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 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
100 static inline int speed_max(mddev_t *mddev)
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
106 static struct ctl_table_header *raid_table_header;
108 static ctl_table raid_table[] = {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = &proc_dointvec,
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = &proc_dointvec,
128 static ctl_table raid_dir_table[] = {
130 .ctl_name = DEV_RAID,
133 .mode = S_IRUGO|S_IXUGO,
139 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static struct block_device_operations md_fops;
152 static int start_readonly;
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
172 EXPORT_SYMBOL_GPL(md_new_event);
174 /* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
177 static void md_new_event_inintr(mddev_t *mddev)
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
198 #define for_each_mddev(mddev,tmp) \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
214 static int md_fail_request (struct request_queue *q, struct bio *bio)
220 static inline mddev_t *mddev_get(mddev_t *mddev)
222 atomic_inc(&mddev->active);
226 static void mddev_put(mddev_t *mddev)
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
232 spin_unlock(&all_mddevs_lock);
233 blk_cleanup_queue(mddev->queue);
234 kobject_put(&mddev->kobj);
236 spin_unlock(&all_mddevs_lock);
239 static mddev_t * mddev_find(dev_t unit)
241 mddev_t *mddev, *new = NULL;
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
248 spin_unlock(&all_mddevs_lock);
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
258 spin_unlock(&all_mddevs_lock);
260 new = kzalloc(sizeof(*new), GFP_KERNEL);
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
270 mutex_init(&new->reconfig_mutex);
271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
275 spin_lock_init(&new->write_lock);
276 init_waitqueue_head(&new->sb_wait);
277 new->reshape_position = MaxSector;
278 new->resync_max = MaxSector;
279 new->level = LEVEL_NONE;
281 new->queue = blk_alloc_queue(GFP_KERNEL);
286 /* Can be unlocked because the queue is new: no concurrency */
287 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
289 blk_queue_make_request(new->queue, md_fail_request);
294 static inline int mddev_lock(mddev_t * mddev)
296 return mutex_lock_interruptible(&mddev->reconfig_mutex);
299 static inline int mddev_trylock(mddev_t * mddev)
301 return mutex_trylock(&mddev->reconfig_mutex);
304 static inline void mddev_unlock(mddev_t * mddev)
306 mutex_unlock(&mddev->reconfig_mutex);
308 md_wakeup_thread(mddev->thread);
311 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
314 struct list_head *tmp;
316 rdev_for_each(rdev, tmp, mddev) {
317 if (rdev->desc_nr == nr)
323 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
325 struct list_head *tmp;
328 rdev_for_each(rdev, tmp, mddev) {
329 if (rdev->bdev->bd_dev == dev)
335 static struct mdk_personality *find_pers(int level, char *clevel)
337 struct mdk_personality *pers;
338 list_for_each_entry(pers, &pers_list, list) {
339 if (level != LEVEL_NONE && pers->level == level)
341 if (strcmp(pers->name, clevel)==0)
347 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
349 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
350 return MD_NEW_SIZE_BLOCKS(size);
353 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
357 size = rdev->sb_offset;
360 size &= ~((sector_t)chunk_size/1024 - 1);
364 static int alloc_disk_sb(mdk_rdev_t * rdev)
369 rdev->sb_page = alloc_page(GFP_KERNEL);
370 if (!rdev->sb_page) {
371 printk(KERN_ALERT "md: out of memory.\n");
378 static void free_disk_sb(mdk_rdev_t * rdev)
381 put_page(rdev->sb_page);
383 rdev->sb_page = NULL;
390 static void super_written(struct bio *bio, int error)
392 mdk_rdev_t *rdev = bio->bi_private;
393 mddev_t *mddev = rdev->mddev;
395 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
396 printk("md: super_written gets error=%d, uptodate=%d\n",
397 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
398 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
399 md_error(mddev, rdev);
402 if (atomic_dec_and_test(&mddev->pending_writes))
403 wake_up(&mddev->sb_wait);
407 static void super_written_barrier(struct bio *bio, int error)
409 struct bio *bio2 = bio->bi_private;
410 mdk_rdev_t *rdev = bio2->bi_private;
411 mddev_t *mddev = rdev->mddev;
413 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
414 error == -EOPNOTSUPP) {
416 /* barriers don't appear to be supported :-( */
417 set_bit(BarriersNotsupp, &rdev->flags);
418 mddev->barriers_work = 0;
419 spin_lock_irqsave(&mddev->write_lock, flags);
420 bio2->bi_next = mddev->biolist;
421 mddev->biolist = bio2;
422 spin_unlock_irqrestore(&mddev->write_lock, flags);
423 wake_up(&mddev->sb_wait);
427 bio->bi_private = rdev;
428 super_written(bio, error);
432 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
433 sector_t sector, int size, struct page *page)
435 /* write first size bytes of page to sector of rdev
436 * Increment mddev->pending_writes before returning
437 * and decrement it on completion, waking up sb_wait
438 * if zero is reached.
439 * If an error occurred, call md_error
441 * As we might need to resubmit the request if BIO_RW_BARRIER
442 * causes ENOTSUPP, we allocate a spare bio...
444 struct bio *bio = bio_alloc(GFP_NOIO, 1);
445 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
447 bio->bi_bdev = rdev->bdev;
448 bio->bi_sector = sector;
449 bio_add_page(bio, page, size, 0);
450 bio->bi_private = rdev;
451 bio->bi_end_io = super_written;
454 atomic_inc(&mddev->pending_writes);
455 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
457 rw |= (1<<BIO_RW_BARRIER);
458 rbio = bio_clone(bio, GFP_NOIO);
459 rbio->bi_private = bio;
460 rbio->bi_end_io = super_written_barrier;
461 submit_bio(rw, rbio);
466 void md_super_wait(mddev_t *mddev)
468 /* wait for all superblock writes that were scheduled to complete.
469 * if any had to be retried (due to BARRIER problems), retry them
473 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
474 if (atomic_read(&mddev->pending_writes)==0)
476 while (mddev->biolist) {
478 spin_lock_irq(&mddev->write_lock);
479 bio = mddev->biolist;
480 mddev->biolist = bio->bi_next ;
482 spin_unlock_irq(&mddev->write_lock);
483 submit_bio(bio->bi_rw, bio);
487 finish_wait(&mddev->sb_wait, &wq);
490 static void bi_complete(struct bio *bio, int error)
492 complete((struct completion*)bio->bi_private);
495 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
496 struct page *page, int rw)
498 struct bio *bio = bio_alloc(GFP_NOIO, 1);
499 struct completion event;
502 rw |= (1 << BIO_RW_SYNC);
505 bio->bi_sector = sector;
506 bio_add_page(bio, page, size, 0);
507 init_completion(&event);
508 bio->bi_private = &event;
509 bio->bi_end_io = bi_complete;
511 wait_for_completion(&event);
513 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
517 EXPORT_SYMBOL_GPL(sync_page_io);
519 static int read_disk_sb(mdk_rdev_t * rdev, int size)
521 char b[BDEVNAME_SIZE];
522 if (!rdev->sb_page) {
530 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
536 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
537 bdevname(rdev->bdev,b));
541 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
543 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
544 (sb1->set_uuid1 == sb2->set_uuid1) &&
545 (sb1->set_uuid2 == sb2->set_uuid2) &&
546 (sb1->set_uuid3 == sb2->set_uuid3))
554 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
557 mdp_super_t *tmp1, *tmp2;
559 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
560 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
562 if (!tmp1 || !tmp2) {
564 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
572 * nr_disks is not constant
577 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
589 static u32 md_csum_fold(u32 csum)
591 csum = (csum & 0xffff) + (csum >> 16);
592 return (csum & 0xffff) + (csum >> 16);
595 static unsigned int calc_sb_csum(mdp_super_t * sb)
598 u32 *sb32 = (u32*)sb;
600 unsigned int disk_csum, csum;
602 disk_csum = sb->sb_csum;
605 for (i = 0; i < MD_SB_BYTES/4 ; i++)
607 csum = (newcsum & 0xffffffff) + (newcsum>>32);
611 /* This used to use csum_partial, which was wrong for several
612 * reasons including that different results are returned on
613 * different architectures. It isn't critical that we get exactly
614 * the same return value as before (we always csum_fold before
615 * testing, and that removes any differences). However as we
616 * know that csum_partial always returned a 16bit value on
617 * alphas, do a fold to maximise conformity to previous behaviour.
619 sb->sb_csum = md_csum_fold(disk_csum);
621 sb->sb_csum = disk_csum;
628 * Handle superblock details.
629 * We want to be able to handle multiple superblock formats
630 * so we have a common interface to them all, and an array of
631 * different handlers.
632 * We rely on user-space to write the initial superblock, and support
633 * reading and updating of superblocks.
634 * Interface methods are:
635 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
636 * loads and validates a superblock on dev.
637 * if refdev != NULL, compare superblocks on both devices
639 * 0 - dev has a superblock that is compatible with refdev
640 * 1 - dev has a superblock that is compatible and newer than refdev
641 * so dev should be used as the refdev in future
642 * -EINVAL superblock incompatible or invalid
643 * -othererror e.g. -EIO
645 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
646 * Verify that dev is acceptable into mddev.
647 * The first time, mddev->raid_disks will be 0, and data from
648 * dev should be merged in. Subsequent calls check that dev
649 * is new enough. Return 0 or -EINVAL
651 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
652 * Update the superblock for rdev with data in mddev
653 * This does not write to disc.
659 struct module *owner;
660 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
661 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
662 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
666 * load_super for 0.90.0
668 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
670 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
676 * Calculate the position of the superblock,
677 * it's at the end of the disk.
679 * It also happens to be a multiple of 4Kb.
681 sb_offset = calc_dev_sboffset(rdev->bdev);
682 rdev->sb_offset = sb_offset;
684 ret = read_disk_sb(rdev, MD_SB_BYTES);
689 bdevname(rdev->bdev, b);
690 sb = (mdp_super_t*)page_address(rdev->sb_page);
692 if (sb->md_magic != MD_SB_MAGIC) {
693 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
698 if (sb->major_version != 0 ||
699 sb->minor_version < 90 ||
700 sb->minor_version > 91) {
701 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
702 sb->major_version, sb->minor_version,
707 if (sb->raid_disks <= 0)
710 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
711 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
716 rdev->preferred_minor = sb->md_minor;
717 rdev->data_offset = 0;
718 rdev->sb_size = MD_SB_BYTES;
720 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
721 if (sb->level != 1 && sb->level != 4
722 && sb->level != 5 && sb->level != 6
723 && sb->level != 10) {
724 /* FIXME use a better test */
726 "md: bitmaps not supported for this level.\n");
731 if (sb->level == LEVEL_MULTIPATH)
734 rdev->desc_nr = sb->this_disk.number;
740 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
741 if (!uuid_equal(refsb, sb)) {
742 printk(KERN_WARNING "md: %s has different UUID to %s\n",
743 b, bdevname(refdev->bdev,b2));
746 if (!sb_equal(refsb, sb)) {
747 printk(KERN_WARNING "md: %s has same UUID"
748 " but different superblock to %s\n",
749 b, bdevname(refdev->bdev, b2));
753 ev2 = md_event(refsb);
759 rdev->size = calc_dev_size(rdev, sb->chunk_size);
761 if (rdev->size < sb->size && sb->level > 1)
762 /* "this cannot possibly happen" ... */
770 * validate_super for 0.90.0
772 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
775 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
776 __u64 ev1 = md_event(sb);
778 rdev->raid_disk = -1;
779 clear_bit(Faulty, &rdev->flags);
780 clear_bit(In_sync, &rdev->flags);
781 clear_bit(WriteMostly, &rdev->flags);
782 clear_bit(BarriersNotsupp, &rdev->flags);
784 if (mddev->raid_disks == 0) {
785 mddev->major_version = 0;
786 mddev->minor_version = sb->minor_version;
787 mddev->patch_version = sb->patch_version;
789 mddev->chunk_size = sb->chunk_size;
790 mddev->ctime = sb->ctime;
791 mddev->utime = sb->utime;
792 mddev->level = sb->level;
793 mddev->clevel[0] = 0;
794 mddev->layout = sb->layout;
795 mddev->raid_disks = sb->raid_disks;
796 mddev->size = sb->size;
798 mddev->bitmap_offset = 0;
799 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
801 if (mddev->minor_version >= 91) {
802 mddev->reshape_position = sb->reshape_position;
803 mddev->delta_disks = sb->delta_disks;
804 mddev->new_level = sb->new_level;
805 mddev->new_layout = sb->new_layout;
806 mddev->new_chunk = sb->new_chunk;
808 mddev->reshape_position = MaxSector;
809 mddev->delta_disks = 0;
810 mddev->new_level = mddev->level;
811 mddev->new_layout = mddev->layout;
812 mddev->new_chunk = mddev->chunk_size;
815 if (sb->state & (1<<MD_SB_CLEAN))
816 mddev->recovery_cp = MaxSector;
818 if (sb->events_hi == sb->cp_events_hi &&
819 sb->events_lo == sb->cp_events_lo) {
820 mddev->recovery_cp = sb->recovery_cp;
822 mddev->recovery_cp = 0;
825 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
826 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
827 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
828 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
830 mddev->max_disks = MD_SB_DISKS;
832 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
833 mddev->bitmap_file == NULL)
834 mddev->bitmap_offset = mddev->default_bitmap_offset;
836 } else if (mddev->pers == NULL) {
837 /* Insist on good event counter while assembling */
839 if (ev1 < mddev->events)
841 } else if (mddev->bitmap) {
842 /* if adding to array with a bitmap, then we can accept an
843 * older device ... but not too old.
845 if (ev1 < mddev->bitmap->events_cleared)
848 if (ev1 < mddev->events)
849 /* just a hot-add of a new device, leave raid_disk at -1 */
853 if (mddev->level != LEVEL_MULTIPATH) {
854 desc = sb->disks + rdev->desc_nr;
856 if (desc->state & (1<<MD_DISK_FAULTY))
857 set_bit(Faulty, &rdev->flags);
858 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
859 desc->raid_disk < mddev->raid_disks */) {
860 set_bit(In_sync, &rdev->flags);
861 rdev->raid_disk = desc->raid_disk;
863 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
864 set_bit(WriteMostly, &rdev->flags);
865 } else /* MULTIPATH are always insync */
866 set_bit(In_sync, &rdev->flags);
871 * sync_super for 0.90.0
873 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
876 struct list_head *tmp;
878 int next_spare = mddev->raid_disks;
881 /* make rdev->sb match mddev data..
884 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
885 * 3/ any empty disks < next_spare become removed
887 * disks[0] gets initialised to REMOVED because
888 * we cannot be sure from other fields if it has
889 * been initialised or not.
892 int active=0, working=0,failed=0,spare=0,nr_disks=0;
894 rdev->sb_size = MD_SB_BYTES;
896 sb = (mdp_super_t*)page_address(rdev->sb_page);
898 memset(sb, 0, sizeof(*sb));
900 sb->md_magic = MD_SB_MAGIC;
901 sb->major_version = mddev->major_version;
902 sb->patch_version = mddev->patch_version;
903 sb->gvalid_words = 0; /* ignored */
904 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
905 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
906 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
907 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
909 sb->ctime = mddev->ctime;
910 sb->level = mddev->level;
911 sb->size = mddev->size;
912 sb->raid_disks = mddev->raid_disks;
913 sb->md_minor = mddev->md_minor;
914 sb->not_persistent = 0;
915 sb->utime = mddev->utime;
917 sb->events_hi = (mddev->events>>32);
918 sb->events_lo = (u32)mddev->events;
920 if (mddev->reshape_position == MaxSector)
921 sb->minor_version = 90;
923 sb->minor_version = 91;
924 sb->reshape_position = mddev->reshape_position;
925 sb->new_level = mddev->new_level;
926 sb->delta_disks = mddev->delta_disks;
927 sb->new_layout = mddev->new_layout;
928 sb->new_chunk = mddev->new_chunk;
930 mddev->minor_version = sb->minor_version;
933 sb->recovery_cp = mddev->recovery_cp;
934 sb->cp_events_hi = (mddev->events>>32);
935 sb->cp_events_lo = (u32)mddev->events;
936 if (mddev->recovery_cp == MaxSector)
937 sb->state = (1<< MD_SB_CLEAN);
941 sb->layout = mddev->layout;
942 sb->chunk_size = mddev->chunk_size;
944 if (mddev->bitmap && mddev->bitmap_file == NULL)
945 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
947 sb->disks[0].state = (1<<MD_DISK_REMOVED);
948 rdev_for_each(rdev2, tmp, mddev) {
951 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
952 && !test_bit(Faulty, &rdev2->flags))
953 desc_nr = rdev2->raid_disk;
955 desc_nr = next_spare++;
956 rdev2->desc_nr = desc_nr;
957 d = &sb->disks[rdev2->desc_nr];
959 d->number = rdev2->desc_nr;
960 d->major = MAJOR(rdev2->bdev->bd_dev);
961 d->minor = MINOR(rdev2->bdev->bd_dev);
962 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
963 && !test_bit(Faulty, &rdev2->flags))
964 d->raid_disk = rdev2->raid_disk;
966 d->raid_disk = rdev2->desc_nr; /* compatibility */
967 if (test_bit(Faulty, &rdev2->flags))
968 d->state = (1<<MD_DISK_FAULTY);
969 else if (test_bit(In_sync, &rdev2->flags)) {
970 d->state = (1<<MD_DISK_ACTIVE);
971 d->state |= (1<<MD_DISK_SYNC);
979 if (test_bit(WriteMostly, &rdev2->flags))
980 d->state |= (1<<MD_DISK_WRITEMOSTLY);
982 /* now set the "removed" and "faulty" bits on any missing devices */
983 for (i=0 ; i < mddev->raid_disks ; i++) {
984 mdp_disk_t *d = &sb->disks[i];
985 if (d->state == 0 && d->number == 0) {
988 d->state = (1<<MD_DISK_REMOVED);
989 d->state |= (1<<MD_DISK_FAULTY);
993 sb->nr_disks = nr_disks;
994 sb->active_disks = active;
995 sb->working_disks = working;
996 sb->failed_disks = failed;
997 sb->spare_disks = spare;
999 sb->this_disk = sb->disks[rdev->desc_nr];
1000 sb->sb_csum = calc_sb_csum(sb);
1004 * version 1 superblock
1007 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1011 unsigned long long newcsum;
1012 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1013 __le32 *isuper = (__le32*)sb;
1016 disk_csum = sb->sb_csum;
1019 for (i=0; size>=4; size -= 4 )
1020 newcsum += le32_to_cpu(*isuper++);
1023 newcsum += le16_to_cpu(*(__le16*) isuper);
1025 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1026 sb->sb_csum = disk_csum;
1027 return cpu_to_le32(csum);
1030 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1032 struct mdp_superblock_1 *sb;
1035 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1039 * Calculate the position of the superblock.
1040 * It is always aligned to a 4K boundary and
1041 * depeding on minor_version, it can be:
1042 * 0: At least 8K, but less than 12K, from end of device
1043 * 1: At start of device
1044 * 2: 4K from start of device.
1046 switch(minor_version) {
1048 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1050 sb_offset &= ~(sector_t)(4*2-1);
1051 /* convert from sectors to K */
1063 rdev->sb_offset = sb_offset;
1065 /* superblock is rarely larger than 1K, but it can be larger,
1066 * and it is safe to read 4k, so we do that
1068 ret = read_disk_sb(rdev, 4096);
1069 if (ret) return ret;
1072 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1074 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1075 sb->major_version != cpu_to_le32(1) ||
1076 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1077 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1078 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1081 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1082 printk("md: invalid superblock checksum on %s\n",
1083 bdevname(rdev->bdev,b));
1086 if (le64_to_cpu(sb->data_size) < 10) {
1087 printk("md: data_size too small on %s\n",
1088 bdevname(rdev->bdev,b));
1091 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1092 if (sb->level != cpu_to_le32(1) &&
1093 sb->level != cpu_to_le32(4) &&
1094 sb->level != cpu_to_le32(5) &&
1095 sb->level != cpu_to_le32(6) &&
1096 sb->level != cpu_to_le32(10)) {
1098 "md: bitmaps not supported for this level.\n");
1103 rdev->preferred_minor = 0xffff;
1104 rdev->data_offset = le64_to_cpu(sb->data_offset);
1105 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1107 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1108 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1109 if (rdev->sb_size & bmask)
1110 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1113 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1116 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1119 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1125 struct mdp_superblock_1 *refsb =
1126 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1128 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1129 sb->level != refsb->level ||
1130 sb->layout != refsb->layout ||
1131 sb->chunksize != refsb->chunksize) {
1132 printk(KERN_WARNING "md: %s has strangely different"
1133 " superblock to %s\n",
1134 bdevname(rdev->bdev,b),
1135 bdevname(refdev->bdev,b2));
1138 ev1 = le64_to_cpu(sb->events);
1139 ev2 = le64_to_cpu(refsb->events);
1147 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1149 rdev->size = rdev->sb_offset;
1150 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1152 rdev->size = le64_to_cpu(sb->data_size)/2;
1153 if (le32_to_cpu(sb->chunksize))
1154 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1156 if (le64_to_cpu(sb->size) > rdev->size*2)
1161 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1163 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1164 __u64 ev1 = le64_to_cpu(sb->events);
1166 rdev->raid_disk = -1;
1167 clear_bit(Faulty, &rdev->flags);
1168 clear_bit(In_sync, &rdev->flags);
1169 clear_bit(WriteMostly, &rdev->flags);
1170 clear_bit(BarriersNotsupp, &rdev->flags);
1172 if (mddev->raid_disks == 0) {
1173 mddev->major_version = 1;
1174 mddev->patch_version = 0;
1175 mddev->external = 0;
1176 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1177 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1178 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1179 mddev->level = le32_to_cpu(sb->level);
1180 mddev->clevel[0] = 0;
1181 mddev->layout = le32_to_cpu(sb->layout);
1182 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1183 mddev->size = le64_to_cpu(sb->size)/2;
1184 mddev->events = ev1;
1185 mddev->bitmap_offset = 0;
1186 mddev->default_bitmap_offset = 1024 >> 9;
1188 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1189 memcpy(mddev->uuid, sb->set_uuid, 16);
1191 mddev->max_disks = (4096-256)/2;
1193 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1194 mddev->bitmap_file == NULL )
1195 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1197 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1198 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1199 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1200 mddev->new_level = le32_to_cpu(sb->new_level);
1201 mddev->new_layout = le32_to_cpu(sb->new_layout);
1202 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1204 mddev->reshape_position = MaxSector;
1205 mddev->delta_disks = 0;
1206 mddev->new_level = mddev->level;
1207 mddev->new_layout = mddev->layout;
1208 mddev->new_chunk = mddev->chunk_size;
1211 } else if (mddev->pers == NULL) {
1212 /* Insist of good event counter while assembling */
1214 if (ev1 < mddev->events)
1216 } else if (mddev->bitmap) {
1217 /* If adding to array with a bitmap, then we can accept an
1218 * older device, but not too old.
1220 if (ev1 < mddev->bitmap->events_cleared)
1223 if (ev1 < mddev->events)
1224 /* just a hot-add of a new device, leave raid_disk at -1 */
1227 if (mddev->level != LEVEL_MULTIPATH) {
1229 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1231 case 0xffff: /* spare */
1233 case 0xfffe: /* faulty */
1234 set_bit(Faulty, &rdev->flags);
1237 if ((le32_to_cpu(sb->feature_map) &
1238 MD_FEATURE_RECOVERY_OFFSET))
1239 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1241 set_bit(In_sync, &rdev->flags);
1242 rdev->raid_disk = role;
1245 if (sb->devflags & WriteMostly1)
1246 set_bit(WriteMostly, &rdev->flags);
1247 } else /* MULTIPATH are always insync */
1248 set_bit(In_sync, &rdev->flags);
1253 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1255 struct mdp_superblock_1 *sb;
1256 struct list_head *tmp;
1259 /* make rdev->sb match mddev and rdev data. */
1261 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1263 sb->feature_map = 0;
1265 sb->recovery_offset = cpu_to_le64(0);
1266 memset(sb->pad1, 0, sizeof(sb->pad1));
1267 memset(sb->pad2, 0, sizeof(sb->pad2));
1268 memset(sb->pad3, 0, sizeof(sb->pad3));
1270 sb->utime = cpu_to_le64((__u64)mddev->utime);
1271 sb->events = cpu_to_le64(mddev->events);
1273 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1275 sb->resync_offset = cpu_to_le64(0);
1277 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1279 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1280 sb->size = cpu_to_le64(mddev->size<<1);
1282 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1283 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1284 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1287 if (rdev->raid_disk >= 0 &&
1288 !test_bit(In_sync, &rdev->flags) &&
1289 rdev->recovery_offset > 0) {
1290 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1291 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1294 if (mddev->reshape_position != MaxSector) {
1295 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1296 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1297 sb->new_layout = cpu_to_le32(mddev->new_layout);
1298 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1299 sb->new_level = cpu_to_le32(mddev->new_level);
1300 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1304 rdev_for_each(rdev2, tmp, mddev)
1305 if (rdev2->desc_nr+1 > max_dev)
1306 max_dev = rdev2->desc_nr+1;
1308 if (max_dev > le32_to_cpu(sb->max_dev))
1309 sb->max_dev = cpu_to_le32(max_dev);
1310 for (i=0; i<max_dev;i++)
1311 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1313 rdev_for_each(rdev2, tmp, mddev) {
1315 if (test_bit(Faulty, &rdev2->flags))
1316 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1317 else if (test_bit(In_sync, &rdev2->flags))
1318 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1319 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1320 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1322 sb->dev_roles[i] = cpu_to_le16(0xffff);
1325 sb->sb_csum = calc_sb_1_csum(sb);
1329 static struct super_type super_types[] = {
1332 .owner = THIS_MODULE,
1333 .load_super = super_90_load,
1334 .validate_super = super_90_validate,
1335 .sync_super = super_90_sync,
1339 .owner = THIS_MODULE,
1340 .load_super = super_1_load,
1341 .validate_super = super_1_validate,
1342 .sync_super = super_1_sync,
1346 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1348 struct list_head *tmp, *tmp2;
1349 mdk_rdev_t *rdev, *rdev2;
1351 rdev_for_each(rdev, tmp, mddev1)
1352 rdev_for_each(rdev2, tmp2, mddev2)
1353 if (rdev->bdev->bd_contains ==
1354 rdev2->bdev->bd_contains)
1360 static LIST_HEAD(pending_raid_disks);
1362 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1364 char b[BDEVNAME_SIZE];
1374 /* prevent duplicates */
1375 if (find_rdev(mddev, rdev->bdev->bd_dev))
1378 /* make sure rdev->size exceeds mddev->size */
1379 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1381 /* Cannot change size, so fail
1382 * If mddev->level <= 0, then we don't care
1383 * about aligning sizes (e.g. linear)
1385 if (mddev->level > 0)
1388 mddev->size = rdev->size;
1391 /* Verify rdev->desc_nr is unique.
1392 * If it is -1, assign a free number, else
1393 * check number is not in use
1395 if (rdev->desc_nr < 0) {
1397 if (mddev->pers) choice = mddev->raid_disks;
1398 while (find_rdev_nr(mddev, choice))
1400 rdev->desc_nr = choice;
1402 if (find_rdev_nr(mddev, rdev->desc_nr))
1405 bdevname(rdev->bdev,b);
1406 while ( (s=strchr(b, '/')) != NULL)
1409 rdev->mddev = mddev;
1410 printk(KERN_INFO "md: bind<%s>\n", b);
1412 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1415 if (rdev->bdev->bd_part)
1416 ko = &rdev->bdev->bd_part->dev.kobj;
1418 ko = &rdev->bdev->bd_disk->dev.kobj;
1419 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1420 kobject_del(&rdev->kobj);
1423 list_add(&rdev->same_set, &mddev->disks);
1424 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1428 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1433 static void md_delayed_delete(struct work_struct *ws)
1435 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1436 kobject_del(&rdev->kobj);
1437 kobject_put(&rdev->kobj);
1440 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1442 char b[BDEVNAME_SIZE];
1447 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1448 list_del_init(&rdev->same_set);
1449 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1451 sysfs_remove_link(&rdev->kobj, "block");
1453 /* We need to delay this, otherwise we can deadlock when
1454 * writing to 'remove' to "dev/state"
1456 INIT_WORK(&rdev->del_work, md_delayed_delete);
1457 kobject_get(&rdev->kobj);
1458 schedule_work(&rdev->del_work);
1462 * prevent the device from being mounted, repartitioned or
1463 * otherwise reused by a RAID array (or any other kernel
1464 * subsystem), by bd_claiming the device.
1466 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1469 struct block_device *bdev;
1470 char b[BDEVNAME_SIZE];
1472 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1474 printk(KERN_ERR "md: could not open %s.\n",
1475 __bdevname(dev, b));
1476 return PTR_ERR(bdev);
1478 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1480 printk(KERN_ERR "md: could not bd_claim %s.\n",
1486 set_bit(AllReserved, &rdev->flags);
1491 static void unlock_rdev(mdk_rdev_t *rdev)
1493 struct block_device *bdev = rdev->bdev;
1501 void md_autodetect_dev(dev_t dev);
1503 static void export_rdev(mdk_rdev_t * rdev)
1505 char b[BDEVNAME_SIZE];
1506 printk(KERN_INFO "md: export_rdev(%s)\n",
1507 bdevname(rdev->bdev,b));
1511 list_del_init(&rdev->same_set);
1513 if (test_bit(AutoDetected, &rdev->flags))
1514 md_autodetect_dev(rdev->bdev->bd_dev);
1517 kobject_put(&rdev->kobj);
1520 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1522 unbind_rdev_from_array(rdev);
1526 static void export_array(mddev_t *mddev)
1528 struct list_head *tmp;
1531 rdev_for_each(rdev, tmp, mddev) {
1536 kick_rdev_from_array(rdev);
1538 if (!list_empty(&mddev->disks))
1540 mddev->raid_disks = 0;
1541 mddev->major_version = 0;
1544 static void print_desc(mdp_disk_t *desc)
1546 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1547 desc->major,desc->minor,desc->raid_disk,desc->state);
1550 static void print_sb(mdp_super_t *sb)
1555 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1556 sb->major_version, sb->minor_version, sb->patch_version,
1557 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1559 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1560 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1561 sb->md_minor, sb->layout, sb->chunk_size);
1562 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1563 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1564 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1565 sb->failed_disks, sb->spare_disks,
1566 sb->sb_csum, (unsigned long)sb->events_lo);
1569 for (i = 0; i < MD_SB_DISKS; i++) {
1572 desc = sb->disks + i;
1573 if (desc->number || desc->major || desc->minor ||
1574 desc->raid_disk || (desc->state && (desc->state != 4))) {
1575 printk(" D %2d: ", i);
1579 printk(KERN_INFO "md: THIS: ");
1580 print_desc(&sb->this_disk);
1584 static void print_rdev(mdk_rdev_t *rdev)
1586 char b[BDEVNAME_SIZE];
1587 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1588 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1589 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1591 if (rdev->sb_loaded) {
1592 printk(KERN_INFO "md: rdev superblock:\n");
1593 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1595 printk(KERN_INFO "md: no rdev superblock!\n");
1598 static void md_print_devices(void)
1600 struct list_head *tmp, *tmp2;
1603 char b[BDEVNAME_SIZE];
1606 printk("md: **********************************\n");
1607 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1608 printk("md: **********************************\n");
1609 for_each_mddev(mddev, tmp) {
1612 bitmap_print_sb(mddev->bitmap);
1614 printk("%s: ", mdname(mddev));
1615 rdev_for_each(rdev, tmp2, mddev)
1616 printk("<%s>", bdevname(rdev->bdev,b));
1619 rdev_for_each(rdev, tmp2, mddev)
1622 printk("md: **********************************\n");
1627 static void sync_sbs(mddev_t * mddev, int nospares)
1629 /* Update each superblock (in-memory image), but
1630 * if we are allowed to, skip spares which already
1631 * have the right event counter, or have one earlier
1632 * (which would mean they aren't being marked as dirty
1633 * with the rest of the array)
1636 struct list_head *tmp;
1638 rdev_for_each(rdev, tmp, mddev) {
1639 if (rdev->sb_events == mddev->events ||
1641 rdev->raid_disk < 0 &&
1642 (rdev->sb_events&1)==0 &&
1643 rdev->sb_events+1 == mddev->events)) {
1644 /* Don't update this superblock */
1645 rdev->sb_loaded = 2;
1647 super_types[mddev->major_version].
1648 sync_super(mddev, rdev);
1649 rdev->sb_loaded = 1;
1654 static void md_update_sb(mddev_t * mddev, int force_change)
1656 struct list_head *tmp;
1661 if (mddev->external)
1664 spin_lock_irq(&mddev->write_lock);
1666 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1667 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1669 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1670 /* just a clean<-> dirty transition, possibly leave spares alone,
1671 * though if events isn't the right even/odd, we will have to do
1677 if (mddev->degraded)
1678 /* If the array is degraded, then skipping spares is both
1679 * dangerous and fairly pointless.
1680 * Dangerous because a device that was removed from the array
1681 * might have a event_count that still looks up-to-date,
1682 * so it can be re-added without a resync.
1683 * Pointless because if there are any spares to skip,
1684 * then a recovery will happen and soon that array won't
1685 * be degraded any more and the spare can go back to sleep then.
1689 sync_req = mddev->in_sync;
1690 mddev->utime = get_seconds();
1692 /* If this is just a dirty<->clean transition, and the array is clean
1693 * and 'events' is odd, we can roll back to the previous clean state */
1695 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1696 && (mddev->events & 1)
1697 && mddev->events != 1)
1700 /* otherwise we have to go forward and ... */
1702 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1703 /* .. if the array isn't clean, insist on an odd 'events' */
1704 if ((mddev->events&1)==0) {
1709 /* otherwise insist on an even 'events' (for clean states) */
1710 if ((mddev->events&1)) {
1717 if (!mddev->events) {
1719 * oops, this 64-bit counter should never wrap.
1720 * Either we are in around ~1 trillion A.C., assuming
1721 * 1 reboot per second, or we have a bug:
1728 * do not write anything to disk if using
1729 * nonpersistent superblocks
1731 if (!mddev->persistent) {
1732 if (!mddev->external)
1733 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1735 spin_unlock_irq(&mddev->write_lock);
1736 wake_up(&mddev->sb_wait);
1739 sync_sbs(mddev, nospares);
1740 spin_unlock_irq(&mddev->write_lock);
1743 "md: updating %s RAID superblock on device (in sync %d)\n",
1744 mdname(mddev),mddev->in_sync);
1746 bitmap_update_sb(mddev->bitmap);
1747 rdev_for_each(rdev, tmp, mddev) {
1748 char b[BDEVNAME_SIZE];
1749 dprintk(KERN_INFO "md: ");
1750 if (rdev->sb_loaded != 1)
1751 continue; /* no noise on spare devices */
1752 if (test_bit(Faulty, &rdev->flags))
1753 dprintk("(skipping faulty ");
1755 dprintk("%s ", bdevname(rdev->bdev,b));
1756 if (!test_bit(Faulty, &rdev->flags)) {
1757 md_super_write(mddev,rdev,
1758 rdev->sb_offset<<1, rdev->sb_size,
1760 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1761 bdevname(rdev->bdev,b),
1762 (unsigned long long)rdev->sb_offset);
1763 rdev->sb_events = mddev->events;
1767 if (mddev->level == LEVEL_MULTIPATH)
1768 /* only need to write one superblock... */
1771 md_super_wait(mddev);
1772 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1774 spin_lock_irq(&mddev->write_lock);
1775 if (mddev->in_sync != sync_req ||
1776 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1777 /* have to write it out again */
1778 spin_unlock_irq(&mddev->write_lock);
1781 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1782 spin_unlock_irq(&mddev->write_lock);
1783 wake_up(&mddev->sb_wait);
1787 /* words written to sysfs files may, or my not, be \n terminated.
1788 * We want to accept with case. For this we use cmd_match.
1790 static int cmd_match(const char *cmd, const char *str)
1792 /* See if cmd, written into a sysfs file, matches
1793 * str. They must either be the same, or cmd can
1794 * have a trailing newline
1796 while (*cmd && *str && *cmd == *str) {
1807 struct rdev_sysfs_entry {
1808 struct attribute attr;
1809 ssize_t (*show)(mdk_rdev_t *, char *);
1810 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1814 state_show(mdk_rdev_t *rdev, char *page)
1819 if (test_bit(Faulty, &rdev->flags)) {
1820 len+= sprintf(page+len, "%sfaulty",sep);
1823 if (test_bit(In_sync, &rdev->flags)) {
1824 len += sprintf(page+len, "%sin_sync",sep);
1827 if (test_bit(WriteMostly, &rdev->flags)) {
1828 len += sprintf(page+len, "%swrite_mostly",sep);
1831 if (!test_bit(Faulty, &rdev->flags) &&
1832 !test_bit(In_sync, &rdev->flags)) {
1833 len += sprintf(page+len, "%sspare", sep);
1836 return len+sprintf(page+len, "\n");
1840 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1843 * faulty - simulates and error
1844 * remove - disconnects the device
1845 * writemostly - sets write_mostly
1846 * -writemostly - clears write_mostly
1849 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1850 md_error(rdev->mddev, rdev);
1852 } else if (cmd_match(buf, "remove")) {
1853 if (rdev->raid_disk >= 0)
1856 mddev_t *mddev = rdev->mddev;
1857 kick_rdev_from_array(rdev);
1859 md_update_sb(mddev, 1);
1860 md_new_event(mddev);
1863 } else if (cmd_match(buf, "writemostly")) {
1864 set_bit(WriteMostly, &rdev->flags);
1866 } else if (cmd_match(buf, "-writemostly")) {
1867 clear_bit(WriteMostly, &rdev->flags);
1870 return err ? err : len;
1872 static struct rdev_sysfs_entry rdev_state =
1873 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1876 errors_show(mdk_rdev_t *rdev, char *page)
1878 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1882 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1885 unsigned long n = simple_strtoul(buf, &e, 10);
1886 if (*buf && (*e == 0 || *e == '\n')) {
1887 atomic_set(&rdev->corrected_errors, n);
1892 static struct rdev_sysfs_entry rdev_errors =
1893 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1896 slot_show(mdk_rdev_t *rdev, char *page)
1898 if (rdev->raid_disk < 0)
1899 return sprintf(page, "none\n");
1901 return sprintf(page, "%d\n", rdev->raid_disk);
1905 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1910 int slot = simple_strtoul(buf, &e, 10);
1911 if (strncmp(buf, "none", 4)==0)
1913 else if (e==buf || (*e && *e!= '\n'))
1915 if (rdev->mddev->pers) {
1916 /* Setting 'slot' on an active array requires also
1917 * updating the 'rd%d' link, and communicating
1918 * with the personality with ->hot_*_disk.
1919 * For now we only support removing
1920 * failed/spare devices. This normally happens automatically,
1921 * but not when the metadata is externally managed.
1925 if (rdev->raid_disk == -1)
1927 /* personality does all needed checks */
1928 if (rdev->mddev->pers->hot_add_disk == NULL)
1930 err = rdev->mddev->pers->
1931 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1934 sprintf(nm, "rd%d", rdev->raid_disk);
1935 sysfs_remove_link(&rdev->mddev->kobj, nm);
1936 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1937 md_wakeup_thread(rdev->mddev->thread);
1939 if (slot >= rdev->mddev->raid_disks)
1941 rdev->raid_disk = slot;
1942 /* assume it is working */
1943 clear_bit(Faulty, &rdev->flags);
1944 clear_bit(WriteMostly, &rdev->flags);
1945 set_bit(In_sync, &rdev->flags);
1951 static struct rdev_sysfs_entry rdev_slot =
1952 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1955 offset_show(mdk_rdev_t *rdev, char *page)
1957 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1961 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1964 unsigned long long offset = simple_strtoull(buf, &e, 10);
1965 if (e==buf || (*e && *e != '\n'))
1967 if (rdev->mddev->pers)
1969 if (rdev->size && rdev->mddev->external)
1970 /* Must set offset before size, so overlap checks
1973 rdev->data_offset = offset;
1977 static struct rdev_sysfs_entry rdev_offset =
1978 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1981 rdev_size_show(mdk_rdev_t *rdev, char *page)
1983 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1986 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1988 /* check if two start/length pairs overlap */
1997 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2000 unsigned long long size = simple_strtoull(buf, &e, 10);
2001 unsigned long long oldsize = rdev->size;
2002 mddev_t *my_mddev = rdev->mddev;
2004 if (e==buf || (*e && *e != '\n'))
2009 if (size > oldsize && rdev->mddev->external) {
2010 /* need to check that all other rdevs with the same ->bdev
2011 * do not overlap. We need to unlock the mddev to avoid
2012 * a deadlock. We have already changed rdev->size, and if
2013 * we have to change it back, we will have the lock again.
2017 struct list_head *tmp, *tmp2;
2019 mddev_unlock(my_mddev);
2020 for_each_mddev(mddev, tmp) {
2024 rdev_for_each(rdev2, tmp2, mddev)
2025 if (test_bit(AllReserved, &rdev2->flags) ||
2026 (rdev->bdev == rdev2->bdev &&
2028 overlaps(rdev->data_offset, rdev->size,
2029 rdev2->data_offset, rdev2->size))) {
2033 mddev_unlock(mddev);
2039 mddev_lock(my_mddev);
2041 /* Someone else could have slipped in a size
2042 * change here, but doing so is just silly.
2043 * We put oldsize back because we *know* it is
2044 * safe, and trust userspace not to race with
2047 rdev->size = oldsize;
2051 if (size < my_mddev->size || my_mddev->size == 0)
2052 my_mddev->size = size;
2056 static struct rdev_sysfs_entry rdev_size =
2057 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2059 static struct attribute *rdev_default_attrs[] = {
2068 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2070 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2071 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2072 mddev_t *mddev = rdev->mddev;
2078 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2080 if (rdev->mddev == NULL)
2083 rv = entry->show(rdev, page);
2084 mddev_unlock(mddev);
2090 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2091 const char *page, size_t length)
2093 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2094 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2096 mddev_t *mddev = rdev->mddev;
2100 if (!capable(CAP_SYS_ADMIN))
2102 rv = mddev ? mddev_lock(mddev): -EBUSY;
2104 if (rdev->mddev == NULL)
2107 rv = entry->store(rdev, page, length);
2108 mddev_unlock(mddev);
2113 static void rdev_free(struct kobject *ko)
2115 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2118 static struct sysfs_ops rdev_sysfs_ops = {
2119 .show = rdev_attr_show,
2120 .store = rdev_attr_store,
2122 static struct kobj_type rdev_ktype = {
2123 .release = rdev_free,
2124 .sysfs_ops = &rdev_sysfs_ops,
2125 .default_attrs = rdev_default_attrs,
2129 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2131 * mark the device faulty if:
2133 * - the device is nonexistent (zero size)
2134 * - the device has no valid superblock
2136 * a faulty rdev _never_ has rdev->sb set.
2138 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2140 char b[BDEVNAME_SIZE];
2145 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2147 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2148 return ERR_PTR(-ENOMEM);
2151 if ((err = alloc_disk_sb(rdev)))
2154 err = lock_rdev(rdev, newdev, super_format == -2);
2158 kobject_init(&rdev->kobj, &rdev_ktype);
2161 rdev->saved_raid_disk = -1;
2162 rdev->raid_disk = -1;
2164 rdev->data_offset = 0;
2165 rdev->sb_events = 0;
2166 atomic_set(&rdev->nr_pending, 0);
2167 atomic_set(&rdev->read_errors, 0);
2168 atomic_set(&rdev->corrected_errors, 0);
2170 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2173 "md: %s has zero or unknown size, marking faulty!\n",
2174 bdevname(rdev->bdev,b));
2179 if (super_format >= 0) {
2180 err = super_types[super_format].
2181 load_super(rdev, NULL, super_minor);
2182 if (err == -EINVAL) {
2184 "md: %s does not have a valid v%d.%d "
2185 "superblock, not importing!\n",
2186 bdevname(rdev->bdev,b),
2187 super_format, super_minor);
2192 "md: could not read %s's sb, not importing!\n",
2193 bdevname(rdev->bdev,b));
2197 INIT_LIST_HEAD(&rdev->same_set);
2202 if (rdev->sb_page) {
2208 return ERR_PTR(err);
2212 * Check a full RAID array for plausibility
2216 static void analyze_sbs(mddev_t * mddev)
2219 struct list_head *tmp;
2220 mdk_rdev_t *rdev, *freshest;
2221 char b[BDEVNAME_SIZE];
2224 rdev_for_each(rdev, tmp, mddev)
2225 switch (super_types[mddev->major_version].
2226 load_super(rdev, freshest, mddev->minor_version)) {
2234 "md: fatal superblock inconsistency in %s"
2235 " -- removing from array\n",
2236 bdevname(rdev->bdev,b));
2237 kick_rdev_from_array(rdev);
2241 super_types[mddev->major_version].
2242 validate_super(mddev, freshest);
2245 rdev_for_each(rdev, tmp, mddev) {
2246 if (rdev != freshest)
2247 if (super_types[mddev->major_version].
2248 validate_super(mddev, rdev)) {
2249 printk(KERN_WARNING "md: kicking non-fresh %s"
2251 bdevname(rdev->bdev,b));
2252 kick_rdev_from_array(rdev);
2255 if (mddev->level == LEVEL_MULTIPATH) {
2256 rdev->desc_nr = i++;
2257 rdev->raid_disk = rdev->desc_nr;
2258 set_bit(In_sync, &rdev->flags);
2259 } else if (rdev->raid_disk >= mddev->raid_disks) {
2260 rdev->raid_disk = -1;
2261 clear_bit(In_sync, &rdev->flags);
2267 if (mddev->recovery_cp != MaxSector &&
2269 printk(KERN_ERR "md: %s: raid array is not clean"
2270 " -- starting background reconstruction\n",
2276 safe_delay_show(mddev_t *mddev, char *page)
2278 int msec = (mddev->safemode_delay*1000)/HZ;
2279 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2282 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2290 /* remove a period, and count digits after it */
2291 if (len >= sizeof(buf))
2293 strlcpy(buf, cbuf, len);
2295 for (i=0; i<len; i++) {
2297 if (isdigit(buf[i])) {
2302 } else if (buf[i] == '.') {
2307 msec = simple_strtoul(buf, &e, 10);
2308 if (e == buf || (*e && *e != '\n'))
2310 msec = (msec * 1000) / scale;
2312 mddev->safemode_delay = 0;
2314 mddev->safemode_delay = (msec*HZ)/1000;
2315 if (mddev->safemode_delay == 0)
2316 mddev->safemode_delay = 1;
2320 static struct md_sysfs_entry md_safe_delay =
2321 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2324 level_show(mddev_t *mddev, char *page)
2326 struct mdk_personality *p = mddev->pers;
2328 return sprintf(page, "%s\n", p->name);
2329 else if (mddev->clevel[0])
2330 return sprintf(page, "%s\n", mddev->clevel);
2331 else if (mddev->level != LEVEL_NONE)
2332 return sprintf(page, "%d\n", mddev->level);
2338 level_store(mddev_t *mddev, const char *buf, size_t len)
2345 if (len >= sizeof(mddev->clevel))
2347 strncpy(mddev->clevel, buf, len);
2348 if (mddev->clevel[len-1] == '\n')
2350 mddev->clevel[len] = 0;
2351 mddev->level = LEVEL_NONE;
2355 static struct md_sysfs_entry md_level =
2356 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2360 layout_show(mddev_t *mddev, char *page)
2362 /* just a number, not meaningful for all levels */
2363 if (mddev->reshape_position != MaxSector &&
2364 mddev->layout != mddev->new_layout)
2365 return sprintf(page, "%d (%d)\n",
2366 mddev->new_layout, mddev->layout);
2367 return sprintf(page, "%d\n", mddev->layout);
2371 layout_store(mddev_t *mddev, const char *buf, size_t len)
2374 unsigned long n = simple_strtoul(buf, &e, 10);
2376 if (!*buf || (*e && *e != '\n'))
2381 if (mddev->reshape_position != MaxSector)
2382 mddev->new_layout = n;
2387 static struct md_sysfs_entry md_layout =
2388 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2392 raid_disks_show(mddev_t *mddev, char *page)
2394 if (mddev->raid_disks == 0)
2396 if (mddev->reshape_position != MaxSector &&
2397 mddev->delta_disks != 0)
2398 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2399 mddev->raid_disks - mddev->delta_disks);
2400 return sprintf(page, "%d\n", mddev->raid_disks);
2403 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2406 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2410 unsigned long n = simple_strtoul(buf, &e, 10);
2412 if (!*buf || (*e && *e != '\n'))
2416 rv = update_raid_disks(mddev, n);
2417 else if (mddev->reshape_position != MaxSector) {
2418 int olddisks = mddev->raid_disks - mddev->delta_disks;
2419 mddev->delta_disks = n - olddisks;
2420 mddev->raid_disks = n;
2422 mddev->raid_disks = n;
2423 return rv ? rv : len;
2425 static struct md_sysfs_entry md_raid_disks =
2426 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2429 chunk_size_show(mddev_t *mddev, char *page)
2431 if (mddev->reshape_position != MaxSector &&
2432 mddev->chunk_size != mddev->new_chunk)
2433 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2435 return sprintf(page, "%d\n", mddev->chunk_size);
2439 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2441 /* can only set chunk_size if array is not yet active */
2443 unsigned long n = simple_strtoul(buf, &e, 10);
2445 if (!*buf || (*e && *e != '\n'))
2450 else if (mddev->reshape_position != MaxSector)
2451 mddev->new_chunk = n;
2453 mddev->chunk_size = n;
2456 static struct md_sysfs_entry md_chunk_size =
2457 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2460 resync_start_show(mddev_t *mddev, char *page)
2462 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2466 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2469 unsigned long long n = simple_strtoull(buf, &e, 10);
2473 if (!*buf || (*e && *e != '\n'))
2476 mddev->recovery_cp = n;
2479 static struct md_sysfs_entry md_resync_start =
2480 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2483 * The array state can be:
2486 * No devices, no size, no level
2487 * Equivalent to STOP_ARRAY ioctl
2489 * May have some settings, but array is not active
2490 * all IO results in error
2491 * When written, doesn't tear down array, but just stops it
2492 * suspended (not supported yet)
2493 * All IO requests will block. The array can be reconfigured.
2494 * Writing this, if accepted, will block until array is quiessent
2496 * no resync can happen. no superblocks get written.
2497 * write requests fail
2499 * like readonly, but behaves like 'clean' on a write request.
2501 * clean - no pending writes, but otherwise active.
2502 * When written to inactive array, starts without resync
2503 * If a write request arrives then
2504 * if metadata is known, mark 'dirty' and switch to 'active'.
2505 * if not known, block and switch to write-pending
2506 * If written to an active array that has pending writes, then fails.
2508 * fully active: IO and resync can be happening.
2509 * When written to inactive array, starts with resync
2512 * clean, but writes are blocked waiting for 'active' to be written.
2515 * like active, but no writes have been seen for a while (100msec).
2518 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2519 write_pending, active_idle, bad_word};
2520 static char *array_states[] = {
2521 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2522 "write-pending", "active-idle", NULL };
2524 static int match_word(const char *word, char **list)
2527 for (n=0; list[n]; n++)
2528 if (cmd_match(word, list[n]))
2534 array_state_show(mddev_t *mddev, char *page)
2536 enum array_state st = inactive;
2549 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2551 else if (mddev->safemode)
2557 if (list_empty(&mddev->disks) &&
2558 mddev->raid_disks == 0 &&
2564 return sprintf(page, "%s\n", array_states[st]);
2567 static int do_md_stop(mddev_t * mddev, int ro);
2568 static int do_md_run(mddev_t * mddev);
2569 static int restart_array(mddev_t *mddev);
2572 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2575 enum array_state st = match_word(buf, array_states);
2580 /* stopping an active array */
2581 if (atomic_read(&mddev->active) > 1)
2583 err = do_md_stop(mddev, 0);
2586 /* stopping an active array */
2588 if (atomic_read(&mddev->active) > 1)
2590 err = do_md_stop(mddev, 2);
2592 err = 0; /* already inactive */
2595 break; /* not supported yet */
2598 err = do_md_stop(mddev, 1);
2601 set_disk_ro(mddev->gendisk, 1);
2602 err = do_md_run(mddev);
2608 err = do_md_stop(mddev, 1);
2610 err = restart_array(mddev);
2613 set_disk_ro(mddev->gendisk, 0);
2617 err = do_md_run(mddev);
2622 restart_array(mddev);
2623 spin_lock_irq(&mddev->write_lock);
2624 if (atomic_read(&mddev->writes_pending) == 0) {
2625 if (mddev->in_sync == 0) {
2627 if (mddev->safemode == 1)
2628 mddev->safemode = 0;
2629 if (mddev->persistent)
2630 set_bit(MD_CHANGE_CLEAN,
2636 spin_unlock_irq(&mddev->write_lock);
2639 mddev->recovery_cp = MaxSector;
2640 err = do_md_run(mddev);
2645 restart_array(mddev);
2646 if (mddev->external)
2647 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2648 wake_up(&mddev->sb_wait);
2652 set_disk_ro(mddev->gendisk, 0);
2653 err = do_md_run(mddev);
2658 /* these cannot be set */
2666 static struct md_sysfs_entry md_array_state =
2667 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2670 null_show(mddev_t *mddev, char *page)
2676 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2678 /* buf must be %d:%d\n? giving major and minor numbers */
2679 /* The new device is added to the array.
2680 * If the array has a persistent superblock, we read the
2681 * superblock to initialise info and check validity.
2682 * Otherwise, only checking done is that in bind_rdev_to_array,
2683 * which mainly checks size.
2686 int major = simple_strtoul(buf, &e, 10);
2692 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2694 minor = simple_strtoul(e+1, &e, 10);
2695 if (*e && *e != '\n')
2697 dev = MKDEV(major, minor);
2698 if (major != MAJOR(dev) ||
2699 minor != MINOR(dev))
2703 if (mddev->persistent) {
2704 rdev = md_import_device(dev, mddev->major_version,
2705 mddev->minor_version);
2706 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2707 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2708 mdk_rdev_t, same_set);
2709 err = super_types[mddev->major_version]
2710 .load_super(rdev, rdev0, mddev->minor_version);
2714 } else if (mddev->external)
2715 rdev = md_import_device(dev, -2, -1);
2717 rdev = md_import_device(dev, -1, -1);
2720 return PTR_ERR(rdev);
2721 err = bind_rdev_to_array(rdev, mddev);
2725 return err ? err : len;
2728 static struct md_sysfs_entry md_new_device =
2729 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2732 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2735 unsigned long chunk, end_chunk;
2739 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2741 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2742 if (buf == end) break;
2743 if (*end == '-') { /* range */
2745 end_chunk = simple_strtoul(buf, &end, 0);
2746 if (buf == end) break;
2748 if (*end && !isspace(*end)) break;
2749 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2751 while (isspace(*buf)) buf++;
2753 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2758 static struct md_sysfs_entry md_bitmap =
2759 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2762 size_show(mddev_t *mddev, char *page)
2764 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2767 static int update_size(mddev_t *mddev, unsigned long size);
2770 size_store(mddev_t *mddev, const char *buf, size_t len)
2772 /* If array is inactive, we can reduce the component size, but
2773 * not increase it (except from 0).
2774 * If array is active, we can try an on-line resize
2778 unsigned long long size = simple_strtoull(buf, &e, 10);
2779 if (!*buf || *buf == '\n' ||
2784 err = update_size(mddev, size);
2785 md_update_sb(mddev, 1);
2787 if (mddev->size == 0 ||
2793 return err ? err : len;
2796 static struct md_sysfs_entry md_size =
2797 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2802 * 'none' for arrays with no metadata (good luck...)
2803 * 'external' for arrays with externally managed metadata,
2804 * or N.M for internally known formats
2807 metadata_show(mddev_t *mddev, char *page)
2809 if (mddev->persistent)
2810 return sprintf(page, "%d.%d\n",
2811 mddev->major_version, mddev->minor_version);
2812 else if (mddev->external)
2813 return sprintf(page, "external:%s\n", mddev->metadata_type);
2815 return sprintf(page, "none\n");
2819 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2823 if (!list_empty(&mddev->disks))
2826 if (cmd_match(buf, "none")) {
2827 mddev->persistent = 0;
2828 mddev->external = 0;
2829 mddev->major_version = 0;
2830 mddev->minor_version = 90;
2833 if (strncmp(buf, "external:", 9) == 0) {
2834 size_t namelen = len-9;
2835 if (namelen >= sizeof(mddev->metadata_type))
2836 namelen = sizeof(mddev->metadata_type)-1;
2837 strncpy(mddev->metadata_type, buf+9, namelen);
2838 mddev->metadata_type[namelen] = 0;
2839 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2840 mddev->metadata_type[--namelen] = 0;
2841 mddev->persistent = 0;
2842 mddev->external = 1;
2843 mddev->major_version = 0;
2844 mddev->minor_version = 90;
2847 major = simple_strtoul(buf, &e, 10);
2848 if (e==buf || *e != '.')
2851 minor = simple_strtoul(buf, &e, 10);
2852 if (e==buf || (*e && *e != '\n') )
2854 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2856 mddev->major_version = major;
2857 mddev->minor_version = minor;
2858 mddev->persistent = 1;
2859 mddev->external = 0;
2863 static struct md_sysfs_entry md_metadata =
2864 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2867 action_show(mddev_t *mddev, char *page)
2869 char *type = "idle";
2870 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2871 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2872 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2874 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2875 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2877 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2884 return sprintf(page, "%s\n", type);
2888 action_store(mddev_t *mddev, const char *page, size_t len)
2890 if (!mddev->pers || !mddev->pers->sync_request)
2893 if (cmd_match(page, "idle")) {
2894 if (mddev->sync_thread) {
2895 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2896 md_unregister_thread(mddev->sync_thread);
2897 mddev->sync_thread = NULL;
2898 mddev->recovery = 0;
2900 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2901 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2903 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2904 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2905 else if (cmd_match(page, "reshape")) {
2907 if (mddev->pers->start_reshape == NULL)
2909 err = mddev->pers->start_reshape(mddev);
2913 if (cmd_match(page, "check"))
2914 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2915 else if (!cmd_match(page, "repair"))
2917 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2918 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2920 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2921 md_wakeup_thread(mddev->thread);
2926 mismatch_cnt_show(mddev_t *mddev, char *page)
2928 return sprintf(page, "%llu\n",
2929 (unsigned long long) mddev->resync_mismatches);
2932 static struct md_sysfs_entry md_scan_mode =
2933 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2936 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2939 sync_min_show(mddev_t *mddev, char *page)
2941 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2942 mddev->sync_speed_min ? "local": "system");
2946 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2950 if (strncmp(buf, "system", 6)==0) {
2951 mddev->sync_speed_min = 0;
2954 min = simple_strtoul(buf, &e, 10);
2955 if (buf == e || (*e && *e != '\n') || min <= 0)
2957 mddev->sync_speed_min = min;
2961 static struct md_sysfs_entry md_sync_min =
2962 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2965 sync_max_show(mddev_t *mddev, char *page)
2967 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2968 mddev->sync_speed_max ? "local": "system");
2972 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2976 if (strncmp(buf, "system", 6)==0) {
2977 mddev->sync_speed_max = 0;
2980 max = simple_strtoul(buf, &e, 10);
2981 if (buf == e || (*e && *e != '\n') || max <= 0)
2983 mddev->sync_speed_max = max;
2987 static struct md_sysfs_entry md_sync_max =
2988 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2991 degraded_show(mddev_t *mddev, char *page)
2993 return sprintf(page, "%d\n", mddev->degraded);
2995 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2998 sync_speed_show(mddev_t *mddev, char *page)
3000 unsigned long resync, dt, db;
3001 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
3002 dt = ((jiffies - mddev->resync_mark) / HZ);
3004 db = resync - (mddev->resync_mark_cnt);
3005 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
3008 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3011 sync_completed_show(mddev_t *mddev, char *page)
3013 unsigned long max_blocks, resync;
3015 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3016 max_blocks = mddev->resync_max_sectors;
3018 max_blocks = mddev->size << 1;
3020 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3021 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3024 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3027 max_sync_show(mddev_t *mddev, char *page)
3029 if (mddev->resync_max == MaxSector)
3030 return sprintf(page, "max\n");
3032 return sprintf(page, "%llu\n",
3033 (unsigned long long)mddev->resync_max);
3036 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3038 if (strncmp(buf, "max", 3) == 0)
3039 mddev->resync_max = MaxSector;
3042 unsigned long long max = simple_strtoull(buf, &ep, 10);
3043 if (ep == buf || (*ep != 0 && *ep != '\n'))
3045 if (max < mddev->resync_max &&
3046 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3049 /* Must be a multiple of chunk_size */
3050 if (mddev->chunk_size) {
3051 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3054 mddev->resync_max = max;
3056 wake_up(&mddev->recovery_wait);
3060 static struct md_sysfs_entry md_max_sync =
3061 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3064 suspend_lo_show(mddev_t *mddev, char *page)
3066 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3070 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3073 unsigned long long new = simple_strtoull(buf, &e, 10);
3075 if (mddev->pers->quiesce == NULL)
3077 if (buf == e || (*e && *e != '\n'))
3079 if (new >= mddev->suspend_hi ||
3080 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3081 mddev->suspend_lo = new;
3082 mddev->pers->quiesce(mddev, 2);
3087 static struct md_sysfs_entry md_suspend_lo =
3088 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3092 suspend_hi_show(mddev_t *mddev, char *page)
3094 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3098 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3101 unsigned long long new = simple_strtoull(buf, &e, 10);
3103 if (mddev->pers->quiesce == NULL)
3105 if (buf == e || (*e && *e != '\n'))
3107 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3108 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3109 mddev->suspend_hi = new;
3110 mddev->pers->quiesce(mddev, 1);
3111 mddev->pers->quiesce(mddev, 0);
3116 static struct md_sysfs_entry md_suspend_hi =
3117 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3120 reshape_position_show(mddev_t *mddev, char *page)
3122 if (mddev->reshape_position != MaxSector)
3123 return sprintf(page, "%llu\n",
3124 (unsigned long long)mddev->reshape_position);
3125 strcpy(page, "none\n");
3130 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3133 unsigned long long new = simple_strtoull(buf, &e, 10);
3136 if (buf == e || (*e && *e != '\n'))
3138 mddev->reshape_position = new;
3139 mddev->delta_disks = 0;
3140 mddev->new_level = mddev->level;
3141 mddev->new_layout = mddev->layout;
3142 mddev->new_chunk = mddev->chunk_size;
3146 static struct md_sysfs_entry md_reshape_position =
3147 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3148 reshape_position_store);
3151 static struct attribute *md_default_attrs[] = {
3154 &md_raid_disks.attr,
3155 &md_chunk_size.attr,
3157 &md_resync_start.attr,
3159 &md_new_device.attr,
3160 &md_safe_delay.attr,
3161 &md_array_state.attr,
3162 &md_reshape_position.attr,
3166 static struct attribute *md_redundancy_attrs[] = {
3168 &md_mismatches.attr,
3171 &md_sync_speed.attr,
3172 &md_sync_completed.attr,
3174 &md_suspend_lo.attr,
3175 &md_suspend_hi.attr,
3180 static struct attribute_group md_redundancy_group = {
3182 .attrs = md_redundancy_attrs,
3187 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3189 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3190 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3195 rv = mddev_lock(mddev);
3197 rv = entry->show(mddev, page);
3198 mddev_unlock(mddev);
3204 md_attr_store(struct kobject *kobj, struct attribute *attr,
3205 const char *page, size_t length)
3207 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3208 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3213 if (!capable(CAP_SYS_ADMIN))
3215 rv = mddev_lock(mddev);
3217 rv = entry->store(mddev, page, length);
3218 mddev_unlock(mddev);
3223 static void md_free(struct kobject *ko)
3225 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3229 static struct sysfs_ops md_sysfs_ops = {
3230 .show = md_attr_show,
3231 .store = md_attr_store,
3233 static struct kobj_type md_ktype = {
3235 .sysfs_ops = &md_sysfs_ops,
3236 .default_attrs = md_default_attrs,
3241 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3243 static DEFINE_MUTEX(disks_mutex);
3244 mddev_t *mddev = mddev_find(dev);
3245 struct gendisk *disk;
3246 int partitioned = (MAJOR(dev) != MD_MAJOR);
3247 int shift = partitioned ? MdpMinorShift : 0;
3248 int unit = MINOR(dev) >> shift;
3254 mutex_lock(&disks_mutex);
3255 if (mddev->gendisk) {
3256 mutex_unlock(&disks_mutex);
3260 disk = alloc_disk(1 << shift);
3262 mutex_unlock(&disks_mutex);
3266 disk->major = MAJOR(dev);
3267 disk->first_minor = unit << shift;
3269 sprintf(disk->disk_name, "md_d%d", unit);
3271 sprintf(disk->disk_name, "md%d", unit);
3272 disk->fops = &md_fops;
3273 disk->private_data = mddev;
3274 disk->queue = mddev->queue;
3276 mddev->gendisk = disk;
3277 mutex_unlock(&disks_mutex);
3278 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3281 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3284 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3288 static void md_safemode_timeout(unsigned long data)
3290 mddev_t *mddev = (mddev_t *) data;
3292 mddev->safemode = 1;
3293 md_wakeup_thread(mddev->thread);
3296 static int start_dirty_degraded;
3298 static int do_md_run(mddev_t * mddev)
3302 struct list_head *tmp;
3304 struct gendisk *disk;
3305 struct mdk_personality *pers;
3306 char b[BDEVNAME_SIZE];
3308 if (list_empty(&mddev->disks))
3309 /* cannot run an array with no devices.. */
3316 * Analyze all RAID superblock(s)
3318 if (!mddev->raid_disks) {
3319 if (!mddev->persistent)
3324 chunk_size = mddev->chunk_size;
3327 if (chunk_size > MAX_CHUNK_SIZE) {
3328 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3329 chunk_size, MAX_CHUNK_SIZE);
3333 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3335 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3336 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3339 if (chunk_size < PAGE_SIZE) {
3340 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3341 chunk_size, PAGE_SIZE);
3345 /* devices must have minimum size of one chunk */
3346 rdev_for_each(rdev, tmp, mddev) {
3347 if (test_bit(Faulty, &rdev->flags))
3349 if (rdev->size < chunk_size / 1024) {
3351 "md: Dev %s smaller than chunk_size:"
3353 bdevname(rdev->bdev,b),
3354 (unsigned long long)rdev->size,
3362 if (mddev->level != LEVEL_NONE)
3363 request_module("md-level-%d", mddev->level);
3364 else if (mddev->clevel[0])
3365 request_module("md-%s", mddev->clevel);
3369 * Drop all container device buffers, from now on
3370 * the only valid external interface is through the md
3373 rdev_for_each(rdev, tmp, mddev) {
3374 if (test_bit(Faulty, &rdev->flags))
3376 sync_blockdev(rdev->bdev);
3377 invalidate_bdev(rdev->bdev);
3379 /* perform some consistency tests on the device.
3380 * We don't want the data to overlap the metadata,
3381 * Internal Bitmap issues has handled elsewhere.
3383 if (rdev->data_offset < rdev->sb_offset) {
3385 rdev->data_offset + mddev->size*2
3386 > rdev->sb_offset*2) {
3387 printk("md: %s: data overlaps metadata\n",
3392 if (rdev->sb_offset*2 + rdev->sb_size/512
3393 > rdev->data_offset) {
3394 printk("md: %s: metadata overlaps data\n",
3401 md_probe(mddev->unit, NULL, NULL);
3402 disk = mddev->gendisk;
3406 spin_lock(&pers_lock);
3407 pers = find_pers(mddev->level, mddev->clevel);
3408 if (!pers || !try_module_get(pers->owner)) {
3409 spin_unlock(&pers_lock);
3410 if (mddev->level != LEVEL_NONE)
3411 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3414 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3419 spin_unlock(&pers_lock);
3420 mddev->level = pers->level;
3421 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3423 if (mddev->reshape_position != MaxSector &&
3424 pers->start_reshape == NULL) {
3425 /* This personality cannot handle reshaping... */
3427 module_put(pers->owner);
3431 if (pers->sync_request) {
3432 /* Warn if this is a potentially silly
3435 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3437 struct list_head *tmp2;
3439 rdev_for_each(rdev, tmp, mddev) {
3440 rdev_for_each(rdev2, tmp2, mddev) {
3442 rdev->bdev->bd_contains ==
3443 rdev2->bdev->bd_contains) {
3445 "%s: WARNING: %s appears to be"
3446 " on the same physical disk as"
3449 bdevname(rdev->bdev,b),
3450 bdevname(rdev2->bdev,b2));
3457 "True protection against single-disk"
3458 " failure might be compromised.\n");
3461 mddev->recovery = 0;
3462 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3463 mddev->barriers_work = 1;
3464 mddev->ok_start_degraded = start_dirty_degraded;
3467 mddev->ro = 2; /* read-only, but switch on first write */
3469 err = mddev->pers->run(mddev);
3470 if (!err && mddev->pers->sync_request) {
3471 err = bitmap_create(mddev);
3473 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3474 mdname(mddev), err);
3475 mddev->pers->stop(mddev);
3479 printk(KERN_ERR "md: pers->run() failed ...\n");
3480 module_put(mddev->pers->owner);
3482 bitmap_destroy(mddev);
3485 if (mddev->pers->sync_request) {
3486 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3488 "md: cannot register extra attributes for %s\n",
3490 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3493 atomic_set(&mddev->writes_pending,0);
3494 mddev->safemode = 0;
3495 mddev->safemode_timer.function = md_safemode_timeout;
3496 mddev->safemode_timer.data = (unsigned long) mddev;
3497 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3500 rdev_for_each(rdev, tmp, mddev)
3501 if (rdev->raid_disk >= 0) {
3503 sprintf(nm, "rd%d", rdev->raid_disk);
3504 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3505 printk("md: cannot register %s for %s\n",
3509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3512 md_update_sb(mddev, 0);
3514 set_capacity(disk, mddev->array_size<<1);
3516 /* If we call blk_queue_make_request here, it will
3517 * re-initialise max_sectors etc which may have been
3518 * refined inside -> run. So just set the bits we need to set.
3519 * Most initialisation happended when we called
3520 * blk_queue_make_request(..., md_fail_request)
3523 mddev->queue->queuedata = mddev;
3524 mddev->queue->make_request_fn = mddev->pers->make_request;
3526 /* If there is a partially-recovered drive we need to
3527 * start recovery here. If we leave it to md_check_recovery,
3528 * it will remove the drives and not do the right thing
3530 if (mddev->degraded && !mddev->sync_thread) {
3531 struct list_head *rtmp;
3533 rdev_for_each(rdev, rtmp, mddev)
3534 if (rdev->raid_disk >= 0 &&
3535 !test_bit(In_sync, &rdev->flags) &&
3536 !test_bit(Faulty, &rdev->flags))
3537 /* complete an interrupted recovery */
3539 if (spares && mddev->pers->sync_request) {
3540 mddev->recovery = 0;
3541 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3542 mddev->sync_thread = md_register_thread(md_do_sync,
3545 if (!mddev->sync_thread) {
3546 printk(KERN_ERR "%s: could not start resync"
3549 /* leave the spares where they are, it shouldn't hurt */
3550 mddev->recovery = 0;
3554 md_wakeup_thread(mddev->thread);
3555 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3558 md_new_event(mddev);
3559 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3563 static int restart_array(mddev_t *mddev)
3565 struct gendisk *disk = mddev->gendisk;
3569 * Complain if it has no devices
3572 if (list_empty(&mddev->disks))
3580 mddev->safemode = 0;
3582 set_disk_ro(disk, 0);
3584 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3587 * Kick recovery or resync if necessary
3589 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3590 md_wakeup_thread(mddev->thread);
3591 md_wakeup_thread(mddev->sync_thread);
3600 /* similar to deny_write_access, but accounts for our holding a reference
3601 * to the file ourselves */
3602 static int deny_bitmap_write_access(struct file * file)
3604 struct inode *inode = file->f_mapping->host;
3606 spin_lock(&inode->i_lock);
3607 if (atomic_read(&inode->i_writecount) > 1) {
3608 spin_unlock(&inode->i_lock);
3611 atomic_set(&inode->i_writecount, -1);
3612 spin_unlock(&inode->i_lock);
3617 static void restore_bitmap_write_access(struct file *file)
3619 struct inode *inode = file->f_mapping->host;
3621 spin_lock(&inode->i_lock);
3622 atomic_set(&inode->i_writecount, 1);
3623 spin_unlock(&inode->i_lock);
3627 * 0 - completely stop and dis-assemble array
3628 * 1 - switch to readonly
3629 * 2 - stop but do not disassemble array
3631 static int do_md_stop(mddev_t * mddev, int mode)
3634 struct gendisk *disk = mddev->gendisk;
3637 if (atomic_read(&mddev->active)>2) {
3638 printk("md: %s still in use.\n",mdname(mddev));
3642 if (mddev->sync_thread) {
3643 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3644 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3645 md_unregister_thread(mddev->sync_thread);
3646 mddev->sync_thread = NULL;
3649 del_timer_sync(&mddev->safemode_timer);
3651 invalidate_partition(disk, 0);
3654 case 1: /* readonly */
3660 case 0: /* disassemble */
3662 bitmap_flush(mddev);
3663 md_super_wait(mddev);
3665 set_disk_ro(disk, 0);
3666 blk_queue_make_request(mddev->queue, md_fail_request);
3667 mddev->pers->stop(mddev);
3668 mddev->queue->merge_bvec_fn = NULL;
3669 mddev->queue->unplug_fn = NULL;
3670 mddev->queue->backing_dev_info.congested_fn = NULL;
3671 if (mddev->pers->sync_request)
3672 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3674 module_put(mddev->pers->owner);
3677 set_capacity(disk, 0);
3683 if (!mddev->in_sync || mddev->flags) {
3684 /* mark array as shutdown cleanly */
3686 md_update_sb(mddev, 1);
3689 set_disk_ro(disk, 1);
3690 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3694 * Free resources if final stop
3698 struct list_head *tmp;
3700 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3702 bitmap_destroy(mddev);
3703 if (mddev->bitmap_file) {
3704 restore_bitmap_write_access(mddev->bitmap_file);
3705 fput(mddev->bitmap_file);
3706 mddev->bitmap_file = NULL;
3708 mddev->bitmap_offset = 0;
3710 rdev_for_each(rdev, tmp, mddev)
3711 if (rdev->raid_disk >= 0) {
3713 sprintf(nm, "rd%d", rdev->raid_disk);
3714 sysfs_remove_link(&mddev->kobj, nm);
3717 /* make sure all md_delayed_delete calls have finished */
3718 flush_scheduled_work();
3720 export_array(mddev);
3722 mddev->array_size = 0;
3724 mddev->raid_disks = 0;
3725 mddev->recovery_cp = 0;
3726 mddev->resync_max = MaxSector;
3727 mddev->reshape_position = MaxSector;
3728 mddev->external = 0;
3729 mddev->persistent = 0;
3730 mddev->level = LEVEL_NONE;
3731 mddev->clevel[0] = 0;
3734 mddev->metadata_type[0] = 0;
3735 mddev->chunk_size = 0;
3736 mddev->ctime = mddev->utime = 0;
3738 mddev->max_disks = 0;
3740 mddev->delta_disks = 0;
3741 mddev->new_level = LEVEL_NONE;
3742 mddev->new_layout = 0;
3743 mddev->new_chunk = 0;
3744 mddev->curr_resync = 0;
3745 mddev->resync_mismatches = 0;
3746 mddev->suspend_lo = mddev->suspend_hi = 0;
3747 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3748 mddev->recovery = 0;
3751 mddev->degraded = 0;
3752 mddev->barriers_work = 0;
3753 mddev->safemode = 0;
3755 } else if (mddev->pers)
3756 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3759 md_new_event(mddev);
3765 static void autorun_array(mddev_t *mddev)
3768 struct list_head *tmp;
3771 if (list_empty(&mddev->disks))
3774 printk(KERN_INFO "md: running: ");
3776 rdev_for_each(rdev, tmp, mddev) {
3777 char b[BDEVNAME_SIZE];
3778 printk("<%s>", bdevname(rdev->bdev,b));
3782 err = do_md_run (mddev);
3784 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3785 do_md_stop (mddev, 0);
3790 * lets try to run arrays based on all disks that have arrived
3791 * until now. (those are in pending_raid_disks)
3793 * the method: pick the first pending disk, collect all disks with
3794 * the same UUID, remove all from the pending list and put them into
3795 * the 'same_array' list. Then order this list based on superblock
3796 * update time (freshest comes first), kick out 'old' disks and
3797 * compare superblocks. If everything's fine then run it.
3799 * If "unit" is allocated, then bump its reference count
3801 static void autorun_devices(int part)
3803 struct list_head *tmp;
3804 mdk_rdev_t *rdev0, *rdev;
3806 char b[BDEVNAME_SIZE];
3808 printk(KERN_INFO "md: autorun ...\n");
3809 while (!list_empty(&pending_raid_disks)) {
3812 LIST_HEAD(candidates);
3813 rdev0 = list_entry(pending_raid_disks.next,
3814 mdk_rdev_t, same_set);
3816 printk(KERN_INFO "md: considering %s ...\n",
3817 bdevname(rdev0->bdev,b));
3818 INIT_LIST_HEAD(&candidates);
3819 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3820 if (super_90_load(rdev, rdev0, 0) >= 0) {
3821 printk(KERN_INFO "md: adding %s ...\n",
3822 bdevname(rdev->bdev,b));
3823 list_move(&rdev->same_set, &candidates);
3826 * now we have a set of devices, with all of them having
3827 * mostly sane superblocks. It's time to allocate the
3831 dev = MKDEV(mdp_major,
3832 rdev0->preferred_minor << MdpMinorShift);
3833 unit = MINOR(dev) >> MdpMinorShift;
3835 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3838 if (rdev0->preferred_minor != unit) {
3839 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3840 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3844 md_probe(dev, NULL, NULL);
3845 mddev = mddev_find(dev);
3848 "md: cannot allocate memory for md drive.\n");
3851 if (mddev_lock(mddev))
3852 printk(KERN_WARNING "md: %s locked, cannot run\n",
3854 else if (mddev->raid_disks || mddev->major_version
3855 || !list_empty(&mddev->disks)) {
3857 "md: %s already running, cannot run %s\n",
3858 mdname(mddev), bdevname(rdev0->bdev,b));
3859 mddev_unlock(mddev);
3861 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3862 mddev->persistent = 1;
3863 rdev_for_each_list(rdev, tmp, candidates) {
3864 list_del_init(&rdev->same_set);
3865 if (bind_rdev_to_array(rdev, mddev))
3868 autorun_array(mddev);
3869 mddev_unlock(mddev);
3871 /* on success, candidates will be empty, on error
3874 rdev_for_each_list(rdev, tmp, candidates)
3878 printk(KERN_INFO "md: ... autorun DONE.\n");
3880 #endif /* !MODULE */
3882 static int get_version(void __user * arg)
3886 ver.major = MD_MAJOR_VERSION;
3887 ver.minor = MD_MINOR_VERSION;
3888 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3890 if (copy_to_user(arg, &ver, sizeof(ver)))
3896 static int get_array_info(mddev_t * mddev, void __user * arg)
3898 mdu_array_info_t info;
3899 int nr,working,active,failed,spare;
3901 struct list_head *tmp;
3903 nr=working=active=failed=spare=0;
3904 rdev_for_each(rdev, tmp, mddev) {
3906 if (test_bit(Faulty, &rdev->flags))
3910 if (test_bit(In_sync, &rdev->flags))
3917 info.major_version = mddev->major_version;
3918 info.minor_version = mddev->minor_version;
3919 info.patch_version = MD_PATCHLEVEL_VERSION;
3920 info.ctime = mddev->ctime;
3921 info.level = mddev->level;
3922 info.size = mddev->size;
3923 if (info.size != mddev->size) /* overflow */
3926 info.raid_disks = mddev->raid_disks;
3927 info.md_minor = mddev->md_minor;
3928 info.not_persistent= !mddev->persistent;
3930 info.utime = mddev->utime;
3933 info.state = (1<<MD_SB_CLEAN);
3934 if (mddev->bitmap && mddev->bitmap_offset)
3935 info.state = (1<<MD_SB_BITMAP_PRESENT);
3936 info.active_disks = active;
3937 info.working_disks = working;
3938 info.failed_disks = failed;
3939 info.spare_disks = spare;
3941 info.layout = mddev->layout;
3942 info.chunk_size = mddev->chunk_size;
3944 if (copy_to_user(arg, &info, sizeof(info)))
3950 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3952 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3953 char *ptr, *buf = NULL;
3956 md_allow_write(mddev);
3958 file = kmalloc(sizeof(*file), GFP_KERNEL);
3962 /* bitmap disabled, zero the first byte and copy out */
3963 if (!mddev->bitmap || !mddev->bitmap->file) {
3964 file->pathname[0] = '\0';
3968 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3972 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3976 strcpy(file->pathname, ptr);
3980 if (copy_to_user(arg, file, sizeof(*file)))
3988 static int get_disk_info(mddev_t * mddev, void __user * arg)
3990 mdu_disk_info_t info;
3994 if (copy_from_user(&info, arg, sizeof(info)))
3999 rdev = find_rdev_nr(mddev, nr);
4001 info.major = MAJOR(rdev->bdev->bd_dev);
4002 info.minor = MINOR(rdev->bdev->bd_dev);
4003 info.raid_disk = rdev->raid_disk;
4005 if (test_bit(Faulty, &rdev->flags))
4006 info.state |= (1<<MD_DISK_FAULTY);
4007 else if (test_bit(In_sync, &rdev->flags)) {
4008 info.state |= (1<<MD_DISK_ACTIVE);
4009 info.state |= (1<<MD_DISK_SYNC);
4011 if (test_bit(WriteMostly, &rdev->flags))
4012 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4014 info.major = info.minor = 0;
4015 info.raid_disk = -1;
4016 info.state = (1<<MD_DISK_REMOVED);
4019 if (copy_to_user(arg, &info, sizeof(info)))
4025 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4027 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4029 dev_t dev = MKDEV(info->major,info->minor);
4031 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4034 if (!mddev->raid_disks) {
4036 /* expecting a device which has a superblock */
4037 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4040 "md: md_import_device returned %ld\n",
4042 return PTR_ERR(rdev);
4044 if (!list_empty(&mddev->disks)) {
4045 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4046 mdk_rdev_t, same_set);
4047 int err = super_types[mddev->major_version]
4048 .load_super(rdev, rdev0, mddev->minor_version);
4051 "md: %s has different UUID to %s\n",
4052 bdevname(rdev->bdev,b),
4053 bdevname(rdev0->bdev,b2));
4058 err = bind_rdev_to_array(rdev, mddev);
4065 * add_new_disk can be used once the array is assembled
4066 * to add "hot spares". They must already have a superblock
4071 if (!mddev->pers->hot_add_disk) {
4073 "%s: personality does not support diskops!\n",
4077 if (mddev->persistent)
4078 rdev = md_import_device(dev, mddev->major_version,
4079 mddev->minor_version);
4081 rdev = md_import_device(dev, -1, -1);
4084 "md: md_import_device returned %ld\n",
4086 return PTR_ERR(rdev);
4088 /* set save_raid_disk if appropriate */
4089 if (!mddev->persistent) {
4090 if (info->state & (1<<MD_DISK_SYNC) &&
4091 info->raid_disk < mddev->raid_disks)
4092 rdev->raid_disk = info->raid_disk;
4094 rdev->raid_disk = -1;
4096 super_types[mddev->major_version].
4097 validate_super(mddev, rdev);
4098 rdev->saved_raid_disk = rdev->raid_disk;
4100 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4101 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4102 set_bit(WriteMostly, &rdev->flags);
4104 rdev->raid_disk = -1;
4105 err = bind_rdev_to_array(rdev, mddev);
4106 if (!err && !mddev->pers->hot_remove_disk) {
4107 /* If there is hot_add_disk but no hot_remove_disk
4108 * then added disks for geometry changes,
4109 * and should be added immediately.
4111 super_types[mddev->major_version].
4112 validate_super(mddev, rdev);
4113 err = mddev->pers->hot_add_disk(mddev, rdev);
4115 unbind_rdev_from_array(rdev);
4120 md_update_sb(mddev, 1);
4121 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4122 md_wakeup_thread(mddev->thread);
4126 /* otherwise, add_new_disk is only allowed
4127 * for major_version==0 superblocks
4129 if (mddev->major_version != 0) {
4130 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4135 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4137 rdev = md_import_device (dev, -1, 0);
4140 "md: error, md_import_device() returned %ld\n",
4142 return PTR_ERR(rdev);
4144 rdev->desc_nr = info->number;
4145 if (info->raid_disk < mddev->raid_disks)
4146 rdev->raid_disk = info->raid_disk;
4148 rdev->raid_disk = -1;
4150 if (rdev->raid_disk < mddev->raid_disks)
4151 if (info->state & (1<<MD_DISK_SYNC))
4152 set_bit(In_sync, &rdev->flags);
4154 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4155 set_bit(WriteMostly, &rdev->flags);
4157 if (!mddev->persistent) {
4158 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4159 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4161 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4162 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4164 err = bind_rdev_to_array(rdev, mddev);
4174 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4176 char b[BDEVNAME_SIZE];
4182 rdev = find_rdev(mddev, dev);
4186 if (rdev->raid_disk >= 0)
4189 kick_rdev_from_array(rdev);
4190 md_update_sb(mddev, 1);
4191 md_new_event(mddev);
4195 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4196 bdevname(rdev->bdev,b), mdname(mddev));
4200 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4202 char b[BDEVNAME_SIZE];
4210 if (mddev->major_version != 0) {
4211 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4212 " version-0 superblocks.\n",
4216 if (!mddev->pers->hot_add_disk) {
4218 "%s: personality does not support diskops!\n",
4223 rdev = md_import_device (dev, -1, 0);
4226 "md: error, md_import_device() returned %ld\n",
4231 if (mddev->persistent)
4232 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4235 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4237 size = calc_dev_size(rdev, mddev->chunk_size);
4240 if (test_bit(Faulty, &rdev->flags)) {
4242 "md: can not hot-add faulty %s disk to %s!\n",
4243 bdevname(rdev->bdev,b), mdname(mddev));
4247 clear_bit(In_sync, &rdev->flags);
4249 rdev->saved_raid_disk = -1;
4250 err = bind_rdev_to_array(rdev, mddev);
4255 * The rest should better be atomic, we can have disk failures
4256 * noticed in interrupt contexts ...
4259 if (rdev->desc_nr == mddev->max_disks) {
4260 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4263 goto abort_unbind_export;
4266 rdev->raid_disk = -1;
4268 md_update_sb(mddev, 1);
4271 * Kick recovery, maybe this spare has to be added to the
4272 * array immediately.
4274 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4275 md_wakeup_thread(mddev->thread);
4276 md_new_event(mddev);
4279 abort_unbind_export:
4280 unbind_rdev_from_array(rdev);
4287 static int set_bitmap_file(mddev_t *mddev, int fd)
4292 if (!mddev->pers->quiesce)
4294 if (mddev->recovery || mddev->sync_thread)
4296 /* we should be able to change the bitmap.. */
4302 return -EEXIST; /* cannot add when bitmap is present */
4303 mddev->bitmap_file = fget(fd);
4305 if (mddev->bitmap_file == NULL) {
4306 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4311 err = deny_bitmap_write_access(mddev->bitmap_file);
4313 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4315 fput(mddev->bitmap_file);
4316 mddev->bitmap_file = NULL;
4319 mddev->bitmap_offset = 0; /* file overrides offset */
4320 } else if (mddev->bitmap == NULL)
4321 return -ENOENT; /* cannot remove what isn't there */
4324 mddev->pers->quiesce(mddev, 1);
4326 err = bitmap_create(mddev);
4327 if (fd < 0 || err) {
4328 bitmap_destroy(mddev);
4329 fd = -1; /* make sure to put the file */
4331 mddev->pers->quiesce(mddev, 0);
4334 if (mddev->bitmap_file) {
4335 restore_bitmap_write_access(mddev->bitmap_file);
4336 fput(mddev->bitmap_file);
4338 mddev->bitmap_file = NULL;
4345 * set_array_info is used two different ways
4346 * The original usage is when creating a new array.
4347 * In this usage, raid_disks is > 0 and it together with
4348 * level, size, not_persistent,layout,chunksize determine the
4349 * shape of the array.
4350 * This will always create an array with a type-0.90.0 superblock.
4351 * The newer usage is when assembling an array.
4352 * In this case raid_disks will be 0, and the major_version field is
4353 * use to determine which style super-blocks are to be found on the devices.
4354 * The minor and patch _version numbers are also kept incase the
4355 * super_block handler wishes to interpret them.
4357 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4360 if (info->raid_disks == 0) {
4361 /* just setting version number for superblock loading */
4362 if (info->major_version < 0 ||
4363 info->major_version >= ARRAY_SIZE(super_types) ||
4364 super_types[info->major_version].name == NULL) {
4365 /* maybe try to auto-load a module? */
4367 "md: superblock version %d not known\n",
4368 info->major_version);
4371 mddev->major_version = info->major_version;
4372 mddev->minor_version = info->minor_version;
4373 mddev->patch_version = info->patch_version;
4374 mddev->persistent = !info->not_persistent;
4377 mddev->major_version = MD_MAJOR_VERSION;
4378 mddev->minor_version = MD_MINOR_VERSION;
4379 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4380 mddev->ctime = get_seconds();
4382 mddev->level = info->level;
4383 mddev->clevel[0] = 0;
4384 mddev->size = info->size;
4385 mddev->raid_disks = info->raid_disks;
4386 /* don't set md_minor, it is determined by which /dev/md* was
4389 if (info->state & (1<<MD_SB_CLEAN))
4390 mddev->recovery_cp = MaxSector;
4392 mddev->recovery_cp = 0;
4393 mddev->persistent = ! info->not_persistent;
4394 mddev->external = 0;
4396 mddev->layout = info->layout;
4397 mddev->chunk_size = info->chunk_size;
4399 mddev->max_disks = MD_SB_DISKS;
4401 if (mddev->persistent)
4403 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4405 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4406 mddev->bitmap_offset = 0;
4408 mddev->reshape_position = MaxSector;
4411 * Generate a 128 bit UUID
4413 get_random_bytes(mddev->uuid, 16);
4415 mddev->new_level = mddev->level;
4416 mddev->new_chunk = mddev->chunk_size;
4417 mddev->new_layout = mddev->layout;
4418 mddev->delta_disks = 0;
4423 static int update_size(mddev_t *mddev, unsigned long size)
4427 struct list_head *tmp;
4428 int fit = (size == 0);
4430 if (mddev->pers->resize == NULL)
4432 /* The "size" is the amount of each device that is used.
4433 * This can only make sense for arrays with redundancy.
4434 * linear and raid0 always use whatever space is available
4435 * We can only consider changing the size if no resync
4436 * or reconstruction is happening, and if the new size
4437 * is acceptable. It must fit before the sb_offset or,
4438 * if that is <data_offset, it must fit before the
4439 * size of each device.
4440 * If size is zero, we find the largest size that fits.
4442 if (mddev->sync_thread)
4444 rdev_for_each(rdev, tmp, mddev) {
4446 avail = rdev->size * 2;
4448 if (fit && (size == 0 || size > avail/2))
4450 if (avail < ((sector_t)size << 1))
4453 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4455 struct block_device *bdev;
4457 bdev = bdget_disk(mddev->gendisk, 0);
4459 mutex_lock(&bdev->bd_inode->i_mutex);
4460 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4461 mutex_unlock(&bdev->bd_inode->i_mutex);
4468 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4471 /* change the number of raid disks */
4472 if (mddev->pers->check_reshape == NULL)
4474 if (raid_disks <= 0 ||
4475 raid_disks >= mddev->max_disks)
4477 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4479 mddev->delta_disks = raid_disks - mddev->raid_disks;
4481 rv = mddev->pers->check_reshape(mddev);
4487 * update_array_info is used to change the configuration of an
4489 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4490 * fields in the info are checked against the array.
4491 * Any differences that cannot be handled will cause an error.
4492 * Normally, only one change can be managed at a time.
4494 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4500 /* calculate expected state,ignoring low bits */
4501 if (mddev->bitmap && mddev->bitmap_offset)
4502 state |= (1 << MD_SB_BITMAP_PRESENT);
4504 if (mddev->major_version != info->major_version ||
4505 mddev->minor_version != info->minor_version ||
4506 /* mddev->patch_version != info->patch_version || */
4507 mddev->ctime != info->ctime ||
4508 mddev->level != info->level ||
4509 /* mddev->layout != info->layout || */
4510 !mddev->persistent != info->not_persistent||
4511 mddev->chunk_size != info->chunk_size ||
4512 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4513 ((state^info->state) & 0xfffffe00)
4516 /* Check there is only one change */
4517 if (info->size >= 0 && mddev->size != info->size) cnt++;
4518 if (mddev->raid_disks != info->raid_disks) cnt++;
4519 if (mddev->layout != info->layout) cnt++;
4520 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4521 if (cnt == 0) return 0;
4522 if (cnt > 1) return -EINVAL;
4524 if (mddev->layout != info->layout) {
4526 * we don't need to do anything at the md level, the
4527 * personality will take care of it all.
4529 if (mddev->pers->reconfig == NULL)
4532 return mddev->pers->reconfig(mddev, info->layout, -1);
4534 if (info->size >= 0 && mddev->size != info->size)
4535 rv = update_size(mddev, info->size);
4537 if (mddev->raid_disks != info->raid_disks)
4538 rv = update_raid_disks(mddev, info->raid_disks);
4540 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4541 if (mddev->pers->quiesce == NULL)
4543 if (mddev->recovery || mddev->sync_thread)
4545 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4546 /* add the bitmap */
4549 if (mddev->default_bitmap_offset == 0)
4551 mddev->bitmap_offset = mddev->default_bitmap_offset;
4552 mddev->pers->quiesce(mddev, 1);
4553 rv = bitmap_create(mddev);
4555 bitmap_destroy(mddev);
4556 mddev->pers->quiesce(mddev, 0);
4558 /* remove the bitmap */
4561 if (mddev->bitmap->file)
4563 mddev->pers->quiesce(mddev, 1);
4564 bitmap_destroy(mddev);
4565 mddev->pers->quiesce(mddev, 0);
4566 mddev->bitmap_offset = 0;
4569 md_update_sb(mddev, 1);
4573 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4577 if (mddev->pers == NULL)
4580 rdev = find_rdev(mddev, dev);
4584 md_error(mddev, rdev);
4588 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4590 mddev_t *mddev = bdev->bd_disk->private_data;
4594 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4598 static int md_ioctl(struct inode *inode, struct file *file,
4599 unsigned int cmd, unsigned long arg)
4602 void __user *argp = (void __user *)arg;
4603 mddev_t *mddev = NULL;
4605 if (!capable(CAP_SYS_ADMIN))
4609 * Commands dealing with the RAID driver but not any
4615 err = get_version(argp);
4618 case PRINT_RAID_DEBUG:
4626 autostart_arrays(arg);
4633 * Commands creating/starting a new array:
4636 mddev = inode->i_bdev->bd_disk->private_data;
4643 err = mddev_lock(mddev);
4646 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4653 case SET_ARRAY_INFO:
4655 mdu_array_info_t info;
4657 memset(&info, 0, sizeof(info));
4658 else if (copy_from_user(&info, argp, sizeof(info))) {
4663 err = update_array_info(mddev, &info);
4665 printk(KERN_WARNING "md: couldn't update"
4666 " array info. %d\n", err);
4671 if (!list_empty(&mddev->disks)) {
4673 "md: array %s already has disks!\n",
4678 if (mddev->raid_disks) {
4680 "md: array %s already initialised!\n",
4685 err = set_array_info(mddev, &info);
4687 printk(KERN_WARNING "md: couldn't set"
4688 " array info. %d\n", err);
4698 * Commands querying/configuring an existing array:
4700 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4701 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4702 if ((!mddev->raid_disks && !mddev->external)
4703 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4704 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4705 && cmd != GET_BITMAP_FILE) {
4711 * Commands even a read-only array can execute:
4715 case GET_ARRAY_INFO:
4716 err = get_array_info(mddev, argp);
4719 case GET_BITMAP_FILE:
4720 err = get_bitmap_file(mddev, argp);
4724 err = get_disk_info(mddev, argp);
4727 case RESTART_ARRAY_RW:
4728 err = restart_array(mddev);
4732 err = do_md_stop (mddev, 0);
4736 err = do_md_stop (mddev, 1);
4740 * We have a problem here : there is no easy way to give a CHS
4741 * virtual geometry. We currently pretend that we have a 2 heads
4742 * 4 sectors (with a BIG number of cylinders...). This drives
4743 * dosfs just mad... ;-)
4748 * The remaining ioctls are changing the state of the
4749 * superblock, so we do not allow them on read-only arrays.
4750 * However non-MD ioctls (e.g. get-size) will still come through
4751 * here and hit the 'default' below, so only disallow
4752 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4754 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4755 mddev->ro && mddev->pers) {
4756 if (mddev->ro == 2) {
4758 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4759 md_wakeup_thread(mddev->thread);
4771 mdu_disk_info_t info;
4772 if (copy_from_user(&info, argp, sizeof(info)))
4775 err = add_new_disk(mddev, &info);
4779 case HOT_REMOVE_DISK:
4780 err = hot_remove_disk(mddev, new_decode_dev(arg));
4784 err = hot_add_disk(mddev, new_decode_dev(arg));
4787 case SET_DISK_FAULTY:
4788 err = set_disk_faulty(mddev, new_decode_dev(arg));
4792 err = do_md_run (mddev);
4795 case SET_BITMAP_FILE:
4796 err = set_bitmap_file(mddev, (int)arg);
4806 mddev_unlock(mddev);
4816 static int md_open(struct inode *inode, struct file *file)
4819 * Succeed if we can lock the mddev, which confirms that
4820 * it isn't being stopped right now.
4822 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4825 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4830 mddev_unlock(mddev);
4832 check_disk_change(inode->i_bdev);
4837 static int md_release(struct inode *inode, struct file * file)
4839 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4847 static int md_media_changed(struct gendisk *disk)
4849 mddev_t *mddev = disk->private_data;
4851 return mddev->changed;
4854 static int md_revalidate(struct gendisk *disk)
4856 mddev_t *mddev = disk->private_data;
4861 static struct block_device_operations md_fops =
4863 .owner = THIS_MODULE,
4865 .release = md_release,
4867 .getgeo = md_getgeo,
4868 .media_changed = md_media_changed,
4869 .revalidate_disk= md_revalidate,
4872 static int md_thread(void * arg)
4874 mdk_thread_t *thread = arg;
4877 * md_thread is a 'system-thread', it's priority should be very
4878 * high. We avoid resource deadlocks individually in each
4879 * raid personality. (RAID5 does preallocation) We also use RR and
4880 * the very same RT priority as kswapd, thus we will never get
4881 * into a priority inversion deadlock.
4883 * we definitely have to have equal or higher priority than
4884 * bdflush, otherwise bdflush will deadlock if there are too
4885 * many dirty RAID5 blocks.
4888 allow_signal(SIGKILL);
4889 while (!kthread_should_stop()) {
4891 /* We need to wait INTERRUPTIBLE so that
4892 * we don't add to the load-average.
4893 * That means we need to be sure no signals are
4896 if (signal_pending(current))
4897 flush_signals(current);
4899 wait_event_interruptible_timeout
4901 test_bit(THREAD_WAKEUP, &thread->flags)
4902 || kthread_should_stop(),
4905 clear_bit(THREAD_WAKEUP, &thread->flags);
4907 thread->run(thread->mddev);
4913 void md_wakeup_thread(mdk_thread_t *thread)
4916 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4917 set_bit(THREAD_WAKEUP, &thread->flags);
4918 wake_up(&thread->wqueue);
4922 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4925 mdk_thread_t *thread;
4927 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4931 init_waitqueue_head(&thread->wqueue);
4934 thread->mddev = mddev;
4935 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4936 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4937 if (IS_ERR(thread->tsk)) {
4944 void md_unregister_thread(mdk_thread_t *thread)
4946 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4948 kthread_stop(thread->tsk);
4952 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4959 if (!rdev || test_bit(Faulty, &rdev->flags))
4962 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4964 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4965 __builtin_return_address(0),__builtin_return_address(1),
4966 __builtin_return_address(2),__builtin_return_address(3));
4970 if (!mddev->pers->error_handler)
4972 mddev->pers->error_handler(mddev,rdev);
4973 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4974 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4975 md_wakeup_thread(mddev->thread);
4976 md_new_event_inintr(mddev);
4979 /* seq_file implementation /proc/mdstat */
4981 static void status_unused(struct seq_file *seq)
4985 struct list_head *tmp;
4987 seq_printf(seq, "unused devices: ");
4989 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
4990 char b[BDEVNAME_SIZE];
4992 seq_printf(seq, "%s ",
4993 bdevname(rdev->bdev,b));
4996 seq_printf(seq, "<none>");
4998 seq_printf(seq, "\n");
5002 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5004 sector_t max_blocks, resync, res;
5005 unsigned long dt, db, rt;
5007 unsigned int per_milli;
5009 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5011 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5012 max_blocks = mddev->resync_max_sectors >> 1;
5014 max_blocks = mddev->size;
5017 * Should not happen.
5023 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5024 * in a sector_t, and (max_blocks>>scale) will fit in a
5025 * u32, as those are the requirements for sector_div.
5026 * Thus 'scale' must be at least 10
5029 if (sizeof(sector_t) > sizeof(unsigned long)) {
5030 while ( max_blocks/2 > (1ULL<<(scale+32)))
5033 res = (resync>>scale)*1000;
5034 sector_div(res, (u32)((max_blocks>>scale)+1));
5038 int i, x = per_milli/50, y = 20-x;
5039 seq_printf(seq, "[");
5040 for (i = 0; i < x; i++)
5041 seq_printf(seq, "=");
5042 seq_printf(seq, ">");
5043 for (i = 0; i < y; i++)
5044 seq_printf(seq, ".");
5045 seq_printf(seq, "] ");
5047 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5048 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5050 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5052 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5053 "resync" : "recovery"))),
5054 per_milli/10, per_milli % 10,
5055 (unsigned long long) resync,
5056 (unsigned long long) max_blocks);
5059 * We do not want to overflow, so the order of operands and
5060 * the * 100 / 100 trick are important. We do a +1 to be
5061 * safe against division by zero. We only estimate anyway.
5063 * dt: time from mark until now
5064 * db: blocks written from mark until now
5065 * rt: remaining time
5067 dt = ((jiffies - mddev->resync_mark) / HZ);
5069 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5070 - mddev->resync_mark_cnt;
5071 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5073 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5075 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5078 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5080 struct list_head *tmp;
5090 spin_lock(&all_mddevs_lock);
5091 list_for_each(tmp,&all_mddevs)
5093 mddev = list_entry(tmp, mddev_t, all_mddevs);
5095 spin_unlock(&all_mddevs_lock);
5098 spin_unlock(&all_mddevs_lock);
5100 return (void*)2;/* tail */
5104 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5106 struct list_head *tmp;
5107 mddev_t *next_mddev, *mddev = v;
5113 spin_lock(&all_mddevs_lock);
5115 tmp = all_mddevs.next;
5117 tmp = mddev->all_mddevs.next;
5118 if (tmp != &all_mddevs)
5119 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5121 next_mddev = (void*)2;
5124 spin_unlock(&all_mddevs_lock);
5132 static void md_seq_stop(struct seq_file *seq, void *v)
5136 if (mddev && v != (void*)1 && v != (void*)2)
5140 struct mdstat_info {
5144 static int md_seq_show(struct seq_file *seq, void *v)
5148 struct list_head *tmp2;
5150 struct mdstat_info *mi = seq->private;
5151 struct bitmap *bitmap;
5153 if (v == (void*)1) {
5154 struct mdk_personality *pers;
5155 seq_printf(seq, "Personalities : ");
5156 spin_lock(&pers_lock);
5157 list_for_each_entry(pers, &pers_list, list)
5158 seq_printf(seq, "[%s] ", pers->name);
5160 spin_unlock(&pers_lock);
5161 seq_printf(seq, "\n");
5162 mi->event = atomic_read(&md_event_count);
5165 if (v == (void*)2) {
5170 if (mddev_lock(mddev) < 0)
5173 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5174 seq_printf(seq, "%s : %sactive", mdname(mddev),
5175 mddev->pers ? "" : "in");
5178 seq_printf(seq, " (read-only)");
5180 seq_printf(seq, " (auto-read-only)");
5181 seq_printf(seq, " %s", mddev->pers->name);
5185 rdev_for_each(rdev, tmp2, mddev) {
5186 char b[BDEVNAME_SIZE];
5187 seq_printf(seq, " %s[%d]",
5188 bdevname(rdev->bdev,b), rdev->desc_nr);
5189 if (test_bit(WriteMostly, &rdev->flags))
5190 seq_printf(seq, "(W)");
5191 if (test_bit(Faulty, &rdev->flags)) {
5192 seq_printf(seq, "(F)");
5194 } else if (rdev->raid_disk < 0)
5195 seq_printf(seq, "(S)"); /* spare */
5199 if (!list_empty(&mddev->disks)) {
5201 seq_printf(seq, "\n %llu blocks",
5202 (unsigned long long)mddev->array_size);
5204 seq_printf(seq, "\n %llu blocks",
5205 (unsigned long long)size);
5207 if (mddev->persistent) {
5208 if (mddev->major_version != 0 ||
5209 mddev->minor_version != 90) {
5210 seq_printf(seq," super %d.%d",
5211 mddev->major_version,
5212 mddev->minor_version);
5214 } else if (mddev->external)
5215 seq_printf(seq, " super external:%s",
5216 mddev->metadata_type);
5218 seq_printf(seq, " super non-persistent");
5221 mddev->pers->status (seq, mddev);
5222 seq_printf(seq, "\n ");
5223 if (mddev->pers->sync_request) {
5224 if (mddev->curr_resync > 2) {
5225 status_resync (seq, mddev);
5226 seq_printf(seq, "\n ");
5227 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5228 seq_printf(seq, "\tresync=DELAYED\n ");
5229 else if (mddev->recovery_cp < MaxSector)
5230 seq_printf(seq, "\tresync=PENDING\n ");
5233 seq_printf(seq, "\n ");
5235 if ((bitmap = mddev->bitmap)) {
5236 unsigned long chunk_kb;
5237 unsigned long flags;
5238 spin_lock_irqsave(&bitmap->lock, flags);
5239 chunk_kb = bitmap->chunksize >> 10;
5240 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5242 bitmap->pages - bitmap->missing_pages,
5244 (bitmap->pages - bitmap->missing_pages)
5245 << (PAGE_SHIFT - 10),
5246 chunk_kb ? chunk_kb : bitmap->chunksize,
5247 chunk_kb ? "KB" : "B");
5249 seq_printf(seq, ", file: ");
5250 seq_path(seq, &bitmap->file->f_path, " \t\n");
5253 seq_printf(seq, "\n");
5254 spin_unlock_irqrestore(&bitmap->lock, flags);
5257 seq_printf(seq, "\n");
5259 mddev_unlock(mddev);
5264 static struct seq_operations md_seq_ops = {
5265 .start = md_seq_start,
5266 .next = md_seq_next,
5267 .stop = md_seq_stop,
5268 .show = md_seq_show,
5271 static int md_seq_open(struct inode *inode, struct file *file)
5274 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5278 error = seq_open(file, &md_seq_ops);
5282 struct seq_file *p = file->private_data;
5284 mi->event = atomic_read(&md_event_count);
5289 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5291 struct seq_file *m = filp->private_data;
5292 struct mdstat_info *mi = m->private;
5295 poll_wait(filp, &md_event_waiters, wait);
5297 /* always allow read */
5298 mask = POLLIN | POLLRDNORM;
5300 if (mi->event != atomic_read(&md_event_count))
5301 mask |= POLLERR | POLLPRI;
5305 static const struct file_operations md_seq_fops = {
5306 .owner = THIS_MODULE,
5307 .open = md_seq_open,
5309 .llseek = seq_lseek,
5310 .release = seq_release_private,
5311 .poll = mdstat_poll,
5314 int register_md_personality(struct mdk_personality *p)
5316 spin_lock(&pers_lock);
5317 list_add_tail(&p->list, &pers_list);
5318 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5319 spin_unlock(&pers_lock);
5323 int unregister_md_personality(struct mdk_personality *p)
5325 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5326 spin_lock(&pers_lock);
5327 list_del_init(&p->list);
5328 spin_unlock(&pers_lock);
5332 static int is_mddev_idle(mddev_t *mddev)
5335 struct list_head *tmp;
5340 rdev_for_each(rdev, tmp, mddev) {
5341 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5342 curr_events = disk_stat_read(disk, sectors[0]) +
5343 disk_stat_read(disk, sectors[1]) -
5344 atomic_read(&disk->sync_io);
5345 /* sync IO will cause sync_io to increase before the disk_stats
5346 * as sync_io is counted when a request starts, and
5347 * disk_stats is counted when it completes.
5348 * So resync activity will cause curr_events to be smaller than
5349 * when there was no such activity.
5350 * non-sync IO will cause disk_stat to increase without
5351 * increasing sync_io so curr_events will (eventually)
5352 * be larger than it was before. Once it becomes
5353 * substantially larger, the test below will cause
5354 * the array to appear non-idle, and resync will slow
5356 * If there is a lot of outstanding resync activity when
5357 * we set last_event to curr_events, then all that activity
5358 * completing might cause the array to appear non-idle
5359 * and resync will be slowed down even though there might
5360 * not have been non-resync activity. This will only
5361 * happen once though. 'last_events' will soon reflect
5362 * the state where there is little or no outstanding
5363 * resync requests, and further resync activity will
5364 * always make curr_events less than last_events.
5367 if (curr_events - rdev->last_events > 4096) {
5368 rdev->last_events = curr_events;
5375 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5377 /* another "blocks" (512byte) blocks have been synced */
5378 atomic_sub(blocks, &mddev->recovery_active);
5379 wake_up(&mddev->recovery_wait);
5381 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5382 md_wakeup_thread(mddev->thread);
5383 // stop recovery, signal do_sync ....
5388 /* md_write_start(mddev, bi)
5389 * If we need to update some array metadata (e.g. 'active' flag
5390 * in superblock) before writing, schedule a superblock update
5391 * and wait for it to complete.
5393 void md_write_start(mddev_t *mddev, struct bio *bi)
5395 if (bio_data_dir(bi) != WRITE)
5398 BUG_ON(mddev->ro == 1);
5399 if (mddev->ro == 2) {
5400 /* need to switch to read/write */
5402 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5403 md_wakeup_thread(mddev->thread);
5404 md_wakeup_thread(mddev->sync_thread);
5406 atomic_inc(&mddev->writes_pending);
5407 if (mddev->safemode == 1)
5408 mddev->safemode = 0;
5409 if (mddev->in_sync) {
5410 spin_lock_irq(&mddev->write_lock);
5411 if (mddev->in_sync) {
5413 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5414 md_wakeup_thread(mddev->thread);
5416 spin_unlock_irq(&mddev->write_lock);
5418 wait_event(mddev->sb_wait, mddev->flags==0);
5421 void md_write_end(mddev_t *mddev)
5423 if (atomic_dec_and_test(&mddev->writes_pending)) {
5424 if (mddev->safemode == 2)
5425 md_wakeup_thread(mddev->thread);
5426 else if (mddev->safemode_delay)
5427 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5431 /* md_allow_write(mddev)
5432 * Calling this ensures that the array is marked 'active' so that writes
5433 * may proceed without blocking. It is important to call this before
5434 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5435 * Must be called with mddev_lock held.
5437 void md_allow_write(mddev_t *mddev)
5444 spin_lock_irq(&mddev->write_lock);
5445 if (mddev->in_sync) {
5447 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5448 if (mddev->safemode_delay &&
5449 mddev->safemode == 0)
5450 mddev->safemode = 1;
5451 spin_unlock_irq(&mddev->write_lock);
5452 md_update_sb(mddev, 0);
5454 spin_unlock_irq(&mddev->write_lock);
5456 EXPORT_SYMBOL_GPL(md_allow_write);
5458 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5460 #define SYNC_MARKS 10
5461 #define SYNC_MARK_STEP (3*HZ)
5462 void md_do_sync(mddev_t *mddev)
5465 unsigned int currspeed = 0,
5467 sector_t max_sectors,j, io_sectors;
5468 unsigned long mark[SYNC_MARKS];
5469 sector_t mark_cnt[SYNC_MARKS];
5471 struct list_head *tmp;
5472 sector_t last_check;
5474 struct list_head *rtmp;
5478 /* just incase thread restarts... */
5479 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5481 if (mddev->ro) /* never try to sync a read-only array */
5484 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5485 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5486 desc = "data-check";
5487 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5488 desc = "requested-resync";
5491 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5496 /* we overload curr_resync somewhat here.
5497 * 0 == not engaged in resync at all
5498 * 2 == checking that there is no conflict with another sync
5499 * 1 == like 2, but have yielded to allow conflicting resync to
5501 * other == active in resync - this many blocks
5503 * Before starting a resync we must have set curr_resync to
5504 * 2, and then checked that every "conflicting" array has curr_resync
5505 * less than ours. When we find one that is the same or higher
5506 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5507 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5508 * This will mean we have to start checking from the beginning again.
5513 mddev->curr_resync = 2;
5516 if (kthread_should_stop()) {
5517 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5520 for_each_mddev(mddev2, tmp) {
5521 if (mddev2 == mddev)
5523 if (mddev2->curr_resync &&
5524 match_mddev_units(mddev,mddev2)) {
5526 if (mddev < mddev2 && mddev->curr_resync == 2) {
5527 /* arbitrarily yield */
5528 mddev->curr_resync = 1;
5529 wake_up(&resync_wait);
5531 if (mddev > mddev2 && mddev->curr_resync == 1)
5532 /* no need to wait here, we can wait the next
5533 * time 'round when curr_resync == 2
5536 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5537 if (!kthread_should_stop() &&
5538 mddev2->curr_resync >= mddev->curr_resync) {
5539 printk(KERN_INFO "md: delaying %s of %s"
5540 " until %s has finished (they"
5541 " share one or more physical units)\n",
5542 desc, mdname(mddev), mdname(mddev2));
5545 finish_wait(&resync_wait, &wq);
5548 finish_wait(&resync_wait, &wq);
5551 } while (mddev->curr_resync < 2);
5554 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5555 /* resync follows the size requested by the personality,
5556 * which defaults to physical size, but can be virtual size
5558 max_sectors = mddev->resync_max_sectors;
5559 mddev->resync_mismatches = 0;
5560 /* we don't use the checkpoint if there's a bitmap */
5561 if (!mddev->bitmap &&
5562 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5563 j = mddev->recovery_cp;
5564 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5565 max_sectors = mddev->size << 1;
5567 /* recovery follows the physical size of devices */
5568 max_sectors = mddev->size << 1;
5570 rdev_for_each(rdev, rtmp, mddev)
5571 if (rdev->raid_disk >= 0 &&
5572 !test_bit(Faulty, &rdev->flags) &&
5573 !test_bit(In_sync, &rdev->flags) &&
5574 rdev->recovery_offset < j)
5575 j = rdev->recovery_offset;
5578 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5579 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5580 " %d KB/sec/disk.\n", speed_min(mddev));
5581 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5582 "(but not more than %d KB/sec) for %s.\n",
5583 speed_max(mddev), desc);
5585 is_mddev_idle(mddev); /* this also initializes IO event counters */
5588 for (m = 0; m < SYNC_MARKS; m++) {
5590 mark_cnt[m] = io_sectors;
5593 mddev->resync_mark = mark[last_mark];
5594 mddev->resync_mark_cnt = mark_cnt[last_mark];
5597 * Tune reconstruction:
5599 window = 32*(PAGE_SIZE/512);
5600 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5601 window/2,(unsigned long long) max_sectors/2);
5603 atomic_set(&mddev->recovery_active, 0);
5604 init_waitqueue_head(&mddev->recovery_wait);
5609 "md: resuming %s of %s from checkpoint.\n",
5610 desc, mdname(mddev));
5611 mddev->curr_resync = j;
5614 while (j < max_sectors) {
5618 if (j >= mddev->resync_max) {
5619 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5620 wait_event(mddev->recovery_wait,
5621 mddev->resync_max > j
5622 || kthread_should_stop());
5624 if (kthread_should_stop())
5626 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5627 currspeed < speed_min(mddev));
5629 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5633 if (!skipped) { /* actual IO requested */
5634 io_sectors += sectors;
5635 atomic_add(sectors, &mddev->recovery_active);
5639 if (j>1) mddev->curr_resync = j;
5640 mddev->curr_mark_cnt = io_sectors;
5641 if (last_check == 0)
5642 /* this is the earliers that rebuilt will be
5643 * visible in /proc/mdstat
5645 md_new_event(mddev);
5647 if (last_check + window > io_sectors || j == max_sectors)
5650 last_check = io_sectors;
5652 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5653 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5657 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5659 int next = (last_mark+1) % SYNC_MARKS;
5661 mddev->resync_mark = mark[next];
5662 mddev->resync_mark_cnt = mark_cnt[next];
5663 mark[next] = jiffies;
5664 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5669 if (kthread_should_stop())
5674 * this loop exits only if either when we are slower than
5675 * the 'hard' speed limit, or the system was IO-idle for
5677 * the system might be non-idle CPU-wise, but we only care
5678 * about not overloading the IO subsystem. (things like an
5679 * e2fsck being done on the RAID array should execute fast)
5681 blk_unplug(mddev->queue);
5684 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5685 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5687 if (currspeed > speed_min(mddev)) {
5688 if ((currspeed > speed_max(mddev)) ||
5689 !is_mddev_idle(mddev)) {
5695 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5697 * this also signals 'finished resyncing' to md_stop
5700 blk_unplug(mddev->queue);
5702 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5704 /* tell personality that we are finished */
5705 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5707 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5708 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5709 mddev->curr_resync > 2) {
5710 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5711 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5712 if (mddev->curr_resync >= mddev->recovery_cp) {
5714 "md: checkpointing %s of %s.\n",
5715 desc, mdname(mddev));
5716 mddev->recovery_cp = mddev->curr_resync;
5719 mddev->recovery_cp = MaxSector;
5721 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5722 mddev->curr_resync = MaxSector;
5723 rdev_for_each(rdev, rtmp, mddev)
5724 if (rdev->raid_disk >= 0 &&
5725 !test_bit(Faulty, &rdev->flags) &&
5726 !test_bit(In_sync, &rdev->flags) &&
5727 rdev->recovery_offset < mddev->curr_resync)
5728 rdev->recovery_offset = mddev->curr_resync;
5731 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5734 mddev->curr_resync = 0;
5735 mddev->resync_max = MaxSector;
5736 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5737 wake_up(&resync_wait);
5738 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5739 md_wakeup_thread(mddev->thread);
5744 * got a signal, exit.
5747 "md: md_do_sync() got signal ... exiting\n");
5748 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5752 EXPORT_SYMBOL_GPL(md_do_sync);
5755 static int remove_and_add_spares(mddev_t *mddev)
5758 struct list_head *rtmp;
5761 rdev_for_each(rdev, rtmp, mddev)
5762 if (rdev->raid_disk >= 0 &&
5764 (test_bit(Faulty, &rdev->flags) ||
5765 ! test_bit(In_sync, &rdev->flags)) &&
5766 atomic_read(&rdev->nr_pending)==0) {
5767 if (mddev->pers->hot_remove_disk(
5768 mddev, rdev->raid_disk)==0) {
5770 sprintf(nm,"rd%d", rdev->raid_disk);
5771 sysfs_remove_link(&mddev->kobj, nm);
5772 rdev->raid_disk = -1;
5776 if (mddev->degraded) {
5777 rdev_for_each(rdev, rtmp, mddev)
5778 if (rdev->raid_disk < 0
5779 && !test_bit(Faulty, &rdev->flags)) {
5780 rdev->recovery_offset = 0;
5781 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5783 sprintf(nm, "rd%d", rdev->raid_disk);
5784 if (sysfs_create_link(&mddev->kobj,
5787 "md: cannot register "
5791 md_new_event(mddev);
5799 * This routine is regularly called by all per-raid-array threads to
5800 * deal with generic issues like resync and super-block update.
5801 * Raid personalities that don't have a thread (linear/raid0) do not
5802 * need this as they never do any recovery or update the superblock.
5804 * It does not do any resync itself, but rather "forks" off other threads
5805 * to do that as needed.
5806 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5807 * "->recovery" and create a thread at ->sync_thread.
5808 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5809 * and wakeups up this thread which will reap the thread and finish up.
5810 * This thread also removes any faulty devices (with nr_pending == 0).
5812 * The overall approach is:
5813 * 1/ if the superblock needs updating, update it.
5814 * 2/ If a recovery thread is running, don't do anything else.
5815 * 3/ If recovery has finished, clean up, possibly marking spares active.
5816 * 4/ If there are any faulty devices, remove them.
5817 * 5/ If array is degraded, try to add spares devices
5818 * 6/ If array has spares or is not in-sync, start a resync thread.
5820 void md_check_recovery(mddev_t *mddev)
5823 struct list_head *rtmp;
5827 bitmap_daemon_work(mddev->bitmap);
5832 if (signal_pending(current)) {
5833 if (mddev->pers->sync_request && !mddev->external) {
5834 printk(KERN_INFO "md: %s in immediate safe mode\n",
5836 mddev->safemode = 2;
5838 flush_signals(current);
5842 (mddev->flags && !mddev->external) ||
5843 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5844 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5845 (mddev->external == 0 && mddev->safemode == 1) ||
5846 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5847 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5851 if (mddev_trylock(mddev)) {
5854 if (!mddev->external) {
5855 spin_lock_irq(&mddev->write_lock);
5856 if (mddev->safemode &&
5857 !atomic_read(&mddev->writes_pending) &&
5859 mddev->recovery_cp == MaxSector) {
5861 if (mddev->persistent)
5862 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5864 if (mddev->safemode == 1)
5865 mddev->safemode = 0;
5866 spin_unlock_irq(&mddev->write_lock);
5870 md_update_sb(mddev, 0);
5873 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5874 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5875 /* resync/recovery still happening */
5876 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5879 if (mddev->sync_thread) {
5880 /* resync has finished, collect result */
5881 md_unregister_thread(mddev->sync_thread);
5882 mddev->sync_thread = NULL;
5883 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5884 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5886 /* activate any spares */
5887 mddev->pers->spare_active(mddev);
5889 md_update_sb(mddev, 1);
5891 /* if array is no-longer degraded, then any saved_raid_disk
5892 * information must be scrapped
5894 if (!mddev->degraded)
5895 rdev_for_each(rdev, rtmp, mddev)
5896 rdev->saved_raid_disk = -1;
5898 mddev->recovery = 0;
5899 /* flag recovery needed just to double check */
5900 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5901 md_new_event(mddev);
5904 /* Clear some bits that don't mean anything, but
5907 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5908 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5909 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5910 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5912 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5914 /* no recovery is running.
5915 * remove any failed drives, then
5916 * add spares if possible.
5917 * Spare are also removed and re-added, to allow
5918 * the personality to fail the re-add.
5921 if (mddev->reshape_position != MaxSector) {
5922 if (mddev->pers->check_reshape(mddev) != 0)
5923 /* Cannot proceed */
5925 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5926 } else if ((spares = remove_and_add_spares(mddev))) {
5927 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5928 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5929 } else if (mddev->recovery_cp < MaxSector) {
5930 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5931 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5932 /* nothing to be done ... */
5935 if (mddev->pers->sync_request) {
5936 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5937 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5938 /* We are adding a device or devices to an array
5939 * which has the bitmap stored on all devices.
5940 * So make sure all bitmap pages get written
5942 bitmap_write_all(mddev->bitmap);
5944 mddev->sync_thread = md_register_thread(md_do_sync,
5947 if (!mddev->sync_thread) {
5948 printk(KERN_ERR "%s: could not start resync"
5951 /* leave the spares where they are, it shouldn't hurt */
5952 mddev->recovery = 0;
5954 md_wakeup_thread(mddev->sync_thread);
5955 md_new_event(mddev);
5958 mddev_unlock(mddev);
5962 static int md_notify_reboot(struct notifier_block *this,
5963 unsigned long code, void *x)
5965 struct list_head *tmp;
5968 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5970 printk(KERN_INFO "md: stopping all md devices.\n");
5972 for_each_mddev(mddev, tmp)
5973 if (mddev_trylock(mddev)) {
5974 do_md_stop (mddev, 1);
5975 mddev_unlock(mddev);
5978 * certain more exotic SCSI devices are known to be
5979 * volatile wrt too early system reboots. While the
5980 * right place to handle this issue is the given
5981 * driver, we do want to have a safe RAID driver ...
5988 static struct notifier_block md_notifier = {
5989 .notifier_call = md_notify_reboot,
5991 .priority = INT_MAX, /* before any real devices */
5994 static void md_geninit(void)
5996 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5998 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6001 static int __init md_init(void)
6003 if (register_blkdev(MAJOR_NR, "md"))
6005 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6006 unregister_blkdev(MAJOR_NR, "md");
6009 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6010 md_probe, NULL, NULL);
6011 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6012 md_probe, NULL, NULL);
6014 register_reboot_notifier(&md_notifier);
6015 raid_table_header = register_sysctl_table(raid_root_table);
6025 * Searches all registered partitions for autorun RAID arrays
6029 static LIST_HEAD(all_detected_devices);
6030 struct detected_devices_node {
6031 struct list_head list;
6035 void md_autodetect_dev(dev_t dev)
6037 struct detected_devices_node *node_detected_dev;
6039 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6040 if (node_detected_dev) {
6041 node_detected_dev->dev = dev;
6042 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6044 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6045 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6050 static void autostart_arrays(int part)
6053 struct detected_devices_node *node_detected_dev;
6055 int i_scanned, i_passed;
6060 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6062 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6064 node_detected_dev = list_entry(all_detected_devices.next,
6065 struct detected_devices_node, list);
6066 list_del(&node_detected_dev->list);
6067 dev = node_detected_dev->dev;
6068 kfree(node_detected_dev);
6069 rdev = md_import_device(dev,0, 90);
6073 if (test_bit(Faulty, &rdev->flags)) {
6077 set_bit(AutoDetected, &rdev->flags);
6078 list_add(&rdev->same_set, &pending_raid_disks);
6082 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6083 i_scanned, i_passed);
6085 autorun_devices(part);
6088 #endif /* !MODULE */
6090 static __exit void md_exit(void)
6093 struct list_head *tmp;
6095 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6096 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6098 unregister_blkdev(MAJOR_NR,"md");
6099 unregister_blkdev(mdp_major, "mdp");
6100 unregister_reboot_notifier(&md_notifier);
6101 unregister_sysctl_table(raid_table_header);
6102 remove_proc_entry("mdstat", NULL);
6103 for_each_mddev(mddev, tmp) {
6104 struct gendisk *disk = mddev->gendisk;
6107 export_array(mddev);
6110 mddev->gendisk = NULL;
6115 subsys_initcall(md_init);
6116 module_exit(md_exit)
6118 static int get_ro(char *buffer, struct kernel_param *kp)
6120 return sprintf(buffer, "%d", start_readonly);
6122 static int set_ro(const char *val, struct kernel_param *kp)
6125 int num = simple_strtoul(val, &e, 10);
6126 if (*val && (*e == '\0' || *e == '\n')) {
6127 start_readonly = num;
6133 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6134 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6137 EXPORT_SYMBOL(register_md_personality);
6138 EXPORT_SYMBOL(unregister_md_personality);
6139 EXPORT_SYMBOL(md_error);
6140 EXPORT_SYMBOL(md_done_sync);
6141 EXPORT_SYMBOL(md_write_start);
6142 EXPORT_SYMBOL(md_write_end);
6143 EXPORT_SYMBOL(md_register_thread);
6144 EXPORT_SYMBOL(md_unregister_thread);
6145 EXPORT_SYMBOL(md_wakeup_thread);
6146 EXPORT_SYMBOL(md_check_recovery);
6147 MODULE_LICENSE("GPL");
6149 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);