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);
173 EXPORT_SYMBOL_GPL(md_new_event);
175 /* Alternate version that can be called from interrupts
176 * when calling sysfs_notify isn't needed.
178 static void md_new_event_inintr(mddev_t *mddev)
180 atomic_inc(&md_event_count);
181 wake_up(&md_event_waiters);
185 * Enables to iterate over all existing md arrays
186 * all_mddevs_lock protects this list.
188 static LIST_HEAD(all_mddevs);
189 static DEFINE_SPINLOCK(all_mddevs_lock);
193 * iterates through all used mddevs in the system.
194 * We take care to grab the all_mddevs_lock whenever navigating
195 * the list, and to always hold a refcount when unlocked.
196 * Any code which breaks out of this loop while own
197 * a reference to the current mddev and must mddev_put it.
199 #define for_each_mddev(mddev,tmp) \
201 for (({ spin_lock(&all_mddevs_lock); \
202 tmp = all_mddevs.next; \
204 ({ if (tmp != &all_mddevs) \
205 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
206 spin_unlock(&all_mddevs_lock); \
207 if (mddev) mddev_put(mddev); \
208 mddev = list_entry(tmp, mddev_t, all_mddevs); \
209 tmp != &all_mddevs;}); \
210 ({ spin_lock(&all_mddevs_lock); \
215 static int md_fail_request (struct request_queue *q, struct bio *bio)
221 static inline mddev_t *mddev_get(mddev_t *mddev)
223 atomic_inc(&mddev->active);
227 static void mddev_put(mddev_t *mddev)
229 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
231 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
232 list_del(&mddev->all_mddevs);
233 spin_unlock(&all_mddevs_lock);
234 blk_cleanup_queue(mddev->queue);
235 kobject_put(&mddev->kobj);
237 spin_unlock(&all_mddevs_lock);
240 static mddev_t * mddev_find(dev_t unit)
242 mddev_t *mddev, *new = NULL;
245 spin_lock(&all_mddevs_lock);
246 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
247 if (mddev->unit == unit) {
249 spin_unlock(&all_mddevs_lock);
255 list_add(&new->all_mddevs, &all_mddevs);
256 spin_unlock(&all_mddevs_lock);
259 spin_unlock(&all_mddevs_lock);
261 new = kzalloc(sizeof(*new), GFP_KERNEL);
266 if (MAJOR(unit) == MD_MAJOR)
267 new->md_minor = MINOR(unit);
269 new->md_minor = MINOR(unit) >> MdpMinorShift;
271 mutex_init(&new->reconfig_mutex);
272 INIT_LIST_HEAD(&new->disks);
273 INIT_LIST_HEAD(&new->all_mddevs);
274 init_timer(&new->safemode_timer);
275 atomic_set(&new->active, 1);
276 spin_lock_init(&new->write_lock);
277 init_waitqueue_head(&new->sb_wait);
278 init_waitqueue_head(&new->recovery_wait);
279 new->reshape_position = MaxSector;
281 new->resync_max = MaxSector;
282 new->level = LEVEL_NONE;
284 new->queue = blk_alloc_queue(GFP_KERNEL);
289 /* Can be unlocked because the queue is new: no concurrency */
290 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
292 blk_queue_make_request(new->queue, md_fail_request);
297 static inline int mddev_lock(mddev_t * mddev)
299 return mutex_lock_interruptible(&mddev->reconfig_mutex);
302 static inline int mddev_trylock(mddev_t * mddev)
304 return mutex_trylock(&mddev->reconfig_mutex);
307 static inline void mddev_unlock(mddev_t * mddev)
309 mutex_unlock(&mddev->reconfig_mutex);
311 md_wakeup_thread(mddev->thread);
314 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
317 struct list_head *tmp;
319 rdev_for_each(rdev, tmp, mddev) {
320 if (rdev->desc_nr == nr)
326 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
328 struct list_head *tmp;
331 rdev_for_each(rdev, tmp, mddev) {
332 if (rdev->bdev->bd_dev == dev)
338 static struct mdk_personality *find_pers(int level, char *clevel)
340 struct mdk_personality *pers;
341 list_for_each_entry(pers, &pers_list, list) {
342 if (level != LEVEL_NONE && pers->level == level)
344 if (strcmp(pers->name, clevel)==0)
350 /* return the offset of the super block in 512byte sectors */
351 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
353 sector_t num_sectors = bdev->bd_inode->i_size / 512;
354 return MD_NEW_SIZE_SECTORS(num_sectors);
357 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
359 sector_t num_sectors = rdev->sb_start;
362 num_sectors &= ~((sector_t)chunk_size/512 - 1);
366 static int alloc_disk_sb(mdk_rdev_t * rdev)
371 rdev->sb_page = alloc_page(GFP_KERNEL);
372 if (!rdev->sb_page) {
373 printk(KERN_ALERT "md: out of memory.\n");
380 static void free_disk_sb(mdk_rdev_t * rdev)
383 put_page(rdev->sb_page);
385 rdev->sb_page = NULL;
392 static void super_written(struct bio *bio, int error)
394 mdk_rdev_t *rdev = bio->bi_private;
395 mddev_t *mddev = rdev->mddev;
397 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
398 printk("md: super_written gets error=%d, uptodate=%d\n",
399 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
400 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
401 md_error(mddev, rdev);
404 if (atomic_dec_and_test(&mddev->pending_writes))
405 wake_up(&mddev->sb_wait);
409 static void super_written_barrier(struct bio *bio, int error)
411 struct bio *bio2 = bio->bi_private;
412 mdk_rdev_t *rdev = bio2->bi_private;
413 mddev_t *mddev = rdev->mddev;
415 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
416 error == -EOPNOTSUPP) {
418 /* barriers don't appear to be supported :-( */
419 set_bit(BarriersNotsupp, &rdev->flags);
420 mddev->barriers_work = 0;
421 spin_lock_irqsave(&mddev->write_lock, flags);
422 bio2->bi_next = mddev->biolist;
423 mddev->biolist = bio2;
424 spin_unlock_irqrestore(&mddev->write_lock, flags);
425 wake_up(&mddev->sb_wait);
429 bio->bi_private = rdev;
430 super_written(bio, error);
434 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
435 sector_t sector, int size, struct page *page)
437 /* write first size bytes of page to sector of rdev
438 * Increment mddev->pending_writes before returning
439 * and decrement it on completion, waking up sb_wait
440 * if zero is reached.
441 * If an error occurred, call md_error
443 * As we might need to resubmit the request if BIO_RW_BARRIER
444 * causes ENOTSUPP, we allocate a spare bio...
446 struct bio *bio = bio_alloc(GFP_NOIO, 1);
447 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
449 bio->bi_bdev = rdev->bdev;
450 bio->bi_sector = sector;
451 bio_add_page(bio, page, size, 0);
452 bio->bi_private = rdev;
453 bio->bi_end_io = super_written;
456 atomic_inc(&mddev->pending_writes);
457 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
459 rw |= (1<<BIO_RW_BARRIER);
460 rbio = bio_clone(bio, GFP_NOIO);
461 rbio->bi_private = bio;
462 rbio->bi_end_io = super_written_barrier;
463 submit_bio(rw, rbio);
468 void md_super_wait(mddev_t *mddev)
470 /* wait for all superblock writes that were scheduled to complete.
471 * if any had to be retried (due to BARRIER problems), retry them
475 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
476 if (atomic_read(&mddev->pending_writes)==0)
478 while (mddev->biolist) {
480 spin_lock_irq(&mddev->write_lock);
481 bio = mddev->biolist;
482 mddev->biolist = bio->bi_next ;
484 spin_unlock_irq(&mddev->write_lock);
485 submit_bio(bio->bi_rw, bio);
489 finish_wait(&mddev->sb_wait, &wq);
492 static void bi_complete(struct bio *bio, int error)
494 complete((struct completion*)bio->bi_private);
497 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
498 struct page *page, int rw)
500 struct bio *bio = bio_alloc(GFP_NOIO, 1);
501 struct completion event;
504 rw |= (1 << BIO_RW_SYNC);
507 bio->bi_sector = sector;
508 bio_add_page(bio, page, size, 0);
509 init_completion(&event);
510 bio->bi_private = &event;
511 bio->bi_end_io = bi_complete;
513 wait_for_completion(&event);
515 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
519 EXPORT_SYMBOL_GPL(sync_page_io);
521 static int read_disk_sb(mdk_rdev_t * rdev, int size)
523 char b[BDEVNAME_SIZE];
524 if (!rdev->sb_page) {
532 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
538 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
539 bdevname(rdev->bdev,b));
543 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
545 return sb1->set_uuid0 == sb2->set_uuid0 &&
546 sb1->set_uuid1 == sb2->set_uuid1 &&
547 sb1->set_uuid2 == sb2->set_uuid2 &&
548 sb1->set_uuid3 == sb2->set_uuid3;
551 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
554 mdp_super_t *tmp1, *tmp2;
556 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
557 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
559 if (!tmp1 || !tmp2) {
561 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
569 * nr_disks is not constant
574 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
582 static u32 md_csum_fold(u32 csum)
584 csum = (csum & 0xffff) + (csum >> 16);
585 return (csum & 0xffff) + (csum >> 16);
588 static unsigned int calc_sb_csum(mdp_super_t * sb)
591 u32 *sb32 = (u32*)sb;
593 unsigned int disk_csum, csum;
595 disk_csum = sb->sb_csum;
598 for (i = 0; i < MD_SB_BYTES/4 ; i++)
600 csum = (newcsum & 0xffffffff) + (newcsum>>32);
604 /* This used to use csum_partial, which was wrong for several
605 * reasons including that different results are returned on
606 * different architectures. It isn't critical that we get exactly
607 * the same return value as before (we always csum_fold before
608 * testing, and that removes any differences). However as we
609 * know that csum_partial always returned a 16bit value on
610 * alphas, do a fold to maximise conformity to previous behaviour.
612 sb->sb_csum = md_csum_fold(disk_csum);
614 sb->sb_csum = disk_csum;
621 * Handle superblock details.
622 * We want to be able to handle multiple superblock formats
623 * so we have a common interface to them all, and an array of
624 * different handlers.
625 * We rely on user-space to write the initial superblock, and support
626 * reading and updating of superblocks.
627 * Interface methods are:
628 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
629 * loads and validates a superblock on dev.
630 * if refdev != NULL, compare superblocks on both devices
632 * 0 - dev has a superblock that is compatible with refdev
633 * 1 - dev has a superblock that is compatible and newer than refdev
634 * so dev should be used as the refdev in future
635 * -EINVAL superblock incompatible or invalid
636 * -othererror e.g. -EIO
638 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
639 * Verify that dev is acceptable into mddev.
640 * The first time, mddev->raid_disks will be 0, and data from
641 * dev should be merged in. Subsequent calls check that dev
642 * is new enough. Return 0 or -EINVAL
644 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
645 * Update the superblock for rdev with data in mddev
646 * This does not write to disc.
652 struct module *owner;
653 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
655 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
656 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
657 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
658 sector_t num_sectors);
662 * load_super for 0.90.0
664 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
666 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
671 * Calculate the position of the superblock (512byte sectors),
672 * it's at the end of the disk.
674 * It also happens to be a multiple of 4Kb.
676 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
678 ret = read_disk_sb(rdev, MD_SB_BYTES);
683 bdevname(rdev->bdev, b);
684 sb = (mdp_super_t*)page_address(rdev->sb_page);
686 if (sb->md_magic != MD_SB_MAGIC) {
687 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
692 if (sb->major_version != 0 ||
693 sb->minor_version < 90 ||
694 sb->minor_version > 91) {
695 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
696 sb->major_version, sb->minor_version,
701 if (sb->raid_disks <= 0)
704 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
705 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
710 rdev->preferred_minor = sb->md_minor;
711 rdev->data_offset = 0;
712 rdev->sb_size = MD_SB_BYTES;
714 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
715 if (sb->level != 1 && sb->level != 4
716 && sb->level != 5 && sb->level != 6
717 && sb->level != 10) {
718 /* FIXME use a better test */
720 "md: bitmaps not supported for this level.\n");
725 if (sb->level == LEVEL_MULTIPATH)
728 rdev->desc_nr = sb->this_disk.number;
734 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
735 if (!uuid_equal(refsb, sb)) {
736 printk(KERN_WARNING "md: %s has different UUID to %s\n",
737 b, bdevname(refdev->bdev,b2));
740 if (!sb_equal(refsb, sb)) {
741 printk(KERN_WARNING "md: %s has same UUID"
742 " but different superblock to %s\n",
743 b, bdevname(refdev->bdev, b2));
747 ev2 = md_event(refsb);
753 rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
755 if (rdev->size < sb->size && sb->level > 1)
756 /* "this cannot possibly happen" ... */
764 * validate_super for 0.90.0
766 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
769 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
770 __u64 ev1 = md_event(sb);
772 rdev->raid_disk = -1;
773 clear_bit(Faulty, &rdev->flags);
774 clear_bit(In_sync, &rdev->flags);
775 clear_bit(WriteMostly, &rdev->flags);
776 clear_bit(BarriersNotsupp, &rdev->flags);
778 if (mddev->raid_disks == 0) {
779 mddev->major_version = 0;
780 mddev->minor_version = sb->minor_version;
781 mddev->patch_version = sb->patch_version;
783 mddev->chunk_size = sb->chunk_size;
784 mddev->ctime = sb->ctime;
785 mddev->utime = sb->utime;
786 mddev->level = sb->level;
787 mddev->clevel[0] = 0;
788 mddev->layout = sb->layout;
789 mddev->raid_disks = sb->raid_disks;
790 mddev->size = sb->size;
792 mddev->bitmap_offset = 0;
793 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
795 if (mddev->minor_version >= 91) {
796 mddev->reshape_position = sb->reshape_position;
797 mddev->delta_disks = sb->delta_disks;
798 mddev->new_level = sb->new_level;
799 mddev->new_layout = sb->new_layout;
800 mddev->new_chunk = sb->new_chunk;
802 mddev->reshape_position = MaxSector;
803 mddev->delta_disks = 0;
804 mddev->new_level = mddev->level;
805 mddev->new_layout = mddev->layout;
806 mddev->new_chunk = mddev->chunk_size;
809 if (sb->state & (1<<MD_SB_CLEAN))
810 mddev->recovery_cp = MaxSector;
812 if (sb->events_hi == sb->cp_events_hi &&
813 sb->events_lo == sb->cp_events_lo) {
814 mddev->recovery_cp = sb->recovery_cp;
816 mddev->recovery_cp = 0;
819 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
820 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
821 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
822 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
824 mddev->max_disks = MD_SB_DISKS;
826 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
827 mddev->bitmap_file == NULL)
828 mddev->bitmap_offset = mddev->default_bitmap_offset;
830 } else if (mddev->pers == NULL) {
831 /* Insist on good event counter while assembling */
833 if (ev1 < mddev->events)
835 } else if (mddev->bitmap) {
836 /* if adding to array with a bitmap, then we can accept an
837 * older device ... but not too old.
839 if (ev1 < mddev->bitmap->events_cleared)
842 if (ev1 < mddev->events)
843 /* just a hot-add of a new device, leave raid_disk at -1 */
847 if (mddev->level != LEVEL_MULTIPATH) {
848 desc = sb->disks + rdev->desc_nr;
850 if (desc->state & (1<<MD_DISK_FAULTY))
851 set_bit(Faulty, &rdev->flags);
852 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
853 desc->raid_disk < mddev->raid_disks */) {
854 set_bit(In_sync, &rdev->flags);
855 rdev->raid_disk = desc->raid_disk;
857 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
858 set_bit(WriteMostly, &rdev->flags);
859 } else /* MULTIPATH are always insync */
860 set_bit(In_sync, &rdev->flags);
865 * sync_super for 0.90.0
867 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
870 struct list_head *tmp;
872 int next_spare = mddev->raid_disks;
875 /* make rdev->sb match mddev data..
878 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
879 * 3/ any empty disks < next_spare become removed
881 * disks[0] gets initialised to REMOVED because
882 * we cannot be sure from other fields if it has
883 * been initialised or not.
886 int active=0, working=0,failed=0,spare=0,nr_disks=0;
888 rdev->sb_size = MD_SB_BYTES;
890 sb = (mdp_super_t*)page_address(rdev->sb_page);
892 memset(sb, 0, sizeof(*sb));
894 sb->md_magic = MD_SB_MAGIC;
895 sb->major_version = mddev->major_version;
896 sb->patch_version = mddev->patch_version;
897 sb->gvalid_words = 0; /* ignored */
898 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
899 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
900 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
901 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
903 sb->ctime = mddev->ctime;
904 sb->level = mddev->level;
905 sb->size = mddev->size;
906 sb->raid_disks = mddev->raid_disks;
907 sb->md_minor = mddev->md_minor;
908 sb->not_persistent = 0;
909 sb->utime = mddev->utime;
911 sb->events_hi = (mddev->events>>32);
912 sb->events_lo = (u32)mddev->events;
914 if (mddev->reshape_position == MaxSector)
915 sb->minor_version = 90;
917 sb->minor_version = 91;
918 sb->reshape_position = mddev->reshape_position;
919 sb->new_level = mddev->new_level;
920 sb->delta_disks = mddev->delta_disks;
921 sb->new_layout = mddev->new_layout;
922 sb->new_chunk = mddev->new_chunk;
924 mddev->minor_version = sb->minor_version;
927 sb->recovery_cp = mddev->recovery_cp;
928 sb->cp_events_hi = (mddev->events>>32);
929 sb->cp_events_lo = (u32)mddev->events;
930 if (mddev->recovery_cp == MaxSector)
931 sb->state = (1<< MD_SB_CLEAN);
935 sb->layout = mddev->layout;
936 sb->chunk_size = mddev->chunk_size;
938 if (mddev->bitmap && mddev->bitmap_file == NULL)
939 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
941 sb->disks[0].state = (1<<MD_DISK_REMOVED);
942 rdev_for_each(rdev2, tmp, mddev) {
945 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
946 && !test_bit(Faulty, &rdev2->flags))
947 desc_nr = rdev2->raid_disk;
949 desc_nr = next_spare++;
950 rdev2->desc_nr = desc_nr;
951 d = &sb->disks[rdev2->desc_nr];
953 d->number = rdev2->desc_nr;
954 d->major = MAJOR(rdev2->bdev->bd_dev);
955 d->minor = MINOR(rdev2->bdev->bd_dev);
956 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
957 && !test_bit(Faulty, &rdev2->flags))
958 d->raid_disk = rdev2->raid_disk;
960 d->raid_disk = rdev2->desc_nr; /* compatibility */
961 if (test_bit(Faulty, &rdev2->flags))
962 d->state = (1<<MD_DISK_FAULTY);
963 else if (test_bit(In_sync, &rdev2->flags)) {
964 d->state = (1<<MD_DISK_ACTIVE);
965 d->state |= (1<<MD_DISK_SYNC);
973 if (test_bit(WriteMostly, &rdev2->flags))
974 d->state |= (1<<MD_DISK_WRITEMOSTLY);
976 /* now set the "removed" and "faulty" bits on any missing devices */
977 for (i=0 ; i < mddev->raid_disks ; i++) {
978 mdp_disk_t *d = &sb->disks[i];
979 if (d->state == 0 && d->number == 0) {
982 d->state = (1<<MD_DISK_REMOVED);
983 d->state |= (1<<MD_DISK_FAULTY);
987 sb->nr_disks = nr_disks;
988 sb->active_disks = active;
989 sb->working_disks = working;
990 sb->failed_disks = failed;
991 sb->spare_disks = spare;
993 sb->this_disk = sb->disks[rdev->desc_nr];
994 sb->sb_csum = calc_sb_csum(sb);
998 * rdev_size_change for 0.90.0
1000 static unsigned long long
1001 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1003 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1004 return 0; /* component must fit device */
1005 if (rdev->mddev->bitmap_offset)
1006 return 0; /* can't move bitmap */
1007 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1008 if (!num_sectors || num_sectors > rdev->sb_start)
1009 num_sectors = rdev->sb_start;
1010 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1012 md_super_wait(rdev->mddev);
1013 return num_sectors / 2; /* kB for sysfs */
1018 * version 1 superblock
1021 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1025 unsigned long long newcsum;
1026 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1027 __le32 *isuper = (__le32*)sb;
1030 disk_csum = sb->sb_csum;
1033 for (i=0; size>=4; size -= 4 )
1034 newcsum += le32_to_cpu(*isuper++);
1037 newcsum += le16_to_cpu(*(__le16*) isuper);
1039 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1040 sb->sb_csum = disk_csum;
1041 return cpu_to_le32(csum);
1044 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1046 struct mdp_superblock_1 *sb;
1049 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1053 * Calculate the position of the superblock in 512byte sectors.
1054 * It is always aligned to a 4K boundary and
1055 * depeding on minor_version, it can be:
1056 * 0: At least 8K, but less than 12K, from end of device
1057 * 1: At start of device
1058 * 2: 4K from start of device.
1060 switch(minor_version) {
1062 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1064 sb_start &= ~(sector_t)(4*2-1);
1075 rdev->sb_start = sb_start;
1077 /* superblock is rarely larger than 1K, but it can be larger,
1078 * and it is safe to read 4k, so we do that
1080 ret = read_disk_sb(rdev, 4096);
1081 if (ret) return ret;
1084 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1086 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1087 sb->major_version != cpu_to_le32(1) ||
1088 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1089 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1090 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1093 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1094 printk("md: invalid superblock checksum on %s\n",
1095 bdevname(rdev->bdev,b));
1098 if (le64_to_cpu(sb->data_size) < 10) {
1099 printk("md: data_size too small on %s\n",
1100 bdevname(rdev->bdev,b));
1103 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1104 if (sb->level != cpu_to_le32(1) &&
1105 sb->level != cpu_to_le32(4) &&
1106 sb->level != cpu_to_le32(5) &&
1107 sb->level != cpu_to_le32(6) &&
1108 sb->level != cpu_to_le32(10)) {
1110 "md: bitmaps not supported for this level.\n");
1115 rdev->preferred_minor = 0xffff;
1116 rdev->data_offset = le64_to_cpu(sb->data_offset);
1117 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1119 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1120 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1121 if (rdev->sb_size & bmask)
1122 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1125 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1128 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1131 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1137 struct mdp_superblock_1 *refsb =
1138 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1140 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1141 sb->level != refsb->level ||
1142 sb->layout != refsb->layout ||
1143 sb->chunksize != refsb->chunksize) {
1144 printk(KERN_WARNING "md: %s has strangely different"
1145 " superblock to %s\n",
1146 bdevname(rdev->bdev,b),
1147 bdevname(refdev->bdev,b2));
1150 ev1 = le64_to_cpu(sb->events);
1151 ev2 = le64_to_cpu(refsb->events);
1159 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1161 rdev->size = rdev->sb_start / 2;
1162 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1164 rdev->size = le64_to_cpu(sb->data_size)/2;
1165 if (le32_to_cpu(sb->chunksize))
1166 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1168 if (le64_to_cpu(sb->size) > rdev->size*2)
1173 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1175 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1176 __u64 ev1 = le64_to_cpu(sb->events);
1178 rdev->raid_disk = -1;
1179 clear_bit(Faulty, &rdev->flags);
1180 clear_bit(In_sync, &rdev->flags);
1181 clear_bit(WriteMostly, &rdev->flags);
1182 clear_bit(BarriersNotsupp, &rdev->flags);
1184 if (mddev->raid_disks == 0) {
1185 mddev->major_version = 1;
1186 mddev->patch_version = 0;
1187 mddev->external = 0;
1188 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1189 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1190 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1191 mddev->level = le32_to_cpu(sb->level);
1192 mddev->clevel[0] = 0;
1193 mddev->layout = le32_to_cpu(sb->layout);
1194 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1195 mddev->size = le64_to_cpu(sb->size)/2;
1196 mddev->events = ev1;
1197 mddev->bitmap_offset = 0;
1198 mddev->default_bitmap_offset = 1024 >> 9;
1200 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1201 memcpy(mddev->uuid, sb->set_uuid, 16);
1203 mddev->max_disks = (4096-256)/2;
1205 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1206 mddev->bitmap_file == NULL )
1207 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1209 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1210 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1211 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1212 mddev->new_level = le32_to_cpu(sb->new_level);
1213 mddev->new_layout = le32_to_cpu(sb->new_layout);
1214 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1216 mddev->reshape_position = MaxSector;
1217 mddev->delta_disks = 0;
1218 mddev->new_level = mddev->level;
1219 mddev->new_layout = mddev->layout;
1220 mddev->new_chunk = mddev->chunk_size;
1223 } else if (mddev->pers == NULL) {
1224 /* Insist of good event counter while assembling */
1226 if (ev1 < mddev->events)
1228 } else if (mddev->bitmap) {
1229 /* If adding to array with a bitmap, then we can accept an
1230 * older device, but not too old.
1232 if (ev1 < mddev->bitmap->events_cleared)
1235 if (ev1 < mddev->events)
1236 /* just a hot-add of a new device, leave raid_disk at -1 */
1239 if (mddev->level != LEVEL_MULTIPATH) {
1241 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1243 case 0xffff: /* spare */
1245 case 0xfffe: /* faulty */
1246 set_bit(Faulty, &rdev->flags);
1249 if ((le32_to_cpu(sb->feature_map) &
1250 MD_FEATURE_RECOVERY_OFFSET))
1251 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1253 set_bit(In_sync, &rdev->flags);
1254 rdev->raid_disk = role;
1257 if (sb->devflags & WriteMostly1)
1258 set_bit(WriteMostly, &rdev->flags);
1259 } else /* MULTIPATH are always insync */
1260 set_bit(In_sync, &rdev->flags);
1265 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1267 struct mdp_superblock_1 *sb;
1268 struct list_head *tmp;
1271 /* make rdev->sb match mddev and rdev data. */
1273 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1275 sb->feature_map = 0;
1277 sb->recovery_offset = cpu_to_le64(0);
1278 memset(sb->pad1, 0, sizeof(sb->pad1));
1279 memset(sb->pad2, 0, sizeof(sb->pad2));
1280 memset(sb->pad3, 0, sizeof(sb->pad3));
1282 sb->utime = cpu_to_le64((__u64)mddev->utime);
1283 sb->events = cpu_to_le64(mddev->events);
1285 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1287 sb->resync_offset = cpu_to_le64(0);
1289 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1291 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1292 sb->size = cpu_to_le64(mddev->size<<1);
1294 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1295 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1296 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1299 if (rdev->raid_disk >= 0 &&
1300 !test_bit(In_sync, &rdev->flags) &&
1301 rdev->recovery_offset > 0) {
1302 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1303 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1306 if (mddev->reshape_position != MaxSector) {
1307 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1308 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1309 sb->new_layout = cpu_to_le32(mddev->new_layout);
1310 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1311 sb->new_level = cpu_to_le32(mddev->new_level);
1312 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1316 rdev_for_each(rdev2, tmp, mddev)
1317 if (rdev2->desc_nr+1 > max_dev)
1318 max_dev = rdev2->desc_nr+1;
1320 if (max_dev > le32_to_cpu(sb->max_dev))
1321 sb->max_dev = cpu_to_le32(max_dev);
1322 for (i=0; i<max_dev;i++)
1323 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1325 rdev_for_each(rdev2, tmp, mddev) {
1327 if (test_bit(Faulty, &rdev2->flags))
1328 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1329 else if (test_bit(In_sync, &rdev2->flags))
1330 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1331 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1332 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1334 sb->dev_roles[i] = cpu_to_le16(0xffff);
1337 sb->sb_csum = calc_sb_1_csum(sb);
1340 static unsigned long long
1341 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1343 struct mdp_superblock_1 *sb;
1344 sector_t max_sectors;
1345 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1346 return 0; /* component must fit device */
1347 if (rdev->sb_start < rdev->data_offset) {
1348 /* minor versions 1 and 2; superblock before data */
1349 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1350 max_sectors -= rdev->data_offset;
1351 if (!num_sectors || num_sectors > max_sectors)
1352 num_sectors = max_sectors;
1353 } else if (rdev->mddev->bitmap_offset) {
1354 /* minor version 0 with bitmap we can't move */
1357 /* minor version 0; superblock after data */
1359 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1360 sb_start &= ~(sector_t)(4*2 - 1);
1361 max_sectors = rdev->size * 2 + sb_start - rdev->sb_start;
1362 if (!num_sectors || num_sectors > max_sectors)
1363 num_sectors = max_sectors;
1364 rdev->sb_start = sb_start;
1366 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1367 sb->data_size = cpu_to_le64(num_sectors);
1368 sb->super_offset = rdev->sb_start;
1369 sb->sb_csum = calc_sb_1_csum(sb);
1370 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1372 md_super_wait(rdev->mddev);
1373 return num_sectors / 2; /* kB for sysfs */
1376 static struct super_type super_types[] = {
1379 .owner = THIS_MODULE,
1380 .load_super = super_90_load,
1381 .validate_super = super_90_validate,
1382 .sync_super = super_90_sync,
1383 .rdev_size_change = super_90_rdev_size_change,
1387 .owner = THIS_MODULE,
1388 .load_super = super_1_load,
1389 .validate_super = super_1_validate,
1390 .sync_super = super_1_sync,
1391 .rdev_size_change = super_1_rdev_size_change,
1395 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1397 struct list_head *tmp, *tmp2;
1398 mdk_rdev_t *rdev, *rdev2;
1400 rdev_for_each(rdev, tmp, mddev1)
1401 rdev_for_each(rdev2, tmp2, mddev2)
1402 if (rdev->bdev->bd_contains ==
1403 rdev2->bdev->bd_contains)
1409 static LIST_HEAD(pending_raid_disks);
1411 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1413 char b[BDEVNAME_SIZE];
1423 /* prevent duplicates */
1424 if (find_rdev(mddev, rdev->bdev->bd_dev))
1427 /* make sure rdev->size exceeds mddev->size */
1428 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1430 /* Cannot change size, so fail
1431 * If mddev->level <= 0, then we don't care
1432 * about aligning sizes (e.g. linear)
1434 if (mddev->level > 0)
1437 mddev->size = rdev->size;
1440 /* Verify rdev->desc_nr is unique.
1441 * If it is -1, assign a free number, else
1442 * check number is not in use
1444 if (rdev->desc_nr < 0) {
1446 if (mddev->pers) choice = mddev->raid_disks;
1447 while (find_rdev_nr(mddev, choice))
1449 rdev->desc_nr = choice;
1451 if (find_rdev_nr(mddev, rdev->desc_nr))
1454 bdevname(rdev->bdev,b);
1455 while ( (s=strchr(b, '/')) != NULL)
1458 rdev->mddev = mddev;
1459 printk(KERN_INFO "md: bind<%s>\n", b);
1461 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1464 if (rdev->bdev->bd_part)
1465 ko = &rdev->bdev->bd_part->dev.kobj;
1467 ko = &rdev->bdev->bd_disk->dev.kobj;
1468 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1469 kobject_del(&rdev->kobj);
1472 list_add(&rdev->same_set, &mddev->disks);
1473 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1477 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1482 static void md_delayed_delete(struct work_struct *ws)
1484 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1485 kobject_del(&rdev->kobj);
1486 kobject_put(&rdev->kobj);
1489 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1491 char b[BDEVNAME_SIZE];
1496 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1497 list_del_init(&rdev->same_set);
1498 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1500 sysfs_remove_link(&rdev->kobj, "block");
1502 /* We need to delay this, otherwise we can deadlock when
1503 * writing to 'remove' to "dev/state"
1505 INIT_WORK(&rdev->del_work, md_delayed_delete);
1506 kobject_get(&rdev->kobj);
1507 schedule_work(&rdev->del_work);
1511 * prevent the device from being mounted, repartitioned or
1512 * otherwise reused by a RAID array (or any other kernel
1513 * subsystem), by bd_claiming the device.
1515 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1518 struct block_device *bdev;
1519 char b[BDEVNAME_SIZE];
1521 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1523 printk(KERN_ERR "md: could not open %s.\n",
1524 __bdevname(dev, b));
1525 return PTR_ERR(bdev);
1527 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1529 printk(KERN_ERR "md: could not bd_claim %s.\n",
1535 set_bit(AllReserved, &rdev->flags);
1540 static void unlock_rdev(mdk_rdev_t *rdev)
1542 struct block_device *bdev = rdev->bdev;
1550 void md_autodetect_dev(dev_t dev);
1552 static void export_rdev(mdk_rdev_t * rdev)
1554 char b[BDEVNAME_SIZE];
1555 printk(KERN_INFO "md: export_rdev(%s)\n",
1556 bdevname(rdev->bdev,b));
1560 list_del_init(&rdev->same_set);
1562 if (test_bit(AutoDetected, &rdev->flags))
1563 md_autodetect_dev(rdev->bdev->bd_dev);
1566 kobject_put(&rdev->kobj);
1569 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1571 unbind_rdev_from_array(rdev);
1575 static void export_array(mddev_t *mddev)
1577 struct list_head *tmp;
1580 rdev_for_each(rdev, tmp, mddev) {
1585 kick_rdev_from_array(rdev);
1587 if (!list_empty(&mddev->disks))
1589 mddev->raid_disks = 0;
1590 mddev->major_version = 0;
1593 static void print_desc(mdp_disk_t *desc)
1595 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1596 desc->major,desc->minor,desc->raid_disk,desc->state);
1599 static void print_sb(mdp_super_t *sb)
1604 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1605 sb->major_version, sb->minor_version, sb->patch_version,
1606 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1608 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1609 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1610 sb->md_minor, sb->layout, sb->chunk_size);
1611 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1612 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1613 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1614 sb->failed_disks, sb->spare_disks,
1615 sb->sb_csum, (unsigned long)sb->events_lo);
1618 for (i = 0; i < MD_SB_DISKS; i++) {
1621 desc = sb->disks + i;
1622 if (desc->number || desc->major || desc->minor ||
1623 desc->raid_disk || (desc->state && (desc->state != 4))) {
1624 printk(" D %2d: ", i);
1628 printk(KERN_INFO "md: THIS: ");
1629 print_desc(&sb->this_disk);
1633 static void print_rdev(mdk_rdev_t *rdev)
1635 char b[BDEVNAME_SIZE];
1636 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1637 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1638 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1640 if (rdev->sb_loaded) {
1641 printk(KERN_INFO "md: rdev superblock:\n");
1642 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1644 printk(KERN_INFO "md: no rdev superblock!\n");
1647 static void md_print_devices(void)
1649 struct list_head *tmp, *tmp2;
1652 char b[BDEVNAME_SIZE];
1655 printk("md: **********************************\n");
1656 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1657 printk("md: **********************************\n");
1658 for_each_mddev(mddev, tmp) {
1661 bitmap_print_sb(mddev->bitmap);
1663 printk("%s: ", mdname(mddev));
1664 rdev_for_each(rdev, tmp2, mddev)
1665 printk("<%s>", bdevname(rdev->bdev,b));
1668 rdev_for_each(rdev, tmp2, mddev)
1671 printk("md: **********************************\n");
1676 static void sync_sbs(mddev_t * mddev, int nospares)
1678 /* Update each superblock (in-memory image), but
1679 * if we are allowed to, skip spares which already
1680 * have the right event counter, or have one earlier
1681 * (which would mean they aren't being marked as dirty
1682 * with the rest of the array)
1685 struct list_head *tmp;
1687 rdev_for_each(rdev, tmp, mddev) {
1688 if (rdev->sb_events == mddev->events ||
1690 rdev->raid_disk < 0 &&
1691 (rdev->sb_events&1)==0 &&
1692 rdev->sb_events+1 == mddev->events)) {
1693 /* Don't update this superblock */
1694 rdev->sb_loaded = 2;
1696 super_types[mddev->major_version].
1697 sync_super(mddev, rdev);
1698 rdev->sb_loaded = 1;
1703 static void md_update_sb(mddev_t * mddev, int force_change)
1705 struct list_head *tmp;
1710 if (mddev->external)
1713 spin_lock_irq(&mddev->write_lock);
1715 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1716 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1718 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1719 /* just a clean<-> dirty transition, possibly leave spares alone,
1720 * though if events isn't the right even/odd, we will have to do
1726 if (mddev->degraded)
1727 /* If the array is degraded, then skipping spares is both
1728 * dangerous and fairly pointless.
1729 * Dangerous because a device that was removed from the array
1730 * might have a event_count that still looks up-to-date,
1731 * so it can be re-added without a resync.
1732 * Pointless because if there are any spares to skip,
1733 * then a recovery will happen and soon that array won't
1734 * be degraded any more and the spare can go back to sleep then.
1738 sync_req = mddev->in_sync;
1739 mddev->utime = get_seconds();
1741 /* If this is just a dirty<->clean transition, and the array is clean
1742 * and 'events' is odd, we can roll back to the previous clean state */
1744 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1745 && (mddev->events & 1)
1746 && mddev->events != 1)
1749 /* otherwise we have to go forward and ... */
1751 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1752 /* .. if the array isn't clean, insist on an odd 'events' */
1753 if ((mddev->events&1)==0) {
1758 /* otherwise insist on an even 'events' (for clean states) */
1759 if ((mddev->events&1)) {
1766 if (!mddev->events) {
1768 * oops, this 64-bit counter should never wrap.
1769 * Either we are in around ~1 trillion A.C., assuming
1770 * 1 reboot per second, or we have a bug:
1777 * do not write anything to disk if using
1778 * nonpersistent superblocks
1780 if (!mddev->persistent) {
1781 if (!mddev->external)
1782 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1784 spin_unlock_irq(&mddev->write_lock);
1785 wake_up(&mddev->sb_wait);
1788 sync_sbs(mddev, nospares);
1789 spin_unlock_irq(&mddev->write_lock);
1792 "md: updating %s RAID superblock on device (in sync %d)\n",
1793 mdname(mddev),mddev->in_sync);
1795 bitmap_update_sb(mddev->bitmap);
1796 rdev_for_each(rdev, tmp, mddev) {
1797 char b[BDEVNAME_SIZE];
1798 dprintk(KERN_INFO "md: ");
1799 if (rdev->sb_loaded != 1)
1800 continue; /* no noise on spare devices */
1801 if (test_bit(Faulty, &rdev->flags))
1802 dprintk("(skipping faulty ");
1804 dprintk("%s ", bdevname(rdev->bdev,b));
1805 if (!test_bit(Faulty, &rdev->flags)) {
1806 md_super_write(mddev,rdev,
1807 rdev->sb_start, rdev->sb_size,
1809 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1810 bdevname(rdev->bdev,b),
1811 (unsigned long long)rdev->sb_start);
1812 rdev->sb_events = mddev->events;
1816 if (mddev->level == LEVEL_MULTIPATH)
1817 /* only need to write one superblock... */
1820 md_super_wait(mddev);
1821 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1823 spin_lock_irq(&mddev->write_lock);
1824 if (mddev->in_sync != sync_req ||
1825 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1826 /* have to write it out again */
1827 spin_unlock_irq(&mddev->write_lock);
1830 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1831 spin_unlock_irq(&mddev->write_lock);
1832 wake_up(&mddev->sb_wait);
1836 /* words written to sysfs files may, or may not, be \n terminated.
1837 * We want to accept with case. For this we use cmd_match.
1839 static int cmd_match(const char *cmd, const char *str)
1841 /* See if cmd, written into a sysfs file, matches
1842 * str. They must either be the same, or cmd can
1843 * have a trailing newline
1845 while (*cmd && *str && *cmd == *str) {
1856 struct rdev_sysfs_entry {
1857 struct attribute attr;
1858 ssize_t (*show)(mdk_rdev_t *, char *);
1859 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1863 state_show(mdk_rdev_t *rdev, char *page)
1868 if (test_bit(Faulty, &rdev->flags)) {
1869 len+= sprintf(page+len, "%sfaulty",sep);
1872 if (test_bit(In_sync, &rdev->flags)) {
1873 len += sprintf(page+len, "%sin_sync",sep);
1876 if (test_bit(WriteMostly, &rdev->flags)) {
1877 len += sprintf(page+len, "%swrite_mostly",sep);
1880 if (test_bit(Blocked, &rdev->flags)) {
1881 len += sprintf(page+len, "%sblocked", sep);
1884 if (!test_bit(Faulty, &rdev->flags) &&
1885 !test_bit(In_sync, &rdev->flags)) {
1886 len += sprintf(page+len, "%sspare", sep);
1889 return len+sprintf(page+len, "\n");
1893 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1896 * faulty - simulates and error
1897 * remove - disconnects the device
1898 * writemostly - sets write_mostly
1899 * -writemostly - clears write_mostly
1900 * blocked - sets the Blocked flag
1901 * -blocked - clears the Blocked flag
1904 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1905 md_error(rdev->mddev, rdev);
1907 } else if (cmd_match(buf, "remove")) {
1908 if (rdev->raid_disk >= 0)
1911 mddev_t *mddev = rdev->mddev;
1912 kick_rdev_from_array(rdev);
1914 md_update_sb(mddev, 1);
1915 md_new_event(mddev);
1918 } else if (cmd_match(buf, "writemostly")) {
1919 set_bit(WriteMostly, &rdev->flags);
1921 } else if (cmd_match(buf, "-writemostly")) {
1922 clear_bit(WriteMostly, &rdev->flags);
1924 } else if (cmd_match(buf, "blocked")) {
1925 set_bit(Blocked, &rdev->flags);
1927 } else if (cmd_match(buf, "-blocked")) {
1928 clear_bit(Blocked, &rdev->flags);
1929 wake_up(&rdev->blocked_wait);
1930 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1931 md_wakeup_thread(rdev->mddev->thread);
1936 sysfs_notify(&rdev->kobj, NULL, "state");
1937 return err ? err : len;
1939 static struct rdev_sysfs_entry rdev_state =
1940 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1943 errors_show(mdk_rdev_t *rdev, char *page)
1945 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1949 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1952 unsigned long n = simple_strtoul(buf, &e, 10);
1953 if (*buf && (*e == 0 || *e == '\n')) {
1954 atomic_set(&rdev->corrected_errors, n);
1959 static struct rdev_sysfs_entry rdev_errors =
1960 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1963 slot_show(mdk_rdev_t *rdev, char *page)
1965 if (rdev->raid_disk < 0)
1966 return sprintf(page, "none\n");
1968 return sprintf(page, "%d\n", rdev->raid_disk);
1972 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1977 int slot = simple_strtoul(buf, &e, 10);
1978 if (strncmp(buf, "none", 4)==0)
1980 else if (e==buf || (*e && *e!= '\n'))
1982 if (rdev->mddev->pers && slot == -1) {
1983 /* Setting 'slot' on an active array requires also
1984 * updating the 'rd%d' link, and communicating
1985 * with the personality with ->hot_*_disk.
1986 * For now we only support removing
1987 * failed/spare devices. This normally happens automatically,
1988 * but not when the metadata is externally managed.
1990 if (rdev->raid_disk == -1)
1992 /* personality does all needed checks */
1993 if (rdev->mddev->pers->hot_add_disk == NULL)
1995 err = rdev->mddev->pers->
1996 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1999 sprintf(nm, "rd%d", rdev->raid_disk);
2000 sysfs_remove_link(&rdev->mddev->kobj, nm);
2001 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2002 md_wakeup_thread(rdev->mddev->thread);
2003 } else if (rdev->mddev->pers) {
2005 struct list_head *tmp;
2006 /* Activating a spare .. or possibly reactivating
2007 * if we every get bitmaps working here.
2010 if (rdev->raid_disk != -1)
2013 if (rdev->mddev->pers->hot_add_disk == NULL)
2016 rdev_for_each(rdev2, tmp, rdev->mddev)
2017 if (rdev2->raid_disk == slot)
2020 rdev->raid_disk = slot;
2021 if (test_bit(In_sync, &rdev->flags))
2022 rdev->saved_raid_disk = slot;
2024 rdev->saved_raid_disk = -1;
2025 err = rdev->mddev->pers->
2026 hot_add_disk(rdev->mddev, rdev);
2028 rdev->raid_disk = -1;
2031 sysfs_notify(&rdev->kobj, NULL, "state");
2032 sprintf(nm, "rd%d", rdev->raid_disk);
2033 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2035 "md: cannot register "
2037 nm, mdname(rdev->mddev));
2039 /* don't wakeup anyone, leave that to userspace. */
2041 if (slot >= rdev->mddev->raid_disks)
2043 rdev->raid_disk = slot;
2044 /* assume it is working */
2045 clear_bit(Faulty, &rdev->flags);
2046 clear_bit(WriteMostly, &rdev->flags);
2047 set_bit(In_sync, &rdev->flags);
2048 sysfs_notify(&rdev->kobj, NULL, "state");
2054 static struct rdev_sysfs_entry rdev_slot =
2055 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2058 offset_show(mdk_rdev_t *rdev, char *page)
2060 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2064 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2067 unsigned long long offset = simple_strtoull(buf, &e, 10);
2068 if (e==buf || (*e && *e != '\n'))
2070 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2072 if (rdev->size && rdev->mddev->external)
2073 /* Must set offset before size, so overlap checks
2076 rdev->data_offset = offset;
2080 static struct rdev_sysfs_entry rdev_offset =
2081 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2084 rdev_size_show(mdk_rdev_t *rdev, char *page)
2086 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2089 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2091 /* check if two start/length pairs overlap */
2100 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2102 unsigned long long size;
2103 unsigned long long oldsize = rdev->size;
2104 mddev_t *my_mddev = rdev->mddev;
2106 if (strict_strtoull(buf, 10, &size) < 0)
2108 if (size < my_mddev->size)
2110 if (my_mddev->pers && rdev->raid_disk >= 0) {
2111 if (my_mddev->persistent) {
2112 size = super_types[my_mddev->major_version].
2113 rdev_size_change(rdev, size * 2);
2117 size = (rdev->bdev->bd_inode->i_size >> 10);
2118 size -= rdev->data_offset/2;
2120 if (size < my_mddev->size)
2121 return -EINVAL; /* component must fit device */
2125 if (size > oldsize && my_mddev->external) {
2126 /* need to check that all other rdevs with the same ->bdev
2127 * do not overlap. We need to unlock the mddev to avoid
2128 * a deadlock. We have already changed rdev->size, and if
2129 * we have to change it back, we will have the lock again.
2133 struct list_head *tmp, *tmp2;
2135 mddev_unlock(my_mddev);
2136 for_each_mddev(mddev, tmp) {
2140 rdev_for_each(rdev2, tmp2, mddev)
2141 if (test_bit(AllReserved, &rdev2->flags) ||
2142 (rdev->bdev == rdev2->bdev &&
2144 overlaps(rdev->data_offset, rdev->size * 2,
2146 rdev2->size * 2))) {
2150 mddev_unlock(mddev);
2156 mddev_lock(my_mddev);
2158 /* Someone else could have slipped in a size
2159 * change here, but doing so is just silly.
2160 * We put oldsize back because we *know* it is
2161 * safe, and trust userspace not to race with
2164 rdev->size = oldsize;
2171 static struct rdev_sysfs_entry rdev_size =
2172 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2174 static struct attribute *rdev_default_attrs[] = {
2183 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2185 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2186 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2187 mddev_t *mddev = rdev->mddev;
2193 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2195 if (rdev->mddev == NULL)
2198 rv = entry->show(rdev, page);
2199 mddev_unlock(mddev);
2205 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2206 const char *page, size_t length)
2208 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2209 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2211 mddev_t *mddev = rdev->mddev;
2215 if (!capable(CAP_SYS_ADMIN))
2217 rv = mddev ? mddev_lock(mddev): -EBUSY;
2219 if (rdev->mddev == NULL)
2222 rv = entry->store(rdev, page, length);
2223 mddev_unlock(mddev);
2228 static void rdev_free(struct kobject *ko)
2230 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2233 static struct sysfs_ops rdev_sysfs_ops = {
2234 .show = rdev_attr_show,
2235 .store = rdev_attr_store,
2237 static struct kobj_type rdev_ktype = {
2238 .release = rdev_free,
2239 .sysfs_ops = &rdev_sysfs_ops,
2240 .default_attrs = rdev_default_attrs,
2244 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2246 * mark the device faulty if:
2248 * - the device is nonexistent (zero size)
2249 * - the device has no valid superblock
2251 * a faulty rdev _never_ has rdev->sb set.
2253 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2255 char b[BDEVNAME_SIZE];
2260 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2262 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2263 return ERR_PTR(-ENOMEM);
2266 if ((err = alloc_disk_sb(rdev)))
2269 err = lock_rdev(rdev, newdev, super_format == -2);
2273 kobject_init(&rdev->kobj, &rdev_ktype);
2276 rdev->saved_raid_disk = -1;
2277 rdev->raid_disk = -1;
2279 rdev->data_offset = 0;
2280 rdev->sb_events = 0;
2281 atomic_set(&rdev->nr_pending, 0);
2282 atomic_set(&rdev->read_errors, 0);
2283 atomic_set(&rdev->corrected_errors, 0);
2285 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2288 "md: %s has zero or unknown size, marking faulty!\n",
2289 bdevname(rdev->bdev,b));
2294 if (super_format >= 0) {
2295 err = super_types[super_format].
2296 load_super(rdev, NULL, super_minor);
2297 if (err == -EINVAL) {
2299 "md: %s does not have a valid v%d.%d "
2300 "superblock, not importing!\n",
2301 bdevname(rdev->bdev,b),
2302 super_format, super_minor);
2307 "md: could not read %s's sb, not importing!\n",
2308 bdevname(rdev->bdev,b));
2313 INIT_LIST_HEAD(&rdev->same_set);
2314 init_waitqueue_head(&rdev->blocked_wait);
2319 if (rdev->sb_page) {
2325 return ERR_PTR(err);
2329 * Check a full RAID array for plausibility
2333 static void analyze_sbs(mddev_t * mddev)
2336 struct list_head *tmp;
2337 mdk_rdev_t *rdev, *freshest;
2338 char b[BDEVNAME_SIZE];
2341 rdev_for_each(rdev, tmp, mddev)
2342 switch (super_types[mddev->major_version].
2343 load_super(rdev, freshest, mddev->minor_version)) {
2351 "md: fatal superblock inconsistency in %s"
2352 " -- removing from array\n",
2353 bdevname(rdev->bdev,b));
2354 kick_rdev_from_array(rdev);
2358 super_types[mddev->major_version].
2359 validate_super(mddev, freshest);
2362 rdev_for_each(rdev, tmp, mddev) {
2363 if (rdev != freshest)
2364 if (super_types[mddev->major_version].
2365 validate_super(mddev, rdev)) {
2366 printk(KERN_WARNING "md: kicking non-fresh %s"
2368 bdevname(rdev->bdev,b));
2369 kick_rdev_from_array(rdev);
2372 if (mddev->level == LEVEL_MULTIPATH) {
2373 rdev->desc_nr = i++;
2374 rdev->raid_disk = rdev->desc_nr;
2375 set_bit(In_sync, &rdev->flags);
2376 } else if (rdev->raid_disk >= mddev->raid_disks) {
2377 rdev->raid_disk = -1;
2378 clear_bit(In_sync, &rdev->flags);
2384 if (mddev->recovery_cp != MaxSector &&
2386 printk(KERN_ERR "md: %s: raid array is not clean"
2387 " -- starting background reconstruction\n",
2393 safe_delay_show(mddev_t *mddev, char *page)
2395 int msec = (mddev->safemode_delay*1000)/HZ;
2396 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2399 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2407 /* remove a period, and count digits after it */
2408 if (len >= sizeof(buf))
2410 strlcpy(buf, cbuf, len);
2412 for (i=0; i<len; i++) {
2414 if (isdigit(buf[i])) {
2419 } else if (buf[i] == '.') {
2424 msec = simple_strtoul(buf, &e, 10);
2425 if (e == buf || (*e && *e != '\n'))
2427 msec = (msec * 1000) / scale;
2429 mddev->safemode_delay = 0;
2431 mddev->safemode_delay = (msec*HZ)/1000;
2432 if (mddev->safemode_delay == 0)
2433 mddev->safemode_delay = 1;
2437 static struct md_sysfs_entry md_safe_delay =
2438 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2441 level_show(mddev_t *mddev, char *page)
2443 struct mdk_personality *p = mddev->pers;
2445 return sprintf(page, "%s\n", p->name);
2446 else if (mddev->clevel[0])
2447 return sprintf(page, "%s\n", mddev->clevel);
2448 else if (mddev->level != LEVEL_NONE)
2449 return sprintf(page, "%d\n", mddev->level);
2455 level_store(mddev_t *mddev, const char *buf, size_t len)
2462 if (len >= sizeof(mddev->clevel))
2464 strncpy(mddev->clevel, buf, len);
2465 if (mddev->clevel[len-1] == '\n')
2467 mddev->clevel[len] = 0;
2468 mddev->level = LEVEL_NONE;
2472 static struct md_sysfs_entry md_level =
2473 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2477 layout_show(mddev_t *mddev, char *page)
2479 /* just a number, not meaningful for all levels */
2480 if (mddev->reshape_position != MaxSector &&
2481 mddev->layout != mddev->new_layout)
2482 return sprintf(page, "%d (%d)\n",
2483 mddev->new_layout, mddev->layout);
2484 return sprintf(page, "%d\n", mddev->layout);
2488 layout_store(mddev_t *mddev, const char *buf, size_t len)
2491 unsigned long n = simple_strtoul(buf, &e, 10);
2493 if (!*buf || (*e && *e != '\n'))
2498 if (mddev->reshape_position != MaxSector)
2499 mddev->new_layout = n;
2504 static struct md_sysfs_entry md_layout =
2505 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2509 raid_disks_show(mddev_t *mddev, char *page)
2511 if (mddev->raid_disks == 0)
2513 if (mddev->reshape_position != MaxSector &&
2514 mddev->delta_disks != 0)
2515 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2516 mddev->raid_disks - mddev->delta_disks);
2517 return sprintf(page, "%d\n", mddev->raid_disks);
2520 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2523 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2527 unsigned long n = simple_strtoul(buf, &e, 10);
2529 if (!*buf || (*e && *e != '\n'))
2533 rv = update_raid_disks(mddev, n);
2534 else if (mddev->reshape_position != MaxSector) {
2535 int olddisks = mddev->raid_disks - mddev->delta_disks;
2536 mddev->delta_disks = n - olddisks;
2537 mddev->raid_disks = n;
2539 mddev->raid_disks = n;
2540 return rv ? rv : len;
2542 static struct md_sysfs_entry md_raid_disks =
2543 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2546 chunk_size_show(mddev_t *mddev, char *page)
2548 if (mddev->reshape_position != MaxSector &&
2549 mddev->chunk_size != mddev->new_chunk)
2550 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2552 return sprintf(page, "%d\n", mddev->chunk_size);
2556 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2558 /* can only set chunk_size if array is not yet active */
2560 unsigned long n = simple_strtoul(buf, &e, 10);
2562 if (!*buf || (*e && *e != '\n'))
2567 else if (mddev->reshape_position != MaxSector)
2568 mddev->new_chunk = n;
2570 mddev->chunk_size = n;
2573 static struct md_sysfs_entry md_chunk_size =
2574 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2577 resync_start_show(mddev_t *mddev, char *page)
2579 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2583 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2586 unsigned long long n = simple_strtoull(buf, &e, 10);
2590 if (!*buf || (*e && *e != '\n'))
2593 mddev->recovery_cp = n;
2596 static struct md_sysfs_entry md_resync_start =
2597 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2600 * The array state can be:
2603 * No devices, no size, no level
2604 * Equivalent to STOP_ARRAY ioctl
2606 * May have some settings, but array is not active
2607 * all IO results in error
2608 * When written, doesn't tear down array, but just stops it
2609 * suspended (not supported yet)
2610 * All IO requests will block. The array can be reconfigured.
2611 * Writing this, if accepted, will block until array is quiescent
2613 * no resync can happen. no superblocks get written.
2614 * write requests fail
2616 * like readonly, but behaves like 'clean' on a write request.
2618 * clean - no pending writes, but otherwise active.
2619 * When written to inactive array, starts without resync
2620 * If a write request arrives then
2621 * if metadata is known, mark 'dirty' and switch to 'active'.
2622 * if not known, block and switch to write-pending
2623 * If written to an active array that has pending writes, then fails.
2625 * fully active: IO and resync can be happening.
2626 * When written to inactive array, starts with resync
2629 * clean, but writes are blocked waiting for 'active' to be written.
2632 * like active, but no writes have been seen for a while (100msec).
2635 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2636 write_pending, active_idle, bad_word};
2637 static char *array_states[] = {
2638 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2639 "write-pending", "active-idle", NULL };
2641 static int match_word(const char *word, char **list)
2644 for (n=0; list[n]; n++)
2645 if (cmd_match(word, list[n]))
2651 array_state_show(mddev_t *mddev, char *page)
2653 enum array_state st = inactive;
2666 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2668 else if (mddev->safemode)
2674 if (list_empty(&mddev->disks) &&
2675 mddev->raid_disks == 0 &&
2681 return sprintf(page, "%s\n", array_states[st]);
2684 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2685 static int do_md_run(mddev_t * mddev);
2686 static int restart_array(mddev_t *mddev);
2689 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2692 enum array_state st = match_word(buf, array_states);
2697 /* stopping an active array */
2698 if (atomic_read(&mddev->active) > 1)
2700 err = do_md_stop(mddev, 0, 0);
2703 /* stopping an active array */
2705 if (atomic_read(&mddev->active) > 1)
2707 err = do_md_stop(mddev, 2, 0);
2709 err = 0; /* already inactive */
2712 break; /* not supported yet */
2715 err = do_md_stop(mddev, 1, 0);
2718 set_disk_ro(mddev->gendisk, 1);
2719 err = do_md_run(mddev);
2725 err = do_md_stop(mddev, 1, 0);
2727 err = restart_array(mddev);
2730 set_disk_ro(mddev->gendisk, 0);
2734 err = do_md_run(mddev);
2739 restart_array(mddev);
2740 spin_lock_irq(&mddev->write_lock);
2741 if (atomic_read(&mddev->writes_pending) == 0) {
2742 if (mddev->in_sync == 0) {
2744 if (mddev->safemode == 1)
2745 mddev->safemode = 0;
2746 if (mddev->persistent)
2747 set_bit(MD_CHANGE_CLEAN,
2753 spin_unlock_irq(&mddev->write_lock);
2756 mddev->recovery_cp = MaxSector;
2757 err = do_md_run(mddev);
2762 restart_array(mddev);
2763 if (mddev->external)
2764 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2765 wake_up(&mddev->sb_wait);
2769 set_disk_ro(mddev->gendisk, 0);
2770 err = do_md_run(mddev);
2775 /* these cannot be set */
2781 sysfs_notify(&mddev->kobj, NULL, "array_state");
2785 static struct md_sysfs_entry md_array_state =
2786 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2789 null_show(mddev_t *mddev, char *page)
2795 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2797 /* buf must be %d:%d\n? giving major and minor numbers */
2798 /* The new device is added to the array.
2799 * If the array has a persistent superblock, we read the
2800 * superblock to initialise info and check validity.
2801 * Otherwise, only checking done is that in bind_rdev_to_array,
2802 * which mainly checks size.
2805 int major = simple_strtoul(buf, &e, 10);
2811 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2813 minor = simple_strtoul(e+1, &e, 10);
2814 if (*e && *e != '\n')
2816 dev = MKDEV(major, minor);
2817 if (major != MAJOR(dev) ||
2818 minor != MINOR(dev))
2822 if (mddev->persistent) {
2823 rdev = md_import_device(dev, mddev->major_version,
2824 mddev->minor_version);
2825 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2826 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2827 mdk_rdev_t, same_set);
2828 err = super_types[mddev->major_version]
2829 .load_super(rdev, rdev0, mddev->minor_version);
2833 } else if (mddev->external)
2834 rdev = md_import_device(dev, -2, -1);
2836 rdev = md_import_device(dev, -1, -1);
2839 return PTR_ERR(rdev);
2840 err = bind_rdev_to_array(rdev, mddev);
2844 return err ? err : len;
2847 static struct md_sysfs_entry md_new_device =
2848 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2851 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2854 unsigned long chunk, end_chunk;
2858 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2860 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2861 if (buf == end) break;
2862 if (*end == '-') { /* range */
2864 end_chunk = simple_strtoul(buf, &end, 0);
2865 if (buf == end) break;
2867 if (*end && !isspace(*end)) break;
2868 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2870 while (isspace(*buf)) buf++;
2872 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2877 static struct md_sysfs_entry md_bitmap =
2878 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2881 size_show(mddev_t *mddev, char *page)
2883 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2886 static int update_size(mddev_t *mddev, sector_t num_sectors);
2889 size_store(mddev_t *mddev, const char *buf, size_t len)
2891 /* If array is inactive, we can reduce the component size, but
2892 * not increase it (except from 0).
2893 * If array is active, we can try an on-line resize
2897 unsigned long long size = simple_strtoull(buf, &e, 10);
2898 if (!*buf || *buf == '\n' ||
2903 err = update_size(mddev, size * 2);
2904 md_update_sb(mddev, 1);
2906 if (mddev->size == 0 ||
2912 return err ? err : len;
2915 static struct md_sysfs_entry md_size =
2916 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2921 * 'none' for arrays with no metadata (good luck...)
2922 * 'external' for arrays with externally managed metadata,
2923 * or N.M for internally known formats
2926 metadata_show(mddev_t *mddev, char *page)
2928 if (mddev->persistent)
2929 return sprintf(page, "%d.%d\n",
2930 mddev->major_version, mddev->minor_version);
2931 else if (mddev->external)
2932 return sprintf(page, "external:%s\n", mddev->metadata_type);
2934 return sprintf(page, "none\n");
2938 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2942 if (!list_empty(&mddev->disks))
2945 if (cmd_match(buf, "none")) {
2946 mddev->persistent = 0;
2947 mddev->external = 0;
2948 mddev->major_version = 0;
2949 mddev->minor_version = 90;
2952 if (strncmp(buf, "external:", 9) == 0) {
2953 size_t namelen = len-9;
2954 if (namelen >= sizeof(mddev->metadata_type))
2955 namelen = sizeof(mddev->metadata_type)-1;
2956 strncpy(mddev->metadata_type, buf+9, namelen);
2957 mddev->metadata_type[namelen] = 0;
2958 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2959 mddev->metadata_type[--namelen] = 0;
2960 mddev->persistent = 0;
2961 mddev->external = 1;
2962 mddev->major_version = 0;
2963 mddev->minor_version = 90;
2966 major = simple_strtoul(buf, &e, 10);
2967 if (e==buf || *e != '.')
2970 minor = simple_strtoul(buf, &e, 10);
2971 if (e==buf || (*e && *e != '\n') )
2973 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2975 mddev->major_version = major;
2976 mddev->minor_version = minor;
2977 mddev->persistent = 1;
2978 mddev->external = 0;
2982 static struct md_sysfs_entry md_metadata =
2983 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2986 action_show(mddev_t *mddev, char *page)
2988 char *type = "idle";
2989 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2990 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2991 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2993 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2994 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2996 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3000 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3003 return sprintf(page, "%s\n", type);
3007 action_store(mddev_t *mddev, const char *page, size_t len)
3009 if (!mddev->pers || !mddev->pers->sync_request)
3012 if (cmd_match(page, "idle")) {
3013 if (mddev->sync_thread) {
3014 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3015 md_unregister_thread(mddev->sync_thread);
3016 mddev->sync_thread = NULL;
3017 mddev->recovery = 0;
3019 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3020 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3022 else if (cmd_match(page, "resync"))
3023 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3024 else if (cmd_match(page, "recover")) {
3025 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3026 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3027 } else if (cmd_match(page, "reshape")) {
3029 if (mddev->pers->start_reshape == NULL)
3031 err = mddev->pers->start_reshape(mddev);
3034 sysfs_notify(&mddev->kobj, NULL, "degraded");
3036 if (cmd_match(page, "check"))
3037 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3038 else if (!cmd_match(page, "repair"))
3040 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3041 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3043 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3044 md_wakeup_thread(mddev->thread);
3045 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3050 mismatch_cnt_show(mddev_t *mddev, char *page)
3052 return sprintf(page, "%llu\n",
3053 (unsigned long long) mddev->resync_mismatches);
3056 static struct md_sysfs_entry md_scan_mode =
3057 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3060 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3063 sync_min_show(mddev_t *mddev, char *page)
3065 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3066 mddev->sync_speed_min ? "local": "system");
3070 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3074 if (strncmp(buf, "system", 6)==0) {
3075 mddev->sync_speed_min = 0;
3078 min = simple_strtoul(buf, &e, 10);
3079 if (buf == e || (*e && *e != '\n') || min <= 0)
3081 mddev->sync_speed_min = min;
3085 static struct md_sysfs_entry md_sync_min =
3086 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3089 sync_max_show(mddev_t *mddev, char *page)
3091 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3092 mddev->sync_speed_max ? "local": "system");
3096 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3100 if (strncmp(buf, "system", 6)==0) {
3101 mddev->sync_speed_max = 0;
3104 max = simple_strtoul(buf, &e, 10);
3105 if (buf == e || (*e && *e != '\n') || max <= 0)
3107 mddev->sync_speed_max = max;
3111 static struct md_sysfs_entry md_sync_max =
3112 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3115 degraded_show(mddev_t *mddev, char *page)
3117 return sprintf(page, "%d\n", mddev->degraded);
3119 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3122 sync_force_parallel_show(mddev_t *mddev, char *page)
3124 return sprintf(page, "%d\n", mddev->parallel_resync);
3128 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3132 if (strict_strtol(buf, 10, &n))
3135 if (n != 0 && n != 1)
3138 mddev->parallel_resync = n;
3140 if (mddev->sync_thread)
3141 wake_up(&resync_wait);
3146 /* force parallel resync, even with shared block devices */
3147 static struct md_sysfs_entry md_sync_force_parallel =
3148 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3149 sync_force_parallel_show, sync_force_parallel_store);
3152 sync_speed_show(mddev_t *mddev, char *page)
3154 unsigned long resync, dt, db;
3155 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3156 dt = (jiffies - mddev->resync_mark) / HZ;
3158 db = resync - mddev->resync_mark_cnt;
3159 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3162 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3165 sync_completed_show(mddev_t *mddev, char *page)
3167 unsigned long max_blocks, resync;
3169 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3170 max_blocks = mddev->resync_max_sectors;
3172 max_blocks = mddev->size << 1;
3174 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3175 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3178 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3181 min_sync_show(mddev_t *mddev, char *page)
3183 return sprintf(page, "%llu\n",
3184 (unsigned long long)mddev->resync_min);
3187 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3189 unsigned long long min;
3190 if (strict_strtoull(buf, 10, &min))
3192 if (min > mddev->resync_max)
3194 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3197 /* Must be a multiple of chunk_size */
3198 if (mddev->chunk_size) {
3199 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3202 mddev->resync_min = min;
3207 static struct md_sysfs_entry md_min_sync =
3208 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3211 max_sync_show(mddev_t *mddev, char *page)
3213 if (mddev->resync_max == MaxSector)
3214 return sprintf(page, "max\n");
3216 return sprintf(page, "%llu\n",
3217 (unsigned long long)mddev->resync_max);
3220 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3222 if (strncmp(buf, "max", 3) == 0)
3223 mddev->resync_max = MaxSector;
3225 unsigned long long max;
3226 if (strict_strtoull(buf, 10, &max))
3228 if (max < mddev->resync_min)
3230 if (max < mddev->resync_max &&
3231 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3234 /* Must be a multiple of chunk_size */
3235 if (mddev->chunk_size) {
3236 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3239 mddev->resync_max = max;
3241 wake_up(&mddev->recovery_wait);
3245 static struct md_sysfs_entry md_max_sync =
3246 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3249 suspend_lo_show(mddev_t *mddev, char *page)
3251 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3255 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3258 unsigned long long new = simple_strtoull(buf, &e, 10);
3260 if (mddev->pers->quiesce == NULL)
3262 if (buf == e || (*e && *e != '\n'))
3264 if (new >= mddev->suspend_hi ||
3265 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3266 mddev->suspend_lo = new;
3267 mddev->pers->quiesce(mddev, 2);
3272 static struct md_sysfs_entry md_suspend_lo =
3273 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3277 suspend_hi_show(mddev_t *mddev, char *page)
3279 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3283 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3286 unsigned long long new = simple_strtoull(buf, &e, 10);
3288 if (mddev->pers->quiesce == NULL)
3290 if (buf == e || (*e && *e != '\n'))
3292 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3293 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3294 mddev->suspend_hi = new;
3295 mddev->pers->quiesce(mddev, 1);
3296 mddev->pers->quiesce(mddev, 0);
3301 static struct md_sysfs_entry md_suspend_hi =
3302 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3305 reshape_position_show(mddev_t *mddev, char *page)
3307 if (mddev->reshape_position != MaxSector)
3308 return sprintf(page, "%llu\n",
3309 (unsigned long long)mddev->reshape_position);
3310 strcpy(page, "none\n");
3315 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3318 unsigned long long new = simple_strtoull(buf, &e, 10);
3321 if (buf == e || (*e && *e != '\n'))
3323 mddev->reshape_position = new;
3324 mddev->delta_disks = 0;
3325 mddev->new_level = mddev->level;
3326 mddev->new_layout = mddev->layout;
3327 mddev->new_chunk = mddev->chunk_size;
3331 static struct md_sysfs_entry md_reshape_position =
3332 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3333 reshape_position_store);
3336 static struct attribute *md_default_attrs[] = {
3339 &md_raid_disks.attr,
3340 &md_chunk_size.attr,
3342 &md_resync_start.attr,
3344 &md_new_device.attr,
3345 &md_safe_delay.attr,
3346 &md_array_state.attr,
3347 &md_reshape_position.attr,
3351 static struct attribute *md_redundancy_attrs[] = {
3353 &md_mismatches.attr,
3356 &md_sync_speed.attr,
3357 &md_sync_force_parallel.attr,
3358 &md_sync_completed.attr,
3361 &md_suspend_lo.attr,
3362 &md_suspend_hi.attr,
3367 static struct attribute_group md_redundancy_group = {
3369 .attrs = md_redundancy_attrs,
3374 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3376 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3377 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3382 rv = mddev_lock(mddev);
3384 rv = entry->show(mddev, page);
3385 mddev_unlock(mddev);
3391 md_attr_store(struct kobject *kobj, struct attribute *attr,
3392 const char *page, size_t length)
3394 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3395 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3400 if (!capable(CAP_SYS_ADMIN))
3402 rv = mddev_lock(mddev);
3404 rv = entry->store(mddev, page, length);
3405 mddev_unlock(mddev);
3410 static void md_free(struct kobject *ko)
3412 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3416 static struct sysfs_ops md_sysfs_ops = {
3417 .show = md_attr_show,
3418 .store = md_attr_store,
3420 static struct kobj_type md_ktype = {
3422 .sysfs_ops = &md_sysfs_ops,
3423 .default_attrs = md_default_attrs,
3428 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3430 static DEFINE_MUTEX(disks_mutex);
3431 mddev_t *mddev = mddev_find(dev);
3432 struct gendisk *disk;
3433 int partitioned = (MAJOR(dev) != MD_MAJOR);
3434 int shift = partitioned ? MdpMinorShift : 0;
3435 int unit = MINOR(dev) >> shift;
3441 mutex_lock(&disks_mutex);
3442 if (mddev->gendisk) {
3443 mutex_unlock(&disks_mutex);
3447 disk = alloc_disk(1 << shift);
3449 mutex_unlock(&disks_mutex);
3453 disk->major = MAJOR(dev);
3454 disk->first_minor = unit << shift;
3456 sprintf(disk->disk_name, "md_d%d", unit);
3458 sprintf(disk->disk_name, "md%d", unit);
3459 disk->fops = &md_fops;
3460 disk->private_data = mddev;
3461 disk->queue = mddev->queue;
3463 mddev->gendisk = disk;
3464 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3466 mutex_unlock(&disks_mutex);
3468 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3471 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3475 static void md_safemode_timeout(unsigned long data)
3477 mddev_t *mddev = (mddev_t *) data;
3479 if (!atomic_read(&mddev->writes_pending)) {
3480 mddev->safemode = 1;
3481 if (mddev->external)
3482 sysfs_notify(&mddev->kobj, NULL, "array_state");
3484 md_wakeup_thread(mddev->thread);
3487 static int start_dirty_degraded;
3489 static int do_md_run(mddev_t * mddev)
3493 struct list_head *tmp;
3495 struct gendisk *disk;
3496 struct mdk_personality *pers;
3497 char b[BDEVNAME_SIZE];
3499 if (list_empty(&mddev->disks))
3500 /* cannot run an array with no devices.. */
3507 * Analyze all RAID superblock(s)
3509 if (!mddev->raid_disks) {
3510 if (!mddev->persistent)
3515 chunk_size = mddev->chunk_size;
3518 if (chunk_size > MAX_CHUNK_SIZE) {
3519 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3520 chunk_size, MAX_CHUNK_SIZE);
3524 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3526 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3527 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3530 if (chunk_size < PAGE_SIZE) {
3531 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3532 chunk_size, PAGE_SIZE);
3536 /* devices must have minimum size of one chunk */
3537 rdev_for_each(rdev, tmp, mddev) {
3538 if (test_bit(Faulty, &rdev->flags))
3540 if (rdev->size < chunk_size / 1024) {
3542 "md: Dev %s smaller than chunk_size:"
3544 bdevname(rdev->bdev,b),
3545 (unsigned long long)rdev->size,
3553 if (mddev->level != LEVEL_NONE)
3554 request_module("md-level-%d", mddev->level);
3555 else if (mddev->clevel[0])
3556 request_module("md-%s", mddev->clevel);
3560 * Drop all container device buffers, from now on
3561 * the only valid external interface is through the md
3564 rdev_for_each(rdev, tmp, mddev) {
3565 if (test_bit(Faulty, &rdev->flags))
3567 sync_blockdev(rdev->bdev);
3568 invalidate_bdev(rdev->bdev);
3570 /* perform some consistency tests on the device.
3571 * We don't want the data to overlap the metadata,
3572 * Internal Bitmap issues has handled elsewhere.
3574 if (rdev->data_offset < rdev->sb_start) {
3576 rdev->data_offset + mddev->size*2
3578 printk("md: %s: data overlaps metadata\n",
3583 if (rdev->sb_start + rdev->sb_size/512
3584 > rdev->data_offset) {
3585 printk("md: %s: metadata overlaps data\n",
3590 sysfs_notify(&rdev->kobj, NULL, "state");
3593 md_probe(mddev->unit, NULL, NULL);
3594 disk = mddev->gendisk;
3598 spin_lock(&pers_lock);
3599 pers = find_pers(mddev->level, mddev->clevel);
3600 if (!pers || !try_module_get(pers->owner)) {
3601 spin_unlock(&pers_lock);
3602 if (mddev->level != LEVEL_NONE)
3603 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3606 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3611 spin_unlock(&pers_lock);
3612 mddev->level = pers->level;
3613 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3615 if (mddev->reshape_position != MaxSector &&
3616 pers->start_reshape == NULL) {
3617 /* This personality cannot handle reshaping... */
3619 module_put(pers->owner);
3623 if (pers->sync_request) {
3624 /* Warn if this is a potentially silly
3627 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3629 struct list_head *tmp2;
3631 rdev_for_each(rdev, tmp, mddev) {
3632 rdev_for_each(rdev2, tmp2, mddev) {
3634 rdev->bdev->bd_contains ==
3635 rdev2->bdev->bd_contains) {
3637 "%s: WARNING: %s appears to be"
3638 " on the same physical disk as"
3641 bdevname(rdev->bdev,b),
3642 bdevname(rdev2->bdev,b2));
3649 "True protection against single-disk"
3650 " failure might be compromised.\n");
3653 mddev->recovery = 0;
3654 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3655 mddev->barriers_work = 1;
3656 mddev->ok_start_degraded = start_dirty_degraded;
3659 mddev->ro = 2; /* read-only, but switch on first write */
3661 err = mddev->pers->run(mddev);
3663 printk(KERN_ERR "md: pers->run() failed ...\n");
3664 else if (mddev->pers->sync_request) {
3665 err = bitmap_create(mddev);
3667 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3668 mdname(mddev), err);
3669 mddev->pers->stop(mddev);
3673 module_put(mddev->pers->owner);
3675 bitmap_destroy(mddev);
3678 if (mddev->pers->sync_request) {
3679 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3681 "md: cannot register extra attributes for %s\n",
3683 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3686 atomic_set(&mddev->writes_pending,0);
3687 mddev->safemode = 0;
3688 mddev->safemode_timer.function = md_safemode_timeout;
3689 mddev->safemode_timer.data = (unsigned long) mddev;
3690 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3693 rdev_for_each(rdev, tmp, mddev)
3694 if (rdev->raid_disk >= 0) {
3696 sprintf(nm, "rd%d", rdev->raid_disk);
3697 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3698 printk("md: cannot register %s for %s\n",
3702 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3705 md_update_sb(mddev, 0);
3707 set_capacity(disk, mddev->array_size<<1);
3709 /* If we call blk_queue_make_request here, it will
3710 * re-initialise max_sectors etc which may have been
3711 * refined inside -> run. So just set the bits we need to set.
3712 * Most initialisation happended when we called
3713 * blk_queue_make_request(..., md_fail_request)
3716 mddev->queue->queuedata = mddev;
3717 mddev->queue->make_request_fn = mddev->pers->make_request;
3719 /* If there is a partially-recovered drive we need to
3720 * start recovery here. If we leave it to md_check_recovery,
3721 * it will remove the drives and not do the right thing
3723 if (mddev->degraded && !mddev->sync_thread) {
3724 struct list_head *rtmp;
3726 rdev_for_each(rdev, rtmp, mddev)
3727 if (rdev->raid_disk >= 0 &&
3728 !test_bit(In_sync, &rdev->flags) &&
3729 !test_bit(Faulty, &rdev->flags))
3730 /* complete an interrupted recovery */
3732 if (spares && mddev->pers->sync_request) {
3733 mddev->recovery = 0;
3734 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3735 mddev->sync_thread = md_register_thread(md_do_sync,
3738 if (!mddev->sync_thread) {
3739 printk(KERN_ERR "%s: could not start resync"
3742 /* leave the spares where they are, it shouldn't hurt */
3743 mddev->recovery = 0;
3747 md_wakeup_thread(mddev->thread);
3748 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3751 md_new_event(mddev);
3752 sysfs_notify(&mddev->kobj, NULL, "array_state");
3753 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3754 sysfs_notify(&mddev->kobj, NULL, "degraded");
3755 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3759 static int restart_array(mddev_t *mddev)
3761 struct gendisk *disk = mddev->gendisk;
3763 /* Complain if it has no devices */
3764 if (list_empty(&mddev->disks))
3770 mddev->safemode = 0;
3772 set_disk_ro(disk, 0);
3773 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3775 /* Kick recovery or resync if necessary */
3776 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3777 md_wakeup_thread(mddev->thread);
3778 md_wakeup_thread(mddev->sync_thread);
3779 sysfs_notify(&mddev->kobj, NULL, "array_state");
3783 /* similar to deny_write_access, but accounts for our holding a reference
3784 * to the file ourselves */
3785 static int deny_bitmap_write_access(struct file * file)
3787 struct inode *inode = file->f_mapping->host;
3789 spin_lock(&inode->i_lock);
3790 if (atomic_read(&inode->i_writecount) > 1) {
3791 spin_unlock(&inode->i_lock);
3794 atomic_set(&inode->i_writecount, -1);
3795 spin_unlock(&inode->i_lock);
3800 static void restore_bitmap_write_access(struct file *file)
3802 struct inode *inode = file->f_mapping->host;
3804 spin_lock(&inode->i_lock);
3805 atomic_set(&inode->i_writecount, 1);
3806 spin_unlock(&inode->i_lock);
3810 * 0 - completely stop and dis-assemble array
3811 * 1 - switch to readonly
3812 * 2 - stop but do not disassemble array
3814 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3817 struct gendisk *disk = mddev->gendisk;
3819 if (atomic_read(&mddev->active) > 1 + is_open) {
3820 printk("md: %s still in use.\n",mdname(mddev));
3826 if (mddev->sync_thread) {
3827 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3828 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3829 md_unregister_thread(mddev->sync_thread);
3830 mddev->sync_thread = NULL;
3833 del_timer_sync(&mddev->safemode_timer);
3835 invalidate_partition(disk, 0);
3838 case 1: /* readonly */
3844 case 0: /* disassemble */
3846 bitmap_flush(mddev);
3847 md_super_wait(mddev);
3849 set_disk_ro(disk, 0);
3850 blk_queue_make_request(mddev->queue, md_fail_request);
3851 mddev->pers->stop(mddev);
3852 mddev->queue->merge_bvec_fn = NULL;
3853 mddev->queue->unplug_fn = NULL;
3854 mddev->queue->backing_dev_info.congested_fn = NULL;
3855 if (mddev->pers->sync_request)
3856 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3858 module_put(mddev->pers->owner);
3860 /* tell userspace to handle 'inactive' */
3861 sysfs_notify(&mddev->kobj, NULL, "array_state");
3863 set_capacity(disk, 0);
3869 if (!mddev->in_sync || mddev->flags) {
3870 /* mark array as shutdown cleanly */
3872 md_update_sb(mddev, 1);
3875 set_disk_ro(disk, 1);
3876 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3880 * Free resources if final stop
3884 struct list_head *tmp;
3886 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3888 bitmap_destroy(mddev);
3889 if (mddev->bitmap_file) {
3890 restore_bitmap_write_access(mddev->bitmap_file);
3891 fput(mddev->bitmap_file);
3892 mddev->bitmap_file = NULL;
3894 mddev->bitmap_offset = 0;
3896 rdev_for_each(rdev, tmp, mddev)
3897 if (rdev->raid_disk >= 0) {
3899 sprintf(nm, "rd%d", rdev->raid_disk);
3900 sysfs_remove_link(&mddev->kobj, nm);
3903 /* make sure all md_delayed_delete calls have finished */
3904 flush_scheduled_work();
3906 export_array(mddev);
3908 mddev->array_size = 0;
3910 mddev->raid_disks = 0;
3911 mddev->recovery_cp = 0;
3912 mddev->resync_min = 0;
3913 mddev->resync_max = MaxSector;
3914 mddev->reshape_position = MaxSector;
3915 mddev->external = 0;
3916 mddev->persistent = 0;
3917 mddev->level = LEVEL_NONE;
3918 mddev->clevel[0] = 0;
3921 mddev->metadata_type[0] = 0;
3922 mddev->chunk_size = 0;
3923 mddev->ctime = mddev->utime = 0;
3925 mddev->max_disks = 0;
3927 mddev->delta_disks = 0;
3928 mddev->new_level = LEVEL_NONE;
3929 mddev->new_layout = 0;
3930 mddev->new_chunk = 0;
3931 mddev->curr_resync = 0;
3932 mddev->resync_mismatches = 0;
3933 mddev->suspend_lo = mddev->suspend_hi = 0;
3934 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3935 mddev->recovery = 0;
3938 mddev->degraded = 0;
3939 mddev->barriers_work = 0;
3940 mddev->safemode = 0;
3942 } else if (mddev->pers)
3943 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3946 md_new_event(mddev);
3947 sysfs_notify(&mddev->kobj, NULL, "array_state");
3953 static void autorun_array(mddev_t *mddev)
3956 struct list_head *tmp;
3959 if (list_empty(&mddev->disks))
3962 printk(KERN_INFO "md: running: ");
3964 rdev_for_each(rdev, tmp, mddev) {
3965 char b[BDEVNAME_SIZE];
3966 printk("<%s>", bdevname(rdev->bdev,b));
3970 err = do_md_run (mddev);
3972 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3973 do_md_stop (mddev, 0, 0);
3978 * lets try to run arrays based on all disks that have arrived
3979 * until now. (those are in pending_raid_disks)
3981 * the method: pick the first pending disk, collect all disks with
3982 * the same UUID, remove all from the pending list and put them into
3983 * the 'same_array' list. Then order this list based on superblock
3984 * update time (freshest comes first), kick out 'old' disks and
3985 * compare superblocks. If everything's fine then run it.
3987 * If "unit" is allocated, then bump its reference count
3989 static void autorun_devices(int part)
3991 struct list_head *tmp;
3992 mdk_rdev_t *rdev0, *rdev;
3994 char b[BDEVNAME_SIZE];
3996 printk(KERN_INFO "md: autorun ...\n");
3997 while (!list_empty(&pending_raid_disks)) {
4000 LIST_HEAD(candidates);
4001 rdev0 = list_entry(pending_raid_disks.next,
4002 mdk_rdev_t, same_set);
4004 printk(KERN_INFO "md: considering %s ...\n",
4005 bdevname(rdev0->bdev,b));
4006 INIT_LIST_HEAD(&candidates);
4007 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4008 if (super_90_load(rdev, rdev0, 0) >= 0) {
4009 printk(KERN_INFO "md: adding %s ...\n",
4010 bdevname(rdev->bdev,b));
4011 list_move(&rdev->same_set, &candidates);
4014 * now we have a set of devices, with all of them having
4015 * mostly sane superblocks. It's time to allocate the
4019 dev = MKDEV(mdp_major,
4020 rdev0->preferred_minor << MdpMinorShift);
4021 unit = MINOR(dev) >> MdpMinorShift;
4023 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4026 if (rdev0->preferred_minor != unit) {
4027 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4028 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4032 md_probe(dev, NULL, NULL);
4033 mddev = mddev_find(dev);
4034 if (!mddev || !mddev->gendisk) {
4038 "md: cannot allocate memory for md drive.\n");
4041 if (mddev_lock(mddev))
4042 printk(KERN_WARNING "md: %s locked, cannot run\n",
4044 else if (mddev->raid_disks || mddev->major_version
4045 || !list_empty(&mddev->disks)) {
4047 "md: %s already running, cannot run %s\n",
4048 mdname(mddev), bdevname(rdev0->bdev,b));
4049 mddev_unlock(mddev);
4051 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4052 mddev->persistent = 1;
4053 rdev_for_each_list(rdev, tmp, candidates) {
4054 list_del_init(&rdev->same_set);
4055 if (bind_rdev_to_array(rdev, mddev))
4058 autorun_array(mddev);
4059 mddev_unlock(mddev);
4061 /* on success, candidates will be empty, on error
4064 rdev_for_each_list(rdev, tmp, candidates)
4068 printk(KERN_INFO "md: ... autorun DONE.\n");
4070 #endif /* !MODULE */
4072 static int get_version(void __user * arg)
4076 ver.major = MD_MAJOR_VERSION;
4077 ver.minor = MD_MINOR_VERSION;
4078 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4080 if (copy_to_user(arg, &ver, sizeof(ver)))
4086 static int get_array_info(mddev_t * mddev, void __user * arg)
4088 mdu_array_info_t info;
4089 int nr,working,active,failed,spare;
4091 struct list_head *tmp;
4093 nr=working=active=failed=spare=0;
4094 rdev_for_each(rdev, tmp, mddev) {
4096 if (test_bit(Faulty, &rdev->flags))
4100 if (test_bit(In_sync, &rdev->flags))
4107 info.major_version = mddev->major_version;
4108 info.minor_version = mddev->minor_version;
4109 info.patch_version = MD_PATCHLEVEL_VERSION;
4110 info.ctime = mddev->ctime;
4111 info.level = mddev->level;
4112 info.size = mddev->size;
4113 if (info.size != mddev->size) /* overflow */
4116 info.raid_disks = mddev->raid_disks;
4117 info.md_minor = mddev->md_minor;
4118 info.not_persistent= !mddev->persistent;
4120 info.utime = mddev->utime;
4123 info.state = (1<<MD_SB_CLEAN);
4124 if (mddev->bitmap && mddev->bitmap_offset)
4125 info.state = (1<<MD_SB_BITMAP_PRESENT);
4126 info.active_disks = active;
4127 info.working_disks = working;
4128 info.failed_disks = failed;
4129 info.spare_disks = spare;
4131 info.layout = mddev->layout;
4132 info.chunk_size = mddev->chunk_size;
4134 if (copy_to_user(arg, &info, sizeof(info)))
4140 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4142 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4143 char *ptr, *buf = NULL;
4146 if (md_allow_write(mddev))
4147 file = kmalloc(sizeof(*file), GFP_NOIO);
4149 file = kmalloc(sizeof(*file), GFP_KERNEL);
4154 /* bitmap disabled, zero the first byte and copy out */
4155 if (!mddev->bitmap || !mddev->bitmap->file) {
4156 file->pathname[0] = '\0';
4160 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4164 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4168 strcpy(file->pathname, ptr);
4172 if (copy_to_user(arg, file, sizeof(*file)))
4180 static int get_disk_info(mddev_t * mddev, void __user * arg)
4182 mdu_disk_info_t info;
4185 if (copy_from_user(&info, arg, sizeof(info)))
4188 rdev = find_rdev_nr(mddev, info.number);
4190 info.major = MAJOR(rdev->bdev->bd_dev);
4191 info.minor = MINOR(rdev->bdev->bd_dev);
4192 info.raid_disk = rdev->raid_disk;
4194 if (test_bit(Faulty, &rdev->flags))
4195 info.state |= (1<<MD_DISK_FAULTY);
4196 else if (test_bit(In_sync, &rdev->flags)) {
4197 info.state |= (1<<MD_DISK_ACTIVE);
4198 info.state |= (1<<MD_DISK_SYNC);
4200 if (test_bit(WriteMostly, &rdev->flags))
4201 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4203 info.major = info.minor = 0;
4204 info.raid_disk = -1;
4205 info.state = (1<<MD_DISK_REMOVED);
4208 if (copy_to_user(arg, &info, sizeof(info)))
4214 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4216 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4218 dev_t dev = MKDEV(info->major,info->minor);
4220 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4223 if (!mddev->raid_disks) {
4225 /* expecting a device which has a superblock */
4226 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4229 "md: md_import_device returned %ld\n",
4231 return PTR_ERR(rdev);
4233 if (!list_empty(&mddev->disks)) {
4234 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4235 mdk_rdev_t, same_set);
4236 int err = super_types[mddev->major_version]
4237 .load_super(rdev, rdev0, mddev->minor_version);
4240 "md: %s has different UUID to %s\n",
4241 bdevname(rdev->bdev,b),
4242 bdevname(rdev0->bdev,b2));
4247 err = bind_rdev_to_array(rdev, mddev);
4254 * add_new_disk can be used once the array is assembled
4255 * to add "hot spares". They must already have a superblock
4260 if (!mddev->pers->hot_add_disk) {
4262 "%s: personality does not support diskops!\n",
4266 if (mddev->persistent)
4267 rdev = md_import_device(dev, mddev->major_version,
4268 mddev->minor_version);
4270 rdev = md_import_device(dev, -1, -1);
4273 "md: md_import_device returned %ld\n",
4275 return PTR_ERR(rdev);
4277 /* set save_raid_disk if appropriate */
4278 if (!mddev->persistent) {
4279 if (info->state & (1<<MD_DISK_SYNC) &&
4280 info->raid_disk < mddev->raid_disks)
4281 rdev->raid_disk = info->raid_disk;
4283 rdev->raid_disk = -1;
4285 super_types[mddev->major_version].
4286 validate_super(mddev, rdev);
4287 rdev->saved_raid_disk = rdev->raid_disk;
4289 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4290 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4291 set_bit(WriteMostly, &rdev->flags);
4293 rdev->raid_disk = -1;
4294 err = bind_rdev_to_array(rdev, mddev);
4295 if (!err && !mddev->pers->hot_remove_disk) {
4296 /* If there is hot_add_disk but no hot_remove_disk
4297 * then added disks for geometry changes,
4298 * and should be added immediately.
4300 super_types[mddev->major_version].
4301 validate_super(mddev, rdev);
4302 err = mddev->pers->hot_add_disk(mddev, rdev);
4304 unbind_rdev_from_array(rdev);
4309 sysfs_notify(&rdev->kobj, NULL, "state");
4311 md_update_sb(mddev, 1);
4312 if (mddev->degraded)
4313 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4314 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4315 md_wakeup_thread(mddev->thread);
4319 /* otherwise, add_new_disk is only allowed
4320 * for major_version==0 superblocks
4322 if (mddev->major_version != 0) {
4323 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4328 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4330 rdev = md_import_device (dev, -1, 0);
4333 "md: error, md_import_device() returned %ld\n",
4335 return PTR_ERR(rdev);
4337 rdev->desc_nr = info->number;
4338 if (info->raid_disk < mddev->raid_disks)
4339 rdev->raid_disk = info->raid_disk;
4341 rdev->raid_disk = -1;
4343 if (rdev->raid_disk < mddev->raid_disks)
4344 if (info->state & (1<<MD_DISK_SYNC))
4345 set_bit(In_sync, &rdev->flags);
4347 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4348 set_bit(WriteMostly, &rdev->flags);
4350 if (!mddev->persistent) {
4351 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4352 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4354 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4355 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4357 err = bind_rdev_to_array(rdev, mddev);
4367 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4369 char b[BDEVNAME_SIZE];
4372 rdev = find_rdev(mddev, dev);
4376 if (rdev->raid_disk >= 0)
4379 kick_rdev_from_array(rdev);
4380 md_update_sb(mddev, 1);
4381 md_new_event(mddev);
4385 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4386 bdevname(rdev->bdev,b), mdname(mddev));
4390 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4392 char b[BDEVNAME_SIZE];
4399 if (mddev->major_version != 0) {
4400 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4401 " version-0 superblocks.\n",
4405 if (!mddev->pers->hot_add_disk) {
4407 "%s: personality does not support diskops!\n",
4412 rdev = md_import_device (dev, -1, 0);
4415 "md: error, md_import_device() returned %ld\n",
4420 if (mddev->persistent)
4421 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4423 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4425 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4427 if (test_bit(Faulty, &rdev->flags)) {
4429 "md: can not hot-add faulty %s disk to %s!\n",
4430 bdevname(rdev->bdev,b), mdname(mddev));
4434 clear_bit(In_sync, &rdev->flags);
4436 rdev->saved_raid_disk = -1;
4437 err = bind_rdev_to_array(rdev, mddev);
4442 * The rest should better be atomic, we can have disk failures
4443 * noticed in interrupt contexts ...
4446 if (rdev->desc_nr == mddev->max_disks) {
4447 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4450 goto abort_unbind_export;
4453 rdev->raid_disk = -1;
4455 md_update_sb(mddev, 1);
4458 * Kick recovery, maybe this spare has to be added to the
4459 * array immediately.
4461 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4462 md_wakeup_thread(mddev->thread);
4463 md_new_event(mddev);
4466 abort_unbind_export:
4467 unbind_rdev_from_array(rdev);
4474 static int set_bitmap_file(mddev_t *mddev, int fd)
4479 if (!mddev->pers->quiesce)
4481 if (mddev->recovery || mddev->sync_thread)
4483 /* we should be able to change the bitmap.. */
4489 return -EEXIST; /* cannot add when bitmap is present */
4490 mddev->bitmap_file = fget(fd);
4492 if (mddev->bitmap_file == NULL) {
4493 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4498 err = deny_bitmap_write_access(mddev->bitmap_file);
4500 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4502 fput(mddev->bitmap_file);
4503 mddev->bitmap_file = NULL;
4506 mddev->bitmap_offset = 0; /* file overrides offset */
4507 } else if (mddev->bitmap == NULL)
4508 return -ENOENT; /* cannot remove what isn't there */
4511 mddev->pers->quiesce(mddev, 1);
4513 err = bitmap_create(mddev);
4514 if (fd < 0 || err) {
4515 bitmap_destroy(mddev);
4516 fd = -1; /* make sure to put the file */
4518 mddev->pers->quiesce(mddev, 0);
4521 if (mddev->bitmap_file) {
4522 restore_bitmap_write_access(mddev->bitmap_file);
4523 fput(mddev->bitmap_file);
4525 mddev->bitmap_file = NULL;
4532 * set_array_info is used two different ways
4533 * The original usage is when creating a new array.
4534 * In this usage, raid_disks is > 0 and it together with
4535 * level, size, not_persistent,layout,chunksize determine the
4536 * shape of the array.
4537 * This will always create an array with a type-0.90.0 superblock.
4538 * The newer usage is when assembling an array.
4539 * In this case raid_disks will be 0, and the major_version field is
4540 * use to determine which style super-blocks are to be found on the devices.
4541 * The minor and patch _version numbers are also kept incase the
4542 * super_block handler wishes to interpret them.
4544 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4547 if (info->raid_disks == 0) {
4548 /* just setting version number for superblock loading */
4549 if (info->major_version < 0 ||
4550 info->major_version >= ARRAY_SIZE(super_types) ||
4551 super_types[info->major_version].name == NULL) {
4552 /* maybe try to auto-load a module? */
4554 "md: superblock version %d not known\n",
4555 info->major_version);
4558 mddev->major_version = info->major_version;
4559 mddev->minor_version = info->minor_version;
4560 mddev->patch_version = info->patch_version;
4561 mddev->persistent = !info->not_persistent;
4564 mddev->major_version = MD_MAJOR_VERSION;
4565 mddev->minor_version = MD_MINOR_VERSION;
4566 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4567 mddev->ctime = get_seconds();
4569 mddev->level = info->level;
4570 mddev->clevel[0] = 0;
4571 mddev->size = info->size;
4572 mddev->raid_disks = info->raid_disks;
4573 /* don't set md_minor, it is determined by which /dev/md* was
4576 if (info->state & (1<<MD_SB_CLEAN))
4577 mddev->recovery_cp = MaxSector;
4579 mddev->recovery_cp = 0;
4580 mddev->persistent = ! info->not_persistent;
4581 mddev->external = 0;
4583 mddev->layout = info->layout;
4584 mddev->chunk_size = info->chunk_size;
4586 mddev->max_disks = MD_SB_DISKS;
4588 if (mddev->persistent)
4590 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4592 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4593 mddev->bitmap_offset = 0;
4595 mddev->reshape_position = MaxSector;
4598 * Generate a 128 bit UUID
4600 get_random_bytes(mddev->uuid, 16);
4602 mddev->new_level = mddev->level;
4603 mddev->new_chunk = mddev->chunk_size;
4604 mddev->new_layout = mddev->layout;
4605 mddev->delta_disks = 0;
4610 static int update_size(mddev_t *mddev, sector_t num_sectors)
4614 struct list_head *tmp;
4615 int fit = (num_sectors == 0);
4617 if (mddev->pers->resize == NULL)
4619 /* The "num_sectors" is the number of sectors of each device that
4620 * is used. This can only make sense for arrays with redundancy.
4621 * linear and raid0 always use whatever space is available. We can only
4622 * consider changing this number if no resync or reconstruction is
4623 * happening, and if the new size is acceptable. It must fit before the
4624 * sb_start or, if that is <data_offset, it must fit before the size
4625 * of each device. If num_sectors is zero, we find the largest size
4629 if (mddev->sync_thread)
4631 rdev_for_each(rdev, tmp, mddev) {
4633 avail = rdev->size * 2;
4635 if (fit && (num_sectors == 0 || num_sectors > avail))
4636 num_sectors = avail;
4637 if (avail < num_sectors)
4640 rv = mddev->pers->resize(mddev, num_sectors);
4642 struct block_device *bdev;
4644 bdev = bdget_disk(mddev->gendisk, 0);
4646 mutex_lock(&bdev->bd_inode->i_mutex);
4647 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4648 mutex_unlock(&bdev->bd_inode->i_mutex);
4655 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4658 /* change the number of raid disks */
4659 if (mddev->pers->check_reshape == NULL)
4661 if (raid_disks <= 0 ||
4662 raid_disks >= mddev->max_disks)
4664 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4666 mddev->delta_disks = raid_disks - mddev->raid_disks;
4668 rv = mddev->pers->check_reshape(mddev);
4674 * update_array_info is used to change the configuration of an
4676 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4677 * fields in the info are checked against the array.
4678 * Any differences that cannot be handled will cause an error.
4679 * Normally, only one change can be managed at a time.
4681 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4687 /* calculate expected state,ignoring low bits */
4688 if (mddev->bitmap && mddev->bitmap_offset)
4689 state |= (1 << MD_SB_BITMAP_PRESENT);
4691 if (mddev->major_version != info->major_version ||
4692 mddev->minor_version != info->minor_version ||
4693 /* mddev->patch_version != info->patch_version || */
4694 mddev->ctime != info->ctime ||
4695 mddev->level != info->level ||
4696 /* mddev->layout != info->layout || */
4697 !mddev->persistent != info->not_persistent||
4698 mddev->chunk_size != info->chunk_size ||
4699 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4700 ((state^info->state) & 0xfffffe00)
4703 /* Check there is only one change */
4704 if (info->size >= 0 && mddev->size != info->size) cnt++;
4705 if (mddev->raid_disks != info->raid_disks) cnt++;
4706 if (mddev->layout != info->layout) cnt++;
4707 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4708 if (cnt == 0) return 0;
4709 if (cnt > 1) return -EINVAL;
4711 if (mddev->layout != info->layout) {
4713 * we don't need to do anything at the md level, the
4714 * personality will take care of it all.
4716 if (mddev->pers->reconfig == NULL)
4719 return mddev->pers->reconfig(mddev, info->layout, -1);
4721 if (info->size >= 0 && mddev->size != info->size)
4722 rv = update_size(mddev, (sector_t)info->size * 2);
4724 if (mddev->raid_disks != info->raid_disks)
4725 rv = update_raid_disks(mddev, info->raid_disks);
4727 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4728 if (mddev->pers->quiesce == NULL)
4730 if (mddev->recovery || mddev->sync_thread)
4732 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4733 /* add the bitmap */
4736 if (mddev->default_bitmap_offset == 0)
4738 mddev->bitmap_offset = mddev->default_bitmap_offset;
4739 mddev->pers->quiesce(mddev, 1);
4740 rv = bitmap_create(mddev);
4742 bitmap_destroy(mddev);
4743 mddev->pers->quiesce(mddev, 0);
4745 /* remove the bitmap */
4748 if (mddev->bitmap->file)
4750 mddev->pers->quiesce(mddev, 1);
4751 bitmap_destroy(mddev);
4752 mddev->pers->quiesce(mddev, 0);
4753 mddev->bitmap_offset = 0;
4756 md_update_sb(mddev, 1);
4760 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4764 if (mddev->pers == NULL)
4767 rdev = find_rdev(mddev, dev);
4771 md_error(mddev, rdev);
4776 * We have a problem here : there is no easy way to give a CHS
4777 * virtual geometry. We currently pretend that we have a 2 heads
4778 * 4 sectors (with a BIG number of cylinders...). This drives
4779 * dosfs just mad... ;-)
4781 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4783 mddev_t *mddev = bdev->bd_disk->private_data;
4787 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4791 static int md_ioctl(struct inode *inode, struct file *file,
4792 unsigned int cmd, unsigned long arg)
4795 void __user *argp = (void __user *)arg;
4796 mddev_t *mddev = NULL;
4798 if (!capable(CAP_SYS_ADMIN))
4802 * Commands dealing with the RAID driver but not any
4808 err = get_version(argp);
4811 case PRINT_RAID_DEBUG:
4819 autostart_arrays(arg);
4826 * Commands creating/starting a new array:
4829 mddev = inode->i_bdev->bd_disk->private_data;
4836 err = mddev_lock(mddev);
4839 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4846 case SET_ARRAY_INFO:
4848 mdu_array_info_t info;
4850 memset(&info, 0, sizeof(info));
4851 else if (copy_from_user(&info, argp, sizeof(info))) {
4856 err = update_array_info(mddev, &info);
4858 printk(KERN_WARNING "md: couldn't update"
4859 " array info. %d\n", err);
4864 if (!list_empty(&mddev->disks)) {
4866 "md: array %s already has disks!\n",
4871 if (mddev->raid_disks) {
4873 "md: array %s already initialised!\n",
4878 err = set_array_info(mddev, &info);
4880 printk(KERN_WARNING "md: couldn't set"
4881 " array info. %d\n", err);
4891 * Commands querying/configuring an existing array:
4893 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4894 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4895 if ((!mddev->raid_disks && !mddev->external)
4896 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4897 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4898 && cmd != GET_BITMAP_FILE) {
4904 * Commands even a read-only array can execute:
4908 case GET_ARRAY_INFO:
4909 err = get_array_info(mddev, argp);
4912 case GET_BITMAP_FILE:
4913 err = get_bitmap_file(mddev, argp);
4917 err = get_disk_info(mddev, argp);
4920 case RESTART_ARRAY_RW:
4921 err = restart_array(mddev);
4925 err = do_md_stop (mddev, 0, 1);
4929 err = do_md_stop (mddev, 1, 1);
4935 * The remaining ioctls are changing the state of the
4936 * superblock, so we do not allow them on read-only arrays.
4937 * However non-MD ioctls (e.g. get-size) will still come through
4938 * here and hit the 'default' below, so only disallow
4939 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4941 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4942 if (mddev->ro == 2) {
4944 sysfs_notify(&mddev->kobj, NULL, "array_state");
4945 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4946 md_wakeup_thread(mddev->thread);
4957 mdu_disk_info_t info;
4958 if (copy_from_user(&info, argp, sizeof(info)))
4961 err = add_new_disk(mddev, &info);
4965 case HOT_REMOVE_DISK:
4966 err = hot_remove_disk(mddev, new_decode_dev(arg));
4970 err = hot_add_disk(mddev, new_decode_dev(arg));
4973 case SET_DISK_FAULTY:
4974 err = set_disk_faulty(mddev, new_decode_dev(arg));
4978 err = do_md_run (mddev);
4981 case SET_BITMAP_FILE:
4982 err = set_bitmap_file(mddev, (int)arg);
4992 mddev_unlock(mddev);
5002 static int md_open(struct inode *inode, struct file *file)
5005 * Succeed if we can lock the mddev, which confirms that
5006 * it isn't being stopped right now.
5008 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5011 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5016 mddev_unlock(mddev);
5018 check_disk_change(inode->i_bdev);
5023 static int md_release(struct inode *inode, struct file * file)
5025 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5033 static int md_media_changed(struct gendisk *disk)
5035 mddev_t *mddev = disk->private_data;
5037 return mddev->changed;
5040 static int md_revalidate(struct gendisk *disk)
5042 mddev_t *mddev = disk->private_data;
5047 static struct block_device_operations md_fops =
5049 .owner = THIS_MODULE,
5051 .release = md_release,
5053 .getgeo = md_getgeo,
5054 .media_changed = md_media_changed,
5055 .revalidate_disk= md_revalidate,
5058 static int md_thread(void * arg)
5060 mdk_thread_t *thread = arg;
5063 * md_thread is a 'system-thread', it's priority should be very
5064 * high. We avoid resource deadlocks individually in each
5065 * raid personality. (RAID5 does preallocation) We also use RR and
5066 * the very same RT priority as kswapd, thus we will never get
5067 * into a priority inversion deadlock.
5069 * we definitely have to have equal or higher priority than
5070 * bdflush, otherwise bdflush will deadlock if there are too
5071 * many dirty RAID5 blocks.
5074 allow_signal(SIGKILL);
5075 while (!kthread_should_stop()) {
5077 /* We need to wait INTERRUPTIBLE so that
5078 * we don't add to the load-average.
5079 * That means we need to be sure no signals are
5082 if (signal_pending(current))
5083 flush_signals(current);
5085 wait_event_interruptible_timeout
5087 test_bit(THREAD_WAKEUP, &thread->flags)
5088 || kthread_should_stop(),
5091 clear_bit(THREAD_WAKEUP, &thread->flags);
5093 thread->run(thread->mddev);
5099 void md_wakeup_thread(mdk_thread_t *thread)
5102 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5103 set_bit(THREAD_WAKEUP, &thread->flags);
5104 wake_up(&thread->wqueue);
5108 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5111 mdk_thread_t *thread;
5113 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5117 init_waitqueue_head(&thread->wqueue);
5120 thread->mddev = mddev;
5121 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5122 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5123 if (IS_ERR(thread->tsk)) {
5130 void md_unregister_thread(mdk_thread_t *thread)
5132 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5134 kthread_stop(thread->tsk);
5138 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5145 if (!rdev || test_bit(Faulty, &rdev->flags))
5148 if (mddev->external)
5149 set_bit(Blocked, &rdev->flags);
5151 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5153 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5154 __builtin_return_address(0),__builtin_return_address(1),
5155 __builtin_return_address(2),__builtin_return_address(3));
5159 if (!mddev->pers->error_handler)
5161 mddev->pers->error_handler(mddev,rdev);
5162 if (mddev->degraded)
5163 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5164 set_bit(StateChanged, &rdev->flags);
5165 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5166 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5167 md_wakeup_thread(mddev->thread);
5168 md_new_event_inintr(mddev);
5171 /* seq_file implementation /proc/mdstat */
5173 static void status_unused(struct seq_file *seq)
5177 struct list_head *tmp;
5179 seq_printf(seq, "unused devices: ");
5181 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5182 char b[BDEVNAME_SIZE];
5184 seq_printf(seq, "%s ",
5185 bdevname(rdev->bdev,b));
5188 seq_printf(seq, "<none>");
5190 seq_printf(seq, "\n");
5194 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5196 sector_t max_blocks, resync, res;
5197 unsigned long dt, db, rt;
5199 unsigned int per_milli;
5201 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5203 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5204 max_blocks = mddev->resync_max_sectors >> 1;
5206 max_blocks = mddev->size;
5209 * Should not happen.
5215 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5216 * in a sector_t, and (max_blocks>>scale) will fit in a
5217 * u32, as those are the requirements for sector_div.
5218 * Thus 'scale' must be at least 10
5221 if (sizeof(sector_t) > sizeof(unsigned long)) {
5222 while ( max_blocks/2 > (1ULL<<(scale+32)))
5225 res = (resync>>scale)*1000;
5226 sector_div(res, (u32)((max_blocks>>scale)+1));
5230 int i, x = per_milli/50, y = 20-x;
5231 seq_printf(seq, "[");
5232 for (i = 0; i < x; i++)
5233 seq_printf(seq, "=");
5234 seq_printf(seq, ">");
5235 for (i = 0; i < y; i++)
5236 seq_printf(seq, ".");
5237 seq_printf(seq, "] ");
5239 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5240 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5242 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5244 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5245 "resync" : "recovery"))),
5246 per_milli/10, per_milli % 10,
5247 (unsigned long long) resync,
5248 (unsigned long long) max_blocks);
5251 * We do not want to overflow, so the order of operands and
5252 * the * 100 / 100 trick are important. We do a +1 to be
5253 * safe against division by zero. We only estimate anyway.
5255 * dt: time from mark until now
5256 * db: blocks written from mark until now
5257 * rt: remaining time
5259 dt = ((jiffies - mddev->resync_mark) / HZ);
5261 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5262 - mddev->resync_mark_cnt;
5263 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5265 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5267 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5270 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5272 struct list_head *tmp;
5282 spin_lock(&all_mddevs_lock);
5283 list_for_each(tmp,&all_mddevs)
5285 mddev = list_entry(tmp, mddev_t, all_mddevs);
5287 spin_unlock(&all_mddevs_lock);
5290 spin_unlock(&all_mddevs_lock);
5292 return (void*)2;/* tail */
5296 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5298 struct list_head *tmp;
5299 mddev_t *next_mddev, *mddev = v;
5305 spin_lock(&all_mddevs_lock);
5307 tmp = all_mddevs.next;
5309 tmp = mddev->all_mddevs.next;
5310 if (tmp != &all_mddevs)
5311 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5313 next_mddev = (void*)2;
5316 spin_unlock(&all_mddevs_lock);
5324 static void md_seq_stop(struct seq_file *seq, void *v)
5328 if (mddev && v != (void*)1 && v != (void*)2)
5332 struct mdstat_info {
5336 static int md_seq_show(struct seq_file *seq, void *v)
5340 struct list_head *tmp2;
5342 struct mdstat_info *mi = seq->private;
5343 struct bitmap *bitmap;
5345 if (v == (void*)1) {
5346 struct mdk_personality *pers;
5347 seq_printf(seq, "Personalities : ");
5348 spin_lock(&pers_lock);
5349 list_for_each_entry(pers, &pers_list, list)
5350 seq_printf(seq, "[%s] ", pers->name);
5352 spin_unlock(&pers_lock);
5353 seq_printf(seq, "\n");
5354 mi->event = atomic_read(&md_event_count);
5357 if (v == (void*)2) {
5362 if (mddev_lock(mddev) < 0)
5365 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5366 seq_printf(seq, "%s : %sactive", mdname(mddev),
5367 mddev->pers ? "" : "in");
5370 seq_printf(seq, " (read-only)");
5372 seq_printf(seq, " (auto-read-only)");
5373 seq_printf(seq, " %s", mddev->pers->name);
5377 rdev_for_each(rdev, tmp2, mddev) {
5378 char b[BDEVNAME_SIZE];
5379 seq_printf(seq, " %s[%d]",
5380 bdevname(rdev->bdev,b), rdev->desc_nr);
5381 if (test_bit(WriteMostly, &rdev->flags))
5382 seq_printf(seq, "(W)");
5383 if (test_bit(Faulty, &rdev->flags)) {
5384 seq_printf(seq, "(F)");
5386 } else if (rdev->raid_disk < 0)
5387 seq_printf(seq, "(S)"); /* spare */
5391 if (!list_empty(&mddev->disks)) {
5393 seq_printf(seq, "\n %llu blocks",
5394 (unsigned long long)mddev->array_size);
5396 seq_printf(seq, "\n %llu blocks",
5397 (unsigned long long)size);
5399 if (mddev->persistent) {
5400 if (mddev->major_version != 0 ||
5401 mddev->minor_version != 90) {
5402 seq_printf(seq," super %d.%d",
5403 mddev->major_version,
5404 mddev->minor_version);
5406 } else if (mddev->external)
5407 seq_printf(seq, " super external:%s",
5408 mddev->metadata_type);
5410 seq_printf(seq, " super non-persistent");
5413 mddev->pers->status (seq, mddev);
5414 seq_printf(seq, "\n ");
5415 if (mddev->pers->sync_request) {
5416 if (mddev->curr_resync > 2) {
5417 status_resync (seq, mddev);
5418 seq_printf(seq, "\n ");
5419 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5420 seq_printf(seq, "\tresync=DELAYED\n ");
5421 else if (mddev->recovery_cp < MaxSector)
5422 seq_printf(seq, "\tresync=PENDING\n ");
5425 seq_printf(seq, "\n ");
5427 if ((bitmap = mddev->bitmap)) {
5428 unsigned long chunk_kb;
5429 unsigned long flags;
5430 spin_lock_irqsave(&bitmap->lock, flags);
5431 chunk_kb = bitmap->chunksize >> 10;
5432 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5434 bitmap->pages - bitmap->missing_pages,
5436 (bitmap->pages - bitmap->missing_pages)
5437 << (PAGE_SHIFT - 10),
5438 chunk_kb ? chunk_kb : bitmap->chunksize,
5439 chunk_kb ? "KB" : "B");
5441 seq_printf(seq, ", file: ");
5442 seq_path(seq, &bitmap->file->f_path, " \t\n");
5445 seq_printf(seq, "\n");
5446 spin_unlock_irqrestore(&bitmap->lock, flags);
5449 seq_printf(seq, "\n");
5451 mddev_unlock(mddev);
5456 static struct seq_operations md_seq_ops = {
5457 .start = md_seq_start,
5458 .next = md_seq_next,
5459 .stop = md_seq_stop,
5460 .show = md_seq_show,
5463 static int md_seq_open(struct inode *inode, struct file *file)
5466 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5470 error = seq_open(file, &md_seq_ops);
5474 struct seq_file *p = file->private_data;
5476 mi->event = atomic_read(&md_event_count);
5481 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5483 struct seq_file *m = filp->private_data;
5484 struct mdstat_info *mi = m->private;
5487 poll_wait(filp, &md_event_waiters, wait);
5489 /* always allow read */
5490 mask = POLLIN | POLLRDNORM;
5492 if (mi->event != atomic_read(&md_event_count))
5493 mask |= POLLERR | POLLPRI;
5497 static const struct file_operations md_seq_fops = {
5498 .owner = THIS_MODULE,
5499 .open = md_seq_open,
5501 .llseek = seq_lseek,
5502 .release = seq_release_private,
5503 .poll = mdstat_poll,
5506 int register_md_personality(struct mdk_personality *p)
5508 spin_lock(&pers_lock);
5509 list_add_tail(&p->list, &pers_list);
5510 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5511 spin_unlock(&pers_lock);
5515 int unregister_md_personality(struct mdk_personality *p)
5517 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5518 spin_lock(&pers_lock);
5519 list_del_init(&p->list);
5520 spin_unlock(&pers_lock);
5524 static int is_mddev_idle(mddev_t *mddev)
5527 struct list_head *tmp;
5532 rdev_for_each(rdev, tmp, mddev) {
5533 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5534 curr_events = disk_stat_read(disk, sectors[0]) +
5535 disk_stat_read(disk, sectors[1]) -
5536 atomic_read(&disk->sync_io);
5537 /* sync IO will cause sync_io to increase before the disk_stats
5538 * as sync_io is counted when a request starts, and
5539 * disk_stats is counted when it completes.
5540 * So resync activity will cause curr_events to be smaller than
5541 * when there was no such activity.
5542 * non-sync IO will cause disk_stat to increase without
5543 * increasing sync_io so curr_events will (eventually)
5544 * be larger than it was before. Once it becomes
5545 * substantially larger, the test below will cause
5546 * the array to appear non-idle, and resync will slow
5548 * If there is a lot of outstanding resync activity when
5549 * we set last_event to curr_events, then all that activity
5550 * completing might cause the array to appear non-idle
5551 * and resync will be slowed down even though there might
5552 * not have been non-resync activity. This will only
5553 * happen once though. 'last_events' will soon reflect
5554 * the state where there is little or no outstanding
5555 * resync requests, and further resync activity will
5556 * always make curr_events less than last_events.
5559 if (curr_events - rdev->last_events > 4096) {
5560 rdev->last_events = curr_events;
5567 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5569 /* another "blocks" (512byte) blocks have been synced */
5570 atomic_sub(blocks, &mddev->recovery_active);
5571 wake_up(&mddev->recovery_wait);
5573 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5574 md_wakeup_thread(mddev->thread);
5575 // stop recovery, signal do_sync ....
5580 /* md_write_start(mddev, bi)
5581 * If we need to update some array metadata (e.g. 'active' flag
5582 * in superblock) before writing, schedule a superblock update
5583 * and wait for it to complete.
5585 void md_write_start(mddev_t *mddev, struct bio *bi)
5588 if (bio_data_dir(bi) != WRITE)
5591 BUG_ON(mddev->ro == 1);
5592 if (mddev->ro == 2) {
5593 /* need to switch to read/write */
5595 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5596 md_wakeup_thread(mddev->thread);
5597 md_wakeup_thread(mddev->sync_thread);
5600 atomic_inc(&mddev->writes_pending);
5601 if (mddev->safemode == 1)
5602 mddev->safemode = 0;
5603 if (mddev->in_sync) {
5604 spin_lock_irq(&mddev->write_lock);
5605 if (mddev->in_sync) {
5607 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5608 md_wakeup_thread(mddev->thread);
5611 spin_unlock_irq(&mddev->write_lock);
5614 sysfs_notify(&mddev->kobj, NULL, "array_state");
5615 wait_event(mddev->sb_wait,
5616 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5617 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5620 void md_write_end(mddev_t *mddev)
5622 if (atomic_dec_and_test(&mddev->writes_pending)) {
5623 if (mddev->safemode == 2)
5624 md_wakeup_thread(mddev->thread);
5625 else if (mddev->safemode_delay)
5626 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5630 /* md_allow_write(mddev)
5631 * Calling this ensures that the array is marked 'active' so that writes
5632 * may proceed without blocking. It is important to call this before
5633 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5634 * Must be called with mddev_lock held.
5636 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5637 * is dropped, so return -EAGAIN after notifying userspace.
5639 int md_allow_write(mddev_t *mddev)
5645 if (!mddev->pers->sync_request)
5648 spin_lock_irq(&mddev->write_lock);
5649 if (mddev->in_sync) {
5651 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5652 if (mddev->safemode_delay &&
5653 mddev->safemode == 0)
5654 mddev->safemode = 1;
5655 spin_unlock_irq(&mddev->write_lock);
5656 md_update_sb(mddev, 0);
5657 sysfs_notify(&mddev->kobj, NULL, "array_state");
5659 spin_unlock_irq(&mddev->write_lock);
5661 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5666 EXPORT_SYMBOL_GPL(md_allow_write);
5668 #define SYNC_MARKS 10
5669 #define SYNC_MARK_STEP (3*HZ)
5670 void md_do_sync(mddev_t *mddev)
5673 unsigned int currspeed = 0,
5675 sector_t max_sectors,j, io_sectors;
5676 unsigned long mark[SYNC_MARKS];
5677 sector_t mark_cnt[SYNC_MARKS];
5679 struct list_head *tmp;
5680 sector_t last_check;
5682 struct list_head *rtmp;
5686 /* just incase thread restarts... */
5687 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5689 if (mddev->ro) /* never try to sync a read-only array */
5692 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5693 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5694 desc = "data-check";
5695 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5696 desc = "requested-resync";
5699 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5704 /* we overload curr_resync somewhat here.
5705 * 0 == not engaged in resync at all
5706 * 2 == checking that there is no conflict with another sync
5707 * 1 == like 2, but have yielded to allow conflicting resync to
5709 * other == active in resync - this many blocks
5711 * Before starting a resync we must have set curr_resync to
5712 * 2, and then checked that every "conflicting" array has curr_resync
5713 * less than ours. When we find one that is the same or higher
5714 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5715 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5716 * This will mean we have to start checking from the beginning again.
5721 mddev->curr_resync = 2;
5724 if (kthread_should_stop()) {
5725 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5728 for_each_mddev(mddev2, tmp) {
5729 if (mddev2 == mddev)
5731 if (!mddev->parallel_resync
5732 && mddev2->curr_resync
5733 && match_mddev_units(mddev, mddev2)) {
5735 if (mddev < mddev2 && mddev->curr_resync == 2) {
5736 /* arbitrarily yield */
5737 mddev->curr_resync = 1;
5738 wake_up(&resync_wait);
5740 if (mddev > mddev2 && mddev->curr_resync == 1)
5741 /* no need to wait here, we can wait the next
5742 * time 'round when curr_resync == 2
5745 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5746 if (!kthread_should_stop() &&
5747 mddev2->curr_resync >= mddev->curr_resync) {
5748 printk(KERN_INFO "md: delaying %s of %s"
5749 " until %s has finished (they"
5750 " share one or more physical units)\n",
5751 desc, mdname(mddev), mdname(mddev2));
5754 finish_wait(&resync_wait, &wq);
5757 finish_wait(&resync_wait, &wq);
5760 } while (mddev->curr_resync < 2);
5763 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5764 /* resync follows the size requested by the personality,
5765 * which defaults to physical size, but can be virtual size
5767 max_sectors = mddev->resync_max_sectors;
5768 mddev->resync_mismatches = 0;
5769 /* we don't use the checkpoint if there's a bitmap */
5770 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5771 j = mddev->resync_min;
5772 else if (!mddev->bitmap)
5773 j = mddev->recovery_cp;
5775 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5776 max_sectors = mddev->size << 1;
5778 /* recovery follows the physical size of devices */
5779 max_sectors = mddev->size << 1;
5781 rdev_for_each(rdev, rtmp, mddev)
5782 if (rdev->raid_disk >= 0 &&
5783 !test_bit(Faulty, &rdev->flags) &&
5784 !test_bit(In_sync, &rdev->flags) &&
5785 rdev->recovery_offset < j)
5786 j = rdev->recovery_offset;
5789 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5790 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5791 " %d KB/sec/disk.\n", speed_min(mddev));
5792 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5793 "(but not more than %d KB/sec) for %s.\n",
5794 speed_max(mddev), desc);
5796 is_mddev_idle(mddev); /* this also initializes IO event counters */
5799 for (m = 0; m < SYNC_MARKS; m++) {
5801 mark_cnt[m] = io_sectors;
5804 mddev->resync_mark = mark[last_mark];
5805 mddev->resync_mark_cnt = mark_cnt[last_mark];
5808 * Tune reconstruction:
5810 window = 32*(PAGE_SIZE/512);
5811 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5812 window/2,(unsigned long long) max_sectors/2);
5814 atomic_set(&mddev->recovery_active, 0);
5819 "md: resuming %s of %s from checkpoint.\n",
5820 desc, mdname(mddev));
5821 mddev->curr_resync = j;
5824 while (j < max_sectors) {
5828 if (j >= mddev->resync_max) {
5829 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5830 wait_event(mddev->recovery_wait,
5831 mddev->resync_max > j
5832 || kthread_should_stop());
5834 if (kthread_should_stop())
5836 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5837 currspeed < speed_min(mddev));
5839 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5843 if (!skipped) { /* actual IO requested */
5844 io_sectors += sectors;
5845 atomic_add(sectors, &mddev->recovery_active);
5849 if (j>1) mddev->curr_resync = j;
5850 mddev->curr_mark_cnt = io_sectors;
5851 if (last_check == 0)
5852 /* this is the earliers that rebuilt will be
5853 * visible in /proc/mdstat
5855 md_new_event(mddev);
5857 if (last_check + window > io_sectors || j == max_sectors)
5860 last_check = io_sectors;
5862 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5866 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5868 int next = (last_mark+1) % SYNC_MARKS;
5870 mddev->resync_mark = mark[next];
5871 mddev->resync_mark_cnt = mark_cnt[next];
5872 mark[next] = jiffies;
5873 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5878 if (kthread_should_stop())
5883 * this loop exits only if either when we are slower than
5884 * the 'hard' speed limit, or the system was IO-idle for
5886 * the system might be non-idle CPU-wise, but we only care
5887 * about not overloading the IO subsystem. (things like an
5888 * e2fsck being done on the RAID array should execute fast)
5890 blk_unplug(mddev->queue);
5893 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5894 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5896 if (currspeed > speed_min(mddev)) {
5897 if ((currspeed > speed_max(mddev)) ||
5898 !is_mddev_idle(mddev)) {
5904 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5906 * this also signals 'finished resyncing' to md_stop
5909 blk_unplug(mddev->queue);
5911 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5913 /* tell personality that we are finished */
5914 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5916 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5917 mddev->curr_resync > 2) {
5918 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5919 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5920 if (mddev->curr_resync >= mddev->recovery_cp) {
5922 "md: checkpointing %s of %s.\n",
5923 desc, mdname(mddev));
5924 mddev->recovery_cp = mddev->curr_resync;
5927 mddev->recovery_cp = MaxSector;
5929 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5930 mddev->curr_resync = MaxSector;
5931 rdev_for_each(rdev, rtmp, mddev)
5932 if (rdev->raid_disk >= 0 &&
5933 !test_bit(Faulty, &rdev->flags) &&
5934 !test_bit(In_sync, &rdev->flags) &&
5935 rdev->recovery_offset < mddev->curr_resync)
5936 rdev->recovery_offset = mddev->curr_resync;
5939 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5942 mddev->curr_resync = 0;
5943 mddev->resync_min = 0;
5944 mddev->resync_max = MaxSector;
5945 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5946 wake_up(&resync_wait);
5947 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5948 md_wakeup_thread(mddev->thread);
5953 * got a signal, exit.
5956 "md: md_do_sync() got signal ... exiting\n");
5957 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5961 EXPORT_SYMBOL_GPL(md_do_sync);
5964 static int remove_and_add_spares(mddev_t *mddev)
5967 struct list_head *rtmp;
5970 rdev_for_each(rdev, rtmp, mddev)
5971 if (rdev->raid_disk >= 0 &&
5972 !test_bit(Blocked, &rdev->flags) &&
5973 (test_bit(Faulty, &rdev->flags) ||
5974 ! test_bit(In_sync, &rdev->flags)) &&
5975 atomic_read(&rdev->nr_pending)==0) {
5976 if (mddev->pers->hot_remove_disk(
5977 mddev, rdev->raid_disk)==0) {
5979 sprintf(nm,"rd%d", rdev->raid_disk);
5980 sysfs_remove_link(&mddev->kobj, nm);
5981 rdev->raid_disk = -1;
5985 if (mddev->degraded) {
5986 rdev_for_each(rdev, rtmp, mddev) {
5987 if (rdev->raid_disk >= 0 &&
5988 !test_bit(In_sync, &rdev->flags))
5990 if (rdev->raid_disk < 0
5991 && !test_bit(Faulty, &rdev->flags)) {
5992 rdev->recovery_offset = 0;
5994 hot_add_disk(mddev, rdev) == 0) {
5996 sprintf(nm, "rd%d", rdev->raid_disk);
5997 if (sysfs_create_link(&mddev->kobj,
6000 "md: cannot register "
6004 md_new_event(mddev);
6013 * This routine is regularly called by all per-raid-array threads to
6014 * deal with generic issues like resync and super-block update.
6015 * Raid personalities that don't have a thread (linear/raid0) do not
6016 * need this as they never do any recovery or update the superblock.
6018 * It does not do any resync itself, but rather "forks" off other threads
6019 * to do that as needed.
6020 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6021 * "->recovery" and create a thread at ->sync_thread.
6022 * When the thread finishes it sets MD_RECOVERY_DONE
6023 * and wakeups up this thread which will reap the thread and finish up.
6024 * This thread also removes any faulty devices (with nr_pending == 0).
6026 * The overall approach is:
6027 * 1/ if the superblock needs updating, update it.
6028 * 2/ If a recovery thread is running, don't do anything else.
6029 * 3/ If recovery has finished, clean up, possibly marking spares active.
6030 * 4/ If there are any faulty devices, remove them.
6031 * 5/ If array is degraded, try to add spares devices
6032 * 6/ If array has spares or is not in-sync, start a resync thread.
6034 void md_check_recovery(mddev_t *mddev)
6037 struct list_head *rtmp;
6041 bitmap_daemon_work(mddev->bitmap);
6046 if (signal_pending(current)) {
6047 if (mddev->pers->sync_request && !mddev->external) {
6048 printk(KERN_INFO "md: %s in immediate safe mode\n",
6050 mddev->safemode = 2;
6052 flush_signals(current);
6056 (mddev->flags && !mddev->external) ||
6057 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6058 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6059 (mddev->external == 0 && mddev->safemode == 1) ||
6060 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6061 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6065 if (mddev_trylock(mddev)) {
6068 if (!mddev->external) {
6070 spin_lock_irq(&mddev->write_lock);
6071 if (mddev->safemode &&
6072 !atomic_read(&mddev->writes_pending) &&
6074 mddev->recovery_cp == MaxSector) {
6077 if (mddev->persistent)
6078 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6080 if (mddev->safemode == 1)
6081 mddev->safemode = 0;
6082 spin_unlock_irq(&mddev->write_lock);
6084 sysfs_notify(&mddev->kobj, NULL, "array_state");
6088 md_update_sb(mddev, 0);
6090 rdev_for_each(rdev, rtmp, mddev)
6091 if (test_and_clear_bit(StateChanged, &rdev->flags))
6092 sysfs_notify(&rdev->kobj, NULL, "state");
6095 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6096 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6097 /* resync/recovery still happening */
6098 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6101 if (mddev->sync_thread) {
6102 /* resync has finished, collect result */
6103 md_unregister_thread(mddev->sync_thread);
6104 mddev->sync_thread = NULL;
6105 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6107 /* activate any spares */
6108 if (mddev->pers->spare_active(mddev))
6109 sysfs_notify(&mddev->kobj, NULL,
6112 md_update_sb(mddev, 1);
6114 /* if array is no-longer degraded, then any saved_raid_disk
6115 * information must be scrapped
6117 if (!mddev->degraded)
6118 rdev_for_each(rdev, rtmp, mddev)
6119 rdev->saved_raid_disk = -1;
6121 mddev->recovery = 0;
6122 /* flag recovery needed just to double check */
6123 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6124 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6125 md_new_event(mddev);
6128 /* Set RUNNING before clearing NEEDED to avoid
6129 * any transients in the value of "sync_action".
6131 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6132 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6133 /* Clear some bits that don't mean anything, but
6136 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6137 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6139 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6141 /* no recovery is running.
6142 * remove any failed drives, then
6143 * add spares if possible.
6144 * Spare are also removed and re-added, to allow
6145 * the personality to fail the re-add.
6148 if (mddev->reshape_position != MaxSector) {
6149 if (mddev->pers->check_reshape(mddev) != 0)
6150 /* Cannot proceed */
6152 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6153 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6154 } else if ((spares = remove_and_add_spares(mddev))) {
6155 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6156 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6157 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6158 } else if (mddev->recovery_cp < MaxSector) {
6159 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6160 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6161 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6162 /* nothing to be done ... */
6165 if (mddev->pers->sync_request) {
6166 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6167 /* We are adding a device or devices to an array
6168 * which has the bitmap stored on all devices.
6169 * So make sure all bitmap pages get written
6171 bitmap_write_all(mddev->bitmap);
6173 mddev->sync_thread = md_register_thread(md_do_sync,
6176 if (!mddev->sync_thread) {
6177 printk(KERN_ERR "%s: could not start resync"
6180 /* leave the spares where they are, it shouldn't hurt */
6181 mddev->recovery = 0;
6183 md_wakeup_thread(mddev->sync_thread);
6184 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6185 md_new_event(mddev);
6188 if (!mddev->sync_thread) {
6189 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6190 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6192 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6194 mddev_unlock(mddev);
6198 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6200 sysfs_notify(&rdev->kobj, NULL, "state");
6201 wait_event_timeout(rdev->blocked_wait,
6202 !test_bit(Blocked, &rdev->flags),
6203 msecs_to_jiffies(5000));
6204 rdev_dec_pending(rdev, mddev);
6206 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6208 static int md_notify_reboot(struct notifier_block *this,
6209 unsigned long code, void *x)
6211 struct list_head *tmp;
6214 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6216 printk(KERN_INFO "md: stopping all md devices.\n");
6218 for_each_mddev(mddev, tmp)
6219 if (mddev_trylock(mddev)) {
6220 do_md_stop (mddev, 1, 0);
6221 mddev_unlock(mddev);
6224 * certain more exotic SCSI devices are known to be
6225 * volatile wrt too early system reboots. While the
6226 * right place to handle this issue is the given
6227 * driver, we do want to have a safe RAID driver ...
6234 static struct notifier_block md_notifier = {
6235 .notifier_call = md_notify_reboot,
6237 .priority = INT_MAX, /* before any real devices */
6240 static void md_geninit(void)
6242 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6244 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6247 static int __init md_init(void)
6249 if (register_blkdev(MAJOR_NR, "md"))
6251 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6252 unregister_blkdev(MAJOR_NR, "md");
6255 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6256 md_probe, NULL, NULL);
6257 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6258 md_probe, NULL, NULL);
6260 register_reboot_notifier(&md_notifier);
6261 raid_table_header = register_sysctl_table(raid_root_table);
6271 * Searches all registered partitions for autorun RAID arrays
6275 static LIST_HEAD(all_detected_devices);
6276 struct detected_devices_node {
6277 struct list_head list;
6281 void md_autodetect_dev(dev_t dev)
6283 struct detected_devices_node *node_detected_dev;
6285 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6286 if (node_detected_dev) {
6287 node_detected_dev->dev = dev;
6288 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6290 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6291 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6296 static void autostart_arrays(int part)
6299 struct detected_devices_node *node_detected_dev;
6301 int i_scanned, i_passed;
6306 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6308 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6310 node_detected_dev = list_entry(all_detected_devices.next,
6311 struct detected_devices_node, list);
6312 list_del(&node_detected_dev->list);
6313 dev = node_detected_dev->dev;
6314 kfree(node_detected_dev);
6315 rdev = md_import_device(dev,0, 90);
6319 if (test_bit(Faulty, &rdev->flags)) {
6323 set_bit(AutoDetected, &rdev->flags);
6324 list_add(&rdev->same_set, &pending_raid_disks);
6328 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6329 i_scanned, i_passed);
6331 autorun_devices(part);
6334 #endif /* !MODULE */
6336 static __exit void md_exit(void)
6339 struct list_head *tmp;
6341 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6342 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6344 unregister_blkdev(MAJOR_NR,"md");
6345 unregister_blkdev(mdp_major, "mdp");
6346 unregister_reboot_notifier(&md_notifier);
6347 unregister_sysctl_table(raid_table_header);
6348 remove_proc_entry("mdstat", NULL);
6349 for_each_mddev(mddev, tmp) {
6350 struct gendisk *disk = mddev->gendisk;
6353 export_array(mddev);
6356 mddev->gendisk = NULL;
6361 subsys_initcall(md_init);
6362 module_exit(md_exit)
6364 static int get_ro(char *buffer, struct kernel_param *kp)
6366 return sprintf(buffer, "%d", start_readonly);
6368 static int set_ro(const char *val, struct kernel_param *kp)
6371 int num = simple_strtoul(val, &e, 10);
6372 if (*val && (*e == '\0' || *e == '\n')) {
6373 start_readonly = num;
6379 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6380 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6383 EXPORT_SYMBOL(register_md_personality);
6384 EXPORT_SYMBOL(unregister_md_personality);
6385 EXPORT_SYMBOL(md_error);
6386 EXPORT_SYMBOL(md_done_sync);
6387 EXPORT_SYMBOL(md_write_start);
6388 EXPORT_SYMBOL(md_write_end);
6389 EXPORT_SYMBOL(md_register_thread);
6390 EXPORT_SYMBOL(md_unregister_thread);
6391 EXPORT_SYMBOL(md_wakeup_thread);
6392 EXPORT_SYMBOL(md_check_recovery);
6393 MODULE_LICENSE("GPL");
6395 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);