2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part);
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
75 static void md_print_devices(void);
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
100 static inline int speed_max(mddev_t *mddev)
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
106 static struct ctl_table_header *raid_table_header;
108 static ctl_table raid_table[] = {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = &proc_dointvec,
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = &proc_dointvec,
128 static ctl_table raid_dir_table[] = {
130 .ctl_name = DEV_RAID,
133 .mode = S_IRUGO|S_IXUGO,
139 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static struct block_device_operations md_fops;
152 static int start_readonly;
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
172 EXPORT_SYMBOL_GPL(md_new_event);
174 /* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
177 static void md_new_event_inintr(mddev_t *mddev)
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
198 #define ITERATE_MDDEV(mddev,tmp) \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
214 static int md_fail_request (request_queue_t *q, struct bio *bio)
216 bio_io_error(bio, bio->bi_size);
220 static inline mddev_t *mddev_get(mddev_t *mddev)
222 atomic_inc(&mddev->active);
226 static void mddev_put(mddev_t *mddev)
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
232 spin_unlock(&all_mddevs_lock);
233 blk_cleanup_queue(mddev->queue);
234 kobject_unregister(&mddev->kobj);
236 spin_unlock(&all_mddevs_lock);
239 static mddev_t * mddev_find(dev_t unit)
241 mddev_t *mddev, *new = NULL;
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
248 spin_unlock(&all_mddevs_lock);
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
258 spin_unlock(&all_mddevs_lock);
260 new = kzalloc(sizeof(*new), GFP_KERNEL);
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
270 mutex_init(&new->reconfig_mutex);
271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
275 spin_lock_init(&new->write_lock);
276 init_waitqueue_head(&new->sb_wait);
277 new->reshape_position = MaxSector;
279 new->queue = blk_alloc_queue(GFP_KERNEL);
284 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
286 blk_queue_make_request(new->queue, md_fail_request);
291 static inline int mddev_lock(mddev_t * mddev)
293 return mutex_lock_interruptible(&mddev->reconfig_mutex);
296 static inline int mddev_trylock(mddev_t * mddev)
298 return mutex_trylock(&mddev->reconfig_mutex);
301 static inline void mddev_unlock(mddev_t * mddev)
303 mutex_unlock(&mddev->reconfig_mutex);
305 md_wakeup_thread(mddev->thread);
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
311 struct list_head *tmp;
313 ITERATE_RDEV(mddev,rdev,tmp) {
314 if (rdev->desc_nr == nr)
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
322 struct list_head *tmp;
325 ITERATE_RDEV(mddev,rdev,tmp) {
326 if (rdev->bdev->bd_dev == dev)
332 static struct mdk_personality *find_pers(int level, char *clevel)
334 struct mdk_personality *pers;
335 list_for_each_entry(pers, &pers_list, list) {
336 if (level != LEVEL_NONE && pers->level == level)
338 if (strcmp(pers->name, clevel)==0)
344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
346 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
347 return MD_NEW_SIZE_BLOCKS(size);
350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
354 size = rdev->sb_offset;
357 size &= ~((sector_t)chunk_size/1024 - 1);
361 static int alloc_disk_sb(mdk_rdev_t * rdev)
366 rdev->sb_page = alloc_page(GFP_KERNEL);
367 if (!rdev->sb_page) {
368 printk(KERN_ALERT "md: out of memory.\n");
375 static void free_disk_sb(mdk_rdev_t * rdev)
378 put_page(rdev->sb_page);
380 rdev->sb_page = NULL;
387 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
389 mdk_rdev_t *rdev = bio->bi_private;
390 mddev_t *mddev = rdev->mddev;
394 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
395 printk("md: super_written gets error=%d, uptodate=%d\n",
396 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
397 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
398 md_error(mddev, rdev);
401 if (atomic_dec_and_test(&mddev->pending_writes))
402 wake_up(&mddev->sb_wait);
407 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
409 struct bio *bio2 = bio->bi_private;
410 mdk_rdev_t *rdev = bio2->bi_private;
411 mddev_t *mddev = rdev->mddev;
415 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
416 error == -EOPNOTSUPP) {
418 /* barriers don't appear to be supported :-( */
419 set_bit(BarriersNotsupp, &rdev->flags);
420 mddev->barriers_work = 0;
421 spin_lock_irqsave(&mddev->write_lock, flags);
422 bio2->bi_next = mddev->biolist;
423 mddev->biolist = bio2;
424 spin_unlock_irqrestore(&mddev->write_lock, flags);
425 wake_up(&mddev->sb_wait);
430 bio->bi_private = rdev;
431 return super_written(bio, bytes_done, error);
434 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
435 sector_t sector, int size, struct page *page)
437 /* write first size bytes of page to sector of rdev
438 * Increment mddev->pending_writes before returning
439 * and decrement it on completion, waking up sb_wait
440 * if zero is reached.
441 * If an error occurred, call md_error
443 * As we might need to resubmit the request if BIO_RW_BARRIER
444 * causes ENOTSUPP, we allocate a spare bio...
446 struct bio *bio = bio_alloc(GFP_NOIO, 1);
447 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
449 bio->bi_bdev = rdev->bdev;
450 bio->bi_sector = sector;
451 bio_add_page(bio, page, size, 0);
452 bio->bi_private = rdev;
453 bio->bi_end_io = super_written;
456 atomic_inc(&mddev->pending_writes);
457 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
459 rw |= (1<<BIO_RW_BARRIER);
460 rbio = bio_clone(bio, GFP_NOIO);
461 rbio->bi_private = bio;
462 rbio->bi_end_io = super_written_barrier;
463 submit_bio(rw, rbio);
468 void md_super_wait(mddev_t *mddev)
470 /* wait for all superblock writes that were scheduled to complete.
471 * if any had to be retried (due to BARRIER problems), retry them
475 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
476 if (atomic_read(&mddev->pending_writes)==0)
478 while (mddev->biolist) {
480 spin_lock_irq(&mddev->write_lock);
481 bio = mddev->biolist;
482 mddev->biolist = bio->bi_next ;
484 spin_unlock_irq(&mddev->write_lock);
485 submit_bio(bio->bi_rw, bio);
489 finish_wait(&mddev->sb_wait, &wq);
492 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
497 complete((struct completion*)bio->bi_private);
501 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
502 struct page *page, int rw)
504 struct bio *bio = bio_alloc(GFP_NOIO, 1);
505 struct completion event;
508 rw |= (1 << BIO_RW_SYNC);
511 bio->bi_sector = sector;
512 bio_add_page(bio, page, size, 0);
513 init_completion(&event);
514 bio->bi_private = &event;
515 bio->bi_end_io = bi_complete;
517 wait_for_completion(&event);
519 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
523 EXPORT_SYMBOL_GPL(sync_page_io);
525 static int read_disk_sb(mdk_rdev_t * rdev, int size)
527 char b[BDEVNAME_SIZE];
528 if (!rdev->sb_page) {
536 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
542 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
543 bdevname(rdev->bdev,b));
547 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
549 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
550 (sb1->set_uuid1 == sb2->set_uuid1) &&
551 (sb1->set_uuid2 == sb2->set_uuid2) &&
552 (sb1->set_uuid3 == sb2->set_uuid3))
560 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
563 mdp_super_t *tmp1, *tmp2;
565 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
566 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
568 if (!tmp1 || !tmp2) {
570 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
578 * nr_disks is not constant
583 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
595 static u32 md_csum_fold(u32 csum)
597 csum = (csum & 0xffff) + (csum >> 16);
598 return (csum & 0xffff) + (csum >> 16);
601 static unsigned int calc_sb_csum(mdp_super_t * sb)
604 u32 *sb32 = (u32*)sb;
606 unsigned int disk_csum, csum;
608 disk_csum = sb->sb_csum;
611 for (i = 0; i < MD_SB_BYTES/4 ; i++)
613 csum = (newcsum & 0xffffffff) + (newcsum>>32);
617 /* This used to use csum_partial, which was wrong for several
618 * reasons including that different results are returned on
619 * different architectures. It isn't critical that we get exactly
620 * the same return value as before (we always csum_fold before
621 * testing, and that removes any differences). However as we
622 * know that csum_partial always returned a 16bit value on
623 * alphas, do a fold to maximise conformity to previous behaviour.
625 sb->sb_csum = md_csum_fold(disk_csum);
627 sb->sb_csum = disk_csum;
634 * Handle superblock details.
635 * We want to be able to handle multiple superblock formats
636 * so we have a common interface to them all, and an array of
637 * different handlers.
638 * We rely on user-space to write the initial superblock, and support
639 * reading and updating of superblocks.
640 * Interface methods are:
641 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
642 * loads and validates a superblock on dev.
643 * if refdev != NULL, compare superblocks on both devices
645 * 0 - dev has a superblock that is compatible with refdev
646 * 1 - dev has a superblock that is compatible and newer than refdev
647 * so dev should be used as the refdev in future
648 * -EINVAL superblock incompatible or invalid
649 * -othererror e.g. -EIO
651 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
652 * Verify that dev is acceptable into mddev.
653 * The first time, mddev->raid_disks will be 0, and data from
654 * dev should be merged in. Subsequent calls check that dev
655 * is new enough. Return 0 or -EINVAL
657 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
658 * Update the superblock for rdev with data in mddev
659 * This does not write to disc.
665 struct module *owner;
666 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
667 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
668 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
672 * load_super for 0.90.0
674 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
676 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
682 * Calculate the position of the superblock,
683 * it's at the end of the disk.
685 * It also happens to be a multiple of 4Kb.
687 sb_offset = calc_dev_sboffset(rdev->bdev);
688 rdev->sb_offset = sb_offset;
690 ret = read_disk_sb(rdev, MD_SB_BYTES);
695 bdevname(rdev->bdev, b);
696 sb = (mdp_super_t*)page_address(rdev->sb_page);
698 if (sb->md_magic != MD_SB_MAGIC) {
699 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
704 if (sb->major_version != 0 ||
705 sb->minor_version < 90 ||
706 sb->minor_version > 91) {
707 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
708 sb->major_version, sb->minor_version,
713 if (sb->raid_disks <= 0)
716 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
717 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
722 rdev->preferred_minor = sb->md_minor;
723 rdev->data_offset = 0;
724 rdev->sb_size = MD_SB_BYTES;
726 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
727 if (sb->level != 1 && sb->level != 4
728 && sb->level != 5 && sb->level != 6
729 && sb->level != 10) {
730 /* FIXME use a better test */
732 "md: bitmaps not supported for this level.\n");
737 if (sb->level == LEVEL_MULTIPATH)
740 rdev->desc_nr = sb->this_disk.number;
746 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
747 if (!uuid_equal(refsb, sb)) {
748 printk(KERN_WARNING "md: %s has different UUID to %s\n",
749 b, bdevname(refdev->bdev,b2));
752 if (!sb_equal(refsb, sb)) {
753 printk(KERN_WARNING "md: %s has same UUID"
754 " but different superblock to %s\n",
755 b, bdevname(refdev->bdev, b2));
759 ev2 = md_event(refsb);
765 rdev->size = calc_dev_size(rdev, sb->chunk_size);
767 if (rdev->size < sb->size && sb->level > 1)
768 /* "this cannot possibly happen" ... */
776 * validate_super for 0.90.0
778 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
781 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
782 __u64 ev1 = md_event(sb);
784 rdev->raid_disk = -1;
786 if (mddev->raid_disks == 0) {
787 mddev->major_version = 0;
788 mddev->minor_version = sb->minor_version;
789 mddev->patch_version = sb->patch_version;
790 mddev->persistent = ! sb->not_persistent;
791 mddev->chunk_size = sb->chunk_size;
792 mddev->ctime = sb->ctime;
793 mddev->utime = sb->utime;
794 mddev->level = sb->level;
795 mddev->clevel[0] = 0;
796 mddev->layout = sb->layout;
797 mddev->raid_disks = sb->raid_disks;
798 mddev->size = sb->size;
800 mddev->bitmap_offset = 0;
801 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
803 if (mddev->minor_version >= 91) {
804 mddev->reshape_position = sb->reshape_position;
805 mddev->delta_disks = sb->delta_disks;
806 mddev->new_level = sb->new_level;
807 mddev->new_layout = sb->new_layout;
808 mddev->new_chunk = sb->new_chunk;
810 mddev->reshape_position = MaxSector;
811 mddev->delta_disks = 0;
812 mddev->new_level = mddev->level;
813 mddev->new_layout = mddev->layout;
814 mddev->new_chunk = mddev->chunk_size;
817 if (sb->state & (1<<MD_SB_CLEAN))
818 mddev->recovery_cp = MaxSector;
820 if (sb->events_hi == sb->cp_events_hi &&
821 sb->events_lo == sb->cp_events_lo) {
822 mddev->recovery_cp = sb->recovery_cp;
824 mddev->recovery_cp = 0;
827 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
828 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
829 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
830 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
832 mddev->max_disks = MD_SB_DISKS;
834 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
835 mddev->bitmap_file == NULL)
836 mddev->bitmap_offset = mddev->default_bitmap_offset;
838 } else if (mddev->pers == NULL) {
839 /* Insist on good event counter while assembling */
841 if (ev1 < mddev->events)
843 } else if (mddev->bitmap) {
844 /* if adding to array with a bitmap, then we can accept an
845 * older device ... but not too old.
847 if (ev1 < mddev->bitmap->events_cleared)
850 if (ev1 < mddev->events)
851 /* just a hot-add of a new device, leave raid_disk at -1 */
855 if (mddev->level != LEVEL_MULTIPATH) {
856 desc = sb->disks + rdev->desc_nr;
858 if (desc->state & (1<<MD_DISK_FAULTY))
859 set_bit(Faulty, &rdev->flags);
860 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
861 desc->raid_disk < mddev->raid_disks */) {
862 set_bit(In_sync, &rdev->flags);
863 rdev->raid_disk = desc->raid_disk;
865 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
866 set_bit(WriteMostly, &rdev->flags);
867 } else /* MULTIPATH are always insync */
868 set_bit(In_sync, &rdev->flags);
873 * sync_super for 0.90.0
875 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
878 struct list_head *tmp;
880 int next_spare = mddev->raid_disks;
883 /* make rdev->sb match mddev data..
886 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
887 * 3/ any empty disks < next_spare become removed
889 * disks[0] gets initialised to REMOVED because
890 * we cannot be sure from other fields if it has
891 * been initialised or not.
894 int active=0, working=0,failed=0,spare=0,nr_disks=0;
896 rdev->sb_size = MD_SB_BYTES;
898 sb = (mdp_super_t*)page_address(rdev->sb_page);
900 memset(sb, 0, sizeof(*sb));
902 sb->md_magic = MD_SB_MAGIC;
903 sb->major_version = mddev->major_version;
904 sb->patch_version = mddev->patch_version;
905 sb->gvalid_words = 0; /* ignored */
906 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
907 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
908 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
909 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
911 sb->ctime = mddev->ctime;
912 sb->level = mddev->level;
913 sb->size = mddev->size;
914 sb->raid_disks = mddev->raid_disks;
915 sb->md_minor = mddev->md_minor;
916 sb->not_persistent = !mddev->persistent;
917 sb->utime = mddev->utime;
919 sb->events_hi = (mddev->events>>32);
920 sb->events_lo = (u32)mddev->events;
922 if (mddev->reshape_position == MaxSector)
923 sb->minor_version = 90;
925 sb->minor_version = 91;
926 sb->reshape_position = mddev->reshape_position;
927 sb->new_level = mddev->new_level;
928 sb->delta_disks = mddev->delta_disks;
929 sb->new_layout = mddev->new_layout;
930 sb->new_chunk = mddev->new_chunk;
932 mddev->minor_version = sb->minor_version;
935 sb->recovery_cp = mddev->recovery_cp;
936 sb->cp_events_hi = (mddev->events>>32);
937 sb->cp_events_lo = (u32)mddev->events;
938 if (mddev->recovery_cp == MaxSector)
939 sb->state = (1<< MD_SB_CLEAN);
943 sb->layout = mddev->layout;
944 sb->chunk_size = mddev->chunk_size;
946 if (mddev->bitmap && mddev->bitmap_file == NULL)
947 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
949 sb->disks[0].state = (1<<MD_DISK_REMOVED);
950 ITERATE_RDEV(mddev,rdev2,tmp) {
953 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
954 && !test_bit(Faulty, &rdev2->flags))
955 desc_nr = rdev2->raid_disk;
957 desc_nr = next_spare++;
958 rdev2->desc_nr = desc_nr;
959 d = &sb->disks[rdev2->desc_nr];
961 d->number = rdev2->desc_nr;
962 d->major = MAJOR(rdev2->bdev->bd_dev);
963 d->minor = MINOR(rdev2->bdev->bd_dev);
964 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
965 && !test_bit(Faulty, &rdev2->flags))
966 d->raid_disk = rdev2->raid_disk;
968 d->raid_disk = rdev2->desc_nr; /* compatibility */
969 if (test_bit(Faulty, &rdev2->flags))
970 d->state = (1<<MD_DISK_FAULTY);
971 else if (test_bit(In_sync, &rdev2->flags)) {
972 d->state = (1<<MD_DISK_ACTIVE);
973 d->state |= (1<<MD_DISK_SYNC);
981 if (test_bit(WriteMostly, &rdev2->flags))
982 d->state |= (1<<MD_DISK_WRITEMOSTLY);
984 /* now set the "removed" and "faulty" bits on any missing devices */
985 for (i=0 ; i < mddev->raid_disks ; i++) {
986 mdp_disk_t *d = &sb->disks[i];
987 if (d->state == 0 && d->number == 0) {
990 d->state = (1<<MD_DISK_REMOVED);
991 d->state |= (1<<MD_DISK_FAULTY);
995 sb->nr_disks = nr_disks;
996 sb->active_disks = active;
997 sb->working_disks = working;
998 sb->failed_disks = failed;
999 sb->spare_disks = spare;
1001 sb->this_disk = sb->disks[rdev->desc_nr];
1002 sb->sb_csum = calc_sb_csum(sb);
1006 * version 1 superblock
1009 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1013 unsigned long long newcsum;
1014 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1015 __le32 *isuper = (__le32*)sb;
1018 disk_csum = sb->sb_csum;
1021 for (i=0; size>=4; size -= 4 )
1022 newcsum += le32_to_cpu(*isuper++);
1025 newcsum += le16_to_cpu(*(__le16*) isuper);
1027 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1028 sb->sb_csum = disk_csum;
1029 return cpu_to_le32(csum);
1032 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1034 struct mdp_superblock_1 *sb;
1037 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1041 * Calculate the position of the superblock.
1042 * It is always aligned to a 4K boundary and
1043 * depeding on minor_version, it can be:
1044 * 0: At least 8K, but less than 12K, from end of device
1045 * 1: At start of device
1046 * 2: 4K from start of device.
1048 switch(minor_version) {
1050 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1052 sb_offset &= ~(sector_t)(4*2-1);
1053 /* convert from sectors to K */
1065 rdev->sb_offset = sb_offset;
1067 /* superblock is rarely larger than 1K, but it can be larger,
1068 * and it is safe to read 4k, so we do that
1070 ret = read_disk_sb(rdev, 4096);
1071 if (ret) return ret;
1074 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1076 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1077 sb->major_version != cpu_to_le32(1) ||
1078 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1079 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1080 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1083 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1084 printk("md: invalid superblock checksum on %s\n",
1085 bdevname(rdev->bdev,b));
1088 if (le64_to_cpu(sb->data_size) < 10) {
1089 printk("md: data_size too small on %s\n",
1090 bdevname(rdev->bdev,b));
1093 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1094 if (sb->level != cpu_to_le32(1) &&
1095 sb->level != cpu_to_le32(4) &&
1096 sb->level != cpu_to_le32(5) &&
1097 sb->level != cpu_to_le32(6) &&
1098 sb->level != cpu_to_le32(10)) {
1100 "md: bitmaps not supported for this level.\n");
1105 rdev->preferred_minor = 0xffff;
1106 rdev->data_offset = le64_to_cpu(sb->data_offset);
1107 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1109 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1110 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1111 if (rdev->sb_size & bmask)
1112 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1114 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1117 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1123 struct mdp_superblock_1 *refsb =
1124 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1126 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1127 sb->level != refsb->level ||
1128 sb->layout != refsb->layout ||
1129 sb->chunksize != refsb->chunksize) {
1130 printk(KERN_WARNING "md: %s has strangely different"
1131 " superblock to %s\n",
1132 bdevname(rdev->bdev,b),
1133 bdevname(refdev->bdev,b2));
1136 ev1 = le64_to_cpu(sb->events);
1137 ev2 = le64_to_cpu(refsb->events);
1145 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1147 rdev->size = rdev->sb_offset;
1148 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1150 rdev->size = le64_to_cpu(sb->data_size)/2;
1151 if (le32_to_cpu(sb->chunksize))
1152 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1154 if (le64_to_cpu(sb->size) > rdev->size*2)
1159 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1161 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1162 __u64 ev1 = le64_to_cpu(sb->events);
1164 rdev->raid_disk = -1;
1166 if (mddev->raid_disks == 0) {
1167 mddev->major_version = 1;
1168 mddev->patch_version = 0;
1169 mddev->persistent = 1;
1170 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1171 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1172 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1173 mddev->level = le32_to_cpu(sb->level);
1174 mddev->clevel[0] = 0;
1175 mddev->layout = le32_to_cpu(sb->layout);
1176 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1177 mddev->size = le64_to_cpu(sb->size)/2;
1178 mddev->events = ev1;
1179 mddev->bitmap_offset = 0;
1180 mddev->default_bitmap_offset = 1024 >> 9;
1182 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1183 memcpy(mddev->uuid, sb->set_uuid, 16);
1185 mddev->max_disks = (4096-256)/2;
1187 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1188 mddev->bitmap_file == NULL )
1189 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1191 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1192 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1193 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1194 mddev->new_level = le32_to_cpu(sb->new_level);
1195 mddev->new_layout = le32_to_cpu(sb->new_layout);
1196 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1198 mddev->reshape_position = MaxSector;
1199 mddev->delta_disks = 0;
1200 mddev->new_level = mddev->level;
1201 mddev->new_layout = mddev->layout;
1202 mddev->new_chunk = mddev->chunk_size;
1205 } else if (mddev->pers == NULL) {
1206 /* Insist of good event counter while assembling */
1208 if (ev1 < mddev->events)
1210 } else if (mddev->bitmap) {
1211 /* If adding to array with a bitmap, then we can accept an
1212 * older device, but not too old.
1214 if (ev1 < mddev->bitmap->events_cleared)
1217 if (ev1 < mddev->events)
1218 /* just a hot-add of a new device, leave raid_disk at -1 */
1221 if (mddev->level != LEVEL_MULTIPATH) {
1223 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1225 case 0xffff: /* spare */
1227 case 0xfffe: /* faulty */
1228 set_bit(Faulty, &rdev->flags);
1231 if ((le32_to_cpu(sb->feature_map) &
1232 MD_FEATURE_RECOVERY_OFFSET))
1233 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1235 set_bit(In_sync, &rdev->flags);
1236 rdev->raid_disk = role;
1239 if (sb->devflags & WriteMostly1)
1240 set_bit(WriteMostly, &rdev->flags);
1241 } else /* MULTIPATH are always insync */
1242 set_bit(In_sync, &rdev->flags);
1247 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1249 struct mdp_superblock_1 *sb;
1250 struct list_head *tmp;
1253 /* make rdev->sb match mddev and rdev data. */
1255 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1257 sb->feature_map = 0;
1259 sb->recovery_offset = cpu_to_le64(0);
1260 memset(sb->pad1, 0, sizeof(sb->pad1));
1261 memset(sb->pad2, 0, sizeof(sb->pad2));
1262 memset(sb->pad3, 0, sizeof(sb->pad3));
1264 sb->utime = cpu_to_le64((__u64)mddev->utime);
1265 sb->events = cpu_to_le64(mddev->events);
1267 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1269 sb->resync_offset = cpu_to_le64(0);
1271 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1273 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1274 sb->size = cpu_to_le64(mddev->size<<1);
1276 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1277 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1278 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1281 if (rdev->raid_disk >= 0 &&
1282 !test_bit(In_sync, &rdev->flags) &&
1283 rdev->recovery_offset > 0) {
1284 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1285 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1288 if (mddev->reshape_position != MaxSector) {
1289 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1290 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1291 sb->new_layout = cpu_to_le32(mddev->new_layout);
1292 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1293 sb->new_level = cpu_to_le32(mddev->new_level);
1294 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1298 ITERATE_RDEV(mddev,rdev2,tmp)
1299 if (rdev2->desc_nr+1 > max_dev)
1300 max_dev = rdev2->desc_nr+1;
1302 if (max_dev > le32_to_cpu(sb->max_dev))
1303 sb->max_dev = cpu_to_le32(max_dev);
1304 for (i=0; i<max_dev;i++)
1305 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1307 ITERATE_RDEV(mddev,rdev2,tmp) {
1309 if (test_bit(Faulty, &rdev2->flags))
1310 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1311 else if (test_bit(In_sync, &rdev2->flags))
1312 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1313 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1314 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1316 sb->dev_roles[i] = cpu_to_le16(0xffff);
1319 sb->sb_csum = calc_sb_1_csum(sb);
1323 static struct super_type super_types[] = {
1326 .owner = THIS_MODULE,
1327 .load_super = super_90_load,
1328 .validate_super = super_90_validate,
1329 .sync_super = super_90_sync,
1333 .owner = THIS_MODULE,
1334 .load_super = super_1_load,
1335 .validate_super = super_1_validate,
1336 .sync_super = super_1_sync,
1340 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1342 struct list_head *tmp, *tmp2;
1343 mdk_rdev_t *rdev, *rdev2;
1345 ITERATE_RDEV(mddev1,rdev,tmp)
1346 ITERATE_RDEV(mddev2, rdev2, tmp2)
1347 if (rdev->bdev->bd_contains ==
1348 rdev2->bdev->bd_contains)
1354 static LIST_HEAD(pending_raid_disks);
1356 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1358 char b[BDEVNAME_SIZE];
1367 /* make sure rdev->size exceeds mddev->size */
1368 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1370 /* Cannot change size, so fail
1371 * If mddev->level <= 0, then we don't care
1372 * about aligning sizes (e.g. linear)
1374 if (mddev->level > 0)
1377 mddev->size = rdev->size;
1380 /* Verify rdev->desc_nr is unique.
1381 * If it is -1, assign a free number, else
1382 * check number is not in use
1384 if (rdev->desc_nr < 0) {
1386 if (mddev->pers) choice = mddev->raid_disks;
1387 while (find_rdev_nr(mddev, choice))
1389 rdev->desc_nr = choice;
1391 if (find_rdev_nr(mddev, rdev->desc_nr))
1394 bdevname(rdev->bdev,b);
1395 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1397 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1400 rdev->mddev = mddev;
1401 printk(KERN_INFO "md: bind<%s>\n", b);
1403 rdev->kobj.parent = &mddev->kobj;
1404 if ((err = kobject_add(&rdev->kobj)))
1407 if (rdev->bdev->bd_part)
1408 ko = &rdev->bdev->bd_part->kobj;
1410 ko = &rdev->bdev->bd_disk->kobj;
1411 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1412 kobject_del(&rdev->kobj);
1415 list_add(&rdev->same_set, &mddev->disks);
1416 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1420 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1425 static void delayed_delete(struct work_struct *ws)
1427 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1428 kobject_del(&rdev->kobj);
1431 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1433 char b[BDEVNAME_SIZE];
1438 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1439 list_del_init(&rdev->same_set);
1440 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1442 sysfs_remove_link(&rdev->kobj, "block");
1444 /* We need to delay this, otherwise we can deadlock when
1445 * writing to 'remove' to "dev/state"
1447 INIT_WORK(&rdev->del_work, delayed_delete);
1448 schedule_work(&rdev->del_work);
1452 * prevent the device from being mounted, repartitioned or
1453 * otherwise reused by a RAID array (or any other kernel
1454 * subsystem), by bd_claiming the device.
1456 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1459 struct block_device *bdev;
1460 char b[BDEVNAME_SIZE];
1462 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1464 printk(KERN_ERR "md: could not open %s.\n",
1465 __bdevname(dev, b));
1466 return PTR_ERR(bdev);
1468 err = bd_claim(bdev, rdev);
1470 printk(KERN_ERR "md: could not bd_claim %s.\n",
1479 static void unlock_rdev(mdk_rdev_t *rdev)
1481 struct block_device *bdev = rdev->bdev;
1489 void md_autodetect_dev(dev_t dev);
1491 static void export_rdev(mdk_rdev_t * rdev)
1493 char b[BDEVNAME_SIZE];
1494 printk(KERN_INFO "md: export_rdev(%s)\n",
1495 bdevname(rdev->bdev,b));
1499 list_del_init(&rdev->same_set);
1501 md_autodetect_dev(rdev->bdev->bd_dev);
1504 kobject_put(&rdev->kobj);
1507 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1509 unbind_rdev_from_array(rdev);
1513 static void export_array(mddev_t *mddev)
1515 struct list_head *tmp;
1518 ITERATE_RDEV(mddev,rdev,tmp) {
1523 kick_rdev_from_array(rdev);
1525 if (!list_empty(&mddev->disks))
1527 mddev->raid_disks = 0;
1528 mddev->major_version = 0;
1531 static void print_desc(mdp_disk_t *desc)
1533 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1534 desc->major,desc->minor,desc->raid_disk,desc->state);
1537 static void print_sb(mdp_super_t *sb)
1542 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1543 sb->major_version, sb->minor_version, sb->patch_version,
1544 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1546 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1547 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1548 sb->md_minor, sb->layout, sb->chunk_size);
1549 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1550 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1551 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1552 sb->failed_disks, sb->spare_disks,
1553 sb->sb_csum, (unsigned long)sb->events_lo);
1556 for (i = 0; i < MD_SB_DISKS; i++) {
1559 desc = sb->disks + i;
1560 if (desc->number || desc->major || desc->minor ||
1561 desc->raid_disk || (desc->state && (desc->state != 4))) {
1562 printk(" D %2d: ", i);
1566 printk(KERN_INFO "md: THIS: ");
1567 print_desc(&sb->this_disk);
1571 static void print_rdev(mdk_rdev_t *rdev)
1573 char b[BDEVNAME_SIZE];
1574 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1575 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1576 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1578 if (rdev->sb_loaded) {
1579 printk(KERN_INFO "md: rdev superblock:\n");
1580 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1582 printk(KERN_INFO "md: no rdev superblock!\n");
1585 static void md_print_devices(void)
1587 struct list_head *tmp, *tmp2;
1590 char b[BDEVNAME_SIZE];
1593 printk("md: **********************************\n");
1594 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1595 printk("md: **********************************\n");
1596 ITERATE_MDDEV(mddev,tmp) {
1599 bitmap_print_sb(mddev->bitmap);
1601 printk("%s: ", mdname(mddev));
1602 ITERATE_RDEV(mddev,rdev,tmp2)
1603 printk("<%s>", bdevname(rdev->bdev,b));
1606 ITERATE_RDEV(mddev,rdev,tmp2)
1609 printk("md: **********************************\n");
1614 static void sync_sbs(mddev_t * mddev, int nospares)
1616 /* Update each superblock (in-memory image), but
1617 * if we are allowed to, skip spares which already
1618 * have the right event counter, or have one earlier
1619 * (which would mean they aren't being marked as dirty
1620 * with the rest of the array)
1623 struct list_head *tmp;
1625 ITERATE_RDEV(mddev,rdev,tmp) {
1626 if (rdev->sb_events == mddev->events ||
1628 rdev->raid_disk < 0 &&
1629 (rdev->sb_events&1)==0 &&
1630 rdev->sb_events+1 == mddev->events)) {
1631 /* Don't update this superblock */
1632 rdev->sb_loaded = 2;
1634 super_types[mddev->major_version].
1635 sync_super(mddev, rdev);
1636 rdev->sb_loaded = 1;
1641 static void md_update_sb(mddev_t * mddev, int force_change)
1644 struct list_head *tmp;
1650 spin_lock_irq(&mddev->write_lock);
1652 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1653 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1655 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1656 /* just a clean<-> dirty transition, possibly leave spares alone,
1657 * though if events isn't the right even/odd, we will have to do
1663 if (mddev->degraded)
1664 /* If the array is degraded, then skipping spares is both
1665 * dangerous and fairly pointless.
1666 * Dangerous because a device that was removed from the array
1667 * might have a event_count that still looks up-to-date,
1668 * so it can be re-added without a resync.
1669 * Pointless because if there are any spares to skip,
1670 * then a recovery will happen and soon that array won't
1671 * be degraded any more and the spare can go back to sleep then.
1675 sync_req = mddev->in_sync;
1676 mddev->utime = get_seconds();
1678 /* If this is just a dirty<->clean transition, and the array is clean
1679 * and 'events' is odd, we can roll back to the previous clean state */
1681 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1682 && (mddev->events & 1)
1683 && mddev->events != 1)
1686 /* otherwise we have to go forward and ... */
1688 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1689 /* .. if the array isn't clean, insist on an odd 'events' */
1690 if ((mddev->events&1)==0) {
1695 /* otherwise insist on an even 'events' (for clean states) */
1696 if ((mddev->events&1)) {
1703 if (!mddev->events) {
1705 * oops, this 64-bit counter should never wrap.
1706 * Either we are in around ~1 trillion A.C., assuming
1707 * 1 reboot per second, or we have a bug:
1712 sync_sbs(mddev, nospares);
1715 * do not write anything to disk if using
1716 * nonpersistent superblocks
1718 if (!mddev->persistent) {
1719 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1720 spin_unlock_irq(&mddev->write_lock);
1721 wake_up(&mddev->sb_wait);
1724 spin_unlock_irq(&mddev->write_lock);
1727 "md: updating %s RAID superblock on device (in sync %d)\n",
1728 mdname(mddev),mddev->in_sync);
1730 err = bitmap_update_sb(mddev->bitmap);
1731 ITERATE_RDEV(mddev,rdev,tmp) {
1732 char b[BDEVNAME_SIZE];
1733 dprintk(KERN_INFO "md: ");
1734 if (rdev->sb_loaded != 1)
1735 continue; /* no noise on spare devices */
1736 if (test_bit(Faulty, &rdev->flags))
1737 dprintk("(skipping faulty ");
1739 dprintk("%s ", bdevname(rdev->bdev,b));
1740 if (!test_bit(Faulty, &rdev->flags)) {
1741 md_super_write(mddev,rdev,
1742 rdev->sb_offset<<1, rdev->sb_size,
1744 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1745 bdevname(rdev->bdev,b),
1746 (unsigned long long)rdev->sb_offset);
1747 rdev->sb_events = mddev->events;
1751 if (mddev->level == LEVEL_MULTIPATH)
1752 /* only need to write one superblock... */
1755 md_super_wait(mddev);
1756 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1758 spin_lock_irq(&mddev->write_lock);
1759 if (mddev->in_sync != sync_req ||
1760 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1761 /* have to write it out again */
1762 spin_unlock_irq(&mddev->write_lock);
1765 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1766 spin_unlock_irq(&mddev->write_lock);
1767 wake_up(&mddev->sb_wait);
1771 /* words written to sysfs files may, or my not, be \n terminated.
1772 * We want to accept with case. For this we use cmd_match.
1774 static int cmd_match(const char *cmd, const char *str)
1776 /* See if cmd, written into a sysfs file, matches
1777 * str. They must either be the same, or cmd can
1778 * have a trailing newline
1780 while (*cmd && *str && *cmd == *str) {
1791 struct rdev_sysfs_entry {
1792 struct attribute attr;
1793 ssize_t (*show)(mdk_rdev_t *, char *);
1794 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1798 state_show(mdk_rdev_t *rdev, char *page)
1803 if (test_bit(Faulty, &rdev->flags)) {
1804 len+= sprintf(page+len, "%sfaulty",sep);
1807 if (test_bit(In_sync, &rdev->flags)) {
1808 len += sprintf(page+len, "%sin_sync",sep);
1811 if (test_bit(WriteMostly, &rdev->flags)) {
1812 len += sprintf(page+len, "%swrite_mostly",sep);
1815 if (!test_bit(Faulty, &rdev->flags) &&
1816 !test_bit(In_sync, &rdev->flags)) {
1817 len += sprintf(page+len, "%sspare", sep);
1820 return len+sprintf(page+len, "\n");
1824 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1827 * faulty - simulates and error
1828 * remove - disconnects the device
1829 * writemostly - sets write_mostly
1830 * -writemostly - clears write_mostly
1833 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1834 md_error(rdev->mddev, rdev);
1836 } else if (cmd_match(buf, "remove")) {
1837 if (rdev->raid_disk >= 0)
1840 mddev_t *mddev = rdev->mddev;
1841 kick_rdev_from_array(rdev);
1843 md_update_sb(mddev, 1);
1844 md_new_event(mddev);
1847 } else if (cmd_match(buf, "writemostly")) {
1848 set_bit(WriteMostly, &rdev->flags);
1850 } else if (cmd_match(buf, "-writemostly")) {
1851 clear_bit(WriteMostly, &rdev->flags);
1854 return err ? err : len;
1856 static struct rdev_sysfs_entry rdev_state =
1857 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1860 super_show(mdk_rdev_t *rdev, char *page)
1862 if (rdev->sb_loaded && rdev->sb_size) {
1863 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1864 return rdev->sb_size;
1868 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1871 errors_show(mdk_rdev_t *rdev, char *page)
1873 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1877 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1880 unsigned long n = simple_strtoul(buf, &e, 10);
1881 if (*buf && (*e == 0 || *e == '\n')) {
1882 atomic_set(&rdev->corrected_errors, n);
1887 static struct rdev_sysfs_entry rdev_errors =
1888 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1891 slot_show(mdk_rdev_t *rdev, char *page)
1893 if (rdev->raid_disk < 0)
1894 return sprintf(page, "none\n");
1896 return sprintf(page, "%d\n", rdev->raid_disk);
1900 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1903 int slot = simple_strtoul(buf, &e, 10);
1904 if (strncmp(buf, "none", 4)==0)
1906 else if (e==buf || (*e && *e!= '\n'))
1908 if (rdev->mddev->pers)
1909 /* Cannot set slot in active array (yet) */
1911 if (slot >= rdev->mddev->raid_disks)
1913 rdev->raid_disk = slot;
1914 /* assume it is working */
1916 set_bit(In_sync, &rdev->flags);
1921 static struct rdev_sysfs_entry rdev_slot =
1922 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1925 offset_show(mdk_rdev_t *rdev, char *page)
1927 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1931 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1934 unsigned long long offset = simple_strtoull(buf, &e, 10);
1935 if (e==buf || (*e && *e != '\n'))
1937 if (rdev->mddev->pers)
1939 rdev->data_offset = offset;
1943 static struct rdev_sysfs_entry rdev_offset =
1944 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1947 rdev_size_show(mdk_rdev_t *rdev, char *page)
1949 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1953 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1956 unsigned long long size = simple_strtoull(buf, &e, 10);
1957 if (e==buf || (*e && *e != '\n'))
1959 if (rdev->mddev->pers)
1962 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1963 rdev->mddev->size = size;
1967 static struct rdev_sysfs_entry rdev_size =
1968 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1970 static struct attribute *rdev_default_attrs[] = {
1980 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1982 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1983 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1987 return entry->show(rdev, page);
1991 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1992 const char *page, size_t length)
1994 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1995 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1999 if (!capable(CAP_SYS_ADMIN))
2001 return entry->store(rdev, page, length);
2004 static void rdev_free(struct kobject *ko)
2006 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2009 static struct sysfs_ops rdev_sysfs_ops = {
2010 .show = rdev_attr_show,
2011 .store = rdev_attr_store,
2013 static struct kobj_type rdev_ktype = {
2014 .release = rdev_free,
2015 .sysfs_ops = &rdev_sysfs_ops,
2016 .default_attrs = rdev_default_attrs,
2020 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2022 * mark the device faulty if:
2024 * - the device is nonexistent (zero size)
2025 * - the device has no valid superblock
2027 * a faulty rdev _never_ has rdev->sb set.
2029 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2031 char b[BDEVNAME_SIZE];
2036 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2038 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2039 return ERR_PTR(-ENOMEM);
2042 if ((err = alloc_disk_sb(rdev)))
2045 err = lock_rdev(rdev, newdev);
2049 rdev->kobj.parent = NULL;
2050 rdev->kobj.ktype = &rdev_ktype;
2051 kobject_init(&rdev->kobj);
2054 rdev->saved_raid_disk = -1;
2055 rdev->raid_disk = -1;
2057 rdev->data_offset = 0;
2058 rdev->sb_events = 0;
2059 atomic_set(&rdev->nr_pending, 0);
2060 atomic_set(&rdev->read_errors, 0);
2061 atomic_set(&rdev->corrected_errors, 0);
2063 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2066 "md: %s has zero or unknown size, marking faulty!\n",
2067 bdevname(rdev->bdev,b));
2072 if (super_format >= 0) {
2073 err = super_types[super_format].
2074 load_super(rdev, NULL, super_minor);
2075 if (err == -EINVAL) {
2077 "md: %s does not have a valid v%d.%d "
2078 "superblock, not importing!\n",
2079 bdevname(rdev->bdev,b),
2080 super_format, super_minor);
2085 "md: could not read %s's sb, not importing!\n",
2086 bdevname(rdev->bdev,b));
2090 INIT_LIST_HEAD(&rdev->same_set);
2095 if (rdev->sb_page) {
2101 return ERR_PTR(err);
2105 * Check a full RAID array for plausibility
2109 static void analyze_sbs(mddev_t * mddev)
2112 struct list_head *tmp;
2113 mdk_rdev_t *rdev, *freshest;
2114 char b[BDEVNAME_SIZE];
2117 ITERATE_RDEV(mddev,rdev,tmp)
2118 switch (super_types[mddev->major_version].
2119 load_super(rdev, freshest, mddev->minor_version)) {
2127 "md: fatal superblock inconsistency in %s"
2128 " -- removing from array\n",
2129 bdevname(rdev->bdev,b));
2130 kick_rdev_from_array(rdev);
2134 super_types[mddev->major_version].
2135 validate_super(mddev, freshest);
2138 ITERATE_RDEV(mddev,rdev,tmp) {
2139 if (rdev != freshest)
2140 if (super_types[mddev->major_version].
2141 validate_super(mddev, rdev)) {
2142 printk(KERN_WARNING "md: kicking non-fresh %s"
2144 bdevname(rdev->bdev,b));
2145 kick_rdev_from_array(rdev);
2148 if (mddev->level == LEVEL_MULTIPATH) {
2149 rdev->desc_nr = i++;
2150 rdev->raid_disk = rdev->desc_nr;
2151 set_bit(In_sync, &rdev->flags);
2152 } else if (rdev->raid_disk >= mddev->raid_disks) {
2153 rdev->raid_disk = -1;
2154 clear_bit(In_sync, &rdev->flags);
2160 if (mddev->recovery_cp != MaxSector &&
2162 printk(KERN_ERR "md: %s: raid array is not clean"
2163 " -- starting background reconstruction\n",
2169 safe_delay_show(mddev_t *mddev, char *page)
2171 int msec = (mddev->safemode_delay*1000)/HZ;
2172 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2175 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2183 /* remove a period, and count digits after it */
2184 if (len >= sizeof(buf))
2186 strlcpy(buf, cbuf, len);
2188 for (i=0; i<len; i++) {
2190 if (isdigit(buf[i])) {
2195 } else if (buf[i] == '.') {
2200 msec = simple_strtoul(buf, &e, 10);
2201 if (e == buf || (*e && *e != '\n'))
2203 msec = (msec * 1000) / scale;
2205 mddev->safemode_delay = 0;
2207 mddev->safemode_delay = (msec*HZ)/1000;
2208 if (mddev->safemode_delay == 0)
2209 mddev->safemode_delay = 1;
2213 static struct md_sysfs_entry md_safe_delay =
2214 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2217 level_show(mddev_t *mddev, char *page)
2219 struct mdk_personality *p = mddev->pers;
2221 return sprintf(page, "%s\n", p->name);
2222 else if (mddev->clevel[0])
2223 return sprintf(page, "%s\n", mddev->clevel);
2224 else if (mddev->level != LEVEL_NONE)
2225 return sprintf(page, "%d\n", mddev->level);
2231 level_store(mddev_t *mddev, const char *buf, size_t len)
2238 if (len >= sizeof(mddev->clevel))
2240 strncpy(mddev->clevel, buf, len);
2241 if (mddev->clevel[len-1] == '\n')
2243 mddev->clevel[len] = 0;
2244 mddev->level = LEVEL_NONE;
2248 static struct md_sysfs_entry md_level =
2249 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2253 layout_show(mddev_t *mddev, char *page)
2255 /* just a number, not meaningful for all levels */
2256 if (mddev->reshape_position != MaxSector &&
2257 mddev->layout != mddev->new_layout)
2258 return sprintf(page, "%d (%d)\n",
2259 mddev->new_layout, mddev->layout);
2260 return sprintf(page, "%d\n", mddev->layout);
2264 layout_store(mddev_t *mddev, const char *buf, size_t len)
2267 unsigned long n = simple_strtoul(buf, &e, 10);
2269 if (!*buf || (*e && *e != '\n'))
2274 if (mddev->reshape_position != MaxSector)
2275 mddev->new_layout = n;
2280 static struct md_sysfs_entry md_layout =
2281 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2285 raid_disks_show(mddev_t *mddev, char *page)
2287 if (mddev->raid_disks == 0)
2289 if (mddev->reshape_position != MaxSector &&
2290 mddev->delta_disks != 0)
2291 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2292 mddev->raid_disks - mddev->delta_disks);
2293 return sprintf(page, "%d\n", mddev->raid_disks);
2296 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2299 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2303 unsigned long n = simple_strtoul(buf, &e, 10);
2305 if (!*buf || (*e && *e != '\n'))
2309 rv = update_raid_disks(mddev, n);
2310 else if (mddev->reshape_position != MaxSector) {
2311 int olddisks = mddev->raid_disks - mddev->delta_disks;
2312 mddev->delta_disks = n - olddisks;
2313 mddev->raid_disks = n;
2315 mddev->raid_disks = n;
2316 return rv ? rv : len;
2318 static struct md_sysfs_entry md_raid_disks =
2319 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2322 chunk_size_show(mddev_t *mddev, char *page)
2324 if (mddev->reshape_position != MaxSector &&
2325 mddev->chunk_size != mddev->new_chunk)
2326 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2328 return sprintf(page, "%d\n", mddev->chunk_size);
2332 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2334 /* can only set chunk_size if array is not yet active */
2336 unsigned long n = simple_strtoul(buf, &e, 10);
2338 if (!*buf || (*e && *e != '\n'))
2343 else if (mddev->reshape_position != MaxSector)
2344 mddev->new_chunk = n;
2346 mddev->chunk_size = n;
2349 static struct md_sysfs_entry md_chunk_size =
2350 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2353 resync_start_show(mddev_t *mddev, char *page)
2355 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2359 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2361 /* can only set chunk_size if array is not yet active */
2363 unsigned long long n = simple_strtoull(buf, &e, 10);
2367 if (!*buf || (*e && *e != '\n'))
2370 mddev->recovery_cp = n;
2373 static struct md_sysfs_entry md_resync_start =
2374 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2377 * The array state can be:
2380 * No devices, no size, no level
2381 * Equivalent to STOP_ARRAY ioctl
2383 * May have some settings, but array is not active
2384 * all IO results in error
2385 * When written, doesn't tear down array, but just stops it
2386 * suspended (not supported yet)
2387 * All IO requests will block. The array can be reconfigured.
2388 * Writing this, if accepted, will block until array is quiessent
2390 * no resync can happen. no superblocks get written.
2391 * write requests fail
2393 * like readonly, but behaves like 'clean' on a write request.
2395 * clean - no pending writes, but otherwise active.
2396 * When written to inactive array, starts without resync
2397 * If a write request arrives then
2398 * if metadata is known, mark 'dirty' and switch to 'active'.
2399 * if not known, block and switch to write-pending
2400 * If written to an active array that has pending writes, then fails.
2402 * fully active: IO and resync can be happening.
2403 * When written to inactive array, starts with resync
2406 * clean, but writes are blocked waiting for 'active' to be written.
2409 * like active, but no writes have been seen for a while (100msec).
2412 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2413 write_pending, active_idle, bad_word};
2414 static char *array_states[] = {
2415 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2416 "write-pending", "active-idle", NULL };
2418 static int match_word(const char *word, char **list)
2421 for (n=0; list[n]; n++)
2422 if (cmd_match(word, list[n]))
2428 array_state_show(mddev_t *mddev, char *page)
2430 enum array_state st = inactive;
2443 else if (mddev->safemode)
2449 if (list_empty(&mddev->disks) &&
2450 mddev->raid_disks == 0 &&
2456 return sprintf(page, "%s\n", array_states[st]);
2459 static int do_md_stop(mddev_t * mddev, int ro);
2460 static int do_md_run(mddev_t * mddev);
2461 static int restart_array(mddev_t *mddev);
2464 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2467 enum array_state st = match_word(buf, array_states);
2472 /* stopping an active array */
2474 if (atomic_read(&mddev->active) > 1)
2476 err = do_md_stop(mddev, 0);
2480 /* stopping an active array */
2482 if (atomic_read(&mddev->active) > 1)
2484 err = do_md_stop(mddev, 2);
2488 break; /* not supported yet */
2491 err = do_md_stop(mddev, 1);
2494 err = do_md_run(mddev);
2498 /* stopping an active array */
2500 err = do_md_stop(mddev, 1);
2502 mddev->ro = 2; /* FIXME mark devices writable */
2505 err = do_md_run(mddev);
2510 restart_array(mddev);
2511 spin_lock_irq(&mddev->write_lock);
2512 if (atomic_read(&mddev->writes_pending) == 0) {
2514 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2516 spin_unlock_irq(&mddev->write_lock);
2519 mddev->recovery_cp = MaxSector;
2520 err = do_md_run(mddev);
2525 restart_array(mddev);
2526 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2527 wake_up(&mddev->sb_wait);
2531 err = do_md_run(mddev);
2536 /* these cannot be set */
2544 static struct md_sysfs_entry md_array_state =
2545 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2548 null_show(mddev_t *mddev, char *page)
2554 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2556 /* buf must be %d:%d\n? giving major and minor numbers */
2557 /* The new device is added to the array.
2558 * If the array has a persistent superblock, we read the
2559 * superblock to initialise info and check validity.
2560 * Otherwise, only checking done is that in bind_rdev_to_array,
2561 * which mainly checks size.
2564 int major = simple_strtoul(buf, &e, 10);
2570 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2572 minor = simple_strtoul(e+1, &e, 10);
2573 if (*e && *e != '\n')
2575 dev = MKDEV(major, minor);
2576 if (major != MAJOR(dev) ||
2577 minor != MINOR(dev))
2581 if (mddev->persistent) {
2582 rdev = md_import_device(dev, mddev->major_version,
2583 mddev->minor_version);
2584 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2585 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2586 mdk_rdev_t, same_set);
2587 err = super_types[mddev->major_version]
2588 .load_super(rdev, rdev0, mddev->minor_version);
2593 rdev = md_import_device(dev, -1, -1);
2596 return PTR_ERR(rdev);
2597 err = bind_rdev_to_array(rdev, mddev);
2601 return err ? err : len;
2604 static struct md_sysfs_entry md_new_device =
2605 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2608 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2611 unsigned long chunk, end_chunk;
2615 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2617 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2618 if (buf == end) break;
2619 if (*end == '-') { /* range */
2621 end_chunk = simple_strtoul(buf, &end, 0);
2622 if (buf == end) break;
2624 if (*end && !isspace(*end)) break;
2625 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2627 while (isspace(*buf)) buf++;
2629 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2634 static struct md_sysfs_entry md_bitmap =
2635 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2638 size_show(mddev_t *mddev, char *page)
2640 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2643 static int update_size(mddev_t *mddev, unsigned long size);
2646 size_store(mddev_t *mddev, const char *buf, size_t len)
2648 /* If array is inactive, we can reduce the component size, but
2649 * not increase it (except from 0).
2650 * If array is active, we can try an on-line resize
2654 unsigned long long size = simple_strtoull(buf, &e, 10);
2655 if (!*buf || *buf == '\n' ||
2660 err = update_size(mddev, size);
2661 md_update_sb(mddev, 1);
2663 if (mddev->size == 0 ||
2669 return err ? err : len;
2672 static struct md_sysfs_entry md_size =
2673 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2677 * This is either 'none' for arrays with externally managed metadata,
2678 * or N.M for internally known formats
2681 metadata_show(mddev_t *mddev, char *page)
2683 if (mddev->persistent)
2684 return sprintf(page, "%d.%d\n",
2685 mddev->major_version, mddev->minor_version);
2687 return sprintf(page, "none\n");
2691 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2695 if (!list_empty(&mddev->disks))
2698 if (cmd_match(buf, "none")) {
2699 mddev->persistent = 0;
2700 mddev->major_version = 0;
2701 mddev->minor_version = 90;
2704 major = simple_strtoul(buf, &e, 10);
2705 if (e==buf || *e != '.')
2708 minor = simple_strtoul(buf, &e, 10);
2709 if (e==buf || (*e && *e != '\n') )
2711 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2713 mddev->major_version = major;
2714 mddev->minor_version = minor;
2715 mddev->persistent = 1;
2719 static struct md_sysfs_entry md_metadata =
2720 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2723 action_show(mddev_t *mddev, char *page)
2725 char *type = "idle";
2726 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2727 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2728 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2730 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2731 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2733 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2740 return sprintf(page, "%s\n", type);
2744 action_store(mddev_t *mddev, const char *page, size_t len)
2746 if (!mddev->pers || !mddev->pers->sync_request)
2749 if (cmd_match(page, "idle")) {
2750 if (mddev->sync_thread) {
2751 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2752 md_unregister_thread(mddev->sync_thread);
2753 mddev->sync_thread = NULL;
2754 mddev->recovery = 0;
2756 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2757 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2759 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2760 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2761 else if (cmd_match(page, "reshape")) {
2763 if (mddev->pers->start_reshape == NULL)
2765 err = mddev->pers->start_reshape(mddev);
2769 if (cmd_match(page, "check"))
2770 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2771 else if (!cmd_match(page, "repair"))
2773 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2774 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2776 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2777 md_wakeup_thread(mddev->thread);
2782 mismatch_cnt_show(mddev_t *mddev, char *page)
2784 return sprintf(page, "%llu\n",
2785 (unsigned long long) mddev->resync_mismatches);
2788 static struct md_sysfs_entry md_scan_mode =
2789 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2792 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2795 sync_min_show(mddev_t *mddev, char *page)
2797 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2798 mddev->sync_speed_min ? "local": "system");
2802 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2806 if (strncmp(buf, "system", 6)==0) {
2807 mddev->sync_speed_min = 0;
2810 min = simple_strtoul(buf, &e, 10);
2811 if (buf == e || (*e && *e != '\n') || min <= 0)
2813 mddev->sync_speed_min = min;
2817 static struct md_sysfs_entry md_sync_min =
2818 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2821 sync_max_show(mddev_t *mddev, char *page)
2823 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2824 mddev->sync_speed_max ? "local": "system");
2828 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2832 if (strncmp(buf, "system", 6)==0) {
2833 mddev->sync_speed_max = 0;
2836 max = simple_strtoul(buf, &e, 10);
2837 if (buf == e || (*e && *e != '\n') || max <= 0)
2839 mddev->sync_speed_max = max;
2843 static struct md_sysfs_entry md_sync_max =
2844 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2848 sync_speed_show(mddev_t *mddev, char *page)
2850 unsigned long resync, dt, db;
2851 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2852 dt = ((jiffies - mddev->resync_mark) / HZ);
2854 db = resync - (mddev->resync_mark_cnt);
2855 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2858 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2861 sync_completed_show(mddev_t *mddev, char *page)
2863 unsigned long max_blocks, resync;
2865 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2866 max_blocks = mddev->resync_max_sectors;
2868 max_blocks = mddev->size << 1;
2870 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2871 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2874 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2877 suspend_lo_show(mddev_t *mddev, char *page)
2879 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2883 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2886 unsigned long long new = simple_strtoull(buf, &e, 10);
2888 if (mddev->pers->quiesce == NULL)
2890 if (buf == e || (*e && *e != '\n'))
2892 if (new >= mddev->suspend_hi ||
2893 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2894 mddev->suspend_lo = new;
2895 mddev->pers->quiesce(mddev, 2);
2900 static struct md_sysfs_entry md_suspend_lo =
2901 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2905 suspend_hi_show(mddev_t *mddev, char *page)
2907 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2911 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2914 unsigned long long new = simple_strtoull(buf, &e, 10);
2916 if (mddev->pers->quiesce == NULL)
2918 if (buf == e || (*e && *e != '\n'))
2920 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2921 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2922 mddev->suspend_hi = new;
2923 mddev->pers->quiesce(mddev, 1);
2924 mddev->pers->quiesce(mddev, 0);
2929 static struct md_sysfs_entry md_suspend_hi =
2930 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2933 reshape_position_show(mddev_t *mddev, char *page)
2935 if (mddev->reshape_position != MaxSector)
2936 return sprintf(page, "%llu\n",
2937 (unsigned long long)mddev->reshape_position);
2938 strcpy(page, "none\n");
2943 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2946 unsigned long long new = simple_strtoull(buf, &e, 10);
2949 if (buf == e || (*e && *e != '\n'))
2951 mddev->reshape_position = new;
2952 mddev->delta_disks = 0;
2953 mddev->new_level = mddev->level;
2954 mddev->new_layout = mddev->layout;
2955 mddev->new_chunk = mddev->chunk_size;
2959 static struct md_sysfs_entry md_reshape_position =
2960 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2961 reshape_position_store);
2964 static struct attribute *md_default_attrs[] = {
2967 &md_raid_disks.attr,
2968 &md_chunk_size.attr,
2970 &md_resync_start.attr,
2972 &md_new_device.attr,
2973 &md_safe_delay.attr,
2974 &md_array_state.attr,
2975 &md_reshape_position.attr,
2979 static struct attribute *md_redundancy_attrs[] = {
2981 &md_mismatches.attr,
2984 &md_sync_speed.attr,
2985 &md_sync_completed.attr,
2986 &md_suspend_lo.attr,
2987 &md_suspend_hi.attr,
2991 static struct attribute_group md_redundancy_group = {
2993 .attrs = md_redundancy_attrs,
2998 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3000 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3001 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3006 rv = mddev_lock(mddev);
3008 rv = entry->show(mddev, page);
3009 mddev_unlock(mddev);
3015 md_attr_store(struct kobject *kobj, struct attribute *attr,
3016 const char *page, size_t length)
3018 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3019 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3024 if (!capable(CAP_SYS_ADMIN))
3026 rv = mddev_lock(mddev);
3028 rv = entry->store(mddev, page, length);
3029 mddev_unlock(mddev);
3034 static void md_free(struct kobject *ko)
3036 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3040 static struct sysfs_ops md_sysfs_ops = {
3041 .show = md_attr_show,
3042 .store = md_attr_store,
3044 static struct kobj_type md_ktype = {
3046 .sysfs_ops = &md_sysfs_ops,
3047 .default_attrs = md_default_attrs,
3052 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3054 static DEFINE_MUTEX(disks_mutex);
3055 mddev_t *mddev = mddev_find(dev);
3056 struct gendisk *disk;
3057 int partitioned = (MAJOR(dev) != MD_MAJOR);
3058 int shift = partitioned ? MdpMinorShift : 0;
3059 int unit = MINOR(dev) >> shift;
3064 mutex_lock(&disks_mutex);
3065 if (mddev->gendisk) {
3066 mutex_unlock(&disks_mutex);
3070 disk = alloc_disk(1 << shift);
3072 mutex_unlock(&disks_mutex);
3076 disk->major = MAJOR(dev);
3077 disk->first_minor = unit << shift;
3079 sprintf(disk->disk_name, "md_d%d", unit);
3081 sprintf(disk->disk_name, "md%d", unit);
3082 disk->fops = &md_fops;
3083 disk->private_data = mddev;
3084 disk->queue = mddev->queue;
3086 mddev->gendisk = disk;
3087 mutex_unlock(&disks_mutex);
3088 mddev->kobj.parent = &disk->kobj;
3089 mddev->kobj.k_name = NULL;
3090 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
3091 mddev->kobj.ktype = &md_ktype;
3092 if (kobject_register(&mddev->kobj))
3093 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3098 static void md_safemode_timeout(unsigned long data)
3100 mddev_t *mddev = (mddev_t *) data;
3102 mddev->safemode = 1;
3103 md_wakeup_thread(mddev->thread);
3106 static int start_dirty_degraded;
3108 static int do_md_run(mddev_t * mddev)
3112 struct list_head *tmp;
3114 struct gendisk *disk;
3115 struct mdk_personality *pers;
3116 char b[BDEVNAME_SIZE];
3118 if (list_empty(&mddev->disks))
3119 /* cannot run an array with no devices.. */
3126 * Analyze all RAID superblock(s)
3128 if (!mddev->raid_disks)
3131 chunk_size = mddev->chunk_size;
3134 if (chunk_size > MAX_CHUNK_SIZE) {
3135 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3136 chunk_size, MAX_CHUNK_SIZE);
3140 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3142 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3143 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3146 if (chunk_size < PAGE_SIZE) {
3147 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3148 chunk_size, PAGE_SIZE);
3152 /* devices must have minimum size of one chunk */
3153 ITERATE_RDEV(mddev,rdev,tmp) {
3154 if (test_bit(Faulty, &rdev->flags))
3156 if (rdev->size < chunk_size / 1024) {
3158 "md: Dev %s smaller than chunk_size:"
3160 bdevname(rdev->bdev,b),
3161 (unsigned long long)rdev->size,
3169 if (mddev->level != LEVEL_NONE)
3170 request_module("md-level-%d", mddev->level);
3171 else if (mddev->clevel[0])
3172 request_module("md-%s", mddev->clevel);
3176 * Drop all container device buffers, from now on
3177 * the only valid external interface is through the md
3179 * Also find largest hardsector size
3181 ITERATE_RDEV(mddev,rdev,tmp) {
3182 if (test_bit(Faulty, &rdev->flags))
3184 sync_blockdev(rdev->bdev);
3185 invalidate_bdev(rdev->bdev);
3188 md_probe(mddev->unit, NULL, NULL);
3189 disk = mddev->gendisk;
3193 spin_lock(&pers_lock);
3194 pers = find_pers(mddev->level, mddev->clevel);
3195 if (!pers || !try_module_get(pers->owner)) {
3196 spin_unlock(&pers_lock);
3197 if (mddev->level != LEVEL_NONE)
3198 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3201 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3206 spin_unlock(&pers_lock);
3207 mddev->level = pers->level;
3208 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3210 if (mddev->reshape_position != MaxSector &&
3211 pers->start_reshape == NULL) {
3212 /* This personality cannot handle reshaping... */
3214 module_put(pers->owner);
3218 if (pers->sync_request) {
3219 /* Warn if this is a potentially silly
3222 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3224 struct list_head *tmp2;
3226 ITERATE_RDEV(mddev, rdev, tmp) {
3227 ITERATE_RDEV(mddev, rdev2, tmp2) {
3229 rdev->bdev->bd_contains ==
3230 rdev2->bdev->bd_contains) {
3232 "%s: WARNING: %s appears to be"
3233 " on the same physical disk as"
3236 bdevname(rdev->bdev,b),
3237 bdevname(rdev2->bdev,b2));
3244 "True protection against single-disk"
3245 " failure might be compromised.\n");
3248 mddev->recovery = 0;
3249 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3250 mddev->barriers_work = 1;
3251 mddev->ok_start_degraded = start_dirty_degraded;
3254 mddev->ro = 2; /* read-only, but switch on first write */
3256 err = mddev->pers->run(mddev);
3257 if (!err && mddev->pers->sync_request) {
3258 err = bitmap_create(mddev);
3260 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3261 mdname(mddev), err);
3262 mddev->pers->stop(mddev);
3266 printk(KERN_ERR "md: pers->run() failed ...\n");
3267 module_put(mddev->pers->owner);
3269 bitmap_destroy(mddev);
3272 if (mddev->pers->sync_request) {
3273 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3275 "md: cannot register extra attributes for %s\n",
3277 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3280 atomic_set(&mddev->writes_pending,0);
3281 mddev->safemode = 0;
3282 mddev->safemode_timer.function = md_safemode_timeout;
3283 mddev->safemode_timer.data = (unsigned long) mddev;
3284 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3287 ITERATE_RDEV(mddev,rdev,tmp)
3288 if (rdev->raid_disk >= 0) {
3290 sprintf(nm, "rd%d", rdev->raid_disk);
3291 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3292 printk("md: cannot register %s for %s\n",
3296 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3299 md_update_sb(mddev, 0);
3301 set_capacity(disk, mddev->array_size<<1);
3303 /* If we call blk_queue_make_request here, it will
3304 * re-initialise max_sectors etc which may have been
3305 * refined inside -> run. So just set the bits we need to set.
3306 * Most initialisation happended when we called
3307 * blk_queue_make_request(..., md_fail_request)
3310 mddev->queue->queuedata = mddev;
3311 mddev->queue->make_request_fn = mddev->pers->make_request;
3313 /* If there is a partially-recovered drive we need to
3314 * start recovery here. If we leave it to md_check_recovery,
3315 * it will remove the drives and not do the right thing
3317 if (mddev->degraded && !mddev->sync_thread) {
3318 struct list_head *rtmp;
3320 ITERATE_RDEV(mddev,rdev,rtmp)
3321 if (rdev->raid_disk >= 0 &&
3322 !test_bit(In_sync, &rdev->flags) &&
3323 !test_bit(Faulty, &rdev->flags))
3324 /* complete an interrupted recovery */
3326 if (spares && mddev->pers->sync_request) {
3327 mddev->recovery = 0;
3328 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3329 mddev->sync_thread = md_register_thread(md_do_sync,
3332 if (!mddev->sync_thread) {
3333 printk(KERN_ERR "%s: could not start resync"
3336 /* leave the spares where they are, it shouldn't hurt */
3337 mddev->recovery = 0;
3341 md_wakeup_thread(mddev->thread);
3342 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3345 md_new_event(mddev);
3346 kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3350 static int restart_array(mddev_t *mddev)
3352 struct gendisk *disk = mddev->gendisk;
3356 * Complain if it has no devices
3359 if (list_empty(&mddev->disks))
3367 mddev->safemode = 0;
3369 set_disk_ro(disk, 0);
3371 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3374 * Kick recovery or resync if necessary
3376 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3377 md_wakeup_thread(mddev->thread);
3378 md_wakeup_thread(mddev->sync_thread);
3387 /* similar to deny_write_access, but accounts for our holding a reference
3388 * to the file ourselves */
3389 static int deny_bitmap_write_access(struct file * file)
3391 struct inode *inode = file->f_mapping->host;
3393 spin_lock(&inode->i_lock);
3394 if (atomic_read(&inode->i_writecount) > 1) {
3395 spin_unlock(&inode->i_lock);
3398 atomic_set(&inode->i_writecount, -1);
3399 spin_unlock(&inode->i_lock);
3404 static void restore_bitmap_write_access(struct file *file)
3406 struct inode *inode = file->f_mapping->host;
3408 spin_lock(&inode->i_lock);
3409 atomic_set(&inode->i_writecount, 1);
3410 spin_unlock(&inode->i_lock);
3414 * 0 - completely stop and dis-assemble array
3415 * 1 - switch to readonly
3416 * 2 - stop but do not disassemble array
3418 static int do_md_stop(mddev_t * mddev, int mode)
3421 struct gendisk *disk = mddev->gendisk;
3424 if (atomic_read(&mddev->active)>2) {
3425 printk("md: %s still in use.\n",mdname(mddev));
3429 if (mddev->sync_thread) {
3430 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3431 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3432 md_unregister_thread(mddev->sync_thread);
3433 mddev->sync_thread = NULL;
3436 del_timer_sync(&mddev->safemode_timer);
3438 invalidate_partition(disk, 0);
3441 case 1: /* readonly */
3447 case 0: /* disassemble */
3449 bitmap_flush(mddev);
3450 md_super_wait(mddev);
3452 set_disk_ro(disk, 0);
3453 blk_queue_make_request(mddev->queue, md_fail_request);
3454 mddev->pers->stop(mddev);
3455 mddev->queue->merge_bvec_fn = NULL;
3456 mddev->queue->unplug_fn = NULL;
3457 mddev->queue->issue_flush_fn = NULL;
3458 mddev->queue->backing_dev_info.congested_fn = NULL;
3459 if (mddev->pers->sync_request)
3460 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3462 module_put(mddev->pers->owner);
3465 set_capacity(disk, 0);
3471 if (!mddev->in_sync || mddev->flags) {
3472 /* mark array as shutdown cleanly */
3474 md_update_sb(mddev, 1);
3477 set_disk_ro(disk, 1);
3478 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3482 * Free resources if final stop
3486 struct list_head *tmp;
3488 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3490 bitmap_destroy(mddev);
3491 if (mddev->bitmap_file) {
3492 restore_bitmap_write_access(mddev->bitmap_file);
3493 fput(mddev->bitmap_file);
3494 mddev->bitmap_file = NULL;
3496 mddev->bitmap_offset = 0;
3498 ITERATE_RDEV(mddev,rdev,tmp)
3499 if (rdev->raid_disk >= 0) {
3501 sprintf(nm, "rd%d", rdev->raid_disk);
3502 sysfs_remove_link(&mddev->kobj, nm);
3505 /* make sure all delayed_delete calls have finished */
3506 flush_scheduled_work();
3508 export_array(mddev);
3510 mddev->array_size = 0;
3512 mddev->raid_disks = 0;
3513 mddev->recovery_cp = 0;
3514 mddev->reshape_position = MaxSector;
3516 } else if (mddev->pers)
3517 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3520 md_new_event(mddev);
3526 static void autorun_array(mddev_t *mddev)
3529 struct list_head *tmp;
3532 if (list_empty(&mddev->disks))
3535 printk(KERN_INFO "md: running: ");
3537 ITERATE_RDEV(mddev,rdev,tmp) {
3538 char b[BDEVNAME_SIZE];
3539 printk("<%s>", bdevname(rdev->bdev,b));
3543 err = do_md_run (mddev);
3545 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3546 do_md_stop (mddev, 0);
3551 * lets try to run arrays based on all disks that have arrived
3552 * until now. (those are in pending_raid_disks)
3554 * the method: pick the first pending disk, collect all disks with
3555 * the same UUID, remove all from the pending list and put them into
3556 * the 'same_array' list. Then order this list based on superblock
3557 * update time (freshest comes first), kick out 'old' disks and
3558 * compare superblocks. If everything's fine then run it.
3560 * If "unit" is allocated, then bump its reference count
3562 static void autorun_devices(int part)
3564 struct list_head *tmp;
3565 mdk_rdev_t *rdev0, *rdev;
3567 char b[BDEVNAME_SIZE];
3569 printk(KERN_INFO "md: autorun ...\n");
3570 while (!list_empty(&pending_raid_disks)) {
3573 LIST_HEAD(candidates);
3574 rdev0 = list_entry(pending_raid_disks.next,
3575 mdk_rdev_t, same_set);
3577 printk(KERN_INFO "md: considering %s ...\n",
3578 bdevname(rdev0->bdev,b));
3579 INIT_LIST_HEAD(&candidates);
3580 ITERATE_RDEV_PENDING(rdev,tmp)
3581 if (super_90_load(rdev, rdev0, 0) >= 0) {
3582 printk(KERN_INFO "md: adding %s ...\n",
3583 bdevname(rdev->bdev,b));
3584 list_move(&rdev->same_set, &candidates);
3587 * now we have a set of devices, with all of them having
3588 * mostly sane superblocks. It's time to allocate the
3592 dev = MKDEV(mdp_major,
3593 rdev0->preferred_minor << MdpMinorShift);
3594 unit = MINOR(dev) >> MdpMinorShift;
3596 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3599 if (rdev0->preferred_minor != unit) {
3600 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3601 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3605 md_probe(dev, NULL, NULL);
3606 mddev = mddev_find(dev);
3609 "md: cannot allocate memory for md drive.\n");
3612 if (mddev_lock(mddev))
3613 printk(KERN_WARNING "md: %s locked, cannot run\n",
3615 else if (mddev->raid_disks || mddev->major_version
3616 || !list_empty(&mddev->disks)) {
3618 "md: %s already running, cannot run %s\n",
3619 mdname(mddev), bdevname(rdev0->bdev,b));
3620 mddev_unlock(mddev);
3622 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3623 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3624 list_del_init(&rdev->same_set);
3625 if (bind_rdev_to_array(rdev, mddev))
3628 autorun_array(mddev);
3629 mddev_unlock(mddev);
3631 /* on success, candidates will be empty, on error
3634 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3638 printk(KERN_INFO "md: ... autorun DONE.\n");
3640 #endif /* !MODULE */
3642 static int get_version(void __user * arg)
3646 ver.major = MD_MAJOR_VERSION;
3647 ver.minor = MD_MINOR_VERSION;
3648 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3650 if (copy_to_user(arg, &ver, sizeof(ver)))
3656 static int get_array_info(mddev_t * mddev, void __user * arg)
3658 mdu_array_info_t info;
3659 int nr,working,active,failed,spare;
3661 struct list_head *tmp;
3663 nr=working=active=failed=spare=0;
3664 ITERATE_RDEV(mddev,rdev,tmp) {
3666 if (test_bit(Faulty, &rdev->flags))
3670 if (test_bit(In_sync, &rdev->flags))
3677 info.major_version = mddev->major_version;
3678 info.minor_version = mddev->minor_version;
3679 info.patch_version = MD_PATCHLEVEL_VERSION;
3680 info.ctime = mddev->ctime;
3681 info.level = mddev->level;
3682 info.size = mddev->size;
3683 if (info.size != mddev->size) /* overflow */
3686 info.raid_disks = mddev->raid_disks;
3687 info.md_minor = mddev->md_minor;
3688 info.not_persistent= !mddev->persistent;
3690 info.utime = mddev->utime;
3693 info.state = (1<<MD_SB_CLEAN);
3694 if (mddev->bitmap && mddev->bitmap_offset)
3695 info.state = (1<<MD_SB_BITMAP_PRESENT);
3696 info.active_disks = active;
3697 info.working_disks = working;
3698 info.failed_disks = failed;
3699 info.spare_disks = spare;
3701 info.layout = mddev->layout;
3702 info.chunk_size = mddev->chunk_size;
3704 if (copy_to_user(arg, &info, sizeof(info)))
3710 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3712 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3713 char *ptr, *buf = NULL;
3716 md_allow_write(mddev);
3718 file = kmalloc(sizeof(*file), GFP_KERNEL);
3722 /* bitmap disabled, zero the first byte and copy out */
3723 if (!mddev->bitmap || !mddev->bitmap->file) {
3724 file->pathname[0] = '\0';
3728 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3732 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3736 strcpy(file->pathname, ptr);
3740 if (copy_to_user(arg, file, sizeof(*file)))
3748 static int get_disk_info(mddev_t * mddev, void __user * arg)
3750 mdu_disk_info_t info;
3754 if (copy_from_user(&info, arg, sizeof(info)))
3759 rdev = find_rdev_nr(mddev, nr);
3761 info.major = MAJOR(rdev->bdev->bd_dev);
3762 info.minor = MINOR(rdev->bdev->bd_dev);
3763 info.raid_disk = rdev->raid_disk;
3765 if (test_bit(Faulty, &rdev->flags))
3766 info.state |= (1<<MD_DISK_FAULTY);
3767 else if (test_bit(In_sync, &rdev->flags)) {
3768 info.state |= (1<<MD_DISK_ACTIVE);
3769 info.state |= (1<<MD_DISK_SYNC);
3771 if (test_bit(WriteMostly, &rdev->flags))
3772 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3774 info.major = info.minor = 0;
3775 info.raid_disk = -1;
3776 info.state = (1<<MD_DISK_REMOVED);
3779 if (copy_to_user(arg, &info, sizeof(info)))
3785 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3787 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3789 dev_t dev = MKDEV(info->major,info->minor);
3791 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3794 if (!mddev->raid_disks) {
3796 /* expecting a device which has a superblock */
3797 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3800 "md: md_import_device returned %ld\n",
3802 return PTR_ERR(rdev);
3804 if (!list_empty(&mddev->disks)) {
3805 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3806 mdk_rdev_t, same_set);
3807 int err = super_types[mddev->major_version]
3808 .load_super(rdev, rdev0, mddev->minor_version);
3811 "md: %s has different UUID to %s\n",
3812 bdevname(rdev->bdev,b),
3813 bdevname(rdev0->bdev,b2));
3818 err = bind_rdev_to_array(rdev, mddev);
3825 * add_new_disk can be used once the array is assembled
3826 * to add "hot spares". They must already have a superblock
3831 if (!mddev->pers->hot_add_disk) {
3833 "%s: personality does not support diskops!\n",
3837 if (mddev->persistent)
3838 rdev = md_import_device(dev, mddev->major_version,
3839 mddev->minor_version);
3841 rdev = md_import_device(dev, -1, -1);
3844 "md: md_import_device returned %ld\n",
3846 return PTR_ERR(rdev);
3848 /* set save_raid_disk if appropriate */
3849 if (!mddev->persistent) {
3850 if (info->state & (1<<MD_DISK_SYNC) &&
3851 info->raid_disk < mddev->raid_disks)
3852 rdev->raid_disk = info->raid_disk;
3854 rdev->raid_disk = -1;
3856 super_types[mddev->major_version].
3857 validate_super(mddev, rdev);
3858 rdev->saved_raid_disk = rdev->raid_disk;
3860 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3861 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3862 set_bit(WriteMostly, &rdev->flags);
3864 rdev->raid_disk = -1;
3865 err = bind_rdev_to_array(rdev, mddev);
3866 if (!err && !mddev->pers->hot_remove_disk) {
3867 /* If there is hot_add_disk but no hot_remove_disk
3868 * then added disks for geometry changes,
3869 * and should be added immediately.
3871 super_types[mddev->major_version].
3872 validate_super(mddev, rdev);
3873 err = mddev->pers->hot_add_disk(mddev, rdev);
3875 unbind_rdev_from_array(rdev);
3880 md_update_sb(mddev, 1);
3881 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3882 md_wakeup_thread(mddev->thread);
3886 /* otherwise, add_new_disk is only allowed
3887 * for major_version==0 superblocks
3889 if (mddev->major_version != 0) {
3890 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3895 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3897 rdev = md_import_device (dev, -1, 0);
3900 "md: error, md_import_device() returned %ld\n",
3902 return PTR_ERR(rdev);
3904 rdev->desc_nr = info->number;
3905 if (info->raid_disk < mddev->raid_disks)
3906 rdev->raid_disk = info->raid_disk;
3908 rdev->raid_disk = -1;
3912 if (rdev->raid_disk < mddev->raid_disks)
3913 if (info->state & (1<<MD_DISK_SYNC))
3914 set_bit(In_sync, &rdev->flags);
3916 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3917 set_bit(WriteMostly, &rdev->flags);
3919 if (!mddev->persistent) {
3920 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3921 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3923 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3924 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3926 err = bind_rdev_to_array(rdev, mddev);
3936 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3938 char b[BDEVNAME_SIZE];
3944 rdev = find_rdev(mddev, dev);
3948 if (rdev->raid_disk >= 0)
3951 kick_rdev_from_array(rdev);
3952 md_update_sb(mddev, 1);
3953 md_new_event(mddev);
3957 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3958 bdevname(rdev->bdev,b), mdname(mddev));
3962 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3964 char b[BDEVNAME_SIZE];
3972 if (mddev->major_version != 0) {
3973 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3974 " version-0 superblocks.\n",
3978 if (!mddev->pers->hot_add_disk) {
3980 "%s: personality does not support diskops!\n",
3985 rdev = md_import_device (dev, -1, 0);
3988 "md: error, md_import_device() returned %ld\n",
3993 if (mddev->persistent)
3994 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3997 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3999 size = calc_dev_size(rdev, mddev->chunk_size);
4002 if (test_bit(Faulty, &rdev->flags)) {
4004 "md: can not hot-add faulty %s disk to %s!\n",
4005 bdevname(rdev->bdev,b), mdname(mddev));
4009 clear_bit(In_sync, &rdev->flags);
4011 rdev->saved_raid_disk = -1;
4012 err = bind_rdev_to_array(rdev, mddev);
4017 * The rest should better be atomic, we can have disk failures
4018 * noticed in interrupt contexts ...
4021 if (rdev->desc_nr == mddev->max_disks) {
4022 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4025 goto abort_unbind_export;
4028 rdev->raid_disk = -1;
4030 md_update_sb(mddev, 1);
4033 * Kick recovery, maybe this spare has to be added to the
4034 * array immediately.
4036 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4037 md_wakeup_thread(mddev->thread);
4038 md_new_event(mddev);
4041 abort_unbind_export:
4042 unbind_rdev_from_array(rdev);
4049 static int set_bitmap_file(mddev_t *mddev, int fd)
4054 if (!mddev->pers->quiesce)
4056 if (mddev->recovery || mddev->sync_thread)
4058 /* we should be able to change the bitmap.. */
4064 return -EEXIST; /* cannot add when bitmap is present */
4065 mddev->bitmap_file = fget(fd);
4067 if (mddev->bitmap_file == NULL) {
4068 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4073 err = deny_bitmap_write_access(mddev->bitmap_file);
4075 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4077 fput(mddev->bitmap_file);
4078 mddev->bitmap_file = NULL;
4081 mddev->bitmap_offset = 0; /* file overrides offset */
4082 } else if (mddev->bitmap == NULL)
4083 return -ENOENT; /* cannot remove what isn't there */
4086 mddev->pers->quiesce(mddev, 1);
4088 err = bitmap_create(mddev);
4089 if (fd < 0 || err) {
4090 bitmap_destroy(mddev);
4091 fd = -1; /* make sure to put the file */
4093 mddev->pers->quiesce(mddev, 0);
4096 if (mddev->bitmap_file) {
4097 restore_bitmap_write_access(mddev->bitmap_file);
4098 fput(mddev->bitmap_file);
4100 mddev->bitmap_file = NULL;
4107 * set_array_info is used two different ways
4108 * The original usage is when creating a new array.
4109 * In this usage, raid_disks is > 0 and it together with
4110 * level, size, not_persistent,layout,chunksize determine the
4111 * shape of the array.
4112 * This will always create an array with a type-0.90.0 superblock.
4113 * The newer usage is when assembling an array.
4114 * In this case raid_disks will be 0, and the major_version field is
4115 * use to determine which style super-blocks are to be found on the devices.
4116 * The minor and patch _version numbers are also kept incase the
4117 * super_block handler wishes to interpret them.
4119 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4122 if (info->raid_disks == 0) {
4123 /* just setting version number for superblock loading */
4124 if (info->major_version < 0 ||
4125 info->major_version >= ARRAY_SIZE(super_types) ||
4126 super_types[info->major_version].name == NULL) {
4127 /* maybe try to auto-load a module? */
4129 "md: superblock version %d not known\n",
4130 info->major_version);
4133 mddev->major_version = info->major_version;
4134 mddev->minor_version = info->minor_version;
4135 mddev->patch_version = info->patch_version;
4136 mddev->persistent = !info->not_persistent;
4139 mddev->major_version = MD_MAJOR_VERSION;
4140 mddev->minor_version = MD_MINOR_VERSION;
4141 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4142 mddev->ctime = get_seconds();
4144 mddev->level = info->level;
4145 mddev->clevel[0] = 0;
4146 mddev->size = info->size;
4147 mddev->raid_disks = info->raid_disks;
4148 /* don't set md_minor, it is determined by which /dev/md* was
4151 if (info->state & (1<<MD_SB_CLEAN))
4152 mddev->recovery_cp = MaxSector;
4154 mddev->recovery_cp = 0;
4155 mddev->persistent = ! info->not_persistent;
4157 mddev->layout = info->layout;
4158 mddev->chunk_size = info->chunk_size;
4160 mddev->max_disks = MD_SB_DISKS;
4163 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4165 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4166 mddev->bitmap_offset = 0;
4168 mddev->reshape_position = MaxSector;
4171 * Generate a 128 bit UUID
4173 get_random_bytes(mddev->uuid, 16);
4175 mddev->new_level = mddev->level;
4176 mddev->new_chunk = mddev->chunk_size;
4177 mddev->new_layout = mddev->layout;
4178 mddev->delta_disks = 0;
4183 static int update_size(mddev_t *mddev, unsigned long size)
4187 struct list_head *tmp;
4188 int fit = (size == 0);
4190 if (mddev->pers->resize == NULL)
4192 /* The "size" is the amount of each device that is used.
4193 * This can only make sense for arrays with redundancy.
4194 * linear and raid0 always use whatever space is available
4195 * We can only consider changing the size if no resync
4196 * or reconstruction is happening, and if the new size
4197 * is acceptable. It must fit before the sb_offset or,
4198 * if that is <data_offset, it must fit before the
4199 * size of each device.
4200 * If size is zero, we find the largest size that fits.
4202 if (mddev->sync_thread)
4204 ITERATE_RDEV(mddev,rdev,tmp) {
4206 avail = rdev->size * 2;
4208 if (fit && (size == 0 || size > avail/2))
4210 if (avail < ((sector_t)size << 1))
4213 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4215 struct block_device *bdev;
4217 bdev = bdget_disk(mddev->gendisk, 0);
4219 mutex_lock(&bdev->bd_inode->i_mutex);
4220 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4221 mutex_unlock(&bdev->bd_inode->i_mutex);
4228 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4231 /* change the number of raid disks */
4232 if (mddev->pers->check_reshape == NULL)
4234 if (raid_disks <= 0 ||
4235 raid_disks >= mddev->max_disks)
4237 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4239 mddev->delta_disks = raid_disks - mddev->raid_disks;
4241 rv = mddev->pers->check_reshape(mddev);
4247 * update_array_info is used to change the configuration of an
4249 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4250 * fields in the info are checked against the array.
4251 * Any differences that cannot be handled will cause an error.
4252 * Normally, only one change can be managed at a time.
4254 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4260 /* calculate expected state,ignoring low bits */
4261 if (mddev->bitmap && mddev->bitmap_offset)
4262 state |= (1 << MD_SB_BITMAP_PRESENT);
4264 if (mddev->major_version != info->major_version ||
4265 mddev->minor_version != info->minor_version ||
4266 /* mddev->patch_version != info->patch_version || */
4267 mddev->ctime != info->ctime ||
4268 mddev->level != info->level ||
4269 /* mddev->layout != info->layout || */
4270 !mddev->persistent != info->not_persistent||
4271 mddev->chunk_size != info->chunk_size ||
4272 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4273 ((state^info->state) & 0xfffffe00)
4276 /* Check there is only one change */
4277 if (info->size >= 0 && mddev->size != info->size) cnt++;
4278 if (mddev->raid_disks != info->raid_disks) cnt++;
4279 if (mddev->layout != info->layout) cnt++;
4280 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4281 if (cnt == 0) return 0;
4282 if (cnt > 1) return -EINVAL;
4284 if (mddev->layout != info->layout) {
4286 * we don't need to do anything at the md level, the
4287 * personality will take care of it all.
4289 if (mddev->pers->reconfig == NULL)
4292 return mddev->pers->reconfig(mddev, info->layout, -1);
4294 if (info->size >= 0 && mddev->size != info->size)
4295 rv = update_size(mddev, info->size);
4297 if (mddev->raid_disks != info->raid_disks)
4298 rv = update_raid_disks(mddev, info->raid_disks);
4300 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4301 if (mddev->pers->quiesce == NULL)
4303 if (mddev->recovery || mddev->sync_thread)
4305 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4306 /* add the bitmap */
4309 if (mddev->default_bitmap_offset == 0)
4311 mddev->bitmap_offset = mddev->default_bitmap_offset;
4312 mddev->pers->quiesce(mddev, 1);
4313 rv = bitmap_create(mddev);
4315 bitmap_destroy(mddev);
4316 mddev->pers->quiesce(mddev, 0);
4318 /* remove the bitmap */
4321 if (mddev->bitmap->file)
4323 mddev->pers->quiesce(mddev, 1);
4324 bitmap_destroy(mddev);
4325 mddev->pers->quiesce(mddev, 0);
4326 mddev->bitmap_offset = 0;
4329 md_update_sb(mddev, 1);
4333 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4337 if (mddev->pers == NULL)
4340 rdev = find_rdev(mddev, dev);
4344 md_error(mddev, rdev);
4348 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4350 mddev_t *mddev = bdev->bd_disk->private_data;
4354 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4358 static int md_ioctl(struct inode *inode, struct file *file,
4359 unsigned int cmd, unsigned long arg)
4362 void __user *argp = (void __user *)arg;
4363 mddev_t *mddev = NULL;
4365 if (!capable(CAP_SYS_ADMIN))
4369 * Commands dealing with the RAID driver but not any
4375 err = get_version(argp);
4378 case PRINT_RAID_DEBUG:
4386 autostart_arrays(arg);
4393 * Commands creating/starting a new array:
4396 mddev = inode->i_bdev->bd_disk->private_data;
4403 err = mddev_lock(mddev);
4406 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4413 case SET_ARRAY_INFO:
4415 mdu_array_info_t info;
4417 memset(&info, 0, sizeof(info));
4418 else if (copy_from_user(&info, argp, sizeof(info))) {
4423 err = update_array_info(mddev, &info);
4425 printk(KERN_WARNING "md: couldn't update"
4426 " array info. %d\n", err);
4431 if (!list_empty(&mddev->disks)) {
4433 "md: array %s already has disks!\n",
4438 if (mddev->raid_disks) {
4440 "md: array %s already initialised!\n",
4445 err = set_array_info(mddev, &info);
4447 printk(KERN_WARNING "md: couldn't set"
4448 " array info. %d\n", err);
4458 * Commands querying/configuring an existing array:
4460 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4461 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4462 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4463 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4464 && cmd != GET_BITMAP_FILE) {
4470 * Commands even a read-only array can execute:
4474 case GET_ARRAY_INFO:
4475 err = get_array_info(mddev, argp);
4478 case GET_BITMAP_FILE:
4479 err = get_bitmap_file(mddev, argp);
4483 err = get_disk_info(mddev, argp);
4486 case RESTART_ARRAY_RW:
4487 err = restart_array(mddev);
4491 err = do_md_stop (mddev, 0);
4495 err = do_md_stop (mddev, 1);
4499 * We have a problem here : there is no easy way to give a CHS
4500 * virtual geometry. We currently pretend that we have a 2 heads
4501 * 4 sectors (with a BIG number of cylinders...). This drives
4502 * dosfs just mad... ;-)
4507 * The remaining ioctls are changing the state of the
4508 * superblock, so we do not allow them on read-only arrays.
4509 * However non-MD ioctls (e.g. get-size) will still come through
4510 * here and hit the 'default' below, so only disallow
4511 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4513 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4514 mddev->ro && mddev->pers) {
4515 if (mddev->ro == 2) {
4517 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4518 md_wakeup_thread(mddev->thread);
4530 mdu_disk_info_t info;
4531 if (copy_from_user(&info, argp, sizeof(info)))
4534 err = add_new_disk(mddev, &info);
4538 case HOT_REMOVE_DISK:
4539 err = hot_remove_disk(mddev, new_decode_dev(arg));
4543 err = hot_add_disk(mddev, new_decode_dev(arg));
4546 case SET_DISK_FAULTY:
4547 err = set_disk_faulty(mddev, new_decode_dev(arg));
4551 err = do_md_run (mddev);
4554 case SET_BITMAP_FILE:
4555 err = set_bitmap_file(mddev, (int)arg);
4565 mddev_unlock(mddev);
4575 static int md_open(struct inode *inode, struct file *file)
4578 * Succeed if we can lock the mddev, which confirms that
4579 * it isn't being stopped right now.
4581 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4584 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4589 mddev_unlock(mddev);
4591 check_disk_change(inode->i_bdev);
4596 static int md_release(struct inode *inode, struct file * file)
4598 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4606 static int md_media_changed(struct gendisk *disk)
4608 mddev_t *mddev = disk->private_data;
4610 return mddev->changed;
4613 static int md_revalidate(struct gendisk *disk)
4615 mddev_t *mddev = disk->private_data;
4620 static struct block_device_operations md_fops =
4622 .owner = THIS_MODULE,
4624 .release = md_release,
4626 .getgeo = md_getgeo,
4627 .media_changed = md_media_changed,
4628 .revalidate_disk= md_revalidate,
4631 static int md_thread(void * arg)
4633 mdk_thread_t *thread = arg;
4636 * md_thread is a 'system-thread', it's priority should be very
4637 * high. We avoid resource deadlocks individually in each
4638 * raid personality. (RAID5 does preallocation) We also use RR and
4639 * the very same RT priority as kswapd, thus we will never get
4640 * into a priority inversion deadlock.
4642 * we definitely have to have equal or higher priority than
4643 * bdflush, otherwise bdflush will deadlock if there are too
4644 * many dirty RAID5 blocks.
4647 allow_signal(SIGKILL);
4648 while (!kthread_should_stop()) {
4650 /* We need to wait INTERRUPTIBLE so that
4651 * we don't add to the load-average.
4652 * That means we need to be sure no signals are
4655 if (signal_pending(current))
4656 flush_signals(current);
4658 wait_event_interruptible_timeout
4660 test_bit(THREAD_WAKEUP, &thread->flags)
4661 || kthread_should_stop(),
4664 clear_bit(THREAD_WAKEUP, &thread->flags);
4666 thread->run(thread->mddev);
4672 void md_wakeup_thread(mdk_thread_t *thread)
4675 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4676 set_bit(THREAD_WAKEUP, &thread->flags);
4677 wake_up(&thread->wqueue);
4681 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4684 mdk_thread_t *thread;
4686 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4690 init_waitqueue_head(&thread->wqueue);
4693 thread->mddev = mddev;
4694 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4695 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4696 if (IS_ERR(thread->tsk)) {
4703 void md_unregister_thread(mdk_thread_t *thread)
4705 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4707 kthread_stop(thread->tsk);
4711 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4718 if (!rdev || test_bit(Faulty, &rdev->flags))
4721 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4723 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4724 __builtin_return_address(0),__builtin_return_address(1),
4725 __builtin_return_address(2),__builtin_return_address(3));
4729 if (!mddev->pers->error_handler)
4731 mddev->pers->error_handler(mddev,rdev);
4732 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4733 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4734 md_wakeup_thread(mddev->thread);
4735 md_new_event_inintr(mddev);
4738 /* seq_file implementation /proc/mdstat */
4740 static void status_unused(struct seq_file *seq)
4744 struct list_head *tmp;
4746 seq_printf(seq, "unused devices: ");
4748 ITERATE_RDEV_PENDING(rdev,tmp) {
4749 char b[BDEVNAME_SIZE];
4751 seq_printf(seq, "%s ",
4752 bdevname(rdev->bdev,b));
4755 seq_printf(seq, "<none>");
4757 seq_printf(seq, "\n");
4761 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4763 sector_t max_blocks, resync, res;
4764 unsigned long dt, db, rt;
4766 unsigned int per_milli;
4768 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4770 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4771 max_blocks = mddev->resync_max_sectors >> 1;
4773 max_blocks = mddev->size;
4776 * Should not happen.
4782 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4783 * in a sector_t, and (max_blocks>>scale) will fit in a
4784 * u32, as those are the requirements for sector_div.
4785 * Thus 'scale' must be at least 10
4788 if (sizeof(sector_t) > sizeof(unsigned long)) {
4789 while ( max_blocks/2 > (1ULL<<(scale+32)))
4792 res = (resync>>scale)*1000;
4793 sector_div(res, (u32)((max_blocks>>scale)+1));
4797 int i, x = per_milli/50, y = 20-x;
4798 seq_printf(seq, "[");
4799 for (i = 0; i < x; i++)
4800 seq_printf(seq, "=");
4801 seq_printf(seq, ">");
4802 for (i = 0; i < y; i++)
4803 seq_printf(seq, ".");
4804 seq_printf(seq, "] ");
4806 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4807 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4809 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4811 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4812 "resync" : "recovery"))),
4813 per_milli/10, per_milli % 10,
4814 (unsigned long long) resync,
4815 (unsigned long long) max_blocks);
4818 * We do not want to overflow, so the order of operands and
4819 * the * 100 / 100 trick are important. We do a +1 to be
4820 * safe against division by zero. We only estimate anyway.
4822 * dt: time from mark until now
4823 * db: blocks written from mark until now
4824 * rt: remaining time
4826 dt = ((jiffies - mddev->resync_mark) / HZ);
4828 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4829 - mddev->resync_mark_cnt;
4830 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4832 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4834 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4837 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4839 struct list_head *tmp;
4849 spin_lock(&all_mddevs_lock);
4850 list_for_each(tmp,&all_mddevs)
4852 mddev = list_entry(tmp, mddev_t, all_mddevs);
4854 spin_unlock(&all_mddevs_lock);
4857 spin_unlock(&all_mddevs_lock);
4859 return (void*)2;/* tail */
4863 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4865 struct list_head *tmp;
4866 mddev_t *next_mddev, *mddev = v;
4872 spin_lock(&all_mddevs_lock);
4874 tmp = all_mddevs.next;
4876 tmp = mddev->all_mddevs.next;
4877 if (tmp != &all_mddevs)
4878 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4880 next_mddev = (void*)2;
4883 spin_unlock(&all_mddevs_lock);
4891 static void md_seq_stop(struct seq_file *seq, void *v)
4895 if (mddev && v != (void*)1 && v != (void*)2)
4899 struct mdstat_info {
4903 static int md_seq_show(struct seq_file *seq, void *v)
4907 struct list_head *tmp2;
4909 struct mdstat_info *mi = seq->private;
4910 struct bitmap *bitmap;
4912 if (v == (void*)1) {
4913 struct mdk_personality *pers;
4914 seq_printf(seq, "Personalities : ");
4915 spin_lock(&pers_lock);
4916 list_for_each_entry(pers, &pers_list, list)
4917 seq_printf(seq, "[%s] ", pers->name);
4919 spin_unlock(&pers_lock);
4920 seq_printf(seq, "\n");
4921 mi->event = atomic_read(&md_event_count);
4924 if (v == (void*)2) {
4929 if (mddev_lock(mddev) < 0)
4932 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4933 seq_printf(seq, "%s : %sactive", mdname(mddev),
4934 mddev->pers ? "" : "in");
4937 seq_printf(seq, " (read-only)");
4939 seq_printf(seq, "(auto-read-only)");
4940 seq_printf(seq, " %s", mddev->pers->name);
4944 ITERATE_RDEV(mddev,rdev,tmp2) {
4945 char b[BDEVNAME_SIZE];
4946 seq_printf(seq, " %s[%d]",
4947 bdevname(rdev->bdev,b), rdev->desc_nr);
4948 if (test_bit(WriteMostly, &rdev->flags))
4949 seq_printf(seq, "(W)");
4950 if (test_bit(Faulty, &rdev->flags)) {
4951 seq_printf(seq, "(F)");
4953 } else if (rdev->raid_disk < 0)
4954 seq_printf(seq, "(S)"); /* spare */
4958 if (!list_empty(&mddev->disks)) {
4960 seq_printf(seq, "\n %llu blocks",
4961 (unsigned long long)mddev->array_size);
4963 seq_printf(seq, "\n %llu blocks",
4964 (unsigned long long)size);
4966 if (mddev->persistent) {
4967 if (mddev->major_version != 0 ||
4968 mddev->minor_version != 90) {
4969 seq_printf(seq," super %d.%d",
4970 mddev->major_version,
4971 mddev->minor_version);
4974 seq_printf(seq, " super non-persistent");
4977 mddev->pers->status (seq, mddev);
4978 seq_printf(seq, "\n ");
4979 if (mddev->pers->sync_request) {
4980 if (mddev->curr_resync > 2) {
4981 status_resync (seq, mddev);
4982 seq_printf(seq, "\n ");
4983 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4984 seq_printf(seq, "\tresync=DELAYED\n ");
4985 else if (mddev->recovery_cp < MaxSector)
4986 seq_printf(seq, "\tresync=PENDING\n ");
4989 seq_printf(seq, "\n ");
4991 if ((bitmap = mddev->bitmap)) {
4992 unsigned long chunk_kb;
4993 unsigned long flags;
4994 spin_lock_irqsave(&bitmap->lock, flags);
4995 chunk_kb = bitmap->chunksize >> 10;
4996 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4998 bitmap->pages - bitmap->missing_pages,
5000 (bitmap->pages - bitmap->missing_pages)
5001 << (PAGE_SHIFT - 10),
5002 chunk_kb ? chunk_kb : bitmap->chunksize,
5003 chunk_kb ? "KB" : "B");
5005 seq_printf(seq, ", file: ");
5006 seq_path(seq, bitmap->file->f_path.mnt,
5007 bitmap->file->f_path.dentry," \t\n");
5010 seq_printf(seq, "\n");
5011 spin_unlock_irqrestore(&bitmap->lock, flags);
5014 seq_printf(seq, "\n");
5016 mddev_unlock(mddev);
5021 static struct seq_operations md_seq_ops = {
5022 .start = md_seq_start,
5023 .next = md_seq_next,
5024 .stop = md_seq_stop,
5025 .show = md_seq_show,
5028 static int md_seq_open(struct inode *inode, struct file *file)
5031 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5035 error = seq_open(file, &md_seq_ops);
5039 struct seq_file *p = file->private_data;
5041 mi->event = atomic_read(&md_event_count);
5046 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5048 struct seq_file *m = filp->private_data;
5049 struct mdstat_info *mi = m->private;
5052 poll_wait(filp, &md_event_waiters, wait);
5054 /* always allow read */
5055 mask = POLLIN | POLLRDNORM;
5057 if (mi->event != atomic_read(&md_event_count))
5058 mask |= POLLERR | POLLPRI;
5062 static const struct file_operations md_seq_fops = {
5063 .owner = THIS_MODULE,
5064 .open = md_seq_open,
5066 .llseek = seq_lseek,
5067 .release = seq_release_private,
5068 .poll = mdstat_poll,
5071 int register_md_personality(struct mdk_personality *p)
5073 spin_lock(&pers_lock);
5074 list_add_tail(&p->list, &pers_list);
5075 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5076 spin_unlock(&pers_lock);
5080 int unregister_md_personality(struct mdk_personality *p)
5082 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5083 spin_lock(&pers_lock);
5084 list_del_init(&p->list);
5085 spin_unlock(&pers_lock);
5089 static int is_mddev_idle(mddev_t *mddev)
5092 struct list_head *tmp;
5097 ITERATE_RDEV(mddev,rdev,tmp) {
5098 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5099 curr_events = disk_stat_read(disk, sectors[0]) +
5100 disk_stat_read(disk, sectors[1]) -
5101 atomic_read(&disk->sync_io);
5102 /* sync IO will cause sync_io to increase before the disk_stats
5103 * as sync_io is counted when a request starts, and
5104 * disk_stats is counted when it completes.
5105 * So resync activity will cause curr_events to be smaller than
5106 * when there was no such activity.
5107 * non-sync IO will cause disk_stat to increase without
5108 * increasing sync_io so curr_events will (eventually)
5109 * be larger than it was before. Once it becomes
5110 * substantially larger, the test below will cause
5111 * the array to appear non-idle, and resync will slow
5113 * If there is a lot of outstanding resync activity when
5114 * we set last_event to curr_events, then all that activity
5115 * completing might cause the array to appear non-idle
5116 * and resync will be slowed down even though there might
5117 * not have been non-resync activity. This will only
5118 * happen once though. 'last_events' will soon reflect
5119 * the state where there is little or no outstanding
5120 * resync requests, and further resync activity will
5121 * always make curr_events less than last_events.
5124 if (curr_events - rdev->last_events > 4096) {
5125 rdev->last_events = curr_events;
5132 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5134 /* another "blocks" (512byte) blocks have been synced */
5135 atomic_sub(blocks, &mddev->recovery_active);
5136 wake_up(&mddev->recovery_wait);
5138 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5139 md_wakeup_thread(mddev->thread);
5140 // stop recovery, signal do_sync ....
5145 /* md_write_start(mddev, bi)
5146 * If we need to update some array metadata (e.g. 'active' flag
5147 * in superblock) before writing, schedule a superblock update
5148 * and wait for it to complete.
5150 void md_write_start(mddev_t *mddev, struct bio *bi)
5152 if (bio_data_dir(bi) != WRITE)
5155 BUG_ON(mddev->ro == 1);
5156 if (mddev->ro == 2) {
5157 /* need to switch to read/write */
5159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5160 md_wakeup_thread(mddev->thread);
5162 atomic_inc(&mddev->writes_pending);
5163 if (mddev->in_sync) {
5164 spin_lock_irq(&mddev->write_lock);
5165 if (mddev->in_sync) {
5167 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5168 md_wakeup_thread(mddev->thread);
5170 spin_unlock_irq(&mddev->write_lock);
5172 wait_event(mddev->sb_wait, mddev->flags==0);
5175 void md_write_end(mddev_t *mddev)
5177 if (atomic_dec_and_test(&mddev->writes_pending)) {
5178 if (mddev->safemode == 2)
5179 md_wakeup_thread(mddev->thread);
5180 else if (mddev->safemode_delay)
5181 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5185 /* md_allow_write(mddev)
5186 * Calling this ensures that the array is marked 'active' so that writes
5187 * may proceed without blocking. It is important to call this before
5188 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5189 * Must be called with mddev_lock held.
5191 void md_allow_write(mddev_t *mddev)
5198 spin_lock_irq(&mddev->write_lock);
5199 if (mddev->in_sync) {
5201 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5202 if (mddev->safemode_delay &&
5203 mddev->safemode == 0)
5204 mddev->safemode = 1;
5205 spin_unlock_irq(&mddev->write_lock);
5206 md_update_sb(mddev, 0);
5208 spin_unlock_irq(&mddev->write_lock);
5210 EXPORT_SYMBOL_GPL(md_allow_write);
5212 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5214 #define SYNC_MARKS 10
5215 #define SYNC_MARK_STEP (3*HZ)
5216 void md_do_sync(mddev_t *mddev)
5219 unsigned int currspeed = 0,
5221 sector_t max_sectors,j, io_sectors;
5222 unsigned long mark[SYNC_MARKS];
5223 sector_t mark_cnt[SYNC_MARKS];
5225 struct list_head *tmp;
5226 sector_t last_check;
5228 struct list_head *rtmp;
5232 /* just incase thread restarts... */
5233 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5235 if (mddev->ro) /* never try to sync a read-only array */
5238 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5239 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5240 desc = "data-check";
5241 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5242 desc = "requested-resync";
5245 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5250 /* we overload curr_resync somewhat here.
5251 * 0 == not engaged in resync at all
5252 * 2 == checking that there is no conflict with another sync
5253 * 1 == like 2, but have yielded to allow conflicting resync to
5255 * other == active in resync - this many blocks
5257 * Before starting a resync we must have set curr_resync to
5258 * 2, and then checked that every "conflicting" array has curr_resync
5259 * less than ours. When we find one that is the same or higher
5260 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5261 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5262 * This will mean we have to start checking from the beginning again.
5267 mddev->curr_resync = 2;
5270 if (kthread_should_stop()) {
5271 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5274 ITERATE_MDDEV(mddev2,tmp) {
5275 if (mddev2 == mddev)
5277 if (mddev2->curr_resync &&
5278 match_mddev_units(mddev,mddev2)) {
5280 if (mddev < mddev2 && mddev->curr_resync == 2) {
5281 /* arbitrarily yield */
5282 mddev->curr_resync = 1;
5283 wake_up(&resync_wait);
5285 if (mddev > mddev2 && mddev->curr_resync == 1)
5286 /* no need to wait here, we can wait the next
5287 * time 'round when curr_resync == 2
5290 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5291 if (!kthread_should_stop() &&
5292 mddev2->curr_resync >= mddev->curr_resync) {
5293 printk(KERN_INFO "md: delaying %s of %s"
5294 " until %s has finished (they"
5295 " share one or more physical units)\n",
5296 desc, mdname(mddev), mdname(mddev2));
5299 finish_wait(&resync_wait, &wq);
5302 finish_wait(&resync_wait, &wq);
5305 } while (mddev->curr_resync < 2);
5308 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5309 /* resync follows the size requested by the personality,
5310 * which defaults to physical size, but can be virtual size
5312 max_sectors = mddev->resync_max_sectors;
5313 mddev->resync_mismatches = 0;
5314 /* we don't use the checkpoint if there's a bitmap */
5315 if (!mddev->bitmap &&
5316 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5317 j = mddev->recovery_cp;
5318 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5319 max_sectors = mddev->size << 1;
5321 /* recovery follows the physical size of devices */
5322 max_sectors = mddev->size << 1;
5324 ITERATE_RDEV(mddev,rdev,rtmp)
5325 if (rdev->raid_disk >= 0 &&
5326 !test_bit(Faulty, &rdev->flags) &&
5327 !test_bit(In_sync, &rdev->flags) &&
5328 rdev->recovery_offset < j)
5329 j = rdev->recovery_offset;
5332 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5333 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5334 " %d KB/sec/disk.\n", speed_min(mddev));
5335 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5336 "(but not more than %d KB/sec) for %s.\n",
5337 speed_max(mddev), desc);
5339 is_mddev_idle(mddev); /* this also initializes IO event counters */
5342 for (m = 0; m < SYNC_MARKS; m++) {
5344 mark_cnt[m] = io_sectors;
5347 mddev->resync_mark = mark[last_mark];
5348 mddev->resync_mark_cnt = mark_cnt[last_mark];
5351 * Tune reconstruction:
5353 window = 32*(PAGE_SIZE/512);
5354 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5355 window/2,(unsigned long long) max_sectors/2);
5357 atomic_set(&mddev->recovery_active, 0);
5358 init_waitqueue_head(&mddev->recovery_wait);
5363 "md: resuming %s of %s from checkpoint.\n",
5364 desc, mdname(mddev));
5365 mddev->curr_resync = j;
5368 while (j < max_sectors) {
5372 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5373 currspeed < speed_min(mddev));
5375 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5379 if (!skipped) { /* actual IO requested */
5380 io_sectors += sectors;
5381 atomic_add(sectors, &mddev->recovery_active);
5385 if (j>1) mddev->curr_resync = j;
5386 mddev->curr_mark_cnt = io_sectors;
5387 if (last_check == 0)
5388 /* this is the earliers that rebuilt will be
5389 * visible in /proc/mdstat
5391 md_new_event(mddev);
5393 if (last_check + window > io_sectors || j == max_sectors)
5396 last_check = io_sectors;
5398 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5399 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5403 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5405 int next = (last_mark+1) % SYNC_MARKS;
5407 mddev->resync_mark = mark[next];
5408 mddev->resync_mark_cnt = mark_cnt[next];
5409 mark[next] = jiffies;
5410 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5415 if (kthread_should_stop()) {
5417 * got a signal, exit.
5420 "md: md_do_sync() got signal ... exiting\n");
5421 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5426 * this loop exits only if either when we are slower than
5427 * the 'hard' speed limit, or the system was IO-idle for
5429 * the system might be non-idle CPU-wise, but we only care
5430 * about not overloading the IO subsystem. (things like an
5431 * e2fsck being done on the RAID array should execute fast)
5433 mddev->queue->unplug_fn(mddev->queue);
5436 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5437 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5439 if (currspeed > speed_min(mddev)) {
5440 if ((currspeed > speed_max(mddev)) ||
5441 !is_mddev_idle(mddev)) {
5447 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5449 * this also signals 'finished resyncing' to md_stop
5452 mddev->queue->unplug_fn(mddev->queue);
5454 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5456 /* tell personality that we are finished */
5457 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5459 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5460 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5461 mddev->curr_resync > 2) {
5462 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5463 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5464 if (mddev->curr_resync >= mddev->recovery_cp) {
5466 "md: checkpointing %s of %s.\n",
5467 desc, mdname(mddev));
5468 mddev->recovery_cp = mddev->curr_resync;
5471 mddev->recovery_cp = MaxSector;
5473 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5474 mddev->curr_resync = MaxSector;
5475 ITERATE_RDEV(mddev,rdev,rtmp)
5476 if (rdev->raid_disk >= 0 &&
5477 !test_bit(Faulty, &rdev->flags) &&
5478 !test_bit(In_sync, &rdev->flags) &&
5479 rdev->recovery_offset < mddev->curr_resync)
5480 rdev->recovery_offset = mddev->curr_resync;
5483 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5486 mddev->curr_resync = 0;
5487 wake_up(&resync_wait);
5488 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5489 md_wakeup_thread(mddev->thread);
5491 EXPORT_SYMBOL_GPL(md_do_sync);
5494 static int remove_and_add_spares(mddev_t *mddev)
5497 struct list_head *rtmp;
5500 ITERATE_RDEV(mddev,rdev,rtmp)
5501 if (rdev->raid_disk >= 0 &&
5502 (test_bit(Faulty, &rdev->flags) ||
5503 ! test_bit(In_sync, &rdev->flags)) &&
5504 atomic_read(&rdev->nr_pending)==0) {
5505 if (mddev->pers->hot_remove_disk(
5506 mddev, rdev->raid_disk)==0) {
5508 sprintf(nm,"rd%d", rdev->raid_disk);
5509 sysfs_remove_link(&mddev->kobj, nm);
5510 rdev->raid_disk = -1;
5514 if (mddev->degraded) {
5515 ITERATE_RDEV(mddev,rdev,rtmp)
5516 if (rdev->raid_disk < 0
5517 && !test_bit(Faulty, &rdev->flags)) {
5518 rdev->recovery_offset = 0;
5519 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5521 sprintf(nm, "rd%d", rdev->raid_disk);
5522 if (sysfs_create_link(&mddev->kobj,
5525 "md: cannot register "
5529 md_new_event(mddev);
5537 * This routine is regularly called by all per-raid-array threads to
5538 * deal with generic issues like resync and super-block update.
5539 * Raid personalities that don't have a thread (linear/raid0) do not
5540 * need this as they never do any recovery or update the superblock.
5542 * It does not do any resync itself, but rather "forks" off other threads
5543 * to do that as needed.
5544 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5545 * "->recovery" and create a thread at ->sync_thread.
5546 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5547 * and wakeups up this thread which will reap the thread and finish up.
5548 * This thread also removes any faulty devices (with nr_pending == 0).
5550 * The overall approach is:
5551 * 1/ if the superblock needs updating, update it.
5552 * 2/ If a recovery thread is running, don't do anything else.
5553 * 3/ If recovery has finished, clean up, possibly marking spares active.
5554 * 4/ If there are any faulty devices, remove them.
5555 * 5/ If array is degraded, try to add spares devices
5556 * 6/ If array has spares or is not in-sync, start a resync thread.
5558 void md_check_recovery(mddev_t *mddev)
5561 struct list_head *rtmp;
5565 bitmap_daemon_work(mddev->bitmap);
5570 if (signal_pending(current)) {
5571 if (mddev->pers->sync_request) {
5572 printk(KERN_INFO "md: %s in immediate safe mode\n",
5574 mddev->safemode = 2;
5576 flush_signals(current);
5581 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5582 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5583 (mddev->safemode == 1) ||
5584 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5585 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5589 if (mddev_trylock(mddev)) {
5592 spin_lock_irq(&mddev->write_lock);
5593 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5594 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5596 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5598 if (mddev->safemode == 1)
5599 mddev->safemode = 0;
5600 spin_unlock_irq(&mddev->write_lock);
5603 md_update_sb(mddev, 0);
5606 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5607 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5608 /* resync/recovery still happening */
5609 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5612 if (mddev->sync_thread) {
5613 /* resync has finished, collect result */
5614 md_unregister_thread(mddev->sync_thread);
5615 mddev->sync_thread = NULL;
5616 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5617 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5619 /* activate any spares */
5620 mddev->pers->spare_active(mddev);
5622 md_update_sb(mddev, 1);
5624 /* if array is no-longer degraded, then any saved_raid_disk
5625 * information must be scrapped
5627 if (!mddev->degraded)
5628 ITERATE_RDEV(mddev,rdev,rtmp)
5629 rdev->saved_raid_disk = -1;
5631 mddev->recovery = 0;
5632 /* flag recovery needed just to double check */
5633 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5634 md_new_event(mddev);
5637 /* Clear some bits that don't mean anything, but
5640 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5641 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5642 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5643 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5645 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5647 /* no recovery is running.
5648 * remove any failed drives, then
5649 * add spares if possible.
5650 * Spare are also removed and re-added, to allow
5651 * the personality to fail the re-add.
5654 if (mddev->reshape_position != MaxSector) {
5655 if (mddev->pers->check_reshape(mddev) != 0)
5656 /* Cannot proceed */
5658 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5659 } else if ((spares = remove_and_add_spares(mddev))) {
5660 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5661 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5662 } else if (mddev->recovery_cp < MaxSector) {
5663 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5664 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5665 /* nothing to be done ... */
5668 if (mddev->pers->sync_request) {
5669 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5670 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5671 /* We are adding a device or devices to an array
5672 * which has the bitmap stored on all devices.
5673 * So make sure all bitmap pages get written
5675 bitmap_write_all(mddev->bitmap);
5677 mddev->sync_thread = md_register_thread(md_do_sync,
5680 if (!mddev->sync_thread) {
5681 printk(KERN_ERR "%s: could not start resync"
5684 /* leave the spares where they are, it shouldn't hurt */
5685 mddev->recovery = 0;
5687 md_wakeup_thread(mddev->sync_thread);
5688 md_new_event(mddev);
5691 mddev_unlock(mddev);
5695 static int md_notify_reboot(struct notifier_block *this,
5696 unsigned long code, void *x)
5698 struct list_head *tmp;
5701 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5703 printk(KERN_INFO "md: stopping all md devices.\n");
5705 ITERATE_MDDEV(mddev,tmp)
5706 if (mddev_trylock(mddev)) {
5707 do_md_stop (mddev, 1);
5708 mddev_unlock(mddev);
5711 * certain more exotic SCSI devices are known to be
5712 * volatile wrt too early system reboots. While the
5713 * right place to handle this issue is the given
5714 * driver, we do want to have a safe RAID driver ...
5721 static struct notifier_block md_notifier = {
5722 .notifier_call = md_notify_reboot,
5724 .priority = INT_MAX, /* before any real devices */
5727 static void md_geninit(void)
5729 struct proc_dir_entry *p;
5731 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5733 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5735 p->proc_fops = &md_seq_fops;
5738 static int __init md_init(void)
5740 if (register_blkdev(MAJOR_NR, "md"))
5742 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5743 unregister_blkdev(MAJOR_NR, "md");
5746 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5747 md_probe, NULL, NULL);
5748 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5749 md_probe, NULL, NULL);
5751 register_reboot_notifier(&md_notifier);
5752 raid_table_header = register_sysctl_table(raid_root_table);
5762 * Searches all registered partitions for autorun RAID arrays
5765 static dev_t detected_devices[128];
5768 void md_autodetect_dev(dev_t dev)
5770 if (dev_cnt >= 0 && dev_cnt < 127)
5771 detected_devices[dev_cnt++] = dev;
5775 static void autostart_arrays(int part)
5780 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5782 for (i = 0; i < dev_cnt; i++) {
5783 dev_t dev = detected_devices[i];
5785 rdev = md_import_device(dev,0, 90);
5789 if (test_bit(Faulty, &rdev->flags)) {
5793 list_add(&rdev->same_set, &pending_raid_disks);
5797 autorun_devices(part);
5800 #endif /* !MODULE */
5802 static __exit void md_exit(void)
5805 struct list_head *tmp;
5807 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5808 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5810 unregister_blkdev(MAJOR_NR,"md");
5811 unregister_blkdev(mdp_major, "mdp");
5812 unregister_reboot_notifier(&md_notifier);
5813 unregister_sysctl_table(raid_table_header);
5814 remove_proc_entry("mdstat", NULL);
5815 ITERATE_MDDEV(mddev,tmp) {
5816 struct gendisk *disk = mddev->gendisk;
5819 export_array(mddev);
5822 mddev->gendisk = NULL;
5827 subsys_initcall(md_init);
5828 module_exit(md_exit)
5830 static int get_ro(char *buffer, struct kernel_param *kp)
5832 return sprintf(buffer, "%d", start_readonly);
5834 static int set_ro(const char *val, struct kernel_param *kp)
5837 int num = simple_strtoul(val, &e, 10);
5838 if (*val && (*e == '\0' || *e == '\n')) {
5839 start_readonly = num;
5845 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5846 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5849 EXPORT_SYMBOL(register_md_personality);
5850 EXPORT_SYMBOL(unregister_md_personality);
5851 EXPORT_SYMBOL(md_error);
5852 EXPORT_SYMBOL(md_done_sync);
5853 EXPORT_SYMBOL(md_write_start);
5854 EXPORT_SYMBOL(md_write_end);
5855 EXPORT_SYMBOL(md_register_thread);
5856 EXPORT_SYMBOL(md_unregister_thread);
5857 EXPORT_SYMBOL(md_wakeup_thread);
5858 EXPORT_SYMBOL(md_check_recovery);
5859 MODULE_LICENSE("GPL");
5861 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);