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>
47 #include <linux/init.h>
49 #include <linux/file.h>
52 #include <linux/kmod.h>
55 #include <asm/unaligned.h>
57 #define MAJOR_NR MD_MAJOR
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
68 static void autostart_arrays (int part);
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
75 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
76 * is 1000 KB/sec, so the extra system load does not show up that much.
77 * Increase it if you want to have more _guaranteed_ speed. Note that
78 * the RAID driver will use the maximum available bandwidth if the IO
79 * subsystem is idle. There is also an 'absolute maximum' reconstruction
80 * speed limit - in case reconstruction slows down your system despite
83 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
84 * or /sys/block/mdX/md/sync_speed_{min,max}
87 static int sysctl_speed_limit_min = 1000;
88 static int sysctl_speed_limit_max = 200000;
89 static inline int speed_min(mddev_t *mddev)
91 return mddev->sync_speed_min ?
92 mddev->sync_speed_min : sysctl_speed_limit_min;
95 static inline int speed_max(mddev_t *mddev)
97 return mddev->sync_speed_max ?
98 mddev->sync_speed_max : sysctl_speed_limit_max;
101 static struct ctl_table_header *raid_table_header;
103 static ctl_table raid_table[] = {
105 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
106 .procname = "speed_limit_min",
107 .data = &sysctl_speed_limit_min,
108 .maxlen = sizeof(int),
110 .proc_handler = &proc_dointvec,
113 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
114 .procname = "speed_limit_max",
115 .data = &sysctl_speed_limit_max,
116 .maxlen = sizeof(int),
118 .proc_handler = &proc_dointvec,
123 static ctl_table raid_dir_table[] = {
125 .ctl_name = DEV_RAID,
134 static ctl_table raid_root_table[] = {
140 .child = raid_dir_table,
145 static struct block_device_operations md_fops;
147 static int start_readonly;
150 * We have a system wide 'event count' that is incremented
151 * on any 'interesting' event, and readers of /proc/mdstat
152 * can use 'poll' or 'select' to find out when the event
156 * start array, stop array, error, add device, remove device,
157 * start build, activate spare
159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
160 static atomic_t md_event_count;
161 static void md_new_event(mddev_t *mddev)
163 atomic_inc(&md_event_count);
164 wake_up(&md_event_waiters);
168 * Enables to iterate over all existing md arrays
169 * all_mddevs_lock protects this list.
171 static LIST_HEAD(all_mddevs);
172 static DEFINE_SPINLOCK(all_mddevs_lock);
176 * iterates through all used mddevs in the system.
177 * We take care to grab the all_mddevs_lock whenever navigating
178 * the list, and to always hold a refcount when unlocked.
179 * Any code which breaks out of this loop while own
180 * a reference to the current mddev and must mddev_put it.
182 #define ITERATE_MDDEV(mddev,tmp) \
184 for (({ spin_lock(&all_mddevs_lock); \
185 tmp = all_mddevs.next; \
187 ({ if (tmp != &all_mddevs) \
188 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
189 spin_unlock(&all_mddevs_lock); \
190 if (mddev) mddev_put(mddev); \
191 mddev = list_entry(tmp, mddev_t, all_mddevs); \
192 tmp != &all_mddevs;}); \
193 ({ spin_lock(&all_mddevs_lock); \
198 static int md_fail_request (request_queue_t *q, struct bio *bio)
200 bio_io_error(bio, bio->bi_size);
204 static inline mddev_t *mddev_get(mddev_t *mddev)
206 atomic_inc(&mddev->active);
210 static void mddev_put(mddev_t *mddev)
212 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
214 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
215 list_del(&mddev->all_mddevs);
216 blk_put_queue(mddev->queue);
217 kobject_unregister(&mddev->kobj);
219 spin_unlock(&all_mddevs_lock);
222 static mddev_t * mddev_find(dev_t unit)
224 mddev_t *mddev, *new = NULL;
227 spin_lock(&all_mddevs_lock);
228 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
229 if (mddev->unit == unit) {
231 spin_unlock(&all_mddevs_lock);
237 list_add(&new->all_mddevs, &all_mddevs);
238 spin_unlock(&all_mddevs_lock);
241 spin_unlock(&all_mddevs_lock);
243 new = kzalloc(sizeof(*new), GFP_KERNEL);
248 if (MAJOR(unit) == MD_MAJOR)
249 new->md_minor = MINOR(unit);
251 new->md_minor = MINOR(unit) >> MdpMinorShift;
253 init_MUTEX(&new->reconfig_sem);
254 INIT_LIST_HEAD(&new->disks);
255 INIT_LIST_HEAD(&new->all_mddevs);
256 init_timer(&new->safemode_timer);
257 atomic_set(&new->active, 1);
258 spin_lock_init(&new->write_lock);
259 init_waitqueue_head(&new->sb_wait);
261 new->queue = blk_alloc_queue(GFP_KERNEL);
267 blk_queue_make_request(new->queue, md_fail_request);
272 static inline int mddev_lock(mddev_t * mddev)
274 return down_interruptible(&mddev->reconfig_sem);
277 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
279 down(&mddev->reconfig_sem);
282 static inline int mddev_trylock(mddev_t * mddev)
284 return down_trylock(&mddev->reconfig_sem);
287 static inline void mddev_unlock(mddev_t * mddev)
289 up(&mddev->reconfig_sem);
291 md_wakeup_thread(mddev->thread);
294 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
297 struct list_head *tmp;
299 ITERATE_RDEV(mddev,rdev,tmp) {
300 if (rdev->desc_nr == nr)
306 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
308 struct list_head *tmp;
311 ITERATE_RDEV(mddev,rdev,tmp) {
312 if (rdev->bdev->bd_dev == dev)
318 static struct mdk_personality *find_pers(int level, char *clevel)
320 struct mdk_personality *pers;
321 list_for_each_entry(pers, &pers_list, list) {
322 if (level != LEVEL_NONE && pers->level == level)
324 if (strcmp(pers->name, clevel)==0)
330 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
332 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
333 return MD_NEW_SIZE_BLOCKS(size);
336 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
340 size = rdev->sb_offset;
343 size &= ~((sector_t)chunk_size/1024 - 1);
347 static int alloc_disk_sb(mdk_rdev_t * rdev)
352 rdev->sb_page = alloc_page(GFP_KERNEL);
353 if (!rdev->sb_page) {
354 printk(KERN_ALERT "md: out of memory.\n");
361 static void free_disk_sb(mdk_rdev_t * rdev)
364 put_page(rdev->sb_page);
366 rdev->sb_page = NULL;
373 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
375 mdk_rdev_t *rdev = bio->bi_private;
376 mddev_t *mddev = rdev->mddev;
380 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
381 md_error(mddev, rdev);
383 if (atomic_dec_and_test(&mddev->pending_writes))
384 wake_up(&mddev->sb_wait);
389 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
391 struct bio *bio2 = bio->bi_private;
392 mdk_rdev_t *rdev = bio2->bi_private;
393 mddev_t *mddev = rdev->mddev;
397 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
398 error == -EOPNOTSUPP) {
400 /* barriers don't appear to be supported :-( */
401 set_bit(BarriersNotsupp, &rdev->flags);
402 mddev->barriers_work = 0;
403 spin_lock_irqsave(&mddev->write_lock, flags);
404 bio2->bi_next = mddev->biolist;
405 mddev->biolist = bio2;
406 spin_unlock_irqrestore(&mddev->write_lock, flags);
407 wake_up(&mddev->sb_wait);
412 bio->bi_private = rdev;
413 return super_written(bio, bytes_done, error);
416 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
417 sector_t sector, int size, struct page *page)
419 /* write first size bytes of page to sector of rdev
420 * Increment mddev->pending_writes before returning
421 * and decrement it on completion, waking up sb_wait
422 * if zero is reached.
423 * If an error occurred, call md_error
425 * As we might need to resubmit the request if BIO_RW_BARRIER
426 * causes ENOTSUPP, we allocate a spare bio...
428 struct bio *bio = bio_alloc(GFP_NOIO, 1);
429 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
431 bio->bi_bdev = rdev->bdev;
432 bio->bi_sector = sector;
433 bio_add_page(bio, page, size, 0);
434 bio->bi_private = rdev;
435 bio->bi_end_io = super_written;
438 atomic_inc(&mddev->pending_writes);
439 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
441 rw |= (1<<BIO_RW_BARRIER);
442 rbio = bio_clone(bio, GFP_NOIO);
443 rbio->bi_private = bio;
444 rbio->bi_end_io = super_written_barrier;
445 submit_bio(rw, rbio);
450 void md_super_wait(mddev_t *mddev)
452 /* wait for all superblock writes that were scheduled to complete.
453 * if any had to be retried (due to BARRIER problems), retry them
457 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
458 if (atomic_read(&mddev->pending_writes)==0)
460 while (mddev->biolist) {
462 spin_lock_irq(&mddev->write_lock);
463 bio = mddev->biolist;
464 mddev->biolist = bio->bi_next ;
466 spin_unlock_irq(&mddev->write_lock);
467 submit_bio(bio->bi_rw, bio);
471 finish_wait(&mddev->sb_wait, &wq);
474 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
479 complete((struct completion*)bio->bi_private);
483 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
484 struct page *page, int rw)
486 struct bio *bio = bio_alloc(GFP_NOIO, 1);
487 struct completion event;
490 rw |= (1 << BIO_RW_SYNC);
493 bio->bi_sector = sector;
494 bio_add_page(bio, page, size, 0);
495 init_completion(&event);
496 bio->bi_private = &event;
497 bio->bi_end_io = bi_complete;
499 wait_for_completion(&event);
501 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
505 EXPORT_SYMBOL_GPL(sync_page_io);
507 static int read_disk_sb(mdk_rdev_t * rdev, int size)
509 char b[BDEVNAME_SIZE];
510 if (!rdev->sb_page) {
518 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
524 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
525 bdevname(rdev->bdev,b));
529 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
531 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
532 (sb1->set_uuid1 == sb2->set_uuid1) &&
533 (sb1->set_uuid2 == sb2->set_uuid2) &&
534 (sb1->set_uuid3 == sb2->set_uuid3))
542 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
545 mdp_super_t *tmp1, *tmp2;
547 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
548 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
550 if (!tmp1 || !tmp2) {
552 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
560 * nr_disks is not constant
565 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
576 static unsigned int calc_sb_csum(mdp_super_t * sb)
578 unsigned int disk_csum, csum;
580 disk_csum = sb->sb_csum;
582 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
583 sb->sb_csum = disk_csum;
589 * Handle superblock details.
590 * We want to be able to handle multiple superblock formats
591 * so we have a common interface to them all, and an array of
592 * different handlers.
593 * We rely on user-space to write the initial superblock, and support
594 * reading and updating of superblocks.
595 * Interface methods are:
596 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
597 * loads and validates a superblock on dev.
598 * if refdev != NULL, compare superblocks on both devices
600 * 0 - dev has a superblock that is compatible with refdev
601 * 1 - dev has a superblock that is compatible and newer than refdev
602 * so dev should be used as the refdev in future
603 * -EINVAL superblock incompatible or invalid
604 * -othererror e.g. -EIO
606 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
607 * Verify that dev is acceptable into mddev.
608 * The first time, mddev->raid_disks will be 0, and data from
609 * dev should be merged in. Subsequent calls check that dev
610 * is new enough. Return 0 or -EINVAL
612 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
613 * Update the superblock for rdev with data in mddev
614 * This does not write to disc.
620 struct module *owner;
621 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
622 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
623 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
627 * load_super for 0.90.0
629 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
631 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
637 * Calculate the position of the superblock,
638 * it's at the end of the disk.
640 * It also happens to be a multiple of 4Kb.
642 sb_offset = calc_dev_sboffset(rdev->bdev);
643 rdev->sb_offset = sb_offset;
645 ret = read_disk_sb(rdev, MD_SB_BYTES);
650 bdevname(rdev->bdev, b);
651 sb = (mdp_super_t*)page_address(rdev->sb_page);
653 if (sb->md_magic != MD_SB_MAGIC) {
654 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
659 if (sb->major_version != 0 ||
660 sb->minor_version != 90) {
661 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
662 sb->major_version, sb->minor_version,
667 if (sb->raid_disks <= 0)
670 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
671 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
676 rdev->preferred_minor = sb->md_minor;
677 rdev->data_offset = 0;
678 rdev->sb_size = MD_SB_BYTES;
680 if (sb->level == LEVEL_MULTIPATH)
683 rdev->desc_nr = sb->this_disk.number;
689 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
690 if (!uuid_equal(refsb, sb)) {
691 printk(KERN_WARNING "md: %s has different UUID to %s\n",
692 b, bdevname(refdev->bdev,b2));
695 if (!sb_equal(refsb, sb)) {
696 printk(KERN_WARNING "md: %s has same UUID"
697 " but different superblock to %s\n",
698 b, bdevname(refdev->bdev, b2));
702 ev2 = md_event(refsb);
708 rdev->size = calc_dev_size(rdev, sb->chunk_size);
710 if (rdev->size < sb->size && sb->level > 1)
711 /* "this cannot possibly happen" ... */
719 * validate_super for 0.90.0
721 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
724 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
726 rdev->raid_disk = -1;
728 if (mddev->raid_disks == 0) {
729 mddev->major_version = 0;
730 mddev->minor_version = sb->minor_version;
731 mddev->patch_version = sb->patch_version;
732 mddev->persistent = ! sb->not_persistent;
733 mddev->chunk_size = sb->chunk_size;
734 mddev->ctime = sb->ctime;
735 mddev->utime = sb->utime;
736 mddev->level = sb->level;
737 mddev->clevel[0] = 0;
738 mddev->layout = sb->layout;
739 mddev->raid_disks = sb->raid_disks;
740 mddev->size = sb->size;
741 mddev->events = md_event(sb);
742 mddev->bitmap_offset = 0;
743 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
745 if (sb->state & (1<<MD_SB_CLEAN))
746 mddev->recovery_cp = MaxSector;
748 if (sb->events_hi == sb->cp_events_hi &&
749 sb->events_lo == sb->cp_events_lo) {
750 mddev->recovery_cp = sb->recovery_cp;
752 mddev->recovery_cp = 0;
755 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
756 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
757 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
758 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
760 mddev->max_disks = MD_SB_DISKS;
762 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
763 mddev->bitmap_file == NULL) {
764 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
765 && mddev->level != 10) {
766 /* FIXME use a better test */
767 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
770 mddev->bitmap_offset = mddev->default_bitmap_offset;
773 } else if (mddev->pers == NULL) {
774 /* Insist on good event counter while assembling */
775 __u64 ev1 = md_event(sb);
777 if (ev1 < mddev->events)
779 } else if (mddev->bitmap) {
780 /* if adding to array with a bitmap, then we can accept an
781 * older device ... but not too old.
783 __u64 ev1 = md_event(sb);
784 if (ev1 < mddev->bitmap->events_cleared)
786 } else /* just a hot-add of a new device, leave raid_disk at -1 */
789 if (mddev->level != LEVEL_MULTIPATH) {
790 desc = sb->disks + rdev->desc_nr;
792 if (desc->state & (1<<MD_DISK_FAULTY))
793 set_bit(Faulty, &rdev->flags);
794 else if (desc->state & (1<<MD_DISK_SYNC) &&
795 desc->raid_disk < mddev->raid_disks) {
796 set_bit(In_sync, &rdev->flags);
797 rdev->raid_disk = desc->raid_disk;
799 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
800 set_bit(WriteMostly, &rdev->flags);
801 } else /* MULTIPATH are always insync */
802 set_bit(In_sync, &rdev->flags);
807 * sync_super for 0.90.0
809 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
812 struct list_head *tmp;
814 int next_spare = mddev->raid_disks;
817 /* make rdev->sb match mddev data..
820 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
821 * 3/ any empty disks < next_spare become removed
823 * disks[0] gets initialised to REMOVED because
824 * we cannot be sure from other fields if it has
825 * been initialised or not.
828 int active=0, working=0,failed=0,spare=0,nr_disks=0;
830 rdev->sb_size = MD_SB_BYTES;
832 sb = (mdp_super_t*)page_address(rdev->sb_page);
834 memset(sb, 0, sizeof(*sb));
836 sb->md_magic = MD_SB_MAGIC;
837 sb->major_version = mddev->major_version;
838 sb->minor_version = mddev->minor_version;
839 sb->patch_version = mddev->patch_version;
840 sb->gvalid_words = 0; /* ignored */
841 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
842 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
843 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
844 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
846 sb->ctime = mddev->ctime;
847 sb->level = mddev->level;
848 sb->size = mddev->size;
849 sb->raid_disks = mddev->raid_disks;
850 sb->md_minor = mddev->md_minor;
851 sb->not_persistent = !mddev->persistent;
852 sb->utime = mddev->utime;
854 sb->events_hi = (mddev->events>>32);
855 sb->events_lo = (u32)mddev->events;
859 sb->recovery_cp = mddev->recovery_cp;
860 sb->cp_events_hi = (mddev->events>>32);
861 sb->cp_events_lo = (u32)mddev->events;
862 if (mddev->recovery_cp == MaxSector)
863 sb->state = (1<< MD_SB_CLEAN);
867 sb->layout = mddev->layout;
868 sb->chunk_size = mddev->chunk_size;
870 if (mddev->bitmap && mddev->bitmap_file == NULL)
871 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
873 sb->disks[0].state = (1<<MD_DISK_REMOVED);
874 ITERATE_RDEV(mddev,rdev2,tmp) {
877 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
878 && !test_bit(Faulty, &rdev2->flags))
879 desc_nr = rdev2->raid_disk;
881 desc_nr = next_spare++;
882 rdev2->desc_nr = desc_nr;
883 d = &sb->disks[rdev2->desc_nr];
885 d->number = rdev2->desc_nr;
886 d->major = MAJOR(rdev2->bdev->bd_dev);
887 d->minor = MINOR(rdev2->bdev->bd_dev);
888 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
889 && !test_bit(Faulty, &rdev2->flags))
890 d->raid_disk = rdev2->raid_disk;
892 d->raid_disk = rdev2->desc_nr; /* compatibility */
893 if (test_bit(Faulty, &rdev2->flags)) {
894 d->state = (1<<MD_DISK_FAULTY);
896 } else if (test_bit(In_sync, &rdev2->flags)) {
897 d->state = (1<<MD_DISK_ACTIVE);
898 d->state |= (1<<MD_DISK_SYNC);
906 if (test_bit(WriteMostly, &rdev2->flags))
907 d->state |= (1<<MD_DISK_WRITEMOSTLY);
909 /* now set the "removed" and "faulty" bits on any missing devices */
910 for (i=0 ; i < mddev->raid_disks ; i++) {
911 mdp_disk_t *d = &sb->disks[i];
912 if (d->state == 0 && d->number == 0) {
915 d->state = (1<<MD_DISK_REMOVED);
916 d->state |= (1<<MD_DISK_FAULTY);
920 sb->nr_disks = nr_disks;
921 sb->active_disks = active;
922 sb->working_disks = working;
923 sb->failed_disks = failed;
924 sb->spare_disks = spare;
926 sb->this_disk = sb->disks[rdev->desc_nr];
927 sb->sb_csum = calc_sb_csum(sb);
931 * version 1 superblock
934 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
936 unsigned int disk_csum, csum;
937 unsigned long long newcsum;
938 int size = 256 + le32_to_cpu(sb->max_dev)*2;
939 unsigned int *isuper = (unsigned int*)sb;
942 disk_csum = sb->sb_csum;
945 for (i=0; size>=4; size -= 4 )
946 newcsum += le32_to_cpu(*isuper++);
949 newcsum += le16_to_cpu(*(unsigned short*) isuper);
951 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
952 sb->sb_csum = disk_csum;
953 return cpu_to_le32(csum);
956 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
958 struct mdp_superblock_1 *sb;
961 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
965 * Calculate the position of the superblock.
966 * It is always aligned to a 4K boundary and
967 * depeding on minor_version, it can be:
968 * 0: At least 8K, but less than 12K, from end of device
969 * 1: At start of device
970 * 2: 4K from start of device.
972 switch(minor_version) {
974 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
976 sb_offset &= ~(sector_t)(4*2-1);
977 /* convert from sectors to K */
989 rdev->sb_offset = sb_offset;
991 /* superblock is rarely larger than 1K, but it can be larger,
992 * and it is safe to read 4k, so we do that
994 ret = read_disk_sb(rdev, 4096);
998 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1000 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1001 sb->major_version != cpu_to_le32(1) ||
1002 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1003 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1004 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1007 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1008 printk("md: invalid superblock checksum on %s\n",
1009 bdevname(rdev->bdev,b));
1012 if (le64_to_cpu(sb->data_size) < 10) {
1013 printk("md: data_size too small on %s\n",
1014 bdevname(rdev->bdev,b));
1017 rdev->preferred_minor = 0xffff;
1018 rdev->data_offset = le64_to_cpu(sb->data_offset);
1019 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1021 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1022 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1023 if (rdev->sb_size & bmask)
1024 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1030 struct mdp_superblock_1 *refsb =
1031 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1033 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1034 sb->level != refsb->level ||
1035 sb->layout != refsb->layout ||
1036 sb->chunksize != refsb->chunksize) {
1037 printk(KERN_WARNING "md: %s has strangely different"
1038 " superblock to %s\n",
1039 bdevname(rdev->bdev,b),
1040 bdevname(refdev->bdev,b2));
1043 ev1 = le64_to_cpu(sb->events);
1044 ev2 = le64_to_cpu(refsb->events);
1050 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1052 rdev->size = rdev->sb_offset;
1053 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1055 rdev->size = le64_to_cpu(sb->data_size)/2;
1056 if (le32_to_cpu(sb->chunksize))
1057 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1059 if (le32_to_cpu(sb->size) > rdev->size*2)
1064 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1066 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1068 rdev->raid_disk = -1;
1070 if (mddev->raid_disks == 0) {
1071 mddev->major_version = 1;
1072 mddev->patch_version = 0;
1073 mddev->persistent = 1;
1074 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1075 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1076 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1077 mddev->level = le32_to_cpu(sb->level);
1078 mddev->clevel[0] = 0;
1079 mddev->layout = le32_to_cpu(sb->layout);
1080 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1081 mddev->size = le64_to_cpu(sb->size)/2;
1082 mddev->events = le64_to_cpu(sb->events);
1083 mddev->bitmap_offset = 0;
1084 mddev->default_bitmap_offset = 1024 >> 9;
1086 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1087 memcpy(mddev->uuid, sb->set_uuid, 16);
1089 mddev->max_disks = (4096-256)/2;
1091 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1092 mddev->bitmap_file == NULL ) {
1093 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1094 && mddev->level != 10) {
1095 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1098 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1100 } else if (mddev->pers == NULL) {
1101 /* Insist of good event counter while assembling */
1102 __u64 ev1 = le64_to_cpu(sb->events);
1104 if (ev1 < mddev->events)
1106 } else if (mddev->bitmap) {
1107 /* If adding to array with a bitmap, then we can accept an
1108 * older device, but not too old.
1110 __u64 ev1 = le64_to_cpu(sb->events);
1111 if (ev1 < mddev->bitmap->events_cleared)
1113 } else /* just a hot-add of a new device, leave raid_disk at -1 */
1116 if (mddev->level != LEVEL_MULTIPATH) {
1118 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1119 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1121 case 0xffff: /* spare */
1123 case 0xfffe: /* faulty */
1124 set_bit(Faulty, &rdev->flags);
1127 set_bit(In_sync, &rdev->flags);
1128 rdev->raid_disk = role;
1131 if (sb->devflags & WriteMostly1)
1132 set_bit(WriteMostly, &rdev->flags);
1133 } else /* MULTIPATH are always insync */
1134 set_bit(In_sync, &rdev->flags);
1139 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1141 struct mdp_superblock_1 *sb;
1142 struct list_head *tmp;
1145 /* make rdev->sb match mddev and rdev data. */
1147 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1149 sb->feature_map = 0;
1151 memset(sb->pad1, 0, sizeof(sb->pad1));
1152 memset(sb->pad2, 0, sizeof(sb->pad2));
1153 memset(sb->pad3, 0, sizeof(sb->pad3));
1155 sb->utime = cpu_to_le64((__u64)mddev->utime);
1156 sb->events = cpu_to_le64(mddev->events);
1158 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1160 sb->resync_offset = cpu_to_le64(0);
1162 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1164 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1165 sb->size = cpu_to_le64(mddev->size<<1);
1167 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1168 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1169 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1173 ITERATE_RDEV(mddev,rdev2,tmp)
1174 if (rdev2->desc_nr+1 > max_dev)
1175 max_dev = rdev2->desc_nr+1;
1177 sb->max_dev = cpu_to_le32(max_dev);
1178 for (i=0; i<max_dev;i++)
1179 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1181 ITERATE_RDEV(mddev,rdev2,tmp) {
1183 if (test_bit(Faulty, &rdev2->flags))
1184 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1185 else if (test_bit(In_sync, &rdev2->flags))
1186 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1188 sb->dev_roles[i] = cpu_to_le16(0xffff);
1191 sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1192 sb->sb_csum = calc_sb_1_csum(sb);
1196 static struct super_type super_types[] = {
1199 .owner = THIS_MODULE,
1200 .load_super = super_90_load,
1201 .validate_super = super_90_validate,
1202 .sync_super = super_90_sync,
1206 .owner = THIS_MODULE,
1207 .load_super = super_1_load,
1208 .validate_super = super_1_validate,
1209 .sync_super = super_1_sync,
1213 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1215 struct list_head *tmp;
1218 ITERATE_RDEV(mddev,rdev,tmp)
1219 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1225 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1227 struct list_head *tmp;
1230 ITERATE_RDEV(mddev1,rdev,tmp)
1231 if (match_dev_unit(mddev2, rdev))
1237 static LIST_HEAD(pending_raid_disks);
1239 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1241 mdk_rdev_t *same_pdev;
1242 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1250 /* make sure rdev->size exceeds mddev->size */
1251 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1253 /* Cannot change size, so fail */
1256 mddev->size = rdev->size;
1258 same_pdev = match_dev_unit(mddev, rdev);
1261 "%s: WARNING: %s appears to be on the same physical"
1262 " disk as %s. True\n protection against single-disk"
1263 " failure might be compromised.\n",
1264 mdname(mddev), bdevname(rdev->bdev,b),
1265 bdevname(same_pdev->bdev,b2));
1267 /* Verify rdev->desc_nr is unique.
1268 * If it is -1, assign a free number, else
1269 * check number is not in use
1271 if (rdev->desc_nr < 0) {
1273 if (mddev->pers) choice = mddev->raid_disks;
1274 while (find_rdev_nr(mddev, choice))
1276 rdev->desc_nr = choice;
1278 if (find_rdev_nr(mddev, rdev->desc_nr))
1281 bdevname(rdev->bdev,b);
1282 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1284 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1287 list_add(&rdev->same_set, &mddev->disks);
1288 rdev->mddev = mddev;
1289 printk(KERN_INFO "md: bind<%s>\n", b);
1291 rdev->kobj.parent = &mddev->kobj;
1292 kobject_add(&rdev->kobj);
1294 if (rdev->bdev->bd_part)
1295 ko = &rdev->bdev->bd_part->kobj;
1297 ko = &rdev->bdev->bd_disk->kobj;
1298 sysfs_create_link(&rdev->kobj, ko, "block");
1302 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1304 char b[BDEVNAME_SIZE];
1309 list_del_init(&rdev->same_set);
1310 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1312 sysfs_remove_link(&rdev->kobj, "block");
1313 kobject_del(&rdev->kobj);
1317 * prevent the device from being mounted, repartitioned or
1318 * otherwise reused by a RAID array (or any other kernel
1319 * subsystem), by bd_claiming the device.
1321 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1324 struct block_device *bdev;
1325 char b[BDEVNAME_SIZE];
1327 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1329 printk(KERN_ERR "md: could not open %s.\n",
1330 __bdevname(dev, b));
1331 return PTR_ERR(bdev);
1333 err = bd_claim(bdev, rdev);
1335 printk(KERN_ERR "md: could not bd_claim %s.\n",
1344 static void unlock_rdev(mdk_rdev_t *rdev)
1346 struct block_device *bdev = rdev->bdev;
1354 void md_autodetect_dev(dev_t dev);
1356 static void export_rdev(mdk_rdev_t * rdev)
1358 char b[BDEVNAME_SIZE];
1359 printk(KERN_INFO "md: export_rdev(%s)\n",
1360 bdevname(rdev->bdev,b));
1364 list_del_init(&rdev->same_set);
1366 md_autodetect_dev(rdev->bdev->bd_dev);
1369 kobject_put(&rdev->kobj);
1372 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1374 unbind_rdev_from_array(rdev);
1378 static void export_array(mddev_t *mddev)
1380 struct list_head *tmp;
1383 ITERATE_RDEV(mddev,rdev,tmp) {
1388 kick_rdev_from_array(rdev);
1390 if (!list_empty(&mddev->disks))
1392 mddev->raid_disks = 0;
1393 mddev->major_version = 0;
1396 static void print_desc(mdp_disk_t *desc)
1398 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1399 desc->major,desc->minor,desc->raid_disk,desc->state);
1402 static void print_sb(mdp_super_t *sb)
1407 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1408 sb->major_version, sb->minor_version, sb->patch_version,
1409 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1411 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1412 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1413 sb->md_minor, sb->layout, sb->chunk_size);
1414 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1415 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1416 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1417 sb->failed_disks, sb->spare_disks,
1418 sb->sb_csum, (unsigned long)sb->events_lo);
1421 for (i = 0; i < MD_SB_DISKS; i++) {
1424 desc = sb->disks + i;
1425 if (desc->number || desc->major || desc->minor ||
1426 desc->raid_disk || (desc->state && (desc->state != 4))) {
1427 printk(" D %2d: ", i);
1431 printk(KERN_INFO "md: THIS: ");
1432 print_desc(&sb->this_disk);
1436 static void print_rdev(mdk_rdev_t *rdev)
1438 char b[BDEVNAME_SIZE];
1439 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1440 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1441 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1443 if (rdev->sb_loaded) {
1444 printk(KERN_INFO "md: rdev superblock:\n");
1445 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1447 printk(KERN_INFO "md: no rdev superblock!\n");
1450 void md_print_devices(void)
1452 struct list_head *tmp, *tmp2;
1455 char b[BDEVNAME_SIZE];
1458 printk("md: **********************************\n");
1459 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1460 printk("md: **********************************\n");
1461 ITERATE_MDDEV(mddev,tmp) {
1464 bitmap_print_sb(mddev->bitmap);
1466 printk("%s: ", mdname(mddev));
1467 ITERATE_RDEV(mddev,rdev,tmp2)
1468 printk("<%s>", bdevname(rdev->bdev,b));
1471 ITERATE_RDEV(mddev,rdev,tmp2)
1474 printk("md: **********************************\n");
1479 static void sync_sbs(mddev_t * mddev)
1482 struct list_head *tmp;
1484 ITERATE_RDEV(mddev,rdev,tmp) {
1485 super_types[mddev->major_version].
1486 sync_super(mddev, rdev);
1487 rdev->sb_loaded = 1;
1491 static void md_update_sb(mddev_t * mddev)
1494 struct list_head *tmp;
1499 spin_lock_irq(&mddev->write_lock);
1500 sync_req = mddev->in_sync;
1501 mddev->utime = get_seconds();
1504 if (!mddev->events) {
1506 * oops, this 64-bit counter should never wrap.
1507 * Either we are in around ~1 trillion A.C., assuming
1508 * 1 reboot per second, or we have a bug:
1513 mddev->sb_dirty = 2;
1517 * do not write anything to disk if using
1518 * nonpersistent superblocks
1520 if (!mddev->persistent) {
1521 mddev->sb_dirty = 0;
1522 spin_unlock_irq(&mddev->write_lock);
1523 wake_up(&mddev->sb_wait);
1526 spin_unlock_irq(&mddev->write_lock);
1529 "md: updating %s RAID superblock on device (in sync %d)\n",
1530 mdname(mddev),mddev->in_sync);
1532 err = bitmap_update_sb(mddev->bitmap);
1533 ITERATE_RDEV(mddev,rdev,tmp) {
1534 char b[BDEVNAME_SIZE];
1535 dprintk(KERN_INFO "md: ");
1536 if (test_bit(Faulty, &rdev->flags))
1537 dprintk("(skipping faulty ");
1539 dprintk("%s ", bdevname(rdev->bdev,b));
1540 if (!test_bit(Faulty, &rdev->flags)) {
1541 md_super_write(mddev,rdev,
1542 rdev->sb_offset<<1, rdev->sb_size,
1544 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1545 bdevname(rdev->bdev,b),
1546 (unsigned long long)rdev->sb_offset);
1550 if (mddev->level == LEVEL_MULTIPATH)
1551 /* only need to write one superblock... */
1554 md_super_wait(mddev);
1555 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1557 spin_lock_irq(&mddev->write_lock);
1558 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1559 /* have to write it out again */
1560 spin_unlock_irq(&mddev->write_lock);
1563 mddev->sb_dirty = 0;
1564 spin_unlock_irq(&mddev->write_lock);
1565 wake_up(&mddev->sb_wait);
1569 /* words written to sysfs files may, or my not, be \n terminated.
1570 * We want to accept with case. For this we use cmd_match.
1572 static int cmd_match(const char *cmd, const char *str)
1574 /* See if cmd, written into a sysfs file, matches
1575 * str. They must either be the same, or cmd can
1576 * have a trailing newline
1578 while (*cmd && *str && *cmd == *str) {
1589 struct rdev_sysfs_entry {
1590 struct attribute attr;
1591 ssize_t (*show)(mdk_rdev_t *, char *);
1592 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1596 state_show(mdk_rdev_t *rdev, char *page)
1601 if (test_bit(Faulty, &rdev->flags)) {
1602 len+= sprintf(page+len, "%sfaulty",sep);
1605 if (test_bit(In_sync, &rdev->flags)) {
1606 len += sprintf(page+len, "%sin_sync",sep);
1609 if (!test_bit(Faulty, &rdev->flags) &&
1610 !test_bit(In_sync, &rdev->flags)) {
1611 len += sprintf(page+len, "%sspare", sep);
1614 return len+sprintf(page+len, "\n");
1617 static struct rdev_sysfs_entry
1618 rdev_state = __ATTR_RO(state);
1621 super_show(mdk_rdev_t *rdev, char *page)
1623 if (rdev->sb_loaded && rdev->sb_size) {
1624 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1625 return rdev->sb_size;
1629 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1632 errors_show(mdk_rdev_t *rdev, char *page)
1634 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1638 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1641 unsigned long n = simple_strtoul(buf, &e, 10);
1642 if (*buf && (*e == 0 || *e == '\n')) {
1643 atomic_set(&rdev->corrected_errors, n);
1648 static struct rdev_sysfs_entry rdev_errors =
1649 __ATTR(errors, 0644, errors_show, errors_store);
1652 slot_show(mdk_rdev_t *rdev, char *page)
1654 if (rdev->raid_disk < 0)
1655 return sprintf(page, "none\n");
1657 return sprintf(page, "%d\n", rdev->raid_disk);
1661 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1664 int slot = simple_strtoul(buf, &e, 10);
1665 if (strncmp(buf, "none", 4)==0)
1667 else if (e==buf || (*e && *e!= '\n'))
1669 if (rdev->mddev->pers)
1670 /* Cannot set slot in active array (yet) */
1672 if (slot >= rdev->mddev->raid_disks)
1674 rdev->raid_disk = slot;
1675 /* assume it is working */
1677 set_bit(In_sync, &rdev->flags);
1682 static struct rdev_sysfs_entry rdev_slot =
1683 __ATTR(slot, 0644, slot_show, slot_store);
1686 offset_show(mdk_rdev_t *rdev, char *page)
1688 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1692 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1695 unsigned long long offset = simple_strtoull(buf, &e, 10);
1696 if (e==buf || (*e && *e != '\n'))
1698 if (rdev->mddev->pers)
1700 rdev->data_offset = offset;
1704 static struct rdev_sysfs_entry rdev_offset =
1705 __ATTR(offset, 0644, offset_show, offset_store);
1708 rdev_size_show(mdk_rdev_t *rdev, char *page)
1710 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1714 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1717 unsigned long long size = simple_strtoull(buf, &e, 10);
1718 if (e==buf || (*e && *e != '\n'))
1720 if (rdev->mddev->pers)
1723 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1724 rdev->mddev->size = size;
1728 static struct rdev_sysfs_entry rdev_size =
1729 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1731 static struct attribute *rdev_default_attrs[] = {
1741 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1743 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1744 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1748 return entry->show(rdev, page);
1752 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1753 const char *page, size_t length)
1755 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1756 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1760 return entry->store(rdev, page, length);
1763 static void rdev_free(struct kobject *ko)
1765 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1768 static struct sysfs_ops rdev_sysfs_ops = {
1769 .show = rdev_attr_show,
1770 .store = rdev_attr_store,
1772 static struct kobj_type rdev_ktype = {
1773 .release = rdev_free,
1774 .sysfs_ops = &rdev_sysfs_ops,
1775 .default_attrs = rdev_default_attrs,
1779 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1781 * mark the device faulty if:
1783 * - the device is nonexistent (zero size)
1784 * - the device has no valid superblock
1786 * a faulty rdev _never_ has rdev->sb set.
1788 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1790 char b[BDEVNAME_SIZE];
1795 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1797 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1798 return ERR_PTR(-ENOMEM);
1801 if ((err = alloc_disk_sb(rdev)))
1804 err = lock_rdev(rdev, newdev);
1808 rdev->kobj.parent = NULL;
1809 rdev->kobj.ktype = &rdev_ktype;
1810 kobject_init(&rdev->kobj);
1814 rdev->data_offset = 0;
1815 atomic_set(&rdev->nr_pending, 0);
1816 atomic_set(&rdev->read_errors, 0);
1817 atomic_set(&rdev->corrected_errors, 0);
1819 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1822 "md: %s has zero or unknown size, marking faulty!\n",
1823 bdevname(rdev->bdev,b));
1828 if (super_format >= 0) {
1829 err = super_types[super_format].
1830 load_super(rdev, NULL, super_minor);
1831 if (err == -EINVAL) {
1833 "md: %s has invalid sb, not importing!\n",
1834 bdevname(rdev->bdev,b));
1839 "md: could not read %s's sb, not importing!\n",
1840 bdevname(rdev->bdev,b));
1844 INIT_LIST_HEAD(&rdev->same_set);
1849 if (rdev->sb_page) {
1855 return ERR_PTR(err);
1859 * Check a full RAID array for plausibility
1863 static void analyze_sbs(mddev_t * mddev)
1866 struct list_head *tmp;
1867 mdk_rdev_t *rdev, *freshest;
1868 char b[BDEVNAME_SIZE];
1871 ITERATE_RDEV(mddev,rdev,tmp)
1872 switch (super_types[mddev->major_version].
1873 load_super(rdev, freshest, mddev->minor_version)) {
1881 "md: fatal superblock inconsistency in %s"
1882 " -- removing from array\n",
1883 bdevname(rdev->bdev,b));
1884 kick_rdev_from_array(rdev);
1888 super_types[mddev->major_version].
1889 validate_super(mddev, freshest);
1892 ITERATE_RDEV(mddev,rdev,tmp) {
1893 if (rdev != freshest)
1894 if (super_types[mddev->major_version].
1895 validate_super(mddev, rdev)) {
1896 printk(KERN_WARNING "md: kicking non-fresh %s"
1898 bdevname(rdev->bdev,b));
1899 kick_rdev_from_array(rdev);
1902 if (mddev->level == LEVEL_MULTIPATH) {
1903 rdev->desc_nr = i++;
1904 rdev->raid_disk = rdev->desc_nr;
1905 set_bit(In_sync, &rdev->flags);
1911 if (mddev->recovery_cp != MaxSector &&
1913 printk(KERN_ERR "md: %s: raid array is not clean"
1914 " -- starting background reconstruction\n",
1920 level_show(mddev_t *mddev, char *page)
1922 struct mdk_personality *p = mddev->pers;
1924 return sprintf(page, "%s\n", p->name);
1925 else if (mddev->clevel[0])
1926 return sprintf(page, "%s\n", mddev->clevel);
1927 else if (mddev->level != LEVEL_NONE)
1928 return sprintf(page, "%d\n", mddev->level);
1934 level_store(mddev_t *mddev, const char *buf, size_t len)
1941 if (len >= sizeof(mddev->clevel))
1943 strncpy(mddev->clevel, buf, len);
1944 if (mddev->clevel[len-1] == '\n')
1946 mddev->clevel[len] = 0;
1947 mddev->level = LEVEL_NONE;
1951 static struct md_sysfs_entry md_level =
1952 __ATTR(level, 0644, level_show, level_store);
1955 raid_disks_show(mddev_t *mddev, char *page)
1957 if (mddev->raid_disks == 0)
1959 return sprintf(page, "%d\n", mddev->raid_disks);
1962 static int update_raid_disks(mddev_t *mddev, int raid_disks);
1965 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
1967 /* can only set raid_disks if array is not yet active */
1970 unsigned long n = simple_strtoul(buf, &e, 10);
1972 if (!*buf || (*e && *e != '\n'))
1976 rv = update_raid_disks(mddev, n);
1978 mddev->raid_disks = n;
1979 return rv ? rv : len;
1981 static struct md_sysfs_entry md_raid_disks =
1982 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
1985 chunk_size_show(mddev_t *mddev, char *page)
1987 return sprintf(page, "%d\n", mddev->chunk_size);
1991 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
1993 /* can only set chunk_size if array is not yet active */
1995 unsigned long n = simple_strtoul(buf, &e, 10);
1999 if (!*buf || (*e && *e != '\n'))
2002 mddev->chunk_size = n;
2005 static struct md_sysfs_entry md_chunk_size =
2006 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2009 null_show(mddev_t *mddev, char *page)
2015 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2017 /* buf must be %d:%d\n? giving major and minor numbers */
2018 /* The new device is added to the array.
2019 * If the array has a persistent superblock, we read the
2020 * superblock to initialise info and check validity.
2021 * Otherwise, only checking done is that in bind_rdev_to_array,
2022 * which mainly checks size.
2025 int major = simple_strtoul(buf, &e, 10);
2031 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2033 minor = simple_strtoul(e+1, &e, 10);
2034 if (*e && *e != '\n')
2036 dev = MKDEV(major, minor);
2037 if (major != MAJOR(dev) ||
2038 minor != MINOR(dev))
2042 if (mddev->persistent) {
2043 rdev = md_import_device(dev, mddev->major_version,
2044 mddev->minor_version);
2045 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2046 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2047 mdk_rdev_t, same_set);
2048 err = super_types[mddev->major_version]
2049 .load_super(rdev, rdev0, mddev->minor_version);
2054 rdev = md_import_device(dev, -1, -1);
2057 return PTR_ERR(rdev);
2058 err = bind_rdev_to_array(rdev, mddev);
2062 return err ? err : len;
2065 static struct md_sysfs_entry md_new_device =
2066 __ATTR(new_dev, 0200, null_show, new_dev_store);
2069 size_show(mddev_t *mddev, char *page)
2071 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2074 static int update_size(mddev_t *mddev, unsigned long size);
2077 size_store(mddev_t *mddev, const char *buf, size_t len)
2079 /* If array is inactive, we can reduce the component size, but
2080 * not increase it (except from 0).
2081 * If array is active, we can try an on-line resize
2085 unsigned long long size = simple_strtoull(buf, &e, 10);
2086 if (!*buf || *buf == '\n' ||
2091 err = update_size(mddev, size);
2092 md_update_sb(mddev);
2094 if (mddev->size == 0 ||
2100 return err ? err : len;
2103 static struct md_sysfs_entry md_size =
2104 __ATTR(component_size, 0644, size_show, size_store);
2108 * This is either 'none' for arrays with externally managed metadata,
2109 * or N.M for internally known formats
2112 metadata_show(mddev_t *mddev, char *page)
2114 if (mddev->persistent)
2115 return sprintf(page, "%d.%d\n",
2116 mddev->major_version, mddev->minor_version);
2118 return sprintf(page, "none\n");
2122 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2126 if (!list_empty(&mddev->disks))
2129 if (cmd_match(buf, "none")) {
2130 mddev->persistent = 0;
2131 mddev->major_version = 0;
2132 mddev->minor_version = 90;
2135 major = simple_strtoul(buf, &e, 10);
2136 if (e==buf || *e != '.')
2139 minor = simple_strtoul(buf, &e, 10);
2140 if (e==buf || *e != '\n')
2142 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2143 super_types[major].name == NULL)
2145 mddev->major_version = major;
2146 mddev->minor_version = minor;
2147 mddev->persistent = 1;
2151 static struct md_sysfs_entry md_metadata =
2152 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2155 action_show(mddev_t *mddev, char *page)
2157 char *type = "idle";
2158 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2159 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2160 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2161 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2163 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2170 return sprintf(page, "%s\n", type);
2174 action_store(mddev_t *mddev, const char *page, size_t len)
2176 if (!mddev->pers || !mddev->pers->sync_request)
2179 if (cmd_match(page, "idle")) {
2180 if (mddev->sync_thread) {
2181 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2182 md_unregister_thread(mddev->sync_thread);
2183 mddev->sync_thread = NULL;
2184 mddev->recovery = 0;
2186 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2187 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2189 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2192 if (cmd_match(page, "check"))
2193 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2194 else if (cmd_match(page, "repair"))
2196 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2197 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2199 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2200 md_wakeup_thread(mddev->thread);
2205 mismatch_cnt_show(mddev_t *mddev, char *page)
2207 return sprintf(page, "%llu\n",
2208 (unsigned long long) mddev->resync_mismatches);
2211 static struct md_sysfs_entry
2212 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2215 static struct md_sysfs_entry
2216 md_mismatches = __ATTR_RO(mismatch_cnt);
2219 sync_min_show(mddev_t *mddev, char *page)
2221 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2222 mddev->sync_speed_min ? "local": "system");
2226 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2230 if (strncmp(buf, "system", 6)==0) {
2231 mddev->sync_speed_min = 0;
2234 min = simple_strtoul(buf, &e, 10);
2235 if (buf == e || (*e && *e != '\n') || min <= 0)
2237 mddev->sync_speed_min = min;
2241 static struct md_sysfs_entry md_sync_min =
2242 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2245 sync_max_show(mddev_t *mddev, char *page)
2247 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2248 mddev->sync_speed_max ? "local": "system");
2252 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2256 if (strncmp(buf, "system", 6)==0) {
2257 mddev->sync_speed_max = 0;
2260 max = simple_strtoul(buf, &e, 10);
2261 if (buf == e || (*e && *e != '\n') || max <= 0)
2263 mddev->sync_speed_max = max;
2267 static struct md_sysfs_entry md_sync_max =
2268 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2272 sync_speed_show(mddev_t *mddev, char *page)
2274 unsigned long resync, dt, db;
2275 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2276 dt = ((jiffies - mddev->resync_mark) / HZ);
2278 db = resync - (mddev->resync_mark_cnt);
2279 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2282 static struct md_sysfs_entry
2283 md_sync_speed = __ATTR_RO(sync_speed);
2286 sync_completed_show(mddev_t *mddev, char *page)
2288 unsigned long max_blocks, resync;
2290 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2291 max_blocks = mddev->resync_max_sectors;
2293 max_blocks = mddev->size << 1;
2295 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2296 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2299 static struct md_sysfs_entry
2300 md_sync_completed = __ATTR_RO(sync_completed);
2302 static struct attribute *md_default_attrs[] = {
2304 &md_raid_disks.attr,
2305 &md_chunk_size.attr,
2308 &md_new_device.attr,
2312 static struct attribute *md_redundancy_attrs[] = {
2314 &md_mismatches.attr,
2317 &md_sync_speed.attr,
2318 &md_sync_completed.attr,
2321 static struct attribute_group md_redundancy_group = {
2323 .attrs = md_redundancy_attrs,
2328 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2330 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2331 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2337 rv = entry->show(mddev, page);
2338 mddev_unlock(mddev);
2343 md_attr_store(struct kobject *kobj, struct attribute *attr,
2344 const char *page, size_t length)
2346 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2347 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2353 rv = entry->store(mddev, page, length);
2354 mddev_unlock(mddev);
2358 static void md_free(struct kobject *ko)
2360 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2364 static struct sysfs_ops md_sysfs_ops = {
2365 .show = md_attr_show,
2366 .store = md_attr_store,
2368 static struct kobj_type md_ktype = {
2370 .sysfs_ops = &md_sysfs_ops,
2371 .default_attrs = md_default_attrs,
2376 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2378 static DECLARE_MUTEX(disks_sem);
2379 mddev_t *mddev = mddev_find(dev);
2380 struct gendisk *disk;
2381 int partitioned = (MAJOR(dev) != MD_MAJOR);
2382 int shift = partitioned ? MdpMinorShift : 0;
2383 int unit = MINOR(dev) >> shift;
2389 if (mddev->gendisk) {
2394 disk = alloc_disk(1 << shift);
2400 disk->major = MAJOR(dev);
2401 disk->first_minor = unit << shift;
2403 sprintf(disk->disk_name, "md_d%d", unit);
2404 sprintf(disk->devfs_name, "md/d%d", unit);
2406 sprintf(disk->disk_name, "md%d", unit);
2407 sprintf(disk->devfs_name, "md/%d", unit);
2409 disk->fops = &md_fops;
2410 disk->private_data = mddev;
2411 disk->queue = mddev->queue;
2413 mddev->gendisk = disk;
2415 mddev->kobj.parent = &disk->kobj;
2416 mddev->kobj.k_name = NULL;
2417 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2418 mddev->kobj.ktype = &md_ktype;
2419 kobject_register(&mddev->kobj);
2423 void md_wakeup_thread(mdk_thread_t *thread);
2425 static void md_safemode_timeout(unsigned long data)
2427 mddev_t *mddev = (mddev_t *) data;
2429 mddev->safemode = 1;
2430 md_wakeup_thread(mddev->thread);
2433 static int start_dirty_degraded;
2435 static int do_md_run(mddev_t * mddev)
2439 struct list_head *tmp;
2441 struct gendisk *disk;
2442 struct mdk_personality *pers;
2443 char b[BDEVNAME_SIZE];
2445 if (list_empty(&mddev->disks))
2446 /* cannot run an array with no devices.. */
2453 * Analyze all RAID superblock(s)
2455 if (!mddev->raid_disks)
2458 chunk_size = mddev->chunk_size;
2461 if (chunk_size > MAX_CHUNK_SIZE) {
2462 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2463 chunk_size, MAX_CHUNK_SIZE);
2467 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2469 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2470 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2473 if (chunk_size < PAGE_SIZE) {
2474 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2475 chunk_size, PAGE_SIZE);
2479 /* devices must have minimum size of one chunk */
2480 ITERATE_RDEV(mddev,rdev,tmp) {
2481 if (test_bit(Faulty, &rdev->flags))
2483 if (rdev->size < chunk_size / 1024) {
2485 "md: Dev %s smaller than chunk_size:"
2487 bdevname(rdev->bdev,b),
2488 (unsigned long long)rdev->size,
2496 if (mddev->level != LEVEL_NONE)
2497 request_module("md-level-%d", mddev->level);
2498 else if (mddev->clevel[0])
2499 request_module("md-%s", mddev->clevel);
2503 * Drop all container device buffers, from now on
2504 * the only valid external interface is through the md
2506 * Also find largest hardsector size
2508 ITERATE_RDEV(mddev,rdev,tmp) {
2509 if (test_bit(Faulty, &rdev->flags))
2511 sync_blockdev(rdev->bdev);
2512 invalidate_bdev(rdev->bdev, 0);
2515 md_probe(mddev->unit, NULL, NULL);
2516 disk = mddev->gendisk;
2520 spin_lock(&pers_lock);
2521 pers = find_pers(mddev->level, mddev->clevel);
2522 if (!pers || !try_module_get(pers->owner)) {
2523 spin_unlock(&pers_lock);
2524 if (mddev->level != LEVEL_NONE)
2525 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2528 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2533 spin_unlock(&pers_lock);
2534 mddev->level = pers->level;
2535 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2537 mddev->recovery = 0;
2538 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2539 mddev->barriers_work = 1;
2540 mddev->ok_start_degraded = start_dirty_degraded;
2543 mddev->ro = 2; /* read-only, but switch on first write */
2545 err = mddev->pers->run(mddev);
2546 if (!err && mddev->pers->sync_request) {
2547 err = bitmap_create(mddev);
2549 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2550 mdname(mddev), err);
2551 mddev->pers->stop(mddev);
2555 printk(KERN_ERR "md: pers->run() failed ...\n");
2556 module_put(mddev->pers->owner);
2558 bitmap_destroy(mddev);
2561 if (mddev->pers->sync_request)
2562 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2563 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2566 atomic_set(&mddev->writes_pending,0);
2567 mddev->safemode = 0;
2568 mddev->safemode_timer.function = md_safemode_timeout;
2569 mddev->safemode_timer.data = (unsigned long) mddev;
2570 mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2573 ITERATE_RDEV(mddev,rdev,tmp)
2574 if (rdev->raid_disk >= 0) {
2576 sprintf(nm, "rd%d", rdev->raid_disk);
2577 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2580 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2581 md_wakeup_thread(mddev->thread);
2583 if (mddev->sb_dirty)
2584 md_update_sb(mddev);
2586 set_capacity(disk, mddev->array_size<<1);
2588 /* If we call blk_queue_make_request here, it will
2589 * re-initialise max_sectors etc which may have been
2590 * refined inside -> run. So just set the bits we need to set.
2591 * Most initialisation happended when we called
2592 * blk_queue_make_request(..., md_fail_request)
2595 mddev->queue->queuedata = mddev;
2596 mddev->queue->make_request_fn = mddev->pers->make_request;
2599 md_new_event(mddev);
2603 static int restart_array(mddev_t *mddev)
2605 struct gendisk *disk = mddev->gendisk;
2609 * Complain if it has no devices
2612 if (list_empty(&mddev->disks))
2620 mddev->safemode = 0;
2622 set_disk_ro(disk, 0);
2624 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2627 * Kick recovery or resync if necessary
2629 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2630 md_wakeup_thread(mddev->thread);
2633 printk(KERN_ERR "md: %s has no personality assigned.\n",
2642 static int do_md_stop(mddev_t * mddev, int ro)
2645 struct gendisk *disk = mddev->gendisk;
2648 if (atomic_read(&mddev->active)>2) {
2649 printk("md: %s still in use.\n",mdname(mddev));
2653 if (mddev->sync_thread) {
2654 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2655 md_unregister_thread(mddev->sync_thread);
2656 mddev->sync_thread = NULL;
2659 del_timer_sync(&mddev->safemode_timer);
2661 invalidate_partition(disk, 0);
2669 bitmap_flush(mddev);
2670 md_super_wait(mddev);
2672 set_disk_ro(disk, 0);
2673 blk_queue_make_request(mddev->queue, md_fail_request);
2674 mddev->pers->stop(mddev);
2675 if (mddev->pers->sync_request)
2676 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2678 module_put(mddev->pers->owner);
2683 if (!mddev->in_sync) {
2684 /* mark array as shutdown cleanly */
2686 md_update_sb(mddev);
2689 set_disk_ro(disk, 1);
2693 * Free resources if final stop
2697 struct list_head *tmp;
2698 struct gendisk *disk;
2699 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2701 bitmap_destroy(mddev);
2702 if (mddev->bitmap_file) {
2703 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2704 fput(mddev->bitmap_file);
2705 mddev->bitmap_file = NULL;
2707 mddev->bitmap_offset = 0;
2709 ITERATE_RDEV(mddev,rdev,tmp)
2710 if (rdev->raid_disk >= 0) {
2712 sprintf(nm, "rd%d", rdev->raid_disk);
2713 sysfs_remove_link(&mddev->kobj, nm);
2716 export_array(mddev);
2718 mddev->array_size = 0;
2719 disk = mddev->gendisk;
2721 set_capacity(disk, 0);
2724 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2727 md_new_event(mddev);
2732 static void autorun_array(mddev_t *mddev)
2735 struct list_head *tmp;
2738 if (list_empty(&mddev->disks))
2741 printk(KERN_INFO "md: running: ");
2743 ITERATE_RDEV(mddev,rdev,tmp) {
2744 char b[BDEVNAME_SIZE];
2745 printk("<%s>", bdevname(rdev->bdev,b));
2749 err = do_md_run (mddev);
2751 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2752 do_md_stop (mddev, 0);
2757 * lets try to run arrays based on all disks that have arrived
2758 * until now. (those are in pending_raid_disks)
2760 * the method: pick the first pending disk, collect all disks with
2761 * the same UUID, remove all from the pending list and put them into
2762 * the 'same_array' list. Then order this list based on superblock
2763 * update time (freshest comes first), kick out 'old' disks and
2764 * compare superblocks. If everything's fine then run it.
2766 * If "unit" is allocated, then bump its reference count
2768 static void autorun_devices(int part)
2770 struct list_head candidates;
2771 struct list_head *tmp;
2772 mdk_rdev_t *rdev0, *rdev;
2774 char b[BDEVNAME_SIZE];
2776 printk(KERN_INFO "md: autorun ...\n");
2777 while (!list_empty(&pending_raid_disks)) {
2779 rdev0 = list_entry(pending_raid_disks.next,
2780 mdk_rdev_t, same_set);
2782 printk(KERN_INFO "md: considering %s ...\n",
2783 bdevname(rdev0->bdev,b));
2784 INIT_LIST_HEAD(&candidates);
2785 ITERATE_RDEV_PENDING(rdev,tmp)
2786 if (super_90_load(rdev, rdev0, 0) >= 0) {
2787 printk(KERN_INFO "md: adding %s ...\n",
2788 bdevname(rdev->bdev,b));
2789 list_move(&rdev->same_set, &candidates);
2792 * now we have a set of devices, with all of them having
2793 * mostly sane superblocks. It's time to allocate the
2796 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2797 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2798 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2802 dev = MKDEV(mdp_major,
2803 rdev0->preferred_minor << MdpMinorShift);
2805 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2807 md_probe(dev, NULL, NULL);
2808 mddev = mddev_find(dev);
2811 "md: cannot allocate memory for md drive.\n");
2814 if (mddev_lock(mddev))
2815 printk(KERN_WARNING "md: %s locked, cannot run\n",
2817 else if (mddev->raid_disks || mddev->major_version
2818 || !list_empty(&mddev->disks)) {
2820 "md: %s already running, cannot run %s\n",
2821 mdname(mddev), bdevname(rdev0->bdev,b));
2822 mddev_unlock(mddev);
2824 printk(KERN_INFO "md: created %s\n", mdname(mddev));
2825 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2826 list_del_init(&rdev->same_set);
2827 if (bind_rdev_to_array(rdev, mddev))
2830 autorun_array(mddev);
2831 mddev_unlock(mddev);
2833 /* on success, candidates will be empty, on error
2836 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2840 printk(KERN_INFO "md: ... autorun DONE.\n");
2844 * import RAID devices based on one partition
2845 * if possible, the array gets run as well.
2848 static int autostart_array(dev_t startdev)
2850 char b[BDEVNAME_SIZE];
2851 int err = -EINVAL, i;
2852 mdp_super_t *sb = NULL;
2853 mdk_rdev_t *start_rdev = NULL, *rdev;
2855 start_rdev = md_import_device(startdev, 0, 0);
2856 if (IS_ERR(start_rdev))
2860 /* NOTE: this can only work for 0.90.0 superblocks */
2861 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2862 if (sb->major_version != 0 ||
2863 sb->minor_version != 90 ) {
2864 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2865 export_rdev(start_rdev);
2869 if (test_bit(Faulty, &start_rdev->flags)) {
2871 "md: can not autostart based on faulty %s!\n",
2872 bdevname(start_rdev->bdev,b));
2873 export_rdev(start_rdev);
2876 list_add(&start_rdev->same_set, &pending_raid_disks);
2878 for (i = 0; i < MD_SB_DISKS; i++) {
2879 mdp_disk_t *desc = sb->disks + i;
2880 dev_t dev = MKDEV(desc->major, desc->minor);
2884 if (dev == startdev)
2886 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2888 rdev = md_import_device(dev, 0, 0);
2892 list_add(&rdev->same_set, &pending_raid_disks);
2896 * possibly return codes
2904 static int get_version(void __user * arg)
2908 ver.major = MD_MAJOR_VERSION;
2909 ver.minor = MD_MINOR_VERSION;
2910 ver.patchlevel = MD_PATCHLEVEL_VERSION;
2912 if (copy_to_user(arg, &ver, sizeof(ver)))
2918 static int get_array_info(mddev_t * mddev, void __user * arg)
2920 mdu_array_info_t info;
2921 int nr,working,active,failed,spare;
2923 struct list_head *tmp;
2925 nr=working=active=failed=spare=0;
2926 ITERATE_RDEV(mddev,rdev,tmp) {
2928 if (test_bit(Faulty, &rdev->flags))
2932 if (test_bit(In_sync, &rdev->flags))
2939 info.major_version = mddev->major_version;
2940 info.minor_version = mddev->minor_version;
2941 info.patch_version = MD_PATCHLEVEL_VERSION;
2942 info.ctime = mddev->ctime;
2943 info.level = mddev->level;
2944 info.size = mddev->size;
2945 if (info.size != mddev->size) /* overflow */
2948 info.raid_disks = mddev->raid_disks;
2949 info.md_minor = mddev->md_minor;
2950 info.not_persistent= !mddev->persistent;
2952 info.utime = mddev->utime;
2955 info.state = (1<<MD_SB_CLEAN);
2956 if (mddev->bitmap && mddev->bitmap_offset)
2957 info.state = (1<<MD_SB_BITMAP_PRESENT);
2958 info.active_disks = active;
2959 info.working_disks = working;
2960 info.failed_disks = failed;
2961 info.spare_disks = spare;
2963 info.layout = mddev->layout;
2964 info.chunk_size = mddev->chunk_size;
2966 if (copy_to_user(arg, &info, sizeof(info)))
2972 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2974 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2975 char *ptr, *buf = NULL;
2978 file = kmalloc(sizeof(*file), GFP_KERNEL);
2982 /* bitmap disabled, zero the first byte and copy out */
2983 if (!mddev->bitmap || !mddev->bitmap->file) {
2984 file->pathname[0] = '\0';
2988 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2992 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2996 strcpy(file->pathname, ptr);
3000 if (copy_to_user(arg, file, sizeof(*file)))
3008 static int get_disk_info(mddev_t * mddev, void __user * arg)
3010 mdu_disk_info_t info;
3014 if (copy_from_user(&info, arg, sizeof(info)))
3019 rdev = find_rdev_nr(mddev, nr);
3021 info.major = MAJOR(rdev->bdev->bd_dev);
3022 info.minor = MINOR(rdev->bdev->bd_dev);
3023 info.raid_disk = rdev->raid_disk;
3025 if (test_bit(Faulty, &rdev->flags))
3026 info.state |= (1<<MD_DISK_FAULTY);
3027 else if (test_bit(In_sync, &rdev->flags)) {
3028 info.state |= (1<<MD_DISK_ACTIVE);
3029 info.state |= (1<<MD_DISK_SYNC);
3031 if (test_bit(WriteMostly, &rdev->flags))
3032 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3034 info.major = info.minor = 0;
3035 info.raid_disk = -1;
3036 info.state = (1<<MD_DISK_REMOVED);
3039 if (copy_to_user(arg, &info, sizeof(info)))
3045 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3047 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3049 dev_t dev = MKDEV(info->major,info->minor);
3051 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3054 if (!mddev->raid_disks) {
3056 /* expecting a device which has a superblock */
3057 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3060 "md: md_import_device returned %ld\n",
3062 return PTR_ERR(rdev);
3064 if (!list_empty(&mddev->disks)) {
3065 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3066 mdk_rdev_t, same_set);
3067 int err = super_types[mddev->major_version]
3068 .load_super(rdev, rdev0, mddev->minor_version);
3071 "md: %s has different UUID to %s\n",
3072 bdevname(rdev->bdev,b),
3073 bdevname(rdev0->bdev,b2));
3078 err = bind_rdev_to_array(rdev, mddev);
3085 * add_new_disk can be used once the array is assembled
3086 * to add "hot spares". They must already have a superblock
3091 if (!mddev->pers->hot_add_disk) {
3093 "%s: personality does not support diskops!\n",
3097 if (mddev->persistent)
3098 rdev = md_import_device(dev, mddev->major_version,
3099 mddev->minor_version);
3101 rdev = md_import_device(dev, -1, -1);
3104 "md: md_import_device returned %ld\n",
3106 return PTR_ERR(rdev);
3108 /* set save_raid_disk if appropriate */
3109 if (!mddev->persistent) {
3110 if (info->state & (1<<MD_DISK_SYNC) &&
3111 info->raid_disk < mddev->raid_disks)
3112 rdev->raid_disk = info->raid_disk;
3114 rdev->raid_disk = -1;
3116 super_types[mddev->major_version].
3117 validate_super(mddev, rdev);
3118 rdev->saved_raid_disk = rdev->raid_disk;
3120 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3121 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3122 set_bit(WriteMostly, &rdev->flags);
3124 rdev->raid_disk = -1;
3125 err = bind_rdev_to_array(rdev, mddev);
3129 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3130 md_wakeup_thread(mddev->thread);
3134 /* otherwise, add_new_disk is only allowed
3135 * for major_version==0 superblocks
3137 if (mddev->major_version != 0) {
3138 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3143 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3145 rdev = md_import_device (dev, -1, 0);
3148 "md: error, md_import_device() returned %ld\n",
3150 return PTR_ERR(rdev);
3152 rdev->desc_nr = info->number;
3153 if (info->raid_disk < mddev->raid_disks)
3154 rdev->raid_disk = info->raid_disk;
3156 rdev->raid_disk = -1;
3160 if (rdev->raid_disk < mddev->raid_disks)
3161 if (info->state & (1<<MD_DISK_SYNC))
3162 set_bit(In_sync, &rdev->flags);
3164 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3165 set_bit(WriteMostly, &rdev->flags);
3167 if (!mddev->persistent) {
3168 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3169 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3171 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3172 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3174 err = bind_rdev_to_array(rdev, mddev);
3184 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3186 char b[BDEVNAME_SIZE];
3192 rdev = find_rdev(mddev, dev);
3196 if (rdev->raid_disk >= 0)
3199 kick_rdev_from_array(rdev);
3200 md_update_sb(mddev);
3201 md_new_event(mddev);
3205 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3206 bdevname(rdev->bdev,b), mdname(mddev));
3210 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3212 char b[BDEVNAME_SIZE];
3220 if (mddev->major_version != 0) {
3221 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3222 " version-0 superblocks.\n",
3226 if (!mddev->pers->hot_add_disk) {
3228 "%s: personality does not support diskops!\n",
3233 rdev = md_import_device (dev, -1, 0);
3236 "md: error, md_import_device() returned %ld\n",
3241 if (mddev->persistent)
3242 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3245 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3247 size = calc_dev_size(rdev, mddev->chunk_size);
3250 if (test_bit(Faulty, &rdev->flags)) {
3252 "md: can not hot-add faulty %s disk to %s!\n",
3253 bdevname(rdev->bdev,b), mdname(mddev));
3257 clear_bit(In_sync, &rdev->flags);
3259 err = bind_rdev_to_array(rdev, mddev);
3264 * The rest should better be atomic, we can have disk failures
3265 * noticed in interrupt contexts ...
3268 if (rdev->desc_nr == mddev->max_disks) {
3269 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3272 goto abort_unbind_export;
3275 rdev->raid_disk = -1;
3277 md_update_sb(mddev);
3280 * Kick recovery, maybe this spare has to be added to the
3281 * array immediately.
3283 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3284 md_wakeup_thread(mddev->thread);
3285 md_new_event(mddev);
3288 abort_unbind_export:
3289 unbind_rdev_from_array(rdev);
3296 /* similar to deny_write_access, but accounts for our holding a reference
3297 * to the file ourselves */
3298 static int deny_bitmap_write_access(struct file * file)
3300 struct inode *inode = file->f_mapping->host;
3302 spin_lock(&inode->i_lock);
3303 if (atomic_read(&inode->i_writecount) > 1) {
3304 spin_unlock(&inode->i_lock);
3307 atomic_set(&inode->i_writecount, -1);
3308 spin_unlock(&inode->i_lock);
3313 static int set_bitmap_file(mddev_t *mddev, int fd)
3318 if (!mddev->pers->quiesce)
3320 if (mddev->recovery || mddev->sync_thread)
3322 /* we should be able to change the bitmap.. */
3328 return -EEXIST; /* cannot add when bitmap is present */
3329 mddev->bitmap_file = fget(fd);
3331 if (mddev->bitmap_file == NULL) {
3332 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3337 err = deny_bitmap_write_access(mddev->bitmap_file);
3339 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3341 fput(mddev->bitmap_file);
3342 mddev->bitmap_file = NULL;
3345 mddev->bitmap_offset = 0; /* file overrides offset */
3346 } else if (mddev->bitmap == NULL)
3347 return -ENOENT; /* cannot remove what isn't there */
3350 mddev->pers->quiesce(mddev, 1);
3352 err = bitmap_create(mddev);
3354 bitmap_destroy(mddev);
3355 mddev->pers->quiesce(mddev, 0);
3356 } else if (fd < 0) {
3357 if (mddev->bitmap_file)
3358 fput(mddev->bitmap_file);
3359 mddev->bitmap_file = NULL;
3366 * set_array_info is used two different ways
3367 * The original usage is when creating a new array.
3368 * In this usage, raid_disks is > 0 and it together with
3369 * level, size, not_persistent,layout,chunksize determine the
3370 * shape of the array.
3371 * This will always create an array with a type-0.90.0 superblock.
3372 * The newer usage is when assembling an array.
3373 * In this case raid_disks will be 0, and the major_version field is
3374 * use to determine which style super-blocks are to be found on the devices.
3375 * The minor and patch _version numbers are also kept incase the
3376 * super_block handler wishes to interpret them.
3378 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3381 if (info->raid_disks == 0) {
3382 /* just setting version number for superblock loading */
3383 if (info->major_version < 0 ||
3384 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3385 super_types[info->major_version].name == NULL) {
3386 /* maybe try to auto-load a module? */
3388 "md: superblock version %d not known\n",
3389 info->major_version);
3392 mddev->major_version = info->major_version;
3393 mddev->minor_version = info->minor_version;
3394 mddev->patch_version = info->patch_version;
3397 mddev->major_version = MD_MAJOR_VERSION;
3398 mddev->minor_version = MD_MINOR_VERSION;
3399 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3400 mddev->ctime = get_seconds();
3402 mddev->level = info->level;
3403 mddev->clevel[0] = 0;
3404 mddev->size = info->size;
3405 mddev->raid_disks = info->raid_disks;
3406 /* don't set md_minor, it is determined by which /dev/md* was
3409 if (info->state & (1<<MD_SB_CLEAN))
3410 mddev->recovery_cp = MaxSector;
3412 mddev->recovery_cp = 0;
3413 mddev->persistent = ! info->not_persistent;
3415 mddev->layout = info->layout;
3416 mddev->chunk_size = info->chunk_size;
3418 mddev->max_disks = MD_SB_DISKS;
3420 mddev->sb_dirty = 1;
3422 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3423 mddev->bitmap_offset = 0;
3426 * Generate a 128 bit UUID
3428 get_random_bytes(mddev->uuid, 16);
3433 static int update_size(mddev_t *mddev, unsigned long size)
3437 struct list_head *tmp;
3439 if (mddev->pers->resize == NULL)
3441 /* The "size" is the amount of each device that is used.
3442 * This can only make sense for arrays with redundancy.
3443 * linear and raid0 always use whatever space is available
3444 * We can only consider changing the size if no resync
3445 * or reconstruction is happening, and if the new size
3446 * is acceptable. It must fit before the sb_offset or,
3447 * if that is <data_offset, it must fit before the
3448 * size of each device.
3449 * If size is zero, we find the largest size that fits.
3451 if (mddev->sync_thread)
3453 ITERATE_RDEV(mddev,rdev,tmp) {
3455 int fit = (size == 0);
3456 if (rdev->sb_offset > rdev->data_offset)
3457 avail = (rdev->sb_offset*2) - rdev->data_offset;
3459 avail = get_capacity(rdev->bdev->bd_disk)
3460 - rdev->data_offset;
3461 if (fit && (size == 0 || size > avail/2))
3463 if (avail < ((sector_t)size << 1))
3466 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3468 struct block_device *bdev;
3470 bdev = bdget_disk(mddev->gendisk, 0);
3472 mutex_lock(&bdev->bd_inode->i_mutex);
3473 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
3474 mutex_unlock(&bdev->bd_inode->i_mutex);
3481 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3484 /* change the number of raid disks */
3485 if (mddev->pers->reshape == NULL)
3487 if (raid_disks <= 0 ||
3488 raid_disks >= mddev->max_disks)
3490 if (mddev->sync_thread)
3492 rv = mddev->pers->reshape(mddev, raid_disks);
3498 * update_array_info is used to change the configuration of an
3500 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3501 * fields in the info are checked against the array.
3502 * Any differences that cannot be handled will cause an error.
3503 * Normally, only one change can be managed at a time.
3505 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3511 /* calculate expected state,ignoring low bits */
3512 if (mddev->bitmap && mddev->bitmap_offset)
3513 state |= (1 << MD_SB_BITMAP_PRESENT);
3515 if (mddev->major_version != info->major_version ||
3516 mddev->minor_version != info->minor_version ||
3517 /* mddev->patch_version != info->patch_version || */
3518 mddev->ctime != info->ctime ||
3519 mddev->level != info->level ||
3520 /* mddev->layout != info->layout || */
3521 !mddev->persistent != info->not_persistent||
3522 mddev->chunk_size != info->chunk_size ||
3523 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3524 ((state^info->state) & 0xfffffe00)
3527 /* Check there is only one change */
3528 if (info->size >= 0 && mddev->size != info->size) cnt++;
3529 if (mddev->raid_disks != info->raid_disks) cnt++;
3530 if (mddev->layout != info->layout) cnt++;
3531 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3532 if (cnt == 0) return 0;
3533 if (cnt > 1) return -EINVAL;
3535 if (mddev->layout != info->layout) {
3537 * we don't need to do anything at the md level, the
3538 * personality will take care of it all.
3540 if (mddev->pers->reconfig == NULL)
3543 return mddev->pers->reconfig(mddev, info->layout, -1);
3545 if (info->size >= 0 && mddev->size != info->size)
3546 rv = update_size(mddev, info->size);
3548 if (mddev->raid_disks != info->raid_disks)
3549 rv = update_raid_disks(mddev, info->raid_disks);
3551 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3552 if (mddev->pers->quiesce == NULL)
3554 if (mddev->recovery || mddev->sync_thread)
3556 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3557 /* add the bitmap */
3560 if (mddev->default_bitmap_offset == 0)
3562 mddev->bitmap_offset = mddev->default_bitmap_offset;
3563 mddev->pers->quiesce(mddev, 1);
3564 rv = bitmap_create(mddev);
3566 bitmap_destroy(mddev);
3567 mddev->pers->quiesce(mddev, 0);
3569 /* remove the bitmap */
3572 if (mddev->bitmap->file)
3574 mddev->pers->quiesce(mddev, 1);
3575 bitmap_destroy(mddev);
3576 mddev->pers->quiesce(mddev, 0);
3577 mddev->bitmap_offset = 0;
3580 md_update_sb(mddev);
3584 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3588 if (mddev->pers == NULL)
3591 rdev = find_rdev(mddev, dev);
3595 md_error(mddev, rdev);
3599 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3601 mddev_t *mddev = bdev->bd_disk->private_data;
3605 geo->cylinders = get_capacity(mddev->gendisk) / 8;
3609 static int md_ioctl(struct inode *inode, struct file *file,
3610 unsigned int cmd, unsigned long arg)
3613 void __user *argp = (void __user *)arg;
3614 mddev_t *mddev = NULL;
3616 if (!capable(CAP_SYS_ADMIN))
3620 * Commands dealing with the RAID driver but not any
3626 err = get_version(argp);
3629 case PRINT_RAID_DEBUG:
3637 autostart_arrays(arg);
3644 * Commands creating/starting a new array:
3647 mddev = inode->i_bdev->bd_disk->private_data;
3655 if (cmd == START_ARRAY) {
3656 /* START_ARRAY doesn't need to lock the array as autostart_array
3657 * does the locking, and it could even be a different array
3662 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3663 "This will not be supported beyond July 2006\n",
3664 current->comm, current->pid);
3667 err = autostart_array(new_decode_dev(arg));
3669 printk(KERN_WARNING "md: autostart failed!\n");
3675 err = mddev_lock(mddev);
3678 "md: ioctl lock interrupted, reason %d, cmd %d\n",
3685 case SET_ARRAY_INFO:
3687 mdu_array_info_t info;
3689 memset(&info, 0, sizeof(info));
3690 else if (copy_from_user(&info, argp, sizeof(info))) {
3695 err = update_array_info(mddev, &info);
3697 printk(KERN_WARNING "md: couldn't update"
3698 " array info. %d\n", err);
3703 if (!list_empty(&mddev->disks)) {
3705 "md: array %s already has disks!\n",
3710 if (mddev->raid_disks) {
3712 "md: array %s already initialised!\n",
3717 err = set_array_info(mddev, &info);
3719 printk(KERN_WARNING "md: couldn't set"
3720 " array info. %d\n", err);
3730 * Commands querying/configuring an existing array:
3732 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3733 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3734 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3735 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3741 * Commands even a read-only array can execute:
3745 case GET_ARRAY_INFO:
3746 err = get_array_info(mddev, argp);
3749 case GET_BITMAP_FILE:
3750 err = get_bitmap_file(mddev, argp);
3754 err = get_disk_info(mddev, argp);
3757 case RESTART_ARRAY_RW:
3758 err = restart_array(mddev);
3762 err = do_md_stop (mddev, 0);
3766 err = do_md_stop (mddev, 1);
3770 * We have a problem here : there is no easy way to give a CHS
3771 * virtual geometry. We currently pretend that we have a 2 heads
3772 * 4 sectors (with a BIG number of cylinders...). This drives
3773 * dosfs just mad... ;-)
3778 * The remaining ioctls are changing the state of the
3779 * superblock, so we do not allow them on read-only arrays.
3780 * However non-MD ioctls (e.g. get-size) will still come through
3781 * here and hit the 'default' below, so only disallow
3782 * 'md' ioctls, and switch to rw mode if started auto-readonly.
3784 if (_IOC_TYPE(cmd) == MD_MAJOR &&
3785 mddev->ro && mddev->pers) {
3786 if (mddev->ro == 2) {
3788 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3789 md_wakeup_thread(mddev->thread);
3801 mdu_disk_info_t info;
3802 if (copy_from_user(&info, argp, sizeof(info)))
3805 err = add_new_disk(mddev, &info);
3809 case HOT_REMOVE_DISK:
3810 err = hot_remove_disk(mddev, new_decode_dev(arg));
3814 err = hot_add_disk(mddev, new_decode_dev(arg));
3817 case SET_DISK_FAULTY:
3818 err = set_disk_faulty(mddev, new_decode_dev(arg));
3822 err = do_md_run (mddev);
3825 case SET_BITMAP_FILE:
3826 err = set_bitmap_file(mddev, (int)arg);
3830 if (_IOC_TYPE(cmd) == MD_MAJOR)
3831 printk(KERN_WARNING "md: %s(pid %d) used"
3832 " obsolete MD ioctl, upgrade your"
3833 " software to use new ictls.\n",
3834 current->comm, current->pid);
3841 mddev_unlock(mddev);
3851 static int md_open(struct inode *inode, struct file *file)
3854 * Succeed if we can lock the mddev, which confirms that
3855 * it isn't being stopped right now.
3857 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3860 if ((err = mddev_lock(mddev)))
3865 mddev_unlock(mddev);
3867 check_disk_change(inode->i_bdev);
3872 static int md_release(struct inode *inode, struct file * file)
3874 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3883 static int md_media_changed(struct gendisk *disk)
3885 mddev_t *mddev = disk->private_data;
3887 return mddev->changed;
3890 static int md_revalidate(struct gendisk *disk)
3892 mddev_t *mddev = disk->private_data;
3897 static struct block_device_operations md_fops =
3899 .owner = THIS_MODULE,
3901 .release = md_release,
3903 .getgeo = md_getgeo,
3904 .media_changed = md_media_changed,
3905 .revalidate_disk= md_revalidate,
3908 static int md_thread(void * arg)
3910 mdk_thread_t *thread = arg;
3913 * md_thread is a 'system-thread', it's priority should be very
3914 * high. We avoid resource deadlocks individually in each
3915 * raid personality. (RAID5 does preallocation) We also use RR and
3916 * the very same RT priority as kswapd, thus we will never get
3917 * into a priority inversion deadlock.
3919 * we definitely have to have equal or higher priority than
3920 * bdflush, otherwise bdflush will deadlock if there are too
3921 * many dirty RAID5 blocks.
3924 allow_signal(SIGKILL);
3925 while (!kthread_should_stop()) {
3927 /* We need to wait INTERRUPTIBLE so that
3928 * we don't add to the load-average.
3929 * That means we need to be sure no signals are
3932 if (signal_pending(current))
3933 flush_signals(current);
3935 wait_event_interruptible_timeout
3937 test_bit(THREAD_WAKEUP, &thread->flags)
3938 || kthread_should_stop(),
3942 clear_bit(THREAD_WAKEUP, &thread->flags);
3944 thread->run(thread->mddev);
3950 void md_wakeup_thread(mdk_thread_t *thread)
3953 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3954 set_bit(THREAD_WAKEUP, &thread->flags);
3955 wake_up(&thread->wqueue);
3959 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3962 mdk_thread_t *thread;
3964 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3968 init_waitqueue_head(&thread->wqueue);
3971 thread->mddev = mddev;
3972 thread->timeout = MAX_SCHEDULE_TIMEOUT;
3973 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3974 if (IS_ERR(thread->tsk)) {
3981 void md_unregister_thread(mdk_thread_t *thread)
3983 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3985 kthread_stop(thread->tsk);
3989 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3996 if (!rdev || test_bit(Faulty, &rdev->flags))
3999 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4001 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4002 __builtin_return_address(0),__builtin_return_address(1),
4003 __builtin_return_address(2),__builtin_return_address(3));
4005 if (!mddev->pers->error_handler)
4007 mddev->pers->error_handler(mddev,rdev);
4008 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4009 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4010 md_wakeup_thread(mddev->thread);
4011 md_new_event(mddev);
4014 /* seq_file implementation /proc/mdstat */
4016 static void status_unused(struct seq_file *seq)
4020 struct list_head *tmp;
4022 seq_printf(seq, "unused devices: ");
4024 ITERATE_RDEV_PENDING(rdev,tmp) {
4025 char b[BDEVNAME_SIZE];
4027 seq_printf(seq, "%s ",
4028 bdevname(rdev->bdev,b));
4031 seq_printf(seq, "<none>");
4033 seq_printf(seq, "\n");
4037 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4039 unsigned long max_blocks, resync, res, dt, db, rt;
4041 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4043 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4044 max_blocks = mddev->resync_max_sectors >> 1;
4046 max_blocks = mddev->size;
4049 * Should not happen.
4055 res = (resync/1024)*1000/(max_blocks/1024 + 1);
4057 int i, x = res/50, y = 20-x;
4058 seq_printf(seq, "[");
4059 for (i = 0; i < x; i++)
4060 seq_printf(seq, "=");
4061 seq_printf(seq, ">");
4062 for (i = 0; i < y; i++)
4063 seq_printf(seq, ".");
4064 seq_printf(seq, "] ");
4066 seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
4067 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4068 "resync" : "recovery"),
4069 res/10, res % 10, resync, max_blocks);
4072 * We do not want to overflow, so the order of operands and
4073 * the * 100 / 100 trick are important. We do a +1 to be
4074 * safe against division by zero. We only estimate anyway.
4076 * dt: time from mark until now
4077 * db: blocks written from mark until now
4078 * rt: remaining time
4080 dt = ((jiffies - mddev->resync_mark) / HZ);
4082 db = resync - (mddev->resync_mark_cnt/2);
4083 rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
4085 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4087 seq_printf(seq, " speed=%ldK/sec", db/dt);
4090 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4092 struct list_head *tmp;
4102 spin_lock(&all_mddevs_lock);
4103 list_for_each(tmp,&all_mddevs)
4105 mddev = list_entry(tmp, mddev_t, all_mddevs);
4107 spin_unlock(&all_mddevs_lock);
4110 spin_unlock(&all_mddevs_lock);
4112 return (void*)2;/* tail */
4116 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4118 struct list_head *tmp;
4119 mddev_t *next_mddev, *mddev = v;
4125 spin_lock(&all_mddevs_lock);
4127 tmp = all_mddevs.next;
4129 tmp = mddev->all_mddevs.next;
4130 if (tmp != &all_mddevs)
4131 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4133 next_mddev = (void*)2;
4136 spin_unlock(&all_mddevs_lock);
4144 static void md_seq_stop(struct seq_file *seq, void *v)
4148 if (mddev && v != (void*)1 && v != (void*)2)
4152 struct mdstat_info {
4156 static int md_seq_show(struct seq_file *seq, void *v)
4160 struct list_head *tmp2;
4162 struct mdstat_info *mi = seq->private;
4163 struct bitmap *bitmap;
4165 if (v == (void*)1) {
4166 struct mdk_personality *pers;
4167 seq_printf(seq, "Personalities : ");
4168 spin_lock(&pers_lock);
4169 list_for_each_entry(pers, &pers_list, list)
4170 seq_printf(seq, "[%s] ", pers->name);
4172 spin_unlock(&pers_lock);
4173 seq_printf(seq, "\n");
4174 mi->event = atomic_read(&md_event_count);
4177 if (v == (void*)2) {
4182 if (mddev_lock(mddev)!=0)
4184 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4185 seq_printf(seq, "%s : %sactive", mdname(mddev),
4186 mddev->pers ? "" : "in");
4189 seq_printf(seq, " (read-only)");
4191 seq_printf(seq, "(auto-read-only)");
4192 seq_printf(seq, " %s", mddev->pers->name);
4196 ITERATE_RDEV(mddev,rdev,tmp2) {
4197 char b[BDEVNAME_SIZE];
4198 seq_printf(seq, " %s[%d]",
4199 bdevname(rdev->bdev,b), rdev->desc_nr);
4200 if (test_bit(WriteMostly, &rdev->flags))
4201 seq_printf(seq, "(W)");
4202 if (test_bit(Faulty, &rdev->flags)) {
4203 seq_printf(seq, "(F)");
4205 } else if (rdev->raid_disk < 0)
4206 seq_printf(seq, "(S)"); /* spare */
4210 if (!list_empty(&mddev->disks)) {
4212 seq_printf(seq, "\n %llu blocks",
4213 (unsigned long long)mddev->array_size);
4215 seq_printf(seq, "\n %llu blocks",
4216 (unsigned long long)size);
4218 if (mddev->persistent) {
4219 if (mddev->major_version != 0 ||
4220 mddev->minor_version != 90) {
4221 seq_printf(seq," super %d.%d",
4222 mddev->major_version,
4223 mddev->minor_version);
4226 seq_printf(seq, " super non-persistent");
4229 mddev->pers->status (seq, mddev);
4230 seq_printf(seq, "\n ");
4231 if (mddev->pers->sync_request) {
4232 if (mddev->curr_resync > 2) {
4233 status_resync (seq, mddev);
4234 seq_printf(seq, "\n ");
4235 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4236 seq_printf(seq, "\tresync=DELAYED\n ");
4237 else if (mddev->recovery_cp < MaxSector)
4238 seq_printf(seq, "\tresync=PENDING\n ");
4241 seq_printf(seq, "\n ");
4243 if ((bitmap = mddev->bitmap)) {
4244 unsigned long chunk_kb;
4245 unsigned long flags;
4246 spin_lock_irqsave(&bitmap->lock, flags);
4247 chunk_kb = bitmap->chunksize >> 10;
4248 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4250 bitmap->pages - bitmap->missing_pages,
4252 (bitmap->pages - bitmap->missing_pages)
4253 << (PAGE_SHIFT - 10),
4254 chunk_kb ? chunk_kb : bitmap->chunksize,
4255 chunk_kb ? "KB" : "B");
4257 seq_printf(seq, ", file: ");
4258 seq_path(seq, bitmap->file->f_vfsmnt,
4259 bitmap->file->f_dentry," \t\n");
4262 seq_printf(seq, "\n");
4263 spin_unlock_irqrestore(&bitmap->lock, flags);
4266 seq_printf(seq, "\n");
4268 mddev_unlock(mddev);
4273 static struct seq_operations md_seq_ops = {
4274 .start = md_seq_start,
4275 .next = md_seq_next,
4276 .stop = md_seq_stop,
4277 .show = md_seq_show,
4280 static int md_seq_open(struct inode *inode, struct file *file)
4283 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4287 error = seq_open(file, &md_seq_ops);
4291 struct seq_file *p = file->private_data;
4293 mi->event = atomic_read(&md_event_count);
4298 static int md_seq_release(struct inode *inode, struct file *file)
4300 struct seq_file *m = file->private_data;
4301 struct mdstat_info *mi = m->private;
4304 return seq_release(inode, file);
4307 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4309 struct seq_file *m = filp->private_data;
4310 struct mdstat_info *mi = m->private;
4313 poll_wait(filp, &md_event_waiters, wait);
4315 /* always allow read */
4316 mask = POLLIN | POLLRDNORM;
4318 if (mi->event != atomic_read(&md_event_count))
4319 mask |= POLLERR | POLLPRI;
4323 static struct file_operations md_seq_fops = {
4324 .open = md_seq_open,
4326 .llseek = seq_lseek,
4327 .release = md_seq_release,
4328 .poll = mdstat_poll,
4331 int register_md_personality(struct mdk_personality *p)
4333 spin_lock(&pers_lock);
4334 list_add_tail(&p->list, &pers_list);
4335 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4336 spin_unlock(&pers_lock);
4340 int unregister_md_personality(struct mdk_personality *p)
4342 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4343 spin_lock(&pers_lock);
4344 list_del_init(&p->list);
4345 spin_unlock(&pers_lock);
4349 static int is_mddev_idle(mddev_t *mddev)
4352 struct list_head *tmp;
4354 unsigned long curr_events;
4357 ITERATE_RDEV(mddev,rdev,tmp) {
4358 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4359 curr_events = disk_stat_read(disk, sectors[0]) +
4360 disk_stat_read(disk, sectors[1]) -
4361 atomic_read(&disk->sync_io);
4362 /* The difference between curr_events and last_events
4363 * will be affected by any new non-sync IO (making
4364 * curr_events bigger) and any difference in the amount of
4365 * in-flight syncio (making current_events bigger or smaller)
4366 * The amount in-flight is currently limited to
4367 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4368 * which is at most 4096 sectors.
4369 * These numbers are fairly fragile and should be made
4370 * more robust, probably by enforcing the
4371 * 'window size' that md_do_sync sort-of uses.
4373 * Note: the following is an unsigned comparison.
4375 if ((curr_events - rdev->last_events + 4096) > 8192) {
4376 rdev->last_events = curr_events;
4383 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4385 /* another "blocks" (512byte) blocks have been synced */
4386 atomic_sub(blocks, &mddev->recovery_active);
4387 wake_up(&mddev->recovery_wait);
4389 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4390 md_wakeup_thread(mddev->thread);
4391 // stop recovery, signal do_sync ....
4396 /* md_write_start(mddev, bi)
4397 * If we need to update some array metadata (e.g. 'active' flag
4398 * in superblock) before writing, schedule a superblock update
4399 * and wait for it to complete.
4401 void md_write_start(mddev_t *mddev, struct bio *bi)
4403 if (bio_data_dir(bi) != WRITE)
4406 BUG_ON(mddev->ro == 1);
4407 if (mddev->ro == 2) {
4408 /* need to switch to read/write */
4410 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4411 md_wakeup_thread(mddev->thread);
4413 atomic_inc(&mddev->writes_pending);
4414 if (mddev->in_sync) {
4415 spin_lock_irq(&mddev->write_lock);
4416 if (mddev->in_sync) {
4418 mddev->sb_dirty = 1;
4419 md_wakeup_thread(mddev->thread);
4421 spin_unlock_irq(&mddev->write_lock);
4423 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4426 void md_write_end(mddev_t *mddev)
4428 if (atomic_dec_and_test(&mddev->writes_pending)) {
4429 if (mddev->safemode == 2)
4430 md_wakeup_thread(mddev->thread);
4432 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4436 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4438 #define SYNC_MARKS 10
4439 #define SYNC_MARK_STEP (3*HZ)
4440 static void md_do_sync(mddev_t *mddev)
4443 unsigned int currspeed = 0,
4445 sector_t max_sectors,j, io_sectors;
4446 unsigned long mark[SYNC_MARKS];
4447 sector_t mark_cnt[SYNC_MARKS];
4449 struct list_head *tmp;
4450 sector_t last_check;
4453 /* just incase thread restarts... */
4454 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4457 /* we overload curr_resync somewhat here.
4458 * 0 == not engaged in resync at all
4459 * 2 == checking that there is no conflict with another sync
4460 * 1 == like 2, but have yielded to allow conflicting resync to
4462 * other == active in resync - this many blocks
4464 * Before starting a resync we must have set curr_resync to
4465 * 2, and then checked that every "conflicting" array has curr_resync
4466 * less than ours. When we find one that is the same or higher
4467 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4468 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4469 * This will mean we have to start checking from the beginning again.
4474 mddev->curr_resync = 2;
4477 if (kthread_should_stop()) {
4478 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4481 ITERATE_MDDEV(mddev2,tmp) {
4482 if (mddev2 == mddev)
4484 if (mddev2->curr_resync &&
4485 match_mddev_units(mddev,mddev2)) {
4487 if (mddev < mddev2 && mddev->curr_resync == 2) {
4488 /* arbitrarily yield */
4489 mddev->curr_resync = 1;
4490 wake_up(&resync_wait);
4492 if (mddev > mddev2 && mddev->curr_resync == 1)
4493 /* no need to wait here, we can wait the next
4494 * time 'round when curr_resync == 2
4497 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4498 if (!kthread_should_stop() &&
4499 mddev2->curr_resync >= mddev->curr_resync) {
4500 printk(KERN_INFO "md: delaying resync of %s"
4501 " until %s has finished resync (they"
4502 " share one or more physical units)\n",
4503 mdname(mddev), mdname(mddev2));
4506 finish_wait(&resync_wait, &wq);
4509 finish_wait(&resync_wait, &wq);
4512 } while (mddev->curr_resync < 2);
4514 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4515 /* resync follows the size requested by the personality,
4516 * which defaults to physical size, but can be virtual size
4518 max_sectors = mddev->resync_max_sectors;
4519 mddev->resync_mismatches = 0;
4521 /* recovery follows the physical size of devices */
4522 max_sectors = mddev->size << 1;
4524 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4525 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4526 " %d KB/sec/disc.\n", speed_min(mddev));
4527 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4528 "(but not more than %d KB/sec) for reconstruction.\n",
4531 is_mddev_idle(mddev); /* this also initializes IO event counters */
4532 /* we don't use the checkpoint if there's a bitmap */
4533 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4534 && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4535 j = mddev->recovery_cp;
4539 for (m = 0; m < SYNC_MARKS; m++) {
4541 mark_cnt[m] = io_sectors;
4544 mddev->resync_mark = mark[last_mark];
4545 mddev->resync_mark_cnt = mark_cnt[last_mark];
4548 * Tune reconstruction:
4550 window = 32*(PAGE_SIZE/512);
4551 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4552 window/2,(unsigned long long) max_sectors/2);
4554 atomic_set(&mddev->recovery_active, 0);
4555 init_waitqueue_head(&mddev->recovery_wait);
4560 "md: resuming recovery of %s from checkpoint.\n",
4562 mddev->curr_resync = j;
4565 while (j < max_sectors) {
4569 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4570 currspeed < speed_min(mddev));
4572 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4576 if (!skipped) { /* actual IO requested */
4577 io_sectors += sectors;
4578 atomic_add(sectors, &mddev->recovery_active);
4582 if (j>1) mddev->curr_resync = j;
4583 if (last_check == 0)
4584 /* this is the earliers that rebuilt will be
4585 * visible in /proc/mdstat
4587 md_new_event(mddev);
4589 if (last_check + window > io_sectors || j == max_sectors)
4592 last_check = io_sectors;
4594 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4595 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4599 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4601 int next = (last_mark+1) % SYNC_MARKS;
4603 mddev->resync_mark = mark[next];
4604 mddev->resync_mark_cnt = mark_cnt[next];
4605 mark[next] = jiffies;
4606 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4611 if (kthread_should_stop()) {
4613 * got a signal, exit.
4616 "md: md_do_sync() got signal ... exiting\n");
4617 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4622 * this loop exits only if either when we are slower than
4623 * the 'hard' speed limit, or the system was IO-idle for
4625 * the system might be non-idle CPU-wise, but we only care
4626 * about not overloading the IO subsystem. (things like an
4627 * e2fsck being done on the RAID array should execute fast)
4629 mddev->queue->unplug_fn(mddev->queue);
4632 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4633 /((jiffies-mddev->resync_mark)/HZ +1) +1;
4635 if (currspeed > speed_min(mddev)) {
4636 if ((currspeed > speed_max(mddev)) ||
4637 !is_mddev_idle(mddev)) {
4643 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4645 * this also signals 'finished resyncing' to md_stop
4648 mddev->queue->unplug_fn(mddev->queue);
4650 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4652 /* tell personality that we are finished */
4653 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4655 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4656 mddev->curr_resync > 2 &&
4657 mddev->curr_resync >= mddev->recovery_cp) {
4658 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4660 "md: checkpointing recovery of %s.\n",
4662 mddev->recovery_cp = mddev->curr_resync;
4664 mddev->recovery_cp = MaxSector;
4668 mddev->curr_resync = 0;
4669 wake_up(&resync_wait);
4670 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4671 md_wakeup_thread(mddev->thread);
4676 * This routine is regularly called by all per-raid-array threads to
4677 * deal with generic issues like resync and super-block update.
4678 * Raid personalities that don't have a thread (linear/raid0) do not
4679 * need this as they never do any recovery or update the superblock.
4681 * It does not do any resync itself, but rather "forks" off other threads
4682 * to do that as needed.
4683 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4684 * "->recovery" and create a thread at ->sync_thread.
4685 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4686 * and wakeups up this thread which will reap the thread and finish up.
4687 * This thread also removes any faulty devices (with nr_pending == 0).
4689 * The overall approach is:
4690 * 1/ if the superblock needs updating, update it.
4691 * 2/ If a recovery thread is running, don't do anything else.
4692 * 3/ If recovery has finished, clean up, possibly marking spares active.
4693 * 4/ If there are any faulty devices, remove them.
4694 * 5/ If array is degraded, try to add spares devices
4695 * 6/ If array has spares or is not in-sync, start a resync thread.
4697 void md_check_recovery(mddev_t *mddev)
4700 struct list_head *rtmp;
4704 bitmap_daemon_work(mddev->bitmap);
4709 if (signal_pending(current)) {
4710 if (mddev->pers->sync_request) {
4711 printk(KERN_INFO "md: %s in immediate safe mode\n",
4713 mddev->safemode = 2;
4715 flush_signals(current);
4720 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4721 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4722 (mddev->safemode == 1) ||
4723 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4724 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4728 if (mddev_trylock(mddev)==0) {
4731 spin_lock_irq(&mddev->write_lock);
4732 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4733 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4735 mddev->sb_dirty = 1;
4737 if (mddev->safemode == 1)
4738 mddev->safemode = 0;
4739 spin_unlock_irq(&mddev->write_lock);
4741 if (mddev->sb_dirty)
4742 md_update_sb(mddev);
4745 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4746 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4747 /* resync/recovery still happening */
4748 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4751 if (mddev->sync_thread) {
4752 /* resync has finished, collect result */
4753 md_unregister_thread(mddev->sync_thread);
4754 mddev->sync_thread = NULL;
4755 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4756 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4758 /* activate any spares */
4759 mddev->pers->spare_active(mddev);
4761 md_update_sb(mddev);
4763 /* if array is no-longer degraded, then any saved_raid_disk
4764 * information must be scrapped
4766 if (!mddev->degraded)
4767 ITERATE_RDEV(mddev,rdev,rtmp)
4768 rdev->saved_raid_disk = -1;
4770 mddev->recovery = 0;
4771 /* flag recovery needed just to double check */
4772 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4773 md_new_event(mddev);
4776 /* Clear some bits that don't mean anything, but
4779 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4780 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4781 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4782 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4784 /* no recovery is running.
4785 * remove any failed drives, then
4786 * add spares if possible.
4787 * Spare are also removed and re-added, to allow
4788 * the personality to fail the re-add.
4790 ITERATE_RDEV(mddev,rdev,rtmp)
4791 if (rdev->raid_disk >= 0 &&
4792 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4793 atomic_read(&rdev->nr_pending)==0) {
4794 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4796 sprintf(nm,"rd%d", rdev->raid_disk);
4797 sysfs_remove_link(&mddev->kobj, nm);
4798 rdev->raid_disk = -1;
4802 if (mddev->degraded) {
4803 ITERATE_RDEV(mddev,rdev,rtmp)
4804 if (rdev->raid_disk < 0
4805 && !test_bit(Faulty, &rdev->flags)) {
4806 if (mddev->pers->hot_add_disk(mddev,rdev)) {
4808 sprintf(nm, "rd%d", rdev->raid_disk);
4809 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4811 md_new_event(mddev);
4818 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4819 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4820 } else if (mddev->recovery_cp < MaxSector) {
4821 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4822 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4823 /* nothing to be done ... */
4826 if (mddev->pers->sync_request) {
4827 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4828 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4829 /* We are adding a device or devices to an array
4830 * which has the bitmap stored on all devices.
4831 * So make sure all bitmap pages get written
4833 bitmap_write_all(mddev->bitmap);
4835 mddev->sync_thread = md_register_thread(md_do_sync,
4838 if (!mddev->sync_thread) {
4839 printk(KERN_ERR "%s: could not start resync"
4842 /* leave the spares where they are, it shouldn't hurt */
4843 mddev->recovery = 0;
4845 md_wakeup_thread(mddev->sync_thread);
4846 md_new_event(mddev);
4849 mddev_unlock(mddev);
4853 static int md_notify_reboot(struct notifier_block *this,
4854 unsigned long code, void *x)
4856 struct list_head *tmp;
4859 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4861 printk(KERN_INFO "md: stopping all md devices.\n");
4863 ITERATE_MDDEV(mddev,tmp)
4864 if (mddev_trylock(mddev)==0)
4865 do_md_stop (mddev, 1);
4867 * certain more exotic SCSI devices are known to be
4868 * volatile wrt too early system reboots. While the
4869 * right place to handle this issue is the given
4870 * driver, we do want to have a safe RAID driver ...
4877 static struct notifier_block md_notifier = {
4878 .notifier_call = md_notify_reboot,
4880 .priority = INT_MAX, /* before any real devices */
4883 static void md_geninit(void)
4885 struct proc_dir_entry *p;
4887 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4889 p = create_proc_entry("mdstat", S_IRUGO, NULL);
4891 p->proc_fops = &md_seq_fops;
4894 static int __init md_init(void)
4898 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4899 " MD_SB_DISKS=%d\n",
4900 MD_MAJOR_VERSION, MD_MINOR_VERSION,
4901 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4902 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4905 if (register_blkdev(MAJOR_NR, "md"))
4907 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4908 unregister_blkdev(MAJOR_NR, "md");
4912 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4913 md_probe, NULL, NULL);
4914 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4915 md_probe, NULL, NULL);
4917 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4918 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4919 S_IFBLK|S_IRUSR|S_IWUSR,
4922 for (minor=0; minor < MAX_MD_DEVS; ++minor)
4923 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4924 S_IFBLK|S_IRUSR|S_IWUSR,
4928 register_reboot_notifier(&md_notifier);
4929 raid_table_header = register_sysctl_table(raid_root_table, 1);
4939 * Searches all registered partitions for autorun RAID arrays
4942 static dev_t detected_devices[128];
4945 void md_autodetect_dev(dev_t dev)
4947 if (dev_cnt >= 0 && dev_cnt < 127)
4948 detected_devices[dev_cnt++] = dev;
4952 static void autostart_arrays(int part)
4957 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4959 for (i = 0; i < dev_cnt; i++) {
4960 dev_t dev = detected_devices[i];
4962 rdev = md_import_device(dev,0, 0);
4966 if (test_bit(Faulty, &rdev->flags)) {
4970 list_add(&rdev->same_set, &pending_raid_disks);
4974 autorun_devices(part);
4979 static __exit void md_exit(void)
4982 struct list_head *tmp;
4984 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4985 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4986 for (i=0; i < MAX_MD_DEVS; i++)
4987 devfs_remove("md/%d", i);
4988 for (i=0; i < MAX_MD_DEVS; i++)
4989 devfs_remove("md/d%d", i);
4993 unregister_blkdev(MAJOR_NR,"md");
4994 unregister_blkdev(mdp_major, "mdp");
4995 unregister_reboot_notifier(&md_notifier);
4996 unregister_sysctl_table(raid_table_header);
4997 remove_proc_entry("mdstat", NULL);
4998 ITERATE_MDDEV(mddev,tmp) {
4999 struct gendisk *disk = mddev->gendisk;
5002 export_array(mddev);
5005 mddev->gendisk = NULL;
5010 module_init(md_init)
5011 module_exit(md_exit)
5013 static int get_ro(char *buffer, struct kernel_param *kp)
5015 return sprintf(buffer, "%d", start_readonly);
5017 static int set_ro(const char *val, struct kernel_param *kp)
5020 int num = simple_strtoul(val, &e, 10);
5021 if (*val && (*e == '\0' || *e == '\n')) {
5022 start_readonly = num;
5028 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5029 module_param(start_dirty_degraded, int, 0644);
5032 EXPORT_SYMBOL(register_md_personality);
5033 EXPORT_SYMBOL(unregister_md_personality);
5034 EXPORT_SYMBOL(md_error);
5035 EXPORT_SYMBOL(md_done_sync);
5036 EXPORT_SYMBOL(md_write_start);
5037 EXPORT_SYMBOL(md_write_end);
5038 EXPORT_SYMBOL(md_register_thread);
5039 EXPORT_SYMBOL(md_unregister_thread);
5040 EXPORT_SYMBOL(md_wakeup_thread);
5041 EXPORT_SYMBOL(md_print_devices);
5042 EXPORT_SYMBOL(md_check_recovery);
5043 MODULE_LICENSE("GPL");
5045 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);