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 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
352 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
353 return MD_NEW_SIZE_BLOCKS(size);
356 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
360 size = rdev->sb_offset;
363 size &= ~((sector_t)chunk_size/1024 - 1);
367 static int alloc_disk_sb(mdk_rdev_t * rdev)
372 rdev->sb_page = alloc_page(GFP_KERNEL);
373 if (!rdev->sb_page) {
374 printk(KERN_ALERT "md: out of memory.\n");
381 static void free_disk_sb(mdk_rdev_t * rdev)
384 put_page(rdev->sb_page);
386 rdev->sb_page = NULL;
393 static void super_written(struct bio *bio, int error)
395 mdk_rdev_t *rdev = bio->bi_private;
396 mddev_t *mddev = rdev->mddev;
398 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
399 printk("md: super_written gets error=%d, uptodate=%d\n",
400 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
401 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
402 md_error(mddev, rdev);
405 if (atomic_dec_and_test(&mddev->pending_writes))
406 wake_up(&mddev->sb_wait);
410 static void super_written_barrier(struct bio *bio, int error)
412 struct bio *bio2 = bio->bi_private;
413 mdk_rdev_t *rdev = bio2->bi_private;
414 mddev_t *mddev = rdev->mddev;
416 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
417 error == -EOPNOTSUPP) {
419 /* barriers don't appear to be supported :-( */
420 set_bit(BarriersNotsupp, &rdev->flags);
421 mddev->barriers_work = 0;
422 spin_lock_irqsave(&mddev->write_lock, flags);
423 bio2->bi_next = mddev->biolist;
424 mddev->biolist = bio2;
425 spin_unlock_irqrestore(&mddev->write_lock, flags);
426 wake_up(&mddev->sb_wait);
430 bio->bi_private = rdev;
431 super_written(bio, error);
435 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
436 sector_t sector, int size, struct page *page)
438 /* write first size bytes of page to sector of rdev
439 * Increment mddev->pending_writes before returning
440 * and decrement it on completion, waking up sb_wait
441 * if zero is reached.
442 * If an error occurred, call md_error
444 * As we might need to resubmit the request if BIO_RW_BARRIER
445 * causes ENOTSUPP, we allocate a spare bio...
447 struct bio *bio = bio_alloc(GFP_NOIO, 1);
448 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
450 bio->bi_bdev = rdev->bdev;
451 bio->bi_sector = sector;
452 bio_add_page(bio, page, size, 0);
453 bio->bi_private = rdev;
454 bio->bi_end_io = super_written;
457 atomic_inc(&mddev->pending_writes);
458 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
460 rw |= (1<<BIO_RW_BARRIER);
461 rbio = bio_clone(bio, GFP_NOIO);
462 rbio->bi_private = bio;
463 rbio->bi_end_io = super_written_barrier;
464 submit_bio(rw, rbio);
469 void md_super_wait(mddev_t *mddev)
471 /* wait for all superblock writes that were scheduled to complete.
472 * if any had to be retried (due to BARRIER problems), retry them
476 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
477 if (atomic_read(&mddev->pending_writes)==0)
479 while (mddev->biolist) {
481 spin_lock_irq(&mddev->write_lock);
482 bio = mddev->biolist;
483 mddev->biolist = bio->bi_next ;
485 spin_unlock_irq(&mddev->write_lock);
486 submit_bio(bio->bi_rw, bio);
490 finish_wait(&mddev->sb_wait, &wq);
493 static void bi_complete(struct bio *bio, int error)
495 complete((struct completion*)bio->bi_private);
498 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
499 struct page *page, int rw)
501 struct bio *bio = bio_alloc(GFP_NOIO, 1);
502 struct completion event;
505 rw |= (1 << BIO_RW_SYNC);
508 bio->bi_sector = sector;
509 bio_add_page(bio, page, size, 0);
510 init_completion(&event);
511 bio->bi_private = &event;
512 bio->bi_end_io = bi_complete;
514 wait_for_completion(&event);
516 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
520 EXPORT_SYMBOL_GPL(sync_page_io);
522 static int read_disk_sb(mdk_rdev_t * rdev, int size)
524 char b[BDEVNAME_SIZE];
525 if (!rdev->sb_page) {
533 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
539 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
540 bdevname(rdev->bdev,b));
544 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
546 return sb1->set_uuid0 == sb2->set_uuid0 &&
547 sb1->set_uuid1 == sb2->set_uuid1 &&
548 sb1->set_uuid2 == sb2->set_uuid2 &&
549 sb1->set_uuid3 == sb2->set_uuid3;
552 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
555 mdp_super_t *tmp1, *tmp2;
557 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
558 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
560 if (!tmp1 || !tmp2) {
562 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
570 * nr_disks is not constant
575 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
583 static u32 md_csum_fold(u32 csum)
585 csum = (csum & 0xffff) + (csum >> 16);
586 return (csum & 0xffff) + (csum >> 16);
589 static unsigned int calc_sb_csum(mdp_super_t * sb)
592 u32 *sb32 = (u32*)sb;
594 unsigned int disk_csum, csum;
596 disk_csum = sb->sb_csum;
599 for (i = 0; i < MD_SB_BYTES/4 ; i++)
601 csum = (newcsum & 0xffffffff) + (newcsum>>32);
605 /* This used to use csum_partial, which was wrong for several
606 * reasons including that different results are returned on
607 * different architectures. It isn't critical that we get exactly
608 * the same return value as before (we always csum_fold before
609 * testing, and that removes any differences). However as we
610 * know that csum_partial always returned a 16bit value on
611 * alphas, do a fold to maximise conformity to previous behaviour.
613 sb->sb_csum = md_csum_fold(disk_csum);
615 sb->sb_csum = disk_csum;
622 * Handle superblock details.
623 * We want to be able to handle multiple superblock formats
624 * so we have a common interface to them all, and an array of
625 * different handlers.
626 * We rely on user-space to write the initial superblock, and support
627 * reading and updating of superblocks.
628 * Interface methods are:
629 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
630 * loads and validates a superblock on dev.
631 * if refdev != NULL, compare superblocks on both devices
633 * 0 - dev has a superblock that is compatible with refdev
634 * 1 - dev has a superblock that is compatible and newer than refdev
635 * so dev should be used as the refdev in future
636 * -EINVAL superblock incompatible or invalid
637 * -othererror e.g. -EIO
639 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
640 * Verify that dev is acceptable into mddev.
641 * The first time, mddev->raid_disks will be 0, and data from
642 * dev should be merged in. Subsequent calls check that dev
643 * is new enough. Return 0 or -EINVAL
645 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
646 * Update the superblock for rdev with data in mddev
647 * This does not write to disc.
653 struct module *owner;
654 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
656 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
657 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
658 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
659 unsigned long long size);
663 * load_super for 0.90.0
665 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
667 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
673 * Calculate the position of the superblock,
674 * it's at the end of the disk.
676 * It also happens to be a multiple of 4Kb.
678 sb_offset = calc_dev_sboffset(rdev->bdev);
679 rdev->sb_offset = sb_offset;
681 ret = read_disk_sb(rdev, MD_SB_BYTES);
686 bdevname(rdev->bdev, b);
687 sb = (mdp_super_t*)page_address(rdev->sb_page);
689 if (sb->md_magic != MD_SB_MAGIC) {
690 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
695 if (sb->major_version != 0 ||
696 sb->minor_version < 90 ||
697 sb->minor_version > 91) {
698 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
699 sb->major_version, sb->minor_version,
704 if (sb->raid_disks <= 0)
707 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
708 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
713 rdev->preferred_minor = sb->md_minor;
714 rdev->data_offset = 0;
715 rdev->sb_size = MD_SB_BYTES;
717 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
718 if (sb->level != 1 && sb->level != 4
719 && sb->level != 5 && sb->level != 6
720 && sb->level != 10) {
721 /* FIXME use a better test */
723 "md: bitmaps not supported for this level.\n");
728 if (sb->level == LEVEL_MULTIPATH)
731 rdev->desc_nr = sb->this_disk.number;
737 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
738 if (!uuid_equal(refsb, sb)) {
739 printk(KERN_WARNING "md: %s has different UUID to %s\n",
740 b, bdevname(refdev->bdev,b2));
743 if (!sb_equal(refsb, sb)) {
744 printk(KERN_WARNING "md: %s has same UUID"
745 " but different superblock to %s\n",
746 b, bdevname(refdev->bdev, b2));
750 ev2 = md_event(refsb);
756 rdev->size = calc_dev_size(rdev, sb->chunk_size);
758 if (rdev->size < sb->size && sb->level > 1)
759 /* "this cannot possibly happen" ... */
767 * validate_super for 0.90.0
769 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
772 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
773 __u64 ev1 = md_event(sb);
775 rdev->raid_disk = -1;
776 clear_bit(Faulty, &rdev->flags);
777 clear_bit(In_sync, &rdev->flags);
778 clear_bit(WriteMostly, &rdev->flags);
779 clear_bit(BarriersNotsupp, &rdev->flags);
781 if (mddev->raid_disks == 0) {
782 mddev->major_version = 0;
783 mddev->minor_version = sb->minor_version;
784 mddev->patch_version = sb->patch_version;
786 mddev->chunk_size = sb->chunk_size;
787 mddev->ctime = sb->ctime;
788 mddev->utime = sb->utime;
789 mddev->level = sb->level;
790 mddev->clevel[0] = 0;
791 mddev->layout = sb->layout;
792 mddev->raid_disks = sb->raid_disks;
793 mddev->size = sb->size;
795 mddev->bitmap_offset = 0;
796 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
798 if (mddev->minor_version >= 91) {
799 mddev->reshape_position = sb->reshape_position;
800 mddev->delta_disks = sb->delta_disks;
801 mddev->new_level = sb->new_level;
802 mddev->new_layout = sb->new_layout;
803 mddev->new_chunk = sb->new_chunk;
805 mddev->reshape_position = MaxSector;
806 mddev->delta_disks = 0;
807 mddev->new_level = mddev->level;
808 mddev->new_layout = mddev->layout;
809 mddev->new_chunk = mddev->chunk_size;
812 if (sb->state & (1<<MD_SB_CLEAN))
813 mddev->recovery_cp = MaxSector;
815 if (sb->events_hi == sb->cp_events_hi &&
816 sb->events_lo == sb->cp_events_lo) {
817 mddev->recovery_cp = sb->recovery_cp;
819 mddev->recovery_cp = 0;
822 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
823 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
824 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
825 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
827 mddev->max_disks = MD_SB_DISKS;
829 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
830 mddev->bitmap_file == NULL)
831 mddev->bitmap_offset = mddev->default_bitmap_offset;
833 } else if (mddev->pers == NULL) {
834 /* Insist on good event counter while assembling */
836 if (ev1 < mddev->events)
838 } else if (mddev->bitmap) {
839 /* if adding to array with a bitmap, then we can accept an
840 * older device ... but not too old.
842 if (ev1 < mddev->bitmap->events_cleared)
845 if (ev1 < mddev->events)
846 /* just a hot-add of a new device, leave raid_disk at -1 */
850 if (mddev->level != LEVEL_MULTIPATH) {
851 desc = sb->disks + rdev->desc_nr;
853 if (desc->state & (1<<MD_DISK_FAULTY))
854 set_bit(Faulty, &rdev->flags);
855 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
856 desc->raid_disk < mddev->raid_disks */) {
857 set_bit(In_sync, &rdev->flags);
858 rdev->raid_disk = desc->raid_disk;
860 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
861 set_bit(WriteMostly, &rdev->flags);
862 } else /* MULTIPATH are always insync */
863 set_bit(In_sync, &rdev->flags);
868 * sync_super for 0.90.0
870 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
873 struct list_head *tmp;
875 int next_spare = mddev->raid_disks;
878 /* make rdev->sb match mddev data..
881 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
882 * 3/ any empty disks < next_spare become removed
884 * disks[0] gets initialised to REMOVED because
885 * we cannot be sure from other fields if it has
886 * been initialised or not.
889 int active=0, working=0,failed=0,spare=0,nr_disks=0;
891 rdev->sb_size = MD_SB_BYTES;
893 sb = (mdp_super_t*)page_address(rdev->sb_page);
895 memset(sb, 0, sizeof(*sb));
897 sb->md_magic = MD_SB_MAGIC;
898 sb->major_version = mddev->major_version;
899 sb->patch_version = mddev->patch_version;
900 sb->gvalid_words = 0; /* ignored */
901 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
902 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
903 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
904 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
906 sb->ctime = mddev->ctime;
907 sb->level = mddev->level;
908 sb->size = mddev->size;
909 sb->raid_disks = mddev->raid_disks;
910 sb->md_minor = mddev->md_minor;
911 sb->not_persistent = 0;
912 sb->utime = mddev->utime;
914 sb->events_hi = (mddev->events>>32);
915 sb->events_lo = (u32)mddev->events;
917 if (mddev->reshape_position == MaxSector)
918 sb->minor_version = 90;
920 sb->minor_version = 91;
921 sb->reshape_position = mddev->reshape_position;
922 sb->new_level = mddev->new_level;
923 sb->delta_disks = mddev->delta_disks;
924 sb->new_layout = mddev->new_layout;
925 sb->new_chunk = mddev->new_chunk;
927 mddev->minor_version = sb->minor_version;
930 sb->recovery_cp = mddev->recovery_cp;
931 sb->cp_events_hi = (mddev->events>>32);
932 sb->cp_events_lo = (u32)mddev->events;
933 if (mddev->recovery_cp == MaxSector)
934 sb->state = (1<< MD_SB_CLEAN);
938 sb->layout = mddev->layout;
939 sb->chunk_size = mddev->chunk_size;
941 if (mddev->bitmap && mddev->bitmap_file == NULL)
942 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
944 sb->disks[0].state = (1<<MD_DISK_REMOVED);
945 rdev_for_each(rdev2, tmp, mddev) {
948 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
949 && !test_bit(Faulty, &rdev2->flags))
950 desc_nr = rdev2->raid_disk;
952 desc_nr = next_spare++;
953 rdev2->desc_nr = desc_nr;
954 d = &sb->disks[rdev2->desc_nr];
956 d->number = rdev2->desc_nr;
957 d->major = MAJOR(rdev2->bdev->bd_dev);
958 d->minor = MINOR(rdev2->bdev->bd_dev);
959 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
960 && !test_bit(Faulty, &rdev2->flags))
961 d->raid_disk = rdev2->raid_disk;
963 d->raid_disk = rdev2->desc_nr; /* compatibility */
964 if (test_bit(Faulty, &rdev2->flags))
965 d->state = (1<<MD_DISK_FAULTY);
966 else if (test_bit(In_sync, &rdev2->flags)) {
967 d->state = (1<<MD_DISK_ACTIVE);
968 d->state |= (1<<MD_DISK_SYNC);
976 if (test_bit(WriteMostly, &rdev2->flags))
977 d->state |= (1<<MD_DISK_WRITEMOSTLY);
979 /* now set the "removed" and "faulty" bits on any missing devices */
980 for (i=0 ; i < mddev->raid_disks ; i++) {
981 mdp_disk_t *d = &sb->disks[i];
982 if (d->state == 0 && d->number == 0) {
985 d->state = (1<<MD_DISK_REMOVED);
986 d->state |= (1<<MD_DISK_FAULTY);
990 sb->nr_disks = nr_disks;
991 sb->active_disks = active;
992 sb->working_disks = working;
993 sb->failed_disks = failed;
994 sb->spare_disks = spare;
996 sb->this_disk = sb->disks[rdev->desc_nr];
997 sb->sb_csum = calc_sb_csum(sb);
1001 * rdev_size_change for 0.90.0
1003 static unsigned long long
1004 super_90_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1006 if (size && size < rdev->mddev->size)
1007 return 0; /* component must fit device */
1008 size *= 2; /* convert to sectors */
1009 if (rdev->mddev->bitmap_offset)
1010 return 0; /* can't move bitmap */
1011 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
1012 if (!size || size > rdev->sb_offset*2)
1013 size = rdev->sb_offset*2;
1014 md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1016 md_super_wait(rdev->mddev);
1017 return size/2; /* kB for sysfs */
1022 * version 1 superblock
1025 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1029 unsigned long long newcsum;
1030 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1031 __le32 *isuper = (__le32*)sb;
1034 disk_csum = sb->sb_csum;
1037 for (i=0; size>=4; size -= 4 )
1038 newcsum += le32_to_cpu(*isuper++);
1041 newcsum += le16_to_cpu(*(__le16*) isuper);
1043 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1044 sb->sb_csum = disk_csum;
1045 return cpu_to_le32(csum);
1048 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1050 struct mdp_superblock_1 *sb;
1053 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1057 * Calculate the position of the superblock.
1058 * It is always aligned to a 4K boundary and
1059 * depeding on minor_version, it can be:
1060 * 0: At least 8K, but less than 12K, from end of device
1061 * 1: At start of device
1062 * 2: 4K from start of device.
1064 switch(minor_version) {
1066 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1068 sb_offset &= ~(sector_t)(4*2-1);
1069 /* convert from sectors to K */
1081 rdev->sb_offset = sb_offset;
1083 /* superblock is rarely larger than 1K, but it can be larger,
1084 * and it is safe to read 4k, so we do that
1086 ret = read_disk_sb(rdev, 4096);
1087 if (ret) return ret;
1090 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1092 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1093 sb->major_version != cpu_to_le32(1) ||
1094 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1095 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1096 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1099 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1100 printk("md: invalid superblock checksum on %s\n",
1101 bdevname(rdev->bdev,b));
1104 if (le64_to_cpu(sb->data_size) < 10) {
1105 printk("md: data_size too small on %s\n",
1106 bdevname(rdev->bdev,b));
1109 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1110 if (sb->level != cpu_to_le32(1) &&
1111 sb->level != cpu_to_le32(4) &&
1112 sb->level != cpu_to_le32(5) &&
1113 sb->level != cpu_to_le32(6) &&
1114 sb->level != cpu_to_le32(10)) {
1116 "md: bitmaps not supported for this level.\n");
1121 rdev->preferred_minor = 0xffff;
1122 rdev->data_offset = le64_to_cpu(sb->data_offset);
1123 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1125 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1126 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1127 if (rdev->sb_size & bmask)
1128 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1131 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1134 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1137 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1143 struct mdp_superblock_1 *refsb =
1144 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1146 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1147 sb->level != refsb->level ||
1148 sb->layout != refsb->layout ||
1149 sb->chunksize != refsb->chunksize) {
1150 printk(KERN_WARNING "md: %s has strangely different"
1151 " superblock to %s\n",
1152 bdevname(rdev->bdev,b),
1153 bdevname(refdev->bdev,b2));
1156 ev1 = le64_to_cpu(sb->events);
1157 ev2 = le64_to_cpu(refsb->events);
1165 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1167 rdev->size = rdev->sb_offset;
1168 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1170 rdev->size = le64_to_cpu(sb->data_size)/2;
1171 if (le32_to_cpu(sb->chunksize))
1172 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1174 if (le64_to_cpu(sb->size) > rdev->size*2)
1179 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1181 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1182 __u64 ev1 = le64_to_cpu(sb->events);
1184 rdev->raid_disk = -1;
1185 clear_bit(Faulty, &rdev->flags);
1186 clear_bit(In_sync, &rdev->flags);
1187 clear_bit(WriteMostly, &rdev->flags);
1188 clear_bit(BarriersNotsupp, &rdev->flags);
1190 if (mddev->raid_disks == 0) {
1191 mddev->major_version = 1;
1192 mddev->patch_version = 0;
1193 mddev->external = 0;
1194 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1195 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1196 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1197 mddev->level = le32_to_cpu(sb->level);
1198 mddev->clevel[0] = 0;
1199 mddev->layout = le32_to_cpu(sb->layout);
1200 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1201 mddev->size = le64_to_cpu(sb->size)/2;
1202 mddev->events = ev1;
1203 mddev->bitmap_offset = 0;
1204 mddev->default_bitmap_offset = 1024 >> 9;
1206 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1207 memcpy(mddev->uuid, sb->set_uuid, 16);
1209 mddev->max_disks = (4096-256)/2;
1211 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1212 mddev->bitmap_file == NULL )
1213 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1215 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1216 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1217 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1218 mddev->new_level = le32_to_cpu(sb->new_level);
1219 mddev->new_layout = le32_to_cpu(sb->new_layout);
1220 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1222 mddev->reshape_position = MaxSector;
1223 mddev->delta_disks = 0;
1224 mddev->new_level = mddev->level;
1225 mddev->new_layout = mddev->layout;
1226 mddev->new_chunk = mddev->chunk_size;
1229 } else if (mddev->pers == NULL) {
1230 /* Insist of good event counter while assembling */
1232 if (ev1 < mddev->events)
1234 } else if (mddev->bitmap) {
1235 /* If adding to array with a bitmap, then we can accept an
1236 * older device, but not too old.
1238 if (ev1 < mddev->bitmap->events_cleared)
1241 if (ev1 < mddev->events)
1242 /* just a hot-add of a new device, leave raid_disk at -1 */
1245 if (mddev->level != LEVEL_MULTIPATH) {
1247 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1249 case 0xffff: /* spare */
1251 case 0xfffe: /* faulty */
1252 set_bit(Faulty, &rdev->flags);
1255 if ((le32_to_cpu(sb->feature_map) &
1256 MD_FEATURE_RECOVERY_OFFSET))
1257 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1259 set_bit(In_sync, &rdev->flags);
1260 rdev->raid_disk = role;
1263 if (sb->devflags & WriteMostly1)
1264 set_bit(WriteMostly, &rdev->flags);
1265 } else /* MULTIPATH are always insync */
1266 set_bit(In_sync, &rdev->flags);
1271 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1273 struct mdp_superblock_1 *sb;
1274 struct list_head *tmp;
1277 /* make rdev->sb match mddev and rdev data. */
1279 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1281 sb->feature_map = 0;
1283 sb->recovery_offset = cpu_to_le64(0);
1284 memset(sb->pad1, 0, sizeof(sb->pad1));
1285 memset(sb->pad2, 0, sizeof(sb->pad2));
1286 memset(sb->pad3, 0, sizeof(sb->pad3));
1288 sb->utime = cpu_to_le64((__u64)mddev->utime);
1289 sb->events = cpu_to_le64(mddev->events);
1291 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1293 sb->resync_offset = cpu_to_le64(0);
1295 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1297 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1298 sb->size = cpu_to_le64(mddev->size<<1);
1300 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1301 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1302 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1305 if (rdev->raid_disk >= 0 &&
1306 !test_bit(In_sync, &rdev->flags) &&
1307 rdev->recovery_offset > 0) {
1308 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1309 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1312 if (mddev->reshape_position != MaxSector) {
1313 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1314 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1315 sb->new_layout = cpu_to_le32(mddev->new_layout);
1316 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1317 sb->new_level = cpu_to_le32(mddev->new_level);
1318 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1322 rdev_for_each(rdev2, tmp, mddev)
1323 if (rdev2->desc_nr+1 > max_dev)
1324 max_dev = rdev2->desc_nr+1;
1326 if (max_dev > le32_to_cpu(sb->max_dev))
1327 sb->max_dev = cpu_to_le32(max_dev);
1328 for (i=0; i<max_dev;i++)
1329 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1331 rdev_for_each(rdev2, tmp, mddev) {
1333 if (test_bit(Faulty, &rdev2->flags))
1334 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1335 else if (test_bit(In_sync, &rdev2->flags))
1336 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1337 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1338 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1340 sb->dev_roles[i] = cpu_to_le16(0xffff);
1343 sb->sb_csum = calc_sb_1_csum(sb);
1346 static unsigned long long
1347 super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1349 struct mdp_superblock_1 *sb;
1350 unsigned long long max_size;
1351 if (size && size < rdev->mddev->size)
1352 return 0; /* component must fit device */
1353 size *= 2; /* convert to sectors */
1354 if (rdev->sb_offset < rdev->data_offset/2) {
1355 /* minor versions 1 and 2; superblock before data */
1356 max_size = (rdev->bdev->bd_inode->i_size >> 9);
1357 max_size -= rdev->data_offset;
1358 if (!size || size > max_size)
1360 } else if (rdev->mddev->bitmap_offset) {
1361 /* minor version 0 with bitmap we can't move */
1364 /* minor version 0; superblock after data */
1366 sb_offset = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1367 sb_offset &= ~(sector_t)(4*2 - 1);
1368 max_size = rdev->size*2 + sb_offset - rdev->sb_offset*2;
1369 if (!size || size > max_size)
1371 rdev->sb_offset = sb_offset/2;
1373 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1374 sb->data_size = cpu_to_le64(size);
1375 sb->super_offset = rdev->sb_offset*2;
1376 sb->sb_csum = calc_sb_1_csum(sb);
1377 md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1379 md_super_wait(rdev->mddev);
1380 return size/2; /* kB for sysfs */
1383 static struct super_type super_types[] = {
1386 .owner = THIS_MODULE,
1387 .load_super = super_90_load,
1388 .validate_super = super_90_validate,
1389 .sync_super = super_90_sync,
1390 .rdev_size_change = super_90_rdev_size_change,
1394 .owner = THIS_MODULE,
1395 .load_super = super_1_load,
1396 .validate_super = super_1_validate,
1397 .sync_super = super_1_sync,
1398 .rdev_size_change = super_1_rdev_size_change,
1402 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1404 struct list_head *tmp, *tmp2;
1405 mdk_rdev_t *rdev, *rdev2;
1407 rdev_for_each(rdev, tmp, mddev1)
1408 rdev_for_each(rdev2, tmp2, mddev2)
1409 if (rdev->bdev->bd_contains ==
1410 rdev2->bdev->bd_contains)
1416 static LIST_HEAD(pending_raid_disks);
1418 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1420 char b[BDEVNAME_SIZE];
1430 /* prevent duplicates */
1431 if (find_rdev(mddev, rdev->bdev->bd_dev))
1434 /* make sure rdev->size exceeds mddev->size */
1435 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1437 /* Cannot change size, so fail
1438 * If mddev->level <= 0, then we don't care
1439 * about aligning sizes (e.g. linear)
1441 if (mddev->level > 0)
1444 mddev->size = rdev->size;
1447 /* Verify rdev->desc_nr is unique.
1448 * If it is -1, assign a free number, else
1449 * check number is not in use
1451 if (rdev->desc_nr < 0) {
1453 if (mddev->pers) choice = mddev->raid_disks;
1454 while (find_rdev_nr(mddev, choice))
1456 rdev->desc_nr = choice;
1458 if (find_rdev_nr(mddev, rdev->desc_nr))
1461 bdevname(rdev->bdev,b);
1462 while ( (s=strchr(b, '/')) != NULL)
1465 rdev->mddev = mddev;
1466 printk(KERN_INFO "md: bind<%s>\n", b);
1468 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1471 if (rdev->bdev->bd_part)
1472 ko = &rdev->bdev->bd_part->dev.kobj;
1474 ko = &rdev->bdev->bd_disk->dev.kobj;
1475 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1476 kobject_del(&rdev->kobj);
1479 list_add(&rdev->same_set, &mddev->disks);
1480 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1484 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1489 static void md_delayed_delete(struct work_struct *ws)
1491 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1492 kobject_del(&rdev->kobj);
1493 kobject_put(&rdev->kobj);
1496 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1498 char b[BDEVNAME_SIZE];
1503 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1504 list_del_init(&rdev->same_set);
1505 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1507 sysfs_remove_link(&rdev->kobj, "block");
1509 /* We need to delay this, otherwise we can deadlock when
1510 * writing to 'remove' to "dev/state"
1512 INIT_WORK(&rdev->del_work, md_delayed_delete);
1513 kobject_get(&rdev->kobj);
1514 schedule_work(&rdev->del_work);
1518 * prevent the device from being mounted, repartitioned or
1519 * otherwise reused by a RAID array (or any other kernel
1520 * subsystem), by bd_claiming the device.
1522 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1525 struct block_device *bdev;
1526 char b[BDEVNAME_SIZE];
1528 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1530 printk(KERN_ERR "md: could not open %s.\n",
1531 __bdevname(dev, b));
1532 return PTR_ERR(bdev);
1534 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1536 printk(KERN_ERR "md: could not bd_claim %s.\n",
1542 set_bit(AllReserved, &rdev->flags);
1547 static void unlock_rdev(mdk_rdev_t *rdev)
1549 struct block_device *bdev = rdev->bdev;
1557 void md_autodetect_dev(dev_t dev);
1559 static void export_rdev(mdk_rdev_t * rdev)
1561 char b[BDEVNAME_SIZE];
1562 printk(KERN_INFO "md: export_rdev(%s)\n",
1563 bdevname(rdev->bdev,b));
1567 list_del_init(&rdev->same_set);
1569 if (test_bit(AutoDetected, &rdev->flags))
1570 md_autodetect_dev(rdev->bdev->bd_dev);
1573 kobject_put(&rdev->kobj);
1576 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1578 unbind_rdev_from_array(rdev);
1582 static void export_array(mddev_t *mddev)
1584 struct list_head *tmp;
1587 rdev_for_each(rdev, tmp, mddev) {
1592 kick_rdev_from_array(rdev);
1594 if (!list_empty(&mddev->disks))
1596 mddev->raid_disks = 0;
1597 mddev->major_version = 0;
1600 static void print_desc(mdp_disk_t *desc)
1602 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1603 desc->major,desc->minor,desc->raid_disk,desc->state);
1606 static void print_sb(mdp_super_t *sb)
1611 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1612 sb->major_version, sb->minor_version, sb->patch_version,
1613 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1615 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1616 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1617 sb->md_minor, sb->layout, sb->chunk_size);
1618 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1619 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1620 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1621 sb->failed_disks, sb->spare_disks,
1622 sb->sb_csum, (unsigned long)sb->events_lo);
1625 for (i = 0; i < MD_SB_DISKS; i++) {
1628 desc = sb->disks + i;
1629 if (desc->number || desc->major || desc->minor ||
1630 desc->raid_disk || (desc->state && (desc->state != 4))) {
1631 printk(" D %2d: ", i);
1635 printk(KERN_INFO "md: THIS: ");
1636 print_desc(&sb->this_disk);
1640 static void print_rdev(mdk_rdev_t *rdev)
1642 char b[BDEVNAME_SIZE];
1643 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1644 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1645 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1647 if (rdev->sb_loaded) {
1648 printk(KERN_INFO "md: rdev superblock:\n");
1649 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1651 printk(KERN_INFO "md: no rdev superblock!\n");
1654 static void md_print_devices(void)
1656 struct list_head *tmp, *tmp2;
1659 char b[BDEVNAME_SIZE];
1662 printk("md: **********************************\n");
1663 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1664 printk("md: **********************************\n");
1665 for_each_mddev(mddev, tmp) {
1668 bitmap_print_sb(mddev->bitmap);
1670 printk("%s: ", mdname(mddev));
1671 rdev_for_each(rdev, tmp2, mddev)
1672 printk("<%s>", bdevname(rdev->bdev,b));
1675 rdev_for_each(rdev, tmp2, mddev)
1678 printk("md: **********************************\n");
1683 static void sync_sbs(mddev_t * mddev, int nospares)
1685 /* Update each superblock (in-memory image), but
1686 * if we are allowed to, skip spares which already
1687 * have the right event counter, or have one earlier
1688 * (which would mean they aren't being marked as dirty
1689 * with the rest of the array)
1692 struct list_head *tmp;
1694 rdev_for_each(rdev, tmp, mddev) {
1695 if (rdev->sb_events == mddev->events ||
1697 rdev->raid_disk < 0 &&
1698 (rdev->sb_events&1)==0 &&
1699 rdev->sb_events+1 == mddev->events)) {
1700 /* Don't update this superblock */
1701 rdev->sb_loaded = 2;
1703 super_types[mddev->major_version].
1704 sync_super(mddev, rdev);
1705 rdev->sb_loaded = 1;
1710 static void md_update_sb(mddev_t * mddev, int force_change)
1712 struct list_head *tmp;
1717 if (mddev->external)
1720 spin_lock_irq(&mddev->write_lock);
1722 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1723 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1725 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1726 /* just a clean<-> dirty transition, possibly leave spares alone,
1727 * though if events isn't the right even/odd, we will have to do
1733 if (mddev->degraded)
1734 /* If the array is degraded, then skipping spares is both
1735 * dangerous and fairly pointless.
1736 * Dangerous because a device that was removed from the array
1737 * might have a event_count that still looks up-to-date,
1738 * so it can be re-added without a resync.
1739 * Pointless because if there are any spares to skip,
1740 * then a recovery will happen and soon that array won't
1741 * be degraded any more and the spare can go back to sleep then.
1745 sync_req = mddev->in_sync;
1746 mddev->utime = get_seconds();
1748 /* If this is just a dirty<->clean transition, and the array is clean
1749 * and 'events' is odd, we can roll back to the previous clean state */
1751 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1752 && (mddev->events & 1)
1753 && mddev->events != 1)
1756 /* otherwise we have to go forward and ... */
1758 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1759 /* .. if the array isn't clean, insist on an odd 'events' */
1760 if ((mddev->events&1)==0) {
1765 /* otherwise insist on an even 'events' (for clean states) */
1766 if ((mddev->events&1)) {
1773 if (!mddev->events) {
1775 * oops, this 64-bit counter should never wrap.
1776 * Either we are in around ~1 trillion A.C., assuming
1777 * 1 reboot per second, or we have a bug:
1784 * do not write anything to disk if using
1785 * nonpersistent superblocks
1787 if (!mddev->persistent) {
1788 if (!mddev->external)
1789 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1791 spin_unlock_irq(&mddev->write_lock);
1792 wake_up(&mddev->sb_wait);
1795 sync_sbs(mddev, nospares);
1796 spin_unlock_irq(&mddev->write_lock);
1799 "md: updating %s RAID superblock on device (in sync %d)\n",
1800 mdname(mddev),mddev->in_sync);
1802 bitmap_update_sb(mddev->bitmap);
1803 rdev_for_each(rdev, tmp, mddev) {
1804 char b[BDEVNAME_SIZE];
1805 dprintk(KERN_INFO "md: ");
1806 if (rdev->sb_loaded != 1)
1807 continue; /* no noise on spare devices */
1808 if (test_bit(Faulty, &rdev->flags))
1809 dprintk("(skipping faulty ");
1811 dprintk("%s ", bdevname(rdev->bdev,b));
1812 if (!test_bit(Faulty, &rdev->flags)) {
1813 md_super_write(mddev,rdev,
1814 rdev->sb_offset<<1, rdev->sb_size,
1816 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1817 bdevname(rdev->bdev,b),
1818 (unsigned long long)rdev->sb_offset);
1819 rdev->sb_events = mddev->events;
1823 if (mddev->level == LEVEL_MULTIPATH)
1824 /* only need to write one superblock... */
1827 md_super_wait(mddev);
1828 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1830 spin_lock_irq(&mddev->write_lock);
1831 if (mddev->in_sync != sync_req ||
1832 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1833 /* have to write it out again */
1834 spin_unlock_irq(&mddev->write_lock);
1837 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1838 spin_unlock_irq(&mddev->write_lock);
1839 wake_up(&mddev->sb_wait);
1843 /* words written to sysfs files may, or may not, be \n terminated.
1844 * We want to accept with case. For this we use cmd_match.
1846 static int cmd_match(const char *cmd, const char *str)
1848 /* See if cmd, written into a sysfs file, matches
1849 * str. They must either be the same, or cmd can
1850 * have a trailing newline
1852 while (*cmd && *str && *cmd == *str) {
1863 struct rdev_sysfs_entry {
1864 struct attribute attr;
1865 ssize_t (*show)(mdk_rdev_t *, char *);
1866 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1870 state_show(mdk_rdev_t *rdev, char *page)
1875 if (test_bit(Faulty, &rdev->flags)) {
1876 len+= sprintf(page+len, "%sfaulty",sep);
1879 if (test_bit(In_sync, &rdev->flags)) {
1880 len += sprintf(page+len, "%sin_sync",sep);
1883 if (test_bit(WriteMostly, &rdev->flags)) {
1884 len += sprintf(page+len, "%swrite_mostly",sep);
1887 if (test_bit(Blocked, &rdev->flags)) {
1888 len += sprintf(page+len, "%sblocked", sep);
1891 if (!test_bit(Faulty, &rdev->flags) &&
1892 !test_bit(In_sync, &rdev->flags)) {
1893 len += sprintf(page+len, "%sspare", sep);
1896 return len+sprintf(page+len, "\n");
1900 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1903 * faulty - simulates and error
1904 * remove - disconnects the device
1905 * writemostly - sets write_mostly
1906 * -writemostly - clears write_mostly
1907 * blocked - sets the Blocked flag
1908 * -blocked - clears the Blocked flag
1911 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1912 md_error(rdev->mddev, rdev);
1914 } else if (cmd_match(buf, "remove")) {
1915 if (rdev->raid_disk >= 0)
1918 mddev_t *mddev = rdev->mddev;
1919 kick_rdev_from_array(rdev);
1921 md_update_sb(mddev, 1);
1922 md_new_event(mddev);
1925 } else if (cmd_match(buf, "writemostly")) {
1926 set_bit(WriteMostly, &rdev->flags);
1928 } else if (cmd_match(buf, "-writemostly")) {
1929 clear_bit(WriteMostly, &rdev->flags);
1931 } else if (cmd_match(buf, "blocked")) {
1932 set_bit(Blocked, &rdev->flags);
1934 } else if (cmd_match(buf, "-blocked")) {
1935 clear_bit(Blocked, &rdev->flags);
1936 wake_up(&rdev->blocked_wait);
1937 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1938 md_wakeup_thread(rdev->mddev->thread);
1943 sysfs_notify(&rdev->kobj, NULL, "state");
1944 return err ? err : len;
1946 static struct rdev_sysfs_entry rdev_state =
1947 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1950 errors_show(mdk_rdev_t *rdev, char *page)
1952 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1956 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1959 unsigned long n = simple_strtoul(buf, &e, 10);
1960 if (*buf && (*e == 0 || *e == '\n')) {
1961 atomic_set(&rdev->corrected_errors, n);
1966 static struct rdev_sysfs_entry rdev_errors =
1967 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1970 slot_show(mdk_rdev_t *rdev, char *page)
1972 if (rdev->raid_disk < 0)
1973 return sprintf(page, "none\n");
1975 return sprintf(page, "%d\n", rdev->raid_disk);
1979 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1984 int slot = simple_strtoul(buf, &e, 10);
1985 if (strncmp(buf, "none", 4)==0)
1987 else if (e==buf || (*e && *e!= '\n'))
1989 if (rdev->mddev->pers && slot == -1) {
1990 /* Setting 'slot' on an active array requires also
1991 * updating the 'rd%d' link, and communicating
1992 * with the personality with ->hot_*_disk.
1993 * For now we only support removing
1994 * failed/spare devices. This normally happens automatically,
1995 * but not when the metadata is externally managed.
1997 if (rdev->raid_disk == -1)
1999 /* personality does all needed checks */
2000 if (rdev->mddev->pers->hot_add_disk == NULL)
2002 err = rdev->mddev->pers->
2003 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2006 sprintf(nm, "rd%d", rdev->raid_disk);
2007 sysfs_remove_link(&rdev->mddev->kobj, nm);
2008 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2009 md_wakeup_thread(rdev->mddev->thread);
2010 } else if (rdev->mddev->pers) {
2012 struct list_head *tmp;
2013 /* Activating a spare .. or possibly reactivating
2014 * if we every get bitmaps working here.
2017 if (rdev->raid_disk != -1)
2020 if (rdev->mddev->pers->hot_add_disk == NULL)
2023 rdev_for_each(rdev2, tmp, rdev->mddev)
2024 if (rdev2->raid_disk == slot)
2027 rdev->raid_disk = slot;
2028 if (test_bit(In_sync, &rdev->flags))
2029 rdev->saved_raid_disk = slot;
2031 rdev->saved_raid_disk = -1;
2032 err = rdev->mddev->pers->
2033 hot_add_disk(rdev->mddev, rdev);
2035 rdev->raid_disk = -1;
2038 sysfs_notify(&rdev->kobj, NULL, "state");
2039 sprintf(nm, "rd%d", rdev->raid_disk);
2040 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2042 "md: cannot register "
2044 nm, mdname(rdev->mddev));
2046 /* don't wakeup anyone, leave that to userspace. */
2048 if (slot >= rdev->mddev->raid_disks)
2050 rdev->raid_disk = slot;
2051 /* assume it is working */
2052 clear_bit(Faulty, &rdev->flags);
2053 clear_bit(WriteMostly, &rdev->flags);
2054 set_bit(In_sync, &rdev->flags);
2055 sysfs_notify(&rdev->kobj, NULL, "state");
2061 static struct rdev_sysfs_entry rdev_slot =
2062 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2065 offset_show(mdk_rdev_t *rdev, char *page)
2067 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2071 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2074 unsigned long long offset = simple_strtoull(buf, &e, 10);
2075 if (e==buf || (*e && *e != '\n'))
2077 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2079 if (rdev->size && rdev->mddev->external)
2080 /* Must set offset before size, so overlap checks
2083 rdev->data_offset = offset;
2087 static struct rdev_sysfs_entry rdev_offset =
2088 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2091 rdev_size_show(mdk_rdev_t *rdev, char *page)
2093 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2096 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2098 /* check if two start/length pairs overlap */
2107 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2110 unsigned long long size = simple_strtoull(buf, &e, 10);
2111 unsigned long long oldsize = rdev->size;
2112 mddev_t *my_mddev = rdev->mddev;
2114 if (e==buf || (*e && *e != '\n'))
2116 if (my_mddev->pers && rdev->raid_disk >= 0) {
2117 if (rdev->mddev->persistent) {
2118 size = super_types[rdev->mddev->major_version].
2119 rdev_size_change(rdev, size);
2123 size = (rdev->bdev->bd_inode->i_size >> 10);
2124 size -= rdev->data_offset/2;
2126 if (size < rdev->mddev->size)
2127 return -EINVAL; /* component must fit device */
2131 if (size > oldsize && rdev->mddev->external) {
2132 /* need to check that all other rdevs with the same ->bdev
2133 * do not overlap. We need to unlock the mddev to avoid
2134 * a deadlock. We have already changed rdev->size, and if
2135 * we have to change it back, we will have the lock again.
2139 struct list_head *tmp, *tmp2;
2141 mddev_unlock(my_mddev);
2142 for_each_mddev(mddev, tmp) {
2146 rdev_for_each(rdev2, tmp2, mddev)
2147 if (test_bit(AllReserved, &rdev2->flags) ||
2148 (rdev->bdev == rdev2->bdev &&
2150 overlaps(rdev->data_offset, rdev->size,
2151 rdev2->data_offset, rdev2->size))) {
2155 mddev_unlock(mddev);
2161 mddev_lock(my_mddev);
2163 /* Someone else could have slipped in a size
2164 * change here, but doing so is just silly.
2165 * We put oldsize back because we *know* it is
2166 * safe, and trust userspace not to race with
2169 rdev->size = oldsize;
2173 if (size < my_mddev->size || my_mddev->size == 0)
2174 my_mddev->size = size;
2178 static struct rdev_sysfs_entry rdev_size =
2179 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2181 static struct attribute *rdev_default_attrs[] = {
2190 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2192 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2193 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2194 mddev_t *mddev = rdev->mddev;
2200 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2202 if (rdev->mddev == NULL)
2205 rv = entry->show(rdev, page);
2206 mddev_unlock(mddev);
2212 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2213 const char *page, size_t length)
2215 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2216 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2218 mddev_t *mddev = rdev->mddev;
2222 if (!capable(CAP_SYS_ADMIN))
2224 rv = mddev ? mddev_lock(mddev): -EBUSY;
2226 if (rdev->mddev == NULL)
2229 rv = entry->store(rdev, page, length);
2230 mddev_unlock(mddev);
2235 static void rdev_free(struct kobject *ko)
2237 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2240 static struct sysfs_ops rdev_sysfs_ops = {
2241 .show = rdev_attr_show,
2242 .store = rdev_attr_store,
2244 static struct kobj_type rdev_ktype = {
2245 .release = rdev_free,
2246 .sysfs_ops = &rdev_sysfs_ops,
2247 .default_attrs = rdev_default_attrs,
2251 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2253 * mark the device faulty if:
2255 * - the device is nonexistent (zero size)
2256 * - the device has no valid superblock
2258 * a faulty rdev _never_ has rdev->sb set.
2260 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2262 char b[BDEVNAME_SIZE];
2267 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2269 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2270 return ERR_PTR(-ENOMEM);
2273 if ((err = alloc_disk_sb(rdev)))
2276 err = lock_rdev(rdev, newdev, super_format == -2);
2280 kobject_init(&rdev->kobj, &rdev_ktype);
2283 rdev->saved_raid_disk = -1;
2284 rdev->raid_disk = -1;
2286 rdev->data_offset = 0;
2287 rdev->sb_events = 0;
2288 atomic_set(&rdev->nr_pending, 0);
2289 atomic_set(&rdev->read_errors, 0);
2290 atomic_set(&rdev->corrected_errors, 0);
2292 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2295 "md: %s has zero or unknown size, marking faulty!\n",
2296 bdevname(rdev->bdev,b));
2301 if (super_format >= 0) {
2302 err = super_types[super_format].
2303 load_super(rdev, NULL, super_minor);
2304 if (err == -EINVAL) {
2306 "md: %s does not have a valid v%d.%d "
2307 "superblock, not importing!\n",
2308 bdevname(rdev->bdev,b),
2309 super_format, super_minor);
2314 "md: could not read %s's sb, not importing!\n",
2315 bdevname(rdev->bdev,b));
2320 INIT_LIST_HEAD(&rdev->same_set);
2321 init_waitqueue_head(&rdev->blocked_wait);
2326 if (rdev->sb_page) {
2332 return ERR_PTR(err);
2336 * Check a full RAID array for plausibility
2340 static void analyze_sbs(mddev_t * mddev)
2343 struct list_head *tmp;
2344 mdk_rdev_t *rdev, *freshest;
2345 char b[BDEVNAME_SIZE];
2348 rdev_for_each(rdev, tmp, mddev)
2349 switch (super_types[mddev->major_version].
2350 load_super(rdev, freshest, mddev->minor_version)) {
2358 "md: fatal superblock inconsistency in %s"
2359 " -- removing from array\n",
2360 bdevname(rdev->bdev,b));
2361 kick_rdev_from_array(rdev);
2365 super_types[mddev->major_version].
2366 validate_super(mddev, freshest);
2369 rdev_for_each(rdev, tmp, mddev) {
2370 if (rdev != freshest)
2371 if (super_types[mddev->major_version].
2372 validate_super(mddev, rdev)) {
2373 printk(KERN_WARNING "md: kicking non-fresh %s"
2375 bdevname(rdev->bdev,b));
2376 kick_rdev_from_array(rdev);
2379 if (mddev->level == LEVEL_MULTIPATH) {
2380 rdev->desc_nr = i++;
2381 rdev->raid_disk = rdev->desc_nr;
2382 set_bit(In_sync, &rdev->flags);
2383 } else if (rdev->raid_disk >= mddev->raid_disks) {
2384 rdev->raid_disk = -1;
2385 clear_bit(In_sync, &rdev->flags);
2391 if (mddev->recovery_cp != MaxSector &&
2393 printk(KERN_ERR "md: %s: raid array is not clean"
2394 " -- starting background reconstruction\n",
2400 safe_delay_show(mddev_t *mddev, char *page)
2402 int msec = (mddev->safemode_delay*1000)/HZ;
2403 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2406 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2414 /* remove a period, and count digits after it */
2415 if (len >= sizeof(buf))
2417 strlcpy(buf, cbuf, len);
2419 for (i=0; i<len; i++) {
2421 if (isdigit(buf[i])) {
2426 } else if (buf[i] == '.') {
2431 msec = simple_strtoul(buf, &e, 10);
2432 if (e == buf || (*e && *e != '\n'))
2434 msec = (msec * 1000) / scale;
2436 mddev->safemode_delay = 0;
2438 mddev->safemode_delay = (msec*HZ)/1000;
2439 if (mddev->safemode_delay == 0)
2440 mddev->safemode_delay = 1;
2444 static struct md_sysfs_entry md_safe_delay =
2445 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2448 level_show(mddev_t *mddev, char *page)
2450 struct mdk_personality *p = mddev->pers;
2452 return sprintf(page, "%s\n", p->name);
2453 else if (mddev->clevel[0])
2454 return sprintf(page, "%s\n", mddev->clevel);
2455 else if (mddev->level != LEVEL_NONE)
2456 return sprintf(page, "%d\n", mddev->level);
2462 level_store(mddev_t *mddev, const char *buf, size_t len)
2469 if (len >= sizeof(mddev->clevel))
2471 strncpy(mddev->clevel, buf, len);
2472 if (mddev->clevel[len-1] == '\n')
2474 mddev->clevel[len] = 0;
2475 mddev->level = LEVEL_NONE;
2479 static struct md_sysfs_entry md_level =
2480 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2484 layout_show(mddev_t *mddev, char *page)
2486 /* just a number, not meaningful for all levels */
2487 if (mddev->reshape_position != MaxSector &&
2488 mddev->layout != mddev->new_layout)
2489 return sprintf(page, "%d (%d)\n",
2490 mddev->new_layout, mddev->layout);
2491 return sprintf(page, "%d\n", mddev->layout);
2495 layout_store(mddev_t *mddev, const char *buf, size_t len)
2498 unsigned long n = simple_strtoul(buf, &e, 10);
2500 if (!*buf || (*e && *e != '\n'))
2505 if (mddev->reshape_position != MaxSector)
2506 mddev->new_layout = n;
2511 static struct md_sysfs_entry md_layout =
2512 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2516 raid_disks_show(mddev_t *mddev, char *page)
2518 if (mddev->raid_disks == 0)
2520 if (mddev->reshape_position != MaxSector &&
2521 mddev->delta_disks != 0)
2522 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2523 mddev->raid_disks - mddev->delta_disks);
2524 return sprintf(page, "%d\n", mddev->raid_disks);
2527 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2530 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2534 unsigned long n = simple_strtoul(buf, &e, 10);
2536 if (!*buf || (*e && *e != '\n'))
2540 rv = update_raid_disks(mddev, n);
2541 else if (mddev->reshape_position != MaxSector) {
2542 int olddisks = mddev->raid_disks - mddev->delta_disks;
2543 mddev->delta_disks = n - olddisks;
2544 mddev->raid_disks = n;
2546 mddev->raid_disks = n;
2547 return rv ? rv : len;
2549 static struct md_sysfs_entry md_raid_disks =
2550 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2553 chunk_size_show(mddev_t *mddev, char *page)
2555 if (mddev->reshape_position != MaxSector &&
2556 mddev->chunk_size != mddev->new_chunk)
2557 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2559 return sprintf(page, "%d\n", mddev->chunk_size);
2563 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2565 /* can only set chunk_size if array is not yet active */
2567 unsigned long n = simple_strtoul(buf, &e, 10);
2569 if (!*buf || (*e && *e != '\n'))
2574 else if (mddev->reshape_position != MaxSector)
2575 mddev->new_chunk = n;
2577 mddev->chunk_size = n;
2580 static struct md_sysfs_entry md_chunk_size =
2581 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2584 resync_start_show(mddev_t *mddev, char *page)
2586 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2590 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2593 unsigned long long n = simple_strtoull(buf, &e, 10);
2597 if (!*buf || (*e && *e != '\n'))
2600 mddev->recovery_cp = n;
2603 static struct md_sysfs_entry md_resync_start =
2604 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2607 * The array state can be:
2610 * No devices, no size, no level
2611 * Equivalent to STOP_ARRAY ioctl
2613 * May have some settings, but array is not active
2614 * all IO results in error
2615 * When written, doesn't tear down array, but just stops it
2616 * suspended (not supported yet)
2617 * All IO requests will block. The array can be reconfigured.
2618 * Writing this, if accepted, will block until array is quiescent
2620 * no resync can happen. no superblocks get written.
2621 * write requests fail
2623 * like readonly, but behaves like 'clean' on a write request.
2625 * clean - no pending writes, but otherwise active.
2626 * When written to inactive array, starts without resync
2627 * If a write request arrives then
2628 * if metadata is known, mark 'dirty' and switch to 'active'.
2629 * if not known, block and switch to write-pending
2630 * If written to an active array that has pending writes, then fails.
2632 * fully active: IO and resync can be happening.
2633 * When written to inactive array, starts with resync
2636 * clean, but writes are blocked waiting for 'active' to be written.
2639 * like active, but no writes have been seen for a while (100msec).
2642 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2643 write_pending, active_idle, bad_word};
2644 static char *array_states[] = {
2645 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2646 "write-pending", "active-idle", NULL };
2648 static int match_word(const char *word, char **list)
2651 for (n=0; list[n]; n++)
2652 if (cmd_match(word, list[n]))
2658 array_state_show(mddev_t *mddev, char *page)
2660 enum array_state st = inactive;
2673 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2675 else if (mddev->safemode)
2681 if (list_empty(&mddev->disks) &&
2682 mddev->raid_disks == 0 &&
2688 return sprintf(page, "%s\n", array_states[st]);
2691 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2692 static int do_md_run(mddev_t * mddev);
2693 static int restart_array(mddev_t *mddev);
2696 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2699 enum array_state st = match_word(buf, array_states);
2704 /* stopping an active array */
2705 if (atomic_read(&mddev->active) > 1)
2707 err = do_md_stop(mddev, 0, 0);
2710 /* stopping an active array */
2712 if (atomic_read(&mddev->active) > 1)
2714 err = do_md_stop(mddev, 2, 0);
2716 err = 0; /* already inactive */
2719 break; /* not supported yet */
2722 err = do_md_stop(mddev, 1, 0);
2725 set_disk_ro(mddev->gendisk, 1);
2726 err = do_md_run(mddev);
2732 err = do_md_stop(mddev, 1, 0);
2734 err = restart_array(mddev);
2737 set_disk_ro(mddev->gendisk, 0);
2741 err = do_md_run(mddev);
2746 restart_array(mddev);
2747 spin_lock_irq(&mddev->write_lock);
2748 if (atomic_read(&mddev->writes_pending) == 0) {
2749 if (mddev->in_sync == 0) {
2751 if (mddev->safemode == 1)
2752 mddev->safemode = 0;
2753 if (mddev->persistent)
2754 set_bit(MD_CHANGE_CLEAN,
2760 spin_unlock_irq(&mddev->write_lock);
2763 mddev->recovery_cp = MaxSector;
2764 err = do_md_run(mddev);
2769 restart_array(mddev);
2770 if (mddev->external)
2771 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2772 wake_up(&mddev->sb_wait);
2776 set_disk_ro(mddev->gendisk, 0);
2777 err = do_md_run(mddev);
2782 /* these cannot be set */
2788 sysfs_notify(&mddev->kobj, NULL, "array_state");
2792 static struct md_sysfs_entry md_array_state =
2793 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2796 null_show(mddev_t *mddev, char *page)
2802 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2804 /* buf must be %d:%d\n? giving major and minor numbers */
2805 /* The new device is added to the array.
2806 * If the array has a persistent superblock, we read the
2807 * superblock to initialise info and check validity.
2808 * Otherwise, only checking done is that in bind_rdev_to_array,
2809 * which mainly checks size.
2812 int major = simple_strtoul(buf, &e, 10);
2818 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2820 minor = simple_strtoul(e+1, &e, 10);
2821 if (*e && *e != '\n')
2823 dev = MKDEV(major, minor);
2824 if (major != MAJOR(dev) ||
2825 minor != MINOR(dev))
2829 if (mddev->persistent) {
2830 rdev = md_import_device(dev, mddev->major_version,
2831 mddev->minor_version);
2832 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2833 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2834 mdk_rdev_t, same_set);
2835 err = super_types[mddev->major_version]
2836 .load_super(rdev, rdev0, mddev->minor_version);
2840 } else if (mddev->external)
2841 rdev = md_import_device(dev, -2, -1);
2843 rdev = md_import_device(dev, -1, -1);
2846 return PTR_ERR(rdev);
2847 err = bind_rdev_to_array(rdev, mddev);
2851 return err ? err : len;
2854 static struct md_sysfs_entry md_new_device =
2855 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2858 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2861 unsigned long chunk, end_chunk;
2865 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2867 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2868 if (buf == end) break;
2869 if (*end == '-') { /* range */
2871 end_chunk = simple_strtoul(buf, &end, 0);
2872 if (buf == end) break;
2874 if (*end && !isspace(*end)) break;
2875 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2877 while (isspace(*buf)) buf++;
2879 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2884 static struct md_sysfs_entry md_bitmap =
2885 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2888 size_show(mddev_t *mddev, char *page)
2890 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2893 static int update_size(mddev_t *mddev, sector_t num_sectors);
2896 size_store(mddev_t *mddev, const char *buf, size_t len)
2898 /* If array is inactive, we can reduce the component size, but
2899 * not increase it (except from 0).
2900 * If array is active, we can try an on-line resize
2904 unsigned long long size = simple_strtoull(buf, &e, 10);
2905 if (!*buf || *buf == '\n' ||
2910 err = update_size(mddev, size * 2);
2911 md_update_sb(mddev, 1);
2913 if (mddev->size == 0 ||
2919 return err ? err : len;
2922 static struct md_sysfs_entry md_size =
2923 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2928 * 'none' for arrays with no metadata (good luck...)
2929 * 'external' for arrays with externally managed metadata,
2930 * or N.M for internally known formats
2933 metadata_show(mddev_t *mddev, char *page)
2935 if (mddev->persistent)
2936 return sprintf(page, "%d.%d\n",
2937 mddev->major_version, mddev->minor_version);
2938 else if (mddev->external)
2939 return sprintf(page, "external:%s\n", mddev->metadata_type);
2941 return sprintf(page, "none\n");
2945 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2949 if (!list_empty(&mddev->disks))
2952 if (cmd_match(buf, "none")) {
2953 mddev->persistent = 0;
2954 mddev->external = 0;
2955 mddev->major_version = 0;
2956 mddev->minor_version = 90;
2959 if (strncmp(buf, "external:", 9) == 0) {
2960 size_t namelen = len-9;
2961 if (namelen >= sizeof(mddev->metadata_type))
2962 namelen = sizeof(mddev->metadata_type)-1;
2963 strncpy(mddev->metadata_type, buf+9, namelen);
2964 mddev->metadata_type[namelen] = 0;
2965 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2966 mddev->metadata_type[--namelen] = 0;
2967 mddev->persistent = 0;
2968 mddev->external = 1;
2969 mddev->major_version = 0;
2970 mddev->minor_version = 90;
2973 major = simple_strtoul(buf, &e, 10);
2974 if (e==buf || *e != '.')
2977 minor = simple_strtoul(buf, &e, 10);
2978 if (e==buf || (*e && *e != '\n') )
2980 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2982 mddev->major_version = major;
2983 mddev->minor_version = minor;
2984 mddev->persistent = 1;
2985 mddev->external = 0;
2989 static struct md_sysfs_entry md_metadata =
2990 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2993 action_show(mddev_t *mddev, char *page)
2995 char *type = "idle";
2996 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2997 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2998 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3000 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3001 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3003 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3007 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3010 return sprintf(page, "%s\n", type);
3014 action_store(mddev_t *mddev, const char *page, size_t len)
3016 if (!mddev->pers || !mddev->pers->sync_request)
3019 if (cmd_match(page, "idle")) {
3020 if (mddev->sync_thread) {
3021 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3022 md_unregister_thread(mddev->sync_thread);
3023 mddev->sync_thread = NULL;
3024 mddev->recovery = 0;
3026 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3027 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3029 else if (cmd_match(page, "resync"))
3030 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3031 else if (cmd_match(page, "recover")) {
3032 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3033 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3034 } else if (cmd_match(page, "reshape")) {
3036 if (mddev->pers->start_reshape == NULL)
3038 err = mddev->pers->start_reshape(mddev);
3041 sysfs_notify(&mddev->kobj, NULL, "degraded");
3043 if (cmd_match(page, "check"))
3044 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3045 else if (!cmd_match(page, "repair"))
3047 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3048 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3050 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3051 md_wakeup_thread(mddev->thread);
3052 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3057 mismatch_cnt_show(mddev_t *mddev, char *page)
3059 return sprintf(page, "%llu\n",
3060 (unsigned long long) mddev->resync_mismatches);
3063 static struct md_sysfs_entry md_scan_mode =
3064 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3067 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3070 sync_min_show(mddev_t *mddev, char *page)
3072 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3073 mddev->sync_speed_min ? "local": "system");
3077 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3081 if (strncmp(buf, "system", 6)==0) {
3082 mddev->sync_speed_min = 0;
3085 min = simple_strtoul(buf, &e, 10);
3086 if (buf == e || (*e && *e != '\n') || min <= 0)
3088 mddev->sync_speed_min = min;
3092 static struct md_sysfs_entry md_sync_min =
3093 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3096 sync_max_show(mddev_t *mddev, char *page)
3098 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3099 mddev->sync_speed_max ? "local": "system");
3103 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3107 if (strncmp(buf, "system", 6)==0) {
3108 mddev->sync_speed_max = 0;
3111 max = simple_strtoul(buf, &e, 10);
3112 if (buf == e || (*e && *e != '\n') || max <= 0)
3114 mddev->sync_speed_max = max;
3118 static struct md_sysfs_entry md_sync_max =
3119 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3122 degraded_show(mddev_t *mddev, char *page)
3124 return sprintf(page, "%d\n", mddev->degraded);
3126 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3129 sync_force_parallel_show(mddev_t *mddev, char *page)
3131 return sprintf(page, "%d\n", mddev->parallel_resync);
3135 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3139 if (strict_strtol(buf, 10, &n))
3142 if (n != 0 && n != 1)
3145 mddev->parallel_resync = n;
3147 if (mddev->sync_thread)
3148 wake_up(&resync_wait);
3153 /* force parallel resync, even with shared block devices */
3154 static struct md_sysfs_entry md_sync_force_parallel =
3155 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3156 sync_force_parallel_show, sync_force_parallel_store);
3159 sync_speed_show(mddev_t *mddev, char *page)
3161 unsigned long resync, dt, db;
3162 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3163 dt = (jiffies - mddev->resync_mark) / HZ;
3165 db = resync - mddev->resync_mark_cnt;
3166 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3169 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3172 sync_completed_show(mddev_t *mddev, char *page)
3174 unsigned long max_blocks, resync;
3176 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3177 max_blocks = mddev->resync_max_sectors;
3179 max_blocks = mddev->size << 1;
3181 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3182 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3185 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3188 min_sync_show(mddev_t *mddev, char *page)
3190 return sprintf(page, "%llu\n",
3191 (unsigned long long)mddev->resync_min);
3194 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3196 unsigned long long min;
3197 if (strict_strtoull(buf, 10, &min))
3199 if (min > mddev->resync_max)
3201 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3204 /* Must be a multiple of chunk_size */
3205 if (mddev->chunk_size) {
3206 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3209 mddev->resync_min = min;
3214 static struct md_sysfs_entry md_min_sync =
3215 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3218 max_sync_show(mddev_t *mddev, char *page)
3220 if (mddev->resync_max == MaxSector)
3221 return sprintf(page, "max\n");
3223 return sprintf(page, "%llu\n",
3224 (unsigned long long)mddev->resync_max);
3227 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3229 if (strncmp(buf, "max", 3) == 0)
3230 mddev->resync_max = MaxSector;
3232 unsigned long long max;
3233 if (strict_strtoull(buf, 10, &max))
3235 if (max < mddev->resync_min)
3237 if (max < mddev->resync_max &&
3238 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3241 /* Must be a multiple of chunk_size */
3242 if (mddev->chunk_size) {
3243 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3246 mddev->resync_max = max;
3248 wake_up(&mddev->recovery_wait);
3252 static struct md_sysfs_entry md_max_sync =
3253 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3256 suspend_lo_show(mddev_t *mddev, char *page)
3258 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3262 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3265 unsigned long long new = simple_strtoull(buf, &e, 10);
3267 if (mddev->pers->quiesce == NULL)
3269 if (buf == e || (*e && *e != '\n'))
3271 if (new >= mddev->suspend_hi ||
3272 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3273 mddev->suspend_lo = new;
3274 mddev->pers->quiesce(mddev, 2);
3279 static struct md_sysfs_entry md_suspend_lo =
3280 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3284 suspend_hi_show(mddev_t *mddev, char *page)
3286 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3290 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3293 unsigned long long new = simple_strtoull(buf, &e, 10);
3295 if (mddev->pers->quiesce == NULL)
3297 if (buf == e || (*e && *e != '\n'))
3299 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3300 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3301 mddev->suspend_hi = new;
3302 mddev->pers->quiesce(mddev, 1);
3303 mddev->pers->quiesce(mddev, 0);
3308 static struct md_sysfs_entry md_suspend_hi =
3309 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3312 reshape_position_show(mddev_t *mddev, char *page)
3314 if (mddev->reshape_position != MaxSector)
3315 return sprintf(page, "%llu\n",
3316 (unsigned long long)mddev->reshape_position);
3317 strcpy(page, "none\n");
3322 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3325 unsigned long long new = simple_strtoull(buf, &e, 10);
3328 if (buf == e || (*e && *e != '\n'))
3330 mddev->reshape_position = new;
3331 mddev->delta_disks = 0;
3332 mddev->new_level = mddev->level;
3333 mddev->new_layout = mddev->layout;
3334 mddev->new_chunk = mddev->chunk_size;
3338 static struct md_sysfs_entry md_reshape_position =
3339 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3340 reshape_position_store);
3343 static struct attribute *md_default_attrs[] = {
3346 &md_raid_disks.attr,
3347 &md_chunk_size.attr,
3349 &md_resync_start.attr,
3351 &md_new_device.attr,
3352 &md_safe_delay.attr,
3353 &md_array_state.attr,
3354 &md_reshape_position.attr,
3358 static struct attribute *md_redundancy_attrs[] = {
3360 &md_mismatches.attr,
3363 &md_sync_speed.attr,
3364 &md_sync_force_parallel.attr,
3365 &md_sync_completed.attr,
3368 &md_suspend_lo.attr,
3369 &md_suspend_hi.attr,
3374 static struct attribute_group md_redundancy_group = {
3376 .attrs = md_redundancy_attrs,
3381 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3383 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3384 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3389 rv = mddev_lock(mddev);
3391 rv = entry->show(mddev, page);
3392 mddev_unlock(mddev);
3398 md_attr_store(struct kobject *kobj, struct attribute *attr,
3399 const char *page, size_t length)
3401 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3402 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3407 if (!capable(CAP_SYS_ADMIN))
3409 rv = mddev_lock(mddev);
3411 rv = entry->store(mddev, page, length);
3412 mddev_unlock(mddev);
3417 static void md_free(struct kobject *ko)
3419 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3423 static struct sysfs_ops md_sysfs_ops = {
3424 .show = md_attr_show,
3425 .store = md_attr_store,
3427 static struct kobj_type md_ktype = {
3429 .sysfs_ops = &md_sysfs_ops,
3430 .default_attrs = md_default_attrs,
3435 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3437 static DEFINE_MUTEX(disks_mutex);
3438 mddev_t *mddev = mddev_find(dev);
3439 struct gendisk *disk;
3440 int partitioned = (MAJOR(dev) != MD_MAJOR);
3441 int shift = partitioned ? MdpMinorShift : 0;
3442 int unit = MINOR(dev) >> shift;
3448 mutex_lock(&disks_mutex);
3449 if (mddev->gendisk) {
3450 mutex_unlock(&disks_mutex);
3454 disk = alloc_disk(1 << shift);
3456 mutex_unlock(&disks_mutex);
3460 disk->major = MAJOR(dev);
3461 disk->first_minor = unit << shift;
3463 sprintf(disk->disk_name, "md_d%d", unit);
3465 sprintf(disk->disk_name, "md%d", unit);
3466 disk->fops = &md_fops;
3467 disk->private_data = mddev;
3468 disk->queue = mddev->queue;
3470 mddev->gendisk = disk;
3471 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3473 mutex_unlock(&disks_mutex);
3475 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3478 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3482 static void md_safemode_timeout(unsigned long data)
3484 mddev_t *mddev = (mddev_t *) data;
3486 if (!atomic_read(&mddev->writes_pending)) {
3487 mddev->safemode = 1;
3488 if (mddev->external)
3489 sysfs_notify(&mddev->kobj, NULL, "array_state");
3491 md_wakeup_thread(mddev->thread);
3494 static int start_dirty_degraded;
3496 static int do_md_run(mddev_t * mddev)
3500 struct list_head *tmp;
3502 struct gendisk *disk;
3503 struct mdk_personality *pers;
3504 char b[BDEVNAME_SIZE];
3506 if (list_empty(&mddev->disks))
3507 /* cannot run an array with no devices.. */
3514 * Analyze all RAID superblock(s)
3516 if (!mddev->raid_disks) {
3517 if (!mddev->persistent)
3522 chunk_size = mddev->chunk_size;
3525 if (chunk_size > MAX_CHUNK_SIZE) {
3526 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3527 chunk_size, MAX_CHUNK_SIZE);
3531 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3533 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3534 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3537 if (chunk_size < PAGE_SIZE) {
3538 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3539 chunk_size, PAGE_SIZE);
3543 /* devices must have minimum size of one chunk */
3544 rdev_for_each(rdev, tmp, mddev) {
3545 if (test_bit(Faulty, &rdev->flags))
3547 if (rdev->size < chunk_size / 1024) {
3549 "md: Dev %s smaller than chunk_size:"
3551 bdevname(rdev->bdev,b),
3552 (unsigned long long)rdev->size,
3560 if (mddev->level != LEVEL_NONE)
3561 request_module("md-level-%d", mddev->level);
3562 else if (mddev->clevel[0])
3563 request_module("md-%s", mddev->clevel);
3567 * Drop all container device buffers, from now on
3568 * the only valid external interface is through the md
3571 rdev_for_each(rdev, tmp, mddev) {
3572 if (test_bit(Faulty, &rdev->flags))
3574 sync_blockdev(rdev->bdev);
3575 invalidate_bdev(rdev->bdev);
3577 /* perform some consistency tests on the device.
3578 * We don't want the data to overlap the metadata,
3579 * Internal Bitmap issues has handled elsewhere.
3581 if (rdev->data_offset < rdev->sb_offset) {
3583 rdev->data_offset + mddev->size*2
3584 > rdev->sb_offset*2) {
3585 printk("md: %s: data overlaps metadata\n",
3590 if (rdev->sb_offset*2 + rdev->sb_size/512
3591 > rdev->data_offset) {
3592 printk("md: %s: metadata overlaps data\n",
3597 sysfs_notify(&rdev->kobj, NULL, "state");
3600 md_probe(mddev->unit, NULL, NULL);
3601 disk = mddev->gendisk;
3605 spin_lock(&pers_lock);
3606 pers = find_pers(mddev->level, mddev->clevel);
3607 if (!pers || !try_module_get(pers->owner)) {
3608 spin_unlock(&pers_lock);
3609 if (mddev->level != LEVEL_NONE)
3610 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3613 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3618 spin_unlock(&pers_lock);
3619 mddev->level = pers->level;
3620 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3622 if (mddev->reshape_position != MaxSector &&
3623 pers->start_reshape == NULL) {
3624 /* This personality cannot handle reshaping... */
3626 module_put(pers->owner);
3630 if (pers->sync_request) {
3631 /* Warn if this is a potentially silly
3634 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3636 struct list_head *tmp2;
3638 rdev_for_each(rdev, tmp, mddev) {
3639 rdev_for_each(rdev2, tmp2, mddev) {
3641 rdev->bdev->bd_contains ==
3642 rdev2->bdev->bd_contains) {
3644 "%s: WARNING: %s appears to be"
3645 " on the same physical disk as"
3648 bdevname(rdev->bdev,b),
3649 bdevname(rdev2->bdev,b2));
3656 "True protection against single-disk"
3657 " failure might be compromised.\n");
3660 mddev->recovery = 0;
3661 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3662 mddev->barriers_work = 1;
3663 mddev->ok_start_degraded = start_dirty_degraded;
3666 mddev->ro = 2; /* read-only, but switch on first write */
3668 err = mddev->pers->run(mddev);
3670 printk(KERN_ERR "md: pers->run() failed ...\n");
3671 else if (mddev->pers->sync_request) {
3672 err = bitmap_create(mddev);
3674 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3675 mdname(mddev), err);
3676 mddev->pers->stop(mddev);
3680 module_put(mddev->pers->owner);
3682 bitmap_destroy(mddev);
3685 if (mddev->pers->sync_request) {
3686 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3688 "md: cannot register extra attributes for %s\n",
3690 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3693 atomic_set(&mddev->writes_pending,0);
3694 mddev->safemode = 0;
3695 mddev->safemode_timer.function = md_safemode_timeout;
3696 mddev->safemode_timer.data = (unsigned long) mddev;
3697 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3700 rdev_for_each(rdev, tmp, mddev)
3701 if (rdev->raid_disk >= 0) {
3703 sprintf(nm, "rd%d", rdev->raid_disk);
3704 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3705 printk("md: cannot register %s for %s\n",
3709 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3712 md_update_sb(mddev, 0);
3714 set_capacity(disk, mddev->array_size<<1);
3716 /* If we call blk_queue_make_request here, it will
3717 * re-initialise max_sectors etc which may have been
3718 * refined inside -> run. So just set the bits we need to set.
3719 * Most initialisation happended when we called
3720 * blk_queue_make_request(..., md_fail_request)
3723 mddev->queue->queuedata = mddev;
3724 mddev->queue->make_request_fn = mddev->pers->make_request;
3726 /* If there is a partially-recovered drive we need to
3727 * start recovery here. If we leave it to md_check_recovery,
3728 * it will remove the drives and not do the right thing
3730 if (mddev->degraded && !mddev->sync_thread) {
3731 struct list_head *rtmp;
3733 rdev_for_each(rdev, rtmp, mddev)
3734 if (rdev->raid_disk >= 0 &&
3735 !test_bit(In_sync, &rdev->flags) &&
3736 !test_bit(Faulty, &rdev->flags))
3737 /* complete an interrupted recovery */
3739 if (spares && mddev->pers->sync_request) {
3740 mddev->recovery = 0;
3741 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3742 mddev->sync_thread = md_register_thread(md_do_sync,
3745 if (!mddev->sync_thread) {
3746 printk(KERN_ERR "%s: could not start resync"
3749 /* leave the spares where they are, it shouldn't hurt */
3750 mddev->recovery = 0;
3754 md_wakeup_thread(mddev->thread);
3755 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3758 md_new_event(mddev);
3759 sysfs_notify(&mddev->kobj, NULL, "array_state");
3760 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3761 sysfs_notify(&mddev->kobj, NULL, "degraded");
3762 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3766 static int restart_array(mddev_t *mddev)
3768 struct gendisk *disk = mddev->gendisk;
3770 /* Complain if it has no devices */
3771 if (list_empty(&mddev->disks))
3777 mddev->safemode = 0;
3779 set_disk_ro(disk, 0);
3780 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3782 /* Kick recovery or resync if necessary */
3783 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3784 md_wakeup_thread(mddev->thread);
3785 md_wakeup_thread(mddev->sync_thread);
3786 sysfs_notify(&mddev->kobj, NULL, "array_state");
3790 /* similar to deny_write_access, but accounts for our holding a reference
3791 * to the file ourselves */
3792 static int deny_bitmap_write_access(struct file * file)
3794 struct inode *inode = file->f_mapping->host;
3796 spin_lock(&inode->i_lock);
3797 if (atomic_read(&inode->i_writecount) > 1) {
3798 spin_unlock(&inode->i_lock);
3801 atomic_set(&inode->i_writecount, -1);
3802 spin_unlock(&inode->i_lock);
3807 static void restore_bitmap_write_access(struct file *file)
3809 struct inode *inode = file->f_mapping->host;
3811 spin_lock(&inode->i_lock);
3812 atomic_set(&inode->i_writecount, 1);
3813 spin_unlock(&inode->i_lock);
3817 * 0 - completely stop and dis-assemble array
3818 * 1 - switch to readonly
3819 * 2 - stop but do not disassemble array
3821 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3824 struct gendisk *disk = mddev->gendisk;
3826 if (atomic_read(&mddev->active) > 1 + is_open) {
3827 printk("md: %s still in use.\n",mdname(mddev));
3833 if (mddev->sync_thread) {
3834 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3835 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3836 md_unregister_thread(mddev->sync_thread);
3837 mddev->sync_thread = NULL;
3840 del_timer_sync(&mddev->safemode_timer);
3842 invalidate_partition(disk, 0);
3845 case 1: /* readonly */
3851 case 0: /* disassemble */
3853 bitmap_flush(mddev);
3854 md_super_wait(mddev);
3856 set_disk_ro(disk, 0);
3857 blk_queue_make_request(mddev->queue, md_fail_request);
3858 mddev->pers->stop(mddev);
3859 mddev->queue->merge_bvec_fn = NULL;
3860 mddev->queue->unplug_fn = NULL;
3861 mddev->queue->backing_dev_info.congested_fn = NULL;
3862 if (mddev->pers->sync_request)
3863 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3865 module_put(mddev->pers->owner);
3867 /* tell userspace to handle 'inactive' */
3868 sysfs_notify(&mddev->kobj, NULL, "array_state");
3870 set_capacity(disk, 0);
3876 if (!mddev->in_sync || mddev->flags) {
3877 /* mark array as shutdown cleanly */
3879 md_update_sb(mddev, 1);
3882 set_disk_ro(disk, 1);
3883 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3887 * Free resources if final stop
3891 struct list_head *tmp;
3893 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3895 bitmap_destroy(mddev);
3896 if (mddev->bitmap_file) {
3897 restore_bitmap_write_access(mddev->bitmap_file);
3898 fput(mddev->bitmap_file);
3899 mddev->bitmap_file = NULL;
3901 mddev->bitmap_offset = 0;
3903 rdev_for_each(rdev, tmp, mddev)
3904 if (rdev->raid_disk >= 0) {
3906 sprintf(nm, "rd%d", rdev->raid_disk);
3907 sysfs_remove_link(&mddev->kobj, nm);
3910 /* make sure all md_delayed_delete calls have finished */
3911 flush_scheduled_work();
3913 export_array(mddev);
3915 mddev->array_size = 0;
3917 mddev->raid_disks = 0;
3918 mddev->recovery_cp = 0;
3919 mddev->resync_min = 0;
3920 mddev->resync_max = MaxSector;
3921 mddev->reshape_position = MaxSector;
3922 mddev->external = 0;
3923 mddev->persistent = 0;
3924 mddev->level = LEVEL_NONE;
3925 mddev->clevel[0] = 0;
3928 mddev->metadata_type[0] = 0;
3929 mddev->chunk_size = 0;
3930 mddev->ctime = mddev->utime = 0;
3932 mddev->max_disks = 0;
3934 mddev->delta_disks = 0;
3935 mddev->new_level = LEVEL_NONE;
3936 mddev->new_layout = 0;
3937 mddev->new_chunk = 0;
3938 mddev->curr_resync = 0;
3939 mddev->resync_mismatches = 0;
3940 mddev->suspend_lo = mddev->suspend_hi = 0;
3941 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3942 mddev->recovery = 0;
3945 mddev->degraded = 0;
3946 mddev->barriers_work = 0;
3947 mddev->safemode = 0;
3949 } else if (mddev->pers)
3950 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3953 md_new_event(mddev);
3954 sysfs_notify(&mddev->kobj, NULL, "array_state");
3960 static void autorun_array(mddev_t *mddev)
3963 struct list_head *tmp;
3966 if (list_empty(&mddev->disks))
3969 printk(KERN_INFO "md: running: ");
3971 rdev_for_each(rdev, tmp, mddev) {
3972 char b[BDEVNAME_SIZE];
3973 printk("<%s>", bdevname(rdev->bdev,b));
3977 err = do_md_run (mddev);
3979 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3980 do_md_stop (mddev, 0, 0);
3985 * lets try to run arrays based on all disks that have arrived
3986 * until now. (those are in pending_raid_disks)
3988 * the method: pick the first pending disk, collect all disks with
3989 * the same UUID, remove all from the pending list and put them into
3990 * the 'same_array' list. Then order this list based on superblock
3991 * update time (freshest comes first), kick out 'old' disks and
3992 * compare superblocks. If everything's fine then run it.
3994 * If "unit" is allocated, then bump its reference count
3996 static void autorun_devices(int part)
3998 struct list_head *tmp;
3999 mdk_rdev_t *rdev0, *rdev;
4001 char b[BDEVNAME_SIZE];
4003 printk(KERN_INFO "md: autorun ...\n");
4004 while (!list_empty(&pending_raid_disks)) {
4007 LIST_HEAD(candidates);
4008 rdev0 = list_entry(pending_raid_disks.next,
4009 mdk_rdev_t, same_set);
4011 printk(KERN_INFO "md: considering %s ...\n",
4012 bdevname(rdev0->bdev,b));
4013 INIT_LIST_HEAD(&candidates);
4014 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4015 if (super_90_load(rdev, rdev0, 0) >= 0) {
4016 printk(KERN_INFO "md: adding %s ...\n",
4017 bdevname(rdev->bdev,b));
4018 list_move(&rdev->same_set, &candidates);
4021 * now we have a set of devices, with all of them having
4022 * mostly sane superblocks. It's time to allocate the
4026 dev = MKDEV(mdp_major,
4027 rdev0->preferred_minor << MdpMinorShift);
4028 unit = MINOR(dev) >> MdpMinorShift;
4030 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4033 if (rdev0->preferred_minor != unit) {
4034 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4035 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4039 md_probe(dev, NULL, NULL);
4040 mddev = mddev_find(dev);
4041 if (!mddev || !mddev->gendisk) {
4045 "md: cannot allocate memory for md drive.\n");
4048 if (mddev_lock(mddev))
4049 printk(KERN_WARNING "md: %s locked, cannot run\n",
4051 else if (mddev->raid_disks || mddev->major_version
4052 || !list_empty(&mddev->disks)) {
4054 "md: %s already running, cannot run %s\n",
4055 mdname(mddev), bdevname(rdev0->bdev,b));
4056 mddev_unlock(mddev);
4058 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4059 mddev->persistent = 1;
4060 rdev_for_each_list(rdev, tmp, candidates) {
4061 list_del_init(&rdev->same_set);
4062 if (bind_rdev_to_array(rdev, mddev))
4065 autorun_array(mddev);
4066 mddev_unlock(mddev);
4068 /* on success, candidates will be empty, on error
4071 rdev_for_each_list(rdev, tmp, candidates)
4075 printk(KERN_INFO "md: ... autorun DONE.\n");
4077 #endif /* !MODULE */
4079 static int get_version(void __user * arg)
4083 ver.major = MD_MAJOR_VERSION;
4084 ver.minor = MD_MINOR_VERSION;
4085 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4087 if (copy_to_user(arg, &ver, sizeof(ver)))
4093 static int get_array_info(mddev_t * mddev, void __user * arg)
4095 mdu_array_info_t info;
4096 int nr,working,active,failed,spare;
4098 struct list_head *tmp;
4100 nr=working=active=failed=spare=0;
4101 rdev_for_each(rdev, tmp, mddev) {
4103 if (test_bit(Faulty, &rdev->flags))
4107 if (test_bit(In_sync, &rdev->flags))
4114 info.major_version = mddev->major_version;
4115 info.minor_version = mddev->minor_version;
4116 info.patch_version = MD_PATCHLEVEL_VERSION;
4117 info.ctime = mddev->ctime;
4118 info.level = mddev->level;
4119 info.size = mddev->size;
4120 if (info.size != mddev->size) /* overflow */
4123 info.raid_disks = mddev->raid_disks;
4124 info.md_minor = mddev->md_minor;
4125 info.not_persistent= !mddev->persistent;
4127 info.utime = mddev->utime;
4130 info.state = (1<<MD_SB_CLEAN);
4131 if (mddev->bitmap && mddev->bitmap_offset)
4132 info.state = (1<<MD_SB_BITMAP_PRESENT);
4133 info.active_disks = active;
4134 info.working_disks = working;
4135 info.failed_disks = failed;
4136 info.spare_disks = spare;
4138 info.layout = mddev->layout;
4139 info.chunk_size = mddev->chunk_size;
4141 if (copy_to_user(arg, &info, sizeof(info)))
4147 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4149 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4150 char *ptr, *buf = NULL;
4153 if (md_allow_write(mddev))
4154 file = kmalloc(sizeof(*file), GFP_NOIO);
4156 file = kmalloc(sizeof(*file), GFP_KERNEL);
4161 /* bitmap disabled, zero the first byte and copy out */
4162 if (!mddev->bitmap || !mddev->bitmap->file) {
4163 file->pathname[0] = '\0';
4167 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4171 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4175 strcpy(file->pathname, ptr);
4179 if (copy_to_user(arg, file, sizeof(*file)))
4187 static int get_disk_info(mddev_t * mddev, void __user * arg)
4189 mdu_disk_info_t info;
4192 if (copy_from_user(&info, arg, sizeof(info)))
4195 rdev = find_rdev_nr(mddev, info.number);
4197 info.major = MAJOR(rdev->bdev->bd_dev);
4198 info.minor = MINOR(rdev->bdev->bd_dev);
4199 info.raid_disk = rdev->raid_disk;
4201 if (test_bit(Faulty, &rdev->flags))
4202 info.state |= (1<<MD_DISK_FAULTY);
4203 else if (test_bit(In_sync, &rdev->flags)) {
4204 info.state |= (1<<MD_DISK_ACTIVE);
4205 info.state |= (1<<MD_DISK_SYNC);
4207 if (test_bit(WriteMostly, &rdev->flags))
4208 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4210 info.major = info.minor = 0;
4211 info.raid_disk = -1;
4212 info.state = (1<<MD_DISK_REMOVED);
4215 if (copy_to_user(arg, &info, sizeof(info)))
4221 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4223 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4225 dev_t dev = MKDEV(info->major,info->minor);
4227 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4230 if (!mddev->raid_disks) {
4232 /* expecting a device which has a superblock */
4233 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4236 "md: md_import_device returned %ld\n",
4238 return PTR_ERR(rdev);
4240 if (!list_empty(&mddev->disks)) {
4241 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4242 mdk_rdev_t, same_set);
4243 int err = super_types[mddev->major_version]
4244 .load_super(rdev, rdev0, mddev->minor_version);
4247 "md: %s has different UUID to %s\n",
4248 bdevname(rdev->bdev,b),
4249 bdevname(rdev0->bdev,b2));
4254 err = bind_rdev_to_array(rdev, mddev);
4261 * add_new_disk can be used once the array is assembled
4262 * to add "hot spares". They must already have a superblock
4267 if (!mddev->pers->hot_add_disk) {
4269 "%s: personality does not support diskops!\n",
4273 if (mddev->persistent)
4274 rdev = md_import_device(dev, mddev->major_version,
4275 mddev->minor_version);
4277 rdev = md_import_device(dev, -1, -1);
4280 "md: md_import_device returned %ld\n",
4282 return PTR_ERR(rdev);
4284 /* set save_raid_disk if appropriate */
4285 if (!mddev->persistent) {
4286 if (info->state & (1<<MD_DISK_SYNC) &&
4287 info->raid_disk < mddev->raid_disks)
4288 rdev->raid_disk = info->raid_disk;
4290 rdev->raid_disk = -1;
4292 super_types[mddev->major_version].
4293 validate_super(mddev, rdev);
4294 rdev->saved_raid_disk = rdev->raid_disk;
4296 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4297 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4298 set_bit(WriteMostly, &rdev->flags);
4300 rdev->raid_disk = -1;
4301 err = bind_rdev_to_array(rdev, mddev);
4302 if (!err && !mddev->pers->hot_remove_disk) {
4303 /* If there is hot_add_disk but no hot_remove_disk
4304 * then added disks for geometry changes,
4305 * and should be added immediately.
4307 super_types[mddev->major_version].
4308 validate_super(mddev, rdev);
4309 err = mddev->pers->hot_add_disk(mddev, rdev);
4311 unbind_rdev_from_array(rdev);
4316 sysfs_notify(&rdev->kobj, NULL, "state");
4318 md_update_sb(mddev, 1);
4319 if (mddev->degraded)
4320 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4321 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4322 md_wakeup_thread(mddev->thread);
4326 /* otherwise, add_new_disk is only allowed
4327 * for major_version==0 superblocks
4329 if (mddev->major_version != 0) {
4330 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4335 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4337 rdev = md_import_device (dev, -1, 0);
4340 "md: error, md_import_device() returned %ld\n",
4342 return PTR_ERR(rdev);
4344 rdev->desc_nr = info->number;
4345 if (info->raid_disk < mddev->raid_disks)
4346 rdev->raid_disk = info->raid_disk;
4348 rdev->raid_disk = -1;
4350 if (rdev->raid_disk < mddev->raid_disks)
4351 if (info->state & (1<<MD_DISK_SYNC))
4352 set_bit(In_sync, &rdev->flags);
4354 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4355 set_bit(WriteMostly, &rdev->flags);
4357 if (!mddev->persistent) {
4358 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4359 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4361 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4362 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4364 err = bind_rdev_to_array(rdev, mddev);
4374 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4376 char b[BDEVNAME_SIZE];
4379 rdev = find_rdev(mddev, dev);
4383 if (rdev->raid_disk >= 0)
4386 kick_rdev_from_array(rdev);
4387 md_update_sb(mddev, 1);
4388 md_new_event(mddev);
4392 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4393 bdevname(rdev->bdev,b), mdname(mddev));
4397 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4399 char b[BDEVNAME_SIZE];
4407 if (mddev->major_version != 0) {
4408 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4409 " version-0 superblocks.\n",
4413 if (!mddev->pers->hot_add_disk) {
4415 "%s: personality does not support diskops!\n",
4420 rdev = md_import_device (dev, -1, 0);
4423 "md: error, md_import_device() returned %ld\n",
4428 if (mddev->persistent)
4429 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4432 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4434 size = calc_dev_size(rdev, mddev->chunk_size);
4437 if (test_bit(Faulty, &rdev->flags)) {
4439 "md: can not hot-add faulty %s disk to %s!\n",
4440 bdevname(rdev->bdev,b), mdname(mddev));
4444 clear_bit(In_sync, &rdev->flags);
4446 rdev->saved_raid_disk = -1;
4447 err = bind_rdev_to_array(rdev, mddev);
4452 * The rest should better be atomic, we can have disk failures
4453 * noticed in interrupt contexts ...
4456 if (rdev->desc_nr == mddev->max_disks) {
4457 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4460 goto abort_unbind_export;
4463 rdev->raid_disk = -1;
4465 md_update_sb(mddev, 1);
4468 * Kick recovery, maybe this spare has to be added to the
4469 * array immediately.
4471 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4472 md_wakeup_thread(mddev->thread);
4473 md_new_event(mddev);
4476 abort_unbind_export:
4477 unbind_rdev_from_array(rdev);
4484 static int set_bitmap_file(mddev_t *mddev, int fd)
4489 if (!mddev->pers->quiesce)
4491 if (mddev->recovery || mddev->sync_thread)
4493 /* we should be able to change the bitmap.. */
4499 return -EEXIST; /* cannot add when bitmap is present */
4500 mddev->bitmap_file = fget(fd);
4502 if (mddev->bitmap_file == NULL) {
4503 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4508 err = deny_bitmap_write_access(mddev->bitmap_file);
4510 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4512 fput(mddev->bitmap_file);
4513 mddev->bitmap_file = NULL;
4516 mddev->bitmap_offset = 0; /* file overrides offset */
4517 } else if (mddev->bitmap == NULL)
4518 return -ENOENT; /* cannot remove what isn't there */
4521 mddev->pers->quiesce(mddev, 1);
4523 err = bitmap_create(mddev);
4524 if (fd < 0 || err) {
4525 bitmap_destroy(mddev);
4526 fd = -1; /* make sure to put the file */
4528 mddev->pers->quiesce(mddev, 0);
4531 if (mddev->bitmap_file) {
4532 restore_bitmap_write_access(mddev->bitmap_file);
4533 fput(mddev->bitmap_file);
4535 mddev->bitmap_file = NULL;
4542 * set_array_info is used two different ways
4543 * The original usage is when creating a new array.
4544 * In this usage, raid_disks is > 0 and it together with
4545 * level, size, not_persistent,layout,chunksize determine the
4546 * shape of the array.
4547 * This will always create an array with a type-0.90.0 superblock.
4548 * The newer usage is when assembling an array.
4549 * In this case raid_disks will be 0, and the major_version field is
4550 * use to determine which style super-blocks are to be found on the devices.
4551 * The minor and patch _version numbers are also kept incase the
4552 * super_block handler wishes to interpret them.
4554 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4557 if (info->raid_disks == 0) {
4558 /* just setting version number for superblock loading */
4559 if (info->major_version < 0 ||
4560 info->major_version >= ARRAY_SIZE(super_types) ||
4561 super_types[info->major_version].name == NULL) {
4562 /* maybe try to auto-load a module? */
4564 "md: superblock version %d not known\n",
4565 info->major_version);
4568 mddev->major_version = info->major_version;
4569 mddev->minor_version = info->minor_version;
4570 mddev->patch_version = info->patch_version;
4571 mddev->persistent = !info->not_persistent;
4574 mddev->major_version = MD_MAJOR_VERSION;
4575 mddev->minor_version = MD_MINOR_VERSION;
4576 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4577 mddev->ctime = get_seconds();
4579 mddev->level = info->level;
4580 mddev->clevel[0] = 0;
4581 mddev->size = info->size;
4582 mddev->raid_disks = info->raid_disks;
4583 /* don't set md_minor, it is determined by which /dev/md* was
4586 if (info->state & (1<<MD_SB_CLEAN))
4587 mddev->recovery_cp = MaxSector;
4589 mddev->recovery_cp = 0;
4590 mddev->persistent = ! info->not_persistent;
4591 mddev->external = 0;
4593 mddev->layout = info->layout;
4594 mddev->chunk_size = info->chunk_size;
4596 mddev->max_disks = MD_SB_DISKS;
4598 if (mddev->persistent)
4600 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4602 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4603 mddev->bitmap_offset = 0;
4605 mddev->reshape_position = MaxSector;
4608 * Generate a 128 bit UUID
4610 get_random_bytes(mddev->uuid, 16);
4612 mddev->new_level = mddev->level;
4613 mddev->new_chunk = mddev->chunk_size;
4614 mddev->new_layout = mddev->layout;
4615 mddev->delta_disks = 0;
4620 static int update_size(mddev_t *mddev, sector_t num_sectors)
4624 struct list_head *tmp;
4625 int fit = (num_sectors == 0);
4627 if (mddev->pers->resize == NULL)
4629 /* The "num_sectors" is the number of sectors of each device that
4630 * is used. This can only make sense for arrays with redundancy.
4631 * linear and raid0 always use whatever space is available. We can only
4632 * consider changing this number if no resync or reconstruction is
4633 * happening, and if the new size is acceptable. It must fit before the
4634 * sb_offset or, if that is <data_offset, it must fit before the size
4635 * of each device. If num_sectors is zero, we find the largest size
4639 if (mddev->sync_thread)
4641 rdev_for_each(rdev, tmp, mddev) {
4643 avail = rdev->size * 2;
4645 if (fit && (num_sectors == 0 || num_sectors > avail))
4646 num_sectors = avail;
4647 if (avail < num_sectors)
4650 rv = mddev->pers->resize(mddev, num_sectors);
4652 struct block_device *bdev;
4654 bdev = bdget_disk(mddev->gendisk, 0);
4656 mutex_lock(&bdev->bd_inode->i_mutex);
4657 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4658 mutex_unlock(&bdev->bd_inode->i_mutex);
4665 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4668 /* change the number of raid disks */
4669 if (mddev->pers->check_reshape == NULL)
4671 if (raid_disks <= 0 ||
4672 raid_disks >= mddev->max_disks)
4674 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4676 mddev->delta_disks = raid_disks - mddev->raid_disks;
4678 rv = mddev->pers->check_reshape(mddev);
4684 * update_array_info is used to change the configuration of an
4686 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4687 * fields in the info are checked against the array.
4688 * Any differences that cannot be handled will cause an error.
4689 * Normally, only one change can be managed at a time.
4691 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4697 /* calculate expected state,ignoring low bits */
4698 if (mddev->bitmap && mddev->bitmap_offset)
4699 state |= (1 << MD_SB_BITMAP_PRESENT);
4701 if (mddev->major_version != info->major_version ||
4702 mddev->minor_version != info->minor_version ||
4703 /* mddev->patch_version != info->patch_version || */
4704 mddev->ctime != info->ctime ||
4705 mddev->level != info->level ||
4706 /* mddev->layout != info->layout || */
4707 !mddev->persistent != info->not_persistent||
4708 mddev->chunk_size != info->chunk_size ||
4709 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4710 ((state^info->state) & 0xfffffe00)
4713 /* Check there is only one change */
4714 if (info->size >= 0 && mddev->size != info->size) cnt++;
4715 if (mddev->raid_disks != info->raid_disks) cnt++;
4716 if (mddev->layout != info->layout) cnt++;
4717 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4718 if (cnt == 0) return 0;
4719 if (cnt > 1) return -EINVAL;
4721 if (mddev->layout != info->layout) {
4723 * we don't need to do anything at the md level, the
4724 * personality will take care of it all.
4726 if (mddev->pers->reconfig == NULL)
4729 return mddev->pers->reconfig(mddev, info->layout, -1);
4731 if (info->size >= 0 && mddev->size != info->size)
4732 rv = update_size(mddev, (sector_t)info->size * 2);
4734 if (mddev->raid_disks != info->raid_disks)
4735 rv = update_raid_disks(mddev, info->raid_disks);
4737 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4738 if (mddev->pers->quiesce == NULL)
4740 if (mddev->recovery || mddev->sync_thread)
4742 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4743 /* add the bitmap */
4746 if (mddev->default_bitmap_offset == 0)
4748 mddev->bitmap_offset = mddev->default_bitmap_offset;
4749 mddev->pers->quiesce(mddev, 1);
4750 rv = bitmap_create(mddev);
4752 bitmap_destroy(mddev);
4753 mddev->pers->quiesce(mddev, 0);
4755 /* remove the bitmap */
4758 if (mddev->bitmap->file)
4760 mddev->pers->quiesce(mddev, 1);
4761 bitmap_destroy(mddev);
4762 mddev->pers->quiesce(mddev, 0);
4763 mddev->bitmap_offset = 0;
4766 md_update_sb(mddev, 1);
4770 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4774 if (mddev->pers == NULL)
4777 rdev = find_rdev(mddev, dev);
4781 md_error(mddev, rdev);
4786 * We have a problem here : there is no easy way to give a CHS
4787 * virtual geometry. We currently pretend that we have a 2 heads
4788 * 4 sectors (with a BIG number of cylinders...). This drives
4789 * dosfs just mad... ;-)
4791 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4793 mddev_t *mddev = bdev->bd_disk->private_data;
4797 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4801 static int md_ioctl(struct inode *inode, struct file *file,
4802 unsigned int cmd, unsigned long arg)
4805 void __user *argp = (void __user *)arg;
4806 mddev_t *mddev = NULL;
4808 if (!capable(CAP_SYS_ADMIN))
4812 * Commands dealing with the RAID driver but not any
4818 err = get_version(argp);
4821 case PRINT_RAID_DEBUG:
4829 autostart_arrays(arg);
4836 * Commands creating/starting a new array:
4839 mddev = inode->i_bdev->bd_disk->private_data;
4846 err = mddev_lock(mddev);
4849 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4856 case SET_ARRAY_INFO:
4858 mdu_array_info_t info;
4860 memset(&info, 0, sizeof(info));
4861 else if (copy_from_user(&info, argp, sizeof(info))) {
4866 err = update_array_info(mddev, &info);
4868 printk(KERN_WARNING "md: couldn't update"
4869 " array info. %d\n", err);
4874 if (!list_empty(&mddev->disks)) {
4876 "md: array %s already has disks!\n",
4881 if (mddev->raid_disks) {
4883 "md: array %s already initialised!\n",
4888 err = set_array_info(mddev, &info);
4890 printk(KERN_WARNING "md: couldn't set"
4891 " array info. %d\n", err);
4901 * Commands querying/configuring an existing array:
4903 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4904 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4905 if ((!mddev->raid_disks && !mddev->external)
4906 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4907 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4908 && cmd != GET_BITMAP_FILE) {
4914 * Commands even a read-only array can execute:
4918 case GET_ARRAY_INFO:
4919 err = get_array_info(mddev, argp);
4922 case GET_BITMAP_FILE:
4923 err = get_bitmap_file(mddev, argp);
4927 err = get_disk_info(mddev, argp);
4930 case RESTART_ARRAY_RW:
4931 err = restart_array(mddev);
4935 err = do_md_stop (mddev, 0, 1);
4939 err = do_md_stop (mddev, 1, 1);
4945 * The remaining ioctls are changing the state of the
4946 * superblock, so we do not allow them on read-only arrays.
4947 * However non-MD ioctls (e.g. get-size) will still come through
4948 * here and hit the 'default' below, so only disallow
4949 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4951 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4952 if (mddev->ro == 2) {
4954 sysfs_notify(&mddev->kobj, NULL, "array_state");
4955 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4956 md_wakeup_thread(mddev->thread);
4967 mdu_disk_info_t info;
4968 if (copy_from_user(&info, argp, sizeof(info)))
4971 err = add_new_disk(mddev, &info);
4975 case HOT_REMOVE_DISK:
4976 err = hot_remove_disk(mddev, new_decode_dev(arg));
4980 err = hot_add_disk(mddev, new_decode_dev(arg));
4983 case SET_DISK_FAULTY:
4984 err = set_disk_faulty(mddev, new_decode_dev(arg));
4988 err = do_md_run (mddev);
4991 case SET_BITMAP_FILE:
4992 err = set_bitmap_file(mddev, (int)arg);
5002 mddev_unlock(mddev);
5012 static int md_open(struct inode *inode, struct file *file)
5015 * Succeed if we can lock the mddev, which confirms that
5016 * it isn't being stopped right now.
5018 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5021 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5026 mddev_unlock(mddev);
5028 check_disk_change(inode->i_bdev);
5033 static int md_release(struct inode *inode, struct file * file)
5035 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5043 static int md_media_changed(struct gendisk *disk)
5045 mddev_t *mddev = disk->private_data;
5047 return mddev->changed;
5050 static int md_revalidate(struct gendisk *disk)
5052 mddev_t *mddev = disk->private_data;
5057 static struct block_device_operations md_fops =
5059 .owner = THIS_MODULE,
5061 .release = md_release,
5063 .getgeo = md_getgeo,
5064 .media_changed = md_media_changed,
5065 .revalidate_disk= md_revalidate,
5068 static int md_thread(void * arg)
5070 mdk_thread_t *thread = arg;
5073 * md_thread is a 'system-thread', it's priority should be very
5074 * high. We avoid resource deadlocks individually in each
5075 * raid personality. (RAID5 does preallocation) We also use RR and
5076 * the very same RT priority as kswapd, thus we will never get
5077 * into a priority inversion deadlock.
5079 * we definitely have to have equal or higher priority than
5080 * bdflush, otherwise bdflush will deadlock if there are too
5081 * many dirty RAID5 blocks.
5084 allow_signal(SIGKILL);
5085 while (!kthread_should_stop()) {
5087 /* We need to wait INTERRUPTIBLE so that
5088 * we don't add to the load-average.
5089 * That means we need to be sure no signals are
5092 if (signal_pending(current))
5093 flush_signals(current);
5095 wait_event_interruptible_timeout
5097 test_bit(THREAD_WAKEUP, &thread->flags)
5098 || kthread_should_stop(),
5101 clear_bit(THREAD_WAKEUP, &thread->flags);
5103 thread->run(thread->mddev);
5109 void md_wakeup_thread(mdk_thread_t *thread)
5112 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5113 set_bit(THREAD_WAKEUP, &thread->flags);
5114 wake_up(&thread->wqueue);
5118 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5121 mdk_thread_t *thread;
5123 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5127 init_waitqueue_head(&thread->wqueue);
5130 thread->mddev = mddev;
5131 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5132 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5133 if (IS_ERR(thread->tsk)) {
5140 void md_unregister_thread(mdk_thread_t *thread)
5142 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5144 kthread_stop(thread->tsk);
5148 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5155 if (!rdev || test_bit(Faulty, &rdev->flags))
5158 if (mddev->external)
5159 set_bit(Blocked, &rdev->flags);
5161 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5163 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5164 __builtin_return_address(0),__builtin_return_address(1),
5165 __builtin_return_address(2),__builtin_return_address(3));
5169 if (!mddev->pers->error_handler)
5171 mddev->pers->error_handler(mddev,rdev);
5172 if (mddev->degraded)
5173 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5174 set_bit(StateChanged, &rdev->flags);
5175 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5176 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5177 md_wakeup_thread(mddev->thread);
5178 md_new_event_inintr(mddev);
5181 /* seq_file implementation /proc/mdstat */
5183 static void status_unused(struct seq_file *seq)
5187 struct list_head *tmp;
5189 seq_printf(seq, "unused devices: ");
5191 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5192 char b[BDEVNAME_SIZE];
5194 seq_printf(seq, "%s ",
5195 bdevname(rdev->bdev,b));
5198 seq_printf(seq, "<none>");
5200 seq_printf(seq, "\n");
5204 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5206 sector_t max_blocks, resync, res;
5207 unsigned long dt, db, rt;
5209 unsigned int per_milli;
5211 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5213 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5214 max_blocks = mddev->resync_max_sectors >> 1;
5216 max_blocks = mddev->size;
5219 * Should not happen.
5225 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5226 * in a sector_t, and (max_blocks>>scale) will fit in a
5227 * u32, as those are the requirements for sector_div.
5228 * Thus 'scale' must be at least 10
5231 if (sizeof(sector_t) > sizeof(unsigned long)) {
5232 while ( max_blocks/2 > (1ULL<<(scale+32)))
5235 res = (resync>>scale)*1000;
5236 sector_div(res, (u32)((max_blocks>>scale)+1));
5240 int i, x = per_milli/50, y = 20-x;
5241 seq_printf(seq, "[");
5242 for (i = 0; i < x; i++)
5243 seq_printf(seq, "=");
5244 seq_printf(seq, ">");
5245 for (i = 0; i < y; i++)
5246 seq_printf(seq, ".");
5247 seq_printf(seq, "] ");
5249 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5250 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5252 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5254 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5255 "resync" : "recovery"))),
5256 per_milli/10, per_milli % 10,
5257 (unsigned long long) resync,
5258 (unsigned long long) max_blocks);
5261 * We do not want to overflow, so the order of operands and
5262 * the * 100 / 100 trick are important. We do a +1 to be
5263 * safe against division by zero. We only estimate anyway.
5265 * dt: time from mark until now
5266 * db: blocks written from mark until now
5267 * rt: remaining time
5269 dt = ((jiffies - mddev->resync_mark) / HZ);
5271 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5272 - mddev->resync_mark_cnt;
5273 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5275 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5277 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5280 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5282 struct list_head *tmp;
5292 spin_lock(&all_mddevs_lock);
5293 list_for_each(tmp,&all_mddevs)
5295 mddev = list_entry(tmp, mddev_t, all_mddevs);
5297 spin_unlock(&all_mddevs_lock);
5300 spin_unlock(&all_mddevs_lock);
5302 return (void*)2;/* tail */
5306 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5308 struct list_head *tmp;
5309 mddev_t *next_mddev, *mddev = v;
5315 spin_lock(&all_mddevs_lock);
5317 tmp = all_mddevs.next;
5319 tmp = mddev->all_mddevs.next;
5320 if (tmp != &all_mddevs)
5321 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5323 next_mddev = (void*)2;
5326 spin_unlock(&all_mddevs_lock);
5334 static void md_seq_stop(struct seq_file *seq, void *v)
5338 if (mddev && v != (void*)1 && v != (void*)2)
5342 struct mdstat_info {
5346 static int md_seq_show(struct seq_file *seq, void *v)
5350 struct list_head *tmp2;
5352 struct mdstat_info *mi = seq->private;
5353 struct bitmap *bitmap;
5355 if (v == (void*)1) {
5356 struct mdk_personality *pers;
5357 seq_printf(seq, "Personalities : ");
5358 spin_lock(&pers_lock);
5359 list_for_each_entry(pers, &pers_list, list)
5360 seq_printf(seq, "[%s] ", pers->name);
5362 spin_unlock(&pers_lock);
5363 seq_printf(seq, "\n");
5364 mi->event = atomic_read(&md_event_count);
5367 if (v == (void*)2) {
5372 if (mddev_lock(mddev) < 0)
5375 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5376 seq_printf(seq, "%s : %sactive", mdname(mddev),
5377 mddev->pers ? "" : "in");
5380 seq_printf(seq, " (read-only)");
5382 seq_printf(seq, " (auto-read-only)");
5383 seq_printf(seq, " %s", mddev->pers->name);
5387 rdev_for_each(rdev, tmp2, mddev) {
5388 char b[BDEVNAME_SIZE];
5389 seq_printf(seq, " %s[%d]",
5390 bdevname(rdev->bdev,b), rdev->desc_nr);
5391 if (test_bit(WriteMostly, &rdev->flags))
5392 seq_printf(seq, "(W)");
5393 if (test_bit(Faulty, &rdev->flags)) {
5394 seq_printf(seq, "(F)");
5396 } else if (rdev->raid_disk < 0)
5397 seq_printf(seq, "(S)"); /* spare */
5401 if (!list_empty(&mddev->disks)) {
5403 seq_printf(seq, "\n %llu blocks",
5404 (unsigned long long)mddev->array_size);
5406 seq_printf(seq, "\n %llu blocks",
5407 (unsigned long long)size);
5409 if (mddev->persistent) {
5410 if (mddev->major_version != 0 ||
5411 mddev->minor_version != 90) {
5412 seq_printf(seq," super %d.%d",
5413 mddev->major_version,
5414 mddev->minor_version);
5416 } else if (mddev->external)
5417 seq_printf(seq, " super external:%s",
5418 mddev->metadata_type);
5420 seq_printf(seq, " super non-persistent");
5423 mddev->pers->status (seq, mddev);
5424 seq_printf(seq, "\n ");
5425 if (mddev->pers->sync_request) {
5426 if (mddev->curr_resync > 2) {
5427 status_resync (seq, mddev);
5428 seq_printf(seq, "\n ");
5429 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5430 seq_printf(seq, "\tresync=DELAYED\n ");
5431 else if (mddev->recovery_cp < MaxSector)
5432 seq_printf(seq, "\tresync=PENDING\n ");
5435 seq_printf(seq, "\n ");
5437 if ((bitmap = mddev->bitmap)) {
5438 unsigned long chunk_kb;
5439 unsigned long flags;
5440 spin_lock_irqsave(&bitmap->lock, flags);
5441 chunk_kb = bitmap->chunksize >> 10;
5442 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5444 bitmap->pages - bitmap->missing_pages,
5446 (bitmap->pages - bitmap->missing_pages)
5447 << (PAGE_SHIFT - 10),
5448 chunk_kb ? chunk_kb : bitmap->chunksize,
5449 chunk_kb ? "KB" : "B");
5451 seq_printf(seq, ", file: ");
5452 seq_path(seq, &bitmap->file->f_path, " \t\n");
5455 seq_printf(seq, "\n");
5456 spin_unlock_irqrestore(&bitmap->lock, flags);
5459 seq_printf(seq, "\n");
5461 mddev_unlock(mddev);
5466 static struct seq_operations md_seq_ops = {
5467 .start = md_seq_start,
5468 .next = md_seq_next,
5469 .stop = md_seq_stop,
5470 .show = md_seq_show,
5473 static int md_seq_open(struct inode *inode, struct file *file)
5476 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5480 error = seq_open(file, &md_seq_ops);
5484 struct seq_file *p = file->private_data;
5486 mi->event = atomic_read(&md_event_count);
5491 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5493 struct seq_file *m = filp->private_data;
5494 struct mdstat_info *mi = m->private;
5497 poll_wait(filp, &md_event_waiters, wait);
5499 /* always allow read */
5500 mask = POLLIN | POLLRDNORM;
5502 if (mi->event != atomic_read(&md_event_count))
5503 mask |= POLLERR | POLLPRI;
5507 static const struct file_operations md_seq_fops = {
5508 .owner = THIS_MODULE,
5509 .open = md_seq_open,
5511 .llseek = seq_lseek,
5512 .release = seq_release_private,
5513 .poll = mdstat_poll,
5516 int register_md_personality(struct mdk_personality *p)
5518 spin_lock(&pers_lock);
5519 list_add_tail(&p->list, &pers_list);
5520 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5521 spin_unlock(&pers_lock);
5525 int unregister_md_personality(struct mdk_personality *p)
5527 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5528 spin_lock(&pers_lock);
5529 list_del_init(&p->list);
5530 spin_unlock(&pers_lock);
5534 static int is_mddev_idle(mddev_t *mddev)
5537 struct list_head *tmp;
5542 rdev_for_each(rdev, tmp, mddev) {
5543 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5544 curr_events = disk_stat_read(disk, sectors[0]) +
5545 disk_stat_read(disk, sectors[1]) -
5546 atomic_read(&disk->sync_io);
5547 /* sync IO will cause sync_io to increase before the disk_stats
5548 * as sync_io is counted when a request starts, and
5549 * disk_stats is counted when it completes.
5550 * So resync activity will cause curr_events to be smaller than
5551 * when there was no such activity.
5552 * non-sync IO will cause disk_stat to increase without
5553 * increasing sync_io so curr_events will (eventually)
5554 * be larger than it was before. Once it becomes
5555 * substantially larger, the test below will cause
5556 * the array to appear non-idle, and resync will slow
5558 * If there is a lot of outstanding resync activity when
5559 * we set last_event to curr_events, then all that activity
5560 * completing might cause the array to appear non-idle
5561 * and resync will be slowed down even though there might
5562 * not have been non-resync activity. This will only
5563 * happen once though. 'last_events' will soon reflect
5564 * the state where there is little or no outstanding
5565 * resync requests, and further resync activity will
5566 * always make curr_events less than last_events.
5569 if (curr_events - rdev->last_events > 4096) {
5570 rdev->last_events = curr_events;
5577 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5579 /* another "blocks" (512byte) blocks have been synced */
5580 atomic_sub(blocks, &mddev->recovery_active);
5581 wake_up(&mddev->recovery_wait);
5583 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5584 md_wakeup_thread(mddev->thread);
5585 // stop recovery, signal do_sync ....
5590 /* md_write_start(mddev, bi)
5591 * If we need to update some array metadata (e.g. 'active' flag
5592 * in superblock) before writing, schedule a superblock update
5593 * and wait for it to complete.
5595 void md_write_start(mddev_t *mddev, struct bio *bi)
5598 if (bio_data_dir(bi) != WRITE)
5601 BUG_ON(mddev->ro == 1);
5602 if (mddev->ro == 2) {
5603 /* need to switch to read/write */
5605 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5606 md_wakeup_thread(mddev->thread);
5607 md_wakeup_thread(mddev->sync_thread);
5610 atomic_inc(&mddev->writes_pending);
5611 if (mddev->safemode == 1)
5612 mddev->safemode = 0;
5613 if (mddev->in_sync) {
5614 spin_lock_irq(&mddev->write_lock);
5615 if (mddev->in_sync) {
5617 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5618 md_wakeup_thread(mddev->thread);
5621 spin_unlock_irq(&mddev->write_lock);
5624 sysfs_notify(&mddev->kobj, NULL, "array_state");
5625 wait_event(mddev->sb_wait,
5626 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5627 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5630 void md_write_end(mddev_t *mddev)
5632 if (atomic_dec_and_test(&mddev->writes_pending)) {
5633 if (mddev->safemode == 2)
5634 md_wakeup_thread(mddev->thread);
5635 else if (mddev->safemode_delay)
5636 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5640 /* md_allow_write(mddev)
5641 * Calling this ensures that the array is marked 'active' so that writes
5642 * may proceed without blocking. It is important to call this before
5643 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5644 * Must be called with mddev_lock held.
5646 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5647 * is dropped, so return -EAGAIN after notifying userspace.
5649 int md_allow_write(mddev_t *mddev)
5655 if (!mddev->pers->sync_request)
5658 spin_lock_irq(&mddev->write_lock);
5659 if (mddev->in_sync) {
5661 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5662 if (mddev->safemode_delay &&
5663 mddev->safemode == 0)
5664 mddev->safemode = 1;
5665 spin_unlock_irq(&mddev->write_lock);
5666 md_update_sb(mddev, 0);
5667 sysfs_notify(&mddev->kobj, NULL, "array_state");
5669 spin_unlock_irq(&mddev->write_lock);
5671 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5676 EXPORT_SYMBOL_GPL(md_allow_write);
5678 #define SYNC_MARKS 10
5679 #define SYNC_MARK_STEP (3*HZ)
5680 void md_do_sync(mddev_t *mddev)
5683 unsigned int currspeed = 0,
5685 sector_t max_sectors,j, io_sectors;
5686 unsigned long mark[SYNC_MARKS];
5687 sector_t mark_cnt[SYNC_MARKS];
5689 struct list_head *tmp;
5690 sector_t last_check;
5692 struct list_head *rtmp;
5696 /* just incase thread restarts... */
5697 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5699 if (mddev->ro) /* never try to sync a read-only array */
5702 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5703 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5704 desc = "data-check";
5705 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5706 desc = "requested-resync";
5709 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5714 /* we overload curr_resync somewhat here.
5715 * 0 == not engaged in resync at all
5716 * 2 == checking that there is no conflict with another sync
5717 * 1 == like 2, but have yielded to allow conflicting resync to
5719 * other == active in resync - this many blocks
5721 * Before starting a resync we must have set curr_resync to
5722 * 2, and then checked that every "conflicting" array has curr_resync
5723 * less than ours. When we find one that is the same or higher
5724 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5725 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5726 * This will mean we have to start checking from the beginning again.
5731 mddev->curr_resync = 2;
5734 if (kthread_should_stop()) {
5735 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5738 for_each_mddev(mddev2, tmp) {
5739 if (mddev2 == mddev)
5741 if (!mddev->parallel_resync
5742 && mddev2->curr_resync
5743 && match_mddev_units(mddev, mddev2)) {
5745 if (mddev < mddev2 && mddev->curr_resync == 2) {
5746 /* arbitrarily yield */
5747 mddev->curr_resync = 1;
5748 wake_up(&resync_wait);
5750 if (mddev > mddev2 && mddev->curr_resync == 1)
5751 /* no need to wait here, we can wait the next
5752 * time 'round when curr_resync == 2
5755 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5756 if (!kthread_should_stop() &&
5757 mddev2->curr_resync >= mddev->curr_resync) {
5758 printk(KERN_INFO "md: delaying %s of %s"
5759 " until %s has finished (they"
5760 " share one or more physical units)\n",
5761 desc, mdname(mddev), mdname(mddev2));
5764 finish_wait(&resync_wait, &wq);
5767 finish_wait(&resync_wait, &wq);
5770 } while (mddev->curr_resync < 2);
5773 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5774 /* resync follows the size requested by the personality,
5775 * which defaults to physical size, but can be virtual size
5777 max_sectors = mddev->resync_max_sectors;
5778 mddev->resync_mismatches = 0;
5779 /* we don't use the checkpoint if there's a bitmap */
5780 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5781 j = mddev->resync_min;
5782 else if (!mddev->bitmap)
5783 j = mddev->recovery_cp;
5785 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5786 max_sectors = mddev->size << 1;
5788 /* recovery follows the physical size of devices */
5789 max_sectors = mddev->size << 1;
5791 rdev_for_each(rdev, rtmp, mddev)
5792 if (rdev->raid_disk >= 0 &&
5793 !test_bit(Faulty, &rdev->flags) &&
5794 !test_bit(In_sync, &rdev->flags) &&
5795 rdev->recovery_offset < j)
5796 j = rdev->recovery_offset;
5799 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5800 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5801 " %d KB/sec/disk.\n", speed_min(mddev));
5802 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5803 "(but not more than %d KB/sec) for %s.\n",
5804 speed_max(mddev), desc);
5806 is_mddev_idle(mddev); /* this also initializes IO event counters */
5809 for (m = 0; m < SYNC_MARKS; m++) {
5811 mark_cnt[m] = io_sectors;
5814 mddev->resync_mark = mark[last_mark];
5815 mddev->resync_mark_cnt = mark_cnt[last_mark];
5818 * Tune reconstruction:
5820 window = 32*(PAGE_SIZE/512);
5821 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5822 window/2,(unsigned long long) max_sectors/2);
5824 atomic_set(&mddev->recovery_active, 0);
5829 "md: resuming %s of %s from checkpoint.\n",
5830 desc, mdname(mddev));
5831 mddev->curr_resync = j;
5834 while (j < max_sectors) {
5838 if (j >= mddev->resync_max) {
5839 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5840 wait_event(mddev->recovery_wait,
5841 mddev->resync_max > j
5842 || kthread_should_stop());
5844 if (kthread_should_stop())
5846 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5847 currspeed < speed_min(mddev));
5849 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5853 if (!skipped) { /* actual IO requested */
5854 io_sectors += sectors;
5855 atomic_add(sectors, &mddev->recovery_active);
5859 if (j>1) mddev->curr_resync = j;
5860 mddev->curr_mark_cnt = io_sectors;
5861 if (last_check == 0)
5862 /* this is the earliers that rebuilt will be
5863 * visible in /proc/mdstat
5865 md_new_event(mddev);
5867 if (last_check + window > io_sectors || j == max_sectors)
5870 last_check = io_sectors;
5872 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5876 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5878 int next = (last_mark+1) % SYNC_MARKS;
5880 mddev->resync_mark = mark[next];
5881 mddev->resync_mark_cnt = mark_cnt[next];
5882 mark[next] = jiffies;
5883 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5888 if (kthread_should_stop())
5893 * this loop exits only if either when we are slower than
5894 * the 'hard' speed limit, or the system was IO-idle for
5896 * the system might be non-idle CPU-wise, but we only care
5897 * about not overloading the IO subsystem. (things like an
5898 * e2fsck being done on the RAID array should execute fast)
5900 blk_unplug(mddev->queue);
5903 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5904 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5906 if (currspeed > speed_min(mddev)) {
5907 if ((currspeed > speed_max(mddev)) ||
5908 !is_mddev_idle(mddev)) {
5914 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5916 * this also signals 'finished resyncing' to md_stop
5919 blk_unplug(mddev->queue);
5921 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5923 /* tell personality that we are finished */
5924 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5926 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5927 mddev->curr_resync > 2) {
5928 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5929 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5930 if (mddev->curr_resync >= mddev->recovery_cp) {
5932 "md: checkpointing %s of %s.\n",
5933 desc, mdname(mddev));
5934 mddev->recovery_cp = mddev->curr_resync;
5937 mddev->recovery_cp = MaxSector;
5939 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5940 mddev->curr_resync = MaxSector;
5941 rdev_for_each(rdev, rtmp, mddev)
5942 if (rdev->raid_disk >= 0 &&
5943 !test_bit(Faulty, &rdev->flags) &&
5944 !test_bit(In_sync, &rdev->flags) &&
5945 rdev->recovery_offset < mddev->curr_resync)
5946 rdev->recovery_offset = mddev->curr_resync;
5949 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5952 mddev->curr_resync = 0;
5953 mddev->resync_min = 0;
5954 mddev->resync_max = MaxSector;
5955 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5956 wake_up(&resync_wait);
5957 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5958 md_wakeup_thread(mddev->thread);
5963 * got a signal, exit.
5966 "md: md_do_sync() got signal ... exiting\n");
5967 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5971 EXPORT_SYMBOL_GPL(md_do_sync);
5974 static int remove_and_add_spares(mddev_t *mddev)
5977 struct list_head *rtmp;
5980 rdev_for_each(rdev, rtmp, mddev)
5981 if (rdev->raid_disk >= 0 &&
5982 !test_bit(Blocked, &rdev->flags) &&
5983 (test_bit(Faulty, &rdev->flags) ||
5984 ! test_bit(In_sync, &rdev->flags)) &&
5985 atomic_read(&rdev->nr_pending)==0) {
5986 if (mddev->pers->hot_remove_disk(
5987 mddev, rdev->raid_disk)==0) {
5989 sprintf(nm,"rd%d", rdev->raid_disk);
5990 sysfs_remove_link(&mddev->kobj, nm);
5991 rdev->raid_disk = -1;
5995 if (mddev->degraded) {
5996 rdev_for_each(rdev, rtmp, mddev) {
5997 if (rdev->raid_disk >= 0 &&
5998 !test_bit(In_sync, &rdev->flags))
6000 if (rdev->raid_disk < 0
6001 && !test_bit(Faulty, &rdev->flags)) {
6002 rdev->recovery_offset = 0;
6004 hot_add_disk(mddev, rdev) == 0) {
6006 sprintf(nm, "rd%d", rdev->raid_disk);
6007 if (sysfs_create_link(&mddev->kobj,
6010 "md: cannot register "
6014 md_new_event(mddev);
6023 * This routine is regularly called by all per-raid-array threads to
6024 * deal with generic issues like resync and super-block update.
6025 * Raid personalities that don't have a thread (linear/raid0) do not
6026 * need this as they never do any recovery or update the superblock.
6028 * It does not do any resync itself, but rather "forks" off other threads
6029 * to do that as needed.
6030 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6031 * "->recovery" and create a thread at ->sync_thread.
6032 * When the thread finishes it sets MD_RECOVERY_DONE
6033 * and wakeups up this thread which will reap the thread and finish up.
6034 * This thread also removes any faulty devices (with nr_pending == 0).
6036 * The overall approach is:
6037 * 1/ if the superblock needs updating, update it.
6038 * 2/ If a recovery thread is running, don't do anything else.
6039 * 3/ If recovery has finished, clean up, possibly marking spares active.
6040 * 4/ If there are any faulty devices, remove them.
6041 * 5/ If array is degraded, try to add spares devices
6042 * 6/ If array has spares or is not in-sync, start a resync thread.
6044 void md_check_recovery(mddev_t *mddev)
6047 struct list_head *rtmp;
6051 bitmap_daemon_work(mddev->bitmap);
6056 if (signal_pending(current)) {
6057 if (mddev->pers->sync_request && !mddev->external) {
6058 printk(KERN_INFO "md: %s in immediate safe mode\n",
6060 mddev->safemode = 2;
6062 flush_signals(current);
6066 (mddev->flags && !mddev->external) ||
6067 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6068 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6069 (mddev->external == 0 && mddev->safemode == 1) ||
6070 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6071 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6075 if (mddev_trylock(mddev)) {
6078 if (!mddev->external) {
6080 spin_lock_irq(&mddev->write_lock);
6081 if (mddev->safemode &&
6082 !atomic_read(&mddev->writes_pending) &&
6084 mddev->recovery_cp == MaxSector) {
6087 if (mddev->persistent)
6088 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6090 if (mddev->safemode == 1)
6091 mddev->safemode = 0;
6092 spin_unlock_irq(&mddev->write_lock);
6094 sysfs_notify(&mddev->kobj, NULL, "array_state");
6098 md_update_sb(mddev, 0);
6100 rdev_for_each(rdev, rtmp, mddev)
6101 if (test_and_clear_bit(StateChanged, &rdev->flags))
6102 sysfs_notify(&rdev->kobj, NULL, "state");
6105 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6106 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6107 /* resync/recovery still happening */
6108 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6111 if (mddev->sync_thread) {
6112 /* resync has finished, collect result */
6113 md_unregister_thread(mddev->sync_thread);
6114 mddev->sync_thread = NULL;
6115 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6117 /* activate any spares */
6118 if (mddev->pers->spare_active(mddev))
6119 sysfs_notify(&mddev->kobj, NULL,
6122 md_update_sb(mddev, 1);
6124 /* if array is no-longer degraded, then any saved_raid_disk
6125 * information must be scrapped
6127 if (!mddev->degraded)
6128 rdev_for_each(rdev, rtmp, mddev)
6129 rdev->saved_raid_disk = -1;
6131 mddev->recovery = 0;
6132 /* flag recovery needed just to double check */
6133 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6134 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6135 md_new_event(mddev);
6138 /* Set RUNNING before clearing NEEDED to avoid
6139 * any transients in the value of "sync_action".
6141 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6142 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6143 /* Clear some bits that don't mean anything, but
6146 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6147 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6149 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6151 /* no recovery is running.
6152 * remove any failed drives, then
6153 * add spares if possible.
6154 * Spare are also removed and re-added, to allow
6155 * the personality to fail the re-add.
6158 if (mddev->reshape_position != MaxSector) {
6159 if (mddev->pers->check_reshape(mddev) != 0)
6160 /* Cannot proceed */
6162 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6163 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6164 } else if ((spares = remove_and_add_spares(mddev))) {
6165 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6166 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6167 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6168 } else if (mddev->recovery_cp < MaxSector) {
6169 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6170 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6171 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6172 /* nothing to be done ... */
6175 if (mddev->pers->sync_request) {
6176 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6177 /* We are adding a device or devices to an array
6178 * which has the bitmap stored on all devices.
6179 * So make sure all bitmap pages get written
6181 bitmap_write_all(mddev->bitmap);
6183 mddev->sync_thread = md_register_thread(md_do_sync,
6186 if (!mddev->sync_thread) {
6187 printk(KERN_ERR "%s: could not start resync"
6190 /* leave the spares where they are, it shouldn't hurt */
6191 mddev->recovery = 0;
6193 md_wakeup_thread(mddev->sync_thread);
6194 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6195 md_new_event(mddev);
6198 if (!mddev->sync_thread) {
6199 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6200 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6202 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6204 mddev_unlock(mddev);
6208 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6210 sysfs_notify(&rdev->kobj, NULL, "state");
6211 wait_event_timeout(rdev->blocked_wait,
6212 !test_bit(Blocked, &rdev->flags),
6213 msecs_to_jiffies(5000));
6214 rdev_dec_pending(rdev, mddev);
6216 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6218 static int md_notify_reboot(struct notifier_block *this,
6219 unsigned long code, void *x)
6221 struct list_head *tmp;
6224 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6226 printk(KERN_INFO "md: stopping all md devices.\n");
6228 for_each_mddev(mddev, tmp)
6229 if (mddev_trylock(mddev)) {
6230 do_md_stop (mddev, 1, 0);
6231 mddev_unlock(mddev);
6234 * certain more exotic SCSI devices are known to be
6235 * volatile wrt too early system reboots. While the
6236 * right place to handle this issue is the given
6237 * driver, we do want to have a safe RAID driver ...
6244 static struct notifier_block md_notifier = {
6245 .notifier_call = md_notify_reboot,
6247 .priority = INT_MAX, /* before any real devices */
6250 static void md_geninit(void)
6252 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6254 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6257 static int __init md_init(void)
6259 if (register_blkdev(MAJOR_NR, "md"))
6261 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6262 unregister_blkdev(MAJOR_NR, "md");
6265 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6266 md_probe, NULL, NULL);
6267 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6268 md_probe, NULL, NULL);
6270 register_reboot_notifier(&md_notifier);
6271 raid_table_header = register_sysctl_table(raid_root_table);
6281 * Searches all registered partitions for autorun RAID arrays
6285 static LIST_HEAD(all_detected_devices);
6286 struct detected_devices_node {
6287 struct list_head list;
6291 void md_autodetect_dev(dev_t dev)
6293 struct detected_devices_node *node_detected_dev;
6295 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6296 if (node_detected_dev) {
6297 node_detected_dev->dev = dev;
6298 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6300 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6301 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6306 static void autostart_arrays(int part)
6309 struct detected_devices_node *node_detected_dev;
6311 int i_scanned, i_passed;
6316 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6318 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6320 node_detected_dev = list_entry(all_detected_devices.next,
6321 struct detected_devices_node, list);
6322 list_del(&node_detected_dev->list);
6323 dev = node_detected_dev->dev;
6324 kfree(node_detected_dev);
6325 rdev = md_import_device(dev,0, 90);
6329 if (test_bit(Faulty, &rdev->flags)) {
6333 set_bit(AutoDetected, &rdev->flags);
6334 list_add(&rdev->same_set, &pending_raid_disks);
6338 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6339 i_scanned, i_passed);
6341 autorun_devices(part);
6344 #endif /* !MODULE */
6346 static __exit void md_exit(void)
6349 struct list_head *tmp;
6351 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6352 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6354 unregister_blkdev(MAJOR_NR,"md");
6355 unregister_blkdev(mdp_major, "mdp");
6356 unregister_reboot_notifier(&md_notifier);
6357 unregister_sysctl_table(raid_table_header);
6358 remove_proc_entry("mdstat", NULL);
6359 for_each_mddev(mddev, tmp) {
6360 struct gendisk *disk = mddev->gendisk;
6363 export_array(mddev);
6366 mddev->gendisk = NULL;
6371 subsys_initcall(md_init);
6372 module_exit(md_exit)
6374 static int get_ro(char *buffer, struct kernel_param *kp)
6376 return sprintf(buffer, "%d", start_readonly);
6378 static int set_ro(const char *val, struct kernel_param *kp)
6381 int num = simple_strtoul(val, &e, 10);
6382 if (*val && (*e == '\0' || *e == '\n')) {
6383 start_readonly = num;
6389 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6390 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6393 EXPORT_SYMBOL(register_md_personality);
6394 EXPORT_SYMBOL(unregister_md_personality);
6395 EXPORT_SYMBOL(md_error);
6396 EXPORT_SYMBOL(md_done_sync);
6397 EXPORT_SYMBOL(md_write_start);
6398 EXPORT_SYMBOL(md_write_end);
6399 EXPORT_SYMBOL(md_register_thread);
6400 EXPORT_SYMBOL(md_unregister_thread);
6401 EXPORT_SYMBOL(md_wakeup_thread);
6402 EXPORT_SYMBOL(md_check_recovery);
6403 MODULE_LICENSE("GPL");
6405 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);