2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part);
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
75 static void md_print_devices(void);
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
100 static inline int speed_max(mddev_t *mddev)
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
106 static struct ctl_table_header *raid_table_header;
108 static ctl_table raid_table[] = {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = &proc_dointvec,
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = &proc_dointvec,
128 static ctl_table raid_dir_table[] = {
130 .ctl_name = DEV_RAID,
133 .mode = S_IRUGO|S_IXUGO,
139 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static struct block_device_operations md_fops;
152 static int start_readonly;
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
172 EXPORT_SYMBOL_GPL(md_new_event);
174 /* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
177 static void md_new_event_inintr(mddev_t *mddev)
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
198 #define for_each_mddev(mddev,tmp) \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
214 static int md_fail_request (struct request_queue *q, struct bio *bio)
220 static inline mddev_t *mddev_get(mddev_t *mddev)
222 atomic_inc(&mddev->active);
226 static void mddev_put(mddev_t *mddev)
228 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
230 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231 list_del(&mddev->all_mddevs);
232 spin_unlock(&all_mddevs_lock);
233 blk_cleanup_queue(mddev->queue);
234 kobject_put(&mddev->kobj);
236 spin_unlock(&all_mddevs_lock);
239 static mddev_t * mddev_find(dev_t unit)
241 mddev_t *mddev, *new = NULL;
244 spin_lock(&all_mddevs_lock);
245 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246 if (mddev->unit == unit) {
248 spin_unlock(&all_mddevs_lock);
254 list_add(&new->all_mddevs, &all_mddevs);
255 spin_unlock(&all_mddevs_lock);
258 spin_unlock(&all_mddevs_lock);
260 new = kzalloc(sizeof(*new), GFP_KERNEL);
265 if (MAJOR(unit) == MD_MAJOR)
266 new->md_minor = MINOR(unit);
268 new->md_minor = MINOR(unit) >> MdpMinorShift;
270 mutex_init(&new->reconfig_mutex);
271 INIT_LIST_HEAD(&new->disks);
272 INIT_LIST_HEAD(&new->all_mddevs);
273 init_timer(&new->safemode_timer);
274 atomic_set(&new->active, 1);
275 spin_lock_init(&new->write_lock);
276 init_waitqueue_head(&new->sb_wait);
277 new->reshape_position = MaxSector;
278 new->resync_max = MaxSector;
279 new->level = LEVEL_NONE;
281 new->queue = blk_alloc_queue(GFP_KERNEL);
286 /* Can be unlocked because the queue is new: no concurrency */
287 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
289 blk_queue_make_request(new->queue, md_fail_request);
294 static inline int mddev_lock(mddev_t * mddev)
296 return mutex_lock_interruptible(&mddev->reconfig_mutex);
299 static inline int mddev_trylock(mddev_t * mddev)
301 return mutex_trylock(&mddev->reconfig_mutex);
304 static inline void mddev_unlock(mddev_t * mddev)
306 mutex_unlock(&mddev->reconfig_mutex);
308 md_wakeup_thread(mddev->thread);
311 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
314 struct list_head *tmp;
316 rdev_for_each(rdev, tmp, mddev) {
317 if (rdev->desc_nr == nr)
323 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
325 struct list_head *tmp;
328 rdev_for_each(rdev, tmp, mddev) {
329 if (rdev->bdev->bd_dev == dev)
335 static struct mdk_personality *find_pers(int level, char *clevel)
337 struct mdk_personality *pers;
338 list_for_each_entry(pers, &pers_list, list) {
339 if (level != LEVEL_NONE && pers->level == level)
341 if (strcmp(pers->name, clevel)==0)
347 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
349 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
350 return MD_NEW_SIZE_BLOCKS(size);
353 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
357 size = rdev->sb_offset;
360 size &= ~((sector_t)chunk_size/1024 - 1);
364 static int alloc_disk_sb(mdk_rdev_t * rdev)
369 rdev->sb_page = alloc_page(GFP_KERNEL);
370 if (!rdev->sb_page) {
371 printk(KERN_ALERT "md: out of memory.\n");
378 static void free_disk_sb(mdk_rdev_t * rdev)
381 put_page(rdev->sb_page);
383 rdev->sb_page = NULL;
390 static void super_written(struct bio *bio, int error)
392 mdk_rdev_t *rdev = bio->bi_private;
393 mddev_t *mddev = rdev->mddev;
395 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
396 printk("md: super_written gets error=%d, uptodate=%d\n",
397 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
398 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
399 md_error(mddev, rdev);
402 if (atomic_dec_and_test(&mddev->pending_writes))
403 wake_up(&mddev->sb_wait);
407 static void super_written_barrier(struct bio *bio, int error)
409 struct bio *bio2 = bio->bi_private;
410 mdk_rdev_t *rdev = bio2->bi_private;
411 mddev_t *mddev = rdev->mddev;
413 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
414 error == -EOPNOTSUPP) {
416 /* barriers don't appear to be supported :-( */
417 set_bit(BarriersNotsupp, &rdev->flags);
418 mddev->barriers_work = 0;
419 spin_lock_irqsave(&mddev->write_lock, flags);
420 bio2->bi_next = mddev->biolist;
421 mddev->biolist = bio2;
422 spin_unlock_irqrestore(&mddev->write_lock, flags);
423 wake_up(&mddev->sb_wait);
427 bio->bi_private = rdev;
428 super_written(bio, error);
432 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
433 sector_t sector, int size, struct page *page)
435 /* write first size bytes of page to sector of rdev
436 * Increment mddev->pending_writes before returning
437 * and decrement it on completion, waking up sb_wait
438 * if zero is reached.
439 * If an error occurred, call md_error
441 * As we might need to resubmit the request if BIO_RW_BARRIER
442 * causes ENOTSUPP, we allocate a spare bio...
444 struct bio *bio = bio_alloc(GFP_NOIO, 1);
445 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
447 bio->bi_bdev = rdev->bdev;
448 bio->bi_sector = sector;
449 bio_add_page(bio, page, size, 0);
450 bio->bi_private = rdev;
451 bio->bi_end_io = super_written;
454 atomic_inc(&mddev->pending_writes);
455 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
457 rw |= (1<<BIO_RW_BARRIER);
458 rbio = bio_clone(bio, GFP_NOIO);
459 rbio->bi_private = bio;
460 rbio->bi_end_io = super_written_barrier;
461 submit_bio(rw, rbio);
466 void md_super_wait(mddev_t *mddev)
468 /* wait for all superblock writes that were scheduled to complete.
469 * if any had to be retried (due to BARRIER problems), retry them
473 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
474 if (atomic_read(&mddev->pending_writes)==0)
476 while (mddev->biolist) {
478 spin_lock_irq(&mddev->write_lock);
479 bio = mddev->biolist;
480 mddev->biolist = bio->bi_next ;
482 spin_unlock_irq(&mddev->write_lock);
483 submit_bio(bio->bi_rw, bio);
487 finish_wait(&mddev->sb_wait, &wq);
490 static void bi_complete(struct bio *bio, int error)
492 complete((struct completion*)bio->bi_private);
495 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
496 struct page *page, int rw)
498 struct bio *bio = bio_alloc(GFP_NOIO, 1);
499 struct completion event;
502 rw |= (1 << BIO_RW_SYNC);
505 bio->bi_sector = sector;
506 bio_add_page(bio, page, size, 0);
507 init_completion(&event);
508 bio->bi_private = &event;
509 bio->bi_end_io = bi_complete;
511 wait_for_completion(&event);
513 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
517 EXPORT_SYMBOL_GPL(sync_page_io);
519 static int read_disk_sb(mdk_rdev_t * rdev, int size)
521 char b[BDEVNAME_SIZE];
522 if (!rdev->sb_page) {
530 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
536 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
537 bdevname(rdev->bdev,b));
541 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
543 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
544 (sb1->set_uuid1 == sb2->set_uuid1) &&
545 (sb1->set_uuid2 == sb2->set_uuid2) &&
546 (sb1->set_uuid3 == sb2->set_uuid3))
554 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
557 mdp_super_t *tmp1, *tmp2;
559 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
560 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
562 if (!tmp1 || !tmp2) {
564 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
572 * nr_disks is not constant
577 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
589 static u32 md_csum_fold(u32 csum)
591 csum = (csum & 0xffff) + (csum >> 16);
592 return (csum & 0xffff) + (csum >> 16);
595 static unsigned int calc_sb_csum(mdp_super_t * sb)
598 u32 *sb32 = (u32*)sb;
600 unsigned int disk_csum, csum;
602 disk_csum = sb->sb_csum;
605 for (i = 0; i < MD_SB_BYTES/4 ; i++)
607 csum = (newcsum & 0xffffffff) + (newcsum>>32);
611 /* This used to use csum_partial, which was wrong for several
612 * reasons including that different results are returned on
613 * different architectures. It isn't critical that we get exactly
614 * the same return value as before (we always csum_fold before
615 * testing, and that removes any differences). However as we
616 * know that csum_partial always returned a 16bit value on
617 * alphas, do a fold to maximise conformity to previous behaviour.
619 sb->sb_csum = md_csum_fold(disk_csum);
621 sb->sb_csum = disk_csum;
628 * Handle superblock details.
629 * We want to be able to handle multiple superblock formats
630 * so we have a common interface to them all, and an array of
631 * different handlers.
632 * We rely on user-space to write the initial superblock, and support
633 * reading and updating of superblocks.
634 * Interface methods are:
635 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
636 * loads and validates a superblock on dev.
637 * if refdev != NULL, compare superblocks on both devices
639 * 0 - dev has a superblock that is compatible with refdev
640 * 1 - dev has a superblock that is compatible and newer than refdev
641 * so dev should be used as the refdev in future
642 * -EINVAL superblock incompatible or invalid
643 * -othererror e.g. -EIO
645 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
646 * Verify that dev is acceptable into mddev.
647 * The first time, mddev->raid_disks will be 0, and data from
648 * dev should be merged in. Subsequent calls check that dev
649 * is new enough. Return 0 or -EINVAL
651 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
652 * Update the superblock for rdev with data in mddev
653 * This does not write to disc.
659 struct module *owner;
660 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
661 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
662 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
666 * load_super for 0.90.0
668 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
670 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
676 * Calculate the position of the superblock,
677 * it's at the end of the disk.
679 * It also happens to be a multiple of 4Kb.
681 sb_offset = calc_dev_sboffset(rdev->bdev);
682 rdev->sb_offset = sb_offset;
684 ret = read_disk_sb(rdev, MD_SB_BYTES);
689 bdevname(rdev->bdev, b);
690 sb = (mdp_super_t*)page_address(rdev->sb_page);
692 if (sb->md_magic != MD_SB_MAGIC) {
693 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
698 if (sb->major_version != 0 ||
699 sb->minor_version < 90 ||
700 sb->minor_version > 91) {
701 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
702 sb->major_version, sb->minor_version,
707 if (sb->raid_disks <= 0)
710 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
711 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
716 rdev->preferred_minor = sb->md_minor;
717 rdev->data_offset = 0;
718 rdev->sb_size = MD_SB_BYTES;
720 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
721 if (sb->level != 1 && sb->level != 4
722 && sb->level != 5 && sb->level != 6
723 && sb->level != 10) {
724 /* FIXME use a better test */
726 "md: bitmaps not supported for this level.\n");
731 if (sb->level == LEVEL_MULTIPATH)
734 rdev->desc_nr = sb->this_disk.number;
740 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
741 if (!uuid_equal(refsb, sb)) {
742 printk(KERN_WARNING "md: %s has different UUID to %s\n",
743 b, bdevname(refdev->bdev,b2));
746 if (!sb_equal(refsb, sb)) {
747 printk(KERN_WARNING "md: %s has same UUID"
748 " but different superblock to %s\n",
749 b, bdevname(refdev->bdev, b2));
753 ev2 = md_event(refsb);
759 rdev->size = calc_dev_size(rdev, sb->chunk_size);
761 if (rdev->size < sb->size && sb->level > 1)
762 /* "this cannot possibly happen" ... */
770 * validate_super for 0.90.0
772 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
775 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
776 __u64 ev1 = md_event(sb);
778 rdev->raid_disk = -1;
779 clear_bit(Faulty, &rdev->flags);
780 clear_bit(In_sync, &rdev->flags);
781 clear_bit(WriteMostly, &rdev->flags);
782 clear_bit(BarriersNotsupp, &rdev->flags);
784 if (mddev->raid_disks == 0) {
785 mddev->major_version = 0;
786 mddev->minor_version = sb->minor_version;
787 mddev->patch_version = sb->patch_version;
789 mddev->chunk_size = sb->chunk_size;
790 mddev->ctime = sb->ctime;
791 mddev->utime = sb->utime;
792 mddev->level = sb->level;
793 mddev->clevel[0] = 0;
794 mddev->layout = sb->layout;
795 mddev->raid_disks = sb->raid_disks;
796 mddev->size = sb->size;
798 mddev->bitmap_offset = 0;
799 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
801 if (mddev->minor_version >= 91) {
802 mddev->reshape_position = sb->reshape_position;
803 mddev->delta_disks = sb->delta_disks;
804 mddev->new_level = sb->new_level;
805 mddev->new_layout = sb->new_layout;
806 mddev->new_chunk = sb->new_chunk;
808 mddev->reshape_position = MaxSector;
809 mddev->delta_disks = 0;
810 mddev->new_level = mddev->level;
811 mddev->new_layout = mddev->layout;
812 mddev->new_chunk = mddev->chunk_size;
815 if (sb->state & (1<<MD_SB_CLEAN))
816 mddev->recovery_cp = MaxSector;
818 if (sb->events_hi == sb->cp_events_hi &&
819 sb->events_lo == sb->cp_events_lo) {
820 mddev->recovery_cp = sb->recovery_cp;
822 mddev->recovery_cp = 0;
825 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
826 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
827 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
828 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
830 mddev->max_disks = MD_SB_DISKS;
832 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
833 mddev->bitmap_file == NULL)
834 mddev->bitmap_offset = mddev->default_bitmap_offset;
836 } else if (mddev->pers == NULL) {
837 /* Insist on good event counter while assembling */
839 if (ev1 < mddev->events)
841 } else if (mddev->bitmap) {
842 /* if adding to array with a bitmap, then we can accept an
843 * older device ... but not too old.
845 if (ev1 < mddev->bitmap->events_cleared)
848 if (ev1 < mddev->events)
849 /* just a hot-add of a new device, leave raid_disk at -1 */
853 if (mddev->level != LEVEL_MULTIPATH) {
854 desc = sb->disks + rdev->desc_nr;
856 if (desc->state & (1<<MD_DISK_FAULTY))
857 set_bit(Faulty, &rdev->flags);
858 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
859 desc->raid_disk < mddev->raid_disks */) {
860 set_bit(In_sync, &rdev->flags);
861 rdev->raid_disk = desc->raid_disk;
863 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
864 set_bit(WriteMostly, &rdev->flags);
865 } else /* MULTIPATH are always insync */
866 set_bit(In_sync, &rdev->flags);
871 * sync_super for 0.90.0
873 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
876 struct list_head *tmp;
878 int next_spare = mddev->raid_disks;
881 /* make rdev->sb match mddev data..
884 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
885 * 3/ any empty disks < next_spare become removed
887 * disks[0] gets initialised to REMOVED because
888 * we cannot be sure from other fields if it has
889 * been initialised or not.
892 int active=0, working=0,failed=0,spare=0,nr_disks=0;
894 rdev->sb_size = MD_SB_BYTES;
896 sb = (mdp_super_t*)page_address(rdev->sb_page);
898 memset(sb, 0, sizeof(*sb));
900 sb->md_magic = MD_SB_MAGIC;
901 sb->major_version = mddev->major_version;
902 sb->patch_version = mddev->patch_version;
903 sb->gvalid_words = 0; /* ignored */
904 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
905 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
906 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
907 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
909 sb->ctime = mddev->ctime;
910 sb->level = mddev->level;
911 sb->size = mddev->size;
912 sb->raid_disks = mddev->raid_disks;
913 sb->md_minor = mddev->md_minor;
914 sb->not_persistent = 0;
915 sb->utime = mddev->utime;
917 sb->events_hi = (mddev->events>>32);
918 sb->events_lo = (u32)mddev->events;
920 if (mddev->reshape_position == MaxSector)
921 sb->minor_version = 90;
923 sb->minor_version = 91;
924 sb->reshape_position = mddev->reshape_position;
925 sb->new_level = mddev->new_level;
926 sb->delta_disks = mddev->delta_disks;
927 sb->new_layout = mddev->new_layout;
928 sb->new_chunk = mddev->new_chunk;
930 mddev->minor_version = sb->minor_version;
933 sb->recovery_cp = mddev->recovery_cp;
934 sb->cp_events_hi = (mddev->events>>32);
935 sb->cp_events_lo = (u32)mddev->events;
936 if (mddev->recovery_cp == MaxSector)
937 sb->state = (1<< MD_SB_CLEAN);
941 sb->layout = mddev->layout;
942 sb->chunk_size = mddev->chunk_size;
944 if (mddev->bitmap && mddev->bitmap_file == NULL)
945 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
947 sb->disks[0].state = (1<<MD_DISK_REMOVED);
948 rdev_for_each(rdev2, tmp, mddev) {
951 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
952 && !test_bit(Faulty, &rdev2->flags))
953 desc_nr = rdev2->raid_disk;
955 desc_nr = next_spare++;
956 rdev2->desc_nr = desc_nr;
957 d = &sb->disks[rdev2->desc_nr];
959 d->number = rdev2->desc_nr;
960 d->major = MAJOR(rdev2->bdev->bd_dev);
961 d->minor = MINOR(rdev2->bdev->bd_dev);
962 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
963 && !test_bit(Faulty, &rdev2->flags))
964 d->raid_disk = rdev2->raid_disk;
966 d->raid_disk = rdev2->desc_nr; /* compatibility */
967 if (test_bit(Faulty, &rdev2->flags))
968 d->state = (1<<MD_DISK_FAULTY);
969 else if (test_bit(In_sync, &rdev2->flags)) {
970 d->state = (1<<MD_DISK_ACTIVE);
971 d->state |= (1<<MD_DISK_SYNC);
979 if (test_bit(WriteMostly, &rdev2->flags))
980 d->state |= (1<<MD_DISK_WRITEMOSTLY);
982 /* now set the "removed" and "faulty" bits on any missing devices */
983 for (i=0 ; i < mddev->raid_disks ; i++) {
984 mdp_disk_t *d = &sb->disks[i];
985 if (d->state == 0 && d->number == 0) {
988 d->state = (1<<MD_DISK_REMOVED);
989 d->state |= (1<<MD_DISK_FAULTY);
993 sb->nr_disks = nr_disks;
994 sb->active_disks = active;
995 sb->working_disks = working;
996 sb->failed_disks = failed;
997 sb->spare_disks = spare;
999 sb->this_disk = sb->disks[rdev->desc_nr];
1000 sb->sb_csum = calc_sb_csum(sb);
1004 * version 1 superblock
1007 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1011 unsigned long long newcsum;
1012 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1013 __le32 *isuper = (__le32*)sb;
1016 disk_csum = sb->sb_csum;
1019 for (i=0; size>=4; size -= 4 )
1020 newcsum += le32_to_cpu(*isuper++);
1023 newcsum += le16_to_cpu(*(__le16*) isuper);
1025 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1026 sb->sb_csum = disk_csum;
1027 return cpu_to_le32(csum);
1030 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1032 struct mdp_superblock_1 *sb;
1035 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1039 * Calculate the position of the superblock.
1040 * It is always aligned to a 4K boundary and
1041 * depeding on minor_version, it can be:
1042 * 0: At least 8K, but less than 12K, from end of device
1043 * 1: At start of device
1044 * 2: 4K from start of device.
1046 switch(minor_version) {
1048 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1050 sb_offset &= ~(sector_t)(4*2-1);
1051 /* convert from sectors to K */
1063 rdev->sb_offset = sb_offset;
1065 /* superblock is rarely larger than 1K, but it can be larger,
1066 * and it is safe to read 4k, so we do that
1068 ret = read_disk_sb(rdev, 4096);
1069 if (ret) return ret;
1072 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1074 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1075 sb->major_version != cpu_to_le32(1) ||
1076 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1077 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1078 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1081 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1082 printk("md: invalid superblock checksum on %s\n",
1083 bdevname(rdev->bdev,b));
1086 if (le64_to_cpu(sb->data_size) < 10) {
1087 printk("md: data_size too small on %s\n",
1088 bdevname(rdev->bdev,b));
1091 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1092 if (sb->level != cpu_to_le32(1) &&
1093 sb->level != cpu_to_le32(4) &&
1094 sb->level != cpu_to_le32(5) &&
1095 sb->level != cpu_to_le32(6) &&
1096 sb->level != cpu_to_le32(10)) {
1098 "md: bitmaps not supported for this level.\n");
1103 rdev->preferred_minor = 0xffff;
1104 rdev->data_offset = le64_to_cpu(sb->data_offset);
1105 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1107 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1108 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1109 if (rdev->sb_size & bmask)
1110 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1113 && rdev->data_offset < sb_offset + (rdev->sb_size/512))
1116 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1119 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1125 struct mdp_superblock_1 *refsb =
1126 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1128 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1129 sb->level != refsb->level ||
1130 sb->layout != refsb->layout ||
1131 sb->chunksize != refsb->chunksize) {
1132 printk(KERN_WARNING "md: %s has strangely different"
1133 " superblock to %s\n",
1134 bdevname(rdev->bdev,b),
1135 bdevname(refdev->bdev,b2));
1138 ev1 = le64_to_cpu(sb->events);
1139 ev2 = le64_to_cpu(refsb->events);
1147 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1149 rdev->size = rdev->sb_offset;
1150 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1152 rdev->size = le64_to_cpu(sb->data_size)/2;
1153 if (le32_to_cpu(sb->chunksize))
1154 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1156 if (le64_to_cpu(sb->size) > rdev->size*2)
1161 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1163 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1164 __u64 ev1 = le64_to_cpu(sb->events);
1166 rdev->raid_disk = -1;
1167 clear_bit(Faulty, &rdev->flags);
1168 clear_bit(In_sync, &rdev->flags);
1169 clear_bit(WriteMostly, &rdev->flags);
1170 clear_bit(BarriersNotsupp, &rdev->flags);
1172 if (mddev->raid_disks == 0) {
1173 mddev->major_version = 1;
1174 mddev->patch_version = 0;
1175 mddev->external = 0;
1176 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1177 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1178 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1179 mddev->level = le32_to_cpu(sb->level);
1180 mddev->clevel[0] = 0;
1181 mddev->layout = le32_to_cpu(sb->layout);
1182 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1183 mddev->size = le64_to_cpu(sb->size)/2;
1184 mddev->events = ev1;
1185 mddev->bitmap_offset = 0;
1186 mddev->default_bitmap_offset = 1024 >> 9;
1188 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1189 memcpy(mddev->uuid, sb->set_uuid, 16);
1191 mddev->max_disks = (4096-256)/2;
1193 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1194 mddev->bitmap_file == NULL )
1195 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1197 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1198 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1199 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1200 mddev->new_level = le32_to_cpu(sb->new_level);
1201 mddev->new_layout = le32_to_cpu(sb->new_layout);
1202 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1204 mddev->reshape_position = MaxSector;
1205 mddev->delta_disks = 0;
1206 mddev->new_level = mddev->level;
1207 mddev->new_layout = mddev->layout;
1208 mddev->new_chunk = mddev->chunk_size;
1211 } else if (mddev->pers == NULL) {
1212 /* Insist of good event counter while assembling */
1214 if (ev1 < mddev->events)
1216 } else if (mddev->bitmap) {
1217 /* If adding to array with a bitmap, then we can accept an
1218 * older device, but not too old.
1220 if (ev1 < mddev->bitmap->events_cleared)
1223 if (ev1 < mddev->events)
1224 /* just a hot-add of a new device, leave raid_disk at -1 */
1227 if (mddev->level != LEVEL_MULTIPATH) {
1229 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1231 case 0xffff: /* spare */
1233 case 0xfffe: /* faulty */
1234 set_bit(Faulty, &rdev->flags);
1237 if ((le32_to_cpu(sb->feature_map) &
1238 MD_FEATURE_RECOVERY_OFFSET))
1239 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1241 set_bit(In_sync, &rdev->flags);
1242 rdev->raid_disk = role;
1245 if (sb->devflags & WriteMostly1)
1246 set_bit(WriteMostly, &rdev->flags);
1247 } else /* MULTIPATH are always insync */
1248 set_bit(In_sync, &rdev->flags);
1253 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1255 struct mdp_superblock_1 *sb;
1256 struct list_head *tmp;
1259 /* make rdev->sb match mddev and rdev data. */
1261 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1263 sb->feature_map = 0;
1265 sb->recovery_offset = cpu_to_le64(0);
1266 memset(sb->pad1, 0, sizeof(sb->pad1));
1267 memset(sb->pad2, 0, sizeof(sb->pad2));
1268 memset(sb->pad3, 0, sizeof(sb->pad3));
1270 sb->utime = cpu_to_le64((__u64)mddev->utime);
1271 sb->events = cpu_to_le64(mddev->events);
1273 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1275 sb->resync_offset = cpu_to_le64(0);
1277 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1279 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1280 sb->size = cpu_to_le64(mddev->size<<1);
1282 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1283 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1284 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1287 if (rdev->raid_disk >= 0 &&
1288 !test_bit(In_sync, &rdev->flags) &&
1289 rdev->recovery_offset > 0) {
1290 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1291 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1294 if (mddev->reshape_position != MaxSector) {
1295 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1296 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1297 sb->new_layout = cpu_to_le32(mddev->new_layout);
1298 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1299 sb->new_level = cpu_to_le32(mddev->new_level);
1300 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1304 rdev_for_each(rdev2, tmp, mddev)
1305 if (rdev2->desc_nr+1 > max_dev)
1306 max_dev = rdev2->desc_nr+1;
1308 if (max_dev > le32_to_cpu(sb->max_dev))
1309 sb->max_dev = cpu_to_le32(max_dev);
1310 for (i=0; i<max_dev;i++)
1311 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1313 rdev_for_each(rdev2, tmp, mddev) {
1315 if (test_bit(Faulty, &rdev2->flags))
1316 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1317 else if (test_bit(In_sync, &rdev2->flags))
1318 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1319 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1320 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1322 sb->dev_roles[i] = cpu_to_le16(0xffff);
1325 sb->sb_csum = calc_sb_1_csum(sb);
1329 static struct super_type super_types[] = {
1332 .owner = THIS_MODULE,
1333 .load_super = super_90_load,
1334 .validate_super = super_90_validate,
1335 .sync_super = super_90_sync,
1339 .owner = THIS_MODULE,
1340 .load_super = super_1_load,
1341 .validate_super = super_1_validate,
1342 .sync_super = super_1_sync,
1346 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1348 struct list_head *tmp, *tmp2;
1349 mdk_rdev_t *rdev, *rdev2;
1351 rdev_for_each(rdev, tmp, mddev1)
1352 rdev_for_each(rdev2, tmp2, mddev2)
1353 if (rdev->bdev->bd_contains ==
1354 rdev2->bdev->bd_contains)
1360 static LIST_HEAD(pending_raid_disks);
1362 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1364 char b[BDEVNAME_SIZE];
1373 /* make sure rdev->size exceeds mddev->size */
1374 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1376 /* Cannot change size, so fail
1377 * If mddev->level <= 0, then we don't care
1378 * about aligning sizes (e.g. linear)
1380 if (mddev->level > 0)
1383 mddev->size = rdev->size;
1386 /* Verify rdev->desc_nr is unique.
1387 * If it is -1, assign a free number, else
1388 * check number is not in use
1390 if (rdev->desc_nr < 0) {
1392 if (mddev->pers) choice = mddev->raid_disks;
1393 while (find_rdev_nr(mddev, choice))
1395 rdev->desc_nr = choice;
1397 if (find_rdev_nr(mddev, rdev->desc_nr))
1400 bdevname(rdev->bdev,b);
1401 while ( (s=strchr(b, '/')) != NULL)
1404 rdev->mddev = mddev;
1405 printk(KERN_INFO "md: bind<%s>\n", b);
1407 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1410 if (rdev->bdev->bd_part)
1411 ko = &rdev->bdev->bd_part->dev.kobj;
1413 ko = &rdev->bdev->bd_disk->dev.kobj;
1414 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1415 kobject_del(&rdev->kobj);
1418 list_add(&rdev->same_set, &mddev->disks);
1419 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1423 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1428 static void md_delayed_delete(struct work_struct *ws)
1430 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1431 kobject_del(&rdev->kobj);
1432 kobject_put(&rdev->kobj);
1435 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1437 char b[BDEVNAME_SIZE];
1442 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1443 list_del_init(&rdev->same_set);
1444 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1446 sysfs_remove_link(&rdev->kobj, "block");
1448 /* We need to delay this, otherwise we can deadlock when
1449 * writing to 'remove' to "dev/state"
1451 INIT_WORK(&rdev->del_work, md_delayed_delete);
1452 kobject_get(&rdev->kobj);
1453 schedule_work(&rdev->del_work);
1457 * prevent the device from being mounted, repartitioned or
1458 * otherwise reused by a RAID array (or any other kernel
1459 * subsystem), by bd_claiming the device.
1461 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1464 struct block_device *bdev;
1465 char b[BDEVNAME_SIZE];
1467 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1469 printk(KERN_ERR "md: could not open %s.\n",
1470 __bdevname(dev, b));
1471 return PTR_ERR(bdev);
1473 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1475 printk(KERN_ERR "md: could not bd_claim %s.\n",
1481 set_bit(AllReserved, &rdev->flags);
1486 static void unlock_rdev(mdk_rdev_t *rdev)
1488 struct block_device *bdev = rdev->bdev;
1496 void md_autodetect_dev(dev_t dev);
1498 static void export_rdev(mdk_rdev_t * rdev)
1500 char b[BDEVNAME_SIZE];
1501 printk(KERN_INFO "md: export_rdev(%s)\n",
1502 bdevname(rdev->bdev,b));
1506 list_del_init(&rdev->same_set);
1508 if (test_bit(AutoDetected, &rdev->flags))
1509 md_autodetect_dev(rdev->bdev->bd_dev);
1512 kobject_put(&rdev->kobj);
1515 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1517 unbind_rdev_from_array(rdev);
1521 static void export_array(mddev_t *mddev)
1523 struct list_head *tmp;
1526 rdev_for_each(rdev, tmp, mddev) {
1531 kick_rdev_from_array(rdev);
1533 if (!list_empty(&mddev->disks))
1535 mddev->raid_disks = 0;
1536 mddev->major_version = 0;
1539 static void print_desc(mdp_disk_t *desc)
1541 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1542 desc->major,desc->minor,desc->raid_disk,desc->state);
1545 static void print_sb(mdp_super_t *sb)
1550 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1551 sb->major_version, sb->minor_version, sb->patch_version,
1552 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1554 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1555 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1556 sb->md_minor, sb->layout, sb->chunk_size);
1557 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1558 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1559 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1560 sb->failed_disks, sb->spare_disks,
1561 sb->sb_csum, (unsigned long)sb->events_lo);
1564 for (i = 0; i < MD_SB_DISKS; i++) {
1567 desc = sb->disks + i;
1568 if (desc->number || desc->major || desc->minor ||
1569 desc->raid_disk || (desc->state && (desc->state != 4))) {
1570 printk(" D %2d: ", i);
1574 printk(KERN_INFO "md: THIS: ");
1575 print_desc(&sb->this_disk);
1579 static void print_rdev(mdk_rdev_t *rdev)
1581 char b[BDEVNAME_SIZE];
1582 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1583 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1584 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1586 if (rdev->sb_loaded) {
1587 printk(KERN_INFO "md: rdev superblock:\n");
1588 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1590 printk(KERN_INFO "md: no rdev superblock!\n");
1593 static void md_print_devices(void)
1595 struct list_head *tmp, *tmp2;
1598 char b[BDEVNAME_SIZE];
1601 printk("md: **********************************\n");
1602 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1603 printk("md: **********************************\n");
1604 for_each_mddev(mddev, tmp) {
1607 bitmap_print_sb(mddev->bitmap);
1609 printk("%s: ", mdname(mddev));
1610 rdev_for_each(rdev, tmp2, mddev)
1611 printk("<%s>", bdevname(rdev->bdev,b));
1614 rdev_for_each(rdev, tmp2, mddev)
1617 printk("md: **********************************\n");
1622 static void sync_sbs(mddev_t * mddev, int nospares)
1624 /* Update each superblock (in-memory image), but
1625 * if we are allowed to, skip spares which already
1626 * have the right event counter, or have one earlier
1627 * (which would mean they aren't being marked as dirty
1628 * with the rest of the array)
1631 struct list_head *tmp;
1633 rdev_for_each(rdev, tmp, mddev) {
1634 if (rdev->sb_events == mddev->events ||
1636 rdev->raid_disk < 0 &&
1637 (rdev->sb_events&1)==0 &&
1638 rdev->sb_events+1 == mddev->events)) {
1639 /* Don't update this superblock */
1640 rdev->sb_loaded = 2;
1642 super_types[mddev->major_version].
1643 sync_super(mddev, rdev);
1644 rdev->sb_loaded = 1;
1649 static void md_update_sb(mddev_t * mddev, int force_change)
1651 struct list_head *tmp;
1656 if (mddev->external)
1659 spin_lock_irq(&mddev->write_lock);
1661 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1662 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1664 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1665 /* just a clean<-> dirty transition, possibly leave spares alone,
1666 * though if events isn't the right even/odd, we will have to do
1672 if (mddev->degraded)
1673 /* If the array is degraded, then skipping spares is both
1674 * dangerous and fairly pointless.
1675 * Dangerous because a device that was removed from the array
1676 * might have a event_count that still looks up-to-date,
1677 * so it can be re-added without a resync.
1678 * Pointless because if there are any spares to skip,
1679 * then a recovery will happen and soon that array won't
1680 * be degraded any more and the spare can go back to sleep then.
1684 sync_req = mddev->in_sync;
1685 mddev->utime = get_seconds();
1687 /* If this is just a dirty<->clean transition, and the array is clean
1688 * and 'events' is odd, we can roll back to the previous clean state */
1690 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1691 && (mddev->events & 1)
1692 && mddev->events != 1)
1695 /* otherwise we have to go forward and ... */
1697 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1698 /* .. if the array isn't clean, insist on an odd 'events' */
1699 if ((mddev->events&1)==0) {
1704 /* otherwise insist on an even 'events' (for clean states) */
1705 if ((mddev->events&1)) {
1712 if (!mddev->events) {
1714 * oops, this 64-bit counter should never wrap.
1715 * Either we are in around ~1 trillion A.C., assuming
1716 * 1 reboot per second, or we have a bug:
1723 * do not write anything to disk if using
1724 * nonpersistent superblocks
1726 if (!mddev->persistent) {
1727 if (!mddev->external)
1728 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1730 spin_unlock_irq(&mddev->write_lock);
1731 wake_up(&mddev->sb_wait);
1734 sync_sbs(mddev, nospares);
1735 spin_unlock_irq(&mddev->write_lock);
1738 "md: updating %s RAID superblock on device (in sync %d)\n",
1739 mdname(mddev),mddev->in_sync);
1741 bitmap_update_sb(mddev->bitmap);
1742 rdev_for_each(rdev, tmp, mddev) {
1743 char b[BDEVNAME_SIZE];
1744 dprintk(KERN_INFO "md: ");
1745 if (rdev->sb_loaded != 1)
1746 continue; /* no noise on spare devices */
1747 if (test_bit(Faulty, &rdev->flags))
1748 dprintk("(skipping faulty ");
1750 dprintk("%s ", bdevname(rdev->bdev,b));
1751 if (!test_bit(Faulty, &rdev->flags)) {
1752 md_super_write(mddev,rdev,
1753 rdev->sb_offset<<1, rdev->sb_size,
1755 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1756 bdevname(rdev->bdev,b),
1757 (unsigned long long)rdev->sb_offset);
1758 rdev->sb_events = mddev->events;
1762 if (mddev->level == LEVEL_MULTIPATH)
1763 /* only need to write one superblock... */
1766 md_super_wait(mddev);
1767 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1769 spin_lock_irq(&mddev->write_lock);
1770 if (mddev->in_sync != sync_req ||
1771 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1772 /* have to write it out again */
1773 spin_unlock_irq(&mddev->write_lock);
1776 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1777 spin_unlock_irq(&mddev->write_lock);
1778 wake_up(&mddev->sb_wait);
1782 /* words written to sysfs files may, or my not, be \n terminated.
1783 * We want to accept with case. For this we use cmd_match.
1785 static int cmd_match(const char *cmd, const char *str)
1787 /* See if cmd, written into a sysfs file, matches
1788 * str. They must either be the same, or cmd can
1789 * have a trailing newline
1791 while (*cmd && *str && *cmd == *str) {
1802 struct rdev_sysfs_entry {
1803 struct attribute attr;
1804 ssize_t (*show)(mdk_rdev_t *, char *);
1805 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1809 state_show(mdk_rdev_t *rdev, char *page)
1814 if (test_bit(Faulty, &rdev->flags)) {
1815 len+= sprintf(page+len, "%sfaulty",sep);
1818 if (test_bit(In_sync, &rdev->flags)) {
1819 len += sprintf(page+len, "%sin_sync",sep);
1822 if (test_bit(WriteMostly, &rdev->flags)) {
1823 len += sprintf(page+len, "%swrite_mostly",sep);
1826 if (!test_bit(Faulty, &rdev->flags) &&
1827 !test_bit(In_sync, &rdev->flags)) {
1828 len += sprintf(page+len, "%sspare", sep);
1831 return len+sprintf(page+len, "\n");
1835 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1838 * faulty - simulates and error
1839 * remove - disconnects the device
1840 * writemostly - sets write_mostly
1841 * -writemostly - clears write_mostly
1844 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1845 md_error(rdev->mddev, rdev);
1847 } else if (cmd_match(buf, "remove")) {
1848 if (rdev->raid_disk >= 0)
1851 mddev_t *mddev = rdev->mddev;
1852 kick_rdev_from_array(rdev);
1854 md_update_sb(mddev, 1);
1855 md_new_event(mddev);
1858 } else if (cmd_match(buf, "writemostly")) {
1859 set_bit(WriteMostly, &rdev->flags);
1861 } else if (cmd_match(buf, "-writemostly")) {
1862 clear_bit(WriteMostly, &rdev->flags);
1865 return err ? err : len;
1867 static struct rdev_sysfs_entry rdev_state =
1868 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1871 errors_show(mdk_rdev_t *rdev, char *page)
1873 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1877 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1880 unsigned long n = simple_strtoul(buf, &e, 10);
1881 if (*buf && (*e == 0 || *e == '\n')) {
1882 atomic_set(&rdev->corrected_errors, n);
1887 static struct rdev_sysfs_entry rdev_errors =
1888 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1891 slot_show(mdk_rdev_t *rdev, char *page)
1893 if (rdev->raid_disk < 0)
1894 return sprintf(page, "none\n");
1896 return sprintf(page, "%d\n", rdev->raid_disk);
1900 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1905 int slot = simple_strtoul(buf, &e, 10);
1906 if (strncmp(buf, "none", 4)==0)
1908 else if (e==buf || (*e && *e!= '\n'))
1910 if (rdev->mddev->pers) {
1911 /* Setting 'slot' on an active array requires also
1912 * updating the 'rd%d' link, and communicating
1913 * with the personality with ->hot_*_disk.
1914 * For now we only support removing
1915 * failed/spare devices. This normally happens automatically,
1916 * but not when the metadata is externally managed.
1920 if (rdev->raid_disk == -1)
1922 /* personality does all needed checks */
1923 if (rdev->mddev->pers->hot_add_disk == NULL)
1925 err = rdev->mddev->pers->
1926 hot_remove_disk(rdev->mddev, rdev->raid_disk);
1929 sprintf(nm, "rd%d", rdev->raid_disk);
1930 sysfs_remove_link(&rdev->mddev->kobj, nm);
1931 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1932 md_wakeup_thread(rdev->mddev->thread);
1934 if (slot >= rdev->mddev->raid_disks)
1936 rdev->raid_disk = slot;
1937 /* assume it is working */
1938 clear_bit(Faulty, &rdev->flags);
1939 clear_bit(WriteMostly, &rdev->flags);
1940 set_bit(In_sync, &rdev->flags);
1946 static struct rdev_sysfs_entry rdev_slot =
1947 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1950 offset_show(mdk_rdev_t *rdev, char *page)
1952 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1956 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1959 unsigned long long offset = simple_strtoull(buf, &e, 10);
1960 if (e==buf || (*e && *e != '\n'))
1962 if (rdev->mddev->pers)
1964 if (rdev->size && rdev->mddev->external)
1965 /* Must set offset before size, so overlap checks
1968 rdev->data_offset = offset;
1972 static struct rdev_sysfs_entry rdev_offset =
1973 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1976 rdev_size_show(mdk_rdev_t *rdev, char *page)
1978 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1981 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
1983 /* check if two start/length pairs overlap */
1992 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1995 unsigned long long size = simple_strtoull(buf, &e, 10);
1996 unsigned long long oldsize = rdev->size;
1997 mddev_t *my_mddev = rdev->mddev;
1999 if (e==buf || (*e && *e != '\n'))
2004 if (size > oldsize && rdev->mddev->external) {
2005 /* need to check that all other rdevs with the same ->bdev
2006 * do not overlap. We need to unlock the mddev to avoid
2007 * a deadlock. We have already changed rdev->size, and if
2008 * we have to change it back, we will have the lock again.
2012 struct list_head *tmp, *tmp2;
2014 mddev_unlock(my_mddev);
2015 for_each_mddev(mddev, tmp) {
2019 rdev_for_each(rdev2, tmp2, mddev)
2020 if (test_bit(AllReserved, &rdev2->flags) ||
2021 (rdev->bdev == rdev2->bdev &&
2023 overlaps(rdev->data_offset, rdev->size,
2024 rdev2->data_offset, rdev2->size))) {
2028 mddev_unlock(mddev);
2034 mddev_lock(my_mddev);
2036 /* Someone else could have slipped in a size
2037 * change here, but doing so is just silly.
2038 * We put oldsize back because we *know* it is
2039 * safe, and trust userspace not to race with
2042 rdev->size = oldsize;
2046 if (size < my_mddev->size || my_mddev->size == 0)
2047 my_mddev->size = size;
2051 static struct rdev_sysfs_entry rdev_size =
2052 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2054 static struct attribute *rdev_default_attrs[] = {
2063 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2065 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2066 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2067 mddev_t *mddev = rdev->mddev;
2073 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2075 if (rdev->mddev == NULL)
2078 rv = entry->show(rdev, page);
2079 mddev_unlock(mddev);
2085 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2086 const char *page, size_t length)
2088 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2089 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2091 mddev_t *mddev = rdev->mddev;
2095 if (!capable(CAP_SYS_ADMIN))
2097 rv = mddev ? mddev_lock(mddev): -EBUSY;
2099 if (rdev->mddev == NULL)
2102 rv = entry->store(rdev, page, length);
2103 mddev_unlock(mddev);
2108 static void rdev_free(struct kobject *ko)
2110 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2113 static struct sysfs_ops rdev_sysfs_ops = {
2114 .show = rdev_attr_show,
2115 .store = rdev_attr_store,
2117 static struct kobj_type rdev_ktype = {
2118 .release = rdev_free,
2119 .sysfs_ops = &rdev_sysfs_ops,
2120 .default_attrs = rdev_default_attrs,
2124 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2126 * mark the device faulty if:
2128 * - the device is nonexistent (zero size)
2129 * - the device has no valid superblock
2131 * a faulty rdev _never_ has rdev->sb set.
2133 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2135 char b[BDEVNAME_SIZE];
2140 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2142 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2143 return ERR_PTR(-ENOMEM);
2146 if ((err = alloc_disk_sb(rdev)))
2149 err = lock_rdev(rdev, newdev, super_format == -2);
2153 kobject_init(&rdev->kobj, &rdev_ktype);
2156 rdev->saved_raid_disk = -1;
2157 rdev->raid_disk = -1;
2159 rdev->data_offset = 0;
2160 rdev->sb_events = 0;
2161 atomic_set(&rdev->nr_pending, 0);
2162 atomic_set(&rdev->read_errors, 0);
2163 atomic_set(&rdev->corrected_errors, 0);
2165 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2168 "md: %s has zero or unknown size, marking faulty!\n",
2169 bdevname(rdev->bdev,b));
2174 if (super_format >= 0) {
2175 err = super_types[super_format].
2176 load_super(rdev, NULL, super_minor);
2177 if (err == -EINVAL) {
2179 "md: %s does not have a valid v%d.%d "
2180 "superblock, not importing!\n",
2181 bdevname(rdev->bdev,b),
2182 super_format, super_minor);
2187 "md: could not read %s's sb, not importing!\n",
2188 bdevname(rdev->bdev,b));
2192 INIT_LIST_HEAD(&rdev->same_set);
2197 if (rdev->sb_page) {
2203 return ERR_PTR(err);
2207 * Check a full RAID array for plausibility
2211 static void analyze_sbs(mddev_t * mddev)
2214 struct list_head *tmp;
2215 mdk_rdev_t *rdev, *freshest;
2216 char b[BDEVNAME_SIZE];
2219 rdev_for_each(rdev, tmp, mddev)
2220 switch (super_types[mddev->major_version].
2221 load_super(rdev, freshest, mddev->minor_version)) {
2229 "md: fatal superblock inconsistency in %s"
2230 " -- removing from array\n",
2231 bdevname(rdev->bdev,b));
2232 kick_rdev_from_array(rdev);
2236 super_types[mddev->major_version].
2237 validate_super(mddev, freshest);
2240 rdev_for_each(rdev, tmp, mddev) {
2241 if (rdev != freshest)
2242 if (super_types[mddev->major_version].
2243 validate_super(mddev, rdev)) {
2244 printk(KERN_WARNING "md: kicking non-fresh %s"
2246 bdevname(rdev->bdev,b));
2247 kick_rdev_from_array(rdev);
2250 if (mddev->level == LEVEL_MULTIPATH) {
2251 rdev->desc_nr = i++;
2252 rdev->raid_disk = rdev->desc_nr;
2253 set_bit(In_sync, &rdev->flags);
2254 } else if (rdev->raid_disk >= mddev->raid_disks) {
2255 rdev->raid_disk = -1;
2256 clear_bit(In_sync, &rdev->flags);
2262 if (mddev->recovery_cp != MaxSector &&
2264 printk(KERN_ERR "md: %s: raid array is not clean"
2265 " -- starting background reconstruction\n",
2271 safe_delay_show(mddev_t *mddev, char *page)
2273 int msec = (mddev->safemode_delay*1000)/HZ;
2274 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2277 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2285 /* remove a period, and count digits after it */
2286 if (len >= sizeof(buf))
2288 strlcpy(buf, cbuf, len);
2290 for (i=0; i<len; i++) {
2292 if (isdigit(buf[i])) {
2297 } else if (buf[i] == '.') {
2302 msec = simple_strtoul(buf, &e, 10);
2303 if (e == buf || (*e && *e != '\n'))
2305 msec = (msec * 1000) / scale;
2307 mddev->safemode_delay = 0;
2309 mddev->safemode_delay = (msec*HZ)/1000;
2310 if (mddev->safemode_delay == 0)
2311 mddev->safemode_delay = 1;
2315 static struct md_sysfs_entry md_safe_delay =
2316 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2319 level_show(mddev_t *mddev, char *page)
2321 struct mdk_personality *p = mddev->pers;
2323 return sprintf(page, "%s\n", p->name);
2324 else if (mddev->clevel[0])
2325 return sprintf(page, "%s\n", mddev->clevel);
2326 else if (mddev->level != LEVEL_NONE)
2327 return sprintf(page, "%d\n", mddev->level);
2333 level_store(mddev_t *mddev, const char *buf, size_t len)
2340 if (len >= sizeof(mddev->clevel))
2342 strncpy(mddev->clevel, buf, len);
2343 if (mddev->clevel[len-1] == '\n')
2345 mddev->clevel[len] = 0;
2346 mddev->level = LEVEL_NONE;
2350 static struct md_sysfs_entry md_level =
2351 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2355 layout_show(mddev_t *mddev, char *page)
2357 /* just a number, not meaningful for all levels */
2358 if (mddev->reshape_position != MaxSector &&
2359 mddev->layout != mddev->new_layout)
2360 return sprintf(page, "%d (%d)\n",
2361 mddev->new_layout, mddev->layout);
2362 return sprintf(page, "%d\n", mddev->layout);
2366 layout_store(mddev_t *mddev, const char *buf, size_t len)
2369 unsigned long n = simple_strtoul(buf, &e, 10);
2371 if (!*buf || (*e && *e != '\n'))
2376 if (mddev->reshape_position != MaxSector)
2377 mddev->new_layout = n;
2382 static struct md_sysfs_entry md_layout =
2383 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2387 raid_disks_show(mddev_t *mddev, char *page)
2389 if (mddev->raid_disks == 0)
2391 if (mddev->reshape_position != MaxSector &&
2392 mddev->delta_disks != 0)
2393 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2394 mddev->raid_disks - mddev->delta_disks);
2395 return sprintf(page, "%d\n", mddev->raid_disks);
2398 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2401 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2405 unsigned long n = simple_strtoul(buf, &e, 10);
2407 if (!*buf || (*e && *e != '\n'))
2411 rv = update_raid_disks(mddev, n);
2412 else if (mddev->reshape_position != MaxSector) {
2413 int olddisks = mddev->raid_disks - mddev->delta_disks;
2414 mddev->delta_disks = n - olddisks;
2415 mddev->raid_disks = n;
2417 mddev->raid_disks = n;
2418 return rv ? rv : len;
2420 static struct md_sysfs_entry md_raid_disks =
2421 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2424 chunk_size_show(mddev_t *mddev, char *page)
2426 if (mddev->reshape_position != MaxSector &&
2427 mddev->chunk_size != mddev->new_chunk)
2428 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2430 return sprintf(page, "%d\n", mddev->chunk_size);
2434 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2436 /* can only set chunk_size if array is not yet active */
2438 unsigned long n = simple_strtoul(buf, &e, 10);
2440 if (!*buf || (*e && *e != '\n'))
2445 else if (mddev->reshape_position != MaxSector)
2446 mddev->new_chunk = n;
2448 mddev->chunk_size = n;
2451 static struct md_sysfs_entry md_chunk_size =
2452 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2455 resync_start_show(mddev_t *mddev, char *page)
2457 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2461 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2463 /* can only set chunk_size if array is not yet active */
2465 unsigned long long n = simple_strtoull(buf, &e, 10);
2469 if (!*buf || (*e && *e != '\n'))
2472 mddev->recovery_cp = n;
2475 static struct md_sysfs_entry md_resync_start =
2476 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2479 * The array state can be:
2482 * No devices, no size, no level
2483 * Equivalent to STOP_ARRAY ioctl
2485 * May have some settings, but array is not active
2486 * all IO results in error
2487 * When written, doesn't tear down array, but just stops it
2488 * suspended (not supported yet)
2489 * All IO requests will block. The array can be reconfigured.
2490 * Writing this, if accepted, will block until array is quiessent
2492 * no resync can happen. no superblocks get written.
2493 * write requests fail
2495 * like readonly, but behaves like 'clean' on a write request.
2497 * clean - no pending writes, but otherwise active.
2498 * When written to inactive array, starts without resync
2499 * If a write request arrives then
2500 * if metadata is known, mark 'dirty' and switch to 'active'.
2501 * if not known, block and switch to write-pending
2502 * If written to an active array that has pending writes, then fails.
2504 * fully active: IO and resync can be happening.
2505 * When written to inactive array, starts with resync
2508 * clean, but writes are blocked waiting for 'active' to be written.
2511 * like active, but no writes have been seen for a while (100msec).
2514 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2515 write_pending, active_idle, bad_word};
2516 static char *array_states[] = {
2517 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2518 "write-pending", "active-idle", NULL };
2520 static int match_word(const char *word, char **list)
2523 for (n=0; list[n]; n++)
2524 if (cmd_match(word, list[n]))
2530 array_state_show(mddev_t *mddev, char *page)
2532 enum array_state st = inactive;
2545 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2547 else if (mddev->safemode)
2553 if (list_empty(&mddev->disks) &&
2554 mddev->raid_disks == 0 &&
2560 return sprintf(page, "%s\n", array_states[st]);
2563 static int do_md_stop(mddev_t * mddev, int ro);
2564 static int do_md_run(mddev_t * mddev);
2565 static int restart_array(mddev_t *mddev);
2568 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2571 enum array_state st = match_word(buf, array_states);
2576 /* stopping an active array */
2577 if (atomic_read(&mddev->active) > 1)
2579 err = do_md_stop(mddev, 0);
2582 /* stopping an active array */
2584 if (atomic_read(&mddev->active) > 1)
2586 err = do_md_stop(mddev, 2);
2588 err = 0; /* already inactive */
2591 break; /* not supported yet */
2594 err = do_md_stop(mddev, 1);
2597 err = do_md_run(mddev);
2601 /* stopping an active array */
2603 err = do_md_stop(mddev, 1);
2605 mddev->ro = 2; /* FIXME mark devices writable */
2608 err = do_md_run(mddev);
2613 restart_array(mddev);
2614 spin_lock_irq(&mddev->write_lock);
2615 if (atomic_read(&mddev->writes_pending) == 0) {
2616 if (mddev->in_sync == 0) {
2618 if (mddev->safemode == 1)
2619 mddev->safemode = 0;
2620 if (mddev->persistent)
2621 set_bit(MD_CHANGE_CLEAN,
2627 spin_unlock_irq(&mddev->write_lock);
2630 mddev->recovery_cp = MaxSector;
2631 err = do_md_run(mddev);
2636 restart_array(mddev);
2637 if (mddev->external)
2638 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2639 wake_up(&mddev->sb_wait);
2643 err = do_md_run(mddev);
2648 /* these cannot be set */
2656 static struct md_sysfs_entry md_array_state =
2657 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2660 null_show(mddev_t *mddev, char *page)
2666 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2668 /* buf must be %d:%d\n? giving major and minor numbers */
2669 /* The new device is added to the array.
2670 * If the array has a persistent superblock, we read the
2671 * superblock to initialise info and check validity.
2672 * Otherwise, only checking done is that in bind_rdev_to_array,
2673 * which mainly checks size.
2676 int major = simple_strtoul(buf, &e, 10);
2682 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2684 minor = simple_strtoul(e+1, &e, 10);
2685 if (*e && *e != '\n')
2687 dev = MKDEV(major, minor);
2688 if (major != MAJOR(dev) ||
2689 minor != MINOR(dev))
2693 if (mddev->persistent) {
2694 rdev = md_import_device(dev, mddev->major_version,
2695 mddev->minor_version);
2696 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2697 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2698 mdk_rdev_t, same_set);
2699 err = super_types[mddev->major_version]
2700 .load_super(rdev, rdev0, mddev->minor_version);
2704 } else if (mddev->external)
2705 rdev = md_import_device(dev, -2, -1);
2707 rdev = md_import_device(dev, -1, -1);
2710 return PTR_ERR(rdev);
2711 err = bind_rdev_to_array(rdev, mddev);
2715 return err ? err : len;
2718 static struct md_sysfs_entry md_new_device =
2719 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2722 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2725 unsigned long chunk, end_chunk;
2729 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2731 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2732 if (buf == end) break;
2733 if (*end == '-') { /* range */
2735 end_chunk = simple_strtoul(buf, &end, 0);
2736 if (buf == end) break;
2738 if (*end && !isspace(*end)) break;
2739 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2741 while (isspace(*buf)) buf++;
2743 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2748 static struct md_sysfs_entry md_bitmap =
2749 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2752 size_show(mddev_t *mddev, char *page)
2754 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2757 static int update_size(mddev_t *mddev, unsigned long size);
2760 size_store(mddev_t *mddev, const char *buf, size_t len)
2762 /* If array is inactive, we can reduce the component size, but
2763 * not increase it (except from 0).
2764 * If array is active, we can try an on-line resize
2768 unsigned long long size = simple_strtoull(buf, &e, 10);
2769 if (!*buf || *buf == '\n' ||
2774 err = update_size(mddev, size);
2775 md_update_sb(mddev, 1);
2777 if (mddev->size == 0 ||
2783 return err ? err : len;
2786 static struct md_sysfs_entry md_size =
2787 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2792 * 'none' for arrays with no metadata (good luck...)
2793 * 'external' for arrays with externally managed metadata,
2794 * or N.M for internally known formats
2797 metadata_show(mddev_t *mddev, char *page)
2799 if (mddev->persistent)
2800 return sprintf(page, "%d.%d\n",
2801 mddev->major_version, mddev->minor_version);
2802 else if (mddev->external)
2803 return sprintf(page, "external:%s\n", mddev->metadata_type);
2805 return sprintf(page, "none\n");
2809 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2813 if (!list_empty(&mddev->disks))
2816 if (cmd_match(buf, "none")) {
2817 mddev->persistent = 0;
2818 mddev->external = 0;
2819 mddev->major_version = 0;
2820 mddev->minor_version = 90;
2823 if (strncmp(buf, "external:", 9) == 0) {
2824 size_t namelen = len-9;
2825 if (namelen >= sizeof(mddev->metadata_type))
2826 namelen = sizeof(mddev->metadata_type)-1;
2827 strncpy(mddev->metadata_type, buf+9, namelen);
2828 mddev->metadata_type[namelen] = 0;
2829 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2830 mddev->metadata_type[--namelen] = 0;
2831 mddev->persistent = 0;
2832 mddev->external = 1;
2833 mddev->major_version = 0;
2834 mddev->minor_version = 90;
2837 major = simple_strtoul(buf, &e, 10);
2838 if (e==buf || *e != '.')
2841 minor = simple_strtoul(buf, &e, 10);
2842 if (e==buf || (*e && *e != '\n') )
2844 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2846 mddev->major_version = major;
2847 mddev->minor_version = minor;
2848 mddev->persistent = 1;
2849 mddev->external = 0;
2853 static struct md_sysfs_entry md_metadata =
2854 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2857 action_show(mddev_t *mddev, char *page)
2859 char *type = "idle";
2860 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2861 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2862 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2864 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2865 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2867 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2874 return sprintf(page, "%s\n", type);
2878 action_store(mddev_t *mddev, const char *page, size_t len)
2880 if (!mddev->pers || !mddev->pers->sync_request)
2883 if (cmd_match(page, "idle")) {
2884 if (mddev->sync_thread) {
2885 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2886 md_unregister_thread(mddev->sync_thread);
2887 mddev->sync_thread = NULL;
2888 mddev->recovery = 0;
2890 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2891 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2893 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2894 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2895 else if (cmd_match(page, "reshape")) {
2897 if (mddev->pers->start_reshape == NULL)
2899 err = mddev->pers->start_reshape(mddev);
2903 if (cmd_match(page, "check"))
2904 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2905 else if (!cmd_match(page, "repair"))
2907 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2908 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2910 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2911 md_wakeup_thread(mddev->thread);
2916 mismatch_cnt_show(mddev_t *mddev, char *page)
2918 return sprintf(page, "%llu\n",
2919 (unsigned long long) mddev->resync_mismatches);
2922 static struct md_sysfs_entry md_scan_mode =
2923 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2926 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2929 sync_min_show(mddev_t *mddev, char *page)
2931 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2932 mddev->sync_speed_min ? "local": "system");
2936 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2940 if (strncmp(buf, "system", 6)==0) {
2941 mddev->sync_speed_min = 0;
2944 min = simple_strtoul(buf, &e, 10);
2945 if (buf == e || (*e && *e != '\n') || min <= 0)
2947 mddev->sync_speed_min = min;
2951 static struct md_sysfs_entry md_sync_min =
2952 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2955 sync_max_show(mddev_t *mddev, char *page)
2957 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2958 mddev->sync_speed_max ? "local": "system");
2962 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2966 if (strncmp(buf, "system", 6)==0) {
2967 mddev->sync_speed_max = 0;
2970 max = simple_strtoul(buf, &e, 10);
2971 if (buf == e || (*e && *e != '\n') || max <= 0)
2973 mddev->sync_speed_max = max;
2977 static struct md_sysfs_entry md_sync_max =
2978 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2981 degraded_show(mddev_t *mddev, char *page)
2983 return sprintf(page, "%d\n", mddev->degraded);
2985 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2988 sync_speed_show(mddev_t *mddev, char *page)
2990 unsigned long resync, dt, db;
2991 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2992 dt = ((jiffies - mddev->resync_mark) / HZ);
2994 db = resync - (mddev->resync_mark_cnt);
2995 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2998 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3001 sync_completed_show(mddev_t *mddev, char *page)
3003 unsigned long max_blocks, resync;
3005 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3006 max_blocks = mddev->resync_max_sectors;
3008 max_blocks = mddev->size << 1;
3010 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3011 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3014 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3017 max_sync_show(mddev_t *mddev, char *page)
3019 if (mddev->resync_max == MaxSector)
3020 return sprintf(page, "max\n");
3022 return sprintf(page, "%llu\n",
3023 (unsigned long long)mddev->resync_max);
3026 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3028 if (strncmp(buf, "max", 3) == 0)
3029 mddev->resync_max = MaxSector;
3032 unsigned long long max = simple_strtoull(buf, &ep, 10);
3033 if (ep == buf || (*ep != 0 && *ep != '\n'))
3035 if (max < mddev->resync_max &&
3036 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3039 /* Must be a multiple of chunk_size */
3040 if (mddev->chunk_size) {
3041 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3044 mddev->resync_max = max;
3046 wake_up(&mddev->recovery_wait);
3050 static struct md_sysfs_entry md_max_sync =
3051 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3054 suspend_lo_show(mddev_t *mddev, char *page)
3056 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3060 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3063 unsigned long long new = simple_strtoull(buf, &e, 10);
3065 if (mddev->pers->quiesce == NULL)
3067 if (buf == e || (*e && *e != '\n'))
3069 if (new >= mddev->suspend_hi ||
3070 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3071 mddev->suspend_lo = new;
3072 mddev->pers->quiesce(mddev, 2);
3077 static struct md_sysfs_entry md_suspend_lo =
3078 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3082 suspend_hi_show(mddev_t *mddev, char *page)
3084 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3088 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3091 unsigned long long new = simple_strtoull(buf, &e, 10);
3093 if (mddev->pers->quiesce == NULL)
3095 if (buf == e || (*e && *e != '\n'))
3097 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3098 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3099 mddev->suspend_hi = new;
3100 mddev->pers->quiesce(mddev, 1);
3101 mddev->pers->quiesce(mddev, 0);
3106 static struct md_sysfs_entry md_suspend_hi =
3107 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3110 reshape_position_show(mddev_t *mddev, char *page)
3112 if (mddev->reshape_position != MaxSector)
3113 return sprintf(page, "%llu\n",
3114 (unsigned long long)mddev->reshape_position);
3115 strcpy(page, "none\n");
3120 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3123 unsigned long long new = simple_strtoull(buf, &e, 10);
3126 if (buf == e || (*e && *e != '\n'))
3128 mddev->reshape_position = new;
3129 mddev->delta_disks = 0;
3130 mddev->new_level = mddev->level;
3131 mddev->new_layout = mddev->layout;
3132 mddev->new_chunk = mddev->chunk_size;
3136 static struct md_sysfs_entry md_reshape_position =
3137 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3138 reshape_position_store);
3141 static struct attribute *md_default_attrs[] = {
3144 &md_raid_disks.attr,
3145 &md_chunk_size.attr,
3147 &md_resync_start.attr,
3149 &md_new_device.attr,
3150 &md_safe_delay.attr,
3151 &md_array_state.attr,
3152 &md_reshape_position.attr,
3156 static struct attribute *md_redundancy_attrs[] = {
3158 &md_mismatches.attr,
3161 &md_sync_speed.attr,
3162 &md_sync_completed.attr,
3164 &md_suspend_lo.attr,
3165 &md_suspend_hi.attr,
3170 static struct attribute_group md_redundancy_group = {
3172 .attrs = md_redundancy_attrs,
3177 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3179 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3180 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3185 rv = mddev_lock(mddev);
3187 rv = entry->show(mddev, page);
3188 mddev_unlock(mddev);
3194 md_attr_store(struct kobject *kobj, struct attribute *attr,
3195 const char *page, size_t length)
3197 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3198 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3203 if (!capable(CAP_SYS_ADMIN))
3205 rv = mddev_lock(mddev);
3207 rv = entry->store(mddev, page, length);
3208 mddev_unlock(mddev);
3213 static void md_free(struct kobject *ko)
3215 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3219 static struct sysfs_ops md_sysfs_ops = {
3220 .show = md_attr_show,
3221 .store = md_attr_store,
3223 static struct kobj_type md_ktype = {
3225 .sysfs_ops = &md_sysfs_ops,
3226 .default_attrs = md_default_attrs,
3231 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3233 static DEFINE_MUTEX(disks_mutex);
3234 mddev_t *mddev = mddev_find(dev);
3235 struct gendisk *disk;
3236 int partitioned = (MAJOR(dev) != MD_MAJOR);
3237 int shift = partitioned ? MdpMinorShift : 0;
3238 int unit = MINOR(dev) >> shift;
3244 mutex_lock(&disks_mutex);
3245 if (mddev->gendisk) {
3246 mutex_unlock(&disks_mutex);
3250 disk = alloc_disk(1 << shift);
3252 mutex_unlock(&disks_mutex);
3256 disk->major = MAJOR(dev);
3257 disk->first_minor = unit << shift;
3259 sprintf(disk->disk_name, "md_d%d", unit);
3261 sprintf(disk->disk_name, "md%d", unit);
3262 disk->fops = &md_fops;
3263 disk->private_data = mddev;
3264 disk->queue = mddev->queue;
3266 mddev->gendisk = disk;
3267 mutex_unlock(&disks_mutex);
3268 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3271 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3274 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3278 static void md_safemode_timeout(unsigned long data)
3280 mddev_t *mddev = (mddev_t *) data;
3282 mddev->safemode = 1;
3283 md_wakeup_thread(mddev->thread);
3286 static int start_dirty_degraded;
3288 static int do_md_run(mddev_t * mddev)
3292 struct list_head *tmp;
3294 struct gendisk *disk;
3295 struct mdk_personality *pers;
3296 char b[BDEVNAME_SIZE];
3298 if (list_empty(&mddev->disks))
3299 /* cannot run an array with no devices.. */
3306 * Analyze all RAID superblock(s)
3308 if (!mddev->raid_disks) {
3309 if (!mddev->persistent)
3314 chunk_size = mddev->chunk_size;
3317 if (chunk_size > MAX_CHUNK_SIZE) {
3318 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3319 chunk_size, MAX_CHUNK_SIZE);
3323 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3325 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3326 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3329 if (chunk_size < PAGE_SIZE) {
3330 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3331 chunk_size, PAGE_SIZE);
3335 /* devices must have minimum size of one chunk */
3336 rdev_for_each(rdev, tmp, mddev) {
3337 if (test_bit(Faulty, &rdev->flags))
3339 if (rdev->size < chunk_size / 1024) {
3341 "md: Dev %s smaller than chunk_size:"
3343 bdevname(rdev->bdev,b),
3344 (unsigned long long)rdev->size,
3352 if (mddev->level != LEVEL_NONE)
3353 request_module("md-level-%d", mddev->level);
3354 else if (mddev->clevel[0])
3355 request_module("md-%s", mddev->clevel);
3359 * Drop all container device buffers, from now on
3360 * the only valid external interface is through the md
3363 rdev_for_each(rdev, tmp, mddev) {
3364 if (test_bit(Faulty, &rdev->flags))
3366 sync_blockdev(rdev->bdev);
3367 invalidate_bdev(rdev->bdev);
3369 /* perform some consistency tests on the device.
3370 * We don't want the data to overlap the metadata,
3371 * Internal Bitmap issues has handled elsewhere.
3373 if (rdev->data_offset < rdev->sb_offset) {
3375 rdev->data_offset + mddev->size*2
3376 > rdev->sb_offset*2) {
3377 printk("md: %s: data overlaps metadata\n",
3382 if (rdev->sb_offset*2 + rdev->sb_size/512
3383 > rdev->data_offset) {
3384 printk("md: %s: metadata overlaps data\n",
3391 md_probe(mddev->unit, NULL, NULL);
3392 disk = mddev->gendisk;
3396 spin_lock(&pers_lock);
3397 pers = find_pers(mddev->level, mddev->clevel);
3398 if (!pers || !try_module_get(pers->owner)) {
3399 spin_unlock(&pers_lock);
3400 if (mddev->level != LEVEL_NONE)
3401 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3404 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3409 spin_unlock(&pers_lock);
3410 mddev->level = pers->level;
3411 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3413 if (mddev->reshape_position != MaxSector &&
3414 pers->start_reshape == NULL) {
3415 /* This personality cannot handle reshaping... */
3417 module_put(pers->owner);
3421 if (pers->sync_request) {
3422 /* Warn if this is a potentially silly
3425 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3427 struct list_head *tmp2;
3429 rdev_for_each(rdev, tmp, mddev) {
3430 rdev_for_each(rdev2, tmp2, mddev) {
3432 rdev->bdev->bd_contains ==
3433 rdev2->bdev->bd_contains) {
3435 "%s: WARNING: %s appears to be"
3436 " on the same physical disk as"
3439 bdevname(rdev->bdev,b),
3440 bdevname(rdev2->bdev,b2));
3447 "True protection against single-disk"
3448 " failure might be compromised.\n");
3451 mddev->recovery = 0;
3452 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3453 mddev->barriers_work = 1;
3454 mddev->ok_start_degraded = start_dirty_degraded;
3457 mddev->ro = 2; /* read-only, but switch on first write */
3459 err = mddev->pers->run(mddev);
3460 if (!err && mddev->pers->sync_request) {
3461 err = bitmap_create(mddev);
3463 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3464 mdname(mddev), err);
3465 mddev->pers->stop(mddev);
3469 printk(KERN_ERR "md: pers->run() failed ...\n");
3470 module_put(mddev->pers->owner);
3472 bitmap_destroy(mddev);
3475 if (mddev->pers->sync_request) {
3476 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3478 "md: cannot register extra attributes for %s\n",
3480 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3483 atomic_set(&mddev->writes_pending,0);
3484 mddev->safemode = 0;
3485 mddev->safemode_timer.function = md_safemode_timeout;
3486 mddev->safemode_timer.data = (unsigned long) mddev;
3487 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3490 rdev_for_each(rdev, tmp, mddev)
3491 if (rdev->raid_disk >= 0) {
3493 sprintf(nm, "rd%d", rdev->raid_disk);
3494 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3495 printk("md: cannot register %s for %s\n",
3499 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3502 md_update_sb(mddev, 0);
3504 set_capacity(disk, mddev->array_size<<1);
3506 /* If we call blk_queue_make_request here, it will
3507 * re-initialise max_sectors etc which may have been
3508 * refined inside -> run. So just set the bits we need to set.
3509 * Most initialisation happended when we called
3510 * blk_queue_make_request(..., md_fail_request)
3513 mddev->queue->queuedata = mddev;
3514 mddev->queue->make_request_fn = mddev->pers->make_request;
3516 /* If there is a partially-recovered drive we need to
3517 * start recovery here. If we leave it to md_check_recovery,
3518 * it will remove the drives and not do the right thing
3520 if (mddev->degraded && !mddev->sync_thread) {
3521 struct list_head *rtmp;
3523 rdev_for_each(rdev, rtmp, mddev)
3524 if (rdev->raid_disk >= 0 &&
3525 !test_bit(In_sync, &rdev->flags) &&
3526 !test_bit(Faulty, &rdev->flags))
3527 /* complete an interrupted recovery */
3529 if (spares && mddev->pers->sync_request) {
3530 mddev->recovery = 0;
3531 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3532 mddev->sync_thread = md_register_thread(md_do_sync,
3535 if (!mddev->sync_thread) {
3536 printk(KERN_ERR "%s: could not start resync"
3539 /* leave the spares where they are, it shouldn't hurt */
3540 mddev->recovery = 0;
3544 md_wakeup_thread(mddev->thread);
3545 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3548 md_new_event(mddev);
3549 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3553 static int restart_array(mddev_t *mddev)
3555 struct gendisk *disk = mddev->gendisk;
3559 * Complain if it has no devices
3562 if (list_empty(&mddev->disks))
3570 mddev->safemode = 0;
3572 set_disk_ro(disk, 0);
3574 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3577 * Kick recovery or resync if necessary
3579 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3580 md_wakeup_thread(mddev->thread);
3581 md_wakeup_thread(mddev->sync_thread);
3590 /* similar to deny_write_access, but accounts for our holding a reference
3591 * to the file ourselves */
3592 static int deny_bitmap_write_access(struct file * file)
3594 struct inode *inode = file->f_mapping->host;
3596 spin_lock(&inode->i_lock);
3597 if (atomic_read(&inode->i_writecount) > 1) {
3598 spin_unlock(&inode->i_lock);
3601 atomic_set(&inode->i_writecount, -1);
3602 spin_unlock(&inode->i_lock);
3607 static void restore_bitmap_write_access(struct file *file)
3609 struct inode *inode = file->f_mapping->host;
3611 spin_lock(&inode->i_lock);
3612 atomic_set(&inode->i_writecount, 1);
3613 spin_unlock(&inode->i_lock);
3617 * 0 - completely stop and dis-assemble array
3618 * 1 - switch to readonly
3619 * 2 - stop but do not disassemble array
3621 static int do_md_stop(mddev_t * mddev, int mode)
3624 struct gendisk *disk = mddev->gendisk;
3627 if (atomic_read(&mddev->active)>2) {
3628 printk("md: %s still in use.\n",mdname(mddev));
3632 if (mddev->sync_thread) {
3633 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3634 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3635 md_unregister_thread(mddev->sync_thread);
3636 mddev->sync_thread = NULL;
3639 del_timer_sync(&mddev->safemode_timer);
3641 invalidate_partition(disk, 0);
3644 case 1: /* readonly */
3650 case 0: /* disassemble */
3652 bitmap_flush(mddev);
3653 md_super_wait(mddev);
3655 set_disk_ro(disk, 0);
3656 blk_queue_make_request(mddev->queue, md_fail_request);
3657 mddev->pers->stop(mddev);
3658 mddev->queue->merge_bvec_fn = NULL;
3659 mddev->queue->unplug_fn = NULL;
3660 mddev->queue->backing_dev_info.congested_fn = NULL;
3661 if (mddev->pers->sync_request)
3662 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3664 module_put(mddev->pers->owner);
3667 set_capacity(disk, 0);
3673 if (!mddev->in_sync || mddev->flags) {
3674 /* mark array as shutdown cleanly */
3676 md_update_sb(mddev, 1);
3679 set_disk_ro(disk, 1);
3680 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3684 * Free resources if final stop
3688 struct list_head *tmp;
3690 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3692 bitmap_destroy(mddev);
3693 if (mddev->bitmap_file) {
3694 restore_bitmap_write_access(mddev->bitmap_file);
3695 fput(mddev->bitmap_file);
3696 mddev->bitmap_file = NULL;
3698 mddev->bitmap_offset = 0;
3700 rdev_for_each(rdev, tmp, mddev)
3701 if (rdev->raid_disk >= 0) {
3703 sprintf(nm, "rd%d", rdev->raid_disk);
3704 sysfs_remove_link(&mddev->kobj, nm);
3707 /* make sure all md_delayed_delete calls have finished */
3708 flush_scheduled_work();
3710 export_array(mddev);
3712 mddev->array_size = 0;
3714 mddev->raid_disks = 0;
3715 mddev->recovery_cp = 0;
3716 mddev->resync_max = MaxSector;
3717 mddev->reshape_position = MaxSector;
3718 mddev->external = 0;
3719 mddev->persistent = 0;
3720 mddev->level = LEVEL_NONE;
3721 mddev->clevel[0] = 0;
3724 mddev->metadata_type[0] = 0;
3725 mddev->chunk_size = 0;
3726 mddev->ctime = mddev->utime = 0;
3728 mddev->max_disks = 0;
3730 mddev->delta_disks = 0;
3731 mddev->new_level = LEVEL_NONE;
3732 mddev->new_layout = 0;
3733 mddev->new_chunk = 0;
3734 mddev->curr_resync = 0;
3735 mddev->resync_mismatches = 0;
3736 mddev->suspend_lo = mddev->suspend_hi = 0;
3737 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3738 mddev->recovery = 0;
3741 mddev->degraded = 0;
3742 mddev->barriers_work = 0;
3743 mddev->safemode = 0;
3745 } else if (mddev->pers)
3746 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3749 md_new_event(mddev);
3755 static void autorun_array(mddev_t *mddev)
3758 struct list_head *tmp;
3761 if (list_empty(&mddev->disks))
3764 printk(KERN_INFO "md: running: ");
3766 rdev_for_each(rdev, tmp, mddev) {
3767 char b[BDEVNAME_SIZE];
3768 printk("<%s>", bdevname(rdev->bdev,b));
3772 err = do_md_run (mddev);
3774 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3775 do_md_stop (mddev, 0);
3780 * lets try to run arrays based on all disks that have arrived
3781 * until now. (those are in pending_raid_disks)
3783 * the method: pick the first pending disk, collect all disks with
3784 * the same UUID, remove all from the pending list and put them into
3785 * the 'same_array' list. Then order this list based on superblock
3786 * update time (freshest comes first), kick out 'old' disks and
3787 * compare superblocks. If everything's fine then run it.
3789 * If "unit" is allocated, then bump its reference count
3791 static void autorun_devices(int part)
3793 struct list_head *tmp;
3794 mdk_rdev_t *rdev0, *rdev;
3796 char b[BDEVNAME_SIZE];
3798 printk(KERN_INFO "md: autorun ...\n");
3799 while (!list_empty(&pending_raid_disks)) {
3802 LIST_HEAD(candidates);
3803 rdev0 = list_entry(pending_raid_disks.next,
3804 mdk_rdev_t, same_set);
3806 printk(KERN_INFO "md: considering %s ...\n",
3807 bdevname(rdev0->bdev,b));
3808 INIT_LIST_HEAD(&candidates);
3809 rdev_for_each_list(rdev, tmp, pending_raid_disks)
3810 if (super_90_load(rdev, rdev0, 0) >= 0) {
3811 printk(KERN_INFO "md: adding %s ...\n",
3812 bdevname(rdev->bdev,b));
3813 list_move(&rdev->same_set, &candidates);
3816 * now we have a set of devices, with all of them having
3817 * mostly sane superblocks. It's time to allocate the
3821 dev = MKDEV(mdp_major,
3822 rdev0->preferred_minor << MdpMinorShift);
3823 unit = MINOR(dev) >> MdpMinorShift;
3825 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3828 if (rdev0->preferred_minor != unit) {
3829 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3830 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3834 md_probe(dev, NULL, NULL);
3835 mddev = mddev_find(dev);
3838 "md: cannot allocate memory for md drive.\n");
3841 if (mddev_lock(mddev))
3842 printk(KERN_WARNING "md: %s locked, cannot run\n",
3844 else if (mddev->raid_disks || mddev->major_version
3845 || !list_empty(&mddev->disks)) {
3847 "md: %s already running, cannot run %s\n",
3848 mdname(mddev), bdevname(rdev0->bdev,b));
3849 mddev_unlock(mddev);
3851 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3852 mddev->persistent = 1;
3853 rdev_for_each_list(rdev, tmp, candidates) {
3854 list_del_init(&rdev->same_set);
3855 if (bind_rdev_to_array(rdev, mddev))
3858 autorun_array(mddev);
3859 mddev_unlock(mddev);
3861 /* on success, candidates will be empty, on error
3864 rdev_for_each_list(rdev, tmp, candidates)
3868 printk(KERN_INFO "md: ... autorun DONE.\n");
3870 #endif /* !MODULE */
3872 static int get_version(void __user * arg)
3876 ver.major = MD_MAJOR_VERSION;
3877 ver.minor = MD_MINOR_VERSION;
3878 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3880 if (copy_to_user(arg, &ver, sizeof(ver)))
3886 static int get_array_info(mddev_t * mddev, void __user * arg)
3888 mdu_array_info_t info;
3889 int nr,working,active,failed,spare;
3891 struct list_head *tmp;
3893 nr=working=active=failed=spare=0;
3894 rdev_for_each(rdev, tmp, mddev) {
3896 if (test_bit(Faulty, &rdev->flags))
3900 if (test_bit(In_sync, &rdev->flags))
3907 info.major_version = mddev->major_version;
3908 info.minor_version = mddev->minor_version;
3909 info.patch_version = MD_PATCHLEVEL_VERSION;
3910 info.ctime = mddev->ctime;
3911 info.level = mddev->level;
3912 info.size = mddev->size;
3913 if (info.size != mddev->size) /* overflow */
3916 info.raid_disks = mddev->raid_disks;
3917 info.md_minor = mddev->md_minor;
3918 info.not_persistent= !mddev->persistent;
3920 info.utime = mddev->utime;
3923 info.state = (1<<MD_SB_CLEAN);
3924 if (mddev->bitmap && mddev->bitmap_offset)
3925 info.state = (1<<MD_SB_BITMAP_PRESENT);
3926 info.active_disks = active;
3927 info.working_disks = working;
3928 info.failed_disks = failed;
3929 info.spare_disks = spare;
3931 info.layout = mddev->layout;
3932 info.chunk_size = mddev->chunk_size;
3934 if (copy_to_user(arg, &info, sizeof(info)))
3940 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3942 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3943 char *ptr, *buf = NULL;
3946 md_allow_write(mddev);
3948 file = kmalloc(sizeof(*file), GFP_KERNEL);
3952 /* bitmap disabled, zero the first byte and copy out */
3953 if (!mddev->bitmap || !mddev->bitmap->file) {
3954 file->pathname[0] = '\0';
3958 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3962 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3966 strcpy(file->pathname, ptr);
3970 if (copy_to_user(arg, file, sizeof(*file)))
3978 static int get_disk_info(mddev_t * mddev, void __user * arg)
3980 mdu_disk_info_t info;
3984 if (copy_from_user(&info, arg, sizeof(info)))
3989 rdev = find_rdev_nr(mddev, nr);
3991 info.major = MAJOR(rdev->bdev->bd_dev);
3992 info.minor = MINOR(rdev->bdev->bd_dev);
3993 info.raid_disk = rdev->raid_disk;
3995 if (test_bit(Faulty, &rdev->flags))
3996 info.state |= (1<<MD_DISK_FAULTY);
3997 else if (test_bit(In_sync, &rdev->flags)) {
3998 info.state |= (1<<MD_DISK_ACTIVE);
3999 info.state |= (1<<MD_DISK_SYNC);
4001 if (test_bit(WriteMostly, &rdev->flags))
4002 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4004 info.major = info.minor = 0;
4005 info.raid_disk = -1;
4006 info.state = (1<<MD_DISK_REMOVED);
4009 if (copy_to_user(arg, &info, sizeof(info)))
4015 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4017 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4019 dev_t dev = MKDEV(info->major,info->minor);
4021 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4024 if (!mddev->raid_disks) {
4026 /* expecting a device which has a superblock */
4027 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4030 "md: md_import_device returned %ld\n",
4032 return PTR_ERR(rdev);
4034 if (!list_empty(&mddev->disks)) {
4035 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4036 mdk_rdev_t, same_set);
4037 int err = super_types[mddev->major_version]
4038 .load_super(rdev, rdev0, mddev->minor_version);
4041 "md: %s has different UUID to %s\n",
4042 bdevname(rdev->bdev,b),
4043 bdevname(rdev0->bdev,b2));
4048 err = bind_rdev_to_array(rdev, mddev);
4055 * add_new_disk can be used once the array is assembled
4056 * to add "hot spares". They must already have a superblock
4061 if (!mddev->pers->hot_add_disk) {
4063 "%s: personality does not support diskops!\n",
4067 if (mddev->persistent)
4068 rdev = md_import_device(dev, mddev->major_version,
4069 mddev->minor_version);
4071 rdev = md_import_device(dev, -1, -1);
4074 "md: md_import_device returned %ld\n",
4076 return PTR_ERR(rdev);
4078 /* set save_raid_disk if appropriate */
4079 if (!mddev->persistent) {
4080 if (info->state & (1<<MD_DISK_SYNC) &&
4081 info->raid_disk < mddev->raid_disks)
4082 rdev->raid_disk = info->raid_disk;
4084 rdev->raid_disk = -1;
4086 super_types[mddev->major_version].
4087 validate_super(mddev, rdev);
4088 rdev->saved_raid_disk = rdev->raid_disk;
4090 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4091 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4092 set_bit(WriteMostly, &rdev->flags);
4094 rdev->raid_disk = -1;
4095 err = bind_rdev_to_array(rdev, mddev);
4096 if (!err && !mddev->pers->hot_remove_disk) {
4097 /* If there is hot_add_disk but no hot_remove_disk
4098 * then added disks for geometry changes,
4099 * and should be added immediately.
4101 super_types[mddev->major_version].
4102 validate_super(mddev, rdev);
4103 err = mddev->pers->hot_add_disk(mddev, rdev);
4105 unbind_rdev_from_array(rdev);
4110 md_update_sb(mddev, 1);
4111 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4112 md_wakeup_thread(mddev->thread);
4116 /* otherwise, add_new_disk is only allowed
4117 * for major_version==0 superblocks
4119 if (mddev->major_version != 0) {
4120 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4125 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4127 rdev = md_import_device (dev, -1, 0);
4130 "md: error, md_import_device() returned %ld\n",
4132 return PTR_ERR(rdev);
4134 rdev->desc_nr = info->number;
4135 if (info->raid_disk < mddev->raid_disks)
4136 rdev->raid_disk = info->raid_disk;
4138 rdev->raid_disk = -1;
4140 if (rdev->raid_disk < mddev->raid_disks)
4141 if (info->state & (1<<MD_DISK_SYNC))
4142 set_bit(In_sync, &rdev->flags);
4144 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4145 set_bit(WriteMostly, &rdev->flags);
4147 if (!mddev->persistent) {
4148 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4149 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4151 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4152 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
4154 err = bind_rdev_to_array(rdev, mddev);
4164 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4166 char b[BDEVNAME_SIZE];
4172 rdev = find_rdev(mddev, dev);
4176 if (rdev->raid_disk >= 0)
4179 kick_rdev_from_array(rdev);
4180 md_update_sb(mddev, 1);
4181 md_new_event(mddev);
4185 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4186 bdevname(rdev->bdev,b), mdname(mddev));
4190 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4192 char b[BDEVNAME_SIZE];
4200 if (mddev->major_version != 0) {
4201 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4202 " version-0 superblocks.\n",
4206 if (!mddev->pers->hot_add_disk) {
4208 "%s: personality does not support diskops!\n",
4213 rdev = md_import_device (dev, -1, 0);
4216 "md: error, md_import_device() returned %ld\n",
4221 if (mddev->persistent)
4222 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4225 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4227 size = calc_dev_size(rdev, mddev->chunk_size);
4230 if (test_bit(Faulty, &rdev->flags)) {
4232 "md: can not hot-add faulty %s disk to %s!\n",
4233 bdevname(rdev->bdev,b), mdname(mddev));
4237 clear_bit(In_sync, &rdev->flags);
4239 rdev->saved_raid_disk = -1;
4240 err = bind_rdev_to_array(rdev, mddev);
4245 * The rest should better be atomic, we can have disk failures
4246 * noticed in interrupt contexts ...
4249 if (rdev->desc_nr == mddev->max_disks) {
4250 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4253 goto abort_unbind_export;
4256 rdev->raid_disk = -1;
4258 md_update_sb(mddev, 1);
4261 * Kick recovery, maybe this spare has to be added to the
4262 * array immediately.
4264 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4265 md_wakeup_thread(mddev->thread);
4266 md_new_event(mddev);
4269 abort_unbind_export:
4270 unbind_rdev_from_array(rdev);
4277 static int set_bitmap_file(mddev_t *mddev, int fd)
4282 if (!mddev->pers->quiesce)
4284 if (mddev->recovery || mddev->sync_thread)
4286 /* we should be able to change the bitmap.. */
4292 return -EEXIST; /* cannot add when bitmap is present */
4293 mddev->bitmap_file = fget(fd);
4295 if (mddev->bitmap_file == NULL) {
4296 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4301 err = deny_bitmap_write_access(mddev->bitmap_file);
4303 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4305 fput(mddev->bitmap_file);
4306 mddev->bitmap_file = NULL;
4309 mddev->bitmap_offset = 0; /* file overrides offset */
4310 } else if (mddev->bitmap == NULL)
4311 return -ENOENT; /* cannot remove what isn't there */
4314 mddev->pers->quiesce(mddev, 1);
4316 err = bitmap_create(mddev);
4317 if (fd < 0 || err) {
4318 bitmap_destroy(mddev);
4319 fd = -1; /* make sure to put the file */
4321 mddev->pers->quiesce(mddev, 0);
4324 if (mddev->bitmap_file) {
4325 restore_bitmap_write_access(mddev->bitmap_file);
4326 fput(mddev->bitmap_file);
4328 mddev->bitmap_file = NULL;
4335 * set_array_info is used two different ways
4336 * The original usage is when creating a new array.
4337 * In this usage, raid_disks is > 0 and it together with
4338 * level, size, not_persistent,layout,chunksize determine the
4339 * shape of the array.
4340 * This will always create an array with a type-0.90.0 superblock.
4341 * The newer usage is when assembling an array.
4342 * In this case raid_disks will be 0, and the major_version field is
4343 * use to determine which style super-blocks are to be found on the devices.
4344 * The minor and patch _version numbers are also kept incase the
4345 * super_block handler wishes to interpret them.
4347 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4350 if (info->raid_disks == 0) {
4351 /* just setting version number for superblock loading */
4352 if (info->major_version < 0 ||
4353 info->major_version >= ARRAY_SIZE(super_types) ||
4354 super_types[info->major_version].name == NULL) {
4355 /* maybe try to auto-load a module? */
4357 "md: superblock version %d not known\n",
4358 info->major_version);
4361 mddev->major_version = info->major_version;
4362 mddev->minor_version = info->minor_version;
4363 mddev->patch_version = info->patch_version;
4364 mddev->persistent = !info->not_persistent;
4367 mddev->major_version = MD_MAJOR_VERSION;
4368 mddev->minor_version = MD_MINOR_VERSION;
4369 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4370 mddev->ctime = get_seconds();
4372 mddev->level = info->level;
4373 mddev->clevel[0] = 0;
4374 mddev->size = info->size;
4375 mddev->raid_disks = info->raid_disks;
4376 /* don't set md_minor, it is determined by which /dev/md* was
4379 if (info->state & (1<<MD_SB_CLEAN))
4380 mddev->recovery_cp = MaxSector;
4382 mddev->recovery_cp = 0;
4383 mddev->persistent = ! info->not_persistent;
4384 mddev->external = 0;
4386 mddev->layout = info->layout;
4387 mddev->chunk_size = info->chunk_size;
4389 mddev->max_disks = MD_SB_DISKS;
4391 if (mddev->persistent)
4393 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4395 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4396 mddev->bitmap_offset = 0;
4398 mddev->reshape_position = MaxSector;
4401 * Generate a 128 bit UUID
4403 get_random_bytes(mddev->uuid, 16);
4405 mddev->new_level = mddev->level;
4406 mddev->new_chunk = mddev->chunk_size;
4407 mddev->new_layout = mddev->layout;
4408 mddev->delta_disks = 0;
4413 static int update_size(mddev_t *mddev, unsigned long size)
4417 struct list_head *tmp;
4418 int fit = (size == 0);
4420 if (mddev->pers->resize == NULL)
4422 /* The "size" is the amount of each device that is used.
4423 * This can only make sense for arrays with redundancy.
4424 * linear and raid0 always use whatever space is available
4425 * We can only consider changing the size if no resync
4426 * or reconstruction is happening, and if the new size
4427 * is acceptable. It must fit before the sb_offset or,
4428 * if that is <data_offset, it must fit before the
4429 * size of each device.
4430 * If size is zero, we find the largest size that fits.
4432 if (mddev->sync_thread)
4434 rdev_for_each(rdev, tmp, mddev) {
4436 avail = rdev->size * 2;
4438 if (fit && (size == 0 || size > avail/2))
4440 if (avail < ((sector_t)size << 1))
4443 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4445 struct block_device *bdev;
4447 bdev = bdget_disk(mddev->gendisk, 0);
4449 mutex_lock(&bdev->bd_inode->i_mutex);
4450 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4451 mutex_unlock(&bdev->bd_inode->i_mutex);
4458 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4461 /* change the number of raid disks */
4462 if (mddev->pers->check_reshape == NULL)
4464 if (raid_disks <= 0 ||
4465 raid_disks >= mddev->max_disks)
4467 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4469 mddev->delta_disks = raid_disks - mddev->raid_disks;
4471 rv = mddev->pers->check_reshape(mddev);
4477 * update_array_info is used to change the configuration of an
4479 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4480 * fields in the info are checked against the array.
4481 * Any differences that cannot be handled will cause an error.
4482 * Normally, only one change can be managed at a time.
4484 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4490 /* calculate expected state,ignoring low bits */
4491 if (mddev->bitmap && mddev->bitmap_offset)
4492 state |= (1 << MD_SB_BITMAP_PRESENT);
4494 if (mddev->major_version != info->major_version ||
4495 mddev->minor_version != info->minor_version ||
4496 /* mddev->patch_version != info->patch_version || */
4497 mddev->ctime != info->ctime ||
4498 mddev->level != info->level ||
4499 /* mddev->layout != info->layout || */
4500 !mddev->persistent != info->not_persistent||
4501 mddev->chunk_size != info->chunk_size ||
4502 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4503 ((state^info->state) & 0xfffffe00)
4506 /* Check there is only one change */
4507 if (info->size >= 0 && mddev->size != info->size) cnt++;
4508 if (mddev->raid_disks != info->raid_disks) cnt++;
4509 if (mddev->layout != info->layout) cnt++;
4510 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4511 if (cnt == 0) return 0;
4512 if (cnt > 1) return -EINVAL;
4514 if (mddev->layout != info->layout) {
4516 * we don't need to do anything at the md level, the
4517 * personality will take care of it all.
4519 if (mddev->pers->reconfig == NULL)
4522 return mddev->pers->reconfig(mddev, info->layout, -1);
4524 if (info->size >= 0 && mddev->size != info->size)
4525 rv = update_size(mddev, info->size);
4527 if (mddev->raid_disks != info->raid_disks)
4528 rv = update_raid_disks(mddev, info->raid_disks);
4530 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4531 if (mddev->pers->quiesce == NULL)
4533 if (mddev->recovery || mddev->sync_thread)
4535 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4536 /* add the bitmap */
4539 if (mddev->default_bitmap_offset == 0)
4541 mddev->bitmap_offset = mddev->default_bitmap_offset;
4542 mddev->pers->quiesce(mddev, 1);
4543 rv = bitmap_create(mddev);
4545 bitmap_destroy(mddev);
4546 mddev->pers->quiesce(mddev, 0);
4548 /* remove the bitmap */
4551 if (mddev->bitmap->file)
4553 mddev->pers->quiesce(mddev, 1);
4554 bitmap_destroy(mddev);
4555 mddev->pers->quiesce(mddev, 0);
4556 mddev->bitmap_offset = 0;
4559 md_update_sb(mddev, 1);
4563 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4567 if (mddev->pers == NULL)
4570 rdev = find_rdev(mddev, dev);
4574 md_error(mddev, rdev);
4578 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4580 mddev_t *mddev = bdev->bd_disk->private_data;
4584 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4588 static int md_ioctl(struct inode *inode, struct file *file,
4589 unsigned int cmd, unsigned long arg)
4592 void __user *argp = (void __user *)arg;
4593 mddev_t *mddev = NULL;
4595 if (!capable(CAP_SYS_ADMIN))
4599 * Commands dealing with the RAID driver but not any
4605 err = get_version(argp);
4608 case PRINT_RAID_DEBUG:
4616 autostart_arrays(arg);
4623 * Commands creating/starting a new array:
4626 mddev = inode->i_bdev->bd_disk->private_data;
4633 err = mddev_lock(mddev);
4636 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4643 case SET_ARRAY_INFO:
4645 mdu_array_info_t info;
4647 memset(&info, 0, sizeof(info));
4648 else if (copy_from_user(&info, argp, sizeof(info))) {
4653 err = update_array_info(mddev, &info);
4655 printk(KERN_WARNING "md: couldn't update"
4656 " array info. %d\n", err);
4661 if (!list_empty(&mddev->disks)) {
4663 "md: array %s already has disks!\n",
4668 if (mddev->raid_disks) {
4670 "md: array %s already initialised!\n",
4675 err = set_array_info(mddev, &info);
4677 printk(KERN_WARNING "md: couldn't set"
4678 " array info. %d\n", err);
4688 * Commands querying/configuring an existing array:
4690 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4691 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4692 if ((!mddev->raid_disks && !mddev->external)
4693 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4694 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4695 && cmd != GET_BITMAP_FILE) {
4701 * Commands even a read-only array can execute:
4705 case GET_ARRAY_INFO:
4706 err = get_array_info(mddev, argp);
4709 case GET_BITMAP_FILE:
4710 err = get_bitmap_file(mddev, argp);
4714 err = get_disk_info(mddev, argp);
4717 case RESTART_ARRAY_RW:
4718 err = restart_array(mddev);
4722 err = do_md_stop (mddev, 0);
4726 err = do_md_stop (mddev, 1);
4730 * We have a problem here : there is no easy way to give a CHS
4731 * virtual geometry. We currently pretend that we have a 2 heads
4732 * 4 sectors (with a BIG number of cylinders...). This drives
4733 * dosfs just mad... ;-)
4738 * The remaining ioctls are changing the state of the
4739 * superblock, so we do not allow them on read-only arrays.
4740 * However non-MD ioctls (e.g. get-size) will still come through
4741 * here and hit the 'default' below, so only disallow
4742 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4744 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4745 mddev->ro && mddev->pers) {
4746 if (mddev->ro == 2) {
4748 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4749 md_wakeup_thread(mddev->thread);
4761 mdu_disk_info_t info;
4762 if (copy_from_user(&info, argp, sizeof(info)))
4765 err = add_new_disk(mddev, &info);
4769 case HOT_REMOVE_DISK:
4770 err = hot_remove_disk(mddev, new_decode_dev(arg));
4774 err = hot_add_disk(mddev, new_decode_dev(arg));
4777 case SET_DISK_FAULTY:
4778 err = set_disk_faulty(mddev, new_decode_dev(arg));
4782 err = do_md_run (mddev);
4785 case SET_BITMAP_FILE:
4786 err = set_bitmap_file(mddev, (int)arg);
4796 mddev_unlock(mddev);
4806 static int md_open(struct inode *inode, struct file *file)
4809 * Succeed if we can lock the mddev, which confirms that
4810 * it isn't being stopped right now.
4812 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4815 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4820 mddev_unlock(mddev);
4822 check_disk_change(inode->i_bdev);
4827 static int md_release(struct inode *inode, struct file * file)
4829 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4837 static int md_media_changed(struct gendisk *disk)
4839 mddev_t *mddev = disk->private_data;
4841 return mddev->changed;
4844 static int md_revalidate(struct gendisk *disk)
4846 mddev_t *mddev = disk->private_data;
4851 static struct block_device_operations md_fops =
4853 .owner = THIS_MODULE,
4855 .release = md_release,
4857 .getgeo = md_getgeo,
4858 .media_changed = md_media_changed,
4859 .revalidate_disk= md_revalidate,
4862 static int md_thread(void * arg)
4864 mdk_thread_t *thread = arg;
4867 * md_thread is a 'system-thread', it's priority should be very
4868 * high. We avoid resource deadlocks individually in each
4869 * raid personality. (RAID5 does preallocation) We also use RR and
4870 * the very same RT priority as kswapd, thus we will never get
4871 * into a priority inversion deadlock.
4873 * we definitely have to have equal or higher priority than
4874 * bdflush, otherwise bdflush will deadlock if there are too
4875 * many dirty RAID5 blocks.
4878 allow_signal(SIGKILL);
4879 while (!kthread_should_stop()) {
4881 /* We need to wait INTERRUPTIBLE so that
4882 * we don't add to the load-average.
4883 * That means we need to be sure no signals are
4886 if (signal_pending(current))
4887 flush_signals(current);
4889 wait_event_interruptible_timeout
4891 test_bit(THREAD_WAKEUP, &thread->flags)
4892 || kthread_should_stop(),
4895 clear_bit(THREAD_WAKEUP, &thread->flags);
4897 thread->run(thread->mddev);
4903 void md_wakeup_thread(mdk_thread_t *thread)
4906 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4907 set_bit(THREAD_WAKEUP, &thread->flags);
4908 wake_up(&thread->wqueue);
4912 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4915 mdk_thread_t *thread;
4917 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4921 init_waitqueue_head(&thread->wqueue);
4924 thread->mddev = mddev;
4925 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4926 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4927 if (IS_ERR(thread->tsk)) {
4934 void md_unregister_thread(mdk_thread_t *thread)
4936 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4938 kthread_stop(thread->tsk);
4942 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4949 if (!rdev || test_bit(Faulty, &rdev->flags))
4952 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4954 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4955 __builtin_return_address(0),__builtin_return_address(1),
4956 __builtin_return_address(2),__builtin_return_address(3));
4960 if (!mddev->pers->error_handler)
4962 mddev->pers->error_handler(mddev,rdev);
4963 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4964 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4965 md_wakeup_thread(mddev->thread);
4966 md_new_event_inintr(mddev);
4969 /* seq_file implementation /proc/mdstat */
4971 static void status_unused(struct seq_file *seq)
4975 struct list_head *tmp;
4977 seq_printf(seq, "unused devices: ");
4979 rdev_for_each_list(rdev, tmp, pending_raid_disks) {
4980 char b[BDEVNAME_SIZE];
4982 seq_printf(seq, "%s ",
4983 bdevname(rdev->bdev,b));
4986 seq_printf(seq, "<none>");
4988 seq_printf(seq, "\n");
4992 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4994 sector_t max_blocks, resync, res;
4995 unsigned long dt, db, rt;
4997 unsigned int per_milli;
4999 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5001 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5002 max_blocks = mddev->resync_max_sectors >> 1;
5004 max_blocks = mddev->size;
5007 * Should not happen.
5013 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5014 * in a sector_t, and (max_blocks>>scale) will fit in a
5015 * u32, as those are the requirements for sector_div.
5016 * Thus 'scale' must be at least 10
5019 if (sizeof(sector_t) > sizeof(unsigned long)) {
5020 while ( max_blocks/2 > (1ULL<<(scale+32)))
5023 res = (resync>>scale)*1000;
5024 sector_div(res, (u32)((max_blocks>>scale)+1));
5028 int i, x = per_milli/50, y = 20-x;
5029 seq_printf(seq, "[");
5030 for (i = 0; i < x; i++)
5031 seq_printf(seq, "=");
5032 seq_printf(seq, ">");
5033 for (i = 0; i < y; i++)
5034 seq_printf(seq, ".");
5035 seq_printf(seq, "] ");
5037 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5038 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5040 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5042 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5043 "resync" : "recovery"))),
5044 per_milli/10, per_milli % 10,
5045 (unsigned long long) resync,
5046 (unsigned long long) max_blocks);
5049 * We do not want to overflow, so the order of operands and
5050 * the * 100 / 100 trick are important. We do a +1 to be
5051 * safe against division by zero. We only estimate anyway.
5053 * dt: time from mark until now
5054 * db: blocks written from mark until now
5055 * rt: remaining time
5057 dt = ((jiffies - mddev->resync_mark) / HZ);
5059 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5060 - mddev->resync_mark_cnt;
5061 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5063 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5065 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5068 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5070 struct list_head *tmp;
5080 spin_lock(&all_mddevs_lock);
5081 list_for_each(tmp,&all_mddevs)
5083 mddev = list_entry(tmp, mddev_t, all_mddevs);
5085 spin_unlock(&all_mddevs_lock);
5088 spin_unlock(&all_mddevs_lock);
5090 return (void*)2;/* tail */
5094 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5096 struct list_head *tmp;
5097 mddev_t *next_mddev, *mddev = v;
5103 spin_lock(&all_mddevs_lock);
5105 tmp = all_mddevs.next;
5107 tmp = mddev->all_mddevs.next;
5108 if (tmp != &all_mddevs)
5109 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5111 next_mddev = (void*)2;
5114 spin_unlock(&all_mddevs_lock);
5122 static void md_seq_stop(struct seq_file *seq, void *v)
5126 if (mddev && v != (void*)1 && v != (void*)2)
5130 struct mdstat_info {
5134 static int md_seq_show(struct seq_file *seq, void *v)
5138 struct list_head *tmp2;
5140 struct mdstat_info *mi = seq->private;
5141 struct bitmap *bitmap;
5143 if (v == (void*)1) {
5144 struct mdk_personality *pers;
5145 seq_printf(seq, "Personalities : ");
5146 spin_lock(&pers_lock);
5147 list_for_each_entry(pers, &pers_list, list)
5148 seq_printf(seq, "[%s] ", pers->name);
5150 spin_unlock(&pers_lock);
5151 seq_printf(seq, "\n");
5152 mi->event = atomic_read(&md_event_count);
5155 if (v == (void*)2) {
5160 if (mddev_lock(mddev) < 0)
5163 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5164 seq_printf(seq, "%s : %sactive", mdname(mddev),
5165 mddev->pers ? "" : "in");
5168 seq_printf(seq, " (read-only)");
5170 seq_printf(seq, " (auto-read-only)");
5171 seq_printf(seq, " %s", mddev->pers->name);
5175 rdev_for_each(rdev, tmp2, mddev) {
5176 char b[BDEVNAME_SIZE];
5177 seq_printf(seq, " %s[%d]",
5178 bdevname(rdev->bdev,b), rdev->desc_nr);
5179 if (test_bit(WriteMostly, &rdev->flags))
5180 seq_printf(seq, "(W)");
5181 if (test_bit(Faulty, &rdev->flags)) {
5182 seq_printf(seq, "(F)");
5184 } else if (rdev->raid_disk < 0)
5185 seq_printf(seq, "(S)"); /* spare */
5189 if (!list_empty(&mddev->disks)) {
5191 seq_printf(seq, "\n %llu blocks",
5192 (unsigned long long)mddev->array_size);
5194 seq_printf(seq, "\n %llu blocks",
5195 (unsigned long long)size);
5197 if (mddev->persistent) {
5198 if (mddev->major_version != 0 ||
5199 mddev->minor_version != 90) {
5200 seq_printf(seq," super %d.%d",
5201 mddev->major_version,
5202 mddev->minor_version);
5204 } else if (mddev->external)
5205 seq_printf(seq, " super external:%s",
5206 mddev->metadata_type);
5208 seq_printf(seq, " super non-persistent");
5211 mddev->pers->status (seq, mddev);
5212 seq_printf(seq, "\n ");
5213 if (mddev->pers->sync_request) {
5214 if (mddev->curr_resync > 2) {
5215 status_resync (seq, mddev);
5216 seq_printf(seq, "\n ");
5217 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5218 seq_printf(seq, "\tresync=DELAYED\n ");
5219 else if (mddev->recovery_cp < MaxSector)
5220 seq_printf(seq, "\tresync=PENDING\n ");
5223 seq_printf(seq, "\n ");
5225 if ((bitmap = mddev->bitmap)) {
5226 unsigned long chunk_kb;
5227 unsigned long flags;
5228 spin_lock_irqsave(&bitmap->lock, flags);
5229 chunk_kb = bitmap->chunksize >> 10;
5230 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5232 bitmap->pages - bitmap->missing_pages,
5234 (bitmap->pages - bitmap->missing_pages)
5235 << (PAGE_SHIFT - 10),
5236 chunk_kb ? chunk_kb : bitmap->chunksize,
5237 chunk_kb ? "KB" : "B");
5239 seq_printf(seq, ", file: ");
5240 seq_path(seq, &bitmap->file->f_path, " \t\n");
5243 seq_printf(seq, "\n");
5244 spin_unlock_irqrestore(&bitmap->lock, flags);
5247 seq_printf(seq, "\n");
5249 mddev_unlock(mddev);
5254 static struct seq_operations md_seq_ops = {
5255 .start = md_seq_start,
5256 .next = md_seq_next,
5257 .stop = md_seq_stop,
5258 .show = md_seq_show,
5261 static int md_seq_open(struct inode *inode, struct file *file)
5264 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5268 error = seq_open(file, &md_seq_ops);
5272 struct seq_file *p = file->private_data;
5274 mi->event = atomic_read(&md_event_count);
5279 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5281 struct seq_file *m = filp->private_data;
5282 struct mdstat_info *mi = m->private;
5285 poll_wait(filp, &md_event_waiters, wait);
5287 /* always allow read */
5288 mask = POLLIN | POLLRDNORM;
5290 if (mi->event != atomic_read(&md_event_count))
5291 mask |= POLLERR | POLLPRI;
5295 static const struct file_operations md_seq_fops = {
5296 .owner = THIS_MODULE,
5297 .open = md_seq_open,
5299 .llseek = seq_lseek,
5300 .release = seq_release_private,
5301 .poll = mdstat_poll,
5304 int register_md_personality(struct mdk_personality *p)
5306 spin_lock(&pers_lock);
5307 list_add_tail(&p->list, &pers_list);
5308 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5309 spin_unlock(&pers_lock);
5313 int unregister_md_personality(struct mdk_personality *p)
5315 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5316 spin_lock(&pers_lock);
5317 list_del_init(&p->list);
5318 spin_unlock(&pers_lock);
5322 static int is_mddev_idle(mddev_t *mddev)
5325 struct list_head *tmp;
5330 rdev_for_each(rdev, tmp, mddev) {
5331 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5332 curr_events = disk_stat_read(disk, sectors[0]) +
5333 disk_stat_read(disk, sectors[1]) -
5334 atomic_read(&disk->sync_io);
5335 /* sync IO will cause sync_io to increase before the disk_stats
5336 * as sync_io is counted when a request starts, and
5337 * disk_stats is counted when it completes.
5338 * So resync activity will cause curr_events to be smaller than
5339 * when there was no such activity.
5340 * non-sync IO will cause disk_stat to increase without
5341 * increasing sync_io so curr_events will (eventually)
5342 * be larger than it was before. Once it becomes
5343 * substantially larger, the test below will cause
5344 * the array to appear non-idle, and resync will slow
5346 * If there is a lot of outstanding resync activity when
5347 * we set last_event to curr_events, then all that activity
5348 * completing might cause the array to appear non-idle
5349 * and resync will be slowed down even though there might
5350 * not have been non-resync activity. This will only
5351 * happen once though. 'last_events' will soon reflect
5352 * the state where there is little or no outstanding
5353 * resync requests, and further resync activity will
5354 * always make curr_events less than last_events.
5357 if (curr_events - rdev->last_events > 4096) {
5358 rdev->last_events = curr_events;
5365 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5367 /* another "blocks" (512byte) blocks have been synced */
5368 atomic_sub(blocks, &mddev->recovery_active);
5369 wake_up(&mddev->recovery_wait);
5371 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5372 md_wakeup_thread(mddev->thread);
5373 // stop recovery, signal do_sync ....
5378 /* md_write_start(mddev, bi)
5379 * If we need to update some array metadata (e.g. 'active' flag
5380 * in superblock) before writing, schedule a superblock update
5381 * and wait for it to complete.
5383 void md_write_start(mddev_t *mddev, struct bio *bi)
5385 if (bio_data_dir(bi) != WRITE)
5388 BUG_ON(mddev->ro == 1);
5389 if (mddev->ro == 2) {
5390 /* need to switch to read/write */
5392 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5393 md_wakeup_thread(mddev->thread);
5394 md_wakeup_thread(mddev->sync_thread);
5396 atomic_inc(&mddev->writes_pending);
5397 if (mddev->safemode == 1)
5398 mddev->safemode = 0;
5399 if (mddev->in_sync) {
5400 spin_lock_irq(&mddev->write_lock);
5401 if (mddev->in_sync) {
5403 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5404 md_wakeup_thread(mddev->thread);
5406 spin_unlock_irq(&mddev->write_lock);
5408 wait_event(mddev->sb_wait, mddev->flags==0);
5411 void md_write_end(mddev_t *mddev)
5413 if (atomic_dec_and_test(&mddev->writes_pending)) {
5414 if (mddev->safemode == 2)
5415 md_wakeup_thread(mddev->thread);
5416 else if (mddev->safemode_delay)
5417 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5421 /* md_allow_write(mddev)
5422 * Calling this ensures that the array is marked 'active' so that writes
5423 * may proceed without blocking. It is important to call this before
5424 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5425 * Must be called with mddev_lock held.
5427 void md_allow_write(mddev_t *mddev)
5434 spin_lock_irq(&mddev->write_lock);
5435 if (mddev->in_sync) {
5437 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5438 if (mddev->safemode_delay &&
5439 mddev->safemode == 0)
5440 mddev->safemode = 1;
5441 spin_unlock_irq(&mddev->write_lock);
5442 md_update_sb(mddev, 0);
5444 spin_unlock_irq(&mddev->write_lock);
5446 EXPORT_SYMBOL_GPL(md_allow_write);
5448 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5450 #define SYNC_MARKS 10
5451 #define SYNC_MARK_STEP (3*HZ)
5452 void md_do_sync(mddev_t *mddev)
5455 unsigned int currspeed = 0,
5457 sector_t max_sectors,j, io_sectors;
5458 unsigned long mark[SYNC_MARKS];
5459 sector_t mark_cnt[SYNC_MARKS];
5461 struct list_head *tmp;
5462 sector_t last_check;
5464 struct list_head *rtmp;
5468 /* just incase thread restarts... */
5469 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5471 if (mddev->ro) /* never try to sync a read-only array */
5474 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5475 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5476 desc = "data-check";
5477 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5478 desc = "requested-resync";
5481 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5486 /* we overload curr_resync somewhat here.
5487 * 0 == not engaged in resync at all
5488 * 2 == checking that there is no conflict with another sync
5489 * 1 == like 2, but have yielded to allow conflicting resync to
5491 * other == active in resync - this many blocks
5493 * Before starting a resync we must have set curr_resync to
5494 * 2, and then checked that every "conflicting" array has curr_resync
5495 * less than ours. When we find one that is the same or higher
5496 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5497 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5498 * This will mean we have to start checking from the beginning again.
5503 mddev->curr_resync = 2;
5506 if (kthread_should_stop()) {
5507 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5510 for_each_mddev(mddev2, tmp) {
5511 if (mddev2 == mddev)
5513 if (mddev2->curr_resync &&
5514 match_mddev_units(mddev,mddev2)) {
5516 if (mddev < mddev2 && mddev->curr_resync == 2) {
5517 /* arbitrarily yield */
5518 mddev->curr_resync = 1;
5519 wake_up(&resync_wait);
5521 if (mddev > mddev2 && mddev->curr_resync == 1)
5522 /* no need to wait here, we can wait the next
5523 * time 'round when curr_resync == 2
5526 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5527 if (!kthread_should_stop() &&
5528 mddev2->curr_resync >= mddev->curr_resync) {
5529 printk(KERN_INFO "md: delaying %s of %s"
5530 " until %s has finished (they"
5531 " share one or more physical units)\n",
5532 desc, mdname(mddev), mdname(mddev2));
5535 finish_wait(&resync_wait, &wq);
5538 finish_wait(&resync_wait, &wq);
5541 } while (mddev->curr_resync < 2);
5544 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5545 /* resync follows the size requested by the personality,
5546 * which defaults to physical size, but can be virtual size
5548 max_sectors = mddev->resync_max_sectors;
5549 mddev->resync_mismatches = 0;
5550 /* we don't use the checkpoint if there's a bitmap */
5551 if (!mddev->bitmap &&
5552 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5553 j = mddev->recovery_cp;
5554 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5555 max_sectors = mddev->size << 1;
5557 /* recovery follows the physical size of devices */
5558 max_sectors = mddev->size << 1;
5560 rdev_for_each(rdev, rtmp, mddev)
5561 if (rdev->raid_disk >= 0 &&
5562 !test_bit(Faulty, &rdev->flags) &&
5563 !test_bit(In_sync, &rdev->flags) &&
5564 rdev->recovery_offset < j)
5565 j = rdev->recovery_offset;
5568 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5569 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5570 " %d KB/sec/disk.\n", speed_min(mddev));
5571 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5572 "(but not more than %d KB/sec) for %s.\n",
5573 speed_max(mddev), desc);
5575 is_mddev_idle(mddev); /* this also initializes IO event counters */
5578 for (m = 0; m < SYNC_MARKS; m++) {
5580 mark_cnt[m] = io_sectors;
5583 mddev->resync_mark = mark[last_mark];
5584 mddev->resync_mark_cnt = mark_cnt[last_mark];
5587 * Tune reconstruction:
5589 window = 32*(PAGE_SIZE/512);
5590 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5591 window/2,(unsigned long long) max_sectors/2);
5593 atomic_set(&mddev->recovery_active, 0);
5594 init_waitqueue_head(&mddev->recovery_wait);
5599 "md: resuming %s of %s from checkpoint.\n",
5600 desc, mdname(mddev));
5601 mddev->curr_resync = j;
5604 while (j < max_sectors) {
5608 if (j >= mddev->resync_max) {
5609 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5610 wait_event(mddev->recovery_wait,
5611 mddev->resync_max > j
5612 || kthread_should_stop());
5614 if (kthread_should_stop())
5616 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5617 currspeed < speed_min(mddev));
5619 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5623 if (!skipped) { /* actual IO requested */
5624 io_sectors += sectors;
5625 atomic_add(sectors, &mddev->recovery_active);
5629 if (j>1) mddev->curr_resync = j;
5630 mddev->curr_mark_cnt = io_sectors;
5631 if (last_check == 0)
5632 /* this is the earliers that rebuilt will be
5633 * visible in /proc/mdstat
5635 md_new_event(mddev);
5637 if (last_check + window > io_sectors || j == max_sectors)
5640 last_check = io_sectors;
5642 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5643 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5647 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5649 int next = (last_mark+1) % SYNC_MARKS;
5651 mddev->resync_mark = mark[next];
5652 mddev->resync_mark_cnt = mark_cnt[next];
5653 mark[next] = jiffies;
5654 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5659 if (kthread_should_stop())
5664 * this loop exits only if either when we are slower than
5665 * the 'hard' speed limit, or the system was IO-idle for
5667 * the system might be non-idle CPU-wise, but we only care
5668 * about not overloading the IO subsystem. (things like an
5669 * e2fsck being done on the RAID array should execute fast)
5671 blk_unplug(mddev->queue);
5674 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5675 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5677 if (currspeed > speed_min(mddev)) {
5678 if ((currspeed > speed_max(mddev)) ||
5679 !is_mddev_idle(mddev)) {
5685 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5687 * this also signals 'finished resyncing' to md_stop
5690 blk_unplug(mddev->queue);
5692 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5694 /* tell personality that we are finished */
5695 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5697 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5698 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5699 mddev->curr_resync > 2) {
5700 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5701 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5702 if (mddev->curr_resync >= mddev->recovery_cp) {
5704 "md: checkpointing %s of %s.\n",
5705 desc, mdname(mddev));
5706 mddev->recovery_cp = mddev->curr_resync;
5709 mddev->recovery_cp = MaxSector;
5711 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5712 mddev->curr_resync = MaxSector;
5713 rdev_for_each(rdev, rtmp, mddev)
5714 if (rdev->raid_disk >= 0 &&
5715 !test_bit(Faulty, &rdev->flags) &&
5716 !test_bit(In_sync, &rdev->flags) &&
5717 rdev->recovery_offset < mddev->curr_resync)
5718 rdev->recovery_offset = mddev->curr_resync;
5721 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5724 mddev->curr_resync = 0;
5725 mddev->resync_max = MaxSector;
5726 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5727 wake_up(&resync_wait);
5728 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5729 md_wakeup_thread(mddev->thread);
5734 * got a signal, exit.
5737 "md: md_do_sync() got signal ... exiting\n");
5738 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5742 EXPORT_SYMBOL_GPL(md_do_sync);
5745 static int remove_and_add_spares(mddev_t *mddev)
5748 struct list_head *rtmp;
5751 rdev_for_each(rdev, rtmp, mddev)
5752 if (rdev->raid_disk >= 0 &&
5754 (test_bit(Faulty, &rdev->flags) ||
5755 ! test_bit(In_sync, &rdev->flags)) &&
5756 atomic_read(&rdev->nr_pending)==0) {
5757 if (mddev->pers->hot_remove_disk(
5758 mddev, rdev->raid_disk)==0) {
5760 sprintf(nm,"rd%d", rdev->raid_disk);
5761 sysfs_remove_link(&mddev->kobj, nm);
5762 rdev->raid_disk = -1;
5766 if (mddev->degraded) {
5767 rdev_for_each(rdev, rtmp, mddev)
5768 if (rdev->raid_disk < 0
5769 && !test_bit(Faulty, &rdev->flags)) {
5770 rdev->recovery_offset = 0;
5771 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5773 sprintf(nm, "rd%d", rdev->raid_disk);
5774 if (sysfs_create_link(&mddev->kobj,
5777 "md: cannot register "
5781 md_new_event(mddev);
5789 * This routine is regularly called by all per-raid-array threads to
5790 * deal with generic issues like resync and super-block update.
5791 * Raid personalities that don't have a thread (linear/raid0) do not
5792 * need this as they never do any recovery or update the superblock.
5794 * It does not do any resync itself, but rather "forks" off other threads
5795 * to do that as needed.
5796 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5797 * "->recovery" and create a thread at ->sync_thread.
5798 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5799 * and wakeups up this thread which will reap the thread and finish up.
5800 * This thread also removes any faulty devices (with nr_pending == 0).
5802 * The overall approach is:
5803 * 1/ if the superblock needs updating, update it.
5804 * 2/ If a recovery thread is running, don't do anything else.
5805 * 3/ If recovery has finished, clean up, possibly marking spares active.
5806 * 4/ If there are any faulty devices, remove them.
5807 * 5/ If array is degraded, try to add spares devices
5808 * 6/ If array has spares or is not in-sync, start a resync thread.
5810 void md_check_recovery(mddev_t *mddev)
5813 struct list_head *rtmp;
5817 bitmap_daemon_work(mddev->bitmap);
5822 if (signal_pending(current)) {
5823 if (mddev->pers->sync_request && !mddev->external) {
5824 printk(KERN_INFO "md: %s in immediate safe mode\n",
5826 mddev->safemode = 2;
5828 flush_signals(current);
5832 (mddev->flags && !mddev->external) ||
5833 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5834 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5835 (mddev->external == 0 && mddev->safemode == 1) ||
5836 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5837 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5841 if (mddev_trylock(mddev)) {
5844 if (!mddev->external) {
5845 spin_lock_irq(&mddev->write_lock);
5846 if (mddev->safemode &&
5847 !atomic_read(&mddev->writes_pending) &&
5849 mddev->recovery_cp == MaxSector) {
5851 if (mddev->persistent)
5852 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5854 if (mddev->safemode == 1)
5855 mddev->safemode = 0;
5856 spin_unlock_irq(&mddev->write_lock);
5860 md_update_sb(mddev, 0);
5863 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5864 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5865 /* resync/recovery still happening */
5866 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5869 if (mddev->sync_thread) {
5870 /* resync has finished, collect result */
5871 md_unregister_thread(mddev->sync_thread);
5872 mddev->sync_thread = NULL;
5873 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5874 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5876 /* activate any spares */
5877 mddev->pers->spare_active(mddev);
5879 md_update_sb(mddev, 1);
5881 /* if array is no-longer degraded, then any saved_raid_disk
5882 * information must be scrapped
5884 if (!mddev->degraded)
5885 rdev_for_each(rdev, rtmp, mddev)
5886 rdev->saved_raid_disk = -1;
5888 mddev->recovery = 0;
5889 /* flag recovery needed just to double check */
5890 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5891 md_new_event(mddev);
5894 /* Clear some bits that don't mean anything, but
5897 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5898 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5899 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5900 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5902 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5904 /* no recovery is running.
5905 * remove any failed drives, then
5906 * add spares if possible.
5907 * Spare are also removed and re-added, to allow
5908 * the personality to fail the re-add.
5911 if (mddev->reshape_position != MaxSector) {
5912 if (mddev->pers->check_reshape(mddev) != 0)
5913 /* Cannot proceed */
5915 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5916 } else if ((spares = remove_and_add_spares(mddev))) {
5917 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5918 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5919 } else if (mddev->recovery_cp < MaxSector) {
5920 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5921 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5922 /* nothing to be done ... */
5925 if (mddev->pers->sync_request) {
5926 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5927 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5928 /* We are adding a device or devices to an array
5929 * which has the bitmap stored on all devices.
5930 * So make sure all bitmap pages get written
5932 bitmap_write_all(mddev->bitmap);
5934 mddev->sync_thread = md_register_thread(md_do_sync,
5937 if (!mddev->sync_thread) {
5938 printk(KERN_ERR "%s: could not start resync"
5941 /* leave the spares where they are, it shouldn't hurt */
5942 mddev->recovery = 0;
5944 md_wakeup_thread(mddev->sync_thread);
5945 md_new_event(mddev);
5948 mddev_unlock(mddev);
5952 static int md_notify_reboot(struct notifier_block *this,
5953 unsigned long code, void *x)
5955 struct list_head *tmp;
5958 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5960 printk(KERN_INFO "md: stopping all md devices.\n");
5962 for_each_mddev(mddev, tmp)
5963 if (mddev_trylock(mddev)) {
5964 do_md_stop (mddev, 1);
5965 mddev_unlock(mddev);
5968 * certain more exotic SCSI devices are known to be
5969 * volatile wrt too early system reboots. While the
5970 * right place to handle this issue is the given
5971 * driver, we do want to have a safe RAID driver ...
5978 static struct notifier_block md_notifier = {
5979 .notifier_call = md_notify_reboot,
5981 .priority = INT_MAX, /* before any real devices */
5984 static void md_geninit(void)
5986 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5988 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
5991 static int __init md_init(void)
5993 if (register_blkdev(MAJOR_NR, "md"))
5995 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5996 unregister_blkdev(MAJOR_NR, "md");
5999 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
6000 md_probe, NULL, NULL);
6001 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6002 md_probe, NULL, NULL);
6004 register_reboot_notifier(&md_notifier);
6005 raid_table_header = register_sysctl_table(raid_root_table);
6015 * Searches all registered partitions for autorun RAID arrays
6019 static LIST_HEAD(all_detected_devices);
6020 struct detected_devices_node {
6021 struct list_head list;
6025 void md_autodetect_dev(dev_t dev)
6027 struct detected_devices_node *node_detected_dev;
6029 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6030 if (node_detected_dev) {
6031 node_detected_dev->dev = dev;
6032 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6034 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6035 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6040 static void autostart_arrays(int part)
6043 struct detected_devices_node *node_detected_dev;
6045 int i_scanned, i_passed;
6050 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6052 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6054 node_detected_dev = list_entry(all_detected_devices.next,
6055 struct detected_devices_node, list);
6056 list_del(&node_detected_dev->list);
6057 dev = node_detected_dev->dev;
6058 kfree(node_detected_dev);
6059 rdev = md_import_device(dev,0, 90);
6063 if (test_bit(Faulty, &rdev->flags)) {
6067 set_bit(AutoDetected, &rdev->flags);
6068 list_add(&rdev->same_set, &pending_raid_disks);
6072 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6073 i_scanned, i_passed);
6075 autorun_devices(part);
6078 #endif /* !MODULE */
6080 static __exit void md_exit(void)
6083 struct list_head *tmp;
6085 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
6086 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6088 unregister_blkdev(MAJOR_NR,"md");
6089 unregister_blkdev(mdp_major, "mdp");
6090 unregister_reboot_notifier(&md_notifier);
6091 unregister_sysctl_table(raid_table_header);
6092 remove_proc_entry("mdstat", NULL);
6093 for_each_mddev(mddev, tmp) {
6094 struct gendisk *disk = mddev->gendisk;
6097 export_array(mddev);
6100 mddev->gendisk = NULL;
6105 subsys_initcall(md_init);
6106 module_exit(md_exit)
6108 static int get_ro(char *buffer, struct kernel_param *kp)
6110 return sprintf(buffer, "%d", start_readonly);
6112 static int set_ro(const char *val, struct kernel_param *kp)
6115 int num = simple_strtoul(val, &e, 10);
6116 if (*val && (*e == '\0' || *e == '\n')) {
6117 start_readonly = num;
6123 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6124 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6127 EXPORT_SYMBOL(register_md_personality);
6128 EXPORT_SYMBOL(unregister_md_personality);
6129 EXPORT_SYMBOL(md_error);
6130 EXPORT_SYMBOL(md_done_sync);
6131 EXPORT_SYMBOL(md_write_start);
6132 EXPORT_SYMBOL(md_write_end);
6133 EXPORT_SYMBOL(md_register_thread);
6134 EXPORT_SYMBOL(md_unregister_thread);
6135 EXPORT_SYMBOL(md_wakeup_thread);
6136 EXPORT_SYMBOL(md_check_recovery);
6137 MODULE_LICENSE("GPL");
6139 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);