(mddev_t * mddev)
1531 struct list_head *tmp;
1533 struct gendisk *disk;
1534 char b[BDEVNAME_SIZE];
1536 if (list_empty(&mddev->disks))
1537 /* cannot run an array with no devices.. */
1544 * Analyze all RAID superblock(s)
1546 if (!mddev->raid_disks)
1549 chunk_size = mddev->chunk_size;
1550 pnum = level_to_pers(mddev->level);
1552 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1555 * 'default chunksize' in the old md code used to
1556 * be PAGE_SIZE, baaad.
1557 * we abort here to be on the safe side. We don't
1558 * want to continue the bad practice.
1561 "no chunksize specified, see 'man raidtab'\n");
1564 if (chunk_size > MAX_CHUNK_SIZE) {
1565 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1566 chunk_size, MAX_CHUNK_SIZE);
1570 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1572 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1573 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1576 if (chunk_size < PAGE_SIZE) {
1577 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1578 chunk_size, PAGE_SIZE);
1582 /* devices must have minimum size of one chunk */
1583 ITERATE_RDEV(mddev,rdev,tmp) {
1586 if (rdev->size < chunk_size / 1024) {
1588 "md: Dev %s smaller than chunk_size:"
1590 bdevname(rdev->bdev,b),
1591 (unsigned long long)rdev->size,
1601 request_module("md-personality-%d", pnum);
1606 * Drop all container device buffers, from now on
1607 * the only valid external interface is through the md
1609 * Also find largest hardsector size
1611 ITERATE_RDEV(mddev,rdev,tmp) {
1614 sync_blockdev(rdev->bdev);
1615 invalidate_bdev(rdev->bdev, 0);
1618 md_probe(mddev->unit, NULL, NULL);
1619 disk = mddev->gendisk;
1623 spin_lock(&pers_lock);
1624 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1625 spin_unlock(&pers_lock);
1626 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1631 mddev->pers = pers[pnum];
1632 spin_unlock(&pers_lock);
1634 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1636 /* before we start the array running, initialise the bitmap */
1637 err = bitmap_create(mddev);
1639 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1640 mdname(mddev), err);
1642 err = mddev->pers->run(mddev);
1644 printk(KERN_ERR "md: pers->run() failed ...\n");
1645 module_put(mddev->pers->owner);
1647 bitmap_destroy(mddev);
1650 atomic_set(&mddev->writes_pending,0);
1651 mddev->safemode = 0;
1652 mddev->safemode_timer.function = md_safemode_timeout;
1653 mddev->safemode_timer.data = (unsigned long) mddev;
1654 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1657 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1659 if (mddev->sb_dirty)
1660 md_update_sb(mddev);
1662 set_capacity(disk, mddev->array_size<<1);
1664 /* If we call blk_queue_make_request here, it will
1665 * re-initialise max_sectors etc which may have been
1666 * refined inside -> run. So just set the bits we need to set.
1667 * Most initialisation happended when we called
1668 * blk_queue_make_request(..., md_fail_request)
1671 mddev->queue->queuedata = mddev;
1672 mddev->queue->make_request_fn = mddev->pers->make_request;
1678 static int restart_array(mddev_t *mddev)
1680 struct gendisk *disk = mddev->gendisk;
1684 * Complain if it has no devices
1687 if (list_empty(&mddev->disks))
1695 mddev->safemode = 0;
1697 set_disk_ro(disk, 0);
1699 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1702 * Kick recovery or resync if necessary
1704 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1705 md_wakeup_thread(mddev->thread);
1708 printk(KERN_ERR "md: %s has no personality assigned.\n",
1717 static int do_md_stop(mddev_t * mddev, int ro)
1720 struct gendisk *disk = mddev->gendisk;
1723 if (atomic_read(&mddev->active)>2) {
1724 printk("md: %s still in use.\n",mdname(mddev));
1728 if (mddev->sync_thread) {
1729 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1730 md_unregister_thread(mddev->sync_thread);
1731 mddev->sync_thread = NULL;
1734 del_timer_sync(&mddev->safemode_timer);
1736 invalidate_partition(disk, 0);
1745 set_disk_ro(disk, 0);
1746 blk_queue_make_request(mddev->queue, md_fail_request);
1747 mddev->pers->stop(mddev);
1748 module_put(mddev->pers->owner);
1753 if (!mddev->in_sync) {
1754 /* mark array as shutdown cleanly */
1756 md_update_sb(mddev);
1759 set_disk_ro(disk, 1);
1762 bitmap_destroy(mddev);
1763 if (mddev->bitmap_file) {
1764 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1765 fput(mddev->bitmap_file);
1766 mddev->bitmap_file = NULL;
1770 * Free resources if final stop
1773 struct gendisk *disk;
1774 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1776 export_array(mddev);
1778 mddev->array_size = 0;
1779 disk = mddev->gendisk;
1781 set_capacity(disk, 0);
1784 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1791 static void autorun_array(mddev_t *mddev)
1794 struct list_head *tmp;
1797 if (list_empty(&mddev->disks))
1800 printk(KERN_INFO "md: running: ");
1802 ITERATE_RDEV(mddev,rdev,tmp) {
1803 char b[BDEVNAME_SIZE];
1804 printk("<%s>", bdevname(rdev->bdev,b));
1808 err = do_md_run (mddev);
1810 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1811 do_md_stop (mddev, 0);
1816 * lets try to run arrays based on all disks that have arrived
1817 * until now. (those are in pending_raid_disks)
1819 * the method: pick the first pending disk, collect all disks with
1820 * the same UUID, remove all from the pending list and put them into
1821 * the 'same_array' list. Then order this list based on superblock
1822 * update time (freshest comes first), kick out 'old' disks and
1823 * compare superblocks. If everything's fine then run it.
1825 * If "unit" is allocated, then bump its reference count
1827 static void autorun_devices(int part)
1829 struct list_head candidates;
1830 struct list_head *tmp;
1831 mdk_rdev_t *rdev0, *rdev;
1833 char b[BDEVNAME_SIZE];
1835 printk(KERN_INFO "md: autorun ...\n");
1836 while (!list_empty(&pending_raid_disks)) {
1838 rdev0 = list_entry(pending_raid_disks.next,
1839 mdk_rdev_t, same_set);
1841 printk(KERN_INFO "md: considering %s ...\n",
1842 bdevname(rdev0->bdev,b));
1843 INIT_LIST_HEAD(&candidates);
1844 ITERATE_RDEV_PENDING(rdev,tmp)
1845 if (super_90_load(rdev, rdev0, 0) >= 0) {
1846 printk(KERN_INFO "md: adding %s ...\n",
1847 bdevname(rdev->bdev,b));
1848 list_move(&rdev->same_set, &candidates);
1851 * now we have a set of devices, with all of them having
1852 * mostly sane superblocks. It's time to allocate the
1855 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1856 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1857 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1861 dev = MKDEV(mdp_major,
1862 rdev0->preferred_minor << MdpMinorShift);
1864 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1866 md_probe(dev, NULL, NULL);
1867 mddev = mddev_find(dev);
1870 "md: cannot allocate memory for md drive.\n");
1873 if (mddev_lock(mddev))
1874 printk(KERN_WARNING "md: %s locked, cannot run\n",
1876 else if (mddev->raid_disks || mddev->major_version
1877 || !list_empty(&mddev->disks)) {
1879 "md: %s already running, cannot run %s\n",
1880 mdname(mddev), bdevname(rdev0->bdev,b));
1881 mddev_unlock(mddev);
1883 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1884 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1885 list_del_init(&rdev->same_set);
1886 if (bind_rdev_to_array(rdev, mddev))
1889 autorun_array(mddev);
1890 mddev_unlock(mddev);
1892 /* on success, candidates will be empty, on error
1895 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1899 printk(KERN_INFO "md: ... autorun DONE.\n");
1903 * import RAID devices based on one partition
1904 * if possible, the array gets run as well.
1907 static int autostart_array(dev_t startdev)
1909 char b[BDEVNAME_SIZE];
1910 int err = -EINVAL, i;
1911 mdp_super_t *sb = NULL;
1912 mdk_rdev_t *start_rdev = NULL, *rdev;
1914 start_rdev = md_import_device(startdev, 0, 0);
1915 if (IS_ERR(start_rdev))
1919 /* NOTE: this can only work for 0.90.0 superblocks */
1920 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1921 if (sb->major_version != 0 ||
1922 sb->minor_version != 90 ) {
1923 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1924 export_rdev(start_rdev);
1928 if (start_rdev->faulty) {
1930 "md: can not autostart based on faulty %s!\n",
1931 bdevname(start_rdev->bdev,b));
1932 export_rdev(start_rdev);
1935 list_add(&start_rdev->same_set, &pending_raid_disks);
1937 for (i = 0; i < MD_SB_DISKS; i++) {
1938 mdp_disk_t *desc = sb->disks + i;
1939 dev_t dev = MKDEV(desc->major, desc->minor);
1943 if (dev == startdev)
1945 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
1947 rdev = md_import_device(dev, 0, 0);
1951 list_add(&rdev->same_set, &pending_raid_disks);
1955 * possibly return codes
1963 static int get_version(void __user * arg)
1967 ver.major = MD_MAJOR_VERSION;
1968 ver.minor = MD_MINOR_VERSION;
1969 ver.patchlevel = MD_PATCHLEVEL_VERSION;
1971 if (copy_to_user(arg, &ver, sizeof(ver)))
1977 static int get_array_info(mddev_t * mddev, void __user * arg)
1979 mdu_array_info_t info;
1980 int nr,working,active,failed,spare;
1982 struct list_head *tmp;
1984 nr=working=active=failed=spare=0;
1985 ITERATE_RDEV(mddev,rdev,tmp) {
1998 info.major_version = mddev->major_version;
1999 info.minor_version = mddev->minor_version;
2000 info.patch_version = MD_PATCHLEVEL_VERSION;
2001 info.ctime = mddev->ctime;
2002 info.level = mddev->level;
2003 info.size = mddev->size;
2005 info.raid_disks = mddev->raid_disks;
2006 info.md_minor = mddev->md_minor;
2007 info.not_persistent= !mddev->persistent;
2009 info.utime = mddev->utime;
2012 info.state = (1<<MD_SB_CLEAN);
2013 info.active_disks = active;
2014 info.working_disks = working;
2015 info.failed_disks = failed;
2016 info.spare_disks = spare;
2018 info.layout = mddev->layout;
2019 info.chunk_size = mddev->chunk_size;
2021 if (copy_to_user(arg, &info, sizeof(info)))
2027 static int get_bitmap_file(mddev_t * mddev, void * arg)
2029 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2030 char *ptr, *buf = NULL;
2033 file = kmalloc(sizeof(*file), GFP_KERNEL);
2037 /* bitmap disabled, zero the first byte and copy out */
2038 if (!mddev->bitmap || !mddev->bitmap->file) {
2039 file->pathname[0] = '\0';
2043 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2047 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2051 strcpy(file->pathname, ptr);
2055 if (copy_to_user(arg, file, sizeof(*file)))
2063 static int get_disk_info(mddev_t * mddev, void __user * arg)
2065 mdu_disk_info_t info;
2069 if (copy_from_user(&info, arg, sizeof(info)))
2074 rdev = find_rdev_nr(mddev, nr);
2076 info.major = MAJOR(rdev->bdev->bd_dev);
2077 info.minor = MINOR(rdev->bdev->bd_dev);
2078 info.raid_disk = rdev->raid_disk;
2081 info.state |= (1<<MD_DISK_FAULTY);
2082 else if (rdev->in_sync) {
2083 info.state |= (1<<MD_DISK_ACTIVE);
2084 info.state |= (1<<MD_DISK_SYNC);
2087 info.major = info.minor = 0;
2088 info.raid_disk = -1;
2089 info.state = (1<<MD_DISK_REMOVED);
2092 if (copy_to_user(arg, &info, sizeof(info)))
2098 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2100 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2102 dev_t dev = MKDEV(info->major,info->minor);
2104 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2107 if (!mddev->raid_disks) {
2109 /* expecting a device which has a superblock */
2110 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2113 "md: md_import_device returned %ld\n",
2115 return PTR_ERR(rdev);
2117 if (!list_empty(&mddev->disks)) {
2118 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2119 mdk_rdev_t, same_set);
2120 int err = super_types[mddev->major_version]
2121 .load_super(rdev, rdev0, mddev->minor_version);
2124 "md: %s has different UUID to %s\n",
2125 bdevname(rdev->bdev,b),
2126 bdevname(rdev0->bdev,b2));
2131 err = bind_rdev_to_array(rdev, mddev);
2138 * add_new_disk can be used once the array is assembled
2139 * to add "hot spares". They must already have a superblock
2144 if (!mddev->pers->hot_add_disk) {
2146 "%s: personality does not support diskops!\n",
2150 rdev = md_import_device(dev, mddev->major_version,
2151 mddev->minor_version);
2154 "md: md_import_device returned %ld\n",
2156 return PTR_ERR(rdev);
2158 rdev->in_sync = 0; /* just to be sure */
2159 rdev->raid_disk = -1;
2160 err = bind_rdev_to_array(rdev, mddev);
2164 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2166 md_wakeup_thread(mddev->thread);
2170 /* otherwise, add_new_disk is only allowed
2171 * for major_version==0 superblocks
2173 if (mddev->major_version != 0) {
2174 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2179 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2181 rdev = md_import_device (dev, -1, 0);
2184 "md: error, md_import_device() returned %ld\n",
2186 return PTR_ERR(rdev);
2188 rdev->desc_nr = info->number;
2189 if (info->raid_disk < mddev->raid_disks)
2190 rdev->raid_disk = info->raid_disk;
2192 rdev->raid_disk = -1;
2195 if (rdev->raid_disk < mddev->raid_disks)
2196 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2200 err = bind_rdev_to_array(rdev, mddev);
2206 if (!mddev->persistent) {
2207 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2208 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2210 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2211 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2213 if (!mddev->size || (mddev->size > rdev->size))
2214 mddev->size = rdev->size;
2220 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2222 char b[BDEVNAME_SIZE];
2228 rdev = find_rdev(mddev, dev);
2232 if (rdev->raid_disk >= 0)
2235 kick_rdev_from_array(rdev);
2236 md_update_sb(mddev);
2240 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2241 bdevname(rdev->bdev,b), mdname(mddev));
2245 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2247 char b[BDEVNAME_SIZE];
2255 if (mddev->major_version != 0) {
2256 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2257 " version-0 superblocks.\n",
2261 if (!mddev->pers->hot_add_disk) {
2263 "%s: personality does not support diskops!\n",
2268 rdev = md_import_device (dev, -1, 0);
2271 "md: error, md_import_device() returned %ld\n",
2276 if (mddev->persistent)
2277 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2280 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2282 size = calc_dev_size(rdev, mddev->chunk_size);
2285 if (size < mddev->size) {
2287 "%s: disk size %llu blocks < array size %llu\n",
2288 mdname(mddev), (unsigned long long)size,
2289 (unsigned long long)mddev->size);
2296 "md: can not hot-add faulty %s disk to %s!\n",
2297 bdevname(rdev->bdev,b), mdname(mddev));
2303 bind_rdev_to_array(rdev, mddev);
2306 * The rest should better be atomic, we can have disk failures
2307 * noticed in interrupt contexts ...
2310 if (rdev->desc_nr == mddev->max_disks) {
2311 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2314 goto abort_unbind_export;
2317 rdev->raid_disk = -1;
2319 md_update_sb(mddev);
2322 * Kick recovery, maybe this spare has to be added to the
2323 * array immediately.
2325 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2326 md_wakeup_thread(mddev->thread);
2330 abort_unbind_export:
2331 unbind_rdev_from_array(rdev);
2338 /* similar to deny_write_access, but accounts for our holding a reference
2339 * to the file ourselves */
2340 static int deny_bitmap_write_access(struct file * file)
2342 struct inode *inode = file->f_mapping->host;
2344 spin_lock(&inode->i_lock);
2345 if (atomic_read(&inode->i_writecount) > 1) {
2346 spin_unlock(&inode->i_lock);
2349 atomic_set(&inode->i_writecount, -1);
2350 spin_unlock(&inode->i_lock);
2355 static int set_bitmap_file(mddev_t *mddev, int fd)
2362 mddev->bitmap_file = fget(fd);
2364 if (mddev->bitmap_file == NULL) {
2365 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2370 err = deny_bitmap_write_access(mddev->bitmap_file);
2372 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2374 fput(mddev->bitmap_file);
2375 mddev->bitmap_file = NULL;
2381 * set_array_info is used two different ways
2382 * The original usage is when creating a new array.
2383 * In this usage, raid_disks is > 0 and it together with
2384 * level, size, not_persistent,layout,chunksize determine the
2385 * shape of the array.
2386 * This will always create an array with a type-0.90.0 superblock.
2387 * The newer usage is when assembling an array.
2388 * In this case raid_disks will be 0, and the major_version field is
2389 * use to determine which style super-blocks are to be found on the devices.
2390 * The minor and patch _version numbers are also kept incase the
2391 * super_block handler wishes to interpret them.
2393 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2396 if (info->raid_disks == 0) {
2397 /* just setting version number for superblock loading */
2398 if (info->major_version < 0 ||
2399 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2400 super_types[info->major_version].name == NULL) {
2401 /* maybe try to auto-load a module? */
2403 "md: superblock version %d not known\n",
2404 info->major_version);
2407 mddev->major_version = info->major_version;
2408 mddev->minor_version = info->minor_version;
2409 mddev->patch_version = info->patch_version;
2412 mddev->major_version = MD_MAJOR_VERSION;
2413 mddev->minor_version = MD_MINOR_VERSION;
2414 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2415 mddev->ctime = get_seconds();
2417 mddev->level = info->level;
2418 mddev->size = info->size;
2419 mddev->raid_disks = info->raid_disks;
2420 /* don't set md_minor, it is determined by which /dev/md* was
2423 if (info->state & (1<<MD_SB_CLEAN))
2424 mddev->recovery_cp = MaxSector;
2426 mddev->recovery_cp = 0;
2427 mddev->persistent = ! info->not_persistent;
2429 mddev->layout = info->layout;
2430 mddev->chunk_size = info->chunk_size;
2432 mddev->max_disks = MD_SB_DISKS;
2434 mddev->sb_dirty = 1;
2437 * Generate a 128 bit UUID
2439 get_random_bytes(mddev->uuid, 16);
2445 * update_array_info is used to change the configuration of an
2447 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2448 * fields in the info are checked against the array.
2449 * Any differences that cannot be handled will cause an error.
2450 * Normally, only one change can be managed at a time.
2452 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2457 if (mddev->major_version != info->major_version ||
2458 mddev->minor_version != info->minor_version ||
2459 /* mddev->patch_version != info->patch_version || */
2460 mddev->ctime != info->ctime ||
2461 mddev->level != info->level ||
2462 /* mddev->layout != info->layout || */
2463 !mddev->persistent != info->not_persistent||
2464 mddev->chunk_size != info->chunk_size )
2466 /* Check there is only one change */
2467 if (mddev->size != info->size) cnt++;
2468 if (mddev->raid_disks != info->raid_disks) cnt++;
2469 if (mddev->layout != info->layout) cnt++;
2470 if (cnt == 0) return 0;
2471 if (cnt > 1) return -EINVAL;
2473 if (mddev->layout != info->layout) {
2475 * we don't need to do anything at the md level, the
2476 * personality will take care of it all.
2478 if (mddev->pers->reconfig == NULL)
2481 return mddev->pers->reconfig(mddev, info->layout, -1);
2483 if (mddev->size != info->size) {
2485 struct list_head *tmp;
2486 if (mddev->pers->resize == NULL)
2488 /* The "size" is the amount of each device that is used.
2489 * This can only make sense for arrays with redundancy.
2490 * linear and raid0 always use whatever space is available
2491 * We can only consider changing the size if no resync
2492 * or reconstruction is happening, and if the new size
2493 * is acceptable. It must fit before the sb_offset or,
2494 * if that is <data_offset, it must fit before the
2495 * size of each device.
2496 * If size is zero, we find the largest size that fits.
2498 if (mddev->sync_thread)
2500 ITERATE_RDEV(mddev,rdev,tmp) {
2502 int fit = (info->size == 0);
2503 if (rdev->sb_offset > rdev->data_offset)
2504 avail = (rdev->sb_offset*2) - rdev->data_offset;
2506 avail = get_capacity(rdev->bdev->bd_disk)
2507 - rdev->data_offset;
2508 if (fit && (info->size == 0 || info->size > avail/2))
2509 info->size = avail/2;
2510 if (avail < ((sector_t)info->size << 1))
2513 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2515 struct block_device *bdev;
2517 bdev = bdget_disk(mddev->gendisk, 0);
2519 down(&bdev->bd_inode->i_sem);
2520 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2521 up(&bdev->bd_inode->i_sem);
2526 if (mddev->raid_disks != info->raid_disks) {
2527 /* change the number of raid disks */
2528 if (mddev->pers->reshape == NULL)
2530 if (info->raid_disks <= 0 ||
2531 info->raid_disks >= mddev->max_disks)
2533 if (mddev->sync_thread)
2535 rv = mddev->pers->reshape(mddev, info->raid_disks);
2537 struct block_device *bdev;
2539 bdev = bdget_disk(mddev->gendisk, 0);
2541 down(&bdev->bd_inode->i_sem);
2542 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2543 up(&bdev->bd_inode->i_sem);
2548 md_update_sb(mddev);
2552 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2556 if (mddev->pers == NULL)
2559 rdev = find_rdev(mddev, dev);
2563 md_error(mddev, rdev);
2567 static int md_ioctl(struct inode *inode, struct file *file,
2568 unsigned int cmd, unsigned long arg)
2571 void __user *argp = (void __user *)arg;
2572 struct hd_geometry __user *loc = argp;
2573 mddev_t *mddev = NULL;
2575 if (!capable(CAP_SYS_ADMIN))
2579 * Commands dealing with the RAID driver but not any
2585 err = get_version(argp);
2588 case PRINT_RAID_DEBUG:
2596 autostart_arrays(arg);
2603 * Commands creating/starting a new array:
2606 mddev = inode->i_bdev->bd_disk->private_data;
2614 if (cmd == START_ARRAY) {
2615 /* START_ARRAY doesn't need to lock the array as autostart_array
2616 * does the locking, and it could even be a different array
2621 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2622 "This will not be supported beyond 2.6\n",
2623 current->comm, current->pid);
2626 err = autostart_array(new_decode_dev(arg));
2628 printk(KERN_WARNING "md: autostart failed!\n");
2634 err = mddev_lock(mddev);
2637 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2644 case SET_ARRAY_INFO:
2646 mdu_array_info_t info;
2648 memset(&info, 0, sizeof(info));
2649 else if (copy_from_user(&info, argp, sizeof(info))) {
2654 err = update_array_info(mddev, &info);
2656 printk(KERN_WARNING "md: couldn't update"
2657 " array info. %d\n", err);
2662 if (!list_empty(&mddev->disks)) {
2664 "md: array %s already has disks!\n",
2669 if (mddev->raid_disks) {
2671 "md: array %s already initialised!\n",
2676 err = set_array_info(mddev, &info);
2678 printk(KERN_WARNING "md: couldn't set"
2679 " array info. %d\n", err);
2689 * Commands querying/configuring an existing array:
2691 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2692 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2693 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2694 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2700 * Commands even a read-only array can execute:
2704 case GET_ARRAY_INFO:
2705 err = get_array_info(mddev, argp);
2708 case GET_BITMAP_FILE:
2709 err = get_bitmap_file(mddev, (void *)arg);
2713 err = get_disk_info(mddev, argp);
2716 case RESTART_ARRAY_RW:
2717 err = restart_array(mddev);
2721 err = do_md_stop (mddev, 0);
2725 err = do_md_stop (mddev, 1);
2729 * We have a problem here : there is no easy way to give a CHS
2730 * virtual geometry. We currently pretend that we have a 2 heads
2731 * 4 sectors (with a BIG number of cylinders...). This drives
2732 * dosfs just mad... ;-)
2739 err = put_user (2, (char __user *) &loc->heads);
2742 err = put_user (4, (char __user *) &loc->sectors);
2745 err = put_user(get_capacity(mddev->gendisk)/8,
2746 (short __user *) &loc->cylinders);
2749 err = put_user (get_start_sect(inode->i_bdev),
2750 (long __user *) &loc->start);
2755 * The remaining ioctls are changing the state of the
2756 * superblock, so we do not allow read-only arrays
2768 mdu_disk_info_t info;
2769 if (copy_from_user(&info, argp, sizeof(info)))
2772 err = add_new_disk(mddev, &info);
2776 case HOT_REMOVE_DISK:
2777 err = hot_remove_disk(mddev, new_decode_dev(arg));
2781 err = hot_add_disk(mddev, new_decode_dev(arg));
2784 case SET_DISK_FAULTY:
2785 err = set_disk_faulty(mddev, new_decode_dev(arg));
2789 err = do_md_run (mddev);
2792 case SET_BITMAP_FILE:
2793 err = set_bitmap_file(mddev, (int)arg);
2797 if (_IOC_TYPE(cmd) == MD_MAJOR)
2798 printk(KERN_WARNING "md: %s(pid %d) used"
2799 " obsolete MD ioctl, upgrade your"
2800 " software to use new ictls.\n",
2801 current->comm, current->pid);
2808 mddev_unlock(mddev);
2818 static int md_open(struct inode *inode, struct file *file)
2821 * Succeed if we can lock the mddev, which confirms that
2822 * it isn't being stopped right now.
2824 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2827 if ((err = mddev_lock(mddev)))
2832 mddev_unlock(mddev);
2834 check_disk_change(inode->i_bdev);
2839 static int md_release(struct inode *inode, struct file * file)
2841 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2850 static int md_media_changed(struct gendisk *disk)
2852 mddev_t *mddev = disk->private_data;
2854 return mddev->changed;
2857 static int md_revalidate(struct gendisk *disk)
2859 mddev_t *mddev = disk->private_data;
2864 static struct block_device_operations md_fops =
2866 .owner = THIS_MODULE,
2868 .release = md_release,
2870 .media_changed = md_media_changed,
2871 .revalidate_disk= md_revalidate,
2874 static int md_thread(void * arg)
2876 mdk_thread_t *thread = arg;
2884 daemonize(thread->name, mdname(thread->mddev));
2886 current->exit_signal = SIGCHLD;
2887 allow_signal(SIGKILL);
2888 thread->tsk = current;
2891 * md_thread is a 'system-thread', it's priority should be very
2892 * high. We avoid resource deadlocks individually in each
2893 * raid personality. (RAID5 does preallocation) We also use RR and
2894 * the very same RT priority as kswapd, thus we will never get
2895 * into a priority inversion deadlock.
2897 * we definitely have to have equal or higher priority than
2898 * bdflush, otherwise bdflush will deadlock if there are too
2899 * many dirty RAID5 blocks.
2903 complete(thread->event);
2904 while (thread->run) {
2905 void (*run)(mddev_t *);
2907 wait_event_interruptible_timeout(thread->wqueue,
2908 test_bit(THREAD_WAKEUP, &thread->flags),
2910 if (current->flags & PF_FREEZE)
2911 refrigerator(PF_FREEZE);
2913 clear_bit(THREAD_WAKEUP, &thread->flags);
2919 if (signal_pending(current))
2920 flush_signals(current);
2922 complete(thread->event);
2926 void md_wakeup_thread(mdk_thread_t *thread)
2929 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2930 set_bit(THREAD_WAKEUP, &thread->flags);
2931 wake_up(&thread->wqueue);
2935 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
2938 mdk_thread_t *thread;
2940 struct completion event;
2942 thread = (mdk_thread_t *) kmalloc
2943 (sizeof(mdk_thread_t), GFP_KERNEL);
2947 memset(thread, 0, sizeof(mdk_thread_t));
2948 init_waitqueue_head(&thread->wqueue);
2950 init_completion(&event);
2951 thread->event = &event;
2953 thread->mddev = mddev;
2954 thread->name = name;
2955 thread->timeout = MAX_SCHEDULE_TIMEOUT;
2956 ret = kernel_thread(md_thread, thread, 0);
2961 wait_for_completion(&event);
2965 void md_unregister_thread(mdk_thread_t *thread)
2967 struct completion event;
2969 init_completion(&event);
2971 thread->event = &event;
2973 /* As soon as ->run is set to NULL, the task could disappear,
2974 * so we need to hold tasklist_lock until we have sent the signal
2976 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
2977 read_lock(&tasklist_lock);
2979 send_sig(SIGKILL, thread->tsk, 1);
2980 read_unlock(&tasklist_lock);
2981 wait_for_completion(&event);
2985 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
2992 if (!rdev || rdev->faulty)
2995 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
2997 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
2998 __builtin_return_address(0),__builtin_return_address(1),
2999 __builtin_return_address(2),__builtin_return_address(3));
3001 if (!mddev->pers->error_handler)
3003 mddev->pers->error_handler(mddev,rdev);
3004 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3005 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3006 md_wakeup_thread(mddev->thread);
3009 /* seq_file implementation /proc/mdstat */
3011 static void status_unused(struct seq_file *seq)
3015 struct list_head *tmp;
3017 seq_printf(seq, "unused devices: ");
3019 ITERATE_RDEV_PENDING(rdev,tmp) {
3020 char b[BDEVNAME_SIZE];
3022 seq_printf(seq, "%s ",
3023 bdevname(rdev->bdev,b));
3026 seq_printf(seq, "<none>");
3028 seq_printf(seq, "\n");
3032 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3034 unsigned long max_blocks, resync, res, dt, db, rt;
3036 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3038 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3039 max_blocks = mddev->resync_max_sectors >> 1;
3041 max_blocks = mddev->size;
3044 * Should not happen.
3050 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3052 int i, x = res/50, y = 20-x;
3053 seq_printf(seq, "[");
3054 for (i = 0; i < x; i++)
3055 seq_printf(seq, "=");
3056 seq_printf(seq, ">");
3057 for (i = 0; i < y; i++)
3058 seq_printf(seq, ".");
3059 seq_printf(seq, "] ");
3061 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3062 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3063 "resync" : "recovery"),
3064 res/10, res % 10, resync, max_blocks);
3067 * We do not want to overflow, so the order of operands and
3068 * the * 100 / 100 trick are important. We do a +1 to be
3069 * safe against division by zero. We only estimate anyway.
3071 * dt: time from mark until now
3072 * db: blocks written from mark until now
3073 * rt: remaining time
3075 dt = ((jiffies - mddev->resync_mark) / HZ);
3077 db = resync - (mddev->resync_mark_cnt/2);
3078 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3080 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3082 seq_printf(seq, " speed=%ldK/sec", db/dt);
3085 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3087 struct list_head *tmp;
3097 spin_lock(&all_mddevs_lock);
3098 list_for_each(tmp,&all_mddevs)
3100 mddev = list_entry(tmp, mddev_t, all_mddevs);
3102 spin_unlock(&all_mddevs_lock);
3105 spin_unlock(&all_mddevs_lock);
3107 return (void*)2;/* tail */
3111 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3113 struct list_head *tmp;
3114 mddev_t *next_mddev, *mddev = v;
3120 spin_lock(&all_mddevs_lock);
3122 tmp = all_mddevs.next;
3124 tmp = mddev->all_mddevs.next;
3125 if (tmp != &all_mddevs)
3126 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3128 next_mddev = (void*)2;
3131 spin_unlock(&all_mddevs_lock);
3139 static void md_seq_stop(struct seq_file *seq, void *v)
3143 if (mddev && v != (void*)1 && v != (void*)2)
3147 static int md_seq_show(struct seq_file *seq, void *v)
3151 struct list_head *tmp2;
3154 struct bitmap *bitmap;
3156 if (v == (void*)1) {
3157 seq_printf(seq, "Personalities : ");
3158 spin_lock(&pers_lock);
3159 for (i = 0; i < MAX_PERSONALITY; i++)
3161 seq_printf(seq, "[%s] ", pers[i]->name);
3163 spin_unlock(&pers_lock);
3164 seq_printf(seq, "\n");
3167 if (v == (void*)2) {
3172 if (mddev_lock(mddev)!=0)
3174 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3175 seq_printf(seq, "%s : %sactive", mdname(mddev),
3176 mddev->pers ? "" : "in");
3179 seq_printf(seq, " (read-only)");
3180 seq_printf(seq, " %s", mddev->pers->name);
3184 ITERATE_RDEV(mddev,rdev,tmp2) {
3185 char b[BDEVNAME_SIZE];
3186 seq_printf(seq, " %s[%d]",
3187 bdevname(rdev->bdev,b), rdev->desc_nr);
3189 seq_printf(seq, "(F)");
3195 if (!list_empty(&mddev->disks)) {
3197 seq_printf(seq, "\n %llu blocks",
3198 (unsigned long long)mddev->array_size);
3200 seq_printf(seq, "\n %llu blocks",
3201 (unsigned long long)size);
3205 mddev->pers->status (seq, mddev);
3206 seq_printf(seq, "\n ");
3207 if (mddev->curr_resync > 2) {
3208 status_resync (seq, mddev);
3209 seq_printf(seq, "\n ");
3210 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3211 seq_printf(seq, " resync=DELAYED\n ");
3213 seq_printf(seq, "\n ");
3215 if ((bitmap = mddev->bitmap)) {
3216 unsigned long chunk_kb;
3217 unsigned long flags;
3218 spin_lock_irqsave(&bitmap->lock, flags);
3219 chunk_kb = bitmap->chunksize >> 10;
3220 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3222 bitmap->pages - bitmap->missing_pages,
3224 (bitmap->pages - bitmap->missing_pages)
3225 << (PAGE_SHIFT - 10),
3226 chunk_kb ? chunk_kb : bitmap->chunksize,
3227 chunk_kb ? "KB" : "B");
3229 seq_printf(seq, ", file: ");
3230 seq_path(seq, bitmap->file->f_vfsmnt,
3231 bitmap->file->f_dentry," \t\n");
3234 seq_printf(seq, "\n");
3235 spin_unlock_irqrestore(&bitmap->lock, flags);
3238 seq_printf(seq, "\n");
3240 mddev_unlock(mddev);
3245 static struct seq_operations md_seq_ops = {
3246 .start = md_seq_start,
3247 .next = md_seq_next,
3248 .stop = md_seq_stop,
3249 .show = md_seq_show,
3252 static int md_seq_open(struct inode *inode, struct file *file)
3256 error = seq_open(file, &md_seq_ops);
3260 static struct file_operations md_seq_fops = {
3261 .open = md_seq_open,
3263 .llseek = seq_lseek,
3264 .release = seq_release,
3267 int register_md_personality(int pnum, mdk_personality_t *p)
3269 if (pnum >= MAX_PERSONALITY) {
3271 "md: tried to install personality %s as nr %d, but max is %lu\n",
3272 p->name, pnum, MAX_PERSONALITY-1);
3276 spin_lock(&pers_lock);
3278 spin_unlock(&pers_lock);
3283 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3284 spin_unlock(&pers_lock);
3288 int unregister_md_personality(int pnum)
3290 if (pnum >= MAX_PERSONALITY)
3293 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3294 spin_lock(&pers_lock);
3296 spin_unlock(&pers_lock);
3300 static int is_mddev_idle(mddev_t *mddev)
3303 struct list_head *tmp;
3305 unsigned long curr_events;
3308 ITERATE_RDEV(mddev,rdev,tmp) {
3309 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3310 curr_events = disk_stat_read(disk, read_sectors) +
3311 disk_stat_read(disk, write_sectors) -
3312 atomic_read(&disk->sync_io);
3313 /* Allow some slack between valud of curr_events and last_events,
3314 * as there are some uninteresting races.
3315 * Note: the following is an unsigned comparison.
3317 if ((curr_events - rdev->last_events + 32) > 64) {
3318 rdev->last_events = curr_events;
3325 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3327 /* another "blocks" (512byte) blocks have been synced */
3328 atomic_sub(blocks, &mddev->recovery_active);
3329 wake_up(&mddev->recovery_wait);
3331 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3332 md_wakeup_thread(mddev->thread);
3333 // stop recovery, signal do_sync ....
3338 /* md_write_start(mddev, bi)
3339 * If we need to update some array metadata (e.g. 'active' flag
3340 * in superblock) before writing, queue bi for later writing
3341 * and return 0, else return 1 and it will be written now
3343 int md_write_start(mddev_t *mddev, struct bio *bi)
3345 if (bio_data_dir(bi) != WRITE)
3348 atomic_inc(&mddev->writes_pending);
3349 spin_lock(&mddev->write_lock);
3350 if (mddev->in_sync == 0 && mddev->sb_dirty == 0) {
3351 spin_unlock(&mddev->write_lock);
3354 bio_list_add(&mddev->write_list, bi);
3356 if (mddev->in_sync) {
3358 mddev->sb_dirty = 1;
3360 spin_unlock(&mddev->write_lock);
3361 md_wakeup_thread(mddev->thread);
3365 void md_write_end(mddev_t *mddev)
3367 if (atomic_dec_and_test(&mddev->writes_pending)) {
3368 if (mddev->safemode == 2)
3369 md_wakeup_thread(mddev->thread);
3371 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3375 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3377 #define SYNC_MARKS 10
3378 #define SYNC_MARK_STEP (3*HZ)
3379 static void md_do_sync(mddev_t *mddev)
3382 unsigned int currspeed = 0,
3384 sector_t max_sectors,j, io_sectors;
3385 unsigned long mark[SYNC_MARKS];
3386 sector_t mark_cnt[SYNC_MARKS];
3388 struct list_head *tmp;
3389 sector_t last_check;
3392 /* just incase thread restarts... */
3393 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3396 /* we overload curr_resync somewhat here.
3397 * 0 == not engaged in resync at all
3398 * 2 == checking that there is no conflict with another sync
3399 * 1 == like 2, but have yielded to allow conflicting resync to
3401 * other == active in resync - this many blocks
3403 * Before starting a resync we must have set curr_resync to
3404 * 2, and then checked that every "conflicting" array has curr_resync
3405 * less than ours. When we find one that is the same or higher
3406 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3407 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3408 * This will mean we have to start checking from the beginning again.
3413 mddev->curr_resync = 2;
3416 if (signal_pending(current)) {
3417 flush_signals(current);
3420 ITERATE_MDDEV(mddev2,tmp) {
3422 if (mddev2 == mddev)
3424 if (mddev2->curr_resync &&
3425 match_mddev_units(mddev,mddev2)) {
3427 if (mddev < mddev2 && mddev->curr_resync == 2) {
3428 /* arbitrarily yield */
3429 mddev->curr_resync = 1;
3430 wake_up(&resync_wait);
3432 if (mddev > mddev2 && mddev->curr_resync == 1)
3433 /* no need to wait here, we can wait the next
3434 * time 'round when curr_resync == 2
3437 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3438 if (!signal_pending(current)
3439 && mddev2->curr_resync >= mddev->curr_resync) {
3440 printk(KERN_INFO "md: delaying resync of %s"
3441 " until %s has finished resync (they"
3442 " share one or more physical units)\n",
3443 mdname(mddev), mdname(mddev2));
3446 finish_wait(&resync_wait, &wq);
3449 finish_wait(&resync_wait, &wq);
3452 } while (mddev->curr_resync < 2);
3454 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3455 /* resync follows the size requested by the personality,
3456 * which defaults to physical size, but can be virtual size
3458 max_sectors = mddev->resync_max_sectors;
3460 /* recovery follows the physical size of devices */
3461 max_sectors = mddev->size << 1;
3463 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3464 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3465 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3466 printk(KERN_INFO "md: using maximum available idle IO bandwith "
3467 "(but not more than %d KB/sec) for reconstruction.\n",
3468 sysctl_speed_limit_max);
3470 is_mddev_idle(mddev); /* this also initializes IO event counters */
3471 /* we don't use the checkpoint if there's a bitmap */
3472 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3473 j = mddev->recovery_cp;
3477 for (m = 0; m < SYNC_MARKS; m++) {
3479 mark_cnt[m] = io_sectors;
3482 mddev->resync_mark = mark[last_mark];
3483 mddev->resync_mark_cnt = mark_cnt[last_mark];
3486 * Tune reconstruction:
3488 window = 32*(PAGE_SIZE/512);
3489 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3490 window/2,(unsigned long long) max_sectors/2);
3492 atomic_set(&mddev->recovery_active, 0);
3493 init_waitqueue_head(&mddev->recovery_wait);
3498 "md: resuming recovery of %s from checkpoint.\n",
3500 mddev->curr_resync = j;
3503 while (j < max_sectors) {
3507 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3508 currspeed < sysctl_speed_limit_min);
3510 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3514 if (!skipped) { /* actual IO requested */
3515 io_sectors += sectors;
3516 atomic_add(sectors, &mddev->recovery_active);
3520 if (j>1) mddev->curr_resync = j;
3523 if (last_check + window > io_sectors || j == max_sectors)
3526 last_check = io_sectors;
3528 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3529 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3533 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3535 int next = (last_mark+1) % SYNC_MARKS;
3537 mddev->resync_mark = mark[next];
3538 mddev->resync_mark_cnt = mark_cnt[next];
3539 mark[next] = jiffies;
3540 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3545 if (signal_pending(current)) {
3547 * got a signal, exit.
3550 "md: md_do_sync() got signal ... exiting\n");
3551 flush_signals(current);
3552 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3557 * this loop exits only if either when we are slower than
3558 * the 'hard' speed limit, or the system was IO-idle for
3560 * the system might be non-idle CPU-wise, but we only care
3561 * about not overloading the IO subsystem. (things like an
3562 * e2fsck being done on the RAID array should execute fast)
3564 mddev->queue->unplug_fn(mddev->queue);
3567 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3568 /((jiffies-mddev->resync_mark)/HZ +1) +1;
3570 if (currspeed > sysctl_speed_limit_min) {
3571 if ((currspeed > sysctl_speed_limit_max) ||
3572 !is_mddev_idle(mddev)) {
3573 msleep_interruptible(250);
3578 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3580 * this also signals 'finished resyncing' to md_stop
3583 mddev->queue->unplug_fn(mddev->queue);
3585 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3587 /* tell personality that we are finished */
3588 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3590 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3591 mddev->curr_resync > 2 &&
3592 mddev->curr_resync >= mddev->recovery_cp) {
3593 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3595 "md: checkpointing recovery of %s.\n",
3597 mddev->recovery_cp = mddev->curr_resync;
3599 mddev->recovery_cp = MaxSector;
3603 mddev->curr_resync = 0;
3604 wake_up(&resync_wait);
3605 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3606 md_wakeup_thread(mddev->thread);
3611 * This routine is regularly called by all per-raid-array threads to
3612 * deal with generic issues like resync and super-block update.
3613 * Raid personalities that don't have a thread (linear/raid0) do not
3614 * need this as they never do any recovery or update the superblock.
3616 * It does not do any resync itself, but rather "forks" off other threads
3617 * to do that as needed.
3618 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3619 * "->recovery" and create a thread at ->sync_thread.
3620 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3621 * and wakeups up this thread which will reap the thread and finish up.
3622 * This thread also removes any faulty devices (with nr_pending == 0).
3624 * The overall approach is:
3625 * 1/ if the superblock needs updating, update it.
3626 * 2/ If a recovery thread is running, don't do anything else.
3627 * 3/ If recovery has finished, clean up, possibly marking spares active.
3628 * 4/ If there are any faulty devices, remove them.
3629 * 5/ If array is degraded, try to add spares devices
3630 * 6/ If array has spares or is not in-sync, start a resync thread.
3632 void md_check_recovery(mddev_t *mddev)
3635 struct list_head *rtmp;
3638 dprintk(KERN_INFO "md: recovery thread got woken up ...\n");
3643 if (signal_pending(current)) {
3644 if (mddev->pers->sync_request) {
3645 printk(KERN_INFO "md: %s in immediate safe mode\n",
3647 mddev->safemode = 2;
3649 flush_signals(current);
3654 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3655 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3656 mddev->write_list.head ||
3657 (mddev->safemode == 1) ||
3658 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3659 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3663 if (mddev_trylock(mddev)==0) {
3667 spin_lock(&mddev->write_lock);
3668 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3669 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3671 mddev->sb_dirty = 1;
3673 if (mddev->safemode == 1)
3674 mddev->safemode = 0;
3675 blist = bio_list_get(&mddev->write_list);
3676 spin_unlock(&mddev->write_lock);
3678 if (mddev->sb_dirty)
3679 md_update_sb(mddev);
3682 struct bio *b = blist;
3683 blist = blist->bi_next;
3685 generic_make_request(b);
3686 /* we already counted this, so need to un-count */
3687 md_write_end(mddev);
3691 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3692 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3693 /* resync/recovery still happening */
3694 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3697 if (mddev->sync_thread) {
3698 /* resync has finished, collect result */
3699 md_unregister_thread(mddev->sync_thread);
3700 mddev->sync_thread = NULL;
3701 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3702 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3704 /* activate any spares */
3705 mddev->pers->spare_active(mddev);
3707 md_update_sb(mddev);
3708 mddev->recovery = 0;
3709 /* flag recovery needed just to double check */
3710 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3713 if (mddev->recovery)
3714 /* probably just the RECOVERY_NEEDED flag */
3715 mddev->recovery = 0;
3717 /* no recovery is running.
3718 * remove any failed drives, then
3719 * add spares if possible.
3720 * Spare are also removed and re-added, to allow
3721 * the personality to fail the re-add.
3723 ITERATE_RDEV(mddev,rdev,rtmp)
3724 if (rdev->raid_disk >= 0 &&
3725 (rdev->faulty || ! rdev->in_sync) &&
3726 atomic_read(&rdev->nr_pending)==0) {
3727 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3728 rdev->raid_disk = -1;
3731 if (mddev->degraded) {
3732 ITERATE_RDEV(mddev,rdev,rtmp)
3733 if (rdev->raid_disk < 0
3735 if (mddev->pers->hot_add_disk(mddev,rdev))
3742 if (!spares && (mddev->recovery_cp == MaxSector )) {
3743 /* nothing we can do ... */
3746 if (mddev->pers->sync_request) {
3747 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3749 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3750 mddev->sync_thread = md_register_thread(md_do_sync,
3753 if (!mddev->sync_thread) {
3754 printk(KERN_ERR "%s: could not start resync"
3757 /* leave the spares where they are, it shouldn't hurt */
3758 mddev->recovery = 0;
3760 md_wakeup_thread(mddev->sync_thread);
3764 mddev_unlock(mddev);
3768 static int md_notify_reboot(struct notifier_block *this,
3769 unsigned long code, void *x)
3771 struct list_head *tmp;
3774 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3776 printk(KERN_INFO "md: stopping all md devices.\n");
3778 ITERATE_MDDEV(mddev,tmp)
3779 if (mddev_trylock(mddev)==0)
3780 do_md_stop (mddev, 1);
3782 * certain more exotic SCSI devices are known to be
3783 * volatile wrt too early system reboots. While the
3784 * right place to handle this issue is the given
3785 * driver, we do want to have a safe RAID driver ...
3792 static struct notifier_block md_notifier = {
3793 .notifier_call = md_notify_reboot,
3795 .priority = INT_MAX, /* before any real devices */
3798 static void md_geninit(void)
3800 struct proc_dir_entry *p;
3802 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3804 p = create_proc_entry("mdstat", S_IRUGO, NULL);
3806 p->proc_fops = &md_seq_fops;
3809 static int __init md_init(void)
3813 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3814 " MD_SB_DISKS=%d\n",
3815 MD_MAJOR_VERSION, MD_MINOR_VERSION,
3816 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3817 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3820 if (register_blkdev(MAJOR_NR, "md"))
3822 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3823 unregister_blkdev(MAJOR_NR, "md");
3827 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3828 md_probe, NULL, NULL);
3829 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3830 md_probe, NULL, NULL);
3832 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3833 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3834 S_IFBLK|S_IRUSR|S_IWUSR,
3837 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3838 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3839 S_IFBLK|S_IRUSR|S_IWUSR,
3843 register_reboot_notifier(&md_notifier);
3844 raid_table_header = register_sysctl_table(raid_root_table, 1);
3854 * Searches all registered partitions for autorun RAID arrays
3857 static dev_t detected_devices[128];
3860 void md_autodetect_dev(dev_t dev)
3862 if (dev_cnt >= 0 && dev_cnt < 127)
3863 detected_devices[dev_cnt++] = dev;
3867 static void autostart_arrays(int part)
3872 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3874 for (i = 0; i < dev_cnt; i++) {
3875 dev_t dev = detected_devices[i];
3877 rdev = md_import_device(dev,0, 0);
3885 list_add(&rdev->same_set, &pending_raid_disks);
3889 autorun_devices(part);
3894 static __exit void md_exit(void)
3897 struct list_head *tmp;
3899 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3900 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3901 for (i=0; i < MAX_MD_DEVS; i++)
3902 devfs_remove("md/%d", i);
3903 for (i=0; i < MAX_MD_DEVS; i++)
3904 devfs_remove("md/d%d", i);
3908 unregister_blkdev(MAJOR_NR,"md");
3909 unregister_blkdev(mdp_major, "mdp");
3910 unregister_reboot_notifier(&md_notifier);
3911 unregister_sysctl_table(raid_table_header);
3912 remove_proc_entry("mdstat", NULL);
3913 ITERATE_MDDEV(mddev,tmp) {
3914 struct gendisk *disk = mddev->gendisk;
3917 export_array(mddev);
3920 mddev->gendisk = NULL;
3925 module_init(md_init)
3926 module_exit(md_exit)
3928 EXPORT_SYMBOL(register_md_personality);
3929 EXPORT_SYMBOL(unregister_md_personality);
3930 EXPORT_SYMBOL(md_error);
3931 EXPORT_SYMBOL(md_done_sync);
3932 EXPORT_SYMBOL(md_write_start);
3933 EXPORT_SYMBOL(md_write_end);
3934 EXPORT_SYMBOL(md_register_thread);
3935 EXPORT_SYMBOL(md_unregister_thread);
3936 EXPORT_SYMBOL(md_wakeup_thread);
3937 EXPORT_SYMBOL(md_print_devices);
3938 EXPORT_SYMBOL(md_check_recovery);
3939 MODULE_LICENSE("GPL");