2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
46 #include <linux/init.h>
48 #include <linux/file.h>
51 #include <linux/kmod.h>
54 #include <asm/unaligned.h>
56 #define MAJOR_NR MD_MAJOR
59 /* 63 partitions with the alternate major number (mdp) */
60 #define MdpMinorShift 6
63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
67 static void autostart_arrays (int part);
70 static mdk_personality_t *pers[MAX_PERSONALITY];
71 static DEFINE_SPINLOCK(pers_lock);
74 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
75 * is 1000 KB/sec, so the extra system load does not show up that much.
76 * Increase it if you want to have more _guaranteed_ speed. Note that
77 * the RAID driver will use the maximum available bandwidth if the IO
78 * subsystem is idle. There is also an 'absolute maximum' reconstruction
79 * speed limit - in case reconstruction slows down your system despite
82 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
85 static int sysctl_speed_limit_min = 1000;
86 static int sysctl_speed_limit_max = 200000;
88 static struct ctl_table_header *raid_table_header;
90 static ctl_table raid_table[] = {
92 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
93 .procname = "speed_limit_min",
94 .data = &sysctl_speed_limit_min,
95 .maxlen = sizeof(int),
97 .proc_handler = &proc_dointvec,
100 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
101 .procname = "speed_limit_max",
102 .data = &sysctl_speed_limit_max,
103 .maxlen = sizeof(int),
105 .proc_handler = &proc_dointvec,
110 static ctl_table raid_dir_table[] = {
112 .ctl_name = DEV_RAID,
121 static ctl_table raid_root_table[] = {
127 .child = raid_dir_table,
132 static struct block_device_operations md_fops;
135 * Enables to iterate over all existing md arrays
136 * all_mddevs_lock protects this list.
138 static LIST_HEAD(all_mddevs);
139 static DEFINE_SPINLOCK(all_mddevs_lock);
143 * iterates through all used mddevs in the system.
144 * We take care to grab the all_mddevs_lock whenever navigating
145 * the list, and to always hold a refcount when unlocked.
146 * Any code which breaks out of this loop while own
147 * a reference to the current mddev and must mddev_put it.
149 #define ITERATE_MDDEV(mddev,tmp) \
151 for (({ spin_lock(&all_mddevs_lock); \
152 tmp = all_mddevs.next; \
154 ({ if (tmp != &all_mddevs) \
155 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
156 spin_unlock(&all_mddevs_lock); \
157 if (mddev) mddev_put(mddev); \
158 mddev = list_entry(tmp, mddev_t, all_mddevs); \
159 tmp != &all_mddevs;}); \
160 ({ spin_lock(&all_mddevs_lock); \
165 static int md_fail_request (request_queue_t *q, struct bio *bio)
167 bio_io_error(bio, bio->bi_size);
171 static inline mddev_t *mddev_get(mddev_t *mddev)
173 atomic_inc(&mddev->active);
177 static void mddev_put(mddev_t *mddev)
179 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
181 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
182 list_del(&mddev->all_mddevs);
183 blk_put_queue(mddev->queue);
184 kobject_unregister(&mddev->kobj);
186 spin_unlock(&all_mddevs_lock);
189 static mddev_t * mddev_find(dev_t unit)
191 mddev_t *mddev, *new = NULL;
194 spin_lock(&all_mddevs_lock);
195 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
196 if (mddev->unit == unit) {
198 spin_unlock(&all_mddevs_lock);
204 list_add(&new->all_mddevs, &all_mddevs);
205 spin_unlock(&all_mddevs_lock);
208 spin_unlock(&all_mddevs_lock);
210 new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
214 memset(new, 0, sizeof(*new));
217 if (MAJOR(unit) == MD_MAJOR)
218 new->md_minor = MINOR(unit);
220 new->md_minor = MINOR(unit) >> MdpMinorShift;
222 init_MUTEX(&new->reconfig_sem);
223 INIT_LIST_HEAD(&new->disks);
224 INIT_LIST_HEAD(&new->all_mddevs);
225 init_timer(&new->safemode_timer);
226 atomic_set(&new->active, 1);
227 spin_lock_init(&new->write_lock);
228 init_waitqueue_head(&new->sb_wait);
230 new->queue = blk_alloc_queue(GFP_KERNEL);
236 blk_queue_make_request(new->queue, md_fail_request);
241 static inline int mddev_lock(mddev_t * mddev)
243 return down_interruptible(&mddev->reconfig_sem);
246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
248 down(&mddev->reconfig_sem);
251 static inline int mddev_trylock(mddev_t * mddev)
253 return down_trylock(&mddev->reconfig_sem);
256 static inline void mddev_unlock(mddev_t * mddev)
258 up(&mddev->reconfig_sem);
260 md_wakeup_thread(mddev->thread);
263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
266 struct list_head *tmp;
268 ITERATE_RDEV(mddev,rdev,tmp) {
269 if (rdev->desc_nr == nr)
275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
277 struct list_head *tmp;
280 ITERATE_RDEV(mddev,rdev,tmp) {
281 if (rdev->bdev->bd_dev == dev)
287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
289 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
290 return MD_NEW_SIZE_BLOCKS(size);
293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
297 size = rdev->sb_offset;
300 size &= ~((sector_t)chunk_size/1024 - 1);
304 static int alloc_disk_sb(mdk_rdev_t * rdev)
309 rdev->sb_page = alloc_page(GFP_KERNEL);
310 if (!rdev->sb_page) {
311 printk(KERN_ALERT "md: out of memory.\n");
318 static void free_disk_sb(mdk_rdev_t * rdev)
321 page_cache_release(rdev->sb_page);
323 rdev->sb_page = NULL;
330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
332 mdk_rdev_t *rdev = bio->bi_private;
336 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
337 md_error(rdev->mddev, rdev);
339 if (atomic_dec_and_test(&rdev->mddev->pending_writes))
340 wake_up(&rdev->mddev->sb_wait);
345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
346 sector_t sector, int size, struct page *page)
348 /* write first size bytes of page to sector of rdev
349 * Increment mddev->pending_writes before returning
350 * and decrement it on completion, waking up sb_wait
351 * if zero is reached.
352 * If an error occurred, call md_error
354 struct bio *bio = bio_alloc(GFP_NOIO, 1);
356 bio->bi_bdev = rdev->bdev;
357 bio->bi_sector = sector;
358 bio_add_page(bio, page, size, 0);
359 bio->bi_private = rdev;
360 bio->bi_end_io = super_written;
361 atomic_inc(&mddev->pending_writes);
362 submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
370 complete((struct completion*)bio->bi_private);
374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
375 struct page *page, int rw)
377 struct bio *bio = bio_alloc(GFP_NOIO, 1);
378 struct completion event;
381 rw |= (1 << BIO_RW_SYNC);
384 bio->bi_sector = sector;
385 bio_add_page(bio, page, size, 0);
386 init_completion(&event);
387 bio->bi_private = &event;
388 bio->bi_end_io = bi_complete;
390 wait_for_completion(&event);
392 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
397 static int read_disk_sb(mdk_rdev_t * rdev, int size)
399 char b[BDEVNAME_SIZE];
400 if (!rdev->sb_page) {
408 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
414 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
415 bdevname(rdev->bdev,b));
419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
421 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
422 (sb1->set_uuid1 == sb2->set_uuid1) &&
423 (sb1->set_uuid2 == sb2->set_uuid2) &&
424 (sb1->set_uuid3 == sb2->set_uuid3))
432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
435 mdp_super_t *tmp1, *tmp2;
437 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
438 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
440 if (!tmp1 || !tmp2) {
442 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
450 * nr_disks is not constant
455 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
466 static unsigned int calc_sb_csum(mdp_super_t * sb)
468 unsigned int disk_csum, csum;
470 disk_csum = sb->sb_csum;
472 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
473 sb->sb_csum = disk_csum;
479 * Handle superblock details.
480 * We want to be able to handle multiple superblock formats
481 * so we have a common interface to them all, and an array of
482 * different handlers.
483 * We rely on user-space to write the initial superblock, and support
484 * reading and updating of superblocks.
485 * Interface methods are:
486 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
487 * loads and validates a superblock on dev.
488 * if refdev != NULL, compare superblocks on both devices
490 * 0 - dev has a superblock that is compatible with refdev
491 * 1 - dev has a superblock that is compatible and newer than refdev
492 * so dev should be used as the refdev in future
493 * -EINVAL superblock incompatible or invalid
494 * -othererror e.g. -EIO
496 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
497 * Verify that dev is acceptable into mddev.
498 * The first time, mddev->raid_disks will be 0, and data from
499 * dev should be merged in. Subsequent calls check that dev
500 * is new enough. Return 0 or -EINVAL
502 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
503 * Update the superblock for rdev with data in mddev
504 * This does not write to disc.
510 struct module *owner;
511 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
512 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
513 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
517 * load_super for 0.90.0
519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
521 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
527 * Calculate the position of the superblock,
528 * it's at the end of the disk.
530 * It also happens to be a multiple of 4Kb.
532 sb_offset = calc_dev_sboffset(rdev->bdev);
533 rdev->sb_offset = sb_offset;
535 ret = read_disk_sb(rdev, MD_SB_BYTES);
540 bdevname(rdev->bdev, b);
541 sb = (mdp_super_t*)page_address(rdev->sb_page);
543 if (sb->md_magic != MD_SB_MAGIC) {
544 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
549 if (sb->major_version != 0 ||
550 sb->minor_version != 90) {
551 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
552 sb->major_version, sb->minor_version,
557 if (sb->raid_disks <= 0)
560 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
561 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
566 rdev->preferred_minor = sb->md_minor;
567 rdev->data_offset = 0;
568 rdev->sb_size = MD_SB_BYTES;
570 if (sb->level == LEVEL_MULTIPATH)
573 rdev->desc_nr = sb->this_disk.number;
579 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
580 if (!uuid_equal(refsb, sb)) {
581 printk(KERN_WARNING "md: %s has different UUID to %s\n",
582 b, bdevname(refdev->bdev,b2));
585 if (!sb_equal(refsb, sb)) {
586 printk(KERN_WARNING "md: %s has same UUID"
587 " but different superblock to %s\n",
588 b, bdevname(refdev->bdev, b2));
592 ev2 = md_event(refsb);
598 rdev->size = calc_dev_size(rdev, sb->chunk_size);
605 * validate_super for 0.90.0
607 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
610 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
612 rdev->raid_disk = -1;
614 if (mddev->raid_disks == 0) {
615 mddev->major_version = 0;
616 mddev->minor_version = sb->minor_version;
617 mddev->patch_version = sb->patch_version;
618 mddev->persistent = ! sb->not_persistent;
619 mddev->chunk_size = sb->chunk_size;
620 mddev->ctime = sb->ctime;
621 mddev->utime = sb->utime;
622 mddev->level = sb->level;
623 mddev->layout = sb->layout;
624 mddev->raid_disks = sb->raid_disks;
625 mddev->size = sb->size;
626 mddev->events = md_event(sb);
627 mddev->bitmap_offset = 0;
628 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
630 if (sb->state & (1<<MD_SB_CLEAN))
631 mddev->recovery_cp = MaxSector;
633 if (sb->events_hi == sb->cp_events_hi &&
634 sb->events_lo == sb->cp_events_lo) {
635 mddev->recovery_cp = sb->recovery_cp;
637 mddev->recovery_cp = 0;
640 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
641 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
642 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
643 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
645 mddev->max_disks = MD_SB_DISKS;
647 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
648 mddev->bitmap_file == NULL) {
649 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
650 /* FIXME use a better test */
651 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
654 mddev->bitmap_offset = mddev->default_bitmap_offset;
657 } else if (mddev->pers == NULL) {
658 /* Insist on good event counter while assembling */
659 __u64 ev1 = md_event(sb);
661 if (ev1 < mddev->events)
663 } else if (mddev->bitmap) {
664 /* if adding to array with a bitmap, then we can accept an
665 * older device ... but not too old.
667 __u64 ev1 = md_event(sb);
668 if (ev1 < mddev->bitmap->events_cleared)
670 } else /* just a hot-add of a new device, leave raid_disk at -1 */
673 if (mddev->level != LEVEL_MULTIPATH) {
676 desc = sb->disks + rdev->desc_nr;
678 if (desc->state & (1<<MD_DISK_FAULTY))
680 else if (desc->state & (1<<MD_DISK_SYNC) &&
681 desc->raid_disk < mddev->raid_disks) {
683 rdev->raid_disk = desc->raid_disk;
685 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
686 set_bit(WriteMostly, &rdev->flags);
687 } else /* MULTIPATH are always insync */
693 * sync_super for 0.90.0
695 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
698 struct list_head *tmp;
700 int next_spare = mddev->raid_disks;
702 /* make rdev->sb match mddev data..
705 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
706 * 3/ any empty disks < next_spare become removed
708 * disks[0] gets initialised to REMOVED because
709 * we cannot be sure from other fields if it has
710 * been initialised or not.
713 int active=0, working=0,failed=0,spare=0,nr_disks=0;
715 rdev->sb_size = MD_SB_BYTES;
717 sb = (mdp_super_t*)page_address(rdev->sb_page);
719 memset(sb, 0, sizeof(*sb));
721 sb->md_magic = MD_SB_MAGIC;
722 sb->major_version = mddev->major_version;
723 sb->minor_version = mddev->minor_version;
724 sb->patch_version = mddev->patch_version;
725 sb->gvalid_words = 0; /* ignored */
726 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
727 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
728 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
729 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
731 sb->ctime = mddev->ctime;
732 sb->level = mddev->level;
733 sb->size = mddev->size;
734 sb->raid_disks = mddev->raid_disks;
735 sb->md_minor = mddev->md_minor;
736 sb->not_persistent = !mddev->persistent;
737 sb->utime = mddev->utime;
739 sb->events_hi = (mddev->events>>32);
740 sb->events_lo = (u32)mddev->events;
744 sb->recovery_cp = mddev->recovery_cp;
745 sb->cp_events_hi = (mddev->events>>32);
746 sb->cp_events_lo = (u32)mddev->events;
747 if (mddev->recovery_cp == MaxSector)
748 sb->state = (1<< MD_SB_CLEAN);
752 sb->layout = mddev->layout;
753 sb->chunk_size = mddev->chunk_size;
755 if (mddev->bitmap && mddev->bitmap_file == NULL)
756 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
758 sb->disks[0].state = (1<<MD_DISK_REMOVED);
759 ITERATE_RDEV(mddev,rdev2,tmp) {
761 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
762 rdev2->desc_nr = rdev2->raid_disk;
764 rdev2->desc_nr = next_spare++;
765 d = &sb->disks[rdev2->desc_nr];
767 d->number = rdev2->desc_nr;
768 d->major = MAJOR(rdev2->bdev->bd_dev);
769 d->minor = MINOR(rdev2->bdev->bd_dev);
770 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
771 d->raid_disk = rdev2->raid_disk;
773 d->raid_disk = rdev2->desc_nr; /* compatibility */
775 d->state = (1<<MD_DISK_FAULTY);
777 } else if (rdev2->in_sync) {
778 d->state = (1<<MD_DISK_ACTIVE);
779 d->state |= (1<<MD_DISK_SYNC);
787 if (test_bit(WriteMostly, &rdev2->flags))
788 d->state |= (1<<MD_DISK_WRITEMOSTLY);
791 /* now set the "removed" and "faulty" bits on any missing devices */
792 for (i=0 ; i < mddev->raid_disks ; i++) {
793 mdp_disk_t *d = &sb->disks[i];
794 if (d->state == 0 && d->number == 0) {
797 d->state = (1<<MD_DISK_REMOVED);
798 d->state |= (1<<MD_DISK_FAULTY);
802 sb->nr_disks = nr_disks;
803 sb->active_disks = active;
804 sb->working_disks = working;
805 sb->failed_disks = failed;
806 sb->spare_disks = spare;
808 sb->this_disk = sb->disks[rdev->desc_nr];
809 sb->sb_csum = calc_sb_csum(sb);
813 * version 1 superblock
816 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
818 unsigned int disk_csum, csum;
819 unsigned long long newcsum;
820 int size = 256 + le32_to_cpu(sb->max_dev)*2;
821 unsigned int *isuper = (unsigned int*)sb;
824 disk_csum = sb->sb_csum;
827 for (i=0; size>=4; size -= 4 )
828 newcsum += le32_to_cpu(*isuper++);
831 newcsum += le16_to_cpu(*(unsigned short*) isuper);
833 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
834 sb->sb_csum = disk_csum;
835 return cpu_to_le32(csum);
838 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
840 struct mdp_superblock_1 *sb;
843 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
847 * Calculate the position of the superblock.
848 * It is always aligned to a 4K boundary and
849 * depeding on minor_version, it can be:
850 * 0: At least 8K, but less than 12K, from end of device
851 * 1: At start of device
852 * 2: 4K from start of device.
854 switch(minor_version) {
856 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
858 sb_offset &= ~(sector_t)(4*2-1);
859 /* convert from sectors to K */
871 rdev->sb_offset = sb_offset;
873 /* superblock is rarely larger than 1K, but it can be larger,
874 * and it is safe to read 4k, so we do that
876 ret = read_disk_sb(rdev, 4096);
880 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
882 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
883 sb->major_version != cpu_to_le32(1) ||
884 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
885 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
886 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
889 if (calc_sb_1_csum(sb) != sb->sb_csum) {
890 printk("md: invalid superblock checksum on %s\n",
891 bdevname(rdev->bdev,b));
894 if (le64_to_cpu(sb->data_size) < 10) {
895 printk("md: data_size too small on %s\n",
896 bdevname(rdev->bdev,b));
899 rdev->preferred_minor = 0xffff;
900 rdev->data_offset = le64_to_cpu(sb->data_offset);
902 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
903 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
904 if (rdev->sb_size & bmask)
905 rdev-> sb_size = (rdev->sb_size | bmask)+1;
911 struct mdp_superblock_1 *refsb =
912 (struct mdp_superblock_1*)page_address(refdev->sb_page);
914 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
915 sb->level != refsb->level ||
916 sb->layout != refsb->layout ||
917 sb->chunksize != refsb->chunksize) {
918 printk(KERN_WARNING "md: %s has strangely different"
919 " superblock to %s\n",
920 bdevname(rdev->bdev,b),
921 bdevname(refdev->bdev,b2));
924 ev1 = le64_to_cpu(sb->events);
925 ev2 = le64_to_cpu(refsb->events);
931 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
933 rdev->size = rdev->sb_offset;
934 if (rdev->size < le64_to_cpu(sb->data_size)/2)
936 rdev->size = le64_to_cpu(sb->data_size)/2;
937 if (le32_to_cpu(sb->chunksize))
938 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
942 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
944 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
946 rdev->raid_disk = -1;
948 if (mddev->raid_disks == 0) {
949 mddev->major_version = 1;
950 mddev->patch_version = 0;
951 mddev->persistent = 1;
952 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
953 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
954 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
955 mddev->level = le32_to_cpu(sb->level);
956 mddev->layout = le32_to_cpu(sb->layout);
957 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
958 mddev->size = le64_to_cpu(sb->size)/2;
959 mddev->events = le64_to_cpu(sb->events);
960 mddev->bitmap_offset = 0;
961 mddev->default_bitmap_offset = 0;
962 mddev->default_bitmap_offset = 1024;
964 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
965 memcpy(mddev->uuid, sb->set_uuid, 16);
967 mddev->max_disks = (4096-256)/2;
969 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
970 mddev->bitmap_file == NULL ) {
971 if (mddev->level != 1) {
972 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
975 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
977 } else if (mddev->pers == NULL) {
978 /* Insist of good event counter while assembling */
979 __u64 ev1 = le64_to_cpu(sb->events);
981 if (ev1 < mddev->events)
983 } else if (mddev->bitmap) {
984 /* If adding to array with a bitmap, then we can accept an
985 * older device, but not too old.
987 __u64 ev1 = le64_to_cpu(sb->events);
988 if (ev1 < mddev->bitmap->events_cleared)
990 } else /* just a hot-add of a new device, leave raid_disk at -1 */
993 if (mddev->level != LEVEL_MULTIPATH) {
995 rdev->desc_nr = le32_to_cpu(sb->dev_number);
996 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
998 case 0xffff: /* spare */
1001 case 0xfffe: /* faulty */
1007 rdev->raid_disk = role;
1011 if (sb->devflags & WriteMostly1)
1012 set_bit(WriteMostly, &rdev->flags);
1013 } else /* MULTIPATH are always insync */
1019 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1021 struct mdp_superblock_1 *sb;
1022 struct list_head *tmp;
1025 /* make rdev->sb match mddev and rdev data. */
1027 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1029 sb->feature_map = 0;
1031 memset(sb->pad1, 0, sizeof(sb->pad1));
1032 memset(sb->pad2, 0, sizeof(sb->pad2));
1033 memset(sb->pad3, 0, sizeof(sb->pad3));
1035 sb->utime = cpu_to_le64((__u64)mddev->utime);
1036 sb->events = cpu_to_le64(mddev->events);
1038 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1040 sb->resync_offset = cpu_to_le64(0);
1042 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1043 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1044 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1048 ITERATE_RDEV(mddev,rdev2,tmp)
1049 if (rdev2->desc_nr+1 > max_dev)
1050 max_dev = rdev2->desc_nr+1;
1052 sb->max_dev = cpu_to_le32(max_dev);
1053 for (i=0; i<max_dev;i++)
1054 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1056 ITERATE_RDEV(mddev,rdev2,tmp) {
1059 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1060 else if (rdev2->in_sync)
1061 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1063 sb->dev_roles[i] = cpu_to_le16(0xffff);
1066 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1067 sb->sb_csum = calc_sb_1_csum(sb);
1071 static struct super_type super_types[] = {
1074 .owner = THIS_MODULE,
1075 .load_super = super_90_load,
1076 .validate_super = super_90_validate,
1077 .sync_super = super_90_sync,
1081 .owner = THIS_MODULE,
1082 .load_super = super_1_load,
1083 .validate_super = super_1_validate,
1084 .sync_super = super_1_sync,
1088 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1090 struct list_head *tmp;
1093 ITERATE_RDEV(mddev,rdev,tmp)
1094 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1100 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1102 struct list_head *tmp;
1105 ITERATE_RDEV(mddev1,rdev,tmp)
1106 if (match_dev_unit(mddev2, rdev))
1112 static LIST_HEAD(pending_raid_disks);
1114 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1116 mdk_rdev_t *same_pdev;
1117 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1123 same_pdev = match_dev_unit(mddev, rdev);
1126 "%s: WARNING: %s appears to be on the same physical"
1127 " disk as %s. True\n protection against single-disk"
1128 " failure might be compromised.\n",
1129 mdname(mddev), bdevname(rdev->bdev,b),
1130 bdevname(same_pdev->bdev,b2));
1132 /* Verify rdev->desc_nr is unique.
1133 * If it is -1, assign a free number, else
1134 * check number is not in use
1136 if (rdev->desc_nr < 0) {
1138 if (mddev->pers) choice = mddev->raid_disks;
1139 while (find_rdev_nr(mddev, choice))
1141 rdev->desc_nr = choice;
1143 if (find_rdev_nr(mddev, rdev->desc_nr))
1147 list_add(&rdev->same_set, &mddev->disks);
1148 rdev->mddev = mddev;
1149 printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1153 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1155 char b[BDEVNAME_SIZE];
1160 list_del_init(&rdev->same_set);
1161 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1166 * prevent the device from being mounted, repartitioned or
1167 * otherwise reused by a RAID array (or any other kernel
1168 * subsystem), by bd_claiming the device.
1170 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1173 struct block_device *bdev;
1174 char b[BDEVNAME_SIZE];
1176 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1178 printk(KERN_ERR "md: could not open %s.\n",
1179 __bdevname(dev, b));
1180 return PTR_ERR(bdev);
1182 err = bd_claim(bdev, rdev);
1184 printk(KERN_ERR "md: could not bd_claim %s.\n",
1193 static void unlock_rdev(mdk_rdev_t *rdev)
1195 struct block_device *bdev = rdev->bdev;
1203 void md_autodetect_dev(dev_t dev);
1205 static void export_rdev(mdk_rdev_t * rdev)
1207 char b[BDEVNAME_SIZE];
1208 printk(KERN_INFO "md: export_rdev(%s)\n",
1209 bdevname(rdev->bdev,b));
1213 list_del_init(&rdev->same_set);
1215 md_autodetect_dev(rdev->bdev->bd_dev);
1221 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1223 unbind_rdev_from_array(rdev);
1227 static void export_array(mddev_t *mddev)
1229 struct list_head *tmp;
1232 ITERATE_RDEV(mddev,rdev,tmp) {
1237 kick_rdev_from_array(rdev);
1239 if (!list_empty(&mddev->disks))
1241 mddev->raid_disks = 0;
1242 mddev->major_version = 0;
1245 static void print_desc(mdp_disk_t *desc)
1247 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1248 desc->major,desc->minor,desc->raid_disk,desc->state);
1251 static void print_sb(mdp_super_t *sb)
1256 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1257 sb->major_version, sb->minor_version, sb->patch_version,
1258 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1260 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1261 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1262 sb->md_minor, sb->layout, sb->chunk_size);
1263 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1264 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1265 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1266 sb->failed_disks, sb->spare_disks,
1267 sb->sb_csum, (unsigned long)sb->events_lo);
1270 for (i = 0; i < MD_SB_DISKS; i++) {
1273 desc = sb->disks + i;
1274 if (desc->number || desc->major || desc->minor ||
1275 desc->raid_disk || (desc->state && (desc->state != 4))) {
1276 printk(" D %2d: ", i);
1280 printk(KERN_INFO "md: THIS: ");
1281 print_desc(&sb->this_disk);
1285 static void print_rdev(mdk_rdev_t *rdev)
1287 char b[BDEVNAME_SIZE];
1288 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1289 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1290 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1291 if (rdev->sb_loaded) {
1292 printk(KERN_INFO "md: rdev superblock:\n");
1293 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1295 printk(KERN_INFO "md: no rdev superblock!\n");
1298 void md_print_devices(void)
1300 struct list_head *tmp, *tmp2;
1303 char b[BDEVNAME_SIZE];
1306 printk("md: **********************************\n");
1307 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1308 printk("md: **********************************\n");
1309 ITERATE_MDDEV(mddev,tmp) {
1312 bitmap_print_sb(mddev->bitmap);
1314 printk("%s: ", mdname(mddev));
1315 ITERATE_RDEV(mddev,rdev,tmp2)
1316 printk("<%s>", bdevname(rdev->bdev,b));
1319 ITERATE_RDEV(mddev,rdev,tmp2)
1322 printk("md: **********************************\n");
1327 static void sync_sbs(mddev_t * mddev)
1330 struct list_head *tmp;
1332 ITERATE_RDEV(mddev,rdev,tmp) {
1333 super_types[mddev->major_version].
1334 sync_super(mddev, rdev);
1335 rdev->sb_loaded = 1;
1339 static void md_update_sb(mddev_t * mddev)
1342 struct list_head *tmp;
1347 spin_lock(&mddev->write_lock);
1348 sync_req = mddev->in_sync;
1349 mddev->utime = get_seconds();
1352 if (!mddev->events) {
1354 * oops, this 64-bit counter should never wrap.
1355 * Either we are in around ~1 trillion A.C., assuming
1356 * 1 reboot per second, or we have a bug:
1361 mddev->sb_dirty = 2;
1365 * do not write anything to disk if using
1366 * nonpersistent superblocks
1368 if (!mddev->persistent) {
1369 mddev->sb_dirty = 0;
1370 spin_unlock(&mddev->write_lock);
1371 wake_up(&mddev->sb_wait);
1374 spin_unlock(&mddev->write_lock);
1377 "md: updating %s RAID superblock on device (in sync %d)\n",
1378 mdname(mddev),mddev->in_sync);
1380 err = bitmap_update_sb(mddev->bitmap);
1381 ITERATE_RDEV(mddev,rdev,tmp) {
1382 char b[BDEVNAME_SIZE];
1383 dprintk(KERN_INFO "md: ");
1385 dprintk("(skipping faulty ");
1387 dprintk("%s ", bdevname(rdev->bdev,b));
1388 if (!rdev->faulty) {
1389 md_super_write(mddev,rdev,
1390 rdev->sb_offset<<1, rdev->sb_size,
1392 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1393 bdevname(rdev->bdev,b),
1394 (unsigned long long)rdev->sb_offset);
1398 if (mddev->level == LEVEL_MULTIPATH)
1399 /* only need to write one superblock... */
1402 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1403 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1405 spin_lock(&mddev->write_lock);
1406 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1407 /* have to write it out again */
1408 spin_unlock(&mddev->write_lock);
1411 mddev->sb_dirty = 0;
1412 spin_unlock(&mddev->write_lock);
1413 wake_up(&mddev->sb_wait);
1418 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1420 * mark the device faulty if:
1422 * - the device is nonexistent (zero size)
1423 * - the device has no valid superblock
1425 * a faulty rdev _never_ has rdev->sb set.
1427 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1429 char b[BDEVNAME_SIZE];
1434 rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1436 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1437 return ERR_PTR(-ENOMEM);
1439 memset(rdev, 0, sizeof(*rdev));
1441 if ((err = alloc_disk_sb(rdev)))
1444 err = lock_rdev(rdev, newdev);
1451 rdev->data_offset = 0;
1452 atomic_set(&rdev->nr_pending, 0);
1454 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1457 "md: %s has zero or unknown size, marking faulty!\n",
1458 bdevname(rdev->bdev,b));
1463 if (super_format >= 0) {
1464 err = super_types[super_format].
1465 load_super(rdev, NULL, super_minor);
1466 if (err == -EINVAL) {
1468 "md: %s has invalid sb, not importing!\n",
1469 bdevname(rdev->bdev,b));
1474 "md: could not read %s's sb, not importing!\n",
1475 bdevname(rdev->bdev,b));
1479 INIT_LIST_HEAD(&rdev->same_set);
1484 if (rdev->sb_page) {
1490 return ERR_PTR(err);
1494 * Check a full RAID array for plausibility
1498 static void analyze_sbs(mddev_t * mddev)
1501 struct list_head *tmp;
1502 mdk_rdev_t *rdev, *freshest;
1503 char b[BDEVNAME_SIZE];
1506 ITERATE_RDEV(mddev,rdev,tmp)
1507 switch (super_types[mddev->major_version].
1508 load_super(rdev, freshest, mddev->minor_version)) {
1516 "md: fatal superblock inconsistency in %s"
1517 " -- removing from array\n",
1518 bdevname(rdev->bdev,b));
1519 kick_rdev_from_array(rdev);
1523 super_types[mddev->major_version].
1524 validate_super(mddev, freshest);
1527 ITERATE_RDEV(mddev,rdev,tmp) {
1528 if (rdev != freshest)
1529 if (super_types[mddev->major_version].
1530 validate_super(mddev, rdev)) {
1531 printk(KERN_WARNING "md: kicking non-fresh %s"
1533 bdevname(rdev->bdev,b));
1534 kick_rdev_from_array(rdev);
1537 if (mddev->level == LEVEL_MULTIPATH) {
1538 rdev->desc_nr = i++;
1539 rdev->raid_disk = rdev->desc_nr;
1546 if (mddev->recovery_cp != MaxSector &&
1548 printk(KERN_ERR "md: %s: raid array is not clean"
1549 " -- starting background reconstruction\n",
1554 struct md_sysfs_entry {
1555 struct attribute attr;
1556 ssize_t (*show)(mddev_t *, char *);
1557 ssize_t (*store)(mddev_t *, const char *, size_t);
1561 md_show_level(mddev_t *mddev, char *page)
1563 mdk_personality_t *p = mddev->pers;
1566 if (mddev->level >= 0)
1567 return sprintf(page, "RAID-%d\n", mddev->level);
1569 return sprintf(page, "%s\n", p->name);
1572 static struct md_sysfs_entry md_level = {
1573 .attr = {.name = "level", .mode = S_IRUGO },
1574 .show = md_show_level,
1578 md_show_rdisks(mddev_t *mddev, char *page)
1580 return sprintf(page, "%d\n", mddev->raid_disks);
1583 static struct md_sysfs_entry md_raid_disks = {
1584 .attr = {.name = "raid_disks", .mode = S_IRUGO },
1585 .show = md_show_rdisks,
1588 static struct attribute *md_default_attrs[] = {
1590 &md_raid_disks.attr,
1595 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1597 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1598 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1602 return entry->show(mddev, page);
1606 md_attr_store(struct kobject *kobj, struct attribute *attr,
1607 const char *page, size_t length)
1609 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1610 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1614 return entry->store(mddev, page, length);
1617 static void md_free(struct kobject *ko)
1619 mddev_t *mddev = container_of(ko, mddev_t, kobj);
1623 static struct sysfs_ops md_sysfs_ops = {
1624 .show = md_attr_show,
1625 .store = md_attr_store,
1627 static struct kobj_type md_ktype = {
1629 .sysfs_ops = &md_sysfs_ops,
1630 .default_attrs = md_default_attrs,
1635 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1637 static DECLARE_MUTEX(disks_sem);
1638 mddev_t *mddev = mddev_find(dev);
1639 struct gendisk *disk;
1640 int partitioned = (MAJOR(dev) != MD_MAJOR);
1641 int shift = partitioned ? MdpMinorShift : 0;
1642 int unit = MINOR(dev) >> shift;
1648 if (mddev->gendisk) {
1653 disk = alloc_disk(1 << shift);
1659 disk->major = MAJOR(dev);
1660 disk->first_minor = unit << shift;
1662 sprintf(disk->disk_name, "md_d%d", unit);
1663 sprintf(disk->devfs_name, "md/d%d", unit);
1665 sprintf(disk->disk_name, "md%d", unit);
1666 sprintf(disk->devfs_name, "md/%d", unit);
1668 disk->fops = &md_fops;
1669 disk->private_data = mddev;
1670 disk->queue = mddev->queue;
1672 mddev->gendisk = disk;
1674 mddev->kobj.parent = kobject_get(&disk->kobj);
1675 mddev->kobj.k_name = NULL;
1676 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1677 mddev->kobj.ktype = &md_ktype;
1678 kobject_register(&mddev->kobj);
1682 void md_wakeup_thread(mdk_thread_t *thread);
1684 static void md_safemode_timeout(unsigned long data)
1686 mddev_t *mddev = (mddev_t *) data;
1688 mddev->safemode = 1;
1689 md_wakeup_thread(mddev->thread);
1693 static int do_md_run(mddev_t * mddev)
1697 struct list_head *tmp;
1699 struct gendisk *disk;
1700 char b[BDEVNAME_SIZE];
1702 if (list_empty(&mddev->disks))
1703 /* cannot run an array with no devices.. */
1710 * Analyze all RAID superblock(s)
1712 if (!mddev->raid_disks)
1715 chunk_size = mddev->chunk_size;
1716 pnum = level_to_pers(mddev->level);
1718 if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1721 * 'default chunksize' in the old md code used to
1722 * be PAGE_SIZE, baaad.
1723 * we abort here to be on the safe side. We don't
1724 * want to continue the bad practice.
1727 "no chunksize specified, see 'man raidtab'\n");
1730 if (chunk_size > MAX_CHUNK_SIZE) {
1731 printk(KERN_ERR "too big chunk_size: %d > %d\n",
1732 chunk_size, MAX_CHUNK_SIZE);
1736 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1738 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1739 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1742 if (chunk_size < PAGE_SIZE) {
1743 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1744 chunk_size, PAGE_SIZE);
1748 /* devices must have minimum size of one chunk */
1749 ITERATE_RDEV(mddev,rdev,tmp) {
1752 if (rdev->size < chunk_size / 1024) {
1754 "md: Dev %s smaller than chunk_size:"
1756 bdevname(rdev->bdev,b),
1757 (unsigned long long)rdev->size,
1767 request_module("md-personality-%d", pnum);
1772 * Drop all container device buffers, from now on
1773 * the only valid external interface is through the md
1775 * Also find largest hardsector size
1777 ITERATE_RDEV(mddev,rdev,tmp) {
1780 sync_blockdev(rdev->bdev);
1781 invalidate_bdev(rdev->bdev, 0);
1784 md_probe(mddev->unit, NULL, NULL);
1785 disk = mddev->gendisk;
1789 spin_lock(&pers_lock);
1790 if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1791 spin_unlock(&pers_lock);
1792 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1797 mddev->pers = pers[pnum];
1798 spin_unlock(&pers_lock);
1800 mddev->recovery = 0;
1801 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1803 /* before we start the array running, initialise the bitmap */
1804 err = bitmap_create(mddev);
1806 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1807 mdname(mddev), err);
1809 err = mddev->pers->run(mddev);
1811 printk(KERN_ERR "md: pers->run() failed ...\n");
1812 module_put(mddev->pers->owner);
1814 bitmap_destroy(mddev);
1817 atomic_set(&mddev->writes_pending,0);
1818 mddev->safemode = 0;
1819 mddev->safemode_timer.function = md_safemode_timeout;
1820 mddev->safemode_timer.data = (unsigned long) mddev;
1821 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1824 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1825 md_wakeup_thread(mddev->thread);
1827 if (mddev->sb_dirty)
1828 md_update_sb(mddev);
1830 set_capacity(disk, mddev->array_size<<1);
1832 /* If we call blk_queue_make_request here, it will
1833 * re-initialise max_sectors etc which may have been
1834 * refined inside -> run. So just set the bits we need to set.
1835 * Most initialisation happended when we called
1836 * blk_queue_make_request(..., md_fail_request)
1839 mddev->queue->queuedata = mddev;
1840 mddev->queue->make_request_fn = mddev->pers->make_request;
1846 static int restart_array(mddev_t *mddev)
1848 struct gendisk *disk = mddev->gendisk;
1852 * Complain if it has no devices
1855 if (list_empty(&mddev->disks))
1863 mddev->safemode = 0;
1865 set_disk_ro(disk, 0);
1867 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1870 * Kick recovery or resync if necessary
1872 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1873 md_wakeup_thread(mddev->thread);
1876 printk(KERN_ERR "md: %s has no personality assigned.\n",
1885 static int do_md_stop(mddev_t * mddev, int ro)
1888 struct gendisk *disk = mddev->gendisk;
1891 if (atomic_read(&mddev->active)>2) {
1892 printk("md: %s still in use.\n",mdname(mddev));
1896 if (mddev->sync_thread) {
1897 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1898 md_unregister_thread(mddev->sync_thread);
1899 mddev->sync_thread = NULL;
1902 del_timer_sync(&mddev->safemode_timer);
1904 invalidate_partition(disk, 0);
1912 bitmap_flush(mddev);
1913 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1915 set_disk_ro(disk, 0);
1916 blk_queue_make_request(mddev->queue, md_fail_request);
1917 mddev->pers->stop(mddev);
1918 module_put(mddev->pers->owner);
1923 if (!mddev->in_sync) {
1924 /* mark array as shutdown cleanly */
1926 md_update_sb(mddev);
1929 set_disk_ro(disk, 1);
1932 bitmap_destroy(mddev);
1933 if (mddev->bitmap_file) {
1934 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1935 fput(mddev->bitmap_file);
1936 mddev->bitmap_file = NULL;
1938 mddev->bitmap_offset = 0;
1941 * Free resources if final stop
1944 struct gendisk *disk;
1945 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1947 export_array(mddev);
1949 mddev->array_size = 0;
1950 disk = mddev->gendisk;
1952 set_capacity(disk, 0);
1955 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1962 static void autorun_array(mddev_t *mddev)
1965 struct list_head *tmp;
1968 if (list_empty(&mddev->disks))
1971 printk(KERN_INFO "md: running: ");
1973 ITERATE_RDEV(mddev,rdev,tmp) {
1974 char b[BDEVNAME_SIZE];
1975 printk("<%s>", bdevname(rdev->bdev,b));
1979 err = do_md_run (mddev);
1981 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1982 do_md_stop (mddev, 0);
1987 * lets try to run arrays based on all disks that have arrived
1988 * until now. (those are in pending_raid_disks)
1990 * the method: pick the first pending disk, collect all disks with
1991 * the same UUID, remove all from the pending list and put them into
1992 * the 'same_array' list. Then order this list based on superblock
1993 * update time (freshest comes first), kick out 'old' disks and
1994 * compare superblocks. If everything's fine then run it.
1996 * If "unit" is allocated, then bump its reference count
1998 static void autorun_devices(int part)
2000 struct list_head candidates;
2001 struct list_head *tmp;
2002 mdk_rdev_t *rdev0, *rdev;
2004 char b[BDEVNAME_SIZE];
2006 printk(KERN_INFO "md: autorun ...\n");
2007 while (!list_empty(&pending_raid_disks)) {
2009 rdev0 = list_entry(pending_raid_disks.next,
2010 mdk_rdev_t, same_set);
2012 printk(KERN_INFO "md: considering %s ...\n",
2013 bdevname(rdev0->bdev,b));
2014 INIT_LIST_HEAD(&candidates);
2015 ITERATE_RDEV_PENDING(rdev,tmp)
2016 if (super_90_load(rdev, rdev0, 0) >= 0) {
2017 printk(KERN_INFO "md: adding %s ...\n",
2018 bdevname(rdev->bdev,b));
2019 list_move(&rdev->same_set, &candidates);
2022 * now we have a set of devices, with all of them having
2023 * mostly sane superblocks. It's time to allocate the
2026 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2027 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2028 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2032 dev = MKDEV(mdp_major,
2033 rdev0->preferred_minor << MdpMinorShift);
2035 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2037 md_probe(dev, NULL, NULL);
2038 mddev = mddev_find(dev);
2041 "md: cannot allocate memory for md drive.\n");
2044 if (mddev_lock(mddev))
2045 printk(KERN_WARNING "md: %s locked, cannot run\n",
2047 else if (mddev->raid_disks || mddev->major_version
2048 || !list_empty(&mddev->disks)) {
2050 "md: %s already running, cannot run %s\n",
2051 mdname(mddev), bdevname(rdev0->bdev,b));
2052 mddev_unlock(mddev);
2054 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2055 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2056 list_del_init(&rdev->same_set);
2057 if (bind_rdev_to_array(rdev, mddev))
2060 autorun_array(mddev);
2061 mddev_unlock(mddev);
2063 /* on success, candidates will be empty, on error
2066 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2070 printk(KERN_INFO "md: ... autorun DONE.\n");
2074 * import RAID devices based on one partition
2075 * if possible, the array gets run as well.
2078 static int autostart_array(dev_t startdev)
2080 char b[BDEVNAME_SIZE];
2081 int err = -EINVAL, i;
2082 mdp_super_t *sb = NULL;
2083 mdk_rdev_t *start_rdev = NULL, *rdev;
2085 start_rdev = md_import_device(startdev, 0, 0);
2086 if (IS_ERR(start_rdev))
2090 /* NOTE: this can only work for 0.90.0 superblocks */
2091 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2092 if (sb->major_version != 0 ||
2093 sb->minor_version != 90 ) {
2094 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2095 export_rdev(start_rdev);
2099 if (start_rdev->faulty) {
2101 "md: can not autostart based on faulty %s!\n",
2102 bdevname(start_rdev->bdev,b));
2103 export_rdev(start_rdev);
2106 list_add(&start_rdev->same_set, &pending_raid_disks);
2108 for (i = 0; i < MD_SB_DISKS; i++) {
2109 mdp_disk_t *desc = sb->disks + i;
2110 dev_t dev = MKDEV(desc->major, desc->minor);
2114 if (dev == startdev)
2116 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2118 rdev = md_import_device(dev, 0, 0);
2122 list_add(&rdev->same_set, &pending_raid_disks);
2126 * possibly return codes
2134 static int get_version(void __user * arg)
2138 ver.major = MD_MAJOR_VERSION;
2139 ver.minor = MD_MINOR_VERSION;
2140 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2142 if (copy_to_user(arg, &ver, sizeof(ver)))
2148 static int get_array_info(mddev_t * mddev, void __user * arg)
2150 mdu_array_info_t info;
2151 int nr,working,active,failed,spare;
2153 struct list_head *tmp;
2155 nr=working=active=failed=spare=0;
2156 ITERATE_RDEV(mddev,rdev,tmp) {
2169 info.major_version = mddev->major_version;
2170 info.minor_version = mddev->minor_version;
2171 info.patch_version = MD_PATCHLEVEL_VERSION;
2172 info.ctime = mddev->ctime;
2173 info.level = mddev->level;
2174 info.size = mddev->size;
2176 info.raid_disks = mddev->raid_disks;
2177 info.md_minor = mddev->md_minor;
2178 info.not_persistent= !mddev->persistent;
2180 info.utime = mddev->utime;
2183 info.state = (1<<MD_SB_CLEAN);
2184 if (mddev->bitmap && mddev->bitmap_offset)
2185 info.state = (1<<MD_SB_BITMAP_PRESENT);
2186 info.active_disks = active;
2187 info.working_disks = working;
2188 info.failed_disks = failed;
2189 info.spare_disks = spare;
2191 info.layout = mddev->layout;
2192 info.chunk_size = mddev->chunk_size;
2194 if (copy_to_user(arg, &info, sizeof(info)))
2200 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2202 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2203 char *ptr, *buf = NULL;
2206 file = kmalloc(sizeof(*file), GFP_KERNEL);
2210 /* bitmap disabled, zero the first byte and copy out */
2211 if (!mddev->bitmap || !mddev->bitmap->file) {
2212 file->pathname[0] = '\0';
2216 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2220 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2224 strcpy(file->pathname, ptr);
2228 if (copy_to_user(arg, file, sizeof(*file)))
2236 static int get_disk_info(mddev_t * mddev, void __user * arg)
2238 mdu_disk_info_t info;
2242 if (copy_from_user(&info, arg, sizeof(info)))
2247 rdev = find_rdev_nr(mddev, nr);
2249 info.major = MAJOR(rdev->bdev->bd_dev);
2250 info.minor = MINOR(rdev->bdev->bd_dev);
2251 info.raid_disk = rdev->raid_disk;
2254 info.state |= (1<<MD_DISK_FAULTY);
2255 else if (rdev->in_sync) {
2256 info.state |= (1<<MD_DISK_ACTIVE);
2257 info.state |= (1<<MD_DISK_SYNC);
2259 if (test_bit(WriteMostly, &rdev->flags))
2260 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2262 info.major = info.minor = 0;
2263 info.raid_disk = -1;
2264 info.state = (1<<MD_DISK_REMOVED);
2267 if (copy_to_user(arg, &info, sizeof(info)))
2273 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2275 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2277 dev_t dev = MKDEV(info->major,info->minor);
2279 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2282 if (!mddev->raid_disks) {
2284 /* expecting a device which has a superblock */
2285 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2288 "md: md_import_device returned %ld\n",
2290 return PTR_ERR(rdev);
2292 if (!list_empty(&mddev->disks)) {
2293 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2294 mdk_rdev_t, same_set);
2295 int err = super_types[mddev->major_version]
2296 .load_super(rdev, rdev0, mddev->minor_version);
2299 "md: %s has different UUID to %s\n",
2300 bdevname(rdev->bdev,b),
2301 bdevname(rdev0->bdev,b2));
2306 err = bind_rdev_to_array(rdev, mddev);
2313 * add_new_disk can be used once the array is assembled
2314 * to add "hot spares". They must already have a superblock
2319 if (!mddev->pers->hot_add_disk) {
2321 "%s: personality does not support diskops!\n",
2325 if (mddev->persistent)
2326 rdev = md_import_device(dev, mddev->major_version,
2327 mddev->minor_version);
2329 rdev = md_import_device(dev, -1, -1);
2332 "md: md_import_device returned %ld\n",
2334 return PTR_ERR(rdev);
2336 /* set save_raid_disk if appropriate */
2337 if (!mddev->persistent) {
2338 if (info->state & (1<<MD_DISK_SYNC) &&
2339 info->raid_disk < mddev->raid_disks)
2340 rdev->raid_disk = info->raid_disk;
2342 rdev->raid_disk = -1;
2344 super_types[mddev->major_version].
2345 validate_super(mddev, rdev);
2346 rdev->saved_raid_disk = rdev->raid_disk;
2348 rdev->in_sync = 0; /* just to be sure */
2349 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2350 set_bit(WriteMostly, &rdev->flags);
2352 rdev->raid_disk = -1;
2353 err = bind_rdev_to_array(rdev, mddev);
2357 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2358 md_wakeup_thread(mddev->thread);
2362 /* otherwise, add_new_disk is only allowed
2363 * for major_version==0 superblocks
2365 if (mddev->major_version != 0) {
2366 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2371 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2373 rdev = md_import_device (dev, -1, 0);
2376 "md: error, md_import_device() returned %ld\n",
2378 return PTR_ERR(rdev);
2380 rdev->desc_nr = info->number;
2381 if (info->raid_disk < mddev->raid_disks)
2382 rdev->raid_disk = info->raid_disk;
2384 rdev->raid_disk = -1;
2387 if (rdev->raid_disk < mddev->raid_disks)
2388 rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2392 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2393 set_bit(WriteMostly, &rdev->flags);
2395 err = bind_rdev_to_array(rdev, mddev);
2401 if (!mddev->persistent) {
2402 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2403 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2405 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2406 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2408 if (!mddev->size || (mddev->size > rdev->size))
2409 mddev->size = rdev->size;
2415 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2417 char b[BDEVNAME_SIZE];
2423 rdev = find_rdev(mddev, dev);
2427 if (rdev->raid_disk >= 0)
2430 kick_rdev_from_array(rdev);
2431 md_update_sb(mddev);
2435 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2436 bdevname(rdev->bdev,b), mdname(mddev));
2440 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2442 char b[BDEVNAME_SIZE];
2450 if (mddev->major_version != 0) {
2451 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2452 " version-0 superblocks.\n",
2456 if (!mddev->pers->hot_add_disk) {
2458 "%s: personality does not support diskops!\n",
2463 rdev = md_import_device (dev, -1, 0);
2466 "md: error, md_import_device() returned %ld\n",
2471 if (mddev->persistent)
2472 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2475 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2477 size = calc_dev_size(rdev, mddev->chunk_size);
2480 if (size < mddev->size) {
2482 "%s: disk size %llu blocks < array size %llu\n",
2483 mdname(mddev), (unsigned long long)size,
2484 (unsigned long long)mddev->size);
2491 "md: can not hot-add faulty %s disk to %s!\n",
2492 bdevname(rdev->bdev,b), mdname(mddev));
2498 bind_rdev_to_array(rdev, mddev);
2501 * The rest should better be atomic, we can have disk failures
2502 * noticed in interrupt contexts ...
2505 if (rdev->desc_nr == mddev->max_disks) {
2506 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2509 goto abort_unbind_export;
2512 rdev->raid_disk = -1;
2514 md_update_sb(mddev);
2517 * Kick recovery, maybe this spare has to be added to the
2518 * array immediately.
2520 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2521 md_wakeup_thread(mddev->thread);
2525 abort_unbind_export:
2526 unbind_rdev_from_array(rdev);
2533 /* similar to deny_write_access, but accounts for our holding a reference
2534 * to the file ourselves */
2535 static int deny_bitmap_write_access(struct file * file)
2537 struct inode *inode = file->f_mapping->host;
2539 spin_lock(&inode->i_lock);
2540 if (atomic_read(&inode->i_writecount) > 1) {
2541 spin_unlock(&inode->i_lock);
2544 atomic_set(&inode->i_writecount, -1);
2545 spin_unlock(&inode->i_lock);
2550 static int set_bitmap_file(mddev_t *mddev, int fd)
2555 if (!mddev->pers->quiesce)
2557 if (mddev->recovery || mddev->sync_thread)
2559 /* we should be able to change the bitmap.. */
2565 return -EEXIST; /* cannot add when bitmap is present */
2566 mddev->bitmap_file = fget(fd);
2568 if (mddev->bitmap_file == NULL) {
2569 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2574 err = deny_bitmap_write_access(mddev->bitmap_file);
2576 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2578 fput(mddev->bitmap_file);
2579 mddev->bitmap_file = NULL;
2582 mddev->bitmap_offset = 0; /* file overrides offset */
2583 } else if (mddev->bitmap == NULL)
2584 return -ENOENT; /* cannot remove what isn't there */
2587 mddev->pers->quiesce(mddev, 1);
2589 err = bitmap_create(mddev);
2591 bitmap_destroy(mddev);
2592 mddev->pers->quiesce(mddev, 0);
2593 } else if (fd < 0) {
2594 if (mddev->bitmap_file)
2595 fput(mddev->bitmap_file);
2596 mddev->bitmap_file = NULL;
2603 * set_array_info is used two different ways
2604 * The original usage is when creating a new array.
2605 * In this usage, raid_disks is > 0 and it together with
2606 * level, size, not_persistent,layout,chunksize determine the
2607 * shape of the array.
2608 * This will always create an array with a type-0.90.0 superblock.
2609 * The newer usage is when assembling an array.
2610 * In this case raid_disks will be 0, and the major_version field is
2611 * use to determine which style super-blocks are to be found on the devices.
2612 * The minor and patch _version numbers are also kept incase the
2613 * super_block handler wishes to interpret them.
2615 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2618 if (info->raid_disks == 0) {
2619 /* just setting version number for superblock loading */
2620 if (info->major_version < 0 ||
2621 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2622 super_types[info->major_version].name == NULL) {
2623 /* maybe try to auto-load a module? */
2625 "md: superblock version %d not known\n",
2626 info->major_version);
2629 mddev->major_version = info->major_version;
2630 mddev->minor_version = info->minor_version;
2631 mddev->patch_version = info->patch_version;
2634 mddev->major_version = MD_MAJOR_VERSION;
2635 mddev->minor_version = MD_MINOR_VERSION;
2636 mddev->patch_version = MD_PATCHLEVEL_VERSION;
2637 mddev->ctime = get_seconds();
2639 mddev->level = info->level;
2640 mddev->size = info->size;
2641 mddev->raid_disks = info->raid_disks;
2642 /* don't set md_minor, it is determined by which /dev/md* was
2645 if (info->state & (1<<MD_SB_CLEAN))
2646 mddev->recovery_cp = MaxSector;
2648 mddev->recovery_cp = 0;
2649 mddev->persistent = ! info->not_persistent;
2651 mddev->layout = info->layout;
2652 mddev->chunk_size = info->chunk_size;
2654 mddev->max_disks = MD_SB_DISKS;
2656 mddev->sb_dirty = 1;
2659 * Generate a 128 bit UUID
2661 get_random_bytes(mddev->uuid, 16);
2667 * update_array_info is used to change the configuration of an
2669 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2670 * fields in the info are checked against the array.
2671 * Any differences that cannot be handled will cause an error.
2672 * Normally, only one change can be managed at a time.
2674 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2680 /* calculate expected state,ignoring low bits */
2681 if (mddev->bitmap && mddev->bitmap_offset)
2682 state |= (1 << MD_SB_BITMAP_PRESENT);
2684 if (mddev->major_version != info->major_version ||
2685 mddev->minor_version != info->minor_version ||
2686 /* mddev->patch_version != info->patch_version || */
2687 mddev->ctime != info->ctime ||
2688 mddev->level != info->level ||
2689 /* mddev->layout != info->layout || */
2690 !mddev->persistent != info->not_persistent||
2691 mddev->chunk_size != info->chunk_size ||
2692 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2693 ((state^info->state) & 0xfffffe00)
2696 /* Check there is only one change */
2697 if (mddev->size != info->size) cnt++;
2698 if (mddev->raid_disks != info->raid_disks) cnt++;
2699 if (mddev->layout != info->layout) cnt++;
2700 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2701 if (cnt == 0) return 0;
2702 if (cnt > 1) return -EINVAL;
2704 if (mddev->layout != info->layout) {
2706 * we don't need to do anything at the md level, the
2707 * personality will take care of it all.
2709 if (mddev->pers->reconfig == NULL)
2712 return mddev->pers->reconfig(mddev, info->layout, -1);
2714 if (mddev->size != info->size) {
2716 struct list_head *tmp;
2717 if (mddev->pers->resize == NULL)
2719 /* The "size" is the amount of each device that is used.
2720 * This can only make sense for arrays with redundancy.
2721 * linear and raid0 always use whatever space is available
2722 * We can only consider changing the size if no resync
2723 * or reconstruction is happening, and if the new size
2724 * is acceptable. It must fit before the sb_offset or,
2725 * if that is <data_offset, it must fit before the
2726 * size of each device.
2727 * If size is zero, we find the largest size that fits.
2729 if (mddev->sync_thread)
2731 ITERATE_RDEV(mddev,rdev,tmp) {
2733 int fit = (info->size == 0);
2734 if (rdev->sb_offset > rdev->data_offset)
2735 avail = (rdev->sb_offset*2) - rdev->data_offset;
2737 avail = get_capacity(rdev->bdev->bd_disk)
2738 - rdev->data_offset;
2739 if (fit && (info->size == 0 || info->size > avail/2))
2740 info->size = avail/2;
2741 if (avail < ((sector_t)info->size << 1))
2744 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2746 struct block_device *bdev;
2748 bdev = bdget_disk(mddev->gendisk, 0);
2750 down(&bdev->bd_inode->i_sem);
2751 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2752 up(&bdev->bd_inode->i_sem);
2757 if (mddev->raid_disks != info->raid_disks) {
2758 /* change the number of raid disks */
2759 if (mddev->pers->reshape == NULL)
2761 if (info->raid_disks <= 0 ||
2762 info->raid_disks >= mddev->max_disks)
2764 if (mddev->sync_thread)
2766 rv = mddev->pers->reshape(mddev, info->raid_disks);
2768 struct block_device *bdev;
2770 bdev = bdget_disk(mddev->gendisk, 0);
2772 down(&bdev->bd_inode->i_sem);
2773 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2774 up(&bdev->bd_inode->i_sem);
2779 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2780 if (mddev->pers->quiesce == NULL)
2782 if (mddev->recovery || mddev->sync_thread)
2784 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2785 /* add the bitmap */
2788 if (mddev->default_bitmap_offset == 0)
2790 mddev->bitmap_offset = mddev->default_bitmap_offset;
2791 mddev->pers->quiesce(mddev, 1);
2792 rv = bitmap_create(mddev);
2794 bitmap_destroy(mddev);
2795 mddev->pers->quiesce(mddev, 0);
2797 /* remove the bitmap */
2800 if (mddev->bitmap->file)
2802 mddev->pers->quiesce(mddev, 1);
2803 bitmap_destroy(mddev);
2804 mddev->pers->quiesce(mddev, 0);
2805 mddev->bitmap_offset = 0;
2808 md_update_sb(mddev);
2812 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2816 if (mddev->pers == NULL)
2819 rdev = find_rdev(mddev, dev);
2823 md_error(mddev, rdev);
2827 static int md_ioctl(struct inode *inode, struct file *file,
2828 unsigned int cmd, unsigned long arg)
2831 void __user *argp = (void __user *)arg;
2832 struct hd_geometry __user *loc = argp;
2833 mddev_t *mddev = NULL;
2835 if (!capable(CAP_SYS_ADMIN))
2839 * Commands dealing with the RAID driver but not any
2845 err = get_version(argp);
2848 case PRINT_RAID_DEBUG:
2856 autostart_arrays(arg);
2863 * Commands creating/starting a new array:
2866 mddev = inode->i_bdev->bd_disk->private_data;
2874 if (cmd == START_ARRAY) {
2875 /* START_ARRAY doesn't need to lock the array as autostart_array
2876 * does the locking, and it could even be a different array
2881 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2882 "This will not be supported beyond 2.6\n",
2883 current->comm, current->pid);
2886 err = autostart_array(new_decode_dev(arg));
2888 printk(KERN_WARNING "md: autostart failed!\n");
2894 err = mddev_lock(mddev);
2897 "md: ioctl lock interrupted, reason %d, cmd %d\n",
2904 case SET_ARRAY_INFO:
2906 mdu_array_info_t info;
2908 memset(&info, 0, sizeof(info));
2909 else if (copy_from_user(&info, argp, sizeof(info))) {
2914 err = update_array_info(mddev, &info);
2916 printk(KERN_WARNING "md: couldn't update"
2917 " array info. %d\n", err);
2922 if (!list_empty(&mddev->disks)) {
2924 "md: array %s already has disks!\n",
2929 if (mddev->raid_disks) {
2931 "md: array %s already initialised!\n",
2936 err = set_array_info(mddev, &info);
2938 printk(KERN_WARNING "md: couldn't set"
2939 " array info. %d\n", err);
2949 * Commands querying/configuring an existing array:
2951 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2952 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2953 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2954 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2960 * Commands even a read-only array can execute:
2964 case GET_ARRAY_INFO:
2965 err = get_array_info(mddev, argp);
2968 case GET_BITMAP_FILE:
2969 err = get_bitmap_file(mddev, argp);
2973 err = get_disk_info(mddev, argp);
2976 case RESTART_ARRAY_RW:
2977 err = restart_array(mddev);
2981 err = do_md_stop (mddev, 0);
2985 err = do_md_stop (mddev, 1);
2989 * We have a problem here : there is no easy way to give a CHS
2990 * virtual geometry. We currently pretend that we have a 2 heads
2991 * 4 sectors (with a BIG number of cylinders...). This drives
2992 * dosfs just mad... ;-)
2999 err = put_user (2, (char __user *) &loc->heads);
3002 err = put_user (4, (char __user *) &loc->sectors);
3005 err = put_user(get_capacity(mddev->gendisk)/8,
3006 (short __user *) &loc->cylinders);
3009 err = put_user (get_start_sect(inode->i_bdev),
3010 (long __user *) &loc->start);
3015 * The remaining ioctls are changing the state of the
3016 * superblock, so we do not allow read-only arrays
3028 mdu_disk_info_t info;
3029 if (copy_from_user(&info, argp, sizeof(info)))
3032 err = add_new_disk(mddev, &info);
3036 case HOT_REMOVE_DISK:
3037 err = hot_remove_disk(mddev, new_decode_dev(arg));
3041 err = hot_add_disk(mddev, new_decode_dev(arg));
3044 case SET_DISK_FAULTY:
3045 err = set_disk_faulty(mddev, new_decode_dev(arg));
3049 err = do_md_run (mddev);
3052 case SET_BITMAP_FILE:
3053 err = set_bitmap_file(mddev, (int)arg);
3057 if (_IOC_TYPE(cmd) == MD_MAJOR)
3058 printk(KERN_WARNING "md: %s(pid %d) used"
3059 " obsolete MD ioctl, upgrade your"
3060 " software to use new ictls.\n",
3061 current->comm, current->pid);
3068 mddev_unlock(mddev);
3078 static int md_open(struct inode *inode, struct file *file)
3081 * Succeed if we can lock the mddev, which confirms that
3082 * it isn't being stopped right now.
3084 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3087 if ((err = mddev_lock(mddev)))
3092 mddev_unlock(mddev);
3094 check_disk_change(inode->i_bdev);
3099 static int md_release(struct inode *inode, struct file * file)
3101 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3110 static int md_media_changed(struct gendisk *disk)
3112 mddev_t *mddev = disk->private_data;
3114 return mddev->changed;
3117 static int md_revalidate(struct gendisk *disk)
3119 mddev_t *mddev = disk->private_data;
3124 static struct block_device_operations md_fops =
3126 .owner = THIS_MODULE,
3128 .release = md_release,
3130 .media_changed = md_media_changed,
3131 .revalidate_disk= md_revalidate,
3134 static int md_thread(void * arg)
3136 mdk_thread_t *thread = arg;
3139 * md_thread is a 'system-thread', it's priority should be very
3140 * high. We avoid resource deadlocks individually in each
3141 * raid personality. (RAID5 does preallocation) We also use RR and
3142 * the very same RT priority as kswapd, thus we will never get
3143 * into a priority inversion deadlock.
3145 * we definitely have to have equal or higher priority than
3146 * bdflush, otherwise bdflush will deadlock if there are too
3147 * many dirty RAID5 blocks.
3150 allow_signal(SIGKILL);
3151 complete(thread->event);
3152 while (!kthread_should_stop()) {
3153 void (*run)(mddev_t *);
3155 wait_event_interruptible_timeout(thread->wqueue,
3156 test_bit(THREAD_WAKEUP, &thread->flags)
3157 || kthread_should_stop(),
3161 clear_bit(THREAD_WAKEUP, &thread->flags);
3171 void md_wakeup_thread(mdk_thread_t *thread)
3174 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3175 set_bit(THREAD_WAKEUP, &thread->flags);
3176 wake_up(&thread->wqueue);
3180 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3183 mdk_thread_t *thread;
3184 struct completion event;
3186 thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3190 memset(thread, 0, sizeof(mdk_thread_t));
3191 init_waitqueue_head(&thread->wqueue);
3193 init_completion(&event);
3194 thread->event = &event;
3196 thread->mddev = mddev;
3197 thread->name = name;
3198 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3199 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3200 if (IS_ERR(thread->tsk)) {
3204 wait_for_completion(&event);
3208 void md_unregister_thread(mdk_thread_t *thread)
3210 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3212 kthread_stop(thread->tsk);
3216 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3223 if (!rdev || rdev->faulty)
3226 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3228 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3229 __builtin_return_address(0),__builtin_return_address(1),
3230 __builtin_return_address(2),__builtin_return_address(3));
3232 if (!mddev->pers->error_handler)
3234 mddev->pers->error_handler(mddev,rdev);
3235 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3236 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3237 md_wakeup_thread(mddev->thread);
3240 /* seq_file implementation /proc/mdstat */
3242 static void status_unused(struct seq_file *seq)
3246 struct list_head *tmp;
3248 seq_printf(seq, "unused devices: ");
3250 ITERATE_RDEV_PENDING(rdev,tmp) {
3251 char b[BDEVNAME_SIZE];
3253 seq_printf(seq, "%s ",
3254 bdevname(rdev->bdev,b));
3257 seq_printf(seq, "<none>");
3259 seq_printf(seq, "\n");
3263 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3265 unsigned long max_blocks, resync, res, dt, db, rt;
3267 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3269 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3270 max_blocks = mddev->resync_max_sectors >> 1;
3272 max_blocks = mddev->size;
3275 * Should not happen.
3281 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3283 int i, x = res/50, y = 20-x;
3284 seq_printf(seq, "[");
3285 for (i = 0; i < x; i++)
3286 seq_printf(seq, "=");
3287 seq_printf(seq, ">");
3288 for (i = 0; i < y; i++)
3289 seq_printf(seq, ".");
3290 seq_printf(seq, "] ");
3292 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3293 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3294 "resync" : "recovery"),
3295 res/10, res % 10, resync, max_blocks);
3298 * We do not want to overflow, so the order of operands and
3299 * the * 100 / 100 trick are important. We do a +1 to be
3300 * safe against division by zero. We only estimate anyway.
3302 * dt: time from mark until now
3303 * db: blocks written from mark until now
3304 * rt: remaining time
3306 dt = ((jiffies - mddev->resync_mark) / HZ);
3308 db = resync - (mddev->resync_mark_cnt/2);
3309 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3311 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3313 seq_printf(seq, " speed=%ldK/sec", db/dt);
3316 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3318 struct list_head *tmp;
3328 spin_lock(&all_mddevs_lock);
3329 list_for_each(tmp,&all_mddevs)
3331 mddev = list_entry(tmp, mddev_t, all_mddevs);
3333 spin_unlock(&all_mddevs_lock);
3336 spin_unlock(&all_mddevs_lock);
3338 return (void*)2;/* tail */
3342 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3344 struct list_head *tmp;
3345 mddev_t *next_mddev, *mddev = v;
3351 spin_lock(&all_mddevs_lock);
3353 tmp = all_mddevs.next;
3355 tmp = mddev->all_mddevs.next;
3356 if (tmp != &all_mddevs)
3357 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3359 next_mddev = (void*)2;
3362 spin_unlock(&all_mddevs_lock);
3370 static void md_seq_stop(struct seq_file *seq, void *v)
3374 if (mddev && v != (void*)1 && v != (void*)2)
3378 static int md_seq_show(struct seq_file *seq, void *v)
3382 struct list_head *tmp2;
3385 struct bitmap *bitmap;
3387 if (v == (void*)1) {
3388 seq_printf(seq, "Personalities : ");
3389 spin_lock(&pers_lock);
3390 for (i = 0; i < MAX_PERSONALITY; i++)
3392 seq_printf(seq, "[%s] ", pers[i]->name);
3394 spin_unlock(&pers_lock);
3395 seq_printf(seq, "\n");
3398 if (v == (void*)2) {
3403 if (mddev_lock(mddev)!=0)
3405 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3406 seq_printf(seq, "%s : %sactive", mdname(mddev),
3407 mddev->pers ? "" : "in");
3410 seq_printf(seq, " (read-only)");
3411 seq_printf(seq, " %s", mddev->pers->name);
3415 ITERATE_RDEV(mddev,rdev,tmp2) {
3416 char b[BDEVNAME_SIZE];
3417 seq_printf(seq, " %s[%d]",
3418 bdevname(rdev->bdev,b), rdev->desc_nr);
3419 if (test_bit(WriteMostly, &rdev->flags))
3420 seq_printf(seq, "(W)");
3422 seq_printf(seq, "(F)");
3424 } else if (rdev->raid_disk < 0)
3425 seq_printf(seq, "(S)"); /* spare */
3429 if (!list_empty(&mddev->disks)) {
3431 seq_printf(seq, "\n %llu blocks",
3432 (unsigned long long)mddev->array_size);
3434 seq_printf(seq, "\n %llu blocks",
3435 (unsigned long long)size);
3437 if (mddev->persistent) {
3438 if (mddev->major_version != 0 ||
3439 mddev->minor_version != 90) {
3440 seq_printf(seq," super %d.%d",
3441 mddev->major_version,
3442 mddev->minor_version);
3445 seq_printf(seq, " super non-persistent");
3448 mddev->pers->status (seq, mddev);
3449 seq_printf(seq, "\n ");
3450 if (mddev->curr_resync > 2) {
3451 status_resync (seq, mddev);
3452 seq_printf(seq, "\n ");
3453 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3454 seq_printf(seq, " resync=DELAYED\n ");
3456 seq_printf(seq, "\n ");
3458 if ((bitmap = mddev->bitmap)) {
3459 unsigned long chunk_kb;
3460 unsigned long flags;
3461 spin_lock_irqsave(&bitmap->lock, flags);
3462 chunk_kb = bitmap->chunksize >> 10;
3463 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3465 bitmap->pages - bitmap->missing_pages,
3467 (bitmap->pages - bitmap->missing_pages)
3468 << (PAGE_SHIFT - 10),
3469 chunk_kb ? chunk_kb : bitmap->chunksize,
3470 chunk_kb ? "KB" : "B");
3472 seq_printf(seq, ", file: ");
3473 seq_path(seq, bitmap->file->f_vfsmnt,
3474 bitmap->file->f_dentry," \t\n");
3477 seq_printf(seq, "\n");
3478 spin_unlock_irqrestore(&bitmap->lock, flags);
3481 seq_printf(seq, "\n");
3483 mddev_unlock(mddev);
3488 static struct seq_operations md_seq_ops = {
3489 .start = md_seq_start,
3490 .next = md_seq_next,
3491 .stop = md_seq_stop,
3492 .show = md_seq_show,
3495 static int md_seq_open(struct inode *inode, struct file *file)
3499 error = seq_open(file, &md_seq_ops);
3503 static struct file_operations md_seq_fops = {
3504 .open = md_seq_open,
3506 .llseek = seq_lseek,
3507 .release = seq_release,
3510 int register_md_personality(int pnum, mdk_personality_t *p)
3512 if (pnum >= MAX_PERSONALITY) {
3514 "md: tried to install personality %s as nr %d, but max is %lu\n",
3515 p->name, pnum, MAX_PERSONALITY-1);
3519 spin_lock(&pers_lock);
3521 spin_unlock(&pers_lock);
3526 printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3527 spin_unlock(&pers_lock);
3531 int unregister_md_personality(int pnum)
3533 if (pnum >= MAX_PERSONALITY)
3536 printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3537 spin_lock(&pers_lock);
3539 spin_unlock(&pers_lock);
3543 static int is_mddev_idle(mddev_t *mddev)
3546 struct list_head *tmp;
3548 unsigned long curr_events;
3551 ITERATE_RDEV(mddev,rdev,tmp) {
3552 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3553 curr_events = disk_stat_read(disk, sectors[0]) +
3554 disk_stat_read(disk, sectors[1]) -
3555 atomic_read(&disk->sync_io);
3556 /* Allow some slack between valud of curr_events and last_events,
3557 * as there are some uninteresting races.
3558 * Note: the following is an unsigned comparison.
3560 if ((curr_events - rdev->last_events + 32) > 64) {
3561 rdev->last_events = curr_events;
3568 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3570 /* another "blocks" (512byte) blocks have been synced */
3571 atomic_sub(blocks, &mddev->recovery_active);
3572 wake_up(&mddev->recovery_wait);
3574 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3575 md_wakeup_thread(mddev->thread);
3576 // stop recovery, signal do_sync ....
3581 /* md_write_start(mddev, bi)
3582 * If we need to update some array metadata (e.g. 'active' flag
3583 * in superblock) before writing, schedule a superblock update
3584 * and wait for it to complete.
3586 void md_write_start(mddev_t *mddev, struct bio *bi)
3588 if (bio_data_dir(bi) != WRITE)
3591 atomic_inc(&mddev->writes_pending);
3592 if (mddev->in_sync) {
3593 spin_lock(&mddev->write_lock);
3594 if (mddev->in_sync) {
3596 mddev->sb_dirty = 1;
3597 md_wakeup_thread(mddev->thread);
3599 spin_unlock(&mddev->write_lock);
3601 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3604 void md_write_end(mddev_t *mddev)
3606 if (atomic_dec_and_test(&mddev->writes_pending)) {
3607 if (mddev->safemode == 2)
3608 md_wakeup_thread(mddev->thread);
3610 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3614 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3616 #define SYNC_MARKS 10
3617 #define SYNC_MARK_STEP (3*HZ)
3618 static void md_do_sync(mddev_t *mddev)
3621 unsigned int currspeed = 0,
3623 sector_t max_sectors,j, io_sectors;
3624 unsigned long mark[SYNC_MARKS];
3625 sector_t mark_cnt[SYNC_MARKS];
3627 struct list_head *tmp;
3628 sector_t last_check;
3631 /* just incase thread restarts... */
3632 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3635 /* we overload curr_resync somewhat here.
3636 * 0 == not engaged in resync at all
3637 * 2 == checking that there is no conflict with another sync
3638 * 1 == like 2, but have yielded to allow conflicting resync to
3640 * other == active in resync - this many blocks
3642 * Before starting a resync we must have set curr_resync to
3643 * 2, and then checked that every "conflicting" array has curr_resync
3644 * less than ours. When we find one that is the same or higher
3645 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
3646 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3647 * This will mean we have to start checking from the beginning again.
3652 mddev->curr_resync = 2;
3655 if (signal_pending(current) ||
3656 kthread_should_stop()) {
3657 flush_signals(current);
3658 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3661 ITERATE_MDDEV(mddev2,tmp) {
3662 if (mddev2 == mddev)
3664 if (mddev2->curr_resync &&
3665 match_mddev_units(mddev,mddev2)) {
3667 if (mddev < mddev2 && mddev->curr_resync == 2) {
3668 /* arbitrarily yield */
3669 mddev->curr_resync = 1;
3670 wake_up(&resync_wait);
3672 if (mddev > mddev2 && mddev->curr_resync == 1)
3673 /* no need to wait here, we can wait the next
3674 * time 'round when curr_resync == 2
3677 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3678 if (!signal_pending(current) &&
3679 !kthread_should_stop() &&
3680 mddev2->curr_resync >= mddev->curr_resync) {
3681 printk(KERN_INFO "md: delaying resync of %s"
3682 " until %s has finished resync (they"
3683 " share one or more physical units)\n",
3684 mdname(mddev), mdname(mddev2));
3687 finish_wait(&resync_wait, &wq);
3690 finish_wait(&resync_wait, &wq);
3693 } while (mddev->curr_resync < 2);
3695 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3696 /* resync follows the size requested by the personality,
3697 * which defaults to physical size, but can be virtual size
3699 max_sectors = mddev->resync_max_sectors;
3701 /* recovery follows the physical size of devices */
3702 max_sectors = mddev->size << 1;
3704 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3705 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3706 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3707 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3708 "(but not more than %d KB/sec) for reconstruction.\n",
3709 sysctl_speed_limit_max);
3711 is_mddev_idle(mddev); /* this also initializes IO event counters */
3712 /* we don't use the checkpoint if there's a bitmap */
3713 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3714 j = mddev->recovery_cp;
3718 for (m = 0; m < SYNC_MARKS; m++) {
3720 mark_cnt[m] = io_sectors;
3723 mddev->resync_mark = mark[last_mark];
3724 mddev->resync_mark_cnt = mark_cnt[last_mark];
3727 * Tune reconstruction:
3729 window = 32*(PAGE_SIZE/512);
3730 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3731 window/2,(unsigned long long) max_sectors/2);
3733 atomic_set(&mddev->recovery_active, 0);
3734 init_waitqueue_head(&mddev->recovery_wait);
3739 "md: resuming recovery of %s from checkpoint.\n",
3741 mddev->curr_resync = j;
3744 while (j < max_sectors) {
3748 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3749 currspeed < sysctl_speed_limit_min);
3751 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3755 if (!skipped) { /* actual IO requested */
3756 io_sectors += sectors;
3757 atomic_add(sectors, &mddev->recovery_active);
3761 if (j>1) mddev->curr_resync = j;
3764 if (last_check + window > io_sectors || j == max_sectors)
3767 last_check = io_sectors;
3769 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3770 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3774 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3776 int next = (last_mark+1) % SYNC_MARKS;
3778 mddev->resync_mark = mark[next];
3779 mddev->resync_mark_cnt = mark_cnt[next];
3780 mark[next] = jiffies;
3781 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3786 if (signal_pending(current) || kthread_should_stop()) {
3788 * got a signal, exit.
3791 "md: md_do_sync() got signal ... exiting\n");
3792 flush_signals(current);
3793 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3798 * this loop exits only if either when we are slower than
3799 * the 'hard' speed limit, or the system was IO-idle for
3801 * the system might be non-idle CPU-wise, but we only care
3802 * about not overloading the IO subsystem. (things like an
3803 * e2fsck being done on the RAID array should execute fast)
3805 mddev->queue->unplug_fn(mddev->queue);
3808 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3809 /((jiffies-mddev->resync_mark)/HZ +1) +1;
3811 if (currspeed > sysctl_speed_limit_min) {
3812 if ((currspeed > sysctl_speed_limit_max) ||
3813 !is_mddev_idle(mddev)) {
3814 msleep_interruptible(250);
3819 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3821 * this also signals 'finished resyncing' to md_stop
3824 mddev->queue->unplug_fn(mddev->queue);
3826 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3828 /* tell personality that we are finished */
3829 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3831 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3832 mddev->curr_resync > 2 &&
3833 mddev->curr_resync >= mddev->recovery_cp) {
3834 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3836 "md: checkpointing recovery of %s.\n",
3838 mddev->recovery_cp = mddev->curr_resync;
3840 mddev->recovery_cp = MaxSector;
3844 mddev->curr_resync = 0;
3845 wake_up(&resync_wait);
3846 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3847 md_wakeup_thread(mddev->thread);
3852 * This routine is regularly called by all per-raid-array threads to
3853 * deal with generic issues like resync and super-block update.
3854 * Raid personalities that don't have a thread (linear/raid0) do not
3855 * need this as they never do any recovery or update the superblock.
3857 * It does not do any resync itself, but rather "forks" off other threads
3858 * to do that as needed.
3859 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3860 * "->recovery" and create a thread at ->sync_thread.
3861 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3862 * and wakeups up this thread which will reap the thread and finish up.
3863 * This thread also removes any faulty devices (with nr_pending == 0).
3865 * The overall approach is:
3866 * 1/ if the superblock needs updating, update it.
3867 * 2/ If a recovery thread is running, don't do anything else.
3868 * 3/ If recovery has finished, clean up, possibly marking spares active.
3869 * 4/ If there are any faulty devices, remove them.
3870 * 5/ If array is degraded, try to add spares devices
3871 * 6/ If array has spares or is not in-sync, start a resync thread.
3873 void md_check_recovery(mddev_t *mddev)
3876 struct list_head *rtmp;
3880 bitmap_daemon_work(mddev->bitmap);
3885 if (signal_pending(current)) {
3886 if (mddev->pers->sync_request) {
3887 printk(KERN_INFO "md: %s in immediate safe mode\n",
3889 mddev->safemode = 2;
3891 flush_signals(current);
3896 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3897 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3898 (mddev->safemode == 1) ||
3899 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3900 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3904 if (mddev_trylock(mddev)==0) {
3907 spin_lock(&mddev->write_lock);
3908 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3909 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3911 mddev->sb_dirty = 1;
3913 if (mddev->safemode == 1)
3914 mddev->safemode = 0;
3915 spin_unlock(&mddev->write_lock);
3917 if (mddev->sb_dirty)
3918 md_update_sb(mddev);
3921 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3922 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3923 /* resync/recovery still happening */
3924 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3927 if (mddev->sync_thread) {
3928 /* resync has finished, collect result */
3929 md_unregister_thread(mddev->sync_thread);
3930 mddev->sync_thread = NULL;
3931 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3932 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3934 /* activate any spares */
3935 mddev->pers->spare_active(mddev);
3937 md_update_sb(mddev);
3939 /* if array is no-longer degraded, then any saved_raid_disk
3940 * information must be scrapped
3942 if (!mddev->degraded)
3943 ITERATE_RDEV(mddev,rdev,rtmp)
3944 rdev->saved_raid_disk = -1;
3946 mddev->recovery = 0;
3947 /* flag recovery needed just to double check */
3948 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3951 if (mddev->recovery)
3952 /* probably just the RECOVERY_NEEDED flag */
3953 mddev->recovery = 0;
3955 /* no recovery is running.
3956 * remove any failed drives, then
3957 * add spares if possible.
3958 * Spare are also removed and re-added, to allow
3959 * the personality to fail the re-add.
3961 ITERATE_RDEV(mddev,rdev,rtmp)
3962 if (rdev->raid_disk >= 0 &&
3963 (rdev->faulty || ! rdev->in_sync) &&
3964 atomic_read(&rdev->nr_pending)==0) {
3965 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3966 rdev->raid_disk = -1;
3969 if (mddev->degraded) {
3970 ITERATE_RDEV(mddev,rdev,rtmp)
3971 if (rdev->raid_disk < 0
3973 if (mddev->pers->hot_add_disk(mddev,rdev))
3980 if (!spares && (mddev->recovery_cp == MaxSector )) {
3981 /* nothing we can do ... */
3984 if (mddev->pers->sync_request) {
3985 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3987 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3988 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3989 /* We are adding a device or devices to an array
3990 * which has the bitmap stored on all devices.
3991 * So make sure all bitmap pages get written
3993 bitmap_write_all(mddev->bitmap);
3995 mddev->sync_thread = md_register_thread(md_do_sync,
3998 if (!mddev->sync_thread) {
3999 printk(KERN_ERR "%s: could not start resync"
4002 /* leave the spares where they are, it shouldn't hurt */
4003 mddev->recovery = 0;
4005 md_wakeup_thread(mddev->sync_thread);
4009 mddev_unlock(mddev);
4013 static int md_notify_reboot(struct notifier_block *this,
4014 unsigned long code, void *x)
4016 struct list_head *tmp;
4019 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4021 printk(KERN_INFO "md: stopping all md devices.\n");
4023 ITERATE_MDDEV(mddev,tmp)
4024 if (mddev_trylock(mddev)==0)
4025 do_md_stop (mddev, 1);
4027 * certain more exotic SCSI devices are known to be
4028 * volatile wrt too early system reboots. While the
4029 * right place to handle this issue is the given
4030 * driver, we do want to have a safe RAID driver ...
4037 static struct notifier_block md_notifier = {
4038 .notifier_call = md_notify_reboot,
4040 .priority = INT_MAX, /* before any real devices */
4043 static void md_geninit(void)
4045 struct proc_dir_entry *p;
4047 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4049 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4051 p->proc_fops = &md_seq_fops;
4054 static int __init md_init(void)
4058 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4059 " MD_SB_DISKS=%d\n",
4060 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4061 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4062 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
4065 if (register_blkdev(MAJOR_NR, "md"))
4067 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4068 unregister_blkdev(MAJOR_NR, "md");
4072 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4073 md_probe, NULL, NULL);
4074 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4075 md_probe, NULL, NULL);
4077 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4078 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4079 S_IFBLK|S_IRUSR|S_IWUSR,
4082 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4083 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4084 S_IFBLK|S_IRUSR|S_IWUSR,
4088 register_reboot_notifier(&md_notifier);
4089 raid_table_header = register_sysctl_table(raid_root_table, 1);
4099 * Searches all registered partitions for autorun RAID arrays
4102 static dev_t detected_devices[128];
4105 void md_autodetect_dev(dev_t dev)
4107 if (dev_cnt >= 0 && dev_cnt < 127)
4108 detected_devices[dev_cnt++] = dev;
4112 static void autostart_arrays(int part)
4117 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4119 for (i = 0; i < dev_cnt; i++) {
4120 dev_t dev = detected_devices[i];
4122 rdev = md_import_device(dev,0, 0);
4130 list_add(&rdev->same_set, &pending_raid_disks);
4134 autorun_devices(part);
4139 static __exit void md_exit(void)
4142 struct list_head *tmp;
4144 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4145 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4146 for (i=0; i < MAX_MD_DEVS; i++)
4147 devfs_remove("md/%d", i);
4148 for (i=0; i < MAX_MD_DEVS; i++)
4149 devfs_remove("md/d%d", i);
4153 unregister_blkdev(MAJOR_NR,"md");
4154 unregister_blkdev(mdp_major, "mdp");
4155 unregister_reboot_notifier(&md_notifier);
4156 unregister_sysctl_table(raid_table_header);
4157 remove_proc_entry("mdstat", NULL);
4158 ITERATE_MDDEV(mddev,tmp) {
4159 struct gendisk *disk = mddev->gendisk;
4162 export_array(mddev);
4165 mddev->gendisk = NULL;
4170 module_init(md_init)
4171 module_exit(md_exit)
4173 EXPORT_SYMBOL(register_md_personality);
4174 EXPORT_SYMBOL(unregister_md_personality);
4175 EXPORT_SYMBOL(md_error);
4176 EXPORT_SYMBOL(md_done_sync);
4177 EXPORT_SYMBOL(md_write_start);
4178 EXPORT_SYMBOL(md_write_end);
4179 EXPORT_SYMBOL(md_register_thread);
4180 EXPORT_SYMBOL(md_unregister_thread);
4181 EXPORT_SYMBOL(md_wakeup_thread);
4182 EXPORT_SYMBOL(md_print_devices);
4183 EXPORT_SYMBOL(md_check_recovery);
4184 MODULE_LICENSE("GPL");
4186 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);