2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part);
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
75 static void md_print_devices(void);
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
100 static inline int speed_max(mddev_t *mddev)
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
106 static struct ctl_table_header *raid_table_header;
108 static ctl_table raid_table[] = {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = &proc_dointvec,
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = &proc_dointvec,
128 static ctl_table raid_dir_table[] = {
130 .ctl_name = DEV_RAID,
133 .mode = S_IRUGO|S_IXUGO,
139 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static struct block_device_operations md_fops;
152 static int start_readonly;
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
172 EXPORT_SYMBOL_GPL(md_new_event);
174 /* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
177 static void md_new_event_inintr(mddev_t *mddev)
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
198 #define for_each_mddev(mddev,tmp) \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
214 static int md_fail_request (struct request_queue *q, struct bio *bio)
220 static inline mddev_t *mddev_get(mddev_t *mddev)
222 atomic_inc(&mddev->active);
226 static void mddev_put(mddev_t *mddev)
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
232 spin_unlock(&all_mddevs_lock);
233 blk_cleanup_queue(mddev->queue);
234 kobject_put(&mddev->kobj);
236 spin_unlock(&all_mddevs_lock);
239 static mddev_t * mddev_find(dev_t unit)
241 mddev_t *mddev, *new = NULL;
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
248 spin_unlock(&all_mddevs_lock);
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
258 spin_unlock(&all_mddevs_lock);
260 new = kzalloc(sizeof(*new), GFP_KERNEL);
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
270 mutex_init(&new->reconfig_mutex);
271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
275 spin_lock_init(&new->write_lock);
276 init_waitqueue_head(&new->sb_wait);
277 new->reshape_position = MaxSector;
278 new->resync_max = MaxSector;
280 new->queue = blk_alloc_queue(GFP_KERNEL);
285 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
287 blk_queue_make_request(new->queue, md_fail_request);
292 static inline int mddev_lock(mddev_t * mddev)
294 return mutex_lock_interruptible(&mddev->reconfig_mutex);
297 static inline int mddev_trylock(mddev_t * mddev)
299 return mutex_trylock(&mddev->reconfig_mutex);
302 static inline void mddev_unlock(mddev_t * mddev)
304 mutex_unlock(&mddev->reconfig_mutex);
306 md_wakeup_thread(mddev->thread);
309 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
312 struct list_head *tmp;
314 rdev_for_each(rdev, tmp, mddev) {
315 if (rdev->desc_nr == nr)
321 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
323 struct list_head *tmp;
326 rdev_for_each(rdev, tmp, mddev) {
327 if (rdev->bdev->bd_dev == dev)
333 static struct mdk_personality *find_pers(int level, char *clevel)
335 struct mdk_personality *pers;
336 list_for_each_entry(pers, &pers_list, list) {
337 if (level != LEVEL_NONE && pers->level == level)
339 if (strcmp(pers->name, clevel)==0)
345 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
347 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
348 return MD_NEW_SIZE_BLOCKS(size);
351 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
355 size = rdev->sb_offset;
358 size &= ~((sector_t)chunk_size/1024 - 1);
362 static int alloc_disk_sb(mdk_rdev_t * rdev)
367 rdev->sb_page = alloc_page(GFP_KERNEL);
368 if (!rdev->sb_page) {
369 printk(KERN_ALERT "md: out of memory.\n");
376 static void free_disk_sb(mdk_rdev_t * rdev)
379 put_page(rdev->sb_page);
381 rdev->sb_page = NULL;
388 static void super_written(struct bio *bio, int error)
390 mdk_rdev_t *rdev = bio->bi_private;
391 mddev_t *mddev = rdev->mddev;
393 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
394 printk("md: super_written gets error=%d, uptodate=%d\n",
395 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
396 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
397 md_error(mddev, rdev);
400 if (atomic_dec_and_test(&mddev->pending_writes))
401 wake_up(&mddev->sb_wait);
405 static void super_written_barrier(struct bio *bio, int error)
407 struct bio *bio2 = bio->bi_private;
408 mdk_rdev_t *rdev = bio2->bi_private;
409 mddev_t *mddev = rdev->mddev;
411 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
412 error == -EOPNOTSUPP) {
414 /* barriers don't appear to be supported :-( */
415 set_bit(BarriersNotsupp, &rdev->flags);
416 mddev->barriers_work = 0;
417 spin_lock_irqsave(&mddev->write_lock, flags);
418 bio2->bi_next = mddev->biolist;
419 mddev->biolist = bio2;
420 spin_unlock_irqrestore(&mddev->write_lock, flags);
421 wake_up(&mddev->sb_wait);
425 bio->bi_private = rdev;
426 super_written(bio, error);
430 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
431 sector_t sector, int size, struct page *page)
433 /* write first size bytes of page to sector of rdev
434 * Increment mddev->pending_writes before returning
435 * and decrement it on completion, waking up sb_wait
436 * if zero is reached.
437 * If an error occurred, call md_error
439 * As we might need to resubmit the request if BIO_RW_BARRIER
440 * causes ENOTSUPP, we allocate a spare bio...
442 struct bio *bio = bio_alloc(GFP_NOIO, 1);
443 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
445 bio->bi_bdev = rdev->bdev;
446 bio->bi_sector = sector;
447 bio_add_page(bio, page, size, 0);
448 bio->bi_private = rdev;
449 bio->bi_end_io = super_written;
452 atomic_inc(&mddev->pending_writes);
453 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
455 rw |= (1<<BIO_RW_BARRIER);
456 rbio = bio_clone(bio, GFP_NOIO);
457 rbio->bi_private = bio;
458 rbio->bi_end_io = super_written_barrier;
459 submit_bio(rw, rbio);
464 void md_super_wait(mddev_t *mddev)
466 /* wait for all superblock writes that were scheduled to complete.
467 * if any had to be retried (due to BARRIER problems), retry them
471 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
472 if (atomic_read(&mddev->pending_writes)==0)
474 while (mddev->biolist) {
476 spin_lock_irq(&mddev->write_lock);
477 bio = mddev->biolist;
478 mddev->biolist = bio->bi_next ;
480 spin_unlock_irq(&mddev->write_lock);
481 submit_bio(bio->bi_rw, bio);
485 finish_wait(&mddev->sb_wait, &wq);
488 static void bi_complete(struct bio *bio, int error)
490 complete((struct completion*)bio->bi_private);
493 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
494 struct page *page, int rw)
496 struct bio *bio = bio_alloc(GFP_NOIO, 1);
497 struct completion event;
500 rw |= (1 << BIO_RW_SYNC);
503 bio->bi_sector = sector;
504 bio_add_page(bio, page, size, 0);
505 init_completion(&event);
506 bio->bi_private = &event;
507 bio->bi_end_io = bi_complete;
509 wait_for_completion(&event);
511 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
515 EXPORT_SYMBOL_GPL(sync_page_io);
517 static int read_disk_sb(mdk_rdev_t * rdev, int size)
519 char b[BDEVNAME_SIZE];
520 if (!rdev->sb_page) {
528 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
534 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
535 bdevname(rdev->bdev,b));
539 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
541 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
542 (sb1->set_uuid1 == sb2->set_uuid1) &&
543 (sb1->set_uuid2 == sb2->set_uuid2) &&
544 (sb1->set_uuid3 == sb2->set_uuid3))
552 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
555 mdp_super_t *tmp1, *tmp2;
557 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
558 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
560 if (!tmp1 || !tmp2) {
562 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
570 * nr_disks is not constant
575 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
587 static u32 md_csum_fold(u32 csum)
589 csum = (csum & 0xffff) + (csum >> 16);
590 return (csum & 0xffff) + (csum >> 16);
593 static unsigned int calc_sb_csum(mdp_super_t * sb)
596 u32 *sb32 = (u32*)sb;
598 unsigned int disk_csum, csum;
600 disk_csum = sb->sb_csum;
603 for (i = 0; i < MD_SB_BYTES/4 ; i++)
605 csum = (newcsum & 0xffffffff) + (newcsum>>32);
609 /* This used to use csum_partial, which was wrong for several
610 * reasons including that different results are returned on
611 * different architectures. It isn't critical that we get exactly
612 * the same return value as before (we always csum_fold before
613 * testing, and that removes any differences). However as we
614 * know that csum_partial always returned a 16bit value on
615 * alphas, do a fold to maximise conformity to previous behaviour.
617 sb->sb_csum = md_csum_fold(disk_csum);
619 sb->sb_csum = disk_csum;
626 * Handle superblock details.
627 * We want to be able to handle multiple superblock formats
628 * so we have a common interface to them all, and an array of
629 * different handlers.
630 * We rely on user-space to write the initial superblock, and support
631 * reading and updating of superblocks.
632 * Interface methods are:
633 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
634 * loads and validates a superblock on dev.
635 * if refdev != NULL, compare superblocks on both devices
637 * 0 - dev has a superblock that is compatible with refdev
638 * 1 - dev has a superblock that is compatible and newer than refdev
639 * so dev should be used as the refdev in future
640 * -EINVAL superblock incompatible or invalid
641 * -othererror e.g. -EIO
643 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
644 * Verify that dev is acceptable into mddev.
645 * The first time, mddev->raid_disks will be 0, and data from
646 * dev should be merged in. Subsequent calls check that dev
647 * is new enough. Return 0 or -EINVAL
649 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
650 * Update the superblock for rdev with data in mddev
651 * This does not write to disc.
657 struct module *owner;
658 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
659 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
660 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
664 * load_super for 0.90.0
666 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
668 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
674 * Calculate the position of the superblock,
675 * it's at the end of the disk.
677 * It also happens to be a multiple of 4Kb.
679 sb_offset = calc_dev_sboffset(rdev->bdev);
680 rdev->sb_offset = sb_offset;
682 ret = read_disk_sb(rdev, MD_SB_BYTES);
687 bdevname(rdev->bdev, b);
688 sb = (mdp_super_t*)page_address(rdev->sb_page);
690 if (sb->md_magic != MD_SB_MAGIC) {
691 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
696 if (sb->major_version != 0 ||
697 sb->minor_version < 90 ||
698 sb->minor_version > 91) {
699 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
700 sb->major_version, sb->minor_version,
705 if (sb->raid_disks <= 0)
708 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
709 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
714 rdev->preferred_minor = sb->md_minor;
715 rdev->data_offset = 0;
716 rdev->sb_size = MD_SB_BYTES;
718 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
719 if (sb->level != 1 && sb->level != 4
720 && sb->level != 5 && sb->level != 6
721 && sb->level != 10) {
722 /* FIXME use a better test */
724 "md: bitmaps not supported for this level.\n");
729 if (sb->level == LEVEL_MULTIPATH)
732 rdev->desc_nr = sb->this_disk.number;
738 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
739 if (!uuid_equal(refsb, sb)) {
740 printk(KERN_WARNING "md: %s has different UUID to %s\n",
741 b, bdevname(refdev->bdev,b2));
744 if (!sb_equal(refsb, sb)) {
745 printk(KERN_WARNING "md: %s has same UUID"
746 " but different superblock to %s\n",
747 b, bdevname(refdev->bdev, b2));
751 ev2 = md_event(refsb);
757 rdev->size = calc_dev_size(rdev, sb->chunk_size);
759 if (rdev->size < sb->size && sb->level > 1)
760 /* "this cannot possibly happen" ... */
768 * validate_super for 0.90.0
770 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
773 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
774 __u64 ev1 = md_event(sb);
776 rdev->raid_disk = -1;
777 clear_bit(Faulty, &rdev->flags);
778 clear_bit(In_sync, &rdev->flags);
779 clear_bit(WriteMostly, &rdev->flags);
780 clear_bit(BarriersNotsupp, &rdev->flags);
782 if (mddev->raid_disks == 0) {
783 mddev->major_version = 0;
784 mddev->minor_version = sb->minor_version;
785 mddev->patch_version = sb->patch_version;
787 mddev->chunk_size = sb->chunk_size;
788 mddev->ctime = sb->ctime;
789 mddev->utime = sb->utime;
790 mddev->level = sb->level;
791 mddev->clevel[0] = 0;
792 mddev->layout = sb->layout;
793 mddev->raid_disks = sb->raid_disks;
794 mddev->size = sb->size;
796 mddev->bitmap_offset = 0;
797 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
799 if (mddev->minor_version >= 91) {
800 mddev->reshape_position = sb->reshape_position;
801 mddev->delta_disks = sb->delta_disks;
802 mddev->new_level = sb->new_level;
803 mddev->new_layout = sb->new_layout;
804 mddev->new_chunk = sb->new_chunk;
806 mddev->reshape_position = MaxSector;
807 mddev->delta_disks = 0;
808 mddev->new_level = mddev->level;
809 mddev->new_layout = mddev->layout;
810 mddev->new_chunk = mddev->chunk_size;
813 if (sb->state & (1<<MD_SB_CLEAN))
814 mddev->recovery_cp = MaxSector;
816 if (sb->events_hi == sb->cp_events_hi &&
817 sb->events_lo == sb->cp_events_lo) {
818 mddev->recovery_cp = sb->recovery_cp;
820 mddev->recovery_cp = 0;
823 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
824 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
825 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
826 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
828 mddev->max_disks = MD_SB_DISKS;
830 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
831 mddev->bitmap_file == NULL)
832 mddev->bitmap_offset = mddev->default_bitmap_offset;
834 } else if (mddev->pers == NULL) {
835 /* Insist on good event counter while assembling */
837 if (ev1 < mddev->events)
839 } else if (mddev->bitmap) {
840 /* if adding to array with a bitmap, then we can accept an
841 * older device ... but not too old.
843 if (ev1 < mddev->bitmap->events_cleared)
846 if (ev1 < mddev->events)
847 /* just a hot-add of a new device, leave raid_disk at -1 */
851 if (mddev->level != LEVEL_MULTIPATH) {
852 desc = sb->disks + rdev->desc_nr;
854 if (desc->state & (1<<MD_DISK_FAULTY))
855 set_bit(Faulty, &rdev->flags);
856 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
857 desc->raid_disk < mddev->raid_disks */) {
858 set_bit(In_sync, &rdev->flags);
859 rdev->raid_disk = desc->raid_disk;
861 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
862 set_bit(WriteMostly, &rdev->flags);
863 } else /* MULTIPATH are always insync */
864 set_bit(In_sync, &rdev->flags);
869 * sync_super for 0.90.0
871 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
874 struct list_head *tmp;
876 int next_spare = mddev->raid_disks;
879 /* make rdev->sb match mddev data..
882 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
883 * 3/ any empty disks < next_spare become removed
885 * disks[0] gets initialised to REMOVED because
886 * we cannot be sure from other fields if it has
887 * been initialised or not.
890 int active=0, working=0,failed=0,spare=0,nr_disks=0;
892 rdev->sb_size = MD_SB_BYTES;
894 sb = (mdp_super_t*)page_address(rdev->sb_page);
896 memset(sb, 0, sizeof(*sb));
898 sb->md_magic = MD_SB_MAGIC;
899 sb->major_version = mddev->major_version;
900 sb->patch_version = mddev->patch_version;
901 sb->gvalid_words = 0; /* ignored */
902 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
903 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
904 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
905 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
907 sb->ctime = mddev->ctime;
908 sb->level = mddev->level;
909 sb->size = mddev->size;
910 sb->raid_disks = mddev->raid_disks;
911 sb->md_minor = mddev->md_minor;
912 sb->not_persistent = 0;
913 sb->utime = mddev->utime;
915 sb->events_hi = (mddev->events>>32);
916 sb->events_lo = (u32)mddev->events;
918 if (mddev->reshape_position == MaxSector)
919 sb->minor_version = 90;
921 sb->minor_version = 91;
922 sb->reshape_position = mddev->reshape_position;
923 sb->new_level = mddev->new_level;
924 sb->delta_disks = mddev->delta_disks;
925 sb->new_layout = mddev->new_layout;
926 sb->new_chunk = mddev->new_chunk;
928 mddev->minor_version = sb->minor_version;
931 sb->recovery_cp = mddev->recovery_cp;
932 sb->cp_events_hi = (mddev->events>>32);
933 sb->cp_events_lo = (u32)mddev->events;
934 if (mddev->recovery_cp == MaxSector)
935 sb->state = (1<< MD_SB_CLEAN);
939 sb->layout = mddev->layout;
940 sb->chunk_size = mddev->chunk_size;
942 if (mddev->bitmap && mddev->bitmap_file == NULL)
943 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
945 sb->disks[0].state = (1<<MD_DISK_REMOVED);
946 rdev_for_each(rdev2, tmp, mddev) {
949 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
950 && !test_bit(Faulty, &rdev2->flags))
951 desc_nr = rdev2->raid_disk;
953 desc_nr = next_spare++;
954 rdev2->desc_nr = desc_nr;
955 d = &sb->disks[rdev2->desc_nr];
957 d->number = rdev2->desc_nr;
958 d->major = MAJOR(rdev2->bdev->bd_dev);
959 d->minor = MINOR(rdev2->bdev->bd_dev);
960 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
961 && !test_bit(Faulty, &rdev2->flags))
962 d->raid_disk = rdev2->raid_disk;
964 d->raid_disk = rdev2->desc_nr; /* compatibility */
965 if (test_bit(Faulty, &rdev2->flags))
966 d->state = (1<<MD_DISK_FAULTY);
967 else if (test_bit(In_sync, &rdev2->flags)) {
968 d->state = (1<<MD_DISK_ACTIVE);
969 d->state |= (1<<MD_DISK_SYNC);
977 if (test_bit(WriteMostly, &rdev2->flags))
978 d->state |= (1<<MD_DISK_WRITEMOSTLY);
980 /* now set the "removed" and "faulty" bits on any missing devices */
981 for (i=0 ; i < mddev->raid_disks ; i++) {
982 mdp_disk_t *d = &sb->disks[i];
983 if (d->state == 0 && d->number == 0) {
986 d->state = (1<<MD_DISK_REMOVED);
987 d->state |= (1<<MD_DISK_FAULTY);
991 sb->nr_disks = nr_disks;
992 sb->active_disks = active;
993 sb->working_disks = working;
994 sb->failed_disks = failed;
995 sb->spare_disks = spare;
997 sb->this_disk = sb->disks[rdev->desc_nr];
998 sb->sb_csum = calc_sb_csum(sb);
1002 * version 1 superblock
1005 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1009 unsigned long long newcsum;
1010 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1011 __le32 *isuper = (__le32*)sb;
1014 disk_csum = sb->sb_csum;
1017 for (i=0; size>=4; size -= 4 )
1018 newcsum += le32_to_cpu(*isuper++);
1021 newcsum += le16_to_cpu(*(__le16*) isuper);
1023 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1024 sb->sb_csum = disk_csum;
1025 return cpu_to_le32(csum);
1028 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1030 struct mdp_superblock_1 *sb;
1033 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1037 * Calculate the position of the superblock.
1038 * It is always aligned to a 4K boundary and
1039 * depeding on minor_version, it can be:
1040 * 0: At least 8K, but less than 12K, from end of device
1041 * 1: At start of device
1042 * 2: 4K from start of device.
1044 switch(minor_version) {
1046 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1048 sb_offset &= ~(sector_t)(4*2-1);
1049 /* convert from sectors to K */
1061 rdev->sb_offset = sb_offset;
1063 /* superblock is rarely larger than 1K, but it can be larger,
1064 * and it is safe to read 4k, so we do that
1066 ret = read_disk_sb(rdev, 4096);
1067 if (ret) return ret;
1070 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1072 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1073 sb->major_version != cpu_to_le32(1) ||
1074 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1075 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1076 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1079 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1080 printk("md: invalid superblock checksum on %s\n",
1081 bdevname(rdev->bdev,b));
1084 if (le64_to_cpu(sb->data_size) < 10) {
1085 printk("md: data_size too small on %s\n",
1086 bdevname(rdev->bdev,b));
1089 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1090 if (sb->level != cpu_to_le32(1) &&
1091 sb->level != cpu_to_le32(4) &&
1092 sb->level != cpu_to_le32(5) &&
1093 sb->level != cpu_to_le32(6) &&
1094 sb->level != cpu_to_le32(10)) {
1096 "md: bitmaps not supported for this level.\n");
1101 rdev->preferred_minor = 0xffff;
1102 rdev->data_offset = le64_to_cpu(sb->data_offset);
1103 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1105 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1106 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1107 if (rdev->sb_size & bmask)
1108 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1111 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
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;
1165 clear_bit(Faulty, &rdev->flags);
1166 clear_bit(In_sync, &rdev->flags);
1167 clear_bit(WriteMostly, &rdev->flags);
1168 clear_bit(BarriersNotsupp, &rdev->flags);
1170 if (mddev->raid_disks == 0) {
1171 mddev->major_version = 1;
1172 mddev->patch_version = 0;
1173 mddev->external = 0;
1174 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1175 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1176 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1177 mddev->level = le32_to_cpu(sb->level);
1178 mddev->clevel[0] = 0;
1179 mddev->layout = le32_to_cpu(sb->layout);
1180 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1181 mddev->size = le64_to_cpu(sb->size)/2;
1182 mddev->events = ev1;
1183 mddev->bitmap_offset = 0;
1184 mddev->default_bitmap_offset = 1024 >> 9;
1186 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1187 memcpy(mddev->uuid, sb->set_uuid, 16);
1189 mddev->max_disks = (4096-256)/2;
1191 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1192 mddev->bitmap_file == NULL )
1193 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1195 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1196 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1197 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1198 mddev->new_level = le32_to_cpu(sb->new_level);
1199 mddev->new_layout = le32_to_cpu(sb->new_layout);
1200 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1202 mddev->reshape_position = MaxSector;
1203 mddev->delta_disks = 0;
1204 mddev->new_level = mddev->level;
1205 mddev->new_layout = mddev->layout;
1206 mddev->new_chunk = mddev->chunk_size;
1209 } else if (mddev->pers == NULL) {
1210 /* Insist of good event counter while assembling */
1212 if (ev1 < mddev->events)
1214 } else if (mddev->bitmap) {
1215 /* If adding to array with a bitmap, then we can accept an
1216 * older device, but not too old.
1218 if (ev1 < mddev->bitmap->events_cleared)
1221 if (ev1 < mddev->events)
1222 /* just a hot-add of a new device, leave raid_disk at -1 */
1225 if (mddev->level != LEVEL_MULTIPATH) {
1227 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1229 case 0xffff: /* spare */
1231 case 0xfffe: /* faulty */
1232 set_bit(Faulty, &rdev->flags);
1235 if ((le32_to_cpu(sb->feature_map) &
1236 MD_FEATURE_RECOVERY_OFFSET))
1237 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1239 set_bit(In_sync, &rdev->flags);
1240 rdev->raid_disk = role;
1243 if (sb->devflags & WriteMostly1)
1244 set_bit(WriteMostly, &rdev->flags);
1245 } else /* MULTIPATH are always insync */
1246 set_bit(In_sync, &rdev->flags);
1251 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1253 struct mdp_superblock_1 *sb;
1254 struct list_head *tmp;
1257 /* make rdev->sb match mddev and rdev data. */
1259 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1261 sb->feature_map = 0;
1263 sb->recovery_offset = cpu_to_le64(0);
1264 memset(sb->pad1, 0, sizeof(sb->pad1));
1265 memset(sb->pad2, 0, sizeof(sb->pad2));
1266 memset(sb->pad3, 0, sizeof(sb->pad3));
1268 sb->utime = cpu_to_le64((__u64)mddev->utime);
1269 sb->events = cpu_to_le64(mddev->events);
1271 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1273 sb->resync_offset = cpu_to_le64(0);
1275 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1277 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1278 sb->size = cpu_to_le64(mddev->size<<1);
1280 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1281 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1282 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1285 if (rdev->raid_disk >= 0 &&
1286 !test_bit(In_sync, &rdev->flags) &&
1287 rdev->recovery_offset > 0) {
1288 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1289 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1292 if (mddev->reshape_position != MaxSector) {
1293 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1294 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1295 sb->new_layout = cpu_to_le32(mddev->new_layout);
1296 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1297 sb->new_level = cpu_to_le32(mddev->new_level);
1298 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1302 rdev_for_each(rdev2, tmp, mddev)
1303 if (rdev2->desc_nr+1 > max_dev)
1304 max_dev = rdev2->desc_nr+1;
1306 if (max_dev > le32_to_cpu(sb->max_dev))
1307 sb->max_dev = cpu_to_le32(max_dev);
1308 for (i=0; i<max_dev;i++)
1309 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1311 rdev_for_each(rdev2, tmp, mddev) {
1313 if (test_bit(Faulty, &rdev2->flags))
1314 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1315 else if (test_bit(In_sync, &rdev2->flags))
1316 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1317 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1318 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1320 sb->dev_roles[i] = cpu_to_le16(0xffff);
1323 sb->sb_csum = calc_sb_1_csum(sb);
1327 static struct super_type super_types[] = {
1330 .owner = THIS_MODULE,
1331 .load_super = super_90_load,
1332 .validate_super = super_90_validate,
1333 .sync_super = super_90_sync,
1337 .owner = THIS_MODULE,
1338 .load_super = super_1_load,
1339 .validate_super = super_1_validate,
1340 .sync_super = super_1_sync,
1344 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1346 struct list_head *tmp, *tmp2;
1347 mdk_rdev_t *rdev, *rdev2;
1349 rdev_for_each(rdev, tmp, mddev1)
1350 rdev_for_each(rdev2, tmp2, mddev2)
1351 if (rdev->bdev->bd_contains ==
1352 rdev2->bdev->bd_contains)
1358 static LIST_HEAD(pending_raid_disks);
1360 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1362 char b[BDEVNAME_SIZE];
1371 /* make sure rdev->size exceeds mddev->size */
1372 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1374 /* Cannot change size, so fail
1375 * If mddev->level <= 0, then we don't care
1376 * about aligning sizes (e.g. linear)
1378 if (mddev->level > 0)
1381 mddev->size = rdev->size;
1384 /* Verify rdev->desc_nr is unique.
1385 * If it is -1, assign a free number, else
1386 * check number is not in use
1388 if (rdev->desc_nr < 0) {
1390 if (mddev->pers) choice = mddev->raid_disks;
1391 while (find_rdev_nr(mddev, choice))
1393 rdev->desc_nr = choice;
1395 if (find_rdev_nr(mddev, rdev->desc_nr))
1398 bdevname(rdev->bdev,b);
1399 while ( (s=strchr(b, '/')) != NULL)
1402 rdev->mddev = mddev;
1403 printk(KERN_INFO "md: bind<%s>\n", b);
1405 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1408 if (rdev->bdev->bd_part)
1409 ko = &rdev->bdev->bd_part->dev.kobj;
1411 ko = &rdev->bdev->bd_disk->dev.kobj;
1412 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1413 kobject_del(&rdev->kobj);
1416 list_add(&rdev->same_set, &mddev->disks);
1417 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1421 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1426 static void md_delayed_delete(struct work_struct *ws)
1428 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1429 kobject_del(&rdev->kobj);
1430 kobject_put(&rdev->kobj);
1433 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1435 char b[BDEVNAME_SIZE];
1440 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1441 list_del_init(&rdev->same_set);
1442 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1444 sysfs_remove_link(&rdev->kobj, "block");
1446 /* We need to delay this, otherwise we can deadlock when
1447 * writing to 'remove' to "dev/state"
1449 INIT_WORK(&rdev->del_work, md_delayed_delete);
1450 kobject_get(&rdev->kobj);
1451 schedule_work(&rdev->del_work);
1455 * prevent the device from being mounted, repartitioned or
1456 * otherwise reused by a RAID array (or any other kernel
1457 * subsystem), by bd_claiming the device.
1459 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1462 struct block_device *bdev;
1463 char b[BDEVNAME_SIZE];
1465 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1467 printk(KERN_ERR "md: could not open %s.\n",
1468 __bdevname(dev, b));
1469 return PTR_ERR(bdev);
1471 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1473 printk(KERN_ERR "md: could not bd_claim %s.\n",
1479 set_bit(AllReserved, &rdev->flags);
1484 static void unlock_rdev(mdk_rdev_t *rdev)
1486 struct block_device *bdev = rdev->bdev;
1494 void md_autodetect_dev(dev_t dev);
1496 static void export_rdev(mdk_rdev_t * rdev)
1498 char b[BDEVNAME_SIZE];
1499 printk(KERN_INFO "md: export_rdev(%s)\n",
1500 bdevname(rdev->bdev,b));
1504 list_del_init(&rdev->same_set);
1506 if (test_bit(AutoDetected, &rdev->flags))
1507 md_autodetect_dev(rdev->bdev->bd_dev);
1510 kobject_put(&rdev->kobj);
1513 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1515 unbind_rdev_from_array(rdev);
1519 static void export_array(mddev_t *mddev)
1521 struct list_head *tmp;
1524 rdev_for_each(rdev, tmp, mddev) {
1529 kick_rdev_from_array(rdev);
1531 if (!list_empty(&mddev->disks))
1533 mddev->raid_disks = 0;
1534 mddev->major_version = 0;
1537 static void print_desc(mdp_disk_t *desc)
1539 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1540 desc->major,desc->minor,desc->raid_disk,desc->state);
1543 static void print_sb(mdp_super_t *sb)
1548 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1549 sb->major_version, sb->minor_version, sb->patch_version,
1550 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1552 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1553 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1554 sb->md_minor, sb->layout, sb->chunk_size);
1555 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1556 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1557 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1558 sb->failed_disks, sb->spare_disks,
1559 sb->sb_csum, (unsigned long)sb->events_lo);
1562 for (i = 0; i < MD_SB_DISKS; i++) {
1565 desc = sb->disks + i;
1566 if (desc->number || desc->major || desc->minor ||
1567 desc->raid_disk || (desc->state && (desc->state != 4))) {
1568 printk(" D %2d: ", i);
1572 printk(KERN_INFO "md: THIS: ");
1573 print_desc(&sb->this_disk);
1577 static void print_rdev(mdk_rdev_t *rdev)
1579 char b[BDEVNAME_SIZE];
1580 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1581 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1582 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1584 if (rdev->sb_loaded) {
1585 printk(KERN_INFO "md: rdev superblock:\n");
1586 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1588 printk(KERN_INFO "md: no rdev superblock!\n");
1591 static void md_print_devices(void)
1593 struct list_head *tmp, *tmp2;
1596 char b[BDEVNAME_SIZE];
1599 printk("md: **********************************\n");
1600 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1601 printk("md: **********************************\n");
1602 for_each_mddev(mddev, tmp) {
1605 bitmap_print_sb(mddev->bitmap);
1607 printk("%s: ", mdname(mddev));
1608 rdev_for_each(rdev, tmp2, mddev)
1609 printk("<%s>", bdevname(rdev->bdev,b));
1612 rdev_for_each(rdev, tmp2, mddev)
1615 printk("md: **********************************\n");
1620 static void sync_sbs(mddev_t * mddev, int nospares)
1622 /* Update each superblock (in-memory image), but
1623 * if we are allowed to, skip spares which already
1624 * have the right event counter, or have one earlier
1625 * (which would mean they aren't being marked as dirty
1626 * with the rest of the array)
1629 struct list_head *tmp;
1631 rdev_for_each(rdev, tmp, mddev) {
1632 if (rdev->sb_events == mddev->events ||
1634 rdev->raid_disk < 0 &&
1635 (rdev->sb_events&1)==0 &&
1636 rdev->sb_events+1 == mddev->events)) {
1637 /* Don't update this superblock */
1638 rdev->sb_loaded = 2;
1640 super_types[mddev->major_version].
1641 sync_super(mddev, rdev);
1642 rdev->sb_loaded = 1;
1647 static void md_update_sb(mddev_t * mddev, int force_change)
1649 struct list_head *tmp;
1655 spin_lock_irq(&mddev->write_lock);
1657 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1658 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1660 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1661 /* just a clean<-> dirty transition, possibly leave spares alone,
1662 * though if events isn't the right even/odd, we will have to do
1668 if (mddev->degraded)
1669 /* If the array is degraded, then skipping spares is both
1670 * dangerous and fairly pointless.
1671 * Dangerous because a device that was removed from the array
1672 * might have a event_count that still looks up-to-date,
1673 * so it can be re-added without a resync.
1674 * Pointless because if there are any spares to skip,
1675 * then a recovery will happen and soon that array won't
1676 * be degraded any more and the spare can go back to sleep then.
1680 sync_req = mddev->in_sync;
1681 mddev->utime = get_seconds();
1683 /* If this is just a dirty<->clean transition, and the array is clean
1684 * and 'events' is odd, we can roll back to the previous clean state */
1686 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1687 && (mddev->events & 1)
1688 && mddev->events != 1)
1691 /* otherwise we have to go forward and ... */
1693 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1694 /* .. if the array isn't clean, insist on an odd 'events' */
1695 if ((mddev->events&1)==0) {
1700 /* otherwise insist on an even 'events' (for clean states) */
1701 if ((mddev->events&1)) {
1708 if (!mddev->events) {
1710 * oops, this 64-bit counter should never wrap.
1711 * Either we are in around ~1 trillion A.C., assuming
1712 * 1 reboot per second, or we have a bug:
1719 * do not write anything to disk if using
1720 * nonpersistent superblocks
1722 if (!mddev->persistent) {
1723 if (!mddev->external)
1724 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1726 spin_unlock_irq(&mddev->write_lock);
1727 wake_up(&mddev->sb_wait);
1730 sync_sbs(mddev, nospares);
1731 spin_unlock_irq(&mddev->write_lock);
1734 "md: updating %s RAID superblock on device (in sync %d)\n",
1735 mdname(mddev),mddev->in_sync);
1737 bitmap_update_sb(mddev->bitmap);
1738 rdev_for_each(rdev, tmp, mddev) {
1739 char b[BDEVNAME_SIZE];
1740 dprintk(KERN_INFO "md: ");
1741 if (rdev->sb_loaded != 1)
1742 continue; /* no noise on spare devices */
1743 if (test_bit(Faulty, &rdev->flags))
1744 dprintk("(skipping faulty ");
1746 dprintk("%s ", bdevname(rdev->bdev,b));
1747 if (!test_bit(Faulty, &rdev->flags)) {
1748 md_super_write(mddev,rdev,
1749 rdev->sb_offset<<1, rdev->sb_size,
1751 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1752 bdevname(rdev->bdev,b),
1753 (unsigned long long)rdev->sb_offset);
1754 rdev->sb_events = mddev->events;
1758 if (mddev->level == LEVEL_MULTIPATH)
1759 /* only need to write one superblock... */
1762 md_super_wait(mddev);
1763 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1765 spin_lock_irq(&mddev->write_lock);
1766 if (mddev->in_sync != sync_req ||
1767 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1768 /* have to write it out again */
1769 spin_unlock_irq(&mddev->write_lock);
1772 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1773 spin_unlock_irq(&mddev->write_lock);
1774 wake_up(&mddev->sb_wait);
1778 /* words written to sysfs files may, or my not, be \n terminated.
1779 * We want to accept with case. For this we use cmd_match.
1781 static int cmd_match(const char *cmd, const char *str)
1783 /* See if cmd, written into a sysfs file, matches
1784 * str. They must either be the same, or cmd can
1785 * have a trailing newline
1787 while (*cmd && *str && *cmd == *str) {
1798 struct rdev_sysfs_entry {
1799 struct attribute attr;
1800 ssize_t (*show)(mdk_rdev_t *, char *);
1801 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1805 state_show(mdk_rdev_t *rdev, char *page)
1810 if (test_bit(Faulty, &rdev->flags)) {
1811 len+= sprintf(page+len, "%sfaulty",sep);
1814 if (test_bit(In_sync, &rdev->flags)) {
1815 len += sprintf(page+len, "%sin_sync",sep);
1818 if (test_bit(WriteMostly, &rdev->flags)) {
1819 len += sprintf(page+len, "%swrite_mostly",sep);
1822 if (!test_bit(Faulty, &rdev->flags) &&
1823 !test_bit(In_sync, &rdev->flags)) {
1824 len += sprintf(page+len, "%sspare", sep);
1827 return len+sprintf(page+len, "\n");
1831 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1834 * faulty - simulates and error
1835 * remove - disconnects the device
1836 * writemostly - sets write_mostly
1837 * -writemostly - clears write_mostly
1840 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1841 md_error(rdev->mddev, rdev);
1843 } else if (cmd_match(buf, "remove")) {
1844 if (rdev->raid_disk >= 0)
1847 mddev_t *mddev = rdev->mddev;
1848 kick_rdev_from_array(rdev);
1850 md_update_sb(mddev, 1);
1851 md_new_event(mddev);
1854 } else if (cmd_match(buf, "writemostly")) {
1855 set_bit(WriteMostly, &rdev->flags);
1857 } else if (cmd_match(buf, "-writemostly")) {
1858 clear_bit(WriteMostly, &rdev->flags);
1861 return err ? err : len;
1863 static struct rdev_sysfs_entry rdev_state =
1864 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1867 super_show(mdk_rdev_t *rdev, char *page)
1869 if (rdev->sb_loaded && rdev->sb_size) {
1870 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1871 return rdev->sb_size;
1875 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1878 errors_show(mdk_rdev_t *rdev, char *page)
1880 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1884 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1887 unsigned long n = simple_strtoul(buf, &e, 10);
1888 if (*buf && (*e == 0 || *e == '\n')) {
1889 atomic_set(&rdev->corrected_errors, n);
1894 static struct rdev_sysfs_entry rdev_errors =
1895 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1898 slot_show(mdk_rdev_t *rdev, char *page)
1900 if (rdev->raid_disk < 0)
1901 return sprintf(page, "none\n");
1903 return sprintf(page, "%d\n", rdev->raid_disk);
1907 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1912 int slot = simple_strtoul(buf, &e, 10);
1913 if (strncmp(buf, "none", 4)==0)
1915 else if (e==buf || (*e && *e!= '\n'))
1917 if (rdev->mddev->pers) {
1918 /* Setting 'slot' on an active array requires also
1919 * updating the 'rd%d' link, and communicating
1920 * with the personality with ->hot_*_disk.
1921 * For now we only support removing
1922 * failed/spare devices. This normally happens automatically,
1923 * but not when the metadata is externally managed.
1927 if (rdev->raid_disk == -1)
1929 /* personality does all needed checks */
1930 if (rdev->mddev->pers->hot_add_disk == NULL)
1932 err = rdev->mddev->pers->
1933 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1936 sprintf(nm, "rd%d", rdev->raid_disk);
1937 sysfs_remove_link(&rdev->mddev->kobj, nm);
1938 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1939 md_wakeup_thread(rdev->mddev->thread);
1941 if (slot >= rdev->mddev->raid_disks)
1943 rdev->raid_disk = slot;
1944 /* assume it is working */
1945 clear_bit(Faulty, &rdev->flags);
1946 clear_bit(WriteMostly, &rdev->flags);
1947 set_bit(In_sync, &rdev->flags);
1953 static struct rdev_sysfs_entry rdev_slot =
1954 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1957 offset_show(mdk_rdev_t *rdev, char *page)
1959 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1963 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1966 unsigned long long offset = simple_strtoull(buf, &e, 10);
1967 if (e==buf || (*e && *e != '\n'))
1969 if (rdev->mddev->pers)
1971 if (rdev->size && rdev->mddev->external)
1972 /* Must set offset before size, so overlap checks
1975 rdev->data_offset = offset;
1979 static struct rdev_sysfs_entry rdev_offset =
1980 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1983 rdev_size_show(mdk_rdev_t *rdev, char *page)
1985 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1988 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1990 /* check if two start/length pairs overlap */
1999 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2002 unsigned long long size = simple_strtoull(buf, &e, 10);
2003 unsigned long long oldsize = rdev->size;
2004 if (e==buf || (*e && *e != '\n'))
2006 if (rdev->mddev->pers)
2009 if (size > oldsize && rdev->mddev->external) {
2010 /* need to check that all other rdevs with the same ->bdev
2011 * do not overlap. We need to unlock the mddev to avoid
2012 * a deadlock. We have already changed rdev->size, and if
2013 * we have to change it back, we will have the lock again.
2017 struct list_head *tmp, *tmp2;
2019 mddev_unlock(rdev->mddev);
2020 for_each_mddev(mddev, tmp) {
2024 rdev_for_each(rdev2, tmp2, mddev)
2025 if (test_bit(AllReserved, &rdev2->flags) ||
2026 (rdev->bdev == rdev2->bdev &&
2028 overlaps(rdev->data_offset, rdev->size,
2029 rdev2->data_offset, rdev2->size))) {
2033 mddev_unlock(mddev);
2039 mddev_lock(rdev->mddev);
2041 /* Someone else could have slipped in a size
2042 * change here, but doing so is just silly.
2043 * We put oldsize back because we *know* it is
2044 * safe, and trust userspace not to race with
2047 rdev->size = oldsize;
2051 if (size < rdev->mddev->size || rdev->mddev->size == 0)
2052 rdev->mddev->size = size;
2056 static struct rdev_sysfs_entry rdev_size =
2057 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2059 static struct attribute *rdev_default_attrs[] = {
2069 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2071 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2072 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2076 return entry->show(rdev, page);
2080 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2081 const char *page, size_t length)
2083 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2084 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2089 if (!capable(CAP_SYS_ADMIN))
2091 rv = mddev_lock(rdev->mddev);
2093 rv = entry->store(rdev, page, length);
2094 mddev_unlock(rdev->mddev);
2099 static void rdev_free(struct kobject *ko)
2101 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2104 static struct sysfs_ops rdev_sysfs_ops = {
2105 .show = rdev_attr_show,
2106 .store = rdev_attr_store,
2108 static struct kobj_type rdev_ktype = {
2109 .release = rdev_free,
2110 .sysfs_ops = &rdev_sysfs_ops,
2111 .default_attrs = rdev_default_attrs,
2115 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2117 * mark the device faulty if:
2119 * - the device is nonexistent (zero size)
2120 * - the device has no valid superblock
2122 * a faulty rdev _never_ has rdev->sb set.
2124 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2126 char b[BDEVNAME_SIZE];
2131 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2133 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2134 return ERR_PTR(-ENOMEM);
2137 if ((err = alloc_disk_sb(rdev)))
2140 err = lock_rdev(rdev, newdev, super_format == -2);
2144 kobject_init(&rdev->kobj, &rdev_ktype);
2147 rdev->saved_raid_disk = -1;
2148 rdev->raid_disk = -1;
2150 rdev->data_offset = 0;
2151 rdev->sb_events = 0;
2152 atomic_set(&rdev->nr_pending, 0);
2153 atomic_set(&rdev->read_errors, 0);
2154 atomic_set(&rdev->corrected_errors, 0);
2156 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2159 "md: %s has zero or unknown size, marking faulty!\n",
2160 bdevname(rdev->bdev,b));
2165 if (super_format >= 0) {
2166 err = super_types[super_format].
2167 load_super(rdev, NULL, super_minor);
2168 if (err == -EINVAL) {
2170 "md: %s does not have a valid v%d.%d "
2171 "superblock, not importing!\n",
2172 bdevname(rdev->bdev,b),
2173 super_format, super_minor);
2178 "md: could not read %s's sb, not importing!\n",
2179 bdevname(rdev->bdev,b));
2183 INIT_LIST_HEAD(&rdev->same_set);
2188 if (rdev->sb_page) {
2194 return ERR_PTR(err);
2198 * Check a full RAID array for plausibility
2202 static void analyze_sbs(mddev_t * mddev)
2205 struct list_head *tmp;
2206 mdk_rdev_t *rdev, *freshest;
2207 char b[BDEVNAME_SIZE];
2210 rdev_for_each(rdev, tmp, mddev)
2211 switch (super_types[mddev->major_version].
2212 load_super(rdev, freshest, mddev->minor_version)) {
2220 "md: fatal superblock inconsistency in %s"
2221 " -- removing from array\n",
2222 bdevname(rdev->bdev,b));
2223 kick_rdev_from_array(rdev);
2227 super_types[mddev->major_version].
2228 validate_super(mddev, freshest);
2231 rdev_for_each(rdev, tmp, mddev) {
2232 if (rdev != freshest)
2233 if (super_types[mddev->major_version].
2234 validate_super(mddev, rdev)) {
2235 printk(KERN_WARNING "md: kicking non-fresh %s"
2237 bdevname(rdev->bdev,b));
2238 kick_rdev_from_array(rdev);
2241 if (mddev->level == LEVEL_MULTIPATH) {
2242 rdev->desc_nr = i++;
2243 rdev->raid_disk = rdev->desc_nr;
2244 set_bit(In_sync, &rdev->flags);
2245 } else if (rdev->raid_disk >= mddev->raid_disks) {
2246 rdev->raid_disk = -1;
2247 clear_bit(In_sync, &rdev->flags);
2253 if (mddev->recovery_cp != MaxSector &&
2255 printk(KERN_ERR "md: %s: raid array is not clean"
2256 " -- starting background reconstruction\n",
2262 safe_delay_show(mddev_t *mddev, char *page)
2264 int msec = (mddev->safemode_delay*1000)/HZ;
2265 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2268 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2276 /* remove a period, and count digits after it */
2277 if (len >= sizeof(buf))
2279 strlcpy(buf, cbuf, len);
2281 for (i=0; i<len; i++) {
2283 if (isdigit(buf[i])) {
2288 } else if (buf[i] == '.') {
2293 msec = simple_strtoul(buf, &e, 10);
2294 if (e == buf || (*e && *e != '\n'))
2296 msec = (msec * 1000) / scale;
2298 mddev->safemode_delay = 0;
2300 mddev->safemode_delay = (msec*HZ)/1000;
2301 if (mddev->safemode_delay == 0)
2302 mddev->safemode_delay = 1;
2306 static struct md_sysfs_entry md_safe_delay =
2307 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2310 level_show(mddev_t *mddev, char *page)
2312 struct mdk_personality *p = mddev->pers;
2314 return sprintf(page, "%s\n", p->name);
2315 else if (mddev->clevel[0])
2316 return sprintf(page, "%s\n", mddev->clevel);
2317 else if (mddev->level != LEVEL_NONE)
2318 return sprintf(page, "%d\n", mddev->level);
2324 level_store(mddev_t *mddev, const char *buf, size_t len)
2331 if (len >= sizeof(mddev->clevel))
2333 strncpy(mddev->clevel, buf, len);
2334 if (mddev->clevel[len-1] == '\n')
2336 mddev->clevel[len] = 0;
2337 mddev->level = LEVEL_NONE;
2341 static struct md_sysfs_entry md_level =
2342 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2346 layout_show(mddev_t *mddev, char *page)
2348 /* just a number, not meaningful for all levels */
2349 if (mddev->reshape_position != MaxSector &&
2350 mddev->layout != mddev->new_layout)
2351 return sprintf(page, "%d (%d)\n",
2352 mddev->new_layout, mddev->layout);
2353 return sprintf(page, "%d\n", mddev->layout);
2357 layout_store(mddev_t *mddev, const char *buf, size_t len)
2360 unsigned long n = simple_strtoul(buf, &e, 10);
2362 if (!*buf || (*e && *e != '\n'))
2367 if (mddev->reshape_position != MaxSector)
2368 mddev->new_layout = n;
2373 static struct md_sysfs_entry md_layout =
2374 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2378 raid_disks_show(mddev_t *mddev, char *page)
2380 if (mddev->raid_disks == 0)
2382 if (mddev->reshape_position != MaxSector &&
2383 mddev->delta_disks != 0)
2384 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2385 mddev->raid_disks - mddev->delta_disks);
2386 return sprintf(page, "%d\n", mddev->raid_disks);
2389 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2392 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2396 unsigned long n = simple_strtoul(buf, &e, 10);
2398 if (!*buf || (*e && *e != '\n'))
2402 rv = update_raid_disks(mddev, n);
2403 else if (mddev->reshape_position != MaxSector) {
2404 int olddisks = mddev->raid_disks - mddev->delta_disks;
2405 mddev->delta_disks = n - olddisks;
2406 mddev->raid_disks = n;
2408 mddev->raid_disks = n;
2409 return rv ? rv : len;
2411 static struct md_sysfs_entry md_raid_disks =
2412 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2415 chunk_size_show(mddev_t *mddev, char *page)
2417 if (mddev->reshape_position != MaxSector &&
2418 mddev->chunk_size != mddev->new_chunk)
2419 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2421 return sprintf(page, "%d\n", mddev->chunk_size);
2425 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2427 /* can only set chunk_size if array is not yet active */
2429 unsigned long n = simple_strtoul(buf, &e, 10);
2431 if (!*buf || (*e && *e != '\n'))
2436 else if (mddev->reshape_position != MaxSector)
2437 mddev->new_chunk = n;
2439 mddev->chunk_size = n;
2442 static struct md_sysfs_entry md_chunk_size =
2443 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2446 resync_start_show(mddev_t *mddev, char *page)
2448 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2452 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2454 /* can only set chunk_size if array is not yet active */
2456 unsigned long long n = simple_strtoull(buf, &e, 10);
2460 if (!*buf || (*e && *e != '\n'))
2463 mddev->recovery_cp = n;
2466 static struct md_sysfs_entry md_resync_start =
2467 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2470 * The array state can be:
2473 * No devices, no size, no level
2474 * Equivalent to STOP_ARRAY ioctl
2476 * May have some settings, but array is not active
2477 * all IO results in error
2478 * When written, doesn't tear down array, but just stops it
2479 * suspended (not supported yet)
2480 * All IO requests will block. The array can be reconfigured.
2481 * Writing this, if accepted, will block until array is quiessent
2483 * no resync can happen. no superblocks get written.
2484 * write requests fail
2486 * like readonly, but behaves like 'clean' on a write request.
2488 * clean - no pending writes, but otherwise active.
2489 * When written to inactive array, starts without resync
2490 * If a write request arrives then
2491 * if metadata is known, mark 'dirty' and switch to 'active'.
2492 * if not known, block and switch to write-pending
2493 * If written to an active array that has pending writes, then fails.
2495 * fully active: IO and resync can be happening.
2496 * When written to inactive array, starts with resync
2499 * clean, but writes are blocked waiting for 'active' to be written.
2502 * like active, but no writes have been seen for a while (100msec).
2505 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2506 write_pending, active_idle, bad_word};
2507 static char *array_states[] = {
2508 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2509 "write-pending", "active-idle", NULL };
2511 static int match_word(const char *word, char **list)
2514 for (n=0; list[n]; n++)
2515 if (cmd_match(word, list[n]))
2521 array_state_show(mddev_t *mddev, char *page)
2523 enum array_state st = inactive;
2536 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2538 else if (mddev->safemode)
2544 if (list_empty(&mddev->disks) &&
2545 mddev->raid_disks == 0 &&
2551 return sprintf(page, "%s\n", array_states[st]);
2554 static int do_md_stop(mddev_t * mddev, int ro);
2555 static int do_md_run(mddev_t * mddev);
2556 static int restart_array(mddev_t *mddev);
2559 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2562 enum array_state st = match_word(buf, array_states);
2567 /* stopping an active array */
2568 if (atomic_read(&mddev->active) > 1)
2570 err = do_md_stop(mddev, 0);
2573 /* stopping an active array */
2575 if (atomic_read(&mddev->active) > 1)
2577 err = do_md_stop(mddev, 2);
2579 err = 0; /* already inactive */
2582 break; /* not supported yet */
2585 err = do_md_stop(mddev, 1);
2588 err = do_md_run(mddev);
2592 /* stopping an active array */
2594 err = do_md_stop(mddev, 1);
2596 mddev->ro = 2; /* FIXME mark devices writable */
2599 err = do_md_run(mddev);
2604 restart_array(mddev);
2605 spin_lock_irq(&mddev->write_lock);
2606 if (atomic_read(&mddev->writes_pending) == 0) {
2607 if (mddev->in_sync == 0) {
2609 if (mddev->persistent)
2610 set_bit(MD_CHANGE_CLEAN,
2616 spin_unlock_irq(&mddev->write_lock);
2619 mddev->recovery_cp = MaxSector;
2620 err = do_md_run(mddev);
2625 restart_array(mddev);
2626 if (mddev->external)
2627 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2628 wake_up(&mddev->sb_wait);
2632 err = do_md_run(mddev);
2637 /* these cannot be set */
2645 static struct md_sysfs_entry md_array_state =
2646 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2649 null_show(mddev_t *mddev, char *page)
2655 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2657 /* buf must be %d:%d\n? giving major and minor numbers */
2658 /* The new device is added to the array.
2659 * If the array has a persistent superblock, we read the
2660 * superblock to initialise info and check validity.
2661 * Otherwise, only checking done is that in bind_rdev_to_array,
2662 * which mainly checks size.
2665 int major = simple_strtoul(buf, &e, 10);
2671 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2673 minor = simple_strtoul(e+1, &e, 10);
2674 if (*e && *e != '\n')
2676 dev = MKDEV(major, minor);
2677 if (major != MAJOR(dev) ||
2678 minor != MINOR(dev))
2682 if (mddev->persistent) {
2683 rdev = md_import_device(dev, mddev->major_version,
2684 mddev->minor_version);
2685 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2686 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2687 mdk_rdev_t, same_set);
2688 err = super_types[mddev->major_version]
2689 .load_super(rdev, rdev0, mddev->minor_version);
2693 } else if (mddev->external)
2694 rdev = md_import_device(dev, -2, -1);
2696 rdev = md_import_device(dev, -1, -1);
2699 return PTR_ERR(rdev);
2700 err = bind_rdev_to_array(rdev, mddev);
2704 return err ? err : len;
2707 static struct md_sysfs_entry md_new_device =
2708 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2711 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2714 unsigned long chunk, end_chunk;
2718 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2720 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2721 if (buf == end) break;
2722 if (*end == '-') { /* range */
2724 end_chunk = simple_strtoul(buf, &end, 0);
2725 if (buf == end) break;
2727 if (*end && !isspace(*end)) break;
2728 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2730 while (isspace(*buf)) buf++;
2732 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2737 static struct md_sysfs_entry md_bitmap =
2738 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2741 size_show(mddev_t *mddev, char *page)
2743 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2746 static int update_size(mddev_t *mddev, unsigned long size);
2749 size_store(mddev_t *mddev, const char *buf, size_t len)
2751 /* If array is inactive, we can reduce the component size, but
2752 * not increase it (except from 0).
2753 * If array is active, we can try an on-line resize
2757 unsigned long long size = simple_strtoull(buf, &e, 10);
2758 if (!*buf || *buf == '\n' ||
2763 err = update_size(mddev, size);
2764 md_update_sb(mddev, 1);
2766 if (mddev->size == 0 ||
2772 return err ? err : len;
2775 static struct md_sysfs_entry md_size =
2776 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2781 * 'none' for arrays with no metadata (good luck...)
2782 * 'external' for arrays with externally managed metadata,
2783 * or N.M for internally known formats
2786 metadata_show(mddev_t *mddev, char *page)
2788 if (mddev->persistent)
2789 return sprintf(page, "%d.%d\n",
2790 mddev->major_version, mddev->minor_version);
2791 else if (mddev->external)
2792 return sprintf(page, "external:%s\n", mddev->metadata_type);
2794 return sprintf(page, "none\n");
2798 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2802 if (!list_empty(&mddev->disks))
2805 if (cmd_match(buf, "none")) {
2806 mddev->persistent = 0;
2807 mddev->external = 0;
2808 mddev->major_version = 0;
2809 mddev->minor_version = 90;
2812 if (strncmp(buf, "external:", 9) == 0) {
2813 size_t namelen = len-9;
2814 if (namelen >= sizeof(mddev->metadata_type))
2815 namelen = sizeof(mddev->metadata_type)-1;
2816 strncpy(mddev->metadata_type, buf+9, namelen);
2817 mddev->metadata_type[namelen] = 0;
2818 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2819 mddev->metadata_type[--namelen] = 0;
2820 mddev->persistent = 0;
2821 mddev->external = 1;
2822 mddev->major_version = 0;
2823 mddev->minor_version = 90;
2826 major = simple_strtoul(buf, &e, 10);
2827 if (e==buf || *e != '.')
2830 minor = simple_strtoul(buf, &e, 10);
2831 if (e==buf || (*e && *e != '\n') )
2833 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2835 mddev->major_version = major;
2836 mddev->minor_version = minor;
2837 mddev->persistent = 1;
2838 mddev->external = 0;
2842 static struct md_sysfs_entry md_metadata =
2843 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2846 action_show(mddev_t *mddev, char *page)
2848 char *type = "idle";
2849 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2850 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2851 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2853 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2854 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2856 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2863 return sprintf(page, "%s\n", type);
2867 action_store(mddev_t *mddev, const char *page, size_t len)
2869 if (!mddev->pers || !mddev->pers->sync_request)
2872 if (cmd_match(page, "idle")) {
2873 if (mddev->sync_thread) {
2874 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2875 md_unregister_thread(mddev->sync_thread);
2876 mddev->sync_thread = NULL;
2877 mddev->recovery = 0;
2879 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2880 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2882 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2883 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2884 else if (cmd_match(page, "reshape")) {
2886 if (mddev->pers->start_reshape == NULL)
2888 err = mddev->pers->start_reshape(mddev);
2892 if (cmd_match(page, "check"))
2893 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2894 else if (!cmd_match(page, "repair"))
2896 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2897 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2899 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2900 md_wakeup_thread(mddev->thread);
2905 mismatch_cnt_show(mddev_t *mddev, char *page)
2907 return sprintf(page, "%llu\n",
2908 (unsigned long long) mddev->resync_mismatches);
2911 static struct md_sysfs_entry md_scan_mode =
2912 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2915 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2918 sync_min_show(mddev_t *mddev, char *page)
2920 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2921 mddev->sync_speed_min ? "local": "system");
2925 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2929 if (strncmp(buf, "system", 6)==0) {
2930 mddev->sync_speed_min = 0;
2933 min = simple_strtoul(buf, &e, 10);
2934 if (buf == e || (*e && *e != '\n') || min <= 0)
2936 mddev->sync_speed_min = min;
2940 static struct md_sysfs_entry md_sync_min =
2941 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2944 sync_max_show(mddev_t *mddev, char *page)
2946 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2947 mddev->sync_speed_max ? "local": "system");
2951 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2955 if (strncmp(buf, "system", 6)==0) {
2956 mddev->sync_speed_max = 0;
2959 max = simple_strtoul(buf, &e, 10);
2960 if (buf == e || (*e && *e != '\n') || max <= 0)
2962 mddev->sync_speed_max = max;
2966 static struct md_sysfs_entry md_sync_max =
2967 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2970 degraded_show(mddev_t *mddev, char *page)
2972 return sprintf(page, "%d\n", mddev->degraded);
2974 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2977 sync_speed_show(mddev_t *mddev, char *page)
2979 unsigned long resync, dt, db;
2980 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2981 dt = ((jiffies - mddev->resync_mark) / HZ);
2983 db = resync - (mddev->resync_mark_cnt);
2984 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2987 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2990 sync_completed_show(mddev_t *mddev, char *page)
2992 unsigned long max_blocks, resync;
2994 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2995 max_blocks = mddev->resync_max_sectors;
2997 max_blocks = mddev->size << 1;
2999 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3000 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3003 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3006 max_sync_show(mddev_t *mddev, char *page)
3008 if (mddev->resync_max == MaxSector)
3009 return sprintf(page, "max\n");
3011 return sprintf(page, "%llu\n",
3012 (unsigned long long)mddev->resync_max);
3015 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3017 if (strncmp(buf, "max", 3) == 0)
3018 mddev->resync_max = MaxSector;
3021 unsigned long long max = simple_strtoull(buf, &ep, 10);
3022 if (ep == buf || (*ep != 0 && *ep != '\n'))
3024 if (max < mddev->resync_max &&
3025 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3028 /* Must be a multiple of chunk_size */
3029 if (mddev->chunk_size) {
3030 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3033 mddev->resync_max = max;
3035 wake_up(&mddev->recovery_wait);
3039 static struct md_sysfs_entry md_max_sync =
3040 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3043 suspend_lo_show(mddev_t *mddev, char *page)
3045 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3049 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3052 unsigned long long new = simple_strtoull(buf, &e, 10);
3054 if (mddev->pers->quiesce == NULL)
3056 if (buf == e || (*e && *e != '\n'))
3058 if (new >= mddev->suspend_hi ||
3059 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3060 mddev->suspend_lo = new;
3061 mddev->pers->quiesce(mddev, 2);
3066 static struct md_sysfs_entry md_suspend_lo =
3067 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3071 suspend_hi_show(mddev_t *mddev, char *page)
3073 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3077 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3080 unsigned long long new = simple_strtoull(buf, &e, 10);
3082 if (mddev->pers->quiesce == NULL)
3084 if (buf == e || (*e && *e != '\n'))
3086 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3087 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3088 mddev->suspend_hi = new;
3089 mddev->pers->quiesce(mddev, 1);
3090 mddev->pers->quiesce(mddev, 0);
3095 static struct md_sysfs_entry md_suspend_hi =
3096 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3099 reshape_position_show(mddev_t *mddev, char *page)
3101 if (mddev->reshape_position != MaxSector)
3102 return sprintf(page, "%llu\n",
3103 (unsigned long long)mddev->reshape_position);
3104 strcpy(page, "none\n");
3109 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3112 unsigned long long new = simple_strtoull(buf, &e, 10);
3115 if (buf == e || (*e && *e != '\n'))
3117 mddev->reshape_position = new;
3118 mddev->delta_disks = 0;
3119 mddev->new_level = mddev->level;
3120 mddev->new_layout = mddev->layout;
3121 mddev->new_chunk = mddev->chunk_size;
3125 static struct md_sysfs_entry md_reshape_position =
3126 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3127 reshape_position_store);
3130 static struct attribute *md_default_attrs[] = {
3133 &md_raid_disks.attr,
3134 &md_chunk_size.attr,
3136 &md_resync_start.attr,
3138 &md_new_device.attr,
3139 &md_safe_delay.attr,
3140 &md_array_state.attr,
3141 &md_reshape_position.attr,
3145 static struct attribute *md_redundancy_attrs[] = {
3147 &md_mismatches.attr,
3150 &md_sync_speed.attr,
3151 &md_sync_completed.attr,
3153 &md_suspend_lo.attr,
3154 &md_suspend_hi.attr,
3159 static struct attribute_group md_redundancy_group = {
3161 .attrs = md_redundancy_attrs,
3166 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3168 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3169 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3174 rv = mddev_lock(mddev);
3176 rv = entry->show(mddev, page);
3177 mddev_unlock(mddev);
3183 md_attr_store(struct kobject *kobj, struct attribute *attr,
3184 const char *page, size_t length)
3186 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3187 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3192 if (!capable(CAP_SYS_ADMIN))
3194 rv = mddev_lock(mddev);
3196 rv = entry->store(mddev, page, length);
3197 mddev_unlock(mddev);
3202 static void md_free(struct kobject *ko)
3204 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3208 static struct sysfs_ops md_sysfs_ops = {
3209 .show = md_attr_show,
3210 .store = md_attr_store,
3212 static struct kobj_type md_ktype = {
3214 .sysfs_ops = &md_sysfs_ops,
3215 .default_attrs = md_default_attrs,
3220 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3222 static DEFINE_MUTEX(disks_mutex);
3223 mddev_t *mddev = mddev_find(dev);
3224 struct gendisk *disk;
3225 int partitioned = (MAJOR(dev) != MD_MAJOR);
3226 int shift = partitioned ? MdpMinorShift : 0;
3227 int unit = MINOR(dev) >> shift;
3233 mutex_lock(&disks_mutex);
3234 if (mddev->gendisk) {
3235 mutex_unlock(&disks_mutex);
3239 disk = alloc_disk(1 << shift);
3241 mutex_unlock(&disks_mutex);
3245 disk->major = MAJOR(dev);
3246 disk->first_minor = unit << shift;
3248 sprintf(disk->disk_name, "md_d%d", unit);
3250 sprintf(disk->disk_name, "md%d", unit);
3251 disk->fops = &md_fops;
3252 disk->private_data = mddev;
3253 disk->queue = mddev->queue;
3255 mddev->gendisk = disk;
3256 mutex_unlock(&disks_mutex);
3257 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3260 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3263 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3267 static void md_safemode_timeout(unsigned long data)
3269 mddev_t *mddev = (mddev_t *) data;
3271 mddev->safemode = 1;
3272 md_wakeup_thread(mddev->thread);
3275 static int start_dirty_degraded;
3277 static int do_md_run(mddev_t * mddev)
3281 struct list_head *tmp;
3283 struct gendisk *disk;
3284 struct mdk_personality *pers;
3285 char b[BDEVNAME_SIZE];
3287 if (list_empty(&mddev->disks))
3288 /* cannot run an array with no devices.. */
3295 * Analyze all RAID superblock(s)
3297 if (!mddev->raid_disks) {
3298 if (!mddev->persistent)
3303 chunk_size = mddev->chunk_size;
3306 if (chunk_size > MAX_CHUNK_SIZE) {
3307 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3308 chunk_size, MAX_CHUNK_SIZE);
3312 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3314 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3315 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3318 if (chunk_size < PAGE_SIZE) {
3319 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3320 chunk_size, PAGE_SIZE);
3324 /* devices must have minimum size of one chunk */
3325 rdev_for_each(rdev, tmp, mddev) {
3326 if (test_bit(Faulty, &rdev->flags))
3328 if (rdev->size < chunk_size / 1024) {
3330 "md: Dev %s smaller than chunk_size:"
3332 bdevname(rdev->bdev,b),
3333 (unsigned long long)rdev->size,
3341 if (mddev->level != LEVEL_NONE)
3342 request_module("md-level-%d", mddev->level);
3343 else if (mddev->clevel[0])
3344 request_module("md-%s", mddev->clevel);
3348 * Drop all container device buffers, from now on
3349 * the only valid external interface is through the md
3352 rdev_for_each(rdev, tmp, mddev) {
3353 if (test_bit(Faulty, &rdev->flags))
3355 sync_blockdev(rdev->bdev);
3356 invalidate_bdev(rdev->bdev);
3358 /* perform some consistency tests on the device.
3359 * We don't want the data to overlap the metadata,
3360 * Internal Bitmap issues has handled elsewhere.
3362 if (rdev->data_offset < rdev->sb_offset) {
3364 rdev->data_offset + mddev->size*2
3365 > rdev->sb_offset*2) {
3366 printk("md: %s: data overlaps metadata\n",
3371 if (rdev->sb_offset*2 + rdev->sb_size/512
3372 > rdev->data_offset) {
3373 printk("md: %s: metadata overlaps data\n",
3380 md_probe(mddev->unit, NULL, NULL);
3381 disk = mddev->gendisk;
3385 spin_lock(&pers_lock);
3386 pers = find_pers(mddev->level, mddev->clevel);
3387 if (!pers || !try_module_get(pers->owner)) {
3388 spin_unlock(&pers_lock);
3389 if (mddev->level != LEVEL_NONE)
3390 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3393 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3398 spin_unlock(&pers_lock);
3399 mddev->level = pers->level;
3400 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3402 if (mddev->reshape_position != MaxSector &&
3403 pers->start_reshape == NULL) {
3404 /* This personality cannot handle reshaping... */
3406 module_put(pers->owner);
3410 if (pers->sync_request) {
3411 /* Warn if this is a potentially silly
3414 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3416 struct list_head *tmp2;
3418 rdev_for_each(rdev, tmp, mddev) {
3419 rdev_for_each(rdev2, tmp2, mddev) {
3421 rdev->bdev->bd_contains ==
3422 rdev2->bdev->bd_contains) {
3424 "%s: WARNING: %s appears to be"
3425 " on the same physical disk as"
3428 bdevname(rdev->bdev,b),
3429 bdevname(rdev2->bdev,b2));
3436 "True protection against single-disk"
3437 " failure might be compromised.\n");
3440 mddev->recovery = 0;
3441 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3442 mddev->barriers_work = 1;
3443 mddev->ok_start_degraded = start_dirty_degraded;
3446 mddev->ro = 2; /* read-only, but switch on first write */
3448 err = mddev->pers->run(mddev);
3449 if (!err && mddev->pers->sync_request) {
3450 err = bitmap_create(mddev);
3452 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3453 mdname(mddev), err);
3454 mddev->pers->stop(mddev);
3458 printk(KERN_ERR "md: pers->run() failed ...\n");
3459 module_put(mddev->pers->owner);
3461 bitmap_destroy(mddev);
3464 if (mddev->pers->sync_request) {
3465 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3467 "md: cannot register extra attributes for %s\n",
3469 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3472 atomic_set(&mddev->writes_pending,0);
3473 mddev->safemode = 0;
3474 mddev->safemode_timer.function = md_safemode_timeout;
3475 mddev->safemode_timer.data = (unsigned long) mddev;
3476 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3479 rdev_for_each(rdev, tmp, mddev)
3480 if (rdev->raid_disk >= 0) {
3482 sprintf(nm, "rd%d", rdev->raid_disk);
3483 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3484 printk("md: cannot register %s for %s\n",
3488 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3491 md_update_sb(mddev, 0);
3493 set_capacity(disk, mddev->array_size<<1);
3495 /* If we call blk_queue_make_request here, it will
3496 * re-initialise max_sectors etc which may have been
3497 * refined inside -> run. So just set the bits we need to set.
3498 * Most initialisation happended when we called
3499 * blk_queue_make_request(..., md_fail_request)
3502 mddev->queue->queuedata = mddev;
3503 mddev->queue->make_request_fn = mddev->pers->make_request;
3505 /* If there is a partially-recovered drive we need to
3506 * start recovery here. If we leave it to md_check_recovery,
3507 * it will remove the drives and not do the right thing
3509 if (mddev->degraded && !mddev->sync_thread) {
3510 struct list_head *rtmp;
3512 rdev_for_each(rdev, rtmp, mddev)
3513 if (rdev->raid_disk >= 0 &&
3514 !test_bit(In_sync, &rdev->flags) &&
3515 !test_bit(Faulty, &rdev->flags))
3516 /* complete an interrupted recovery */
3518 if (spares && mddev->pers->sync_request) {
3519 mddev->recovery = 0;
3520 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3521 mddev->sync_thread = md_register_thread(md_do_sync,
3524 if (!mddev->sync_thread) {
3525 printk(KERN_ERR "%s: could not start resync"
3528 /* leave the spares where they are, it shouldn't hurt */
3529 mddev->recovery = 0;
3533 md_wakeup_thread(mddev->thread);
3534 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3537 md_new_event(mddev);
3538 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3542 static int restart_array(mddev_t *mddev)
3544 struct gendisk *disk = mddev->gendisk;
3548 * Complain if it has no devices
3551 if (list_empty(&mddev->disks))
3559 mddev->safemode = 0;
3561 set_disk_ro(disk, 0);
3563 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3566 * Kick recovery or resync if necessary
3568 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3569 md_wakeup_thread(mddev->thread);
3570 md_wakeup_thread(mddev->sync_thread);
3579 /* similar to deny_write_access, but accounts for our holding a reference
3580 * to the file ourselves */
3581 static int deny_bitmap_write_access(struct file * file)
3583 struct inode *inode = file->f_mapping->host;
3585 spin_lock(&inode->i_lock);
3586 if (atomic_read(&inode->i_writecount) > 1) {
3587 spin_unlock(&inode->i_lock);
3590 atomic_set(&inode->i_writecount, -1);
3591 spin_unlock(&inode->i_lock);
3596 static void restore_bitmap_write_access(struct file *file)
3598 struct inode *inode = file->f_mapping->host;
3600 spin_lock(&inode->i_lock);
3601 atomic_set(&inode->i_writecount, 1);
3602 spin_unlock(&inode->i_lock);
3606 * 0 - completely stop and dis-assemble array
3607 * 1 - switch to readonly
3608 * 2 - stop but do not disassemble array
3610 static int do_md_stop(mddev_t * mddev, int mode)
3613 struct gendisk *disk = mddev->gendisk;
3616 if (atomic_read(&mddev->active)>2) {
3617 printk("md: %s still in use.\n",mdname(mddev));
3621 if (mddev->sync_thread) {
3622 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3623 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3624 md_unregister_thread(mddev->sync_thread);
3625 mddev->sync_thread = NULL;
3628 del_timer_sync(&mddev->safemode_timer);
3630 invalidate_partition(disk, 0);
3633 case 1: /* readonly */
3639 case 0: /* disassemble */
3641 bitmap_flush(mddev);
3642 md_super_wait(mddev);
3644 set_disk_ro(disk, 0);
3645 blk_queue_make_request(mddev->queue, md_fail_request);
3646 mddev->pers->stop(mddev);
3647 mddev->queue->merge_bvec_fn = NULL;
3648 mddev->queue->unplug_fn = NULL;
3649 mddev->queue->backing_dev_info.congested_fn = NULL;
3650 if (mddev->pers->sync_request)
3651 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3653 module_put(mddev->pers->owner);
3656 set_capacity(disk, 0);
3662 if (!mddev->in_sync || mddev->flags) {
3663 /* mark array as shutdown cleanly */
3665 md_update_sb(mddev, 1);
3668 set_disk_ro(disk, 1);
3669 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3673 * Free resources if final stop
3677 struct list_head *tmp;
3679 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3681 bitmap_destroy(mddev);
3682 if (mddev->bitmap_file) {
3683 restore_bitmap_write_access(mddev->bitmap_file);
3684 fput(mddev->bitmap_file);
3685 mddev->bitmap_file = NULL;
3687 mddev->bitmap_offset = 0;
3689 rdev_for_each(rdev, tmp, mddev)
3690 if (rdev->raid_disk >= 0) {
3692 sprintf(nm, "rd%d", rdev->raid_disk);
3693 sysfs_remove_link(&mddev->kobj, nm);
3696 /* make sure all md_delayed_delete calls have finished */
3697 flush_scheduled_work();
3699 export_array(mddev);
3701 mddev->array_size = 0;
3703 mddev->raid_disks = 0;
3704 mddev->recovery_cp = 0;
3705 mddev->resync_max = MaxSector;
3706 mddev->reshape_position = MaxSector;
3707 mddev->external = 0;
3708 mddev->persistent = 0;
3710 } else if (mddev->pers)
3711 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3714 md_new_event(mddev);
3720 static void autorun_array(mddev_t *mddev)
3723 struct list_head *tmp;
3726 if (list_empty(&mddev->disks))
3729 printk(KERN_INFO "md: running: ");
3731 rdev_for_each(rdev, tmp, mddev) {
3732 char b[BDEVNAME_SIZE];
3733 printk("<%s>", bdevname(rdev->bdev,b));
3737 err = do_md_run (mddev);
3739 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3740 do_md_stop (mddev, 0);
3745 * lets try to run arrays based on all disks that have arrived
3746 * until now. (those are in pending_raid_disks)
3748 * the method: pick the first pending disk, collect all disks with
3749 * the same UUID, remove all from the pending list and put them into
3750 * the 'same_array' list. Then order this list based on superblock
3751 * update time (freshest comes first), kick out 'old' disks and
3752 * compare superblocks. If everything's fine then run it.
3754 * If "unit" is allocated, then bump its reference count
3756 static void autorun_devices(int part)
3758 struct list_head *tmp;
3759 mdk_rdev_t *rdev0, *rdev;
3761 char b[BDEVNAME_SIZE];
3763 printk(KERN_INFO "md: autorun ...\n");
3764 while (!list_empty(&pending_raid_disks)) {
3767 LIST_HEAD(candidates);
3768 rdev0 = list_entry(pending_raid_disks.next,
3769 mdk_rdev_t, same_set);
3771 printk(KERN_INFO "md: considering %s ...\n",
3772 bdevname(rdev0->bdev,b));
3773 INIT_LIST_HEAD(&candidates);
3774 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3775 if (super_90_load(rdev, rdev0, 0) >= 0) {
3776 printk(KERN_INFO "md: adding %s ...\n",
3777 bdevname(rdev->bdev,b));
3778 list_move(&rdev->same_set, &candidates);
3781 * now we have a set of devices, with all of them having
3782 * mostly sane superblocks. It's time to allocate the
3786 dev = MKDEV(mdp_major,
3787 rdev0->preferred_minor << MdpMinorShift);
3788 unit = MINOR(dev) >> MdpMinorShift;
3790 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3793 if (rdev0->preferred_minor != unit) {
3794 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3795 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3799 md_probe(dev, NULL, NULL);
3800 mddev = mddev_find(dev);
3803 "md: cannot allocate memory for md drive.\n");
3806 if (mddev_lock(mddev))
3807 printk(KERN_WARNING "md: %s locked, cannot run\n",
3809 else if (mddev->raid_disks || mddev->major_version
3810 || !list_empty(&mddev->disks)) {
3812 "md: %s already running, cannot run %s\n",
3813 mdname(mddev), bdevname(rdev0->bdev,b));
3814 mddev_unlock(mddev);
3816 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3817 mddev->persistent = 1;
3818 rdev_for_each_list(rdev, tmp, candidates) {
3819 list_del_init(&rdev->same_set);
3820 if (bind_rdev_to_array(rdev, mddev))
3823 autorun_array(mddev);
3824 mddev_unlock(mddev);
3826 /* on success, candidates will be empty, on error
3829 rdev_for_each_list(rdev, tmp, candidates)
3833 printk(KERN_INFO "md: ... autorun DONE.\n");
3835 #endif /* !MODULE */
3837 static int get_version(void __user * arg)
3841 ver.major = MD_MAJOR_VERSION;
3842 ver.minor = MD_MINOR_VERSION;
3843 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3845 if (copy_to_user(arg, &ver, sizeof(ver)))
3851 static int get_array_info(mddev_t * mddev, void __user * arg)
3853 mdu_array_info_t info;
3854 int nr,working,active,failed,spare;
3856 struct list_head *tmp;
3858 nr=working=active=failed=spare=0;
3859 rdev_for_each(rdev, tmp, mddev) {
3861 if (test_bit(Faulty, &rdev->flags))
3865 if (test_bit(In_sync, &rdev->flags))
3872 info.major_version = mddev->major_version;
3873 info.minor_version = mddev->minor_version;
3874 info.patch_version = MD_PATCHLEVEL_VERSION;
3875 info.ctime = mddev->ctime;
3876 info.level = mddev->level;
3877 info.size = mddev->size;
3878 if (info.size != mddev->size) /* overflow */
3881 info.raid_disks = mddev->raid_disks;
3882 info.md_minor = mddev->md_minor;
3883 info.not_persistent= !mddev->persistent;
3885 info.utime = mddev->utime;
3888 info.state = (1<<MD_SB_CLEAN);
3889 if (mddev->bitmap && mddev->bitmap_offset)
3890 info.state = (1<<MD_SB_BITMAP_PRESENT);
3891 info.active_disks = active;
3892 info.working_disks = working;
3893 info.failed_disks = failed;
3894 info.spare_disks = spare;
3896 info.layout = mddev->layout;
3897 info.chunk_size = mddev->chunk_size;
3899 if (copy_to_user(arg, &info, sizeof(info)))
3905 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3907 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3908 char *ptr, *buf = NULL;
3911 md_allow_write(mddev);
3913 file = kmalloc(sizeof(*file), GFP_KERNEL);
3917 /* bitmap disabled, zero the first byte and copy out */
3918 if (!mddev->bitmap || !mddev->bitmap->file) {
3919 file->pathname[0] = '\0';
3923 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3927 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3931 strcpy(file->pathname, ptr);
3935 if (copy_to_user(arg, file, sizeof(*file)))
3943 static int get_disk_info(mddev_t * mddev, void __user * arg)
3945 mdu_disk_info_t info;
3949 if (copy_from_user(&info, arg, sizeof(info)))
3954 rdev = find_rdev_nr(mddev, nr);
3956 info.major = MAJOR(rdev->bdev->bd_dev);
3957 info.minor = MINOR(rdev->bdev->bd_dev);
3958 info.raid_disk = rdev->raid_disk;
3960 if (test_bit(Faulty, &rdev->flags))
3961 info.state |= (1<<MD_DISK_FAULTY);
3962 else if (test_bit(In_sync, &rdev->flags)) {
3963 info.state |= (1<<MD_DISK_ACTIVE);
3964 info.state |= (1<<MD_DISK_SYNC);
3966 if (test_bit(WriteMostly, &rdev->flags))
3967 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3969 info.major = info.minor = 0;
3970 info.raid_disk = -1;
3971 info.state = (1<<MD_DISK_REMOVED);
3974 if (copy_to_user(arg, &info, sizeof(info)))
3980 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3982 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3984 dev_t dev = MKDEV(info->major,info->minor);
3986 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3989 if (!mddev->raid_disks) {
3991 /* expecting a device which has a superblock */
3992 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3995 "md: md_import_device returned %ld\n",
3997 return PTR_ERR(rdev);
3999 if (!list_empty(&mddev->disks)) {
4000 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4001 mdk_rdev_t, same_set);
4002 int err = super_types[mddev->major_version]
4003 .load_super(rdev, rdev0, mddev->minor_version);
4006 "md: %s has different UUID to %s\n",
4007 bdevname(rdev->bdev,b),
4008 bdevname(rdev0->bdev,b2));
4013 err = bind_rdev_to_array(rdev, mddev);
4020 * add_new_disk can be used once the array is assembled
4021 * to add "hot spares". They must already have a superblock
4026 if (!mddev->pers->hot_add_disk) {
4028 "%s: personality does not support diskops!\n",
4032 if (mddev->persistent)
4033 rdev = md_import_device(dev, mddev->major_version,
4034 mddev->minor_version);
4036 rdev = md_import_device(dev, -1, -1);
4039 "md: md_import_device returned %ld\n",
4041 return PTR_ERR(rdev);
4043 /* set save_raid_disk if appropriate */
4044 if (!mddev->persistent) {
4045 if (info->state & (1<<MD_DISK_SYNC) &&
4046 info->raid_disk < mddev->raid_disks)
4047 rdev->raid_disk = info->raid_disk;
4049 rdev->raid_disk = -1;
4051 super_types[mddev->major_version].
4052 validate_super(mddev, rdev);
4053 rdev->saved_raid_disk = rdev->raid_disk;
4055 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4056 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4057 set_bit(WriteMostly, &rdev->flags);
4059 rdev->raid_disk = -1;
4060 err = bind_rdev_to_array(rdev, mddev);
4061 if (!err && !mddev->pers->hot_remove_disk) {
4062 /* If there is hot_add_disk but no hot_remove_disk
4063 * then added disks for geometry changes,
4064 * and should be added immediately.
4066 super_types[mddev->major_version].
4067 validate_super(mddev, rdev);
4068 err = mddev->pers->hot_add_disk(mddev, rdev);
4070 unbind_rdev_from_array(rdev);
4075 md_update_sb(mddev, 1);
4076 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4077 md_wakeup_thread(mddev->thread);
4081 /* otherwise, add_new_disk is only allowed
4082 * for major_version==0 superblocks
4084 if (mddev->major_version != 0) {
4085 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4090 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4092 rdev = md_import_device (dev, -1, 0);
4095 "md: error, md_import_device() returned %ld\n",
4097 return PTR_ERR(rdev);
4099 rdev->desc_nr = info->number;
4100 if (info->raid_disk < mddev->raid_disks)
4101 rdev->raid_disk = info->raid_disk;
4103 rdev->raid_disk = -1;
4105 if (rdev->raid_disk < mddev->raid_disks)
4106 if (info->state & (1<<MD_DISK_SYNC))
4107 set_bit(In_sync, &rdev->flags);
4109 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4110 set_bit(WriteMostly, &rdev->flags);
4112 if (!mddev->persistent) {
4113 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4114 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4116 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4117 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4119 err = bind_rdev_to_array(rdev, mddev);
4129 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4131 char b[BDEVNAME_SIZE];
4137 rdev = find_rdev(mddev, dev);
4141 if (rdev->raid_disk >= 0)
4144 kick_rdev_from_array(rdev);
4145 md_update_sb(mddev, 1);
4146 md_new_event(mddev);
4150 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
4151 bdevname(rdev->bdev,b), mdname(mddev));
4155 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4157 char b[BDEVNAME_SIZE];
4165 if (mddev->major_version != 0) {
4166 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4167 " version-0 superblocks.\n",
4171 if (!mddev->pers->hot_add_disk) {
4173 "%s: personality does not support diskops!\n",
4178 rdev = md_import_device (dev, -1, 0);
4181 "md: error, md_import_device() returned %ld\n",
4186 if (mddev->persistent)
4187 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4190 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4192 size = calc_dev_size(rdev, mddev->chunk_size);
4195 if (test_bit(Faulty, &rdev->flags)) {
4197 "md: can not hot-add faulty %s disk to %s!\n",
4198 bdevname(rdev->bdev,b), mdname(mddev));
4202 clear_bit(In_sync, &rdev->flags);
4204 rdev->saved_raid_disk = -1;
4205 err = bind_rdev_to_array(rdev, mddev);
4210 * The rest should better be atomic, we can have disk failures
4211 * noticed in interrupt contexts ...
4214 if (rdev->desc_nr == mddev->max_disks) {
4215 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4218 goto abort_unbind_export;
4221 rdev->raid_disk = -1;
4223 md_update_sb(mddev, 1);
4226 * Kick recovery, maybe this spare has to be added to the
4227 * array immediately.
4229 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4230 md_wakeup_thread(mddev->thread);
4231 md_new_event(mddev);
4234 abort_unbind_export:
4235 unbind_rdev_from_array(rdev);
4242 static int set_bitmap_file(mddev_t *mddev, int fd)
4247 if (!mddev->pers->quiesce)
4249 if (mddev->recovery || mddev->sync_thread)
4251 /* we should be able to change the bitmap.. */
4257 return -EEXIST; /* cannot add when bitmap is present */
4258 mddev->bitmap_file = fget(fd);
4260 if (mddev->bitmap_file == NULL) {
4261 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4266 err = deny_bitmap_write_access(mddev->bitmap_file);
4268 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4270 fput(mddev->bitmap_file);
4271 mddev->bitmap_file = NULL;
4274 mddev->bitmap_offset = 0; /* file overrides offset */
4275 } else if (mddev->bitmap == NULL)
4276 return -ENOENT; /* cannot remove what isn't there */
4279 mddev->pers->quiesce(mddev, 1);
4281 err = bitmap_create(mddev);
4282 if (fd < 0 || err) {
4283 bitmap_destroy(mddev);
4284 fd = -1; /* make sure to put the file */
4286 mddev->pers->quiesce(mddev, 0);
4289 if (mddev->bitmap_file) {
4290 restore_bitmap_write_access(mddev->bitmap_file);
4291 fput(mddev->bitmap_file);
4293 mddev->bitmap_file = NULL;
4300 * set_array_info is used two different ways
4301 * The original usage is when creating a new array.
4302 * In this usage, raid_disks is > 0 and it together with
4303 * level, size, not_persistent,layout,chunksize determine the
4304 * shape of the array.
4305 * This will always create an array with a type-0.90.0 superblock.
4306 * The newer usage is when assembling an array.
4307 * In this case raid_disks will be 0, and the major_version field is
4308 * use to determine which style super-blocks are to be found on the devices.
4309 * The minor and patch _version numbers are also kept incase the
4310 * super_block handler wishes to interpret them.
4312 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4315 if (info->raid_disks == 0) {
4316 /* just setting version number for superblock loading */
4317 if (info->major_version < 0 ||
4318 info->major_version >= ARRAY_SIZE(super_types) ||
4319 super_types[info->major_version].name == NULL) {
4320 /* maybe try to auto-load a module? */
4322 "md: superblock version %d not known\n",
4323 info->major_version);
4326 mddev->major_version = info->major_version;
4327 mddev->minor_version = info->minor_version;
4328 mddev->patch_version = info->patch_version;
4329 mddev->persistent = !info->not_persistent;
4332 mddev->major_version = MD_MAJOR_VERSION;
4333 mddev->minor_version = MD_MINOR_VERSION;
4334 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4335 mddev->ctime = get_seconds();
4337 mddev->level = info->level;
4338 mddev->clevel[0] = 0;
4339 mddev->size = info->size;
4340 mddev->raid_disks = info->raid_disks;
4341 /* don't set md_minor, it is determined by which /dev/md* was
4344 if (info->state & (1<<MD_SB_CLEAN))
4345 mddev->recovery_cp = MaxSector;
4347 mddev->recovery_cp = 0;
4348 mddev->persistent = ! info->not_persistent;
4349 mddev->external = 0;
4351 mddev->layout = info->layout;
4352 mddev->chunk_size = info->chunk_size;
4354 mddev->max_disks = MD_SB_DISKS;
4356 if (mddev->persistent)
4358 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4360 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4361 mddev->bitmap_offset = 0;
4363 mddev->reshape_position = MaxSector;
4366 * Generate a 128 bit UUID
4368 get_random_bytes(mddev->uuid, 16);
4370 mddev->new_level = mddev->level;
4371 mddev->new_chunk = mddev->chunk_size;
4372 mddev->new_layout = mddev->layout;
4373 mddev->delta_disks = 0;
4378 static int update_size(mddev_t *mddev, unsigned long size)
4382 struct list_head *tmp;
4383 int fit = (size == 0);
4385 if (mddev->pers->resize == NULL)
4387 /* The "size" is the amount of each device that is used.
4388 * This can only make sense for arrays with redundancy.
4389 * linear and raid0 always use whatever space is available
4390 * We can only consider changing the size if no resync
4391 * or reconstruction is happening, and if the new size
4392 * is acceptable. It must fit before the sb_offset or,
4393 * if that is <data_offset, it must fit before the
4394 * size of each device.
4395 * If size is zero, we find the largest size that fits.
4397 if (mddev->sync_thread)
4399 rdev_for_each(rdev, tmp, mddev) {
4401 avail = rdev->size * 2;
4403 if (fit && (size == 0 || size > avail/2))
4405 if (avail < ((sector_t)size << 1))
4408 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4410 struct block_device *bdev;
4412 bdev = bdget_disk(mddev->gendisk, 0);
4414 mutex_lock(&bdev->bd_inode->i_mutex);
4415 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4416 mutex_unlock(&bdev->bd_inode->i_mutex);
4423 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4426 /* change the number of raid disks */
4427 if (mddev->pers->check_reshape == NULL)
4429 if (raid_disks <= 0 ||
4430 raid_disks >= mddev->max_disks)
4432 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4434 mddev->delta_disks = raid_disks - mddev->raid_disks;
4436 rv = mddev->pers->check_reshape(mddev);
4442 * update_array_info is used to change the configuration of an
4444 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4445 * fields in the info are checked against the array.
4446 * Any differences that cannot be handled will cause an error.
4447 * Normally, only one change can be managed at a time.
4449 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4455 /* calculate expected state,ignoring low bits */
4456 if (mddev->bitmap && mddev->bitmap_offset)
4457 state |= (1 << MD_SB_BITMAP_PRESENT);
4459 if (mddev->major_version != info->major_version ||
4460 mddev->minor_version != info->minor_version ||
4461 /* mddev->patch_version != info->patch_version || */
4462 mddev->ctime != info->ctime ||
4463 mddev->level != info->level ||
4464 /* mddev->layout != info->layout || */
4465 !mddev->persistent != info->not_persistent||
4466 mddev->chunk_size != info->chunk_size ||
4467 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4468 ((state^info->state) & 0xfffffe00)
4471 /* Check there is only one change */
4472 if (info->size >= 0 && mddev->size != info->size) cnt++;
4473 if (mddev->raid_disks != info->raid_disks) cnt++;
4474 if (mddev->layout != info->layout) cnt++;
4475 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4476 if (cnt == 0) return 0;
4477 if (cnt > 1) return -EINVAL;
4479 if (mddev->layout != info->layout) {
4481 * we don't need to do anything at the md level, the
4482 * personality will take care of it all.
4484 if (mddev->pers->reconfig == NULL)
4487 return mddev->pers->reconfig(mddev, info->layout, -1);
4489 if (info->size >= 0 && mddev->size != info->size)
4490 rv = update_size(mddev, info->size);
4492 if (mddev->raid_disks != info->raid_disks)
4493 rv = update_raid_disks(mddev, info->raid_disks);
4495 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4496 if (mddev->pers->quiesce == NULL)
4498 if (mddev->recovery || mddev->sync_thread)
4500 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4501 /* add the bitmap */
4504 if (mddev->default_bitmap_offset == 0)
4506 mddev->bitmap_offset = mddev->default_bitmap_offset;
4507 mddev->pers->quiesce(mddev, 1);
4508 rv = bitmap_create(mddev);
4510 bitmap_destroy(mddev);
4511 mddev->pers->quiesce(mddev, 0);
4513 /* remove the bitmap */
4516 if (mddev->bitmap->file)
4518 mddev->pers->quiesce(mddev, 1);
4519 bitmap_destroy(mddev);
4520 mddev->pers->quiesce(mddev, 0);
4521 mddev->bitmap_offset = 0;
4524 md_update_sb(mddev, 1);
4528 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4532 if (mddev->pers == NULL)
4535 rdev = find_rdev(mddev, dev);
4539 md_error(mddev, rdev);
4543 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4545 mddev_t *mddev = bdev->bd_disk->private_data;
4549 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4553 static int md_ioctl(struct inode *inode, struct file *file,
4554 unsigned int cmd, unsigned long arg)
4557 void __user *argp = (void __user *)arg;
4558 mddev_t *mddev = NULL;
4560 if (!capable(CAP_SYS_ADMIN))
4564 * Commands dealing with the RAID driver but not any
4570 err = get_version(argp);
4573 case PRINT_RAID_DEBUG:
4581 autostart_arrays(arg);
4588 * Commands creating/starting a new array:
4591 mddev = inode->i_bdev->bd_disk->private_data;
4598 err = mddev_lock(mddev);
4601 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4608 case SET_ARRAY_INFO:
4610 mdu_array_info_t info;
4612 memset(&info, 0, sizeof(info));
4613 else if (copy_from_user(&info, argp, sizeof(info))) {
4618 err = update_array_info(mddev, &info);
4620 printk(KERN_WARNING "md: couldn't update"
4621 " array info. %d\n", err);
4626 if (!list_empty(&mddev->disks)) {
4628 "md: array %s already has disks!\n",
4633 if (mddev->raid_disks) {
4635 "md: array %s already initialised!\n",
4640 err = set_array_info(mddev, &info);
4642 printk(KERN_WARNING "md: couldn't set"
4643 " array info. %d\n", err);
4653 * Commands querying/configuring an existing array:
4655 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4656 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4657 if ((!mddev->raid_disks && !mddev->external)
4658 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4659 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4660 && cmd != GET_BITMAP_FILE) {
4666 * Commands even a read-only array can execute:
4670 case GET_ARRAY_INFO:
4671 err = get_array_info(mddev, argp);
4674 case GET_BITMAP_FILE:
4675 err = get_bitmap_file(mddev, argp);
4679 err = get_disk_info(mddev, argp);
4682 case RESTART_ARRAY_RW:
4683 err = restart_array(mddev);
4687 err = do_md_stop (mddev, 0);
4691 err = do_md_stop (mddev, 1);
4695 * We have a problem here : there is no easy way to give a CHS
4696 * virtual geometry. We currently pretend that we have a 2 heads
4697 * 4 sectors (with a BIG number of cylinders...). This drives
4698 * dosfs just mad... ;-)
4703 * The remaining ioctls are changing the state of the
4704 * superblock, so we do not allow them on read-only arrays.
4705 * However non-MD ioctls (e.g. get-size) will still come through
4706 * here and hit the 'default' below, so only disallow
4707 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4709 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4710 mddev->ro && mddev->pers) {
4711 if (mddev->ro == 2) {
4713 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4714 md_wakeup_thread(mddev->thread);
4726 mdu_disk_info_t info;
4727 if (copy_from_user(&info, argp, sizeof(info)))
4730 err = add_new_disk(mddev, &info);
4734 case HOT_REMOVE_DISK:
4735 err = hot_remove_disk(mddev, new_decode_dev(arg));
4739 err = hot_add_disk(mddev, new_decode_dev(arg));
4742 case SET_DISK_FAULTY:
4743 err = set_disk_faulty(mddev, new_decode_dev(arg));
4747 err = do_md_run (mddev);
4750 case SET_BITMAP_FILE:
4751 err = set_bitmap_file(mddev, (int)arg);
4761 mddev_unlock(mddev);
4771 static int md_open(struct inode *inode, struct file *file)
4774 * Succeed if we can lock the mddev, which confirms that
4775 * it isn't being stopped right now.
4777 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4780 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4785 mddev_unlock(mddev);
4787 check_disk_change(inode->i_bdev);
4792 static int md_release(struct inode *inode, struct file * file)
4794 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4802 static int md_media_changed(struct gendisk *disk)
4804 mddev_t *mddev = disk->private_data;
4806 return mddev->changed;
4809 static int md_revalidate(struct gendisk *disk)
4811 mddev_t *mddev = disk->private_data;
4816 static struct block_device_operations md_fops =
4818 .owner = THIS_MODULE,
4820 .release = md_release,
4822 .getgeo = md_getgeo,
4823 .media_changed = md_media_changed,
4824 .revalidate_disk= md_revalidate,
4827 static int md_thread(void * arg)
4829 mdk_thread_t *thread = arg;
4832 * md_thread is a 'system-thread', it's priority should be very
4833 * high. We avoid resource deadlocks individually in each
4834 * raid personality. (RAID5 does preallocation) We also use RR and
4835 * the very same RT priority as kswapd, thus we will never get
4836 * into a priority inversion deadlock.
4838 * we definitely have to have equal or higher priority than
4839 * bdflush, otherwise bdflush will deadlock if there are too
4840 * many dirty RAID5 blocks.
4843 allow_signal(SIGKILL);
4844 while (!kthread_should_stop()) {
4846 /* We need to wait INTERRUPTIBLE so that
4847 * we don't add to the load-average.
4848 * That means we need to be sure no signals are
4851 if (signal_pending(current))
4852 flush_signals(current);
4854 wait_event_interruptible_timeout
4856 test_bit(THREAD_WAKEUP, &thread->flags)
4857 || kthread_should_stop(),
4860 clear_bit(THREAD_WAKEUP, &thread->flags);
4862 thread->run(thread->mddev);
4868 void md_wakeup_thread(mdk_thread_t *thread)
4871 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4872 set_bit(THREAD_WAKEUP, &thread->flags);
4873 wake_up(&thread->wqueue);
4877 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4880 mdk_thread_t *thread;
4882 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4886 init_waitqueue_head(&thread->wqueue);
4889 thread->mddev = mddev;
4890 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4891 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4892 if (IS_ERR(thread->tsk)) {
4899 void md_unregister_thread(mdk_thread_t *thread)
4901 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4903 kthread_stop(thread->tsk);
4907 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4914 if (!rdev || test_bit(Faulty, &rdev->flags))
4917 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4919 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4920 __builtin_return_address(0),__builtin_return_address(1),
4921 __builtin_return_address(2),__builtin_return_address(3));
4925 if (!mddev->pers->error_handler)
4927 mddev->pers->error_handler(mddev,rdev);
4928 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4929 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4930 md_wakeup_thread(mddev->thread);
4931 md_new_event_inintr(mddev);
4934 /* seq_file implementation /proc/mdstat */
4936 static void status_unused(struct seq_file *seq)
4940 struct list_head *tmp;
4942 seq_printf(seq, "unused devices: ");
4944 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
4945 char b[BDEVNAME_SIZE];
4947 seq_printf(seq, "%s ",
4948 bdevname(rdev->bdev,b));
4951 seq_printf(seq, "<none>");
4953 seq_printf(seq, "\n");
4957 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4959 sector_t max_blocks, resync, res;
4960 unsigned long dt, db, rt;
4962 unsigned int per_milli;
4964 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4966 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4967 max_blocks = mddev->resync_max_sectors >> 1;
4969 max_blocks = mddev->size;
4972 * Should not happen.
4978 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4979 * in a sector_t, and (max_blocks>>scale) will fit in a
4980 * u32, as those are the requirements for sector_div.
4981 * Thus 'scale' must be at least 10
4984 if (sizeof(sector_t) > sizeof(unsigned long)) {
4985 while ( max_blocks/2 > (1ULL<<(scale+32)))
4988 res = (resync>>scale)*1000;
4989 sector_div(res, (u32)((max_blocks>>scale)+1));
4993 int i, x = per_milli/50, y = 20-x;
4994 seq_printf(seq, "[");
4995 for (i = 0; i < x; i++)
4996 seq_printf(seq, "=");
4997 seq_printf(seq, ">");
4998 for (i = 0; i < y; i++)
4999 seq_printf(seq, ".");
5000 seq_printf(seq, "] ");
5002 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5003 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5005 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5007 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5008 "resync" : "recovery"))),
5009 per_milli/10, per_milli % 10,
5010 (unsigned long long) resync,
5011 (unsigned long long) max_blocks);
5014 * We do not want to overflow, so the order of operands and
5015 * the * 100 / 100 trick are important. We do a +1 to be
5016 * safe against division by zero. We only estimate anyway.
5018 * dt: time from mark until now
5019 * db: blocks written from mark until now
5020 * rt: remaining time
5022 dt = ((jiffies - mddev->resync_mark) / HZ);
5024 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5025 - mddev->resync_mark_cnt;
5026 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5028 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5030 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5033 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5035 struct list_head *tmp;
5045 spin_lock(&all_mddevs_lock);
5046 list_for_each(tmp,&all_mddevs)
5048 mddev = list_entry(tmp, mddev_t, all_mddevs);
5050 spin_unlock(&all_mddevs_lock);
5053 spin_unlock(&all_mddevs_lock);
5055 return (void*)2;/* tail */
5059 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5061 struct list_head *tmp;
5062 mddev_t *next_mddev, *mddev = v;
5068 spin_lock(&all_mddevs_lock);
5070 tmp = all_mddevs.next;
5072 tmp = mddev->all_mddevs.next;
5073 if (tmp != &all_mddevs)
5074 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5076 next_mddev = (void*)2;
5079 spin_unlock(&all_mddevs_lock);
5087 static void md_seq_stop(struct seq_file *seq, void *v)
5091 if (mddev && v != (void*)1 && v != (void*)2)
5095 struct mdstat_info {
5099 static int md_seq_show(struct seq_file *seq, void *v)
5103 struct list_head *tmp2;
5105 struct mdstat_info *mi = seq->private;
5106 struct bitmap *bitmap;
5108 if (v == (void*)1) {
5109 struct mdk_personality *pers;
5110 seq_printf(seq, "Personalities : ");
5111 spin_lock(&pers_lock);
5112 list_for_each_entry(pers, &pers_list, list)
5113 seq_printf(seq, "[%s] ", pers->name);
5115 spin_unlock(&pers_lock);
5116 seq_printf(seq, "\n");
5117 mi->event = atomic_read(&md_event_count);
5120 if (v == (void*)2) {
5125 if (mddev_lock(mddev) < 0)
5128 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5129 seq_printf(seq, "%s : %sactive", mdname(mddev),
5130 mddev->pers ? "" : "in");
5133 seq_printf(seq, " (read-only)");
5135 seq_printf(seq, "(auto-read-only)");
5136 seq_printf(seq, " %s", mddev->pers->name);
5140 rdev_for_each(rdev, tmp2, mddev) {
5141 char b[BDEVNAME_SIZE];
5142 seq_printf(seq, " %s[%d]",
5143 bdevname(rdev->bdev,b), rdev->desc_nr);
5144 if (test_bit(WriteMostly, &rdev->flags))
5145 seq_printf(seq, "(W)");
5146 if (test_bit(Faulty, &rdev->flags)) {
5147 seq_printf(seq, "(F)");
5149 } else if (rdev->raid_disk < 0)
5150 seq_printf(seq, "(S)"); /* spare */
5154 if (!list_empty(&mddev->disks)) {
5156 seq_printf(seq, "\n %llu blocks",
5157 (unsigned long long)mddev->array_size);
5159 seq_printf(seq, "\n %llu blocks",
5160 (unsigned long long)size);
5162 if (mddev->persistent) {
5163 if (mddev->major_version != 0 ||
5164 mddev->minor_version != 90) {
5165 seq_printf(seq," super %d.%d",
5166 mddev->major_version,
5167 mddev->minor_version);
5169 } else if (mddev->external)
5170 seq_printf(seq, " super external:%s",
5171 mddev->metadata_type);
5173 seq_printf(seq, " super non-persistent");
5176 mddev->pers->status (seq, mddev);
5177 seq_printf(seq, "\n ");
5178 if (mddev->pers->sync_request) {
5179 if (mddev->curr_resync > 2) {
5180 status_resync (seq, mddev);
5181 seq_printf(seq, "\n ");
5182 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5183 seq_printf(seq, "\tresync=DELAYED\n ");
5184 else if (mddev->recovery_cp < MaxSector)
5185 seq_printf(seq, "\tresync=PENDING\n ");
5188 seq_printf(seq, "\n ");
5190 if ((bitmap = mddev->bitmap)) {
5191 unsigned long chunk_kb;
5192 unsigned long flags;
5193 spin_lock_irqsave(&bitmap->lock, flags);
5194 chunk_kb = bitmap->chunksize >> 10;
5195 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5197 bitmap->pages - bitmap->missing_pages,
5199 (bitmap->pages - bitmap->missing_pages)
5200 << (PAGE_SHIFT - 10),
5201 chunk_kb ? chunk_kb : bitmap->chunksize,
5202 chunk_kb ? "KB" : "B");
5204 seq_printf(seq, ", file: ");
5205 seq_path(seq, &bitmap->file->f_path, " \t\n");
5208 seq_printf(seq, "\n");
5209 spin_unlock_irqrestore(&bitmap->lock, flags);
5212 seq_printf(seq, "\n");
5214 mddev_unlock(mddev);
5219 static struct seq_operations md_seq_ops = {
5220 .start = md_seq_start,
5221 .next = md_seq_next,
5222 .stop = md_seq_stop,
5223 .show = md_seq_show,
5226 static int md_seq_open(struct inode *inode, struct file *file)
5229 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5233 error = seq_open(file, &md_seq_ops);
5237 struct seq_file *p = file->private_data;
5239 mi->event = atomic_read(&md_event_count);
5244 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5246 struct seq_file *m = filp->private_data;
5247 struct mdstat_info *mi = m->private;
5250 poll_wait(filp, &md_event_waiters, wait);
5252 /* always allow read */
5253 mask = POLLIN | POLLRDNORM;
5255 if (mi->event != atomic_read(&md_event_count))
5256 mask |= POLLERR | POLLPRI;
5260 static const struct file_operations md_seq_fops = {
5261 .owner = THIS_MODULE,
5262 .open = md_seq_open,
5264 .llseek = seq_lseek,
5265 .release = seq_release_private,
5266 .poll = mdstat_poll,
5269 int register_md_personality(struct mdk_personality *p)
5271 spin_lock(&pers_lock);
5272 list_add_tail(&p->list, &pers_list);
5273 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5274 spin_unlock(&pers_lock);
5278 int unregister_md_personality(struct mdk_personality *p)
5280 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5281 spin_lock(&pers_lock);
5282 list_del_init(&p->list);
5283 spin_unlock(&pers_lock);
5287 static int is_mddev_idle(mddev_t *mddev)
5290 struct list_head *tmp;
5295 rdev_for_each(rdev, tmp, mddev) {
5296 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5297 curr_events = disk_stat_read(disk, sectors[0]) +
5298 disk_stat_read(disk, sectors[1]) -
5299 atomic_read(&disk->sync_io);
5300 /* sync IO will cause sync_io to increase before the disk_stats
5301 * as sync_io is counted when a request starts, and
5302 * disk_stats is counted when it completes.
5303 * So resync activity will cause curr_events to be smaller than
5304 * when there was no such activity.
5305 * non-sync IO will cause disk_stat to increase without
5306 * increasing sync_io so curr_events will (eventually)
5307 * be larger than it was before. Once it becomes
5308 * substantially larger, the test below will cause
5309 * the array to appear non-idle, and resync will slow
5311 * If there is a lot of outstanding resync activity when
5312 * we set last_event to curr_events, then all that activity
5313 * completing might cause the array to appear non-idle
5314 * and resync will be slowed down even though there might
5315 * not have been non-resync activity. This will only
5316 * happen once though. 'last_events' will soon reflect
5317 * the state where there is little or no outstanding
5318 * resync requests, and further resync activity will
5319 * always make curr_events less than last_events.
5322 if (curr_events - rdev->last_events > 4096) {
5323 rdev->last_events = curr_events;
5330 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5332 /* another "blocks" (512byte) blocks have been synced */
5333 atomic_sub(blocks, &mddev->recovery_active);
5334 wake_up(&mddev->recovery_wait);
5336 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5337 md_wakeup_thread(mddev->thread);
5338 // stop recovery, signal do_sync ....
5343 /* md_write_start(mddev, bi)
5344 * If we need to update some array metadata (e.g. 'active' flag
5345 * in superblock) before writing, schedule a superblock update
5346 * and wait for it to complete.
5348 void md_write_start(mddev_t *mddev, struct bio *bi)
5350 if (bio_data_dir(bi) != WRITE)
5353 BUG_ON(mddev->ro == 1);
5354 if (mddev->ro == 2) {
5355 /* need to switch to read/write */
5357 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5358 md_wakeup_thread(mddev->thread);
5359 md_wakeup_thread(mddev->sync_thread);
5361 atomic_inc(&mddev->writes_pending);
5362 if (mddev->in_sync) {
5363 spin_lock_irq(&mddev->write_lock);
5364 if (mddev->in_sync) {
5366 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5367 md_wakeup_thread(mddev->thread);
5369 spin_unlock_irq(&mddev->write_lock);
5371 wait_event(mddev->sb_wait, mddev->flags==0);
5374 void md_write_end(mddev_t *mddev)
5376 if (atomic_dec_and_test(&mddev->writes_pending)) {
5377 if (mddev->safemode == 2)
5378 md_wakeup_thread(mddev->thread);
5379 else if (mddev->safemode_delay)
5380 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5384 /* md_allow_write(mddev)
5385 * Calling this ensures that the array is marked 'active' so that writes
5386 * may proceed without blocking. It is important to call this before
5387 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5388 * Must be called with mddev_lock held.
5390 void md_allow_write(mddev_t *mddev)
5397 spin_lock_irq(&mddev->write_lock);
5398 if (mddev->in_sync) {
5400 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5401 if (mddev->safemode_delay &&
5402 mddev->safemode == 0)
5403 mddev->safemode = 1;
5404 spin_unlock_irq(&mddev->write_lock);
5405 md_update_sb(mddev, 0);
5407 spin_unlock_irq(&mddev->write_lock);
5409 EXPORT_SYMBOL_GPL(md_allow_write);
5411 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5413 #define SYNC_MARKS 10
5414 #define SYNC_MARK_STEP (3*HZ)
5415 void md_do_sync(mddev_t *mddev)
5418 unsigned int currspeed = 0,
5420 sector_t max_sectors,j, io_sectors;
5421 unsigned long mark[SYNC_MARKS];
5422 sector_t mark_cnt[SYNC_MARKS];
5424 struct list_head *tmp;
5425 sector_t last_check;
5427 struct list_head *rtmp;
5431 /* just incase thread restarts... */
5432 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5434 if (mddev->ro) /* never try to sync a read-only array */
5437 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5438 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5439 desc = "data-check";
5440 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5441 desc = "requested-resync";
5444 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5449 /* we overload curr_resync somewhat here.
5450 * 0 == not engaged in resync at all
5451 * 2 == checking that there is no conflict with another sync
5452 * 1 == like 2, but have yielded to allow conflicting resync to
5454 * other == active in resync - this many blocks
5456 * Before starting a resync we must have set curr_resync to
5457 * 2, and then checked that every "conflicting" array has curr_resync
5458 * less than ours. When we find one that is the same or higher
5459 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5460 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5461 * This will mean we have to start checking from the beginning again.
5466 mddev->curr_resync = 2;
5469 if (kthread_should_stop()) {
5470 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5473 for_each_mddev(mddev2, tmp) {
5474 if (mddev2 == mddev)
5476 if (mddev2->curr_resync &&
5477 match_mddev_units(mddev,mddev2)) {
5479 if (mddev < mddev2 && mddev->curr_resync == 2) {
5480 /* arbitrarily yield */
5481 mddev->curr_resync = 1;
5482 wake_up(&resync_wait);
5484 if (mddev > mddev2 && mddev->curr_resync == 1)
5485 /* no need to wait here, we can wait the next
5486 * time 'round when curr_resync == 2
5489 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5490 if (!kthread_should_stop() &&
5491 mddev2->curr_resync >= mddev->curr_resync) {
5492 printk(KERN_INFO "md: delaying %s of %s"
5493 " until %s has finished (they"
5494 " share one or more physical units)\n",
5495 desc, mdname(mddev), mdname(mddev2));
5498 finish_wait(&resync_wait, &wq);
5501 finish_wait(&resync_wait, &wq);
5504 } while (mddev->curr_resync < 2);
5507 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5508 /* resync follows the size requested by the personality,
5509 * which defaults to physical size, but can be virtual size
5511 max_sectors = mddev->resync_max_sectors;
5512 mddev->resync_mismatches = 0;
5513 /* we don't use the checkpoint if there's a bitmap */
5514 if (!mddev->bitmap &&
5515 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5516 j = mddev->recovery_cp;
5517 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5518 max_sectors = mddev->size << 1;
5520 /* recovery follows the physical size of devices */
5521 max_sectors = mddev->size << 1;
5523 rdev_for_each(rdev, rtmp, mddev)
5524 if (rdev->raid_disk >= 0 &&
5525 !test_bit(Faulty, &rdev->flags) &&
5526 !test_bit(In_sync, &rdev->flags) &&
5527 rdev->recovery_offset < j)
5528 j = rdev->recovery_offset;
5531 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5532 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5533 " %d KB/sec/disk.\n", speed_min(mddev));
5534 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5535 "(but not more than %d KB/sec) for %s.\n",
5536 speed_max(mddev), desc);
5538 is_mddev_idle(mddev); /* this also initializes IO event counters */
5541 for (m = 0; m < SYNC_MARKS; m++) {
5543 mark_cnt[m] = io_sectors;
5546 mddev->resync_mark = mark[last_mark];
5547 mddev->resync_mark_cnt = mark_cnt[last_mark];
5550 * Tune reconstruction:
5552 window = 32*(PAGE_SIZE/512);
5553 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5554 window/2,(unsigned long long) max_sectors/2);
5556 atomic_set(&mddev->recovery_active, 0);
5557 init_waitqueue_head(&mddev->recovery_wait);
5562 "md: resuming %s of %s from checkpoint.\n",
5563 desc, mdname(mddev));
5564 mddev->curr_resync = j;
5567 while (j < max_sectors) {
5571 if (j >= mddev->resync_max) {
5572 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5573 wait_event(mddev->recovery_wait,
5574 mddev->resync_max > j
5575 || kthread_should_stop());
5577 if (kthread_should_stop())
5579 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5580 currspeed < speed_min(mddev));
5582 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5586 if (!skipped) { /* actual IO requested */
5587 io_sectors += sectors;
5588 atomic_add(sectors, &mddev->recovery_active);
5592 if (j>1) mddev->curr_resync = j;
5593 mddev->curr_mark_cnt = io_sectors;
5594 if (last_check == 0)
5595 /* this is the earliers that rebuilt will be
5596 * visible in /proc/mdstat
5598 md_new_event(mddev);
5600 if (last_check + window > io_sectors || j == max_sectors)
5603 last_check = io_sectors;
5605 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5606 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5610 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5612 int next = (last_mark+1) % SYNC_MARKS;
5614 mddev->resync_mark = mark[next];
5615 mddev->resync_mark_cnt = mark_cnt[next];
5616 mark[next] = jiffies;
5617 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5622 if (kthread_should_stop())
5627 * this loop exits only if either when we are slower than
5628 * the 'hard' speed limit, or the system was IO-idle for
5630 * the system might be non-idle CPU-wise, but we only care
5631 * about not overloading the IO subsystem. (things like an
5632 * e2fsck being done on the RAID array should execute fast)
5634 blk_unplug(mddev->queue);
5637 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5638 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5640 if (currspeed > speed_min(mddev)) {
5641 if ((currspeed > speed_max(mddev)) ||
5642 !is_mddev_idle(mddev)) {
5648 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5650 * this also signals 'finished resyncing' to md_stop
5653 blk_unplug(mddev->queue);
5655 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5657 /* tell personality that we are finished */
5658 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5660 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5661 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5662 mddev->curr_resync > 2) {
5663 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5664 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5665 if (mddev->curr_resync >= mddev->recovery_cp) {
5667 "md: checkpointing %s of %s.\n",
5668 desc, mdname(mddev));
5669 mddev->recovery_cp = mddev->curr_resync;
5672 mddev->recovery_cp = MaxSector;
5674 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5675 mddev->curr_resync = MaxSector;
5676 rdev_for_each(rdev, rtmp, mddev)
5677 if (rdev->raid_disk >= 0 &&
5678 !test_bit(Faulty, &rdev->flags) &&
5679 !test_bit(In_sync, &rdev->flags) &&
5680 rdev->recovery_offset < mddev->curr_resync)
5681 rdev->recovery_offset = mddev->curr_resync;
5684 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5687 mddev->curr_resync = 0;
5688 mddev->resync_max = MaxSector;
5689 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5690 wake_up(&resync_wait);
5691 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5692 md_wakeup_thread(mddev->thread);
5697 * got a signal, exit.
5700 "md: md_do_sync() got signal ... exiting\n");
5701 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5705 EXPORT_SYMBOL_GPL(md_do_sync);
5708 static int remove_and_add_spares(mddev_t *mddev)
5711 struct list_head *rtmp;
5714 rdev_for_each(rdev, rtmp, mddev)
5715 if (rdev->raid_disk >= 0 &&
5717 (test_bit(Faulty, &rdev->flags) ||
5718 ! test_bit(In_sync, &rdev->flags)) &&
5719 atomic_read(&rdev->nr_pending)==0) {
5720 if (mddev->pers->hot_remove_disk(
5721 mddev, rdev->raid_disk)==0) {
5723 sprintf(nm,"rd%d", rdev->raid_disk);
5724 sysfs_remove_link(&mddev->kobj, nm);
5725 rdev->raid_disk = -1;
5729 if (mddev->degraded) {
5730 rdev_for_each(rdev, rtmp, mddev)
5731 if (rdev->raid_disk < 0
5732 && !test_bit(Faulty, &rdev->flags)) {
5733 rdev->recovery_offset = 0;
5734 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5736 sprintf(nm, "rd%d", rdev->raid_disk);
5737 if (sysfs_create_link(&mddev->kobj,
5740 "md: cannot register "
5744 md_new_event(mddev);
5752 * This routine is regularly called by all per-raid-array threads to
5753 * deal with generic issues like resync and super-block update.
5754 * Raid personalities that don't have a thread (linear/raid0) do not
5755 * need this as they never do any recovery or update the superblock.
5757 * It does not do any resync itself, but rather "forks" off other threads
5758 * to do that as needed.
5759 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5760 * "->recovery" and create a thread at ->sync_thread.
5761 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5762 * and wakeups up this thread which will reap the thread and finish up.
5763 * This thread also removes any faulty devices (with nr_pending == 0).
5765 * The overall approach is:
5766 * 1/ if the superblock needs updating, update it.
5767 * 2/ If a recovery thread is running, don't do anything else.
5768 * 3/ If recovery has finished, clean up, possibly marking spares active.
5769 * 4/ If there are any faulty devices, remove them.
5770 * 5/ If array is degraded, try to add spares devices
5771 * 6/ If array has spares or is not in-sync, start a resync thread.
5773 void md_check_recovery(mddev_t *mddev)
5776 struct list_head *rtmp;
5780 bitmap_daemon_work(mddev->bitmap);
5785 if (signal_pending(current)) {
5786 if (mddev->pers->sync_request) {
5787 printk(KERN_INFO "md: %s in immediate safe mode\n",
5789 mddev->safemode = 2;
5791 flush_signals(current);
5795 (mddev->flags && !mddev->external) ||
5796 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5797 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5798 (mddev->safemode == 1) ||
5799 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5800 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5804 if (mddev_trylock(mddev)) {
5807 spin_lock_irq(&mddev->write_lock);
5808 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5809 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5811 if (mddev->persistent)
5812 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5814 if (mddev->safemode == 1)
5815 mddev->safemode = 0;
5816 spin_unlock_irq(&mddev->write_lock);
5819 md_update_sb(mddev, 0);
5822 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5823 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5824 /* resync/recovery still happening */
5825 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5828 if (mddev->sync_thread) {
5829 /* resync has finished, collect result */
5830 md_unregister_thread(mddev->sync_thread);
5831 mddev->sync_thread = NULL;
5832 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5833 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5835 /* activate any spares */
5836 mddev->pers->spare_active(mddev);
5838 md_update_sb(mddev, 1);
5840 /* if array is no-longer degraded, then any saved_raid_disk
5841 * information must be scrapped
5843 if (!mddev->degraded)
5844 rdev_for_each(rdev, rtmp, mddev)
5845 rdev->saved_raid_disk = -1;
5847 mddev->recovery = 0;
5848 /* flag recovery needed just to double check */
5849 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5850 md_new_event(mddev);
5853 /* Clear some bits that don't mean anything, but
5856 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5857 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5858 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5859 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5861 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5863 /* no recovery is running.
5864 * remove any failed drives, then
5865 * add spares if possible.
5866 * Spare are also removed and re-added, to allow
5867 * the personality to fail the re-add.
5870 if (mddev->reshape_position != MaxSector) {
5871 if (mddev->pers->check_reshape(mddev) != 0)
5872 /* Cannot proceed */
5874 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5875 } else if ((spares = remove_and_add_spares(mddev))) {
5876 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5877 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5878 } else if (mddev->recovery_cp < MaxSector) {
5879 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5880 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5881 /* nothing to be done ... */
5884 if (mddev->pers->sync_request) {
5885 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5886 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5887 /* We are adding a device or devices to an array
5888 * which has the bitmap stored on all devices.
5889 * So make sure all bitmap pages get written
5891 bitmap_write_all(mddev->bitmap);
5893 mddev->sync_thread = md_register_thread(md_do_sync,
5896 if (!mddev->sync_thread) {
5897 printk(KERN_ERR "%s: could not start resync"
5900 /* leave the spares where they are, it shouldn't hurt */
5901 mddev->recovery = 0;
5903 md_wakeup_thread(mddev->sync_thread);
5904 md_new_event(mddev);
5907 mddev_unlock(mddev);
5911 static int md_notify_reboot(struct notifier_block *this,
5912 unsigned long code, void *x)
5914 struct list_head *tmp;
5917 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5919 printk(KERN_INFO "md: stopping all md devices.\n");
5921 for_each_mddev(mddev, tmp)
5922 if (mddev_trylock(mddev)) {
5923 do_md_stop (mddev, 1);
5924 mddev_unlock(mddev);
5927 * certain more exotic SCSI devices are known to be
5928 * volatile wrt too early system reboots. While the
5929 * right place to handle this issue is the given
5930 * driver, we do want to have a safe RAID driver ...
5937 static struct notifier_block md_notifier = {
5938 .notifier_call = md_notify_reboot,
5940 .priority = INT_MAX, /* before any real devices */
5943 static void md_geninit(void)
5945 struct proc_dir_entry *p;
5947 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5949 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5951 p->proc_fops = &md_seq_fops;
5954 static int __init md_init(void)
5956 if (register_blkdev(MAJOR_NR, "md"))
5958 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5959 unregister_blkdev(MAJOR_NR, "md");
5962 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5963 md_probe, NULL, NULL);
5964 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5965 md_probe, NULL, NULL);
5967 register_reboot_notifier(&md_notifier);
5968 raid_table_header = register_sysctl_table(raid_root_table);
5978 * Searches all registered partitions for autorun RAID arrays
5982 static LIST_HEAD(all_detected_devices);
5983 struct detected_devices_node {
5984 struct list_head list;
5988 void md_autodetect_dev(dev_t dev)
5990 struct detected_devices_node *node_detected_dev;
5992 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5993 if (node_detected_dev) {
5994 node_detected_dev->dev = dev;
5995 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5997 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5998 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6003 static void autostart_arrays(int part)
6006 struct detected_devices_node *node_detected_dev;
6008 int i_scanned, i_passed;
6013 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6015 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6017 node_detected_dev = list_entry(all_detected_devices.next,
6018 struct detected_devices_node, list);
6019 list_del(&node_detected_dev->list);
6020 dev = node_detected_dev->dev;
6021 kfree(node_detected_dev);
6022 rdev = md_import_device(dev,0, 90);
6026 if (test_bit(Faulty, &rdev->flags)) {
6030 set_bit(AutoDetected, &rdev->flags);
6031 list_add(&rdev->same_set, &pending_raid_disks);
6035 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6036 i_scanned, i_passed);
6038 autorun_devices(part);
6041 #endif /* !MODULE */
6043 static __exit void md_exit(void)
6046 struct list_head *tmp;
6048 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6049 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6051 unregister_blkdev(MAJOR_NR,"md");
6052 unregister_blkdev(mdp_major, "mdp");
6053 unregister_reboot_notifier(&md_notifier);
6054 unregister_sysctl_table(raid_table_header);
6055 remove_proc_entry("mdstat", NULL);
6056 for_each_mddev(mddev, tmp) {
6057 struct gendisk *disk = mddev->gendisk;
6060 export_array(mddev);
6063 mddev->gendisk = NULL;
6068 subsys_initcall(md_init);
6069 module_exit(md_exit)
6071 static int get_ro(char *buffer, struct kernel_param *kp)
6073 return sprintf(buffer, "%d", start_readonly);
6075 static int set_ro(const char *val, struct kernel_param *kp)
6078 int num = simple_strtoul(val, &e, 10);
6079 if (*val && (*e == '\0' || *e == '\n')) {
6080 start_readonly = num;
6086 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6087 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6090 EXPORT_SYMBOL(register_md_personality);
6091 EXPORT_SYMBOL(unregister_md_personality);
6092 EXPORT_SYMBOL(md_error);
6093 EXPORT_SYMBOL(md_done_sync);
6094 EXPORT_SYMBOL(md_write_start);
6095 EXPORT_SYMBOL(md_write_end);
6096 EXPORT_SYMBOL(md_register_thread);
6097 EXPORT_SYMBOL(md_unregister_thread);
6098 EXPORT_SYMBOL(md_wakeup_thread);
6099 EXPORT_SYMBOL(md_check_recovery);
6100 MODULE_LICENSE("GPL");
6102 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);