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);
259 md_wakeup_thread(mddev->thread);
262 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
265 struct list_head *tmp;
267 ITERATE_RDEV(mddev,rdev,tmp) {
268 if (rdev->desc_nr == nr)
274 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
276 struct list_head *tmp;
279 ITERATE_RDEV(mddev,rdev,tmp) {
280 if (rdev->bdev->bd_dev == dev)
286 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
288 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
289 return MD_NEW_SIZE_BLOCKS(size);
292 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
296 size = rdev->sb_offset;
299 size &= ~((sector_t)chunk_size/1024 - 1);
303 static int alloc_disk_sb(mdk_rdev_t * rdev)
308 rdev->sb_page = alloc_page(GFP_KERNEL);
309 if (!rdev->sb_page) {
310 printk(KERN_ALERT "md: out of memory.\n");
317 static void free_disk_sb(mdk_rdev_t * rdev)
320 page_cache_release(rdev->sb_page);
322 rdev->sb_page = NULL;
329 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
331 mdk_rdev_t *rdev = bio->bi_private;
335 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
336 md_error(rdev->mddev, rdev);
338 if (atomic_dec_and_test(&rdev->mddev->pending_writes))
339 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);
625 mddev->bitmap_offset = 0;
626 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
628 if (sb->state & (1<<MD_SB_CLEAN))
629 mddev->recovery_cp = MaxSector;
631 if (sb->events_hi == sb->cp_events_hi &&
632 sb->events_lo == sb->cp_events_lo) {
633 mddev->recovery_cp = sb->recovery_cp;
635 mddev->recovery_cp = 0;
638 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
639 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
640 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
641 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
643 mddev->max_disks = MD_SB_DISKS;
645 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
646 mddev->bitmap_file == NULL) {
647 if (mddev->level != 1) {
648 /* FIXME use a better test */
649 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
652 mddev->bitmap_offset = mddev->default_bitmap_offset;
655 } else if (mddev->pers == NULL) {
656 /* Insist on good event counter while assembling */
657 __u64 ev1 = md_event(sb);
659 if (ev1 < mddev->events)
661 } else if (mddev->bitmap) {
662 /* if adding to array with a bitmap, then we can accept an
663 * older device ... but not too old.
665 __u64 ev1 = md_event(sb);
666 if (ev1 < mddev->bitmap->events_cleared)
668 } else /* just a hot-add of a new device, leave raid_disk at -1 */
671 if (mddev->level != LEVEL_MULTIPATH) {
673 desc = sb->disks + rdev->desc_nr;
675 if (desc->state & (1<<MD_DISK_FAULTY))
677 else if (desc->state & (1<<MD_DISK_SYNC) &&
678 desc->raid_disk < mddev->raid_disks) {
680 rdev->raid_disk = desc->raid_disk;
682 } else /* MULTIPATH are always insync */
688 * sync_super for 0.90.0
690 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
693 struct list_head *tmp;
695 int next_spare = mddev->raid_disks;
697 /* make rdev->sb match mddev data..
700 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
701 * 3/ any empty disks < next_spare become removed
703 * disks[0] gets initialised to REMOVED because
704 * we cannot be sure from other fields if it has
705 * been initialised or not.
708 int active=0, working=0,failed=0,spare=0,nr_disks=0;
710 sb = (mdp_super_t*)page_address(rdev->sb_page);
712 memset(sb, 0, sizeof(*sb));
714 sb->md_magic = MD_SB_MAGIC;
715 sb->major_version = mddev->major_version;
716 sb->minor_version = mddev->minor_version;
717 sb->patch_version = mddev->patch_version;
718 sb->gvalid_words = 0; /* ignored */
719 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
720 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
721 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
722 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
724 sb->ctime = mddev->ctime;
725 sb->level = mddev->level;
726 sb->size = mddev->size;
727 sb->raid_disks = mddev->raid_disks;
728 sb->md_minor = mddev->md_minor;
729 sb->not_persistent = !mddev->persistent;
730 sb->utime = mddev->utime;
732 sb->events_hi = (mddev->events>>32);
733 sb->events_lo = (u32)mddev->events;
737 sb->recovery_cp = mddev->recovery_cp;
738 sb->cp_events_hi = (mddev->events>>32);
739 sb->cp_events_lo = (u32)mddev->events;
740 if (mddev->recovery_cp == MaxSector)
741 sb->state = (1<< MD_SB_CLEAN);
745 sb->layout = mddev->layout;
746 sb->chunk_size = mddev->chunk_size;
748 if (mddev->bitmap && mddev->bitmap_file == NULL)
749 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
751 sb->disks[0].state = (1<<MD_DISK_REMOVED);
752 ITERATE_RDEV(mddev,rdev2,tmp) {
754 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
755 rdev2->desc_nr = rdev2->raid_disk;
757 rdev2->desc_nr = next_spare++;
758 d = &sb->disks[rdev2->desc_nr];
760 d->number = rdev2->desc_nr;
761 d->major = MAJOR(rdev2->bdev->bd_dev);
762 d->minor = MINOR(rdev2->bdev->bd_dev);
763 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
764 d->raid_disk = rdev2->raid_disk;
766 d->raid_disk = rdev2->desc_nr; /* compatibility */
768 d->state = (1<<MD_DISK_FAULTY);
770 } else if (rdev2->in_sync) {
771 d->state = (1<<MD_DISK_ACTIVE);
772 d->state |= (1<<MD_DISK_SYNC);
782 /* now set the "removed" and "faulty" bits on any missing devices */
783 for (i=0 ; i < mddev->raid_disks ; i++) {
784 mdp_disk_t *d = &sb->disks[i];
785 if (d->state == 0 && d->number == 0) {
788 d->state = (1<<MD_DISK_REMOVED);
789 d->state |= (1<<MD_DISK_FAULTY);
793 sb->nr_disks = nr_disks;
794 sb->active_disks = active;
795 sb->working_disks = working;
796 sb->failed_disks = failed;
797 sb->spare_disks = spare;
799 sb->this_disk = sb->disks[rdev->desc_nr];
800 sb->sb_csum = calc_sb_csum(sb);
804 * version 1 superblock
807 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
809 unsigned int disk_csum, csum;
810 unsigned long long newcsum;
811 int size = 256 + le32_to_cpu(sb->max_dev)*2;
812 unsigned int *isuper = (unsigned int*)sb;
815 disk_csum = sb->sb_csum;
818 for (i=0; size>=4; size -= 4 )
819 newcsum += le32_to_cpu(*isuper++);
822 newcsum += le16_to_cpu(*(unsigned short*) isuper);
824 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
825 sb->sb_csum = disk_csum;
826 return cpu_to_le32(csum);
829 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
831 struct mdp_superblock_1 *sb;
834 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
837 * Calculate the position of the superblock.
838 * It is always aligned to a 4K boundary and
839 * depeding on minor_version, it can be:
840 * 0: At least 8K, but less than 12K, from end of device
841 * 1: At start of device
842 * 2: 4K from start of device.
844 switch(minor_version) {
846 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
848 sb_offset &= ~(sector_t)(4*2-1);
849 /* convert from sectors to K */
861 rdev->sb_offset = sb_offset;
863 ret = read_disk_sb(rdev);
867 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
869 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
870 sb->major_version != cpu_to_le32(1) ||
871 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
872 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
873 sb->feature_map != 0)
876 if (calc_sb_1_csum(sb) != sb->sb_csum) {
877 printk("md: invalid superblock checksum on %s\n",
878 bdevname(rdev->bdev,b));
881 if (le64_to_cpu(sb->data_size) < 10) {
882 printk("md: data_size too small on %s\n",
883 bdevname(rdev->bdev,b));
886 rdev->preferred_minor = 0xffff;
887 rdev->data_offset = le64_to_cpu(sb->data_offset);
893 struct mdp_superblock_1 *refsb =
894 (struct mdp_superblock_1*)page_address(refdev->sb_page);
896 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
897 sb->level != refsb->level ||
898 sb->layout != refsb->layout ||
899 sb->chunksize != refsb->chunksize) {
900 printk(KERN_WARNING "md: %s has strangely different"
901 " superblock to %s\n",
902 bdevname(rdev->bdev,b),
903 bdevname(refdev->bdev,b2));
906 ev1 = le64_to_cpu(sb->events);
907 ev2 = le64_to_cpu(refsb->events);
913 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
915 rdev->size = rdev->sb_offset;
916 if (rdev->size < le64_to_cpu(sb->data_size)/2)
918 rdev->size = le64_to_cpu(sb->data_size)/2;
919 if (le32_to_cpu(sb->chunksize))
920 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
924 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
926 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
928 rdev->raid_disk = -1;
930 if (mddev->raid_disks == 0) {
931 mddev->major_version = 1;
932 mddev->patch_version = 0;
933 mddev->persistent = 1;
934 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
935 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
936 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
937 mddev->level = le32_to_cpu(sb->level);
938 mddev->layout = le32_to_cpu(sb->layout);
939 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
940 mddev->size = le64_to_cpu(sb->size)/2;
941 mddev->events = le64_to_cpu(sb->events);
942 mddev->bitmap_offset = 0;
943 mddev->default_bitmap_offset = 0;
944 if (mddev->minor_version == 0)
945 mddev->default_bitmap_offset = -(64*1024)/512;
947 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
948 memcpy(mddev->uuid, sb->set_uuid, 16);
950 mddev->max_disks = (4096-256)/2;
952 if ((le32_to_cpu(sb->feature_map) & 1) &&
953 mddev->bitmap_file == NULL ) {
954 if (mddev->level != 1) {
955 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
958 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
960 } else if (mddev->pers == NULL) {
961 /* Insist of good event counter while assembling */
962 __u64 ev1 = le64_to_cpu(sb->events);
964 if (ev1 < mddev->events)
966 } else if (mddev->bitmap) {
967 /* If adding to array with a bitmap, then we can accept an
968 * older device, but not too old.
970 __u64 ev1 = le64_to_cpu(sb->events);
971 if (ev1 < mddev->bitmap->events_cleared)
973 } else /* just a hot-add of a new device, leave raid_disk at -1 */
976 if (mddev->level != LEVEL_MULTIPATH) {
978 rdev->desc_nr = le32_to_cpu(sb->dev_number);
979 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
981 case 0xffff: /* spare */
984 case 0xfffe: /* faulty */
990 rdev->raid_disk = role;
993 } else /* MULTIPATH are always insync */
999 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1001 struct mdp_superblock_1 *sb;
1002 struct list_head *tmp;
1005 /* make rdev->sb match mddev and rdev data. */
1007 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1009 sb->feature_map = 0;
1011 memset(sb->pad1, 0, sizeof(sb->pad1));
1012 memset(sb->pad2, 0, sizeof(sb->pad2));
1013 memset(sb->pad3, 0, sizeof(sb->pad3));
1015 sb->utime = cpu_to_le64((__u64)mddev->utime);
1016 sb->events = cpu_to_le64(mddev->events);
1018 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1020 sb->resync_offset = cpu_to_le64(0);
1022 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1023 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1024 sb->feature_map = cpu_to_le32(1);
1028 ITERATE_RDEV(mddev,rdev2,tmp)
1029 if (rdev2->desc_nr+1 > max_dev)
1030 max_dev = rdev2->desc_nr+1;
1032 sb->max_dev = cpu_to_le32(max_dev);
1033 for (i=0; i<max_dev;i++)
1034 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1036 ITERATE_RDEV(mddev,rdev2,tmp) {
1039 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1040 else if (rdev2->in_sync)
1041 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1043 sb->dev_roles[i] = cpu_to_le16(0xffff);
1046 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1047 sb->sb_csum = calc_sb_1_csum(sb);
1051 static struct super_type super_types[] = {
1054 .owner = THIS_MODULE,
1055 .load_super = super_90_load,
1056 .validate_super = super_90_validate,
1057 .sync_super = super_90_sync,
1061 .owner = THIS_MODULE,
1062 .load_super = super_1_load,
1063 .validate_super = super_1_validate,
1064 .sync_super = super_1_sync,
1068 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1070 struct list_head *tmp;
1073 ITERATE_RDEV(mddev,rdev,tmp)
1074 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1080 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1082 struct list_head *tmp;
1085 ITERATE_RDEV(mddev1,rdev,tmp)
1086 if (match_dev_unit(mddev2, rdev))
1092 static LIST_HEAD(pending_raid_disks);
1094 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1096 mdk_rdev_t *same_pdev;
1097 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1103 same_pdev = match_dev_unit(mddev, rdev);
1106 "%s: WARNING: %s appears to be on the same physical"
1107 " disk as %s. True\n protection against single-disk"
1108 " failure might be compromised.\n",
1109 mdname(mddev), bdevname(rdev->bdev,b),
1110 bdevname(same_pdev->bdev,b2));
1112 /* Verify rdev->desc_nr is unique.
1113 * If it is -1, assign a free number, else
1114 * check number is not in use
1116 if (rdev->desc_nr < 0) {
1118 if (mddev->pers) choice = mddev->raid_disks;
1119 while (find_rdev_nr(mddev, choice))
1121 rdev->desc_nr = choice;
1123 if (find_rdev_nr(mddev, rdev->desc_nr))
1127 list_add(&rdev->same_set, &mddev->disks);
1128 rdev->mddev = mddev;
1129 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1133 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1135 char b[BDEVNAME_SIZE];
1140 list_del_init(&rdev->same_set);
1141 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1146 * prevent the device from being mounted, repartitioned or
1147 * otherwise reused by a RAID array (or any other kernel
1148 * subsystem), by bd_claiming the device.
1150 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1153 struct block_device *bdev;
1154 char b[BDEVNAME_SIZE];
1156 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1158 printk(KERN_ERR "md: could not open %s.\n",
1159 __bdevname(dev, b));
1160 return PTR_ERR(bdev);
1162 err = bd_claim(bdev, rdev);
1164 printk(KERN_ERR "md: could not bd_claim %s.\n",
1173 static void unlock_rdev(mdk_rdev_t *rdev)
1175 struct block_device *bdev = rdev->bdev;
1183 void md_autodetect_dev(dev_t dev);
1185 static void export_rdev(mdk_rdev_t * rdev)
1187 char b[BDEVNAME_SIZE];
1188 printk(KERN_INFO "md: export_rdev(%s)\n",
1189 bdevname(rdev->bdev,b));
1193 list_del_init(&rdev->same_set);
1195 md_autodetect_dev(rdev->bdev->bd_dev);
1201 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1203 unbind_rdev_from_array(rdev);
1207 static void export_array(mddev_t *mddev)
1209 struct list_head *tmp;
1212 ITERATE_RDEV(mddev,rdev,tmp) {
1217 kick_rdev_from_array(rdev);
1219 if (!list_empty(&mddev->disks))
1221 mddev->raid_disks = 0;
1222 mddev->major_version = 0;
1225 static void print_desc(mdp_disk_t *desc)
1227 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1228 desc->major,desc->minor,desc->raid_disk,desc->state);
1231 static void print_sb(mdp_super_t *sb)
1236 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1237 sb->major_version, sb->minor_version, sb->patch_version,
1238 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1240 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1241 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1242 sb->md_minor, sb->layout, sb->chunk_size);
1243 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1244 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1245 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1246 sb->failed_disks, sb->spare_disks,
1247 sb->sb_csum, (unsigned long)sb->events_lo);
1250 for (i = 0; i < MD_SB_DISKS; i++) {
1253 desc = sb->disks + i;
1254 if (desc->number || desc->major || desc->minor ||
1255 desc->raid_disk || (desc->state && (desc->state != 4))) {
1256 printk(" D %2d: ", i);
1260 printk(KERN_INFO "md: THIS: ");
1261 print_desc(&sb->this_disk);
1265 static void print_rdev(mdk_rdev_t *rdev)
1267 char b[BDEVNAME_SIZE];
1268 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1269 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1270 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1271 if (rdev->sb_loaded) {
1272 printk(KERN_INFO "md: rdev superblock:\n");
1273 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1275 printk(KERN_INFO "md: no rdev superblock!\n");
1278 void md_print_devices(void)
1280 struct list_head *tmp, *tmp2;
1283 char b[BDEVNAME_SIZE];
1286 printk("md: **********************************\n");
1287 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1288 printk("md: **********************************\n");
1289 ITERATE_MDDEV(mddev,tmp) {
1292 bitmap_print_sb(mddev->bitmap);
1294 printk("%s: ", mdname(mddev));
1295 ITERATE_RDEV(mddev,rdev,tmp2)
1296 printk("<%s>", bdevname(rdev->bdev,b));
1299 ITERATE_RDEV(mddev,rdev,tmp2)
1302 printk("md: **********************************\n");
1307 static void sync_sbs(mddev_t * mddev)
1310 struct list_head *tmp;
1312 ITERATE_RDEV(mddev,rdev,tmp) {
1313 super_types[mddev->major_version].
1314 sync_super(mddev, rdev);
1315 rdev->sb_loaded = 1;
1319 static void md_update_sb(mddev_t * mddev)
1322 struct list_head *tmp;
1327 spin_lock(&mddev->write_lock);
1328 sync_req = mddev->in_sync;
1329 mddev->utime = get_seconds();
1332 if (!mddev->events) {
1334 * oops, this 64-bit counter should never wrap.
1335 * Either we are in around ~1 trillion A.C., assuming
1336 * 1 reboot per second, or we have a bug:
1341 mddev->sb_dirty = 2;
1345 * do not write anything to disk if using
1346 * nonpersistent superblocks
1348 if (!mddev->persistent) {
1349 mddev->sb_dirty = 0;
1350 spin_unlock(&mddev->write_lock);
1351 wake_up(&mddev->sb_wait);
1354 spin_unlock(&mddev->write_lock);
1357 "md: updating %s RAID superblock on device (in sync %d)\n",
1358 mdname(mddev),mddev->in_sync);
1360 err = bitmap_update_sb(mddev->bitmap);
1361 ITERATE_RDEV(mddev,rdev,tmp) {
1362 char b[BDEVNAME_SIZE];
1363 dprintk(KERN_INFO "md: ");
1365 dprintk("(skipping faulty ");
1367 dprintk("%s ", bdevname(rdev->bdev,b));
1368 if (!rdev->faulty) {
1369 md_super_write(mddev,rdev,
1370 rdev->sb_offset<<1, MD_SB_BYTES,
1372 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1373 bdevname(rdev->bdev,b),
1374 (unsigned long long)rdev->sb_offset);
1378 if (mddev->level == LEVEL_MULTIPATH)
1379 /* only need to write one superblock... */
1382 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1383 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1385 spin_lock(&mddev->write_lock);
1386 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1387 /* have to write it out again */
1388 spin_unlock(&mddev->write_lock);
1391 mddev->sb_dirty = 0;
1392 spin_unlock(&mddev->write_lock);
1393 wake_up(&mddev->sb_wait);
1398 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1400 * mark the device faulty if:
1402 * - the device is nonexistent (zero size)
1403 * - the device has no valid superblock
1405 * a faulty rdev _never_ has rdev->sb set.
1407 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1409 char b[BDEVNAME_SIZE];
1414 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1416 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1417 return ERR_PTR(-ENOMEM);
1419 memset(rdev, 0, sizeof(*rdev));
1421 if ((err = alloc_disk_sb(rdev)))
1424 err = lock_rdev(rdev, newdev);
1431 rdev->data_offset = 0;
1432 atomic_set(&rdev->nr_pending, 0);
1434 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1437 "md: %s has zero or unknown size, marking faulty!\n",
1438 bdevname(rdev->bdev,b));
1443 if (super_format >= 0) {
1444 err = super_types[super_format].
1445 load_super(rdev, NULL, super_minor);
1446 if (err == -EINVAL) {
1448 "md: %s has invalid sb, not importing!\n",
1449 bdevname(rdev->bdev,b));
1454 "md: could not read %s's sb, not importing!\n",
1455 bdevname(rdev->bdev,b));
1459 INIT_LIST_HEAD(&rdev->same_set);
1464 if (rdev->sb_page) {
1470 return ERR_PTR(err);
1474 * Check a full RAID array for plausibility
1478 static void analyze_sbs(mddev_t * mddev)
1481 struct list_head *tmp;
1482 mdk_rdev_t *rdev, *freshest;
1483 char b[BDEVNAME_SIZE];
1486 ITERATE_RDEV(mddev,rdev,tmp)
1487 switch (super_types[mddev->major_version].
1488 load_super(rdev, freshest, mddev->minor_version)) {
1496 "md: fatal superblock inconsistency in %s"
1497 " -- removing from array\n",
1498 bdevname(rdev->bdev,b));
1499 kick_rdev_from_array(rdev);
1503 super_types[mddev->major_version].
1504 validate_super(mddev, freshest);
1507 ITERATE_RDEV(mddev,rdev,tmp) {
1508 if (rdev != freshest)
1509 if (super_types[mddev->major_version].
1510 validate_super(mddev, rdev)) {
1511 printk(KERN_WARNING "md: kicking non-fresh %s"
1513 bdevname(rdev->bdev,b));
1514 kick_rdev_from_array(rdev);
1517 if (mddev->level == LEVEL_MULTIPATH) {
1518 rdev->desc_nr = i++;
1519 rdev->raid_disk = rdev->desc_nr;
1526 if (mddev->recovery_cp != MaxSector &&
1528 printk(KERN_ERR "md: %s: raid array is not clean"
1529 " -- starting background reconstruction\n",
1536 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1538 static DECLARE_MUTEX(disks_sem);
1539 mddev_t *mddev = mddev_find(dev);
1540 struct gendisk *disk;
1541 int partitioned = (MAJOR(dev) != MD_MAJOR);
1542 int shift = partitioned ? MdpMinorShift : 0;
1543 int unit = MINOR(dev) >> shift;
1549 if (mddev->gendisk) {
1554 disk = alloc_disk(1 << shift);
1560 disk->major = MAJOR(dev);
1561 disk->first_minor = unit << shift;
1563 sprintf(disk->disk_name, "md_d%d", unit);
1564 sprintf(disk->devfs_name, "md/d%d", unit);
1566 sprintf(disk->disk_name, "md%d", unit);
1567 sprintf(disk->devfs_name, "md/%d", unit);
1569 disk->fops = &md_fops;
1570 disk->private_data = mddev;
1571 disk->queue = mddev->queue;
1573 mddev->gendisk = disk;
1578 void md_wakeup_thread(mdk_thread_t *thread);
1580 static void md_safemode_timeout(unsigned long data)
1582 mddev_t *mddev = (mddev_t *) data;
1584 mddev->safemode = 1;
1585 md_wakeup_thread(mddev->thread);
1589 static int do_md_run(mddev_t * mddev)
1593 struct list_head *tmp;
1595 struct gendisk *disk;
1596 char b[BDEVNAME_SIZE];
1598 if (list_empty(&mddev->disks))
1599 /* cannot run an array with no devices.. */
1606 * Analyze all RAID superblock(s)
1608 if (!mddev->raid_disks)
1611 chunk_size = mddev->chunk_size;
1612 pnum = level_to_pers(mddev->level);
1614 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1617 * 'default chunksize' in the old md code used to
1618 * be PAGE_SIZE, baaad.
1619 * we abort here to be on the safe side. We don't
1620 * want to continue the bad practice.
1623 "no chunksize specified, see 'man raidtab'\n");
1626 if (chunk_size > MAX_CHUNK_SIZE) {
1627 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1628 chunk_size, MAX_CHUNK_SIZE);
1632 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1634 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1635 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1638 if (chunk_size < PAGE_SIZE) {
1639 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1640 chunk_size, PAGE_SIZE);
1644 /* devices must have minimum size of one chunk */
1645 ITERATE_RDEV(mddev,rdev,tmp) {
1648 if (rdev->size < chunk_size / 1024) {
1650 "md: Dev %s smaller than chunk_size:"
1652 bdevname(rdev->bdev,b),
1653 (unsigned long long)rdev->size,
1663 request_module("md-personality-%d", pnum);
1668 * Drop all container device buffers, from now on
1669 * the only valid external interface is through the md
1671 * Also find largest hardsector size
1673 ITERATE_RDEV(mddev,rdev,tmp) {
1676 sync_blockdev(rdev->bdev);
1677 invalidate_bdev(rdev->bdev, 0);
1680 md_probe(mddev->unit, NULL, NULL);
1681 disk = mddev->gendisk;
1685 spin_lock(&pers_lock);
1686 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1687 spin_unlock(&pers_lock);
1688 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1693 mddev->pers = pers[pnum];
1694 spin_unlock(&pers_lock);
1696 mddev->recovery = 0;
1697 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1699 /* before we start the array running, initialise the bitmap */
1700 err = bitmap_create(mddev);
1702 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1703 mdname(mddev), err);
1705 err = mddev->pers->run(mddev);
1707 printk(KERN_ERR "md: pers->run() failed ...\n");
1708 module_put(mddev->pers->owner);
1710 bitmap_destroy(mddev);
1713 atomic_set(&mddev->writes_pending,0);
1714 mddev->safemode = 0;
1715 mddev->safemode_timer.function = md_safemode_timeout;
1716 mddev->safemode_timer.data = (unsigned long) mddev;
1717 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1720 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1721 md_wakeup_thread(mddev->thread);
1723 if (mddev->sb_dirty)
1724 md_update_sb(mddev);
1726 set_capacity(disk, mddev->array_size<<1);
1728 /* If we call blk_queue_make_request here, it will
1729 * re-initialise max_sectors etc which may have been
1730 * refined inside -> run. So just set the bits we need to set.
1731 * Most initialisation happended when we called
1732 * blk_queue_make_request(..., md_fail_request)
1735 mddev->queue->queuedata = mddev;
1736 mddev->queue->make_request_fn = mddev->pers->make_request;
1742 static int restart_array(mddev_t *mddev)
1744 struct gendisk *disk = mddev->gendisk;
1748 * Complain if it has no devices
1751 if (list_empty(&mddev->disks))
1759 mddev->safemode = 0;
1761 set_disk_ro(disk, 0);
1763 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1766 * Kick recovery or resync if necessary
1768 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1769 md_wakeup_thread(mddev->thread);
1772 printk(KERN_ERR "md: %s has no personality assigned.\n",
1781 static int do_md_stop(mddev_t * mddev, int ro)
1784 struct gendisk *disk = mddev->gendisk;
1787 if (atomic_read(&mddev->active)>2) {
1788 printk("md: %s still in use.\n",mdname(mddev));
1792 if (mddev->sync_thread) {
1793 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1794 md_unregister_thread(mddev->sync_thread);
1795 mddev->sync_thread = NULL;
1798 del_timer_sync(&mddev->safemode_timer);
1800 invalidate_partition(disk, 0);
1808 bitmap_flush(mddev);
1809 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1811 set_disk_ro(disk, 0);
1812 blk_queue_make_request(mddev->queue, md_fail_request);
1813 mddev->pers->stop(mddev);
1814 module_put(mddev->pers->owner);
1819 if (!mddev->in_sync) {
1820 /* mark array as shutdown cleanly */
1822 md_update_sb(mddev);
1825 set_disk_ro(disk, 1);
1828 bitmap_destroy(mddev);
1829 if (mddev->bitmap_file) {
1830 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1831 fput(mddev->bitmap_file);
1832 mddev->bitmap_file = NULL;
1834 mddev->bitmap_offset = 0;
1837 * Free resources if final stop
1840 struct gendisk *disk;
1841 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1843 export_array(mddev);
1845 mddev->array_size = 0;
1846 disk = mddev->gendisk;
1848 set_capacity(disk, 0);
1851 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1858 static void autorun_array(mddev_t *mddev)
1861 struct list_head *tmp;
1864 if (list_empty(&mddev->disks))
1867 printk(KERN_INFO "md: running: ");
1869 ITERATE_RDEV(mddev,rdev,tmp) {
1870 char b[BDEVNAME_SIZE];
1871 printk("<%s>", bdevname(rdev->bdev,b));
1875 err = do_md_run (mddev);
1877 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1878 do_md_stop (mddev, 0);
1883 * lets try to run arrays based on all disks that have arrived
1884 * until now. (those are in pending_raid_disks)
1886 * the method: pick the first pending disk, collect all disks with
1887 * the same UUID, remove all from the pending list and put them into
1888 * the 'same_array' list. Then order this list based on superblock
1889 * update time (freshest comes first), kick out 'old' disks and
1890 * compare superblocks. If everything's fine then run it.
1892 * If "unit" is allocated, then bump its reference count
1894 static void autorun_devices(int part)
1896 struct list_head candidates;
1897 struct list_head *tmp;
1898 mdk_rdev_t *rdev0, *rdev;
1900 char b[BDEVNAME_SIZE];
1902 printk(KERN_INFO "md: autorun ...\n");
1903 while (!list_empty(&pending_raid_disks)) {
1905 rdev0 = list_entry(pending_raid_disks.next,
1906 mdk_rdev_t, same_set);
1908 printk(KERN_INFO "md: considering %s ...\n",
1909 bdevname(rdev0->bdev,b));
1910 INIT_LIST_HEAD(&candidates);
1911 ITERATE_RDEV_PENDING(rdev,tmp)
1912 if (super_90_load(rdev, rdev0, 0) >= 0) {
1913 printk(KERN_INFO "md: adding %s ...\n",
1914 bdevname(rdev->bdev,b));
1915 list_move(&rdev->same_set, &candidates);
1918 * now we have a set of devices, with all of them having
1919 * mostly sane superblocks. It's time to allocate the
1922 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1923 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1924 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1928 dev = MKDEV(mdp_major,
1929 rdev0->preferred_minor << MdpMinorShift);
1931 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1933 md_probe(dev, NULL, NULL);
1934 mddev = mddev_find(dev);
1937 "md: cannot allocate memory for md drive.\n");
1940 if (mddev_lock(mddev))
1941 printk(KERN_WARNING "md: %s locked, cannot run\n",
1943 else if (mddev->raid_disks || mddev->major_version
1944 || !list_empty(&mddev->disks)) {
1946 "md: %s already running, cannot run %s\n",
1947 mdname(mddev), bdevname(rdev0->bdev,b));
1948 mddev_unlock(mddev);
1950 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1951 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1952 list_del_init(&rdev->same_set);
1953 if (bind_rdev_to_array(rdev, mddev))
1956 autorun_array(mddev);
1957 mddev_unlock(mddev);
1959 /* on success, candidates will be empty, on error
1962 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1966 printk(KERN_INFO "md: ... autorun DONE.\n");
1970 * import RAID devices based on one partition
1971 * if possible, the array gets run as well.
1974 static int autostart_array(dev_t startdev)
1976 char b[BDEVNAME_SIZE];
1977 int err = -EINVAL, i;
1978 mdp_super_t *sb = NULL;
1979 mdk_rdev_t *start_rdev = NULL, *rdev;
1981 start_rdev = md_import_device(startdev, 0, 0);
1982 if (IS_ERR(start_rdev))
1986 /* NOTE: this can only work for 0.90.0 superblocks */
1987 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1988 if (sb->major_version != 0 ||
1989 sb->minor_version != 90 ) {
1990 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1991 export_rdev(start_rdev);
1995 if (start_rdev->faulty) {
1997 "md: can not autostart based on faulty %s!\n",
1998 bdevname(start_rdev->bdev,b));
1999 export_rdev(start_rdev);
2002 list_add(&start_rdev->same_set, &pending_raid_disks);
2004 for (i = 0; i < MD_SB_DISKS; i++) {
2005 mdp_disk_t *desc = sb->disks + i;
2006 dev_t dev = MKDEV(desc->major, desc->minor);
2010 if (dev == startdev)
2012 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2014 rdev = md_import_device(dev, 0, 0);
2018 list_add(&rdev->same_set, &pending_raid_disks);
2022 * possibly return codes
2030 static int get_version(void __user * arg)
2034 ver.major = MD_MAJOR_VERSION;
2035 ver.minor = MD_MINOR_VERSION;
2036 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2038 if (copy_to_user(arg, &ver, sizeof(ver)))
2044 static int get_array_info(mddev_t * mddev, void __user * arg)
2046 mdu_array_info_t info;
2047 int nr,working,active,failed,spare;
2049 struct list_head *tmp;
2051 nr=working=active=failed=spare=0;
2052 ITERATE_RDEV(mddev,rdev,tmp) {
2065 info.major_version = mddev->major_version;
2066 info.minor_version = mddev->minor_version;
2067 info.patch_version = MD_PATCHLEVEL_VERSION;
2068 info.ctime = mddev->ctime;
2069 info.level = mddev->level;
2070 info.size = mddev->size;
2072 info.raid_disks = mddev->raid_disks;
2073 info.md_minor = mddev->md_minor;
2074 info.not_persistent= !mddev->persistent;
2076 info.utime = mddev->utime;
2079 info.state = (1<<MD_SB_CLEAN);
2080 if (mddev->bitmap && mddev->bitmap_offset)
2081 info.state = (1<<MD_SB_BITMAP_PRESENT);
2082 info.active_disks = active;
2083 info.working_disks = working;
2084 info.failed_disks = failed;
2085 info.spare_disks = spare;
2087 info.layout = mddev->layout;
2088 info.chunk_size = mddev->chunk_size;
2090 if (copy_to_user(arg, &info, sizeof(info)))
2096 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2098 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2099 char *ptr, *buf = NULL;
2102 file = kmalloc(sizeof(*file), GFP_KERNEL);
2106 /* bitmap disabled, zero the first byte and copy out */
2107 if (!mddev->bitmap || !mddev->bitmap->file) {
2108 file->pathname[0] = '\0';
2112 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2116 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2120 strcpy(file->pathname, ptr);
2124 if (copy_to_user(arg, file, sizeof(*file)))
2132 static int get_disk_info(mddev_t * mddev, void __user * arg)
2134 mdu_disk_info_t info;
2138 if (copy_from_user(&info, arg, sizeof(info)))
2143 rdev = find_rdev_nr(mddev, nr);
2145 info.major = MAJOR(rdev->bdev->bd_dev);
2146 info.minor = MINOR(rdev->bdev->bd_dev);
2147 info.raid_disk = rdev->raid_disk;
2150 info.state |= (1<<MD_DISK_FAULTY);
2151 else if (rdev->in_sync) {
2152 info.state |= (1<<MD_DISK_ACTIVE);
2153 info.state |= (1<<MD_DISK_SYNC);
2156 info.major = info.minor = 0;
2157 info.raid_disk = -1;
2158 info.state = (1<<MD_DISK_REMOVED);
2161 if (copy_to_user(arg, &info, sizeof(info)))
2167 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2169 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2171 dev_t dev = MKDEV(info->major,info->minor);
2173 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2176 if (!mddev->raid_disks) {
2178 /* expecting a device which has a superblock */
2179 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2182 "md: md_import_device returned %ld\n",
2184 return PTR_ERR(rdev);
2186 if (!list_empty(&mddev->disks)) {
2187 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2188 mdk_rdev_t, same_set);
2189 int err = super_types[mddev->major_version]
2190 .load_super(rdev, rdev0, mddev->minor_version);
2193 "md: %s has different UUID to %s\n",
2194 bdevname(rdev->bdev,b),
2195 bdevname(rdev0->bdev,b2));
2200 err = bind_rdev_to_array(rdev, mddev);
2207 * add_new_disk can be used once the array is assembled
2208 * to add "hot spares". They must already have a superblock
2213 if (!mddev->pers->hot_add_disk) {
2215 "%s: personality does not support diskops!\n",
2219 rdev = md_import_device(dev, mddev->major_version,
2220 mddev->minor_version);
2223 "md: md_import_device returned %ld\n",
2225 return PTR_ERR(rdev);
2227 /* set save_raid_disk if appropriate */
2228 if (!mddev->persistent) {
2229 if (info->state & (1<<MD_DISK_SYNC) &&
2230 info->raid_disk < mddev->raid_disks)
2231 rdev->raid_disk = info->raid_disk;
2233 rdev->raid_disk = -1;
2235 super_types[mddev->major_version].
2236 validate_super(mddev, rdev);
2237 rdev->saved_raid_disk = rdev->raid_disk;
2239 rdev->in_sync = 0; /* just to be sure */
2240 rdev->raid_disk = -1;
2241 err = bind_rdev_to_array(rdev, mddev);
2245 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2246 md_wakeup_thread(mddev->thread);
2250 /* otherwise, add_new_disk is only allowed
2251 * for major_version==0 superblocks
2253 if (mddev->major_version != 0) {
2254 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2259 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2261 rdev = md_import_device (dev, -1, 0);
2264 "md: error, md_import_device() returned %ld\n",
2266 return PTR_ERR(rdev);
2268 rdev->desc_nr = info->number;
2269 if (info->raid_disk < mddev->raid_disks)
2270 rdev->raid_disk = info->raid_disk;
2272 rdev->raid_disk = -1;
2275 if (rdev->raid_disk < mddev->raid_disks)
2276 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2280 err = bind_rdev_to_array(rdev, mddev);
2286 if (!mddev->persistent) {
2287 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2288 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2290 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2291 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2293 if (!mddev->size || (mddev->size > rdev->size))
2294 mddev->size = rdev->size;
2300 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2302 char b[BDEVNAME_SIZE];
2308 rdev = find_rdev(mddev, dev);
2312 if (rdev->raid_disk >= 0)
2315 kick_rdev_from_array(rdev);
2316 md_update_sb(mddev);
2320 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2321 bdevname(rdev->bdev,b), mdname(mddev));
2325 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2327 char b[BDEVNAME_SIZE];
2335 if (mddev->major_version != 0) {
2336 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2337 " version-0 superblocks.\n",
2341 if (!mddev->pers->hot_add_disk) {
2343 "%s: personality does not support diskops!\n",
2348 rdev = md_import_device (dev, -1, 0);
2351 "md: error, md_import_device() returned %ld\n",
2356 if (mddev->persistent)
2357 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2360 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2362 size = calc_dev_size(rdev, mddev->chunk_size);
2365 if (size < mddev->size) {
2367 "%s: disk size %llu blocks < array size %llu\n",
2368 mdname(mddev), (unsigned long long)size,
2369 (unsigned long long)mddev->size);
2376 "md: can not hot-add faulty %s disk to %s!\n",
2377 bdevname(rdev->bdev,b), mdname(mddev));
2383 bind_rdev_to_array(rdev, mddev);
2386 * The rest should better be atomic, we can have disk failures
2387 * noticed in interrupt contexts ...
2390 if (rdev->desc_nr == mddev->max_disks) {
2391 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2394 goto abort_unbind_export;
2397 rdev->raid_disk = -1;
2399 md_update_sb(mddev);
2402 * Kick recovery, maybe this spare has to be added to the
2403 * array immediately.
2405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2406 md_wakeup_thread(mddev->thread);
2410 abort_unbind_export:
2411 unbind_rdev_from_array(rdev);
2418 /* similar to deny_write_access, but accounts for our holding a reference
2419 * to the file ourselves */
2420 static int deny_bitmap_write_access(struct file * file)
2422 struct inode *inode = file->f_mapping->host;
2424 spin_lock(&inode->i_lock);
2425 if (atomic_read(&inode->i_writecount) > 1) {
2426 spin_unlock(&inode->i_lock);
2429 atomic_set(&inode->i_writecount, -1);
2430 spin_unlock(&inode->i_lock);
2435 static int set_bitmap_file(mddev_t *mddev, int fd)
2440 if (!mddev->pers->quiesce)
2442 if (mddev->recovery || mddev->sync_thread)
2444 /* we should be able to change the bitmap.. */
2450 return -EEXIST; /* cannot add when bitmap is present */
2451 mddev->bitmap_file = fget(fd);
2453 if (mddev->bitmap_file == NULL) {
2454 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2459 err = deny_bitmap_write_access(mddev->bitmap_file);
2461 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2463 fput(mddev->bitmap_file);
2464 mddev->bitmap_file = NULL;
2467 mddev->bitmap_offset = 0; /* file overrides offset */
2468 } else if (mddev->bitmap == NULL)
2469 return -ENOENT; /* cannot remove what isn't there */
2472 mddev->pers->quiesce(mddev, 1);
2474 err = bitmap_create(mddev);
2476 bitmap_destroy(mddev);
2477 mddev->pers->quiesce(mddev, 0);
2478 } else if (fd < 0) {
2479 if (mddev->bitmap_file)
2480 fput(mddev->bitmap_file);
2481 mddev->bitmap_file = NULL;
2488 * set_array_info is used two different ways
2489 * The original usage is when creating a new array.
2490 * In this usage, raid_disks is > 0 and it together with
2491 * level, size, not_persistent,layout,chunksize determine the
2492 * shape of the array.
2493 * This will always create an array with a type-0.90.0 superblock.
2494 * The newer usage is when assembling an array.
2495 * In this case raid_disks will be 0, and the major_version field is
2496 * use to determine which style super-blocks are to be found on the devices.
2497 * The minor and patch _version numbers are also kept incase the
2498 * super_block handler wishes to interpret them.
2500 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2503 if (info->raid_disks == 0) {
2504 /* just setting version number for superblock loading */
2505 if (info->major_version < 0 ||
2506 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2507 super_types[info->major_version].name == NULL) {
2508 /* maybe try to auto-load a module? */
2510 "md: superblock version %d not known\n",
2511 info->major_version);
2514 mddev->major_version = info->major_version;
2515 mddev->minor_version = info->minor_version;
2516 mddev->patch_version = info->patch_version;
2519 mddev->major_version = MD_MAJOR_VERSION;
2520 mddev->minor_version = MD_MINOR_VERSION;
2521 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2522 mddev->ctime = get_seconds();
2524 mddev->level = info->level;
2525 mddev->size = info->size;
2526 mddev->raid_disks = info->raid_disks;
2527 /* don't set md_minor, it is determined by which /dev/md* was
2530 if (info->state & (1<<MD_SB_CLEAN))
2531 mddev->recovery_cp = MaxSector;
2533 mddev->recovery_cp = 0;
2534 mddev->persistent = ! info->not_persistent;
2536 mddev->layout = info->layout;
2537 mddev->chunk_size = info->chunk_size;
2539 mddev->max_disks = MD_SB_DISKS;
2541 mddev->sb_dirty = 1;
2544 * Generate a 128 bit UUID
2546 get_random_bytes(mddev->uuid, 16);
2552 * update_array_info is used to change the configuration of an
2554 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2555 * fields in the info are checked against the array.
2556 * Any differences that cannot be handled will cause an error.
2557 * Normally, only one change can be managed at a time.
2559 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2565 /* calculate expected state,ignoring low bits */
2566 if (mddev->bitmap && mddev->bitmap_offset)
2567 state |= (1 << MD_SB_BITMAP_PRESENT);
2569 if (mddev->major_version != info->major_version ||
2570 mddev->minor_version != info->minor_version ||
2571 /* mddev->patch_version != info->patch_version || */
2572 mddev->ctime != info->ctime ||
2573 mddev->level != info->level ||
2574 /* mddev->layout != info->layout || */
2575 !mddev->persistent != info->not_persistent||
2576 mddev->chunk_size != info->chunk_size ||
2577 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2578 ((state^info->state) & 0xfffffe00)
2581 /* Check there is only one change */
2582 if (mddev->size != info->size) cnt++;
2583 if (mddev->raid_disks != info->raid_disks) cnt++;
2584 if (mddev->layout != info->layout) cnt++;
2585 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2586 if (cnt == 0) return 0;
2587 if (cnt > 1) return -EINVAL;
2589 if (mddev->layout != info->layout) {
2591 * we don't need to do anything at the md level, the
2592 * personality will take care of it all.
2594 if (mddev->pers->reconfig == NULL)
2597 return mddev->pers->reconfig(mddev, info->layout, -1);
2599 if (mddev->size != info->size) {
2601 struct list_head *tmp;
2602 if (mddev->pers->resize == NULL)
2604 /* The "size" is the amount of each device that is used.
2605 * This can only make sense for arrays with redundancy.
2606 * linear and raid0 always use whatever space is available
2607 * We can only consider changing the size if no resync
2608 * or reconstruction is happening, and if the new size
2609 * is acceptable. It must fit before the sb_offset or,
2610 * if that is <data_offset, it must fit before the
2611 * size of each device.
2612 * If size is zero, we find the largest size that fits.
2614 if (mddev->sync_thread)
2616 ITERATE_RDEV(mddev,rdev,tmp) {
2618 int fit = (info->size == 0);
2619 if (rdev->sb_offset > rdev->data_offset)
2620 avail = (rdev->sb_offset*2) - rdev->data_offset;
2622 avail = get_capacity(rdev->bdev->bd_disk)
2623 - rdev->data_offset;
2624 if (fit && (info->size == 0 || info->size > avail/2))
2625 info->size = avail/2;
2626 if (avail < ((sector_t)info->size << 1))
2629 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2631 struct block_device *bdev;
2633 bdev = bdget_disk(mddev->gendisk, 0);
2635 down(&bdev->bd_inode->i_sem);
2636 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2637 up(&bdev->bd_inode->i_sem);
2642 if (mddev->raid_disks != info->raid_disks) {
2643 /* change the number of raid disks */
2644 if (mddev->pers->reshape == NULL)
2646 if (info->raid_disks <= 0 ||
2647 info->raid_disks >= mddev->max_disks)
2649 if (mddev->sync_thread)
2651 rv = mddev->pers->reshape(mddev, info->raid_disks);
2653 struct block_device *bdev;
2655 bdev = bdget_disk(mddev->gendisk, 0);
2657 down(&bdev->bd_inode->i_sem);
2658 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2659 up(&bdev->bd_inode->i_sem);
2664 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2665 if (mddev->pers->quiesce == NULL)
2667 if (mddev->recovery || mddev->sync_thread)
2669 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2670 /* add the bitmap */
2673 if (mddev->default_bitmap_offset == 0)
2675 mddev->bitmap_offset = mddev->default_bitmap_offset;
2676 mddev->pers->quiesce(mddev, 1);
2677 rv = bitmap_create(mddev);
2679 bitmap_destroy(mddev);
2680 mddev->pers->quiesce(mddev, 0);
2682 /* remove the bitmap */
2685 if (mddev->bitmap->file)
2687 mddev->pers->quiesce(mddev, 1);
2688 bitmap_destroy(mddev);
2689 mddev->pers->quiesce(mddev, 0);
2690 mddev->bitmap_offset = 0;
2693 md_update_sb(mddev);
2697 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2701 if (mddev->pers == NULL)
2704 rdev = find_rdev(mddev, dev);
2708 md_error(mddev, rdev);
2712 static int md_ioctl(struct inode *inode, struct file *file,
2713 unsigned int cmd, unsigned long arg)
2716 void __user *argp = (void __user *)arg;
2717 struct hd_geometry __user *loc = argp;
2718 mddev_t *mddev = NULL;
2720 if (!capable(CAP_SYS_ADMIN))
2724 * Commands dealing with the RAID driver but not any
2730 err = get_version(argp);
2733 case PRINT_RAID_DEBUG:
2741 autostart_arrays(arg);
2748 * Commands creating/starting a new array:
2751 mddev = inode->i_bdev->bd_disk->private_data;
2759 if (cmd == START_ARRAY) {
2760 /* START_ARRAY doesn't need to lock the array as autostart_array
2761 * does the locking, and it could even be a different array
2766 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2767 "This will not be supported beyond 2.6\n",
2768 current->comm, current->pid);
2771 err = autostart_array(new_decode_dev(arg));
2773 printk(KERN_WARNING "md: autostart failed!\n");
2779 err = mddev_lock(mddev);
2782 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2789 case SET_ARRAY_INFO:
2791 mdu_array_info_t info;
2793 memset(&info, 0, sizeof(info));
2794 else if (copy_from_user(&info, argp, sizeof(info))) {
2799 err = update_array_info(mddev, &info);
2801 printk(KERN_WARNING "md: couldn't update"
2802 " array info. %d\n", err);
2807 if (!list_empty(&mddev->disks)) {
2809 "md: array %s already has disks!\n",
2814 if (mddev->raid_disks) {
2816 "md: array %s already initialised!\n",
2821 err = set_array_info(mddev, &info);
2823 printk(KERN_WARNING "md: couldn't set"
2824 " array info. %d\n", err);
2834 * Commands querying/configuring an existing array:
2836 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2837 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2838 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2839 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2845 * Commands even a read-only array can execute:
2849 case GET_ARRAY_INFO:
2850 err = get_array_info(mddev, argp);
2853 case GET_BITMAP_FILE:
2854 err = get_bitmap_file(mddev, argp);
2858 err = get_disk_info(mddev, argp);
2861 case RESTART_ARRAY_RW:
2862 err = restart_array(mddev);
2866 err = do_md_stop (mddev, 0);
2870 err = do_md_stop (mddev, 1);
2874 * We have a problem here : there is no easy way to give a CHS
2875 * virtual geometry. We currently pretend that we have a 2 heads
2876 * 4 sectors (with a BIG number of cylinders...). This drives
2877 * dosfs just mad... ;-)
2884 err = put_user (2, (char __user *) &loc->heads);
2887 err = put_user (4, (char __user *) &loc->sectors);
2890 err = put_user(get_capacity(mddev->gendisk)/8,
2891 (short __user *) &loc->cylinders);
2894 err = put_user (get_start_sect(inode->i_bdev),
2895 (long __user *) &loc->start);
2900 * The remaining ioctls are changing the state of the
2901 * superblock, so we do not allow read-only arrays
2913 mdu_disk_info_t info;
2914 if (copy_from_user(&info, argp, sizeof(info)))
2917 err = add_new_disk(mddev, &info);
2921 case HOT_REMOVE_DISK:
2922 err = hot_remove_disk(mddev, new_decode_dev(arg));
2926 err = hot_add_disk(mddev, new_decode_dev(arg));
2929 case SET_DISK_FAULTY:
2930 err = set_disk_faulty(mddev, new_decode_dev(arg));
2934 err = do_md_run (mddev);
2937 case SET_BITMAP_FILE:
2938 err = set_bitmap_file(mddev, (int)arg);
2942 if (_IOC_TYPE(cmd) == MD_MAJOR)
2943 printk(KERN_WARNING "md: %s(pid %d) used"
2944 " obsolete MD ioctl, upgrade your"
2945 " software to use new ictls.\n",
2946 current->comm, current->pid);
2953 mddev_unlock(mddev);
2963 static int md_open(struct inode *inode, struct file *file)
2966 * Succeed if we can lock the mddev, which confirms that
2967 * it isn't being stopped right now.
2969 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2972 if ((err = mddev_lock(mddev)))
2977 mddev_unlock(mddev);
2979 check_disk_change(inode->i_bdev);
2984 static int md_release(struct inode *inode, struct file * file)
2986 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2995 static int md_media_changed(struct gendisk *disk)
2997 mddev_t *mddev = disk->private_data;
2999 return mddev->changed;
3002 static int md_revalidate(struct gendisk *disk)
3004 mddev_t *mddev = disk->private_data;
3009 static struct block_device_operations md_fops =
3011 .owner = THIS_MODULE,
3013 .release = md_release,
3015 .media_changed = md_media_changed,
3016 .revalidate_disk= md_revalidate,
3019 static int md_thread(void * arg)
3021 mdk_thread_t *thread = arg;
3029 daemonize(thread->name, mdname(thread->mddev));
3031 current->exit_signal = SIGCHLD;
3032 allow_signal(SIGKILL);
3033 thread->tsk = current;
3036 * md_thread is a 'system-thread', it's priority should be very
3037 * high. We avoid resource deadlocks individually in each
3038 * raid personality. (RAID5 does preallocation) We also use RR and
3039 * the very same RT priority as kswapd, thus we will never get
3040 * into a priority inversion deadlock.
3042 * we definitely have to have equal or higher priority than
3043 * bdflush, otherwise bdflush will deadlock if there are too
3044 * many dirty RAID5 blocks.
3048 complete(thread->event);
3049 while (thread->run) {
3050 void (*run)(mddev_t *);
3052 wait_event_interruptible_timeout(thread->wqueue,
3053 test_bit(THREAD_WAKEUP, &thread->flags),
3057 clear_bit(THREAD_WAKEUP, &thread->flags);
3063 if (signal_pending(current))
3064 flush_signals(current);
3066 complete(thread->event);
3070 void md_wakeup_thread(mdk_thread_t *thread)
3073 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3074 set_bit(THREAD_WAKEUP, &thread->flags);
3075 wake_up(&thread->wqueue);
3079 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3082 mdk_thread_t *thread;
3084 struct completion event;
3086 thread = (mdk_thread_t *) kmalloc
3087 (sizeof(mdk_thread_t), GFP_KERNEL);
3091 memset(thread, 0, sizeof(mdk_thread_t));
3092 init_waitqueue_head(&thread->wqueue);
3094 init_completion(&event);
3095 thread->event = &event;
3097 thread->mddev = mddev;
3098 thread->name = name;
3099 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3100 ret = kernel_thread(md_thread, thread, 0);
3105 wait_for_completion(&event);
3109 void md_unregister_thread(mdk_thread_t *thread)
3111 struct completion event;
3113 init_completion(&event);
3115 thread->event = &event;
3117 /* As soon as ->run is set to NULL, the task could disappear,
3118 * so we need to hold tasklist_lock until we have sent the signal
3120 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3121 read_lock(&tasklist_lock);
3123 send_sig(SIGKILL, thread->tsk, 1);
3124 read_unlock(&tasklist_lock);
3125 wait_for_completion(&event);
3129 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3136 if (!rdev || rdev->faulty)
3139 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3141 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3142 __builtin_return_address(0),__builtin_return_address(1),
3143 __builtin_return_address(2),__builtin_return_address(3));
3145 if (!mddev->pers->error_handler)
3147 mddev->pers->error_handler(mddev,rdev);
3148 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3149 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3150 md_wakeup_thread(mddev->thread);
3153 /* seq_file implementation /proc/mdstat */
3155 static void status_unused(struct seq_file *seq)
3159 struct list_head *tmp;
3161 seq_printf(seq, "unused devices: ");
3163 ITERATE_RDEV_PENDING(rdev,tmp) {
3164 char b[BDEVNAME_SIZE];
3166 seq_printf(seq, "%s ",
3167 bdevname(rdev->bdev,b));
3170 seq_printf(seq, "<none>");
3172 seq_printf(seq, "\n");
3176 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3178 unsigned long max_blocks, resync, res, dt, db, rt;
3180 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3182 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3183 max_blocks = mddev->resync_max_sectors >> 1;
3185 max_blocks = mddev->size;
3188 * Should not happen.
3194 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3196 int i, x = res/50, y = 20-x;
3197 seq_printf(seq, "[");
3198 for (i = 0; i < x; i++)
3199 seq_printf(seq, "=");
3200 seq_printf(seq, ">");
3201 for (i = 0; i < y; i++)
3202 seq_printf(seq, ".");
3203 seq_printf(seq, "] ");
3205 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3206 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3207 "resync" : "recovery"),
3208 res/10, res % 10, resync, max_blocks);
3211 * We do not want to overflow, so the order of operands and
3212 * the * 100 / 100 trick are important. We do a +1 to be
3213 * safe against division by zero. We only estimate anyway.
3215 * dt: time from mark until now
3216 * db: blocks written from mark until now
3217 * rt: remaining time
3219 dt = ((jiffies - mddev->resync_mark) / HZ);
3221 db = resync - (mddev->resync_mark_cnt/2);
3222 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3224 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3226 seq_printf(seq, " speed=%ldK/sec", db/dt);
3229 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3231 struct list_head *tmp;
3241 spin_lock(&all_mddevs_lock);
3242 list_for_each(tmp,&all_mddevs)
3244 mddev = list_entry(tmp, mddev_t, all_mddevs);
3246 spin_unlock(&all_mddevs_lock);
3249 spin_unlock(&all_mddevs_lock);
3251 return (void*)2;/* tail */
3255 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3257 struct list_head *tmp;
3258 mddev_t *next_mddev, *mddev = v;
3264 spin_lock(&all_mddevs_lock);
3266 tmp = all_mddevs.next;
3268 tmp = mddev->all_mddevs.next;
3269 if (tmp != &all_mddevs)
3270 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3272 next_mddev = (void*)2;
3275 spin_unlock(&all_mddevs_lock);
3283 static void md_seq_stop(struct seq_file *seq, void *v)
3287 if (mddev && v != (void*)1 && v != (void*)2)
3291 static int md_seq_show(struct seq_file *seq, void *v)
3295 struct list_head *tmp2;
3298 struct bitmap *bitmap;
3300 if (v == (void*)1) {
3301 seq_printf(seq, "Personalities : ");
3302 spin_lock(&pers_lock);
3303 for (i = 0; i < MAX_PERSONALITY; i++)
3305 seq_printf(seq, "[%s] ", pers[i]->name);
3307 spin_unlock(&pers_lock);
3308 seq_printf(seq, "\n");
3311 if (v == (void*)2) {
3316 if (mddev_lock(mddev)!=0)
3318 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3319 seq_printf(seq, "%s : %sactive", mdname(mddev),
3320 mddev->pers ? "" : "in");
3323 seq_printf(seq, " (read-only)");
3324 seq_printf(seq, " %s", mddev->pers->name);
3328 ITERATE_RDEV(mddev,rdev,tmp2) {
3329 char b[BDEVNAME_SIZE];
3330 seq_printf(seq, " %s[%d]",
3331 bdevname(rdev->bdev,b), rdev->desc_nr);
3333 seq_printf(seq, "(F)");
3339 if (!list_empty(&mddev->disks)) {
3341 seq_printf(seq, "\n %llu blocks",
3342 (unsigned long long)mddev->array_size);
3344 seq_printf(seq, "\n %llu blocks",
3345 (unsigned long long)size);
3349 mddev->pers->status (seq, mddev);
3350 seq_printf(seq, "\n ");
3351 if (mddev->curr_resync > 2) {
3352 status_resync (seq, mddev);
3353 seq_printf(seq, "\n ");
3354 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3355 seq_printf(seq, " resync=DELAYED\n ");
3357 seq_printf(seq, "\n ");
3359 if ((bitmap = mddev->bitmap)) {
3360 unsigned long chunk_kb;
3361 unsigned long flags;
3362 spin_lock_irqsave(&bitmap->lock, flags);
3363 chunk_kb = bitmap->chunksize >> 10;
3364 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3366 bitmap->pages - bitmap->missing_pages,
3368 (bitmap->pages - bitmap->missing_pages)
3369 << (PAGE_SHIFT - 10),
3370 chunk_kb ? chunk_kb : bitmap->chunksize,
3371 chunk_kb ? "KB" : "B");
3373 seq_printf(seq, ", file: ");
3374 seq_path(seq, bitmap->file->f_vfsmnt,
3375 bitmap->file->f_dentry," \t\n");
3378 seq_printf(seq, "\n");
3379 spin_unlock_irqrestore(&bitmap->lock, flags);
3382 seq_printf(seq, "\n");
3384 mddev_unlock(mddev);
3389 static struct seq_operations md_seq_ops = {
3390 .start = md_seq_start,
3391 .next = md_seq_next,
3392 .stop = md_seq_stop,
3393 .show = md_seq_show,
3396 static int md_seq_open(struct inode *inode, struct file *file)
3400 error = seq_open(file, &md_seq_ops);
3404 static struct file_operations md_seq_fops = {
3405 .open = md_seq_open,
3407 .llseek = seq_lseek,
3408 .release = seq_release,
3411 int register_md_personality(int pnum, mdk_personality_t *p)
3413 if (pnum >= MAX_PERSONALITY) {
3415 "md: tried to install personality %s as nr %d, but max is %lu\n",
3416 p->name, pnum, MAX_PERSONALITY-1);
3420 spin_lock(&pers_lock);
3422 spin_unlock(&pers_lock);
3427 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3428 spin_unlock(&pers_lock);
3432 int unregister_md_personality(int pnum)
3434 if (pnum >= MAX_PERSONALITY)
3437 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3438 spin_lock(&pers_lock);
3440 spin_unlock(&pers_lock);
3444 static int is_mddev_idle(mddev_t *mddev)
3447 struct list_head *tmp;
3449 unsigned long curr_events;
3452 ITERATE_RDEV(mddev,rdev,tmp) {
3453 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3454 curr_events = disk_stat_read(disk, read_sectors) +
3455 disk_stat_read(disk, write_sectors) -
3456 atomic_read(&disk->sync_io);
3457 /* Allow some slack between valud of curr_events and last_events,
3458 * as there are some uninteresting races.
3459 * Note: the following is an unsigned comparison.
3461 if ((curr_events - rdev->last_events + 32) > 64) {
3462 rdev->last_events = curr_events;
3469 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3471 /* another "blocks" (512byte) blocks have been synced */
3472 atomic_sub(blocks, &mddev->recovery_active);
3473 wake_up(&mddev->recovery_wait);
3475 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3476 md_wakeup_thread(mddev->thread);
3477 // stop recovery, signal do_sync ....
3482 /* md_write_start(mddev, bi)
3483 * If we need to update some array metadata (e.g. 'active' flag
3484 * in superblock) before writing, schedule a superblock update
3485 * and wait for it to complete.
3487 void md_write_start(mddev_t *mddev, struct bio *bi)
3490 if (bio_data_dir(bi) != WRITE)
3493 atomic_inc(&mddev->writes_pending);
3494 if (mddev->in_sync) {
3495 spin_lock(&mddev->write_lock);
3496 if (mddev->in_sync) {
3498 mddev->sb_dirty = 1;
3499 md_wakeup_thread(mddev->thread);
3501 spin_unlock(&mddev->write_lock);
3503 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3506 void md_write_end(mddev_t *mddev)
3508 if (atomic_dec_and_test(&mddev->writes_pending)) {
3509 if (mddev->safemode == 2)
3510 md_wakeup_thread(mddev->thread);
3512 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3516 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3518 #define SYNC_MARKS 10
3519 #define SYNC_MARK_STEP (3*HZ)
3520 static void md_do_sync(mddev_t *mddev)
3523 unsigned int currspeed = 0,
3525 sector_t max_sectors,j, io_sectors;
3526 unsigned long mark[SYNC_MARKS];
3527 sector_t mark_cnt[SYNC_MARKS];
3529 struct list_head *tmp;
3530 sector_t last_check;
3533 /* just incase thread restarts... */
3534 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3537 /* we overload curr_resync somewhat here.
3538 * 0 == not engaged in resync at all
3539 * 2 == checking that there is no conflict with another sync
3540 * 1 == like 2, but have yielded to allow conflicting resync to
3542 * other == active in resync - this many blocks
3544 * Before starting a resync we must have set curr_resync to
3545 * 2, and then checked that every "conflicting" array has curr_resync
3546 * less than ours. When we find one that is the same or higher
3547 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3548 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3549 * This will mean we have to start checking from the beginning again.
3554 mddev->curr_resync = 2;
3557 if (signal_pending(current)) {
3558 flush_signals(current);
3561 ITERATE_MDDEV(mddev2,tmp) {
3562 if (mddev2 == mddev)
3564 if (mddev2->curr_resync &&
3565 match_mddev_units(mddev,mddev2)) {
3567 if (mddev < mddev2 && mddev->curr_resync == 2) {
3568 /* arbitrarily yield */
3569 mddev->curr_resync = 1;
3570 wake_up(&resync_wait);
3572 if (mddev > mddev2 && mddev->curr_resync == 1)
3573 /* no need to wait here, we can wait the next
3574 * time 'round when curr_resync == 2
3577 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3578 if (!signal_pending(current)
3579 && mddev2->curr_resync >= mddev->curr_resync) {
3580 printk(KERN_INFO "md: delaying resync of %s"
3581 " until %s has finished resync (they"
3582 " share one or more physical units)\n",
3583 mdname(mddev), mdname(mddev2));
3586 finish_wait(&resync_wait, &wq);
3589 finish_wait(&resync_wait, &wq);
3592 } while (mddev->curr_resync < 2);
3594 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3595 /* resync follows the size requested by the personality,
3596 * which defaults to physical size, but can be virtual size
3598 max_sectors = mddev->resync_max_sectors;
3600 /* recovery follows the physical size of devices */
3601 max_sectors = mddev->size << 1;
3603 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3604 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3605 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3606 printk(KERN_INFO "md: using maximum available idle IO bandwith "
3607 "(but not more than %d KB/sec) for reconstruction.\n",
3608 sysctl_speed_limit_max);
3610 is_mddev_idle(mddev); /* this also initializes IO event counters */
3611 /* we don't use the checkpoint if there's a bitmap */
3612 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3613 j = mddev->recovery_cp;
3617 for (m = 0; m < SYNC_MARKS; m++) {
3619 mark_cnt[m] = io_sectors;
3622 mddev->resync_mark = mark[last_mark];
3623 mddev->resync_mark_cnt = mark_cnt[last_mark];
3626 * Tune reconstruction:
3628 window = 32*(PAGE_SIZE/512);
3629 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3630 window/2,(unsigned long long) max_sectors/2);
3632 atomic_set(&mddev->recovery_active, 0);
3633 init_waitqueue_head(&mddev->recovery_wait);
3638 "md: resuming recovery of %s from checkpoint.\n",
3640 mddev->curr_resync = j;
3643 while (j < max_sectors) {
3647 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3648 currspeed < sysctl_speed_limit_min);
3650 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3654 if (!skipped) { /* actual IO requested */
3655 io_sectors += sectors;
3656 atomic_add(sectors, &mddev->recovery_active);
3660 if (j>1) mddev->curr_resync = j;
3663 if (last_check + window > io_sectors || j == max_sectors)
3666 last_check = io_sectors;
3668 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3669 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3673 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3675 int next = (last_mark+1) % SYNC_MARKS;
3677 mddev->resync_mark = mark[next];
3678 mddev->resync_mark_cnt = mark_cnt[next];
3679 mark[next] = jiffies;
3680 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3685 if (signal_pending(current)) {
3687 * got a signal, exit.
3690 "md: md_do_sync() got signal ... exiting\n");
3691 flush_signals(current);
3692 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3697 * this loop exits only if either when we are slower than
3698 * the 'hard' speed limit, or the system was IO-idle for
3700 * the system might be non-idle CPU-wise, but we only care
3701 * about not overloading the IO subsystem. (things like an
3702 * e2fsck being done on the RAID array should execute fast)
3704 mddev->queue->unplug_fn(mddev->queue);
3707 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3708 /((jiffies-mddev->resync_mark)/HZ +1) +1;
3710 if (currspeed > sysctl_speed_limit_min) {
3711 if ((currspeed > sysctl_speed_limit_max) ||
3712 !is_mddev_idle(mddev)) {
3713 msleep_interruptible(250);
3718 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3720 * this also signals 'finished resyncing' to md_stop
3723 mddev->queue->unplug_fn(mddev->queue);
3725 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3727 /* tell personality that we are finished */
3728 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3730 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3731 mddev->curr_resync > 2 &&
3732 mddev->curr_resync >= mddev->recovery_cp) {
3733 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3735 "md: checkpointing recovery of %s.\n",
3737 mddev->recovery_cp = mddev->curr_resync;
3739 mddev->recovery_cp = MaxSector;
3743 mddev->curr_resync = 0;
3744 wake_up(&resync_wait);
3745 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3746 md_wakeup_thread(mddev->thread);
3751 * This routine is regularly called by all per-raid-array threads to
3752 * deal with generic issues like resync and super-block update.
3753 * Raid personalities that don't have a thread (linear/raid0) do not
3754 * need this as they never do any recovery or update the superblock.
3756 * It does not do any resync itself, but rather "forks" off other threads
3757 * to do that as needed.
3758 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3759 * "->recovery" and create a thread at ->sync_thread.
3760 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3761 * and wakeups up this thread which will reap the thread and finish up.
3762 * This thread also removes any faulty devices (with nr_pending == 0).
3764 * The overall approach is:
3765 * 1/ if the superblock needs updating, update it.
3766 * 2/ If a recovery thread is running, don't do anything else.
3767 * 3/ If recovery has finished, clean up, possibly marking spares active.
3768 * 4/ If there are any faulty devices, remove them.
3769 * 5/ If array is degraded, try to add spares devices
3770 * 6/ If array has spares or is not in-sync, start a resync thread.
3772 void md_check_recovery(mddev_t *mddev)
3775 struct list_head *rtmp;
3779 bitmap_daemon_work(mddev->bitmap);
3784 if (signal_pending(current)) {
3785 if (mddev->pers->sync_request) {
3786 printk(KERN_INFO "md: %s in immediate safe mode\n",
3788 mddev->safemode = 2;
3790 flush_signals(current);
3795 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3796 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3797 (mddev->safemode == 1) ||
3798 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3799 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3803 if (mddev_trylock(mddev)==0) {
3806 spin_lock(&mddev->write_lock);
3807 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3808 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3810 mddev->sb_dirty = 1;
3812 if (mddev->safemode == 1)
3813 mddev->safemode = 0;
3814 spin_unlock(&mddev->write_lock);
3816 if (mddev->sb_dirty)
3817 md_update_sb(mddev);
3820 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3821 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3822 /* resync/recovery still happening */
3823 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3826 if (mddev->sync_thread) {
3827 /* resync has finished, collect result */
3828 md_unregister_thread(mddev->sync_thread);
3829 mddev->sync_thread = NULL;
3830 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3831 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3833 /* activate any spares */
3834 mddev->pers->spare_active(mddev);
3836 md_update_sb(mddev);
3838 /* if array is no-longer degraded, then any saved_raid_disk
3839 * information must be scrapped
3841 if (!mddev->degraded)
3842 ITERATE_RDEV(mddev,rdev,rtmp)
3843 rdev->saved_raid_disk = -1;
3845 mddev->recovery = 0;
3846 /* flag recovery needed just to double check */
3847 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3850 if (mddev->recovery)
3851 /* probably just the RECOVERY_NEEDED flag */
3852 mddev->recovery = 0;
3854 /* no recovery is running.
3855 * remove any failed drives, then
3856 * add spares if possible.
3857 * Spare are also removed and re-added, to allow
3858 * the personality to fail the re-add.
3860 ITERATE_RDEV(mddev,rdev,rtmp)
3861 if (rdev->raid_disk >= 0 &&
3862 (rdev->faulty || ! rdev->in_sync) &&
3863 atomic_read(&rdev->nr_pending)==0) {
3864 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3865 rdev->raid_disk = -1;
3868 if (mddev->degraded) {
3869 ITERATE_RDEV(mddev,rdev,rtmp)
3870 if (rdev->raid_disk < 0
3872 if (mddev->pers->hot_add_disk(mddev,rdev))
3879 if (!spares && (mddev->recovery_cp == MaxSector )) {
3880 /* nothing we can do ... */
3883 if (mddev->pers->sync_request) {
3884 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3886 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3887 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3888 /* We are adding a device or devices to an array
3889 * which has the bitmap stored on all devices.
3890 * So make sure all bitmap pages get written
3892 bitmap_write_all(mddev->bitmap);
3894 mddev->sync_thread = md_register_thread(md_do_sync,
3897 if (!mddev->sync_thread) {
3898 printk(KERN_ERR "%s: could not start resync"
3901 /* leave the spares where they are, it shouldn't hurt */
3902 mddev->recovery = 0;
3904 md_wakeup_thread(mddev->sync_thread);
3908 mddev_unlock(mddev);
3912 static int md_notify_reboot(struct notifier_block *this,
3913 unsigned long code, void *x)
3915 struct list_head *tmp;
3918 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3920 printk(KERN_INFO "md: stopping all md devices.\n");
3922 ITERATE_MDDEV(mddev,tmp)
3923 if (mddev_trylock(mddev)==0)
3924 do_md_stop (mddev, 1);
3926 * certain more exotic SCSI devices are known to be
3927 * volatile wrt too early system reboots. While the
3928 * right place to handle this issue is the given
3929 * driver, we do want to have a safe RAID driver ...
3936 static struct notifier_block md_notifier = {
3937 .notifier_call = md_notify_reboot,
3939 .priority = INT_MAX, /* before any real devices */
3942 static void md_geninit(void)
3944 struct proc_dir_entry *p;
3946 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3948 p = create_proc_entry("mdstat", S_IRUGO, NULL);
3950 p->proc_fops = &md_seq_fops;
3953 static int __init md_init(void)
3957 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3958 " MD_SB_DISKS=%d\n",
3959 MD_MAJOR_VERSION, MD_MINOR_VERSION,
3960 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3961 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3964 if (register_blkdev(MAJOR_NR, "md"))
3966 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3967 unregister_blkdev(MAJOR_NR, "md");
3971 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3972 md_probe, NULL, NULL);
3973 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3974 md_probe, NULL, NULL);
3976 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3977 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3978 S_IFBLK|S_IRUSR|S_IWUSR,
3981 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3982 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3983 S_IFBLK|S_IRUSR|S_IWUSR,
3987 register_reboot_notifier(&md_notifier);
3988 raid_table_header = register_sysctl_table(raid_root_table, 1);
3998 * Searches all registered partitions for autorun RAID arrays
4001 static dev_t detected_devices[128];
4004 void md_autodetect_dev(dev_t dev)
4006 if (dev_cnt >= 0 && dev_cnt < 127)
4007 detected_devices[dev_cnt++] = dev;
4011 static void autostart_arrays(int part)
4016 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4018 for (i = 0; i < dev_cnt; i++) {
4019 dev_t dev = detected_devices[i];
4021 rdev = md_import_device(dev,0, 0);
4029 list_add(&rdev->same_set, &pending_raid_disks);
4033 autorun_devices(part);
4038 static __exit void md_exit(void)
4041 struct list_head *tmp;
4043 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4044 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4045 for (i=0; i < MAX_MD_DEVS; i++)
4046 devfs_remove("md/%d", i);
4047 for (i=0; i < MAX_MD_DEVS; i++)
4048 devfs_remove("md/d%d", i);
4052 unregister_blkdev(MAJOR_NR,"md");
4053 unregister_blkdev(mdp_major, "mdp");
4054 unregister_reboot_notifier(&md_notifier);
4055 unregister_sysctl_table(raid_table_header);
4056 remove_proc_entry("mdstat", NULL);
4057 ITERATE_MDDEV(mddev,tmp) {
4058 struct gendisk *disk = mddev->gendisk;
4061 export_array(mddev);
4064 mddev->gendisk = NULL;
4069 module_init(md_init)
4070 module_exit(md_exit)
4072 EXPORT_SYMBOL(register_md_personality);
4073 EXPORT_SYMBOL(unregister_md_personality);
4074 EXPORT_SYMBOL(md_error);
4075 EXPORT_SYMBOL(md_done_sync);
4076 EXPORT_SYMBOL(md_write_start);
4077 EXPORT_SYMBOL(md_write_end);
4078 EXPORT_SYMBOL(md_register_thread);
4079 EXPORT_SYMBOL(md_unregister_thread);
4080 EXPORT_SYMBOL(md_wakeup_thread);
4081 EXPORT_SYMBOL(md_print_devices);
4082 EXPORT_SYMBOL(md_check_recovery);
4083 MODULE_LICENSE("GPL");
4085 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);