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_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
358 sector_t num_sectors = rdev->sb_offset * 2;
361 num_sectors &= ~((sector_t)chunk_size/512 - 1);
365 static int alloc_disk_sb(mdk_rdev_t * rdev)
370 rdev->sb_page = alloc_page(GFP_KERNEL);
371 if (!rdev->sb_page) {
372 printk(KERN_ALERT "md: out of memory.\n");
379 static void free_disk_sb(mdk_rdev_t * rdev)
382 put_page(rdev->sb_page);
384 rdev->sb_page = NULL;
391 static void super_written(struct bio *bio, int error)
393 mdk_rdev_t *rdev = bio->bi_private;
394 mddev_t *mddev = rdev->mddev;
396 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
397 printk("md: super_written gets error=%d, uptodate=%d\n",
398 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
399 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
400 md_error(mddev, rdev);
403 if (atomic_dec_and_test(&mddev->pending_writes))
404 wake_up(&mddev->sb_wait);
408 static void super_written_barrier(struct bio *bio, int error)
410 struct bio *bio2 = bio->bi_private;
411 mdk_rdev_t *rdev = bio2->bi_private;
412 mddev_t *mddev = rdev->mddev;
414 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
415 error == -EOPNOTSUPP) {
417 /* barriers don't appear to be supported :-( */
418 set_bit(BarriersNotsupp, &rdev->flags);
419 mddev->barriers_work = 0;
420 spin_lock_irqsave(&mddev->write_lock, flags);
421 bio2->bi_next = mddev->biolist;
422 mddev->biolist = bio2;
423 spin_unlock_irqrestore(&mddev->write_lock, flags);
424 wake_up(&mddev->sb_wait);
428 bio->bi_private = rdev;
429 super_written(bio, error);
433 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
434 sector_t sector, int size, struct page *page)
436 /* write first size bytes of page to sector of rdev
437 * Increment mddev->pending_writes before returning
438 * and decrement it on completion, waking up sb_wait
439 * if zero is reached.
440 * If an error occurred, call md_error
442 * As we might need to resubmit the request if BIO_RW_BARRIER
443 * causes ENOTSUPP, we allocate a spare bio...
445 struct bio *bio = bio_alloc(GFP_NOIO, 1);
446 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
448 bio->bi_bdev = rdev->bdev;
449 bio->bi_sector = sector;
450 bio_add_page(bio, page, size, 0);
451 bio->bi_private = rdev;
452 bio->bi_end_io = super_written;
455 atomic_inc(&mddev->pending_writes);
456 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
458 rw |= (1<<BIO_RW_BARRIER);
459 rbio = bio_clone(bio, GFP_NOIO);
460 rbio->bi_private = bio;
461 rbio->bi_end_io = super_written_barrier;
462 submit_bio(rw, rbio);
467 void md_super_wait(mddev_t *mddev)
469 /* wait for all superblock writes that were scheduled to complete.
470 * if any had to be retried (due to BARRIER problems), retry them
474 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
475 if (atomic_read(&mddev->pending_writes)==0)
477 while (mddev->biolist) {
479 spin_lock_irq(&mddev->write_lock);
480 bio = mddev->biolist;
481 mddev->biolist = bio->bi_next ;
483 spin_unlock_irq(&mddev->write_lock);
484 submit_bio(bio->bi_rw, bio);
488 finish_wait(&mddev->sb_wait, &wq);
491 static void bi_complete(struct bio *bio, int error)
493 complete((struct completion*)bio->bi_private);
496 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
497 struct page *page, int rw)
499 struct bio *bio = bio_alloc(GFP_NOIO, 1);
500 struct completion event;
503 rw |= (1 << BIO_RW_SYNC);
506 bio->bi_sector = sector;
507 bio_add_page(bio, page, size, 0);
508 init_completion(&event);
509 bio->bi_private = &event;
510 bio->bi_end_io = bi_complete;
512 wait_for_completion(&event);
514 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
518 EXPORT_SYMBOL_GPL(sync_page_io);
520 static int read_disk_sb(mdk_rdev_t * rdev, int size)
522 char b[BDEVNAME_SIZE];
523 if (!rdev->sb_page) {
531 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
537 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
538 bdevname(rdev->bdev,b));
542 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
544 return sb1->set_uuid0 == sb2->set_uuid0 &&
545 sb1->set_uuid1 == sb2->set_uuid1 &&
546 sb1->set_uuid2 == sb2->set_uuid2 &&
547 sb1->set_uuid3 == sb2->set_uuid3;
550 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
553 mdp_super_t *tmp1, *tmp2;
555 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
556 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
558 if (!tmp1 || !tmp2) {
560 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
568 * nr_disks is not constant
573 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
581 static u32 md_csum_fold(u32 csum)
583 csum = (csum & 0xffff) + (csum >> 16);
584 return (csum & 0xffff) + (csum >> 16);
587 static unsigned int calc_sb_csum(mdp_super_t * sb)
590 u32 *sb32 = (u32*)sb;
592 unsigned int disk_csum, csum;
594 disk_csum = sb->sb_csum;
597 for (i = 0; i < MD_SB_BYTES/4 ; i++)
599 csum = (newcsum & 0xffffffff) + (newcsum>>32);
603 /* This used to use csum_partial, which was wrong for several
604 * reasons including that different results are returned on
605 * different architectures. It isn't critical that we get exactly
606 * the same return value as before (we always csum_fold before
607 * testing, and that removes any differences). However as we
608 * know that csum_partial always returned a 16bit value on
609 * alphas, do a fold to maximise conformity to previous behaviour.
611 sb->sb_csum = md_csum_fold(disk_csum);
613 sb->sb_csum = disk_csum;
620 * Handle superblock details.
621 * We want to be able to handle multiple superblock formats
622 * so we have a common interface to them all, and an array of
623 * different handlers.
624 * We rely on user-space to write the initial superblock, and support
625 * reading and updating of superblocks.
626 * Interface methods are:
627 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
628 * loads and validates a superblock on dev.
629 * if refdev != NULL, compare superblocks on both devices
631 * 0 - dev has a superblock that is compatible with refdev
632 * 1 - dev has a superblock that is compatible and newer than refdev
633 * so dev should be used as the refdev in future
634 * -EINVAL superblock incompatible or invalid
635 * -othererror e.g. -EIO
637 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
638 * Verify that dev is acceptable into mddev.
639 * The first time, mddev->raid_disks will be 0, and data from
640 * dev should be merged in. Subsequent calls check that dev
641 * is new enough. Return 0 or -EINVAL
643 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
644 * Update the superblock for rdev with data in mddev
645 * This does not write to disc.
651 struct module *owner;
652 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
654 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
655 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
656 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
657 unsigned long long size);
661 * load_super for 0.90.0
663 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
665 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
671 * Calculate the position of the superblock,
672 * it's at the end of the disk.
674 * It also happens to be a multiple of 4Kb.
676 sb_offset = calc_dev_sboffset(rdev->bdev);
677 rdev->sb_offset = sb_offset;
679 ret = read_disk_sb(rdev, MD_SB_BYTES);
684 bdevname(rdev->bdev, b);
685 sb = (mdp_super_t*)page_address(rdev->sb_page);
687 if (sb->md_magic != MD_SB_MAGIC) {
688 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
693 if (sb->major_version != 0 ||
694 sb->minor_version < 90 ||
695 sb->minor_version > 91) {
696 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
697 sb->major_version, sb->minor_version,
702 if (sb->raid_disks <= 0)
705 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
706 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
711 rdev->preferred_minor = sb->md_minor;
712 rdev->data_offset = 0;
713 rdev->sb_size = MD_SB_BYTES;
715 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
716 if (sb->level != 1 && sb->level != 4
717 && sb->level != 5 && sb->level != 6
718 && sb->level != 10) {
719 /* FIXME use a better test */
721 "md: bitmaps not supported for this level.\n");
726 if (sb->level == LEVEL_MULTIPATH)
729 rdev->desc_nr = sb->this_disk.number;
735 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
736 if (!uuid_equal(refsb, sb)) {
737 printk(KERN_WARNING "md: %s has different UUID to %s\n",
738 b, bdevname(refdev->bdev,b2));
741 if (!sb_equal(refsb, sb)) {
742 printk(KERN_WARNING "md: %s has same UUID"
743 " but different superblock to %s\n",
744 b, bdevname(refdev->bdev, b2));
748 ev2 = md_event(refsb);
754 rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
756 if (rdev->size < sb->size && sb->level > 1)
757 /* "this cannot possibly happen" ... */
765 * validate_super for 0.90.0
767 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
770 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
771 __u64 ev1 = md_event(sb);
773 rdev->raid_disk = -1;
774 clear_bit(Faulty, &rdev->flags);
775 clear_bit(In_sync, &rdev->flags);
776 clear_bit(WriteMostly, &rdev->flags);
777 clear_bit(BarriersNotsupp, &rdev->flags);
779 if (mddev->raid_disks == 0) {
780 mddev->major_version = 0;
781 mddev->minor_version = sb->minor_version;
782 mddev->patch_version = sb->patch_version;
784 mddev->chunk_size = sb->chunk_size;
785 mddev->ctime = sb->ctime;
786 mddev->utime = sb->utime;
787 mddev->level = sb->level;
788 mddev->clevel[0] = 0;
789 mddev->layout = sb->layout;
790 mddev->raid_disks = sb->raid_disks;
791 mddev->size = sb->size;
793 mddev->bitmap_offset = 0;
794 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
796 if (mddev->minor_version >= 91) {
797 mddev->reshape_position = sb->reshape_position;
798 mddev->delta_disks = sb->delta_disks;
799 mddev->new_level = sb->new_level;
800 mddev->new_layout = sb->new_layout;
801 mddev->new_chunk = sb->new_chunk;
803 mddev->reshape_position = MaxSector;
804 mddev->delta_disks = 0;
805 mddev->new_level = mddev->level;
806 mddev->new_layout = mddev->layout;
807 mddev->new_chunk = mddev->chunk_size;
810 if (sb->state & (1<<MD_SB_CLEAN))
811 mddev->recovery_cp = MaxSector;
813 if (sb->events_hi == sb->cp_events_hi &&
814 sb->events_lo == sb->cp_events_lo) {
815 mddev->recovery_cp = sb->recovery_cp;
817 mddev->recovery_cp = 0;
820 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
821 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
822 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
823 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
825 mddev->max_disks = MD_SB_DISKS;
827 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
828 mddev->bitmap_file == NULL)
829 mddev->bitmap_offset = mddev->default_bitmap_offset;
831 } else if (mddev->pers == NULL) {
832 /* Insist on good event counter while assembling */
834 if (ev1 < mddev->events)
836 } else if (mddev->bitmap) {
837 /* if adding to array with a bitmap, then we can accept an
838 * older device ... but not too old.
840 if (ev1 < mddev->bitmap->events_cleared)
843 if (ev1 < mddev->events)
844 /* just a hot-add of a new device, leave raid_disk at -1 */
848 if (mddev->level != LEVEL_MULTIPATH) {
849 desc = sb->disks + rdev->desc_nr;
851 if (desc->state & (1<<MD_DISK_FAULTY))
852 set_bit(Faulty, &rdev->flags);
853 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
854 desc->raid_disk < mddev->raid_disks */) {
855 set_bit(In_sync, &rdev->flags);
856 rdev->raid_disk = desc->raid_disk;
858 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
859 set_bit(WriteMostly, &rdev->flags);
860 } else /* MULTIPATH are always insync */
861 set_bit(In_sync, &rdev->flags);
866 * sync_super for 0.90.0
868 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
871 struct list_head *tmp;
873 int next_spare = mddev->raid_disks;
876 /* make rdev->sb match mddev data..
879 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
880 * 3/ any empty disks < next_spare become removed
882 * disks[0] gets initialised to REMOVED because
883 * we cannot be sure from other fields if it has
884 * been initialised or not.
887 int active=0, working=0,failed=0,spare=0,nr_disks=0;
889 rdev->sb_size = MD_SB_BYTES;
891 sb = (mdp_super_t*)page_address(rdev->sb_page);
893 memset(sb, 0, sizeof(*sb));
895 sb->md_magic = MD_SB_MAGIC;
896 sb->major_version = mddev->major_version;
897 sb->patch_version = mddev->patch_version;
898 sb->gvalid_words = 0; /* ignored */
899 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
900 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
901 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
902 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
904 sb->ctime = mddev->ctime;
905 sb->level = mddev->level;
906 sb->size = mddev->size;
907 sb->raid_disks = mddev->raid_disks;
908 sb->md_minor = mddev->md_minor;
909 sb->not_persistent = 0;
910 sb->utime = mddev->utime;
912 sb->events_hi = (mddev->events>>32);
913 sb->events_lo = (u32)mddev->events;
915 if (mddev->reshape_position == MaxSector)
916 sb->minor_version = 90;
918 sb->minor_version = 91;
919 sb->reshape_position = mddev->reshape_position;
920 sb->new_level = mddev->new_level;
921 sb->delta_disks = mddev->delta_disks;
922 sb->new_layout = mddev->new_layout;
923 sb->new_chunk = mddev->new_chunk;
925 mddev->minor_version = sb->minor_version;
928 sb->recovery_cp = mddev->recovery_cp;
929 sb->cp_events_hi = (mddev->events>>32);
930 sb->cp_events_lo = (u32)mddev->events;
931 if (mddev->recovery_cp == MaxSector)
932 sb->state = (1<< MD_SB_CLEAN);
936 sb->layout = mddev->layout;
937 sb->chunk_size = mddev->chunk_size;
939 if (mddev->bitmap && mddev->bitmap_file == NULL)
940 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
942 sb->disks[0].state = (1<<MD_DISK_REMOVED);
943 rdev_for_each(rdev2, tmp, mddev) {
946 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
947 && !test_bit(Faulty, &rdev2->flags))
948 desc_nr = rdev2->raid_disk;
950 desc_nr = next_spare++;
951 rdev2->desc_nr = desc_nr;
952 d = &sb->disks[rdev2->desc_nr];
954 d->number = rdev2->desc_nr;
955 d->major = MAJOR(rdev2->bdev->bd_dev);
956 d->minor = MINOR(rdev2->bdev->bd_dev);
957 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
958 && !test_bit(Faulty, &rdev2->flags))
959 d->raid_disk = rdev2->raid_disk;
961 d->raid_disk = rdev2->desc_nr; /* compatibility */
962 if (test_bit(Faulty, &rdev2->flags))
963 d->state = (1<<MD_DISK_FAULTY);
964 else if (test_bit(In_sync, &rdev2->flags)) {
965 d->state = (1<<MD_DISK_ACTIVE);
966 d->state |= (1<<MD_DISK_SYNC);
974 if (test_bit(WriteMostly, &rdev2->flags))
975 d->state |= (1<<MD_DISK_WRITEMOSTLY);
977 /* now set the "removed" and "faulty" bits on any missing devices */
978 for (i=0 ; i < mddev->raid_disks ; i++) {
979 mdp_disk_t *d = &sb->disks[i];
980 if (d->state == 0 && d->number == 0) {
983 d->state = (1<<MD_DISK_REMOVED);
984 d->state |= (1<<MD_DISK_FAULTY);
988 sb->nr_disks = nr_disks;
989 sb->active_disks = active;
990 sb->working_disks = working;
991 sb->failed_disks = failed;
992 sb->spare_disks = spare;
994 sb->this_disk = sb->disks[rdev->desc_nr];
995 sb->sb_csum = calc_sb_csum(sb);
999 * rdev_size_change for 0.90.0
1001 static unsigned long long
1002 super_90_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1004 if (size && size < rdev->mddev->size)
1005 return 0; /* component must fit device */
1006 size *= 2; /* convert to sectors */
1007 if (rdev->mddev->bitmap_offset)
1008 return 0; /* can't move bitmap */
1009 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
1010 if (!size || size > rdev->sb_offset*2)
1011 size = rdev->sb_offset*2;
1012 md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1014 md_super_wait(rdev->mddev);
1015 return size/2; /* kB for sysfs */
1020 * version 1 superblock
1023 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1027 unsigned long long newcsum;
1028 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1029 __le32 *isuper = (__le32*)sb;
1032 disk_csum = sb->sb_csum;
1035 for (i=0; size>=4; size -= 4 )
1036 newcsum += le32_to_cpu(*isuper++);
1039 newcsum += le16_to_cpu(*(__le16*) isuper);
1041 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1042 sb->sb_csum = disk_csum;
1043 return cpu_to_le32(csum);
1046 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1048 struct mdp_superblock_1 *sb;
1051 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1055 * Calculate the position of the superblock.
1056 * It is always aligned to a 4K boundary and
1057 * depeding on minor_version, it can be:
1058 * 0: At least 8K, but less than 12K, from end of device
1059 * 1: At start of device
1060 * 2: 4K from start of device.
1062 switch(minor_version) {
1064 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1066 sb_offset &= ~(sector_t)(4*2-1);
1067 /* convert from sectors to K */
1079 rdev->sb_offset = sb_offset;
1081 /* superblock is rarely larger than 1K, but it can be larger,
1082 * and it is safe to read 4k, so we do that
1084 ret = read_disk_sb(rdev, 4096);
1085 if (ret) return ret;
1088 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1090 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1091 sb->major_version != cpu_to_le32(1) ||
1092 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1093 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1094 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1097 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1098 printk("md: invalid superblock checksum on %s\n",
1099 bdevname(rdev->bdev,b));
1102 if (le64_to_cpu(sb->data_size) < 10) {
1103 printk("md: data_size too small on %s\n",
1104 bdevname(rdev->bdev,b));
1107 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1108 if (sb->level != cpu_to_le32(1) &&
1109 sb->level != cpu_to_le32(4) &&
1110 sb->level != cpu_to_le32(5) &&
1111 sb->level != cpu_to_le32(6) &&
1112 sb->level != cpu_to_le32(10)) {
1114 "md: bitmaps not supported for this level.\n");
1119 rdev->preferred_minor = 0xffff;
1120 rdev->data_offset = le64_to_cpu(sb->data_offset);
1121 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1123 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1124 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1125 if (rdev->sb_size & bmask)
1126 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1129 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1132 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1135 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1141 struct mdp_superblock_1 *refsb =
1142 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1144 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1145 sb->level != refsb->level ||
1146 sb->layout != refsb->layout ||
1147 sb->chunksize != refsb->chunksize) {
1148 printk(KERN_WARNING "md: %s has strangely different"
1149 " superblock to %s\n",
1150 bdevname(rdev->bdev,b),
1151 bdevname(refdev->bdev,b2));
1154 ev1 = le64_to_cpu(sb->events);
1155 ev2 = le64_to_cpu(refsb->events);
1163 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1165 rdev->size = rdev->sb_offset;
1166 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1168 rdev->size = le64_to_cpu(sb->data_size)/2;
1169 if (le32_to_cpu(sb->chunksize))
1170 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1172 if (le64_to_cpu(sb->size) > rdev->size*2)
1177 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1179 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1180 __u64 ev1 = le64_to_cpu(sb->events);
1182 rdev->raid_disk = -1;
1183 clear_bit(Faulty, &rdev->flags);
1184 clear_bit(In_sync, &rdev->flags);
1185 clear_bit(WriteMostly, &rdev->flags);
1186 clear_bit(BarriersNotsupp, &rdev->flags);
1188 if (mddev->raid_disks == 0) {
1189 mddev->major_version = 1;
1190 mddev->patch_version = 0;
1191 mddev->external = 0;
1192 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1193 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1194 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1195 mddev->level = le32_to_cpu(sb->level);
1196 mddev->clevel[0] = 0;
1197 mddev->layout = le32_to_cpu(sb->layout);
1198 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1199 mddev->size = le64_to_cpu(sb->size)/2;
1200 mddev->events = ev1;
1201 mddev->bitmap_offset = 0;
1202 mddev->default_bitmap_offset = 1024 >> 9;
1204 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1205 memcpy(mddev->uuid, sb->set_uuid, 16);
1207 mddev->max_disks = (4096-256)/2;
1209 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1210 mddev->bitmap_file == NULL )
1211 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1213 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1214 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1215 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1216 mddev->new_level = le32_to_cpu(sb->new_level);
1217 mddev->new_layout = le32_to_cpu(sb->new_layout);
1218 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1220 mddev->reshape_position = MaxSector;
1221 mddev->delta_disks = 0;
1222 mddev->new_level = mddev->level;
1223 mddev->new_layout = mddev->layout;
1224 mddev->new_chunk = mddev->chunk_size;
1227 } else if (mddev->pers == NULL) {
1228 /* Insist of good event counter while assembling */
1230 if (ev1 < mddev->events)
1232 } else if (mddev->bitmap) {
1233 /* If adding to array with a bitmap, then we can accept an
1234 * older device, but not too old.
1236 if (ev1 < mddev->bitmap->events_cleared)
1239 if (ev1 < mddev->events)
1240 /* just a hot-add of a new device, leave raid_disk at -1 */
1243 if (mddev->level != LEVEL_MULTIPATH) {
1245 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1247 case 0xffff: /* spare */
1249 case 0xfffe: /* faulty */
1250 set_bit(Faulty, &rdev->flags);
1253 if ((le32_to_cpu(sb->feature_map) &
1254 MD_FEATURE_RECOVERY_OFFSET))
1255 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1257 set_bit(In_sync, &rdev->flags);
1258 rdev->raid_disk = role;
1261 if (sb->devflags & WriteMostly1)
1262 set_bit(WriteMostly, &rdev->flags);
1263 } else /* MULTIPATH are always insync */
1264 set_bit(In_sync, &rdev->flags);
1269 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1271 struct mdp_superblock_1 *sb;
1272 struct list_head *tmp;
1275 /* make rdev->sb match mddev and rdev data. */
1277 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1279 sb->feature_map = 0;
1281 sb->recovery_offset = cpu_to_le64(0);
1282 memset(sb->pad1, 0, sizeof(sb->pad1));
1283 memset(sb->pad2, 0, sizeof(sb->pad2));
1284 memset(sb->pad3, 0, sizeof(sb->pad3));
1286 sb->utime = cpu_to_le64((__u64)mddev->utime);
1287 sb->events = cpu_to_le64(mddev->events);
1289 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1291 sb->resync_offset = cpu_to_le64(0);
1293 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1295 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1296 sb->size = cpu_to_le64(mddev->size<<1);
1298 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1299 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1300 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1303 if (rdev->raid_disk >= 0 &&
1304 !test_bit(In_sync, &rdev->flags) &&
1305 rdev->recovery_offset > 0) {
1306 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1307 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1310 if (mddev->reshape_position != MaxSector) {
1311 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1312 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1313 sb->new_layout = cpu_to_le32(mddev->new_layout);
1314 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1315 sb->new_level = cpu_to_le32(mddev->new_level);
1316 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1320 rdev_for_each(rdev2, tmp, mddev)
1321 if (rdev2->desc_nr+1 > max_dev)
1322 max_dev = rdev2->desc_nr+1;
1324 if (max_dev > le32_to_cpu(sb->max_dev))
1325 sb->max_dev = cpu_to_le32(max_dev);
1326 for (i=0; i<max_dev;i++)
1327 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1329 rdev_for_each(rdev2, tmp, mddev) {
1331 if (test_bit(Faulty, &rdev2->flags))
1332 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1333 else if (test_bit(In_sync, &rdev2->flags))
1334 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1335 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1336 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1338 sb->dev_roles[i] = cpu_to_le16(0xffff);
1341 sb->sb_csum = calc_sb_1_csum(sb);
1344 static unsigned long long
1345 super_1_rdev_size_change(mdk_rdev_t *rdev, unsigned long long size)
1347 struct mdp_superblock_1 *sb;
1348 unsigned long long max_size;
1349 if (size && size < rdev->mddev->size)
1350 return 0; /* component must fit device */
1351 size *= 2; /* convert to sectors */
1352 if (rdev->sb_offset < rdev->data_offset/2) {
1353 /* minor versions 1 and 2; superblock before data */
1354 max_size = (rdev->bdev->bd_inode->i_size >> 9);
1355 max_size -= rdev->data_offset;
1356 if (!size || size > max_size)
1358 } else if (rdev->mddev->bitmap_offset) {
1359 /* minor version 0 with bitmap we can't move */
1362 /* minor version 0; superblock after data */
1364 sb_offset = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1365 sb_offset &= ~(sector_t)(4*2 - 1);
1366 max_size = rdev->size*2 + sb_offset - rdev->sb_offset*2;
1367 if (!size || size > max_size)
1369 rdev->sb_offset = sb_offset/2;
1371 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1372 sb->data_size = cpu_to_le64(size);
1373 sb->super_offset = rdev->sb_offset*2;
1374 sb->sb_csum = calc_sb_1_csum(sb);
1375 md_super_write(rdev->mddev, rdev, rdev->sb_offset << 1, rdev->sb_size,
1377 md_super_wait(rdev->mddev);
1378 return size/2; /* kB for sysfs */
1381 static struct super_type super_types[] = {
1384 .owner = THIS_MODULE,
1385 .load_super = super_90_load,
1386 .validate_super = super_90_validate,
1387 .sync_super = super_90_sync,
1388 .rdev_size_change = super_90_rdev_size_change,
1392 .owner = THIS_MODULE,
1393 .load_super = super_1_load,
1394 .validate_super = super_1_validate,
1395 .sync_super = super_1_sync,
1396 .rdev_size_change = super_1_rdev_size_change,
1400 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1402 struct list_head *tmp, *tmp2;
1403 mdk_rdev_t *rdev, *rdev2;
1405 rdev_for_each(rdev, tmp, mddev1)
1406 rdev_for_each(rdev2, tmp2, mddev2)
1407 if (rdev->bdev->bd_contains ==
1408 rdev2->bdev->bd_contains)
1414 static LIST_HEAD(pending_raid_disks);
1416 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1418 char b[BDEVNAME_SIZE];
1428 /* prevent duplicates */
1429 if (find_rdev(mddev, rdev->bdev->bd_dev))
1432 /* make sure rdev->size exceeds mddev->size */
1433 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1435 /* Cannot change size, so fail
1436 * If mddev->level <= 0, then we don't care
1437 * about aligning sizes (e.g. linear)
1439 if (mddev->level > 0)
1442 mddev->size = rdev->size;
1445 /* Verify rdev->desc_nr is unique.
1446 * If it is -1, assign a free number, else
1447 * check number is not in use
1449 if (rdev->desc_nr < 0) {
1451 if (mddev->pers) choice = mddev->raid_disks;
1452 while (find_rdev_nr(mddev, choice))
1454 rdev->desc_nr = choice;
1456 if (find_rdev_nr(mddev, rdev->desc_nr))
1459 bdevname(rdev->bdev,b);
1460 while ( (s=strchr(b, '/')) != NULL)
1463 rdev->mddev = mddev;
1464 printk(KERN_INFO "md: bind<%s>\n", b);
1466 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1469 if (rdev->bdev->bd_part)
1470 ko = &rdev->bdev->bd_part->dev.kobj;
1472 ko = &rdev->bdev->bd_disk->dev.kobj;
1473 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1474 kobject_del(&rdev->kobj);
1477 list_add(&rdev->same_set, &mddev->disks);
1478 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1482 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1487 static void md_delayed_delete(struct work_struct *ws)
1489 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1490 kobject_del(&rdev->kobj);
1491 kobject_put(&rdev->kobj);
1494 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1496 char b[BDEVNAME_SIZE];
1501 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1502 list_del_init(&rdev->same_set);
1503 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1505 sysfs_remove_link(&rdev->kobj, "block");
1507 /* We need to delay this, otherwise we can deadlock when
1508 * writing to 'remove' to "dev/state"
1510 INIT_WORK(&rdev->del_work, md_delayed_delete);
1511 kobject_get(&rdev->kobj);
1512 schedule_work(&rdev->del_work);
1516 * prevent the device from being mounted, repartitioned or
1517 * otherwise reused by a RAID array (or any other kernel
1518 * subsystem), by bd_claiming the device.
1520 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1523 struct block_device *bdev;
1524 char b[BDEVNAME_SIZE];
1526 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1528 printk(KERN_ERR "md: could not open %s.\n",
1529 __bdevname(dev, b));
1530 return PTR_ERR(bdev);
1532 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1534 printk(KERN_ERR "md: could not bd_claim %s.\n",
1540 set_bit(AllReserved, &rdev->flags);
1545 static void unlock_rdev(mdk_rdev_t *rdev)
1547 struct block_device *bdev = rdev->bdev;
1555 void md_autodetect_dev(dev_t dev);
1557 static void export_rdev(mdk_rdev_t * rdev)
1559 char b[BDEVNAME_SIZE];
1560 printk(KERN_INFO "md: export_rdev(%s)\n",
1561 bdevname(rdev->bdev,b));
1565 list_del_init(&rdev->same_set);
1567 if (test_bit(AutoDetected, &rdev->flags))
1568 md_autodetect_dev(rdev->bdev->bd_dev);
1571 kobject_put(&rdev->kobj);
1574 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1576 unbind_rdev_from_array(rdev);
1580 static void export_array(mddev_t *mddev)
1582 struct list_head *tmp;
1585 rdev_for_each(rdev, tmp, mddev) {
1590 kick_rdev_from_array(rdev);
1592 if (!list_empty(&mddev->disks))
1594 mddev->raid_disks = 0;
1595 mddev->major_version = 0;
1598 static void print_desc(mdp_disk_t *desc)
1600 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1601 desc->major,desc->minor,desc->raid_disk,desc->state);
1604 static void print_sb(mdp_super_t *sb)
1609 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1610 sb->major_version, sb->minor_version, sb->patch_version,
1611 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1613 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1614 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1615 sb->md_minor, sb->layout, sb->chunk_size);
1616 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1617 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1618 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1619 sb->failed_disks, sb->spare_disks,
1620 sb->sb_csum, (unsigned long)sb->events_lo);
1623 for (i = 0; i < MD_SB_DISKS; i++) {
1626 desc = sb->disks + i;
1627 if (desc->number || desc->major || desc->minor ||
1628 desc->raid_disk || (desc->state && (desc->state != 4))) {
1629 printk(" D %2d: ", i);
1633 printk(KERN_INFO "md: THIS: ");
1634 print_desc(&sb->this_disk);
1638 static void print_rdev(mdk_rdev_t *rdev)
1640 char b[BDEVNAME_SIZE];
1641 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1642 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1643 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1645 if (rdev->sb_loaded) {
1646 printk(KERN_INFO "md: rdev superblock:\n");
1647 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1649 printk(KERN_INFO "md: no rdev superblock!\n");
1652 static void md_print_devices(void)
1654 struct list_head *tmp, *tmp2;
1657 char b[BDEVNAME_SIZE];
1660 printk("md: **********************************\n");
1661 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1662 printk("md: **********************************\n");
1663 for_each_mddev(mddev, tmp) {
1666 bitmap_print_sb(mddev->bitmap);
1668 printk("%s: ", mdname(mddev));
1669 rdev_for_each(rdev, tmp2, mddev)
1670 printk("<%s>", bdevname(rdev->bdev,b));
1673 rdev_for_each(rdev, tmp2, mddev)
1676 printk("md: **********************************\n");
1681 static void sync_sbs(mddev_t * mddev, int nospares)
1683 /* Update each superblock (in-memory image), but
1684 * if we are allowed to, skip spares which already
1685 * have the right event counter, or have one earlier
1686 * (which would mean they aren't being marked as dirty
1687 * with the rest of the array)
1690 struct list_head *tmp;
1692 rdev_for_each(rdev, tmp, mddev) {
1693 if (rdev->sb_events == mddev->events ||
1695 rdev->raid_disk < 0 &&
1696 (rdev->sb_events&1)==0 &&
1697 rdev->sb_events+1 == mddev->events)) {
1698 /* Don't update this superblock */
1699 rdev->sb_loaded = 2;
1701 super_types[mddev->major_version].
1702 sync_super(mddev, rdev);
1703 rdev->sb_loaded = 1;
1708 static void md_update_sb(mddev_t * mddev, int force_change)
1710 struct list_head *tmp;
1715 if (mddev->external)
1718 spin_lock_irq(&mddev->write_lock);
1720 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1721 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1723 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1724 /* just a clean<-> dirty transition, possibly leave spares alone,
1725 * though if events isn't the right even/odd, we will have to do
1731 if (mddev->degraded)
1732 /* If the array is degraded, then skipping spares is both
1733 * dangerous and fairly pointless.
1734 * Dangerous because a device that was removed from the array
1735 * might have a event_count that still looks up-to-date,
1736 * so it can be re-added without a resync.
1737 * Pointless because if there are any spares to skip,
1738 * then a recovery will happen and soon that array won't
1739 * be degraded any more and the spare can go back to sleep then.
1743 sync_req = mddev->in_sync;
1744 mddev->utime = get_seconds();
1746 /* If this is just a dirty<->clean transition, and the array is clean
1747 * and 'events' is odd, we can roll back to the previous clean state */
1749 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1750 && (mddev->events & 1)
1751 && mddev->events != 1)
1754 /* otherwise we have to go forward and ... */
1756 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1757 /* .. if the array isn't clean, insist on an odd 'events' */
1758 if ((mddev->events&1)==0) {
1763 /* otherwise insist on an even 'events' (for clean states) */
1764 if ((mddev->events&1)) {
1771 if (!mddev->events) {
1773 * oops, this 64-bit counter should never wrap.
1774 * Either we are in around ~1 trillion A.C., assuming
1775 * 1 reboot per second, or we have a bug:
1782 * do not write anything to disk if using
1783 * nonpersistent superblocks
1785 if (!mddev->persistent) {
1786 if (!mddev->external)
1787 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1789 spin_unlock_irq(&mddev->write_lock);
1790 wake_up(&mddev->sb_wait);
1793 sync_sbs(mddev, nospares);
1794 spin_unlock_irq(&mddev->write_lock);
1797 "md: updating %s RAID superblock on device (in sync %d)\n",
1798 mdname(mddev),mddev->in_sync);
1800 bitmap_update_sb(mddev->bitmap);
1801 rdev_for_each(rdev, tmp, mddev) {
1802 char b[BDEVNAME_SIZE];
1803 dprintk(KERN_INFO "md: ");
1804 if (rdev->sb_loaded != 1)
1805 continue; /* no noise on spare devices */
1806 if (test_bit(Faulty, &rdev->flags))
1807 dprintk("(skipping faulty ");
1809 dprintk("%s ", bdevname(rdev->bdev,b));
1810 if (!test_bit(Faulty, &rdev->flags)) {
1811 md_super_write(mddev,rdev,
1812 rdev->sb_offset<<1, rdev->sb_size,
1814 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1815 bdevname(rdev->bdev,b),
1816 (unsigned long long)rdev->sb_offset);
1817 rdev->sb_events = mddev->events;
1821 if (mddev->level == LEVEL_MULTIPATH)
1822 /* only need to write one superblock... */
1825 md_super_wait(mddev);
1826 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1828 spin_lock_irq(&mddev->write_lock);
1829 if (mddev->in_sync != sync_req ||
1830 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1831 /* have to write it out again */
1832 spin_unlock_irq(&mddev->write_lock);
1835 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1836 spin_unlock_irq(&mddev->write_lock);
1837 wake_up(&mddev->sb_wait);
1841 /* words written to sysfs files may, or may not, be \n terminated.
1842 * We want to accept with case. For this we use cmd_match.
1844 static int cmd_match(const char *cmd, const char *str)
1846 /* See if cmd, written into a sysfs file, matches
1847 * str. They must either be the same, or cmd can
1848 * have a trailing newline
1850 while (*cmd && *str && *cmd == *str) {
1861 struct rdev_sysfs_entry {
1862 struct attribute attr;
1863 ssize_t (*show)(mdk_rdev_t *, char *);
1864 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1868 state_show(mdk_rdev_t *rdev, char *page)
1873 if (test_bit(Faulty, &rdev->flags)) {
1874 len+= sprintf(page+len, "%sfaulty",sep);
1877 if (test_bit(In_sync, &rdev->flags)) {
1878 len += sprintf(page+len, "%sin_sync",sep);
1881 if (test_bit(WriteMostly, &rdev->flags)) {
1882 len += sprintf(page+len, "%swrite_mostly",sep);
1885 if (test_bit(Blocked, &rdev->flags)) {
1886 len += sprintf(page+len, "%sblocked", sep);
1889 if (!test_bit(Faulty, &rdev->flags) &&
1890 !test_bit(In_sync, &rdev->flags)) {
1891 len += sprintf(page+len, "%sspare", sep);
1894 return len+sprintf(page+len, "\n");
1898 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1901 * faulty - simulates and error
1902 * remove - disconnects the device
1903 * writemostly - sets write_mostly
1904 * -writemostly - clears write_mostly
1905 * blocked - sets the Blocked flag
1906 * -blocked - clears the Blocked flag
1909 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1910 md_error(rdev->mddev, rdev);
1912 } else if (cmd_match(buf, "remove")) {
1913 if (rdev->raid_disk >= 0)
1916 mddev_t *mddev = rdev->mddev;
1917 kick_rdev_from_array(rdev);
1919 md_update_sb(mddev, 1);
1920 md_new_event(mddev);
1923 } else if (cmd_match(buf, "writemostly")) {
1924 set_bit(WriteMostly, &rdev->flags);
1926 } else if (cmd_match(buf, "-writemostly")) {
1927 clear_bit(WriteMostly, &rdev->flags);
1929 } else if (cmd_match(buf, "blocked")) {
1930 set_bit(Blocked, &rdev->flags);
1932 } else if (cmd_match(buf, "-blocked")) {
1933 clear_bit(Blocked, &rdev->flags);
1934 wake_up(&rdev->blocked_wait);
1935 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1936 md_wakeup_thread(rdev->mddev->thread);
1941 sysfs_notify(&rdev->kobj, NULL, "state");
1942 return err ? err : len;
1944 static struct rdev_sysfs_entry rdev_state =
1945 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1948 errors_show(mdk_rdev_t *rdev, char *page)
1950 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1954 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1957 unsigned long n = simple_strtoul(buf, &e, 10);
1958 if (*buf && (*e == 0 || *e == '\n')) {
1959 atomic_set(&rdev->corrected_errors, n);
1964 static struct rdev_sysfs_entry rdev_errors =
1965 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1968 slot_show(mdk_rdev_t *rdev, char *page)
1970 if (rdev->raid_disk < 0)
1971 return sprintf(page, "none\n");
1973 return sprintf(page, "%d\n", rdev->raid_disk);
1977 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1982 int slot = simple_strtoul(buf, &e, 10);
1983 if (strncmp(buf, "none", 4)==0)
1985 else if (e==buf || (*e && *e!= '\n'))
1987 if (rdev->mddev->pers && slot == -1) {
1988 /* Setting 'slot' on an active array requires also
1989 * updating the 'rd%d' link, and communicating
1990 * with the personality with ->hot_*_disk.
1991 * For now we only support removing
1992 * failed/spare devices. This normally happens automatically,
1993 * but not when the metadata is externally managed.
1995 if (rdev->raid_disk == -1)
1997 /* personality does all needed checks */
1998 if (rdev->mddev->pers->hot_add_disk == NULL)
2000 err = rdev->mddev->pers->
2001 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2004 sprintf(nm, "rd%d", rdev->raid_disk);
2005 sysfs_remove_link(&rdev->mddev->kobj, nm);
2006 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2007 md_wakeup_thread(rdev->mddev->thread);
2008 } else if (rdev->mddev->pers) {
2010 struct list_head *tmp;
2011 /* Activating a spare .. or possibly reactivating
2012 * if we every get bitmaps working here.
2015 if (rdev->raid_disk != -1)
2018 if (rdev->mddev->pers->hot_add_disk == NULL)
2021 rdev_for_each(rdev2, tmp, rdev->mddev)
2022 if (rdev2->raid_disk == slot)
2025 rdev->raid_disk = slot;
2026 if (test_bit(In_sync, &rdev->flags))
2027 rdev->saved_raid_disk = slot;
2029 rdev->saved_raid_disk = -1;
2030 err = rdev->mddev->pers->
2031 hot_add_disk(rdev->mddev, rdev);
2033 rdev->raid_disk = -1;
2036 sysfs_notify(&rdev->kobj, NULL, "state");
2037 sprintf(nm, "rd%d", rdev->raid_disk);
2038 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2040 "md: cannot register "
2042 nm, mdname(rdev->mddev));
2044 /* don't wakeup anyone, leave that to userspace. */
2046 if (slot >= rdev->mddev->raid_disks)
2048 rdev->raid_disk = slot;
2049 /* assume it is working */
2050 clear_bit(Faulty, &rdev->flags);
2051 clear_bit(WriteMostly, &rdev->flags);
2052 set_bit(In_sync, &rdev->flags);
2053 sysfs_notify(&rdev->kobj, NULL, "state");
2059 static struct rdev_sysfs_entry rdev_slot =
2060 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2063 offset_show(mdk_rdev_t *rdev, char *page)
2065 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2069 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2072 unsigned long long offset = simple_strtoull(buf, &e, 10);
2073 if (e==buf || (*e && *e != '\n'))
2075 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2077 if (rdev->size && rdev->mddev->external)
2078 /* Must set offset before size, so overlap checks
2081 rdev->data_offset = offset;
2085 static struct rdev_sysfs_entry rdev_offset =
2086 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2089 rdev_size_show(mdk_rdev_t *rdev, char *page)
2091 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2094 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2096 /* check if two start/length pairs overlap */
2105 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2108 unsigned long long size = simple_strtoull(buf, &e, 10);
2109 unsigned long long oldsize = rdev->size;
2110 mddev_t *my_mddev = rdev->mddev;
2112 if (e==buf || (*e && *e != '\n'))
2114 if (my_mddev->pers && rdev->raid_disk >= 0) {
2115 if (rdev->mddev->persistent) {
2116 size = super_types[rdev->mddev->major_version].
2117 rdev_size_change(rdev, size);
2121 size = (rdev->bdev->bd_inode->i_size >> 10);
2122 size -= rdev->data_offset/2;
2124 if (size < rdev->mddev->size)
2125 return -EINVAL; /* component must fit device */
2129 if (size > oldsize && rdev->mddev->external) {
2130 /* need to check that all other rdevs with the same ->bdev
2131 * do not overlap. We need to unlock the mddev to avoid
2132 * a deadlock. We have already changed rdev->size, and if
2133 * we have to change it back, we will have the lock again.
2137 struct list_head *tmp, *tmp2;
2139 mddev_unlock(my_mddev);
2140 for_each_mddev(mddev, tmp) {
2144 rdev_for_each(rdev2, tmp2, mddev)
2145 if (test_bit(AllReserved, &rdev2->flags) ||
2146 (rdev->bdev == rdev2->bdev &&
2148 overlaps(rdev->data_offset, rdev->size,
2149 rdev2->data_offset, rdev2->size))) {
2153 mddev_unlock(mddev);
2159 mddev_lock(my_mddev);
2161 /* Someone else could have slipped in a size
2162 * change here, but doing so is just silly.
2163 * We put oldsize back because we *know* it is
2164 * safe, and trust userspace not to race with
2167 rdev->size = oldsize;
2171 if (size < my_mddev->size || my_mddev->size == 0)
2172 my_mddev->size = size;
2176 static struct rdev_sysfs_entry rdev_size =
2177 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2179 static struct attribute *rdev_default_attrs[] = {
2188 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2190 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2191 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2192 mddev_t *mddev = rdev->mddev;
2198 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2200 if (rdev->mddev == NULL)
2203 rv = entry->show(rdev, page);
2204 mddev_unlock(mddev);
2210 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2211 const char *page, size_t length)
2213 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2214 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2216 mddev_t *mddev = rdev->mddev;
2220 if (!capable(CAP_SYS_ADMIN))
2222 rv = mddev ? mddev_lock(mddev): -EBUSY;
2224 if (rdev->mddev == NULL)
2227 rv = entry->store(rdev, page, length);
2228 mddev_unlock(mddev);
2233 static void rdev_free(struct kobject *ko)
2235 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2238 static struct sysfs_ops rdev_sysfs_ops = {
2239 .show = rdev_attr_show,
2240 .store = rdev_attr_store,
2242 static struct kobj_type rdev_ktype = {
2243 .release = rdev_free,
2244 .sysfs_ops = &rdev_sysfs_ops,
2245 .default_attrs = rdev_default_attrs,
2249 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2251 * mark the device faulty if:
2253 * - the device is nonexistent (zero size)
2254 * - the device has no valid superblock
2256 * a faulty rdev _never_ has rdev->sb set.
2258 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2260 char b[BDEVNAME_SIZE];
2265 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2267 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2268 return ERR_PTR(-ENOMEM);
2271 if ((err = alloc_disk_sb(rdev)))
2274 err = lock_rdev(rdev, newdev, super_format == -2);
2278 kobject_init(&rdev->kobj, &rdev_ktype);
2281 rdev->saved_raid_disk = -1;
2282 rdev->raid_disk = -1;
2284 rdev->data_offset = 0;
2285 rdev->sb_events = 0;
2286 atomic_set(&rdev->nr_pending, 0);
2287 atomic_set(&rdev->read_errors, 0);
2288 atomic_set(&rdev->corrected_errors, 0);
2290 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2293 "md: %s has zero or unknown size, marking faulty!\n",
2294 bdevname(rdev->bdev,b));
2299 if (super_format >= 0) {
2300 err = super_types[super_format].
2301 load_super(rdev, NULL, super_minor);
2302 if (err == -EINVAL) {
2304 "md: %s does not have a valid v%d.%d "
2305 "superblock, not importing!\n",
2306 bdevname(rdev->bdev,b),
2307 super_format, super_minor);
2312 "md: could not read %s's sb, not importing!\n",
2313 bdevname(rdev->bdev,b));
2318 INIT_LIST_HEAD(&rdev->same_set);
2319 init_waitqueue_head(&rdev->blocked_wait);
2324 if (rdev->sb_page) {
2330 return ERR_PTR(err);
2334 * Check a full RAID array for plausibility
2338 static void analyze_sbs(mddev_t * mddev)
2341 struct list_head *tmp;
2342 mdk_rdev_t *rdev, *freshest;
2343 char b[BDEVNAME_SIZE];
2346 rdev_for_each(rdev, tmp, mddev)
2347 switch (super_types[mddev->major_version].
2348 load_super(rdev, freshest, mddev->minor_version)) {
2356 "md: fatal superblock inconsistency in %s"
2357 " -- removing from array\n",
2358 bdevname(rdev->bdev,b));
2359 kick_rdev_from_array(rdev);
2363 super_types[mddev->major_version].
2364 validate_super(mddev, freshest);
2367 rdev_for_each(rdev, tmp, mddev) {
2368 if (rdev != freshest)
2369 if (super_types[mddev->major_version].
2370 validate_super(mddev, rdev)) {
2371 printk(KERN_WARNING "md: kicking non-fresh %s"
2373 bdevname(rdev->bdev,b));
2374 kick_rdev_from_array(rdev);
2377 if (mddev->level == LEVEL_MULTIPATH) {
2378 rdev->desc_nr = i++;
2379 rdev->raid_disk = rdev->desc_nr;
2380 set_bit(In_sync, &rdev->flags);
2381 } else if (rdev->raid_disk >= mddev->raid_disks) {
2382 rdev->raid_disk = -1;
2383 clear_bit(In_sync, &rdev->flags);
2389 if (mddev->recovery_cp != MaxSector &&
2391 printk(KERN_ERR "md: %s: raid array is not clean"
2392 " -- starting background reconstruction\n",
2398 safe_delay_show(mddev_t *mddev, char *page)
2400 int msec = (mddev->safemode_delay*1000)/HZ;
2401 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2404 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2412 /* remove a period, and count digits after it */
2413 if (len >= sizeof(buf))
2415 strlcpy(buf, cbuf, len);
2417 for (i=0; i<len; i++) {
2419 if (isdigit(buf[i])) {
2424 } else if (buf[i] == '.') {
2429 msec = simple_strtoul(buf, &e, 10);
2430 if (e == buf || (*e && *e != '\n'))
2432 msec = (msec * 1000) / scale;
2434 mddev->safemode_delay = 0;
2436 mddev->safemode_delay = (msec*HZ)/1000;
2437 if (mddev->safemode_delay == 0)
2438 mddev->safemode_delay = 1;
2442 static struct md_sysfs_entry md_safe_delay =
2443 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2446 level_show(mddev_t *mddev, char *page)
2448 struct mdk_personality *p = mddev->pers;
2450 return sprintf(page, "%s\n", p->name);
2451 else if (mddev->clevel[0])
2452 return sprintf(page, "%s\n", mddev->clevel);
2453 else if (mddev->level != LEVEL_NONE)
2454 return sprintf(page, "%d\n", mddev->level);
2460 level_store(mddev_t *mddev, const char *buf, size_t len)
2467 if (len >= sizeof(mddev->clevel))
2469 strncpy(mddev->clevel, buf, len);
2470 if (mddev->clevel[len-1] == '\n')
2472 mddev->clevel[len] = 0;
2473 mddev->level = LEVEL_NONE;
2477 static struct md_sysfs_entry md_level =
2478 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2482 layout_show(mddev_t *mddev, char *page)
2484 /* just a number, not meaningful for all levels */
2485 if (mddev->reshape_position != MaxSector &&
2486 mddev->layout != mddev->new_layout)
2487 return sprintf(page, "%d (%d)\n",
2488 mddev->new_layout, mddev->layout);
2489 return sprintf(page, "%d\n", mddev->layout);
2493 layout_store(mddev_t *mddev, const char *buf, size_t len)
2496 unsigned long n = simple_strtoul(buf, &e, 10);
2498 if (!*buf || (*e && *e != '\n'))
2503 if (mddev->reshape_position != MaxSector)
2504 mddev->new_layout = n;
2509 static struct md_sysfs_entry md_layout =
2510 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2514 raid_disks_show(mddev_t *mddev, char *page)
2516 if (mddev->raid_disks == 0)
2518 if (mddev->reshape_position != MaxSector &&
2519 mddev->delta_disks != 0)
2520 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2521 mddev->raid_disks - mddev->delta_disks);
2522 return sprintf(page, "%d\n", mddev->raid_disks);
2525 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2528 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2532 unsigned long n = simple_strtoul(buf, &e, 10);
2534 if (!*buf || (*e && *e != '\n'))
2538 rv = update_raid_disks(mddev, n);
2539 else if (mddev->reshape_position != MaxSector) {
2540 int olddisks = mddev->raid_disks - mddev->delta_disks;
2541 mddev->delta_disks = n - olddisks;
2542 mddev->raid_disks = n;
2544 mddev->raid_disks = n;
2545 return rv ? rv : len;
2547 static struct md_sysfs_entry md_raid_disks =
2548 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2551 chunk_size_show(mddev_t *mddev, char *page)
2553 if (mddev->reshape_position != MaxSector &&
2554 mddev->chunk_size != mddev->new_chunk)
2555 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2557 return sprintf(page, "%d\n", mddev->chunk_size);
2561 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2563 /* can only set chunk_size if array is not yet active */
2565 unsigned long n = simple_strtoul(buf, &e, 10);
2567 if (!*buf || (*e && *e != '\n'))
2572 else if (mddev->reshape_position != MaxSector)
2573 mddev->new_chunk = n;
2575 mddev->chunk_size = n;
2578 static struct md_sysfs_entry md_chunk_size =
2579 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2582 resync_start_show(mddev_t *mddev, char *page)
2584 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2588 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2591 unsigned long long n = simple_strtoull(buf, &e, 10);
2595 if (!*buf || (*e && *e != '\n'))
2598 mddev->recovery_cp = n;
2601 static struct md_sysfs_entry md_resync_start =
2602 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2605 * The array state can be:
2608 * No devices, no size, no level
2609 * Equivalent to STOP_ARRAY ioctl
2611 * May have some settings, but array is not active
2612 * all IO results in error
2613 * When written, doesn't tear down array, but just stops it
2614 * suspended (not supported yet)
2615 * All IO requests will block. The array can be reconfigured.
2616 * Writing this, if accepted, will block until array is quiescent
2618 * no resync can happen. no superblocks get written.
2619 * write requests fail
2621 * like readonly, but behaves like 'clean' on a write request.
2623 * clean - no pending writes, but otherwise active.
2624 * When written to inactive array, starts without resync
2625 * If a write request arrives then
2626 * if metadata is known, mark 'dirty' and switch to 'active'.
2627 * if not known, block and switch to write-pending
2628 * If written to an active array that has pending writes, then fails.
2630 * fully active: IO and resync can be happening.
2631 * When written to inactive array, starts with resync
2634 * clean, but writes are blocked waiting for 'active' to be written.
2637 * like active, but no writes have been seen for a while (100msec).
2640 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2641 write_pending, active_idle, bad_word};
2642 static char *array_states[] = {
2643 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2644 "write-pending", "active-idle", NULL };
2646 static int match_word(const char *word, char **list)
2649 for (n=0; list[n]; n++)
2650 if (cmd_match(word, list[n]))
2656 array_state_show(mddev_t *mddev, char *page)
2658 enum array_state st = inactive;
2671 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2673 else if (mddev->safemode)
2679 if (list_empty(&mddev->disks) &&
2680 mddev->raid_disks == 0 &&
2686 return sprintf(page, "%s\n", array_states[st]);
2689 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2690 static int do_md_run(mddev_t * mddev);
2691 static int restart_array(mddev_t *mddev);
2694 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2697 enum array_state st = match_word(buf, array_states);
2702 /* stopping an active array */
2703 if (atomic_read(&mddev->active) > 1)
2705 err = do_md_stop(mddev, 0, 0);
2708 /* stopping an active array */
2710 if (atomic_read(&mddev->active) > 1)
2712 err = do_md_stop(mddev, 2, 0);
2714 err = 0; /* already inactive */
2717 break; /* not supported yet */
2720 err = do_md_stop(mddev, 1, 0);
2723 set_disk_ro(mddev->gendisk, 1);
2724 err = do_md_run(mddev);
2730 err = do_md_stop(mddev, 1, 0);
2732 err = restart_array(mddev);
2735 set_disk_ro(mddev->gendisk, 0);
2739 err = do_md_run(mddev);
2744 restart_array(mddev);
2745 spin_lock_irq(&mddev->write_lock);
2746 if (atomic_read(&mddev->writes_pending) == 0) {
2747 if (mddev->in_sync == 0) {
2749 if (mddev->safemode == 1)
2750 mddev->safemode = 0;
2751 if (mddev->persistent)
2752 set_bit(MD_CHANGE_CLEAN,
2758 spin_unlock_irq(&mddev->write_lock);
2761 mddev->recovery_cp = MaxSector;
2762 err = do_md_run(mddev);
2767 restart_array(mddev);
2768 if (mddev->external)
2769 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2770 wake_up(&mddev->sb_wait);
2774 set_disk_ro(mddev->gendisk, 0);
2775 err = do_md_run(mddev);
2780 /* these cannot be set */
2786 sysfs_notify(&mddev->kobj, NULL, "array_state");
2790 static struct md_sysfs_entry md_array_state =
2791 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2794 null_show(mddev_t *mddev, char *page)
2800 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2802 /* buf must be %d:%d\n? giving major and minor numbers */
2803 /* The new device is added to the array.
2804 * If the array has a persistent superblock, we read the
2805 * superblock to initialise info and check validity.
2806 * Otherwise, only checking done is that in bind_rdev_to_array,
2807 * which mainly checks size.
2810 int major = simple_strtoul(buf, &e, 10);
2816 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2818 minor = simple_strtoul(e+1, &e, 10);
2819 if (*e && *e != '\n')
2821 dev = MKDEV(major, minor);
2822 if (major != MAJOR(dev) ||
2823 minor != MINOR(dev))
2827 if (mddev->persistent) {
2828 rdev = md_import_device(dev, mddev->major_version,
2829 mddev->minor_version);
2830 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2831 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2832 mdk_rdev_t, same_set);
2833 err = super_types[mddev->major_version]
2834 .load_super(rdev, rdev0, mddev->minor_version);
2838 } else if (mddev->external)
2839 rdev = md_import_device(dev, -2, -1);
2841 rdev = md_import_device(dev, -1, -1);
2844 return PTR_ERR(rdev);
2845 err = bind_rdev_to_array(rdev, mddev);
2849 return err ? err : len;
2852 static struct md_sysfs_entry md_new_device =
2853 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2856 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2859 unsigned long chunk, end_chunk;
2863 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2865 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2866 if (buf == end) break;
2867 if (*end == '-') { /* range */
2869 end_chunk = simple_strtoul(buf, &end, 0);
2870 if (buf == end) break;
2872 if (*end && !isspace(*end)) break;
2873 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2875 while (isspace(*buf)) buf++;
2877 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2882 static struct md_sysfs_entry md_bitmap =
2883 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2886 size_show(mddev_t *mddev, char *page)
2888 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2891 static int update_size(mddev_t *mddev, sector_t num_sectors);
2894 size_store(mddev_t *mddev, const char *buf, size_t len)
2896 /* If array is inactive, we can reduce the component size, but
2897 * not increase it (except from 0).
2898 * If array is active, we can try an on-line resize
2902 unsigned long long size = simple_strtoull(buf, &e, 10);
2903 if (!*buf || *buf == '\n' ||
2908 err = update_size(mddev, size * 2);
2909 md_update_sb(mddev, 1);
2911 if (mddev->size == 0 ||
2917 return err ? err : len;
2920 static struct md_sysfs_entry md_size =
2921 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2926 * 'none' for arrays with no metadata (good luck...)
2927 * 'external' for arrays with externally managed metadata,
2928 * or N.M for internally known formats
2931 metadata_show(mddev_t *mddev, char *page)
2933 if (mddev->persistent)
2934 return sprintf(page, "%d.%d\n",
2935 mddev->major_version, mddev->minor_version);
2936 else if (mddev->external)
2937 return sprintf(page, "external:%s\n", mddev->metadata_type);
2939 return sprintf(page, "none\n");
2943 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2947 if (!list_empty(&mddev->disks))
2950 if (cmd_match(buf, "none")) {
2951 mddev->persistent = 0;
2952 mddev->external = 0;
2953 mddev->major_version = 0;
2954 mddev->minor_version = 90;
2957 if (strncmp(buf, "external:", 9) == 0) {
2958 size_t namelen = len-9;
2959 if (namelen >= sizeof(mddev->metadata_type))
2960 namelen = sizeof(mddev->metadata_type)-1;
2961 strncpy(mddev->metadata_type, buf+9, namelen);
2962 mddev->metadata_type[namelen] = 0;
2963 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2964 mddev->metadata_type[--namelen] = 0;
2965 mddev->persistent = 0;
2966 mddev->external = 1;
2967 mddev->major_version = 0;
2968 mddev->minor_version = 90;
2971 major = simple_strtoul(buf, &e, 10);
2972 if (e==buf || *e != '.')
2975 minor = simple_strtoul(buf, &e, 10);
2976 if (e==buf || (*e && *e != '\n') )
2978 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2980 mddev->major_version = major;
2981 mddev->minor_version = minor;
2982 mddev->persistent = 1;
2983 mddev->external = 0;
2987 static struct md_sysfs_entry md_metadata =
2988 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2991 action_show(mddev_t *mddev, char *page)
2993 char *type = "idle";
2994 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2995 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2996 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2998 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2999 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3001 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3005 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3008 return sprintf(page, "%s\n", type);
3012 action_store(mddev_t *mddev, const char *page, size_t len)
3014 if (!mddev->pers || !mddev->pers->sync_request)
3017 if (cmd_match(page, "idle")) {
3018 if (mddev->sync_thread) {
3019 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3020 md_unregister_thread(mddev->sync_thread);
3021 mddev->sync_thread = NULL;
3022 mddev->recovery = 0;
3024 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3025 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3027 else if (cmd_match(page, "resync"))
3028 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3029 else if (cmd_match(page, "recover")) {
3030 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3031 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3032 } else if (cmd_match(page, "reshape")) {
3034 if (mddev->pers->start_reshape == NULL)
3036 err = mddev->pers->start_reshape(mddev);
3039 sysfs_notify(&mddev->kobj, NULL, "degraded");
3041 if (cmd_match(page, "check"))
3042 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3043 else if (!cmd_match(page, "repair"))
3045 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3046 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3048 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3049 md_wakeup_thread(mddev->thread);
3050 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3055 mismatch_cnt_show(mddev_t *mddev, char *page)
3057 return sprintf(page, "%llu\n",
3058 (unsigned long long) mddev->resync_mismatches);
3061 static struct md_sysfs_entry md_scan_mode =
3062 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3065 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3068 sync_min_show(mddev_t *mddev, char *page)
3070 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3071 mddev->sync_speed_min ? "local": "system");
3075 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3079 if (strncmp(buf, "system", 6)==0) {
3080 mddev->sync_speed_min = 0;
3083 min = simple_strtoul(buf, &e, 10);
3084 if (buf == e || (*e && *e != '\n') || min <= 0)
3086 mddev->sync_speed_min = min;
3090 static struct md_sysfs_entry md_sync_min =
3091 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3094 sync_max_show(mddev_t *mddev, char *page)
3096 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3097 mddev->sync_speed_max ? "local": "system");
3101 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3105 if (strncmp(buf, "system", 6)==0) {
3106 mddev->sync_speed_max = 0;
3109 max = simple_strtoul(buf, &e, 10);
3110 if (buf == e || (*e && *e != '\n') || max <= 0)
3112 mddev->sync_speed_max = max;
3116 static struct md_sysfs_entry md_sync_max =
3117 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3120 degraded_show(mddev_t *mddev, char *page)
3122 return sprintf(page, "%d\n", mddev->degraded);
3124 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3127 sync_force_parallel_show(mddev_t *mddev, char *page)
3129 return sprintf(page, "%d\n", mddev->parallel_resync);
3133 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3137 if (strict_strtol(buf, 10, &n))
3140 if (n != 0 && n != 1)
3143 mddev->parallel_resync = n;
3145 if (mddev->sync_thread)
3146 wake_up(&resync_wait);
3151 /* force parallel resync, even with shared block devices */
3152 static struct md_sysfs_entry md_sync_force_parallel =
3153 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3154 sync_force_parallel_show, sync_force_parallel_store);
3157 sync_speed_show(mddev_t *mddev, char *page)
3159 unsigned long resync, dt, db;
3160 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3161 dt = (jiffies - mddev->resync_mark) / HZ;
3163 db = resync - mddev->resync_mark_cnt;
3164 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3167 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3170 sync_completed_show(mddev_t *mddev, char *page)
3172 unsigned long max_blocks, resync;
3174 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3175 max_blocks = mddev->resync_max_sectors;
3177 max_blocks = mddev->size << 1;
3179 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3180 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3183 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3186 min_sync_show(mddev_t *mddev, char *page)
3188 return sprintf(page, "%llu\n",
3189 (unsigned long long)mddev->resync_min);
3192 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3194 unsigned long long min;
3195 if (strict_strtoull(buf, 10, &min))
3197 if (min > mddev->resync_max)
3199 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3202 /* Must be a multiple of chunk_size */
3203 if (mddev->chunk_size) {
3204 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3207 mddev->resync_min = min;
3212 static struct md_sysfs_entry md_min_sync =
3213 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3216 max_sync_show(mddev_t *mddev, char *page)
3218 if (mddev->resync_max == MaxSector)
3219 return sprintf(page, "max\n");
3221 return sprintf(page, "%llu\n",
3222 (unsigned long long)mddev->resync_max);
3225 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3227 if (strncmp(buf, "max", 3) == 0)
3228 mddev->resync_max = MaxSector;
3230 unsigned long long max;
3231 if (strict_strtoull(buf, 10, &max))
3233 if (max < mddev->resync_min)
3235 if (max < mddev->resync_max &&
3236 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3239 /* Must be a multiple of chunk_size */
3240 if (mddev->chunk_size) {
3241 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3244 mddev->resync_max = max;
3246 wake_up(&mddev->recovery_wait);
3250 static struct md_sysfs_entry md_max_sync =
3251 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3254 suspend_lo_show(mddev_t *mddev, char *page)
3256 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3260 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3263 unsigned long long new = simple_strtoull(buf, &e, 10);
3265 if (mddev->pers->quiesce == NULL)
3267 if (buf == e || (*e && *e != '\n'))
3269 if (new >= mddev->suspend_hi ||
3270 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3271 mddev->suspend_lo = new;
3272 mddev->pers->quiesce(mddev, 2);
3277 static struct md_sysfs_entry md_suspend_lo =
3278 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3282 suspend_hi_show(mddev_t *mddev, char *page)
3284 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3288 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3291 unsigned long long new = simple_strtoull(buf, &e, 10);
3293 if (mddev->pers->quiesce == NULL)
3295 if (buf == e || (*e && *e != '\n'))
3297 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3298 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3299 mddev->suspend_hi = new;
3300 mddev->pers->quiesce(mddev, 1);
3301 mddev->pers->quiesce(mddev, 0);
3306 static struct md_sysfs_entry md_suspend_hi =
3307 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3310 reshape_position_show(mddev_t *mddev, char *page)
3312 if (mddev->reshape_position != MaxSector)
3313 return sprintf(page, "%llu\n",
3314 (unsigned long long)mddev->reshape_position);
3315 strcpy(page, "none\n");
3320 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3323 unsigned long long new = simple_strtoull(buf, &e, 10);
3326 if (buf == e || (*e && *e != '\n'))
3328 mddev->reshape_position = new;
3329 mddev->delta_disks = 0;
3330 mddev->new_level = mddev->level;
3331 mddev->new_layout = mddev->layout;
3332 mddev->new_chunk = mddev->chunk_size;
3336 static struct md_sysfs_entry md_reshape_position =
3337 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3338 reshape_position_store);
3341 static struct attribute *md_default_attrs[] = {
3344 &md_raid_disks.attr,
3345 &md_chunk_size.attr,
3347 &md_resync_start.attr,
3349 &md_new_device.attr,
3350 &md_safe_delay.attr,
3351 &md_array_state.attr,
3352 &md_reshape_position.attr,
3356 static struct attribute *md_redundancy_attrs[] = {
3358 &md_mismatches.attr,
3361 &md_sync_speed.attr,
3362 &md_sync_force_parallel.attr,
3363 &md_sync_completed.attr,
3366 &md_suspend_lo.attr,
3367 &md_suspend_hi.attr,
3372 static struct attribute_group md_redundancy_group = {
3374 .attrs = md_redundancy_attrs,
3379 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3381 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3382 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3387 rv = mddev_lock(mddev);
3389 rv = entry->show(mddev, page);
3390 mddev_unlock(mddev);
3396 md_attr_store(struct kobject *kobj, struct attribute *attr,
3397 const char *page, size_t length)
3399 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3400 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3405 if (!capable(CAP_SYS_ADMIN))
3407 rv = mddev_lock(mddev);
3409 rv = entry->store(mddev, page, length);
3410 mddev_unlock(mddev);
3415 static void md_free(struct kobject *ko)
3417 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3421 static struct sysfs_ops md_sysfs_ops = {
3422 .show = md_attr_show,
3423 .store = md_attr_store,
3425 static struct kobj_type md_ktype = {
3427 .sysfs_ops = &md_sysfs_ops,
3428 .default_attrs = md_default_attrs,
3433 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3435 static DEFINE_MUTEX(disks_mutex);
3436 mddev_t *mddev = mddev_find(dev);
3437 struct gendisk *disk;
3438 int partitioned = (MAJOR(dev) != MD_MAJOR);
3439 int shift = partitioned ? MdpMinorShift : 0;
3440 int unit = MINOR(dev) >> shift;
3446 mutex_lock(&disks_mutex);
3447 if (mddev->gendisk) {
3448 mutex_unlock(&disks_mutex);
3452 disk = alloc_disk(1 << shift);
3454 mutex_unlock(&disks_mutex);
3458 disk->major = MAJOR(dev);
3459 disk->first_minor = unit << shift;
3461 sprintf(disk->disk_name, "md_d%d", unit);
3463 sprintf(disk->disk_name, "md%d", unit);
3464 disk->fops = &md_fops;
3465 disk->private_data = mddev;
3466 disk->queue = mddev->queue;
3468 mddev->gendisk = disk;
3469 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3471 mutex_unlock(&disks_mutex);
3473 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3476 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3480 static void md_safemode_timeout(unsigned long data)
3482 mddev_t *mddev = (mddev_t *) data;
3484 if (!atomic_read(&mddev->writes_pending)) {
3485 mddev->safemode = 1;
3486 if (mddev->external)
3487 sysfs_notify(&mddev->kobj, NULL, "array_state");
3489 md_wakeup_thread(mddev->thread);
3492 static int start_dirty_degraded;
3494 static int do_md_run(mddev_t * mddev)
3498 struct list_head *tmp;
3500 struct gendisk *disk;
3501 struct mdk_personality *pers;
3502 char b[BDEVNAME_SIZE];
3504 if (list_empty(&mddev->disks))
3505 /* cannot run an array with no devices.. */
3512 * Analyze all RAID superblock(s)
3514 if (!mddev->raid_disks) {
3515 if (!mddev->persistent)
3520 chunk_size = mddev->chunk_size;
3523 if (chunk_size > MAX_CHUNK_SIZE) {
3524 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3525 chunk_size, MAX_CHUNK_SIZE);
3529 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3531 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3532 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3535 if (chunk_size < PAGE_SIZE) {
3536 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3537 chunk_size, PAGE_SIZE);
3541 /* devices must have minimum size of one chunk */
3542 rdev_for_each(rdev, tmp, mddev) {
3543 if (test_bit(Faulty, &rdev->flags))
3545 if (rdev->size < chunk_size / 1024) {
3547 "md: Dev %s smaller than chunk_size:"
3549 bdevname(rdev->bdev,b),
3550 (unsigned long long)rdev->size,
3558 if (mddev->level != LEVEL_NONE)
3559 request_module("md-level-%d", mddev->level);
3560 else if (mddev->clevel[0])
3561 request_module("md-%s", mddev->clevel);
3565 * Drop all container device buffers, from now on
3566 * the only valid external interface is through the md
3569 rdev_for_each(rdev, tmp, mddev) {
3570 if (test_bit(Faulty, &rdev->flags))
3572 sync_blockdev(rdev->bdev);
3573 invalidate_bdev(rdev->bdev);
3575 /* perform some consistency tests on the device.
3576 * We don't want the data to overlap the metadata,
3577 * Internal Bitmap issues has handled elsewhere.
3579 if (rdev->data_offset < rdev->sb_offset) {
3581 rdev->data_offset + mddev->size*2
3582 > rdev->sb_offset*2) {
3583 printk("md: %s: data overlaps metadata\n",
3588 if (rdev->sb_offset*2 + rdev->sb_size/512
3589 > rdev->data_offset) {
3590 printk("md: %s: metadata overlaps data\n",
3595 sysfs_notify(&rdev->kobj, NULL, "state");
3598 md_probe(mddev->unit, NULL, NULL);
3599 disk = mddev->gendisk;
3603 spin_lock(&pers_lock);
3604 pers = find_pers(mddev->level, mddev->clevel);
3605 if (!pers || !try_module_get(pers->owner)) {
3606 spin_unlock(&pers_lock);
3607 if (mddev->level != LEVEL_NONE)
3608 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3611 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3616 spin_unlock(&pers_lock);
3617 mddev->level = pers->level;
3618 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3620 if (mddev->reshape_position != MaxSector &&
3621 pers->start_reshape == NULL) {
3622 /* This personality cannot handle reshaping... */
3624 module_put(pers->owner);
3628 if (pers->sync_request) {
3629 /* Warn if this is a potentially silly
3632 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3634 struct list_head *tmp2;
3636 rdev_for_each(rdev, tmp, mddev) {
3637 rdev_for_each(rdev2, tmp2, mddev) {
3639 rdev->bdev->bd_contains ==
3640 rdev2->bdev->bd_contains) {
3642 "%s: WARNING: %s appears to be"
3643 " on the same physical disk as"
3646 bdevname(rdev->bdev,b),
3647 bdevname(rdev2->bdev,b2));
3654 "True protection against single-disk"
3655 " failure might be compromised.\n");
3658 mddev->recovery = 0;
3659 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3660 mddev->barriers_work = 1;
3661 mddev->ok_start_degraded = start_dirty_degraded;
3664 mddev->ro = 2; /* read-only, but switch on first write */
3666 err = mddev->pers->run(mddev);
3668 printk(KERN_ERR "md: pers->run() failed ...\n");
3669 else if (mddev->pers->sync_request) {
3670 err = bitmap_create(mddev);
3672 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3673 mdname(mddev), err);
3674 mddev->pers->stop(mddev);
3678 module_put(mddev->pers->owner);
3680 bitmap_destroy(mddev);
3683 if (mddev->pers->sync_request) {
3684 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3686 "md: cannot register extra attributes for %s\n",
3688 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3691 atomic_set(&mddev->writes_pending,0);
3692 mddev->safemode = 0;
3693 mddev->safemode_timer.function = md_safemode_timeout;
3694 mddev->safemode_timer.data = (unsigned long) mddev;
3695 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3698 rdev_for_each(rdev, tmp, mddev)
3699 if (rdev->raid_disk >= 0) {
3701 sprintf(nm, "rd%d", rdev->raid_disk);
3702 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3703 printk("md: cannot register %s for %s\n",
3707 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3710 md_update_sb(mddev, 0);
3712 set_capacity(disk, mddev->array_size<<1);
3714 /* If we call blk_queue_make_request here, it will
3715 * re-initialise max_sectors etc which may have been
3716 * refined inside -> run. So just set the bits we need to set.
3717 * Most initialisation happended when we called
3718 * blk_queue_make_request(..., md_fail_request)
3721 mddev->queue->queuedata = mddev;
3722 mddev->queue->make_request_fn = mddev->pers->make_request;
3724 /* If there is a partially-recovered drive we need to
3725 * start recovery here. If we leave it to md_check_recovery,
3726 * it will remove the drives and not do the right thing
3728 if (mddev->degraded && !mddev->sync_thread) {
3729 struct list_head *rtmp;
3731 rdev_for_each(rdev, rtmp, mddev)
3732 if (rdev->raid_disk >= 0 &&
3733 !test_bit(In_sync, &rdev->flags) &&
3734 !test_bit(Faulty, &rdev->flags))
3735 /* complete an interrupted recovery */
3737 if (spares && mddev->pers->sync_request) {
3738 mddev->recovery = 0;
3739 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3740 mddev->sync_thread = md_register_thread(md_do_sync,
3743 if (!mddev->sync_thread) {
3744 printk(KERN_ERR "%s: could not start resync"
3747 /* leave the spares where they are, it shouldn't hurt */
3748 mddev->recovery = 0;
3752 md_wakeup_thread(mddev->thread);
3753 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3756 md_new_event(mddev);
3757 sysfs_notify(&mddev->kobj, NULL, "array_state");
3758 sysfs_notify(&mddev->kobj, NULL, "sync_action");
3759 sysfs_notify(&mddev->kobj, NULL, "degraded");
3760 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3764 static int restart_array(mddev_t *mddev)
3766 struct gendisk *disk = mddev->gendisk;
3768 /* Complain if it has no devices */
3769 if (list_empty(&mddev->disks))
3775 mddev->safemode = 0;
3777 set_disk_ro(disk, 0);
3778 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3780 /* Kick recovery or resync if necessary */
3781 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3782 md_wakeup_thread(mddev->thread);
3783 md_wakeup_thread(mddev->sync_thread);
3784 sysfs_notify(&mddev->kobj, NULL, "array_state");
3788 /* similar to deny_write_access, but accounts for our holding a reference
3789 * to the file ourselves */
3790 static int deny_bitmap_write_access(struct file * file)
3792 struct inode *inode = file->f_mapping->host;
3794 spin_lock(&inode->i_lock);
3795 if (atomic_read(&inode->i_writecount) > 1) {
3796 spin_unlock(&inode->i_lock);
3799 atomic_set(&inode->i_writecount, -1);
3800 spin_unlock(&inode->i_lock);
3805 static void restore_bitmap_write_access(struct file *file)
3807 struct inode *inode = file->f_mapping->host;
3809 spin_lock(&inode->i_lock);
3810 atomic_set(&inode->i_writecount, 1);
3811 spin_unlock(&inode->i_lock);
3815 * 0 - completely stop and dis-assemble array
3816 * 1 - switch to readonly
3817 * 2 - stop but do not disassemble array
3819 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3822 struct gendisk *disk = mddev->gendisk;
3824 if (atomic_read(&mddev->active) > 1 + is_open) {
3825 printk("md: %s still in use.\n",mdname(mddev));
3831 if (mddev->sync_thread) {
3832 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3833 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3834 md_unregister_thread(mddev->sync_thread);
3835 mddev->sync_thread = NULL;
3838 del_timer_sync(&mddev->safemode_timer);
3840 invalidate_partition(disk, 0);
3843 case 1: /* readonly */
3849 case 0: /* disassemble */
3851 bitmap_flush(mddev);
3852 md_super_wait(mddev);
3854 set_disk_ro(disk, 0);
3855 blk_queue_make_request(mddev->queue, md_fail_request);
3856 mddev->pers->stop(mddev);
3857 mddev->queue->merge_bvec_fn = NULL;
3858 mddev->queue->unplug_fn = NULL;
3859 mddev->queue->backing_dev_info.congested_fn = NULL;
3860 if (mddev->pers->sync_request)
3861 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3863 module_put(mddev->pers->owner);
3865 /* tell userspace to handle 'inactive' */
3866 sysfs_notify(&mddev->kobj, NULL, "array_state");
3868 set_capacity(disk, 0);
3874 if (!mddev->in_sync || mddev->flags) {
3875 /* mark array as shutdown cleanly */
3877 md_update_sb(mddev, 1);
3880 set_disk_ro(disk, 1);
3881 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3885 * Free resources if final stop
3889 struct list_head *tmp;
3891 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3893 bitmap_destroy(mddev);
3894 if (mddev->bitmap_file) {
3895 restore_bitmap_write_access(mddev->bitmap_file);
3896 fput(mddev->bitmap_file);
3897 mddev->bitmap_file = NULL;
3899 mddev->bitmap_offset = 0;
3901 rdev_for_each(rdev, tmp, mddev)
3902 if (rdev->raid_disk >= 0) {
3904 sprintf(nm, "rd%d", rdev->raid_disk);
3905 sysfs_remove_link(&mddev->kobj, nm);
3908 /* make sure all md_delayed_delete calls have finished */
3909 flush_scheduled_work();
3911 export_array(mddev);
3913 mddev->array_size = 0;
3915 mddev->raid_disks = 0;
3916 mddev->recovery_cp = 0;
3917 mddev->resync_min = 0;
3918 mddev->resync_max = MaxSector;
3919 mddev->reshape_position = MaxSector;
3920 mddev->external = 0;
3921 mddev->persistent = 0;
3922 mddev->level = LEVEL_NONE;
3923 mddev->clevel[0] = 0;
3926 mddev->metadata_type[0] = 0;
3927 mddev->chunk_size = 0;
3928 mddev->ctime = mddev->utime = 0;
3930 mddev->max_disks = 0;
3932 mddev->delta_disks = 0;
3933 mddev->new_level = LEVEL_NONE;
3934 mddev->new_layout = 0;
3935 mddev->new_chunk = 0;
3936 mddev->curr_resync = 0;
3937 mddev->resync_mismatches = 0;
3938 mddev->suspend_lo = mddev->suspend_hi = 0;
3939 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3940 mddev->recovery = 0;
3943 mddev->degraded = 0;
3944 mddev->barriers_work = 0;
3945 mddev->safemode = 0;
3947 } else if (mddev->pers)
3948 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3951 md_new_event(mddev);
3952 sysfs_notify(&mddev->kobj, NULL, "array_state");
3958 static void autorun_array(mddev_t *mddev)
3961 struct list_head *tmp;
3964 if (list_empty(&mddev->disks))
3967 printk(KERN_INFO "md: running: ");
3969 rdev_for_each(rdev, tmp, mddev) {
3970 char b[BDEVNAME_SIZE];
3971 printk("<%s>", bdevname(rdev->bdev,b));
3975 err = do_md_run (mddev);
3977 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3978 do_md_stop (mddev, 0, 0);
3983 * lets try to run arrays based on all disks that have arrived
3984 * until now. (those are in pending_raid_disks)
3986 * the method: pick the first pending disk, collect all disks with
3987 * the same UUID, remove all from the pending list and put them into
3988 * the 'same_array' list. Then order this list based on superblock
3989 * update time (freshest comes first), kick out 'old' disks and
3990 * compare superblocks. If everything's fine then run it.
3992 * If "unit" is allocated, then bump its reference count
3994 static void autorun_devices(int part)
3996 struct list_head *tmp;
3997 mdk_rdev_t *rdev0, *rdev;
3999 char b[BDEVNAME_SIZE];
4001 printk(KERN_INFO "md: autorun ...\n");
4002 while (!list_empty(&pending_raid_disks)) {
4005 LIST_HEAD(candidates);
4006 rdev0 = list_entry(pending_raid_disks.next,
4007 mdk_rdev_t, same_set);
4009 printk(KERN_INFO "md: considering %s ...\n",
4010 bdevname(rdev0->bdev,b));
4011 INIT_LIST_HEAD(&candidates);
4012 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4013 if (super_90_load(rdev, rdev0, 0) >= 0) {
4014 printk(KERN_INFO "md: adding %s ...\n",
4015 bdevname(rdev->bdev,b));
4016 list_move(&rdev->same_set, &candidates);
4019 * now we have a set of devices, with all of them having
4020 * mostly sane superblocks. It's time to allocate the
4024 dev = MKDEV(mdp_major,
4025 rdev0->preferred_minor << MdpMinorShift);
4026 unit = MINOR(dev) >> MdpMinorShift;
4028 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4031 if (rdev0->preferred_minor != unit) {
4032 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4033 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4037 md_probe(dev, NULL, NULL);
4038 mddev = mddev_find(dev);
4039 if (!mddev || !mddev->gendisk) {
4043 "md: cannot allocate memory for md drive.\n");
4046 if (mddev_lock(mddev))
4047 printk(KERN_WARNING "md: %s locked, cannot run\n",
4049 else if (mddev->raid_disks || mddev->major_version
4050 || !list_empty(&mddev->disks)) {
4052 "md: %s already running, cannot run %s\n",
4053 mdname(mddev), bdevname(rdev0->bdev,b));
4054 mddev_unlock(mddev);
4056 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4057 mddev->persistent = 1;
4058 rdev_for_each_list(rdev, tmp, candidates) {
4059 list_del_init(&rdev->same_set);
4060 if (bind_rdev_to_array(rdev, mddev))
4063 autorun_array(mddev);
4064 mddev_unlock(mddev);
4066 /* on success, candidates will be empty, on error
4069 rdev_for_each_list(rdev, tmp, candidates)
4073 printk(KERN_INFO "md: ... autorun DONE.\n");
4075 #endif /* !MODULE */
4077 static int get_version(void __user * arg)
4081 ver.major = MD_MAJOR_VERSION;
4082 ver.minor = MD_MINOR_VERSION;
4083 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4085 if (copy_to_user(arg, &ver, sizeof(ver)))
4091 static int get_array_info(mddev_t * mddev, void __user * arg)
4093 mdu_array_info_t info;
4094 int nr,working,active,failed,spare;
4096 struct list_head *tmp;
4098 nr=working=active=failed=spare=0;
4099 rdev_for_each(rdev, tmp, mddev) {
4101 if (test_bit(Faulty, &rdev->flags))
4105 if (test_bit(In_sync, &rdev->flags))
4112 info.major_version = mddev->major_version;
4113 info.minor_version = mddev->minor_version;
4114 info.patch_version = MD_PATCHLEVEL_VERSION;
4115 info.ctime = mddev->ctime;
4116 info.level = mddev->level;
4117 info.size = mddev->size;
4118 if (info.size != mddev->size) /* overflow */
4121 info.raid_disks = mddev->raid_disks;
4122 info.md_minor = mddev->md_minor;
4123 info.not_persistent= !mddev->persistent;
4125 info.utime = mddev->utime;
4128 info.state = (1<<MD_SB_CLEAN);
4129 if (mddev->bitmap && mddev->bitmap_offset)
4130 info.state = (1<<MD_SB_BITMAP_PRESENT);
4131 info.active_disks = active;
4132 info.working_disks = working;
4133 info.failed_disks = failed;
4134 info.spare_disks = spare;
4136 info.layout = mddev->layout;
4137 info.chunk_size = mddev->chunk_size;
4139 if (copy_to_user(arg, &info, sizeof(info)))
4145 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4147 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4148 char *ptr, *buf = NULL;
4151 if (md_allow_write(mddev))
4152 file = kmalloc(sizeof(*file), GFP_NOIO);
4154 file = kmalloc(sizeof(*file), GFP_KERNEL);
4159 /* bitmap disabled, zero the first byte and copy out */
4160 if (!mddev->bitmap || !mddev->bitmap->file) {
4161 file->pathname[0] = '\0';
4165 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4169 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4173 strcpy(file->pathname, ptr);
4177 if (copy_to_user(arg, file, sizeof(*file)))
4185 static int get_disk_info(mddev_t * mddev, void __user * arg)
4187 mdu_disk_info_t info;
4190 if (copy_from_user(&info, arg, sizeof(info)))
4193 rdev = find_rdev_nr(mddev, info.number);
4195 info.major = MAJOR(rdev->bdev->bd_dev);
4196 info.minor = MINOR(rdev->bdev->bd_dev);
4197 info.raid_disk = rdev->raid_disk;
4199 if (test_bit(Faulty, &rdev->flags))
4200 info.state |= (1<<MD_DISK_FAULTY);
4201 else if (test_bit(In_sync, &rdev->flags)) {
4202 info.state |= (1<<MD_DISK_ACTIVE);
4203 info.state |= (1<<MD_DISK_SYNC);
4205 if (test_bit(WriteMostly, &rdev->flags))
4206 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4208 info.major = info.minor = 0;
4209 info.raid_disk = -1;
4210 info.state = (1<<MD_DISK_REMOVED);
4213 if (copy_to_user(arg, &info, sizeof(info)))
4219 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4221 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4223 dev_t dev = MKDEV(info->major,info->minor);
4225 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4228 if (!mddev->raid_disks) {
4230 /* expecting a device which has a superblock */
4231 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4234 "md: md_import_device returned %ld\n",
4236 return PTR_ERR(rdev);
4238 if (!list_empty(&mddev->disks)) {
4239 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4240 mdk_rdev_t, same_set);
4241 int err = super_types[mddev->major_version]
4242 .load_super(rdev, rdev0, mddev->minor_version);
4245 "md: %s has different UUID to %s\n",
4246 bdevname(rdev->bdev,b),
4247 bdevname(rdev0->bdev,b2));
4252 err = bind_rdev_to_array(rdev, mddev);
4259 * add_new_disk can be used once the array is assembled
4260 * to add "hot spares". They must already have a superblock
4265 if (!mddev->pers->hot_add_disk) {
4267 "%s: personality does not support diskops!\n",
4271 if (mddev->persistent)
4272 rdev = md_import_device(dev, mddev->major_version,
4273 mddev->minor_version);
4275 rdev = md_import_device(dev, -1, -1);
4278 "md: md_import_device returned %ld\n",
4280 return PTR_ERR(rdev);
4282 /* set save_raid_disk if appropriate */
4283 if (!mddev->persistent) {
4284 if (info->state & (1<<MD_DISK_SYNC) &&
4285 info->raid_disk < mddev->raid_disks)
4286 rdev->raid_disk = info->raid_disk;
4288 rdev->raid_disk = -1;
4290 super_types[mddev->major_version].
4291 validate_super(mddev, rdev);
4292 rdev->saved_raid_disk = rdev->raid_disk;
4294 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4295 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4296 set_bit(WriteMostly, &rdev->flags);
4298 rdev->raid_disk = -1;
4299 err = bind_rdev_to_array(rdev, mddev);
4300 if (!err && !mddev->pers->hot_remove_disk) {
4301 /* If there is hot_add_disk but no hot_remove_disk
4302 * then added disks for geometry changes,
4303 * and should be added immediately.
4305 super_types[mddev->major_version].
4306 validate_super(mddev, rdev);
4307 err = mddev->pers->hot_add_disk(mddev, rdev);
4309 unbind_rdev_from_array(rdev);
4314 sysfs_notify(&rdev->kobj, NULL, "state");
4316 md_update_sb(mddev, 1);
4317 if (mddev->degraded)
4318 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4319 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4320 md_wakeup_thread(mddev->thread);
4324 /* otherwise, add_new_disk is only allowed
4325 * for major_version==0 superblocks
4327 if (mddev->major_version != 0) {
4328 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4333 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4335 rdev = md_import_device (dev, -1, 0);
4338 "md: error, md_import_device() returned %ld\n",
4340 return PTR_ERR(rdev);
4342 rdev->desc_nr = info->number;
4343 if (info->raid_disk < mddev->raid_disks)
4344 rdev->raid_disk = info->raid_disk;
4346 rdev->raid_disk = -1;
4348 if (rdev->raid_disk < mddev->raid_disks)
4349 if (info->state & (1<<MD_DISK_SYNC))
4350 set_bit(In_sync, &rdev->flags);
4352 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4353 set_bit(WriteMostly, &rdev->flags);
4355 if (!mddev->persistent) {
4356 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4357 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4359 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4360 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4362 err = bind_rdev_to_array(rdev, mddev);
4372 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4374 char b[BDEVNAME_SIZE];
4377 rdev = find_rdev(mddev, dev);
4381 if (rdev->raid_disk >= 0)
4384 kick_rdev_from_array(rdev);
4385 md_update_sb(mddev, 1);
4386 md_new_event(mddev);
4390 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4391 bdevname(rdev->bdev,b), mdname(mddev));
4395 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4397 char b[BDEVNAME_SIZE];
4404 if (mddev->major_version != 0) {
4405 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4406 " version-0 superblocks.\n",
4410 if (!mddev->pers->hot_add_disk) {
4412 "%s: personality does not support diskops!\n",
4417 rdev = md_import_device (dev, -1, 0);
4420 "md: error, md_import_device() returned %ld\n",
4425 if (mddev->persistent)
4426 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4429 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4431 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4433 if (test_bit(Faulty, &rdev->flags)) {
4435 "md: can not hot-add faulty %s disk to %s!\n",
4436 bdevname(rdev->bdev,b), mdname(mddev));
4440 clear_bit(In_sync, &rdev->flags);
4442 rdev->saved_raid_disk = -1;
4443 err = bind_rdev_to_array(rdev, mddev);
4448 * The rest should better be atomic, we can have disk failures
4449 * noticed in interrupt contexts ...
4452 if (rdev->desc_nr == mddev->max_disks) {
4453 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4456 goto abort_unbind_export;
4459 rdev->raid_disk = -1;
4461 md_update_sb(mddev, 1);
4464 * Kick recovery, maybe this spare has to be added to the
4465 * array immediately.
4467 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4468 md_wakeup_thread(mddev->thread);
4469 md_new_event(mddev);
4472 abort_unbind_export:
4473 unbind_rdev_from_array(rdev);
4480 static int set_bitmap_file(mddev_t *mddev, int fd)
4485 if (!mddev->pers->quiesce)
4487 if (mddev->recovery || mddev->sync_thread)
4489 /* we should be able to change the bitmap.. */
4495 return -EEXIST; /* cannot add when bitmap is present */
4496 mddev->bitmap_file = fget(fd);
4498 if (mddev->bitmap_file == NULL) {
4499 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4504 err = deny_bitmap_write_access(mddev->bitmap_file);
4506 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4508 fput(mddev->bitmap_file);
4509 mddev->bitmap_file = NULL;
4512 mddev->bitmap_offset = 0; /* file overrides offset */
4513 } else if (mddev->bitmap == NULL)
4514 return -ENOENT; /* cannot remove what isn't there */
4517 mddev->pers->quiesce(mddev, 1);
4519 err = bitmap_create(mddev);
4520 if (fd < 0 || err) {
4521 bitmap_destroy(mddev);
4522 fd = -1; /* make sure to put the file */
4524 mddev->pers->quiesce(mddev, 0);
4527 if (mddev->bitmap_file) {
4528 restore_bitmap_write_access(mddev->bitmap_file);
4529 fput(mddev->bitmap_file);
4531 mddev->bitmap_file = NULL;
4538 * set_array_info is used two different ways
4539 * The original usage is when creating a new array.
4540 * In this usage, raid_disks is > 0 and it together with
4541 * level, size, not_persistent,layout,chunksize determine the
4542 * shape of the array.
4543 * This will always create an array with a type-0.90.0 superblock.
4544 * The newer usage is when assembling an array.
4545 * In this case raid_disks will be 0, and the major_version field is
4546 * use to determine which style super-blocks are to be found on the devices.
4547 * The minor and patch _version numbers are also kept incase the
4548 * super_block handler wishes to interpret them.
4550 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4553 if (info->raid_disks == 0) {
4554 /* just setting version number for superblock loading */
4555 if (info->major_version < 0 ||
4556 info->major_version >= ARRAY_SIZE(super_types) ||
4557 super_types[info->major_version].name == NULL) {
4558 /* maybe try to auto-load a module? */
4560 "md: superblock version %d not known\n",
4561 info->major_version);
4564 mddev->major_version = info->major_version;
4565 mddev->minor_version = info->minor_version;
4566 mddev->patch_version = info->patch_version;
4567 mddev->persistent = !info->not_persistent;
4570 mddev->major_version = MD_MAJOR_VERSION;
4571 mddev->minor_version = MD_MINOR_VERSION;
4572 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4573 mddev->ctime = get_seconds();
4575 mddev->level = info->level;
4576 mddev->clevel[0] = 0;
4577 mddev->size = info->size;
4578 mddev->raid_disks = info->raid_disks;
4579 /* don't set md_minor, it is determined by which /dev/md* was
4582 if (info->state & (1<<MD_SB_CLEAN))
4583 mddev->recovery_cp = MaxSector;
4585 mddev->recovery_cp = 0;
4586 mddev->persistent = ! info->not_persistent;
4587 mddev->external = 0;
4589 mddev->layout = info->layout;
4590 mddev->chunk_size = info->chunk_size;
4592 mddev->max_disks = MD_SB_DISKS;
4594 if (mddev->persistent)
4596 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4598 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4599 mddev->bitmap_offset = 0;
4601 mddev->reshape_position = MaxSector;
4604 * Generate a 128 bit UUID
4606 get_random_bytes(mddev->uuid, 16);
4608 mddev->new_level = mddev->level;
4609 mddev->new_chunk = mddev->chunk_size;
4610 mddev->new_layout = mddev->layout;
4611 mddev->delta_disks = 0;
4616 static int update_size(mddev_t *mddev, sector_t num_sectors)
4620 struct list_head *tmp;
4621 int fit = (num_sectors == 0);
4623 if (mddev->pers->resize == NULL)
4625 /* The "num_sectors" is the number of sectors of each device that
4626 * is used. This can only make sense for arrays with redundancy.
4627 * linear and raid0 always use whatever space is available. We can only
4628 * consider changing this number if no resync or reconstruction is
4629 * happening, and if the new size is acceptable. It must fit before the
4630 * sb_offset or, if that is <data_offset, it must fit before the size
4631 * of each device. If num_sectors is zero, we find the largest size
4635 if (mddev->sync_thread)
4637 rdev_for_each(rdev, tmp, mddev) {
4639 avail = rdev->size * 2;
4641 if (fit && (num_sectors == 0 || num_sectors > avail))
4642 num_sectors = avail;
4643 if (avail < num_sectors)
4646 rv = mddev->pers->resize(mddev, num_sectors);
4648 struct block_device *bdev;
4650 bdev = bdget_disk(mddev->gendisk, 0);
4652 mutex_lock(&bdev->bd_inode->i_mutex);
4653 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4654 mutex_unlock(&bdev->bd_inode->i_mutex);
4661 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4664 /* change the number of raid disks */
4665 if (mddev->pers->check_reshape == NULL)
4667 if (raid_disks <= 0 ||
4668 raid_disks >= mddev->max_disks)
4670 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4672 mddev->delta_disks = raid_disks - mddev->raid_disks;
4674 rv = mddev->pers->check_reshape(mddev);
4680 * update_array_info is used to change the configuration of an
4682 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4683 * fields in the info are checked against the array.
4684 * Any differences that cannot be handled will cause an error.
4685 * Normally, only one change can be managed at a time.
4687 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4693 /* calculate expected state,ignoring low bits */
4694 if (mddev->bitmap && mddev->bitmap_offset)
4695 state |= (1 << MD_SB_BITMAP_PRESENT);
4697 if (mddev->major_version != info->major_version ||
4698 mddev->minor_version != info->minor_version ||
4699 /* mddev->patch_version != info->patch_version || */
4700 mddev->ctime != info->ctime ||
4701 mddev->level != info->level ||
4702 /* mddev->layout != info->layout || */
4703 !mddev->persistent != info->not_persistent||
4704 mddev->chunk_size != info->chunk_size ||
4705 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4706 ((state^info->state) & 0xfffffe00)
4709 /* Check there is only one change */
4710 if (info->size >= 0 && mddev->size != info->size) cnt++;
4711 if (mddev->raid_disks != info->raid_disks) cnt++;
4712 if (mddev->layout != info->layout) cnt++;
4713 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4714 if (cnt == 0) return 0;
4715 if (cnt > 1) return -EINVAL;
4717 if (mddev->layout != info->layout) {
4719 * we don't need to do anything at the md level, the
4720 * personality will take care of it all.
4722 if (mddev->pers->reconfig == NULL)
4725 return mddev->pers->reconfig(mddev, info->layout, -1);
4727 if (info->size >= 0 && mddev->size != info->size)
4728 rv = update_size(mddev, (sector_t)info->size * 2);
4730 if (mddev->raid_disks != info->raid_disks)
4731 rv = update_raid_disks(mddev, info->raid_disks);
4733 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4734 if (mddev->pers->quiesce == NULL)
4736 if (mddev->recovery || mddev->sync_thread)
4738 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4739 /* add the bitmap */
4742 if (mddev->default_bitmap_offset == 0)
4744 mddev->bitmap_offset = mddev->default_bitmap_offset;
4745 mddev->pers->quiesce(mddev, 1);
4746 rv = bitmap_create(mddev);
4748 bitmap_destroy(mddev);
4749 mddev->pers->quiesce(mddev, 0);
4751 /* remove the bitmap */
4754 if (mddev->bitmap->file)
4756 mddev->pers->quiesce(mddev, 1);
4757 bitmap_destroy(mddev);
4758 mddev->pers->quiesce(mddev, 0);
4759 mddev->bitmap_offset = 0;
4762 md_update_sb(mddev, 1);
4766 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4770 if (mddev->pers == NULL)
4773 rdev = find_rdev(mddev, dev);
4777 md_error(mddev, rdev);
4782 * We have a problem here : there is no easy way to give a CHS
4783 * virtual geometry. We currently pretend that we have a 2 heads
4784 * 4 sectors (with a BIG number of cylinders...). This drives
4785 * dosfs just mad... ;-)
4787 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4789 mddev_t *mddev = bdev->bd_disk->private_data;
4793 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4797 static int md_ioctl(struct inode *inode, struct file *file,
4798 unsigned int cmd, unsigned long arg)
4801 void __user *argp = (void __user *)arg;
4802 mddev_t *mddev = NULL;
4804 if (!capable(CAP_SYS_ADMIN))
4808 * Commands dealing with the RAID driver but not any
4814 err = get_version(argp);
4817 case PRINT_RAID_DEBUG:
4825 autostart_arrays(arg);
4832 * Commands creating/starting a new array:
4835 mddev = inode->i_bdev->bd_disk->private_data;
4842 err = mddev_lock(mddev);
4845 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4852 case SET_ARRAY_INFO:
4854 mdu_array_info_t info;
4856 memset(&info, 0, sizeof(info));
4857 else if (copy_from_user(&info, argp, sizeof(info))) {
4862 err = update_array_info(mddev, &info);
4864 printk(KERN_WARNING "md: couldn't update"
4865 " array info. %d\n", err);
4870 if (!list_empty(&mddev->disks)) {
4872 "md: array %s already has disks!\n",
4877 if (mddev->raid_disks) {
4879 "md: array %s already initialised!\n",
4884 err = set_array_info(mddev, &info);
4886 printk(KERN_WARNING "md: couldn't set"
4887 " array info. %d\n", err);
4897 * Commands querying/configuring an existing array:
4899 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4900 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4901 if ((!mddev->raid_disks && !mddev->external)
4902 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4903 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4904 && cmd != GET_BITMAP_FILE) {
4910 * Commands even a read-only array can execute:
4914 case GET_ARRAY_INFO:
4915 err = get_array_info(mddev, argp);
4918 case GET_BITMAP_FILE:
4919 err = get_bitmap_file(mddev, argp);
4923 err = get_disk_info(mddev, argp);
4926 case RESTART_ARRAY_RW:
4927 err = restart_array(mddev);
4931 err = do_md_stop (mddev, 0, 1);
4935 err = do_md_stop (mddev, 1, 1);
4941 * The remaining ioctls are changing the state of the
4942 * superblock, so we do not allow them on read-only arrays.
4943 * However non-MD ioctls (e.g. get-size) will still come through
4944 * here and hit the 'default' below, so only disallow
4945 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4947 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4948 if (mddev->ro == 2) {
4950 sysfs_notify(&mddev->kobj, NULL, "array_state");
4951 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4952 md_wakeup_thread(mddev->thread);
4963 mdu_disk_info_t info;
4964 if (copy_from_user(&info, argp, sizeof(info)))
4967 err = add_new_disk(mddev, &info);
4971 case HOT_REMOVE_DISK:
4972 err = hot_remove_disk(mddev, new_decode_dev(arg));
4976 err = hot_add_disk(mddev, new_decode_dev(arg));
4979 case SET_DISK_FAULTY:
4980 err = set_disk_faulty(mddev, new_decode_dev(arg));
4984 err = do_md_run (mddev);
4987 case SET_BITMAP_FILE:
4988 err = set_bitmap_file(mddev, (int)arg);
4998 mddev_unlock(mddev);
5008 static int md_open(struct inode *inode, struct file *file)
5011 * Succeed if we can lock the mddev, which confirms that
5012 * it isn't being stopped right now.
5014 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5017 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5022 mddev_unlock(mddev);
5024 check_disk_change(inode->i_bdev);
5029 static int md_release(struct inode *inode, struct file * file)
5031 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5039 static int md_media_changed(struct gendisk *disk)
5041 mddev_t *mddev = disk->private_data;
5043 return mddev->changed;
5046 static int md_revalidate(struct gendisk *disk)
5048 mddev_t *mddev = disk->private_data;
5053 static struct block_device_operations md_fops =
5055 .owner = THIS_MODULE,
5057 .release = md_release,
5059 .getgeo = md_getgeo,
5060 .media_changed = md_media_changed,
5061 .revalidate_disk= md_revalidate,
5064 static int md_thread(void * arg)
5066 mdk_thread_t *thread = arg;
5069 * md_thread is a 'system-thread', it's priority should be very
5070 * high. We avoid resource deadlocks individually in each
5071 * raid personality. (RAID5 does preallocation) We also use RR and
5072 * the very same RT priority as kswapd, thus we will never get
5073 * into a priority inversion deadlock.
5075 * we definitely have to have equal or higher priority than
5076 * bdflush, otherwise bdflush will deadlock if there are too
5077 * many dirty RAID5 blocks.
5080 allow_signal(SIGKILL);
5081 while (!kthread_should_stop()) {
5083 /* We need to wait INTERRUPTIBLE so that
5084 * we don't add to the load-average.
5085 * That means we need to be sure no signals are
5088 if (signal_pending(current))
5089 flush_signals(current);
5091 wait_event_interruptible_timeout
5093 test_bit(THREAD_WAKEUP, &thread->flags)
5094 || kthread_should_stop(),
5097 clear_bit(THREAD_WAKEUP, &thread->flags);
5099 thread->run(thread->mddev);
5105 void md_wakeup_thread(mdk_thread_t *thread)
5108 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5109 set_bit(THREAD_WAKEUP, &thread->flags);
5110 wake_up(&thread->wqueue);
5114 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5117 mdk_thread_t *thread;
5119 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5123 init_waitqueue_head(&thread->wqueue);
5126 thread->mddev = mddev;
5127 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5128 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5129 if (IS_ERR(thread->tsk)) {
5136 void md_unregister_thread(mdk_thread_t *thread)
5138 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5140 kthread_stop(thread->tsk);
5144 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5151 if (!rdev || test_bit(Faulty, &rdev->flags))
5154 if (mddev->external)
5155 set_bit(Blocked, &rdev->flags);
5157 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5159 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5160 __builtin_return_address(0),__builtin_return_address(1),
5161 __builtin_return_address(2),__builtin_return_address(3));
5165 if (!mddev->pers->error_handler)
5167 mddev->pers->error_handler(mddev,rdev);
5168 if (mddev->degraded)
5169 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5170 set_bit(StateChanged, &rdev->flags);
5171 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5172 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5173 md_wakeup_thread(mddev->thread);
5174 md_new_event_inintr(mddev);
5177 /* seq_file implementation /proc/mdstat */
5179 static void status_unused(struct seq_file *seq)
5183 struct list_head *tmp;
5185 seq_printf(seq, "unused devices: ");
5187 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5188 char b[BDEVNAME_SIZE];
5190 seq_printf(seq, "%s ",
5191 bdevname(rdev->bdev,b));
5194 seq_printf(seq, "<none>");
5196 seq_printf(seq, "\n");
5200 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5202 sector_t max_blocks, resync, res;
5203 unsigned long dt, db, rt;
5205 unsigned int per_milli;
5207 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5209 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5210 max_blocks = mddev->resync_max_sectors >> 1;
5212 max_blocks = mddev->size;
5215 * Should not happen.
5221 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5222 * in a sector_t, and (max_blocks>>scale) will fit in a
5223 * u32, as those are the requirements for sector_div.
5224 * Thus 'scale' must be at least 10
5227 if (sizeof(sector_t) > sizeof(unsigned long)) {
5228 while ( max_blocks/2 > (1ULL<<(scale+32)))
5231 res = (resync>>scale)*1000;
5232 sector_div(res, (u32)((max_blocks>>scale)+1));
5236 int i, x = per_milli/50, y = 20-x;
5237 seq_printf(seq, "[");
5238 for (i = 0; i < x; i++)
5239 seq_printf(seq, "=");
5240 seq_printf(seq, ">");
5241 for (i = 0; i < y; i++)
5242 seq_printf(seq, ".");
5243 seq_printf(seq, "] ");
5245 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5246 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5248 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5250 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5251 "resync" : "recovery"))),
5252 per_milli/10, per_milli % 10,
5253 (unsigned long long) resync,
5254 (unsigned long long) max_blocks);
5257 * We do not want to overflow, so the order of operands and
5258 * the * 100 / 100 trick are important. We do a +1 to be
5259 * safe against division by zero. We only estimate anyway.
5261 * dt: time from mark until now
5262 * db: blocks written from mark until now
5263 * rt: remaining time
5265 dt = ((jiffies - mddev->resync_mark) / HZ);
5267 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5268 - mddev->resync_mark_cnt;
5269 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5271 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5273 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5276 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5278 struct list_head *tmp;
5288 spin_lock(&all_mddevs_lock);
5289 list_for_each(tmp,&all_mddevs)
5291 mddev = list_entry(tmp, mddev_t, all_mddevs);
5293 spin_unlock(&all_mddevs_lock);
5296 spin_unlock(&all_mddevs_lock);
5298 return (void*)2;/* tail */
5302 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5304 struct list_head *tmp;
5305 mddev_t *next_mddev, *mddev = v;
5311 spin_lock(&all_mddevs_lock);
5313 tmp = all_mddevs.next;
5315 tmp = mddev->all_mddevs.next;
5316 if (tmp != &all_mddevs)
5317 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5319 next_mddev = (void*)2;
5322 spin_unlock(&all_mddevs_lock);
5330 static void md_seq_stop(struct seq_file *seq, void *v)
5334 if (mddev && v != (void*)1 && v != (void*)2)
5338 struct mdstat_info {
5342 static int md_seq_show(struct seq_file *seq, void *v)
5346 struct list_head *tmp2;
5348 struct mdstat_info *mi = seq->private;
5349 struct bitmap *bitmap;
5351 if (v == (void*)1) {
5352 struct mdk_personality *pers;
5353 seq_printf(seq, "Personalities : ");
5354 spin_lock(&pers_lock);
5355 list_for_each_entry(pers, &pers_list, list)
5356 seq_printf(seq, "[%s] ", pers->name);
5358 spin_unlock(&pers_lock);
5359 seq_printf(seq, "\n");
5360 mi->event = atomic_read(&md_event_count);
5363 if (v == (void*)2) {
5368 if (mddev_lock(mddev) < 0)
5371 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5372 seq_printf(seq, "%s : %sactive", mdname(mddev),
5373 mddev->pers ? "" : "in");
5376 seq_printf(seq, " (read-only)");
5378 seq_printf(seq, " (auto-read-only)");
5379 seq_printf(seq, " %s", mddev->pers->name);
5383 rdev_for_each(rdev, tmp2, mddev) {
5384 char b[BDEVNAME_SIZE];
5385 seq_printf(seq, " %s[%d]",
5386 bdevname(rdev->bdev,b), rdev->desc_nr);
5387 if (test_bit(WriteMostly, &rdev->flags))
5388 seq_printf(seq, "(W)");
5389 if (test_bit(Faulty, &rdev->flags)) {
5390 seq_printf(seq, "(F)");
5392 } else if (rdev->raid_disk < 0)
5393 seq_printf(seq, "(S)"); /* spare */
5397 if (!list_empty(&mddev->disks)) {
5399 seq_printf(seq, "\n %llu blocks",
5400 (unsigned long long)mddev->array_size);
5402 seq_printf(seq, "\n %llu blocks",
5403 (unsigned long long)size);
5405 if (mddev->persistent) {
5406 if (mddev->major_version != 0 ||
5407 mddev->minor_version != 90) {
5408 seq_printf(seq," super %d.%d",
5409 mddev->major_version,
5410 mddev->minor_version);
5412 } else if (mddev->external)
5413 seq_printf(seq, " super external:%s",
5414 mddev->metadata_type);
5416 seq_printf(seq, " super non-persistent");
5419 mddev->pers->status (seq, mddev);
5420 seq_printf(seq, "\n ");
5421 if (mddev->pers->sync_request) {
5422 if (mddev->curr_resync > 2) {
5423 status_resync (seq, mddev);
5424 seq_printf(seq, "\n ");
5425 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5426 seq_printf(seq, "\tresync=DELAYED\n ");
5427 else if (mddev->recovery_cp < MaxSector)
5428 seq_printf(seq, "\tresync=PENDING\n ");
5431 seq_printf(seq, "\n ");
5433 if ((bitmap = mddev->bitmap)) {
5434 unsigned long chunk_kb;
5435 unsigned long flags;
5436 spin_lock_irqsave(&bitmap->lock, flags);
5437 chunk_kb = bitmap->chunksize >> 10;
5438 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5440 bitmap->pages - bitmap->missing_pages,
5442 (bitmap->pages - bitmap->missing_pages)
5443 << (PAGE_SHIFT - 10),
5444 chunk_kb ? chunk_kb : bitmap->chunksize,
5445 chunk_kb ? "KB" : "B");
5447 seq_printf(seq, ", file: ");
5448 seq_path(seq, &bitmap->file->f_path, " \t\n");
5451 seq_printf(seq, "\n");
5452 spin_unlock_irqrestore(&bitmap->lock, flags);
5455 seq_printf(seq, "\n");
5457 mddev_unlock(mddev);
5462 static struct seq_operations md_seq_ops = {
5463 .start = md_seq_start,
5464 .next = md_seq_next,
5465 .stop = md_seq_stop,
5466 .show = md_seq_show,
5469 static int md_seq_open(struct inode *inode, struct file *file)
5472 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5476 error = seq_open(file, &md_seq_ops);
5480 struct seq_file *p = file->private_data;
5482 mi->event = atomic_read(&md_event_count);
5487 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5489 struct seq_file *m = filp->private_data;
5490 struct mdstat_info *mi = m->private;
5493 poll_wait(filp, &md_event_waiters, wait);
5495 /* always allow read */
5496 mask = POLLIN | POLLRDNORM;
5498 if (mi->event != atomic_read(&md_event_count))
5499 mask |= POLLERR | POLLPRI;
5503 static const struct file_operations md_seq_fops = {
5504 .owner = THIS_MODULE,
5505 .open = md_seq_open,
5507 .llseek = seq_lseek,
5508 .release = seq_release_private,
5509 .poll = mdstat_poll,
5512 int register_md_personality(struct mdk_personality *p)
5514 spin_lock(&pers_lock);
5515 list_add_tail(&p->list, &pers_list);
5516 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5517 spin_unlock(&pers_lock);
5521 int unregister_md_personality(struct mdk_personality *p)
5523 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5524 spin_lock(&pers_lock);
5525 list_del_init(&p->list);
5526 spin_unlock(&pers_lock);
5530 static int is_mddev_idle(mddev_t *mddev)
5533 struct list_head *tmp;
5538 rdev_for_each(rdev, tmp, mddev) {
5539 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5540 curr_events = disk_stat_read(disk, sectors[0]) +
5541 disk_stat_read(disk, sectors[1]) -
5542 atomic_read(&disk->sync_io);
5543 /* sync IO will cause sync_io to increase before the disk_stats
5544 * as sync_io is counted when a request starts, and
5545 * disk_stats is counted when it completes.
5546 * So resync activity will cause curr_events to be smaller than
5547 * when there was no such activity.
5548 * non-sync IO will cause disk_stat to increase without
5549 * increasing sync_io so curr_events will (eventually)
5550 * be larger than it was before. Once it becomes
5551 * substantially larger, the test below will cause
5552 * the array to appear non-idle, and resync will slow
5554 * If there is a lot of outstanding resync activity when
5555 * we set last_event to curr_events, then all that activity
5556 * completing might cause the array to appear non-idle
5557 * and resync will be slowed down even though there might
5558 * not have been non-resync activity. This will only
5559 * happen once though. 'last_events' will soon reflect
5560 * the state where there is little or no outstanding
5561 * resync requests, and further resync activity will
5562 * always make curr_events less than last_events.
5565 if (curr_events - rdev->last_events > 4096) {
5566 rdev->last_events = curr_events;
5573 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5575 /* another "blocks" (512byte) blocks have been synced */
5576 atomic_sub(blocks, &mddev->recovery_active);
5577 wake_up(&mddev->recovery_wait);
5579 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5580 md_wakeup_thread(mddev->thread);
5581 // stop recovery, signal do_sync ....
5586 /* md_write_start(mddev, bi)
5587 * If we need to update some array metadata (e.g. 'active' flag
5588 * in superblock) before writing, schedule a superblock update
5589 * and wait for it to complete.
5591 void md_write_start(mddev_t *mddev, struct bio *bi)
5594 if (bio_data_dir(bi) != WRITE)
5597 BUG_ON(mddev->ro == 1);
5598 if (mddev->ro == 2) {
5599 /* need to switch to read/write */
5601 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5602 md_wakeup_thread(mddev->thread);
5603 md_wakeup_thread(mddev->sync_thread);
5606 atomic_inc(&mddev->writes_pending);
5607 if (mddev->safemode == 1)
5608 mddev->safemode = 0;
5609 if (mddev->in_sync) {
5610 spin_lock_irq(&mddev->write_lock);
5611 if (mddev->in_sync) {
5613 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5614 md_wakeup_thread(mddev->thread);
5617 spin_unlock_irq(&mddev->write_lock);
5620 sysfs_notify(&mddev->kobj, NULL, "array_state");
5621 wait_event(mddev->sb_wait,
5622 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5623 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5626 void md_write_end(mddev_t *mddev)
5628 if (atomic_dec_and_test(&mddev->writes_pending)) {
5629 if (mddev->safemode == 2)
5630 md_wakeup_thread(mddev->thread);
5631 else if (mddev->safemode_delay)
5632 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5636 /* md_allow_write(mddev)
5637 * Calling this ensures that the array is marked 'active' so that writes
5638 * may proceed without blocking. It is important to call this before
5639 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5640 * Must be called with mddev_lock held.
5642 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5643 * is dropped, so return -EAGAIN after notifying userspace.
5645 int md_allow_write(mddev_t *mddev)
5651 if (!mddev->pers->sync_request)
5654 spin_lock_irq(&mddev->write_lock);
5655 if (mddev->in_sync) {
5657 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5658 if (mddev->safemode_delay &&
5659 mddev->safemode == 0)
5660 mddev->safemode = 1;
5661 spin_unlock_irq(&mddev->write_lock);
5662 md_update_sb(mddev, 0);
5663 sysfs_notify(&mddev->kobj, NULL, "array_state");
5665 spin_unlock_irq(&mddev->write_lock);
5667 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5672 EXPORT_SYMBOL_GPL(md_allow_write);
5674 #define SYNC_MARKS 10
5675 #define SYNC_MARK_STEP (3*HZ)
5676 void md_do_sync(mddev_t *mddev)
5679 unsigned int currspeed = 0,
5681 sector_t max_sectors,j, io_sectors;
5682 unsigned long mark[SYNC_MARKS];
5683 sector_t mark_cnt[SYNC_MARKS];
5685 struct list_head *tmp;
5686 sector_t last_check;
5688 struct list_head *rtmp;
5692 /* just incase thread restarts... */
5693 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5695 if (mddev->ro) /* never try to sync a read-only array */
5698 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5699 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5700 desc = "data-check";
5701 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5702 desc = "requested-resync";
5705 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5710 /* we overload curr_resync somewhat here.
5711 * 0 == not engaged in resync at all
5712 * 2 == checking that there is no conflict with another sync
5713 * 1 == like 2, but have yielded to allow conflicting resync to
5715 * other == active in resync - this many blocks
5717 * Before starting a resync we must have set curr_resync to
5718 * 2, and then checked that every "conflicting" array has curr_resync
5719 * less than ours. When we find one that is the same or higher
5720 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5721 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5722 * This will mean we have to start checking from the beginning again.
5727 mddev->curr_resync = 2;
5730 if (kthread_should_stop()) {
5731 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5734 for_each_mddev(mddev2, tmp) {
5735 if (mddev2 == mddev)
5737 if (!mddev->parallel_resync
5738 && mddev2->curr_resync
5739 && match_mddev_units(mddev, mddev2)) {
5741 if (mddev < mddev2 && mddev->curr_resync == 2) {
5742 /* arbitrarily yield */
5743 mddev->curr_resync = 1;
5744 wake_up(&resync_wait);
5746 if (mddev > mddev2 && mddev->curr_resync == 1)
5747 /* no need to wait here, we can wait the next
5748 * time 'round when curr_resync == 2
5751 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5752 if (!kthread_should_stop() &&
5753 mddev2->curr_resync >= mddev->curr_resync) {
5754 printk(KERN_INFO "md: delaying %s of %s"
5755 " until %s has finished (they"
5756 " share one or more physical units)\n",
5757 desc, mdname(mddev), mdname(mddev2));
5760 finish_wait(&resync_wait, &wq);
5763 finish_wait(&resync_wait, &wq);
5766 } while (mddev->curr_resync < 2);
5769 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5770 /* resync follows the size requested by the personality,
5771 * which defaults to physical size, but can be virtual size
5773 max_sectors = mddev->resync_max_sectors;
5774 mddev->resync_mismatches = 0;
5775 /* we don't use the checkpoint if there's a bitmap */
5776 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5777 j = mddev->resync_min;
5778 else if (!mddev->bitmap)
5779 j = mddev->recovery_cp;
5781 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5782 max_sectors = mddev->size << 1;
5784 /* recovery follows the physical size of devices */
5785 max_sectors = mddev->size << 1;
5787 rdev_for_each(rdev, rtmp, mddev)
5788 if (rdev->raid_disk >= 0 &&
5789 !test_bit(Faulty, &rdev->flags) &&
5790 !test_bit(In_sync, &rdev->flags) &&
5791 rdev->recovery_offset < j)
5792 j = rdev->recovery_offset;
5795 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5796 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5797 " %d KB/sec/disk.\n", speed_min(mddev));
5798 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5799 "(but not more than %d KB/sec) for %s.\n",
5800 speed_max(mddev), desc);
5802 is_mddev_idle(mddev); /* this also initializes IO event counters */
5805 for (m = 0; m < SYNC_MARKS; m++) {
5807 mark_cnt[m] = io_sectors;
5810 mddev->resync_mark = mark[last_mark];
5811 mddev->resync_mark_cnt = mark_cnt[last_mark];
5814 * Tune reconstruction:
5816 window = 32*(PAGE_SIZE/512);
5817 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5818 window/2,(unsigned long long) max_sectors/2);
5820 atomic_set(&mddev->recovery_active, 0);
5825 "md: resuming %s of %s from checkpoint.\n",
5826 desc, mdname(mddev));
5827 mddev->curr_resync = j;
5830 while (j < max_sectors) {
5834 if (j >= mddev->resync_max) {
5835 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5836 wait_event(mddev->recovery_wait,
5837 mddev->resync_max > j
5838 || kthread_should_stop());
5840 if (kthread_should_stop())
5842 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5843 currspeed < speed_min(mddev));
5845 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5849 if (!skipped) { /* actual IO requested */
5850 io_sectors += sectors;
5851 atomic_add(sectors, &mddev->recovery_active);
5855 if (j>1) mddev->curr_resync = j;
5856 mddev->curr_mark_cnt = io_sectors;
5857 if (last_check == 0)
5858 /* this is the earliers that rebuilt will be
5859 * visible in /proc/mdstat
5861 md_new_event(mddev);
5863 if (last_check + window > io_sectors || j == max_sectors)
5866 last_check = io_sectors;
5868 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5872 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5874 int next = (last_mark+1) % SYNC_MARKS;
5876 mddev->resync_mark = mark[next];
5877 mddev->resync_mark_cnt = mark_cnt[next];
5878 mark[next] = jiffies;
5879 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5884 if (kthread_should_stop())
5889 * this loop exits only if either when we are slower than
5890 * the 'hard' speed limit, or the system was IO-idle for
5892 * the system might be non-idle CPU-wise, but we only care
5893 * about not overloading the IO subsystem. (things like an
5894 * e2fsck being done on the RAID array should execute fast)
5896 blk_unplug(mddev->queue);
5899 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5900 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5902 if (currspeed > speed_min(mddev)) {
5903 if ((currspeed > speed_max(mddev)) ||
5904 !is_mddev_idle(mddev)) {
5910 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5912 * this also signals 'finished resyncing' to md_stop
5915 blk_unplug(mddev->queue);
5917 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5919 /* tell personality that we are finished */
5920 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5922 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5923 mddev->curr_resync > 2) {
5924 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5925 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5926 if (mddev->curr_resync >= mddev->recovery_cp) {
5928 "md: checkpointing %s of %s.\n",
5929 desc, mdname(mddev));
5930 mddev->recovery_cp = mddev->curr_resync;
5933 mddev->recovery_cp = MaxSector;
5935 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5936 mddev->curr_resync = MaxSector;
5937 rdev_for_each(rdev, rtmp, mddev)
5938 if (rdev->raid_disk >= 0 &&
5939 !test_bit(Faulty, &rdev->flags) &&
5940 !test_bit(In_sync, &rdev->flags) &&
5941 rdev->recovery_offset < mddev->curr_resync)
5942 rdev->recovery_offset = mddev->curr_resync;
5945 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5948 mddev->curr_resync = 0;
5949 mddev->resync_min = 0;
5950 mddev->resync_max = MaxSector;
5951 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5952 wake_up(&resync_wait);
5953 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5954 md_wakeup_thread(mddev->thread);
5959 * got a signal, exit.
5962 "md: md_do_sync() got signal ... exiting\n");
5963 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5967 EXPORT_SYMBOL_GPL(md_do_sync);
5970 static int remove_and_add_spares(mddev_t *mddev)
5973 struct list_head *rtmp;
5976 rdev_for_each(rdev, rtmp, mddev)
5977 if (rdev->raid_disk >= 0 &&
5978 !test_bit(Blocked, &rdev->flags) &&
5979 (test_bit(Faulty, &rdev->flags) ||
5980 ! test_bit(In_sync, &rdev->flags)) &&
5981 atomic_read(&rdev->nr_pending)==0) {
5982 if (mddev->pers->hot_remove_disk(
5983 mddev, rdev->raid_disk)==0) {
5985 sprintf(nm,"rd%d", rdev->raid_disk);
5986 sysfs_remove_link(&mddev->kobj, nm);
5987 rdev->raid_disk = -1;
5991 if (mddev->degraded) {
5992 rdev_for_each(rdev, rtmp, mddev) {
5993 if (rdev->raid_disk >= 0 &&
5994 !test_bit(In_sync, &rdev->flags))
5996 if (rdev->raid_disk < 0
5997 && !test_bit(Faulty, &rdev->flags)) {
5998 rdev->recovery_offset = 0;
6000 hot_add_disk(mddev, rdev) == 0) {
6002 sprintf(nm, "rd%d", rdev->raid_disk);
6003 if (sysfs_create_link(&mddev->kobj,
6006 "md: cannot register "
6010 md_new_event(mddev);
6019 * This routine is regularly called by all per-raid-array threads to
6020 * deal with generic issues like resync and super-block update.
6021 * Raid personalities that don't have a thread (linear/raid0) do not
6022 * need this as they never do any recovery or update the superblock.
6024 * It does not do any resync itself, but rather "forks" off other threads
6025 * to do that as needed.
6026 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6027 * "->recovery" and create a thread at ->sync_thread.
6028 * When the thread finishes it sets MD_RECOVERY_DONE
6029 * and wakeups up this thread which will reap the thread and finish up.
6030 * This thread also removes any faulty devices (with nr_pending == 0).
6032 * The overall approach is:
6033 * 1/ if the superblock needs updating, update it.
6034 * 2/ If a recovery thread is running, don't do anything else.
6035 * 3/ If recovery has finished, clean up, possibly marking spares active.
6036 * 4/ If there are any faulty devices, remove them.
6037 * 5/ If array is degraded, try to add spares devices
6038 * 6/ If array has spares or is not in-sync, start a resync thread.
6040 void md_check_recovery(mddev_t *mddev)
6043 struct list_head *rtmp;
6047 bitmap_daemon_work(mddev->bitmap);
6052 if (signal_pending(current)) {
6053 if (mddev->pers->sync_request && !mddev->external) {
6054 printk(KERN_INFO "md: %s in immediate safe mode\n",
6056 mddev->safemode = 2;
6058 flush_signals(current);
6062 (mddev->flags && !mddev->external) ||
6063 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6064 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6065 (mddev->external == 0 && mddev->safemode == 1) ||
6066 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6067 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6071 if (mddev_trylock(mddev)) {
6074 if (!mddev->external) {
6076 spin_lock_irq(&mddev->write_lock);
6077 if (mddev->safemode &&
6078 !atomic_read(&mddev->writes_pending) &&
6080 mddev->recovery_cp == MaxSector) {
6083 if (mddev->persistent)
6084 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6086 if (mddev->safemode == 1)
6087 mddev->safemode = 0;
6088 spin_unlock_irq(&mddev->write_lock);
6090 sysfs_notify(&mddev->kobj, NULL, "array_state");
6094 md_update_sb(mddev, 0);
6096 rdev_for_each(rdev, rtmp, mddev)
6097 if (test_and_clear_bit(StateChanged, &rdev->flags))
6098 sysfs_notify(&rdev->kobj, NULL, "state");
6101 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6102 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6103 /* resync/recovery still happening */
6104 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6107 if (mddev->sync_thread) {
6108 /* resync has finished, collect result */
6109 md_unregister_thread(mddev->sync_thread);
6110 mddev->sync_thread = NULL;
6111 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6113 /* activate any spares */
6114 if (mddev->pers->spare_active(mddev))
6115 sysfs_notify(&mddev->kobj, NULL,
6118 md_update_sb(mddev, 1);
6120 /* if array is no-longer degraded, then any saved_raid_disk
6121 * information must be scrapped
6123 if (!mddev->degraded)
6124 rdev_for_each(rdev, rtmp, mddev)
6125 rdev->saved_raid_disk = -1;
6127 mddev->recovery = 0;
6128 /* flag recovery needed just to double check */
6129 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6130 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6131 md_new_event(mddev);
6134 /* Set RUNNING before clearing NEEDED to avoid
6135 * any transients in the value of "sync_action".
6137 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6138 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6139 /* Clear some bits that don't mean anything, but
6142 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6143 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6145 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6147 /* no recovery is running.
6148 * remove any failed drives, then
6149 * add spares if possible.
6150 * Spare are also removed and re-added, to allow
6151 * the personality to fail the re-add.
6154 if (mddev->reshape_position != MaxSector) {
6155 if (mddev->pers->check_reshape(mddev) != 0)
6156 /* Cannot proceed */
6158 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6159 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6160 } else if ((spares = remove_and_add_spares(mddev))) {
6161 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6162 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6163 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6164 } else if (mddev->recovery_cp < MaxSector) {
6165 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6166 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6167 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6168 /* nothing to be done ... */
6171 if (mddev->pers->sync_request) {
6172 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6173 /* We are adding a device or devices to an array
6174 * which has the bitmap stored on all devices.
6175 * So make sure all bitmap pages get written
6177 bitmap_write_all(mddev->bitmap);
6179 mddev->sync_thread = md_register_thread(md_do_sync,
6182 if (!mddev->sync_thread) {
6183 printk(KERN_ERR "%s: could not start resync"
6186 /* leave the spares where they are, it shouldn't hurt */
6187 mddev->recovery = 0;
6189 md_wakeup_thread(mddev->sync_thread);
6190 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6191 md_new_event(mddev);
6194 if (!mddev->sync_thread) {
6195 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6196 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6198 sysfs_notify(&mddev->kobj, NULL, "sync_action");
6200 mddev_unlock(mddev);
6204 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6206 sysfs_notify(&rdev->kobj, NULL, "state");
6207 wait_event_timeout(rdev->blocked_wait,
6208 !test_bit(Blocked, &rdev->flags),
6209 msecs_to_jiffies(5000));
6210 rdev_dec_pending(rdev, mddev);
6212 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6214 static int md_notify_reboot(struct notifier_block *this,
6215 unsigned long code, void *x)
6217 struct list_head *tmp;
6220 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6222 printk(KERN_INFO "md: stopping all md devices.\n");
6224 for_each_mddev(mddev, tmp)
6225 if (mddev_trylock(mddev)) {
6226 do_md_stop (mddev, 1, 0);
6227 mddev_unlock(mddev);
6230 * certain more exotic SCSI devices are known to be
6231 * volatile wrt too early system reboots. While the
6232 * right place to handle this issue is the given
6233 * driver, we do want to have a safe RAID driver ...
6240 static struct notifier_block md_notifier = {
6241 .notifier_call = md_notify_reboot,
6243 .priority = INT_MAX, /* before any real devices */
6246 static void md_geninit(void)
6248 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6250 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6253 static int __init md_init(void)
6255 if (register_blkdev(MAJOR_NR, "md"))
6257 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6258 unregister_blkdev(MAJOR_NR, "md");
6261 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6262 md_probe, NULL, NULL);
6263 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6264 md_probe, NULL, NULL);
6266 register_reboot_notifier(&md_notifier);
6267 raid_table_header = register_sysctl_table(raid_root_table);
6277 * Searches all registered partitions for autorun RAID arrays
6281 static LIST_HEAD(all_detected_devices);
6282 struct detected_devices_node {
6283 struct list_head list;
6287 void md_autodetect_dev(dev_t dev)
6289 struct detected_devices_node *node_detected_dev;
6291 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6292 if (node_detected_dev) {
6293 node_detected_dev->dev = dev;
6294 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6296 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6297 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6302 static void autostart_arrays(int part)
6305 struct detected_devices_node *node_detected_dev;
6307 int i_scanned, i_passed;
6312 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6314 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6316 node_detected_dev = list_entry(all_detected_devices.next,
6317 struct detected_devices_node, list);
6318 list_del(&node_detected_dev->list);
6319 dev = node_detected_dev->dev;
6320 kfree(node_detected_dev);
6321 rdev = md_import_device(dev,0, 90);
6325 if (test_bit(Faulty, &rdev->flags)) {
6329 set_bit(AutoDetected, &rdev->flags);
6330 list_add(&rdev->same_set, &pending_raid_disks);
6334 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6335 i_scanned, i_passed);
6337 autorun_devices(part);
6340 #endif /* !MODULE */
6342 static __exit void md_exit(void)
6345 struct list_head *tmp;
6347 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6348 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6350 unregister_blkdev(MAJOR_NR,"md");
6351 unregister_blkdev(mdp_major, "mdp");
6352 unregister_reboot_notifier(&md_notifier);
6353 unregister_sysctl_table(raid_table_header);
6354 remove_proc_entry("mdstat", NULL);
6355 for_each_mddev(mddev, tmp) {
6356 struct gendisk *disk = mddev->gendisk;
6359 export_array(mddev);
6362 mddev->gendisk = NULL;
6367 subsys_initcall(md_init);
6368 module_exit(md_exit)
6370 static int get_ro(char *buffer, struct kernel_param *kp)
6372 return sprintf(buffer, "%d", start_readonly);
6374 static int set_ro(const char *val, struct kernel_param *kp)
6377 int num = simple_strtoul(val, &e, 10);
6378 if (*val && (*e == '\0' || *e == '\n')) {
6379 start_readonly = num;
6385 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6386 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6389 EXPORT_SYMBOL(register_md_personality);
6390 EXPORT_SYMBOL(unregister_md_personality);
6391 EXPORT_SYMBOL(md_error);
6392 EXPORT_SYMBOL(md_done_sync);
6393 EXPORT_SYMBOL(md_write_start);
6394 EXPORT_SYMBOL(md_write_end);
6395 EXPORT_SYMBOL(md_register_thread);
6396 EXPORT_SYMBOL(md_unregister_thread);
6397 EXPORT_SYMBOL(md_wakeup_thread);
6398 EXPORT_SYMBOL(md_check_recovery);
6399 MODULE_LICENSE("GPL");
6401 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);