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) {
674 desc = sb->disks + rdev->desc_nr;
676 if (desc->state & (1<<MD_DISK_FAULTY))
678 else if (desc->state & (1<<MD_DISK_SYNC) &&
679 desc->raid_disk < mddev->raid_disks) {
681 rdev->raid_disk = desc->raid_disk;
683 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
684 set_bit(WriteMostly, &rdev->flags);
685 } else /* MULTIPATH are always insync */
691 * sync_super for 0.90.0
693 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
696 struct list_head *tmp;
698 int next_spare = mddev->raid_disks;
700 /* make rdev->sb match mddev data..
703 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
704 * 3/ any empty disks < next_spare become removed
706 * disks[0] gets initialised to REMOVED because
707 * we cannot be sure from other fields if it has
708 * been initialised or not.
711 int active=0, working=0,failed=0,spare=0,nr_disks=0;
713 sb = (mdp_super_t*)page_address(rdev->sb_page);
715 memset(sb, 0, sizeof(*sb));
717 sb->md_magic = MD_SB_MAGIC;
718 sb->major_version = mddev->major_version;
719 sb->minor_version = mddev->minor_version;
720 sb->patch_version = mddev->patch_version;
721 sb->gvalid_words = 0; /* ignored */
722 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
723 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
724 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
725 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
727 sb->ctime = mddev->ctime;
728 sb->level = mddev->level;
729 sb->size = mddev->size;
730 sb->raid_disks = mddev->raid_disks;
731 sb->md_minor = mddev->md_minor;
732 sb->not_persistent = !mddev->persistent;
733 sb->utime = mddev->utime;
735 sb->events_hi = (mddev->events>>32);
736 sb->events_lo = (u32)mddev->events;
740 sb->recovery_cp = mddev->recovery_cp;
741 sb->cp_events_hi = (mddev->events>>32);
742 sb->cp_events_lo = (u32)mddev->events;
743 if (mddev->recovery_cp == MaxSector)
744 sb->state = (1<< MD_SB_CLEAN);
748 sb->layout = mddev->layout;
749 sb->chunk_size = mddev->chunk_size;
751 if (mddev->bitmap && mddev->bitmap_file == NULL)
752 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
754 sb->disks[0].state = (1<<MD_DISK_REMOVED);
755 ITERATE_RDEV(mddev,rdev2,tmp) {
757 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
758 rdev2->desc_nr = rdev2->raid_disk;
760 rdev2->desc_nr = next_spare++;
761 d = &sb->disks[rdev2->desc_nr];
763 d->number = rdev2->desc_nr;
764 d->major = MAJOR(rdev2->bdev->bd_dev);
765 d->minor = MINOR(rdev2->bdev->bd_dev);
766 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
767 d->raid_disk = rdev2->raid_disk;
769 d->raid_disk = rdev2->desc_nr; /* compatibility */
771 d->state = (1<<MD_DISK_FAULTY);
773 } else if (rdev2->in_sync) {
774 d->state = (1<<MD_DISK_ACTIVE);
775 d->state |= (1<<MD_DISK_SYNC);
783 if (test_bit(WriteMostly, &rdev2->flags))
784 d->state |= (1<<MD_DISK_WRITEMOSTLY);
787 /* now set the "removed" and "faulty" bits on any missing devices */
788 for (i=0 ; i < mddev->raid_disks ; i++) {
789 mdp_disk_t *d = &sb->disks[i];
790 if (d->state == 0 && d->number == 0) {
793 d->state = (1<<MD_DISK_REMOVED);
794 d->state |= (1<<MD_DISK_FAULTY);
798 sb->nr_disks = nr_disks;
799 sb->active_disks = active;
800 sb->working_disks = working;
801 sb->failed_disks = failed;
802 sb->spare_disks = spare;
804 sb->this_disk = sb->disks[rdev->desc_nr];
805 sb->sb_csum = calc_sb_csum(sb);
809 * version 1 superblock
812 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
814 unsigned int disk_csum, csum;
815 unsigned long long newcsum;
816 int size = 256 + le32_to_cpu(sb->max_dev)*2;
817 unsigned int *isuper = (unsigned int*)sb;
820 disk_csum = sb->sb_csum;
823 for (i=0; size>=4; size -= 4 )
824 newcsum += le32_to_cpu(*isuper++);
827 newcsum += le16_to_cpu(*(unsigned short*) isuper);
829 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
830 sb->sb_csum = disk_csum;
831 return cpu_to_le32(csum);
834 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
836 struct mdp_superblock_1 *sb;
839 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
842 * Calculate the position of the superblock.
843 * It is always aligned to a 4K boundary and
844 * depeding on minor_version, it can be:
845 * 0: At least 8K, but less than 12K, from end of device
846 * 1: At start of device
847 * 2: 4K from start of device.
849 switch(minor_version) {
851 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
853 sb_offset &= ~(sector_t)(4*2-1);
854 /* convert from sectors to K */
866 rdev->sb_offset = sb_offset;
868 ret = read_disk_sb(rdev);
872 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
874 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
875 sb->major_version != cpu_to_le32(1) ||
876 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
877 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
878 sb->feature_map != 0)
881 if (calc_sb_1_csum(sb) != sb->sb_csum) {
882 printk("md: invalid superblock checksum on %s\n",
883 bdevname(rdev->bdev,b));
886 if (le64_to_cpu(sb->data_size) < 10) {
887 printk("md: data_size too small on %s\n",
888 bdevname(rdev->bdev,b));
891 rdev->preferred_minor = 0xffff;
892 rdev->data_offset = le64_to_cpu(sb->data_offset);
898 struct mdp_superblock_1 *refsb =
899 (struct mdp_superblock_1*)page_address(refdev->sb_page);
901 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
902 sb->level != refsb->level ||
903 sb->layout != refsb->layout ||
904 sb->chunksize != refsb->chunksize) {
905 printk(KERN_WARNING "md: %s has strangely different"
906 " superblock to %s\n",
907 bdevname(rdev->bdev,b),
908 bdevname(refdev->bdev,b2));
911 ev1 = le64_to_cpu(sb->events);
912 ev2 = le64_to_cpu(refsb->events);
918 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
920 rdev->size = rdev->sb_offset;
921 if (rdev->size < le64_to_cpu(sb->data_size)/2)
923 rdev->size = le64_to_cpu(sb->data_size)/2;
924 if (le32_to_cpu(sb->chunksize))
925 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
929 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
931 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
933 rdev->raid_disk = -1;
935 if (mddev->raid_disks == 0) {
936 mddev->major_version = 1;
937 mddev->patch_version = 0;
938 mddev->persistent = 1;
939 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
940 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
941 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
942 mddev->level = le32_to_cpu(sb->level);
943 mddev->layout = le32_to_cpu(sb->layout);
944 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
945 mddev->size = le64_to_cpu(sb->size)/2;
946 mddev->events = le64_to_cpu(sb->events);
947 mddev->bitmap_offset = 0;
948 mddev->default_bitmap_offset = 0;
949 if (mddev->minor_version == 0)
950 mddev->default_bitmap_offset = -(64*1024)/512;
952 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
953 memcpy(mddev->uuid, sb->set_uuid, 16);
955 mddev->max_disks = (4096-256)/2;
957 if ((le32_to_cpu(sb->feature_map) & 1) &&
958 mddev->bitmap_file == NULL ) {
959 if (mddev->level != 1) {
960 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
963 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
965 } else if (mddev->pers == NULL) {
966 /* Insist of good event counter while assembling */
967 __u64 ev1 = le64_to_cpu(sb->events);
969 if (ev1 < mddev->events)
971 } else if (mddev->bitmap) {
972 /* If adding to array with a bitmap, then we can accept an
973 * older device, but not too old.
975 __u64 ev1 = le64_to_cpu(sb->events);
976 if (ev1 < mddev->bitmap->events_cleared)
978 } else /* just a hot-add of a new device, leave raid_disk at -1 */
981 if (mddev->level != LEVEL_MULTIPATH) {
983 rdev->desc_nr = le32_to_cpu(sb->dev_number);
984 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
986 case 0xffff: /* spare */
989 case 0xfffe: /* faulty */
995 rdev->raid_disk = role;
999 if (sb->devflags & WriteMostly1)
1000 set_bit(WriteMostly, &rdev->flags);
1001 } else /* MULTIPATH are always insync */
1007 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1009 struct mdp_superblock_1 *sb;
1010 struct list_head *tmp;
1013 /* make rdev->sb match mddev and rdev data. */
1015 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1017 sb->feature_map = 0;
1019 memset(sb->pad1, 0, sizeof(sb->pad1));
1020 memset(sb->pad2, 0, sizeof(sb->pad2));
1021 memset(sb->pad3, 0, sizeof(sb->pad3));
1023 sb->utime = cpu_to_le64((__u64)mddev->utime);
1024 sb->events = cpu_to_le64(mddev->events);
1026 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1028 sb->resync_offset = cpu_to_le64(0);
1030 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1031 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1032 sb->feature_map = cpu_to_le32(1);
1036 ITERATE_RDEV(mddev,rdev2,tmp)
1037 if (rdev2->desc_nr+1 > max_dev)
1038 max_dev = rdev2->desc_nr+1;
1040 sb->max_dev = cpu_to_le32(max_dev);
1041 for (i=0; i<max_dev;i++)
1042 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1044 ITERATE_RDEV(mddev,rdev2,tmp) {
1047 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1048 else if (rdev2->in_sync)
1049 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1051 sb->dev_roles[i] = cpu_to_le16(0xffff);
1054 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1055 sb->sb_csum = calc_sb_1_csum(sb);
1059 static struct super_type super_types[] = {
1062 .owner = THIS_MODULE,
1063 .load_super = super_90_load,
1064 .validate_super = super_90_validate,
1065 .sync_super = super_90_sync,
1069 .owner = THIS_MODULE,
1070 .load_super = super_1_load,
1071 .validate_super = super_1_validate,
1072 .sync_super = super_1_sync,
1076 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1078 struct list_head *tmp;
1081 ITERATE_RDEV(mddev,rdev,tmp)
1082 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1088 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1090 struct list_head *tmp;
1093 ITERATE_RDEV(mddev1,rdev,tmp)
1094 if (match_dev_unit(mddev2, rdev))
1100 static LIST_HEAD(pending_raid_disks);
1102 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1104 mdk_rdev_t *same_pdev;
1105 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1111 same_pdev = match_dev_unit(mddev, rdev);
1114 "%s: WARNING: %s appears to be on the same physical"
1115 " disk as %s. True\n protection against single-disk"
1116 " failure might be compromised.\n",
1117 mdname(mddev), bdevname(rdev->bdev,b),
1118 bdevname(same_pdev->bdev,b2));
1120 /* Verify rdev->desc_nr is unique.
1121 * If it is -1, assign a free number, else
1122 * check number is not in use
1124 if (rdev->desc_nr < 0) {
1126 if (mddev->pers) choice = mddev->raid_disks;
1127 while (find_rdev_nr(mddev, choice))
1129 rdev->desc_nr = choice;
1131 if (find_rdev_nr(mddev, rdev->desc_nr))
1135 list_add(&rdev->same_set, &mddev->disks);
1136 rdev->mddev = mddev;
1137 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1141 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1143 char b[BDEVNAME_SIZE];
1148 list_del_init(&rdev->same_set);
1149 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1154 * prevent the device from being mounted, repartitioned or
1155 * otherwise reused by a RAID array (or any other kernel
1156 * subsystem), by bd_claiming the device.
1158 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1161 struct block_device *bdev;
1162 char b[BDEVNAME_SIZE];
1164 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1166 printk(KERN_ERR "md: could not open %s.\n",
1167 __bdevname(dev, b));
1168 return PTR_ERR(bdev);
1170 err = bd_claim(bdev, rdev);
1172 printk(KERN_ERR "md: could not bd_claim %s.\n",
1181 static void unlock_rdev(mdk_rdev_t *rdev)
1183 struct block_device *bdev = rdev->bdev;
1191 void md_autodetect_dev(dev_t dev);
1193 static void export_rdev(mdk_rdev_t * rdev)
1195 char b[BDEVNAME_SIZE];
1196 printk(KERN_INFO "md: export_rdev(%s)\n",
1197 bdevname(rdev->bdev,b));
1201 list_del_init(&rdev->same_set);
1203 md_autodetect_dev(rdev->bdev->bd_dev);
1209 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1211 unbind_rdev_from_array(rdev);
1215 static void export_array(mddev_t *mddev)
1217 struct list_head *tmp;
1220 ITERATE_RDEV(mddev,rdev,tmp) {
1225 kick_rdev_from_array(rdev);
1227 if (!list_empty(&mddev->disks))
1229 mddev->raid_disks = 0;
1230 mddev->major_version = 0;
1233 static void print_desc(mdp_disk_t *desc)
1235 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1236 desc->major,desc->minor,desc->raid_disk,desc->state);
1239 static void print_sb(mdp_super_t *sb)
1244 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1245 sb->major_version, sb->minor_version, sb->patch_version,
1246 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1248 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1249 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1250 sb->md_minor, sb->layout, sb->chunk_size);
1251 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1252 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1253 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1254 sb->failed_disks, sb->spare_disks,
1255 sb->sb_csum, (unsigned long)sb->events_lo);
1258 for (i = 0; i < MD_SB_DISKS; i++) {
1261 desc = sb->disks + i;
1262 if (desc->number || desc->major || desc->minor ||
1263 desc->raid_disk || (desc->state && (desc->state != 4))) {
1264 printk(" D %2d: ", i);
1268 printk(KERN_INFO "md: THIS: ");
1269 print_desc(&sb->this_disk);
1273 static void print_rdev(mdk_rdev_t *rdev)
1275 char b[BDEVNAME_SIZE];
1276 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1277 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1278 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1279 if (rdev->sb_loaded) {
1280 printk(KERN_INFO "md: rdev superblock:\n");
1281 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1283 printk(KERN_INFO "md: no rdev superblock!\n");
1286 void md_print_devices(void)
1288 struct list_head *tmp, *tmp2;
1291 char b[BDEVNAME_SIZE];
1294 printk("md: **********************************\n");
1295 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1296 printk("md: **********************************\n");
1297 ITERATE_MDDEV(mddev,tmp) {
1300 bitmap_print_sb(mddev->bitmap);
1302 printk("%s: ", mdname(mddev));
1303 ITERATE_RDEV(mddev,rdev,tmp2)
1304 printk("<%s>", bdevname(rdev->bdev,b));
1307 ITERATE_RDEV(mddev,rdev,tmp2)
1310 printk("md: **********************************\n");
1315 static void sync_sbs(mddev_t * mddev)
1318 struct list_head *tmp;
1320 ITERATE_RDEV(mddev,rdev,tmp) {
1321 super_types[mddev->major_version].
1322 sync_super(mddev, rdev);
1323 rdev->sb_loaded = 1;
1327 static void md_update_sb(mddev_t * mddev)
1330 struct list_head *tmp;
1335 spin_lock(&mddev->write_lock);
1336 sync_req = mddev->in_sync;
1337 mddev->utime = get_seconds();
1340 if (!mddev->events) {
1342 * oops, this 64-bit counter should never wrap.
1343 * Either we are in around ~1 trillion A.C., assuming
1344 * 1 reboot per second, or we have a bug:
1349 mddev->sb_dirty = 2;
1353 * do not write anything to disk if using
1354 * nonpersistent superblocks
1356 if (!mddev->persistent) {
1357 mddev->sb_dirty = 0;
1358 spin_unlock(&mddev->write_lock);
1359 wake_up(&mddev->sb_wait);
1362 spin_unlock(&mddev->write_lock);
1365 "md: updating %s RAID superblock on device (in sync %d)\n",
1366 mdname(mddev),mddev->in_sync);
1368 err = bitmap_update_sb(mddev->bitmap);
1369 ITERATE_RDEV(mddev,rdev,tmp) {
1370 char b[BDEVNAME_SIZE];
1371 dprintk(KERN_INFO "md: ");
1373 dprintk("(skipping faulty ");
1375 dprintk("%s ", bdevname(rdev->bdev,b));
1376 if (!rdev->faulty) {
1377 md_super_write(mddev,rdev,
1378 rdev->sb_offset<<1, MD_SB_BYTES,
1380 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1381 bdevname(rdev->bdev,b),
1382 (unsigned long long)rdev->sb_offset);
1386 if (mddev->level == LEVEL_MULTIPATH)
1387 /* only need to write one superblock... */
1390 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1391 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1393 spin_lock(&mddev->write_lock);
1394 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1395 /* have to write it out again */
1396 spin_unlock(&mddev->write_lock);
1399 mddev->sb_dirty = 0;
1400 spin_unlock(&mddev->write_lock);
1401 wake_up(&mddev->sb_wait);
1406 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1408 * mark the device faulty if:
1410 * - the device is nonexistent (zero size)
1411 * - the device has no valid superblock
1413 * a faulty rdev _never_ has rdev->sb set.
1415 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1417 char b[BDEVNAME_SIZE];
1422 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1424 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1425 return ERR_PTR(-ENOMEM);
1427 memset(rdev, 0, sizeof(*rdev));
1429 if ((err = alloc_disk_sb(rdev)))
1432 err = lock_rdev(rdev, newdev);
1439 rdev->data_offset = 0;
1440 atomic_set(&rdev->nr_pending, 0);
1442 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1445 "md: %s has zero or unknown size, marking faulty!\n",
1446 bdevname(rdev->bdev,b));
1451 if (super_format >= 0) {
1452 err = super_types[super_format].
1453 load_super(rdev, NULL, super_minor);
1454 if (err == -EINVAL) {
1456 "md: %s has invalid sb, not importing!\n",
1457 bdevname(rdev->bdev,b));
1462 "md: could not read %s's sb, not importing!\n",
1463 bdevname(rdev->bdev,b));
1467 INIT_LIST_HEAD(&rdev->same_set);
1472 if (rdev->sb_page) {
1478 return ERR_PTR(err);
1482 * Check a full RAID array for plausibility
1486 static void analyze_sbs(mddev_t * mddev)
1489 struct list_head *tmp;
1490 mdk_rdev_t *rdev, *freshest;
1491 char b[BDEVNAME_SIZE];
1494 ITERATE_RDEV(mddev,rdev,tmp)
1495 switch (super_types[mddev->major_version].
1496 load_super(rdev, freshest, mddev->minor_version)) {
1504 "md: fatal superblock inconsistency in %s"
1505 " -- removing from array\n",
1506 bdevname(rdev->bdev,b));
1507 kick_rdev_from_array(rdev);
1511 super_types[mddev->major_version].
1512 validate_super(mddev, freshest);
1515 ITERATE_RDEV(mddev,rdev,tmp) {
1516 if (rdev != freshest)
1517 if (super_types[mddev->major_version].
1518 validate_super(mddev, rdev)) {
1519 printk(KERN_WARNING "md: kicking non-fresh %s"
1521 bdevname(rdev->bdev,b));
1522 kick_rdev_from_array(rdev);
1525 if (mddev->level == LEVEL_MULTIPATH) {
1526 rdev->desc_nr = i++;
1527 rdev->raid_disk = rdev->desc_nr;
1534 if (mddev->recovery_cp != MaxSector &&
1536 printk(KERN_ERR "md: %s: raid array is not clean"
1537 " -- starting background reconstruction\n",
1544 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1546 static DECLARE_MUTEX(disks_sem);
1547 mddev_t *mddev = mddev_find(dev);
1548 struct gendisk *disk;
1549 int partitioned = (MAJOR(dev) != MD_MAJOR);
1550 int shift = partitioned ? MdpMinorShift : 0;
1551 int unit = MINOR(dev) >> shift;
1557 if (mddev->gendisk) {
1562 disk = alloc_disk(1 << shift);
1568 disk->major = MAJOR(dev);
1569 disk->first_minor = unit << shift;
1571 sprintf(disk->disk_name, "md_d%d", unit);
1572 sprintf(disk->devfs_name, "md/d%d", unit);
1574 sprintf(disk->disk_name, "md%d", unit);
1575 sprintf(disk->devfs_name, "md/%d", unit);
1577 disk->fops = &md_fops;
1578 disk->private_data = mddev;
1579 disk->queue = mddev->queue;
1581 mddev->gendisk = disk;
1586 void md_wakeup_thread(mdk_thread_t *thread);
1588 static void md_safemode_timeout(unsigned long data)
1590 mddev_t *mddev = (mddev_t *) data;
1592 mddev->safemode = 1;
1593 md_wakeup_thread(mddev->thread);
1597 static int do_md_run(mddev_t * mddev)
1601 struct list_head *tmp;
1603 struct gendisk *disk;
1604 char b[BDEVNAME_SIZE];
1606 if (list_empty(&mddev->disks))
1607 /* cannot run an array with no devices.. */
1614 * Analyze all RAID superblock(s)
1616 if (!mddev->raid_disks)
1619 chunk_size = mddev->chunk_size;
1620 pnum = level_to_pers(mddev->level);
1622 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1625 * 'default chunksize' in the old md code used to
1626 * be PAGE_SIZE, baaad.
1627 * we abort here to be on the safe side. We don't
1628 * want to continue the bad practice.
1631 "no chunksize specified, see 'man raidtab'\n");
1634 if (chunk_size > MAX_CHUNK_SIZE) {
1635 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1636 chunk_size, MAX_CHUNK_SIZE);
1640 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1642 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1643 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1646 if (chunk_size < PAGE_SIZE) {
1647 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1648 chunk_size, PAGE_SIZE);
1652 /* devices must have minimum size of one chunk */
1653 ITERATE_RDEV(mddev,rdev,tmp) {
1656 if (rdev->size < chunk_size / 1024) {
1658 "md: Dev %s smaller than chunk_size:"
1660 bdevname(rdev->bdev,b),
1661 (unsigned long long)rdev->size,
1671 request_module("md-personality-%d", pnum);
1676 * Drop all container device buffers, from now on
1677 * the only valid external interface is through the md
1679 * Also find largest hardsector size
1681 ITERATE_RDEV(mddev,rdev,tmp) {
1684 sync_blockdev(rdev->bdev);
1685 invalidate_bdev(rdev->bdev, 0);
1688 md_probe(mddev->unit, NULL, NULL);
1689 disk = mddev->gendisk;
1693 spin_lock(&pers_lock);
1694 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1695 spin_unlock(&pers_lock);
1696 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1701 mddev->pers = pers[pnum];
1702 spin_unlock(&pers_lock);
1704 mddev->recovery = 0;
1705 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1707 /* before we start the array running, initialise the bitmap */
1708 err = bitmap_create(mddev);
1710 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1711 mdname(mddev), err);
1713 err = mddev->pers->run(mddev);
1715 printk(KERN_ERR "md: pers->run() failed ...\n");
1716 module_put(mddev->pers->owner);
1718 bitmap_destroy(mddev);
1721 atomic_set(&mddev->writes_pending,0);
1722 mddev->safemode = 0;
1723 mddev->safemode_timer.function = md_safemode_timeout;
1724 mddev->safemode_timer.data = (unsigned long) mddev;
1725 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1728 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1729 md_wakeup_thread(mddev->thread);
1731 if (mddev->sb_dirty)
1732 md_update_sb(mddev);
1734 set_capacity(disk, mddev->array_size<<1);
1736 /* If we call blk_queue_make_request here, it will
1737 * re-initialise max_sectors etc which may have been
1738 * refined inside -> run. So just set the bits we need to set.
1739 * Most initialisation happended when we called
1740 * blk_queue_make_request(..., md_fail_request)
1743 mddev->queue->queuedata = mddev;
1744 mddev->queue->make_request_fn = mddev->pers->make_request;
1750 static int restart_array(mddev_t *mddev)
1752 struct gendisk *disk = mddev->gendisk;
1756 * Complain if it has no devices
1759 if (list_empty(&mddev->disks))
1767 mddev->safemode = 0;
1769 set_disk_ro(disk, 0);
1771 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1774 * Kick recovery or resync if necessary
1776 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1777 md_wakeup_thread(mddev->thread);
1780 printk(KERN_ERR "md: %s has no personality assigned.\n",
1789 static int do_md_stop(mddev_t * mddev, int ro)
1792 struct gendisk *disk = mddev->gendisk;
1795 if (atomic_read(&mddev->active)>2) {
1796 printk("md: %s still in use.\n",mdname(mddev));
1800 if (mddev->sync_thread) {
1801 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1802 md_unregister_thread(mddev->sync_thread);
1803 mddev->sync_thread = NULL;
1806 del_timer_sync(&mddev->safemode_timer);
1808 invalidate_partition(disk, 0);
1816 bitmap_flush(mddev);
1817 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1819 set_disk_ro(disk, 0);
1820 blk_queue_make_request(mddev->queue, md_fail_request);
1821 mddev->pers->stop(mddev);
1822 module_put(mddev->pers->owner);
1827 if (!mddev->in_sync) {
1828 /* mark array as shutdown cleanly */
1830 md_update_sb(mddev);
1833 set_disk_ro(disk, 1);
1836 bitmap_destroy(mddev);
1837 if (mddev->bitmap_file) {
1838 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1839 fput(mddev->bitmap_file);
1840 mddev->bitmap_file = NULL;
1842 mddev->bitmap_offset = 0;
1845 * Free resources if final stop
1848 struct gendisk *disk;
1849 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1851 export_array(mddev);
1853 mddev->array_size = 0;
1854 disk = mddev->gendisk;
1856 set_capacity(disk, 0);
1859 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1866 static void autorun_array(mddev_t *mddev)
1869 struct list_head *tmp;
1872 if (list_empty(&mddev->disks))
1875 printk(KERN_INFO "md: running: ");
1877 ITERATE_RDEV(mddev,rdev,tmp) {
1878 char b[BDEVNAME_SIZE];
1879 printk("<%s>", bdevname(rdev->bdev,b));
1883 err = do_md_run (mddev);
1885 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1886 do_md_stop (mddev, 0);
1891 * lets try to run arrays based on all disks that have arrived
1892 * until now. (those are in pending_raid_disks)
1894 * the method: pick the first pending disk, collect all disks with
1895 * the same UUID, remove all from the pending list and put them into
1896 * the 'same_array' list. Then order this list based on superblock
1897 * update time (freshest comes first), kick out 'old' disks and
1898 * compare superblocks. If everything's fine then run it.
1900 * If "unit" is allocated, then bump its reference count
1902 static void autorun_devices(int part)
1904 struct list_head candidates;
1905 struct list_head *tmp;
1906 mdk_rdev_t *rdev0, *rdev;
1908 char b[BDEVNAME_SIZE];
1910 printk(KERN_INFO "md: autorun ...\n");
1911 while (!list_empty(&pending_raid_disks)) {
1913 rdev0 = list_entry(pending_raid_disks.next,
1914 mdk_rdev_t, same_set);
1916 printk(KERN_INFO "md: considering %s ...\n",
1917 bdevname(rdev0->bdev,b));
1918 INIT_LIST_HEAD(&candidates);
1919 ITERATE_RDEV_PENDING(rdev,tmp)
1920 if (super_90_load(rdev, rdev0, 0) >= 0) {
1921 printk(KERN_INFO "md: adding %s ...\n",
1922 bdevname(rdev->bdev,b));
1923 list_move(&rdev->same_set, &candidates);
1926 * now we have a set of devices, with all of them having
1927 * mostly sane superblocks. It's time to allocate the
1930 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1931 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1932 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1936 dev = MKDEV(mdp_major,
1937 rdev0->preferred_minor << MdpMinorShift);
1939 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1941 md_probe(dev, NULL, NULL);
1942 mddev = mddev_find(dev);
1945 "md: cannot allocate memory for md drive.\n");
1948 if (mddev_lock(mddev))
1949 printk(KERN_WARNING "md: %s locked, cannot run\n",
1951 else if (mddev->raid_disks || mddev->major_version
1952 || !list_empty(&mddev->disks)) {
1954 "md: %s already running, cannot run %s\n",
1955 mdname(mddev), bdevname(rdev0->bdev,b));
1956 mddev_unlock(mddev);
1958 printk(KERN_INFO "md: created %s\n", mdname(mddev));
1959 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1960 list_del_init(&rdev->same_set);
1961 if (bind_rdev_to_array(rdev, mddev))
1964 autorun_array(mddev);
1965 mddev_unlock(mddev);
1967 /* on success, candidates will be empty, on error
1970 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1974 printk(KERN_INFO "md: ... autorun DONE.\n");
1978 * import RAID devices based on one partition
1979 * if possible, the array gets run as well.
1982 static int autostart_array(dev_t startdev)
1984 char b[BDEVNAME_SIZE];
1985 int err = -EINVAL, i;
1986 mdp_super_t *sb = NULL;
1987 mdk_rdev_t *start_rdev = NULL, *rdev;
1989 start_rdev = md_import_device(startdev, 0, 0);
1990 if (IS_ERR(start_rdev))
1994 /* NOTE: this can only work for 0.90.0 superblocks */
1995 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
1996 if (sb->major_version != 0 ||
1997 sb->minor_version != 90 ) {
1998 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
1999 export_rdev(start_rdev);
2003 if (start_rdev->faulty) {
2005 "md: can not autostart based on faulty %s!\n",
2006 bdevname(start_rdev->bdev,b));
2007 export_rdev(start_rdev);
2010 list_add(&start_rdev->same_set, &pending_raid_disks);
2012 for (i = 0; i < MD_SB_DISKS; i++) {
2013 mdp_disk_t *desc = sb->disks + i;
2014 dev_t dev = MKDEV(desc->major, desc->minor);
2018 if (dev == startdev)
2020 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2022 rdev = md_import_device(dev, 0, 0);
2026 list_add(&rdev->same_set, &pending_raid_disks);
2030 * possibly return codes
2038 static int get_version(void __user * arg)
2042 ver.major = MD_MAJOR_VERSION;
2043 ver.minor = MD_MINOR_VERSION;
2044 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2046 if (copy_to_user(arg, &ver, sizeof(ver)))
2052 static int get_array_info(mddev_t * mddev, void __user * arg)
2054 mdu_array_info_t info;
2055 int nr,working,active,failed,spare;
2057 struct list_head *tmp;
2059 nr=working=active=failed=spare=0;
2060 ITERATE_RDEV(mddev,rdev,tmp) {
2073 info.major_version = mddev->major_version;
2074 info.minor_version = mddev->minor_version;
2075 info.patch_version = MD_PATCHLEVEL_VERSION;
2076 info.ctime = mddev->ctime;
2077 info.level = mddev->level;
2078 info.size = mddev->size;
2080 info.raid_disks = mddev->raid_disks;
2081 info.md_minor = mddev->md_minor;
2082 info.not_persistent= !mddev->persistent;
2084 info.utime = mddev->utime;
2087 info.state = (1<<MD_SB_CLEAN);
2088 if (mddev->bitmap && mddev->bitmap_offset)
2089 info.state = (1<<MD_SB_BITMAP_PRESENT);
2090 info.active_disks = active;
2091 info.working_disks = working;
2092 info.failed_disks = failed;
2093 info.spare_disks = spare;
2095 info.layout = mddev->layout;
2096 info.chunk_size = mddev->chunk_size;
2098 if (copy_to_user(arg, &info, sizeof(info)))
2104 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2106 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2107 char *ptr, *buf = NULL;
2110 file = kmalloc(sizeof(*file), GFP_KERNEL);
2114 /* bitmap disabled, zero the first byte and copy out */
2115 if (!mddev->bitmap || !mddev->bitmap->file) {
2116 file->pathname[0] = '\0';
2120 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2124 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2128 strcpy(file->pathname, ptr);
2132 if (copy_to_user(arg, file, sizeof(*file)))
2140 static int get_disk_info(mddev_t * mddev, void __user * arg)
2142 mdu_disk_info_t info;
2146 if (copy_from_user(&info, arg, sizeof(info)))
2151 rdev = find_rdev_nr(mddev, nr);
2153 info.major = MAJOR(rdev->bdev->bd_dev);
2154 info.minor = MINOR(rdev->bdev->bd_dev);
2155 info.raid_disk = rdev->raid_disk;
2158 info.state |= (1<<MD_DISK_FAULTY);
2159 else if (rdev->in_sync) {
2160 info.state |= (1<<MD_DISK_ACTIVE);
2161 info.state |= (1<<MD_DISK_SYNC);
2163 if (test_bit(WriteMostly, &rdev->flags))
2164 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2166 info.major = info.minor = 0;
2167 info.raid_disk = -1;
2168 info.state = (1<<MD_DISK_REMOVED);
2171 if (copy_to_user(arg, &info, sizeof(info)))
2177 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2179 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2181 dev_t dev = MKDEV(info->major,info->minor);
2183 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2186 if (!mddev->raid_disks) {
2188 /* expecting a device which has a superblock */
2189 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2192 "md: md_import_device returned %ld\n",
2194 return PTR_ERR(rdev);
2196 if (!list_empty(&mddev->disks)) {
2197 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2198 mdk_rdev_t, same_set);
2199 int err = super_types[mddev->major_version]
2200 .load_super(rdev, rdev0, mddev->minor_version);
2203 "md: %s has different UUID to %s\n",
2204 bdevname(rdev->bdev,b),
2205 bdevname(rdev0->bdev,b2));
2210 err = bind_rdev_to_array(rdev, mddev);
2217 * add_new_disk can be used once the array is assembled
2218 * to add "hot spares". They must already have a superblock
2223 if (!mddev->pers->hot_add_disk) {
2225 "%s: personality does not support diskops!\n",
2229 rdev = md_import_device(dev, mddev->major_version,
2230 mddev->minor_version);
2233 "md: md_import_device returned %ld\n",
2235 return PTR_ERR(rdev);
2237 /* set save_raid_disk if appropriate */
2238 if (!mddev->persistent) {
2239 if (info->state & (1<<MD_DISK_SYNC) &&
2240 info->raid_disk < mddev->raid_disks)
2241 rdev->raid_disk = info->raid_disk;
2243 rdev->raid_disk = -1;
2245 super_types[mddev->major_version].
2246 validate_super(mddev, rdev);
2247 rdev->saved_raid_disk = rdev->raid_disk;
2249 rdev->in_sync = 0; /* just to be sure */
2250 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2251 set_bit(WriteMostly, &rdev->flags);
2253 rdev->raid_disk = -1;
2254 err = bind_rdev_to_array(rdev, mddev);
2258 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2259 md_wakeup_thread(mddev->thread);
2263 /* otherwise, add_new_disk is only allowed
2264 * for major_version==0 superblocks
2266 if (mddev->major_version != 0) {
2267 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2272 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2274 rdev = md_import_device (dev, -1, 0);
2277 "md: error, md_import_device() returned %ld\n",
2279 return PTR_ERR(rdev);
2281 rdev->desc_nr = info->number;
2282 if (info->raid_disk < mddev->raid_disks)
2283 rdev->raid_disk = info->raid_disk;
2285 rdev->raid_disk = -1;
2288 if (rdev->raid_disk < mddev->raid_disks)
2289 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2293 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2294 set_bit(WriteMostly, &rdev->flags);
2296 err = bind_rdev_to_array(rdev, mddev);
2302 if (!mddev->persistent) {
2303 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2304 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2306 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2307 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2309 if (!mddev->size || (mddev->size > rdev->size))
2310 mddev->size = rdev->size;
2316 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2318 char b[BDEVNAME_SIZE];
2324 rdev = find_rdev(mddev, dev);
2328 if (rdev->raid_disk >= 0)
2331 kick_rdev_from_array(rdev);
2332 md_update_sb(mddev);
2336 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2337 bdevname(rdev->bdev,b), mdname(mddev));
2341 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2343 char b[BDEVNAME_SIZE];
2351 if (mddev->major_version != 0) {
2352 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2353 " version-0 superblocks.\n",
2357 if (!mddev->pers->hot_add_disk) {
2359 "%s: personality does not support diskops!\n",
2364 rdev = md_import_device (dev, -1, 0);
2367 "md: error, md_import_device() returned %ld\n",
2372 if (mddev->persistent)
2373 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2376 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2378 size = calc_dev_size(rdev, mddev->chunk_size);
2381 if (size < mddev->size) {
2383 "%s: disk size %llu blocks < array size %llu\n",
2384 mdname(mddev), (unsigned long long)size,
2385 (unsigned long long)mddev->size);
2392 "md: can not hot-add faulty %s disk to %s!\n",
2393 bdevname(rdev->bdev,b), mdname(mddev));
2399 bind_rdev_to_array(rdev, mddev);
2402 * The rest should better be atomic, we can have disk failures
2403 * noticed in interrupt contexts ...
2406 if (rdev->desc_nr == mddev->max_disks) {
2407 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2410 goto abort_unbind_export;
2413 rdev->raid_disk = -1;
2415 md_update_sb(mddev);
2418 * Kick recovery, maybe this spare has to be added to the
2419 * array immediately.
2421 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2422 md_wakeup_thread(mddev->thread);
2426 abort_unbind_export:
2427 unbind_rdev_from_array(rdev);
2434 /* similar to deny_write_access, but accounts for our holding a reference
2435 * to the file ourselves */
2436 static int deny_bitmap_write_access(struct file * file)
2438 struct inode *inode = file->f_mapping->host;
2440 spin_lock(&inode->i_lock);
2441 if (atomic_read(&inode->i_writecount) > 1) {
2442 spin_unlock(&inode->i_lock);
2445 atomic_set(&inode->i_writecount, -1);
2446 spin_unlock(&inode->i_lock);
2451 static int set_bitmap_file(mddev_t *mddev, int fd)
2456 if (!mddev->pers->quiesce)
2458 if (mddev->recovery || mddev->sync_thread)
2460 /* we should be able to change the bitmap.. */
2466 return -EEXIST; /* cannot add when bitmap is present */
2467 mddev->bitmap_file = fget(fd);
2469 if (mddev->bitmap_file == NULL) {
2470 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2475 err = deny_bitmap_write_access(mddev->bitmap_file);
2477 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2479 fput(mddev->bitmap_file);
2480 mddev->bitmap_file = NULL;
2483 mddev->bitmap_offset = 0; /* file overrides offset */
2484 } else if (mddev->bitmap == NULL)
2485 return -ENOENT; /* cannot remove what isn't there */
2488 mddev->pers->quiesce(mddev, 1);
2490 err = bitmap_create(mddev);
2492 bitmap_destroy(mddev);
2493 mddev->pers->quiesce(mddev, 0);
2494 } else if (fd < 0) {
2495 if (mddev->bitmap_file)
2496 fput(mddev->bitmap_file);
2497 mddev->bitmap_file = NULL;
2504 * set_array_info is used two different ways
2505 * The original usage is when creating a new array.
2506 * In this usage, raid_disks is > 0 and it together with
2507 * level, size, not_persistent,layout,chunksize determine the
2508 * shape of the array.
2509 * This will always create an array with a type-0.90.0 superblock.
2510 * The newer usage is when assembling an array.
2511 * In this case raid_disks will be 0, and the major_version field is
2512 * use to determine which style super-blocks are to be found on the devices.
2513 * The minor and patch _version numbers are also kept incase the
2514 * super_block handler wishes to interpret them.
2516 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2519 if (info->raid_disks == 0) {
2520 /* just setting version number for superblock loading */
2521 if (info->major_version < 0 ||
2522 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2523 super_types[info->major_version].name == NULL) {
2524 /* maybe try to auto-load a module? */
2526 "md: superblock version %d not known\n",
2527 info->major_version);
2530 mddev->major_version = info->major_version;
2531 mddev->minor_version = info->minor_version;
2532 mddev->patch_version = info->patch_version;
2535 mddev->major_version = MD_MAJOR_VERSION;
2536 mddev->minor_version = MD_MINOR_VERSION;
2537 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2538 mddev->ctime = get_seconds();
2540 mddev->level = info->level;
2541 mddev->size = info->size;
2542 mddev->raid_disks = info->raid_disks;
2543 /* don't set md_minor, it is determined by which /dev/md* was
2546 if (info->state & (1<<MD_SB_CLEAN))
2547 mddev->recovery_cp = MaxSector;
2549 mddev->recovery_cp = 0;
2550 mddev->persistent = ! info->not_persistent;
2552 mddev->layout = info->layout;
2553 mddev->chunk_size = info->chunk_size;
2555 mddev->max_disks = MD_SB_DISKS;
2557 mddev->sb_dirty = 1;
2560 * Generate a 128 bit UUID
2562 get_random_bytes(mddev->uuid, 16);
2568 * update_array_info is used to change the configuration of an
2570 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2571 * fields in the info are checked against the array.
2572 * Any differences that cannot be handled will cause an error.
2573 * Normally, only one change can be managed at a time.
2575 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2581 /* calculate expected state,ignoring low bits */
2582 if (mddev->bitmap && mddev->bitmap_offset)
2583 state |= (1 << MD_SB_BITMAP_PRESENT);
2585 if (mddev->major_version != info->major_version ||
2586 mddev->minor_version != info->minor_version ||
2587 /* mddev->patch_version != info->patch_version || */
2588 mddev->ctime != info->ctime ||
2589 mddev->level != info->level ||
2590 /* mddev->layout != info->layout || */
2591 !mddev->persistent != info->not_persistent||
2592 mddev->chunk_size != info->chunk_size ||
2593 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2594 ((state^info->state) & 0xfffffe00)
2597 /* Check there is only one change */
2598 if (mddev->size != info->size) cnt++;
2599 if (mddev->raid_disks != info->raid_disks) cnt++;
2600 if (mddev->layout != info->layout) cnt++;
2601 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2602 if (cnt == 0) return 0;
2603 if (cnt > 1) return -EINVAL;
2605 if (mddev->layout != info->layout) {
2607 * we don't need to do anything at the md level, the
2608 * personality will take care of it all.
2610 if (mddev->pers->reconfig == NULL)
2613 return mddev->pers->reconfig(mddev, info->layout, -1);
2615 if (mddev->size != info->size) {
2617 struct list_head *tmp;
2618 if (mddev->pers->resize == NULL)
2620 /* The "size" is the amount of each device that is used.
2621 * This can only make sense for arrays with redundancy.
2622 * linear and raid0 always use whatever space is available
2623 * We can only consider changing the size if no resync
2624 * or reconstruction is happening, and if the new size
2625 * is acceptable. It must fit before the sb_offset or,
2626 * if that is <data_offset, it must fit before the
2627 * size of each device.
2628 * If size is zero, we find the largest size that fits.
2630 if (mddev->sync_thread)
2632 ITERATE_RDEV(mddev,rdev,tmp) {
2634 int fit = (info->size == 0);
2635 if (rdev->sb_offset > rdev->data_offset)
2636 avail = (rdev->sb_offset*2) - rdev->data_offset;
2638 avail = get_capacity(rdev->bdev->bd_disk)
2639 - rdev->data_offset;
2640 if (fit && (info->size == 0 || info->size > avail/2))
2641 info->size = avail/2;
2642 if (avail < ((sector_t)info->size << 1))
2645 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2647 struct block_device *bdev;
2649 bdev = bdget_disk(mddev->gendisk, 0);
2651 down(&bdev->bd_inode->i_sem);
2652 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2653 up(&bdev->bd_inode->i_sem);
2658 if (mddev->raid_disks != info->raid_disks) {
2659 /* change the number of raid disks */
2660 if (mddev->pers->reshape == NULL)
2662 if (info->raid_disks <= 0 ||
2663 info->raid_disks >= mddev->max_disks)
2665 if (mddev->sync_thread)
2667 rv = mddev->pers->reshape(mddev, info->raid_disks);
2669 struct block_device *bdev;
2671 bdev = bdget_disk(mddev->gendisk, 0);
2673 down(&bdev->bd_inode->i_sem);
2674 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2675 up(&bdev->bd_inode->i_sem);
2680 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2681 if (mddev->pers->quiesce == NULL)
2683 if (mddev->recovery || mddev->sync_thread)
2685 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2686 /* add the bitmap */
2689 if (mddev->default_bitmap_offset == 0)
2691 mddev->bitmap_offset = mddev->default_bitmap_offset;
2692 mddev->pers->quiesce(mddev, 1);
2693 rv = bitmap_create(mddev);
2695 bitmap_destroy(mddev);
2696 mddev->pers->quiesce(mddev, 0);
2698 /* remove the bitmap */
2701 if (mddev->bitmap->file)
2703 mddev->pers->quiesce(mddev, 1);
2704 bitmap_destroy(mddev);
2705 mddev->pers->quiesce(mddev, 0);
2706 mddev->bitmap_offset = 0;
2709 md_update_sb(mddev);
2713 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2717 if (mddev->pers == NULL)
2720 rdev = find_rdev(mddev, dev);
2724 md_error(mddev, rdev);
2728 static int md_ioctl(struct inode *inode, struct file *file,
2729 unsigned int cmd, unsigned long arg)
2732 void __user *argp = (void __user *)arg;
2733 struct hd_geometry __user *loc = argp;
2734 mddev_t *mddev = NULL;
2736 if (!capable(CAP_SYS_ADMIN))
2740 * Commands dealing with the RAID driver but not any
2746 err = get_version(argp);
2749 case PRINT_RAID_DEBUG:
2757 autostart_arrays(arg);
2764 * Commands creating/starting a new array:
2767 mddev = inode->i_bdev->bd_disk->private_data;
2775 if (cmd == START_ARRAY) {
2776 /* START_ARRAY doesn't need to lock the array as autostart_array
2777 * does the locking, and it could even be a different array
2782 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2783 "This will not be supported beyond 2.6\n",
2784 current->comm, current->pid);
2787 err = autostart_array(new_decode_dev(arg));
2789 printk(KERN_WARNING "md: autostart failed!\n");
2795 err = mddev_lock(mddev);
2798 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2805 case SET_ARRAY_INFO:
2807 mdu_array_info_t info;
2809 memset(&info, 0, sizeof(info));
2810 else if (copy_from_user(&info, argp, sizeof(info))) {
2815 err = update_array_info(mddev, &info);
2817 printk(KERN_WARNING "md: couldn't update"
2818 " array info. %d\n", err);
2823 if (!list_empty(&mddev->disks)) {
2825 "md: array %s already has disks!\n",
2830 if (mddev->raid_disks) {
2832 "md: array %s already initialised!\n",
2837 err = set_array_info(mddev, &info);
2839 printk(KERN_WARNING "md: couldn't set"
2840 " array info. %d\n", err);
2850 * Commands querying/configuring an existing array:
2852 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2853 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2854 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2855 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2861 * Commands even a read-only array can execute:
2865 case GET_ARRAY_INFO:
2866 err = get_array_info(mddev, argp);
2869 case GET_BITMAP_FILE:
2870 err = get_bitmap_file(mddev, argp);
2874 err = get_disk_info(mddev, argp);
2877 case RESTART_ARRAY_RW:
2878 err = restart_array(mddev);
2882 err = do_md_stop (mddev, 0);
2886 err = do_md_stop (mddev, 1);
2890 * We have a problem here : there is no easy way to give a CHS
2891 * virtual geometry. We currently pretend that we have a 2 heads
2892 * 4 sectors (with a BIG number of cylinders...). This drives
2893 * dosfs just mad... ;-)
2900 err = put_user (2, (char __user *) &loc->heads);
2903 err = put_user (4, (char __user *) &loc->sectors);
2906 err = put_user(get_capacity(mddev->gendisk)/8,
2907 (short __user *) &loc->cylinders);
2910 err = put_user (get_start_sect(inode->i_bdev),
2911 (long __user *) &loc->start);
2916 * The remaining ioctls are changing the state of the
2917 * superblock, so we do not allow read-only arrays
2929 mdu_disk_info_t info;
2930 if (copy_from_user(&info, argp, sizeof(info)))
2933 err = add_new_disk(mddev, &info);
2937 case HOT_REMOVE_DISK:
2938 err = hot_remove_disk(mddev, new_decode_dev(arg));
2942 err = hot_add_disk(mddev, new_decode_dev(arg));
2945 case SET_DISK_FAULTY:
2946 err = set_disk_faulty(mddev, new_decode_dev(arg));
2950 err = do_md_run (mddev);
2953 case SET_BITMAP_FILE:
2954 err = set_bitmap_file(mddev, (int)arg);
2958 if (_IOC_TYPE(cmd) == MD_MAJOR)
2959 printk(KERN_WARNING "md: %s(pid %d) used"
2960 " obsolete MD ioctl, upgrade your"
2961 " software to use new ictls.\n",
2962 current->comm, current->pid);
2969 mddev_unlock(mddev);
2979 static int md_open(struct inode *inode, struct file *file)
2982 * Succeed if we can lock the mddev, which confirms that
2983 * it isn't being stopped right now.
2985 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
2988 if ((err = mddev_lock(mddev)))
2993 mddev_unlock(mddev);
2995 check_disk_change(inode->i_bdev);
3000 static int md_release(struct inode *inode, struct file * file)
3002 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3011 static int md_media_changed(struct gendisk *disk)
3013 mddev_t *mddev = disk->private_data;
3015 return mddev->changed;
3018 static int md_revalidate(struct gendisk *disk)
3020 mddev_t *mddev = disk->private_data;
3025 static struct block_device_operations md_fops =
3027 .owner = THIS_MODULE,
3029 .release = md_release,
3031 .media_changed = md_media_changed,
3032 .revalidate_disk= md_revalidate,
3035 static int md_thread(void * arg)
3037 mdk_thread_t *thread = arg;
3045 daemonize(thread->name, mdname(thread->mddev));
3047 current->exit_signal = SIGCHLD;
3048 allow_signal(SIGKILL);
3049 thread->tsk = current;
3052 * md_thread is a 'system-thread', it's priority should be very
3053 * high. We avoid resource deadlocks individually in each
3054 * raid personality. (RAID5 does preallocation) We also use RR and
3055 * the very same RT priority as kswapd, thus we will never get
3056 * into a priority inversion deadlock.
3058 * we definitely have to have equal or higher priority than
3059 * bdflush, otherwise bdflush will deadlock if there are too
3060 * many dirty RAID5 blocks.
3064 complete(thread->event);
3065 while (thread->run) {
3066 void (*run)(mddev_t *);
3068 wait_event_interruptible_timeout(thread->wqueue,
3069 test_bit(THREAD_WAKEUP, &thread->flags),
3073 clear_bit(THREAD_WAKEUP, &thread->flags);
3079 if (signal_pending(current))
3080 flush_signals(current);
3082 complete(thread->event);
3086 void md_wakeup_thread(mdk_thread_t *thread)
3089 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3090 set_bit(THREAD_WAKEUP, &thread->flags);
3091 wake_up(&thread->wqueue);
3095 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3098 mdk_thread_t *thread;
3100 struct completion event;
3102 thread = (mdk_thread_t *) kmalloc
3103 (sizeof(mdk_thread_t), GFP_KERNEL);
3107 memset(thread, 0, sizeof(mdk_thread_t));
3108 init_waitqueue_head(&thread->wqueue);
3110 init_completion(&event);
3111 thread->event = &event;
3113 thread->mddev = mddev;
3114 thread->name = name;
3115 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3116 ret = kernel_thread(md_thread, thread, 0);
3121 wait_for_completion(&event);
3125 void md_unregister_thread(mdk_thread_t *thread)
3127 struct completion event;
3129 init_completion(&event);
3131 thread->event = &event;
3133 /* As soon as ->run is set to NULL, the task could disappear,
3134 * so we need to hold tasklist_lock until we have sent the signal
3136 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3137 read_lock(&tasklist_lock);
3139 send_sig(SIGKILL, thread->tsk, 1);
3140 read_unlock(&tasklist_lock);
3141 wait_for_completion(&event);
3145 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3152 if (!rdev || rdev->faulty)
3155 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3157 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3158 __builtin_return_address(0),__builtin_return_address(1),
3159 __builtin_return_address(2),__builtin_return_address(3));
3161 if (!mddev->pers->error_handler)
3163 mddev->pers->error_handler(mddev,rdev);
3164 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3165 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3166 md_wakeup_thread(mddev->thread);
3169 /* seq_file implementation /proc/mdstat */
3171 static void status_unused(struct seq_file *seq)
3175 struct list_head *tmp;
3177 seq_printf(seq, "unused devices: ");
3179 ITERATE_RDEV_PENDING(rdev,tmp) {
3180 char b[BDEVNAME_SIZE];
3182 seq_printf(seq, "%s ",
3183 bdevname(rdev->bdev,b));
3186 seq_printf(seq, "<none>");
3188 seq_printf(seq, "\n");
3192 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3194 unsigned long max_blocks, resync, res, dt, db, rt;
3196 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3198 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3199 max_blocks = mddev->resync_max_sectors >> 1;
3201 max_blocks = mddev->size;
3204 * Should not happen.
3210 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3212 int i, x = res/50, y = 20-x;
3213 seq_printf(seq, "[");
3214 for (i = 0; i < x; i++)
3215 seq_printf(seq, "=");
3216 seq_printf(seq, ">");
3217 for (i = 0; i < y; i++)
3218 seq_printf(seq, ".");
3219 seq_printf(seq, "] ");
3221 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3222 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3223 "resync" : "recovery"),
3224 res/10, res % 10, resync, max_blocks);
3227 * We do not want to overflow, so the order of operands and
3228 * the * 100 / 100 trick are important. We do a +1 to be
3229 * safe against division by zero. We only estimate anyway.
3231 * dt: time from mark until now
3232 * db: blocks written from mark until now
3233 * rt: remaining time
3235 dt = ((jiffies - mddev->resync_mark) / HZ);
3237 db = resync - (mddev->resync_mark_cnt/2);
3238 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3240 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3242 seq_printf(seq, " speed=%ldK/sec", db/dt);
3245 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3247 struct list_head *tmp;
3257 spin_lock(&all_mddevs_lock);
3258 list_for_each(tmp,&all_mddevs)
3260 mddev = list_entry(tmp, mddev_t, all_mddevs);
3262 spin_unlock(&all_mddevs_lock);
3265 spin_unlock(&all_mddevs_lock);
3267 return (void*)2;/* tail */
3271 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3273 struct list_head *tmp;
3274 mddev_t *next_mddev, *mddev = v;
3280 spin_lock(&all_mddevs_lock);
3282 tmp = all_mddevs.next;
3284 tmp = mddev->all_mddevs.next;
3285 if (tmp != &all_mddevs)
3286 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3288 next_mddev = (void*)2;
3291 spin_unlock(&all_mddevs_lock);
3299 static void md_seq_stop(struct seq_file *seq, void *v)
3303 if (mddev && v != (void*)1 && v != (void*)2)
3307 static int md_seq_show(struct seq_file *seq, void *v)
3311 struct list_head *tmp2;
3314 struct bitmap *bitmap;
3316 if (v == (void*)1) {
3317 seq_printf(seq, "Personalities : ");
3318 spin_lock(&pers_lock);
3319 for (i = 0; i < MAX_PERSONALITY; i++)
3321 seq_printf(seq, "[%s] ", pers[i]->name);
3323 spin_unlock(&pers_lock);
3324 seq_printf(seq, "\n");
3327 if (v == (void*)2) {
3332 if (mddev_lock(mddev)!=0)
3334 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3335 seq_printf(seq, "%s : %sactive", mdname(mddev),
3336 mddev->pers ? "" : "in");
3339 seq_printf(seq, " (read-only)");
3340 seq_printf(seq, " %s", mddev->pers->name);
3344 ITERATE_RDEV(mddev,rdev,tmp2) {
3345 char b[BDEVNAME_SIZE];
3346 seq_printf(seq, " %s[%d]",
3347 bdevname(rdev->bdev,b), rdev->desc_nr);
3348 if (test_bit(WriteMostly, &rdev->flags))
3349 seq_printf(seq, "(W)");
3351 seq_printf(seq, "(F)");
3357 if (!list_empty(&mddev->disks)) {
3359 seq_printf(seq, "\n %llu blocks",
3360 (unsigned long long)mddev->array_size);
3362 seq_printf(seq, "\n %llu blocks",
3363 (unsigned long long)size);
3367 mddev->pers->status (seq, mddev);
3368 seq_printf(seq, "\n ");
3369 if (mddev->curr_resync > 2) {
3370 status_resync (seq, mddev);
3371 seq_printf(seq, "\n ");
3372 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3373 seq_printf(seq, " resync=DELAYED\n ");
3375 seq_printf(seq, "\n ");
3377 if ((bitmap = mddev->bitmap)) {
3378 unsigned long chunk_kb;
3379 unsigned long flags;
3380 spin_lock_irqsave(&bitmap->lock, flags);
3381 chunk_kb = bitmap->chunksize >> 10;
3382 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3384 bitmap->pages - bitmap->missing_pages,
3386 (bitmap->pages - bitmap->missing_pages)
3387 << (PAGE_SHIFT - 10),
3388 chunk_kb ? chunk_kb : bitmap->chunksize,
3389 chunk_kb ? "KB" : "B");
3391 seq_printf(seq, ", file: ");
3392 seq_path(seq, bitmap->file->f_vfsmnt,
3393 bitmap->file->f_dentry," \t\n");
3396 seq_printf(seq, "\n");
3397 spin_unlock_irqrestore(&bitmap->lock, flags);
3400 seq_printf(seq, "\n");
3402 mddev_unlock(mddev);
3407 static struct seq_operations md_seq_ops = {
3408 .start = md_seq_start,
3409 .next = md_seq_next,
3410 .stop = md_seq_stop,
3411 .show = md_seq_show,
3414 static int md_seq_open(struct inode *inode, struct file *file)
3418 error = seq_open(file, &md_seq_ops);
3422 static struct file_operations md_seq_fops = {
3423 .open = md_seq_open,
3425 .llseek = seq_lseek,
3426 .release = seq_release,
3429 int register_md_personality(int pnum, mdk_personality_t *p)
3431 if (pnum >= MAX_PERSONALITY) {
3433 "md: tried to install personality %s as nr %d, but max is %lu\n",
3434 p->name, pnum, MAX_PERSONALITY-1);
3438 spin_lock(&pers_lock);
3440 spin_unlock(&pers_lock);
3445 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3446 spin_unlock(&pers_lock);
3450 int unregister_md_personality(int pnum)
3452 if (pnum >= MAX_PERSONALITY)
3455 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3456 spin_lock(&pers_lock);
3458 spin_unlock(&pers_lock);
3462 static int is_mddev_idle(mddev_t *mddev)
3465 struct list_head *tmp;
3467 unsigned long curr_events;
3470 ITERATE_RDEV(mddev,rdev,tmp) {
3471 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3472 curr_events = disk_stat_read(disk, read_sectors) +
3473 disk_stat_read(disk, write_sectors) -
3474 atomic_read(&disk->sync_io);
3475 /* Allow some slack between valud of curr_events and last_events,
3476 * as there are some uninteresting races.
3477 * Note: the following is an unsigned comparison.
3479 if ((curr_events - rdev->last_events + 32) > 64) {
3480 rdev->last_events = curr_events;
3487 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3489 /* another "blocks" (512byte) blocks have been synced */
3490 atomic_sub(blocks, &mddev->recovery_active);
3491 wake_up(&mddev->recovery_wait);
3493 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3494 md_wakeup_thread(mddev->thread);
3495 // stop recovery, signal do_sync ....
3500 /* md_write_start(mddev, bi)
3501 * If we need to update some array metadata (e.g. 'active' flag
3502 * in superblock) before writing, schedule a superblock update
3503 * and wait for it to complete.
3505 void md_write_start(mddev_t *mddev, struct bio *bi)
3508 if (bio_data_dir(bi) != WRITE)
3511 atomic_inc(&mddev->writes_pending);
3512 if (mddev->in_sync) {
3513 spin_lock(&mddev->write_lock);
3514 if (mddev->in_sync) {
3516 mddev->sb_dirty = 1;
3517 md_wakeup_thread(mddev->thread);
3519 spin_unlock(&mddev->write_lock);
3521 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3524 void md_write_end(mddev_t *mddev)
3526 if (atomic_dec_and_test(&mddev->writes_pending)) {
3527 if (mddev->safemode == 2)
3528 md_wakeup_thread(mddev->thread);
3530 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3534 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3536 #define SYNC_MARKS 10
3537 #define SYNC_MARK_STEP (3*HZ)
3538 static void md_do_sync(mddev_t *mddev)
3541 unsigned int currspeed = 0,
3543 sector_t max_sectors,j, io_sectors;
3544 unsigned long mark[SYNC_MARKS];
3545 sector_t mark_cnt[SYNC_MARKS];
3547 struct list_head *tmp;
3548 sector_t last_check;
3551 /* just incase thread restarts... */
3552 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3555 /* we overload curr_resync somewhat here.
3556 * 0 == not engaged in resync at all
3557 * 2 == checking that there is no conflict with another sync
3558 * 1 == like 2, but have yielded to allow conflicting resync to
3560 * other == active in resync - this many blocks
3562 * Before starting a resync we must have set curr_resync to
3563 * 2, and then checked that every "conflicting" array has curr_resync
3564 * less than ours. When we find one that is the same or higher
3565 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3566 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3567 * This will mean we have to start checking from the beginning again.
3572 mddev->curr_resync = 2;
3575 if (signal_pending(current)) {
3576 flush_signals(current);
3579 ITERATE_MDDEV(mddev2,tmp) {
3580 if (mddev2 == mddev)
3582 if (mddev2->curr_resync &&
3583 match_mddev_units(mddev,mddev2)) {
3585 if (mddev < mddev2 && mddev->curr_resync == 2) {
3586 /* arbitrarily yield */
3587 mddev->curr_resync = 1;
3588 wake_up(&resync_wait);
3590 if (mddev > mddev2 && mddev->curr_resync == 1)
3591 /* no need to wait here, we can wait the next
3592 * time 'round when curr_resync == 2
3595 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3596 if (!signal_pending(current)
3597 && mddev2->curr_resync >= mddev->curr_resync) {
3598 printk(KERN_INFO "md: delaying resync of %s"
3599 " until %s has finished resync (they"
3600 " share one or more physical units)\n",
3601 mdname(mddev), mdname(mddev2));
3604 finish_wait(&resync_wait, &wq);
3607 finish_wait(&resync_wait, &wq);
3610 } while (mddev->curr_resync < 2);
3612 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3613 /* resync follows the size requested by the personality,
3614 * which defaults to physical size, but can be virtual size
3616 max_sectors = mddev->resync_max_sectors;
3618 /* recovery follows the physical size of devices */
3619 max_sectors = mddev->size << 1;
3621 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3622 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3623 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3624 printk(KERN_INFO "md: using maximum available idle IO bandwith "
3625 "(but not more than %d KB/sec) for reconstruction.\n",
3626 sysctl_speed_limit_max);
3628 is_mddev_idle(mddev); /* this also initializes IO event counters */
3629 /* we don't use the checkpoint if there's a bitmap */
3630 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3631 j = mddev->recovery_cp;
3635 for (m = 0; m < SYNC_MARKS; m++) {
3637 mark_cnt[m] = io_sectors;
3640 mddev->resync_mark = mark[last_mark];
3641 mddev->resync_mark_cnt = mark_cnt[last_mark];
3644 * Tune reconstruction:
3646 window = 32*(PAGE_SIZE/512);
3647 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3648 window/2,(unsigned long long) max_sectors/2);
3650 atomic_set(&mddev->recovery_active, 0);
3651 init_waitqueue_head(&mddev->recovery_wait);
3656 "md: resuming recovery of %s from checkpoint.\n",
3658 mddev->curr_resync = j;
3661 while (j < max_sectors) {
3665 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3666 currspeed < sysctl_speed_limit_min);
3668 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3672 if (!skipped) { /* actual IO requested */
3673 io_sectors += sectors;
3674 atomic_add(sectors, &mddev->recovery_active);
3678 if (j>1) mddev->curr_resync = j;
3681 if (last_check + window > io_sectors || j == max_sectors)
3684 last_check = io_sectors;
3686 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3687 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3691 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3693 int next = (last_mark+1) % SYNC_MARKS;
3695 mddev->resync_mark = mark[next];
3696 mddev->resync_mark_cnt = mark_cnt[next];
3697 mark[next] = jiffies;
3698 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3703 if (signal_pending(current)) {
3705 * got a signal, exit.
3708 "md: md_do_sync() got signal ... exiting\n");
3709 flush_signals(current);
3710 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3715 * this loop exits only if either when we are slower than
3716 * the 'hard' speed limit, or the system was IO-idle for
3718 * the system might be non-idle CPU-wise, but we only care
3719 * about not overloading the IO subsystem. (things like an
3720 * e2fsck being done on the RAID array should execute fast)
3722 mddev->queue->unplug_fn(mddev->queue);
3725 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3726 /((jiffies-mddev->resync_mark)/HZ +1) +1;
3728 if (currspeed > sysctl_speed_limit_min) {
3729 if ((currspeed > sysctl_speed_limit_max) ||
3730 !is_mddev_idle(mddev)) {
3731 msleep_interruptible(250);
3736 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3738 * this also signals 'finished resyncing' to md_stop
3741 mddev->queue->unplug_fn(mddev->queue);
3743 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3745 /* tell personality that we are finished */
3746 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3748 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3749 mddev->curr_resync > 2 &&
3750 mddev->curr_resync >= mddev->recovery_cp) {
3751 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3753 "md: checkpointing recovery of %s.\n",
3755 mddev->recovery_cp = mddev->curr_resync;
3757 mddev->recovery_cp = MaxSector;
3761 mddev->curr_resync = 0;
3762 wake_up(&resync_wait);
3763 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3764 md_wakeup_thread(mddev->thread);
3769 * This routine is regularly called by all per-raid-array threads to
3770 * deal with generic issues like resync and super-block update.
3771 * Raid personalities that don't have a thread (linear/raid0) do not
3772 * need this as they never do any recovery or update the superblock.
3774 * It does not do any resync itself, but rather "forks" off other threads
3775 * to do that as needed.
3776 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3777 * "->recovery" and create a thread at ->sync_thread.
3778 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3779 * and wakeups up this thread which will reap the thread and finish up.
3780 * This thread also removes any faulty devices (with nr_pending == 0).
3782 * The overall approach is:
3783 * 1/ if the superblock needs updating, update it.
3784 * 2/ If a recovery thread is running, don't do anything else.
3785 * 3/ If recovery has finished, clean up, possibly marking spares active.
3786 * 4/ If there are any faulty devices, remove them.
3787 * 5/ If array is degraded, try to add spares devices
3788 * 6/ If array has spares or is not in-sync, start a resync thread.
3790 void md_check_recovery(mddev_t *mddev)
3793 struct list_head *rtmp;
3797 bitmap_daemon_work(mddev->bitmap);
3802 if (signal_pending(current)) {
3803 if (mddev->pers->sync_request) {
3804 printk(KERN_INFO "md: %s in immediate safe mode\n",
3806 mddev->safemode = 2;
3808 flush_signals(current);
3813 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3814 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3815 (mddev->safemode == 1) ||
3816 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3817 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3821 if (mddev_trylock(mddev)==0) {
3824 spin_lock(&mddev->write_lock);
3825 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3826 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3828 mddev->sb_dirty = 1;
3830 if (mddev->safemode == 1)
3831 mddev->safemode = 0;
3832 spin_unlock(&mddev->write_lock);
3834 if (mddev->sb_dirty)
3835 md_update_sb(mddev);
3838 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3839 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3840 /* resync/recovery still happening */
3841 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3844 if (mddev->sync_thread) {
3845 /* resync has finished, collect result */
3846 md_unregister_thread(mddev->sync_thread);
3847 mddev->sync_thread = NULL;
3848 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3849 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3851 /* activate any spares */
3852 mddev->pers->spare_active(mddev);
3854 md_update_sb(mddev);
3856 /* if array is no-longer degraded, then any saved_raid_disk
3857 * information must be scrapped
3859 if (!mddev->degraded)
3860 ITERATE_RDEV(mddev,rdev,rtmp)
3861 rdev->saved_raid_disk = -1;
3863 mddev->recovery = 0;
3864 /* flag recovery needed just to double check */
3865 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3868 if (mddev->recovery)
3869 /* probably just the RECOVERY_NEEDED flag */
3870 mddev->recovery = 0;
3872 /* no recovery is running.
3873 * remove any failed drives, then
3874 * add spares if possible.
3875 * Spare are also removed and re-added, to allow
3876 * the personality to fail the re-add.
3878 ITERATE_RDEV(mddev,rdev,rtmp)
3879 if (rdev->raid_disk >= 0 &&
3880 (rdev->faulty || ! rdev->in_sync) &&
3881 atomic_read(&rdev->nr_pending)==0) {
3882 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3883 rdev->raid_disk = -1;
3886 if (mddev->degraded) {
3887 ITERATE_RDEV(mddev,rdev,rtmp)
3888 if (rdev->raid_disk < 0
3890 if (mddev->pers->hot_add_disk(mddev,rdev))
3897 if (!spares && (mddev->recovery_cp == MaxSector )) {
3898 /* nothing we can do ... */
3901 if (mddev->pers->sync_request) {
3902 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3904 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3905 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3906 /* We are adding a device or devices to an array
3907 * which has the bitmap stored on all devices.
3908 * So make sure all bitmap pages get written
3910 bitmap_write_all(mddev->bitmap);
3912 mddev->sync_thread = md_register_thread(md_do_sync,
3915 if (!mddev->sync_thread) {
3916 printk(KERN_ERR "%s: could not start resync"
3919 /* leave the spares where they are, it shouldn't hurt */
3920 mddev->recovery = 0;
3922 md_wakeup_thread(mddev->sync_thread);
3926 mddev_unlock(mddev);
3930 static int md_notify_reboot(struct notifier_block *this,
3931 unsigned long code, void *x)
3933 struct list_head *tmp;
3936 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3938 printk(KERN_INFO "md: stopping all md devices.\n");
3940 ITERATE_MDDEV(mddev,tmp)
3941 if (mddev_trylock(mddev)==0)
3942 do_md_stop (mddev, 1);
3944 * certain more exotic SCSI devices are known to be
3945 * volatile wrt too early system reboots. While the
3946 * right place to handle this issue is the given
3947 * driver, we do want to have a safe RAID driver ...
3954 static struct notifier_block md_notifier = {
3955 .notifier_call = md_notify_reboot,
3957 .priority = INT_MAX, /* before any real devices */
3960 static void md_geninit(void)
3962 struct proc_dir_entry *p;
3964 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3966 p = create_proc_entry("mdstat", S_IRUGO, NULL);
3968 p->proc_fops = &md_seq_fops;
3971 static int __init md_init(void)
3975 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3976 " MD_SB_DISKS=%d\n",
3977 MD_MAJOR_VERSION, MD_MINOR_VERSION,
3978 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3979 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3982 if (register_blkdev(MAJOR_NR, "md"))
3984 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3985 unregister_blkdev(MAJOR_NR, "md");
3989 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3990 md_probe, NULL, NULL);
3991 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3992 md_probe, NULL, NULL);
3994 for (minor=0; minor < MAX_MD_DEVS; ++minor)
3995 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3996 S_IFBLK|S_IRUSR|S_IWUSR,
3999 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4000 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4001 S_IFBLK|S_IRUSR|S_IWUSR,
4005 register_reboot_notifier(&md_notifier);
4006 raid_table_header = register_sysctl_table(raid_root_table, 1);
4016 * Searches all registered partitions for autorun RAID arrays
4019 static dev_t detected_devices[128];
4022 void md_autodetect_dev(dev_t dev)
4024 if (dev_cnt >= 0 && dev_cnt < 127)
4025 detected_devices[dev_cnt++] = dev;
4029 static void autostart_arrays(int part)
4034 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4036 for (i = 0; i < dev_cnt; i++) {
4037 dev_t dev = detected_devices[i];
4039 rdev = md_import_device(dev,0, 0);
4047 list_add(&rdev->same_set, &pending_raid_disks);
4051 autorun_devices(part);
4056 static __exit void md_exit(void)
4059 struct list_head *tmp;
4061 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4062 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4063 for (i=0; i < MAX_MD_DEVS; i++)
4064 devfs_remove("md/%d", i);
4065 for (i=0; i < MAX_MD_DEVS; i++)
4066 devfs_remove("md/d%d", i);
4070 unregister_blkdev(MAJOR_NR,"md");
4071 unregister_blkdev(mdp_major, "mdp");
4072 unregister_reboot_notifier(&md_notifier);
4073 unregister_sysctl_table(raid_table_header);
4074 remove_proc_entry("mdstat", NULL);
4075 ITERATE_MDDEV(mddev,tmp) {
4076 struct gendisk *disk = mddev->gendisk;
4079 export_array(mddev);
4082 mddev->gendisk = NULL;
4087 module_init(md_init)
4088 module_exit(md_exit)
4090 EXPORT_SYMBOL(register_md_personality);
4091 EXPORT_SYMBOL(unregister_md_personality);
4092 EXPORT_SYMBOL(md_error);
4093 EXPORT_SYMBOL(md_done_sync);
4094 EXPORT_SYMBOL(md_write_start);
4095 EXPORT_SYMBOL(md_write_end);
4096 EXPORT_SYMBOL(md_register_thread);
4097 EXPORT_SYMBOL(md_unregister_thread);
4098 EXPORT_SYMBOL(md_wakeup_thread);
4099 EXPORT_SYMBOL(md_print_devices);
4100 EXPORT_SYMBOL(md_check_recovery);
4101 MODULE_LICENSE("GPL");
4103 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);