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/kernel.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/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part);
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
75 static void md_print_devices(void);
77 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
79 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
82 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
83 * is 1000 KB/sec, so the extra system load does not show up that much.
84 * Increase it if you want to have more _guaranteed_ speed. Note that
85 * the RAID driver will use the maximum available bandwidth if the IO
86 * subsystem is idle. There is also an 'absolute maximum' reconstruction
87 * speed limit - in case reconstruction slows down your system despite
90 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
91 * or /sys/block/mdX/md/sync_speed_{min,max}
94 static int sysctl_speed_limit_min = 1000;
95 static int sysctl_speed_limit_max = 200000;
96 static inline int speed_min(mddev_t *mddev)
98 return mddev->sync_speed_min ?
99 mddev->sync_speed_min : sysctl_speed_limit_min;
102 static inline int speed_max(mddev_t *mddev)
104 return mddev->sync_speed_max ?
105 mddev->sync_speed_max : sysctl_speed_limit_max;
108 static struct ctl_table_header *raid_table_header;
110 static ctl_table raid_table[] = {
112 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
113 .procname = "speed_limit_min",
114 .data = &sysctl_speed_limit_min,
115 .maxlen = sizeof(int),
116 .mode = S_IRUGO|S_IWUSR,
117 .proc_handler = &proc_dointvec,
120 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
124 .mode = S_IRUGO|S_IWUSR,
125 .proc_handler = &proc_dointvec,
130 static ctl_table raid_dir_table[] = {
132 .ctl_name = DEV_RAID,
135 .mode = S_IRUGO|S_IXUGO,
141 static ctl_table raid_root_table[] = {
147 .child = raid_dir_table,
152 static struct block_device_operations md_fops;
154 static int start_readonly;
157 * We have a system wide 'event count' that is incremented
158 * on any 'interesting' event, and readers of /proc/mdstat
159 * can use 'poll' or 'select' to find out when the event
163 * start array, stop array, error, add device, remove device,
164 * start build, activate spare
166 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
167 static atomic_t md_event_count;
168 void md_new_event(mddev_t *mddev)
170 atomic_inc(&md_event_count);
171 wake_up(&md_event_waiters);
172 sysfs_notify(&mddev->kobj, NULL, "sync_action");
174 EXPORT_SYMBOL_GPL(md_new_event);
176 /* Alternate version that can be called from interrupts
177 * when calling sysfs_notify isn't needed.
179 static void md_new_event_inintr(mddev_t *mddev)
181 atomic_inc(&md_event_count);
182 wake_up(&md_event_waiters);
186 * Enables to iterate over all existing md arrays
187 * all_mddevs_lock protects this list.
189 static LIST_HEAD(all_mddevs);
190 static DEFINE_SPINLOCK(all_mddevs_lock);
194 * iterates through all used mddevs in the system.
195 * We take care to grab the all_mddevs_lock whenever navigating
196 * the list, and to always hold a refcount when unlocked.
197 * Any code which breaks out of this loop while own
198 * a reference to the current mddev and must mddev_put it.
200 #define for_each_mddev(mddev,tmp) \
202 for (({ spin_lock(&all_mddevs_lock); \
203 tmp = all_mddevs.next; \
205 ({ if (tmp != &all_mddevs) \
206 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
207 spin_unlock(&all_mddevs_lock); \
208 if (mddev) mddev_put(mddev); \
209 mddev = list_entry(tmp, mddev_t, all_mddevs); \
210 tmp != &all_mddevs;}); \
211 ({ spin_lock(&all_mddevs_lock); \
216 static int md_fail_request (struct request_queue *q, struct bio *bio)
222 static inline mddev_t *mddev_get(mddev_t *mddev)
224 atomic_inc(&mddev->active);
228 static void mddev_put(mddev_t *mddev)
230 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
232 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
233 list_del(&mddev->all_mddevs);
234 spin_unlock(&all_mddevs_lock);
235 blk_cleanup_queue(mddev->queue);
236 kobject_put(&mddev->kobj);
238 spin_unlock(&all_mddevs_lock);
241 static mddev_t * mddev_find(dev_t unit)
243 mddev_t *mddev, *new = NULL;
246 spin_lock(&all_mddevs_lock);
247 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
248 if (mddev->unit == unit) {
250 spin_unlock(&all_mddevs_lock);
256 list_add(&new->all_mddevs, &all_mddevs);
257 spin_unlock(&all_mddevs_lock);
260 spin_unlock(&all_mddevs_lock);
262 new = kzalloc(sizeof(*new), GFP_KERNEL);
267 if (MAJOR(unit) == MD_MAJOR)
268 new->md_minor = MINOR(unit);
270 new->md_minor = MINOR(unit) >> MdpMinorShift;
272 mutex_init(&new->reconfig_mutex);
273 INIT_LIST_HEAD(&new->disks);
274 INIT_LIST_HEAD(&new->all_mddevs);
275 init_timer(&new->safemode_timer);
276 atomic_set(&new->active, 1);
277 spin_lock_init(&new->write_lock);
278 init_waitqueue_head(&new->sb_wait);
279 init_waitqueue_head(&new->recovery_wait);
280 new->reshape_position = MaxSector;
282 new->resync_max = MaxSector;
283 new->level = LEVEL_NONE;
285 new->queue = blk_alloc_queue(GFP_KERNEL);
290 /* Can be unlocked because the queue is new: no concurrency */
291 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
293 blk_queue_make_request(new->queue, md_fail_request);
298 static inline int mddev_lock(mddev_t * mddev)
300 return mutex_lock_interruptible(&mddev->reconfig_mutex);
303 static inline int mddev_trylock(mddev_t * mddev)
305 return mutex_trylock(&mddev->reconfig_mutex);
308 static inline void mddev_unlock(mddev_t * mddev)
310 mutex_unlock(&mddev->reconfig_mutex);
312 md_wakeup_thread(mddev->thread);
315 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
318 struct list_head *tmp;
320 rdev_for_each(rdev, tmp, mddev) {
321 if (rdev->desc_nr == nr)
327 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
329 struct list_head *tmp;
332 rdev_for_each(rdev, tmp, mddev) {
333 if (rdev->bdev->bd_dev == dev)
339 static struct mdk_personality *find_pers(int level, char *clevel)
341 struct mdk_personality *pers;
342 list_for_each_entry(pers, &pers_list, list) {
343 if (level != LEVEL_NONE && pers->level == level)
345 if (strcmp(pers->name, clevel)==0)
351 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
353 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
354 return MD_NEW_SIZE_BLOCKS(size);
357 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
361 size = rdev->sb_offset;
364 size &= ~((sector_t)chunk_size/1024 - 1);
368 static int alloc_disk_sb(mdk_rdev_t * rdev)
373 rdev->sb_page = alloc_page(GFP_KERNEL);
374 if (!rdev->sb_page) {
375 printk(KERN_ALERT "md: out of memory.\n");
382 static void free_disk_sb(mdk_rdev_t * rdev)
385 put_page(rdev->sb_page);
387 rdev->sb_page = NULL;
394 static void super_written(struct bio *bio, int error)
396 mdk_rdev_t *rdev = bio->bi_private;
397 mddev_t *mddev = rdev->mddev;
399 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
400 printk("md: super_written gets error=%d, uptodate=%d\n",
401 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
402 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
403 md_error(mddev, rdev);
406 if (atomic_dec_and_test(&mddev->pending_writes))
407 wake_up(&mddev->sb_wait);
411 static void super_written_barrier(struct bio *bio, int error)
413 struct bio *bio2 = bio->bi_private;
414 mdk_rdev_t *rdev = bio2->bi_private;
415 mddev_t *mddev = rdev->mddev;
417 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
418 error == -EOPNOTSUPP) {
420 /* barriers don't appear to be supported :-( */
421 set_bit(BarriersNotsupp, &rdev->flags);
422 mddev->barriers_work = 0;
423 spin_lock_irqsave(&mddev->write_lock, flags);
424 bio2->bi_next = mddev->biolist;
425 mddev->biolist = bio2;
426 spin_unlock_irqrestore(&mddev->write_lock, flags);
427 wake_up(&mddev->sb_wait);
431 bio->bi_private = rdev;
432 super_written(bio, error);
436 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
437 sector_t sector, int size, struct page *page)
439 /* write first size bytes of page to sector of rdev
440 * Increment mddev->pending_writes before returning
441 * and decrement it on completion, waking up sb_wait
442 * if zero is reached.
443 * If an error occurred, call md_error
445 * As we might need to resubmit the request if BIO_RW_BARRIER
446 * causes ENOTSUPP, we allocate a spare bio...
448 struct bio *bio = bio_alloc(GFP_NOIO, 1);
449 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
451 bio->bi_bdev = rdev->bdev;
452 bio->bi_sector = sector;
453 bio_add_page(bio, page, size, 0);
454 bio->bi_private = rdev;
455 bio->bi_end_io = super_written;
458 atomic_inc(&mddev->pending_writes);
459 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
461 rw |= (1<<BIO_RW_BARRIER);
462 rbio = bio_clone(bio, GFP_NOIO);
463 rbio->bi_private = bio;
464 rbio->bi_end_io = super_written_barrier;
465 submit_bio(rw, rbio);
470 void md_super_wait(mddev_t *mddev)
472 /* wait for all superblock writes that were scheduled to complete.
473 * if any had to be retried (due to BARRIER problems), retry them
477 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
478 if (atomic_read(&mddev->pending_writes)==0)
480 while (mddev->biolist) {
482 spin_lock_irq(&mddev->write_lock);
483 bio = mddev->biolist;
484 mddev->biolist = bio->bi_next ;
486 spin_unlock_irq(&mddev->write_lock);
487 submit_bio(bio->bi_rw, bio);
491 finish_wait(&mddev->sb_wait, &wq);
494 static void bi_complete(struct bio *bio, int error)
496 complete((struct completion*)bio->bi_private);
499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
500 struct page *page, int rw)
502 struct bio *bio = bio_alloc(GFP_NOIO, 1);
503 struct completion event;
506 rw |= (1 << BIO_RW_SYNC);
509 bio->bi_sector = sector;
510 bio_add_page(bio, page, size, 0);
511 init_completion(&event);
512 bio->bi_private = &event;
513 bio->bi_end_io = bi_complete;
515 wait_for_completion(&event);
517 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
521 EXPORT_SYMBOL_GPL(sync_page_io);
523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
525 char b[BDEVNAME_SIZE];
526 if (!rdev->sb_page) {
534 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
540 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
541 bdevname(rdev->bdev,b));
545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
547 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
548 (sb1->set_uuid1 == sb2->set_uuid1) &&
549 (sb1->set_uuid2 == sb2->set_uuid2) &&
550 (sb1->set_uuid3 == sb2->set_uuid3))
558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
561 mdp_super_t *tmp1, *tmp2;
563 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
564 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
566 if (!tmp1 || !tmp2) {
568 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
576 * nr_disks is not constant
581 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
593 static u32 md_csum_fold(u32 csum)
595 csum = (csum & 0xffff) + (csum >> 16);
596 return (csum & 0xffff) + (csum >> 16);
599 static unsigned int calc_sb_csum(mdp_super_t * sb)
602 u32 *sb32 = (u32*)sb;
604 unsigned int disk_csum, csum;
606 disk_csum = sb->sb_csum;
609 for (i = 0; i < MD_SB_BYTES/4 ; i++)
611 csum = (newcsum & 0xffffffff) + (newcsum>>32);
615 /* This used to use csum_partial, which was wrong for several
616 * reasons including that different results are returned on
617 * different architectures. It isn't critical that we get exactly
618 * the same return value as before (we always csum_fold before
619 * testing, and that removes any differences). However as we
620 * know that csum_partial always returned a 16bit value on
621 * alphas, do a fold to maximise conformity to previous behaviour.
623 sb->sb_csum = md_csum_fold(disk_csum);
625 sb->sb_csum = disk_csum;
632 * Handle superblock details.
633 * We want to be able to handle multiple superblock formats
634 * so we have a common interface to them all, and an array of
635 * different handlers.
636 * We rely on user-space to write the initial superblock, and support
637 * reading and updating of superblocks.
638 * Interface methods are:
639 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
640 * loads and validates a superblock on dev.
641 * if refdev != NULL, compare superblocks on both devices
643 * 0 - dev has a superblock that is compatible with refdev
644 * 1 - dev has a superblock that is compatible and newer than refdev
645 * so dev should be used as the refdev in future
646 * -EINVAL superblock incompatible or invalid
647 * -othererror e.g. -EIO
649 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
650 * Verify that dev is acceptable into mddev.
651 * The first time, mddev->raid_disks will be 0, and data from
652 * dev should be merged in. Subsequent calls check that dev
653 * is new enough. Return 0 or -EINVAL
655 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
656 * Update the superblock for rdev with data in mddev
657 * This does not write to disc.
663 struct module *owner;
664 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
665 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
666 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
670 * load_super for 0.90.0
672 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
674 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
680 * Calculate the position of the superblock,
681 * it's at the end of the disk.
683 * It also happens to be a multiple of 4Kb.
685 sb_offset = calc_dev_sboffset(rdev->bdev);
686 rdev->sb_offset = sb_offset;
688 ret = read_disk_sb(rdev, MD_SB_BYTES);
693 bdevname(rdev->bdev, b);
694 sb = (mdp_super_t*)page_address(rdev->sb_page);
696 if (sb->md_magic != MD_SB_MAGIC) {
697 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
702 if (sb->major_version != 0 ||
703 sb->minor_version < 90 ||
704 sb->minor_version > 91) {
705 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
706 sb->major_version, sb->minor_version,
711 if (sb->raid_disks <= 0)
714 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
715 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
720 rdev->preferred_minor = sb->md_minor;
721 rdev->data_offset = 0;
722 rdev->sb_size = MD_SB_BYTES;
724 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
725 if (sb->level != 1 && sb->level != 4
726 && sb->level != 5 && sb->level != 6
727 && sb->level != 10) {
728 /* FIXME use a better test */
730 "md: bitmaps not supported for this level.\n");
735 if (sb->level == LEVEL_MULTIPATH)
738 rdev->desc_nr = sb->this_disk.number;
744 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
745 if (!uuid_equal(refsb, sb)) {
746 printk(KERN_WARNING "md: %s has different UUID to %s\n",
747 b, bdevname(refdev->bdev,b2));
750 if (!sb_equal(refsb, sb)) {
751 printk(KERN_WARNING "md: %s has same UUID"
752 " but different superblock to %s\n",
753 b, bdevname(refdev->bdev, b2));
757 ev2 = md_event(refsb);
763 rdev->size = calc_dev_size(rdev, sb->chunk_size);
765 if (rdev->size < sb->size && sb->level > 1)
766 /* "this cannot possibly happen" ... */
774 * validate_super for 0.90.0
776 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
779 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
780 __u64 ev1 = md_event(sb);
782 rdev->raid_disk = -1;
783 clear_bit(Faulty, &rdev->flags);
784 clear_bit(In_sync, &rdev->flags);
785 clear_bit(WriteMostly, &rdev->flags);
786 clear_bit(BarriersNotsupp, &rdev->flags);
788 if (mddev->raid_disks == 0) {
789 mddev->major_version = 0;
790 mddev->minor_version = sb->minor_version;
791 mddev->patch_version = sb->patch_version;
793 mddev->chunk_size = sb->chunk_size;
794 mddev->ctime = sb->ctime;
795 mddev->utime = sb->utime;
796 mddev->level = sb->level;
797 mddev->clevel[0] = 0;
798 mddev->layout = sb->layout;
799 mddev->raid_disks = sb->raid_disks;
800 mddev->size = sb->size;
802 mddev->bitmap_offset = 0;
803 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
805 if (mddev->minor_version >= 91) {
806 mddev->reshape_position = sb->reshape_position;
807 mddev->delta_disks = sb->delta_disks;
808 mddev->new_level = sb->new_level;
809 mddev->new_layout = sb->new_layout;
810 mddev->new_chunk = sb->new_chunk;
812 mddev->reshape_position = MaxSector;
813 mddev->delta_disks = 0;
814 mddev->new_level = mddev->level;
815 mddev->new_layout = mddev->layout;
816 mddev->new_chunk = mddev->chunk_size;
819 if (sb->state & (1<<MD_SB_CLEAN))
820 mddev->recovery_cp = MaxSector;
822 if (sb->events_hi == sb->cp_events_hi &&
823 sb->events_lo == sb->cp_events_lo) {
824 mddev->recovery_cp = sb->recovery_cp;
826 mddev->recovery_cp = 0;
829 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
830 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
831 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
832 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
834 mddev->max_disks = MD_SB_DISKS;
836 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
837 mddev->bitmap_file == NULL)
838 mddev->bitmap_offset = mddev->default_bitmap_offset;
840 } else if (mddev->pers == NULL) {
841 /* Insist on good event counter while assembling */
843 if (ev1 < mddev->events)
845 } else if (mddev->bitmap) {
846 /* if adding to array with a bitmap, then we can accept an
847 * older device ... but not too old.
849 if (ev1 < mddev->bitmap->events_cleared)
852 if (ev1 < mddev->events)
853 /* just a hot-add of a new device, leave raid_disk at -1 */
857 if (mddev->level != LEVEL_MULTIPATH) {
858 desc = sb->disks + rdev->desc_nr;
860 if (desc->state & (1<<MD_DISK_FAULTY))
861 set_bit(Faulty, &rdev->flags);
862 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
863 desc->raid_disk < mddev->raid_disks */) {
864 set_bit(In_sync, &rdev->flags);
865 rdev->raid_disk = desc->raid_disk;
867 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
868 set_bit(WriteMostly, &rdev->flags);
869 } else /* MULTIPATH are always insync */
870 set_bit(In_sync, &rdev->flags);
875 * sync_super for 0.90.0
877 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
880 struct list_head *tmp;
882 int next_spare = mddev->raid_disks;
885 /* make rdev->sb match mddev data..
888 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
889 * 3/ any empty disks < next_spare become removed
891 * disks[0] gets initialised to REMOVED because
892 * we cannot be sure from other fields if it has
893 * been initialised or not.
896 int active=0, working=0,failed=0,spare=0,nr_disks=0;
898 rdev->sb_size = MD_SB_BYTES;
900 sb = (mdp_super_t*)page_address(rdev->sb_page);
902 memset(sb, 0, sizeof(*sb));
904 sb->md_magic = MD_SB_MAGIC;
905 sb->major_version = mddev->major_version;
906 sb->patch_version = mddev->patch_version;
907 sb->gvalid_words = 0; /* ignored */
908 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
909 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
910 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
911 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
913 sb->ctime = mddev->ctime;
914 sb->level = mddev->level;
915 sb->size = mddev->size;
916 sb->raid_disks = mddev->raid_disks;
917 sb->md_minor = mddev->md_minor;
918 sb->not_persistent = 0;
919 sb->utime = mddev->utime;
921 sb->events_hi = (mddev->events>>32);
922 sb->events_lo = (u32)mddev->events;
924 if (mddev->reshape_position == MaxSector)
925 sb->minor_version = 90;
927 sb->minor_version = 91;
928 sb->reshape_position = mddev->reshape_position;
929 sb->new_level = mddev->new_level;
930 sb->delta_disks = mddev->delta_disks;
931 sb->new_layout = mddev->new_layout;
932 sb->new_chunk = mddev->new_chunk;
934 mddev->minor_version = sb->minor_version;
937 sb->recovery_cp = mddev->recovery_cp;
938 sb->cp_events_hi = (mddev->events>>32);
939 sb->cp_events_lo = (u32)mddev->events;
940 if (mddev->recovery_cp == MaxSector)
941 sb->state = (1<< MD_SB_CLEAN);
945 sb->layout = mddev->layout;
946 sb->chunk_size = mddev->chunk_size;
948 if (mddev->bitmap && mddev->bitmap_file == NULL)
949 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
951 sb->disks[0].state = (1<<MD_DISK_REMOVED);
952 rdev_for_each(rdev2, tmp, mddev) {
955 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
956 && !test_bit(Faulty, &rdev2->flags))
957 desc_nr = rdev2->raid_disk;
959 desc_nr = next_spare++;
960 rdev2->desc_nr = desc_nr;
961 d = &sb->disks[rdev2->desc_nr];
963 d->number = rdev2->desc_nr;
964 d->major = MAJOR(rdev2->bdev->bd_dev);
965 d->minor = MINOR(rdev2->bdev->bd_dev);
966 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
967 && !test_bit(Faulty, &rdev2->flags))
968 d->raid_disk = rdev2->raid_disk;
970 d->raid_disk = rdev2->desc_nr; /* compatibility */
971 if (test_bit(Faulty, &rdev2->flags))
972 d->state = (1<<MD_DISK_FAULTY);
973 else if (test_bit(In_sync, &rdev2->flags)) {
974 d->state = (1<<MD_DISK_ACTIVE);
975 d->state |= (1<<MD_DISK_SYNC);
983 if (test_bit(WriteMostly, &rdev2->flags))
984 d->state |= (1<<MD_DISK_WRITEMOSTLY);
986 /* now set the "removed" and "faulty" bits on any missing devices */
987 for (i=0 ; i < mddev->raid_disks ; i++) {
988 mdp_disk_t *d = &sb->disks[i];
989 if (d->state == 0 && d->number == 0) {
992 d->state = (1<<MD_DISK_REMOVED);
993 d->state |= (1<<MD_DISK_FAULTY);
997 sb->nr_disks = nr_disks;
998 sb->active_disks = active;
999 sb->working_disks = working;
1000 sb->failed_disks = failed;
1001 sb->spare_disks = spare;
1003 sb->this_disk = sb->disks[rdev->desc_nr];
1004 sb->sb_csum = calc_sb_csum(sb);
1008 * version 1 superblock
1011 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1015 unsigned long long newcsum;
1016 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1017 __le32 *isuper = (__le32*)sb;
1020 disk_csum = sb->sb_csum;
1023 for (i=0; size>=4; size -= 4 )
1024 newcsum += le32_to_cpu(*isuper++);
1027 newcsum += le16_to_cpu(*(__le16*) isuper);
1029 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1030 sb->sb_csum = disk_csum;
1031 return cpu_to_le32(csum);
1034 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1036 struct mdp_superblock_1 *sb;
1039 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1043 * Calculate the position of the superblock.
1044 * It is always aligned to a 4K boundary and
1045 * depeding on minor_version, it can be:
1046 * 0: At least 8K, but less than 12K, from end of device
1047 * 1: At start of device
1048 * 2: 4K from start of device.
1050 switch(minor_version) {
1052 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1054 sb_offset &= ~(sector_t)(4*2-1);
1055 /* convert from sectors to K */
1067 rdev->sb_offset = sb_offset;
1069 /* superblock is rarely larger than 1K, but it can be larger,
1070 * and it is safe to read 4k, so we do that
1072 ret = read_disk_sb(rdev, 4096);
1073 if (ret) return ret;
1076 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1078 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1079 sb->major_version != cpu_to_le32(1) ||
1080 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1081 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1082 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1085 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1086 printk("md: invalid superblock checksum on %s\n",
1087 bdevname(rdev->bdev,b));
1090 if (le64_to_cpu(sb->data_size) < 10) {
1091 printk("md: data_size too small on %s\n",
1092 bdevname(rdev->bdev,b));
1095 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1096 if (sb->level != cpu_to_le32(1) &&
1097 sb->level != cpu_to_le32(4) &&
1098 sb->level != cpu_to_le32(5) &&
1099 sb->level != cpu_to_le32(6) &&
1100 sb->level != cpu_to_le32(10)) {
1102 "md: bitmaps not supported for this level.\n");
1107 rdev->preferred_minor = 0xffff;
1108 rdev->data_offset = le64_to_cpu(sb->data_offset);
1109 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1111 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1112 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1113 if (rdev->sb_size & bmask)
1114 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1117 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1120 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1123 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1129 struct mdp_superblock_1 *refsb =
1130 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1132 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1133 sb->level != refsb->level ||
1134 sb->layout != refsb->layout ||
1135 sb->chunksize != refsb->chunksize) {
1136 printk(KERN_WARNING "md: %s has strangely different"
1137 " superblock to %s\n",
1138 bdevname(rdev->bdev,b),
1139 bdevname(refdev->bdev,b2));
1142 ev1 = le64_to_cpu(sb->events);
1143 ev2 = le64_to_cpu(refsb->events);
1151 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1153 rdev->size = rdev->sb_offset;
1154 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1156 rdev->size = le64_to_cpu(sb->data_size)/2;
1157 if (le32_to_cpu(sb->chunksize))
1158 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1160 if (le64_to_cpu(sb->size) > rdev->size*2)
1165 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1167 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1168 __u64 ev1 = le64_to_cpu(sb->events);
1170 rdev->raid_disk = -1;
1171 clear_bit(Faulty, &rdev->flags);
1172 clear_bit(In_sync, &rdev->flags);
1173 clear_bit(WriteMostly, &rdev->flags);
1174 clear_bit(BarriersNotsupp, &rdev->flags);
1176 if (mddev->raid_disks == 0) {
1177 mddev->major_version = 1;
1178 mddev->patch_version = 0;
1179 mddev->external = 0;
1180 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1181 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1182 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1183 mddev->level = le32_to_cpu(sb->level);
1184 mddev->clevel[0] = 0;
1185 mddev->layout = le32_to_cpu(sb->layout);
1186 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1187 mddev->size = le64_to_cpu(sb->size)/2;
1188 mddev->events = ev1;
1189 mddev->bitmap_offset = 0;
1190 mddev->default_bitmap_offset = 1024 >> 9;
1192 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1193 memcpy(mddev->uuid, sb->set_uuid, 16);
1195 mddev->max_disks = (4096-256)/2;
1197 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1198 mddev->bitmap_file == NULL )
1199 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1201 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1202 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1203 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1204 mddev->new_level = le32_to_cpu(sb->new_level);
1205 mddev->new_layout = le32_to_cpu(sb->new_layout);
1206 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1208 mddev->reshape_position = MaxSector;
1209 mddev->delta_disks = 0;
1210 mddev->new_level = mddev->level;
1211 mddev->new_layout = mddev->layout;
1212 mddev->new_chunk = mddev->chunk_size;
1215 } else if (mddev->pers == NULL) {
1216 /* Insist of good event counter while assembling */
1218 if (ev1 < mddev->events)
1220 } else if (mddev->bitmap) {
1221 /* If adding to array with a bitmap, then we can accept an
1222 * older device, but not too old.
1224 if (ev1 < mddev->bitmap->events_cleared)
1227 if (ev1 < mddev->events)
1228 /* just a hot-add of a new device, leave raid_disk at -1 */
1231 if (mddev->level != LEVEL_MULTIPATH) {
1233 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1235 case 0xffff: /* spare */
1237 case 0xfffe: /* faulty */
1238 set_bit(Faulty, &rdev->flags);
1241 if ((le32_to_cpu(sb->feature_map) &
1242 MD_FEATURE_RECOVERY_OFFSET))
1243 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1245 set_bit(In_sync, &rdev->flags);
1246 rdev->raid_disk = role;
1249 if (sb->devflags & WriteMostly1)
1250 set_bit(WriteMostly, &rdev->flags);
1251 } else /* MULTIPATH are always insync */
1252 set_bit(In_sync, &rdev->flags);
1257 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1259 struct mdp_superblock_1 *sb;
1260 struct list_head *tmp;
1263 /* make rdev->sb match mddev and rdev data. */
1265 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1267 sb->feature_map = 0;
1269 sb->recovery_offset = cpu_to_le64(0);
1270 memset(sb->pad1, 0, sizeof(sb->pad1));
1271 memset(sb->pad2, 0, sizeof(sb->pad2));
1272 memset(sb->pad3, 0, sizeof(sb->pad3));
1274 sb->utime = cpu_to_le64((__u64)mddev->utime);
1275 sb->events = cpu_to_le64(mddev->events);
1277 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1279 sb->resync_offset = cpu_to_le64(0);
1281 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1283 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1284 sb->size = cpu_to_le64(mddev->size<<1);
1286 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1287 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1288 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1291 if (rdev->raid_disk >= 0 &&
1292 !test_bit(In_sync, &rdev->flags) &&
1293 rdev->recovery_offset > 0) {
1294 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1295 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1298 if (mddev->reshape_position != MaxSector) {
1299 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1300 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1301 sb->new_layout = cpu_to_le32(mddev->new_layout);
1302 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1303 sb->new_level = cpu_to_le32(mddev->new_level);
1304 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1308 rdev_for_each(rdev2, tmp, mddev)
1309 if (rdev2->desc_nr+1 > max_dev)
1310 max_dev = rdev2->desc_nr+1;
1312 if (max_dev > le32_to_cpu(sb->max_dev))
1313 sb->max_dev = cpu_to_le32(max_dev);
1314 for (i=0; i<max_dev;i++)
1315 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1317 rdev_for_each(rdev2, tmp, mddev) {
1319 if (test_bit(Faulty, &rdev2->flags))
1320 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1321 else if (test_bit(In_sync, &rdev2->flags))
1322 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1323 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1324 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1326 sb->dev_roles[i] = cpu_to_le16(0xffff);
1329 sb->sb_csum = calc_sb_1_csum(sb);
1333 static struct super_type super_types[] = {
1336 .owner = THIS_MODULE,
1337 .load_super = super_90_load,
1338 .validate_super = super_90_validate,
1339 .sync_super = super_90_sync,
1343 .owner = THIS_MODULE,
1344 .load_super = super_1_load,
1345 .validate_super = super_1_validate,
1346 .sync_super = super_1_sync,
1350 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1352 struct list_head *tmp, *tmp2;
1353 mdk_rdev_t *rdev, *rdev2;
1355 rdev_for_each(rdev, tmp, mddev1)
1356 rdev_for_each(rdev2, tmp2, mddev2)
1357 if (rdev->bdev->bd_contains ==
1358 rdev2->bdev->bd_contains)
1364 static LIST_HEAD(pending_raid_disks);
1366 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1368 char b[BDEVNAME_SIZE];
1378 /* prevent duplicates */
1379 if (find_rdev(mddev, rdev->bdev->bd_dev))
1382 /* make sure rdev->size exceeds mddev->size */
1383 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1385 /* Cannot change size, so fail
1386 * If mddev->level <= 0, then we don't care
1387 * about aligning sizes (e.g. linear)
1389 if (mddev->level > 0)
1392 mddev->size = rdev->size;
1395 /* Verify rdev->desc_nr is unique.
1396 * If it is -1, assign a free number, else
1397 * check number is not in use
1399 if (rdev->desc_nr < 0) {
1401 if (mddev->pers) choice = mddev->raid_disks;
1402 while (find_rdev_nr(mddev, choice))
1404 rdev->desc_nr = choice;
1406 if (find_rdev_nr(mddev, rdev->desc_nr))
1409 bdevname(rdev->bdev,b);
1410 while ( (s=strchr(b, '/')) != NULL)
1413 rdev->mddev = mddev;
1414 printk(KERN_INFO "md: bind<%s>\n", b);
1416 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1419 if (rdev->bdev->bd_part)
1420 ko = &rdev->bdev->bd_part->dev.kobj;
1422 ko = &rdev->bdev->bd_disk->dev.kobj;
1423 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1424 kobject_del(&rdev->kobj);
1427 list_add(&rdev->same_set, &mddev->disks);
1428 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1432 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1437 static void md_delayed_delete(struct work_struct *ws)
1439 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1440 kobject_del(&rdev->kobj);
1441 kobject_put(&rdev->kobj);
1444 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1446 char b[BDEVNAME_SIZE];
1451 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1452 list_del_init(&rdev->same_set);
1453 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1455 sysfs_remove_link(&rdev->kobj, "block");
1457 /* We need to delay this, otherwise we can deadlock when
1458 * writing to 'remove' to "dev/state"
1460 INIT_WORK(&rdev->del_work, md_delayed_delete);
1461 kobject_get(&rdev->kobj);
1462 schedule_work(&rdev->del_work);
1466 * prevent the device from being mounted, repartitioned or
1467 * otherwise reused by a RAID array (or any other kernel
1468 * subsystem), by bd_claiming the device.
1470 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1473 struct block_device *bdev;
1474 char b[BDEVNAME_SIZE];
1476 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1478 printk(KERN_ERR "md: could not open %s.\n",
1479 __bdevname(dev, b));
1480 return PTR_ERR(bdev);
1482 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1484 printk(KERN_ERR "md: could not bd_claim %s.\n",
1490 set_bit(AllReserved, &rdev->flags);
1495 static void unlock_rdev(mdk_rdev_t *rdev)
1497 struct block_device *bdev = rdev->bdev;
1505 void md_autodetect_dev(dev_t dev);
1507 static void export_rdev(mdk_rdev_t * rdev)
1509 char b[BDEVNAME_SIZE];
1510 printk(KERN_INFO "md: export_rdev(%s)\n",
1511 bdevname(rdev->bdev,b));
1515 list_del_init(&rdev->same_set);
1517 if (test_bit(AutoDetected, &rdev->flags))
1518 md_autodetect_dev(rdev->bdev->bd_dev);
1521 kobject_put(&rdev->kobj);
1524 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1526 unbind_rdev_from_array(rdev);
1530 static void export_array(mddev_t *mddev)
1532 struct list_head *tmp;
1535 rdev_for_each(rdev, tmp, mddev) {
1540 kick_rdev_from_array(rdev);
1542 if (!list_empty(&mddev->disks))
1544 mddev->raid_disks = 0;
1545 mddev->major_version = 0;
1548 static void print_desc(mdp_disk_t *desc)
1550 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1551 desc->major,desc->minor,desc->raid_disk,desc->state);
1554 static void print_sb(mdp_super_t *sb)
1559 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1560 sb->major_version, sb->minor_version, sb->patch_version,
1561 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1563 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1564 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1565 sb->md_minor, sb->layout, sb->chunk_size);
1566 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1567 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1568 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1569 sb->failed_disks, sb->spare_disks,
1570 sb->sb_csum, (unsigned long)sb->events_lo);
1573 for (i = 0; i < MD_SB_DISKS; i++) {
1576 desc = sb->disks + i;
1577 if (desc->number || desc->major || desc->minor ||
1578 desc->raid_disk || (desc->state && (desc->state != 4))) {
1579 printk(" D %2d: ", i);
1583 printk(KERN_INFO "md: THIS: ");
1584 print_desc(&sb->this_disk);
1588 static void print_rdev(mdk_rdev_t *rdev)
1590 char b[BDEVNAME_SIZE];
1591 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1592 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1593 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1595 if (rdev->sb_loaded) {
1596 printk(KERN_INFO "md: rdev superblock:\n");
1597 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1599 printk(KERN_INFO "md: no rdev superblock!\n");
1602 static void md_print_devices(void)
1604 struct list_head *tmp, *tmp2;
1607 char b[BDEVNAME_SIZE];
1610 printk("md: **********************************\n");
1611 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1612 printk("md: **********************************\n");
1613 for_each_mddev(mddev, tmp) {
1616 bitmap_print_sb(mddev->bitmap);
1618 printk("%s: ", mdname(mddev));
1619 rdev_for_each(rdev, tmp2, mddev)
1620 printk("<%s>", bdevname(rdev->bdev,b));
1623 rdev_for_each(rdev, tmp2, mddev)
1626 printk("md: **********************************\n");
1631 static void sync_sbs(mddev_t * mddev, int nospares)
1633 /* Update each superblock (in-memory image), but
1634 * if we are allowed to, skip spares which already
1635 * have the right event counter, or have one earlier
1636 * (which would mean they aren't being marked as dirty
1637 * with the rest of the array)
1640 struct list_head *tmp;
1642 rdev_for_each(rdev, tmp, mddev) {
1643 if (rdev->sb_events == mddev->events ||
1645 rdev->raid_disk < 0 &&
1646 (rdev->sb_events&1)==0 &&
1647 rdev->sb_events+1 == mddev->events)) {
1648 /* Don't update this superblock */
1649 rdev->sb_loaded = 2;
1651 super_types[mddev->major_version].
1652 sync_super(mddev, rdev);
1653 rdev->sb_loaded = 1;
1658 static void md_update_sb(mddev_t * mddev, int force_change)
1660 struct list_head *tmp;
1665 if (mddev->external)
1668 spin_lock_irq(&mddev->write_lock);
1670 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1671 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1673 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1674 /* just a clean<-> dirty transition, possibly leave spares alone,
1675 * though if events isn't the right even/odd, we will have to do
1681 if (mddev->degraded)
1682 /* If the array is degraded, then skipping spares is both
1683 * dangerous and fairly pointless.
1684 * Dangerous because a device that was removed from the array
1685 * might have a event_count that still looks up-to-date,
1686 * so it can be re-added without a resync.
1687 * Pointless because if there are any spares to skip,
1688 * then a recovery will happen and soon that array won't
1689 * be degraded any more and the spare can go back to sleep then.
1693 sync_req = mddev->in_sync;
1694 mddev->utime = get_seconds();
1696 /* If this is just a dirty<->clean transition, and the array is clean
1697 * and 'events' is odd, we can roll back to the previous clean state */
1699 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1700 && (mddev->events & 1)
1701 && mddev->events != 1)
1704 /* otherwise we have to go forward and ... */
1706 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1707 /* .. if the array isn't clean, insist on an odd 'events' */
1708 if ((mddev->events&1)==0) {
1713 /* otherwise insist on an even 'events' (for clean states) */
1714 if ((mddev->events&1)) {
1721 if (!mddev->events) {
1723 * oops, this 64-bit counter should never wrap.
1724 * Either we are in around ~1 trillion A.C., assuming
1725 * 1 reboot per second, or we have a bug:
1732 * do not write anything to disk if using
1733 * nonpersistent superblocks
1735 if (!mddev->persistent) {
1736 if (!mddev->external)
1737 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1739 spin_unlock_irq(&mddev->write_lock);
1740 wake_up(&mddev->sb_wait);
1743 sync_sbs(mddev, nospares);
1744 spin_unlock_irq(&mddev->write_lock);
1747 "md: updating %s RAID superblock on device (in sync %d)\n",
1748 mdname(mddev),mddev->in_sync);
1750 bitmap_update_sb(mddev->bitmap);
1751 rdev_for_each(rdev, tmp, mddev) {
1752 char b[BDEVNAME_SIZE];
1753 dprintk(KERN_INFO "md: ");
1754 if (rdev->sb_loaded != 1)
1755 continue; /* no noise on spare devices */
1756 if (test_bit(Faulty, &rdev->flags))
1757 dprintk("(skipping faulty ");
1759 dprintk("%s ", bdevname(rdev->bdev,b));
1760 if (!test_bit(Faulty, &rdev->flags)) {
1761 md_super_write(mddev,rdev,
1762 rdev->sb_offset<<1, rdev->sb_size,
1764 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1765 bdevname(rdev->bdev,b),
1766 (unsigned long long)rdev->sb_offset);
1767 rdev->sb_events = mddev->events;
1771 if (mddev->level == LEVEL_MULTIPATH)
1772 /* only need to write one superblock... */
1775 md_super_wait(mddev);
1776 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1778 spin_lock_irq(&mddev->write_lock);
1779 if (mddev->in_sync != sync_req ||
1780 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1781 /* have to write it out again */
1782 spin_unlock_irq(&mddev->write_lock);
1785 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1786 spin_unlock_irq(&mddev->write_lock);
1787 wake_up(&mddev->sb_wait);
1791 /* words written to sysfs files may, or my not, be \n terminated.
1792 * We want to accept with case. For this we use cmd_match.
1794 static int cmd_match(const char *cmd, const char *str)
1796 /* See if cmd, written into a sysfs file, matches
1797 * str. They must either be the same, or cmd can
1798 * have a trailing newline
1800 while (*cmd && *str && *cmd == *str) {
1811 struct rdev_sysfs_entry {
1812 struct attribute attr;
1813 ssize_t (*show)(mdk_rdev_t *, char *);
1814 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1818 state_show(mdk_rdev_t *rdev, char *page)
1823 if (test_bit(Faulty, &rdev->flags)) {
1824 len+= sprintf(page+len, "%sfaulty",sep);
1827 if (test_bit(In_sync, &rdev->flags)) {
1828 len += sprintf(page+len, "%sin_sync",sep);
1831 if (test_bit(WriteMostly, &rdev->flags)) {
1832 len += sprintf(page+len, "%swrite_mostly",sep);
1835 if (test_bit(Blocked, &rdev->flags)) {
1836 len += sprintf(page+len, "%sblocked", sep);
1839 if (!test_bit(Faulty, &rdev->flags) &&
1840 !test_bit(In_sync, &rdev->flags)) {
1841 len += sprintf(page+len, "%sspare", sep);
1844 return len+sprintf(page+len, "\n");
1848 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1851 * faulty - simulates and error
1852 * remove - disconnects the device
1853 * writemostly - sets write_mostly
1854 * -writemostly - clears write_mostly
1855 * blocked - sets the Blocked flag
1856 * -blocked - clears the Blocked flag
1859 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1860 md_error(rdev->mddev, rdev);
1862 } else if (cmd_match(buf, "remove")) {
1863 if (rdev->raid_disk >= 0)
1866 mddev_t *mddev = rdev->mddev;
1867 kick_rdev_from_array(rdev);
1869 md_update_sb(mddev, 1);
1870 md_new_event(mddev);
1873 } else if (cmd_match(buf, "writemostly")) {
1874 set_bit(WriteMostly, &rdev->flags);
1876 } else if (cmd_match(buf, "-writemostly")) {
1877 clear_bit(WriteMostly, &rdev->flags);
1879 } else if (cmd_match(buf, "blocked")) {
1880 set_bit(Blocked, &rdev->flags);
1882 } else if (cmd_match(buf, "-blocked")) {
1883 clear_bit(Blocked, &rdev->flags);
1884 wake_up(&rdev->blocked_wait);
1885 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1886 md_wakeup_thread(rdev->mddev->thread);
1890 return err ? err : len;
1892 static struct rdev_sysfs_entry rdev_state =
1893 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1896 errors_show(mdk_rdev_t *rdev, char *page)
1898 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1902 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1905 unsigned long n = simple_strtoul(buf, &e, 10);
1906 if (*buf && (*e == 0 || *e == '\n')) {
1907 atomic_set(&rdev->corrected_errors, n);
1912 static struct rdev_sysfs_entry rdev_errors =
1913 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1916 slot_show(mdk_rdev_t *rdev, char *page)
1918 if (rdev->raid_disk < 0)
1919 return sprintf(page, "none\n");
1921 return sprintf(page, "%d\n", rdev->raid_disk);
1925 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1930 int slot = simple_strtoul(buf, &e, 10);
1931 if (strncmp(buf, "none", 4)==0)
1933 else if (e==buf || (*e && *e!= '\n'))
1935 if (rdev->mddev->pers) {
1936 /* Setting 'slot' on an active array requires also
1937 * updating the 'rd%d' link, and communicating
1938 * with the personality with ->hot_*_disk.
1939 * For now we only support removing
1940 * failed/spare devices. This normally happens automatically,
1941 * but not when the metadata is externally managed.
1945 if (rdev->raid_disk == -1)
1947 /* personality does all needed checks */
1948 if (rdev->mddev->pers->hot_add_disk == NULL)
1950 err = rdev->mddev->pers->
1951 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1954 sprintf(nm, "rd%d", rdev->raid_disk);
1955 sysfs_remove_link(&rdev->mddev->kobj, nm);
1956 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1957 md_wakeup_thread(rdev->mddev->thread);
1959 if (slot >= rdev->mddev->raid_disks)
1961 rdev->raid_disk = slot;
1962 /* assume it is working */
1963 clear_bit(Faulty, &rdev->flags);
1964 clear_bit(WriteMostly, &rdev->flags);
1965 set_bit(In_sync, &rdev->flags);
1971 static struct rdev_sysfs_entry rdev_slot =
1972 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1975 offset_show(mdk_rdev_t *rdev, char *page)
1977 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1981 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1984 unsigned long long offset = simple_strtoull(buf, &e, 10);
1985 if (e==buf || (*e && *e != '\n'))
1987 if (rdev->mddev->pers)
1989 if (rdev->size && rdev->mddev->external)
1990 /* Must set offset before size, so overlap checks
1993 rdev->data_offset = offset;
1997 static struct rdev_sysfs_entry rdev_offset =
1998 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2001 rdev_size_show(mdk_rdev_t *rdev, char *page)
2003 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2006 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2008 /* check if two start/length pairs overlap */
2017 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2020 unsigned long long size = simple_strtoull(buf, &e, 10);
2021 unsigned long long oldsize = rdev->size;
2022 mddev_t *my_mddev = rdev->mddev;
2024 if (e==buf || (*e && *e != '\n'))
2029 if (size > oldsize && rdev->mddev->external) {
2030 /* need to check that all other rdevs with the same ->bdev
2031 * do not overlap. We need to unlock the mddev to avoid
2032 * a deadlock. We have already changed rdev->size, and if
2033 * we have to change it back, we will have the lock again.
2037 struct list_head *tmp, *tmp2;
2039 mddev_unlock(my_mddev);
2040 for_each_mddev(mddev, tmp) {
2044 rdev_for_each(rdev2, tmp2, mddev)
2045 if (test_bit(AllReserved, &rdev2->flags) ||
2046 (rdev->bdev == rdev2->bdev &&
2048 overlaps(rdev->data_offset, rdev->size,
2049 rdev2->data_offset, rdev2->size))) {
2053 mddev_unlock(mddev);
2059 mddev_lock(my_mddev);
2061 /* Someone else could have slipped in a size
2062 * change here, but doing so is just silly.
2063 * We put oldsize back because we *know* it is
2064 * safe, and trust userspace not to race with
2067 rdev->size = oldsize;
2071 if (size < my_mddev->size || my_mddev->size == 0)
2072 my_mddev->size = size;
2076 static struct rdev_sysfs_entry rdev_size =
2077 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2079 static struct attribute *rdev_default_attrs[] = {
2088 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2090 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2091 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2092 mddev_t *mddev = rdev->mddev;
2098 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2100 if (rdev->mddev == NULL)
2103 rv = entry->show(rdev, page);
2104 mddev_unlock(mddev);
2110 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2111 const char *page, size_t length)
2113 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2114 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2116 mddev_t *mddev = rdev->mddev;
2120 if (!capable(CAP_SYS_ADMIN))
2122 rv = mddev ? mddev_lock(mddev): -EBUSY;
2124 if (rdev->mddev == NULL)
2127 rv = entry->store(rdev, page, length);
2128 mddev_unlock(mddev);
2133 static void rdev_free(struct kobject *ko)
2135 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2138 static struct sysfs_ops rdev_sysfs_ops = {
2139 .show = rdev_attr_show,
2140 .store = rdev_attr_store,
2142 static struct kobj_type rdev_ktype = {
2143 .release = rdev_free,
2144 .sysfs_ops = &rdev_sysfs_ops,
2145 .default_attrs = rdev_default_attrs,
2149 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2151 * mark the device faulty if:
2153 * - the device is nonexistent (zero size)
2154 * - the device has no valid superblock
2156 * a faulty rdev _never_ has rdev->sb set.
2158 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2160 char b[BDEVNAME_SIZE];
2165 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2167 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2168 return ERR_PTR(-ENOMEM);
2171 if ((err = alloc_disk_sb(rdev)))
2174 err = lock_rdev(rdev, newdev, super_format == -2);
2178 kobject_init(&rdev->kobj, &rdev_ktype);
2181 rdev->saved_raid_disk = -1;
2182 rdev->raid_disk = -1;
2184 rdev->data_offset = 0;
2185 rdev->sb_events = 0;
2186 atomic_set(&rdev->nr_pending, 0);
2187 atomic_set(&rdev->read_errors, 0);
2188 atomic_set(&rdev->corrected_errors, 0);
2190 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2193 "md: %s has zero or unknown size, marking faulty!\n",
2194 bdevname(rdev->bdev,b));
2199 if (super_format >= 0) {
2200 err = super_types[super_format].
2201 load_super(rdev, NULL, super_minor);
2202 if (err == -EINVAL) {
2204 "md: %s does not have a valid v%d.%d "
2205 "superblock, not importing!\n",
2206 bdevname(rdev->bdev,b),
2207 super_format, super_minor);
2212 "md: could not read %s's sb, not importing!\n",
2213 bdevname(rdev->bdev,b));
2218 INIT_LIST_HEAD(&rdev->same_set);
2219 init_waitqueue_head(&rdev->blocked_wait);
2224 if (rdev->sb_page) {
2230 return ERR_PTR(err);
2234 * Check a full RAID array for plausibility
2238 static void analyze_sbs(mddev_t * mddev)
2241 struct list_head *tmp;
2242 mdk_rdev_t *rdev, *freshest;
2243 char b[BDEVNAME_SIZE];
2246 rdev_for_each(rdev, tmp, mddev)
2247 switch (super_types[mddev->major_version].
2248 load_super(rdev, freshest, mddev->minor_version)) {
2256 "md: fatal superblock inconsistency in %s"
2257 " -- removing from array\n",
2258 bdevname(rdev->bdev,b));
2259 kick_rdev_from_array(rdev);
2263 super_types[mddev->major_version].
2264 validate_super(mddev, freshest);
2267 rdev_for_each(rdev, tmp, mddev) {
2268 if (rdev != freshest)
2269 if (super_types[mddev->major_version].
2270 validate_super(mddev, rdev)) {
2271 printk(KERN_WARNING "md: kicking non-fresh %s"
2273 bdevname(rdev->bdev,b));
2274 kick_rdev_from_array(rdev);
2277 if (mddev->level == LEVEL_MULTIPATH) {
2278 rdev->desc_nr = i++;
2279 rdev->raid_disk = rdev->desc_nr;
2280 set_bit(In_sync, &rdev->flags);
2281 } else if (rdev->raid_disk >= mddev->raid_disks) {
2282 rdev->raid_disk = -1;
2283 clear_bit(In_sync, &rdev->flags);
2289 if (mddev->recovery_cp != MaxSector &&
2291 printk(KERN_ERR "md: %s: raid array is not clean"
2292 " -- starting background reconstruction\n",
2298 safe_delay_show(mddev_t *mddev, char *page)
2300 int msec = (mddev->safemode_delay*1000)/HZ;
2301 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2304 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2312 /* remove a period, and count digits after it */
2313 if (len >= sizeof(buf))
2315 strlcpy(buf, cbuf, len);
2317 for (i=0; i<len; i++) {
2319 if (isdigit(buf[i])) {
2324 } else if (buf[i] == '.') {
2329 msec = simple_strtoul(buf, &e, 10);
2330 if (e == buf || (*e && *e != '\n'))
2332 msec = (msec * 1000) / scale;
2334 mddev->safemode_delay = 0;
2336 mddev->safemode_delay = (msec*HZ)/1000;
2337 if (mddev->safemode_delay == 0)
2338 mddev->safemode_delay = 1;
2342 static struct md_sysfs_entry md_safe_delay =
2343 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2346 level_show(mddev_t *mddev, char *page)
2348 struct mdk_personality *p = mddev->pers;
2350 return sprintf(page, "%s\n", p->name);
2351 else if (mddev->clevel[0])
2352 return sprintf(page, "%s\n", mddev->clevel);
2353 else if (mddev->level != LEVEL_NONE)
2354 return sprintf(page, "%d\n", mddev->level);
2360 level_store(mddev_t *mddev, const char *buf, size_t len)
2367 if (len >= sizeof(mddev->clevel))
2369 strncpy(mddev->clevel, buf, len);
2370 if (mddev->clevel[len-1] == '\n')
2372 mddev->clevel[len] = 0;
2373 mddev->level = LEVEL_NONE;
2377 static struct md_sysfs_entry md_level =
2378 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2382 layout_show(mddev_t *mddev, char *page)
2384 /* just a number, not meaningful for all levels */
2385 if (mddev->reshape_position != MaxSector &&
2386 mddev->layout != mddev->new_layout)
2387 return sprintf(page, "%d (%d)\n",
2388 mddev->new_layout, mddev->layout);
2389 return sprintf(page, "%d\n", mddev->layout);
2393 layout_store(mddev_t *mddev, const char *buf, size_t len)
2396 unsigned long n = simple_strtoul(buf, &e, 10);
2398 if (!*buf || (*e && *e != '\n'))
2403 if (mddev->reshape_position != MaxSector)
2404 mddev->new_layout = n;
2409 static struct md_sysfs_entry md_layout =
2410 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2414 raid_disks_show(mddev_t *mddev, char *page)
2416 if (mddev->raid_disks == 0)
2418 if (mddev->reshape_position != MaxSector &&
2419 mddev->delta_disks != 0)
2420 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2421 mddev->raid_disks - mddev->delta_disks);
2422 return sprintf(page, "%d\n", mddev->raid_disks);
2425 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2428 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2432 unsigned long n = simple_strtoul(buf, &e, 10);
2434 if (!*buf || (*e && *e != '\n'))
2438 rv = update_raid_disks(mddev, n);
2439 else if (mddev->reshape_position != MaxSector) {
2440 int olddisks = mddev->raid_disks - mddev->delta_disks;
2441 mddev->delta_disks = n - olddisks;
2442 mddev->raid_disks = n;
2444 mddev->raid_disks = n;
2445 return rv ? rv : len;
2447 static struct md_sysfs_entry md_raid_disks =
2448 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2451 chunk_size_show(mddev_t *mddev, char *page)
2453 if (mddev->reshape_position != MaxSector &&
2454 mddev->chunk_size != mddev->new_chunk)
2455 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2457 return sprintf(page, "%d\n", mddev->chunk_size);
2461 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2463 /* can only set chunk_size if array is not yet active */
2465 unsigned long n = simple_strtoul(buf, &e, 10);
2467 if (!*buf || (*e && *e != '\n'))
2472 else if (mddev->reshape_position != MaxSector)
2473 mddev->new_chunk = n;
2475 mddev->chunk_size = n;
2478 static struct md_sysfs_entry md_chunk_size =
2479 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2482 resync_start_show(mddev_t *mddev, char *page)
2484 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2488 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2491 unsigned long long n = simple_strtoull(buf, &e, 10);
2495 if (!*buf || (*e && *e != '\n'))
2498 mddev->recovery_cp = n;
2501 static struct md_sysfs_entry md_resync_start =
2502 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2505 * The array state can be:
2508 * No devices, no size, no level
2509 * Equivalent to STOP_ARRAY ioctl
2511 * May have some settings, but array is not active
2512 * all IO results in error
2513 * When written, doesn't tear down array, but just stops it
2514 * suspended (not supported yet)
2515 * All IO requests will block. The array can be reconfigured.
2516 * Writing this, if accepted, will block until array is quiessent
2518 * no resync can happen. no superblocks get written.
2519 * write requests fail
2521 * like readonly, but behaves like 'clean' on a write request.
2523 * clean - no pending writes, but otherwise active.
2524 * When written to inactive array, starts without resync
2525 * If a write request arrives then
2526 * if metadata is known, mark 'dirty' and switch to 'active'.
2527 * if not known, block and switch to write-pending
2528 * If written to an active array that has pending writes, then fails.
2530 * fully active: IO and resync can be happening.
2531 * When written to inactive array, starts with resync
2534 * clean, but writes are blocked waiting for 'active' to be written.
2537 * like active, but no writes have been seen for a while (100msec).
2540 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2541 write_pending, active_idle, bad_word};
2542 static char *array_states[] = {
2543 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2544 "write-pending", "active-idle", NULL };
2546 static int match_word(const char *word, char **list)
2549 for (n=0; list[n]; n++)
2550 if (cmd_match(word, list[n]))
2556 array_state_show(mddev_t *mddev, char *page)
2558 enum array_state st = inactive;
2571 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2573 else if (mddev->safemode)
2579 if (list_empty(&mddev->disks) &&
2580 mddev->raid_disks == 0 &&
2586 return sprintf(page, "%s\n", array_states[st]);
2589 static int do_md_stop(mddev_t * mddev, int ro);
2590 static int do_md_run(mddev_t * mddev);
2591 static int restart_array(mddev_t *mddev);
2594 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2597 enum array_state st = match_word(buf, array_states);
2602 /* stopping an active array */
2603 if (atomic_read(&mddev->active) > 1)
2605 err = do_md_stop(mddev, 0);
2608 /* stopping an active array */
2610 if (atomic_read(&mddev->active) > 1)
2612 err = do_md_stop(mddev, 2);
2614 err = 0; /* already inactive */
2617 break; /* not supported yet */
2620 err = do_md_stop(mddev, 1);
2623 set_disk_ro(mddev->gendisk, 1);
2624 err = do_md_run(mddev);
2630 err = do_md_stop(mddev, 1);
2632 err = restart_array(mddev);
2635 set_disk_ro(mddev->gendisk, 0);
2639 err = do_md_run(mddev);
2644 restart_array(mddev);
2645 spin_lock_irq(&mddev->write_lock);
2646 if (atomic_read(&mddev->writes_pending) == 0) {
2647 if (mddev->in_sync == 0) {
2649 if (mddev->safemode == 1)
2650 mddev->safemode = 0;
2651 if (mddev->persistent)
2652 set_bit(MD_CHANGE_CLEAN,
2658 spin_unlock_irq(&mddev->write_lock);
2661 mddev->recovery_cp = MaxSector;
2662 err = do_md_run(mddev);
2667 restart_array(mddev);
2668 if (mddev->external)
2669 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2670 wake_up(&mddev->sb_wait);
2674 set_disk_ro(mddev->gendisk, 0);
2675 err = do_md_run(mddev);
2680 /* these cannot be set */
2688 static struct md_sysfs_entry md_array_state =
2689 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2692 null_show(mddev_t *mddev, char *page)
2698 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2700 /* buf must be %d:%d\n? giving major and minor numbers */
2701 /* The new device is added to the array.
2702 * If the array has a persistent superblock, we read the
2703 * superblock to initialise info and check validity.
2704 * Otherwise, only checking done is that in bind_rdev_to_array,
2705 * which mainly checks size.
2708 int major = simple_strtoul(buf, &e, 10);
2714 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2716 minor = simple_strtoul(e+1, &e, 10);
2717 if (*e && *e != '\n')
2719 dev = MKDEV(major, minor);
2720 if (major != MAJOR(dev) ||
2721 minor != MINOR(dev))
2725 if (mddev->persistent) {
2726 rdev = md_import_device(dev, mddev->major_version,
2727 mddev->minor_version);
2728 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2729 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2730 mdk_rdev_t, same_set);
2731 err = super_types[mddev->major_version]
2732 .load_super(rdev, rdev0, mddev->minor_version);
2736 } else if (mddev->external)
2737 rdev = md_import_device(dev, -2, -1);
2739 rdev = md_import_device(dev, -1, -1);
2742 return PTR_ERR(rdev);
2743 err = bind_rdev_to_array(rdev, mddev);
2747 return err ? err : len;
2750 static struct md_sysfs_entry md_new_device =
2751 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2754 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2757 unsigned long chunk, end_chunk;
2761 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2763 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2764 if (buf == end) break;
2765 if (*end == '-') { /* range */
2767 end_chunk = simple_strtoul(buf, &end, 0);
2768 if (buf == end) break;
2770 if (*end && !isspace(*end)) break;
2771 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2773 while (isspace(*buf)) buf++;
2775 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2780 static struct md_sysfs_entry md_bitmap =
2781 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2784 size_show(mddev_t *mddev, char *page)
2786 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2789 static int update_size(mddev_t *mddev, unsigned long size);
2792 size_store(mddev_t *mddev, const char *buf, size_t len)
2794 /* If array is inactive, we can reduce the component size, but
2795 * not increase it (except from 0).
2796 * If array is active, we can try an on-line resize
2800 unsigned long long size = simple_strtoull(buf, &e, 10);
2801 if (!*buf || *buf == '\n' ||
2806 err = update_size(mddev, size);
2807 md_update_sb(mddev, 1);
2809 if (mddev->size == 0 ||
2815 return err ? err : len;
2818 static struct md_sysfs_entry md_size =
2819 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2824 * 'none' for arrays with no metadata (good luck...)
2825 * 'external' for arrays with externally managed metadata,
2826 * or N.M for internally known formats
2829 metadata_show(mddev_t *mddev, char *page)
2831 if (mddev->persistent)
2832 return sprintf(page, "%d.%d\n",
2833 mddev->major_version, mddev->minor_version);
2834 else if (mddev->external)
2835 return sprintf(page, "external:%s\n", mddev->metadata_type);
2837 return sprintf(page, "none\n");
2841 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2845 if (!list_empty(&mddev->disks))
2848 if (cmd_match(buf, "none")) {
2849 mddev->persistent = 0;
2850 mddev->external = 0;
2851 mddev->major_version = 0;
2852 mddev->minor_version = 90;
2855 if (strncmp(buf, "external:", 9) == 0) {
2856 size_t namelen = len-9;
2857 if (namelen >= sizeof(mddev->metadata_type))
2858 namelen = sizeof(mddev->metadata_type)-1;
2859 strncpy(mddev->metadata_type, buf+9, namelen);
2860 mddev->metadata_type[namelen] = 0;
2861 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2862 mddev->metadata_type[--namelen] = 0;
2863 mddev->persistent = 0;
2864 mddev->external = 1;
2865 mddev->major_version = 0;
2866 mddev->minor_version = 90;
2869 major = simple_strtoul(buf, &e, 10);
2870 if (e==buf || *e != '.')
2873 minor = simple_strtoul(buf, &e, 10);
2874 if (e==buf || (*e && *e != '\n') )
2876 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2878 mddev->major_version = major;
2879 mddev->minor_version = minor;
2880 mddev->persistent = 1;
2881 mddev->external = 0;
2885 static struct md_sysfs_entry md_metadata =
2886 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2889 action_show(mddev_t *mddev, char *page)
2891 char *type = "idle";
2892 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2893 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2894 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2896 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2897 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2899 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2906 return sprintf(page, "%s\n", type);
2910 action_store(mddev_t *mddev, const char *page, size_t len)
2912 if (!mddev->pers || !mddev->pers->sync_request)
2915 if (cmd_match(page, "idle")) {
2916 if (mddev->sync_thread) {
2917 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2918 md_unregister_thread(mddev->sync_thread);
2919 mddev->sync_thread = NULL;
2920 mddev->recovery = 0;
2922 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2923 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2925 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2926 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2927 else if (cmd_match(page, "reshape")) {
2929 if (mddev->pers->start_reshape == NULL)
2931 err = mddev->pers->start_reshape(mddev);
2935 if (cmd_match(page, "check"))
2936 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2937 else if (!cmd_match(page, "repair"))
2939 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2940 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2942 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2943 md_wakeup_thread(mddev->thread);
2948 mismatch_cnt_show(mddev_t *mddev, char *page)
2950 return sprintf(page, "%llu\n",
2951 (unsigned long long) mddev->resync_mismatches);
2954 static struct md_sysfs_entry md_scan_mode =
2955 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2958 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2961 sync_min_show(mddev_t *mddev, char *page)
2963 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2964 mddev->sync_speed_min ? "local": "system");
2968 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2972 if (strncmp(buf, "system", 6)==0) {
2973 mddev->sync_speed_min = 0;
2976 min = simple_strtoul(buf, &e, 10);
2977 if (buf == e || (*e && *e != '\n') || min <= 0)
2979 mddev->sync_speed_min = min;
2983 static struct md_sysfs_entry md_sync_min =
2984 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2987 sync_max_show(mddev_t *mddev, char *page)
2989 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2990 mddev->sync_speed_max ? "local": "system");
2994 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2998 if (strncmp(buf, "system", 6)==0) {
2999 mddev->sync_speed_max = 0;
3002 max = simple_strtoul(buf, &e, 10);
3003 if (buf == e || (*e && *e != '\n') || max <= 0)
3005 mddev->sync_speed_max = max;
3009 static struct md_sysfs_entry md_sync_max =
3010 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3013 degraded_show(mddev_t *mddev, char *page)
3015 return sprintf(page, "%d\n", mddev->degraded);
3017 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3020 sync_force_parallel_show(mddev_t *mddev, char *page)
3022 return sprintf(page, "%d\n", mddev->parallel_resync);
3026 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3030 if (strict_strtol(buf, 10, &n))
3033 if (n != 0 && n != 1)
3036 mddev->parallel_resync = n;
3038 if (mddev->sync_thread)
3039 wake_up(&resync_wait);
3044 /* force parallel resync, even with shared block devices */
3045 static struct md_sysfs_entry md_sync_force_parallel =
3046 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3047 sync_force_parallel_show, sync_force_parallel_store);
3050 sync_speed_show(mddev_t *mddev, char *page)
3052 unsigned long resync, dt, db;
3053 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3054 dt = ((jiffies - mddev->resync_mark) / HZ);
3056 db = resync - (mddev->resync_mark_cnt);
3057 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3060 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3063 sync_completed_show(mddev_t *mddev, char *page)
3065 unsigned long max_blocks, resync;
3067 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3068 max_blocks = mddev->resync_max_sectors;
3070 max_blocks = mddev->size << 1;
3072 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3073 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3076 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3079 min_sync_show(mddev_t *mddev, char *page)
3081 return sprintf(page, "%llu\n",
3082 (unsigned long long)mddev->resync_min);
3085 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3087 unsigned long long min;
3088 if (strict_strtoull(buf, 10, &min))
3090 if (min > mddev->resync_max)
3092 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3095 /* Must be a multiple of chunk_size */
3096 if (mddev->chunk_size) {
3097 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3100 mddev->resync_min = min;
3105 static struct md_sysfs_entry md_min_sync =
3106 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3109 max_sync_show(mddev_t *mddev, char *page)
3111 if (mddev->resync_max == MaxSector)
3112 return sprintf(page, "max\n");
3114 return sprintf(page, "%llu\n",
3115 (unsigned long long)mddev->resync_max);
3118 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3120 if (strncmp(buf, "max", 3) == 0)
3121 mddev->resync_max = MaxSector;
3123 unsigned long long max;
3124 if (strict_strtoull(buf, 10, &max))
3126 if (max < mddev->resync_min)
3128 if (max < mddev->resync_max &&
3129 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3132 /* Must be a multiple of chunk_size */
3133 if (mddev->chunk_size) {
3134 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3137 mddev->resync_max = max;
3139 wake_up(&mddev->recovery_wait);
3143 static struct md_sysfs_entry md_max_sync =
3144 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3147 suspend_lo_show(mddev_t *mddev, char *page)
3149 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3153 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3156 unsigned long long new = simple_strtoull(buf, &e, 10);
3158 if (mddev->pers->quiesce == NULL)
3160 if (buf == e || (*e && *e != '\n'))
3162 if (new >= mddev->suspend_hi ||
3163 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3164 mddev->suspend_lo = new;
3165 mddev->pers->quiesce(mddev, 2);
3170 static struct md_sysfs_entry md_suspend_lo =
3171 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3175 suspend_hi_show(mddev_t *mddev, char *page)
3177 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3181 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3184 unsigned long long new = simple_strtoull(buf, &e, 10);
3186 if (mddev->pers->quiesce == NULL)
3188 if (buf == e || (*e && *e != '\n'))
3190 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3191 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3192 mddev->suspend_hi = new;
3193 mddev->pers->quiesce(mddev, 1);
3194 mddev->pers->quiesce(mddev, 0);
3199 static struct md_sysfs_entry md_suspend_hi =
3200 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3203 reshape_position_show(mddev_t *mddev, char *page)
3205 if (mddev->reshape_position != MaxSector)
3206 return sprintf(page, "%llu\n",
3207 (unsigned long long)mddev->reshape_position);
3208 strcpy(page, "none\n");
3213 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3216 unsigned long long new = simple_strtoull(buf, &e, 10);
3219 if (buf == e || (*e && *e != '\n'))
3221 mddev->reshape_position = new;
3222 mddev->delta_disks = 0;
3223 mddev->new_level = mddev->level;
3224 mddev->new_layout = mddev->layout;
3225 mddev->new_chunk = mddev->chunk_size;
3229 static struct md_sysfs_entry md_reshape_position =
3230 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3231 reshape_position_store);
3234 static struct attribute *md_default_attrs[] = {
3237 &md_raid_disks.attr,
3238 &md_chunk_size.attr,
3240 &md_resync_start.attr,
3242 &md_new_device.attr,
3243 &md_safe_delay.attr,
3244 &md_array_state.attr,
3245 &md_reshape_position.attr,
3249 static struct attribute *md_redundancy_attrs[] = {
3251 &md_mismatches.attr,
3254 &md_sync_speed.attr,
3255 &md_sync_force_parallel.attr,
3256 &md_sync_completed.attr,
3259 &md_suspend_lo.attr,
3260 &md_suspend_hi.attr,
3265 static struct attribute_group md_redundancy_group = {
3267 .attrs = md_redundancy_attrs,
3272 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3274 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3275 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3280 rv = mddev_lock(mddev);
3282 rv = entry->show(mddev, page);
3283 mddev_unlock(mddev);
3289 md_attr_store(struct kobject *kobj, struct attribute *attr,
3290 const char *page, size_t length)
3292 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3293 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3298 if (!capable(CAP_SYS_ADMIN))
3300 rv = mddev_lock(mddev);
3302 rv = entry->store(mddev, page, length);
3303 mddev_unlock(mddev);
3308 static void md_free(struct kobject *ko)
3310 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3314 static struct sysfs_ops md_sysfs_ops = {
3315 .show = md_attr_show,
3316 .store = md_attr_store,
3318 static struct kobj_type md_ktype = {
3320 .sysfs_ops = &md_sysfs_ops,
3321 .default_attrs = md_default_attrs,
3326 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3328 static DEFINE_MUTEX(disks_mutex);
3329 mddev_t *mddev = mddev_find(dev);
3330 struct gendisk *disk;
3331 int partitioned = (MAJOR(dev) != MD_MAJOR);
3332 int shift = partitioned ? MdpMinorShift : 0;
3333 int unit = MINOR(dev) >> shift;
3339 mutex_lock(&disks_mutex);
3340 if (mddev->gendisk) {
3341 mutex_unlock(&disks_mutex);
3345 disk = alloc_disk(1 << shift);
3347 mutex_unlock(&disks_mutex);
3351 disk->major = MAJOR(dev);
3352 disk->first_minor = unit << shift;
3354 sprintf(disk->disk_name, "md_d%d", unit);
3356 sprintf(disk->disk_name, "md%d", unit);
3357 disk->fops = &md_fops;
3358 disk->private_data = mddev;
3359 disk->queue = mddev->queue;
3361 mddev->gendisk = disk;
3362 mutex_unlock(&disks_mutex);
3363 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3366 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3369 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3373 static void md_safemode_timeout(unsigned long data)
3375 mddev_t *mddev = (mddev_t *) data;
3377 mddev->safemode = 1;
3378 md_wakeup_thread(mddev->thread);
3381 static int start_dirty_degraded;
3383 static int do_md_run(mddev_t * mddev)
3387 struct list_head *tmp;
3389 struct gendisk *disk;
3390 struct mdk_personality *pers;
3391 char b[BDEVNAME_SIZE];
3393 if (list_empty(&mddev->disks))
3394 /* cannot run an array with no devices.. */
3401 * Analyze all RAID superblock(s)
3403 if (!mddev->raid_disks) {
3404 if (!mddev->persistent)
3409 chunk_size = mddev->chunk_size;
3412 if (chunk_size > MAX_CHUNK_SIZE) {
3413 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3414 chunk_size, MAX_CHUNK_SIZE);
3418 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3420 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3421 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3424 if (chunk_size < PAGE_SIZE) {
3425 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3426 chunk_size, PAGE_SIZE);
3430 /* devices must have minimum size of one chunk */
3431 rdev_for_each(rdev, tmp, mddev) {
3432 if (test_bit(Faulty, &rdev->flags))
3434 if (rdev->size < chunk_size / 1024) {
3436 "md: Dev %s smaller than chunk_size:"
3438 bdevname(rdev->bdev,b),
3439 (unsigned long long)rdev->size,
3447 if (mddev->level != LEVEL_NONE)
3448 request_module("md-level-%d", mddev->level);
3449 else if (mddev->clevel[0])
3450 request_module("md-%s", mddev->clevel);
3454 * Drop all container device buffers, from now on
3455 * the only valid external interface is through the md
3458 rdev_for_each(rdev, tmp, mddev) {
3459 if (test_bit(Faulty, &rdev->flags))
3461 sync_blockdev(rdev->bdev);
3462 invalidate_bdev(rdev->bdev);
3464 /* perform some consistency tests on the device.
3465 * We don't want the data to overlap the metadata,
3466 * Internal Bitmap issues has handled elsewhere.
3468 if (rdev->data_offset < rdev->sb_offset) {
3470 rdev->data_offset + mddev->size*2
3471 > rdev->sb_offset*2) {
3472 printk("md: %s: data overlaps metadata\n",
3477 if (rdev->sb_offset*2 + rdev->sb_size/512
3478 > rdev->data_offset) {
3479 printk("md: %s: metadata overlaps data\n",
3486 md_probe(mddev->unit, NULL, NULL);
3487 disk = mddev->gendisk;
3491 spin_lock(&pers_lock);
3492 pers = find_pers(mddev->level, mddev->clevel);
3493 if (!pers || !try_module_get(pers->owner)) {
3494 spin_unlock(&pers_lock);
3495 if (mddev->level != LEVEL_NONE)
3496 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3499 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3504 spin_unlock(&pers_lock);
3505 mddev->level = pers->level;
3506 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3508 if (mddev->reshape_position != MaxSector &&
3509 pers->start_reshape == NULL) {
3510 /* This personality cannot handle reshaping... */
3512 module_put(pers->owner);
3516 if (pers->sync_request) {
3517 /* Warn if this is a potentially silly
3520 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3522 struct list_head *tmp2;
3524 rdev_for_each(rdev, tmp, mddev) {
3525 rdev_for_each(rdev2, tmp2, mddev) {
3527 rdev->bdev->bd_contains ==
3528 rdev2->bdev->bd_contains) {
3530 "%s: WARNING: %s appears to be"
3531 " on the same physical disk as"
3534 bdevname(rdev->bdev,b),
3535 bdevname(rdev2->bdev,b2));
3542 "True protection against single-disk"
3543 " failure might be compromised.\n");
3546 mddev->recovery = 0;
3547 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3548 mddev->barriers_work = 1;
3549 mddev->ok_start_degraded = start_dirty_degraded;
3552 mddev->ro = 2; /* read-only, but switch on first write */
3554 err = mddev->pers->run(mddev);
3555 if (!err && mddev->pers->sync_request) {
3556 err = bitmap_create(mddev);
3558 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3559 mdname(mddev), err);
3560 mddev->pers->stop(mddev);
3564 printk(KERN_ERR "md: pers->run() failed ...\n");
3565 module_put(mddev->pers->owner);
3567 bitmap_destroy(mddev);
3570 if (mddev->pers->sync_request) {
3571 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3573 "md: cannot register extra attributes for %s\n",
3575 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3578 atomic_set(&mddev->writes_pending,0);
3579 mddev->safemode = 0;
3580 mddev->safemode_timer.function = md_safemode_timeout;
3581 mddev->safemode_timer.data = (unsigned long) mddev;
3582 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3585 rdev_for_each(rdev, tmp, mddev)
3586 if (rdev->raid_disk >= 0) {
3588 sprintf(nm, "rd%d", rdev->raid_disk);
3589 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3590 printk("md: cannot register %s for %s\n",
3594 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3597 md_update_sb(mddev, 0);
3599 set_capacity(disk, mddev->array_size<<1);
3601 /* If we call blk_queue_make_request here, it will
3602 * re-initialise max_sectors etc which may have been
3603 * refined inside -> run. So just set the bits we need to set.
3604 * Most initialisation happended when we called
3605 * blk_queue_make_request(..., md_fail_request)
3608 mddev->queue->queuedata = mddev;
3609 mddev->queue->make_request_fn = mddev->pers->make_request;
3611 /* If there is a partially-recovered drive we need to
3612 * start recovery here. If we leave it to md_check_recovery,
3613 * it will remove the drives and not do the right thing
3615 if (mddev->degraded && !mddev->sync_thread) {
3616 struct list_head *rtmp;
3618 rdev_for_each(rdev, rtmp, mddev)
3619 if (rdev->raid_disk >= 0 &&
3620 !test_bit(In_sync, &rdev->flags) &&
3621 !test_bit(Faulty, &rdev->flags))
3622 /* complete an interrupted recovery */
3624 if (spares && mddev->pers->sync_request) {
3625 mddev->recovery = 0;
3626 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3627 mddev->sync_thread = md_register_thread(md_do_sync,
3630 if (!mddev->sync_thread) {
3631 printk(KERN_ERR "%s: could not start resync"
3634 /* leave the spares where they are, it shouldn't hurt */
3635 mddev->recovery = 0;
3639 md_wakeup_thread(mddev->thread);
3640 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3643 md_new_event(mddev);
3644 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3648 static int restart_array(mddev_t *mddev)
3650 struct gendisk *disk = mddev->gendisk;
3654 * Complain if it has no devices
3657 if (list_empty(&mddev->disks))
3665 mddev->safemode = 0;
3667 set_disk_ro(disk, 0);
3669 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3672 * Kick recovery or resync if necessary
3674 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3675 md_wakeup_thread(mddev->thread);
3676 md_wakeup_thread(mddev->sync_thread);
3685 /* similar to deny_write_access, but accounts for our holding a reference
3686 * to the file ourselves */
3687 static int deny_bitmap_write_access(struct file * file)
3689 struct inode *inode = file->f_mapping->host;
3691 spin_lock(&inode->i_lock);
3692 if (atomic_read(&inode->i_writecount) > 1) {
3693 spin_unlock(&inode->i_lock);
3696 atomic_set(&inode->i_writecount, -1);
3697 spin_unlock(&inode->i_lock);
3702 static void restore_bitmap_write_access(struct file *file)
3704 struct inode *inode = file->f_mapping->host;
3706 spin_lock(&inode->i_lock);
3707 atomic_set(&inode->i_writecount, 1);
3708 spin_unlock(&inode->i_lock);
3712 * 0 - completely stop and dis-assemble array
3713 * 1 - switch to readonly
3714 * 2 - stop but do not disassemble array
3716 static int do_md_stop(mddev_t * mddev, int mode)
3719 struct gendisk *disk = mddev->gendisk;
3722 if (atomic_read(&mddev->active)>2) {
3723 printk("md: %s still in use.\n",mdname(mddev));
3727 if (mddev->sync_thread) {
3728 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3729 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3730 md_unregister_thread(mddev->sync_thread);
3731 mddev->sync_thread = NULL;
3734 del_timer_sync(&mddev->safemode_timer);
3736 invalidate_partition(disk, 0);
3739 case 1: /* readonly */
3745 case 0: /* disassemble */
3747 bitmap_flush(mddev);
3748 md_super_wait(mddev);
3750 set_disk_ro(disk, 0);
3751 blk_queue_make_request(mddev->queue, md_fail_request);
3752 mddev->pers->stop(mddev);
3753 mddev->queue->merge_bvec_fn = NULL;
3754 mddev->queue->unplug_fn = NULL;
3755 mddev->queue->backing_dev_info.congested_fn = NULL;
3756 if (mddev->pers->sync_request)
3757 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3759 module_put(mddev->pers->owner);
3761 /* tell userspace to handle 'inactive' */
3762 sysfs_notify(&mddev->kobj, NULL, "array_state");
3764 set_capacity(disk, 0);
3770 if (!mddev->in_sync || mddev->flags) {
3771 /* mark array as shutdown cleanly */
3773 md_update_sb(mddev, 1);
3776 set_disk_ro(disk, 1);
3777 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3781 * Free resources if final stop
3785 struct list_head *tmp;
3787 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3789 bitmap_destroy(mddev);
3790 if (mddev->bitmap_file) {
3791 restore_bitmap_write_access(mddev->bitmap_file);
3792 fput(mddev->bitmap_file);
3793 mddev->bitmap_file = NULL;
3795 mddev->bitmap_offset = 0;
3797 rdev_for_each(rdev, tmp, mddev)
3798 if (rdev->raid_disk >= 0) {
3800 sprintf(nm, "rd%d", rdev->raid_disk);
3801 sysfs_remove_link(&mddev->kobj, nm);
3804 /* make sure all md_delayed_delete calls have finished */
3805 flush_scheduled_work();
3807 export_array(mddev);
3809 mddev->array_size = 0;
3811 mddev->raid_disks = 0;
3812 mddev->recovery_cp = 0;
3813 mddev->resync_min = 0;
3814 mddev->resync_max = MaxSector;
3815 mddev->reshape_position = MaxSector;
3816 mddev->external = 0;
3817 mddev->persistent = 0;
3818 mddev->level = LEVEL_NONE;
3819 mddev->clevel[0] = 0;
3822 mddev->metadata_type[0] = 0;
3823 mddev->chunk_size = 0;
3824 mddev->ctime = mddev->utime = 0;
3826 mddev->max_disks = 0;
3828 mddev->delta_disks = 0;
3829 mddev->new_level = LEVEL_NONE;
3830 mddev->new_layout = 0;
3831 mddev->new_chunk = 0;
3832 mddev->curr_resync = 0;
3833 mddev->resync_mismatches = 0;
3834 mddev->suspend_lo = mddev->suspend_hi = 0;
3835 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3836 mddev->recovery = 0;
3839 mddev->degraded = 0;
3840 mddev->barriers_work = 0;
3841 mddev->safemode = 0;
3843 } else if (mddev->pers)
3844 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3847 md_new_event(mddev);
3853 static void autorun_array(mddev_t *mddev)
3856 struct list_head *tmp;
3859 if (list_empty(&mddev->disks))
3862 printk(KERN_INFO "md: running: ");
3864 rdev_for_each(rdev, tmp, mddev) {
3865 char b[BDEVNAME_SIZE];
3866 printk("<%s>", bdevname(rdev->bdev,b));
3870 err = do_md_run (mddev);
3872 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3873 do_md_stop (mddev, 0);
3878 * lets try to run arrays based on all disks that have arrived
3879 * until now. (those are in pending_raid_disks)
3881 * the method: pick the first pending disk, collect all disks with
3882 * the same UUID, remove all from the pending list and put them into
3883 * the 'same_array' list. Then order this list based on superblock
3884 * update time (freshest comes first), kick out 'old' disks and
3885 * compare superblocks. If everything's fine then run it.
3887 * If "unit" is allocated, then bump its reference count
3889 static void autorun_devices(int part)
3891 struct list_head *tmp;
3892 mdk_rdev_t *rdev0, *rdev;
3894 char b[BDEVNAME_SIZE];
3896 printk(KERN_INFO "md: autorun ...\n");
3897 while (!list_empty(&pending_raid_disks)) {
3900 LIST_HEAD(candidates);
3901 rdev0 = list_entry(pending_raid_disks.next,
3902 mdk_rdev_t, same_set);
3904 printk(KERN_INFO "md: considering %s ...\n",
3905 bdevname(rdev0->bdev,b));
3906 INIT_LIST_HEAD(&candidates);
3907 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3908 if (super_90_load(rdev, rdev0, 0) >= 0) {
3909 printk(KERN_INFO "md: adding %s ...\n",
3910 bdevname(rdev->bdev,b));
3911 list_move(&rdev->same_set, &candidates);
3914 * now we have a set of devices, with all of them having
3915 * mostly sane superblocks. It's time to allocate the
3919 dev = MKDEV(mdp_major,
3920 rdev0->preferred_minor << MdpMinorShift);
3921 unit = MINOR(dev) >> MdpMinorShift;
3923 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3926 if (rdev0->preferred_minor != unit) {
3927 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3928 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3932 md_probe(dev, NULL, NULL);
3933 mddev = mddev_find(dev);
3934 if (!mddev || !mddev->gendisk) {
3938 "md: cannot allocate memory for md drive.\n");
3941 if (mddev_lock(mddev))
3942 printk(KERN_WARNING "md: %s locked, cannot run\n",
3944 else if (mddev->raid_disks || mddev->major_version
3945 || !list_empty(&mddev->disks)) {
3947 "md: %s already running, cannot run %s\n",
3948 mdname(mddev), bdevname(rdev0->bdev,b));
3949 mddev_unlock(mddev);
3951 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3952 mddev->persistent = 1;
3953 rdev_for_each_list(rdev, tmp, candidates) {
3954 list_del_init(&rdev->same_set);
3955 if (bind_rdev_to_array(rdev, mddev))
3958 autorun_array(mddev);
3959 mddev_unlock(mddev);
3961 /* on success, candidates will be empty, on error
3964 rdev_for_each_list(rdev, tmp, candidates)
3968 printk(KERN_INFO "md: ... autorun DONE.\n");
3970 #endif /* !MODULE */
3972 static int get_version(void __user * arg)
3976 ver.major = MD_MAJOR_VERSION;
3977 ver.minor = MD_MINOR_VERSION;
3978 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3980 if (copy_to_user(arg, &ver, sizeof(ver)))
3986 static int get_array_info(mddev_t * mddev, void __user * arg)
3988 mdu_array_info_t info;
3989 int nr,working,active,failed,spare;
3991 struct list_head *tmp;
3993 nr=working=active=failed=spare=0;
3994 rdev_for_each(rdev, tmp, mddev) {
3996 if (test_bit(Faulty, &rdev->flags))
4000 if (test_bit(In_sync, &rdev->flags))
4007 info.major_version = mddev->major_version;
4008 info.minor_version = mddev->minor_version;
4009 info.patch_version = MD_PATCHLEVEL_VERSION;
4010 info.ctime = mddev->ctime;
4011 info.level = mddev->level;
4012 info.size = mddev->size;
4013 if (info.size != mddev->size) /* overflow */
4016 info.raid_disks = mddev->raid_disks;
4017 info.md_minor = mddev->md_minor;
4018 info.not_persistent= !mddev->persistent;
4020 info.utime = mddev->utime;
4023 info.state = (1<<MD_SB_CLEAN);
4024 if (mddev->bitmap && mddev->bitmap_offset)
4025 info.state = (1<<MD_SB_BITMAP_PRESENT);
4026 info.active_disks = active;
4027 info.working_disks = working;
4028 info.failed_disks = failed;
4029 info.spare_disks = spare;
4031 info.layout = mddev->layout;
4032 info.chunk_size = mddev->chunk_size;
4034 if (copy_to_user(arg, &info, sizeof(info)))
4040 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4042 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4043 char *ptr, *buf = NULL;
4046 md_allow_write(mddev);
4048 file = kmalloc(sizeof(*file), GFP_KERNEL);
4052 /* bitmap disabled, zero the first byte and copy out */
4053 if (!mddev->bitmap || !mddev->bitmap->file) {
4054 file->pathname[0] = '\0';
4058 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4062 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4066 strcpy(file->pathname, ptr);
4070 if (copy_to_user(arg, file, sizeof(*file)))
4078 static int get_disk_info(mddev_t * mddev, void __user * arg)
4080 mdu_disk_info_t info;
4084 if (copy_from_user(&info, arg, sizeof(info)))
4089 rdev = find_rdev_nr(mddev, nr);
4091 info.major = MAJOR(rdev->bdev->bd_dev);
4092 info.minor = MINOR(rdev->bdev->bd_dev);
4093 info.raid_disk = rdev->raid_disk;
4095 if (test_bit(Faulty, &rdev->flags))
4096 info.state |= (1<<MD_DISK_FAULTY);
4097 else if (test_bit(In_sync, &rdev->flags)) {
4098 info.state |= (1<<MD_DISK_ACTIVE);
4099 info.state |= (1<<MD_DISK_SYNC);
4101 if (test_bit(WriteMostly, &rdev->flags))
4102 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4104 info.major = info.minor = 0;
4105 info.raid_disk = -1;
4106 info.state = (1<<MD_DISK_REMOVED);
4109 if (copy_to_user(arg, &info, sizeof(info)))
4115 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4117 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4119 dev_t dev = MKDEV(info->major,info->minor);
4121 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4124 if (!mddev->raid_disks) {
4126 /* expecting a device which has a superblock */
4127 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4130 "md: md_import_device returned %ld\n",
4132 return PTR_ERR(rdev);
4134 if (!list_empty(&mddev->disks)) {
4135 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4136 mdk_rdev_t, same_set);
4137 int err = super_types[mddev->major_version]
4138 .load_super(rdev, rdev0, mddev->minor_version);
4141 "md: %s has different UUID to %s\n",
4142 bdevname(rdev->bdev,b),
4143 bdevname(rdev0->bdev,b2));
4148 err = bind_rdev_to_array(rdev, mddev);
4155 * add_new_disk can be used once the array is assembled
4156 * to add "hot spares". They must already have a superblock
4161 if (!mddev->pers->hot_add_disk) {
4163 "%s: personality does not support diskops!\n",
4167 if (mddev->persistent)
4168 rdev = md_import_device(dev, mddev->major_version,
4169 mddev->minor_version);
4171 rdev = md_import_device(dev, -1, -1);
4174 "md: md_import_device returned %ld\n",
4176 return PTR_ERR(rdev);
4178 /* set save_raid_disk if appropriate */
4179 if (!mddev->persistent) {
4180 if (info->state & (1<<MD_DISK_SYNC) &&
4181 info->raid_disk < mddev->raid_disks)
4182 rdev->raid_disk = info->raid_disk;
4184 rdev->raid_disk = -1;
4186 super_types[mddev->major_version].
4187 validate_super(mddev, rdev);
4188 rdev->saved_raid_disk = rdev->raid_disk;
4190 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4191 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4192 set_bit(WriteMostly, &rdev->flags);
4194 rdev->raid_disk = -1;
4195 err = bind_rdev_to_array(rdev, mddev);
4196 if (!err && !mddev->pers->hot_remove_disk) {
4197 /* If there is hot_add_disk but no hot_remove_disk
4198 * then added disks for geometry changes,
4199 * and should be added immediately.
4201 super_types[mddev->major_version].
4202 validate_super(mddev, rdev);
4203 err = mddev->pers->hot_add_disk(mddev, rdev);
4205 unbind_rdev_from_array(rdev);
4210 md_update_sb(mddev, 1);
4211 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4212 md_wakeup_thread(mddev->thread);
4216 /* otherwise, add_new_disk is only allowed
4217 * for major_version==0 superblocks
4219 if (mddev->major_version != 0) {
4220 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4225 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4227 rdev = md_import_device (dev, -1, 0);
4230 "md: error, md_import_device() returned %ld\n",
4232 return PTR_ERR(rdev);
4234 rdev->desc_nr = info->number;
4235 if (info->raid_disk < mddev->raid_disks)
4236 rdev->raid_disk = info->raid_disk;
4238 rdev->raid_disk = -1;
4240 if (rdev->raid_disk < mddev->raid_disks)
4241 if (info->state & (1<<MD_DISK_SYNC))
4242 set_bit(In_sync, &rdev->flags);
4244 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4245 set_bit(WriteMostly, &rdev->flags);
4247 if (!mddev->persistent) {
4248 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4249 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4251 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4252 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4254 err = bind_rdev_to_array(rdev, mddev);
4264 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4266 char b[BDEVNAME_SIZE];
4272 rdev = find_rdev(mddev, dev);
4276 if (rdev->raid_disk >= 0)
4279 kick_rdev_from_array(rdev);
4280 md_update_sb(mddev, 1);
4281 md_new_event(mddev);
4285 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4286 bdevname(rdev->bdev,b), mdname(mddev));
4290 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4292 char b[BDEVNAME_SIZE];
4300 if (mddev->major_version != 0) {
4301 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4302 " version-0 superblocks.\n",
4306 if (!mddev->pers->hot_add_disk) {
4308 "%s: personality does not support diskops!\n",
4313 rdev = md_import_device (dev, -1, 0);
4316 "md: error, md_import_device() returned %ld\n",
4321 if (mddev->persistent)
4322 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4325 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4327 size = calc_dev_size(rdev, mddev->chunk_size);
4330 if (test_bit(Faulty, &rdev->flags)) {
4332 "md: can not hot-add faulty %s disk to %s!\n",
4333 bdevname(rdev->bdev,b), mdname(mddev));
4337 clear_bit(In_sync, &rdev->flags);
4339 rdev->saved_raid_disk = -1;
4340 err = bind_rdev_to_array(rdev, mddev);
4345 * The rest should better be atomic, we can have disk failures
4346 * noticed in interrupt contexts ...
4349 if (rdev->desc_nr == mddev->max_disks) {
4350 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4353 goto abort_unbind_export;
4356 rdev->raid_disk = -1;
4358 md_update_sb(mddev, 1);
4361 * Kick recovery, maybe this spare has to be added to the
4362 * array immediately.
4364 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4365 md_wakeup_thread(mddev->thread);
4366 md_new_event(mddev);
4369 abort_unbind_export:
4370 unbind_rdev_from_array(rdev);
4377 static int set_bitmap_file(mddev_t *mddev, int fd)
4382 if (!mddev->pers->quiesce)
4384 if (mddev->recovery || mddev->sync_thread)
4386 /* we should be able to change the bitmap.. */
4392 return -EEXIST; /* cannot add when bitmap is present */
4393 mddev->bitmap_file = fget(fd);
4395 if (mddev->bitmap_file == NULL) {
4396 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4401 err = deny_bitmap_write_access(mddev->bitmap_file);
4403 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4405 fput(mddev->bitmap_file);
4406 mddev->bitmap_file = NULL;
4409 mddev->bitmap_offset = 0; /* file overrides offset */
4410 } else if (mddev->bitmap == NULL)
4411 return -ENOENT; /* cannot remove what isn't there */
4414 mddev->pers->quiesce(mddev, 1);
4416 err = bitmap_create(mddev);
4417 if (fd < 0 || err) {
4418 bitmap_destroy(mddev);
4419 fd = -1; /* make sure to put the file */
4421 mddev->pers->quiesce(mddev, 0);
4424 if (mddev->bitmap_file) {
4425 restore_bitmap_write_access(mddev->bitmap_file);
4426 fput(mddev->bitmap_file);
4428 mddev->bitmap_file = NULL;
4435 * set_array_info is used two different ways
4436 * The original usage is when creating a new array.
4437 * In this usage, raid_disks is > 0 and it together with
4438 * level, size, not_persistent,layout,chunksize determine the
4439 * shape of the array.
4440 * This will always create an array with a type-0.90.0 superblock.
4441 * The newer usage is when assembling an array.
4442 * In this case raid_disks will be 0, and the major_version field is
4443 * use to determine which style super-blocks are to be found on the devices.
4444 * The minor and patch _version numbers are also kept incase the
4445 * super_block handler wishes to interpret them.
4447 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4450 if (info->raid_disks == 0) {
4451 /* just setting version number for superblock loading */
4452 if (info->major_version < 0 ||
4453 info->major_version >= ARRAY_SIZE(super_types) ||
4454 super_types[info->major_version].name == NULL) {
4455 /* maybe try to auto-load a module? */
4457 "md: superblock version %d not known\n",
4458 info->major_version);
4461 mddev->major_version = info->major_version;
4462 mddev->minor_version = info->minor_version;
4463 mddev->patch_version = info->patch_version;
4464 mddev->persistent = !info->not_persistent;
4467 mddev->major_version = MD_MAJOR_VERSION;
4468 mddev->minor_version = MD_MINOR_VERSION;
4469 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4470 mddev->ctime = get_seconds();
4472 mddev->level = info->level;
4473 mddev->clevel[0] = 0;
4474 mddev->size = info->size;
4475 mddev->raid_disks = info->raid_disks;
4476 /* don't set md_minor, it is determined by which /dev/md* was
4479 if (info->state & (1<<MD_SB_CLEAN))
4480 mddev->recovery_cp = MaxSector;
4482 mddev->recovery_cp = 0;
4483 mddev->persistent = ! info->not_persistent;
4484 mddev->external = 0;
4486 mddev->layout = info->layout;
4487 mddev->chunk_size = info->chunk_size;
4489 mddev->max_disks = MD_SB_DISKS;
4491 if (mddev->persistent)
4493 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4495 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4496 mddev->bitmap_offset = 0;
4498 mddev->reshape_position = MaxSector;
4501 * Generate a 128 bit UUID
4503 get_random_bytes(mddev->uuid, 16);
4505 mddev->new_level = mddev->level;
4506 mddev->new_chunk = mddev->chunk_size;
4507 mddev->new_layout = mddev->layout;
4508 mddev->delta_disks = 0;
4513 static int update_size(mddev_t *mddev, unsigned long size)
4517 struct list_head *tmp;
4518 int fit = (size == 0);
4520 if (mddev->pers->resize == NULL)
4522 /* The "size" is the amount of each device that is used.
4523 * This can only make sense for arrays with redundancy.
4524 * linear and raid0 always use whatever space is available
4525 * We can only consider changing the size if no resync
4526 * or reconstruction is happening, and if the new size
4527 * is acceptable. It must fit before the sb_offset or,
4528 * if that is <data_offset, it must fit before the
4529 * size of each device.
4530 * If size is zero, we find the largest size that fits.
4532 if (mddev->sync_thread)
4534 rdev_for_each(rdev, tmp, mddev) {
4536 avail = rdev->size * 2;
4538 if (fit && (size == 0 || size > avail/2))
4540 if (avail < ((sector_t)size << 1))
4543 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4545 struct block_device *bdev;
4547 bdev = bdget_disk(mddev->gendisk, 0);
4549 mutex_lock(&bdev->bd_inode->i_mutex);
4550 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4551 mutex_unlock(&bdev->bd_inode->i_mutex);
4558 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4561 /* change the number of raid disks */
4562 if (mddev->pers->check_reshape == NULL)
4564 if (raid_disks <= 0 ||
4565 raid_disks >= mddev->max_disks)
4567 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4569 mddev->delta_disks = raid_disks - mddev->raid_disks;
4571 rv = mddev->pers->check_reshape(mddev);
4577 * update_array_info is used to change the configuration of an
4579 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4580 * fields in the info are checked against the array.
4581 * Any differences that cannot be handled will cause an error.
4582 * Normally, only one change can be managed at a time.
4584 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4590 /* calculate expected state,ignoring low bits */
4591 if (mddev->bitmap && mddev->bitmap_offset)
4592 state |= (1 << MD_SB_BITMAP_PRESENT);
4594 if (mddev->major_version != info->major_version ||
4595 mddev->minor_version != info->minor_version ||
4596 /* mddev->patch_version != info->patch_version || */
4597 mddev->ctime != info->ctime ||
4598 mddev->level != info->level ||
4599 /* mddev->layout != info->layout || */
4600 !mddev->persistent != info->not_persistent||
4601 mddev->chunk_size != info->chunk_size ||
4602 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4603 ((state^info->state) & 0xfffffe00)
4606 /* Check there is only one change */
4607 if (info->size >= 0 && mddev->size != info->size) cnt++;
4608 if (mddev->raid_disks != info->raid_disks) cnt++;
4609 if (mddev->layout != info->layout) cnt++;
4610 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4611 if (cnt == 0) return 0;
4612 if (cnt > 1) return -EINVAL;
4614 if (mddev->layout != info->layout) {
4616 * we don't need to do anything at the md level, the
4617 * personality will take care of it all.
4619 if (mddev->pers->reconfig == NULL)
4622 return mddev->pers->reconfig(mddev, info->layout, -1);
4624 if (info->size >= 0 && mddev->size != info->size)
4625 rv = update_size(mddev, info->size);
4627 if (mddev->raid_disks != info->raid_disks)
4628 rv = update_raid_disks(mddev, info->raid_disks);
4630 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4631 if (mddev->pers->quiesce == NULL)
4633 if (mddev->recovery || mddev->sync_thread)
4635 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4636 /* add the bitmap */
4639 if (mddev->default_bitmap_offset == 0)
4641 mddev->bitmap_offset = mddev->default_bitmap_offset;
4642 mddev->pers->quiesce(mddev, 1);
4643 rv = bitmap_create(mddev);
4645 bitmap_destroy(mddev);
4646 mddev->pers->quiesce(mddev, 0);
4648 /* remove the bitmap */
4651 if (mddev->bitmap->file)
4653 mddev->pers->quiesce(mddev, 1);
4654 bitmap_destroy(mddev);
4655 mddev->pers->quiesce(mddev, 0);
4656 mddev->bitmap_offset = 0;
4659 md_update_sb(mddev, 1);
4663 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4667 if (mddev->pers == NULL)
4670 rdev = find_rdev(mddev, dev);
4674 md_error(mddev, rdev);
4678 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4680 mddev_t *mddev = bdev->bd_disk->private_data;
4684 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4688 static int md_ioctl(struct inode *inode, struct file *file,
4689 unsigned int cmd, unsigned long arg)
4692 void __user *argp = (void __user *)arg;
4693 mddev_t *mddev = NULL;
4695 if (!capable(CAP_SYS_ADMIN))
4699 * Commands dealing with the RAID driver but not any
4705 err = get_version(argp);
4708 case PRINT_RAID_DEBUG:
4716 autostart_arrays(arg);
4723 * Commands creating/starting a new array:
4726 mddev = inode->i_bdev->bd_disk->private_data;
4733 err = mddev_lock(mddev);
4736 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4743 case SET_ARRAY_INFO:
4745 mdu_array_info_t info;
4747 memset(&info, 0, sizeof(info));
4748 else if (copy_from_user(&info, argp, sizeof(info))) {
4753 err = update_array_info(mddev, &info);
4755 printk(KERN_WARNING "md: couldn't update"
4756 " array info. %d\n", err);
4761 if (!list_empty(&mddev->disks)) {
4763 "md: array %s already has disks!\n",
4768 if (mddev->raid_disks) {
4770 "md: array %s already initialised!\n",
4775 err = set_array_info(mddev, &info);
4777 printk(KERN_WARNING "md: couldn't set"
4778 " array info. %d\n", err);
4788 * Commands querying/configuring an existing array:
4790 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4791 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4792 if ((!mddev->raid_disks && !mddev->external)
4793 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4794 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4795 && cmd != GET_BITMAP_FILE) {
4801 * Commands even a read-only array can execute:
4805 case GET_ARRAY_INFO:
4806 err = get_array_info(mddev, argp);
4809 case GET_BITMAP_FILE:
4810 err = get_bitmap_file(mddev, argp);
4814 err = get_disk_info(mddev, argp);
4817 case RESTART_ARRAY_RW:
4818 err = restart_array(mddev);
4822 err = do_md_stop (mddev, 0);
4826 err = do_md_stop (mddev, 1);
4830 * We have a problem here : there is no easy way to give a CHS
4831 * virtual geometry. We currently pretend that we have a 2 heads
4832 * 4 sectors (with a BIG number of cylinders...). This drives
4833 * dosfs just mad... ;-)
4838 * The remaining ioctls are changing the state of the
4839 * superblock, so we do not allow them on read-only arrays.
4840 * However non-MD ioctls (e.g. get-size) will still come through
4841 * here and hit the 'default' below, so only disallow
4842 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4844 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4845 mddev->ro && mddev->pers) {
4846 if (mddev->ro == 2) {
4848 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4849 md_wakeup_thread(mddev->thread);
4861 mdu_disk_info_t info;
4862 if (copy_from_user(&info, argp, sizeof(info)))
4865 err = add_new_disk(mddev, &info);
4869 case HOT_REMOVE_DISK:
4870 err = hot_remove_disk(mddev, new_decode_dev(arg));
4874 err = hot_add_disk(mddev, new_decode_dev(arg));
4877 case SET_DISK_FAULTY:
4878 err = set_disk_faulty(mddev, new_decode_dev(arg));
4882 err = do_md_run (mddev);
4885 case SET_BITMAP_FILE:
4886 err = set_bitmap_file(mddev, (int)arg);
4896 mddev_unlock(mddev);
4906 static int md_open(struct inode *inode, struct file *file)
4909 * Succeed if we can lock the mddev, which confirms that
4910 * it isn't being stopped right now.
4912 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4915 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4920 mddev_unlock(mddev);
4922 check_disk_change(inode->i_bdev);
4927 static int md_release(struct inode *inode, struct file * file)
4929 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4937 static int md_media_changed(struct gendisk *disk)
4939 mddev_t *mddev = disk->private_data;
4941 return mddev->changed;
4944 static int md_revalidate(struct gendisk *disk)
4946 mddev_t *mddev = disk->private_data;
4951 static struct block_device_operations md_fops =
4953 .owner = THIS_MODULE,
4955 .release = md_release,
4957 .getgeo = md_getgeo,
4958 .media_changed = md_media_changed,
4959 .revalidate_disk= md_revalidate,
4962 static int md_thread(void * arg)
4964 mdk_thread_t *thread = arg;
4967 * md_thread is a 'system-thread', it's priority should be very
4968 * high. We avoid resource deadlocks individually in each
4969 * raid personality. (RAID5 does preallocation) We also use RR and
4970 * the very same RT priority as kswapd, thus we will never get
4971 * into a priority inversion deadlock.
4973 * we definitely have to have equal or higher priority than
4974 * bdflush, otherwise bdflush will deadlock if there are too
4975 * many dirty RAID5 blocks.
4978 allow_signal(SIGKILL);
4979 while (!kthread_should_stop()) {
4981 /* We need to wait INTERRUPTIBLE so that
4982 * we don't add to the load-average.
4983 * That means we need to be sure no signals are
4986 if (signal_pending(current))
4987 flush_signals(current);
4989 wait_event_interruptible_timeout
4991 test_bit(THREAD_WAKEUP, &thread->flags)
4992 || kthread_should_stop(),
4995 clear_bit(THREAD_WAKEUP, &thread->flags);
4997 thread->run(thread->mddev);
5003 void md_wakeup_thread(mdk_thread_t *thread)
5006 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5007 set_bit(THREAD_WAKEUP, &thread->flags);
5008 wake_up(&thread->wqueue);
5012 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5015 mdk_thread_t *thread;
5017 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5021 init_waitqueue_head(&thread->wqueue);
5024 thread->mddev = mddev;
5025 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5026 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5027 if (IS_ERR(thread->tsk)) {
5034 void md_unregister_thread(mdk_thread_t *thread)
5036 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5038 kthread_stop(thread->tsk);
5042 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5049 if (!rdev || test_bit(Faulty, &rdev->flags))
5052 if (mddev->external)
5053 set_bit(Blocked, &rdev->flags);
5055 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5057 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5058 __builtin_return_address(0),__builtin_return_address(1),
5059 __builtin_return_address(2),__builtin_return_address(3));
5063 if (!mddev->pers->error_handler)
5065 mddev->pers->error_handler(mddev,rdev);
5066 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5067 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5068 md_wakeup_thread(mddev->thread);
5069 md_new_event_inintr(mddev);
5072 /* seq_file implementation /proc/mdstat */
5074 static void status_unused(struct seq_file *seq)
5078 struct list_head *tmp;
5080 seq_printf(seq, "unused devices: ");
5082 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5083 char b[BDEVNAME_SIZE];
5085 seq_printf(seq, "%s ",
5086 bdevname(rdev->bdev,b));
5089 seq_printf(seq, "<none>");
5091 seq_printf(seq, "\n");
5095 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5097 sector_t max_blocks, resync, res;
5098 unsigned long dt, db, rt;
5100 unsigned int per_milli;
5102 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5104 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5105 max_blocks = mddev->resync_max_sectors >> 1;
5107 max_blocks = mddev->size;
5110 * Should not happen.
5116 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5117 * in a sector_t, and (max_blocks>>scale) will fit in a
5118 * u32, as those are the requirements for sector_div.
5119 * Thus 'scale' must be at least 10
5122 if (sizeof(sector_t) > sizeof(unsigned long)) {
5123 while ( max_blocks/2 > (1ULL<<(scale+32)))
5126 res = (resync>>scale)*1000;
5127 sector_div(res, (u32)((max_blocks>>scale)+1));
5131 int i, x = per_milli/50, y = 20-x;
5132 seq_printf(seq, "[");
5133 for (i = 0; i < x; i++)
5134 seq_printf(seq, "=");
5135 seq_printf(seq, ">");
5136 for (i = 0; i < y; i++)
5137 seq_printf(seq, ".");
5138 seq_printf(seq, "] ");
5140 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5141 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5143 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5145 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5146 "resync" : "recovery"))),
5147 per_milli/10, per_milli % 10,
5148 (unsigned long long) resync,
5149 (unsigned long long) max_blocks);
5152 * We do not want to overflow, so the order of operands and
5153 * the * 100 / 100 trick are important. We do a +1 to be
5154 * safe against division by zero. We only estimate anyway.
5156 * dt: time from mark until now
5157 * db: blocks written from mark until now
5158 * rt: remaining time
5160 dt = ((jiffies - mddev->resync_mark) / HZ);
5162 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5163 - mddev->resync_mark_cnt;
5164 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5166 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5168 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5171 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5173 struct list_head *tmp;
5183 spin_lock(&all_mddevs_lock);
5184 list_for_each(tmp,&all_mddevs)
5186 mddev = list_entry(tmp, mddev_t, all_mddevs);
5188 spin_unlock(&all_mddevs_lock);
5191 spin_unlock(&all_mddevs_lock);
5193 return (void*)2;/* tail */
5197 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5199 struct list_head *tmp;
5200 mddev_t *next_mddev, *mddev = v;
5206 spin_lock(&all_mddevs_lock);
5208 tmp = all_mddevs.next;
5210 tmp = mddev->all_mddevs.next;
5211 if (tmp != &all_mddevs)
5212 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5214 next_mddev = (void*)2;
5217 spin_unlock(&all_mddevs_lock);
5225 static void md_seq_stop(struct seq_file *seq, void *v)
5229 if (mddev && v != (void*)1 && v != (void*)2)
5233 struct mdstat_info {
5237 static int md_seq_show(struct seq_file *seq, void *v)
5241 struct list_head *tmp2;
5243 struct mdstat_info *mi = seq->private;
5244 struct bitmap *bitmap;
5246 if (v == (void*)1) {
5247 struct mdk_personality *pers;
5248 seq_printf(seq, "Personalities : ");
5249 spin_lock(&pers_lock);
5250 list_for_each_entry(pers, &pers_list, list)
5251 seq_printf(seq, "[%s] ", pers->name);
5253 spin_unlock(&pers_lock);
5254 seq_printf(seq, "\n");
5255 mi->event = atomic_read(&md_event_count);
5258 if (v == (void*)2) {
5263 if (mddev_lock(mddev) < 0)
5266 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5267 seq_printf(seq, "%s : %sactive", mdname(mddev),
5268 mddev->pers ? "" : "in");
5271 seq_printf(seq, " (read-only)");
5273 seq_printf(seq, " (auto-read-only)");
5274 seq_printf(seq, " %s", mddev->pers->name);
5278 rdev_for_each(rdev, tmp2, mddev) {
5279 char b[BDEVNAME_SIZE];
5280 seq_printf(seq, " %s[%d]",
5281 bdevname(rdev->bdev,b), rdev->desc_nr);
5282 if (test_bit(WriteMostly, &rdev->flags))
5283 seq_printf(seq, "(W)");
5284 if (test_bit(Faulty, &rdev->flags)) {
5285 seq_printf(seq, "(F)");
5287 } else if (rdev->raid_disk < 0)
5288 seq_printf(seq, "(S)"); /* spare */
5292 if (!list_empty(&mddev->disks)) {
5294 seq_printf(seq, "\n %llu blocks",
5295 (unsigned long long)mddev->array_size);
5297 seq_printf(seq, "\n %llu blocks",
5298 (unsigned long long)size);
5300 if (mddev->persistent) {
5301 if (mddev->major_version != 0 ||
5302 mddev->minor_version != 90) {
5303 seq_printf(seq," super %d.%d",
5304 mddev->major_version,
5305 mddev->minor_version);
5307 } else if (mddev->external)
5308 seq_printf(seq, " super external:%s",
5309 mddev->metadata_type);
5311 seq_printf(seq, " super non-persistent");
5314 mddev->pers->status (seq, mddev);
5315 seq_printf(seq, "\n ");
5316 if (mddev->pers->sync_request) {
5317 if (mddev->curr_resync > 2) {
5318 status_resync (seq, mddev);
5319 seq_printf(seq, "\n ");
5320 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5321 seq_printf(seq, "\tresync=DELAYED\n ");
5322 else if (mddev->recovery_cp < MaxSector)
5323 seq_printf(seq, "\tresync=PENDING\n ");
5326 seq_printf(seq, "\n ");
5328 if ((bitmap = mddev->bitmap)) {
5329 unsigned long chunk_kb;
5330 unsigned long flags;
5331 spin_lock_irqsave(&bitmap->lock, flags);
5332 chunk_kb = bitmap->chunksize >> 10;
5333 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5335 bitmap->pages - bitmap->missing_pages,
5337 (bitmap->pages - bitmap->missing_pages)
5338 << (PAGE_SHIFT - 10),
5339 chunk_kb ? chunk_kb : bitmap->chunksize,
5340 chunk_kb ? "KB" : "B");
5342 seq_printf(seq, ", file: ");
5343 seq_path(seq, &bitmap->file->f_path, " \t\n");
5346 seq_printf(seq, "\n");
5347 spin_unlock_irqrestore(&bitmap->lock, flags);
5350 seq_printf(seq, "\n");
5352 mddev_unlock(mddev);
5357 static struct seq_operations md_seq_ops = {
5358 .start = md_seq_start,
5359 .next = md_seq_next,
5360 .stop = md_seq_stop,
5361 .show = md_seq_show,
5364 static int md_seq_open(struct inode *inode, struct file *file)
5367 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5371 error = seq_open(file, &md_seq_ops);
5375 struct seq_file *p = file->private_data;
5377 mi->event = atomic_read(&md_event_count);
5382 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5384 struct seq_file *m = filp->private_data;
5385 struct mdstat_info *mi = m->private;
5388 poll_wait(filp, &md_event_waiters, wait);
5390 /* always allow read */
5391 mask = POLLIN | POLLRDNORM;
5393 if (mi->event != atomic_read(&md_event_count))
5394 mask |= POLLERR | POLLPRI;
5398 static const struct file_operations md_seq_fops = {
5399 .owner = THIS_MODULE,
5400 .open = md_seq_open,
5402 .llseek = seq_lseek,
5403 .release = seq_release_private,
5404 .poll = mdstat_poll,
5407 int register_md_personality(struct mdk_personality *p)
5409 spin_lock(&pers_lock);
5410 list_add_tail(&p->list, &pers_list);
5411 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5412 spin_unlock(&pers_lock);
5416 int unregister_md_personality(struct mdk_personality *p)
5418 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5419 spin_lock(&pers_lock);
5420 list_del_init(&p->list);
5421 spin_unlock(&pers_lock);
5425 static int is_mddev_idle(mddev_t *mddev)
5428 struct list_head *tmp;
5433 rdev_for_each(rdev, tmp, mddev) {
5434 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5435 curr_events = disk_stat_read(disk, sectors[0]) +
5436 disk_stat_read(disk, sectors[1]) -
5437 atomic_read(&disk->sync_io);
5438 /* sync IO will cause sync_io to increase before the disk_stats
5439 * as sync_io is counted when a request starts, and
5440 * disk_stats is counted when it completes.
5441 * So resync activity will cause curr_events to be smaller than
5442 * when there was no such activity.
5443 * non-sync IO will cause disk_stat to increase without
5444 * increasing sync_io so curr_events will (eventually)
5445 * be larger than it was before. Once it becomes
5446 * substantially larger, the test below will cause
5447 * the array to appear non-idle, and resync will slow
5449 * If there is a lot of outstanding resync activity when
5450 * we set last_event to curr_events, then all that activity
5451 * completing might cause the array to appear non-idle
5452 * and resync will be slowed down even though there might
5453 * not have been non-resync activity. This will only
5454 * happen once though. 'last_events' will soon reflect
5455 * the state where there is little or no outstanding
5456 * resync requests, and further resync activity will
5457 * always make curr_events less than last_events.
5460 if (curr_events - rdev->last_events > 4096) {
5461 rdev->last_events = curr_events;
5468 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5470 /* another "blocks" (512byte) blocks have been synced */
5471 atomic_sub(blocks, &mddev->recovery_active);
5472 wake_up(&mddev->recovery_wait);
5474 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5475 md_wakeup_thread(mddev->thread);
5476 // stop recovery, signal do_sync ....
5481 /* md_write_start(mddev, bi)
5482 * If we need to update some array metadata (e.g. 'active' flag
5483 * in superblock) before writing, schedule a superblock update
5484 * and wait for it to complete.
5486 void md_write_start(mddev_t *mddev, struct bio *bi)
5488 if (bio_data_dir(bi) != WRITE)
5491 BUG_ON(mddev->ro == 1);
5492 if (mddev->ro == 2) {
5493 /* need to switch to read/write */
5495 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5496 md_wakeup_thread(mddev->thread);
5497 md_wakeup_thread(mddev->sync_thread);
5499 atomic_inc(&mddev->writes_pending);
5500 if (mddev->safemode == 1)
5501 mddev->safemode = 0;
5502 if (mddev->in_sync) {
5503 spin_lock_irq(&mddev->write_lock);
5504 if (mddev->in_sync) {
5506 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5507 md_wakeup_thread(mddev->thread);
5509 spin_unlock_irq(&mddev->write_lock);
5510 sysfs_notify(&mddev->kobj, NULL, "array_state");
5512 wait_event(mddev->sb_wait,
5513 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5514 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5517 void md_write_end(mddev_t *mddev)
5519 if (atomic_dec_and_test(&mddev->writes_pending)) {
5520 if (mddev->safemode == 2)
5521 md_wakeup_thread(mddev->thread);
5522 else if (mddev->safemode_delay)
5523 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5527 /* md_allow_write(mddev)
5528 * Calling this ensures that the array is marked 'active' so that writes
5529 * may proceed without blocking. It is important to call this before
5530 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5531 * Must be called with mddev_lock held.
5533 void md_allow_write(mddev_t *mddev)
5540 spin_lock_irq(&mddev->write_lock);
5541 if (mddev->in_sync) {
5543 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5544 if (mddev->safemode_delay &&
5545 mddev->safemode == 0)
5546 mddev->safemode = 1;
5547 spin_unlock_irq(&mddev->write_lock);
5548 md_update_sb(mddev, 0);
5550 sysfs_notify(&mddev->kobj, NULL, "array_state");
5551 /* wait for the dirty state to be recorded in the metadata */
5552 wait_event(mddev->sb_wait,
5553 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5554 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5556 spin_unlock_irq(&mddev->write_lock);
5558 EXPORT_SYMBOL_GPL(md_allow_write);
5560 #define SYNC_MARKS 10
5561 #define SYNC_MARK_STEP (3*HZ)
5562 void md_do_sync(mddev_t *mddev)
5565 unsigned int currspeed = 0,
5567 sector_t max_sectors,j, io_sectors;
5568 unsigned long mark[SYNC_MARKS];
5569 sector_t mark_cnt[SYNC_MARKS];
5571 struct list_head *tmp;
5572 sector_t last_check;
5574 struct list_head *rtmp;
5578 /* just incase thread restarts... */
5579 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5581 if (mddev->ro) /* never try to sync a read-only array */
5584 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5585 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5586 desc = "data-check";
5587 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5588 desc = "requested-resync";
5591 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5596 /* we overload curr_resync somewhat here.
5597 * 0 == not engaged in resync at all
5598 * 2 == checking that there is no conflict with another sync
5599 * 1 == like 2, but have yielded to allow conflicting resync to
5601 * other == active in resync - this many blocks
5603 * Before starting a resync we must have set curr_resync to
5604 * 2, and then checked that every "conflicting" array has curr_resync
5605 * less than ours. When we find one that is the same or higher
5606 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5607 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5608 * This will mean we have to start checking from the beginning again.
5613 mddev->curr_resync = 2;
5616 if (kthread_should_stop()) {
5617 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5620 for_each_mddev(mddev2, tmp) {
5621 if (mddev2 == mddev)
5623 if (!mddev->parallel_resync
5624 && mddev2->curr_resync
5625 && match_mddev_units(mddev, mddev2)) {
5627 if (mddev < mddev2 && mddev->curr_resync == 2) {
5628 /* arbitrarily yield */
5629 mddev->curr_resync = 1;
5630 wake_up(&resync_wait);
5632 if (mddev > mddev2 && mddev->curr_resync == 1)
5633 /* no need to wait here, we can wait the next
5634 * time 'round when curr_resync == 2
5637 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5638 if (!kthread_should_stop() &&
5639 mddev2->curr_resync >= mddev->curr_resync) {
5640 printk(KERN_INFO "md: delaying %s of %s"
5641 " until %s has finished (they"
5642 " share one or more physical units)\n",
5643 desc, mdname(mddev), mdname(mddev2));
5646 finish_wait(&resync_wait, &wq);
5649 finish_wait(&resync_wait, &wq);
5652 } while (mddev->curr_resync < 2);
5655 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5656 /* resync follows the size requested by the personality,
5657 * which defaults to physical size, but can be virtual size
5659 max_sectors = mddev->resync_max_sectors;
5660 mddev->resync_mismatches = 0;
5661 /* we don't use the checkpoint if there's a bitmap */
5662 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5663 j = mddev->resync_min;
5664 else if (!mddev->bitmap)
5665 j = mddev->recovery_cp;
5667 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5668 max_sectors = mddev->size << 1;
5670 /* recovery follows the physical size of devices */
5671 max_sectors = mddev->size << 1;
5673 rdev_for_each(rdev, rtmp, mddev)
5674 if (rdev->raid_disk >= 0 &&
5675 !test_bit(Faulty, &rdev->flags) &&
5676 !test_bit(In_sync, &rdev->flags) &&
5677 rdev->recovery_offset < j)
5678 j = rdev->recovery_offset;
5681 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5682 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5683 " %d KB/sec/disk.\n", speed_min(mddev));
5684 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5685 "(but not more than %d KB/sec) for %s.\n",
5686 speed_max(mddev), desc);
5688 is_mddev_idle(mddev); /* this also initializes IO event counters */
5691 for (m = 0; m < SYNC_MARKS; m++) {
5693 mark_cnt[m] = io_sectors;
5696 mddev->resync_mark = mark[last_mark];
5697 mddev->resync_mark_cnt = mark_cnt[last_mark];
5700 * Tune reconstruction:
5702 window = 32*(PAGE_SIZE/512);
5703 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5704 window/2,(unsigned long long) max_sectors/2);
5706 atomic_set(&mddev->recovery_active, 0);
5711 "md: resuming %s of %s from checkpoint.\n",
5712 desc, mdname(mddev));
5713 mddev->curr_resync = j;
5716 while (j < max_sectors) {
5720 if (j >= mddev->resync_max) {
5721 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5722 wait_event(mddev->recovery_wait,
5723 mddev->resync_max > j
5724 || kthread_should_stop());
5726 if (kthread_should_stop())
5728 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5729 currspeed < speed_min(mddev));
5731 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5735 if (!skipped) { /* actual IO requested */
5736 io_sectors += sectors;
5737 atomic_add(sectors, &mddev->recovery_active);
5741 if (j>1) mddev->curr_resync = j;
5742 mddev->curr_mark_cnt = io_sectors;
5743 if (last_check == 0)
5744 /* this is the earliers that rebuilt will be
5745 * visible in /proc/mdstat
5747 md_new_event(mddev);
5749 if (last_check + window > io_sectors || j == max_sectors)
5752 last_check = io_sectors;
5754 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5758 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5760 int next = (last_mark+1) % SYNC_MARKS;
5762 mddev->resync_mark = mark[next];
5763 mddev->resync_mark_cnt = mark_cnt[next];
5764 mark[next] = jiffies;
5765 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5770 if (kthread_should_stop())
5775 * this loop exits only if either when we are slower than
5776 * the 'hard' speed limit, or the system was IO-idle for
5778 * the system might be non-idle CPU-wise, but we only care
5779 * about not overloading the IO subsystem. (things like an
5780 * e2fsck being done on the RAID array should execute fast)
5782 blk_unplug(mddev->queue);
5785 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5786 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5788 if (currspeed > speed_min(mddev)) {
5789 if ((currspeed > speed_max(mddev)) ||
5790 !is_mddev_idle(mddev)) {
5796 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5798 * this also signals 'finished resyncing' to md_stop
5801 blk_unplug(mddev->queue);
5803 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5805 /* tell personality that we are finished */
5806 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5808 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5809 mddev->curr_resync > 2) {
5810 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5811 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5812 if (mddev->curr_resync >= mddev->recovery_cp) {
5814 "md: checkpointing %s of %s.\n",
5815 desc, mdname(mddev));
5816 mddev->recovery_cp = mddev->curr_resync;
5819 mddev->recovery_cp = MaxSector;
5821 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5822 mddev->curr_resync = MaxSector;
5823 rdev_for_each(rdev, rtmp, mddev)
5824 if (rdev->raid_disk >= 0 &&
5825 !test_bit(Faulty, &rdev->flags) &&
5826 !test_bit(In_sync, &rdev->flags) &&
5827 rdev->recovery_offset < mddev->curr_resync)
5828 rdev->recovery_offset = mddev->curr_resync;
5831 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5834 mddev->curr_resync = 0;
5835 mddev->resync_min = 0;
5836 mddev->resync_max = MaxSector;
5837 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5838 wake_up(&resync_wait);
5839 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5840 md_wakeup_thread(mddev->thread);
5845 * got a signal, exit.
5848 "md: md_do_sync() got signal ... exiting\n");
5849 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5853 EXPORT_SYMBOL_GPL(md_do_sync);
5856 static int remove_and_add_spares(mddev_t *mddev)
5859 struct list_head *rtmp;
5862 rdev_for_each(rdev, rtmp, mddev)
5863 if (rdev->raid_disk >= 0 &&
5864 !test_bit(Blocked, &rdev->flags) &&
5865 (test_bit(Faulty, &rdev->flags) ||
5866 ! test_bit(In_sync, &rdev->flags)) &&
5867 atomic_read(&rdev->nr_pending)==0) {
5868 if (mddev->pers->hot_remove_disk(
5869 mddev, rdev->raid_disk)==0) {
5871 sprintf(nm,"rd%d", rdev->raid_disk);
5872 sysfs_remove_link(&mddev->kobj, nm);
5873 rdev->raid_disk = -1;
5877 if (mddev->degraded) {
5878 rdev_for_each(rdev, rtmp, mddev) {
5879 if (rdev->raid_disk >= 0 &&
5880 !test_bit(In_sync, &rdev->flags))
5882 if (rdev->raid_disk < 0
5883 && !test_bit(Faulty, &rdev->flags)) {
5884 rdev->recovery_offset = 0;
5885 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5887 sprintf(nm, "rd%d", rdev->raid_disk);
5888 if (sysfs_create_link(&mddev->kobj,
5891 "md: cannot register "
5895 md_new_event(mddev);
5904 * This routine is regularly called by all per-raid-array threads to
5905 * deal with generic issues like resync and super-block update.
5906 * Raid personalities that don't have a thread (linear/raid0) do not
5907 * need this as they never do any recovery or update the superblock.
5909 * It does not do any resync itself, but rather "forks" off other threads
5910 * to do that as needed.
5911 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5912 * "->recovery" and create a thread at ->sync_thread.
5913 * When the thread finishes it sets MD_RECOVERY_DONE
5914 * and wakeups up this thread which will reap the thread and finish up.
5915 * This thread also removes any faulty devices (with nr_pending == 0).
5917 * The overall approach is:
5918 * 1/ if the superblock needs updating, update it.
5919 * 2/ If a recovery thread is running, don't do anything else.
5920 * 3/ If recovery has finished, clean up, possibly marking spares active.
5921 * 4/ If there are any faulty devices, remove them.
5922 * 5/ If array is degraded, try to add spares devices
5923 * 6/ If array has spares or is not in-sync, start a resync thread.
5925 void md_check_recovery(mddev_t *mddev)
5928 struct list_head *rtmp;
5932 bitmap_daemon_work(mddev->bitmap);
5937 if (signal_pending(current)) {
5938 if (mddev->pers->sync_request && !mddev->external) {
5939 printk(KERN_INFO "md: %s in immediate safe mode\n",
5941 mddev->safemode = 2;
5943 flush_signals(current);
5947 (mddev->flags && !mddev->external) ||
5948 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5949 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5950 (mddev->external == 0 && mddev->safemode == 1) ||
5951 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5952 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5956 if (mddev_trylock(mddev)) {
5959 if (!mddev->external) {
5960 spin_lock_irq(&mddev->write_lock);
5961 if (mddev->safemode &&
5962 !atomic_read(&mddev->writes_pending) &&
5964 mddev->recovery_cp == MaxSector) {
5966 if (mddev->persistent)
5967 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5969 if (mddev->safemode == 1)
5970 mddev->safemode = 0;
5971 spin_unlock_irq(&mddev->write_lock);
5975 md_update_sb(mddev, 0);
5978 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5979 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5980 /* resync/recovery still happening */
5981 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5984 if (mddev->sync_thread) {
5985 /* resync has finished, collect result */
5986 md_unregister_thread(mddev->sync_thread);
5987 mddev->sync_thread = NULL;
5988 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5990 /* activate any spares */
5991 mddev->pers->spare_active(mddev);
5993 md_update_sb(mddev, 1);
5995 /* if array is no-longer degraded, then any saved_raid_disk
5996 * information must be scrapped
5998 if (!mddev->degraded)
5999 rdev_for_each(rdev, rtmp, mddev)
6000 rdev->saved_raid_disk = -1;
6002 mddev->recovery = 0;
6003 /* flag recovery needed just to double check */
6004 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6005 md_new_event(mddev);
6008 /* Clear some bits that don't mean anything, but
6011 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6012 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6013 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6015 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6017 /* no recovery is running.
6018 * remove any failed drives, then
6019 * add spares if possible.
6020 * Spare are also removed and re-added, to allow
6021 * the personality to fail the re-add.
6024 if (mddev->reshape_position != MaxSector) {
6025 if (mddev->pers->check_reshape(mddev) != 0)
6026 /* Cannot proceed */
6028 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6029 } else if ((spares = remove_and_add_spares(mddev))) {
6030 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6031 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6032 } else if (mddev->recovery_cp < MaxSector) {
6033 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6034 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6035 /* nothing to be done ... */
6038 if (mddev->pers->sync_request) {
6039 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6040 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6041 /* We are adding a device or devices to an array
6042 * which has the bitmap stored on all devices.
6043 * So make sure all bitmap pages get written
6045 bitmap_write_all(mddev->bitmap);
6047 mddev->sync_thread = md_register_thread(md_do_sync,
6050 if (!mddev->sync_thread) {
6051 printk(KERN_ERR "%s: could not start resync"
6054 /* leave the spares where they are, it shouldn't hurt */
6055 mddev->recovery = 0;
6057 md_wakeup_thread(mddev->sync_thread);
6058 md_new_event(mddev);
6061 mddev_unlock(mddev);
6065 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6067 sysfs_notify(&rdev->kobj, NULL, "state");
6068 wait_event_timeout(rdev->blocked_wait,
6069 !test_bit(Blocked, &rdev->flags),
6070 msecs_to_jiffies(5000));
6071 rdev_dec_pending(rdev, mddev);
6073 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6075 static int md_notify_reboot(struct notifier_block *this,
6076 unsigned long code, void *x)
6078 struct list_head *tmp;
6081 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6083 printk(KERN_INFO "md: stopping all md devices.\n");
6085 for_each_mddev(mddev, tmp)
6086 if (mddev_trylock(mddev)) {
6087 do_md_stop (mddev, 1);
6088 mddev_unlock(mddev);
6091 * certain more exotic SCSI devices are known to be
6092 * volatile wrt too early system reboots. While the
6093 * right place to handle this issue is the given
6094 * driver, we do want to have a safe RAID driver ...
6101 static struct notifier_block md_notifier = {
6102 .notifier_call = md_notify_reboot,
6104 .priority = INT_MAX, /* before any real devices */
6107 static void md_geninit(void)
6109 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6111 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6114 static int __init md_init(void)
6116 if (register_blkdev(MAJOR_NR, "md"))
6118 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6119 unregister_blkdev(MAJOR_NR, "md");
6122 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6123 md_probe, NULL, NULL);
6124 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6125 md_probe, NULL, NULL);
6127 register_reboot_notifier(&md_notifier);
6128 raid_table_header = register_sysctl_table(raid_root_table);
6138 * Searches all registered partitions for autorun RAID arrays
6142 static LIST_HEAD(all_detected_devices);
6143 struct detected_devices_node {
6144 struct list_head list;
6148 void md_autodetect_dev(dev_t dev)
6150 struct detected_devices_node *node_detected_dev;
6152 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6153 if (node_detected_dev) {
6154 node_detected_dev->dev = dev;
6155 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6157 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6158 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6163 static void autostart_arrays(int part)
6166 struct detected_devices_node *node_detected_dev;
6168 int i_scanned, i_passed;
6173 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6175 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6177 node_detected_dev = list_entry(all_detected_devices.next,
6178 struct detected_devices_node, list);
6179 list_del(&node_detected_dev->list);
6180 dev = node_detected_dev->dev;
6181 kfree(node_detected_dev);
6182 rdev = md_import_device(dev,0, 90);
6186 if (test_bit(Faulty, &rdev->flags)) {
6190 set_bit(AutoDetected, &rdev->flags);
6191 list_add(&rdev->same_set, &pending_raid_disks);
6195 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6196 i_scanned, i_passed);
6198 autorun_devices(part);
6201 #endif /* !MODULE */
6203 static __exit void md_exit(void)
6206 struct list_head *tmp;
6208 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6209 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6211 unregister_blkdev(MAJOR_NR,"md");
6212 unregister_blkdev(mdp_major, "mdp");
6213 unregister_reboot_notifier(&md_notifier);
6214 unregister_sysctl_table(raid_table_header);
6215 remove_proc_entry("mdstat", NULL);
6216 for_each_mddev(mddev, tmp) {
6217 struct gendisk *disk = mddev->gendisk;
6220 export_array(mddev);
6223 mddev->gendisk = NULL;
6228 subsys_initcall(md_init);
6229 module_exit(md_exit)
6231 static int get_ro(char *buffer, struct kernel_param *kp)
6233 return sprintf(buffer, "%d", start_readonly);
6235 static int set_ro(const char *val, struct kernel_param *kp)
6238 int num = simple_strtoul(val, &e, 10);
6239 if (*val && (*e == '\0' || *e == '\n')) {
6240 start_readonly = num;
6246 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6247 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6250 EXPORT_SYMBOL(register_md_personality);
6251 EXPORT_SYMBOL(unregister_md_personality);
6252 EXPORT_SYMBOL(md_error);
6253 EXPORT_SYMBOL(md_done_sync);
6254 EXPORT_SYMBOL(md_write_start);
6255 EXPORT_SYMBOL(md_write_end);
6256 EXPORT_SYMBOL(md_register_thread);
6257 EXPORT_SYMBOL(md_unregister_thread);
6258 EXPORT_SYMBOL(md_wakeup_thread);
6259 EXPORT_SYMBOL(md_check_recovery);
6260 MODULE_LICENSE("GPL");
6262 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);