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);
203 list_add(&new->all_mddevs, &all_mddevs);
204 spin_unlock(&all_mddevs_lock);
207 spin_unlock(&all_mddevs_lock);
209 new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
213 memset(new, 0, sizeof(*new));
216 if (MAJOR(unit) == MD_MAJOR)
217 new->md_minor = MINOR(unit);
219 new->md_minor = MINOR(unit) >> MdpMinorShift;
221 init_MUTEX(&new->reconfig_sem);
222 INIT_LIST_HEAD(&new->disks);
223 INIT_LIST_HEAD(&new->all_mddevs);
224 init_timer(&new->safemode_timer);
225 atomic_set(&new->active, 1);
226 spin_lock_init(&new->write_lock);
227 init_waitqueue_head(&new->sb_wait);
229 new->queue = blk_alloc_queue(GFP_KERNEL);
235 blk_queue_make_request(new->queue, md_fail_request);
240 static inline int mddev_lock(mddev_t * mddev)
242 return down_interruptible(&mddev->reconfig_sem);
245 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
247 down(&mddev->reconfig_sem);
250 static inline int mddev_trylock(mddev_t * mddev)
252 return down_trylock(&mddev->reconfig_sem);
255 static inline void mddev_unlock(mddev_t * mddev)
257 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 inline static 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);
344 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
345 sector_t sector, int size, struct page *page)
347 /* write first size bytes of page to sector of rdev
348 * Increment mddev->pending_writes before returning
349 * and decrement it on completion, waking up sb_wait
350 * if zero is reached.
351 * If an error occurred, call md_error
353 struct bio *bio = bio_alloc(GFP_NOIO, 1);
355 bio->bi_bdev = rdev->bdev;
356 bio->bi_sector = sector;
357 bio_add_page(bio, page, size, 0);
358 bio->bi_private = rdev;
359 bio->bi_end_io = super_written;
360 atomic_inc(&mddev->pending_writes);
361 submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
364 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
369 complete((struct completion*)bio->bi_private);
373 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
374 struct page *page, int rw)
376 struct bio *bio = bio_alloc(GFP_NOIO, 1);
377 struct completion event;
380 rw |= (1 << BIO_RW_SYNC);
383 bio->bi_sector = sector;
384 bio_add_page(bio, page, size, 0);
385 init_completion(&event);
386 bio->bi_private = &event;
387 bio->bi_end_io = bi_complete;
389 wait_for_completion(&event);
391 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
396 static int read_disk_sb(mdk_rdev_t * rdev)
398 char b[BDEVNAME_SIZE];
399 if (!rdev->sb_page) {
407 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, MD_SB_BYTES, rdev->sb_page, READ))
413 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
414 bdevname(rdev->bdev,b));
418 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
420 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
421 (sb1->set_uuid1 == sb2->set_uuid1) &&
422 (sb1->set_uuid2 == sb2->set_uuid2) &&
423 (sb1->set_uuid3 == sb2->set_uuid3))
431 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
434 mdp_super_t *tmp1, *tmp2;
436 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
437 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
439 if (!tmp1 || !tmp2) {
441 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
449 * nr_disks is not constant
454 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
465 static unsigned int calc_sb_csum(mdp_super_t * sb)
467 unsigned int disk_csum, csum;
469 disk_csum = sb->sb_csum;
471 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
472 sb->sb_csum = disk_csum;
478 * Handle superblock details.
479 * We want to be able to handle multiple superblock formats
480 * so we have a common interface to them all, and an array of
481 * different handlers.
482 * We rely on user-space to write the initial superblock, and support
483 * reading and updating of superblocks.
484 * Interface methods are:
485 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
486 * loads and validates a superblock on dev.
487 * if refdev != NULL, compare superblocks on both devices
489 * 0 - dev has a superblock that is compatible with refdev
490 * 1 - dev has a superblock that is compatible and newer than refdev
491 * so dev should be used as the refdev in future
492 * -EINVAL superblock incompatible or invalid
493 * -othererror e.g. -EIO
495 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
496 * Verify that dev is acceptable into mddev.
497 * The first time, mddev->raid_disks will be 0, and data from
498 * dev should be merged in. Subsequent calls check that dev
499 * is new enough. Return 0 or -EINVAL
501 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
502 * Update the superblock for rdev with data in mddev
503 * This does not write to disc.
509 struct module *owner;
510 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
511 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
512 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
516 * load_super for 0.90.0
518 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
520 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
526 * Calculate the position of the superblock,
527 * it's at the end of the disk.
529 * It also happens to be a multiple of 4Kb.
531 sb_offset = calc_dev_sboffset(rdev->bdev);
532 rdev->sb_offset = sb_offset;
534 ret = read_disk_sb(rdev);
539 bdevname(rdev->bdev, b);
540 sb = (mdp_super_t*)page_address(rdev->sb_page);
542 if (sb->md_magic != MD_SB_MAGIC) {
543 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
548 if (sb->major_version != 0 ||
549 sb->minor_version != 90) {
550 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
551 sb->major_version, sb->minor_version,
556 if (sb->raid_disks <= 0)
559 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
560 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
565 rdev->preferred_minor = sb->md_minor;
566 rdev->data_offset = 0;
568 if (sb->level == LEVEL_MULTIPATH)
571 rdev->desc_nr = sb->this_disk.number;
577 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
578 if (!uuid_equal(refsb, sb)) {
579 printk(KERN_WARNING "md: %s has different UUID to %s\n",
580 b, bdevname(refdev->bdev,b2));
583 if (!sb_equal(refsb, sb)) {
584 printk(KERN_WARNING "md: %s has same UUID"
585 " but different superblock to %s\n",
586 b, bdevname(refdev->bdev, b2));
590 ev2 = md_event(refsb);
596 rdev->size = calc_dev_size(rdev, sb->chunk_size);
603 * validate_super for 0.90.0
605 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
608 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
610 rdev->raid_disk = -1;
612 if (mddev->raid_disks == 0) {
613 mddev->major_version = 0;
614 mddev->minor_version = sb->minor_version;
615 mddev->patch_version = sb->patch_version;
616 mddev->persistent = ! sb->not_persistent;
617 mddev->chunk_size = sb->chunk_size;
618 mddev->ctime = sb->ctime;
619 mddev->utime = sb->utime;
620 mddev->level = sb->level;
621 mddev->layout = sb->layout;
622 mddev->raid_disks = sb->raid_disks;
623 mddev->size = sb->size;
624 mddev->events = md_event(sb);
626 if (sb->state & (1<<MD_SB_CLEAN))
627 mddev->recovery_cp = MaxSector;
629 if (sb->events_hi == sb->cp_events_hi &&
630 sb->events_lo == sb->cp_events_lo) {
631 mddev->recovery_cp = sb->recovery_cp;
633 mddev->recovery_cp = 0;
636 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
637 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
638 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
639 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
641 mddev->max_disks = MD_SB_DISKS;
643 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
644 mddev->bitmap_file == NULL) {
645 if (mddev->level != 1) {
646 /* FIXME use a better test */
647 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
650 mddev->bitmap_offset = (MD_SB_BYTES >> 9);
653 } else if (mddev->pers == NULL) {
654 /* Insist on good event counter while assembling */
655 __u64 ev1 = md_event(sb);
657 if (ev1 < mddev->events)
659 } else if (mddev->bitmap) {
660 /* if adding to array with a bitmap, then we can accept an
661 * older device ... but not too old.
663 __u64 ev1 = md_event(sb);
664 if (ev1 < mddev->bitmap->events_cleared)
666 } else /* just a hot-add of a new device, leave raid_disk at -1 */
669 if (mddev->level != LEVEL_MULTIPATH) {
671 desc = sb->disks + rdev->desc_nr;
673 if (desc->state & (1<<MD_DISK_FAULTY))
675 else if (desc->state & (1<<MD_DISK_SYNC) &&
676 desc->raid_disk < mddev->raid_disks) {
678 rdev->raid_disk = desc->raid_disk;
680 } else /* MULTIPATH are always insync */
686 * sync_super for 0.90.0
688 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
691 struct list_head *tmp;
693 int next_spare = mddev->raid_disks;
695 /* make rdev->sb match mddev data..
698 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
699 * 3/ any empty disks < next_spare become removed
701 * disks[0] gets initialised to REMOVED because
702 * we cannot be sure from other fields if it has
703 * been initialised or not.
706 int active=0, working=0,failed=0,spare=0,nr_disks=0;
708 sb = (mdp_super_t*)page_address(rdev->sb_page);
710 memset(sb, 0, sizeof(*sb));
712 sb->md_magic = MD_SB_MAGIC;
713 sb->major_version = mddev->major_version;
714 sb->minor_version = mddev->minor_version;
715 sb->patch_version = mddev->patch_version;
716 sb->gvalid_words = 0; /* ignored */
717 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
718 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
719 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
720 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
722 sb->ctime = mddev->ctime;
723 sb->level = mddev->level;
724 sb->size = mddev->size;
725 sb->raid_disks = mddev->raid_disks;
726 sb->md_minor = mddev->md_minor;
727 sb->not_persistent = !mddev->persistent;
728 sb->utime = mddev->utime;
730 sb->events_hi = (mddev->events>>32);
731 sb->events_lo = (u32)mddev->events;
735 sb->recovery_cp = mddev->recovery_cp;
736 sb->cp_events_hi = (mddev->events>>32);
737 sb->cp_events_lo = (u32)mddev->events;
738 if (mddev->recovery_cp == MaxSector)
739 sb->state = (1<< MD_SB_CLEAN);
743 sb->layout = mddev->layout;
744 sb->chunk_size = mddev->chunk_size;
746 if (mddev->bitmap && mddev->bitmap_file == NULL)
747 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
749 sb->disks[0].state = (1<<MD_DISK_REMOVED);
750 ITERATE_RDEV(mddev,rdev2,tmp) {
752 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
753 rdev2->desc_nr = rdev2->raid_disk;
755 rdev2->desc_nr = next_spare++;
756 d = &sb->disks[rdev2->desc_nr];
758 d->number = rdev2->desc_nr;
759 d->major = MAJOR(rdev2->bdev->bd_dev);
760 d->minor = MINOR(rdev2->bdev->bd_dev);
761 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
762 d->raid_disk = rdev2->raid_disk;
764 d->raid_disk = rdev2->desc_nr; /* compatibility */
766 d->state = (1<<MD_DISK_FAULTY);
768 } else if (rdev2->in_sync) {
769 d->state = (1<<MD_DISK_ACTIVE);
770 d->state |= (1<<MD_DISK_SYNC);
780 /* now set the "removed" and "faulty" bits on any missing devices */
781 for (i=0 ; i < mddev->raid_disks ; i++) {
782 mdp_disk_t *d = &sb->disks[i];
783 if (d->state == 0 && d->number == 0) {
786 d->state = (1<<MD_DISK_REMOVED);
787 d->state |= (1<<MD_DISK_FAULTY);
791 sb->nr_disks = nr_disks;
792 sb->active_disks = active;
793 sb->working_disks = working;
794 sb->failed_disks = failed;
795 sb->spare_disks = spare;
797 sb->this_disk = sb->disks[rdev->desc_nr];
798 sb->sb_csum = calc_sb_csum(sb);
802 * version 1 superblock
805 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
807 unsigned int disk_csum, csum;
808 unsigned long long newcsum;
809 int size = 256 + le32_to_cpu(sb->max_dev)*2;
810 unsigned int *isuper = (unsigned int*)sb;
813 disk_csum = sb->sb_csum;
816 for (i=0; size>=4; size -= 4 )
817 newcsum += le32_to_cpu(*isuper++);
820 newcsum += le16_to_cpu(*(unsigned short*) isuper);
822 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
823 sb->sb_csum = disk_csum;
824 return cpu_to_le32(csum);
827 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
829 struct mdp_superblock_1 *sb;
832 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
835 * Calculate the position of the superblock.
836 * It is always aligned to a 4K boundary and
837 * depeding on minor_version, it can be:
838 * 0: At least 8K, but less than 12K, from end of device
839 * 1: At start of device
840 * 2: 4K from start of device.
842 switch(minor_version) {
844 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
846 sb_offset &= ~(sector_t)(4*2-1);
847 /* convert from sectors to K */
859 rdev->sb_offset = sb_offset;
861 ret = read_disk_sb(rdev);
865 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
867 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
868 sb->major_version != cpu_to_le32(1) ||
869 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
870 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
871 sb->feature_map != 0)
874 if (calc_sb_1_csum(sb) != sb->sb_csum) {
875 printk("md: invalid superblock checksum on %s\n",
876 bdevname(rdev->bdev,b));
879 if (le64_to_cpu(sb->data_size) < 10) {
880 printk("md: data_size too small on %s\n",
881 bdevname(rdev->bdev,b));
884 rdev->preferred_minor = 0xffff;
885 rdev->data_offset = le64_to_cpu(sb->data_offset);
891 struct mdp_superblock_1 *refsb =
892 (struct mdp_superblock_1*)page_address(refdev->sb_page);
894 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
895 sb->level != refsb->level ||
896 sb->layout != refsb->layout ||
897 sb->chunksize != refsb->chunksize) {
898 printk(KERN_WARNING "md: %s has strangely different"
899 " superblock to %s\n",
900 bdevname(rdev->bdev,b),
901 bdevname(refdev->bdev,b2));
904 ev1 = le64_to_cpu(sb->events);
905 ev2 = le64_to_cpu(refsb->events);
911 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
913 rdev->size = rdev->sb_offset;
914 if (rdev->size < le64_to_cpu(sb->data_size)/2)
916 rdev->size = le64_to_cpu(sb->data_size)/2;
917 if (le32_to_cpu(sb->chunksize))
918 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
922 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
924 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
926 rdev->raid_disk = -1;
928 if (mddev->raid_disks == 0) {
929 mddev->major_version = 1;
930 mddev->patch_version = 0;
931 mddev->persistent = 1;
932 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
933 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
934 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
935 mddev->level = le32_to_cpu(sb->level);
936 mddev->layout = le32_to_cpu(sb->layout);
937 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
938 mddev->size = le64_to_cpu(sb->size)/2;
939 mddev->events = le64_to_cpu(sb->events);
941 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
942 memcpy(mddev->uuid, sb->set_uuid, 16);
944 mddev->max_disks = (4096-256)/2;
946 if ((le32_to_cpu(sb->feature_map) & 1) &&
947 mddev->bitmap_file == NULL ) {
948 if (mddev->level != 1) {
949 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
952 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
954 } else if (mddev->pers == NULL) {
955 /* Insist of good event counter while assembling */
956 __u64 ev1 = le64_to_cpu(sb->events);
958 if (ev1 < mddev->events)
960 } else if (mddev->bitmap) {
961 /* If adding to array with a bitmap, then we can accept an
962 * older device, but not too old.
964 __u64 ev1 = le64_to_cpu(sb->events);
965 if (ev1 < mddev->bitmap->events_cleared)
967 } else /* just a hot-add of a new device, leave raid_disk at -1 */
970 if (mddev->level != LEVEL_MULTIPATH) {
972 rdev->desc_nr = le32_to_cpu(sb->dev_number);
973 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
975 case 0xffff: /* spare */
978 case 0xfffe: /* faulty */
984 rdev->raid_disk = role;
987 } else /* MULTIPATH are always insync */
993 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
995 struct mdp_superblock_1 *sb;
996 struct list_head *tmp;
999 /* make rdev->sb match mddev and rdev data. */
1001 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1003 sb->feature_map = 0;
1005 memset(sb->pad1, 0, sizeof(sb->pad1));
1006 memset(sb->pad2, 0, sizeof(sb->pad2));
1007 memset(sb->pad3, 0, sizeof(sb->pad3));
1009 sb->utime = cpu_to_le64((__u64)mddev->utime);
1010 sb->events = cpu_to_le64(mddev->events);
1012 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1014 sb->resync_offset = cpu_to_le64(0);
1016 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1017 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1018 sb->feature_map = cpu_to_le32(1);
1022 ITERATE_RDEV(mddev,rdev2,tmp)
1023 if (rdev2->desc_nr+1 > max_dev)
1024 max_dev = rdev2->desc_nr+1;
1026 sb->max_dev = cpu_to_le32(max_dev);
1027 for (i=0; i<max_dev;i++)
1028 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1030 ITERATE_RDEV(mddev,rdev2,tmp) {
1033 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1034 else if (rdev2->in_sync)
1035 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1037 sb->dev_roles[i] = cpu_to_le16(0xffff);
1040 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1041 sb->sb_csum = calc_sb_1_csum(sb);
1045 static struct super_type super_types[] = {
1048 .owner = THIS_MODULE,
1049 .load_super = super_90_load,
1050 .validate_super = super_90_validate,
1051 .sync_super = super_90_sync,
1055 .owner = THIS_MODULE,
1056 .load_super = super_1_load,
1057 .validate_super = super_1_validate,
1058 .sync_super = super_1_sync,
1062 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1064 struct list_head *tmp;
1067 ITERATE_RDEV(mddev,rdev,tmp)
1068 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1074 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1076 struct list_head *tmp;
1079 ITERATE_RDEV(mddev1,rdev,tmp)
1080 if (match_dev_unit(mddev2, rdev))
1086 static LIST_HEAD(pending_raid_disks);
1088 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1090 mdk_rdev_t *same_pdev;
1091 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1097 same_pdev = match_dev_unit(mddev, rdev);
1100 "%s: WARNING: %s appears to be on the same physical"
1101 " disk as %s. True\n protection against single-disk"
1102 " failure might be compromised.\n",
1103 mdname(mddev), bdevname(rdev->bdev,b),
1104 bdevname(same_pdev->bdev,b2));
1106 /* Verify rdev->desc_nr is unique.
1107 * If it is -1, assign a free number, else
1108 * check number is not in use
1110 if (rdev->desc_nr < 0) {
1112 if (mddev->pers) choice = mddev->raid_disks;
1113 while (find_rdev_nr(mddev, choice))
1115 rdev->desc_nr = choice;
1117 if (find_rdev_nr(mddev, rdev->desc_nr))
1121 list_add(&rdev->same_set, &mddev->disks);
1122 rdev->mddev = mddev;
1123 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1127 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1129 char b[BDEVNAME_SIZE];
1134 list_del_init(&rdev->same_set);
1135 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1140 * prevent the device from being mounted, repartitioned or
1141 * otherwise reused by a RAID array (or any other kernel
1142 * subsystem), by bd_claiming the device.
1144 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1147 struct block_device *bdev;
1148 char b[BDEVNAME_SIZE];
1150 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1152 printk(KERN_ERR "md: could not open %s.\n",
1153 __bdevname(dev, b));
1154 return PTR_ERR(bdev);
1156 err = bd_claim(bdev, rdev);
1158 printk(KERN_ERR "md: could not bd_claim %s.\n",
1167 static void unlock_rdev(mdk_rdev_t *rdev)
1169 struct block_device *bdev = rdev->bdev;
1177 void md_autodetect_dev(dev_t dev);
1179 static void export_rdev(mdk_rdev_t * rdev)
1181 char b[BDEVNAME_SIZE];
1182 printk(KERN_INFO "md: export_rdev(%s)\n",
1183 bdevname(rdev->bdev,b));
1187 list_del_init(&rdev->same_set);
1189 md_autodetect_dev(rdev->bdev->bd_dev);
1195 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1197 unbind_rdev_from_array(rdev);
1201 static void export_array(mddev_t *mddev)
1203 struct list_head *tmp;
1206 ITERATE_RDEV(mddev,rdev,tmp) {
1211 kick_rdev_from_array(rdev);
1213 if (!list_empty(&mddev->disks))
1215 mddev->raid_disks = 0;
1216 mddev->major_version = 0;
1219 static void print_desc(mdp_disk_t *desc)
1221 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1222 desc->major,desc->minor,desc->raid_disk,desc->state);
1225 static void print_sb(mdp_super_t *sb)
1230 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1231 sb->major_version, sb->minor_version, sb->patch_version,
1232 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1234 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1235 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1236 sb->md_minor, sb->layout, sb->chunk_size);
1237 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1238 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1239 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1240 sb->failed_disks, sb->spare_disks,
1241 sb->sb_csum, (unsigned long)sb->events_lo);
1244 for (i = 0; i < MD_SB_DISKS; i++) {
1247 desc = sb->disks + i;
1248 if (desc->number || desc->major || desc->minor ||
1249 desc->raid_disk || (desc->state && (desc->state != 4))) {
1250 printk(" D %2d: ", i);
1254 printk(KERN_INFO "md: THIS: ");
1255 print_desc(&sb->this_disk);
1259 static void print_rdev(mdk_rdev_t *rdev)
1261 char b[BDEVNAME_SIZE];
1262 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1263 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1264 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1265 if (rdev->sb_loaded) {
1266 printk(KERN_INFO "md: rdev superblock:\n");
1267 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1269 printk(KERN_INFO "md: no rdev superblock!\n");
1272 void md_print_devices(void)
1274 struct list_head *tmp, *tmp2;
1277 char b[BDEVNAME_SIZE];
1280 printk("md: **********************************\n");
1281 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1282 printk("md: **********************************\n");
1283 ITERATE_MDDEV(mddev,tmp) {
1286 bitmap_print_sb(mddev->bitmap);
1288 printk("%s: ", mdname(mddev));
1289 ITERATE_RDEV(mddev,rdev,tmp2)
1290 printk("<%s>", bdevname(rdev->bdev,b));
1293 ITERATE_RDEV(mddev,rdev,tmp2)
1296 printk("md: **********************************\n");
1301 static void sync_sbs(mddev_t * mddev)
1304 struct list_head *tmp;
1306 ITERATE_RDEV(mddev,rdev,tmp) {
1307 super_types[mddev->major_version].
1308 sync_super(mddev, rdev);
1309 rdev->sb_loaded = 1;
1313 static void md_update_sb(mddev_t * mddev)
1316 struct list_head *tmp;
1321 spin_lock(&mddev->write_lock);
1322 sync_req = mddev->in_sync;
1323 mddev->utime = get_seconds();
1326 if (!mddev->events) {
1328 * oops, this 64-bit counter should never wrap.
1329 * Either we are in around ~1 trillion A.C., assuming
1330 * 1 reboot per second, or we have a bug:
1335 mddev->sb_dirty = 2;
1339 * do not write anything to disk if using
1340 * nonpersistent superblocks
1342 if (!mddev->persistent) {
1343 mddev->sb_dirty = 0;
1344 spin_unlock(&mddev->write_lock);
1345 wake_up(&mddev->sb_wait);
1348 spin_unlock(&mddev->write_lock);
1351 "md: updating %s RAID superblock on device (in sync %d)\n",
1352 mdname(mddev),mddev->in_sync);
1354 err = bitmap_update_sb(mddev->bitmap);
1355 ITERATE_RDEV(mddev,rdev,tmp) {
1356 char b[BDEVNAME_SIZE];
1357 dprintk(KERN_INFO "md: ");
1359 dprintk("(skipping faulty ");
1361 dprintk("%s ", bdevname(rdev->bdev,b));
1362 if (!rdev->faulty) {
1363 md_super_write(mddev,rdev,
1364 rdev->sb_offset<<1, MD_SB_BYTES,
1366 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1367 bdevname(rdev->bdev,b),
1368 (unsigned long long)rdev->sb_offset);
1372 if (mddev->level == LEVEL_MULTIPATH)
1373 /* only need to write one superblock... */
1376 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1377 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1379 spin_lock(&mddev->write_lock);
1380 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1381 /* have to write it out again */
1382 spin_unlock(&mddev->write_lock);
1385 mddev->sb_dirty = 0;
1386 spin_unlock(&mddev->write_lock);
1387 wake_up(&mddev->sb_wait);
1392 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1394 * mark the device faulty if:
1396 * - the device is nonexistent (zero size)
1397 * - the device has no valid superblock
1399 * a faulty rdev _never_ has rdev->sb set.
1401 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1403 char b[BDEVNAME_SIZE];
1408 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1410 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1411 return ERR_PTR(-ENOMEM);
1413 memset(rdev, 0, sizeof(*rdev));
1415 if ((err = alloc_disk_sb(rdev)))
1418 err = lock_rdev(rdev, newdev);
1425 rdev->data_offset = 0;
1426 atomic_set(&rdev->nr_pending, 0);
1428 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1431 "md: %s has zero or unknown size, marking faulty!\n",
1432 bdevname(rdev->bdev,b));
1437 if (super_format >= 0) {
1438 err = super_types[super_format].
1439 load_super(rdev, NULL, super_minor);
1440 if (err == -EINVAL) {
1442 "md: %s has invalid sb, not importing!\n",
1443 bdevname(rdev->bdev,b));
1448 "md: could not read %s's sb, not importing!\n",
1449 bdevname(rdev->bdev,b));
1453 INIT_LIST_HEAD(&rdev->same_set);
1458 if (rdev->sb_page) {
1464 return ERR_PTR(err);
1468 * Check a full RAID array for plausibility
1472 static void analyze_sbs(mddev_t * mddev)
1475 struct list_head *tmp;
1476 mdk_rdev_t *rdev, *freshest;
1477 char b[BDEVNAME_SIZE];
1480 ITERATE_RDEV(mddev,rdev,tmp)
1481 switch (super_types[mddev->major_version].
1482 load_super(rdev, freshest, mddev->minor_version)) {
1490 "md: fatal superblock inconsistency in %s"
1491 " -- removing from array\n",
1492 bdevname(rdev->bdev,b));
1493 kick_rdev_from_array(rdev);
1497 super_types[mddev->major_version].
1498 validate_super(mddev, freshest);
1501 ITERATE_RDEV(mddev,rdev,tmp) {
1502 if (rdev != freshest)
1503 if (super_types[mddev->major_version].
1504 validate_super(mddev, rdev)) {
1505 printk(KERN_WARNING "md: kicking non-fresh %s"
1507 bdevname(rdev->bdev,b));
1508 kick_rdev_from_array(rdev);
1511 if (mddev->level == LEVEL_MULTIPATH) {
1512 rdev->desc_nr = i++;
1513 rdev->raid_disk = rdev->desc_nr;
1520 if (mddev->recovery_cp != MaxSector &&
1522 printk(KERN_ERR "md: %s: raid array is not clean"
1523 " -- starting background reconstruction\n",
1530 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1532 static DECLARE_MUTEX(disks_sem);
1533 mddev_t *mddev = mddev_find(dev);
1534 struct gendisk *disk;
1535 int partitioned = (MAJOR(dev) != MD_MAJOR);
1536 int shift = partitioned ? MdpMinorShift : 0;
1537 int unit = MINOR(dev) >> shift;
1543 if (mddev->gendisk) {
1548 disk = alloc_disk(1 << shift);
1554 disk->major = MAJOR(dev);
1555 disk->first_minor = unit << shift;
1557 sprintf(disk->disk_name, "md_d%d", unit);
1558 sprintf(disk->devfs_name, "md/d%d", unit);
1560 sprintf(disk->disk_name, "md%d", unit);
1561 sprintf(disk->devfs_name, "md/%d", unit);
1563 disk->fops = &md_fops;
1564 disk->private_data = mddev;
1565 disk->queue = mddev->queue;
1567 mddev->gendisk = disk;
1572 void md_wakeup_thread(mdk_thread_t *thread);
1574 static void md_safemode_timeout(unsigned long data)
1576 mddev_t *mddev = (mddev_t *) data;
1578 mddev->safemode = 1;
1579 md_wakeup_thread(mddev->thread);
1583 static int do_md_run(mddev_t * mddev)
1587 struct list_head *tmp;
1589 struct gendisk *disk;
1590 char b[BDEVNAME_SIZE];
1592 if (list_empty(&mddev->disks))
1593 /* cannot run an array with no devices.. */
1600 * Analyze all RAID superblock(s)
1602 if (!mddev->raid_disks)
1605 chunk_size = mddev->chunk_size;
1606 pnum = level_to_pers(mddev->level);
1608 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1611 * 'default chunksize' in the old md code used to
1612 * be PAGE_SIZE, baaad.
1613 * we abort here to be on the safe side. We don't
1614 * want to continue the bad practice.
1617 "no chunksize specified, see 'man raidtab'\n");
1620 if (chunk_size > MAX_CHUNK_SIZE) {
1621 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1622 chunk_size, MAX_CHUNK_SIZE);
1626 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1628 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1629 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1632 if (chunk_size < PAGE_SIZE) {
1633 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1634 chunk_size, PAGE_SIZE);
1638 /* devices must have minimum size of one chunk */
1639 ITERATE_RDEV(mddev,rdev,tmp) {
1642 if (rdev->size < chunk_size / 1024) {
1644 "md: Dev %s smaller than chunk_size:"
1646 bdevname(rdev->bdev,b),
1647 (unsigned long long)rdev->size,
1657 request_module("md-personality-%d", pnum);
1662 * Drop all container device buffers, from now on
1663 * the only valid external interface is through the md
1665 * Also find largest hardsector size
1667 ITERATE_RDEV(mddev,rdev,tmp) {
1670 sync_blockdev(rdev->bdev);
1671 invalidate_bdev(rdev->bdev, 0);
1674 md_probe(mddev->unit, NULL, NULL);
1675 disk = mddev->gendisk;
1679 spin_lock(&pers_lock);
1680 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1681 spin_unlock(&pers_lock);
1682 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1687 mddev->pers = pers[pnum];
1688 spin_unlock(&pers_lock);
1690 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1692 /* before we start the array running, initialise the bitmap */
1693 err = bitmap_create(mddev);
1695 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1696 mdname(mddev), err);
1698 err = mddev->pers->run(mddev);
1700 printk(KERN_ERR "md: pers->run() failed ...\n");
1701 module_put(mddev->pers->owner);
1703 bitmap_destroy(mddev);
1706 atomic_set(&mddev->writes_pending,0);
1707 mddev->safemode = 0;
1708 mddev->safemode_timer.function = md_safemode_timeout;
1709 mddev->safemode_timer.data = (unsigned long) mddev;
1710 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1713 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1715 if (mddev->sb_dirty)
1716 md_update_sb(mddev);
1718 set_capacity(disk, mddev->array_size<<1);
1720 /* If we call blk_queue_make_request here, it will
1721 * re-initialise max_sectors etc which may have been
1722 * refined inside -> run. So just set the bits we need to set.
1723 * Most initialisation happended when we called
1724 * blk_queue_make_request(..., md_fail_request)
1727 mddev->queue->queuedata = mddev;
1728 mddev->queue->make_request_fn = mddev->pers->make_request;
1734 static int restart_array(mddev_t *mddev)
1736 struct gendisk *disk = mddev->gendisk;
1740 * Complain if it has no devices
1743 if (list_empty(&mddev->disks))
1751 mddev->safemode = 0;
1753 set_disk_ro(disk, 0);
1755 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1758 * Kick recovery or resync if necessary
1760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1761 md_wakeup_thread(mddev->thread);
1764 printk(KERN_ERR "md: %s has no personality assigned.\n",
1773 static int do_md_stop(mddev_t * mddev, int ro)
1776 struct gendisk *disk = mddev->gendisk;
1779 if (atomic_read(&mddev->active)>2) {
1780 printk("md: %s still in use.\n",mdname(mddev));
1784 if (mddev->sync_thread) {
1785 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1786 md_unregister_thread(mddev->sync_thread);
1787 mddev->sync_thread = NULL;
1790 del_timer_sync(&mddev->safemode_timer);
1792 invalidate_partition(disk, 0);
1801 set_disk_ro(disk, 0);
1802 blk_queue_make_request(mddev->queue, md_fail_request);
1803 mddev->pers->stop(mddev);
1804 module_put(mddev->pers->owner);
1809 if (!mddev->in_sync) {
1810 /* mark array as shutdown cleanly */
1812 md_update_sb(mddev);
1815 set_disk_ro(disk, 1);
1818 bitmap_destroy(mddev);
1819 if (mddev->bitmap_file) {
1820 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1821 fput(mddev->bitmap_file);
1822 mddev->bitmap_file = NULL;
1826 * Free resources if final stop
1829 struct gendisk *disk;
1830 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1832 export_array(mddev);
1834 mddev->array_size = 0;
1835 disk = mddev->gendisk;
1837 set_capacity(disk, 0);
1840 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1847 static void autorun_array(mddev_t *mddev)
1850 struct list_head *tmp;
1853 if (list_empty(&mddev->disks))
1856 printk(KERN_INFO "md: running: ");
1858 ITERATE_RDEV(mddev,rdev,tmp) {
1859 char b[BDEVNAME_SIZE];
1860 printk("<%s>", bdevname(rdev->bdev,b));
1864 err = do_md_run (mddev);
1866 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1867 do_md_stop (mddev, 0);
1872 * lets try to run arrays based on all disks that have arrived
1873 * until now. (those are in pending_raid_disks)
1875 * the method: pick the first pending disk, collect all disks with
1876 * the same UUID, remove all from the pending list and put them into
1877 * the 'same_array' list. Then order this list based on superblock
1878 * update time (freshest comes first), kick out 'old' disks and
1879 * compare superblocks. If everything's fine then run it.
1881 * If "unit" is allocated, then bump its reference count
1883 static void autorun_devices(int part)
1885 struct list_head candidates;
1886 struct list_head *tmp;
1887 mdk_rdev_t *rdev0, *rdev;
1889 char b[BDEVNAME_SIZE];
1891 printk(KERN_INFO "md: autorun ...\n");
1892 while (!list_empty(&pending_raid_disks)) {
1894 rdev0 = list_entry(pending_raid_disks.next,
1895 mdk_rdev_t, same_set);
1897 printk(KERN_INFO "md: considering %s ...\n",
1898 bdevname(rdev0->bdev,b));
1899 INIT_LIST_HEAD(&candidates);
1900 ITERATE_RDEV_PENDING(rdev,tmp)
1901 if (super_90_load(rdev, rdev0, 0) >= 0) {
1902 printk(KERN_INFO "md: adding %s ...\n",
1903 bdevname(rdev->bdev,b));
1904 list_move(&rdev->same_set, &candidates);
1907 * now we have a set of devices, with all of them having
1908 * mostly sane superblocks. It's time to allocate the
1911 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1912 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1913 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1917 dev = MKDEV(mdp_major,
1918 rdev0->preferred_minor << MdpMinorShift);
1920 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1922 md_probe(dev, NULL, NULL);
1923 mddev = mddev_find(dev);
1926 "md: cannot allocate memory for md drive.\n");
1929 if (mddev_lock(mddev))
1930 printk(KERN_WARNING "md: %s locked, cannot run\n",
1932 else if (mddev->raid_disks || mddev->major_version
1933 || !list_empty(&mddev->disks)) {
1935 "md: %s already running, cannot run %s\n",
1936 mdname(mddev), bdevname(rdev0->bdev,b));
1937 mddev_unlock(mddev);
1939 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1940 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1941 list_del_init(&rdev->same_set);
1942 if (bind_rdev_to_array(rdev, mddev))
1945 autorun_array(mddev);
1946 mddev_unlock(mddev);
1948 /* on success, candidates will be empty, on error
1951 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1955 printk(KERN_INFO "md: ... autorun DONE.\n");
1959 * import RAID devices based on one partition
1960 * if possible, the array gets run as well.
1963 static int autostart_array(dev_t startdev)
1965 char b[BDEVNAME_SIZE];
1966 int err = -EINVAL, i;
1967 mdp_super_t *sb = NULL;
1968 mdk_rdev_t *start_rdev = NULL, *rdev;
1970 start_rdev = md_import_device(startdev, 0, 0);
1971 if (IS_ERR(start_rdev))
1975 /* NOTE: this can only work for 0.90.0 superblocks */
1976 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1977 if (sb->major_version != 0 ||
1978 sb->minor_version != 90 ) {
1979 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1980 export_rdev(start_rdev);
1984 if (start_rdev->faulty) {
1986 "md: can not autostart based on faulty %s!\n",
1987 bdevname(start_rdev->bdev,b));
1988 export_rdev(start_rdev);
1991 list_add(&start_rdev->same_set, &pending_raid_disks);
1993 for (i = 0; i < MD_SB_DISKS; i++) {
1994 mdp_disk_t *desc = sb->disks + i;
1995 dev_t dev = MKDEV(desc->major, desc->minor);
1999 if (dev == startdev)
2001 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2003 rdev = md_import_device(dev, 0, 0);
2007 list_add(&rdev->same_set, &pending_raid_disks);
2011 * possibly return codes
2019 static int get_version(void __user * arg)
2023 ver.major = MD_MAJOR_VERSION;
2024 ver.minor = MD_MINOR_VERSION;
2025 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2027 if (copy_to_user(arg, &ver, sizeof(ver)))
2033 static int get_array_info(mddev_t * mddev, void __user * arg)
2035 mdu_array_info_t info;
2036 int nr,working,active,failed,spare;
2038 struct list_head *tmp;
2040 nr=working=active=failed=spare=0;
2041 ITERATE_RDEV(mddev,rdev,tmp) {
2054 info.major_version = mddev->major_version;
2055 info.minor_version = mddev->minor_version;
2056 info.patch_version = MD_PATCHLEVEL_VERSION;
2057 info.ctime = mddev->ctime;
2058 info.level = mddev->level;
2059 info.size = mddev->size;
2061 info.raid_disks = mddev->raid_disks;
2062 info.md_minor = mddev->md_minor;
2063 info.not_persistent= !mddev->persistent;
2065 info.utime = mddev->utime;
2068 info.state = (1<<MD_SB_CLEAN);
2069 info.active_disks = active;
2070 info.working_disks = working;
2071 info.failed_disks = failed;
2072 info.spare_disks = spare;
2074 info.layout = mddev->layout;
2075 info.chunk_size = mddev->chunk_size;
2077 if (copy_to_user(arg, &info, sizeof(info)))
2083 static int get_bitmap_file(mddev_t * mddev, void * arg)
2085 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2086 char *ptr, *buf = NULL;
2089 file = kmalloc(sizeof(*file), GFP_KERNEL);
2093 /* bitmap disabled, zero the first byte and copy out */
2094 if (!mddev->bitmap || !mddev->bitmap->file) {
2095 file->pathname[0] = '\0';
2099 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2103 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2107 strcpy(file->pathname, ptr);
2111 if (copy_to_user(arg, file, sizeof(*file)))
2119 static int get_disk_info(mddev_t * mddev, void __user * arg)
2121 mdu_disk_info_t info;
2125 if (copy_from_user(&info, arg, sizeof(info)))
2130 rdev = find_rdev_nr(mddev, nr);
2132 info.major = MAJOR(rdev->bdev->bd_dev);
2133 info.minor = MINOR(rdev->bdev->bd_dev);
2134 info.raid_disk = rdev->raid_disk;
2137 info.state |= (1<<MD_DISK_FAULTY);
2138 else if (rdev->in_sync) {
2139 info.state |= (1<<MD_DISK_ACTIVE);
2140 info.state |= (1<<MD_DISK_SYNC);
2143 info.major = info.minor = 0;
2144 info.raid_disk = -1;
2145 info.state = (1<<MD_DISK_REMOVED);
2148 if (copy_to_user(arg, &info, sizeof(info)))
2154 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2156 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2158 dev_t dev = MKDEV(info->major,info->minor);
2160 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2163 if (!mddev->raid_disks) {
2165 /* expecting a device which has a superblock */
2166 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2169 "md: md_import_device returned %ld\n",
2171 return PTR_ERR(rdev);
2173 if (!list_empty(&mddev->disks)) {
2174 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2175 mdk_rdev_t, same_set);
2176 int err = super_types[mddev->major_version]
2177 .load_super(rdev, rdev0, mddev->minor_version);
2180 "md: %s has different UUID to %s\n",
2181 bdevname(rdev->bdev,b),
2182 bdevname(rdev0->bdev,b2));
2187 err = bind_rdev_to_array(rdev, mddev);
2194 * add_new_disk can be used once the array is assembled
2195 * to add "hot spares". They must already have a superblock
2200 if (!mddev->pers->hot_add_disk) {
2202 "%s: personality does not support diskops!\n",
2206 rdev = md_import_device(dev, mddev->major_version,
2207 mddev->minor_version);
2210 "md: md_import_device returned %ld\n",
2212 return PTR_ERR(rdev);
2214 /* set save_raid_disk if appropriate */
2215 if (!mddev->persistent) {
2216 if (info->state & (1<<MD_DISK_SYNC) &&
2217 info->raid_disk < mddev->raid_disks)
2218 rdev->raid_disk = info->raid_disk;
2220 rdev->raid_disk = -1;
2222 super_types[mddev->major_version].
2223 validate_super(mddev, rdev);
2224 rdev->saved_raid_disk = rdev->raid_disk;
2226 rdev->in_sync = 0; /* just to be sure */
2227 rdev->raid_disk = -1;
2228 err = bind_rdev_to_array(rdev, mddev);
2232 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2234 md_wakeup_thread(mddev->thread);
2238 /* otherwise, add_new_disk is only allowed
2239 * for major_version==0 superblocks
2241 if (mddev->major_version != 0) {
2242 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2247 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2249 rdev = md_import_device (dev, -1, 0);
2252 "md: error, md_import_device() returned %ld\n",
2254 return PTR_ERR(rdev);
2256 rdev->desc_nr = info->number;
2257 if (info->raid_disk < mddev->raid_disks)
2258 rdev->raid_disk = info->raid_disk;
2260 rdev->raid_disk = -1;
2263 if (rdev->raid_disk < mddev->raid_disks)
2264 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2268 err = bind_rdev_to_array(rdev, mddev);
2274 if (!mddev->persistent) {
2275 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2276 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2278 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2279 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2281 if (!mddev->size || (mddev->size > rdev->size))
2282 mddev->size = rdev->size;
2288 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2290 char b[BDEVNAME_SIZE];
2296 rdev = find_rdev(mddev, dev);
2300 if (rdev->raid_disk >= 0)
2303 kick_rdev_from_array(rdev);
2304 md_update_sb(mddev);
2308 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2309 bdevname(rdev->bdev,b), mdname(mddev));
2313 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2315 char b[BDEVNAME_SIZE];
2323 if (mddev->major_version != 0) {
2324 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2325 " version-0 superblocks.\n",
2329 if (!mddev->pers->hot_add_disk) {
2331 "%s: personality does not support diskops!\n",
2336 rdev = md_import_device (dev, -1, 0);
2339 "md: error, md_import_device() returned %ld\n",
2344 if (mddev->persistent)
2345 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2348 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2350 size = calc_dev_size(rdev, mddev->chunk_size);
2353 if (size < mddev->size) {
2355 "%s: disk size %llu blocks < array size %llu\n",
2356 mdname(mddev), (unsigned long long)size,
2357 (unsigned long long)mddev->size);
2364 "md: can not hot-add faulty %s disk to %s!\n",
2365 bdevname(rdev->bdev,b), mdname(mddev));
2371 bind_rdev_to_array(rdev, mddev);
2374 * The rest should better be atomic, we can have disk failures
2375 * noticed in interrupt contexts ...
2378 if (rdev->desc_nr == mddev->max_disks) {
2379 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2382 goto abort_unbind_export;
2385 rdev->raid_disk = -1;
2387 md_update_sb(mddev);
2390 * Kick recovery, maybe this spare has to be added to the
2391 * array immediately.
2393 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2394 md_wakeup_thread(mddev->thread);
2398 abort_unbind_export:
2399 unbind_rdev_from_array(rdev);
2406 /* similar to deny_write_access, but accounts for our holding a reference
2407 * to the file ourselves */
2408 static int deny_bitmap_write_access(struct file * file)
2410 struct inode *inode = file->f_mapping->host;
2412 spin_lock(&inode->i_lock);
2413 if (atomic_read(&inode->i_writecount) > 1) {
2414 spin_unlock(&inode->i_lock);
2417 atomic_set(&inode->i_writecount, -1);
2418 spin_unlock(&inode->i_lock);
2423 static int set_bitmap_file(mddev_t *mddev, int fd)
2430 mddev->bitmap_file = fget(fd);
2432 if (mddev->bitmap_file == NULL) {
2433 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2438 err = deny_bitmap_write_access(mddev->bitmap_file);
2440 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2442 fput(mddev->bitmap_file);
2443 mddev->bitmap_file = NULL;
2445 mddev->bitmap_offset = 0; /* file overrides offset */
2450 * set_array_info is used two different ways
2451 * The original usage is when creating a new array.
2452 * In this usage, raid_disks is > 0 and it together with
2453 * level, size, not_persistent,layout,chunksize determine the
2454 * shape of the array.
2455 * This will always create an array with a type-0.90.0 superblock.
2456 * The newer usage is when assembling an array.
2457 * In this case raid_disks will be 0, and the major_version field is
2458 * use to determine which style super-blocks are to be found on the devices.
2459 * The minor and patch _version numbers are also kept incase the
2460 * super_block handler wishes to interpret them.
2462 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2465 if (info->raid_disks == 0) {
2466 /* just setting version number for superblock loading */
2467 if (info->major_version < 0 ||
2468 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2469 super_types[info->major_version].name == NULL) {
2470 /* maybe try to auto-load a module? */
2472 "md: superblock version %d not known\n",
2473 info->major_version);
2476 mddev->major_version = info->major_version;
2477 mddev->minor_version = info->minor_version;
2478 mddev->patch_version = info->patch_version;
2481 mddev->major_version = MD_MAJOR_VERSION;
2482 mddev->minor_version = MD_MINOR_VERSION;
2483 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2484 mddev->ctime = get_seconds();
2486 mddev->level = info->level;
2487 mddev->size = info->size;
2488 mddev->raid_disks = info->raid_disks;
2489 /* don't set md_minor, it is determined by which /dev/md* was
2492 if (info->state & (1<<MD_SB_CLEAN))
2493 mddev->recovery_cp = MaxSector;
2495 mddev->recovery_cp = 0;
2496 mddev->persistent = ! info->not_persistent;
2498 mddev->layout = info->layout;
2499 mddev->chunk_size = info->chunk_size;
2501 mddev->max_disks = MD_SB_DISKS;
2503 mddev->sb_dirty = 1;
2506 * Generate a 128 bit UUID
2508 get_random_bytes(mddev->uuid, 16);
2514 * update_array_info is used to change the configuration of an
2516 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2517 * fields in the info are checked against the array.
2518 * Any differences that cannot be handled will cause an error.
2519 * Normally, only one change can be managed at a time.
2521 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2526 if (mddev->major_version != info->major_version ||
2527 mddev->minor_version != info->minor_version ||
2528 /* mddev->patch_version != info->patch_version || */
2529 mddev->ctime != info->ctime ||
2530 mddev->level != info->level ||
2531 /* mddev->layout != info->layout || */
2532 !mddev->persistent != info->not_persistent||
2533 mddev->chunk_size != info->chunk_size )
2535 /* Check there is only one change */
2536 if (mddev->size != info->size) cnt++;
2537 if (mddev->raid_disks != info->raid_disks) cnt++;
2538 if (mddev->layout != info->layout) cnt++;
2539 if (cnt == 0) return 0;
2540 if (cnt > 1) return -EINVAL;
2542 if (mddev->layout != info->layout) {
2544 * we don't need to do anything at the md level, the
2545 * personality will take care of it all.
2547 if (mddev->pers->reconfig == NULL)
2550 return mddev->pers->reconfig(mddev, info->layout, -1);
2552 if (mddev->size != info->size) {
2554 struct list_head *tmp;
2555 if (mddev->pers->resize == NULL)
2557 /* The "size" is the amount of each device that is used.
2558 * This can only make sense for arrays with redundancy.
2559 * linear and raid0 always use whatever space is available
2560 * We can only consider changing the size if no resync
2561 * or reconstruction is happening, and if the new size
2562 * is acceptable. It must fit before the sb_offset or,
2563 * if that is <data_offset, it must fit before the
2564 * size of each device.
2565 * If size is zero, we find the largest size that fits.
2567 if (mddev->sync_thread)
2569 ITERATE_RDEV(mddev,rdev,tmp) {
2571 int fit = (info->size == 0);
2572 if (rdev->sb_offset > rdev->data_offset)
2573 avail = (rdev->sb_offset*2) - rdev->data_offset;
2575 avail = get_capacity(rdev->bdev->bd_disk)
2576 - rdev->data_offset;
2577 if (fit && (info->size == 0 || info->size > avail/2))
2578 info->size = avail/2;
2579 if (avail < ((sector_t)info->size << 1))
2582 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2584 struct block_device *bdev;
2586 bdev = bdget_disk(mddev->gendisk, 0);
2588 down(&bdev->bd_inode->i_sem);
2589 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2590 up(&bdev->bd_inode->i_sem);
2595 if (mddev->raid_disks != info->raid_disks) {
2596 /* change the number of raid disks */
2597 if (mddev->pers->reshape == NULL)
2599 if (info->raid_disks <= 0 ||
2600 info->raid_disks >= mddev->max_disks)
2602 if (mddev->sync_thread)
2604 rv = mddev->pers->reshape(mddev, info->raid_disks);
2606 struct block_device *bdev;
2608 bdev = bdget_disk(mddev->gendisk, 0);
2610 down(&bdev->bd_inode->i_sem);
2611 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2612 up(&bdev->bd_inode->i_sem);
2617 md_update_sb(mddev);
2621 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2625 if (mddev->pers == NULL)
2628 rdev = find_rdev(mddev, dev);
2632 md_error(mddev, rdev);
2636 static int md_ioctl(struct inode *inode, struct file *file,
2637 unsigned int cmd, unsigned long arg)
2640 void __user *argp = (void __user *)arg;
2641 struct hd_geometry __user *loc = argp;
2642 mddev_t *mddev = NULL;
2644 if (!capable(CAP_SYS_ADMIN))
2648 * Commands dealing with the RAID driver but not any
2654 err = get_version(argp);
2657 case PRINT_RAID_DEBUG:
2665 autostart_arrays(arg);
2672 * Commands creating/starting a new array:
2675 mddev = inode->i_bdev->bd_disk->private_data;
2683 if (cmd == START_ARRAY) {
2684 /* START_ARRAY doesn't need to lock the array as autostart_array
2685 * does the locking, and it could even be a different array
2690 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2691 "This will not be supported beyond 2.6\n",
2692 current->comm, current->pid);
2695 err = autostart_array(new_decode_dev(arg));
2697 printk(KERN_WARNING "md: autostart failed!\n");
2703 err = mddev_lock(mddev);
2706 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2713 case SET_ARRAY_INFO:
2715 mdu_array_info_t info;
2717 memset(&info, 0, sizeof(info));
2718 else if (copy_from_user(&info, argp, sizeof(info))) {
2723 err = update_array_info(mddev, &info);
2725 printk(KERN_WARNING "md: couldn't update"
2726 " array info. %d\n", err);
2731 if (!list_empty(&mddev->disks)) {
2733 "md: array %s already has disks!\n",
2738 if (mddev->raid_disks) {
2740 "md: array %s already initialised!\n",
2745 err = set_array_info(mddev, &info);
2747 printk(KERN_WARNING "md: couldn't set"
2748 " array info. %d\n", err);
2758 * Commands querying/configuring an existing array:
2760 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2761 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2762 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2763 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2769 * Commands even a read-only array can execute:
2773 case GET_ARRAY_INFO:
2774 err = get_array_info(mddev, argp);
2777 case GET_BITMAP_FILE:
2778 err = get_bitmap_file(mddev, (void *)arg);
2782 err = get_disk_info(mddev, argp);
2785 case RESTART_ARRAY_RW:
2786 err = restart_array(mddev);
2790 err = do_md_stop (mddev, 0);
2794 err = do_md_stop (mddev, 1);
2798 * We have a problem here : there is no easy way to give a CHS
2799 * virtual geometry. We currently pretend that we have a 2 heads
2800 * 4 sectors (with a BIG number of cylinders...). This drives
2801 * dosfs just mad... ;-)
2808 err = put_user (2, (char __user *) &loc->heads);
2811 err = put_user (4, (char __user *) &loc->sectors);
2814 err = put_user(get_capacity(mddev->gendisk)/8,
2815 (short __user *) &loc->cylinders);
2818 err = put_user (get_start_sect(inode->i_bdev),
2819 (long __user *) &loc->start);
2824 * The remaining ioctls are changing the state of the
2825 * superblock, so we do not allow read-only arrays
2837 mdu_disk_info_t info;
2838 if (copy_from_user(&info, argp, sizeof(info)))
2841 err = add_new_disk(mddev, &info);
2845 case HOT_REMOVE_DISK:
2846 err = hot_remove_disk(mddev, new_decode_dev(arg));
2850 err = hot_add_disk(mddev, new_decode_dev(arg));
2853 case SET_DISK_FAULTY:
2854 err = set_disk_faulty(mddev, new_decode_dev(arg));
2858 err = do_md_run (mddev);
2861 case SET_BITMAP_FILE:
2862 err = set_bitmap_file(mddev, (int)arg);
2866 if (_IOC_TYPE(cmd) == MD_MAJOR)
2867 printk(KERN_WARNING "md: %s(pid %d) used"
2868 " obsolete MD ioctl, upgrade your"
2869 " software to use new ictls.\n",
2870 current->comm, current->pid);
2877 mddev_unlock(mddev);
2887 static int md_open(struct inode *inode, struct file *file)
2890 * Succeed if we can lock the mddev, which confirms that
2891 * it isn't being stopped right now.
2893 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2896 if ((err = mddev_lock(mddev)))
2901 mddev_unlock(mddev);
2903 check_disk_change(inode->i_bdev);
2908 static int md_release(struct inode *inode, struct file * file)
2910 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2919 static int md_media_changed(struct gendisk *disk)
2921 mddev_t *mddev = disk->private_data;
2923 return mddev->changed;
2926 static int md_revalidate(struct gendisk *disk)
2928 mddev_t *mddev = disk->private_data;
2933 static struct block_device_operations md_fops =
2935 .owner = THIS_MODULE,
2937 .release = md_release,
2939 .media_changed = md_media_changed,
2940 .revalidate_disk= md_revalidate,
2943 static int md_thread(void * arg)
2945 mdk_thread_t *thread = arg;
2953 daemonize(thread->name, mdname(thread->mddev));
2955 current->exit_signal = SIGCHLD;
2956 allow_signal(SIGKILL);
2957 thread->tsk = current;
2960 * md_thread is a 'system-thread', it's priority should be very
2961 * high. We avoid resource deadlocks individually in each
2962 * raid personality. (RAID5 does preallocation) We also use RR and
2963 * the very same RT priority as kswapd, thus we will never get
2964 * into a priority inversion deadlock.
2966 * we definitely have to have equal or higher priority than
2967 * bdflush, otherwise bdflush will deadlock if there are too
2968 * many dirty RAID5 blocks.
2972 complete(thread->event);
2973 while (thread->run) {
2974 void (*run)(mddev_t *);
2976 wait_event_interruptible_timeout(thread->wqueue,
2977 test_bit(THREAD_WAKEUP, &thread->flags),
2981 clear_bit(THREAD_WAKEUP, &thread->flags);
2987 if (signal_pending(current))
2988 flush_signals(current);
2990 complete(thread->event);
2994 void md_wakeup_thread(mdk_thread_t *thread)
2997 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
2998 set_bit(THREAD_WAKEUP, &thread->flags);
2999 wake_up(&thread->wqueue);
3003 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3006 mdk_thread_t *thread;
3008 struct completion event;
3010 thread = (mdk_thread_t *) kmalloc
3011 (sizeof(mdk_thread_t), GFP_KERNEL);
3015 memset(thread, 0, sizeof(mdk_thread_t));
3016 init_waitqueue_head(&thread->wqueue);
3018 init_completion(&event);
3019 thread->event = &event;
3021 thread->mddev = mddev;
3022 thread->name = name;
3023 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3024 ret = kernel_thread(md_thread, thread, 0);
3029 wait_for_completion(&event);
3033 void md_unregister_thread(mdk_thread_t *thread)
3035 struct completion event;
3037 init_completion(&event);
3039 thread->event = &event;
3041 /* As soon as ->run is set to NULL, the task could disappear,
3042 * so we need to hold tasklist_lock until we have sent the signal
3044 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3045 read_lock(&tasklist_lock);
3047 send_sig(SIGKILL, thread->tsk, 1);
3048 read_unlock(&tasklist_lock);
3049 wait_for_completion(&event);
3053 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3060 if (!rdev || rdev->faulty)
3063 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3065 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3066 __builtin_return_address(0),__builtin_return_address(1),
3067 __builtin_return_address(2),__builtin_return_address(3));
3069 if (!mddev->pers->error_handler)
3071 mddev->pers->error_handler(mddev,rdev);
3072 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3073 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3074 md_wakeup_thread(mddev->thread);
3077 /* seq_file implementation /proc/mdstat */
3079 static void status_unused(struct seq_file *seq)
3083 struct list_head *tmp;
3085 seq_printf(seq, "unused devices: ");
3087 ITERATE_RDEV_PENDING(rdev,tmp) {
3088 char b[BDEVNAME_SIZE];
3090 seq_printf(seq, "%s ",
3091 bdevname(rdev->bdev,b));
3094 seq_printf(seq, "<none>");
3096 seq_printf(seq, "\n");
3100 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3102 unsigned long max_blocks, resync, res, dt, db, rt;
3104 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3106 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3107 max_blocks = mddev->resync_max_sectors >> 1;
3109 max_blocks = mddev->size;
3112 * Should not happen.
3118 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3120 int i, x = res/50, y = 20-x;
3121 seq_printf(seq, "[");
3122 for (i = 0; i < x; i++)
3123 seq_printf(seq, "=");
3124 seq_printf(seq, ">");
3125 for (i = 0; i < y; i++)
3126 seq_printf(seq, ".");
3127 seq_printf(seq, "] ");
3129 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3130 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3131 "resync" : "recovery"),
3132 res/10, res % 10, resync, max_blocks);
3135 * We do not want to overflow, so the order of operands and
3136 * the * 100 / 100 trick are important. We do a +1 to be
3137 * safe against division by zero. We only estimate anyway.
3139 * dt: time from mark until now
3140 * db: blocks written from mark until now
3141 * rt: remaining time
3143 dt = ((jiffies - mddev->resync_mark) / HZ);
3145 db = resync - (mddev->resync_mark_cnt/2);
3146 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3148 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3150 seq_printf(seq, " speed=%ldK/sec", db/dt);
3153 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3155 struct list_head *tmp;
3165 spin_lock(&all_mddevs_lock);
3166 list_for_each(tmp,&all_mddevs)
3168 mddev = list_entry(tmp, mddev_t, all_mddevs);
3170 spin_unlock(&all_mddevs_lock);
3173 spin_unlock(&all_mddevs_lock);
3175 return (void*)2;/* tail */
3179 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3181 struct list_head *tmp;
3182 mddev_t *next_mddev, *mddev = v;
3188 spin_lock(&all_mddevs_lock);
3190 tmp = all_mddevs.next;
3192 tmp = mddev->all_mddevs.next;
3193 if (tmp != &all_mddevs)
3194 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3196 next_mddev = (void*)2;
3199 spin_unlock(&all_mddevs_lock);
3207 static void md_seq_stop(struct seq_file *seq, void *v)
3211 if (mddev && v != (void*)1 && v != (void*)2)
3215 static int md_seq_show(struct seq_file *seq, void *v)
3219 struct list_head *tmp2;
3222 struct bitmap *bitmap;
3224 if (v == (void*)1) {
3225 seq_printf(seq, "Personalities : ");
3226 spin_lock(&pers_lock);
3227 for (i = 0; i < MAX_PERSONALITY; i++)
3229 seq_printf(seq, "[%s] ", pers[i]->name);
3231 spin_unlock(&pers_lock);
3232 seq_printf(seq, "\n");
3235 if (v == (void*)2) {
3240 if (mddev_lock(mddev)!=0)
3242 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3243 seq_printf(seq, "%s : %sactive", mdname(mddev),
3244 mddev->pers ? "" : "in");
3247 seq_printf(seq, " (read-only)");
3248 seq_printf(seq, " %s", mddev->pers->name);
3252 ITERATE_RDEV(mddev,rdev,tmp2) {
3253 char b[BDEVNAME_SIZE];
3254 seq_printf(seq, " %s[%d]",
3255 bdevname(rdev->bdev,b), rdev->desc_nr);
3257 seq_printf(seq, "(F)");
3263 if (!list_empty(&mddev->disks)) {
3265 seq_printf(seq, "\n %llu blocks",
3266 (unsigned long long)mddev->array_size);
3268 seq_printf(seq, "\n %llu blocks",
3269 (unsigned long long)size);
3273 mddev->pers->status (seq, mddev);
3274 seq_printf(seq, "\n ");
3275 if (mddev->curr_resync > 2) {
3276 status_resync (seq, mddev);
3277 seq_printf(seq, "\n ");
3278 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3279 seq_printf(seq, " resync=DELAYED\n ");
3281 seq_printf(seq, "\n ");
3283 if ((bitmap = mddev->bitmap)) {
3284 unsigned long chunk_kb;
3285 unsigned long flags;
3286 spin_lock_irqsave(&bitmap->lock, flags);
3287 chunk_kb = bitmap->chunksize >> 10;
3288 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3290 bitmap->pages - bitmap->missing_pages,
3292 (bitmap->pages - bitmap->missing_pages)
3293 << (PAGE_SHIFT - 10),
3294 chunk_kb ? chunk_kb : bitmap->chunksize,
3295 chunk_kb ? "KB" : "B");
3297 seq_printf(seq, ", file: ");
3298 seq_path(seq, bitmap->file->f_vfsmnt,
3299 bitmap->file->f_dentry," \t\n");
3302 seq_printf(seq, "\n");
3303 spin_unlock_irqrestore(&bitmap->lock, flags);
3306 seq_printf(seq, "\n");
3308 mddev_unlock(mddev);
3313 static struct seq_operations md_seq_ops = {
3314 .start = md_seq_start,
3315 .next = md_seq_next,
3316 .stop = md_seq_stop,
3317 .show = md_seq_show,
3320 static int md_seq_open(struct inode *inode, struct file *file)
3324 error = seq_open(file, &md_seq_ops);
3328 static struct file_operations md_seq_fops = {
3329 .open = md_seq_open,
3331 .llseek = seq_lseek,
3332 .release = seq_release,
3335 int register_md_personality(int pnum, mdk_personality_t *p)
3337 if (pnum >= MAX_PERSONALITY) {
3339 "md: tried to install personality %s as nr %d, but max is %lu\n",
3340 p->name, pnum, MAX_PERSONALITY-1);
3344 spin_lock(&pers_lock);
3346 spin_unlock(&pers_lock);
3351 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3352 spin_unlock(&pers_lock);
3356 int unregister_md_personality(int pnum)
3358 if (pnum >= MAX_PERSONALITY)
3361 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3362 spin_lock(&pers_lock);
3364 spin_unlock(&pers_lock);
3368 static int is_mddev_idle(mddev_t *mddev)
3371 struct list_head *tmp;
3373 unsigned long curr_events;
3376 ITERATE_RDEV(mddev,rdev,tmp) {
3377 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3378 curr_events = disk_stat_read(disk, read_sectors) +
3379 disk_stat_read(disk, write_sectors) -
3380 atomic_read(&disk->sync_io);
3381 /* Allow some slack between valud of curr_events and last_events,
3382 * as there are some uninteresting races.
3383 * Note: the following is an unsigned comparison.
3385 if ((curr_events - rdev->last_events + 32) > 64) {
3386 rdev->last_events = curr_events;
3393 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3395 /* another "blocks" (512byte) blocks have been synced */
3396 atomic_sub(blocks, &mddev->recovery_active);
3397 wake_up(&mddev->recovery_wait);
3399 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3400 md_wakeup_thread(mddev->thread);
3401 // stop recovery, signal do_sync ....
3406 /* md_write_start(mddev, bi)
3407 * If we need to update some array metadata (e.g. 'active' flag
3408 * in superblock) before writing, schedule a superblock update
3409 * and wait for it to complete.
3411 void md_write_start(mddev_t *mddev, struct bio *bi)
3414 if (bio_data_dir(bi) != WRITE)
3417 atomic_inc(&mddev->writes_pending);
3418 if (mddev->in_sync) {
3419 spin_lock(&mddev->write_lock);
3420 if (mddev->in_sync) {
3422 mddev->sb_dirty = 1;
3423 md_wakeup_thread(mddev->thread);
3425 spin_unlock(&mddev->write_lock);
3427 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3430 void md_write_end(mddev_t *mddev)
3432 if (atomic_dec_and_test(&mddev->writes_pending)) {
3433 if (mddev->safemode == 2)
3434 md_wakeup_thread(mddev->thread);
3436 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3440 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3442 #define SYNC_MARKS 10
3443 #define SYNC_MARK_STEP (3*HZ)
3444 static void md_do_sync(mddev_t *mddev)
3447 unsigned int currspeed = 0,
3449 sector_t max_sectors,j, io_sectors;
3450 unsigned long mark[SYNC_MARKS];
3451 sector_t mark_cnt[SYNC_MARKS];
3453 struct list_head *tmp;
3454 sector_t last_check;
3457 /* just incase thread restarts... */
3458 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3461 /* we overload curr_resync somewhat here.
3462 * 0 == not engaged in resync at all
3463 * 2 == checking that there is no conflict with another sync
3464 * 1 == like 2, but have yielded to allow conflicting resync to
3466 * other == active in resync - this many blocks
3468 * Before starting a resync we must have set curr_resync to
3469 * 2, and then checked that every "conflicting" array has curr_resync
3470 * less than ours. When we find one that is the same or higher
3471 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3472 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3473 * This will mean we have to start checking from the beginning again.
3478 mddev->curr_resync = 2;
3481 if (signal_pending(current)) {
3482 flush_signals(current);
3485 ITERATE_MDDEV(mddev2,tmp) {
3487 if (mddev2 == mddev)
3489 if (mddev2->curr_resync &&
3490 match_mddev_units(mddev,mddev2)) {
3492 if (mddev < mddev2 && mddev->curr_resync == 2) {
3493 /* arbitrarily yield */
3494 mddev->curr_resync = 1;
3495 wake_up(&resync_wait);
3497 if (mddev > mddev2 && mddev->curr_resync == 1)
3498 /* no need to wait here, we can wait the next
3499 * time 'round when curr_resync == 2
3502 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3503 if (!signal_pending(current)
3504 && mddev2->curr_resync >= mddev->curr_resync) {
3505 printk(KERN_INFO "md: delaying resync of %s"
3506 " until %s has finished resync (they"
3507 " share one or more physical units)\n",
3508 mdname(mddev), mdname(mddev2));
3511 finish_wait(&resync_wait, &wq);
3514 finish_wait(&resync_wait, &wq);
3517 } while (mddev->curr_resync < 2);
3519 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3520 /* resync follows the size requested by the personality,
3521 * which defaults to physical size, but can be virtual size
3523 max_sectors = mddev->resync_max_sectors;
3525 /* recovery follows the physical size of devices */
3526 max_sectors = mddev->size << 1;
3528 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3529 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3530 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3531 printk(KERN_INFO "md: using maximum available idle IO bandwith "
3532 "(but not more than %d KB/sec) for reconstruction.\n",
3533 sysctl_speed_limit_max);
3535 is_mddev_idle(mddev); /* this also initializes IO event counters */
3536 /* we don't use the checkpoint if there's a bitmap */
3537 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3538 j = mddev->recovery_cp;
3542 for (m = 0; m < SYNC_MARKS; m++) {
3544 mark_cnt[m] = io_sectors;
3547 mddev->resync_mark = mark[last_mark];
3548 mddev->resync_mark_cnt = mark_cnt[last_mark];
3551 * Tune reconstruction:
3553 window = 32*(PAGE_SIZE/512);
3554 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3555 window/2,(unsigned long long) max_sectors/2);
3557 atomic_set(&mddev->recovery_active, 0);
3558 init_waitqueue_head(&mddev->recovery_wait);
3563 "md: resuming recovery of %s from checkpoint.\n",
3565 mddev->curr_resync = j;
3568 while (j < max_sectors) {
3572 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3573 currspeed < sysctl_speed_limit_min);
3575 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3579 if (!skipped) { /* actual IO requested */
3580 io_sectors += sectors;
3581 atomic_add(sectors, &mddev->recovery_active);
3585 if (j>1) mddev->curr_resync = j;
3588 if (last_check + window > io_sectors || j == max_sectors)
3591 last_check = io_sectors;
3593 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3594 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3598 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3600 int next = (last_mark+1) % SYNC_MARKS;
3602 mddev->resync_mark = mark[next];
3603 mddev->resync_mark_cnt = mark_cnt[next];
3604 mark[next] = jiffies;
3605 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3610 if (signal_pending(current)) {
3612 * got a signal, exit.
3615 "md: md_do_sync() got signal ... exiting\n");
3616 flush_signals(current);
3617 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3622 * this loop exits only if either when we are slower than
3623 * the 'hard' speed limit, or the system was IO-idle for
3625 * the system might be non-idle CPU-wise, but we only care
3626 * about not overloading the IO subsystem. (things like an
3627 * e2fsck being done on the RAID array should execute fast)
3629 mddev->queue->unplug_fn(mddev->queue);
3632 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3633 /((jiffies-mddev->resync_mark)/HZ +1) +1;
3635 if (currspeed > sysctl_speed_limit_min) {
3636 if ((currspeed > sysctl_speed_limit_max) ||
3637 !is_mddev_idle(mddev)) {
3638 msleep_interruptible(250);
3643 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3645 * this also signals 'finished resyncing' to md_stop
3648 mddev->queue->unplug_fn(mddev->queue);
3650 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3652 /* tell personality that we are finished */
3653 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3655 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3656 mddev->curr_resync > 2 &&
3657 mddev->curr_resync >= mddev->recovery_cp) {
3658 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3660 "md: checkpointing recovery of %s.\n",
3662 mddev->recovery_cp = mddev->curr_resync;
3664 mddev->recovery_cp = MaxSector;
3668 mddev->curr_resync = 0;
3669 wake_up(&resync_wait);
3670 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3671 md_wakeup_thread(mddev->thread);
3676 * This routine is regularly called by all per-raid-array threads to
3677 * deal with generic issues like resync and super-block update.
3678 * Raid personalities that don't have a thread (linear/raid0) do not
3679 * need this as they never do any recovery or update the superblock.
3681 * It does not do any resync itself, but rather "forks" off other threads
3682 * to do that as needed.
3683 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3684 * "->recovery" and create a thread at ->sync_thread.
3685 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3686 * and wakeups up this thread which will reap the thread and finish up.
3687 * This thread also removes any faulty devices (with nr_pending == 0).
3689 * The overall approach is:
3690 * 1/ if the superblock needs updating, update it.
3691 * 2/ If a recovery thread is running, don't do anything else.
3692 * 3/ If recovery has finished, clean up, possibly marking spares active.
3693 * 4/ If there are any faulty devices, remove them.
3694 * 5/ If array is degraded, try to add spares devices
3695 * 6/ If array has spares or is not in-sync, start a resync thread.
3697 void md_check_recovery(mddev_t *mddev)
3700 struct list_head *rtmp;
3704 bitmap_daemon_work(mddev->bitmap);
3709 if (signal_pending(current)) {
3710 if (mddev->pers->sync_request) {
3711 printk(KERN_INFO "md: %s in immediate safe mode\n",
3713 mddev->safemode = 2;
3715 flush_signals(current);
3720 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3721 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3722 (mddev->safemode == 1) ||
3723 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3724 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3728 if (mddev_trylock(mddev)==0) {
3731 spin_lock(&mddev->write_lock);
3732 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3733 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3735 mddev->sb_dirty = 1;
3737 if (mddev->safemode == 1)
3738 mddev->safemode = 0;
3739 spin_unlock(&mddev->write_lock);
3741 if (mddev->sb_dirty)
3742 md_update_sb(mddev);
3745 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3746 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3747 /* resync/recovery still happening */
3748 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3751 if (mddev->sync_thread) {
3752 /* resync has finished, collect result */
3753 md_unregister_thread(mddev->sync_thread);
3754 mddev->sync_thread = NULL;
3755 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3756 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3758 /* activate any spares */
3759 mddev->pers->spare_active(mddev);
3761 md_update_sb(mddev);
3763 /* if array is no-longer degraded, then any saved_raid_disk
3764 * information must be scrapped
3766 if (!mddev->degraded)
3767 ITERATE_RDEV(mddev,rdev,rtmp)
3768 rdev->saved_raid_disk = -1;
3770 mddev->recovery = 0;
3771 /* flag recovery needed just to double check */
3772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3775 if (mddev->recovery)
3776 /* probably just the RECOVERY_NEEDED flag */
3777 mddev->recovery = 0;
3779 /* no recovery is running.
3780 * remove any failed drives, then
3781 * add spares if possible.
3782 * Spare are also removed and re-added, to allow
3783 * the personality to fail the re-add.
3785 ITERATE_RDEV(mddev,rdev,rtmp)
3786 if (rdev->raid_disk >= 0 &&
3787 (rdev->faulty || ! rdev->in_sync) &&
3788 atomic_read(&rdev->nr_pending)==0) {
3789 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3790 rdev->raid_disk = -1;
3793 if (mddev->degraded) {
3794 ITERATE_RDEV(mddev,rdev,rtmp)
3795 if (rdev->raid_disk < 0
3797 if (mddev->pers->hot_add_disk(mddev,rdev))
3804 if (!spares && (mddev->recovery_cp == MaxSector )) {
3805 /* nothing we can do ... */
3808 if (mddev->pers->sync_request) {
3809 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3811 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3812 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3813 /* We are adding a device or devices to an array
3814 * which has the bitmap stored on all devices.
3815 * So make sure all bitmap pages get written
3817 bitmap_write_all(mddev->bitmap);
3819 mddev->sync_thread = md_register_thread(md_do_sync,
3822 if (!mddev->sync_thread) {
3823 printk(KERN_ERR "%s: could not start resync"
3826 /* leave the spares where they are, it shouldn't hurt */
3827 mddev->recovery = 0;
3829 md_wakeup_thread(mddev->sync_thread);
3833 mddev_unlock(mddev);
3837 static int md_notify_reboot(struct notifier_block *this,
3838 unsigned long code, void *x)
3840 struct list_head *tmp;
3843 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3845 printk(KERN_INFO "md: stopping all md devices.\n");
3847 ITERATE_MDDEV(mddev,tmp)
3848 if (mddev_trylock(mddev)==0)
3849 do_md_stop (mddev, 1);
3851 * certain more exotic SCSI devices are known to be
3852 * volatile wrt too early system reboots. While the
3853 * right place to handle this issue is the given
3854 * driver, we do want to have a safe RAID driver ...
3861 static struct notifier_block md_notifier = {
3862 .notifier_call = md_notify_reboot,
3864 .priority = INT_MAX, /* before any real devices */
3867 static void md_geninit(void)
3869 struct proc_dir_entry *p;
3871 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3873 p = create_proc_entry("mdstat", S_IRUGO, NULL);
3875 p->proc_fops = &md_seq_fops;
3878 static int __init md_init(void)
3882 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3883 " MD_SB_DISKS=%d\n",
3884 MD_MAJOR_VERSION, MD_MINOR_VERSION,
3885 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3886 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3889 if (register_blkdev(MAJOR_NR, "md"))
3891 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3892 unregister_blkdev(MAJOR_NR, "md");
3896 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3897 md_probe, NULL, NULL);
3898 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3899 md_probe, NULL, NULL);
3901 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3902 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3903 S_IFBLK|S_IRUSR|S_IWUSR,
3906 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3907 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3908 S_IFBLK|S_IRUSR|S_IWUSR,
3912 register_reboot_notifier(&md_notifier);
3913 raid_table_header = register_sysctl_table(raid_root_table, 1);
3923 * Searches all registered partitions for autorun RAID arrays
3926 static dev_t detected_devices[128];
3929 void md_autodetect_dev(dev_t dev)
3931 if (dev_cnt >= 0 && dev_cnt < 127)
3932 detected_devices[dev_cnt++] = dev;
3936 static void autostart_arrays(int part)
3941 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
3943 for (i = 0; i < dev_cnt; i++) {
3944 dev_t dev = detected_devices[i];
3946 rdev = md_import_device(dev,0, 0);
3954 list_add(&rdev->same_set, &pending_raid_disks);
3958 autorun_devices(part);
3963 static __exit void md_exit(void)
3966 struct list_head *tmp;
3968 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
3969 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
3970 for (i=0; i < MAX_MD_DEVS; i++)
3971 devfs_remove("md/%d", i);
3972 for (i=0; i < MAX_MD_DEVS; i++)
3973 devfs_remove("md/d%d", i);
3977 unregister_blkdev(MAJOR_NR,"md");
3978 unregister_blkdev(mdp_major, "mdp");
3979 unregister_reboot_notifier(&md_notifier);
3980 unregister_sysctl_table(raid_table_header);
3981 remove_proc_entry("mdstat", NULL);
3982 ITERATE_MDDEV(mddev,tmp) {
3983 struct gendisk *disk = mddev->gendisk;
3986 export_array(mddev);
3989 mddev->gendisk = NULL;
3994 module_init(md_init)
3995 module_exit(md_exit)
3997 EXPORT_SYMBOL(register_md_personality);
3998 EXPORT_SYMBOL(unregister_md_personality);
3999 EXPORT_SYMBOL(md_error);
4000 EXPORT_SYMBOL(md_done_sync);
4001 EXPORT_SYMBOL(md_write_start);
4002 EXPORT_SYMBOL(md_write_end);
4003 EXPORT_SYMBOL(md_register_thread);
4004 EXPORT_SYMBOL(md_unregister_thread);
4005 EXPORT_SYMBOL(md_wakeup_thread);
4006 EXPORT_SYMBOL(md_print_devices);
4007 EXPORT_SYMBOL(md_check_recovery);
4008 MODULE_LICENSE("GPL");