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>
45 #include <linux/poll.h>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part);
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
86 static int sysctl_speed_limit_min = 1000;
87 static int sysctl_speed_limit_max = 200000;
89 static struct ctl_table_header *raid_table_header;
91 static ctl_table raid_table[] = {
93 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
94 .procname = "speed_limit_min",
95 .data = &sysctl_speed_limit_min,
96 .maxlen = sizeof(int),
98 .proc_handler = &proc_dointvec,
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
102 .procname = "speed_limit_max",
103 .data = &sysctl_speed_limit_max,
104 .maxlen = sizeof(int),
106 .proc_handler = &proc_dointvec,
111 static ctl_table raid_dir_table[] = {
113 .ctl_name = DEV_RAID,
122 static ctl_table raid_root_table[] = {
128 .child = raid_dir_table,
133 static struct block_device_operations md_fops;
135 static int start_readonly;
138 * We have a system wide 'event count' that is incremented
139 * on any 'interesting' event, and readers of /proc/mdstat
140 * can use 'poll' or 'select' to find out when the event
144 * start array, stop array, error, add device, remove device,
145 * start build, activate spare
147 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
148 static atomic_t md_event_count;
149 static void md_new_event(mddev_t *mddev)
151 atomic_inc(&md_event_count);
152 wake_up(&md_event_waiters);
156 * Enables to iterate over all existing md arrays
157 * all_mddevs_lock protects this list.
159 static LIST_HEAD(all_mddevs);
160 static DEFINE_SPINLOCK(all_mddevs_lock);
164 * iterates through all used mddevs in the system.
165 * We take care to grab the all_mddevs_lock whenever navigating
166 * the list, and to always hold a refcount when unlocked.
167 * Any code which breaks out of this loop while own
168 * a reference to the current mddev and must mddev_put it.
170 #define ITERATE_MDDEV(mddev,tmp) \
172 for (({ spin_lock(&all_mddevs_lock); \
173 tmp = all_mddevs.next; \
175 ({ if (tmp != &all_mddevs) \
176 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
177 spin_unlock(&all_mddevs_lock); \
178 if (mddev) mddev_put(mddev); \
179 mddev = list_entry(tmp, mddev_t, all_mddevs); \
180 tmp != &all_mddevs;}); \
181 ({ spin_lock(&all_mddevs_lock); \
186 static int md_fail_request (request_queue_t *q, struct bio *bio)
188 bio_io_error(bio, bio->bi_size);
192 static inline mddev_t *mddev_get(mddev_t *mddev)
194 atomic_inc(&mddev->active);
198 static void mddev_put(mddev_t *mddev)
200 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
202 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
203 list_del(&mddev->all_mddevs);
204 blk_put_queue(mddev->queue);
205 kobject_unregister(&mddev->kobj);
207 spin_unlock(&all_mddevs_lock);
210 static mddev_t * mddev_find(dev_t unit)
212 mddev_t *mddev, *new = NULL;
215 spin_lock(&all_mddevs_lock);
216 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
217 if (mddev->unit == unit) {
219 spin_unlock(&all_mddevs_lock);
225 list_add(&new->all_mddevs, &all_mddevs);
226 spin_unlock(&all_mddevs_lock);
229 spin_unlock(&all_mddevs_lock);
231 new = kzalloc(sizeof(*new), GFP_KERNEL);
236 if (MAJOR(unit) == MD_MAJOR)
237 new->md_minor = MINOR(unit);
239 new->md_minor = MINOR(unit) >> MdpMinorShift;
241 init_MUTEX(&new->reconfig_sem);
242 INIT_LIST_HEAD(&new->disks);
243 INIT_LIST_HEAD(&new->all_mddevs);
244 init_timer(&new->safemode_timer);
245 atomic_set(&new->active, 1);
246 spin_lock_init(&new->write_lock);
247 init_waitqueue_head(&new->sb_wait);
249 new->queue = blk_alloc_queue(GFP_KERNEL);
255 blk_queue_make_request(new->queue, md_fail_request);
260 static inline int mddev_lock(mddev_t * mddev)
262 return down_interruptible(&mddev->reconfig_sem);
265 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
267 down(&mddev->reconfig_sem);
270 static inline int mddev_trylock(mddev_t * mddev)
272 return down_trylock(&mddev->reconfig_sem);
275 static inline void mddev_unlock(mddev_t * mddev)
277 up(&mddev->reconfig_sem);
279 md_wakeup_thread(mddev->thread);
282 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
285 struct list_head *tmp;
287 ITERATE_RDEV(mddev,rdev,tmp) {
288 if (rdev->desc_nr == nr)
294 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
296 struct list_head *tmp;
299 ITERATE_RDEV(mddev,rdev,tmp) {
300 if (rdev->bdev->bd_dev == dev)
306 static struct mdk_personality *find_pers(int level, char *clevel)
308 struct mdk_personality *pers;
309 list_for_each_entry(pers, &pers_list, list) {
310 if (level != LEVEL_NONE && pers->level == level)
312 if (strcmp(pers->name, clevel)==0)
318 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
320 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
321 return MD_NEW_SIZE_BLOCKS(size);
324 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
328 size = rdev->sb_offset;
331 size &= ~((sector_t)chunk_size/1024 - 1);
335 static int alloc_disk_sb(mdk_rdev_t * rdev)
340 rdev->sb_page = alloc_page(GFP_KERNEL);
341 if (!rdev->sb_page) {
342 printk(KERN_ALERT "md: out of memory.\n");
349 static void free_disk_sb(mdk_rdev_t * rdev)
352 put_page(rdev->sb_page);
354 rdev->sb_page = NULL;
361 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
363 mdk_rdev_t *rdev = bio->bi_private;
364 mddev_t *mddev = rdev->mddev;
368 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
369 md_error(mddev, rdev);
371 if (atomic_dec_and_test(&mddev->pending_writes))
372 wake_up(&mddev->sb_wait);
377 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
379 struct bio *bio2 = bio->bi_private;
380 mdk_rdev_t *rdev = bio2->bi_private;
381 mddev_t *mddev = rdev->mddev;
385 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
386 error == -EOPNOTSUPP) {
388 /* barriers don't appear to be supported :-( */
389 set_bit(BarriersNotsupp, &rdev->flags);
390 mddev->barriers_work = 0;
391 spin_lock_irqsave(&mddev->write_lock, flags);
392 bio2->bi_next = mddev->biolist;
393 mddev->biolist = bio2;
394 spin_unlock_irqrestore(&mddev->write_lock, flags);
395 wake_up(&mddev->sb_wait);
400 bio->bi_private = rdev;
401 return super_written(bio, bytes_done, error);
404 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
405 sector_t sector, int size, struct page *page)
407 /* write first size bytes of page to sector of rdev
408 * Increment mddev->pending_writes before returning
409 * and decrement it on completion, waking up sb_wait
410 * if zero is reached.
411 * If an error occurred, call md_error
413 * As we might need to resubmit the request if BIO_RW_BARRIER
414 * causes ENOTSUPP, we allocate a spare bio...
416 struct bio *bio = bio_alloc(GFP_NOIO, 1);
417 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
419 bio->bi_bdev = rdev->bdev;
420 bio->bi_sector = sector;
421 bio_add_page(bio, page, size, 0);
422 bio->bi_private = rdev;
423 bio->bi_end_io = super_written;
426 atomic_inc(&mddev->pending_writes);
427 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
429 rw |= (1<<BIO_RW_BARRIER);
430 rbio = bio_clone(bio, GFP_NOIO);
431 rbio->bi_private = bio;
432 rbio->bi_end_io = super_written_barrier;
433 submit_bio(rw, rbio);
438 void md_super_wait(mddev_t *mddev)
440 /* wait for all superblock writes that were scheduled to complete.
441 * if any had to be retried (due to BARRIER problems), retry them
445 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
446 if (atomic_read(&mddev->pending_writes)==0)
448 while (mddev->biolist) {
450 spin_lock_irq(&mddev->write_lock);
451 bio = mddev->biolist;
452 mddev->biolist = bio->bi_next ;
454 spin_unlock_irq(&mddev->write_lock);
455 submit_bio(bio->bi_rw, bio);
459 finish_wait(&mddev->sb_wait, &wq);
462 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
467 complete((struct completion*)bio->bi_private);
471 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
472 struct page *page, int rw)
474 struct bio *bio = bio_alloc(GFP_NOIO, 1);
475 struct completion event;
478 rw |= (1 << BIO_RW_SYNC);
481 bio->bi_sector = sector;
482 bio_add_page(bio, page, size, 0);
483 init_completion(&event);
484 bio->bi_private = &event;
485 bio->bi_end_io = bi_complete;
487 wait_for_completion(&event);
489 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
493 EXPORT_SYMBOL_GPL(sync_page_io);
495 static int read_disk_sb(mdk_rdev_t * rdev, int size)
497 char b[BDEVNAME_SIZE];
498 if (!rdev->sb_page) {
506 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
512 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
513 bdevname(rdev->bdev,b));
517 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
519 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
520 (sb1->set_uuid1 == sb2->set_uuid1) &&
521 (sb1->set_uuid2 == sb2->set_uuid2) &&
522 (sb1->set_uuid3 == sb2->set_uuid3))
530 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
533 mdp_super_t *tmp1, *tmp2;
535 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
536 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
538 if (!tmp1 || !tmp2) {
540 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
548 * nr_disks is not constant
553 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
564 static unsigned int calc_sb_csum(mdp_super_t * sb)
566 unsigned int disk_csum, csum;
568 disk_csum = sb->sb_csum;
570 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
571 sb->sb_csum = disk_csum;
577 * Handle superblock details.
578 * We want to be able to handle multiple superblock formats
579 * so we have a common interface to them all, and an array of
580 * different handlers.
581 * We rely on user-space to write the initial superblock, and support
582 * reading and updating of superblocks.
583 * Interface methods are:
584 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
585 * loads and validates a superblock on dev.
586 * if refdev != NULL, compare superblocks on both devices
588 * 0 - dev has a superblock that is compatible with refdev
589 * 1 - dev has a superblock that is compatible and newer than refdev
590 * so dev should be used as the refdev in future
591 * -EINVAL superblock incompatible or invalid
592 * -othererror e.g. -EIO
594 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
595 * Verify that dev is acceptable into mddev.
596 * The first time, mddev->raid_disks will be 0, and data from
597 * dev should be merged in. Subsequent calls check that dev
598 * is new enough. Return 0 or -EINVAL
600 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
601 * Update the superblock for rdev with data in mddev
602 * This does not write to disc.
608 struct module *owner;
609 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
610 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
611 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
615 * load_super for 0.90.0
617 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
619 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
625 * Calculate the position of the superblock,
626 * it's at the end of the disk.
628 * It also happens to be a multiple of 4Kb.
630 sb_offset = calc_dev_sboffset(rdev->bdev);
631 rdev->sb_offset = sb_offset;
633 ret = read_disk_sb(rdev, MD_SB_BYTES);
638 bdevname(rdev->bdev, b);
639 sb = (mdp_super_t*)page_address(rdev->sb_page);
641 if (sb->md_magic != MD_SB_MAGIC) {
642 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
647 if (sb->major_version != 0 ||
648 sb->minor_version != 90) {
649 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
650 sb->major_version, sb->minor_version,
655 if (sb->raid_disks <= 0)
658 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
659 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
664 rdev->preferred_minor = sb->md_minor;
665 rdev->data_offset = 0;
666 rdev->sb_size = MD_SB_BYTES;
668 if (sb->level == LEVEL_MULTIPATH)
671 rdev->desc_nr = sb->this_disk.number;
677 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
678 if (!uuid_equal(refsb, sb)) {
679 printk(KERN_WARNING "md: %s has different UUID to %s\n",
680 b, bdevname(refdev->bdev,b2));
683 if (!sb_equal(refsb, sb)) {
684 printk(KERN_WARNING "md: %s has same UUID"
685 " but different superblock to %s\n",
686 b, bdevname(refdev->bdev, b2));
690 ev2 = md_event(refsb);
696 rdev->size = calc_dev_size(rdev, sb->chunk_size);
703 * validate_super for 0.90.0
705 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
708 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
710 rdev->raid_disk = -1;
712 if (mddev->raid_disks == 0) {
713 mddev->major_version = 0;
714 mddev->minor_version = sb->minor_version;
715 mddev->patch_version = sb->patch_version;
716 mddev->persistent = ! sb->not_persistent;
717 mddev->chunk_size = sb->chunk_size;
718 mddev->ctime = sb->ctime;
719 mddev->utime = sb->utime;
720 mddev->level = sb->level;
721 mddev->clevel[0] = 0;
722 mddev->layout = sb->layout;
723 mddev->raid_disks = sb->raid_disks;
724 mddev->size = sb->size;
725 mddev->events = md_event(sb);
726 mddev->bitmap_offset = 0;
727 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
729 if (sb->state & (1<<MD_SB_CLEAN))
730 mddev->recovery_cp = MaxSector;
732 if (sb->events_hi == sb->cp_events_hi &&
733 sb->events_lo == sb->cp_events_lo) {
734 mddev->recovery_cp = sb->recovery_cp;
736 mddev->recovery_cp = 0;
739 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
740 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
741 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
742 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
744 mddev->max_disks = MD_SB_DISKS;
746 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
747 mddev->bitmap_file == NULL) {
748 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
749 && mddev->level != 10) {
750 /* FIXME use a better test */
751 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
754 mddev->bitmap_offset = mddev->default_bitmap_offset;
757 } else if (mddev->pers == NULL) {
758 /* Insist on good event counter while assembling */
759 __u64 ev1 = md_event(sb);
761 if (ev1 < mddev->events)
763 } else if (mddev->bitmap) {
764 /* if adding to array with a bitmap, then we can accept an
765 * older device ... but not too old.
767 __u64 ev1 = md_event(sb);
768 if (ev1 < mddev->bitmap->events_cleared)
770 } else /* just a hot-add of a new device, leave raid_disk at -1 */
773 if (mddev->level != LEVEL_MULTIPATH) {
774 desc = sb->disks + rdev->desc_nr;
776 if (desc->state & (1<<MD_DISK_FAULTY))
777 set_bit(Faulty, &rdev->flags);
778 else if (desc->state & (1<<MD_DISK_SYNC) &&
779 desc->raid_disk < mddev->raid_disks) {
780 set_bit(In_sync, &rdev->flags);
781 rdev->raid_disk = desc->raid_disk;
783 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
784 set_bit(WriteMostly, &rdev->flags);
785 } else /* MULTIPATH are always insync */
786 set_bit(In_sync, &rdev->flags);
791 * sync_super for 0.90.0
793 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
796 struct list_head *tmp;
798 int next_spare = mddev->raid_disks;
801 /* make rdev->sb match mddev data..
804 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
805 * 3/ any empty disks < next_spare become removed
807 * disks[0] gets initialised to REMOVED because
808 * we cannot be sure from other fields if it has
809 * been initialised or not.
812 int active=0, working=0,failed=0,spare=0,nr_disks=0;
814 rdev->sb_size = MD_SB_BYTES;
816 sb = (mdp_super_t*)page_address(rdev->sb_page);
818 memset(sb, 0, sizeof(*sb));
820 sb->md_magic = MD_SB_MAGIC;
821 sb->major_version = mddev->major_version;
822 sb->minor_version = mddev->minor_version;
823 sb->patch_version = mddev->patch_version;
824 sb->gvalid_words = 0; /* ignored */
825 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
826 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
827 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
828 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
830 sb->ctime = mddev->ctime;
831 sb->level = mddev->level;
832 sb->size = mddev->size;
833 sb->raid_disks = mddev->raid_disks;
834 sb->md_minor = mddev->md_minor;
835 sb->not_persistent = !mddev->persistent;
836 sb->utime = mddev->utime;
838 sb->events_hi = (mddev->events>>32);
839 sb->events_lo = (u32)mddev->events;
843 sb->recovery_cp = mddev->recovery_cp;
844 sb->cp_events_hi = (mddev->events>>32);
845 sb->cp_events_lo = (u32)mddev->events;
846 if (mddev->recovery_cp == MaxSector)
847 sb->state = (1<< MD_SB_CLEAN);
851 sb->layout = mddev->layout;
852 sb->chunk_size = mddev->chunk_size;
854 if (mddev->bitmap && mddev->bitmap_file == NULL)
855 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
857 sb->disks[0].state = (1<<MD_DISK_REMOVED);
858 ITERATE_RDEV(mddev,rdev2,tmp) {
861 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
862 && !test_bit(Faulty, &rdev2->flags))
863 desc_nr = rdev2->raid_disk;
865 desc_nr = next_spare++;
866 rdev2->desc_nr = desc_nr;
867 d = &sb->disks[rdev2->desc_nr];
869 d->number = rdev2->desc_nr;
870 d->major = MAJOR(rdev2->bdev->bd_dev);
871 d->minor = MINOR(rdev2->bdev->bd_dev);
872 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
873 && !test_bit(Faulty, &rdev2->flags))
874 d->raid_disk = rdev2->raid_disk;
876 d->raid_disk = rdev2->desc_nr; /* compatibility */
877 if (test_bit(Faulty, &rdev2->flags)) {
878 d->state = (1<<MD_DISK_FAULTY);
880 } else if (test_bit(In_sync, &rdev2->flags)) {
881 d->state = (1<<MD_DISK_ACTIVE);
882 d->state |= (1<<MD_DISK_SYNC);
890 if (test_bit(WriteMostly, &rdev2->flags))
891 d->state |= (1<<MD_DISK_WRITEMOSTLY);
893 /* now set the "removed" and "faulty" bits on any missing devices */
894 for (i=0 ; i < mddev->raid_disks ; i++) {
895 mdp_disk_t *d = &sb->disks[i];
896 if (d->state == 0 && d->number == 0) {
899 d->state = (1<<MD_DISK_REMOVED);
900 d->state |= (1<<MD_DISK_FAULTY);
904 sb->nr_disks = nr_disks;
905 sb->active_disks = active;
906 sb->working_disks = working;
907 sb->failed_disks = failed;
908 sb->spare_disks = spare;
910 sb->this_disk = sb->disks[rdev->desc_nr];
911 sb->sb_csum = calc_sb_csum(sb);
915 * version 1 superblock
918 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
920 unsigned int disk_csum, csum;
921 unsigned long long newcsum;
922 int size = 256 + le32_to_cpu(sb->max_dev)*2;
923 unsigned int *isuper = (unsigned int*)sb;
926 disk_csum = sb->sb_csum;
929 for (i=0; size>=4; size -= 4 )
930 newcsum += le32_to_cpu(*isuper++);
933 newcsum += le16_to_cpu(*(unsigned short*) isuper);
935 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
936 sb->sb_csum = disk_csum;
937 return cpu_to_le32(csum);
940 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
942 struct mdp_superblock_1 *sb;
945 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
949 * Calculate the position of the superblock.
950 * It is always aligned to a 4K boundary and
951 * depeding on minor_version, it can be:
952 * 0: At least 8K, but less than 12K, from end of device
953 * 1: At start of device
954 * 2: 4K from start of device.
956 switch(minor_version) {
958 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
960 sb_offset &= ~(sector_t)(4*2-1);
961 /* convert from sectors to K */
973 rdev->sb_offset = sb_offset;
975 /* superblock is rarely larger than 1K, but it can be larger,
976 * and it is safe to read 4k, so we do that
978 ret = read_disk_sb(rdev, 4096);
982 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
984 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
985 sb->major_version != cpu_to_le32(1) ||
986 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
987 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
988 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
991 if (calc_sb_1_csum(sb) != sb->sb_csum) {
992 printk("md: invalid superblock checksum on %s\n",
993 bdevname(rdev->bdev,b));
996 if (le64_to_cpu(sb->data_size) < 10) {
997 printk("md: data_size too small on %s\n",
998 bdevname(rdev->bdev,b));
1001 rdev->preferred_minor = 0xffff;
1002 rdev->data_offset = le64_to_cpu(sb->data_offset);
1004 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1005 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1006 if (rdev->sb_size & bmask)
1007 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1013 struct mdp_superblock_1 *refsb =
1014 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1016 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1017 sb->level != refsb->level ||
1018 sb->layout != refsb->layout ||
1019 sb->chunksize != refsb->chunksize) {
1020 printk(KERN_WARNING "md: %s has strangely different"
1021 " superblock to %s\n",
1022 bdevname(rdev->bdev,b),
1023 bdevname(refdev->bdev,b2));
1026 ev1 = le64_to_cpu(sb->events);
1027 ev2 = le64_to_cpu(refsb->events);
1033 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1035 rdev->size = rdev->sb_offset;
1036 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1038 rdev->size = le64_to_cpu(sb->data_size)/2;
1039 if (le32_to_cpu(sb->chunksize))
1040 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1044 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1046 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1048 rdev->raid_disk = -1;
1050 if (mddev->raid_disks == 0) {
1051 mddev->major_version = 1;
1052 mddev->patch_version = 0;
1053 mddev->persistent = 1;
1054 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1055 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1056 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1057 mddev->level = le32_to_cpu(sb->level);
1058 mddev->clevel[0] = 0;
1059 mddev->layout = le32_to_cpu(sb->layout);
1060 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1061 mddev->size = le64_to_cpu(sb->size)/2;
1062 mddev->events = le64_to_cpu(sb->events);
1063 mddev->bitmap_offset = 0;
1064 mddev->default_bitmap_offset = 1024;
1066 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1067 memcpy(mddev->uuid, sb->set_uuid, 16);
1069 mddev->max_disks = (4096-256)/2;
1071 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1072 mddev->bitmap_file == NULL ) {
1073 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1074 && mddev->level != 10) {
1075 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1078 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1080 } else if (mddev->pers == NULL) {
1081 /* Insist of good event counter while assembling */
1082 __u64 ev1 = le64_to_cpu(sb->events);
1084 if (ev1 < mddev->events)
1086 } else if (mddev->bitmap) {
1087 /* If adding to array with a bitmap, then we can accept an
1088 * older device, but not too old.
1090 __u64 ev1 = le64_to_cpu(sb->events);
1091 if (ev1 < mddev->bitmap->events_cleared)
1093 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1096 if (mddev->level != LEVEL_MULTIPATH) {
1098 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1099 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1101 case 0xffff: /* spare */
1103 case 0xfffe: /* faulty */
1104 set_bit(Faulty, &rdev->flags);
1107 set_bit(In_sync, &rdev->flags);
1108 rdev->raid_disk = role;
1111 if (sb->devflags & WriteMostly1)
1112 set_bit(WriteMostly, &rdev->flags);
1113 } else /* MULTIPATH are always insync */
1114 set_bit(In_sync, &rdev->flags);
1119 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1121 struct mdp_superblock_1 *sb;
1122 struct list_head *tmp;
1125 /* make rdev->sb match mddev and rdev data. */
1127 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1129 sb->feature_map = 0;
1131 memset(sb->pad1, 0, sizeof(sb->pad1));
1132 memset(sb->pad2, 0, sizeof(sb->pad2));
1133 memset(sb->pad3, 0, sizeof(sb->pad3));
1135 sb->utime = cpu_to_le64((__u64)mddev->utime);
1136 sb->events = cpu_to_le64(mddev->events);
1138 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1140 sb->resync_offset = cpu_to_le64(0);
1142 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1143 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1144 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1148 ITERATE_RDEV(mddev,rdev2,tmp)
1149 if (rdev2->desc_nr+1 > max_dev)
1150 max_dev = rdev2->desc_nr+1;
1152 sb->max_dev = cpu_to_le32(max_dev);
1153 for (i=0; i<max_dev;i++)
1154 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1156 ITERATE_RDEV(mddev,rdev2,tmp) {
1158 if (test_bit(Faulty, &rdev2->flags))
1159 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1160 else if (test_bit(In_sync, &rdev2->flags))
1161 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1163 sb->dev_roles[i] = cpu_to_le16(0xffff);
1166 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1167 sb->sb_csum = calc_sb_1_csum(sb);
1171 static struct super_type super_types[] = {
1174 .owner = THIS_MODULE,
1175 .load_super = super_90_load,
1176 .validate_super = super_90_validate,
1177 .sync_super = super_90_sync,
1181 .owner = THIS_MODULE,
1182 .load_super = super_1_load,
1183 .validate_super = super_1_validate,
1184 .sync_super = super_1_sync,
1188 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1190 struct list_head *tmp;
1193 ITERATE_RDEV(mddev,rdev,tmp)
1194 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1200 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1202 struct list_head *tmp;
1205 ITERATE_RDEV(mddev1,rdev,tmp)
1206 if (match_dev_unit(mddev2, rdev))
1212 static LIST_HEAD(pending_raid_disks);
1214 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1216 mdk_rdev_t *same_pdev;
1217 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1224 same_pdev = match_dev_unit(mddev, rdev);
1227 "%s: WARNING: %s appears to be on the same physical"
1228 " disk as %s. True\n protection against single-disk"
1229 " failure might be compromised.\n",
1230 mdname(mddev), bdevname(rdev->bdev,b),
1231 bdevname(same_pdev->bdev,b2));
1233 /* Verify rdev->desc_nr is unique.
1234 * If it is -1, assign a free number, else
1235 * check number is not in use
1237 if (rdev->desc_nr < 0) {
1239 if (mddev->pers) choice = mddev->raid_disks;
1240 while (find_rdev_nr(mddev, choice))
1242 rdev->desc_nr = choice;
1244 if (find_rdev_nr(mddev, rdev->desc_nr))
1247 bdevname(rdev->bdev,b);
1248 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1251 list_add(&rdev->same_set, &mddev->disks);
1252 rdev->mddev = mddev;
1253 printk(KERN_INFO "md: bind<%s>\n", b);
1255 rdev->kobj.parent = &mddev->kobj;
1256 kobject_add(&rdev->kobj);
1258 if (rdev->bdev->bd_part)
1259 ko = &rdev->bdev->bd_part->kobj;
1261 ko = &rdev->bdev->bd_disk->kobj;
1262 sysfs_create_link(&rdev->kobj, ko, "block");
1266 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1268 char b[BDEVNAME_SIZE];
1273 list_del_init(&rdev->same_set);
1274 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1276 sysfs_remove_link(&rdev->kobj, "block");
1277 kobject_del(&rdev->kobj);
1281 * prevent the device from being mounted, repartitioned or
1282 * otherwise reused by a RAID array (or any other kernel
1283 * subsystem), by bd_claiming the device.
1285 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1288 struct block_device *bdev;
1289 char b[BDEVNAME_SIZE];
1291 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1293 printk(KERN_ERR "md: could not open %s.\n",
1294 __bdevname(dev, b));
1295 return PTR_ERR(bdev);
1297 err = bd_claim(bdev, rdev);
1299 printk(KERN_ERR "md: could not bd_claim %s.\n",
1308 static void unlock_rdev(mdk_rdev_t *rdev)
1310 struct block_device *bdev = rdev->bdev;
1318 void md_autodetect_dev(dev_t dev);
1320 static void export_rdev(mdk_rdev_t * rdev)
1322 char b[BDEVNAME_SIZE];
1323 printk(KERN_INFO "md: export_rdev(%s)\n",
1324 bdevname(rdev->bdev,b));
1328 list_del_init(&rdev->same_set);
1330 md_autodetect_dev(rdev->bdev->bd_dev);
1333 kobject_put(&rdev->kobj);
1336 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1338 unbind_rdev_from_array(rdev);
1342 static void export_array(mddev_t *mddev)
1344 struct list_head *tmp;
1347 ITERATE_RDEV(mddev,rdev,tmp) {
1352 kick_rdev_from_array(rdev);
1354 if (!list_empty(&mddev->disks))
1356 mddev->raid_disks = 0;
1357 mddev->major_version = 0;
1360 static void print_desc(mdp_disk_t *desc)
1362 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1363 desc->major,desc->minor,desc->raid_disk,desc->state);
1366 static void print_sb(mdp_super_t *sb)
1371 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1372 sb->major_version, sb->minor_version, sb->patch_version,
1373 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1375 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1376 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1377 sb->md_minor, sb->layout, sb->chunk_size);
1378 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1379 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1380 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1381 sb->failed_disks, sb->spare_disks,
1382 sb->sb_csum, (unsigned long)sb->events_lo);
1385 for (i = 0; i < MD_SB_DISKS; i++) {
1388 desc = sb->disks + i;
1389 if (desc->number || desc->major || desc->minor ||
1390 desc->raid_disk || (desc->state && (desc->state != 4))) {
1391 printk(" D %2d: ", i);
1395 printk(KERN_INFO "md: THIS: ");
1396 print_desc(&sb->this_disk);
1400 static void print_rdev(mdk_rdev_t *rdev)
1402 char b[BDEVNAME_SIZE];
1403 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1404 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1405 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1407 if (rdev->sb_loaded) {
1408 printk(KERN_INFO "md: rdev superblock:\n");
1409 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1411 printk(KERN_INFO "md: no rdev superblock!\n");
1414 void md_print_devices(void)
1416 struct list_head *tmp, *tmp2;
1419 char b[BDEVNAME_SIZE];
1422 printk("md: **********************************\n");
1423 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1424 printk("md: **********************************\n");
1425 ITERATE_MDDEV(mddev,tmp) {
1428 bitmap_print_sb(mddev->bitmap);
1430 printk("%s: ", mdname(mddev));
1431 ITERATE_RDEV(mddev,rdev,tmp2)
1432 printk("<%s>", bdevname(rdev->bdev,b));
1435 ITERATE_RDEV(mddev,rdev,tmp2)
1438 printk("md: **********************************\n");
1443 static void sync_sbs(mddev_t * mddev)
1446 struct list_head *tmp;
1448 ITERATE_RDEV(mddev,rdev,tmp) {
1449 super_types[mddev->major_version].
1450 sync_super(mddev, rdev);
1451 rdev->sb_loaded = 1;
1455 static void md_update_sb(mddev_t * mddev)
1458 struct list_head *tmp;
1463 spin_lock_irq(&mddev->write_lock);
1464 sync_req = mddev->in_sync;
1465 mddev->utime = get_seconds();
1468 if (!mddev->events) {
1470 * oops, this 64-bit counter should never wrap.
1471 * Either we are in around ~1 trillion A.C., assuming
1472 * 1 reboot per second, or we have a bug:
1477 mddev->sb_dirty = 2;
1481 * do not write anything to disk if using
1482 * nonpersistent superblocks
1484 if (!mddev->persistent) {
1485 mddev->sb_dirty = 0;
1486 spin_unlock_irq(&mddev->write_lock);
1487 wake_up(&mddev->sb_wait);
1490 spin_unlock_irq(&mddev->write_lock);
1493 "md: updating %s RAID superblock on device (in sync %d)\n",
1494 mdname(mddev),mddev->in_sync);
1496 err = bitmap_update_sb(mddev->bitmap);
1497 ITERATE_RDEV(mddev,rdev,tmp) {
1498 char b[BDEVNAME_SIZE];
1499 dprintk(KERN_INFO "md: ");
1500 if (test_bit(Faulty, &rdev->flags))
1501 dprintk("(skipping faulty ");
1503 dprintk("%s ", bdevname(rdev->bdev,b));
1504 if (!test_bit(Faulty, &rdev->flags)) {
1505 md_super_write(mddev,rdev,
1506 rdev->sb_offset<<1, rdev->sb_size,
1508 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1509 bdevname(rdev->bdev,b),
1510 (unsigned long long)rdev->sb_offset);
1514 if (mddev->level == LEVEL_MULTIPATH)
1515 /* only need to write one superblock... */
1518 md_super_wait(mddev);
1519 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1521 spin_lock_irq(&mddev->write_lock);
1522 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1523 /* have to write it out again */
1524 spin_unlock_irq(&mddev->write_lock);
1527 mddev->sb_dirty = 0;
1528 spin_unlock_irq(&mddev->write_lock);
1529 wake_up(&mddev->sb_wait);
1533 /* words written to sysfs files may, or my not, be \n terminated.
1534 * We want to accept with case. For this we use cmd_match.
1536 static int cmd_match(const char *cmd, const char *str)
1538 /* See if cmd, written into a sysfs file, matches
1539 * str. They must either be the same, or cmd can
1540 * have a trailing newline
1542 while (*cmd && *str && *cmd == *str) {
1553 struct rdev_sysfs_entry {
1554 struct attribute attr;
1555 ssize_t (*show)(mdk_rdev_t *, char *);
1556 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1560 state_show(mdk_rdev_t *rdev, char *page)
1565 if (test_bit(Faulty, &rdev->flags)) {
1566 len+= sprintf(page+len, "%sfaulty",sep);
1569 if (test_bit(In_sync, &rdev->flags)) {
1570 len += sprintf(page+len, "%sin_sync",sep);
1573 if (!test_bit(Faulty, &rdev->flags) &&
1574 !test_bit(In_sync, &rdev->flags)) {
1575 len += sprintf(page+len, "%sspare", sep);
1578 return len+sprintf(page+len, "\n");
1581 static struct rdev_sysfs_entry
1582 rdev_state = __ATTR_RO(state);
1585 super_show(mdk_rdev_t *rdev, char *page)
1587 if (rdev->sb_loaded && rdev->sb_size) {
1588 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1589 return rdev->sb_size;
1593 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1595 static struct attribute *rdev_default_attrs[] = {
1601 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1603 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1604 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1608 return entry->show(rdev, page);
1612 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1613 const char *page, size_t length)
1615 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1616 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1620 return entry->store(rdev, page, length);
1623 static void rdev_free(struct kobject *ko)
1625 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1628 static struct sysfs_ops rdev_sysfs_ops = {
1629 .show = rdev_attr_show,
1630 .store = rdev_attr_store,
1632 static struct kobj_type rdev_ktype = {
1633 .release = rdev_free,
1634 .sysfs_ops = &rdev_sysfs_ops,
1635 .default_attrs = rdev_default_attrs,
1639 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1641 * mark the device faulty if:
1643 * - the device is nonexistent (zero size)
1644 * - the device has no valid superblock
1646 * a faulty rdev _never_ has rdev->sb set.
1648 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1650 char b[BDEVNAME_SIZE];
1655 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1657 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1658 return ERR_PTR(-ENOMEM);
1661 if ((err = alloc_disk_sb(rdev)))
1664 err = lock_rdev(rdev, newdev);
1668 rdev->kobj.parent = NULL;
1669 rdev->kobj.ktype = &rdev_ktype;
1670 kobject_init(&rdev->kobj);
1674 rdev->data_offset = 0;
1675 atomic_set(&rdev->nr_pending, 0);
1676 atomic_set(&rdev->read_errors, 0);
1678 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1681 "md: %s has zero or unknown size, marking faulty!\n",
1682 bdevname(rdev->bdev,b));
1687 if (super_format >= 0) {
1688 err = super_types[super_format].
1689 load_super(rdev, NULL, super_minor);
1690 if (err == -EINVAL) {
1692 "md: %s has invalid sb, not importing!\n",
1693 bdevname(rdev->bdev,b));
1698 "md: could not read %s's sb, not importing!\n",
1699 bdevname(rdev->bdev,b));
1703 INIT_LIST_HEAD(&rdev->same_set);
1708 if (rdev->sb_page) {
1714 return ERR_PTR(err);
1718 * Check a full RAID array for plausibility
1722 static void analyze_sbs(mddev_t * mddev)
1725 struct list_head *tmp;
1726 mdk_rdev_t *rdev, *freshest;
1727 char b[BDEVNAME_SIZE];
1730 ITERATE_RDEV(mddev,rdev,tmp)
1731 switch (super_types[mddev->major_version].
1732 load_super(rdev, freshest, mddev->minor_version)) {
1740 "md: fatal superblock inconsistency in %s"
1741 " -- removing from array\n",
1742 bdevname(rdev->bdev,b));
1743 kick_rdev_from_array(rdev);
1747 super_types[mddev->major_version].
1748 validate_super(mddev, freshest);
1751 ITERATE_RDEV(mddev,rdev,tmp) {
1752 if (rdev != freshest)
1753 if (super_types[mddev->major_version].
1754 validate_super(mddev, rdev)) {
1755 printk(KERN_WARNING "md: kicking non-fresh %s"
1757 bdevname(rdev->bdev,b));
1758 kick_rdev_from_array(rdev);
1761 if (mddev->level == LEVEL_MULTIPATH) {
1762 rdev->desc_nr = i++;
1763 rdev->raid_disk = rdev->desc_nr;
1764 set_bit(In_sync, &rdev->flags);
1770 if (mddev->recovery_cp != MaxSector &&
1772 printk(KERN_ERR "md: %s: raid array is not clean"
1773 " -- starting background reconstruction\n",
1779 level_show(mddev_t *mddev, char *page)
1781 struct mdk_personality *p = mddev->pers;
1783 return sprintf(page, "%s\n", p->name);
1784 else if (mddev->clevel[0])
1785 return sprintf(page, "%s\n", mddev->clevel);
1786 else if (mddev->level != LEVEL_NONE)
1787 return sprintf(page, "%d\n", mddev->level);
1793 level_store(mddev_t *mddev, const char *buf, size_t len)
1800 if (len >= sizeof(mddev->clevel))
1802 strncpy(mddev->clevel, buf, len);
1803 if (mddev->clevel[len-1] == '\n')
1805 mddev->clevel[len] = 0;
1806 mddev->level = LEVEL_NONE;
1810 static struct md_sysfs_entry md_level =
1811 __ATTR(level, 0644, level_show, level_store);
1814 raid_disks_show(mddev_t *mddev, char *page)
1816 if (mddev->raid_disks == 0)
1818 return sprintf(page, "%d\n", mddev->raid_disks);
1821 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1824 chunk_size_show(mddev_t *mddev, char *page)
1826 return sprintf(page, "%d\n", mddev->chunk_size);
1830 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1832 /* can only set chunk_size if array is not yet active */
1834 unsigned long n = simple_strtoul(buf, &e, 10);
1838 if (!*buf || (*e && *e != '\n'))
1841 mddev->chunk_size = n;
1844 static struct md_sysfs_entry md_chunk_size =
1845 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
1849 size_show(mddev_t *mddev, char *page)
1851 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
1854 static int update_size(mddev_t *mddev, unsigned long size);
1857 size_store(mddev_t *mddev, const char *buf, size_t len)
1859 /* If array is inactive, we can reduce the component size, but
1860 * not increase it (except from 0).
1861 * If array is active, we can try an on-line resize
1865 unsigned long long size = simple_strtoull(buf, &e, 10);
1866 if (!*buf || *buf == '\n' ||
1871 err = update_size(mddev, size);
1872 md_update_sb(mddev);
1874 if (mddev->size == 0 ||
1880 return err ? err : len;
1883 static struct md_sysfs_entry md_size =
1884 __ATTR(component_size, 0644, size_show, size_store);
1888 * This is either 'none' for arrays with externally managed metadata,
1889 * or N.M for internally known formats
1892 metadata_show(mddev_t *mddev, char *page)
1894 if (mddev->persistent)
1895 return sprintf(page, "%d.%d\n",
1896 mddev->major_version, mddev->minor_version);
1898 return sprintf(page, "none\n");
1902 metadata_store(mddev_t *mddev, const char *buf, size_t len)
1906 if (!list_empty(&mddev->disks))
1909 if (cmd_match(buf, "none")) {
1910 mddev->persistent = 0;
1911 mddev->major_version = 0;
1912 mddev->minor_version = 90;
1915 major = simple_strtoul(buf, &e, 10);
1916 if (e==buf || *e != '.')
1919 minor = simple_strtoul(buf, &e, 10);
1920 if (e==buf || *e != '\n')
1922 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
1923 super_types[major].name == NULL)
1925 mddev->major_version = major;
1926 mddev->minor_version = minor;
1927 mddev->persistent = 1;
1931 static struct md_sysfs_entry md_metadata =
1932 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
1935 action_show(mddev_t *mddev, char *page)
1937 char *type = "idle";
1938 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1939 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1940 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1941 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1943 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1950 return sprintf(page, "%s\n", type);
1954 action_store(mddev_t *mddev, const char *page, size_t len)
1956 if (!mddev->pers || !mddev->pers->sync_request)
1959 if (cmd_match(page, "idle")) {
1960 if (mddev->sync_thread) {
1961 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1962 md_unregister_thread(mddev->sync_thread);
1963 mddev->sync_thread = NULL;
1964 mddev->recovery = 0;
1966 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1967 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1969 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
1970 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1972 if (cmd_match(page, "check"))
1973 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1974 else if (cmd_match(page, "repair"))
1976 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1977 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1979 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1980 md_wakeup_thread(mddev->thread);
1985 mismatch_cnt_show(mddev_t *mddev, char *page)
1987 return sprintf(page, "%llu\n",
1988 (unsigned long long) mddev->resync_mismatches);
1991 static struct md_sysfs_entry
1992 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
1995 static struct md_sysfs_entry
1996 md_mismatches = __ATTR_RO(mismatch_cnt);
1998 static struct attribute *md_default_attrs[] = {
2000 &md_raid_disks.attr,
2001 &md_chunk_size.attr,
2007 static struct attribute *md_redundancy_attrs[] = {
2009 &md_mismatches.attr,
2012 static struct attribute_group md_redundancy_group = {
2014 .attrs = md_redundancy_attrs,
2019 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2021 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2022 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2028 rv = entry->show(mddev, page);
2029 mddev_unlock(mddev);
2034 md_attr_store(struct kobject *kobj, struct attribute *attr,
2035 const char *page, size_t length)
2037 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2038 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2044 rv = entry->store(mddev, page, length);
2045 mddev_unlock(mddev);
2049 static void md_free(struct kobject *ko)
2051 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2055 static struct sysfs_ops md_sysfs_ops = {
2056 .show = md_attr_show,
2057 .store = md_attr_store,
2059 static struct kobj_type md_ktype = {
2061 .sysfs_ops = &md_sysfs_ops,
2062 .default_attrs = md_default_attrs,
2067 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2069 static DECLARE_MUTEX(disks_sem);
2070 mddev_t *mddev = mddev_find(dev);
2071 struct gendisk *disk;
2072 int partitioned = (MAJOR(dev) != MD_MAJOR);
2073 int shift = partitioned ? MdpMinorShift : 0;
2074 int unit = MINOR(dev) >> shift;
2080 if (mddev->gendisk) {
2085 disk = alloc_disk(1 << shift);
2091 disk->major = MAJOR(dev);
2092 disk->first_minor = unit << shift;
2094 sprintf(disk->disk_name, "md_d%d", unit);
2095 sprintf(disk->devfs_name, "md/d%d", unit);
2097 sprintf(disk->disk_name, "md%d", unit);
2098 sprintf(disk->devfs_name, "md/%d", unit);
2100 disk->fops = &md_fops;
2101 disk->private_data = mddev;
2102 disk->queue = mddev->queue;
2104 mddev->gendisk = disk;
2106 mddev->kobj.parent = &disk->kobj;
2107 mddev->kobj.k_name = NULL;
2108 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2109 mddev->kobj.ktype = &md_ktype;
2110 kobject_register(&mddev->kobj);
2114 void md_wakeup_thread(mdk_thread_t *thread);
2116 static void md_safemode_timeout(unsigned long data)
2118 mddev_t *mddev = (mddev_t *) data;
2120 mddev->safemode = 1;
2121 md_wakeup_thread(mddev->thread);
2124 static int start_dirty_degraded;
2126 static int do_md_run(mddev_t * mddev)
2130 struct list_head *tmp;
2132 struct gendisk *disk;
2133 struct mdk_personality *pers;
2134 char b[BDEVNAME_SIZE];
2136 if (list_empty(&mddev->disks))
2137 /* cannot run an array with no devices.. */
2144 * Analyze all RAID superblock(s)
2146 if (!mddev->raid_disks)
2149 chunk_size = mddev->chunk_size;
2152 if (chunk_size > MAX_CHUNK_SIZE) {
2153 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2154 chunk_size, MAX_CHUNK_SIZE);
2158 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2160 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2161 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2164 if (chunk_size < PAGE_SIZE) {
2165 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2166 chunk_size, PAGE_SIZE);
2170 /* devices must have minimum size of one chunk */
2171 ITERATE_RDEV(mddev,rdev,tmp) {
2172 if (test_bit(Faulty, &rdev->flags))
2174 if (rdev->size < chunk_size / 1024) {
2176 "md: Dev %s smaller than chunk_size:"
2178 bdevname(rdev->bdev,b),
2179 (unsigned long long)rdev->size,
2187 if (mddev->level != LEVEL_NONE)
2188 request_module("md-level-%d", mddev->level);
2189 else if (mddev->clevel[0])
2190 request_module("md-%s", mddev->clevel);
2194 * Drop all container device buffers, from now on
2195 * the only valid external interface is through the md
2197 * Also find largest hardsector size
2199 ITERATE_RDEV(mddev,rdev,tmp) {
2200 if (test_bit(Faulty, &rdev->flags))
2202 sync_blockdev(rdev->bdev);
2203 invalidate_bdev(rdev->bdev, 0);
2206 md_probe(mddev->unit, NULL, NULL);
2207 disk = mddev->gendisk;
2211 spin_lock(&pers_lock);
2212 pers = find_pers(mddev->level, mddev->clevel);
2213 if (!pers || !try_module_get(pers->owner)) {
2214 spin_unlock(&pers_lock);
2215 if (mddev->level != LEVEL_NONE)
2216 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2219 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2224 spin_unlock(&pers_lock);
2225 mddev->level = pers->level;
2226 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2228 mddev->recovery = 0;
2229 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2230 mddev->barriers_work = 1;
2231 mddev->ok_start_degraded = start_dirty_degraded;
2234 mddev->ro = 2; /* read-only, but switch on first write */
2236 err = mddev->pers->run(mddev);
2237 if (!err && mddev->pers->sync_request) {
2238 err = bitmap_create(mddev);
2240 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2241 mdname(mddev), err);
2242 mddev->pers->stop(mddev);
2246 printk(KERN_ERR "md: pers->run() failed ...\n");
2247 module_put(mddev->pers->owner);
2249 bitmap_destroy(mddev);
2252 if (mddev->pers->sync_request)
2253 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2254 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2257 atomic_set(&mddev->writes_pending,0);
2258 mddev->safemode = 0;
2259 mddev->safemode_timer.function = md_safemode_timeout;
2260 mddev->safemode_timer.data = (unsigned long) mddev;
2261 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2264 ITERATE_RDEV(mddev,rdev,tmp)
2265 if (rdev->raid_disk >= 0) {
2267 sprintf(nm, "rd%d", rdev->raid_disk);
2268 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2271 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2272 md_wakeup_thread(mddev->thread);
2274 if (mddev->sb_dirty)
2275 md_update_sb(mddev);
2277 set_capacity(disk, mddev->array_size<<1);
2279 /* If we call blk_queue_make_request here, it will
2280 * re-initialise max_sectors etc which may have been
2281 * refined inside -> run. So just set the bits we need to set.
2282 * Most initialisation happended when we called
2283 * blk_queue_make_request(..., md_fail_request)
2286 mddev->queue->queuedata = mddev;
2287 mddev->queue->make_request_fn = mddev->pers->make_request;
2290 md_new_event(mddev);
2294 static int restart_array(mddev_t *mddev)
2296 struct gendisk *disk = mddev->gendisk;
2300 * Complain if it has no devices
2303 if (list_empty(&mddev->disks))
2311 mddev->safemode = 0;
2313 set_disk_ro(disk, 0);
2315 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2318 * Kick recovery or resync if necessary
2320 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2321 md_wakeup_thread(mddev->thread);
2324 printk(KERN_ERR "md: %s has no personality assigned.\n",
2333 static int do_md_stop(mddev_t * mddev, int ro)
2336 struct gendisk *disk = mddev->gendisk;
2339 if (atomic_read(&mddev->active)>2) {
2340 printk("md: %s still in use.\n",mdname(mddev));
2344 if (mddev->sync_thread) {
2345 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2346 md_unregister_thread(mddev->sync_thread);
2347 mddev->sync_thread = NULL;
2350 del_timer_sync(&mddev->safemode_timer);
2352 invalidate_partition(disk, 0);
2360 bitmap_flush(mddev);
2361 md_super_wait(mddev);
2363 set_disk_ro(disk, 0);
2364 blk_queue_make_request(mddev->queue, md_fail_request);
2365 mddev->pers->stop(mddev);
2366 if (mddev->pers->sync_request)
2367 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2369 module_put(mddev->pers->owner);
2374 if (!mddev->in_sync) {
2375 /* mark array as shutdown cleanly */
2377 md_update_sb(mddev);
2380 set_disk_ro(disk, 1);
2383 bitmap_destroy(mddev);
2384 if (mddev->bitmap_file) {
2385 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2386 fput(mddev->bitmap_file);
2387 mddev->bitmap_file = NULL;
2389 mddev->bitmap_offset = 0;
2392 * Free resources if final stop
2396 struct list_head *tmp;
2397 struct gendisk *disk;
2398 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2400 ITERATE_RDEV(mddev,rdev,tmp)
2401 if (rdev->raid_disk >= 0) {
2403 sprintf(nm, "rd%d", rdev->raid_disk);
2404 sysfs_remove_link(&mddev->kobj, nm);
2407 export_array(mddev);
2409 mddev->array_size = 0;
2410 disk = mddev->gendisk;
2412 set_capacity(disk, 0);
2415 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2418 md_new_event(mddev);
2423 static void autorun_array(mddev_t *mddev)
2426 struct list_head *tmp;
2429 if (list_empty(&mddev->disks))
2432 printk(KERN_INFO "md: running: ");
2434 ITERATE_RDEV(mddev,rdev,tmp) {
2435 char b[BDEVNAME_SIZE];
2436 printk("<%s>", bdevname(rdev->bdev,b));
2440 err = do_md_run (mddev);
2442 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2443 do_md_stop (mddev, 0);
2448 * lets try to run arrays based on all disks that have arrived
2449 * until now. (those are in pending_raid_disks)
2451 * the method: pick the first pending disk, collect all disks with
2452 * the same UUID, remove all from the pending list and put them into
2453 * the 'same_array' list. Then order this list based on superblock
2454 * update time (freshest comes first), kick out 'old' disks and
2455 * compare superblocks. If everything's fine then run it.
2457 * If "unit" is allocated, then bump its reference count
2459 static void autorun_devices(int part)
2461 struct list_head candidates;
2462 struct list_head *tmp;
2463 mdk_rdev_t *rdev0, *rdev;
2465 char b[BDEVNAME_SIZE];
2467 printk(KERN_INFO "md: autorun ...\n");
2468 while (!list_empty(&pending_raid_disks)) {
2470 rdev0 = list_entry(pending_raid_disks.next,
2471 mdk_rdev_t, same_set);
2473 printk(KERN_INFO "md: considering %s ...\n",
2474 bdevname(rdev0->bdev,b));
2475 INIT_LIST_HEAD(&candidates);
2476 ITERATE_RDEV_PENDING(rdev,tmp)
2477 if (super_90_load(rdev, rdev0, 0) >= 0) {
2478 printk(KERN_INFO "md: adding %s ...\n",
2479 bdevname(rdev->bdev,b));
2480 list_move(&rdev->same_set, &candidates);
2483 * now we have a set of devices, with all of them having
2484 * mostly sane superblocks. It's time to allocate the
2487 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2488 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2489 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2493 dev = MKDEV(mdp_major,
2494 rdev0->preferred_minor << MdpMinorShift);
2496 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2498 md_probe(dev, NULL, NULL);
2499 mddev = mddev_find(dev);
2502 "md: cannot allocate memory for md drive.\n");
2505 if (mddev_lock(mddev))
2506 printk(KERN_WARNING "md: %s locked, cannot run\n",
2508 else if (mddev->raid_disks || mddev->major_version
2509 || !list_empty(&mddev->disks)) {
2511 "md: %s already running, cannot run %s\n",
2512 mdname(mddev), bdevname(rdev0->bdev,b));
2513 mddev_unlock(mddev);
2515 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2516 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2517 list_del_init(&rdev->same_set);
2518 if (bind_rdev_to_array(rdev, mddev))
2521 autorun_array(mddev);
2522 mddev_unlock(mddev);
2524 /* on success, candidates will be empty, on error
2527 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2531 printk(KERN_INFO "md: ... autorun DONE.\n");
2535 * import RAID devices based on one partition
2536 * if possible, the array gets run as well.
2539 static int autostart_array(dev_t startdev)
2541 char b[BDEVNAME_SIZE];
2542 int err = -EINVAL, i;
2543 mdp_super_t *sb = NULL;
2544 mdk_rdev_t *start_rdev = NULL, *rdev;
2546 start_rdev = md_import_device(startdev, 0, 0);
2547 if (IS_ERR(start_rdev))
2551 /* NOTE: this can only work for 0.90.0 superblocks */
2552 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2553 if (sb->major_version != 0 ||
2554 sb->minor_version != 90 ) {
2555 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2556 export_rdev(start_rdev);
2560 if (test_bit(Faulty, &start_rdev->flags)) {
2562 "md: can not autostart based on faulty %s!\n",
2563 bdevname(start_rdev->bdev,b));
2564 export_rdev(start_rdev);
2567 list_add(&start_rdev->same_set, &pending_raid_disks);
2569 for (i = 0; i < MD_SB_DISKS; i++) {
2570 mdp_disk_t *desc = sb->disks + i;
2571 dev_t dev = MKDEV(desc->major, desc->minor);
2575 if (dev == startdev)
2577 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2579 rdev = md_import_device(dev, 0, 0);
2583 list_add(&rdev->same_set, &pending_raid_disks);
2587 * possibly return codes
2595 static int get_version(void __user * arg)
2599 ver.major = MD_MAJOR_VERSION;
2600 ver.minor = MD_MINOR_VERSION;
2601 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2603 if (copy_to_user(arg, &ver, sizeof(ver)))
2609 static int get_array_info(mddev_t * mddev, void __user * arg)
2611 mdu_array_info_t info;
2612 int nr,working,active,failed,spare;
2614 struct list_head *tmp;
2616 nr=working=active=failed=spare=0;
2617 ITERATE_RDEV(mddev,rdev,tmp) {
2619 if (test_bit(Faulty, &rdev->flags))
2623 if (test_bit(In_sync, &rdev->flags))
2630 info.major_version = mddev->major_version;
2631 info.minor_version = mddev->minor_version;
2632 info.patch_version = MD_PATCHLEVEL_VERSION;
2633 info.ctime = mddev->ctime;
2634 info.level = mddev->level;
2635 info.size = mddev->size;
2637 info.raid_disks = mddev->raid_disks;
2638 info.md_minor = mddev->md_minor;
2639 info.not_persistent= !mddev->persistent;
2641 info.utime = mddev->utime;
2644 info.state = (1<<MD_SB_CLEAN);
2645 if (mddev->bitmap && mddev->bitmap_offset)
2646 info.state = (1<<MD_SB_BITMAP_PRESENT);
2647 info.active_disks = active;
2648 info.working_disks = working;
2649 info.failed_disks = failed;
2650 info.spare_disks = spare;
2652 info.layout = mddev->layout;
2653 info.chunk_size = mddev->chunk_size;
2655 if (copy_to_user(arg, &info, sizeof(info)))
2661 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2663 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2664 char *ptr, *buf = NULL;
2667 file = kmalloc(sizeof(*file), GFP_KERNEL);
2671 /* bitmap disabled, zero the first byte and copy out */
2672 if (!mddev->bitmap || !mddev->bitmap->file) {
2673 file->pathname[0] = '\0';
2677 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2681 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2685 strcpy(file->pathname, ptr);
2689 if (copy_to_user(arg, file, sizeof(*file)))
2697 static int get_disk_info(mddev_t * mddev, void __user * arg)
2699 mdu_disk_info_t info;
2703 if (copy_from_user(&info, arg, sizeof(info)))
2708 rdev = find_rdev_nr(mddev, nr);
2710 info.major = MAJOR(rdev->bdev->bd_dev);
2711 info.minor = MINOR(rdev->bdev->bd_dev);
2712 info.raid_disk = rdev->raid_disk;
2714 if (test_bit(Faulty, &rdev->flags))
2715 info.state |= (1<<MD_DISK_FAULTY);
2716 else if (test_bit(In_sync, &rdev->flags)) {
2717 info.state |= (1<<MD_DISK_ACTIVE);
2718 info.state |= (1<<MD_DISK_SYNC);
2720 if (test_bit(WriteMostly, &rdev->flags))
2721 info.state |= (1<<MD_DISK_WRITEMOSTLY);
2723 info.major = info.minor = 0;
2724 info.raid_disk = -1;
2725 info.state = (1<<MD_DISK_REMOVED);
2728 if (copy_to_user(arg, &info, sizeof(info)))
2734 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2736 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2738 dev_t dev = MKDEV(info->major,info->minor);
2740 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2743 if (!mddev->raid_disks) {
2745 /* expecting a device which has a superblock */
2746 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2749 "md: md_import_device returned %ld\n",
2751 return PTR_ERR(rdev);
2753 if (!list_empty(&mddev->disks)) {
2754 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2755 mdk_rdev_t, same_set);
2756 int err = super_types[mddev->major_version]
2757 .load_super(rdev, rdev0, mddev->minor_version);
2760 "md: %s has different UUID to %s\n",
2761 bdevname(rdev->bdev,b),
2762 bdevname(rdev0->bdev,b2));
2767 err = bind_rdev_to_array(rdev, mddev);
2774 * add_new_disk can be used once the array is assembled
2775 * to add "hot spares". They must already have a superblock
2780 if (!mddev->pers->hot_add_disk) {
2782 "%s: personality does not support diskops!\n",
2786 if (mddev->persistent)
2787 rdev = md_import_device(dev, mddev->major_version,
2788 mddev->minor_version);
2790 rdev = md_import_device(dev, -1, -1);
2793 "md: md_import_device returned %ld\n",
2795 return PTR_ERR(rdev);
2797 /* set save_raid_disk if appropriate */
2798 if (!mddev->persistent) {
2799 if (info->state & (1<<MD_DISK_SYNC) &&
2800 info->raid_disk < mddev->raid_disks)
2801 rdev->raid_disk = info->raid_disk;
2803 rdev->raid_disk = -1;
2805 super_types[mddev->major_version].
2806 validate_super(mddev, rdev);
2807 rdev->saved_raid_disk = rdev->raid_disk;
2809 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2810 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2811 set_bit(WriteMostly, &rdev->flags);
2813 rdev->raid_disk = -1;
2814 err = bind_rdev_to_array(rdev, mddev);
2818 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2819 md_wakeup_thread(mddev->thread);
2823 /* otherwise, add_new_disk is only allowed
2824 * for major_version==0 superblocks
2826 if (mddev->major_version != 0) {
2827 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2832 if (!(info->state & (1<<MD_DISK_FAULTY))) {
2834 rdev = md_import_device (dev, -1, 0);
2837 "md: error, md_import_device() returned %ld\n",
2839 return PTR_ERR(rdev);
2841 rdev->desc_nr = info->number;
2842 if (info->raid_disk < mddev->raid_disks)
2843 rdev->raid_disk = info->raid_disk;
2845 rdev->raid_disk = -1;
2849 if (rdev->raid_disk < mddev->raid_disks)
2850 if (info->state & (1<<MD_DISK_SYNC))
2851 set_bit(In_sync, &rdev->flags);
2853 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2854 set_bit(WriteMostly, &rdev->flags);
2856 err = bind_rdev_to_array(rdev, mddev);
2862 if (!mddev->persistent) {
2863 printk(KERN_INFO "md: nonpersistent superblock ...\n");
2864 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2866 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2867 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2869 if (!mddev->size || (mddev->size > rdev->size))
2870 mddev->size = rdev->size;
2876 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2878 char b[BDEVNAME_SIZE];
2884 rdev = find_rdev(mddev, dev);
2888 if (rdev->raid_disk >= 0)
2891 kick_rdev_from_array(rdev);
2892 md_update_sb(mddev);
2893 md_new_event(mddev);
2897 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2898 bdevname(rdev->bdev,b), mdname(mddev));
2902 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2904 char b[BDEVNAME_SIZE];
2912 if (mddev->major_version != 0) {
2913 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2914 " version-0 superblocks.\n",
2918 if (!mddev->pers->hot_add_disk) {
2920 "%s: personality does not support diskops!\n",
2925 rdev = md_import_device (dev, -1, 0);
2928 "md: error, md_import_device() returned %ld\n",
2933 if (mddev->persistent)
2934 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2937 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2939 size = calc_dev_size(rdev, mddev->chunk_size);
2942 if (size < mddev->size) {
2944 "%s: disk size %llu blocks < array size %llu\n",
2945 mdname(mddev), (unsigned long long)size,
2946 (unsigned long long)mddev->size);
2951 if (test_bit(Faulty, &rdev->flags)) {
2953 "md: can not hot-add faulty %s disk to %s!\n",
2954 bdevname(rdev->bdev,b), mdname(mddev));
2958 clear_bit(In_sync, &rdev->flags);
2960 bind_rdev_to_array(rdev, mddev);
2963 * The rest should better be atomic, we can have disk failures
2964 * noticed in interrupt contexts ...
2967 if (rdev->desc_nr == mddev->max_disks) {
2968 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2971 goto abort_unbind_export;
2974 rdev->raid_disk = -1;
2976 md_update_sb(mddev);
2979 * Kick recovery, maybe this spare has to be added to the
2980 * array immediately.
2982 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2983 md_wakeup_thread(mddev->thread);
2984 md_new_event(mddev);
2987 abort_unbind_export:
2988 unbind_rdev_from_array(rdev);
2995 /* similar to deny_write_access, but accounts for our holding a reference
2996 * to the file ourselves */
2997 static int deny_bitmap_write_access(struct file * file)
2999 struct inode *inode = file->f_mapping->host;
3001 spin_lock(&inode->i_lock);
3002 if (atomic_read(&inode->i_writecount) > 1) {
3003 spin_unlock(&inode->i_lock);
3006 atomic_set(&inode->i_writecount, -1);
3007 spin_unlock(&inode->i_lock);
3012 static int set_bitmap_file(mddev_t *mddev, int fd)
3017 if (!mddev->pers->quiesce)
3019 if (mddev->recovery || mddev->sync_thread)
3021 /* we should be able to change the bitmap.. */
3027 return -EEXIST; /* cannot add when bitmap is present */
3028 mddev->bitmap_file = fget(fd);
3030 if (mddev->bitmap_file == NULL) {
3031 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3036 err = deny_bitmap_write_access(mddev->bitmap_file);
3038 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3040 fput(mddev->bitmap_file);
3041 mddev->bitmap_file = NULL;
3044 mddev->bitmap_offset = 0; /* file overrides offset */
3045 } else if (mddev->bitmap == NULL)
3046 return -ENOENT; /* cannot remove what isn't there */
3049 mddev->pers->quiesce(mddev, 1);
3051 err = bitmap_create(mddev);
3053 bitmap_destroy(mddev);
3054 mddev->pers->quiesce(mddev, 0);
3055 } else if (fd < 0) {
3056 if (mddev->bitmap_file)
3057 fput(mddev->bitmap_file);
3058 mddev->bitmap_file = NULL;
3065 * set_array_info is used two different ways
3066 * The original usage is when creating a new array.
3067 * In this usage, raid_disks is > 0 and it together with
3068 * level, size, not_persistent,layout,chunksize determine the
3069 * shape of the array.
3070 * This will always create an array with a type-0.90.0 superblock.
3071 * The newer usage is when assembling an array.
3072 * In this case raid_disks will be 0, and the major_version field is
3073 * use to determine which style super-blocks are to be found on the devices.
3074 * The minor and patch _version numbers are also kept incase the
3075 * super_block handler wishes to interpret them.
3077 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3080 if (info->raid_disks == 0) {
3081 /* just setting version number for superblock loading */
3082 if (info->major_version < 0 ||
3083 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3084 super_types[info->major_version].name == NULL) {
3085 /* maybe try to auto-load a module? */
3087 "md: superblock version %d not known\n",
3088 info->major_version);
3091 mddev->major_version = info->major_version;
3092 mddev->minor_version = info->minor_version;
3093 mddev->patch_version = info->patch_version;
3096 mddev->major_version = MD_MAJOR_VERSION;
3097 mddev->minor_version = MD_MINOR_VERSION;
3098 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3099 mddev->ctime = get_seconds();
3101 mddev->level = info->level;
3102 mddev->size = info->size;
3103 mddev->raid_disks = info->raid_disks;
3104 /* don't set md_minor, it is determined by which /dev/md* was
3107 if (info->state & (1<<MD_SB_CLEAN))
3108 mddev->recovery_cp = MaxSector;
3110 mddev->recovery_cp = 0;
3111 mddev->persistent = ! info->not_persistent;
3113 mddev->layout = info->layout;
3114 mddev->chunk_size = info->chunk_size;
3116 mddev->max_disks = MD_SB_DISKS;
3118 mddev->sb_dirty = 1;
3120 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3121 mddev->bitmap_offset = 0;
3124 * Generate a 128 bit UUID
3126 get_random_bytes(mddev->uuid, 16);
3131 static int update_size(mddev_t *mddev, unsigned long size)
3135 struct list_head *tmp;
3137 if (mddev->pers->resize == NULL)
3139 /* The "size" is the amount of each device that is used.
3140 * This can only make sense for arrays with redundancy.
3141 * linear and raid0 always use whatever space is available
3142 * We can only consider changing the size if no resync
3143 * or reconstruction is happening, and if the new size
3144 * is acceptable. It must fit before the sb_offset or,
3145 * if that is <data_offset, it must fit before the
3146 * size of each device.
3147 * If size is zero, we find the largest size that fits.
3149 if (mddev->sync_thread)
3151 ITERATE_RDEV(mddev,rdev,tmp) {
3153 int fit = (size == 0);
3154 if (rdev->sb_offset > rdev->data_offset)
3155 avail = (rdev->sb_offset*2) - rdev->data_offset;
3157 avail = get_capacity(rdev->bdev->bd_disk)
3158 - rdev->data_offset;
3159 if (fit && (size == 0 || size > avail/2))
3161 if (avail < ((sector_t)size << 1))
3164 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3166 struct block_device *bdev;
3168 bdev = bdget_disk(mddev->gendisk, 0);
3170 down(&bdev->bd_inode->i_sem);
3171 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3172 up(&bdev->bd_inode->i_sem);
3180 * update_array_info is used to change the configuration of an
3182 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3183 * fields in the info are checked against the array.
3184 * Any differences that cannot be handled will cause an error.
3185 * Normally, only one change can be managed at a time.
3187 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3193 /* calculate expected state,ignoring low bits */
3194 if (mddev->bitmap && mddev->bitmap_offset)
3195 state |= (1 << MD_SB_BITMAP_PRESENT);
3197 if (mddev->major_version != info->major_version ||
3198 mddev->minor_version != info->minor_version ||
3199 /* mddev->patch_version != info->patch_version || */
3200 mddev->ctime != info->ctime ||
3201 mddev->level != info->level ||
3202 /* mddev->layout != info->layout || */
3203 !mddev->persistent != info->not_persistent||
3204 mddev->chunk_size != info->chunk_size ||
3205 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3206 ((state^info->state) & 0xfffffe00)
3209 /* Check there is only one change */
3210 if (mddev->size != info->size) cnt++;
3211 if (mddev->raid_disks != info->raid_disks) cnt++;
3212 if (mddev->layout != info->layout) cnt++;
3213 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3214 if (cnt == 0) return 0;
3215 if (cnt > 1) return -EINVAL;
3217 if (mddev->layout != info->layout) {
3219 * we don't need to do anything at the md level, the
3220 * personality will take care of it all.
3222 if (mddev->pers->reconfig == NULL)
3225 return mddev->pers->reconfig(mddev, info->layout, -1);
3227 if (mddev->size != info->size)
3228 rv = update_size(mddev, info->size);
3230 if (mddev->raid_disks != info->raid_disks) {
3231 /* change the number of raid disks */
3232 if (mddev->pers->reshape == NULL)
3234 if (info->raid_disks <= 0 ||
3235 info->raid_disks >= mddev->max_disks)
3237 if (mddev->sync_thread)
3239 rv = mddev->pers->reshape(mddev, info->raid_disks);
3241 struct block_device *bdev;
3243 bdev = bdget_disk(mddev->gendisk, 0);
3245 down(&bdev->bd_inode->i_sem);
3246 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3247 up(&bdev->bd_inode->i_sem);
3252 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3253 if (mddev->pers->quiesce == NULL)
3255 if (mddev->recovery || mddev->sync_thread)
3257 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3258 /* add the bitmap */
3261 if (mddev->default_bitmap_offset == 0)
3263 mddev->bitmap_offset = mddev->default_bitmap_offset;
3264 mddev->pers->quiesce(mddev, 1);
3265 rv = bitmap_create(mddev);
3267 bitmap_destroy(mddev);
3268 mddev->pers->quiesce(mddev, 0);
3270 /* remove the bitmap */
3273 if (mddev->bitmap->file)
3275 mddev->pers->quiesce(mddev, 1);
3276 bitmap_destroy(mddev);
3277 mddev->pers->quiesce(mddev, 0);
3278 mddev->bitmap_offset = 0;
3281 md_update_sb(mddev);
3285 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3289 if (mddev->pers == NULL)
3292 rdev = find_rdev(mddev, dev);
3296 md_error(mddev, rdev);
3300 static int md_ioctl(struct inode *inode, struct file *file,
3301 unsigned int cmd, unsigned long arg)
3304 void __user *argp = (void __user *)arg;
3305 struct hd_geometry __user *loc = argp;
3306 mddev_t *mddev = NULL;
3308 if (!capable(CAP_SYS_ADMIN))
3312 * Commands dealing with the RAID driver but not any
3318 err = get_version(argp);
3321 case PRINT_RAID_DEBUG:
3329 autostart_arrays(arg);
3336 * Commands creating/starting a new array:
3339 mddev = inode->i_bdev->bd_disk->private_data;
3347 if (cmd == START_ARRAY) {
3348 /* START_ARRAY doesn't need to lock the array as autostart_array
3349 * does the locking, and it could even be a different array
3354 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3355 "This will not be supported beyond July 2006\n",
3356 current->comm, current->pid);
3359 err = autostart_array(new_decode_dev(arg));
3361 printk(KERN_WARNING "md: autostart failed!\n");
3367 err = mddev_lock(mddev);
3370 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3377 case SET_ARRAY_INFO:
3379 mdu_array_info_t info;
3381 memset(&info, 0, sizeof(info));
3382 else if (copy_from_user(&info, argp, sizeof(info))) {
3387 err = update_array_info(mddev, &info);
3389 printk(KERN_WARNING "md: couldn't update"
3390 " array info. %d\n", err);
3395 if (!list_empty(&mddev->disks)) {
3397 "md: array %s already has disks!\n",
3402 if (mddev->raid_disks) {
3404 "md: array %s already initialised!\n",
3409 err = set_array_info(mddev, &info);
3411 printk(KERN_WARNING "md: couldn't set"
3412 " array info. %d\n", err);
3422 * Commands querying/configuring an existing array:
3424 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3425 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3426 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3427 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3433 * Commands even a read-only array can execute:
3437 case GET_ARRAY_INFO:
3438 err = get_array_info(mddev, argp);
3441 case GET_BITMAP_FILE:
3442 err = get_bitmap_file(mddev, argp);
3446 err = get_disk_info(mddev, argp);
3449 case RESTART_ARRAY_RW:
3450 err = restart_array(mddev);
3454 err = do_md_stop (mddev, 0);
3458 err = do_md_stop (mddev, 1);
3462 * We have a problem here : there is no easy way to give a CHS
3463 * virtual geometry. We currently pretend that we have a 2 heads
3464 * 4 sectors (with a BIG number of cylinders...). This drives
3465 * dosfs just mad... ;-)
3472 err = put_user (2, (char __user *) &loc->heads);
3475 err = put_user (4, (char __user *) &loc->sectors);
3478 err = put_user(get_capacity(mddev->gendisk)/8,
3479 (short __user *) &loc->cylinders);
3482 err = put_user (get_start_sect(inode->i_bdev),
3483 (long __user *) &loc->start);
3488 * The remaining ioctls are changing the state of the
3489 * superblock, so we do not allow them on read-only arrays.
3490 * However non-MD ioctls (e.g. get-size) will still come through
3491 * here and hit the 'default' below, so only disallow
3492 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3494 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3495 mddev->ro && mddev->pers) {
3496 if (mddev->ro == 2) {
3498 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3499 md_wakeup_thread(mddev->thread);
3511 mdu_disk_info_t info;
3512 if (copy_from_user(&info, argp, sizeof(info)))
3515 err = add_new_disk(mddev, &info);
3519 case HOT_REMOVE_DISK:
3520 err = hot_remove_disk(mddev, new_decode_dev(arg));
3524 err = hot_add_disk(mddev, new_decode_dev(arg));
3527 case SET_DISK_FAULTY:
3528 err = set_disk_faulty(mddev, new_decode_dev(arg));
3532 err = do_md_run (mddev);
3535 case SET_BITMAP_FILE:
3536 err = set_bitmap_file(mddev, (int)arg);
3540 if (_IOC_TYPE(cmd) == MD_MAJOR)
3541 printk(KERN_WARNING "md: %s(pid %d) used"
3542 " obsolete MD ioctl, upgrade your"
3543 " software to use new ictls.\n",
3544 current->comm, current->pid);
3551 mddev_unlock(mddev);
3561 static int md_open(struct inode *inode, struct file *file)
3564 * Succeed if we can lock the mddev, which confirms that
3565 * it isn't being stopped right now.
3567 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3570 if ((err = mddev_lock(mddev)))
3575 mddev_unlock(mddev);
3577 check_disk_change(inode->i_bdev);
3582 static int md_release(struct inode *inode, struct file * file)
3584 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3593 static int md_media_changed(struct gendisk *disk)
3595 mddev_t *mddev = disk->private_data;
3597 return mddev->changed;
3600 static int md_revalidate(struct gendisk *disk)
3602 mddev_t *mddev = disk->private_data;
3607 static struct block_device_operations md_fops =
3609 .owner = THIS_MODULE,
3611 .release = md_release,
3613 .media_changed = md_media_changed,
3614 .revalidate_disk= md_revalidate,
3617 static int md_thread(void * arg)
3619 mdk_thread_t *thread = arg;
3622 * md_thread is a 'system-thread', it's priority should be very
3623 * high. We avoid resource deadlocks individually in each
3624 * raid personality. (RAID5 does preallocation) We also use RR and
3625 * the very same RT priority as kswapd, thus we will never get
3626 * into a priority inversion deadlock.
3628 * we definitely have to have equal or higher priority than
3629 * bdflush, otherwise bdflush will deadlock if there are too
3630 * many dirty RAID5 blocks.
3633 allow_signal(SIGKILL);
3634 while (!kthread_should_stop()) {
3636 /* We need to wait INTERRUPTIBLE so that
3637 * we don't add to the load-average.
3638 * That means we need to be sure no signals are
3641 if (signal_pending(current))
3642 flush_signals(current);
3644 wait_event_interruptible_timeout
3646 test_bit(THREAD_WAKEUP, &thread->flags)
3647 || kthread_should_stop(),
3651 clear_bit(THREAD_WAKEUP, &thread->flags);
3653 thread->run(thread->mddev);
3659 void md_wakeup_thread(mdk_thread_t *thread)
3662 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3663 set_bit(THREAD_WAKEUP, &thread->flags);
3664 wake_up(&thread->wqueue);
3668 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3671 mdk_thread_t *thread;
3673 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3677 init_waitqueue_head(&thread->wqueue);
3680 thread->mddev = mddev;
3681 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3682 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3683 if (IS_ERR(thread->tsk)) {
3690 void md_unregister_thread(mdk_thread_t *thread)
3692 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3694 kthread_stop(thread->tsk);
3698 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3705 if (!rdev || test_bit(Faulty, &rdev->flags))
3708 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3710 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3711 __builtin_return_address(0),__builtin_return_address(1),
3712 __builtin_return_address(2),__builtin_return_address(3));
3714 if (!mddev->pers->error_handler)
3716 mddev->pers->error_handler(mddev,rdev);
3717 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3718 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3719 md_wakeup_thread(mddev->thread);
3720 md_new_event(mddev);
3723 /* seq_file implementation /proc/mdstat */
3725 static void status_unused(struct seq_file *seq)
3729 struct list_head *tmp;
3731 seq_printf(seq, "unused devices: ");
3733 ITERATE_RDEV_PENDING(rdev,tmp) {
3734 char b[BDEVNAME_SIZE];
3736 seq_printf(seq, "%s ",
3737 bdevname(rdev->bdev,b));
3740 seq_printf(seq, "<none>");
3742 seq_printf(seq, "\n");
3746 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3748 unsigned long max_blocks, resync, res, dt, db, rt;
3750 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3752 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3753 max_blocks = mddev->resync_max_sectors >> 1;
3755 max_blocks = mddev->size;
3758 * Should not happen.
3764 res = (resync/1024)*1000/(max_blocks/1024 + 1);
3766 int i, x = res/50, y = 20-x;
3767 seq_printf(seq, "[");
3768 for (i = 0; i < x; i++)
3769 seq_printf(seq, "=");
3770 seq_printf(seq, ">");
3771 for (i = 0; i < y; i++)
3772 seq_printf(seq, ".");
3773 seq_printf(seq, "] ");
3775 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3776 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3777 "resync" : "recovery"),
3778 res/10, res % 10, resync, max_blocks);
3781 * We do not want to overflow, so the order of operands and
3782 * the * 100 / 100 trick are important. We do a +1 to be
3783 * safe against division by zero. We only estimate anyway.
3785 * dt: time from mark until now
3786 * db: blocks written from mark until now
3787 * rt: remaining time
3789 dt = ((jiffies - mddev->resync_mark) / HZ);
3791 db = resync - (mddev->resync_mark_cnt/2);
3792 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3794 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3796 seq_printf(seq, " speed=%ldK/sec", db/dt);
3799 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3801 struct list_head *tmp;
3811 spin_lock(&all_mddevs_lock);
3812 list_for_each(tmp,&all_mddevs)
3814 mddev = list_entry(tmp, mddev_t, all_mddevs);
3816 spin_unlock(&all_mddevs_lock);
3819 spin_unlock(&all_mddevs_lock);
3821 return (void*)2;/* tail */
3825 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3827 struct list_head *tmp;
3828 mddev_t *next_mddev, *mddev = v;
3834 spin_lock(&all_mddevs_lock);
3836 tmp = all_mddevs.next;
3838 tmp = mddev->all_mddevs.next;
3839 if (tmp != &all_mddevs)
3840 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3842 next_mddev = (void*)2;
3845 spin_unlock(&all_mddevs_lock);
3853 static void md_seq_stop(struct seq_file *seq, void *v)
3857 if (mddev && v != (void*)1 && v != (void*)2)
3861 struct mdstat_info {
3865 static int md_seq_show(struct seq_file *seq, void *v)
3869 struct list_head *tmp2;
3871 struct mdstat_info *mi = seq->private;
3872 struct bitmap *bitmap;
3874 if (v == (void*)1) {
3875 struct mdk_personality *pers;
3876 seq_printf(seq, "Personalities : ");
3877 spin_lock(&pers_lock);
3878 list_for_each_entry(pers, &pers_list, list)
3879 seq_printf(seq, "[%s] ", pers->name);
3881 spin_unlock(&pers_lock);
3882 seq_printf(seq, "\n");
3883 mi->event = atomic_read(&md_event_count);
3886 if (v == (void*)2) {
3891 if (mddev_lock(mddev)!=0)
3893 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3894 seq_printf(seq, "%s : %sactive", mdname(mddev),
3895 mddev->pers ? "" : "in");
3898 seq_printf(seq, " (read-only)");
3900 seq_printf(seq, "(auto-read-only)");
3901 seq_printf(seq, " %s", mddev->pers->name);
3905 ITERATE_RDEV(mddev,rdev,tmp2) {
3906 char b[BDEVNAME_SIZE];
3907 seq_printf(seq, " %s[%d]",
3908 bdevname(rdev->bdev,b), rdev->desc_nr);
3909 if (test_bit(WriteMostly, &rdev->flags))
3910 seq_printf(seq, "(W)");
3911 if (test_bit(Faulty, &rdev->flags)) {
3912 seq_printf(seq, "(F)");
3914 } else if (rdev->raid_disk < 0)
3915 seq_printf(seq, "(S)"); /* spare */
3919 if (!list_empty(&mddev->disks)) {
3921 seq_printf(seq, "\n %llu blocks",
3922 (unsigned long long)mddev->array_size);
3924 seq_printf(seq, "\n %llu blocks",
3925 (unsigned long long)size);
3927 if (mddev->persistent) {
3928 if (mddev->major_version != 0 ||
3929 mddev->minor_version != 90) {
3930 seq_printf(seq," super %d.%d",
3931 mddev->major_version,
3932 mddev->minor_version);
3935 seq_printf(seq, " super non-persistent");
3938 mddev->pers->status (seq, mddev);
3939 seq_printf(seq, "\n ");
3940 if (mddev->pers->sync_request) {
3941 if (mddev->curr_resync > 2) {
3942 status_resync (seq, mddev);
3943 seq_printf(seq, "\n ");
3944 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3945 seq_printf(seq, "\tresync=DELAYED\n ");
3946 else if (mddev->recovery_cp < MaxSector)
3947 seq_printf(seq, "\tresync=PENDING\n ");
3950 seq_printf(seq, "\n ");
3952 if ((bitmap = mddev->bitmap)) {
3953 unsigned long chunk_kb;
3954 unsigned long flags;
3955 spin_lock_irqsave(&bitmap->lock, flags);
3956 chunk_kb = bitmap->chunksize >> 10;
3957 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3959 bitmap->pages - bitmap->missing_pages,
3961 (bitmap->pages - bitmap->missing_pages)
3962 << (PAGE_SHIFT - 10),
3963 chunk_kb ? chunk_kb : bitmap->chunksize,
3964 chunk_kb ? "KB" : "B");
3966 seq_printf(seq, ", file: ");
3967 seq_path(seq, bitmap->file->f_vfsmnt,
3968 bitmap->file->f_dentry," \t\n");
3971 seq_printf(seq, "\n");
3972 spin_unlock_irqrestore(&bitmap->lock, flags);
3975 seq_printf(seq, "\n");
3977 mddev_unlock(mddev);
3982 static struct seq_operations md_seq_ops = {
3983 .start = md_seq_start,
3984 .next = md_seq_next,
3985 .stop = md_seq_stop,
3986 .show = md_seq_show,
3989 static int md_seq_open(struct inode *inode, struct file *file)
3992 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
3996 error = seq_open(file, &md_seq_ops);
4000 struct seq_file *p = file->private_data;
4002 mi->event = atomic_read(&md_event_count);
4007 static int md_seq_release(struct inode *inode, struct file *file)
4009 struct seq_file *m = file->private_data;
4010 struct mdstat_info *mi = m->private;
4013 return seq_release(inode, file);
4016 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4018 struct seq_file *m = filp->private_data;
4019 struct mdstat_info *mi = m->private;
4022 poll_wait(filp, &md_event_waiters, wait);
4024 /* always allow read */
4025 mask = POLLIN | POLLRDNORM;
4027 if (mi->event != atomic_read(&md_event_count))
4028 mask |= POLLERR | POLLPRI;
4032 static struct file_operations md_seq_fops = {
4033 .open = md_seq_open,
4035 .llseek = seq_lseek,
4036 .release = md_seq_release,
4037 .poll = mdstat_poll,
4040 int register_md_personality(struct mdk_personality *p)
4042 spin_lock(&pers_lock);
4043 list_add_tail(&p->list, &pers_list);
4044 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4045 spin_unlock(&pers_lock);
4049 int unregister_md_personality(struct mdk_personality *p)
4051 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4052 spin_lock(&pers_lock);
4053 list_del_init(&p->list);
4054 spin_unlock(&pers_lock);
4058 static int is_mddev_idle(mddev_t *mddev)
4061 struct list_head *tmp;
4063 unsigned long curr_events;
4066 ITERATE_RDEV(mddev,rdev,tmp) {
4067 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4068 curr_events = disk_stat_read(disk, sectors[0]) +
4069 disk_stat_read(disk, sectors[1]) -
4070 atomic_read(&disk->sync_io);
4071 /* The difference between curr_events and last_events
4072 * will be affected by any new non-sync IO (making
4073 * curr_events bigger) and any difference in the amount of
4074 * in-flight syncio (making current_events bigger or smaller)
4075 * The amount in-flight is currently limited to
4076 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4077 * which is at most 4096 sectors.
4078 * These numbers are fairly fragile and should be made
4079 * more robust, probably by enforcing the
4080 * 'window size' that md_do_sync sort-of uses.
4082 * Note: the following is an unsigned comparison.
4084 if ((curr_events - rdev->last_events + 4096) > 8192) {
4085 rdev->last_events = curr_events;
4092 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4094 /* another "blocks" (512byte) blocks have been synced */
4095 atomic_sub(blocks, &mddev->recovery_active);
4096 wake_up(&mddev->recovery_wait);
4098 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4099 md_wakeup_thread(mddev->thread);
4100 // stop recovery, signal do_sync ....
4105 /* md_write_start(mddev, bi)
4106 * If we need to update some array metadata (e.g. 'active' flag
4107 * in superblock) before writing, schedule a superblock update
4108 * and wait for it to complete.
4110 void md_write_start(mddev_t *mddev, struct bio *bi)
4112 if (bio_data_dir(bi) != WRITE)
4115 BUG_ON(mddev->ro == 1);
4116 if (mddev->ro == 2) {
4117 /* need to switch to read/write */
4119 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4120 md_wakeup_thread(mddev->thread);
4122 atomic_inc(&mddev->writes_pending);
4123 if (mddev->in_sync) {
4124 spin_lock_irq(&mddev->write_lock);
4125 if (mddev->in_sync) {
4127 mddev->sb_dirty = 1;
4128 md_wakeup_thread(mddev->thread);
4130 spin_unlock_irq(&mddev->write_lock);
4132 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4135 void md_write_end(mddev_t *mddev)
4137 if (atomic_dec_and_test(&mddev->writes_pending)) {
4138 if (mddev->safemode == 2)
4139 md_wakeup_thread(mddev->thread);
4141 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4145 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4147 #define SYNC_MARKS 10
4148 #define SYNC_MARK_STEP (3*HZ)
4149 static void md_do_sync(mddev_t *mddev)
4152 unsigned int currspeed = 0,
4154 sector_t max_sectors,j, io_sectors;
4155 unsigned long mark[SYNC_MARKS];
4156 sector_t mark_cnt[SYNC_MARKS];
4158 struct list_head *tmp;
4159 sector_t last_check;
4162 /* just incase thread restarts... */
4163 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4166 /* we overload curr_resync somewhat here.
4167 * 0 == not engaged in resync at all
4168 * 2 == checking that there is no conflict with another sync
4169 * 1 == like 2, but have yielded to allow conflicting resync to
4171 * other == active in resync - this many blocks
4173 * Before starting a resync we must have set curr_resync to
4174 * 2, and then checked that every "conflicting" array has curr_resync
4175 * less than ours. When we find one that is the same or higher
4176 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4177 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4178 * This will mean we have to start checking from the beginning again.
4183 mddev->curr_resync = 2;
4186 if (kthread_should_stop()) {
4187 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4190 ITERATE_MDDEV(mddev2,tmp) {
4191 if (mddev2 == mddev)
4193 if (mddev2->curr_resync &&
4194 match_mddev_units(mddev,mddev2)) {
4196 if (mddev < mddev2 && mddev->curr_resync == 2) {
4197 /* arbitrarily yield */
4198 mddev->curr_resync = 1;
4199 wake_up(&resync_wait);
4201 if (mddev > mddev2 && mddev->curr_resync == 1)
4202 /* no need to wait here, we can wait the next
4203 * time 'round when curr_resync == 2
4206 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4207 if (!kthread_should_stop() &&
4208 mddev2->curr_resync >= mddev->curr_resync) {
4209 printk(KERN_INFO "md: delaying resync of %s"
4210 " until %s has finished resync (they"
4211 " share one or more physical units)\n",
4212 mdname(mddev), mdname(mddev2));
4215 finish_wait(&resync_wait, &wq);
4218 finish_wait(&resync_wait, &wq);
4221 } while (mddev->curr_resync < 2);
4223 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4224 /* resync follows the size requested by the personality,
4225 * which defaults to physical size, but can be virtual size
4227 max_sectors = mddev->resync_max_sectors;
4228 mddev->resync_mismatches = 0;
4230 /* recovery follows the physical size of devices */
4231 max_sectors = mddev->size << 1;
4233 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4234 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4235 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
4236 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4237 "(but not more than %d KB/sec) for reconstruction.\n",
4238 sysctl_speed_limit_max);
4240 is_mddev_idle(mddev); /* this also initializes IO event counters */
4241 /* we don't use the checkpoint if there's a bitmap */
4242 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4243 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4244 j = mddev->recovery_cp;
4248 for (m = 0; m < SYNC_MARKS; m++) {
4250 mark_cnt[m] = io_sectors;
4253 mddev->resync_mark = mark[last_mark];
4254 mddev->resync_mark_cnt = mark_cnt[last_mark];
4257 * Tune reconstruction:
4259 window = 32*(PAGE_SIZE/512);
4260 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4261 window/2,(unsigned long long) max_sectors/2);
4263 atomic_set(&mddev->recovery_active, 0);
4264 init_waitqueue_head(&mddev->recovery_wait);
4269 "md: resuming recovery of %s from checkpoint.\n",
4271 mddev->curr_resync = j;
4274 while (j < max_sectors) {
4278 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4279 currspeed < sysctl_speed_limit_min);
4281 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4285 if (!skipped) { /* actual IO requested */
4286 io_sectors += sectors;
4287 atomic_add(sectors, &mddev->recovery_active);
4291 if (j>1) mddev->curr_resync = j;
4292 if (last_check == 0)
4293 /* this is the earliers that rebuilt will be
4294 * visible in /proc/mdstat
4296 md_new_event(mddev);
4298 if (last_check + window > io_sectors || j == max_sectors)
4301 last_check = io_sectors;
4303 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4304 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4308 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4310 int next = (last_mark+1) % SYNC_MARKS;
4312 mddev->resync_mark = mark[next];
4313 mddev->resync_mark_cnt = mark_cnt[next];
4314 mark[next] = jiffies;
4315 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4320 if (kthread_should_stop()) {
4322 * got a signal, exit.
4325 "md: md_do_sync() got signal ... exiting\n");
4326 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4331 * this loop exits only if either when we are slower than
4332 * the 'hard' speed limit, or the system was IO-idle for
4334 * the system might be non-idle CPU-wise, but we only care
4335 * about not overloading the IO subsystem. (things like an
4336 * e2fsck being done on the RAID array should execute fast)
4338 mddev->queue->unplug_fn(mddev->queue);
4341 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4342 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4344 if (currspeed > sysctl_speed_limit_min) {
4345 if ((currspeed > sysctl_speed_limit_max) ||
4346 !is_mddev_idle(mddev)) {
4352 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4354 * this also signals 'finished resyncing' to md_stop
4357 mddev->queue->unplug_fn(mddev->queue);
4359 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4361 /* tell personality that we are finished */
4362 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4364 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4365 mddev->curr_resync > 2 &&
4366 mddev->curr_resync >= mddev->recovery_cp) {
4367 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4369 "md: checkpointing recovery of %s.\n",
4371 mddev->recovery_cp = mddev->curr_resync;
4373 mddev->recovery_cp = MaxSector;
4377 mddev->curr_resync = 0;
4378 wake_up(&resync_wait);
4379 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4380 md_wakeup_thread(mddev->thread);
4385 * This routine is regularly called by all per-raid-array threads to
4386 * deal with generic issues like resync and super-block update.
4387 * Raid personalities that don't have a thread (linear/raid0) do not
4388 * need this as they never do any recovery or update the superblock.
4390 * It does not do any resync itself, but rather "forks" off other threads
4391 * to do that as needed.
4392 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4393 * "->recovery" and create a thread at ->sync_thread.
4394 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4395 * and wakeups up this thread which will reap the thread and finish up.
4396 * This thread also removes any faulty devices (with nr_pending == 0).
4398 * The overall approach is:
4399 * 1/ if the superblock needs updating, update it.
4400 * 2/ If a recovery thread is running, don't do anything else.
4401 * 3/ If recovery has finished, clean up, possibly marking spares active.
4402 * 4/ If there are any faulty devices, remove them.
4403 * 5/ If array is degraded, try to add spares devices
4404 * 6/ If array has spares or is not in-sync, start a resync thread.
4406 void md_check_recovery(mddev_t *mddev)
4409 struct list_head *rtmp;
4413 bitmap_daemon_work(mddev->bitmap);
4418 if (signal_pending(current)) {
4419 if (mddev->pers->sync_request) {
4420 printk(KERN_INFO "md: %s in immediate safe mode\n",
4422 mddev->safemode = 2;
4424 flush_signals(current);
4429 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4430 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4431 (mddev->safemode == 1) ||
4432 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4433 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4437 if (mddev_trylock(mddev)==0) {
4440 spin_lock_irq(&mddev->write_lock);
4441 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4442 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4444 mddev->sb_dirty = 1;
4446 if (mddev->safemode == 1)
4447 mddev->safemode = 0;
4448 spin_unlock_irq(&mddev->write_lock);
4450 if (mddev->sb_dirty)
4451 md_update_sb(mddev);
4454 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4455 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4456 /* resync/recovery still happening */
4457 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4460 if (mddev->sync_thread) {
4461 /* resync has finished, collect result */
4462 md_unregister_thread(mddev->sync_thread);
4463 mddev->sync_thread = NULL;
4464 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4465 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4467 /* activate any spares */
4468 mddev->pers->spare_active(mddev);
4470 md_update_sb(mddev);
4472 /* if array is no-longer degraded, then any saved_raid_disk
4473 * information must be scrapped
4475 if (!mddev->degraded)
4476 ITERATE_RDEV(mddev,rdev,rtmp)
4477 rdev->saved_raid_disk = -1;
4479 mddev->recovery = 0;
4480 /* flag recovery needed just to double check */
4481 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4482 md_new_event(mddev);
4485 /* Clear some bits that don't mean anything, but
4488 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4489 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4490 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4491 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4493 /* no recovery is running.
4494 * remove any failed drives, then
4495 * add spares if possible.
4496 * Spare are also removed and re-added, to allow
4497 * the personality to fail the re-add.
4499 ITERATE_RDEV(mddev,rdev,rtmp)
4500 if (rdev->raid_disk >= 0 &&
4501 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4502 atomic_read(&rdev->nr_pending)==0) {
4503 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4505 sprintf(nm,"rd%d", rdev->raid_disk);
4506 sysfs_remove_link(&mddev->kobj, nm);
4507 rdev->raid_disk = -1;
4511 if (mddev->degraded) {
4512 ITERATE_RDEV(mddev,rdev,rtmp)
4513 if (rdev->raid_disk < 0
4514 && !test_bit(Faulty, &rdev->flags)) {
4515 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4517 sprintf(nm, "rd%d", rdev->raid_disk);
4518 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4520 md_new_event(mddev);
4527 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4528 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4529 } else if (mddev->recovery_cp < MaxSector) {
4530 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4531 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4532 /* nothing to be done ... */
4535 if (mddev->pers->sync_request) {
4536 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4537 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4538 /* We are adding a device or devices to an array
4539 * which has the bitmap stored on all devices.
4540 * So make sure all bitmap pages get written
4542 bitmap_write_all(mddev->bitmap);
4544 mddev->sync_thread = md_register_thread(md_do_sync,
4547 if (!mddev->sync_thread) {
4548 printk(KERN_ERR "%s: could not start resync"
4551 /* leave the spares where they are, it shouldn't hurt */
4552 mddev->recovery = 0;
4554 md_wakeup_thread(mddev->sync_thread);
4555 md_new_event(mddev);
4558 mddev_unlock(mddev);
4562 static int md_notify_reboot(struct notifier_block *this,
4563 unsigned long code, void *x)
4565 struct list_head *tmp;
4568 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4570 printk(KERN_INFO "md: stopping all md devices.\n");
4572 ITERATE_MDDEV(mddev,tmp)
4573 if (mddev_trylock(mddev)==0)
4574 do_md_stop (mddev, 1);
4576 * certain more exotic SCSI devices are known to be
4577 * volatile wrt too early system reboots. While the
4578 * right place to handle this issue is the given
4579 * driver, we do want to have a safe RAID driver ...
4586 static struct notifier_block md_notifier = {
4587 .notifier_call = md_notify_reboot,
4589 .priority = INT_MAX, /* before any real devices */
4592 static void md_geninit(void)
4594 struct proc_dir_entry *p;
4596 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4598 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4600 p->proc_fops = &md_seq_fops;
4603 static int __init md_init(void)
4607 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4608 " MD_SB_DISKS=%d\n",
4609 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4610 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4611 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4614 if (register_blkdev(MAJOR_NR, "md"))
4616 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4617 unregister_blkdev(MAJOR_NR, "md");
4621 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4622 md_probe, NULL, NULL);
4623 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4624 md_probe, NULL, NULL);
4626 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4627 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4628 S_IFBLK|S_IRUSR|S_IWUSR,
4631 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4632 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4633 S_IFBLK|S_IRUSR|S_IWUSR,
4637 register_reboot_notifier(&md_notifier);
4638 raid_table_header = register_sysctl_table(raid_root_table, 1);
4648 * Searches all registered partitions for autorun RAID arrays
4651 static dev_t detected_devices[128];
4654 void md_autodetect_dev(dev_t dev)
4656 if (dev_cnt >= 0 && dev_cnt < 127)
4657 detected_devices[dev_cnt++] = dev;
4661 static void autostart_arrays(int part)
4666 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4668 for (i = 0; i < dev_cnt; i++) {
4669 dev_t dev = detected_devices[i];
4671 rdev = md_import_device(dev,0, 0);
4675 if (test_bit(Faulty, &rdev->flags)) {
4679 list_add(&rdev->same_set, &pending_raid_disks);
4683 autorun_devices(part);
4688 static __exit void md_exit(void)
4691 struct list_head *tmp;
4693 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4694 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4695 for (i=0; i < MAX_MD_DEVS; i++)
4696 devfs_remove("md/%d", i);
4697 for (i=0; i < MAX_MD_DEVS; i++)
4698 devfs_remove("md/d%d", i);
4702 unregister_blkdev(MAJOR_NR,"md");
4703 unregister_blkdev(mdp_major, "mdp");
4704 unregister_reboot_notifier(&md_notifier);
4705 unregister_sysctl_table(raid_table_header);
4706 remove_proc_entry("mdstat", NULL);
4707 ITERATE_MDDEV(mddev,tmp) {
4708 struct gendisk *disk = mddev->gendisk;
4711 export_array(mddev);
4714 mddev->gendisk = NULL;
4719 module_init(md_init)
4720 module_exit(md_exit)
4722 static int get_ro(char *buffer, struct kernel_param *kp)
4724 return sprintf(buffer, "%d", start_readonly);
4726 static int set_ro(const char *val, struct kernel_param *kp)
4729 int num = simple_strtoul(val, &e, 10);
4730 if (*val && (*e == '\0' || *e == '\n')) {
4731 start_readonly = num;
4737 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4738 module_param(start_dirty_degraded, int, 0644);
4741 EXPORT_SYMBOL(register_md_personality);
4742 EXPORT_SYMBOL(unregister_md_personality);
4743 EXPORT_SYMBOL(md_error);
4744 EXPORT_SYMBOL(md_done_sync);
4745 EXPORT_SYMBOL(md_write_start);
4746 EXPORT_SYMBOL(md_write_end);
4747 EXPORT_SYMBOL(md_register_thread);
4748 EXPORT_SYMBOL(md_unregister_thread);
4749 EXPORT_SYMBOL(md_wakeup_thread);
4750 EXPORT_SYMBOL(md_print_devices);
4751 EXPORT_SYMBOL(md_check_recovery);
4752 MODULE_LICENSE("GPL");
4754 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);