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 unsigned long long size);
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, unsigned long long size)
1003 if (size && size < rdev->mddev->size)
1004 return 0; /* component must fit device */
1005 size *= 2; /* convert to sectors */
1006 if (rdev->mddev->bitmap_offset)
1007 return 0; /* can't move bitmap */
1008 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1009 if (!size || size > rdev->sb_start)
1010 size = rdev->sb_start;
1011 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1013 md_super_wait(rdev->mddev);
1014 return size/2; /* kB for sysfs */
1019 * version 1 superblock
1022 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1026 unsigned long long newcsum;
1027 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1028 __le32 *isuper = (__le32*)sb;
1031 disk_csum = sb->sb_csum;
1034 for (i=0; size>=4; size -= 4 )
1035 newcsum += le32_to_cpu(*isuper++);
1038 newcsum += le16_to_cpu(*(__le16*) isuper);
1040 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1041 sb->sb_csum = disk_csum;
1042 return cpu_to_le32(csum);
1045 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1047 struct mdp_superblock_1 *sb;
1050 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1054 * Calculate the position of the superblock in 512byte sectors.
1055 * It is always aligned to a 4K boundary and
1056 * depeding on minor_version, it can be:
1057 * 0: At least 8K, but less than 12K, from end of device
1058 * 1: At start of device
1059 * 2: 4K from start of device.
1061 switch(minor_version) {
1063 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1065 sb_start &= ~(sector_t)(4*2-1);
1076 rdev->sb_start = sb_start;
1078 /* superblock is rarely larger than 1K, but it can be larger,
1079 * and it is safe to read 4k, so we do that
1081 ret = read_disk_sb(rdev, 4096);
1082 if (ret) return ret;
1085 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1087 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1088 sb->major_version != cpu_to_le32(1) ||
1089 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1090 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1091 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1094 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1095 printk("md: invalid superblock checksum on %s\n",
1096 bdevname(rdev->bdev,b));
1099 if (le64_to_cpu(sb->data_size) < 10) {
1100 printk("md: data_size too small on %s\n",
1101 bdevname(rdev->bdev,b));
1104 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1105 if (sb->level != cpu_to_le32(1) &&
1106 sb->level != cpu_to_le32(4) &&
1107 sb->level != cpu_to_le32(5) &&
1108 sb->level != cpu_to_le32(6) &&
1109 sb->level != cpu_to_le32(10)) {
1111 "md: bitmaps not supported for this level.\n");
1116 rdev->preferred_minor = 0xffff;
1117 rdev->data_offset = le64_to_cpu(sb->data_offset);
1118 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1120 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1121 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1122 if (rdev->sb_size & bmask)
1123 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1126 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1129 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1132 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1138 struct mdp_superblock_1 *refsb =
1139 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1141 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1142 sb->level != refsb->level ||
1143 sb->layout != refsb->layout ||
1144 sb->chunksize != refsb->chunksize) {
1145 printk(KERN_WARNING "md: %s has strangely different"
1146 " superblock to %s\n",
1147 bdevname(rdev->bdev,b),
1148 bdevname(refdev->bdev,b2));
1151 ev1 = le64_to_cpu(sb->events);
1152 ev2 = le64_to_cpu(refsb->events);
1160 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1162 rdev->size = rdev->sb_start / 2;
1163 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1165 rdev->size = le64_to_cpu(sb->data_size)/2;
1166 if (le32_to_cpu(sb->chunksize))
1167 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1169 if (le64_to_cpu(sb->size) > rdev->size*2)
1174 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1176 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1177 __u64 ev1 = le64_to_cpu(sb->events);
1179 rdev->raid_disk = -1;
1180 clear_bit(Faulty, &rdev->flags);
1181 clear_bit(In_sync, &rdev->flags);
1182 clear_bit(WriteMostly, &rdev->flags);
1183 clear_bit(BarriersNotsupp, &rdev->flags);
1185 if (mddev->raid_disks == 0) {
1186 mddev->major_version = 1;
1187 mddev->patch_version = 0;
1188 mddev->external = 0;
1189 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1190 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1191 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1192 mddev->level = le32_to_cpu(sb->level);
1193 mddev->clevel[0] = 0;
1194 mddev->layout = le32_to_cpu(sb->layout);
1195 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1196 mddev->size = le64_to_cpu(sb->size)/2;
1197 mddev->events = ev1;
1198 mddev->bitmap_offset = 0;
1199 mddev->default_bitmap_offset = 1024 >> 9;
1201 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1202 memcpy(mddev->uuid, sb->set_uuid, 16);
1204 mddev->max_disks = (4096-256)/2;
1206 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1207 mddev->bitmap_file == NULL )
1208 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1210 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1211 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1212 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1213 mddev->new_level = le32_to_cpu(sb->new_level);
1214 mddev->new_layout = le32_to_cpu(sb->new_layout);
1215 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1217 mddev->reshape_position = MaxSector;
1218 mddev->delta_disks = 0;
1219 mddev->new_level = mddev->level;
1220 mddev->new_layout = mddev->layout;
1221 mddev->new_chunk = mddev->chunk_size;
1224 } else if (mddev->pers == NULL) {
1225 /* Insist of good event counter while assembling */
1227 if (ev1 < mddev->events)
1229 } else if (mddev->bitmap) {
1230 /* If adding to array with a bitmap, then we can accept an
1231 * older device, but not too old.
1233 if (ev1 < mddev->bitmap->events_cleared)
1236 if (ev1 < mddev->events)
1237 /* just a hot-add of a new device, leave raid_disk at -1 */
1240 if (mddev->level != LEVEL_MULTIPATH) {
1242 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1244 case 0xffff: /* spare */
1246 case 0xfffe: /* faulty */
1247 set_bit(Faulty, &rdev->flags);
1250 if ((le32_to_cpu(sb->feature_map) &
1251 MD_FEATURE_RECOVERY_OFFSET))
1252 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1254 set_bit(In_sync, &rdev->flags);
1255 rdev->raid_disk = role;
1258 if (sb->devflags & WriteMostly1)
1259 set_bit(WriteMostly, &rdev->flags);
1260 } else /* MULTIPATH are always insync */
1261 set_bit(In_sync, &rdev->flags);
1266 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1268 struct mdp_superblock_1 *sb;
1269 struct list_head *tmp;
1272 /* make rdev->sb match mddev and rdev data. */
1274 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1276 sb->feature_map = 0;
1278 sb->recovery_offset = cpu_to_le64(0);
1279 memset(sb->pad1, 0, sizeof(sb->pad1));
1280 memset(sb->pad2, 0, sizeof(sb->pad2));
1281 memset(sb->pad3, 0, sizeof(sb->pad3));
1283 sb->utime = cpu_to_le64((__u64)mddev->utime);
1284 sb->events = cpu_to_le64(mddev->events);
1286 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1288 sb->resync_offset = cpu_to_le64(0);
1290 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1292 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1293 sb->size = cpu_to_le64(mddev->size<<1);
1295 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1296 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1297 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1300 if (rdev->raid_disk >= 0 &&
1301 !test_bit(In_sync, &rdev->flags) &&
1302 rdev->recovery_offset > 0) {
1303 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1304 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1307 if (mddev->reshape_position != MaxSector) {
1308 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1309 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1310 sb->new_layout = cpu_to_le32(mddev->new_layout);
1311 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1312 sb->new_level = cpu_to_le32(mddev->new_level);
1313 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1317 rdev_for_each(rdev2, tmp, mddev)
1318 if (rdev2->desc_nr+1 > max_dev)
1319 max_dev = rdev2->desc_nr+1;
1321 if (max_dev > le32_to_cpu(sb->max_dev))
1322 sb->max_dev = cpu_to_le32(max_dev);
1323 for (i=0; i<max_dev;i++)
1324 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1326 rdev_for_each(rdev2, tmp, mddev) {
1328 if (test_bit(Faulty, &rdev2->flags))
1329 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1330 else if (test_bit(In_sync, &rdev2->flags))
1331 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1332 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1333 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1335 sb->dev_roles[i] = cpu_to_le16(0xffff);
1338 sb->sb_csum = calc_sb_1_csum(sb);
1341 static unsigned long long
1342 super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1344 struct mdp_superblock_1 *sb;
1345 unsigned long long max_size;
1346 if (size && size < rdev->mddev->size)
1347 return 0; /* component must fit device */
1348 size *= 2; /* convert to sectors */
1349 if (rdev->sb_start < rdev->data_offset) {
1350 /* minor versions 1 and 2; superblock before data */
1351 max_size = (rdev->bdev->bd_inode->i_size >> 9);
1352 max_size -= rdev->data_offset;
1353 if (!size || size > max_size)
1355 } else if (rdev->mddev->bitmap_offset) {
1356 /* minor version 0 with bitmap we can't move */
1359 /* minor version 0; superblock after data */
1361 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1362 sb_start &= ~(sector_t)(4*2 - 1);
1363 max_size = rdev->size*2 + sb_start - rdev->sb_start;
1364 if (!size || size > max_size)
1366 rdev->sb_start = sb_start;
1368 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1369 sb->data_size = cpu_to_le64(size);
1370 sb->super_offset = rdev->sb_start;
1371 sb->sb_csum = calc_sb_1_csum(sb);
1372 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1374 md_super_wait(rdev->mddev);
1375 return size/2; /* kB for sysfs */
1378 static struct super_type super_types[] = {
1381 .owner = THIS_MODULE,
1382 .load_super = super_90_load,
1383 .validate_super = super_90_validate,
1384 .sync_super = super_90_sync,
1385 .rdev_size_change = super_90_rdev_size_change,
1389 .owner = THIS_MODULE,
1390 .load_super = super_1_load,
1391 .validate_super = super_1_validate,
1392 .sync_super = super_1_sync,
1393 .rdev_size_change = super_1_rdev_size_change,
1397 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1399 struct list_head *tmp, *tmp2;
1400 mdk_rdev_t *rdev, *rdev2;
1402 rdev_for_each(rdev, tmp, mddev1)
1403 rdev_for_each(rdev2, tmp2, mddev2)
1404 if (rdev->bdev->bd_contains ==
1405 rdev2->bdev->bd_contains)
1411 static LIST_HEAD(pending_raid_disks);
1413 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1415 char b[BDEVNAME_SIZE];
1425 /* prevent duplicates */
1426 if (find_rdev(mddev, rdev->bdev->bd_dev))
1429 /* make sure rdev->size exceeds mddev->size */
1430 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1432 /* Cannot change size, so fail
1433 * If mddev->level <= 0, then we don't care
1434 * about aligning sizes (e.g. linear)
1436 if (mddev->level > 0)
1439 mddev->size = rdev->size;
1442 /* Verify rdev->desc_nr is unique.
1443 * If it is -1, assign a free number, else
1444 * check number is not in use
1446 if (rdev->desc_nr < 0) {
1448 if (mddev->pers) choice = mddev->raid_disks;
1449 while (find_rdev_nr(mddev, choice))
1451 rdev->desc_nr = choice;
1453 if (find_rdev_nr(mddev, rdev->desc_nr))
1456 bdevname(rdev->bdev,b);
1457 while ( (s=strchr(b, '/')) != NULL)
1460 rdev->mddev = mddev;
1461 printk(KERN_INFO "md: bind<%s>\n", b);
1463 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1466 if (rdev->bdev->bd_part)
1467 ko = &rdev->bdev->bd_part->dev.kobj;
1469 ko = &rdev->bdev->bd_disk->dev.kobj;
1470 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1471 kobject_del(&rdev->kobj);
1474 list_add(&rdev->same_set, &mddev->disks);
1475 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1479 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1484 static void md_delayed_delete(struct work_struct *ws)
1486 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1487 kobject_del(&rdev->kobj);
1488 kobject_put(&rdev->kobj);
1491 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1493 char b[BDEVNAME_SIZE];
1498 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1499 list_del_init(&rdev->same_set);
1500 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1502 sysfs_remove_link(&rdev->kobj, "block");
1504 /* We need to delay this, otherwise we can deadlock when
1505 * writing to 'remove' to "dev/state"
1507 INIT_WORK(&rdev->del_work, md_delayed_delete);
1508 kobject_get(&rdev->kobj);
1509 schedule_work(&rdev->del_work);
1513 * prevent the device from being mounted, repartitioned or
1514 * otherwise reused by a RAID array (or any other kernel
1515 * subsystem), by bd_claiming the device.
1517 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1520 struct block_device *bdev;
1521 char b[BDEVNAME_SIZE];
1523 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1525 printk(KERN_ERR "md: could not open %s.\n",
1526 __bdevname(dev, b));
1527 return PTR_ERR(bdev);
1529 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1531 printk(KERN_ERR "md: could not bd_claim %s.\n",
1537 set_bit(AllReserved, &rdev->flags);
1542 static void unlock_rdev(mdk_rdev_t *rdev)
1544 struct block_device *bdev = rdev->bdev;
1552 void md_autodetect_dev(dev_t dev);
1554 static void export_rdev(mdk_rdev_t * rdev)
1556 char b[BDEVNAME_SIZE];
1557 printk(KERN_INFO "md: export_rdev(%s)\n",
1558 bdevname(rdev->bdev,b));
1562 list_del_init(&rdev->same_set);
1564 if (test_bit(AutoDetected, &rdev->flags))
1565 md_autodetect_dev(rdev->bdev->bd_dev);
1568 kobject_put(&rdev->kobj);
1571 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1573 unbind_rdev_from_array(rdev);
1577 static void export_array(mddev_t *mddev)
1579 struct list_head *tmp;
1582 rdev_for_each(rdev, tmp, mddev) {
1587 kick_rdev_from_array(rdev);
1589 if (!list_empty(&mddev->disks))
1591 mddev->raid_disks = 0;
1592 mddev->major_version = 0;
1595 static void print_desc(mdp_disk_t *desc)
1597 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1598 desc->major,desc->minor,desc->raid_disk,desc->state);
1601 static void print_sb(mdp_super_t *sb)
1606 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1607 sb->major_version, sb->minor_version, sb->patch_version,
1608 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1610 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1611 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1612 sb->md_minor, sb->layout, sb->chunk_size);
1613 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1614 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1615 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1616 sb->failed_disks, sb->spare_disks,
1617 sb->sb_csum, (unsigned long)sb->events_lo);
1620 for (i = 0; i < MD_SB_DISKS; i++) {
1623 desc = sb->disks + i;
1624 if (desc->number || desc->major || desc->minor ||
1625 desc->raid_disk || (desc->state && (desc->state != 4))) {
1626 printk(" D %2d: ", i);
1630 printk(KERN_INFO "md: THIS: ");
1631 print_desc(&sb->this_disk);
1635 static void print_rdev(mdk_rdev_t *rdev)
1637 char b[BDEVNAME_SIZE];
1638 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1639 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1640 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1642 if (rdev->sb_loaded) {
1643 printk(KERN_INFO "md: rdev superblock:\n");
1644 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1646 printk(KERN_INFO "md: no rdev superblock!\n");
1649 static void md_print_devices(void)
1651 struct list_head *tmp, *tmp2;
1654 char b[BDEVNAME_SIZE];
1657 printk("md: **********************************\n");
1658 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1659 printk("md: **********************************\n");
1660 for_each_mddev(mddev, tmp) {
1663 bitmap_print_sb(mddev->bitmap);
1665 printk("%s: ", mdname(mddev));
1666 rdev_for_each(rdev, tmp2, mddev)
1667 printk("<%s>", bdevname(rdev->bdev,b));
1670 rdev_for_each(rdev, tmp2, mddev)
1673 printk("md: **********************************\n");
1678 static void sync_sbs(mddev_t * mddev, int nospares)
1680 /* Update each superblock (in-memory image), but
1681 * if we are allowed to, skip spares which already
1682 * have the right event counter, or have one earlier
1683 * (which would mean they aren't being marked as dirty
1684 * with the rest of the array)
1687 struct list_head *tmp;
1689 rdev_for_each(rdev, tmp, mddev) {
1690 if (rdev->sb_events == mddev->events ||
1692 rdev->raid_disk < 0 &&
1693 (rdev->sb_events&1)==0 &&
1694 rdev->sb_events+1 == mddev->events)) {
1695 /* Don't update this superblock */
1696 rdev->sb_loaded = 2;
1698 super_types[mddev->major_version].
1699 sync_super(mddev, rdev);
1700 rdev->sb_loaded = 1;
1705 static void md_update_sb(mddev_t * mddev, int force_change)
1707 struct list_head *tmp;
1712 if (mddev->external)
1715 spin_lock_irq(&mddev->write_lock);
1717 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1718 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1720 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1721 /* just a clean<-> dirty transition, possibly leave spares alone,
1722 * though if events isn't the right even/odd, we will have to do
1728 if (mddev->degraded)
1729 /* If the array is degraded, then skipping spares is both
1730 * dangerous and fairly pointless.
1731 * Dangerous because a device that was removed from the array
1732 * might have a event_count that still looks up-to-date,
1733 * so it can be re-added without a resync.
1734 * Pointless because if there are any spares to skip,
1735 * then a recovery will happen and soon that array won't
1736 * be degraded any more and the spare can go back to sleep then.
1740 sync_req = mddev->in_sync;
1741 mddev->utime = get_seconds();
1743 /* If this is just a dirty<->clean transition, and the array is clean
1744 * and 'events' is odd, we can roll back to the previous clean state */
1746 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1747 && (mddev->events & 1)
1748 && mddev->events != 1)
1751 /* otherwise we have to go forward and ... */
1753 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1754 /* .. if the array isn't clean, insist on an odd 'events' */
1755 if ((mddev->events&1)==0) {
1760 /* otherwise insist on an even 'events' (for clean states) */
1761 if ((mddev->events&1)) {
1768 if (!mddev->events) {
1770 * oops, this 64-bit counter should never wrap.
1771 * Either we are in around ~1 trillion A.C., assuming
1772 * 1 reboot per second, or we have a bug:
1779 * do not write anything to disk if using
1780 * nonpersistent superblocks
1782 if (!mddev->persistent) {
1783 if (!mddev->external)
1784 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1786 spin_unlock_irq(&mddev->write_lock);
1787 wake_up(&mddev->sb_wait);
1790 sync_sbs(mddev, nospares);
1791 spin_unlock_irq(&mddev->write_lock);
1794 "md: updating %s RAID superblock on device (in sync %d)\n",
1795 mdname(mddev),mddev->in_sync);
1797 bitmap_update_sb(mddev->bitmap);
1798 rdev_for_each(rdev, tmp, mddev) {
1799 char b[BDEVNAME_SIZE];
1800 dprintk(KERN_INFO "md: ");
1801 if (rdev->sb_loaded != 1)
1802 continue; /* no noise on spare devices */
1803 if (test_bit(Faulty, &rdev->flags))
1804 dprintk("(skipping faulty ");
1806 dprintk("%s ", bdevname(rdev->bdev,b));
1807 if (!test_bit(Faulty, &rdev->flags)) {
1808 md_super_write(mddev,rdev,
1809 rdev->sb_start, rdev->sb_size,
1811 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1812 bdevname(rdev->bdev,b),
1813 (unsigned long long)rdev->sb_start);
1814 rdev->sb_events = mddev->events;
1818 if (mddev->level == LEVEL_MULTIPATH)
1819 /* only need to write one superblock... */
1822 md_super_wait(mddev);
1823 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1825 spin_lock_irq(&mddev->write_lock);
1826 if (mddev->in_sync != sync_req ||
1827 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1828 /* have to write it out again */
1829 spin_unlock_irq(&mddev->write_lock);
1832 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1833 spin_unlock_irq(&mddev->write_lock);
1834 wake_up(&mddev->sb_wait);
1838 /* words written to sysfs files may, or may not, be \n terminated.
1839 * We want to accept with case. For this we use cmd_match.
1841 static int cmd_match(const char *cmd, const char *str)
1843 /* See if cmd, written into a sysfs file, matches
1844 * str. They must either be the same, or cmd can
1845 * have a trailing newline
1847 while (*cmd && *str && *cmd == *str) {
1858 struct rdev_sysfs_entry {
1859 struct attribute attr;
1860 ssize_t (*show)(mdk_rdev_t *, char *);
1861 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1865 state_show(mdk_rdev_t *rdev, char *page)
1870 if (test_bit(Faulty, &rdev->flags)) {
1871 len+= sprintf(page+len, "%sfaulty",sep);
1874 if (test_bit(In_sync, &rdev->flags)) {
1875 len += sprintf(page+len, "%sin_sync",sep);
1878 if (test_bit(WriteMostly, &rdev->flags)) {
1879 len += sprintf(page+len, "%swrite_mostly",sep);
1882 if (test_bit(Blocked, &rdev->flags)) {
1883 len += sprintf(page+len, "%sblocked", sep);
1886 if (!test_bit(Faulty, &rdev->flags) &&
1887 !test_bit(In_sync, &rdev->flags)) {
1888 len += sprintf(page+len, "%sspare", sep);
1891 return len+sprintf(page+len, "\n");
1895 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1898 * faulty - simulates and error
1899 * remove - disconnects the device
1900 * writemostly - sets write_mostly
1901 * -writemostly - clears write_mostly
1902 * blocked - sets the Blocked flag
1903 * -blocked - clears the Blocked flag
1906 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1907 md_error(rdev->mddev, rdev);
1909 } else if (cmd_match(buf, "remove")) {
1910 if (rdev->raid_disk >= 0)
1913 mddev_t *mddev = rdev->mddev;
1914 kick_rdev_from_array(rdev);
1916 md_update_sb(mddev, 1);
1917 md_new_event(mddev);
1920 } else if (cmd_match(buf, "writemostly")) {
1921 set_bit(WriteMostly, &rdev->flags);
1923 } else if (cmd_match(buf, "-writemostly")) {
1924 clear_bit(WriteMostly, &rdev->flags);
1926 } else if (cmd_match(buf, "blocked")) {
1927 set_bit(Blocked, &rdev->flags);
1929 } else if (cmd_match(buf, "-blocked")) {
1930 clear_bit(Blocked, &rdev->flags);
1931 wake_up(&rdev->blocked_wait);
1932 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1933 md_wakeup_thread(rdev->mddev->thread);
1938 sysfs_notify(&rdev->kobj, NULL, "state");
1939 return err ? err : len;
1941 static struct rdev_sysfs_entry rdev_state =
1942 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1945 errors_show(mdk_rdev_t *rdev, char *page)
1947 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1951 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1954 unsigned long n = simple_strtoul(buf, &e, 10);
1955 if (*buf && (*e == 0 || *e == '\n')) {
1956 atomic_set(&rdev->corrected_errors, n);
1961 static struct rdev_sysfs_entry rdev_errors =
1962 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1965 slot_show(mdk_rdev_t *rdev, char *page)
1967 if (rdev->raid_disk < 0)
1968 return sprintf(page, "none\n");
1970 return sprintf(page, "%d\n", rdev->raid_disk);
1974 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1979 int slot = simple_strtoul(buf, &e, 10);
1980 if (strncmp(buf, "none", 4)==0)
1982 else if (e==buf || (*e && *e!= '\n'))
1984 if (rdev->mddev->pers && slot == -1) {
1985 /* Setting 'slot' on an active array requires also
1986 * updating the 'rd%d' link, and communicating
1987 * with the personality with ->hot_*_disk.
1988 * For now we only support removing
1989 * failed/spare devices. This normally happens automatically,
1990 * but not when the metadata is externally managed.
1992 if (rdev->raid_disk == -1)
1994 /* personality does all needed checks */
1995 if (rdev->mddev->pers->hot_add_disk == NULL)
1997 err = rdev->mddev->pers->
1998 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2001 sprintf(nm, "rd%d", rdev->raid_disk);
2002 sysfs_remove_link(&rdev->mddev->kobj, nm);
2003 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2004 md_wakeup_thread(rdev->mddev->thread);
2005 } else if (rdev->mddev->pers) {
2007 struct list_head *tmp;
2008 /* Activating a spare .. or possibly reactivating
2009 * if we every get bitmaps working here.
2012 if (rdev->raid_disk != -1)
2015 if (rdev->mddev->pers->hot_add_disk == NULL)
2018 rdev_for_each(rdev2, tmp, rdev->mddev)
2019 if (rdev2->raid_disk == slot)
2022 rdev->raid_disk = slot;
2023 if (test_bit(In_sync, &rdev->flags))
2024 rdev->saved_raid_disk = slot;
2026 rdev->saved_raid_disk = -1;
2027 err = rdev->mddev->pers->
2028 hot_add_disk(rdev->mddev, rdev);
2030 rdev->raid_disk = -1;
2033 sysfs_notify(&rdev->kobj, NULL, "state");
2034 sprintf(nm, "rd%d", rdev->raid_disk);
2035 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2037 "md: cannot register "
2039 nm, mdname(rdev->mddev));
2041 /* don't wakeup anyone, leave that to userspace. */
2043 if (slot >= rdev->mddev->raid_disks)
2045 rdev->raid_disk = slot;
2046 /* assume it is working */
2047 clear_bit(Faulty, &rdev->flags);
2048 clear_bit(WriteMostly, &rdev->flags);
2049 set_bit(In_sync, &rdev->flags);
2050 sysfs_notify(&rdev->kobj, NULL, "state");
2056 static struct rdev_sysfs_entry rdev_slot =
2057 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2060 offset_show(mdk_rdev_t *rdev, char *page)
2062 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2066 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2069 unsigned long long offset = simple_strtoull(buf, &e, 10);
2070 if (e==buf || (*e && *e != '\n'))
2072 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2074 if (rdev->size && rdev->mddev->external)
2075 /* Must set offset before size, so overlap checks
2078 rdev->data_offset = offset;
2082 static struct rdev_sysfs_entry rdev_offset =
2083 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2086 rdev_size_show(mdk_rdev_t *rdev, char *page)
2088 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2091 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2093 /* check if two start/length pairs overlap */
2102 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2105 unsigned long long size = simple_strtoull(buf, &e, 10);
2106 unsigned long long oldsize = rdev->size;
2107 mddev_t *my_mddev = rdev->mddev;
2109 if (e==buf || (*e && *e != '\n'))
2111 if (my_mddev->pers && rdev->raid_disk >= 0) {
2112 if (rdev->mddev->persistent) {
2113 size = super_types[rdev->mddev->major_version].
2114 rdev_size_change(rdev, size);
2118 size = (rdev->bdev->bd_inode->i_size >> 10);
2119 size -= rdev->data_offset/2;
2121 if (size < rdev->mddev->size)
2122 return -EINVAL; /* component must fit device */
2126 if (size > oldsize && rdev->mddev->external) {
2127 /* need to check that all other rdevs with the same ->bdev
2128 * do not overlap. We need to unlock the mddev to avoid
2129 * a deadlock. We have already changed rdev->size, and if
2130 * we have to change it back, we will have the lock again.
2134 struct list_head *tmp, *tmp2;
2136 mddev_unlock(my_mddev);
2137 for_each_mddev(mddev, tmp) {
2141 rdev_for_each(rdev2, tmp2, mddev)
2142 if (test_bit(AllReserved, &rdev2->flags) ||
2143 (rdev->bdev == rdev2->bdev &&
2145 overlaps(rdev->data_offset, rdev->size,
2146 rdev2->data_offset, rdev2->size))) {
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;
2168 if (size < my_mddev->size || my_mddev->size == 0)
2169 my_mddev->size = size;
2173 static struct rdev_sysfs_entry rdev_size =
2174 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2176 static struct attribute *rdev_default_attrs[] = {
2185 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2187 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2188 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2189 mddev_t *mddev = rdev->mddev;
2195 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2197 if (rdev->mddev == NULL)
2200 rv = entry->show(rdev, page);
2201 mddev_unlock(mddev);
2207 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2208 const char *page, size_t length)
2210 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2211 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2213 mddev_t *mddev = rdev->mddev;
2217 if (!capable(CAP_SYS_ADMIN))
2219 rv = mddev ? mddev_lock(mddev): -EBUSY;
2221 if (rdev->mddev == NULL)
2224 rv = entry->store(rdev, page, length);
2225 mddev_unlock(mddev);
2230 static void rdev_free(struct kobject *ko)
2232 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2235 static struct sysfs_ops rdev_sysfs_ops = {
2236 .show = rdev_attr_show,
2237 .store = rdev_attr_store,
2239 static struct kobj_type rdev_ktype = {
2240 .release = rdev_free,
2241 .sysfs_ops = &rdev_sysfs_ops,
2242 .default_attrs = rdev_default_attrs,
2246 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2248 * mark the device faulty if:
2250 * - the device is nonexistent (zero size)
2251 * - the device has no valid superblock
2253 * a faulty rdev _never_ has rdev->sb set.
2255 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2257 char b[BDEVNAME_SIZE];
2262 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2264 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2265 return ERR_PTR(-ENOMEM);
2268 if ((err = alloc_disk_sb(rdev)))
2271 err = lock_rdev(rdev, newdev, super_format == -2);
2275 kobject_init(&rdev->kobj, &rdev_ktype);
2278 rdev->saved_raid_disk = -1;
2279 rdev->raid_disk = -1;
2281 rdev->data_offset = 0;
2282 rdev->sb_events = 0;
2283 atomic_set(&rdev->nr_pending, 0);
2284 atomic_set(&rdev->read_errors, 0);
2285 atomic_set(&rdev->corrected_errors, 0);
2287 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2290 "md: %s has zero or unknown size, marking faulty!\n",
2291 bdevname(rdev->bdev,b));
2296 if (super_format >= 0) {
2297 err = super_types[super_format].
2298 load_super(rdev, NULL, super_minor);
2299 if (err == -EINVAL) {
2301 "md: %s does not have a valid v%d.%d "
2302 "superblock, not importing!\n",
2303 bdevname(rdev->bdev,b),
2304 super_format, super_minor);
2309 "md: could not read %s's sb, not importing!\n",
2310 bdevname(rdev->bdev,b));
2315 INIT_LIST_HEAD(&rdev->same_set);
2316 init_waitqueue_head(&rdev->blocked_wait);
2321 if (rdev->sb_page) {
2327 return ERR_PTR(err);
2331 * Check a full RAID array for plausibility
2335 static void analyze_sbs(mddev_t * mddev)
2338 struct list_head *tmp;
2339 mdk_rdev_t *rdev, *freshest;
2340 char b[BDEVNAME_SIZE];
2343 rdev_for_each(rdev, tmp, mddev)
2344 switch (super_types[mddev->major_version].
2345 load_super(rdev, freshest, mddev->minor_version)) {
2353 "md: fatal superblock inconsistency in %s"
2354 " -- removing from array\n",
2355 bdevname(rdev->bdev,b));
2356 kick_rdev_from_array(rdev);
2360 super_types[mddev->major_version].
2361 validate_super(mddev, freshest);
2364 rdev_for_each(rdev, tmp, mddev) {
2365 if (rdev != freshest)
2366 if (super_types[mddev->major_version].
2367 validate_super(mddev, rdev)) {
2368 printk(KERN_WARNING "md: kicking non-fresh %s"
2370 bdevname(rdev->bdev,b));
2371 kick_rdev_from_array(rdev);
2374 if (mddev->level == LEVEL_MULTIPATH) {
2375 rdev->desc_nr = i++;
2376 rdev->raid_disk = rdev->desc_nr;
2377 set_bit(In_sync, &rdev->flags);
2378 } else if (rdev->raid_disk >= mddev->raid_disks) {
2379 rdev->raid_disk = -1;
2380 clear_bit(In_sync, &rdev->flags);
2386 if (mddev->recovery_cp != MaxSector &&
2388 printk(KERN_ERR "md: %s: raid array is not clean"
2389 " -- starting background reconstruction\n",
2395 safe_delay_show(mddev_t *mddev, char *page)
2397 int msec = (mddev->safemode_delay*1000)/HZ;
2398 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2401 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2409 /* remove a period, and count digits after it */
2410 if (len >= sizeof(buf))
2412 strlcpy(buf, cbuf, len);
2414 for (i=0; i<len; i++) {
2416 if (isdigit(buf[i])) {
2421 } else if (buf[i] == '.') {
2426 msec = simple_strtoul(buf, &e, 10);
2427 if (e == buf || (*e && *e != '\n'))
2429 msec = (msec * 1000) / scale;
2431 mddev->safemode_delay = 0;
2433 mddev->safemode_delay = (msec*HZ)/1000;
2434 if (mddev->safemode_delay == 0)
2435 mddev->safemode_delay = 1;
2439 static struct md_sysfs_entry md_safe_delay =
2440 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2443 level_show(mddev_t *mddev, char *page)
2445 struct mdk_personality *p = mddev->pers;
2447 return sprintf(page, "%s\n", p->name);
2448 else if (mddev->clevel[0])
2449 return sprintf(page, "%s\n", mddev->clevel);
2450 else if (mddev->level != LEVEL_NONE)
2451 return sprintf(page, "%d\n", mddev->level);
2457 level_store(mddev_t *mddev, const char *buf, size_t len)
2464 if (len >= sizeof(mddev->clevel))
2466 strncpy(mddev->clevel, buf, len);
2467 if (mddev->clevel[len-1] == '\n')
2469 mddev->clevel[len] = 0;
2470 mddev->level = LEVEL_NONE;
2474 static struct md_sysfs_entry md_level =
2475 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2479 layout_show(mddev_t *mddev, char *page)
2481 /* just a number, not meaningful for all levels */
2482 if (mddev->reshape_position != MaxSector &&
2483 mddev->layout != mddev->new_layout)
2484 return sprintf(page, "%d (%d)\n",
2485 mddev->new_layout, mddev->layout);
2486 return sprintf(page, "%d\n", mddev->layout);
2490 layout_store(mddev_t *mddev, const char *buf, size_t len)
2493 unsigned long n = simple_strtoul(buf, &e, 10);
2495 if (!*buf || (*e && *e != '\n'))
2500 if (mddev->reshape_position != MaxSector)
2501 mddev->new_layout = n;
2506 static struct md_sysfs_entry md_layout =
2507 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2511 raid_disks_show(mddev_t *mddev, char *page)
2513 if (mddev->raid_disks == 0)
2515 if (mddev->reshape_position != MaxSector &&
2516 mddev->delta_disks != 0)
2517 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2518 mddev->raid_disks - mddev->delta_disks);
2519 return sprintf(page, "%d\n", mddev->raid_disks);
2522 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2525 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2529 unsigned long n = simple_strtoul(buf, &e, 10);
2531 if (!*buf || (*e && *e != '\n'))
2535 rv = update_raid_disks(mddev, n);
2536 else if (mddev->reshape_position != MaxSector) {
2537 int olddisks = mddev->raid_disks - mddev->delta_disks;
2538 mddev->delta_disks = n - olddisks;
2539 mddev->raid_disks = n;
2541 mddev->raid_disks = n;
2542 return rv ? rv : len;
2544 static struct md_sysfs_entry md_raid_disks =
2545 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2548 chunk_size_show(mddev_t *mddev, char *page)
2550 if (mddev->reshape_position != MaxSector &&
2551 mddev->chunk_size != mddev->new_chunk)
2552 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2554 return sprintf(page, "%d\n", mddev->chunk_size);
2558 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2560 /* can only set chunk_size if array is not yet active */
2562 unsigned long n = simple_strtoul(buf, &e, 10);
2564 if (!*buf || (*e && *e != '\n'))
2569 else if (mddev->reshape_position != MaxSector)
2570 mddev->new_chunk = n;
2572 mddev->chunk_size = n;
2575 static struct md_sysfs_entry md_chunk_size =
2576 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2579 resync_start_show(mddev_t *mddev, char *page)
2581 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2585 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2588 unsigned long long n = simple_strtoull(buf, &e, 10);
2592 if (!*buf || (*e && *e != '\n'))
2595 mddev->recovery_cp = n;
2598 static struct md_sysfs_entry md_resync_start =
2599 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2602 * The array state can be:
2605 * No devices, no size, no level
2606 * Equivalent to STOP_ARRAY ioctl
2608 * May have some settings, but array is not active
2609 * all IO results in error
2610 * When written, doesn't tear down array, but just stops it
2611 * suspended (not supported yet)
2612 * All IO requests will block. The array can be reconfigured.
2613 * Writing this, if accepted, will block until array is quiescent
2615 * no resync can happen. no superblocks get written.
2616 * write requests fail
2618 * like readonly, but behaves like 'clean' on a write request.
2620 * clean - no pending writes, but otherwise active.
2621 * When written to inactive array, starts without resync
2622 * If a write request arrives then
2623 * if metadata is known, mark 'dirty' and switch to 'active'.
2624 * if not known, block and switch to write-pending
2625 * If written to an active array that has pending writes, then fails.
2627 * fully active: IO and resync can be happening.
2628 * When written to inactive array, starts with resync
2631 * clean, but writes are blocked waiting for 'active' to be written.
2634 * like active, but no writes have been seen for a while (100msec).
2637 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2638 write_pending, active_idle, bad_word};
2639 static char *array_states[] = {
2640 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2641 "write-pending", "active-idle", NULL };
2643 static int match_word(const char *word, char **list)
2646 for (n=0; list[n]; n++)
2647 if (cmd_match(word, list[n]))
2653 array_state_show(mddev_t *mddev, char *page)
2655 enum array_state st = inactive;
2668 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2670 else if (mddev->safemode)
2676 if (list_empty(&mddev->disks) &&
2677 mddev->raid_disks == 0 &&
2683 return sprintf(page, "%s\n", array_states[st]);
2686 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2687 static int do_md_run(mddev_t * mddev);
2688 static int restart_array(mddev_t *mddev);
2691 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2694 enum array_state st = match_word(buf, array_states);
2699 /* stopping an active array */
2700 if (atomic_read(&mddev->active) > 1)
2702 err = do_md_stop(mddev, 0, 0);
2705 /* stopping an active array */
2707 if (atomic_read(&mddev->active) > 1)
2709 err = do_md_stop(mddev, 2, 0);
2711 err = 0; /* already inactive */
2714 break; /* not supported yet */
2717 err = do_md_stop(mddev, 1, 0);
2720 set_disk_ro(mddev->gendisk, 1);
2721 err = do_md_run(mddev);
2727 err = do_md_stop(mddev, 1, 0);
2729 err = restart_array(mddev);
2732 set_disk_ro(mddev->gendisk, 0);
2736 err = do_md_run(mddev);
2741 restart_array(mddev);
2742 spin_lock_irq(&mddev->write_lock);
2743 if (atomic_read(&mddev->writes_pending) == 0) {
2744 if (mddev->in_sync == 0) {
2746 if (mddev->safemode == 1)
2747 mddev->safemode = 0;
2748 if (mddev->persistent)
2749 set_bit(MD_CHANGE_CLEAN,
2755 spin_unlock_irq(&mddev->write_lock);
2758 mddev->recovery_cp = MaxSector;
2759 err = do_md_run(mddev);
2764 restart_array(mddev);
2765 if (mddev->external)
2766 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2767 wake_up(&mddev->sb_wait);
2771 set_disk_ro(mddev->gendisk, 0);
2772 err = do_md_run(mddev);
2777 /* these cannot be set */
2783 sysfs_notify(&mddev->kobj, NULL, "array_state");
2787 static struct md_sysfs_entry md_array_state =
2788 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2791 null_show(mddev_t *mddev, char *page)
2797 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2799 /* buf must be %d:%d\n? giving major and minor numbers */
2800 /* The new device is added to the array.
2801 * If the array has a persistent superblock, we read the
2802 * superblock to initialise info and check validity.
2803 * Otherwise, only checking done is that in bind_rdev_to_array,
2804 * which mainly checks size.
2807 int major = simple_strtoul(buf, &e, 10);
2813 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2815 minor = simple_strtoul(e+1, &e, 10);
2816 if (*e && *e != '\n')
2818 dev = MKDEV(major, minor);
2819 if (major != MAJOR(dev) ||
2820 minor != MINOR(dev))
2824 if (mddev->persistent) {
2825 rdev = md_import_device(dev, mddev->major_version,
2826 mddev->minor_version);
2827 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2828 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2829 mdk_rdev_t, same_set);
2830 err = super_types[mddev->major_version]
2831 .load_super(rdev, rdev0, mddev->minor_version);
2835 } else if (mddev->external)
2836 rdev = md_import_device(dev, -2, -1);
2838 rdev = md_import_device(dev, -1, -1);
2841 return PTR_ERR(rdev);
2842 err = bind_rdev_to_array(rdev, mddev);
2846 return err ? err : len;
2849 static struct md_sysfs_entry md_new_device =
2850 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2853 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2856 unsigned long chunk, end_chunk;
2860 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2862 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2863 if (buf == end) break;
2864 if (*end == '-') { /* range */
2866 end_chunk = simple_strtoul(buf, &end, 0);
2867 if (buf == end) break;
2869 if (*end && !isspace(*end)) break;
2870 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2872 while (isspace(*buf)) buf++;
2874 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2879 static struct md_sysfs_entry md_bitmap =
2880 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2883 size_show(mddev_t *mddev, char *page)
2885 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2888 static int update_size(mddev_t *mddev, sector_t num_sectors);
2891 size_store(mddev_t *mddev, const char *buf, size_t len)
2893 /* If array is inactive, we can reduce the component size, but
2894 * not increase it (except from 0).
2895 * If array is active, we can try an on-line resize
2899 unsigned long long size = simple_strtoull(buf, &e, 10);
2900 if (!*buf || *buf == '\n' ||
2905 err = update_size(mddev, size * 2);
2906 md_update_sb(mddev, 1);
2908 if (mddev->size == 0 ||
2914 return err ? err : len;
2917 static struct md_sysfs_entry md_size =
2918 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2923 * 'none' for arrays with no metadata (good luck...)
2924 * 'external' for arrays with externally managed metadata,
2925 * or N.M for internally known formats
2928 metadata_show(mddev_t *mddev, char *page)
2930 if (mddev->persistent)
2931 return sprintf(page, "%d.%d\n",
2932 mddev->major_version, mddev->minor_version);
2933 else if (mddev->external)
2934 return sprintf(page, "external:%s\n", mddev->metadata_type);
2936 return sprintf(page, "none\n");
2940 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2944 if (!list_empty(&mddev->disks))
2947 if (cmd_match(buf, "none")) {
2948 mddev->persistent = 0;
2949 mddev->external = 0;
2950 mddev->major_version = 0;
2951 mddev->minor_version = 90;
2954 if (strncmp(buf, "external:", 9) == 0) {
2955 size_t namelen = len-9;
2956 if (namelen >= sizeof(mddev->metadata_type))
2957 namelen = sizeof(mddev->metadata_type)-1;
2958 strncpy(mddev->metadata_type, buf+9, namelen);
2959 mddev->metadata_type[namelen] = 0;
2960 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2961 mddev->metadata_type[--namelen] = 0;
2962 mddev->persistent = 0;
2963 mddev->external = 1;
2964 mddev->major_version = 0;
2965 mddev->minor_version = 90;
2968 major = simple_strtoul(buf, &e, 10);
2969 if (e==buf || *e != '.')
2972 minor = simple_strtoul(buf, &e, 10);
2973 if (e==buf || (*e && *e != '\n') )
2975 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2977 mddev->major_version = major;
2978 mddev->minor_version = minor;
2979 mddev->persistent = 1;
2980 mddev->external = 0;
2984 static struct md_sysfs_entry md_metadata =
2985 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2988 action_show(mddev_t *mddev, char *page)
2990 char *type = "idle";
2991 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2992 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2993 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2995 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2996 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2998 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3002 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3005 return sprintf(page, "%s\n", type);
3009 action_store(mddev_t *mddev, const char *page, size_t len)
3011 if (!mddev->pers || !mddev->pers->sync_request)
3014 if (cmd_match(page, "idle")) {
3015 if (mddev->sync_thread) {
3016 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3017 md_unregister_thread(mddev->sync_thread);
3018 mddev->sync_thread = NULL;
3019 mddev->recovery = 0;
3021 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3022 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3024 else if (cmd_match(page, "resync"))
3025 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3026 else if (cmd_match(page, "recover")) {
3027 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3028 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3029 } else if (cmd_match(page, "reshape")) {
3031 if (mddev->pers->start_reshape == NULL)
3033 err = mddev->pers->start_reshape(mddev);
3036 sysfs_notify(&mddev->kobj, NULL, "degraded");
3038 if (cmd_match(page, "check"))
3039 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3040 else if (!cmd_match(page, "repair"))
3042 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3043 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3045 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3046 md_wakeup_thread(mddev->thread);
3047 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3052 mismatch_cnt_show(mddev_t *mddev, char *page)
3054 return sprintf(page, "%llu\n",
3055 (unsigned long long) mddev->resync_mismatches);
3058 static struct md_sysfs_entry md_scan_mode =
3059 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3062 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3065 sync_min_show(mddev_t *mddev, char *page)
3067 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3068 mddev->sync_speed_min ? "local": "system");
3072 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3076 if (strncmp(buf, "system", 6)==0) {
3077 mddev->sync_speed_min = 0;
3080 min = simple_strtoul(buf, &e, 10);
3081 if (buf == e || (*e && *e != '\n') || min <= 0)
3083 mddev->sync_speed_min = min;
3087 static struct md_sysfs_entry md_sync_min =
3088 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3091 sync_max_show(mddev_t *mddev, char *page)
3093 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3094 mddev->sync_speed_max ? "local": "system");
3098 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3102 if (strncmp(buf, "system", 6)==0) {
3103 mddev->sync_speed_max = 0;
3106 max = simple_strtoul(buf, &e, 10);
3107 if (buf == e || (*e && *e != '\n') || max <= 0)
3109 mddev->sync_speed_max = max;
3113 static struct md_sysfs_entry md_sync_max =
3114 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3117 degraded_show(mddev_t *mddev, char *page)
3119 return sprintf(page, "%d\n", mddev->degraded);
3121 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3124 sync_force_parallel_show(mddev_t *mddev, char *page)
3126 return sprintf(page, "%d\n", mddev->parallel_resync);
3130 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3134 if (strict_strtol(buf, 10, &n))
3137 if (n != 0 && n != 1)
3140 mddev->parallel_resync = n;
3142 if (mddev->sync_thread)
3143 wake_up(&resync_wait);
3148 /* force parallel resync, even with shared block devices */
3149 static struct md_sysfs_entry md_sync_force_parallel =
3150 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3151 sync_force_parallel_show, sync_force_parallel_store);
3154 sync_speed_show(mddev_t *mddev, char *page)
3156 unsigned long resync, dt, db;
3157 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3158 dt = (jiffies - mddev->resync_mark) / HZ;
3160 db = resync - mddev->resync_mark_cnt;
3161 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3164 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3167 sync_completed_show(mddev_t *mddev, char *page)
3169 unsigned long max_blocks, resync;
3171 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3172 max_blocks = mddev->resync_max_sectors;
3174 max_blocks = mddev->size << 1;
3176 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3177 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3180 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3183 min_sync_show(mddev_t *mddev, char *page)
3185 return sprintf(page, "%llu\n",
3186 (unsigned long long)mddev->resync_min);
3189 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3191 unsigned long long min;
3192 if (strict_strtoull(buf, 10, &min))
3194 if (min > mddev->resync_max)
3196 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3199 /* Must be a multiple of chunk_size */
3200 if (mddev->chunk_size) {
3201 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3204 mddev->resync_min = min;
3209 static struct md_sysfs_entry md_min_sync =
3210 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3213 max_sync_show(mddev_t *mddev, char *page)
3215 if (mddev->resync_max == MaxSector)
3216 return sprintf(page, "max\n");
3218 return sprintf(page, "%llu\n",
3219 (unsigned long long)mddev->resync_max);
3222 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3224 if (strncmp(buf, "max", 3) == 0)
3225 mddev->resync_max = MaxSector;
3227 unsigned long long max;
3228 if (strict_strtoull(buf, 10, &max))
3230 if (max < mddev->resync_min)
3232 if (max < mddev->resync_max &&
3233 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3236 /* Must be a multiple of chunk_size */
3237 if (mddev->chunk_size) {
3238 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3241 mddev->resync_max = max;
3243 wake_up(&mddev->recovery_wait);
3247 static struct md_sysfs_entry md_max_sync =
3248 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3251 suspend_lo_show(mddev_t *mddev, char *page)
3253 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3257 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3260 unsigned long long new = simple_strtoull(buf, &e, 10);
3262 if (mddev->pers->quiesce == NULL)
3264 if (buf == e || (*e && *e != '\n'))
3266 if (new >= mddev->suspend_hi ||
3267 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3268 mddev->suspend_lo = new;
3269 mddev->pers->quiesce(mddev, 2);
3274 static struct md_sysfs_entry md_suspend_lo =
3275 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3279 suspend_hi_show(mddev_t *mddev, char *page)
3281 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3285 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3288 unsigned long long new = simple_strtoull(buf, &e, 10);
3290 if (mddev->pers->quiesce == NULL)
3292 if (buf == e || (*e && *e != '\n'))
3294 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3295 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3296 mddev->suspend_hi = new;
3297 mddev->pers->quiesce(mddev, 1);
3298 mddev->pers->quiesce(mddev, 0);
3303 static struct md_sysfs_entry md_suspend_hi =
3304 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3307 reshape_position_show(mddev_t *mddev, char *page)
3309 if (mddev->reshape_position != MaxSector)
3310 return sprintf(page, "%llu\n",
3311 (unsigned long long)mddev->reshape_position);
3312 strcpy(page, "none\n");
3317 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3320 unsigned long long new = simple_strtoull(buf, &e, 10);
3323 if (buf == e || (*e && *e != '\n'))
3325 mddev->reshape_position = new;
3326 mddev->delta_disks = 0;
3327 mddev->new_level = mddev->level;
3328 mddev->new_layout = mddev->layout;
3329 mddev->new_chunk = mddev->chunk_size;
3333 static struct md_sysfs_entry md_reshape_position =
3334 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3335 reshape_position_store);
3338 static struct attribute *md_default_attrs[] = {
3341 &md_raid_disks.attr,
3342 &md_chunk_size.attr,
3344 &md_resync_start.attr,
3346 &md_new_device.attr,
3347 &md_safe_delay.attr,
3348 &md_array_state.attr,
3349 &md_reshape_position.attr,
3353 static struct attribute *md_redundancy_attrs[] = {
3355 &md_mismatches.attr,
3358 &md_sync_speed.attr,
3359 &md_sync_force_parallel.attr,
3360 &md_sync_completed.attr,
3363 &md_suspend_lo.attr,
3364 &md_suspend_hi.attr,
3369 static struct attribute_group md_redundancy_group = {
3371 .attrs = md_redundancy_attrs,
3376 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3378 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3379 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3384 rv = mddev_lock(mddev);
3386 rv = entry->show(mddev, page);
3387 mddev_unlock(mddev);
3393 md_attr_store(struct kobject *kobj, struct attribute *attr,
3394 const char *page, size_t length)
3396 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3397 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3402 if (!capable(CAP_SYS_ADMIN))
3404 rv = mddev_lock(mddev);
3406 rv = entry->store(mddev, page, length);
3407 mddev_unlock(mddev);
3412 static void md_free(struct kobject *ko)
3414 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3418 static struct sysfs_ops md_sysfs_ops = {
3419 .show = md_attr_show,
3420 .store = md_attr_store,
3422 static struct kobj_type md_ktype = {
3424 .sysfs_ops = &md_sysfs_ops,
3425 .default_attrs = md_default_attrs,
3430 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3432 static DEFINE_MUTEX(disks_mutex);
3433 mddev_t *mddev = mddev_find(dev);
3434 struct gendisk *disk;
3435 int partitioned = (MAJOR(dev) != MD_MAJOR);
3436 int shift = partitioned ? MdpMinorShift : 0;
3437 int unit = MINOR(dev) >> shift;
3443 mutex_lock(&disks_mutex);
3444 if (mddev->gendisk) {
3445 mutex_unlock(&disks_mutex);
3449 disk = alloc_disk(1 << shift);
3451 mutex_unlock(&disks_mutex);
3455 disk->major = MAJOR(dev);
3456 disk->first_minor = unit << shift;
3458 sprintf(disk->disk_name, "md_d%d", unit);
3460 sprintf(disk->disk_name, "md%d", unit);
3461 disk->fops = &md_fops;
3462 disk->private_data = mddev;
3463 disk->queue = mddev->queue;
3465 mddev->gendisk = disk;
3466 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3468 mutex_unlock(&disks_mutex);
3470 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3473 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3477 static void md_safemode_timeout(unsigned long data)
3479 mddev_t *mddev = (mddev_t *) data;
3481 if (!atomic_read(&mddev->writes_pending)) {
3482 mddev->safemode = 1;
3483 if (mddev->external)
3484 sysfs_notify(&mddev->kobj, NULL, "array_state");
3486 md_wakeup_thread(mddev->thread);
3489 static int start_dirty_degraded;
3491 static int do_md_run(mddev_t * mddev)
3495 struct list_head *tmp;
3497 struct gendisk *disk;
3498 struct mdk_personality *pers;
3499 char b[BDEVNAME_SIZE];
3501 if (list_empty(&mddev->disks))
3502 /* cannot run an array with no devices.. */
3509 * Analyze all RAID superblock(s)
3511 if (!mddev->raid_disks) {
3512 if (!mddev->persistent)
3517 chunk_size = mddev->chunk_size;
3520 if (chunk_size > MAX_CHUNK_SIZE) {
3521 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3522 chunk_size, MAX_CHUNK_SIZE);
3526 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3528 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3529 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3532 if (chunk_size < PAGE_SIZE) {
3533 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3534 chunk_size, PAGE_SIZE);
3538 /* devices must have minimum size of one chunk */
3539 rdev_for_each(rdev, tmp, mddev) {
3540 if (test_bit(Faulty, &rdev->flags))
3542 if (rdev->size < chunk_size / 1024) {
3544 "md: Dev %s smaller than chunk_size:"
3546 bdevname(rdev->bdev,b),
3547 (unsigned long long)rdev->size,
3555 if (mddev->level != LEVEL_NONE)
3556 request_module("md-level-%d", mddev->level);
3557 else if (mddev->clevel[0])
3558 request_module("md-%s", mddev->clevel);
3562 * Drop all container device buffers, from now on
3563 * the only valid external interface is through the md
3566 rdev_for_each(rdev, tmp, mddev) {
3567 if (test_bit(Faulty, &rdev->flags))
3569 sync_blockdev(rdev->bdev);
3570 invalidate_bdev(rdev->bdev);
3572 /* perform some consistency tests on the device.
3573 * We don't want the data to overlap the metadata,
3574 * Internal Bitmap issues has handled elsewhere.
3576 if (rdev->data_offset < rdev->sb_start) {
3578 rdev->data_offset + mddev->size*2
3580 printk("md: %s: data overlaps metadata\n",
3585 if (rdev->sb_start + rdev->sb_size/512
3586 > rdev->data_offset) {
3587 printk("md: %s: metadata overlaps data\n",
3592 sysfs_notify(&rdev->kobj, NULL, "state");
3595 md_probe(mddev->unit, NULL, NULL);
3596 disk = mddev->gendisk;
3600 spin_lock(&pers_lock);
3601 pers = find_pers(mddev->level, mddev->clevel);
3602 if (!pers || !try_module_get(pers->owner)) {
3603 spin_unlock(&pers_lock);
3604 if (mddev->level != LEVEL_NONE)
3605 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3608 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3613 spin_unlock(&pers_lock);
3614 mddev->level = pers->level;
3615 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3617 if (mddev->reshape_position != MaxSector &&
3618 pers->start_reshape == NULL) {
3619 /* This personality cannot handle reshaping... */
3621 module_put(pers->owner);
3625 if (pers->sync_request) {
3626 /* Warn if this is a potentially silly
3629 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3631 struct list_head *tmp2;
3633 rdev_for_each(rdev, tmp, mddev) {
3634 rdev_for_each(rdev2, tmp2, mddev) {
3636 rdev->bdev->bd_contains ==
3637 rdev2->bdev->bd_contains) {
3639 "%s: WARNING: %s appears to be"
3640 " on the same physical disk as"
3643 bdevname(rdev->bdev,b),
3644 bdevname(rdev2->bdev,b2));
3651 "True protection against single-disk"
3652 " failure might be compromised.\n");
3655 mddev->recovery = 0;
3656 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3657 mddev->barriers_work = 1;
3658 mddev->ok_start_degraded = start_dirty_degraded;
3661 mddev->ro = 2; /* read-only, but switch on first write */
3663 err = mddev->pers->run(mddev);
3665 printk(KERN_ERR "md: pers->run() failed ...\n");
3666 else if (mddev->pers->sync_request) {
3667 err = bitmap_create(mddev);
3669 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3670 mdname(mddev), err);
3671 mddev->pers->stop(mddev);
3675 module_put(mddev->pers->owner);
3677 bitmap_destroy(mddev);
3680 if (mddev->pers->sync_request) {
3681 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3683 "md: cannot register extra attributes for %s\n",
3685 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3688 atomic_set(&mddev->writes_pending,0);
3689 mddev->safemode = 0;
3690 mddev->safemode_timer.function = md_safemode_timeout;
3691 mddev->safemode_timer.data = (unsigned long) mddev;
3692 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3695 rdev_for_each(rdev, tmp, mddev)
3696 if (rdev->raid_disk >= 0) {
3698 sprintf(nm, "rd%d", rdev->raid_disk);
3699 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3700 printk("md: cannot register %s for %s\n",
3704 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3707 md_update_sb(mddev, 0);
3709 set_capacity(disk, mddev->array_size<<1);
3711 /* If we call blk_queue_make_request here, it will
3712 * re-initialise max_sectors etc which may have been
3713 * refined inside -> run. So just set the bits we need to set.
3714 * Most initialisation happended when we called
3715 * blk_queue_make_request(..., md_fail_request)
3718 mddev->queue->queuedata = mddev;
3719 mddev->queue->make_request_fn = mddev->pers->make_request;
3721 /* If there is a partially-recovered drive we need to
3722 * start recovery here. If we leave it to md_check_recovery,
3723 * it will remove the drives and not do the right thing
3725 if (mddev->degraded && !mddev->sync_thread) {
3726 struct list_head *rtmp;
3728 rdev_for_each(rdev, rtmp, mddev)
3729 if (rdev->raid_disk >= 0 &&
3730 !test_bit(In_sync, &rdev->flags) &&
3731 !test_bit(Faulty, &rdev->flags))
3732 /* complete an interrupted recovery */
3734 if (spares && mddev->pers->sync_request) {
3735 mddev->recovery = 0;
3736 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3737 mddev->sync_thread = md_register_thread(md_do_sync,
3740 if (!mddev->sync_thread) {
3741 printk(KERN_ERR "%s: could not start resync"
3744 /* leave the spares where they are, it shouldn't hurt */
3745 mddev->recovery = 0;
3749 md_wakeup_thread(mddev->thread);
3750 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3753 md_new_event(mddev);
3754 sysfs_notify(&mddev->kobj, NULL, "array_state");
3755 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3756 sysfs_notify(&mddev->kobj, NULL, "degraded");
3757 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3761 static int restart_array(mddev_t *mddev)
3763 struct gendisk *disk = mddev->gendisk;
3765 /* Complain if it has no devices */
3766 if (list_empty(&mddev->disks))
3772 mddev->safemode = 0;
3774 set_disk_ro(disk, 0);
3775 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3777 /* Kick recovery or resync if necessary */
3778 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3779 md_wakeup_thread(mddev->thread);
3780 md_wakeup_thread(mddev->sync_thread);
3781 sysfs_notify(&mddev->kobj, NULL, "array_state");
3785 /* similar to deny_write_access, but accounts for our holding a reference
3786 * to the file ourselves */
3787 static int deny_bitmap_write_access(struct file * file)
3789 struct inode *inode = file->f_mapping->host;
3791 spin_lock(&inode->i_lock);
3792 if (atomic_read(&inode->i_writecount) > 1) {
3793 spin_unlock(&inode->i_lock);
3796 atomic_set(&inode->i_writecount, -1);
3797 spin_unlock(&inode->i_lock);
3802 static void restore_bitmap_write_access(struct file *file)
3804 struct inode *inode = file->f_mapping->host;
3806 spin_lock(&inode->i_lock);
3807 atomic_set(&inode->i_writecount, 1);
3808 spin_unlock(&inode->i_lock);
3812 * 0 - completely stop and dis-assemble array
3813 * 1 - switch to readonly
3814 * 2 - stop but do not disassemble array
3816 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3819 struct gendisk *disk = mddev->gendisk;
3821 if (atomic_read(&mddev->active) > 1 + is_open) {
3822 printk("md: %s still in use.\n",mdname(mddev));
3828 if (mddev->sync_thread) {
3829 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3830 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3831 md_unregister_thread(mddev->sync_thread);
3832 mddev->sync_thread = NULL;
3835 del_timer_sync(&mddev->safemode_timer);
3837 invalidate_partition(disk, 0);
3840 case 1: /* readonly */
3846 case 0: /* disassemble */
3848 bitmap_flush(mddev);
3849 md_super_wait(mddev);
3851 set_disk_ro(disk, 0);
3852 blk_queue_make_request(mddev->queue, md_fail_request);
3853 mddev->pers->stop(mddev);
3854 mddev->queue->merge_bvec_fn = NULL;
3855 mddev->queue->unplug_fn = NULL;
3856 mddev->queue->backing_dev_info.congested_fn = NULL;
3857 if (mddev->pers->sync_request)
3858 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3860 module_put(mddev->pers->owner);
3862 /* tell userspace to handle 'inactive' */
3863 sysfs_notify(&mddev->kobj, NULL, "array_state");
3865 set_capacity(disk, 0);
3871 if (!mddev->in_sync || mddev->flags) {
3872 /* mark array as shutdown cleanly */
3874 md_update_sb(mddev, 1);
3877 set_disk_ro(disk, 1);
3878 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3882 * Free resources if final stop
3886 struct list_head *tmp;
3888 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3890 bitmap_destroy(mddev);
3891 if (mddev->bitmap_file) {
3892 restore_bitmap_write_access(mddev->bitmap_file);
3893 fput(mddev->bitmap_file);
3894 mddev->bitmap_file = NULL;
3896 mddev->bitmap_offset = 0;
3898 rdev_for_each(rdev, tmp, mddev)
3899 if (rdev->raid_disk >= 0) {
3901 sprintf(nm, "rd%d", rdev->raid_disk);
3902 sysfs_remove_link(&mddev->kobj, nm);
3905 /* make sure all md_delayed_delete calls have finished */
3906 flush_scheduled_work();
3908 export_array(mddev);
3910 mddev->array_size = 0;
3912 mddev->raid_disks = 0;
3913 mddev->recovery_cp = 0;
3914 mddev->resync_min = 0;
3915 mddev->resync_max = MaxSector;
3916 mddev->reshape_position = MaxSector;
3917 mddev->external = 0;
3918 mddev->persistent = 0;
3919 mddev->level = LEVEL_NONE;
3920 mddev->clevel[0] = 0;
3923 mddev->metadata_type[0] = 0;
3924 mddev->chunk_size = 0;
3925 mddev->ctime = mddev->utime = 0;
3927 mddev->max_disks = 0;
3929 mddev->delta_disks = 0;
3930 mddev->new_level = LEVEL_NONE;
3931 mddev->new_layout = 0;
3932 mddev->new_chunk = 0;
3933 mddev->curr_resync = 0;
3934 mddev->resync_mismatches = 0;
3935 mddev->suspend_lo = mddev->suspend_hi = 0;
3936 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3937 mddev->recovery = 0;
3940 mddev->degraded = 0;
3941 mddev->barriers_work = 0;
3942 mddev->safemode = 0;
3944 } else if (mddev->pers)
3945 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3948 md_new_event(mddev);
3949 sysfs_notify(&mddev->kobj, NULL, "array_state");
3955 static void autorun_array(mddev_t *mddev)
3958 struct list_head *tmp;
3961 if (list_empty(&mddev->disks))
3964 printk(KERN_INFO "md: running: ");
3966 rdev_for_each(rdev, tmp, mddev) {
3967 char b[BDEVNAME_SIZE];
3968 printk("<%s>", bdevname(rdev->bdev,b));
3972 err = do_md_run (mddev);
3974 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3975 do_md_stop (mddev, 0, 0);
3980 * lets try to run arrays based on all disks that have arrived
3981 * until now. (those are in pending_raid_disks)
3983 * the method: pick the first pending disk, collect all disks with
3984 * the same UUID, remove all from the pending list and put them into
3985 * the 'same_array' list. Then order this list based on superblock
3986 * update time (freshest comes first), kick out 'old' disks and
3987 * compare superblocks. If everything's fine then run it.
3989 * If "unit" is allocated, then bump its reference count
3991 static void autorun_devices(int part)
3993 struct list_head *tmp;
3994 mdk_rdev_t *rdev0, *rdev;
3996 char b[BDEVNAME_SIZE];
3998 printk(KERN_INFO "md: autorun ...\n");
3999 while (!list_empty(&pending_raid_disks)) {
4002 LIST_HEAD(candidates);
4003 rdev0 = list_entry(pending_raid_disks.next,
4004 mdk_rdev_t, same_set);
4006 printk(KERN_INFO "md: considering %s ...\n",
4007 bdevname(rdev0->bdev,b));
4008 INIT_LIST_HEAD(&candidates);
4009 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4010 if (super_90_load(rdev, rdev0, 0) >= 0) {
4011 printk(KERN_INFO "md: adding %s ...\n",
4012 bdevname(rdev->bdev,b));
4013 list_move(&rdev->same_set, &candidates);
4016 * now we have a set of devices, with all of them having
4017 * mostly sane superblocks. It's time to allocate the
4021 dev = MKDEV(mdp_major,
4022 rdev0->preferred_minor << MdpMinorShift);
4023 unit = MINOR(dev) >> MdpMinorShift;
4025 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4028 if (rdev0->preferred_minor != unit) {
4029 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4030 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4034 md_probe(dev, NULL, NULL);
4035 mddev = mddev_find(dev);
4036 if (!mddev || !mddev->gendisk) {
4040 "md: cannot allocate memory for md drive.\n");
4043 if (mddev_lock(mddev))
4044 printk(KERN_WARNING "md: %s locked, cannot run\n",
4046 else if (mddev->raid_disks || mddev->major_version
4047 || !list_empty(&mddev->disks)) {
4049 "md: %s already running, cannot run %s\n",
4050 mdname(mddev), bdevname(rdev0->bdev,b));
4051 mddev_unlock(mddev);
4053 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4054 mddev->persistent = 1;
4055 rdev_for_each_list(rdev, tmp, candidates) {
4056 list_del_init(&rdev->same_set);
4057 if (bind_rdev_to_array(rdev, mddev))
4060 autorun_array(mddev);
4061 mddev_unlock(mddev);
4063 /* on success, candidates will be empty, on error
4066 rdev_for_each_list(rdev, tmp, candidates)
4070 printk(KERN_INFO "md: ... autorun DONE.\n");
4072 #endif /* !MODULE */
4074 static int get_version(void __user * arg)
4078 ver.major = MD_MAJOR_VERSION;
4079 ver.minor = MD_MINOR_VERSION;
4080 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4082 if (copy_to_user(arg, &ver, sizeof(ver)))
4088 static int get_array_info(mddev_t * mddev, void __user * arg)
4090 mdu_array_info_t info;
4091 int nr,working,active,failed,spare;
4093 struct list_head *tmp;
4095 nr=working=active=failed=spare=0;
4096 rdev_for_each(rdev, tmp, mddev) {
4098 if (test_bit(Faulty, &rdev->flags))
4102 if (test_bit(In_sync, &rdev->flags))
4109 info.major_version = mddev->major_version;
4110 info.minor_version = mddev->minor_version;
4111 info.patch_version = MD_PATCHLEVEL_VERSION;
4112 info.ctime = mddev->ctime;
4113 info.level = mddev->level;
4114 info.size = mddev->size;
4115 if (info.size != mddev->size) /* overflow */
4118 info.raid_disks = mddev->raid_disks;
4119 info.md_minor = mddev->md_minor;
4120 info.not_persistent= !mddev->persistent;
4122 info.utime = mddev->utime;
4125 info.state = (1<<MD_SB_CLEAN);
4126 if (mddev->bitmap && mddev->bitmap_offset)
4127 info.state = (1<<MD_SB_BITMAP_PRESENT);
4128 info.active_disks = active;
4129 info.working_disks = working;
4130 info.failed_disks = failed;
4131 info.spare_disks = spare;
4133 info.layout = mddev->layout;
4134 info.chunk_size = mddev->chunk_size;
4136 if (copy_to_user(arg, &info, sizeof(info)))
4142 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4144 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4145 char *ptr, *buf = NULL;
4148 if (md_allow_write(mddev))
4149 file = kmalloc(sizeof(*file), GFP_NOIO);
4151 file = kmalloc(sizeof(*file), GFP_KERNEL);
4156 /* bitmap disabled, zero the first byte and copy out */
4157 if (!mddev->bitmap || !mddev->bitmap->file) {
4158 file->pathname[0] = '\0';
4162 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4166 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4170 strcpy(file->pathname, ptr);
4174 if (copy_to_user(arg, file, sizeof(*file)))
4182 static int get_disk_info(mddev_t * mddev, void __user * arg)
4184 mdu_disk_info_t info;
4187 if (copy_from_user(&info, arg, sizeof(info)))
4190 rdev = find_rdev_nr(mddev, info.number);
4192 info.major = MAJOR(rdev->bdev->bd_dev);
4193 info.minor = MINOR(rdev->bdev->bd_dev);
4194 info.raid_disk = rdev->raid_disk;
4196 if (test_bit(Faulty, &rdev->flags))
4197 info.state |= (1<<MD_DISK_FAULTY);
4198 else if (test_bit(In_sync, &rdev->flags)) {
4199 info.state |= (1<<MD_DISK_ACTIVE);
4200 info.state |= (1<<MD_DISK_SYNC);
4202 if (test_bit(WriteMostly, &rdev->flags))
4203 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4205 info.major = info.minor = 0;
4206 info.raid_disk = -1;
4207 info.state = (1<<MD_DISK_REMOVED);
4210 if (copy_to_user(arg, &info, sizeof(info)))
4216 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4218 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4220 dev_t dev = MKDEV(info->major,info->minor);
4222 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4225 if (!mddev->raid_disks) {
4227 /* expecting a device which has a superblock */
4228 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4231 "md: md_import_device returned %ld\n",
4233 return PTR_ERR(rdev);
4235 if (!list_empty(&mddev->disks)) {
4236 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4237 mdk_rdev_t, same_set);
4238 int err = super_types[mddev->major_version]
4239 .load_super(rdev, rdev0, mddev->minor_version);
4242 "md: %s has different UUID to %s\n",
4243 bdevname(rdev->bdev,b),
4244 bdevname(rdev0->bdev,b2));
4249 err = bind_rdev_to_array(rdev, mddev);
4256 * add_new_disk can be used once the array is assembled
4257 * to add "hot spares". They must already have a superblock
4262 if (!mddev->pers->hot_add_disk) {
4264 "%s: personality does not support diskops!\n",
4268 if (mddev->persistent)
4269 rdev = md_import_device(dev, mddev->major_version,
4270 mddev->minor_version);
4272 rdev = md_import_device(dev, -1, -1);
4275 "md: md_import_device returned %ld\n",
4277 return PTR_ERR(rdev);
4279 /* set save_raid_disk if appropriate */
4280 if (!mddev->persistent) {
4281 if (info->state & (1<<MD_DISK_SYNC) &&
4282 info->raid_disk < mddev->raid_disks)
4283 rdev->raid_disk = info->raid_disk;
4285 rdev->raid_disk = -1;
4287 super_types[mddev->major_version].
4288 validate_super(mddev, rdev);
4289 rdev->saved_raid_disk = rdev->raid_disk;
4291 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4292 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4293 set_bit(WriteMostly, &rdev->flags);
4295 rdev->raid_disk = -1;
4296 err = bind_rdev_to_array(rdev, mddev);
4297 if (!err && !mddev->pers->hot_remove_disk) {
4298 /* If there is hot_add_disk but no hot_remove_disk
4299 * then added disks for geometry changes,
4300 * and should be added immediately.
4302 super_types[mddev->major_version].
4303 validate_super(mddev, rdev);
4304 err = mddev->pers->hot_add_disk(mddev, rdev);
4306 unbind_rdev_from_array(rdev);
4311 sysfs_notify(&rdev->kobj, NULL, "state");
4313 md_update_sb(mddev, 1);
4314 if (mddev->degraded)
4315 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4316 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4317 md_wakeup_thread(mddev->thread);
4321 /* otherwise, add_new_disk is only allowed
4322 * for major_version==0 superblocks
4324 if (mddev->major_version != 0) {
4325 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4330 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4332 rdev = md_import_device (dev, -1, 0);
4335 "md: error, md_import_device() returned %ld\n",
4337 return PTR_ERR(rdev);
4339 rdev->desc_nr = info->number;
4340 if (info->raid_disk < mddev->raid_disks)
4341 rdev->raid_disk = info->raid_disk;
4343 rdev->raid_disk = -1;
4345 if (rdev->raid_disk < mddev->raid_disks)
4346 if (info->state & (1<<MD_DISK_SYNC))
4347 set_bit(In_sync, &rdev->flags);
4349 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4350 set_bit(WriteMostly, &rdev->flags);
4352 if (!mddev->persistent) {
4353 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4354 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4356 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4357 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4359 err = bind_rdev_to_array(rdev, mddev);
4369 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4371 char b[BDEVNAME_SIZE];
4374 rdev = find_rdev(mddev, dev);
4378 if (rdev->raid_disk >= 0)
4381 kick_rdev_from_array(rdev);
4382 md_update_sb(mddev, 1);
4383 md_new_event(mddev);
4387 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4388 bdevname(rdev->bdev,b), mdname(mddev));
4392 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4394 char b[BDEVNAME_SIZE];
4401 if (mddev->major_version != 0) {
4402 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4403 " version-0 superblocks.\n",
4407 if (!mddev->pers->hot_add_disk) {
4409 "%s: personality does not support diskops!\n",
4414 rdev = md_import_device (dev, -1, 0);
4417 "md: error, md_import_device() returned %ld\n",
4422 if (mddev->persistent)
4423 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4425 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4427 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4429 if (test_bit(Faulty, &rdev->flags)) {
4431 "md: can not hot-add faulty %s disk to %s!\n",
4432 bdevname(rdev->bdev,b), mdname(mddev));
4436 clear_bit(In_sync, &rdev->flags);
4438 rdev->saved_raid_disk = -1;
4439 err = bind_rdev_to_array(rdev, mddev);
4444 * The rest should better be atomic, we can have disk failures
4445 * noticed in interrupt contexts ...
4448 if (rdev->desc_nr == mddev->max_disks) {
4449 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4452 goto abort_unbind_export;
4455 rdev->raid_disk = -1;
4457 md_update_sb(mddev, 1);
4460 * Kick recovery, maybe this spare has to be added to the
4461 * array immediately.
4463 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4464 md_wakeup_thread(mddev->thread);
4465 md_new_event(mddev);
4468 abort_unbind_export:
4469 unbind_rdev_from_array(rdev);
4476 static int set_bitmap_file(mddev_t *mddev, int fd)
4481 if (!mddev->pers->quiesce)
4483 if (mddev->recovery || mddev->sync_thread)
4485 /* we should be able to change the bitmap.. */
4491 return -EEXIST; /* cannot add when bitmap is present */
4492 mddev->bitmap_file = fget(fd);
4494 if (mddev->bitmap_file == NULL) {
4495 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4500 err = deny_bitmap_write_access(mddev->bitmap_file);
4502 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4504 fput(mddev->bitmap_file);
4505 mddev->bitmap_file = NULL;
4508 mddev->bitmap_offset = 0; /* file overrides offset */
4509 } else if (mddev->bitmap == NULL)
4510 return -ENOENT; /* cannot remove what isn't there */
4513 mddev->pers->quiesce(mddev, 1);
4515 err = bitmap_create(mddev);
4516 if (fd < 0 || err) {
4517 bitmap_destroy(mddev);
4518 fd = -1; /* make sure to put the file */
4520 mddev->pers->quiesce(mddev, 0);
4523 if (mddev->bitmap_file) {
4524 restore_bitmap_write_access(mddev->bitmap_file);
4525 fput(mddev->bitmap_file);
4527 mddev->bitmap_file = NULL;
4534 * set_array_info is used two different ways
4535 * The original usage is when creating a new array.
4536 * In this usage, raid_disks is > 0 and it together with
4537 * level, size, not_persistent,layout,chunksize determine the
4538 * shape of the array.
4539 * This will always create an array with a type-0.90.0 superblock.
4540 * The newer usage is when assembling an array.
4541 * In this case raid_disks will be 0, and the major_version field is
4542 * use to determine which style super-blocks are to be found on the devices.
4543 * The minor and patch _version numbers are also kept incase the
4544 * super_block handler wishes to interpret them.
4546 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4549 if (info->raid_disks == 0) {
4550 /* just setting version number for superblock loading */
4551 if (info->major_version < 0 ||
4552 info->major_version >= ARRAY_SIZE(super_types) ||
4553 super_types[info->major_version].name == NULL) {
4554 /* maybe try to auto-load a module? */
4556 "md: superblock version %d not known\n",
4557 info->major_version);
4560 mddev->major_version = info->major_version;
4561 mddev->minor_version = info->minor_version;
4562 mddev->patch_version = info->patch_version;
4563 mddev->persistent = !info->not_persistent;
4566 mddev->major_version = MD_MAJOR_VERSION;
4567 mddev->minor_version = MD_MINOR_VERSION;
4568 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4569 mddev->ctime = get_seconds();
4571 mddev->level = info->level;
4572 mddev->clevel[0] = 0;
4573 mddev->size = info->size;
4574 mddev->raid_disks = info->raid_disks;
4575 /* don't set md_minor, it is determined by which /dev/md* was
4578 if (info->state & (1<<MD_SB_CLEAN))
4579 mddev->recovery_cp = MaxSector;
4581 mddev->recovery_cp = 0;
4582 mddev->persistent = ! info->not_persistent;
4583 mddev->external = 0;
4585 mddev->layout = info->layout;
4586 mddev->chunk_size = info->chunk_size;
4588 mddev->max_disks = MD_SB_DISKS;
4590 if (mddev->persistent)
4592 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4594 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4595 mddev->bitmap_offset = 0;
4597 mddev->reshape_position = MaxSector;
4600 * Generate a 128 bit UUID
4602 get_random_bytes(mddev->uuid, 16);
4604 mddev->new_level = mddev->level;
4605 mddev->new_chunk = mddev->chunk_size;
4606 mddev->new_layout = mddev->layout;
4607 mddev->delta_disks = 0;
4612 static int update_size(mddev_t *mddev, sector_t num_sectors)
4616 struct list_head *tmp;
4617 int fit = (num_sectors == 0);
4619 if (mddev->pers->resize == NULL)
4621 /* The "num_sectors" is the number of sectors of each device that
4622 * is used. This can only make sense for arrays with redundancy.
4623 * linear and raid0 always use whatever space is available. We can only
4624 * consider changing this number if no resync or reconstruction is
4625 * happening, and if the new size is acceptable. It must fit before the
4626 * sb_start or, if that is <data_offset, it must fit before the size
4627 * of each device. If num_sectors is zero, we find the largest size
4631 if (mddev->sync_thread)
4633 rdev_for_each(rdev, tmp, mddev) {
4635 avail = rdev->size * 2;
4637 if (fit && (num_sectors == 0 || num_sectors > avail))
4638 num_sectors = avail;
4639 if (avail < num_sectors)
4642 rv = mddev->pers->resize(mddev, num_sectors);
4644 struct block_device *bdev;
4646 bdev = bdget_disk(mddev->gendisk, 0);
4648 mutex_lock(&bdev->bd_inode->i_mutex);
4649 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4650 mutex_unlock(&bdev->bd_inode->i_mutex);
4657 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4660 /* change the number of raid disks */
4661 if (mddev->pers->check_reshape == NULL)
4663 if (raid_disks <= 0 ||
4664 raid_disks >= mddev->max_disks)
4666 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4668 mddev->delta_disks = raid_disks - mddev->raid_disks;
4670 rv = mddev->pers->check_reshape(mddev);
4676 * update_array_info is used to change the configuration of an
4678 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4679 * fields in the info are checked against the array.
4680 * Any differences that cannot be handled will cause an error.
4681 * Normally, only one change can be managed at a time.
4683 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4689 /* calculate expected state,ignoring low bits */
4690 if (mddev->bitmap && mddev->bitmap_offset)
4691 state |= (1 << MD_SB_BITMAP_PRESENT);
4693 if (mddev->major_version != info->major_version ||
4694 mddev->minor_version != info->minor_version ||
4695 /* mddev->patch_version != info->patch_version || */
4696 mddev->ctime != info->ctime ||
4697 mddev->level != info->level ||
4698 /* mddev->layout != info->layout || */
4699 !mddev->persistent != info->not_persistent||
4700 mddev->chunk_size != info->chunk_size ||
4701 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4702 ((state^info->state) & 0xfffffe00)
4705 /* Check there is only one change */
4706 if (info->size >= 0 && mddev->size != info->size) cnt++;
4707 if (mddev->raid_disks != info->raid_disks) cnt++;
4708 if (mddev->layout != info->layout) cnt++;
4709 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4710 if (cnt == 0) return 0;
4711 if (cnt > 1) return -EINVAL;
4713 if (mddev->layout != info->layout) {
4715 * we don't need to do anything at the md level, the
4716 * personality will take care of it all.
4718 if (mddev->pers->reconfig == NULL)
4721 return mddev->pers->reconfig(mddev, info->layout, -1);
4723 if (info->size >= 0 && mddev->size != info->size)
4724 rv = update_size(mddev, (sector_t)info->size * 2);
4726 if (mddev->raid_disks != info->raid_disks)
4727 rv = update_raid_disks(mddev, info->raid_disks);
4729 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4730 if (mddev->pers->quiesce == NULL)
4732 if (mddev->recovery || mddev->sync_thread)
4734 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4735 /* add the bitmap */
4738 if (mddev->default_bitmap_offset == 0)
4740 mddev->bitmap_offset = mddev->default_bitmap_offset;
4741 mddev->pers->quiesce(mddev, 1);
4742 rv = bitmap_create(mddev);
4744 bitmap_destroy(mddev);
4745 mddev->pers->quiesce(mddev, 0);
4747 /* remove the bitmap */
4750 if (mddev->bitmap->file)
4752 mddev->pers->quiesce(mddev, 1);
4753 bitmap_destroy(mddev);
4754 mddev->pers->quiesce(mddev, 0);
4755 mddev->bitmap_offset = 0;
4758 md_update_sb(mddev, 1);
4762 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4766 if (mddev->pers == NULL)
4769 rdev = find_rdev(mddev, dev);
4773 md_error(mddev, rdev);
4778 * We have a problem here : there is no easy way to give a CHS
4779 * virtual geometry. We currently pretend that we have a 2 heads
4780 * 4 sectors (with a BIG number of cylinders...). This drives
4781 * dosfs just mad... ;-)
4783 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4785 mddev_t *mddev = bdev->bd_disk->private_data;
4789 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4793 static int md_ioctl(struct inode *inode, struct file *file,
4794 unsigned int cmd, unsigned long arg)
4797 void __user *argp = (void __user *)arg;
4798 mddev_t *mddev = NULL;
4800 if (!capable(CAP_SYS_ADMIN))
4804 * Commands dealing with the RAID driver but not any
4810 err = get_version(argp);
4813 case PRINT_RAID_DEBUG:
4821 autostart_arrays(arg);
4828 * Commands creating/starting a new array:
4831 mddev = inode->i_bdev->bd_disk->private_data;
4838 err = mddev_lock(mddev);
4841 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4848 case SET_ARRAY_INFO:
4850 mdu_array_info_t info;
4852 memset(&info, 0, sizeof(info));
4853 else if (copy_from_user(&info, argp, sizeof(info))) {
4858 err = update_array_info(mddev, &info);
4860 printk(KERN_WARNING "md: couldn't update"
4861 " array info. %d\n", err);
4866 if (!list_empty(&mddev->disks)) {
4868 "md: array %s already has disks!\n",
4873 if (mddev->raid_disks) {
4875 "md: array %s already initialised!\n",
4880 err = set_array_info(mddev, &info);
4882 printk(KERN_WARNING "md: couldn't set"
4883 " array info. %d\n", err);
4893 * Commands querying/configuring an existing array:
4895 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4896 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4897 if ((!mddev->raid_disks && !mddev->external)
4898 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4899 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4900 && cmd != GET_BITMAP_FILE) {
4906 * Commands even a read-only array can execute:
4910 case GET_ARRAY_INFO:
4911 err = get_array_info(mddev, argp);
4914 case GET_BITMAP_FILE:
4915 err = get_bitmap_file(mddev, argp);
4919 err = get_disk_info(mddev, argp);
4922 case RESTART_ARRAY_RW:
4923 err = restart_array(mddev);
4927 err = do_md_stop (mddev, 0, 1);
4931 err = do_md_stop (mddev, 1, 1);
4937 * The remaining ioctls are changing the state of the
4938 * superblock, so we do not allow them on read-only arrays.
4939 * However non-MD ioctls (e.g. get-size) will still come through
4940 * here and hit the 'default' below, so only disallow
4941 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4943 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4944 if (mddev->ro == 2) {
4946 sysfs_notify(&mddev->kobj, NULL, "array_state");
4947 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4948 md_wakeup_thread(mddev->thread);
4959 mdu_disk_info_t info;
4960 if (copy_from_user(&info, argp, sizeof(info)))
4963 err = add_new_disk(mddev, &info);
4967 case HOT_REMOVE_DISK:
4968 err = hot_remove_disk(mddev, new_decode_dev(arg));
4972 err = hot_add_disk(mddev, new_decode_dev(arg));
4975 case SET_DISK_FAULTY:
4976 err = set_disk_faulty(mddev, new_decode_dev(arg));
4980 err = do_md_run (mddev);
4983 case SET_BITMAP_FILE:
4984 err = set_bitmap_file(mddev, (int)arg);
4994 mddev_unlock(mddev);
5004 static int md_open(struct inode *inode, struct file *file)
5007 * Succeed if we can lock the mddev, which confirms that
5008 * it isn't being stopped right now.
5010 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5013 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5018 mddev_unlock(mddev);
5020 check_disk_change(inode->i_bdev);
5025 static int md_release(struct inode *inode, struct file * file)
5027 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5035 static int md_media_changed(struct gendisk *disk)
5037 mddev_t *mddev = disk->private_data;
5039 return mddev->changed;
5042 static int md_revalidate(struct gendisk *disk)
5044 mddev_t *mddev = disk->private_data;
5049 static struct block_device_operations md_fops =
5051 .owner = THIS_MODULE,
5053 .release = md_release,
5055 .getgeo = md_getgeo,
5056 .media_changed = md_media_changed,
5057 .revalidate_disk= md_revalidate,
5060 static int md_thread(void * arg)
5062 mdk_thread_t *thread = arg;
5065 * md_thread is a 'system-thread', it's priority should be very
5066 * high. We avoid resource deadlocks individually in each
5067 * raid personality. (RAID5 does preallocation) We also use RR and
5068 * the very same RT priority as kswapd, thus we will never get
5069 * into a priority inversion deadlock.
5071 * we definitely have to have equal or higher priority than
5072 * bdflush, otherwise bdflush will deadlock if there are too
5073 * many dirty RAID5 blocks.
5076 allow_signal(SIGKILL);
5077 while (!kthread_should_stop()) {
5079 /* We need to wait INTERRUPTIBLE so that
5080 * we don't add to the load-average.
5081 * That means we need to be sure no signals are
5084 if (signal_pending(current))
5085 flush_signals(current);
5087 wait_event_interruptible_timeout
5089 test_bit(THREAD_WAKEUP, &thread->flags)
5090 || kthread_should_stop(),
5093 clear_bit(THREAD_WAKEUP, &thread->flags);
5095 thread->run(thread->mddev);
5101 void md_wakeup_thread(mdk_thread_t *thread)
5104 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5105 set_bit(THREAD_WAKEUP, &thread->flags);
5106 wake_up(&thread->wqueue);
5110 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5113 mdk_thread_t *thread;
5115 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5119 init_waitqueue_head(&thread->wqueue);
5122 thread->mddev = mddev;
5123 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5124 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5125 if (IS_ERR(thread->tsk)) {
5132 void md_unregister_thread(mdk_thread_t *thread)
5134 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5136 kthread_stop(thread->tsk);
5140 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5147 if (!rdev || test_bit(Faulty, &rdev->flags))
5150 if (mddev->external)
5151 set_bit(Blocked, &rdev->flags);
5153 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5155 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5156 __builtin_return_address(0),__builtin_return_address(1),
5157 __builtin_return_address(2),__builtin_return_address(3));
5161 if (!mddev->pers->error_handler)
5163 mddev->pers->error_handler(mddev,rdev);
5164 if (mddev->degraded)
5165 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5166 set_bit(StateChanged, &rdev->flags);
5167 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5168 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5169 md_wakeup_thread(mddev->thread);
5170 md_new_event_inintr(mddev);
5173 /* seq_file implementation /proc/mdstat */
5175 static void status_unused(struct seq_file *seq)
5179 struct list_head *tmp;
5181 seq_printf(seq, "unused devices: ");
5183 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5184 char b[BDEVNAME_SIZE];
5186 seq_printf(seq, "%s ",
5187 bdevname(rdev->bdev,b));
5190 seq_printf(seq, "<none>");
5192 seq_printf(seq, "\n");
5196 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5198 sector_t max_blocks, resync, res;
5199 unsigned long dt, db, rt;
5201 unsigned int per_milli;
5203 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5205 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5206 max_blocks = mddev->resync_max_sectors >> 1;
5208 max_blocks = mddev->size;
5211 * Should not happen.
5217 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5218 * in a sector_t, and (max_blocks>>scale) will fit in a
5219 * u32, as those are the requirements for sector_div.
5220 * Thus 'scale' must be at least 10
5223 if (sizeof(sector_t) > sizeof(unsigned long)) {
5224 while ( max_blocks/2 > (1ULL<<(scale+32)))
5227 res = (resync>>scale)*1000;
5228 sector_div(res, (u32)((max_blocks>>scale)+1));
5232 int i, x = per_milli/50, y = 20-x;
5233 seq_printf(seq, "[");
5234 for (i = 0; i < x; i++)
5235 seq_printf(seq, "=");
5236 seq_printf(seq, ">");
5237 for (i = 0; i < y; i++)
5238 seq_printf(seq, ".");
5239 seq_printf(seq, "] ");
5241 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5242 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5244 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5246 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5247 "resync" : "recovery"))),
5248 per_milli/10, per_milli % 10,
5249 (unsigned long long) resync,
5250 (unsigned long long) max_blocks);
5253 * We do not want to overflow, so the order of operands and
5254 * the * 100 / 100 trick are important. We do a +1 to be
5255 * safe against division by zero. We only estimate anyway.
5257 * dt: time from mark until now
5258 * db: blocks written from mark until now
5259 * rt: remaining time
5261 dt = ((jiffies - mddev->resync_mark) / HZ);
5263 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5264 - mddev->resync_mark_cnt;
5265 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5267 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5269 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5272 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5274 struct list_head *tmp;
5284 spin_lock(&all_mddevs_lock);
5285 list_for_each(tmp,&all_mddevs)
5287 mddev = list_entry(tmp, mddev_t, all_mddevs);
5289 spin_unlock(&all_mddevs_lock);
5292 spin_unlock(&all_mddevs_lock);
5294 return (void*)2;/* tail */
5298 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5300 struct list_head *tmp;
5301 mddev_t *next_mddev, *mddev = v;
5307 spin_lock(&all_mddevs_lock);
5309 tmp = all_mddevs.next;
5311 tmp = mddev->all_mddevs.next;
5312 if (tmp != &all_mddevs)
5313 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5315 next_mddev = (void*)2;
5318 spin_unlock(&all_mddevs_lock);
5326 static void md_seq_stop(struct seq_file *seq, void *v)
5330 if (mddev && v != (void*)1 && v != (void*)2)
5334 struct mdstat_info {
5338 static int md_seq_show(struct seq_file *seq, void *v)
5342 struct list_head *tmp2;
5344 struct mdstat_info *mi = seq->private;
5345 struct bitmap *bitmap;
5347 if (v == (void*)1) {
5348 struct mdk_personality *pers;
5349 seq_printf(seq, "Personalities : ");
5350 spin_lock(&pers_lock);
5351 list_for_each_entry(pers, &pers_list, list)
5352 seq_printf(seq, "[%s] ", pers->name);
5354 spin_unlock(&pers_lock);
5355 seq_printf(seq, "\n");
5356 mi->event = atomic_read(&md_event_count);
5359 if (v == (void*)2) {
5364 if (mddev_lock(mddev) < 0)
5367 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5368 seq_printf(seq, "%s : %sactive", mdname(mddev),
5369 mddev->pers ? "" : "in");
5372 seq_printf(seq, " (read-only)");
5374 seq_printf(seq, " (auto-read-only)");
5375 seq_printf(seq, " %s", mddev->pers->name);
5379 rdev_for_each(rdev, tmp2, mddev) {
5380 char b[BDEVNAME_SIZE];
5381 seq_printf(seq, " %s[%d]",
5382 bdevname(rdev->bdev,b), rdev->desc_nr);
5383 if (test_bit(WriteMostly, &rdev->flags))
5384 seq_printf(seq, "(W)");
5385 if (test_bit(Faulty, &rdev->flags)) {
5386 seq_printf(seq, "(F)");
5388 } else if (rdev->raid_disk < 0)
5389 seq_printf(seq, "(S)"); /* spare */
5393 if (!list_empty(&mddev->disks)) {
5395 seq_printf(seq, "\n %llu blocks",
5396 (unsigned long long)mddev->array_size);
5398 seq_printf(seq, "\n %llu blocks",
5399 (unsigned long long)size);
5401 if (mddev->persistent) {
5402 if (mddev->major_version != 0 ||
5403 mddev->minor_version != 90) {
5404 seq_printf(seq," super %d.%d",
5405 mddev->major_version,
5406 mddev->minor_version);
5408 } else if (mddev->external)
5409 seq_printf(seq, " super external:%s",
5410 mddev->metadata_type);
5412 seq_printf(seq, " super non-persistent");
5415 mddev->pers->status (seq, mddev);
5416 seq_printf(seq, "\n ");
5417 if (mddev->pers->sync_request) {
5418 if (mddev->curr_resync > 2) {
5419 status_resync (seq, mddev);
5420 seq_printf(seq, "\n ");
5421 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5422 seq_printf(seq, "\tresync=DELAYED\n ");
5423 else if (mddev->recovery_cp < MaxSector)
5424 seq_printf(seq, "\tresync=PENDING\n ");
5427 seq_printf(seq, "\n ");
5429 if ((bitmap = mddev->bitmap)) {
5430 unsigned long chunk_kb;
5431 unsigned long flags;
5432 spin_lock_irqsave(&bitmap->lock, flags);
5433 chunk_kb = bitmap->chunksize >> 10;
5434 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5436 bitmap->pages - bitmap->missing_pages,
5438 (bitmap->pages - bitmap->missing_pages)
5439 << (PAGE_SHIFT - 10),
5440 chunk_kb ? chunk_kb : bitmap->chunksize,
5441 chunk_kb ? "KB" : "B");
5443 seq_printf(seq, ", file: ");
5444 seq_path(seq, &bitmap->file->f_path, " \t\n");
5447 seq_printf(seq, "\n");
5448 spin_unlock_irqrestore(&bitmap->lock, flags);
5451 seq_printf(seq, "\n");
5453 mddev_unlock(mddev);
5458 static struct seq_operations md_seq_ops = {
5459 .start = md_seq_start,
5460 .next = md_seq_next,
5461 .stop = md_seq_stop,
5462 .show = md_seq_show,
5465 static int md_seq_open(struct inode *inode, struct file *file)
5468 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5472 error = seq_open(file, &md_seq_ops);
5476 struct seq_file *p = file->private_data;
5478 mi->event = atomic_read(&md_event_count);
5483 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5485 struct seq_file *m = filp->private_data;
5486 struct mdstat_info *mi = m->private;
5489 poll_wait(filp, &md_event_waiters, wait);
5491 /* always allow read */
5492 mask = POLLIN | POLLRDNORM;
5494 if (mi->event != atomic_read(&md_event_count))
5495 mask |= POLLERR | POLLPRI;
5499 static const struct file_operations md_seq_fops = {
5500 .owner = THIS_MODULE,
5501 .open = md_seq_open,
5503 .llseek = seq_lseek,
5504 .release = seq_release_private,
5505 .poll = mdstat_poll,
5508 int register_md_personality(struct mdk_personality *p)
5510 spin_lock(&pers_lock);
5511 list_add_tail(&p->list, &pers_list);
5512 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5513 spin_unlock(&pers_lock);
5517 int unregister_md_personality(struct mdk_personality *p)
5519 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5520 spin_lock(&pers_lock);
5521 list_del_init(&p->list);
5522 spin_unlock(&pers_lock);
5526 static int is_mddev_idle(mddev_t *mddev)
5529 struct list_head *tmp;
5534 rdev_for_each(rdev, tmp, mddev) {
5535 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5536 curr_events = disk_stat_read(disk, sectors[0]) +
5537 disk_stat_read(disk, sectors[1]) -
5538 atomic_read(&disk->sync_io);
5539 /* sync IO will cause sync_io to increase before the disk_stats
5540 * as sync_io is counted when a request starts, and
5541 * disk_stats is counted when it completes.
5542 * So resync activity will cause curr_events to be smaller than
5543 * when there was no such activity.
5544 * non-sync IO will cause disk_stat to increase without
5545 * increasing sync_io so curr_events will (eventually)
5546 * be larger than it was before. Once it becomes
5547 * substantially larger, the test below will cause
5548 * the array to appear non-idle, and resync will slow
5550 * If there is a lot of outstanding resync activity when
5551 * we set last_event to curr_events, then all that activity
5552 * completing might cause the array to appear non-idle
5553 * and resync will be slowed down even though there might
5554 * not have been non-resync activity. This will only
5555 * happen once though. 'last_events' will soon reflect
5556 * the state where there is little or no outstanding
5557 * resync requests, and further resync activity will
5558 * always make curr_events less than last_events.
5561 if (curr_events - rdev->last_events > 4096) {
5562 rdev->last_events = curr_events;
5569 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5571 /* another "blocks" (512byte) blocks have been synced */
5572 atomic_sub(blocks, &mddev->recovery_active);
5573 wake_up(&mddev->recovery_wait);
5575 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5576 md_wakeup_thread(mddev->thread);
5577 // stop recovery, signal do_sync ....
5582 /* md_write_start(mddev, bi)
5583 * If we need to update some array metadata (e.g. 'active' flag
5584 * in superblock) before writing, schedule a superblock update
5585 * and wait for it to complete.
5587 void md_write_start(mddev_t *mddev, struct bio *bi)
5590 if (bio_data_dir(bi) != WRITE)
5593 BUG_ON(mddev->ro == 1);
5594 if (mddev->ro == 2) {
5595 /* need to switch to read/write */
5597 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5598 md_wakeup_thread(mddev->thread);
5599 md_wakeup_thread(mddev->sync_thread);
5602 atomic_inc(&mddev->writes_pending);
5603 if (mddev->safemode == 1)
5604 mddev->safemode = 0;
5605 if (mddev->in_sync) {
5606 spin_lock_irq(&mddev->write_lock);
5607 if (mddev->in_sync) {
5609 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5610 md_wakeup_thread(mddev->thread);
5613 spin_unlock_irq(&mddev->write_lock);
5616 sysfs_notify(&mddev->kobj, NULL, "array_state");
5617 wait_event(mddev->sb_wait,
5618 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5619 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5622 void md_write_end(mddev_t *mddev)
5624 if (atomic_dec_and_test(&mddev->writes_pending)) {
5625 if (mddev->safemode == 2)
5626 md_wakeup_thread(mddev->thread);
5627 else if (mddev->safemode_delay)
5628 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5632 /* md_allow_write(mddev)
5633 * Calling this ensures that the array is marked 'active' so that writes
5634 * may proceed without blocking. It is important to call this before
5635 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5636 * Must be called with mddev_lock held.
5638 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5639 * is dropped, so return -EAGAIN after notifying userspace.
5641 int md_allow_write(mddev_t *mddev)
5647 if (!mddev->pers->sync_request)
5650 spin_lock_irq(&mddev->write_lock);
5651 if (mddev->in_sync) {
5653 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5654 if (mddev->safemode_delay &&
5655 mddev->safemode == 0)
5656 mddev->safemode = 1;
5657 spin_unlock_irq(&mddev->write_lock);
5658 md_update_sb(mddev, 0);
5659 sysfs_notify(&mddev->kobj, NULL, "array_state");
5661 spin_unlock_irq(&mddev->write_lock);
5663 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5668 EXPORT_SYMBOL_GPL(md_allow_write);
5670 #define SYNC_MARKS 10
5671 #define SYNC_MARK_STEP (3*HZ)
5672 void md_do_sync(mddev_t *mddev)
5675 unsigned int currspeed = 0,
5677 sector_t max_sectors,j, io_sectors;
5678 unsigned long mark[SYNC_MARKS];
5679 sector_t mark_cnt[SYNC_MARKS];
5681 struct list_head *tmp;
5682 sector_t last_check;
5684 struct list_head *rtmp;
5688 /* just incase thread restarts... */
5689 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5691 if (mddev->ro) /* never try to sync a read-only array */
5694 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5695 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5696 desc = "data-check";
5697 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5698 desc = "requested-resync";
5701 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5706 /* we overload curr_resync somewhat here.
5707 * 0 == not engaged in resync at all
5708 * 2 == checking that there is no conflict with another sync
5709 * 1 == like 2, but have yielded to allow conflicting resync to
5711 * other == active in resync - this many blocks
5713 * Before starting a resync we must have set curr_resync to
5714 * 2, and then checked that every "conflicting" array has curr_resync
5715 * less than ours. When we find one that is the same or higher
5716 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5717 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5718 * This will mean we have to start checking from the beginning again.
5723 mddev->curr_resync = 2;
5726 if (kthread_should_stop()) {
5727 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5730 for_each_mddev(mddev2, tmp) {
5731 if (mddev2 == mddev)
5733 if (!mddev->parallel_resync
5734 && mddev2->curr_resync
5735 && match_mddev_units(mddev, mddev2)) {
5737 if (mddev < mddev2 && mddev->curr_resync == 2) {
5738 /* arbitrarily yield */
5739 mddev->curr_resync = 1;
5740 wake_up(&resync_wait);
5742 if (mddev > mddev2 && mddev->curr_resync == 1)
5743 /* no need to wait here, we can wait the next
5744 * time 'round when curr_resync == 2
5747 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5748 if (!kthread_should_stop() &&
5749 mddev2->curr_resync >= mddev->curr_resync) {
5750 printk(KERN_INFO "md: delaying %s of %s"
5751 " until %s has finished (they"
5752 " share one or more physical units)\n",
5753 desc, mdname(mddev), mdname(mddev2));
5756 finish_wait(&resync_wait, &wq);
5759 finish_wait(&resync_wait, &wq);
5762 } while (mddev->curr_resync < 2);
5765 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5766 /* resync follows the size requested by the personality,
5767 * which defaults to physical size, but can be virtual size
5769 max_sectors = mddev->resync_max_sectors;
5770 mddev->resync_mismatches = 0;
5771 /* we don't use the checkpoint if there's a bitmap */
5772 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5773 j = mddev->resync_min;
5774 else if (!mddev->bitmap)
5775 j = mddev->recovery_cp;
5777 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5778 max_sectors = mddev->size << 1;
5780 /* recovery follows the physical size of devices */
5781 max_sectors = mddev->size << 1;
5783 rdev_for_each(rdev, rtmp, mddev)
5784 if (rdev->raid_disk >= 0 &&
5785 !test_bit(Faulty, &rdev->flags) &&
5786 !test_bit(In_sync, &rdev->flags) &&
5787 rdev->recovery_offset < j)
5788 j = rdev->recovery_offset;
5791 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5792 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5793 " %d KB/sec/disk.\n", speed_min(mddev));
5794 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5795 "(but not more than %d KB/sec) for %s.\n",
5796 speed_max(mddev), desc);
5798 is_mddev_idle(mddev); /* this also initializes IO event counters */
5801 for (m = 0; m < SYNC_MARKS; m++) {
5803 mark_cnt[m] = io_sectors;
5806 mddev->resync_mark = mark[last_mark];
5807 mddev->resync_mark_cnt = mark_cnt[last_mark];
5810 * Tune reconstruction:
5812 window = 32*(PAGE_SIZE/512);
5813 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5814 window/2,(unsigned long long) max_sectors/2);
5816 atomic_set(&mddev->recovery_active, 0);
5821 "md: resuming %s of %s from checkpoint.\n",
5822 desc, mdname(mddev));
5823 mddev->curr_resync = j;
5826 while (j < max_sectors) {
5830 if (j >= mddev->resync_max) {
5831 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5832 wait_event(mddev->recovery_wait,
5833 mddev->resync_max > j
5834 || kthread_should_stop());
5836 if (kthread_should_stop())
5838 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5839 currspeed < speed_min(mddev));
5841 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5845 if (!skipped) { /* actual IO requested */
5846 io_sectors += sectors;
5847 atomic_add(sectors, &mddev->recovery_active);
5851 if (j>1) mddev->curr_resync = j;
5852 mddev->curr_mark_cnt = io_sectors;
5853 if (last_check == 0)
5854 /* this is the earliers that rebuilt will be
5855 * visible in /proc/mdstat
5857 md_new_event(mddev);
5859 if (last_check + window > io_sectors || j == max_sectors)
5862 last_check = io_sectors;
5864 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5868 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5870 int next = (last_mark+1) % SYNC_MARKS;
5872 mddev->resync_mark = mark[next];
5873 mddev->resync_mark_cnt = mark_cnt[next];
5874 mark[next] = jiffies;
5875 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5880 if (kthread_should_stop())
5885 * this loop exits only if either when we are slower than
5886 * the 'hard' speed limit, or the system was IO-idle for
5888 * the system might be non-idle CPU-wise, but we only care
5889 * about not overloading the IO subsystem. (things like an
5890 * e2fsck being done on the RAID array should execute fast)
5892 blk_unplug(mddev->queue);
5895 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5896 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5898 if (currspeed > speed_min(mddev)) {
5899 if ((currspeed > speed_max(mddev)) ||
5900 !is_mddev_idle(mddev)) {
5906 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5908 * this also signals 'finished resyncing' to md_stop
5911 blk_unplug(mddev->queue);
5913 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5915 /* tell personality that we are finished */
5916 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5918 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5919 mddev->curr_resync > 2) {
5920 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5921 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5922 if (mddev->curr_resync >= mddev->recovery_cp) {
5924 "md: checkpointing %s of %s.\n",
5925 desc, mdname(mddev));
5926 mddev->recovery_cp = mddev->curr_resync;
5929 mddev->recovery_cp = MaxSector;
5931 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5932 mddev->curr_resync = MaxSector;
5933 rdev_for_each(rdev, rtmp, mddev)
5934 if (rdev->raid_disk >= 0 &&
5935 !test_bit(Faulty, &rdev->flags) &&
5936 !test_bit(In_sync, &rdev->flags) &&
5937 rdev->recovery_offset < mddev->curr_resync)
5938 rdev->recovery_offset = mddev->curr_resync;
5941 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5944 mddev->curr_resync = 0;
5945 mddev->resync_min = 0;
5946 mddev->resync_max = MaxSector;
5947 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5948 wake_up(&resync_wait);
5949 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5950 md_wakeup_thread(mddev->thread);
5955 * got a signal, exit.
5958 "md: md_do_sync() got signal ... exiting\n");
5959 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5963 EXPORT_SYMBOL_GPL(md_do_sync);
5966 static int remove_and_add_spares(mddev_t *mddev)
5969 struct list_head *rtmp;
5972 rdev_for_each(rdev, rtmp, mddev)
5973 if (rdev->raid_disk >= 0 &&
5974 !test_bit(Blocked, &rdev->flags) &&
5975 (test_bit(Faulty, &rdev->flags) ||
5976 ! test_bit(In_sync, &rdev->flags)) &&
5977 atomic_read(&rdev->nr_pending)==0) {
5978 if (mddev->pers->hot_remove_disk(
5979 mddev, rdev->raid_disk)==0) {
5981 sprintf(nm,"rd%d", rdev->raid_disk);
5982 sysfs_remove_link(&mddev->kobj, nm);
5983 rdev->raid_disk = -1;
5987 if (mddev->degraded) {
5988 rdev_for_each(rdev, rtmp, mddev) {
5989 if (rdev->raid_disk >= 0 &&
5990 !test_bit(In_sync, &rdev->flags))
5992 if (rdev->raid_disk < 0
5993 && !test_bit(Faulty, &rdev->flags)) {
5994 rdev->recovery_offset = 0;
5996 hot_add_disk(mddev, rdev) == 0) {
5998 sprintf(nm, "rd%d", rdev->raid_disk);
5999 if (sysfs_create_link(&mddev->kobj,
6002 "md: cannot register "
6006 md_new_event(mddev);
6015 * This routine is regularly called by all per-raid-array threads to
6016 * deal with generic issues like resync and super-block update.
6017 * Raid personalities that don't have a thread (linear/raid0) do not
6018 * need this as they never do any recovery or update the superblock.
6020 * It does not do any resync itself, but rather "forks" off other threads
6021 * to do that as needed.
6022 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6023 * "->recovery" and create a thread at ->sync_thread.
6024 * When the thread finishes it sets MD_RECOVERY_DONE
6025 * and wakeups up this thread which will reap the thread and finish up.
6026 * This thread also removes any faulty devices (with nr_pending == 0).
6028 * The overall approach is:
6029 * 1/ if the superblock needs updating, update it.
6030 * 2/ If a recovery thread is running, don't do anything else.
6031 * 3/ If recovery has finished, clean up, possibly marking spares active.
6032 * 4/ If there are any faulty devices, remove them.
6033 * 5/ If array is degraded, try to add spares devices
6034 * 6/ If array has spares or is not in-sync, start a resync thread.
6036 void md_check_recovery(mddev_t *mddev)
6039 struct list_head *rtmp;
6043 bitmap_daemon_work(mddev->bitmap);
6048 if (signal_pending(current)) {
6049 if (mddev->pers->sync_request && !mddev->external) {
6050 printk(KERN_INFO "md: %s in immediate safe mode\n",
6052 mddev->safemode = 2;
6054 flush_signals(current);
6058 (mddev->flags && !mddev->external) ||
6059 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6060 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6061 (mddev->external == 0 && mddev->safemode == 1) ||
6062 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6063 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6067 if (mddev_trylock(mddev)) {
6070 if (!mddev->external) {
6072 spin_lock_irq(&mddev->write_lock);
6073 if (mddev->safemode &&
6074 !atomic_read(&mddev->writes_pending) &&
6076 mddev->recovery_cp == MaxSector) {
6079 if (mddev->persistent)
6080 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6082 if (mddev->safemode == 1)
6083 mddev->safemode = 0;
6084 spin_unlock_irq(&mddev->write_lock);
6086 sysfs_notify(&mddev->kobj, NULL, "array_state");
6090 md_update_sb(mddev, 0);
6092 rdev_for_each(rdev, rtmp, mddev)
6093 if (test_and_clear_bit(StateChanged, &rdev->flags))
6094 sysfs_notify(&rdev->kobj, NULL, "state");
6097 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6098 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6099 /* resync/recovery still happening */
6100 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6103 if (mddev->sync_thread) {
6104 /* resync has finished, collect result */
6105 md_unregister_thread(mddev->sync_thread);
6106 mddev->sync_thread = NULL;
6107 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6109 /* activate any spares */
6110 if (mddev->pers->spare_active(mddev))
6111 sysfs_notify(&mddev->kobj, NULL,
6114 md_update_sb(mddev, 1);
6116 /* if array is no-longer degraded, then any saved_raid_disk
6117 * information must be scrapped
6119 if (!mddev->degraded)
6120 rdev_for_each(rdev, rtmp, mddev)
6121 rdev->saved_raid_disk = -1;
6123 mddev->recovery = 0;
6124 /* flag recovery needed just to double check */
6125 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6126 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6127 md_new_event(mddev);
6130 /* Set RUNNING before clearing NEEDED to avoid
6131 * any transients in the value of "sync_action".
6133 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6134 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6135 /* Clear some bits that don't mean anything, but
6138 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6139 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6141 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6143 /* no recovery is running.
6144 * remove any failed drives, then
6145 * add spares if possible.
6146 * Spare are also removed and re-added, to allow
6147 * the personality to fail the re-add.
6150 if (mddev->reshape_position != MaxSector) {
6151 if (mddev->pers->check_reshape(mddev) != 0)
6152 /* Cannot proceed */
6154 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6155 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6156 } else if ((spares = remove_and_add_spares(mddev))) {
6157 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6158 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6159 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6160 } else if (mddev->recovery_cp < MaxSector) {
6161 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6162 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6163 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6164 /* nothing to be done ... */
6167 if (mddev->pers->sync_request) {
6168 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6169 /* We are adding a device or devices to an array
6170 * which has the bitmap stored on all devices.
6171 * So make sure all bitmap pages get written
6173 bitmap_write_all(mddev->bitmap);
6175 mddev->sync_thread = md_register_thread(md_do_sync,
6178 if (!mddev->sync_thread) {
6179 printk(KERN_ERR "%s: could not start resync"
6182 /* leave the spares where they are, it shouldn't hurt */
6183 mddev->recovery = 0;
6185 md_wakeup_thread(mddev->sync_thread);
6186 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6187 md_new_event(mddev);
6190 if (!mddev->sync_thread) {
6191 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6192 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6194 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6196 mddev_unlock(mddev);
6200 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6202 sysfs_notify(&rdev->kobj, NULL, "state");
6203 wait_event_timeout(rdev->blocked_wait,
6204 !test_bit(Blocked, &rdev->flags),
6205 msecs_to_jiffies(5000));
6206 rdev_dec_pending(rdev, mddev);
6208 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6210 static int md_notify_reboot(struct notifier_block *this,
6211 unsigned long code, void *x)
6213 struct list_head *tmp;
6216 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6218 printk(KERN_INFO "md: stopping all md devices.\n");
6220 for_each_mddev(mddev, tmp)
6221 if (mddev_trylock(mddev)) {
6222 do_md_stop (mddev, 1, 0);
6223 mddev_unlock(mddev);
6226 * certain more exotic SCSI devices are known to be
6227 * volatile wrt too early system reboots. While the
6228 * right place to handle this issue is the given
6229 * driver, we do want to have a safe RAID driver ...
6236 static struct notifier_block md_notifier = {
6237 .notifier_call = md_notify_reboot,
6239 .priority = INT_MAX, /* before any real devices */
6242 static void md_geninit(void)
6244 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6246 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6249 static int __init md_init(void)
6251 if (register_blkdev(MAJOR_NR, "md"))
6253 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6254 unregister_blkdev(MAJOR_NR, "md");
6257 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6258 md_probe, NULL, NULL);
6259 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6260 md_probe, NULL, NULL);
6262 register_reboot_notifier(&md_notifier);
6263 raid_table_header = register_sysctl_table(raid_root_table);
6273 * Searches all registered partitions for autorun RAID arrays
6277 static LIST_HEAD(all_detected_devices);
6278 struct detected_devices_node {
6279 struct list_head list;
6283 void md_autodetect_dev(dev_t dev)
6285 struct detected_devices_node *node_detected_dev;
6287 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6288 if (node_detected_dev) {
6289 node_detected_dev->dev = dev;
6290 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6292 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6293 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6298 static void autostart_arrays(int part)
6301 struct detected_devices_node *node_detected_dev;
6303 int i_scanned, i_passed;
6308 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6310 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6312 node_detected_dev = list_entry(all_detected_devices.next,
6313 struct detected_devices_node, list);
6314 list_del(&node_detected_dev->list);
6315 dev = node_detected_dev->dev;
6316 kfree(node_detected_dev);
6317 rdev = md_import_device(dev,0, 90);
6321 if (test_bit(Faulty, &rdev->flags)) {
6325 set_bit(AutoDetected, &rdev->flags);
6326 list_add(&rdev->same_set, &pending_raid_disks);
6330 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6331 i_scanned, i_passed);
6333 autorun_devices(part);
6336 #endif /* !MODULE */
6338 static __exit void md_exit(void)
6341 struct list_head *tmp;
6343 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6344 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6346 unregister_blkdev(MAJOR_NR,"md");
6347 unregister_blkdev(mdp_major, "mdp");
6348 unregister_reboot_notifier(&md_notifier);
6349 unregister_sysctl_table(raid_table_header);
6350 remove_proc_entry("mdstat", NULL);
6351 for_each_mddev(mddev, tmp) {
6352 struct gendisk *disk = mddev->gendisk;
6355 export_array(mddev);
6358 mddev->gendisk = NULL;
6363 subsys_initcall(md_init);
6364 module_exit(md_exit)
6366 static int get_ro(char *buffer, struct kernel_param *kp)
6368 return sprintf(buffer, "%d", start_readonly);
6370 static int set_ro(const char *val, struct kernel_param *kp)
6373 int num = simple_strtoul(val, &e, 10);
6374 if (*val && (*e == '\0' || *e == '\n')) {
6375 start_readonly = num;
6381 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6382 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6385 EXPORT_SYMBOL(register_md_personality);
6386 EXPORT_SYMBOL(unregister_md_personality);
6387 EXPORT_SYMBOL(md_error);
6388 EXPORT_SYMBOL(md_done_sync);
6389 EXPORT_SYMBOL(md_write_start);
6390 EXPORT_SYMBOL(md_write_end);
6391 EXPORT_SYMBOL(md_register_thread);
6392 EXPORT_SYMBOL(md_unregister_thread);
6393 EXPORT_SYMBOL(md_wakeup_thread);
6394 EXPORT_SYMBOL(md_check_recovery);
6395 MODULE_LICENSE("GPL");
6397 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);