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/config.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/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.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);
76 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
77 * is 1000 KB/sec, so the extra system load does not show up that much.
78 * Increase it if you want to have more _guaranteed_ speed. Note that
79 * the RAID driver will use the maximum available bandwidth if the IO
80 * subsystem is idle. There is also an 'absolute maximum' reconstruction
81 * speed limit - in case reconstruction slows down your system despite
84 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
85 * or /sys/block/mdX/md/sync_speed_{min,max}
88 static int sysctl_speed_limit_min = 1000;
89 static int sysctl_speed_limit_max = 200000;
90 static inline int speed_min(mddev_t *mddev)
92 return mddev->sync_speed_min ?
93 mddev->sync_speed_min : sysctl_speed_limit_min;
96 static inline int speed_max(mddev_t *mddev)
98 return mddev->sync_speed_max ?
99 mddev->sync_speed_max : sysctl_speed_limit_max;
102 static struct ctl_table_header *raid_table_header;
104 static ctl_table raid_table[] = {
106 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
107 .procname = "speed_limit_min",
108 .data = &sysctl_speed_limit_min,
109 .maxlen = sizeof(int),
111 .proc_handler = &proc_dointvec,
114 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
115 .procname = "speed_limit_max",
116 .data = &sysctl_speed_limit_max,
117 .maxlen = sizeof(int),
119 .proc_handler = &proc_dointvec,
124 static ctl_table raid_dir_table[] = {
126 .ctl_name = DEV_RAID,
135 static ctl_table raid_root_table[] = {
141 .child = raid_dir_table,
146 static struct block_device_operations md_fops;
148 static int start_readonly;
151 * We have a system wide 'event count' that is incremented
152 * on any 'interesting' event, and readers of /proc/mdstat
153 * can use 'poll' or 'select' to find out when the event
157 * start array, stop array, error, add device, remove device,
158 * start build, activate spare
160 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
161 static atomic_t md_event_count;
162 void md_new_event(mddev_t *mddev)
164 atomic_inc(&md_event_count);
165 wake_up(&md_event_waiters);
167 EXPORT_SYMBOL_GPL(md_new_event);
170 * Enables to iterate over all existing md arrays
171 * all_mddevs_lock protects this list.
173 static LIST_HEAD(all_mddevs);
174 static DEFINE_SPINLOCK(all_mddevs_lock);
178 * iterates through all used mddevs in the system.
179 * We take care to grab the all_mddevs_lock whenever navigating
180 * the list, and to always hold a refcount when unlocked.
181 * Any code which breaks out of this loop while own
182 * a reference to the current mddev and must mddev_put it.
184 #define ITERATE_MDDEV(mddev,tmp) \
186 for (({ spin_lock(&all_mddevs_lock); \
187 tmp = all_mddevs.next; \
189 ({ if (tmp != &all_mddevs) \
190 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
191 spin_unlock(&all_mddevs_lock); \
192 if (mddev) mddev_put(mddev); \
193 mddev = list_entry(tmp, mddev_t, all_mddevs); \
194 tmp != &all_mddevs;}); \
195 ({ spin_lock(&all_mddevs_lock); \
200 static int md_fail_request (request_queue_t *q, struct bio *bio)
202 bio_io_error(bio, bio->bi_size);
206 static inline mddev_t *mddev_get(mddev_t *mddev)
208 atomic_inc(&mddev->active);
212 static void mddev_put(mddev_t *mddev)
214 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
216 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
217 list_del(&mddev->all_mddevs);
219 blk_cleanup_queue(mddev->queue);
220 /* that also blocks */
221 kobject_unregister(&mddev->kobj);
222 /* result blows... */
224 spin_unlock(&all_mddevs_lock);
227 static mddev_t * mddev_find(dev_t unit)
229 mddev_t *mddev, *new = NULL;
232 spin_lock(&all_mddevs_lock);
233 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
234 if (mddev->unit == unit) {
236 spin_unlock(&all_mddevs_lock);
242 list_add(&new->all_mddevs, &all_mddevs);
243 spin_unlock(&all_mddevs_lock);
246 spin_unlock(&all_mddevs_lock);
248 new = kzalloc(sizeof(*new), GFP_KERNEL);
253 if (MAJOR(unit) == MD_MAJOR)
254 new->md_minor = MINOR(unit);
256 new->md_minor = MINOR(unit) >> MdpMinorShift;
258 init_MUTEX(&new->reconfig_sem);
259 INIT_LIST_HEAD(&new->disks);
260 INIT_LIST_HEAD(&new->all_mddevs);
261 init_timer(&new->safemode_timer);
262 atomic_set(&new->active, 1);
263 spin_lock_init(&new->write_lock);
264 init_waitqueue_head(&new->sb_wait);
266 new->queue = blk_alloc_queue(GFP_KERNEL);
271 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
273 blk_queue_make_request(new->queue, md_fail_request);
278 static inline int mddev_lock(mddev_t * mddev)
280 return down_interruptible(&mddev->reconfig_sem);
283 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
285 down(&mddev->reconfig_sem);
288 static inline int mddev_trylock(mddev_t * mddev)
290 return down_trylock(&mddev->reconfig_sem);
293 static inline void mddev_unlock(mddev_t * mddev)
295 up(&mddev->reconfig_sem);
297 md_wakeup_thread(mddev->thread);
300 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
303 struct list_head *tmp;
305 ITERATE_RDEV(mddev,rdev,tmp) {
306 if (rdev->desc_nr == nr)
312 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
314 struct list_head *tmp;
317 ITERATE_RDEV(mddev,rdev,tmp) {
318 if (rdev->bdev->bd_dev == dev)
324 static struct mdk_personality *find_pers(int level, char *clevel)
326 struct mdk_personality *pers;
327 list_for_each_entry(pers, &pers_list, list) {
328 if (level != LEVEL_NONE && pers->level == level)
330 if (strcmp(pers->name, clevel)==0)
336 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
338 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
339 return MD_NEW_SIZE_BLOCKS(size);
342 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
346 size = rdev->sb_offset;
349 size &= ~((sector_t)chunk_size/1024 - 1);
353 static int alloc_disk_sb(mdk_rdev_t * rdev)
358 rdev->sb_page = alloc_page(GFP_KERNEL);
359 if (!rdev->sb_page) {
360 printk(KERN_ALERT "md: out of memory.\n");
367 static void free_disk_sb(mdk_rdev_t * rdev)
370 put_page(rdev->sb_page);
372 rdev->sb_page = NULL;
379 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
381 mdk_rdev_t *rdev = bio->bi_private;
382 mddev_t *mddev = rdev->mddev;
386 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
387 md_error(mddev, rdev);
389 if (atomic_dec_and_test(&mddev->pending_writes))
390 wake_up(&mddev->sb_wait);
395 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
397 struct bio *bio2 = bio->bi_private;
398 mdk_rdev_t *rdev = bio2->bi_private;
399 mddev_t *mddev = rdev->mddev;
403 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
404 error == -EOPNOTSUPP) {
406 /* barriers don't appear to be supported :-( */
407 set_bit(BarriersNotsupp, &rdev->flags);
408 mddev->barriers_work = 0;
409 spin_lock_irqsave(&mddev->write_lock, flags);
410 bio2->bi_next = mddev->biolist;
411 mddev->biolist = bio2;
412 spin_unlock_irqrestore(&mddev->write_lock, flags);
413 wake_up(&mddev->sb_wait);
418 bio->bi_private = rdev;
419 return super_written(bio, bytes_done, error);
422 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
423 sector_t sector, int size, struct page *page)
425 /* write first size bytes of page to sector of rdev
426 * Increment mddev->pending_writes before returning
427 * and decrement it on completion, waking up sb_wait
428 * if zero is reached.
429 * If an error occurred, call md_error
431 * As we might need to resubmit the request if BIO_RW_BARRIER
432 * causes ENOTSUPP, we allocate a spare bio...
434 struct bio *bio = bio_alloc(GFP_NOIO, 1);
435 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
437 bio->bi_bdev = rdev->bdev;
438 bio->bi_sector = sector;
439 bio_add_page(bio, page, size, 0);
440 bio->bi_private = rdev;
441 bio->bi_end_io = super_written;
444 atomic_inc(&mddev->pending_writes);
445 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
447 rw |= (1<<BIO_RW_BARRIER);
448 rbio = bio_clone(bio, GFP_NOIO);
449 rbio->bi_private = bio;
450 rbio->bi_end_io = super_written_barrier;
451 submit_bio(rw, rbio);
456 void md_super_wait(mddev_t *mddev)
458 /* wait for all superblock writes that were scheduled to complete.
459 * if any had to be retried (due to BARRIER problems), retry them
463 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
464 if (atomic_read(&mddev->pending_writes)==0)
466 while (mddev->biolist) {
468 spin_lock_irq(&mddev->write_lock);
469 bio = mddev->biolist;
470 mddev->biolist = bio->bi_next ;
472 spin_unlock_irq(&mddev->write_lock);
473 submit_bio(bio->bi_rw, bio);
477 finish_wait(&mddev->sb_wait, &wq);
480 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
485 complete((struct completion*)bio->bi_private);
489 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
490 struct page *page, int rw)
492 struct bio *bio = bio_alloc(GFP_NOIO, 1);
493 struct completion event;
496 rw |= (1 << BIO_RW_SYNC);
499 bio->bi_sector = sector;
500 bio_add_page(bio, page, size, 0);
501 init_completion(&event);
502 bio->bi_private = &event;
503 bio->bi_end_io = bi_complete;
505 wait_for_completion(&event);
507 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
511 EXPORT_SYMBOL_GPL(sync_page_io);
513 static int read_disk_sb(mdk_rdev_t * rdev, int size)
515 char b[BDEVNAME_SIZE];
516 if (!rdev->sb_page) {
524 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
530 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
531 bdevname(rdev->bdev,b));
535 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
537 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
538 (sb1->set_uuid1 == sb2->set_uuid1) &&
539 (sb1->set_uuid2 == sb2->set_uuid2) &&
540 (sb1->set_uuid3 == sb2->set_uuid3))
548 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
551 mdp_super_t *tmp1, *tmp2;
553 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
554 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
556 if (!tmp1 || !tmp2) {
558 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
566 * nr_disks is not constant
571 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
582 static unsigned int calc_sb_csum(mdp_super_t * sb)
584 unsigned int disk_csum, csum;
586 disk_csum = sb->sb_csum;
588 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
589 sb->sb_csum = disk_csum;
595 * Handle superblock details.
596 * We want to be able to handle multiple superblock formats
597 * so we have a common interface to them all, and an array of
598 * different handlers.
599 * We rely on user-space to write the initial superblock, and support
600 * reading and updating of superblocks.
601 * Interface methods are:
602 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
603 * loads and validates a superblock on dev.
604 * if refdev != NULL, compare superblocks on both devices
606 * 0 - dev has a superblock that is compatible with refdev
607 * 1 - dev has a superblock that is compatible and newer than refdev
608 * so dev should be used as the refdev in future
609 * -EINVAL superblock incompatible or invalid
610 * -othererror e.g. -EIO
612 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
613 * Verify that dev is acceptable into mddev.
614 * The first time, mddev->raid_disks will be 0, and data from
615 * dev should be merged in. Subsequent calls check that dev
616 * is new enough. Return 0 or -EINVAL
618 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
619 * Update the superblock for rdev with data in mddev
620 * This does not write to disc.
626 struct module *owner;
627 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
628 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
629 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
633 * load_super for 0.90.0
635 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
637 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
643 * Calculate the position of the superblock,
644 * it's at the end of the disk.
646 * It also happens to be a multiple of 4Kb.
648 sb_offset = calc_dev_sboffset(rdev->bdev);
649 rdev->sb_offset = sb_offset;
651 ret = read_disk_sb(rdev, MD_SB_BYTES);
656 bdevname(rdev->bdev, b);
657 sb = (mdp_super_t*)page_address(rdev->sb_page);
659 if (sb->md_magic != MD_SB_MAGIC) {
660 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
665 if (sb->major_version != 0 ||
666 sb->minor_version < 90 ||
667 sb->minor_version > 91) {
668 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
669 sb->major_version, sb->minor_version,
674 if (sb->raid_disks <= 0)
677 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
678 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
683 rdev->preferred_minor = sb->md_minor;
684 rdev->data_offset = 0;
685 rdev->sb_size = MD_SB_BYTES;
687 if (sb->level == LEVEL_MULTIPATH)
690 rdev->desc_nr = sb->this_disk.number;
696 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
697 if (!uuid_equal(refsb, sb)) {
698 printk(KERN_WARNING "md: %s has different UUID to %s\n",
699 b, bdevname(refdev->bdev,b2));
702 if (!sb_equal(refsb, sb)) {
703 printk(KERN_WARNING "md: %s has same UUID"
704 " but different superblock to %s\n",
705 b, bdevname(refdev->bdev, b2));
709 ev2 = md_event(refsb);
715 rdev->size = calc_dev_size(rdev, sb->chunk_size);
717 if (rdev->size < sb->size && sb->level > 1)
718 /* "this cannot possibly happen" ... */
726 * validate_super for 0.90.0
728 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
731 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
733 rdev->raid_disk = -1;
735 if (mddev->raid_disks == 0) {
736 mddev->major_version = 0;
737 mddev->minor_version = sb->minor_version;
738 mddev->patch_version = sb->patch_version;
739 mddev->persistent = ! sb->not_persistent;
740 mddev->chunk_size = sb->chunk_size;
741 mddev->ctime = sb->ctime;
742 mddev->utime = sb->utime;
743 mddev->level = sb->level;
744 mddev->clevel[0] = 0;
745 mddev->layout = sb->layout;
746 mddev->raid_disks = sb->raid_disks;
747 mddev->size = sb->size;
748 mddev->events = md_event(sb);
749 mddev->bitmap_offset = 0;
750 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
752 if (mddev->minor_version >= 91) {
753 mddev->reshape_position = sb->reshape_position;
754 mddev->delta_disks = sb->delta_disks;
755 mddev->new_level = sb->new_level;
756 mddev->new_layout = sb->new_layout;
757 mddev->new_chunk = sb->new_chunk;
759 mddev->reshape_position = MaxSector;
760 mddev->delta_disks = 0;
761 mddev->new_level = mddev->level;
762 mddev->new_layout = mddev->layout;
763 mddev->new_chunk = mddev->chunk_size;
766 if (sb->state & (1<<MD_SB_CLEAN))
767 mddev->recovery_cp = MaxSector;
769 if (sb->events_hi == sb->cp_events_hi &&
770 sb->events_lo == sb->cp_events_lo) {
771 mddev->recovery_cp = sb->recovery_cp;
773 mddev->recovery_cp = 0;
776 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
777 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
778 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
779 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
781 mddev->max_disks = MD_SB_DISKS;
783 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
784 mddev->bitmap_file == NULL) {
785 if (mddev->level != 1 && mddev->level != 4
786 && mddev->level != 5 && mddev->level != 6
787 && mddev->level != 10) {
788 /* FIXME use a better test */
789 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
792 mddev->bitmap_offset = mddev->default_bitmap_offset;
795 } else if (mddev->pers == NULL) {
796 /* Insist on good event counter while assembling */
797 __u64 ev1 = md_event(sb);
799 if (ev1 < mddev->events)
801 } else if (mddev->bitmap) {
802 /* if adding to array with a bitmap, then we can accept an
803 * older device ... but not too old.
805 __u64 ev1 = md_event(sb);
806 if (ev1 < mddev->bitmap->events_cleared)
808 } else /* just a hot-add of a new device, leave raid_disk at -1 */
811 if (mddev->level != LEVEL_MULTIPATH) {
812 desc = sb->disks + rdev->desc_nr;
814 if (desc->state & (1<<MD_DISK_FAULTY))
815 set_bit(Faulty, &rdev->flags);
816 else if (desc->state & (1<<MD_DISK_SYNC) &&
817 desc->raid_disk < mddev->raid_disks) {
818 set_bit(In_sync, &rdev->flags);
819 rdev->raid_disk = desc->raid_disk;
821 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
822 set_bit(WriteMostly, &rdev->flags);
823 } else /* MULTIPATH are always insync */
824 set_bit(In_sync, &rdev->flags);
829 * sync_super for 0.90.0
831 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
834 struct list_head *tmp;
836 int next_spare = mddev->raid_disks;
839 /* make rdev->sb match mddev data..
842 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
843 * 3/ any empty disks < next_spare become removed
845 * disks[0] gets initialised to REMOVED because
846 * we cannot be sure from other fields if it has
847 * been initialised or not.
850 int active=0, working=0,failed=0,spare=0,nr_disks=0;
852 rdev->sb_size = MD_SB_BYTES;
854 sb = (mdp_super_t*)page_address(rdev->sb_page);
856 memset(sb, 0, sizeof(*sb));
858 sb->md_magic = MD_SB_MAGIC;
859 sb->major_version = mddev->major_version;
860 sb->patch_version = mddev->patch_version;
861 sb->gvalid_words = 0; /* ignored */
862 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
863 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
864 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
865 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
867 sb->ctime = mddev->ctime;
868 sb->level = mddev->level;
869 sb->size = mddev->size;
870 sb->raid_disks = mddev->raid_disks;
871 sb->md_minor = mddev->md_minor;
872 sb->not_persistent = !mddev->persistent;
873 sb->utime = mddev->utime;
875 sb->events_hi = (mddev->events>>32);
876 sb->events_lo = (u32)mddev->events;
878 if (mddev->reshape_position == MaxSector)
879 sb->minor_version = 90;
881 sb->minor_version = 91;
882 sb->reshape_position = mddev->reshape_position;
883 sb->new_level = mddev->new_level;
884 sb->delta_disks = mddev->delta_disks;
885 sb->new_layout = mddev->new_layout;
886 sb->new_chunk = mddev->new_chunk;
888 mddev->minor_version = sb->minor_version;
891 sb->recovery_cp = mddev->recovery_cp;
892 sb->cp_events_hi = (mddev->events>>32);
893 sb->cp_events_lo = (u32)mddev->events;
894 if (mddev->recovery_cp == MaxSector)
895 sb->state = (1<< MD_SB_CLEAN);
899 sb->layout = mddev->layout;
900 sb->chunk_size = mddev->chunk_size;
902 if (mddev->bitmap && mddev->bitmap_file == NULL)
903 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
905 sb->disks[0].state = (1<<MD_DISK_REMOVED);
906 ITERATE_RDEV(mddev,rdev2,tmp) {
909 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
910 && !test_bit(Faulty, &rdev2->flags))
911 desc_nr = rdev2->raid_disk;
913 desc_nr = next_spare++;
914 rdev2->desc_nr = desc_nr;
915 d = &sb->disks[rdev2->desc_nr];
917 d->number = rdev2->desc_nr;
918 d->major = MAJOR(rdev2->bdev->bd_dev);
919 d->minor = MINOR(rdev2->bdev->bd_dev);
920 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
921 && !test_bit(Faulty, &rdev2->flags))
922 d->raid_disk = rdev2->raid_disk;
924 d->raid_disk = rdev2->desc_nr; /* compatibility */
925 if (test_bit(Faulty, &rdev2->flags))
926 d->state = (1<<MD_DISK_FAULTY);
927 else if (test_bit(In_sync, &rdev2->flags)) {
928 d->state = (1<<MD_DISK_ACTIVE);
929 d->state |= (1<<MD_DISK_SYNC);
937 if (test_bit(WriteMostly, &rdev2->flags))
938 d->state |= (1<<MD_DISK_WRITEMOSTLY);
940 /* now set the "removed" and "faulty" bits on any missing devices */
941 for (i=0 ; i < mddev->raid_disks ; i++) {
942 mdp_disk_t *d = &sb->disks[i];
943 if (d->state == 0 && d->number == 0) {
946 d->state = (1<<MD_DISK_REMOVED);
947 d->state |= (1<<MD_DISK_FAULTY);
951 sb->nr_disks = nr_disks;
952 sb->active_disks = active;
953 sb->working_disks = working;
954 sb->failed_disks = failed;
955 sb->spare_disks = spare;
957 sb->this_disk = sb->disks[rdev->desc_nr];
958 sb->sb_csum = calc_sb_csum(sb);
962 * version 1 superblock
965 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
967 unsigned int disk_csum, csum;
968 unsigned long long newcsum;
969 int size = 256 + le32_to_cpu(sb->max_dev)*2;
970 unsigned int *isuper = (unsigned int*)sb;
973 disk_csum = sb->sb_csum;
976 for (i=0; size>=4; size -= 4 )
977 newcsum += le32_to_cpu(*isuper++);
980 newcsum += le16_to_cpu(*(unsigned short*) isuper);
982 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
983 sb->sb_csum = disk_csum;
984 return cpu_to_le32(csum);
987 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
989 struct mdp_superblock_1 *sb;
992 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
996 * Calculate the position of the superblock.
997 * It is always aligned to a 4K boundary and
998 * depeding on minor_version, it can be:
999 * 0: At least 8K, but less than 12K, from end of device
1000 * 1: At start of device
1001 * 2: 4K from start of device.
1003 switch(minor_version) {
1005 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1007 sb_offset &= ~(sector_t)(4*2-1);
1008 /* convert from sectors to K */
1020 rdev->sb_offset = sb_offset;
1022 /* superblock is rarely larger than 1K, but it can be larger,
1023 * and it is safe to read 4k, so we do that
1025 ret = read_disk_sb(rdev, 4096);
1026 if (ret) return ret;
1029 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1031 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1032 sb->major_version != cpu_to_le32(1) ||
1033 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1034 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1035 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1038 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1039 printk("md: invalid superblock checksum on %s\n",
1040 bdevname(rdev->bdev,b));
1043 if (le64_to_cpu(sb->data_size) < 10) {
1044 printk("md: data_size too small on %s\n",
1045 bdevname(rdev->bdev,b));
1048 rdev->preferred_minor = 0xffff;
1049 rdev->data_offset = le64_to_cpu(sb->data_offset);
1050 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1052 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1053 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1054 if (rdev->sb_size & bmask)
1055 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1061 struct mdp_superblock_1 *refsb =
1062 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1064 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1065 sb->level != refsb->level ||
1066 sb->layout != refsb->layout ||
1067 sb->chunksize != refsb->chunksize) {
1068 printk(KERN_WARNING "md: %s has strangely different"
1069 " superblock to %s\n",
1070 bdevname(rdev->bdev,b),
1071 bdevname(refdev->bdev,b2));
1074 ev1 = le64_to_cpu(sb->events);
1075 ev2 = le64_to_cpu(refsb->events);
1083 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1085 rdev->size = rdev->sb_offset;
1086 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1088 rdev->size = le64_to_cpu(sb->data_size)/2;
1089 if (le32_to_cpu(sb->chunksize))
1090 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1092 if (le32_to_cpu(sb->size) > rdev->size*2)
1097 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1099 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1101 rdev->raid_disk = -1;
1103 if (mddev->raid_disks == 0) {
1104 mddev->major_version = 1;
1105 mddev->patch_version = 0;
1106 mddev->persistent = 1;
1107 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1108 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1109 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1110 mddev->level = le32_to_cpu(sb->level);
1111 mddev->clevel[0] = 0;
1112 mddev->layout = le32_to_cpu(sb->layout);
1113 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1114 mddev->size = le64_to_cpu(sb->size)/2;
1115 mddev->events = le64_to_cpu(sb->events);
1116 mddev->bitmap_offset = 0;
1117 mddev->default_bitmap_offset = 1024 >> 9;
1119 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1120 memcpy(mddev->uuid, sb->set_uuid, 16);
1122 mddev->max_disks = (4096-256)/2;
1124 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1125 mddev->bitmap_file == NULL ) {
1126 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1127 && mddev->level != 10) {
1128 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1131 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1133 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1134 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1135 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1136 mddev->new_level = le32_to_cpu(sb->new_level);
1137 mddev->new_layout = le32_to_cpu(sb->new_layout);
1138 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1140 mddev->reshape_position = MaxSector;
1141 mddev->delta_disks = 0;
1142 mddev->new_level = mddev->level;
1143 mddev->new_layout = mddev->layout;
1144 mddev->new_chunk = mddev->chunk_size;
1147 } else if (mddev->pers == NULL) {
1148 /* Insist of good event counter while assembling */
1149 __u64 ev1 = le64_to_cpu(sb->events);
1151 if (ev1 < mddev->events)
1153 } else if (mddev->bitmap) {
1154 /* If adding to array with a bitmap, then we can accept an
1155 * older device, but not too old.
1157 __u64 ev1 = le64_to_cpu(sb->events);
1158 if (ev1 < mddev->bitmap->events_cleared)
1160 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1163 if (mddev->level != LEVEL_MULTIPATH) {
1165 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1166 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1168 case 0xffff: /* spare */
1170 case 0xfffe: /* faulty */
1171 set_bit(Faulty, &rdev->flags);
1174 set_bit(In_sync, &rdev->flags);
1175 rdev->raid_disk = role;
1178 if (sb->devflags & WriteMostly1)
1179 set_bit(WriteMostly, &rdev->flags);
1180 } else /* MULTIPATH are always insync */
1181 set_bit(In_sync, &rdev->flags);
1186 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1188 struct mdp_superblock_1 *sb;
1189 struct list_head *tmp;
1192 /* make rdev->sb match mddev and rdev data. */
1194 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1196 sb->feature_map = 0;
1198 memset(sb->pad1, 0, sizeof(sb->pad1));
1199 memset(sb->pad2, 0, sizeof(sb->pad2));
1200 memset(sb->pad3, 0, sizeof(sb->pad3));
1202 sb->utime = cpu_to_le64((__u64)mddev->utime);
1203 sb->events = cpu_to_le64(mddev->events);
1205 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1207 sb->resync_offset = cpu_to_le64(0);
1209 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1211 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1212 sb->size = cpu_to_le64(mddev->size<<1);
1214 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1215 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1216 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1218 if (mddev->reshape_position != MaxSector) {
1219 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1220 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1221 sb->new_layout = cpu_to_le32(mddev->new_layout);
1222 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1223 sb->new_level = cpu_to_le32(mddev->new_level);
1224 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1228 ITERATE_RDEV(mddev,rdev2,tmp)
1229 if (rdev2->desc_nr+1 > max_dev)
1230 max_dev = rdev2->desc_nr+1;
1232 sb->max_dev = cpu_to_le32(max_dev);
1233 for (i=0; i<max_dev;i++)
1234 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1236 ITERATE_RDEV(mddev,rdev2,tmp) {
1238 if (test_bit(Faulty, &rdev2->flags))
1239 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1240 else if (test_bit(In_sync, &rdev2->flags))
1241 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1243 sb->dev_roles[i] = cpu_to_le16(0xffff);
1246 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1247 sb->sb_csum = calc_sb_1_csum(sb);
1251 static struct super_type super_types[] = {
1254 .owner = THIS_MODULE,
1255 .load_super = super_90_load,
1256 .validate_super = super_90_validate,
1257 .sync_super = super_90_sync,
1261 .owner = THIS_MODULE,
1262 .load_super = super_1_load,
1263 .validate_super = super_1_validate,
1264 .sync_super = super_1_sync,
1268 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1270 struct list_head *tmp;
1273 ITERATE_RDEV(mddev,rdev,tmp)
1274 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1280 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1282 struct list_head *tmp;
1285 ITERATE_RDEV(mddev1,rdev,tmp)
1286 if (match_dev_unit(mddev2, rdev))
1292 static LIST_HEAD(pending_raid_disks);
1294 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1296 mdk_rdev_t *same_pdev;
1297 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1305 /* make sure rdev->size exceeds mddev->size */
1306 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1308 /* Cannot change size, so fail */
1311 mddev->size = rdev->size;
1313 same_pdev = match_dev_unit(mddev, rdev);
1316 "%s: WARNING: %s appears to be on the same physical"
1317 " disk as %s. True\n protection against single-disk"
1318 " failure might be compromised.\n",
1319 mdname(mddev), bdevname(rdev->bdev,b),
1320 bdevname(same_pdev->bdev,b2));
1322 /* Verify rdev->desc_nr is unique.
1323 * If it is -1, assign a free number, else
1324 * check number is not in use
1326 if (rdev->desc_nr < 0) {
1328 if (mddev->pers) choice = mddev->raid_disks;
1329 while (find_rdev_nr(mddev, choice))
1331 rdev->desc_nr = choice;
1333 if (find_rdev_nr(mddev, rdev->desc_nr))
1336 bdevname(rdev->bdev,b);
1337 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1339 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1342 list_add(&rdev->same_set, &mddev->disks);
1343 rdev->mddev = mddev;
1344 printk(KERN_INFO "md: bind<%s>\n", b);
1346 rdev->kobj.parent = &mddev->kobj;
1347 kobject_add(&rdev->kobj);
1349 if (rdev->bdev->bd_part)
1350 ko = &rdev->bdev->bd_part->kobj;
1352 ko = &rdev->bdev->bd_disk->kobj;
1353 sysfs_create_link(&rdev->kobj, ko, "block");
1354 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1358 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1360 char b[BDEVNAME_SIZE];
1365 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1366 list_del_init(&rdev->same_set);
1367 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1369 sysfs_remove_link(&rdev->kobj, "block");
1370 kobject_del(&rdev->kobj);
1374 * prevent the device from being mounted, repartitioned or
1375 * otherwise reused by a RAID array (or any other kernel
1376 * subsystem), by bd_claiming the device.
1378 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1381 struct block_device *bdev;
1382 char b[BDEVNAME_SIZE];
1384 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1386 printk(KERN_ERR "md: could not open %s.\n",
1387 __bdevname(dev, b));
1388 return PTR_ERR(bdev);
1390 err = bd_claim(bdev, rdev);
1392 printk(KERN_ERR "md: could not bd_claim %s.\n",
1401 static void unlock_rdev(mdk_rdev_t *rdev)
1403 struct block_device *bdev = rdev->bdev;
1411 void md_autodetect_dev(dev_t dev);
1413 static void export_rdev(mdk_rdev_t * rdev)
1415 char b[BDEVNAME_SIZE];
1416 printk(KERN_INFO "md: export_rdev(%s)\n",
1417 bdevname(rdev->bdev,b));
1421 list_del_init(&rdev->same_set);
1423 md_autodetect_dev(rdev->bdev->bd_dev);
1426 kobject_put(&rdev->kobj);
1429 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1431 unbind_rdev_from_array(rdev);
1435 static void export_array(mddev_t *mddev)
1437 struct list_head *tmp;
1440 ITERATE_RDEV(mddev,rdev,tmp) {
1445 kick_rdev_from_array(rdev);
1447 if (!list_empty(&mddev->disks))
1449 mddev->raid_disks = 0;
1450 mddev->major_version = 0;
1453 static void print_desc(mdp_disk_t *desc)
1455 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1456 desc->major,desc->minor,desc->raid_disk,desc->state);
1459 static void print_sb(mdp_super_t *sb)
1464 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1465 sb->major_version, sb->minor_version, sb->patch_version,
1466 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1468 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1469 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1470 sb->md_minor, sb->layout, sb->chunk_size);
1471 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1472 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1473 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1474 sb->failed_disks, sb->spare_disks,
1475 sb->sb_csum, (unsigned long)sb->events_lo);
1478 for (i = 0; i < MD_SB_DISKS; i++) {
1481 desc = sb->disks + i;
1482 if (desc->number || desc->major || desc->minor ||
1483 desc->raid_disk || (desc->state && (desc->state != 4))) {
1484 printk(" D %2d: ", i);
1488 printk(KERN_INFO "md: THIS: ");
1489 print_desc(&sb->this_disk);
1493 static void print_rdev(mdk_rdev_t *rdev)
1495 char b[BDEVNAME_SIZE];
1496 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1497 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1498 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1500 if (rdev->sb_loaded) {
1501 printk(KERN_INFO "md: rdev superblock:\n");
1502 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1504 printk(KERN_INFO "md: no rdev superblock!\n");
1507 void md_print_devices(void)
1509 struct list_head *tmp, *tmp2;
1512 char b[BDEVNAME_SIZE];
1515 printk("md: **********************************\n");
1516 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1517 printk("md: **********************************\n");
1518 ITERATE_MDDEV(mddev,tmp) {
1521 bitmap_print_sb(mddev->bitmap);
1523 printk("%s: ", mdname(mddev));
1524 ITERATE_RDEV(mddev,rdev,tmp2)
1525 printk("<%s>", bdevname(rdev->bdev,b));
1528 ITERATE_RDEV(mddev,rdev,tmp2)
1531 printk("md: **********************************\n");
1536 static void sync_sbs(mddev_t * mddev)
1539 struct list_head *tmp;
1541 ITERATE_RDEV(mddev,rdev,tmp) {
1542 super_types[mddev->major_version].
1543 sync_super(mddev, rdev);
1544 rdev->sb_loaded = 1;
1548 void md_update_sb(mddev_t * mddev)
1551 struct list_head *tmp;
1556 spin_lock_irq(&mddev->write_lock);
1557 sync_req = mddev->in_sync;
1558 mddev->utime = get_seconds();
1561 if (!mddev->events) {
1563 * oops, this 64-bit counter should never wrap.
1564 * Either we are in around ~1 trillion A.C., assuming
1565 * 1 reboot per second, or we have a bug:
1570 mddev->sb_dirty = 2;
1574 * do not write anything to disk if using
1575 * nonpersistent superblocks
1577 if (!mddev->persistent) {
1578 mddev->sb_dirty = 0;
1579 spin_unlock_irq(&mddev->write_lock);
1580 wake_up(&mddev->sb_wait);
1583 spin_unlock_irq(&mddev->write_lock);
1586 "md: updating %s RAID superblock on device (in sync %d)\n",
1587 mdname(mddev),mddev->in_sync);
1589 err = bitmap_update_sb(mddev->bitmap);
1590 ITERATE_RDEV(mddev,rdev,tmp) {
1591 char b[BDEVNAME_SIZE];
1592 dprintk(KERN_INFO "md: ");
1593 if (test_bit(Faulty, &rdev->flags))
1594 dprintk("(skipping faulty ");
1596 dprintk("%s ", bdevname(rdev->bdev,b));
1597 if (!test_bit(Faulty, &rdev->flags)) {
1598 md_super_write(mddev,rdev,
1599 rdev->sb_offset<<1, rdev->sb_size,
1601 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1602 bdevname(rdev->bdev,b),
1603 (unsigned long long)rdev->sb_offset);
1607 if (mddev->level == LEVEL_MULTIPATH)
1608 /* only need to write one superblock... */
1611 md_super_wait(mddev);
1612 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1614 spin_lock_irq(&mddev->write_lock);
1615 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1616 /* have to write it out again */
1617 spin_unlock_irq(&mddev->write_lock);
1620 mddev->sb_dirty = 0;
1621 spin_unlock_irq(&mddev->write_lock);
1622 wake_up(&mddev->sb_wait);
1625 EXPORT_SYMBOL_GPL(md_update_sb);
1627 /* words written to sysfs files may, or my not, be \n terminated.
1628 * We want to accept with case. For this we use cmd_match.
1630 static int cmd_match(const char *cmd, const char *str)
1632 /* See if cmd, written into a sysfs file, matches
1633 * str. They must either be the same, or cmd can
1634 * have a trailing newline
1636 while (*cmd && *str && *cmd == *str) {
1647 struct rdev_sysfs_entry {
1648 struct attribute attr;
1649 ssize_t (*show)(mdk_rdev_t *, char *);
1650 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1654 state_show(mdk_rdev_t *rdev, char *page)
1659 if (test_bit(Faulty, &rdev->flags)) {
1660 len+= sprintf(page+len, "%sfaulty",sep);
1663 if (test_bit(In_sync, &rdev->flags)) {
1664 len += sprintf(page+len, "%sin_sync",sep);
1667 if (!test_bit(Faulty, &rdev->flags) &&
1668 !test_bit(In_sync, &rdev->flags)) {
1669 len += sprintf(page+len, "%sspare", sep);
1672 return len+sprintf(page+len, "\n");
1675 static struct rdev_sysfs_entry
1676 rdev_state = __ATTR_RO(state);
1679 super_show(mdk_rdev_t *rdev, char *page)
1681 if (rdev->sb_loaded && rdev->sb_size) {
1682 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1683 return rdev->sb_size;
1687 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1690 errors_show(mdk_rdev_t *rdev, char *page)
1692 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1696 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1699 unsigned long n = simple_strtoul(buf, &e, 10);
1700 if (*buf && (*e == 0 || *e == '\n')) {
1701 atomic_set(&rdev->corrected_errors, n);
1706 static struct rdev_sysfs_entry rdev_errors =
1707 __ATTR(errors, 0644, errors_show, errors_store);
1710 slot_show(mdk_rdev_t *rdev, char *page)
1712 if (rdev->raid_disk < 0)
1713 return sprintf(page, "none\n");
1715 return sprintf(page, "%d\n", rdev->raid_disk);
1719 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1722 int slot = simple_strtoul(buf, &e, 10);
1723 if (strncmp(buf, "none", 4)==0)
1725 else if (e==buf || (*e && *e!= '\n'))
1727 if (rdev->mddev->pers)
1728 /* Cannot set slot in active array (yet) */
1730 if (slot >= rdev->mddev->raid_disks)
1732 rdev->raid_disk = slot;
1733 /* assume it is working */
1735 set_bit(In_sync, &rdev->flags);
1740 static struct rdev_sysfs_entry rdev_slot =
1741 __ATTR(slot, 0644, slot_show, slot_store);
1744 offset_show(mdk_rdev_t *rdev, char *page)
1746 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1750 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1753 unsigned long long offset = simple_strtoull(buf, &e, 10);
1754 if (e==buf || (*e && *e != '\n'))
1756 if (rdev->mddev->pers)
1758 rdev->data_offset = offset;
1762 static struct rdev_sysfs_entry rdev_offset =
1763 __ATTR(offset, 0644, offset_show, offset_store);
1766 rdev_size_show(mdk_rdev_t *rdev, char *page)
1768 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1772 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1775 unsigned long long size = simple_strtoull(buf, &e, 10);
1776 if (e==buf || (*e && *e != '\n'))
1778 if (rdev->mddev->pers)
1781 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1782 rdev->mddev->size = size;
1786 static struct rdev_sysfs_entry rdev_size =
1787 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1789 static struct attribute *rdev_default_attrs[] = {
1799 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1801 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1802 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1806 return entry->show(rdev, page);
1810 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1811 const char *page, size_t length)
1813 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1814 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1818 return entry->store(rdev, page, length);
1821 static void rdev_free(struct kobject *ko)
1823 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1826 static struct sysfs_ops rdev_sysfs_ops = {
1827 .show = rdev_attr_show,
1828 .store = rdev_attr_store,
1830 static struct kobj_type rdev_ktype = {
1831 .release = rdev_free,
1832 .sysfs_ops = &rdev_sysfs_ops,
1833 .default_attrs = rdev_default_attrs,
1837 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1839 * mark the device faulty if:
1841 * - the device is nonexistent (zero size)
1842 * - the device has no valid superblock
1844 * a faulty rdev _never_ has rdev->sb set.
1846 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1848 char b[BDEVNAME_SIZE];
1853 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1855 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1856 return ERR_PTR(-ENOMEM);
1859 if ((err = alloc_disk_sb(rdev)))
1862 err = lock_rdev(rdev, newdev);
1866 rdev->kobj.parent = NULL;
1867 rdev->kobj.ktype = &rdev_ktype;
1868 kobject_init(&rdev->kobj);
1872 rdev->data_offset = 0;
1873 atomic_set(&rdev->nr_pending, 0);
1874 atomic_set(&rdev->read_errors, 0);
1875 atomic_set(&rdev->corrected_errors, 0);
1877 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1880 "md: %s has zero or unknown size, marking faulty!\n",
1881 bdevname(rdev->bdev,b));
1886 if (super_format >= 0) {
1887 err = super_types[super_format].
1888 load_super(rdev, NULL, super_minor);
1889 if (err == -EINVAL) {
1891 "md: %s has invalid sb, not importing!\n",
1892 bdevname(rdev->bdev,b));
1897 "md: could not read %s's sb, not importing!\n",
1898 bdevname(rdev->bdev,b));
1902 INIT_LIST_HEAD(&rdev->same_set);
1907 if (rdev->sb_page) {
1913 return ERR_PTR(err);
1917 * Check a full RAID array for plausibility
1921 static void analyze_sbs(mddev_t * mddev)
1924 struct list_head *tmp;
1925 mdk_rdev_t *rdev, *freshest;
1926 char b[BDEVNAME_SIZE];
1929 ITERATE_RDEV(mddev,rdev,tmp)
1930 switch (super_types[mddev->major_version].
1931 load_super(rdev, freshest, mddev->minor_version)) {
1939 "md: fatal superblock inconsistency in %s"
1940 " -- removing from array\n",
1941 bdevname(rdev->bdev,b));
1942 kick_rdev_from_array(rdev);
1946 super_types[mddev->major_version].
1947 validate_super(mddev, freshest);
1950 ITERATE_RDEV(mddev,rdev,tmp) {
1951 if (rdev != freshest)
1952 if (super_types[mddev->major_version].
1953 validate_super(mddev, rdev)) {
1954 printk(KERN_WARNING "md: kicking non-fresh %s"
1956 bdevname(rdev->bdev,b));
1957 kick_rdev_from_array(rdev);
1960 if (mddev->level == LEVEL_MULTIPATH) {
1961 rdev->desc_nr = i++;
1962 rdev->raid_disk = rdev->desc_nr;
1963 set_bit(In_sync, &rdev->flags);
1969 if (mddev->recovery_cp != MaxSector &&
1971 printk(KERN_ERR "md: %s: raid array is not clean"
1972 " -- starting background reconstruction\n",
1978 level_show(mddev_t *mddev, char *page)
1980 struct mdk_personality *p = mddev->pers;
1982 return sprintf(page, "%s\n", p->name);
1983 else if (mddev->clevel[0])
1984 return sprintf(page, "%s\n", mddev->clevel);
1985 else if (mddev->level != LEVEL_NONE)
1986 return sprintf(page, "%d\n", mddev->level);
1992 level_store(mddev_t *mddev, const char *buf, size_t len)
1999 if (len >= sizeof(mddev->clevel))
2001 strncpy(mddev->clevel, buf, len);
2002 if (mddev->clevel[len-1] == '\n')
2004 mddev->clevel[len] = 0;
2005 mddev->level = LEVEL_NONE;
2009 static struct md_sysfs_entry md_level =
2010 __ATTR(level, 0644, level_show, level_store);
2013 raid_disks_show(mddev_t *mddev, char *page)
2015 if (mddev->raid_disks == 0)
2017 return sprintf(page, "%d\n", mddev->raid_disks);
2020 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2023 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2025 /* can only set raid_disks if array is not yet active */
2028 unsigned long n = simple_strtoul(buf, &e, 10);
2030 if (!*buf || (*e && *e != '\n'))
2034 rv = update_raid_disks(mddev, n);
2036 mddev->raid_disks = n;
2037 return rv ? rv : len;
2039 static struct md_sysfs_entry md_raid_disks =
2040 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
2043 chunk_size_show(mddev_t *mddev, char *page)
2045 return sprintf(page, "%d\n", mddev->chunk_size);
2049 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2051 /* can only set chunk_size if array is not yet active */
2053 unsigned long n = simple_strtoul(buf, &e, 10);
2057 if (!*buf || (*e && *e != '\n'))
2060 mddev->chunk_size = n;
2063 static struct md_sysfs_entry md_chunk_size =
2064 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2067 null_show(mddev_t *mddev, char *page)
2073 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2075 /* buf must be %d:%d\n? giving major and minor numbers */
2076 /* The new device is added to the array.
2077 * If the array has a persistent superblock, we read the
2078 * superblock to initialise info and check validity.
2079 * Otherwise, only checking done is that in bind_rdev_to_array,
2080 * which mainly checks size.
2083 int major = simple_strtoul(buf, &e, 10);
2089 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2091 minor = simple_strtoul(e+1, &e, 10);
2092 if (*e && *e != '\n')
2094 dev = MKDEV(major, minor);
2095 if (major != MAJOR(dev) ||
2096 minor != MINOR(dev))
2100 if (mddev->persistent) {
2101 rdev = md_import_device(dev, mddev->major_version,
2102 mddev->minor_version);
2103 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2104 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2105 mdk_rdev_t, same_set);
2106 err = super_types[mddev->major_version]
2107 .load_super(rdev, rdev0, mddev->minor_version);
2112 rdev = md_import_device(dev, -1, -1);
2115 return PTR_ERR(rdev);
2116 err = bind_rdev_to_array(rdev, mddev);
2120 return err ? err : len;
2123 static struct md_sysfs_entry md_new_device =
2124 __ATTR(new_dev, 0200, null_show, new_dev_store);
2127 size_show(mddev_t *mddev, char *page)
2129 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2132 static int update_size(mddev_t *mddev, unsigned long size);
2135 size_store(mddev_t *mddev, const char *buf, size_t len)
2137 /* If array is inactive, we can reduce the component size, but
2138 * not increase it (except from 0).
2139 * If array is active, we can try an on-line resize
2143 unsigned long long size = simple_strtoull(buf, &e, 10);
2144 if (!*buf || *buf == '\n' ||
2149 err = update_size(mddev, size);
2150 md_update_sb(mddev);
2152 if (mddev->size == 0 ||
2158 return err ? err : len;
2161 static struct md_sysfs_entry md_size =
2162 __ATTR(component_size, 0644, size_show, size_store);
2166 * This is either 'none' for arrays with externally managed metadata,
2167 * or N.M for internally known formats
2170 metadata_show(mddev_t *mddev, char *page)
2172 if (mddev->persistent)
2173 return sprintf(page, "%d.%d\n",
2174 mddev->major_version, mddev->minor_version);
2176 return sprintf(page, "none\n");
2180 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2184 if (!list_empty(&mddev->disks))
2187 if (cmd_match(buf, "none")) {
2188 mddev->persistent = 0;
2189 mddev->major_version = 0;
2190 mddev->minor_version = 90;
2193 major = simple_strtoul(buf, &e, 10);
2194 if (e==buf || *e != '.')
2197 minor = simple_strtoul(buf, &e, 10);
2198 if (e==buf || *e != '\n')
2200 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2201 super_types[major].name == NULL)
2203 mddev->major_version = major;
2204 mddev->minor_version = minor;
2205 mddev->persistent = 1;
2209 static struct md_sysfs_entry md_metadata =
2210 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2213 action_show(mddev_t *mddev, char *page)
2215 char *type = "idle";
2216 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2217 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2218 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2220 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2221 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2223 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2230 return sprintf(page, "%s\n", type);
2234 action_store(mddev_t *mddev, const char *page, size_t len)
2236 if (!mddev->pers || !mddev->pers->sync_request)
2239 if (cmd_match(page, "idle")) {
2240 if (mddev->sync_thread) {
2241 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2242 md_unregister_thread(mddev->sync_thread);
2243 mddev->sync_thread = NULL;
2244 mddev->recovery = 0;
2246 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2247 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2249 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2250 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2251 else if (cmd_match(page, "reshape")) {
2253 if (mddev->pers->start_reshape == NULL)
2255 err = mddev->pers->start_reshape(mddev);
2259 if (cmd_match(page, "check"))
2260 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2261 else if (cmd_match(page, "repair"))
2263 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2264 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2266 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2267 md_wakeup_thread(mddev->thread);
2272 mismatch_cnt_show(mddev_t *mddev, char *page)
2274 return sprintf(page, "%llu\n",
2275 (unsigned long long) mddev->resync_mismatches);
2278 static struct md_sysfs_entry
2279 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2282 static struct md_sysfs_entry
2283 md_mismatches = __ATTR_RO(mismatch_cnt);
2286 sync_min_show(mddev_t *mddev, char *page)
2288 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2289 mddev->sync_speed_min ? "local": "system");
2293 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2297 if (strncmp(buf, "system", 6)==0) {
2298 mddev->sync_speed_min = 0;
2301 min = simple_strtoul(buf, &e, 10);
2302 if (buf == e || (*e && *e != '\n') || min <= 0)
2304 mddev->sync_speed_min = min;
2308 static struct md_sysfs_entry md_sync_min =
2309 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2312 sync_max_show(mddev_t *mddev, char *page)
2314 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2315 mddev->sync_speed_max ? "local": "system");
2319 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2323 if (strncmp(buf, "system", 6)==0) {
2324 mddev->sync_speed_max = 0;
2327 max = simple_strtoul(buf, &e, 10);
2328 if (buf == e || (*e && *e != '\n') || max <= 0)
2330 mddev->sync_speed_max = max;
2334 static struct md_sysfs_entry md_sync_max =
2335 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2339 sync_speed_show(mddev_t *mddev, char *page)
2341 unsigned long resync, dt, db;
2342 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2343 dt = ((jiffies - mddev->resync_mark) / HZ);
2345 db = resync - (mddev->resync_mark_cnt);
2346 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2349 static struct md_sysfs_entry
2350 md_sync_speed = __ATTR_RO(sync_speed);
2353 sync_completed_show(mddev_t *mddev, char *page)
2355 unsigned long max_blocks, resync;
2357 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2358 max_blocks = mddev->resync_max_sectors;
2360 max_blocks = mddev->size << 1;
2362 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2363 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2366 static struct md_sysfs_entry
2367 md_sync_completed = __ATTR_RO(sync_completed);
2370 suspend_lo_show(mddev_t *mddev, char *page)
2372 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2376 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2379 unsigned long long new = simple_strtoull(buf, &e, 10);
2381 if (mddev->pers->quiesce == NULL)
2383 if (buf == e || (*e && *e != '\n'))
2385 if (new >= mddev->suspend_hi ||
2386 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2387 mddev->suspend_lo = new;
2388 mddev->pers->quiesce(mddev, 2);
2393 static struct md_sysfs_entry md_suspend_lo =
2394 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2398 suspend_hi_show(mddev_t *mddev, char *page)
2400 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2404 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2407 unsigned long long new = simple_strtoull(buf, &e, 10);
2409 if (mddev->pers->quiesce == NULL)
2411 if (buf == e || (*e && *e != '\n'))
2413 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2414 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2415 mddev->suspend_hi = new;
2416 mddev->pers->quiesce(mddev, 1);
2417 mddev->pers->quiesce(mddev, 0);
2422 static struct md_sysfs_entry md_suspend_hi =
2423 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2426 static struct attribute *md_default_attrs[] = {
2428 &md_raid_disks.attr,
2429 &md_chunk_size.attr,
2432 &md_new_device.attr,
2436 static struct attribute *md_redundancy_attrs[] = {
2438 &md_mismatches.attr,
2441 &md_sync_speed.attr,
2442 &md_sync_completed.attr,
2443 &md_suspend_lo.attr,
2444 &md_suspend_hi.attr,
2447 static struct attribute_group md_redundancy_group = {
2449 .attrs = md_redundancy_attrs,
2454 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2456 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2457 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2463 rv = entry->show(mddev, page);
2464 mddev_unlock(mddev);
2469 md_attr_store(struct kobject *kobj, struct attribute *attr,
2470 const char *page, size_t length)
2472 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2473 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2479 rv = entry->store(mddev, page, length);
2480 mddev_unlock(mddev);
2484 static void md_free(struct kobject *ko)
2486 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2490 static struct sysfs_ops md_sysfs_ops = {
2491 .show = md_attr_show,
2492 .store = md_attr_store,
2494 static struct kobj_type md_ktype = {
2496 .sysfs_ops = &md_sysfs_ops,
2497 .default_attrs = md_default_attrs,
2502 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2504 static DEFINE_MUTEX(disks_mutex);
2505 mddev_t *mddev = mddev_find(dev);
2506 struct gendisk *disk;
2507 int partitioned = (MAJOR(dev) != MD_MAJOR);
2508 int shift = partitioned ? MdpMinorShift : 0;
2509 int unit = MINOR(dev) >> shift;
2514 mutex_lock(&disks_mutex);
2515 if (mddev->gendisk) {
2516 mutex_unlock(&disks_mutex);
2520 disk = alloc_disk(1 << shift);
2522 mutex_unlock(&disks_mutex);
2526 disk->major = MAJOR(dev);
2527 disk->first_minor = unit << shift;
2529 sprintf(disk->disk_name, "md_d%d", unit);
2530 sprintf(disk->devfs_name, "md/d%d", unit);
2532 sprintf(disk->disk_name, "md%d", unit);
2533 sprintf(disk->devfs_name, "md/%d", unit);
2535 disk->fops = &md_fops;
2536 disk->private_data = mddev;
2537 disk->queue = mddev->queue;
2539 mddev->gendisk = disk;
2540 mutex_unlock(&disks_mutex);
2541 mddev->kobj.parent = &disk->kobj;
2542 mddev->kobj.k_name = NULL;
2543 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2544 mddev->kobj.ktype = &md_ktype;
2545 kobject_register(&mddev->kobj);
2549 void md_wakeup_thread(mdk_thread_t *thread);
2551 static void md_safemode_timeout(unsigned long data)
2553 mddev_t *mddev = (mddev_t *) data;
2555 mddev->safemode = 1;
2556 md_wakeup_thread(mddev->thread);
2559 static int start_dirty_degraded;
2561 static int do_md_run(mddev_t * mddev)
2565 struct list_head *tmp;
2567 struct gendisk *disk;
2568 struct mdk_personality *pers;
2569 char b[BDEVNAME_SIZE];
2571 if (list_empty(&mddev->disks))
2572 /* cannot run an array with no devices.. */
2579 * Analyze all RAID superblock(s)
2581 if (!mddev->raid_disks)
2584 chunk_size = mddev->chunk_size;
2587 if (chunk_size > MAX_CHUNK_SIZE) {
2588 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2589 chunk_size, MAX_CHUNK_SIZE);
2593 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2595 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2596 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2599 if (chunk_size < PAGE_SIZE) {
2600 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2601 chunk_size, PAGE_SIZE);
2605 /* devices must have minimum size of one chunk */
2606 ITERATE_RDEV(mddev,rdev,tmp) {
2607 if (test_bit(Faulty, &rdev->flags))
2609 if (rdev->size < chunk_size / 1024) {
2611 "md: Dev %s smaller than chunk_size:"
2613 bdevname(rdev->bdev,b),
2614 (unsigned long long)rdev->size,
2622 if (mddev->level != LEVEL_NONE)
2623 request_module("md-level-%d", mddev->level);
2624 else if (mddev->clevel[0])
2625 request_module("md-%s", mddev->clevel);
2629 * Drop all container device buffers, from now on
2630 * the only valid external interface is through the md
2632 * Also find largest hardsector size
2634 ITERATE_RDEV(mddev,rdev,tmp) {
2635 if (test_bit(Faulty, &rdev->flags))
2637 sync_blockdev(rdev->bdev);
2638 invalidate_bdev(rdev->bdev, 0);
2641 md_probe(mddev->unit, NULL, NULL);
2642 disk = mddev->gendisk;
2646 spin_lock(&pers_lock);
2647 pers = find_pers(mddev->level, mddev->clevel);
2648 if (!pers || !try_module_get(pers->owner)) {
2649 spin_unlock(&pers_lock);
2650 if (mddev->level != LEVEL_NONE)
2651 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2654 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2659 spin_unlock(&pers_lock);
2660 mddev->level = pers->level;
2661 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2663 if (mddev->reshape_position != MaxSector &&
2664 pers->start_reshape == NULL) {
2665 /* This personality cannot handle reshaping... */
2667 module_put(pers->owner);
2671 mddev->recovery = 0;
2672 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2673 mddev->barriers_work = 1;
2674 mddev->ok_start_degraded = start_dirty_degraded;
2677 mddev->ro = 2; /* read-only, but switch on first write */
2679 err = mddev->pers->run(mddev);
2680 if (!err && mddev->pers->sync_request) {
2681 err = bitmap_create(mddev);
2683 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2684 mdname(mddev), err);
2685 mddev->pers->stop(mddev);
2689 printk(KERN_ERR "md: pers->run() failed ...\n");
2690 module_put(mddev->pers->owner);
2692 bitmap_destroy(mddev);
2695 if (mddev->pers->sync_request)
2696 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2697 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2700 atomic_set(&mddev->writes_pending,0);
2701 mddev->safemode = 0;
2702 mddev->safemode_timer.function = md_safemode_timeout;
2703 mddev->safemode_timer.data = (unsigned long) mddev;
2704 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2707 ITERATE_RDEV(mddev,rdev,tmp)
2708 if (rdev->raid_disk >= 0) {
2710 sprintf(nm, "rd%d", rdev->raid_disk);
2711 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2714 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2715 md_wakeup_thread(mddev->thread);
2717 if (mddev->sb_dirty)
2718 md_update_sb(mddev);
2720 set_capacity(disk, mddev->array_size<<1);
2722 /* If we call blk_queue_make_request here, it will
2723 * re-initialise max_sectors etc which may have been
2724 * refined inside -> run. So just set the bits we need to set.
2725 * Most initialisation happended when we called
2726 * blk_queue_make_request(..., md_fail_request)
2729 mddev->queue->queuedata = mddev;
2730 mddev->queue->make_request_fn = mddev->pers->make_request;
2733 md_new_event(mddev);
2737 static int restart_array(mddev_t *mddev)
2739 struct gendisk *disk = mddev->gendisk;
2743 * Complain if it has no devices
2746 if (list_empty(&mddev->disks))
2754 mddev->safemode = 0;
2756 set_disk_ro(disk, 0);
2758 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2761 * Kick recovery or resync if necessary
2763 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2764 md_wakeup_thread(mddev->thread);
2767 printk(KERN_ERR "md: %s has no personality assigned.\n",
2776 static int do_md_stop(mddev_t * mddev, int ro)
2779 struct gendisk *disk = mddev->gendisk;
2782 if (atomic_read(&mddev->active)>2) {
2783 printk("md: %s still in use.\n",mdname(mddev));
2787 if (mddev->sync_thread) {
2788 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2789 md_unregister_thread(mddev->sync_thread);
2790 mddev->sync_thread = NULL;
2793 del_timer_sync(&mddev->safemode_timer);
2795 invalidate_partition(disk, 0);
2803 bitmap_flush(mddev);
2804 md_super_wait(mddev);
2806 set_disk_ro(disk, 0);
2807 blk_queue_make_request(mddev->queue, md_fail_request);
2808 mddev->pers->stop(mddev);
2809 if (mddev->pers->sync_request)
2810 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2812 module_put(mddev->pers->owner);
2817 if (!mddev->in_sync) {
2818 /* mark array as shutdown cleanly */
2820 md_update_sb(mddev);
2823 set_disk_ro(disk, 1);
2827 * Free resources if final stop
2831 struct list_head *tmp;
2832 struct gendisk *disk;
2833 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2835 bitmap_destroy(mddev);
2836 if (mddev->bitmap_file) {
2837 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2838 fput(mddev->bitmap_file);
2839 mddev->bitmap_file = NULL;
2841 mddev->bitmap_offset = 0;
2843 ITERATE_RDEV(mddev,rdev,tmp)
2844 if (rdev->raid_disk >= 0) {
2846 sprintf(nm, "rd%d", rdev->raid_disk);
2847 sysfs_remove_link(&mddev->kobj, nm);
2850 export_array(mddev);
2852 mddev->array_size = 0;
2853 disk = mddev->gendisk;
2855 set_capacity(disk, 0);
2858 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2861 md_new_event(mddev);
2866 static void autorun_array(mddev_t *mddev)
2869 struct list_head *tmp;
2872 if (list_empty(&mddev->disks))
2875 printk(KERN_INFO "md: running: ");
2877 ITERATE_RDEV(mddev,rdev,tmp) {
2878 char b[BDEVNAME_SIZE];
2879 printk("<%s>", bdevname(rdev->bdev,b));
2883 err = do_md_run (mddev);
2885 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2886 do_md_stop (mddev, 0);
2891 * lets try to run arrays based on all disks that have arrived
2892 * until now. (those are in pending_raid_disks)
2894 * the method: pick the first pending disk, collect all disks with
2895 * the same UUID, remove all from the pending list and put them into
2896 * the 'same_array' list. Then order this list based on superblock
2897 * update time (freshest comes first), kick out 'old' disks and
2898 * compare superblocks. If everything's fine then run it.
2900 * If "unit" is allocated, then bump its reference count
2902 static void autorun_devices(int part)
2904 struct list_head *tmp;
2905 mdk_rdev_t *rdev0, *rdev;
2907 char b[BDEVNAME_SIZE];
2909 printk(KERN_INFO "md: autorun ...\n");
2910 while (!list_empty(&pending_raid_disks)) {
2912 LIST_HEAD(candidates);
2913 rdev0 = list_entry(pending_raid_disks.next,
2914 mdk_rdev_t, same_set);
2916 printk(KERN_INFO "md: considering %s ...\n",
2917 bdevname(rdev0->bdev,b));
2918 INIT_LIST_HEAD(&candidates);
2919 ITERATE_RDEV_PENDING(rdev,tmp)
2920 if (super_90_load(rdev, rdev0, 0) >= 0) {
2921 printk(KERN_INFO "md: adding %s ...\n",
2922 bdevname(rdev->bdev,b));
2923 list_move(&rdev->same_set, &candidates);
2926 * now we have a set of devices, with all of them having
2927 * mostly sane superblocks. It's time to allocate the
2930 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2931 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2932 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2936 dev = MKDEV(mdp_major,
2937 rdev0->preferred_minor << MdpMinorShift);
2939 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2941 md_probe(dev, NULL, NULL);
2942 mddev = mddev_find(dev);
2945 "md: cannot allocate memory for md drive.\n");
2948 if (mddev_lock(mddev))
2949 printk(KERN_WARNING "md: %s locked, cannot run\n",
2951 else if (mddev->raid_disks || mddev->major_version
2952 || !list_empty(&mddev->disks)) {
2954 "md: %s already running, cannot run %s\n",
2955 mdname(mddev), bdevname(rdev0->bdev,b));
2956 mddev_unlock(mddev);
2958 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2959 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2960 list_del_init(&rdev->same_set);
2961 if (bind_rdev_to_array(rdev, mddev))
2964 autorun_array(mddev);
2965 mddev_unlock(mddev);
2967 /* on success, candidates will be empty, on error
2970 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2974 printk(KERN_INFO "md: ... autorun DONE.\n");
2978 * import RAID devices based on one partition
2979 * if possible, the array gets run as well.
2982 static int autostart_array(dev_t startdev)
2984 char b[BDEVNAME_SIZE];
2985 int err = -EINVAL, i;
2986 mdp_super_t *sb = NULL;
2987 mdk_rdev_t *start_rdev = NULL, *rdev;
2989 start_rdev = md_import_device(startdev, 0, 0);
2990 if (IS_ERR(start_rdev))
2994 /* NOTE: this can only work for 0.90.0 superblocks */
2995 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2996 if (sb->major_version != 0 ||
2997 sb->minor_version != 90 ) {
2998 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2999 export_rdev(start_rdev);
3003 if (test_bit(Faulty, &start_rdev->flags)) {
3005 "md: can not autostart based on faulty %s!\n",
3006 bdevname(start_rdev->bdev,b));
3007 export_rdev(start_rdev);
3010 list_add(&start_rdev->same_set, &pending_raid_disks);
3012 for (i = 0; i < MD_SB_DISKS; i++) {
3013 mdp_disk_t *desc = sb->disks + i;
3014 dev_t dev = MKDEV(desc->major, desc->minor);
3018 if (dev == startdev)
3020 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3022 rdev = md_import_device(dev, 0, 0);
3026 list_add(&rdev->same_set, &pending_raid_disks);
3030 * possibly return codes
3038 static int get_version(void __user * arg)
3042 ver.major = MD_MAJOR_VERSION;
3043 ver.minor = MD_MINOR_VERSION;
3044 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3046 if (copy_to_user(arg, &ver, sizeof(ver)))
3052 static int get_array_info(mddev_t * mddev, void __user * arg)
3054 mdu_array_info_t info;
3055 int nr,working,active,failed,spare;
3057 struct list_head *tmp;
3059 nr=working=active=failed=spare=0;
3060 ITERATE_RDEV(mddev,rdev,tmp) {
3062 if (test_bit(Faulty, &rdev->flags))
3066 if (test_bit(In_sync, &rdev->flags))
3073 info.major_version = mddev->major_version;
3074 info.minor_version = mddev->minor_version;
3075 info.patch_version = MD_PATCHLEVEL_VERSION;
3076 info.ctime = mddev->ctime;
3077 info.level = mddev->level;
3078 info.size = mddev->size;
3079 if (info.size != mddev->size) /* overflow */
3082 info.raid_disks = mddev->raid_disks;
3083 info.md_minor = mddev->md_minor;
3084 info.not_persistent= !mddev->persistent;
3086 info.utime = mddev->utime;
3089 info.state = (1<<MD_SB_CLEAN);
3090 if (mddev->bitmap && mddev->bitmap_offset)
3091 info.state = (1<<MD_SB_BITMAP_PRESENT);
3092 info.active_disks = active;
3093 info.working_disks = working;
3094 info.failed_disks = failed;
3095 info.spare_disks = spare;
3097 info.layout = mddev->layout;
3098 info.chunk_size = mddev->chunk_size;
3100 if (copy_to_user(arg, &info, sizeof(info)))
3106 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3108 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3109 char *ptr, *buf = NULL;
3112 file = kmalloc(sizeof(*file), GFP_KERNEL);
3116 /* bitmap disabled, zero the first byte and copy out */
3117 if (!mddev->bitmap || !mddev->bitmap->file) {
3118 file->pathname[0] = '\0';
3122 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3126 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3130 strcpy(file->pathname, ptr);
3134 if (copy_to_user(arg, file, sizeof(*file)))
3142 static int get_disk_info(mddev_t * mddev, void __user * arg)
3144 mdu_disk_info_t info;
3148 if (copy_from_user(&info, arg, sizeof(info)))
3153 rdev = find_rdev_nr(mddev, nr);
3155 info.major = MAJOR(rdev->bdev->bd_dev);
3156 info.minor = MINOR(rdev->bdev->bd_dev);
3157 info.raid_disk = rdev->raid_disk;
3159 if (test_bit(Faulty, &rdev->flags))
3160 info.state |= (1<<MD_DISK_FAULTY);
3161 else if (test_bit(In_sync, &rdev->flags)) {
3162 info.state |= (1<<MD_DISK_ACTIVE);
3163 info.state |= (1<<MD_DISK_SYNC);
3165 if (test_bit(WriteMostly, &rdev->flags))
3166 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3168 info.major = info.minor = 0;
3169 info.raid_disk = -1;
3170 info.state = (1<<MD_DISK_REMOVED);
3173 if (copy_to_user(arg, &info, sizeof(info)))
3179 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3181 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3183 dev_t dev = MKDEV(info->major,info->minor);
3185 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3188 if (!mddev->raid_disks) {
3190 /* expecting a device which has a superblock */
3191 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3194 "md: md_import_device returned %ld\n",
3196 return PTR_ERR(rdev);
3198 if (!list_empty(&mddev->disks)) {
3199 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3200 mdk_rdev_t, same_set);
3201 int err = super_types[mddev->major_version]
3202 .load_super(rdev, rdev0, mddev->minor_version);
3205 "md: %s has different UUID to %s\n",
3206 bdevname(rdev->bdev,b),
3207 bdevname(rdev0->bdev,b2));
3212 err = bind_rdev_to_array(rdev, mddev);
3219 * add_new_disk can be used once the array is assembled
3220 * to add "hot spares". They must already have a superblock
3225 if (!mddev->pers->hot_add_disk) {
3227 "%s: personality does not support diskops!\n",
3231 if (mddev->persistent)
3232 rdev = md_import_device(dev, mddev->major_version,
3233 mddev->minor_version);
3235 rdev = md_import_device(dev, -1, -1);
3238 "md: md_import_device returned %ld\n",
3240 return PTR_ERR(rdev);
3242 /* set save_raid_disk if appropriate */
3243 if (!mddev->persistent) {
3244 if (info->state & (1<<MD_DISK_SYNC) &&
3245 info->raid_disk < mddev->raid_disks)
3246 rdev->raid_disk = info->raid_disk;
3248 rdev->raid_disk = -1;
3250 super_types[mddev->major_version].
3251 validate_super(mddev, rdev);
3252 rdev->saved_raid_disk = rdev->raid_disk;
3254 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3255 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3256 set_bit(WriteMostly, &rdev->flags);
3258 rdev->raid_disk = -1;
3259 err = bind_rdev_to_array(rdev, mddev);
3263 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3264 md_wakeup_thread(mddev->thread);
3268 /* otherwise, add_new_disk is only allowed
3269 * for major_version==0 superblocks
3271 if (mddev->major_version != 0) {
3272 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3277 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3279 rdev = md_import_device (dev, -1, 0);
3282 "md: error, md_import_device() returned %ld\n",
3284 return PTR_ERR(rdev);
3286 rdev->desc_nr = info->number;
3287 if (info->raid_disk < mddev->raid_disks)
3288 rdev->raid_disk = info->raid_disk;
3290 rdev->raid_disk = -1;
3294 if (rdev->raid_disk < mddev->raid_disks)
3295 if (info->state & (1<<MD_DISK_SYNC))
3296 set_bit(In_sync, &rdev->flags);
3298 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3299 set_bit(WriteMostly, &rdev->flags);
3301 if (!mddev->persistent) {
3302 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3303 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3305 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3306 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3308 err = bind_rdev_to_array(rdev, mddev);
3318 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3320 char b[BDEVNAME_SIZE];
3326 rdev = find_rdev(mddev, dev);
3330 if (rdev->raid_disk >= 0)
3333 kick_rdev_from_array(rdev);
3334 md_update_sb(mddev);
3335 md_new_event(mddev);
3339 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3340 bdevname(rdev->bdev,b), mdname(mddev));
3344 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3346 char b[BDEVNAME_SIZE];
3354 if (mddev->major_version != 0) {
3355 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3356 " version-0 superblocks.\n",
3360 if (!mddev->pers->hot_add_disk) {
3362 "%s: personality does not support diskops!\n",
3367 rdev = md_import_device (dev, -1, 0);
3370 "md: error, md_import_device() returned %ld\n",
3375 if (mddev->persistent)
3376 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3379 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3381 size = calc_dev_size(rdev, mddev->chunk_size);
3384 if (test_bit(Faulty, &rdev->flags)) {
3386 "md: can not hot-add faulty %s disk to %s!\n",
3387 bdevname(rdev->bdev,b), mdname(mddev));
3391 clear_bit(In_sync, &rdev->flags);
3393 err = bind_rdev_to_array(rdev, mddev);
3398 * The rest should better be atomic, we can have disk failures
3399 * noticed in interrupt contexts ...
3402 if (rdev->desc_nr == mddev->max_disks) {
3403 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3406 goto abort_unbind_export;
3409 rdev->raid_disk = -1;
3411 md_update_sb(mddev);
3414 * Kick recovery, maybe this spare has to be added to the
3415 * array immediately.
3417 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3418 md_wakeup_thread(mddev->thread);
3419 md_new_event(mddev);
3422 abort_unbind_export:
3423 unbind_rdev_from_array(rdev);
3430 /* similar to deny_write_access, but accounts for our holding a reference
3431 * to the file ourselves */
3432 static int deny_bitmap_write_access(struct file * file)
3434 struct inode *inode = file->f_mapping->host;
3436 spin_lock(&inode->i_lock);
3437 if (atomic_read(&inode->i_writecount) > 1) {
3438 spin_unlock(&inode->i_lock);
3441 atomic_set(&inode->i_writecount, -1);
3442 spin_unlock(&inode->i_lock);
3447 static int set_bitmap_file(mddev_t *mddev, int fd)
3452 if (!mddev->pers->quiesce)
3454 if (mddev->recovery || mddev->sync_thread)
3456 /* we should be able to change the bitmap.. */
3462 return -EEXIST; /* cannot add when bitmap is present */
3463 mddev->bitmap_file = fget(fd);
3465 if (mddev->bitmap_file == NULL) {
3466 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3471 err = deny_bitmap_write_access(mddev->bitmap_file);
3473 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3475 fput(mddev->bitmap_file);
3476 mddev->bitmap_file = NULL;
3479 mddev->bitmap_offset = 0; /* file overrides offset */
3480 } else if (mddev->bitmap == NULL)
3481 return -ENOENT; /* cannot remove what isn't there */
3484 mddev->pers->quiesce(mddev, 1);
3486 err = bitmap_create(mddev);
3488 bitmap_destroy(mddev);
3489 mddev->pers->quiesce(mddev, 0);
3490 } else if (fd < 0) {
3491 if (mddev->bitmap_file)
3492 fput(mddev->bitmap_file);
3493 mddev->bitmap_file = NULL;
3500 * set_array_info is used two different ways
3501 * The original usage is when creating a new array.
3502 * In this usage, raid_disks is > 0 and it together with
3503 * level, size, not_persistent,layout,chunksize determine the
3504 * shape of the array.
3505 * This will always create an array with a type-0.90.0 superblock.
3506 * The newer usage is when assembling an array.
3507 * In this case raid_disks will be 0, and the major_version field is
3508 * use to determine which style super-blocks are to be found on the devices.
3509 * The minor and patch _version numbers are also kept incase the
3510 * super_block handler wishes to interpret them.
3512 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3515 if (info->raid_disks == 0) {
3516 /* just setting version number for superblock loading */
3517 if (info->major_version < 0 ||
3518 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3519 super_types[info->major_version].name == NULL) {
3520 /* maybe try to auto-load a module? */
3522 "md: superblock version %d not known\n",
3523 info->major_version);
3526 mddev->major_version = info->major_version;
3527 mddev->minor_version = info->minor_version;
3528 mddev->patch_version = info->patch_version;
3531 mddev->major_version = MD_MAJOR_VERSION;
3532 mddev->minor_version = MD_MINOR_VERSION;
3533 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3534 mddev->ctime = get_seconds();
3536 mddev->level = info->level;
3537 mddev->clevel[0] = 0;
3538 mddev->size = info->size;
3539 mddev->raid_disks = info->raid_disks;
3540 /* don't set md_minor, it is determined by which /dev/md* was
3543 if (info->state & (1<<MD_SB_CLEAN))
3544 mddev->recovery_cp = MaxSector;
3546 mddev->recovery_cp = 0;
3547 mddev->persistent = ! info->not_persistent;
3549 mddev->layout = info->layout;
3550 mddev->chunk_size = info->chunk_size;
3552 mddev->max_disks = MD_SB_DISKS;
3554 mddev->sb_dirty = 1;
3556 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3557 mddev->bitmap_offset = 0;
3559 mddev->reshape_position = MaxSector;
3562 * Generate a 128 bit UUID
3564 get_random_bytes(mddev->uuid, 16);
3566 mddev->new_level = mddev->level;
3567 mddev->new_chunk = mddev->chunk_size;
3568 mddev->new_layout = mddev->layout;
3569 mddev->delta_disks = 0;
3574 static int update_size(mddev_t *mddev, unsigned long size)
3578 struct list_head *tmp;
3579 int fit = (size == 0);
3581 if (mddev->pers->resize == NULL)
3583 /* The "size" is the amount of each device that is used.
3584 * This can only make sense for arrays with redundancy.
3585 * linear and raid0 always use whatever space is available
3586 * We can only consider changing the size if no resync
3587 * or reconstruction is happening, and if the new size
3588 * is acceptable. It must fit before the sb_offset or,
3589 * if that is <data_offset, it must fit before the
3590 * size of each device.
3591 * If size is zero, we find the largest size that fits.
3593 if (mddev->sync_thread)
3595 ITERATE_RDEV(mddev,rdev,tmp) {
3597 if (rdev->sb_offset > rdev->data_offset)
3598 avail = (rdev->sb_offset*2) - rdev->data_offset;
3600 avail = get_capacity(rdev->bdev->bd_disk)
3601 - rdev->data_offset;
3602 if (fit && (size == 0 || size > avail/2))
3604 if (avail < ((sector_t)size << 1))
3607 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3609 struct block_device *bdev;
3611 bdev = bdget_disk(mddev->gendisk, 0);
3613 mutex_lock(&bdev->bd_inode->i_mutex);
3614 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
3615 mutex_unlock(&bdev->bd_inode->i_mutex);
3622 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3625 /* change the number of raid disks */
3626 if (mddev->pers->check_reshape == NULL)
3628 if (raid_disks <= 0 ||
3629 raid_disks >= mddev->max_disks)
3631 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
3633 mddev->delta_disks = raid_disks - mddev->raid_disks;
3635 rv = mddev->pers->check_reshape(mddev);
3641 * update_array_info is used to change the configuration of an
3643 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3644 * fields in the info are checked against the array.
3645 * Any differences that cannot be handled will cause an error.
3646 * Normally, only one change can be managed at a time.
3648 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3654 /* calculate expected state,ignoring low bits */
3655 if (mddev->bitmap && mddev->bitmap_offset)
3656 state |= (1 << MD_SB_BITMAP_PRESENT);
3658 if (mddev->major_version != info->major_version ||
3659 mddev->minor_version != info->minor_version ||
3660 /* mddev->patch_version != info->patch_version || */
3661 mddev->ctime != info->ctime ||
3662 mddev->level != info->level ||
3663 /* mddev->layout != info->layout || */
3664 !mddev->persistent != info->not_persistent||
3665 mddev->chunk_size != info->chunk_size ||
3666 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3667 ((state^info->state) & 0xfffffe00)
3670 /* Check there is only one change */
3671 if (info->size >= 0 && mddev->size != info->size) cnt++;
3672 if (mddev->raid_disks != info->raid_disks) cnt++;
3673 if (mddev->layout != info->layout) cnt++;
3674 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3675 if (cnt == 0) return 0;
3676 if (cnt > 1) return -EINVAL;
3678 if (mddev->layout != info->layout) {
3680 * we don't need to do anything at the md level, the
3681 * personality will take care of it all.
3683 if (mddev->pers->reconfig == NULL)
3686 return mddev->pers->reconfig(mddev, info->layout, -1);
3688 if (info->size >= 0 && mddev->size != info->size)
3689 rv = update_size(mddev, info->size);
3691 if (mddev->raid_disks != info->raid_disks)
3692 rv = update_raid_disks(mddev, info->raid_disks);
3694 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3695 if (mddev->pers->quiesce == NULL)
3697 if (mddev->recovery || mddev->sync_thread)
3699 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3700 /* add the bitmap */
3703 if (mddev->default_bitmap_offset == 0)
3705 mddev->bitmap_offset = mddev->default_bitmap_offset;
3706 mddev->pers->quiesce(mddev, 1);
3707 rv = bitmap_create(mddev);
3709 bitmap_destroy(mddev);
3710 mddev->pers->quiesce(mddev, 0);
3712 /* remove the bitmap */
3715 if (mddev->bitmap->file)
3717 mddev->pers->quiesce(mddev, 1);
3718 bitmap_destroy(mddev);
3719 mddev->pers->quiesce(mddev, 0);
3720 mddev->bitmap_offset = 0;
3723 md_update_sb(mddev);
3727 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3731 if (mddev->pers == NULL)
3734 rdev = find_rdev(mddev, dev);
3738 md_error(mddev, rdev);
3742 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3744 mddev_t *mddev = bdev->bd_disk->private_data;
3748 geo->cylinders = get_capacity(mddev->gendisk) / 8;
3752 static int md_ioctl(struct inode *inode, struct file *file,
3753 unsigned int cmd, unsigned long arg)
3756 void __user *argp = (void __user *)arg;
3757 mddev_t *mddev = NULL;
3759 if (!capable(CAP_SYS_ADMIN))
3763 * Commands dealing with the RAID driver but not any
3769 err = get_version(argp);
3772 case PRINT_RAID_DEBUG:
3780 autostart_arrays(arg);
3787 * Commands creating/starting a new array:
3790 mddev = inode->i_bdev->bd_disk->private_data;
3798 if (cmd == START_ARRAY) {
3799 /* START_ARRAY doesn't need to lock the array as autostart_array
3800 * does the locking, and it could even be a different array
3805 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3806 "This will not be supported beyond July 2006\n",
3807 current->comm, current->pid);
3810 err = autostart_array(new_decode_dev(arg));
3812 printk(KERN_WARNING "md: autostart failed!\n");
3818 err = mddev_lock(mddev);
3821 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3828 case SET_ARRAY_INFO:
3830 mdu_array_info_t info;
3832 memset(&info, 0, sizeof(info));
3833 else if (copy_from_user(&info, argp, sizeof(info))) {
3838 err = update_array_info(mddev, &info);
3840 printk(KERN_WARNING "md: couldn't update"
3841 " array info. %d\n", err);
3846 if (!list_empty(&mddev->disks)) {
3848 "md: array %s already has disks!\n",
3853 if (mddev->raid_disks) {
3855 "md: array %s already initialised!\n",
3860 err = set_array_info(mddev, &info);
3862 printk(KERN_WARNING "md: couldn't set"
3863 " array info. %d\n", err);
3873 * Commands querying/configuring an existing array:
3875 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3876 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3877 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3878 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3884 * Commands even a read-only array can execute:
3888 case GET_ARRAY_INFO:
3889 err = get_array_info(mddev, argp);
3892 case GET_BITMAP_FILE:
3893 err = get_bitmap_file(mddev, argp);
3897 err = get_disk_info(mddev, argp);
3900 case RESTART_ARRAY_RW:
3901 err = restart_array(mddev);
3905 err = do_md_stop (mddev, 0);
3909 err = do_md_stop (mddev, 1);
3913 * We have a problem here : there is no easy way to give a CHS
3914 * virtual geometry. We currently pretend that we have a 2 heads
3915 * 4 sectors (with a BIG number of cylinders...). This drives
3916 * dosfs just mad... ;-)
3921 * The remaining ioctls are changing the state of the
3922 * superblock, so we do not allow them on read-only arrays.
3923 * However non-MD ioctls (e.g. get-size) will still come through
3924 * here and hit the 'default' below, so only disallow
3925 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3927 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3928 mddev->ro && mddev->pers) {
3929 if (mddev->ro == 2) {
3931 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3932 md_wakeup_thread(mddev->thread);
3944 mdu_disk_info_t info;
3945 if (copy_from_user(&info, argp, sizeof(info)))
3948 err = add_new_disk(mddev, &info);
3952 case HOT_REMOVE_DISK:
3953 err = hot_remove_disk(mddev, new_decode_dev(arg));
3957 err = hot_add_disk(mddev, new_decode_dev(arg));
3960 case SET_DISK_FAULTY:
3961 err = set_disk_faulty(mddev, new_decode_dev(arg));
3965 err = do_md_run (mddev);
3968 case SET_BITMAP_FILE:
3969 err = set_bitmap_file(mddev, (int)arg);
3973 if (_IOC_TYPE(cmd) == MD_MAJOR)
3974 printk(KERN_WARNING "md: %s(pid %d) used"
3975 " obsolete MD ioctl, upgrade your"
3976 " software to use new ictls.\n",
3977 current->comm, current->pid);
3984 mddev_unlock(mddev);
3994 static int md_open(struct inode *inode, struct file *file)
3997 * Succeed if we can lock the mddev, which confirms that
3998 * it isn't being stopped right now.
4000 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4003 if ((err = mddev_lock(mddev)))
4008 mddev_unlock(mddev);
4010 check_disk_change(inode->i_bdev);
4015 static int md_release(struct inode *inode, struct file * file)
4017 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4026 static int md_media_changed(struct gendisk *disk)
4028 mddev_t *mddev = disk->private_data;
4030 return mddev->changed;
4033 static int md_revalidate(struct gendisk *disk)
4035 mddev_t *mddev = disk->private_data;
4040 static struct block_device_operations md_fops =
4042 .owner = THIS_MODULE,
4044 .release = md_release,
4046 .getgeo = md_getgeo,
4047 .media_changed = md_media_changed,
4048 .revalidate_disk= md_revalidate,
4051 static int md_thread(void * arg)
4053 mdk_thread_t *thread = arg;
4056 * md_thread is a 'system-thread', it's priority should be very
4057 * high. We avoid resource deadlocks individually in each
4058 * raid personality. (RAID5 does preallocation) We also use RR and
4059 * the very same RT priority as kswapd, thus we will never get
4060 * into a priority inversion deadlock.
4062 * we definitely have to have equal or higher priority than
4063 * bdflush, otherwise bdflush will deadlock if there are too
4064 * many dirty RAID5 blocks.
4067 allow_signal(SIGKILL);
4068 while (!kthread_should_stop()) {
4070 /* We need to wait INTERRUPTIBLE so that
4071 * we don't add to the load-average.
4072 * That means we need to be sure no signals are
4075 if (signal_pending(current))
4076 flush_signals(current);
4078 wait_event_interruptible_timeout
4080 test_bit(THREAD_WAKEUP, &thread->flags)
4081 || kthread_should_stop(),
4085 clear_bit(THREAD_WAKEUP, &thread->flags);
4087 thread->run(thread->mddev);
4093 void md_wakeup_thread(mdk_thread_t *thread)
4096 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4097 set_bit(THREAD_WAKEUP, &thread->flags);
4098 wake_up(&thread->wqueue);
4102 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4105 mdk_thread_t *thread;
4107 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4111 init_waitqueue_head(&thread->wqueue);
4114 thread->mddev = mddev;
4115 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4116 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4117 if (IS_ERR(thread->tsk)) {
4124 void md_unregister_thread(mdk_thread_t *thread)
4126 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4128 kthread_stop(thread->tsk);
4132 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4139 if (!rdev || test_bit(Faulty, &rdev->flags))
4142 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4144 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4145 __builtin_return_address(0),__builtin_return_address(1),
4146 __builtin_return_address(2),__builtin_return_address(3));
4148 if (!mddev->pers->error_handler)
4150 mddev->pers->error_handler(mddev,rdev);
4151 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4152 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4153 md_wakeup_thread(mddev->thread);
4154 md_new_event(mddev);
4157 /* seq_file implementation /proc/mdstat */
4159 static void status_unused(struct seq_file *seq)
4163 struct list_head *tmp;
4165 seq_printf(seq, "unused devices: ");
4167 ITERATE_RDEV_PENDING(rdev,tmp) {
4168 char b[BDEVNAME_SIZE];
4170 seq_printf(seq, "%s ",
4171 bdevname(rdev->bdev,b));
4174 seq_printf(seq, "<none>");
4176 seq_printf(seq, "\n");
4180 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4182 sector_t max_blocks, resync, res;
4183 unsigned long dt, db, rt;
4185 unsigned int per_milli;
4187 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4189 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4190 max_blocks = mddev->resync_max_sectors >> 1;
4192 max_blocks = mddev->size;
4195 * Should not happen.
4201 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4202 * in a sector_t, and (max_blocks>>scale) will fit in a
4203 * u32, as those are the requirements for sector_div.
4204 * Thus 'scale' must be at least 10
4207 if (sizeof(sector_t) > sizeof(unsigned long)) {
4208 while ( max_blocks/2 > (1ULL<<(scale+32)))
4211 res = (resync>>scale)*1000;
4212 sector_div(res, (u32)((max_blocks>>scale)+1));
4216 int i, x = per_milli/50, y = 20-x;
4217 seq_printf(seq, "[");
4218 for (i = 0; i < x; i++)
4219 seq_printf(seq, "=");
4220 seq_printf(seq, ">");
4221 for (i = 0; i < y; i++)
4222 seq_printf(seq, ".");
4223 seq_printf(seq, "] ");
4225 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4226 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4228 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4229 "resync" : "recovery")),
4230 per_milli/10, per_milli % 10,
4231 (unsigned long long) resync,
4232 (unsigned long long) max_blocks);
4235 * We do not want to overflow, so the order of operands and
4236 * the * 100 / 100 trick are important. We do a +1 to be
4237 * safe against division by zero. We only estimate anyway.
4239 * dt: time from mark until now
4240 * db: blocks written from mark until now
4241 * rt: remaining time
4243 dt = ((jiffies - mddev->resync_mark) / HZ);
4245 db = resync - (mddev->resync_mark_cnt/2);
4246 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100;
4248 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4250 seq_printf(seq, " speed=%ldK/sec", db/dt);
4253 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4255 struct list_head *tmp;
4265 spin_lock(&all_mddevs_lock);
4266 list_for_each(tmp,&all_mddevs)
4268 mddev = list_entry(tmp, mddev_t, all_mddevs);
4270 spin_unlock(&all_mddevs_lock);
4273 spin_unlock(&all_mddevs_lock);
4275 return (void*)2;/* tail */
4279 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4281 struct list_head *tmp;
4282 mddev_t *next_mddev, *mddev = v;
4288 spin_lock(&all_mddevs_lock);
4290 tmp = all_mddevs.next;
4292 tmp = mddev->all_mddevs.next;
4293 if (tmp != &all_mddevs)
4294 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4296 next_mddev = (void*)2;
4299 spin_unlock(&all_mddevs_lock);
4307 static void md_seq_stop(struct seq_file *seq, void *v)
4311 if (mddev && v != (void*)1 && v != (void*)2)
4315 struct mdstat_info {
4319 static int md_seq_show(struct seq_file *seq, void *v)
4323 struct list_head *tmp2;
4325 struct mdstat_info *mi = seq->private;
4326 struct bitmap *bitmap;
4328 if (v == (void*)1) {
4329 struct mdk_personality *pers;
4330 seq_printf(seq, "Personalities : ");
4331 spin_lock(&pers_lock);
4332 list_for_each_entry(pers, &pers_list, list)
4333 seq_printf(seq, "[%s] ", pers->name);
4335 spin_unlock(&pers_lock);
4336 seq_printf(seq, "\n");
4337 mi->event = atomic_read(&md_event_count);
4340 if (v == (void*)2) {
4345 if (mddev_lock(mddev)!=0)
4347 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4348 seq_printf(seq, "%s : %sactive", mdname(mddev),
4349 mddev->pers ? "" : "in");
4352 seq_printf(seq, " (read-only)");
4354 seq_printf(seq, "(auto-read-only)");
4355 seq_printf(seq, " %s", mddev->pers->name);
4359 ITERATE_RDEV(mddev,rdev,tmp2) {
4360 char b[BDEVNAME_SIZE];
4361 seq_printf(seq, " %s[%d]",
4362 bdevname(rdev->bdev,b), rdev->desc_nr);
4363 if (test_bit(WriteMostly, &rdev->flags))
4364 seq_printf(seq, "(W)");
4365 if (test_bit(Faulty, &rdev->flags)) {
4366 seq_printf(seq, "(F)");
4368 } else if (rdev->raid_disk < 0)
4369 seq_printf(seq, "(S)"); /* spare */
4373 if (!list_empty(&mddev->disks)) {
4375 seq_printf(seq, "\n %llu blocks",
4376 (unsigned long long)mddev->array_size);
4378 seq_printf(seq, "\n %llu blocks",
4379 (unsigned long long)size);
4381 if (mddev->persistent) {
4382 if (mddev->major_version != 0 ||
4383 mddev->minor_version != 90) {
4384 seq_printf(seq," super %d.%d",
4385 mddev->major_version,
4386 mddev->minor_version);
4389 seq_printf(seq, " super non-persistent");
4392 mddev->pers->status (seq, mddev);
4393 seq_printf(seq, "\n ");
4394 if (mddev->pers->sync_request) {
4395 if (mddev->curr_resync > 2) {
4396 status_resync (seq, mddev);
4397 seq_printf(seq, "\n ");
4398 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4399 seq_printf(seq, "\tresync=DELAYED\n ");
4400 else if (mddev->recovery_cp < MaxSector)
4401 seq_printf(seq, "\tresync=PENDING\n ");
4404 seq_printf(seq, "\n ");
4406 if ((bitmap = mddev->bitmap)) {
4407 unsigned long chunk_kb;
4408 unsigned long flags;
4409 spin_lock_irqsave(&bitmap->lock, flags);
4410 chunk_kb = bitmap->chunksize >> 10;
4411 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4413 bitmap->pages - bitmap->missing_pages,
4415 (bitmap->pages - bitmap->missing_pages)
4416 << (PAGE_SHIFT - 10),
4417 chunk_kb ? chunk_kb : bitmap->chunksize,
4418 chunk_kb ? "KB" : "B");
4420 seq_printf(seq, ", file: ");
4421 seq_path(seq, bitmap->file->f_vfsmnt,
4422 bitmap->file->f_dentry," \t\n");
4425 seq_printf(seq, "\n");
4426 spin_unlock_irqrestore(&bitmap->lock, flags);
4429 seq_printf(seq, "\n");
4431 mddev_unlock(mddev);
4436 static struct seq_operations md_seq_ops = {
4437 .start = md_seq_start,
4438 .next = md_seq_next,
4439 .stop = md_seq_stop,
4440 .show = md_seq_show,
4443 static int md_seq_open(struct inode *inode, struct file *file)
4446 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4450 error = seq_open(file, &md_seq_ops);
4454 struct seq_file *p = file->private_data;
4456 mi->event = atomic_read(&md_event_count);
4461 static int md_seq_release(struct inode *inode, struct file *file)
4463 struct seq_file *m = file->private_data;
4464 struct mdstat_info *mi = m->private;
4467 return seq_release(inode, file);
4470 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4472 struct seq_file *m = filp->private_data;
4473 struct mdstat_info *mi = m->private;
4476 poll_wait(filp, &md_event_waiters, wait);
4478 /* always allow read */
4479 mask = POLLIN | POLLRDNORM;
4481 if (mi->event != atomic_read(&md_event_count))
4482 mask |= POLLERR | POLLPRI;
4486 static struct file_operations md_seq_fops = {
4487 .open = md_seq_open,
4489 .llseek = seq_lseek,
4490 .release = md_seq_release,
4491 .poll = mdstat_poll,
4494 int register_md_personality(struct mdk_personality *p)
4496 spin_lock(&pers_lock);
4497 list_add_tail(&p->list, &pers_list);
4498 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4499 spin_unlock(&pers_lock);
4503 int unregister_md_personality(struct mdk_personality *p)
4505 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4506 spin_lock(&pers_lock);
4507 list_del_init(&p->list);
4508 spin_unlock(&pers_lock);
4512 static int is_mddev_idle(mddev_t *mddev)
4515 struct list_head *tmp;
4517 unsigned long curr_events;
4520 ITERATE_RDEV(mddev,rdev,tmp) {
4521 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4522 curr_events = disk_stat_read(disk, sectors[0]) +
4523 disk_stat_read(disk, sectors[1]) -
4524 atomic_read(&disk->sync_io);
4525 /* The difference between curr_events and last_events
4526 * will be affected by any new non-sync IO (making
4527 * curr_events bigger) and any difference in the amount of
4528 * in-flight syncio (making current_events bigger or smaller)
4529 * The amount in-flight is currently limited to
4530 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4531 * which is at most 4096 sectors.
4532 * These numbers are fairly fragile and should be made
4533 * more robust, probably by enforcing the
4534 * 'window size' that md_do_sync sort-of uses.
4536 * Note: the following is an unsigned comparison.
4538 if ((curr_events - rdev->last_events + 4096) > 8192) {
4539 rdev->last_events = curr_events;
4546 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4548 /* another "blocks" (512byte) blocks have been synced */
4549 atomic_sub(blocks, &mddev->recovery_active);
4550 wake_up(&mddev->recovery_wait);
4552 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4553 md_wakeup_thread(mddev->thread);
4554 // stop recovery, signal do_sync ....
4559 /* md_write_start(mddev, bi)
4560 * If we need to update some array metadata (e.g. 'active' flag
4561 * in superblock) before writing, schedule a superblock update
4562 * and wait for it to complete.
4564 void md_write_start(mddev_t *mddev, struct bio *bi)
4566 if (bio_data_dir(bi) != WRITE)
4569 BUG_ON(mddev->ro == 1);
4570 if (mddev->ro == 2) {
4571 /* need to switch to read/write */
4573 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4574 md_wakeup_thread(mddev->thread);
4576 atomic_inc(&mddev->writes_pending);
4577 if (mddev->in_sync) {
4578 spin_lock_irq(&mddev->write_lock);
4579 if (mddev->in_sync) {
4581 mddev->sb_dirty = 1;
4582 md_wakeup_thread(mddev->thread);
4584 spin_unlock_irq(&mddev->write_lock);
4586 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4589 void md_write_end(mddev_t *mddev)
4591 if (atomic_dec_and_test(&mddev->writes_pending)) {
4592 if (mddev->safemode == 2)
4593 md_wakeup_thread(mddev->thread);
4595 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4599 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4601 #define SYNC_MARKS 10
4602 #define SYNC_MARK_STEP (3*HZ)
4603 void md_do_sync(mddev_t *mddev)
4606 unsigned int currspeed = 0,
4608 sector_t max_sectors,j, io_sectors;
4609 unsigned long mark[SYNC_MARKS];
4610 sector_t mark_cnt[SYNC_MARKS];
4612 struct list_head *tmp;
4613 sector_t last_check;
4616 /* just incase thread restarts... */
4617 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4620 /* we overload curr_resync somewhat here.
4621 * 0 == not engaged in resync at all
4622 * 2 == checking that there is no conflict with another sync
4623 * 1 == like 2, but have yielded to allow conflicting resync to
4625 * other == active in resync - this many blocks
4627 * Before starting a resync we must have set curr_resync to
4628 * 2, and then checked that every "conflicting" array has curr_resync
4629 * less than ours. When we find one that is the same or higher
4630 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4631 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4632 * This will mean we have to start checking from the beginning again.
4637 mddev->curr_resync = 2;
4640 if (kthread_should_stop()) {
4641 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4644 ITERATE_MDDEV(mddev2,tmp) {
4645 if (mddev2 == mddev)
4647 if (mddev2->curr_resync &&
4648 match_mddev_units(mddev,mddev2)) {
4650 if (mddev < mddev2 && mddev->curr_resync == 2) {
4651 /* arbitrarily yield */
4652 mddev->curr_resync = 1;
4653 wake_up(&resync_wait);
4655 if (mddev > mddev2 && mddev->curr_resync == 1)
4656 /* no need to wait here, we can wait the next
4657 * time 'round when curr_resync == 2
4660 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4661 if (!kthread_should_stop() &&
4662 mddev2->curr_resync >= mddev->curr_resync) {
4663 printk(KERN_INFO "md: delaying resync of %s"
4664 " until %s has finished resync (they"
4665 " share one or more physical units)\n",
4666 mdname(mddev), mdname(mddev2));
4669 finish_wait(&resync_wait, &wq);
4672 finish_wait(&resync_wait, &wq);
4675 } while (mddev->curr_resync < 2);
4677 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4678 /* resync follows the size requested by the personality,
4679 * which defaults to physical size, but can be virtual size
4681 max_sectors = mddev->resync_max_sectors;
4682 mddev->resync_mismatches = 0;
4683 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4684 max_sectors = mddev->size << 1;
4686 /* recovery follows the physical size of devices */
4687 max_sectors = mddev->size << 1;
4689 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4690 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4691 " %d KB/sec/disc.\n", speed_min(mddev));
4692 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4693 "(but not more than %d KB/sec) for reconstruction.\n",
4696 is_mddev_idle(mddev); /* this also initializes IO event counters */
4697 /* we don't use the checkpoint if there's a bitmap */
4698 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4699 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4700 j = mddev->recovery_cp;
4704 for (m = 0; m < SYNC_MARKS; m++) {
4706 mark_cnt[m] = io_sectors;
4709 mddev->resync_mark = mark[last_mark];
4710 mddev->resync_mark_cnt = mark_cnt[last_mark];
4713 * Tune reconstruction:
4715 window = 32*(PAGE_SIZE/512);
4716 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4717 window/2,(unsigned long long) max_sectors/2);
4719 atomic_set(&mddev->recovery_active, 0);
4720 init_waitqueue_head(&mddev->recovery_wait);
4725 "md: resuming recovery of %s from checkpoint.\n",
4727 mddev->curr_resync = j;
4730 while (j < max_sectors) {
4734 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4735 currspeed < speed_min(mddev));
4737 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4741 if (!skipped) { /* actual IO requested */
4742 io_sectors += sectors;
4743 atomic_add(sectors, &mddev->recovery_active);
4747 if (j>1) mddev->curr_resync = j;
4748 if (last_check == 0)
4749 /* this is the earliers that rebuilt will be
4750 * visible in /proc/mdstat
4752 md_new_event(mddev);
4754 if (last_check + window > io_sectors || j == max_sectors)
4757 last_check = io_sectors;
4759 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4760 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4764 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4766 int next = (last_mark+1) % SYNC_MARKS;
4768 mddev->resync_mark = mark[next];
4769 mddev->resync_mark_cnt = mark_cnt[next];
4770 mark[next] = jiffies;
4771 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4776 if (kthread_should_stop()) {
4778 * got a signal, exit.
4781 "md: md_do_sync() got signal ... exiting\n");
4782 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4787 * this loop exits only if either when we are slower than
4788 * the 'hard' speed limit, or the system was IO-idle for
4790 * the system might be non-idle CPU-wise, but we only care
4791 * about not overloading the IO subsystem. (things like an
4792 * e2fsck being done on the RAID array should execute fast)
4794 mddev->queue->unplug_fn(mddev->queue);
4797 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4798 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4800 if (currspeed > speed_min(mddev)) {
4801 if ((currspeed > speed_max(mddev)) ||
4802 !is_mddev_idle(mddev)) {
4808 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4810 * this also signals 'finished resyncing' to md_stop
4813 mddev->queue->unplug_fn(mddev->queue);
4815 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4817 /* tell personality that we are finished */
4818 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4820 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4821 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
4822 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
4823 mddev->curr_resync > 2 &&
4824 mddev->curr_resync >= mddev->recovery_cp) {
4825 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4827 "md: checkpointing recovery of %s.\n",
4829 mddev->recovery_cp = mddev->curr_resync;
4831 mddev->recovery_cp = MaxSector;
4835 mddev->curr_resync = 0;
4836 wake_up(&resync_wait);
4837 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4838 md_wakeup_thread(mddev->thread);
4840 EXPORT_SYMBOL_GPL(md_do_sync);
4844 * This routine is regularly called by all per-raid-array threads to
4845 * deal with generic issues like resync and super-block update.
4846 * Raid personalities that don't have a thread (linear/raid0) do not
4847 * need this as they never do any recovery or update the superblock.
4849 * It does not do any resync itself, but rather "forks" off other threads
4850 * to do that as needed.
4851 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4852 * "->recovery" and create a thread at ->sync_thread.
4853 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4854 * and wakeups up this thread which will reap the thread and finish up.
4855 * This thread also removes any faulty devices (with nr_pending == 0).
4857 * The overall approach is:
4858 * 1/ if the superblock needs updating, update it.
4859 * 2/ If a recovery thread is running, don't do anything else.
4860 * 3/ If recovery has finished, clean up, possibly marking spares active.
4861 * 4/ If there are any faulty devices, remove them.
4862 * 5/ If array is degraded, try to add spares devices
4863 * 6/ If array has spares or is not in-sync, start a resync thread.
4865 void md_check_recovery(mddev_t *mddev)
4868 struct list_head *rtmp;
4872 bitmap_daemon_work(mddev->bitmap);
4877 if (signal_pending(current)) {
4878 if (mddev->pers->sync_request) {
4879 printk(KERN_INFO "md: %s in immediate safe mode\n",
4881 mddev->safemode = 2;
4883 flush_signals(current);
4888 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4889 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4890 (mddev->safemode == 1) ||
4891 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4892 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4896 if (mddev_trylock(mddev)==0) {
4899 spin_lock_irq(&mddev->write_lock);
4900 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4901 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4903 mddev->sb_dirty = 1;
4905 if (mddev->safemode == 1)
4906 mddev->safemode = 0;
4907 spin_unlock_irq(&mddev->write_lock);
4909 if (mddev->sb_dirty)
4910 md_update_sb(mddev);
4913 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4914 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4915 /* resync/recovery still happening */
4916 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4919 if (mddev->sync_thread) {
4920 /* resync has finished, collect result */
4921 md_unregister_thread(mddev->sync_thread);
4922 mddev->sync_thread = NULL;
4923 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4924 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4926 /* activate any spares */
4927 mddev->pers->spare_active(mddev);
4929 md_update_sb(mddev);
4931 /* if array is no-longer degraded, then any saved_raid_disk
4932 * information must be scrapped
4934 if (!mddev->degraded)
4935 ITERATE_RDEV(mddev,rdev,rtmp)
4936 rdev->saved_raid_disk = -1;
4938 mddev->recovery = 0;
4939 /* flag recovery needed just to double check */
4940 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4941 md_new_event(mddev);
4944 /* Clear some bits that don't mean anything, but
4947 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4948 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4949 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4950 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4952 /* no recovery is running.
4953 * remove any failed drives, then
4954 * add spares if possible.
4955 * Spare are also removed and re-added, to allow
4956 * the personality to fail the re-add.
4958 ITERATE_RDEV(mddev,rdev,rtmp)
4959 if (rdev->raid_disk >= 0 &&
4960 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4961 atomic_read(&rdev->nr_pending)==0) {
4962 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4964 sprintf(nm,"rd%d", rdev->raid_disk);
4965 sysfs_remove_link(&mddev->kobj, nm);
4966 rdev->raid_disk = -1;
4970 if (mddev->degraded) {
4971 ITERATE_RDEV(mddev,rdev,rtmp)
4972 if (rdev->raid_disk < 0
4973 && !test_bit(Faulty, &rdev->flags)) {
4974 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4976 sprintf(nm, "rd%d", rdev->raid_disk);
4977 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4979 md_new_event(mddev);
4986 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4987 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4988 } else if (mddev->recovery_cp < MaxSector) {
4989 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4990 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4991 /* nothing to be done ... */
4994 if (mddev->pers->sync_request) {
4995 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4996 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4997 /* We are adding a device or devices to an array
4998 * which has the bitmap stored on all devices.
4999 * So make sure all bitmap pages get written
5001 bitmap_write_all(mddev->bitmap);
5003 mddev->sync_thread = md_register_thread(md_do_sync,
5006 if (!mddev->sync_thread) {
5007 printk(KERN_ERR "%s: could not start resync"
5010 /* leave the spares where they are, it shouldn't hurt */
5011 mddev->recovery = 0;
5013 md_wakeup_thread(mddev->sync_thread);
5014 md_new_event(mddev);
5017 mddev_unlock(mddev);
5021 static int md_notify_reboot(struct notifier_block *this,
5022 unsigned long code, void *x)
5024 struct list_head *tmp;
5027 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5029 printk(KERN_INFO "md: stopping all md devices.\n");
5031 ITERATE_MDDEV(mddev,tmp)
5032 if (mddev_trylock(mddev)==0)
5033 do_md_stop (mddev, 1);
5035 * certain more exotic SCSI devices are known to be
5036 * volatile wrt too early system reboots. While the
5037 * right place to handle this issue is the given
5038 * driver, we do want to have a safe RAID driver ...
5045 static struct notifier_block md_notifier = {
5046 .notifier_call = md_notify_reboot,
5048 .priority = INT_MAX, /* before any real devices */
5051 static void md_geninit(void)
5053 struct proc_dir_entry *p;
5055 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5057 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5059 p->proc_fops = &md_seq_fops;
5062 static int __init md_init(void)
5066 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5067 " MD_SB_DISKS=%d\n",
5068 MD_MAJOR_VERSION, MD_MINOR_VERSION,
5069 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
5070 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
5073 if (register_blkdev(MAJOR_NR, "md"))
5075 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5076 unregister_blkdev(MAJOR_NR, "md");
5080 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5081 md_probe, NULL, NULL);
5082 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5083 md_probe, NULL, NULL);
5085 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5086 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
5087 S_IFBLK|S_IRUSR|S_IWUSR,
5090 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5091 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
5092 S_IFBLK|S_IRUSR|S_IWUSR,
5096 register_reboot_notifier(&md_notifier);
5097 raid_table_header = register_sysctl_table(raid_root_table, 1);
5107 * Searches all registered partitions for autorun RAID arrays
5110 static dev_t detected_devices[128];
5113 void md_autodetect_dev(dev_t dev)
5115 if (dev_cnt >= 0 && dev_cnt < 127)
5116 detected_devices[dev_cnt++] = dev;
5120 static void autostart_arrays(int part)
5125 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5127 for (i = 0; i < dev_cnt; i++) {
5128 dev_t dev = detected_devices[i];
5130 rdev = md_import_device(dev,0, 0);
5134 if (test_bit(Faulty, &rdev->flags)) {
5138 list_add(&rdev->same_set, &pending_raid_disks);
5142 autorun_devices(part);
5147 static __exit void md_exit(void)
5150 struct list_head *tmp;
5152 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5153 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
5154 for (i=0; i < MAX_MD_DEVS; i++)
5155 devfs_remove("md/%d", i);
5156 for (i=0; i < MAX_MD_DEVS; i++)
5157 devfs_remove("md/d%d", i);
5161 unregister_blkdev(MAJOR_NR,"md");
5162 unregister_blkdev(mdp_major, "mdp");
5163 unregister_reboot_notifier(&md_notifier);
5164 unregister_sysctl_table(raid_table_header);
5165 remove_proc_entry("mdstat", NULL);
5166 ITERATE_MDDEV(mddev,tmp) {
5167 struct gendisk *disk = mddev->gendisk;
5170 export_array(mddev);
5173 mddev->gendisk = NULL;
5178 module_init(md_init)
5179 module_exit(md_exit)
5181 static int get_ro(char *buffer, struct kernel_param *kp)
5183 return sprintf(buffer, "%d", start_readonly);
5185 static int set_ro(const char *val, struct kernel_param *kp)
5188 int num = simple_strtoul(val, &e, 10);
5189 if (*val && (*e == '\0' || *e == '\n')) {
5190 start_readonly = num;
5196 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5197 module_param(start_dirty_degraded, int, 0644);
5200 EXPORT_SYMBOL(register_md_personality);
5201 EXPORT_SYMBOL(unregister_md_personality);
5202 EXPORT_SYMBOL(md_error);
5203 EXPORT_SYMBOL(md_done_sync);
5204 EXPORT_SYMBOL(md_write_start);
5205 EXPORT_SYMBOL(md_write_end);
5206 EXPORT_SYMBOL(md_register_thread);
5207 EXPORT_SYMBOL(md_unregister_thread);
5208 EXPORT_SYMBOL(md_wakeup_thread);
5209 EXPORT_SYMBOL(md_print_devices);
5210 EXPORT_SYMBOL(md_check_recovery);
5211 MODULE_LICENSE("GPL");
5213 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);