2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
48 #include <linux/init.h>
50 #include <linux/file.h>
53 #include <linux/kmod.h>
56 #include <asm/unaligned.h>
58 #define MAJOR_NR MD_MAJOR
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
69 static void autostart_arrays (int part);
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
75 static void md_print_devices(void);
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
80 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81 * is 1000 KB/sec, so the extra system load does not show up that much.
82 * Increase it if you want to have more _guaranteed_ speed. Note that
83 * the RAID driver will use the maximum available bandwidth if the IO
84 * subsystem is idle. There is also an 'absolute maximum' reconstruction
85 * speed limit - in case reconstruction slows down your system despite
88 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89 * or /sys/block/mdX/md/sync_speed_{min,max}
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
96 return mddev->sync_speed_min ?
97 mddev->sync_speed_min : sysctl_speed_limit_min;
100 static inline int speed_max(mddev_t *mddev)
102 return mddev->sync_speed_max ?
103 mddev->sync_speed_max : sysctl_speed_limit_max;
106 static struct ctl_table_header *raid_table_header;
108 static ctl_table raid_table[] = {
110 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
111 .procname = "speed_limit_min",
112 .data = &sysctl_speed_limit_min,
113 .maxlen = sizeof(int),
114 .mode = S_IRUGO|S_IWUSR,
115 .proc_handler = &proc_dointvec,
118 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
119 .procname = "speed_limit_max",
120 .data = &sysctl_speed_limit_max,
121 .maxlen = sizeof(int),
122 .mode = S_IRUGO|S_IWUSR,
123 .proc_handler = &proc_dointvec,
128 static ctl_table raid_dir_table[] = {
130 .ctl_name = DEV_RAID,
133 .mode = S_IRUGO|S_IXUGO,
139 static ctl_table raid_root_table[] = {
145 .child = raid_dir_table,
150 static struct block_device_operations md_fops;
152 static int start_readonly;
155 * We have a system wide 'event count' that is incremented
156 * on any 'interesting' event, and readers of /proc/mdstat
157 * can use 'poll' or 'select' to find out when the event
161 * start array, stop array, error, add device, remove device,
162 * start build, activate spare
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
168 atomic_inc(&md_event_count);
169 wake_up(&md_event_waiters);
170 sysfs_notify(&mddev->kobj, NULL, "sync_action");
172 EXPORT_SYMBOL_GPL(md_new_event);
174 /* Alternate version that can be called from interrupts
175 * when calling sysfs_notify isn't needed.
177 static void md_new_event_inintr(mddev_t *mddev)
179 atomic_inc(&md_event_count);
180 wake_up(&md_event_waiters);
184 * Enables to iterate over all existing md arrays
185 * all_mddevs_lock protects this list.
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
192 * iterates through all used mddevs in the system.
193 * We take care to grab the all_mddevs_lock whenever navigating
194 * the list, and to always hold a refcount when unlocked.
195 * Any code which breaks out of this loop while own
196 * a reference to the current mddev and must mddev_put it.
198 #define ITERATE_MDDEV(mddev,tmp) \
200 for (({ spin_lock(&all_mddevs_lock); \
201 tmp = all_mddevs.next; \
203 ({ if (tmp != &all_mddevs) \
204 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205 spin_unlock(&all_mddevs_lock); \
206 if (mddev) mddev_put(mddev); \
207 mddev = list_entry(tmp, mddev_t, all_mddevs); \
208 tmp != &all_mddevs;}); \
209 ({ spin_lock(&all_mddevs_lock); \
214 static int md_fail_request (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;
279 new->queue = blk_alloc_queue(GFP_KERNEL);
284 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
286 blk_queue_make_request(new->queue, md_fail_request);
291 static inline int mddev_lock(mddev_t * mddev)
293 return mutex_lock_interruptible(&mddev->reconfig_mutex);
296 static inline int mddev_trylock(mddev_t * mddev)
298 return mutex_trylock(&mddev->reconfig_mutex);
301 static inline void mddev_unlock(mddev_t * mddev)
303 mutex_unlock(&mddev->reconfig_mutex);
305 md_wakeup_thread(mddev->thread);
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
311 struct list_head *tmp;
313 ITERATE_RDEV(mddev,rdev,tmp) {
314 if (rdev->desc_nr == nr)
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
322 struct list_head *tmp;
325 ITERATE_RDEV(mddev,rdev,tmp) {
326 if (rdev->bdev->bd_dev == dev)
332 static struct mdk_personality *find_pers(int level, char *clevel)
334 struct mdk_personality *pers;
335 list_for_each_entry(pers, &pers_list, list) {
336 if (level != LEVEL_NONE && pers->level == level)
338 if (strcmp(pers->name, clevel)==0)
344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
346 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
347 return MD_NEW_SIZE_BLOCKS(size);
350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
354 size = rdev->sb_offset;
357 size &= ~((sector_t)chunk_size/1024 - 1);
361 static int alloc_disk_sb(mdk_rdev_t * rdev)
366 rdev->sb_page = alloc_page(GFP_KERNEL);
367 if (!rdev->sb_page) {
368 printk(KERN_ALERT "md: out of memory.\n");
375 static void free_disk_sb(mdk_rdev_t * rdev)
378 put_page(rdev->sb_page);
380 rdev->sb_page = NULL;
387 static void super_written(struct bio *bio, int error)
389 mdk_rdev_t *rdev = bio->bi_private;
390 mddev_t *mddev = rdev->mddev;
392 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
393 printk("md: super_written gets error=%d, uptodate=%d\n",
394 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
395 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
396 md_error(mddev, rdev);
399 if (atomic_dec_and_test(&mddev->pending_writes))
400 wake_up(&mddev->sb_wait);
404 static void super_written_barrier(struct bio *bio, int error)
406 struct bio *bio2 = bio->bi_private;
407 mdk_rdev_t *rdev = bio2->bi_private;
408 mddev_t *mddev = rdev->mddev;
410 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
411 error == -EOPNOTSUPP) {
413 /* barriers don't appear to be supported :-( */
414 set_bit(BarriersNotsupp, &rdev->flags);
415 mddev->barriers_work = 0;
416 spin_lock_irqsave(&mddev->write_lock, flags);
417 bio2->bi_next = mddev->biolist;
418 mddev->biolist = bio2;
419 spin_unlock_irqrestore(&mddev->write_lock, flags);
420 wake_up(&mddev->sb_wait);
424 bio->bi_private = rdev;
425 super_written(bio, error);
429 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
430 sector_t sector, int size, struct page *page)
432 /* write first size bytes of page to sector of rdev
433 * Increment mddev->pending_writes before returning
434 * and decrement it on completion, waking up sb_wait
435 * if zero is reached.
436 * If an error occurred, call md_error
438 * As we might need to resubmit the request if BIO_RW_BARRIER
439 * causes ENOTSUPP, we allocate a spare bio...
441 struct bio *bio = bio_alloc(GFP_NOIO, 1);
442 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
444 bio->bi_bdev = rdev->bdev;
445 bio->bi_sector = sector;
446 bio_add_page(bio, page, size, 0);
447 bio->bi_private = rdev;
448 bio->bi_end_io = super_written;
451 atomic_inc(&mddev->pending_writes);
452 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
454 rw |= (1<<BIO_RW_BARRIER);
455 rbio = bio_clone(bio, GFP_NOIO);
456 rbio->bi_private = bio;
457 rbio->bi_end_io = super_written_barrier;
458 submit_bio(rw, rbio);
463 void md_super_wait(mddev_t *mddev)
465 /* wait for all superblock writes that were scheduled to complete.
466 * if any had to be retried (due to BARRIER problems), retry them
470 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
471 if (atomic_read(&mddev->pending_writes)==0)
473 while (mddev->biolist) {
475 spin_lock_irq(&mddev->write_lock);
476 bio = mddev->biolist;
477 mddev->biolist = bio->bi_next ;
479 spin_unlock_irq(&mddev->write_lock);
480 submit_bio(bio->bi_rw, bio);
484 finish_wait(&mddev->sb_wait, &wq);
487 static void bi_complete(struct bio *bio, int error)
489 complete((struct completion*)bio->bi_private);
492 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
493 struct page *page, int rw)
495 struct bio *bio = bio_alloc(GFP_NOIO, 1);
496 struct completion event;
499 rw |= (1 << BIO_RW_SYNC);
502 bio->bi_sector = sector;
503 bio_add_page(bio, page, size, 0);
504 init_completion(&event);
505 bio->bi_private = &event;
506 bio->bi_end_io = bi_complete;
508 wait_for_completion(&event);
510 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
514 EXPORT_SYMBOL_GPL(sync_page_io);
516 static int read_disk_sb(mdk_rdev_t * rdev, int size)
518 char b[BDEVNAME_SIZE];
519 if (!rdev->sb_page) {
527 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
533 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
534 bdevname(rdev->bdev,b));
538 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
540 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
541 (sb1->set_uuid1 == sb2->set_uuid1) &&
542 (sb1->set_uuid2 == sb2->set_uuid2) &&
543 (sb1->set_uuid3 == sb2->set_uuid3))
551 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
554 mdp_super_t *tmp1, *tmp2;
556 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
557 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
559 if (!tmp1 || !tmp2) {
561 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
569 * nr_disks is not constant
574 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
586 static u32 md_csum_fold(u32 csum)
588 csum = (csum & 0xffff) + (csum >> 16);
589 return (csum & 0xffff) + (csum >> 16);
592 static unsigned int calc_sb_csum(mdp_super_t * sb)
595 u32 *sb32 = (u32*)sb;
597 unsigned int disk_csum, csum;
599 disk_csum = sb->sb_csum;
602 for (i = 0; i < MD_SB_BYTES/4 ; i++)
604 csum = (newcsum & 0xffffffff) + (newcsum>>32);
608 /* This used to use csum_partial, which was wrong for several
609 * reasons including that different results are returned on
610 * different architectures. It isn't critical that we get exactly
611 * the same return value as before (we always csum_fold before
612 * testing, and that removes any differences). However as we
613 * know that csum_partial always returned a 16bit value on
614 * alphas, do a fold to maximise conformity to previous behaviour.
616 sb->sb_csum = md_csum_fold(disk_csum);
618 sb->sb_csum = disk_csum;
625 * Handle superblock details.
626 * We want to be able to handle multiple superblock formats
627 * so we have a common interface to them all, and an array of
628 * different handlers.
629 * We rely on user-space to write the initial superblock, and support
630 * reading and updating of superblocks.
631 * Interface methods are:
632 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
633 * loads and validates a superblock on dev.
634 * if refdev != NULL, compare superblocks on both devices
636 * 0 - dev has a superblock that is compatible with refdev
637 * 1 - dev has a superblock that is compatible and newer than refdev
638 * so dev should be used as the refdev in future
639 * -EINVAL superblock incompatible or invalid
640 * -othererror e.g. -EIO
642 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
643 * Verify that dev is acceptable into mddev.
644 * The first time, mddev->raid_disks will be 0, and data from
645 * dev should be merged in. Subsequent calls check that dev
646 * is new enough. Return 0 or -EINVAL
648 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
649 * Update the superblock for rdev with data in mddev
650 * This does not write to disc.
656 struct module *owner;
657 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
658 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
659 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
663 * load_super for 0.90.0
665 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
667 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
673 * Calculate the position of the superblock,
674 * it's at the end of the disk.
676 * It also happens to be a multiple of 4Kb.
678 sb_offset = calc_dev_sboffset(rdev->bdev);
679 rdev->sb_offset = sb_offset;
681 ret = read_disk_sb(rdev, MD_SB_BYTES);
686 bdevname(rdev->bdev, b);
687 sb = (mdp_super_t*)page_address(rdev->sb_page);
689 if (sb->md_magic != MD_SB_MAGIC) {
690 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
695 if (sb->major_version != 0 ||
696 sb->minor_version < 90 ||
697 sb->minor_version > 91) {
698 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
699 sb->major_version, sb->minor_version,
704 if (sb->raid_disks <= 0)
707 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
708 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
713 rdev->preferred_minor = sb->md_minor;
714 rdev->data_offset = 0;
715 rdev->sb_size = MD_SB_BYTES;
717 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
718 if (sb->level != 1 && sb->level != 4
719 && sb->level != 5 && sb->level != 6
720 && sb->level != 10) {
721 /* FIXME use a better test */
723 "md: bitmaps not supported for this level.\n");
728 if (sb->level == LEVEL_MULTIPATH)
731 rdev->desc_nr = sb->this_disk.number;
737 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
738 if (!uuid_equal(refsb, sb)) {
739 printk(KERN_WARNING "md: %s has different UUID to %s\n",
740 b, bdevname(refdev->bdev,b2));
743 if (!sb_equal(refsb, sb)) {
744 printk(KERN_WARNING "md: %s has same UUID"
745 " but different superblock to %s\n",
746 b, bdevname(refdev->bdev, b2));
750 ev2 = md_event(refsb);
756 rdev->size = calc_dev_size(rdev, sb->chunk_size);
758 if (rdev->size < sb->size && sb->level > 1)
759 /* "this cannot possibly happen" ... */
767 * validate_super for 0.90.0
769 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
772 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
773 __u64 ev1 = md_event(sb);
775 rdev->raid_disk = -1;
777 if (mddev->raid_disks == 0) {
778 mddev->major_version = 0;
779 mddev->minor_version = sb->minor_version;
780 mddev->patch_version = sb->patch_version;
781 mddev->persistent = 1;
783 mddev->chunk_size = sb->chunk_size;
784 mddev->ctime = sb->ctime;
785 mddev->utime = sb->utime;
786 mddev->level = sb->level;
787 mddev->clevel[0] = 0;
788 mddev->layout = sb->layout;
789 mddev->raid_disks = sb->raid_disks;
790 mddev->size = sb->size;
792 mddev->bitmap_offset = 0;
793 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
795 if (mddev->minor_version >= 91) {
796 mddev->reshape_position = sb->reshape_position;
797 mddev->delta_disks = sb->delta_disks;
798 mddev->new_level = sb->new_level;
799 mddev->new_layout = sb->new_layout;
800 mddev->new_chunk = sb->new_chunk;
802 mddev->reshape_position = MaxSector;
803 mddev->delta_disks = 0;
804 mddev->new_level = mddev->level;
805 mddev->new_layout = mddev->layout;
806 mddev->new_chunk = mddev->chunk_size;
809 if (sb->state & (1<<MD_SB_CLEAN))
810 mddev->recovery_cp = MaxSector;
812 if (sb->events_hi == sb->cp_events_hi &&
813 sb->events_lo == sb->cp_events_lo) {
814 mddev->recovery_cp = sb->recovery_cp;
816 mddev->recovery_cp = 0;
819 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
820 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
821 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
822 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
824 mddev->max_disks = MD_SB_DISKS;
826 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
827 mddev->bitmap_file == NULL)
828 mddev->bitmap_offset = mddev->default_bitmap_offset;
830 } else if (mddev->pers == NULL) {
831 /* Insist on good event counter while assembling */
833 if (ev1 < mddev->events)
835 } else if (mddev->bitmap) {
836 /* if adding to array with a bitmap, then we can accept an
837 * older device ... but not too old.
839 if (ev1 < mddev->bitmap->events_cleared)
842 if (ev1 < mddev->events)
843 /* just a hot-add of a new device, leave raid_disk at -1 */
847 if (mddev->level != LEVEL_MULTIPATH) {
848 desc = sb->disks + rdev->desc_nr;
850 if (desc->state & (1<<MD_DISK_FAULTY))
851 set_bit(Faulty, &rdev->flags);
852 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
853 desc->raid_disk < mddev->raid_disks */) {
854 set_bit(In_sync, &rdev->flags);
855 rdev->raid_disk = desc->raid_disk;
857 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
858 set_bit(WriteMostly, &rdev->flags);
859 } else /* MULTIPATH are always insync */
860 set_bit(In_sync, &rdev->flags);
865 * sync_super for 0.90.0
867 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
870 struct list_head *tmp;
872 int next_spare = mddev->raid_disks;
875 /* make rdev->sb match mddev data..
878 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
879 * 3/ any empty disks < next_spare become removed
881 * disks[0] gets initialised to REMOVED because
882 * we cannot be sure from other fields if it has
883 * been initialised or not.
886 int active=0, working=0,failed=0,spare=0,nr_disks=0;
888 rdev->sb_size = MD_SB_BYTES;
890 sb = (mdp_super_t*)page_address(rdev->sb_page);
892 memset(sb, 0, sizeof(*sb));
894 sb->md_magic = MD_SB_MAGIC;
895 sb->major_version = mddev->major_version;
896 sb->patch_version = mddev->patch_version;
897 sb->gvalid_words = 0; /* ignored */
898 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
899 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
900 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
901 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
903 sb->ctime = mddev->ctime;
904 sb->level = mddev->level;
905 sb->size = mddev->size;
906 sb->raid_disks = mddev->raid_disks;
907 sb->md_minor = mddev->md_minor;
908 sb->not_persistent = 0;
909 sb->utime = mddev->utime;
911 sb->events_hi = (mddev->events>>32);
912 sb->events_lo = (u32)mddev->events;
914 if (mddev->reshape_position == MaxSector)
915 sb->minor_version = 90;
917 sb->minor_version = 91;
918 sb->reshape_position = mddev->reshape_position;
919 sb->new_level = mddev->new_level;
920 sb->delta_disks = mddev->delta_disks;
921 sb->new_layout = mddev->new_layout;
922 sb->new_chunk = mddev->new_chunk;
924 mddev->minor_version = sb->minor_version;
927 sb->recovery_cp = mddev->recovery_cp;
928 sb->cp_events_hi = (mddev->events>>32);
929 sb->cp_events_lo = (u32)mddev->events;
930 if (mddev->recovery_cp == MaxSector)
931 sb->state = (1<< MD_SB_CLEAN);
935 sb->layout = mddev->layout;
936 sb->chunk_size = mddev->chunk_size;
938 if (mddev->bitmap && mddev->bitmap_file == NULL)
939 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
941 sb->disks[0].state = (1<<MD_DISK_REMOVED);
942 ITERATE_RDEV(mddev,rdev2,tmp) {
945 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
946 && !test_bit(Faulty, &rdev2->flags))
947 desc_nr = rdev2->raid_disk;
949 desc_nr = next_spare++;
950 rdev2->desc_nr = desc_nr;
951 d = &sb->disks[rdev2->desc_nr];
953 d->number = rdev2->desc_nr;
954 d->major = MAJOR(rdev2->bdev->bd_dev);
955 d->minor = MINOR(rdev2->bdev->bd_dev);
956 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
957 && !test_bit(Faulty, &rdev2->flags))
958 d->raid_disk = rdev2->raid_disk;
960 d->raid_disk = rdev2->desc_nr; /* compatibility */
961 if (test_bit(Faulty, &rdev2->flags))
962 d->state = (1<<MD_DISK_FAULTY);
963 else if (test_bit(In_sync, &rdev2->flags)) {
964 d->state = (1<<MD_DISK_ACTIVE);
965 d->state |= (1<<MD_DISK_SYNC);
973 if (test_bit(WriteMostly, &rdev2->flags))
974 d->state |= (1<<MD_DISK_WRITEMOSTLY);
976 /* now set the "removed" and "faulty" bits on any missing devices */
977 for (i=0 ; i < mddev->raid_disks ; i++) {
978 mdp_disk_t *d = &sb->disks[i];
979 if (d->state == 0 && d->number == 0) {
982 d->state = (1<<MD_DISK_REMOVED);
983 d->state |= (1<<MD_DISK_FAULTY);
987 sb->nr_disks = nr_disks;
988 sb->active_disks = active;
989 sb->working_disks = working;
990 sb->failed_disks = failed;
991 sb->spare_disks = spare;
993 sb->this_disk = sb->disks[rdev->desc_nr];
994 sb->sb_csum = calc_sb_csum(sb);
998 * version 1 superblock
1001 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1005 unsigned long long newcsum;
1006 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1007 __le32 *isuper = (__le32*)sb;
1010 disk_csum = sb->sb_csum;
1013 for (i=0; size>=4; size -= 4 )
1014 newcsum += le32_to_cpu(*isuper++);
1017 newcsum += le16_to_cpu(*(__le16*) isuper);
1019 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1020 sb->sb_csum = disk_csum;
1021 return cpu_to_le32(csum);
1024 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1026 struct mdp_superblock_1 *sb;
1029 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1033 * Calculate the position of the superblock.
1034 * It is always aligned to a 4K boundary and
1035 * depeding on minor_version, it can be:
1036 * 0: At least 8K, but less than 12K, from end of device
1037 * 1: At start of device
1038 * 2: 4K from start of device.
1040 switch(minor_version) {
1042 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1044 sb_offset &= ~(sector_t)(4*2-1);
1045 /* convert from sectors to K */
1057 rdev->sb_offset = sb_offset;
1059 /* superblock is rarely larger than 1K, but it can be larger,
1060 * and it is safe to read 4k, so we do that
1062 ret = read_disk_sb(rdev, 4096);
1063 if (ret) return ret;
1066 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1068 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1069 sb->major_version != cpu_to_le32(1) ||
1070 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1071 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1072 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1075 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1076 printk("md: invalid superblock checksum on %s\n",
1077 bdevname(rdev->bdev,b));
1080 if (le64_to_cpu(sb->data_size) < 10) {
1081 printk("md: data_size too small on %s\n",
1082 bdevname(rdev->bdev,b));
1085 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1086 if (sb->level != cpu_to_le32(1) &&
1087 sb->level != cpu_to_le32(4) &&
1088 sb->level != cpu_to_le32(5) &&
1089 sb->level != cpu_to_le32(6) &&
1090 sb->level != cpu_to_le32(10)) {
1092 "md: bitmaps not supported for this level.\n");
1097 rdev->preferred_minor = 0xffff;
1098 rdev->data_offset = le64_to_cpu(sb->data_offset);
1099 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1101 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1102 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1103 if (rdev->sb_size & bmask)
1104 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1106 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1109 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1115 struct mdp_superblock_1 *refsb =
1116 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1118 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1119 sb->level != refsb->level ||
1120 sb->layout != refsb->layout ||
1121 sb->chunksize != refsb->chunksize) {
1122 printk(KERN_WARNING "md: %s has strangely different"
1123 " superblock to %s\n",
1124 bdevname(rdev->bdev,b),
1125 bdevname(refdev->bdev,b2));
1128 ev1 = le64_to_cpu(sb->events);
1129 ev2 = le64_to_cpu(refsb->events);
1137 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1139 rdev->size = rdev->sb_offset;
1140 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1142 rdev->size = le64_to_cpu(sb->data_size)/2;
1143 if (le32_to_cpu(sb->chunksize))
1144 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1146 if (le64_to_cpu(sb->size) > rdev->size*2)
1151 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1153 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1154 __u64 ev1 = le64_to_cpu(sb->events);
1156 rdev->raid_disk = -1;
1158 if (mddev->raid_disks == 0) {
1159 mddev->major_version = 1;
1160 mddev->patch_version = 0;
1161 mddev->persistent = 1;
1162 mddev->external = 0;
1163 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1164 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1165 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1166 mddev->level = le32_to_cpu(sb->level);
1167 mddev->clevel[0] = 0;
1168 mddev->layout = le32_to_cpu(sb->layout);
1169 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1170 mddev->size = le64_to_cpu(sb->size)/2;
1171 mddev->events = ev1;
1172 mddev->bitmap_offset = 0;
1173 mddev->default_bitmap_offset = 1024 >> 9;
1175 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1176 memcpy(mddev->uuid, sb->set_uuid, 16);
1178 mddev->max_disks = (4096-256)/2;
1180 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1181 mddev->bitmap_file == NULL )
1182 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1184 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1185 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1186 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1187 mddev->new_level = le32_to_cpu(sb->new_level);
1188 mddev->new_layout = le32_to_cpu(sb->new_layout);
1189 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1191 mddev->reshape_position = MaxSector;
1192 mddev->delta_disks = 0;
1193 mddev->new_level = mddev->level;
1194 mddev->new_layout = mddev->layout;
1195 mddev->new_chunk = mddev->chunk_size;
1198 } else if (mddev->pers == NULL) {
1199 /* Insist of good event counter while assembling */
1201 if (ev1 < mddev->events)
1203 } else if (mddev->bitmap) {
1204 /* If adding to array with a bitmap, then we can accept an
1205 * older device, but not too old.
1207 if (ev1 < mddev->bitmap->events_cleared)
1210 if (ev1 < mddev->events)
1211 /* just a hot-add of a new device, leave raid_disk at -1 */
1214 if (mddev->level != LEVEL_MULTIPATH) {
1216 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1218 case 0xffff: /* spare */
1220 case 0xfffe: /* faulty */
1221 set_bit(Faulty, &rdev->flags);
1224 if ((le32_to_cpu(sb->feature_map) &
1225 MD_FEATURE_RECOVERY_OFFSET))
1226 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1228 set_bit(In_sync, &rdev->flags);
1229 rdev->raid_disk = role;
1232 if (sb->devflags & WriteMostly1)
1233 set_bit(WriteMostly, &rdev->flags);
1234 } else /* MULTIPATH are always insync */
1235 set_bit(In_sync, &rdev->flags);
1240 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1242 struct mdp_superblock_1 *sb;
1243 struct list_head *tmp;
1246 /* make rdev->sb match mddev and rdev data. */
1248 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1250 sb->feature_map = 0;
1252 sb->recovery_offset = cpu_to_le64(0);
1253 memset(sb->pad1, 0, sizeof(sb->pad1));
1254 memset(sb->pad2, 0, sizeof(sb->pad2));
1255 memset(sb->pad3, 0, sizeof(sb->pad3));
1257 sb->utime = cpu_to_le64((__u64)mddev->utime);
1258 sb->events = cpu_to_le64(mddev->events);
1260 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1262 sb->resync_offset = cpu_to_le64(0);
1264 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1266 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1267 sb->size = cpu_to_le64(mddev->size<<1);
1269 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1270 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1271 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1274 if (rdev->raid_disk >= 0 &&
1275 !test_bit(In_sync, &rdev->flags) &&
1276 rdev->recovery_offset > 0) {
1277 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1278 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1281 if (mddev->reshape_position != MaxSector) {
1282 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1283 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1284 sb->new_layout = cpu_to_le32(mddev->new_layout);
1285 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1286 sb->new_level = cpu_to_le32(mddev->new_level);
1287 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1291 ITERATE_RDEV(mddev,rdev2,tmp)
1292 if (rdev2->desc_nr+1 > max_dev)
1293 max_dev = rdev2->desc_nr+1;
1295 if (max_dev > le32_to_cpu(sb->max_dev))
1296 sb->max_dev = cpu_to_le32(max_dev);
1297 for (i=0; i<max_dev;i++)
1298 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1300 ITERATE_RDEV(mddev,rdev2,tmp) {
1302 if (test_bit(Faulty, &rdev2->flags))
1303 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1304 else if (test_bit(In_sync, &rdev2->flags))
1305 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1306 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1307 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1309 sb->dev_roles[i] = cpu_to_le16(0xffff);
1312 sb->sb_csum = calc_sb_1_csum(sb);
1316 static struct super_type super_types[] = {
1319 .owner = THIS_MODULE,
1320 .load_super = super_90_load,
1321 .validate_super = super_90_validate,
1322 .sync_super = super_90_sync,
1326 .owner = THIS_MODULE,
1327 .load_super = super_1_load,
1328 .validate_super = super_1_validate,
1329 .sync_super = super_1_sync,
1333 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1335 struct list_head *tmp, *tmp2;
1336 mdk_rdev_t *rdev, *rdev2;
1338 ITERATE_RDEV(mddev1,rdev,tmp)
1339 ITERATE_RDEV(mddev2, rdev2, tmp2)
1340 if (rdev->bdev->bd_contains ==
1341 rdev2->bdev->bd_contains)
1347 static LIST_HEAD(pending_raid_disks);
1349 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1351 char b[BDEVNAME_SIZE];
1360 /* make sure rdev->size exceeds mddev->size */
1361 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1363 /* Cannot change size, so fail
1364 * If mddev->level <= 0, then we don't care
1365 * about aligning sizes (e.g. linear)
1367 if (mddev->level > 0)
1370 mddev->size = rdev->size;
1373 /* Verify rdev->desc_nr is unique.
1374 * If it is -1, assign a free number, else
1375 * check number is not in use
1377 if (rdev->desc_nr < 0) {
1379 if (mddev->pers) choice = mddev->raid_disks;
1380 while (find_rdev_nr(mddev, choice))
1382 rdev->desc_nr = choice;
1384 if (find_rdev_nr(mddev, rdev->desc_nr))
1387 bdevname(rdev->bdev,b);
1388 while ( (s=strchr(b, '/')) != NULL)
1391 rdev->mddev = mddev;
1392 printk(KERN_INFO "md: bind<%s>\n", b);
1394 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1397 if (rdev->bdev->bd_part)
1398 ko = &rdev->bdev->bd_part->dev.kobj;
1400 ko = &rdev->bdev->bd_disk->dev.kobj;
1401 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1402 kobject_del(&rdev->kobj);
1405 list_add(&rdev->same_set, &mddev->disks);
1406 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1410 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1415 static void delayed_delete(struct work_struct *ws)
1417 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1418 kobject_del(&rdev->kobj);
1421 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1423 char b[BDEVNAME_SIZE];
1428 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1429 list_del_init(&rdev->same_set);
1430 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1432 sysfs_remove_link(&rdev->kobj, "block");
1434 /* We need to delay this, otherwise we can deadlock when
1435 * writing to 'remove' to "dev/state"
1437 INIT_WORK(&rdev->del_work, delayed_delete);
1438 schedule_work(&rdev->del_work);
1442 * prevent the device from being mounted, repartitioned or
1443 * otherwise reused by a RAID array (or any other kernel
1444 * subsystem), by bd_claiming the device.
1446 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1449 struct block_device *bdev;
1450 char b[BDEVNAME_SIZE];
1452 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1454 printk(KERN_ERR "md: could not open %s.\n",
1455 __bdevname(dev, b));
1456 return PTR_ERR(bdev);
1458 err = bd_claim(bdev, rdev);
1460 printk(KERN_ERR "md: could not bd_claim %s.\n",
1469 static void unlock_rdev(mdk_rdev_t *rdev)
1471 struct block_device *bdev = rdev->bdev;
1479 void md_autodetect_dev(dev_t dev);
1481 static void export_rdev(mdk_rdev_t * rdev)
1483 char b[BDEVNAME_SIZE];
1484 printk(KERN_INFO "md: export_rdev(%s)\n",
1485 bdevname(rdev->bdev,b));
1489 list_del_init(&rdev->same_set);
1491 md_autodetect_dev(rdev->bdev->bd_dev);
1494 kobject_put(&rdev->kobj);
1497 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1499 unbind_rdev_from_array(rdev);
1503 static void export_array(mddev_t *mddev)
1505 struct list_head *tmp;
1508 ITERATE_RDEV(mddev,rdev,tmp) {
1513 kick_rdev_from_array(rdev);
1515 if (!list_empty(&mddev->disks))
1517 mddev->raid_disks = 0;
1518 mddev->major_version = 0;
1521 static void print_desc(mdp_disk_t *desc)
1523 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1524 desc->major,desc->minor,desc->raid_disk,desc->state);
1527 static void print_sb(mdp_super_t *sb)
1532 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1533 sb->major_version, sb->minor_version, sb->patch_version,
1534 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1536 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1537 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1538 sb->md_minor, sb->layout, sb->chunk_size);
1539 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1540 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1541 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1542 sb->failed_disks, sb->spare_disks,
1543 sb->sb_csum, (unsigned long)sb->events_lo);
1546 for (i = 0; i < MD_SB_DISKS; i++) {
1549 desc = sb->disks + i;
1550 if (desc->number || desc->major || desc->minor ||
1551 desc->raid_disk || (desc->state && (desc->state != 4))) {
1552 printk(" D %2d: ", i);
1556 printk(KERN_INFO "md: THIS: ");
1557 print_desc(&sb->this_disk);
1561 static void print_rdev(mdk_rdev_t *rdev)
1563 char b[BDEVNAME_SIZE];
1564 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1565 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1566 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1568 if (rdev->sb_loaded) {
1569 printk(KERN_INFO "md: rdev superblock:\n");
1570 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1572 printk(KERN_INFO "md: no rdev superblock!\n");
1575 static void md_print_devices(void)
1577 struct list_head *tmp, *tmp2;
1580 char b[BDEVNAME_SIZE];
1583 printk("md: **********************************\n");
1584 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1585 printk("md: **********************************\n");
1586 ITERATE_MDDEV(mddev,tmp) {
1589 bitmap_print_sb(mddev->bitmap);
1591 printk("%s: ", mdname(mddev));
1592 ITERATE_RDEV(mddev,rdev,tmp2)
1593 printk("<%s>", bdevname(rdev->bdev,b));
1596 ITERATE_RDEV(mddev,rdev,tmp2)
1599 printk("md: **********************************\n");
1604 static void sync_sbs(mddev_t * mddev, int nospares)
1606 /* Update each superblock (in-memory image), but
1607 * if we are allowed to, skip spares which already
1608 * have the right event counter, or have one earlier
1609 * (which would mean they aren't being marked as dirty
1610 * with the rest of the array)
1613 struct list_head *tmp;
1615 ITERATE_RDEV(mddev,rdev,tmp) {
1616 if (rdev->sb_events == mddev->events ||
1618 rdev->raid_disk < 0 &&
1619 (rdev->sb_events&1)==0 &&
1620 rdev->sb_events+1 == mddev->events)) {
1621 /* Don't update this superblock */
1622 rdev->sb_loaded = 2;
1624 super_types[mddev->major_version].
1625 sync_super(mddev, rdev);
1626 rdev->sb_loaded = 1;
1631 static void md_update_sb(mddev_t * mddev, int force_change)
1633 struct list_head *tmp;
1639 spin_lock_irq(&mddev->write_lock);
1641 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1642 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1644 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1645 /* just a clean<-> dirty transition, possibly leave spares alone,
1646 * though if events isn't the right even/odd, we will have to do
1652 if (mddev->degraded)
1653 /* If the array is degraded, then skipping spares is both
1654 * dangerous and fairly pointless.
1655 * Dangerous because a device that was removed from the array
1656 * might have a event_count that still looks up-to-date,
1657 * so it can be re-added without a resync.
1658 * Pointless because if there are any spares to skip,
1659 * then a recovery will happen and soon that array won't
1660 * be degraded any more and the spare can go back to sleep then.
1664 sync_req = mddev->in_sync;
1665 mddev->utime = get_seconds();
1667 /* If this is just a dirty<->clean transition, and the array is clean
1668 * and 'events' is odd, we can roll back to the previous clean state */
1670 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1671 && (mddev->events & 1)
1672 && mddev->events != 1)
1675 /* otherwise we have to go forward and ... */
1677 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1678 /* .. if the array isn't clean, insist on an odd 'events' */
1679 if ((mddev->events&1)==0) {
1684 /* otherwise insist on an even 'events' (for clean states) */
1685 if ((mddev->events&1)) {
1692 if (!mddev->events) {
1694 * oops, this 64-bit counter should never wrap.
1695 * Either we are in around ~1 trillion A.C., assuming
1696 * 1 reboot per second, or we have a bug:
1703 * do not write anything to disk if using
1704 * nonpersistent superblocks
1706 if (!mddev->persistent) {
1707 if (!mddev->external)
1708 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1710 spin_unlock_irq(&mddev->write_lock);
1711 wake_up(&mddev->sb_wait);
1714 sync_sbs(mddev, nospares);
1715 spin_unlock_irq(&mddev->write_lock);
1718 "md: updating %s RAID superblock on device (in sync %d)\n",
1719 mdname(mddev),mddev->in_sync);
1721 bitmap_update_sb(mddev->bitmap);
1722 ITERATE_RDEV(mddev,rdev,tmp) {
1723 char b[BDEVNAME_SIZE];
1724 dprintk(KERN_INFO "md: ");
1725 if (rdev->sb_loaded != 1)
1726 continue; /* no noise on spare devices */
1727 if (test_bit(Faulty, &rdev->flags))
1728 dprintk("(skipping faulty ");
1730 dprintk("%s ", bdevname(rdev->bdev,b));
1731 if (!test_bit(Faulty, &rdev->flags)) {
1732 md_super_write(mddev,rdev,
1733 rdev->sb_offset<<1, rdev->sb_size,
1735 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1736 bdevname(rdev->bdev,b),
1737 (unsigned long long)rdev->sb_offset);
1738 rdev->sb_events = mddev->events;
1742 if (mddev->level == LEVEL_MULTIPATH)
1743 /* only need to write one superblock... */
1746 md_super_wait(mddev);
1747 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1749 spin_lock_irq(&mddev->write_lock);
1750 if (mddev->in_sync != sync_req ||
1751 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1752 /* have to write it out again */
1753 spin_unlock_irq(&mddev->write_lock);
1756 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1757 spin_unlock_irq(&mddev->write_lock);
1758 wake_up(&mddev->sb_wait);
1762 /* words written to sysfs files may, or my not, be \n terminated.
1763 * We want to accept with case. For this we use cmd_match.
1765 static int cmd_match(const char *cmd, const char *str)
1767 /* See if cmd, written into a sysfs file, matches
1768 * str. They must either be the same, or cmd can
1769 * have a trailing newline
1771 while (*cmd && *str && *cmd == *str) {
1782 struct rdev_sysfs_entry {
1783 struct attribute attr;
1784 ssize_t (*show)(mdk_rdev_t *, char *);
1785 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1789 state_show(mdk_rdev_t *rdev, char *page)
1794 if (test_bit(Faulty, &rdev->flags)) {
1795 len+= sprintf(page+len, "%sfaulty",sep);
1798 if (test_bit(In_sync, &rdev->flags)) {
1799 len += sprintf(page+len, "%sin_sync",sep);
1802 if (test_bit(WriteMostly, &rdev->flags)) {
1803 len += sprintf(page+len, "%swrite_mostly",sep);
1806 if (!test_bit(Faulty, &rdev->flags) &&
1807 !test_bit(In_sync, &rdev->flags)) {
1808 len += sprintf(page+len, "%sspare", sep);
1811 return len+sprintf(page+len, "\n");
1815 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1818 * faulty - simulates and error
1819 * remove - disconnects the device
1820 * writemostly - sets write_mostly
1821 * -writemostly - clears write_mostly
1824 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1825 md_error(rdev->mddev, rdev);
1827 } else if (cmd_match(buf, "remove")) {
1828 if (rdev->raid_disk >= 0)
1831 mddev_t *mddev = rdev->mddev;
1832 kick_rdev_from_array(rdev);
1834 md_update_sb(mddev, 1);
1835 md_new_event(mddev);
1838 } else if (cmd_match(buf, "writemostly")) {
1839 set_bit(WriteMostly, &rdev->flags);
1841 } else if (cmd_match(buf, "-writemostly")) {
1842 clear_bit(WriteMostly, &rdev->flags);
1845 return err ? err : len;
1847 static struct rdev_sysfs_entry rdev_state =
1848 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1851 super_show(mdk_rdev_t *rdev, char *page)
1853 if (rdev->sb_loaded && rdev->sb_size) {
1854 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1855 return rdev->sb_size;
1859 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1862 errors_show(mdk_rdev_t *rdev, char *page)
1864 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1868 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1871 unsigned long n = simple_strtoul(buf, &e, 10);
1872 if (*buf && (*e == 0 || *e == '\n')) {
1873 atomic_set(&rdev->corrected_errors, n);
1878 static struct rdev_sysfs_entry rdev_errors =
1879 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1882 slot_show(mdk_rdev_t *rdev, char *page)
1884 if (rdev->raid_disk < 0)
1885 return sprintf(page, "none\n");
1887 return sprintf(page, "%d\n", rdev->raid_disk);
1891 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1894 int slot = simple_strtoul(buf, &e, 10);
1895 if (strncmp(buf, "none", 4)==0)
1897 else if (e==buf || (*e && *e!= '\n'))
1899 if (rdev->mddev->pers)
1900 /* Cannot set slot in active array (yet) */
1902 if (slot >= rdev->mddev->raid_disks)
1904 rdev->raid_disk = slot;
1905 /* assume it is working */
1907 set_bit(In_sync, &rdev->flags);
1912 static struct rdev_sysfs_entry rdev_slot =
1913 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1916 offset_show(mdk_rdev_t *rdev, char *page)
1918 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1922 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1925 unsigned long long offset = simple_strtoull(buf, &e, 10);
1926 if (e==buf || (*e && *e != '\n'))
1928 if (rdev->mddev->pers)
1930 rdev->data_offset = offset;
1934 static struct rdev_sysfs_entry rdev_offset =
1935 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1938 rdev_size_show(mdk_rdev_t *rdev, char *page)
1940 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1944 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1947 unsigned long long size = simple_strtoull(buf, &e, 10);
1948 if (e==buf || (*e && *e != '\n'))
1950 if (rdev->mddev->pers)
1953 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1954 rdev->mddev->size = size;
1958 static struct rdev_sysfs_entry rdev_size =
1959 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1961 static struct attribute *rdev_default_attrs[] = {
1971 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1973 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1974 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1978 return entry->show(rdev, page);
1982 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1983 const char *page, size_t length)
1985 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1986 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1990 if (!capable(CAP_SYS_ADMIN))
1992 return entry->store(rdev, page, length);
1995 static void rdev_free(struct kobject *ko)
1997 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2000 static struct sysfs_ops rdev_sysfs_ops = {
2001 .show = rdev_attr_show,
2002 .store = rdev_attr_store,
2004 static struct kobj_type rdev_ktype = {
2005 .release = rdev_free,
2006 .sysfs_ops = &rdev_sysfs_ops,
2007 .default_attrs = rdev_default_attrs,
2011 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2013 * mark the device faulty if:
2015 * - the device is nonexistent (zero size)
2016 * - the device has no valid superblock
2018 * a faulty rdev _never_ has rdev->sb set.
2020 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2022 char b[BDEVNAME_SIZE];
2027 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2029 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2030 return ERR_PTR(-ENOMEM);
2033 if ((err = alloc_disk_sb(rdev)))
2036 err = lock_rdev(rdev, newdev);
2040 kobject_init(&rdev->kobj, &rdev_ktype);
2043 rdev->saved_raid_disk = -1;
2044 rdev->raid_disk = -1;
2046 rdev->data_offset = 0;
2047 rdev->sb_events = 0;
2048 atomic_set(&rdev->nr_pending, 0);
2049 atomic_set(&rdev->read_errors, 0);
2050 atomic_set(&rdev->corrected_errors, 0);
2052 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2055 "md: %s has zero or unknown size, marking faulty!\n",
2056 bdevname(rdev->bdev,b));
2061 if (super_format >= 0) {
2062 err = super_types[super_format].
2063 load_super(rdev, NULL, super_minor);
2064 if (err == -EINVAL) {
2066 "md: %s does not have a valid v%d.%d "
2067 "superblock, not importing!\n",
2068 bdevname(rdev->bdev,b),
2069 super_format, super_minor);
2074 "md: could not read %s's sb, not importing!\n",
2075 bdevname(rdev->bdev,b));
2079 INIT_LIST_HEAD(&rdev->same_set);
2084 if (rdev->sb_page) {
2090 return ERR_PTR(err);
2094 * Check a full RAID array for plausibility
2098 static void analyze_sbs(mddev_t * mddev)
2101 struct list_head *tmp;
2102 mdk_rdev_t *rdev, *freshest;
2103 char b[BDEVNAME_SIZE];
2106 ITERATE_RDEV(mddev,rdev,tmp)
2107 switch (super_types[mddev->major_version].
2108 load_super(rdev, freshest, mddev->minor_version)) {
2116 "md: fatal superblock inconsistency in %s"
2117 " -- removing from array\n",
2118 bdevname(rdev->bdev,b));
2119 kick_rdev_from_array(rdev);
2123 super_types[mddev->major_version].
2124 validate_super(mddev, freshest);
2127 ITERATE_RDEV(mddev,rdev,tmp) {
2128 if (rdev != freshest)
2129 if (super_types[mddev->major_version].
2130 validate_super(mddev, rdev)) {
2131 printk(KERN_WARNING "md: kicking non-fresh %s"
2133 bdevname(rdev->bdev,b));
2134 kick_rdev_from_array(rdev);
2137 if (mddev->level == LEVEL_MULTIPATH) {
2138 rdev->desc_nr = i++;
2139 rdev->raid_disk = rdev->desc_nr;
2140 set_bit(In_sync, &rdev->flags);
2141 } else if (rdev->raid_disk >= mddev->raid_disks) {
2142 rdev->raid_disk = -1;
2143 clear_bit(In_sync, &rdev->flags);
2149 if (mddev->recovery_cp != MaxSector &&
2151 printk(KERN_ERR "md: %s: raid array is not clean"
2152 " -- starting background reconstruction\n",
2158 safe_delay_show(mddev_t *mddev, char *page)
2160 int msec = (mddev->safemode_delay*1000)/HZ;
2161 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2164 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2172 /* remove a period, and count digits after it */
2173 if (len >= sizeof(buf))
2175 strlcpy(buf, cbuf, len);
2177 for (i=0; i<len; i++) {
2179 if (isdigit(buf[i])) {
2184 } else if (buf[i] == '.') {
2189 msec = simple_strtoul(buf, &e, 10);
2190 if (e == buf || (*e && *e != '\n'))
2192 msec = (msec * 1000) / scale;
2194 mddev->safemode_delay = 0;
2196 mddev->safemode_delay = (msec*HZ)/1000;
2197 if (mddev->safemode_delay == 0)
2198 mddev->safemode_delay = 1;
2202 static struct md_sysfs_entry md_safe_delay =
2203 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2206 level_show(mddev_t *mddev, char *page)
2208 struct mdk_personality *p = mddev->pers;
2210 return sprintf(page, "%s\n", p->name);
2211 else if (mddev->clevel[0])
2212 return sprintf(page, "%s\n", mddev->clevel);
2213 else if (mddev->level != LEVEL_NONE)
2214 return sprintf(page, "%d\n", mddev->level);
2220 level_store(mddev_t *mddev, const char *buf, size_t len)
2227 if (len >= sizeof(mddev->clevel))
2229 strncpy(mddev->clevel, buf, len);
2230 if (mddev->clevel[len-1] == '\n')
2232 mddev->clevel[len] = 0;
2233 mddev->level = LEVEL_NONE;
2237 static struct md_sysfs_entry md_level =
2238 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2242 layout_show(mddev_t *mddev, char *page)
2244 /* just a number, not meaningful for all levels */
2245 if (mddev->reshape_position != MaxSector &&
2246 mddev->layout != mddev->new_layout)
2247 return sprintf(page, "%d (%d)\n",
2248 mddev->new_layout, mddev->layout);
2249 return sprintf(page, "%d\n", mddev->layout);
2253 layout_store(mddev_t *mddev, const char *buf, size_t len)
2256 unsigned long n = simple_strtoul(buf, &e, 10);
2258 if (!*buf || (*e && *e != '\n'))
2263 if (mddev->reshape_position != MaxSector)
2264 mddev->new_layout = n;
2269 static struct md_sysfs_entry md_layout =
2270 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2274 raid_disks_show(mddev_t *mddev, char *page)
2276 if (mddev->raid_disks == 0)
2278 if (mddev->reshape_position != MaxSector &&
2279 mddev->delta_disks != 0)
2280 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2281 mddev->raid_disks - mddev->delta_disks);
2282 return sprintf(page, "%d\n", mddev->raid_disks);
2285 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2288 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2292 unsigned long n = simple_strtoul(buf, &e, 10);
2294 if (!*buf || (*e && *e != '\n'))
2298 rv = update_raid_disks(mddev, n);
2299 else if (mddev->reshape_position != MaxSector) {
2300 int olddisks = mddev->raid_disks - mddev->delta_disks;
2301 mddev->delta_disks = n - olddisks;
2302 mddev->raid_disks = n;
2304 mddev->raid_disks = n;
2305 return rv ? rv : len;
2307 static struct md_sysfs_entry md_raid_disks =
2308 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2311 chunk_size_show(mddev_t *mddev, char *page)
2313 if (mddev->reshape_position != MaxSector &&
2314 mddev->chunk_size != mddev->new_chunk)
2315 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2317 return sprintf(page, "%d\n", mddev->chunk_size);
2321 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2323 /* can only set chunk_size if array is not yet active */
2325 unsigned long n = simple_strtoul(buf, &e, 10);
2327 if (!*buf || (*e && *e != '\n'))
2332 else if (mddev->reshape_position != MaxSector)
2333 mddev->new_chunk = n;
2335 mddev->chunk_size = n;
2338 static struct md_sysfs_entry md_chunk_size =
2339 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2342 resync_start_show(mddev_t *mddev, char *page)
2344 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2348 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2350 /* can only set chunk_size if array is not yet active */
2352 unsigned long long n = simple_strtoull(buf, &e, 10);
2356 if (!*buf || (*e && *e != '\n'))
2359 mddev->recovery_cp = n;
2362 static struct md_sysfs_entry md_resync_start =
2363 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2366 * The array state can be:
2369 * No devices, no size, no level
2370 * Equivalent to STOP_ARRAY ioctl
2372 * May have some settings, but array is not active
2373 * all IO results in error
2374 * When written, doesn't tear down array, but just stops it
2375 * suspended (not supported yet)
2376 * All IO requests will block. The array can be reconfigured.
2377 * Writing this, if accepted, will block until array is quiessent
2379 * no resync can happen. no superblocks get written.
2380 * write requests fail
2382 * like readonly, but behaves like 'clean' on a write request.
2384 * clean - no pending writes, but otherwise active.
2385 * When written to inactive array, starts without resync
2386 * If a write request arrives then
2387 * if metadata is known, mark 'dirty' and switch to 'active'.
2388 * if not known, block and switch to write-pending
2389 * If written to an active array that has pending writes, then fails.
2391 * fully active: IO and resync can be happening.
2392 * When written to inactive array, starts with resync
2395 * clean, but writes are blocked waiting for 'active' to be written.
2398 * like active, but no writes have been seen for a while (100msec).
2401 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2402 write_pending, active_idle, bad_word};
2403 static char *array_states[] = {
2404 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2405 "write-pending", "active-idle", NULL };
2407 static int match_word(const char *word, char **list)
2410 for (n=0; list[n]; n++)
2411 if (cmd_match(word, list[n]))
2417 array_state_show(mddev_t *mddev, char *page)
2419 enum array_state st = inactive;
2432 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2434 else if (mddev->safemode)
2440 if (list_empty(&mddev->disks) &&
2441 mddev->raid_disks == 0 &&
2447 return sprintf(page, "%s\n", array_states[st]);
2450 static int do_md_stop(mddev_t * mddev, int ro);
2451 static int do_md_run(mddev_t * mddev);
2452 static int restart_array(mddev_t *mddev);
2455 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2458 enum array_state st = match_word(buf, array_states);
2463 /* stopping an active array */
2464 if (atomic_read(&mddev->active) > 1)
2466 err = do_md_stop(mddev, 0);
2469 /* stopping an active array */
2471 if (atomic_read(&mddev->active) > 1)
2473 err = do_md_stop(mddev, 2);
2475 err = 0; /* already inactive */
2478 break; /* not supported yet */
2481 err = do_md_stop(mddev, 1);
2484 err = do_md_run(mddev);
2488 /* stopping an active array */
2490 err = do_md_stop(mddev, 1);
2492 mddev->ro = 2; /* FIXME mark devices writable */
2495 err = do_md_run(mddev);
2500 restart_array(mddev);
2501 spin_lock_irq(&mddev->write_lock);
2502 if (atomic_read(&mddev->writes_pending) == 0) {
2503 if (mddev->in_sync == 0) {
2505 if (mddev->persistent)
2506 set_bit(MD_CHANGE_CLEAN,
2512 spin_unlock_irq(&mddev->write_lock);
2515 mddev->recovery_cp = MaxSector;
2516 err = do_md_run(mddev);
2521 restart_array(mddev);
2522 if (mddev->external)
2523 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2524 wake_up(&mddev->sb_wait);
2528 err = do_md_run(mddev);
2533 /* these cannot be set */
2541 static struct md_sysfs_entry md_array_state =
2542 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2545 null_show(mddev_t *mddev, char *page)
2551 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2553 /* buf must be %d:%d\n? giving major and minor numbers */
2554 /* The new device is added to the array.
2555 * If the array has a persistent superblock, we read the
2556 * superblock to initialise info and check validity.
2557 * Otherwise, only checking done is that in bind_rdev_to_array,
2558 * which mainly checks size.
2561 int major = simple_strtoul(buf, &e, 10);
2567 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2569 minor = simple_strtoul(e+1, &e, 10);
2570 if (*e && *e != '\n')
2572 dev = MKDEV(major, minor);
2573 if (major != MAJOR(dev) ||
2574 minor != MINOR(dev))
2578 if (mddev->persistent) {
2579 rdev = md_import_device(dev, mddev->major_version,
2580 mddev->minor_version);
2581 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2582 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2583 mdk_rdev_t, same_set);
2584 err = super_types[mddev->major_version]
2585 .load_super(rdev, rdev0, mddev->minor_version);
2590 rdev = md_import_device(dev, -1, -1);
2593 return PTR_ERR(rdev);
2594 err = bind_rdev_to_array(rdev, mddev);
2598 return err ? err : len;
2601 static struct md_sysfs_entry md_new_device =
2602 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2605 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2608 unsigned long chunk, end_chunk;
2612 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2614 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2615 if (buf == end) break;
2616 if (*end == '-') { /* range */
2618 end_chunk = simple_strtoul(buf, &end, 0);
2619 if (buf == end) break;
2621 if (*end && !isspace(*end)) break;
2622 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2624 while (isspace(*buf)) buf++;
2626 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2631 static struct md_sysfs_entry md_bitmap =
2632 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2635 size_show(mddev_t *mddev, char *page)
2637 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2640 static int update_size(mddev_t *mddev, unsigned long size);
2643 size_store(mddev_t *mddev, const char *buf, size_t len)
2645 /* If array is inactive, we can reduce the component size, but
2646 * not increase it (except from 0).
2647 * If array is active, we can try an on-line resize
2651 unsigned long long size = simple_strtoull(buf, &e, 10);
2652 if (!*buf || *buf == '\n' ||
2657 err = update_size(mddev, size);
2658 md_update_sb(mddev, 1);
2660 if (mddev->size == 0 ||
2666 return err ? err : len;
2669 static struct md_sysfs_entry md_size =
2670 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2675 * 'none' for arrays with no metadata (good luck...)
2676 * 'external' for arrays with externally managed metadata,
2677 * or N.M for internally known formats
2680 metadata_show(mddev_t *mddev, char *page)
2682 if (mddev->persistent)
2683 return sprintf(page, "%d.%d\n",
2684 mddev->major_version, mddev->minor_version);
2685 else if (mddev->external)
2686 return sprintf(page, "external:%s\n", mddev->metadata_type);
2688 return sprintf(page, "none\n");
2692 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2696 if (!list_empty(&mddev->disks))
2699 if (cmd_match(buf, "none")) {
2700 mddev->persistent = 0;
2701 mddev->external = 0;
2702 mddev->major_version = 0;
2703 mddev->minor_version = 90;
2706 if (strncmp(buf, "external:", 9) == 0) {
2707 int namelen = len-9;
2708 if (namelen >= sizeof(mddev->metadata_type))
2709 namelen = sizeof(mddev->metadata_type)-1;
2710 strncpy(mddev->metadata_type, buf+9, namelen);
2711 mddev->metadata_type[namelen] = 0;
2712 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2713 mddev->metadata_type[--namelen] = 0;
2714 mddev->persistent = 0;
2715 mddev->external = 1;
2716 mddev->major_version = 0;
2717 mddev->minor_version = 90;
2720 major = simple_strtoul(buf, &e, 10);
2721 if (e==buf || *e != '.')
2724 minor = simple_strtoul(buf, &e, 10);
2725 if (e==buf || (*e && *e != '\n') )
2727 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2729 mddev->major_version = major;
2730 mddev->minor_version = minor;
2731 mddev->persistent = 1;
2732 mddev->external = 0;
2736 static struct md_sysfs_entry md_metadata =
2737 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2740 action_show(mddev_t *mddev, char *page)
2742 char *type = "idle";
2743 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2744 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2745 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2747 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2748 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2750 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2757 return sprintf(page, "%s\n", type);
2761 action_store(mddev_t *mddev, const char *page, size_t len)
2763 if (!mddev->pers || !mddev->pers->sync_request)
2766 if (cmd_match(page, "idle")) {
2767 if (mddev->sync_thread) {
2768 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2769 md_unregister_thread(mddev->sync_thread);
2770 mddev->sync_thread = NULL;
2771 mddev->recovery = 0;
2773 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2774 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2776 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2777 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2778 else if (cmd_match(page, "reshape")) {
2780 if (mddev->pers->start_reshape == NULL)
2782 err = mddev->pers->start_reshape(mddev);
2786 if (cmd_match(page, "check"))
2787 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2788 else if (!cmd_match(page, "repair"))
2790 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2791 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2793 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2794 md_wakeup_thread(mddev->thread);
2799 mismatch_cnt_show(mddev_t *mddev, char *page)
2801 return sprintf(page, "%llu\n",
2802 (unsigned long long) mddev->resync_mismatches);
2805 static struct md_sysfs_entry md_scan_mode =
2806 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2809 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2812 sync_min_show(mddev_t *mddev, char *page)
2814 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2815 mddev->sync_speed_min ? "local": "system");
2819 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2823 if (strncmp(buf, "system", 6)==0) {
2824 mddev->sync_speed_min = 0;
2827 min = simple_strtoul(buf, &e, 10);
2828 if (buf == e || (*e && *e != '\n') || min <= 0)
2830 mddev->sync_speed_min = min;
2834 static struct md_sysfs_entry md_sync_min =
2835 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2838 sync_max_show(mddev_t *mddev, char *page)
2840 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2841 mddev->sync_speed_max ? "local": "system");
2845 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2849 if (strncmp(buf, "system", 6)==0) {
2850 mddev->sync_speed_max = 0;
2853 max = simple_strtoul(buf, &e, 10);
2854 if (buf == e || (*e && *e != '\n') || max <= 0)
2856 mddev->sync_speed_max = max;
2860 static struct md_sysfs_entry md_sync_max =
2861 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2864 degraded_show(mddev_t *mddev, char *page)
2866 return sprintf(page, "%d\n", mddev->degraded);
2868 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
2871 sync_speed_show(mddev_t *mddev, char *page)
2873 unsigned long resync, dt, db;
2874 resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2875 dt = ((jiffies - mddev->resync_mark) / HZ);
2877 db = resync - (mddev->resync_mark_cnt);
2878 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2881 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2884 sync_completed_show(mddev_t *mddev, char *page)
2886 unsigned long max_blocks, resync;
2888 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2889 max_blocks = mddev->resync_max_sectors;
2891 max_blocks = mddev->size << 1;
2893 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2894 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2897 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2900 suspend_lo_show(mddev_t *mddev, char *page)
2902 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2906 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2909 unsigned long long new = simple_strtoull(buf, &e, 10);
2911 if (mddev->pers->quiesce == NULL)
2913 if (buf == e || (*e && *e != '\n'))
2915 if (new >= mddev->suspend_hi ||
2916 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2917 mddev->suspend_lo = new;
2918 mddev->pers->quiesce(mddev, 2);
2923 static struct md_sysfs_entry md_suspend_lo =
2924 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2928 suspend_hi_show(mddev_t *mddev, char *page)
2930 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2934 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2937 unsigned long long new = simple_strtoull(buf, &e, 10);
2939 if (mddev->pers->quiesce == NULL)
2941 if (buf == e || (*e && *e != '\n'))
2943 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2944 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2945 mddev->suspend_hi = new;
2946 mddev->pers->quiesce(mddev, 1);
2947 mddev->pers->quiesce(mddev, 0);
2952 static struct md_sysfs_entry md_suspend_hi =
2953 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2956 reshape_position_show(mddev_t *mddev, char *page)
2958 if (mddev->reshape_position != MaxSector)
2959 return sprintf(page, "%llu\n",
2960 (unsigned long long)mddev->reshape_position);
2961 strcpy(page, "none\n");
2966 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2969 unsigned long long new = simple_strtoull(buf, &e, 10);
2972 if (buf == e || (*e && *e != '\n'))
2974 mddev->reshape_position = new;
2975 mddev->delta_disks = 0;
2976 mddev->new_level = mddev->level;
2977 mddev->new_layout = mddev->layout;
2978 mddev->new_chunk = mddev->chunk_size;
2982 static struct md_sysfs_entry md_reshape_position =
2983 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2984 reshape_position_store);
2987 static struct attribute *md_default_attrs[] = {
2990 &md_raid_disks.attr,
2991 &md_chunk_size.attr,
2993 &md_resync_start.attr,
2995 &md_new_device.attr,
2996 &md_safe_delay.attr,
2997 &md_array_state.attr,
2998 &md_reshape_position.attr,
3002 static struct attribute *md_redundancy_attrs[] = {
3004 &md_mismatches.attr,
3007 &md_sync_speed.attr,
3008 &md_sync_completed.attr,
3009 &md_suspend_lo.attr,
3010 &md_suspend_hi.attr,
3015 static struct attribute_group md_redundancy_group = {
3017 .attrs = md_redundancy_attrs,
3022 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3024 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3025 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3030 rv = mddev_lock(mddev);
3032 rv = entry->show(mddev, page);
3033 mddev_unlock(mddev);
3039 md_attr_store(struct kobject *kobj, struct attribute *attr,
3040 const char *page, size_t length)
3042 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3043 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3048 if (!capable(CAP_SYS_ADMIN))
3050 rv = mddev_lock(mddev);
3052 rv = entry->store(mddev, page, length);
3053 mddev_unlock(mddev);
3058 static void md_free(struct kobject *ko)
3060 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3064 static struct sysfs_ops md_sysfs_ops = {
3065 .show = md_attr_show,
3066 .store = md_attr_store,
3068 static struct kobj_type md_ktype = {
3070 .sysfs_ops = &md_sysfs_ops,
3071 .default_attrs = md_default_attrs,
3076 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3078 static DEFINE_MUTEX(disks_mutex);
3079 mddev_t *mddev = mddev_find(dev);
3080 struct gendisk *disk;
3081 int partitioned = (MAJOR(dev) != MD_MAJOR);
3082 int shift = partitioned ? MdpMinorShift : 0;
3083 int unit = MINOR(dev) >> shift;
3089 mutex_lock(&disks_mutex);
3090 if (mddev->gendisk) {
3091 mutex_unlock(&disks_mutex);
3095 disk = alloc_disk(1 << shift);
3097 mutex_unlock(&disks_mutex);
3101 disk->major = MAJOR(dev);
3102 disk->first_minor = unit << shift;
3104 sprintf(disk->disk_name, "md_d%d", unit);
3106 sprintf(disk->disk_name, "md%d", unit);
3107 disk->fops = &md_fops;
3108 disk->private_data = mddev;
3109 disk->queue = mddev->queue;
3111 mddev->gendisk = disk;
3112 mutex_unlock(&disks_mutex);
3113 error = kobject_init_and_add(&mddev->kobj, &md_ktype, &disk->dev.kobj,
3116 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3119 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3123 static void md_safemode_timeout(unsigned long data)
3125 mddev_t *mddev = (mddev_t *) data;
3127 mddev->safemode = 1;
3128 md_wakeup_thread(mddev->thread);
3131 static int start_dirty_degraded;
3133 static int do_md_run(mddev_t * mddev)
3137 struct list_head *tmp;
3139 struct gendisk *disk;
3140 struct mdk_personality *pers;
3141 char b[BDEVNAME_SIZE];
3143 if (list_empty(&mddev->disks))
3144 /* cannot run an array with no devices.. */
3151 * Analyze all RAID superblock(s)
3153 if (!mddev->raid_disks)
3156 chunk_size = mddev->chunk_size;
3159 if (chunk_size > MAX_CHUNK_SIZE) {
3160 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3161 chunk_size, MAX_CHUNK_SIZE);
3165 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3167 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3168 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3171 if (chunk_size < PAGE_SIZE) {
3172 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3173 chunk_size, PAGE_SIZE);
3177 /* devices must have minimum size of one chunk */
3178 ITERATE_RDEV(mddev,rdev,tmp) {
3179 if (test_bit(Faulty, &rdev->flags))
3181 if (rdev->size < chunk_size / 1024) {
3183 "md: Dev %s smaller than chunk_size:"
3185 bdevname(rdev->bdev,b),
3186 (unsigned long long)rdev->size,
3194 if (mddev->level != LEVEL_NONE)
3195 request_module("md-level-%d", mddev->level);
3196 else if (mddev->clevel[0])
3197 request_module("md-%s", mddev->clevel);
3201 * Drop all container device buffers, from now on
3202 * the only valid external interface is through the md
3205 ITERATE_RDEV(mddev,rdev,tmp) {
3206 if (test_bit(Faulty, &rdev->flags))
3208 sync_blockdev(rdev->bdev);
3209 invalidate_bdev(rdev->bdev);
3211 /* perform some consistency tests on the device.
3212 * We don't want the data to overlap the metadata,
3213 * Internal Bitmap issues has handled elsewhere.
3215 if (rdev->data_offset < rdev->sb_offset) {
3217 rdev->data_offset + mddev->size*2
3218 > rdev->sb_offset*2) {
3219 printk("md: %s: data overlaps metadata\n",
3224 if (rdev->sb_offset*2 + rdev->sb_size/512
3225 > rdev->data_offset) {
3226 printk("md: %s: metadata overlaps data\n",
3233 md_probe(mddev->unit, NULL, NULL);
3234 disk = mddev->gendisk;
3238 spin_lock(&pers_lock);
3239 pers = find_pers(mddev->level, mddev->clevel);
3240 if (!pers || !try_module_get(pers->owner)) {
3241 spin_unlock(&pers_lock);
3242 if (mddev->level != LEVEL_NONE)
3243 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3246 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3251 spin_unlock(&pers_lock);
3252 mddev->level = pers->level;
3253 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3255 if (mddev->reshape_position != MaxSector &&
3256 pers->start_reshape == NULL) {
3257 /* This personality cannot handle reshaping... */
3259 module_put(pers->owner);
3263 if (pers->sync_request) {
3264 /* Warn if this is a potentially silly
3267 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3269 struct list_head *tmp2;
3271 ITERATE_RDEV(mddev, rdev, tmp) {
3272 ITERATE_RDEV(mddev, rdev2, tmp2) {
3274 rdev->bdev->bd_contains ==
3275 rdev2->bdev->bd_contains) {
3277 "%s: WARNING: %s appears to be"
3278 " on the same physical disk as"
3281 bdevname(rdev->bdev,b),
3282 bdevname(rdev2->bdev,b2));
3289 "True protection against single-disk"
3290 " failure might be compromised.\n");
3293 mddev->recovery = 0;
3294 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3295 mddev->barriers_work = 1;
3296 mddev->ok_start_degraded = start_dirty_degraded;
3299 mddev->ro = 2; /* read-only, but switch on first write */
3301 err = mddev->pers->run(mddev);
3302 if (!err && mddev->pers->sync_request) {
3303 err = bitmap_create(mddev);
3305 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3306 mdname(mddev), err);
3307 mddev->pers->stop(mddev);
3311 printk(KERN_ERR "md: pers->run() failed ...\n");
3312 module_put(mddev->pers->owner);
3314 bitmap_destroy(mddev);
3317 if (mddev->pers->sync_request) {
3318 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3320 "md: cannot register extra attributes for %s\n",
3322 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3325 atomic_set(&mddev->writes_pending,0);
3326 mddev->safemode = 0;
3327 mddev->safemode_timer.function = md_safemode_timeout;
3328 mddev->safemode_timer.data = (unsigned long) mddev;
3329 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3332 ITERATE_RDEV(mddev,rdev,tmp)
3333 if (rdev->raid_disk >= 0) {
3335 sprintf(nm, "rd%d", rdev->raid_disk);
3336 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3337 printk("md: cannot register %s for %s\n",
3341 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3344 md_update_sb(mddev, 0);
3346 set_capacity(disk, mddev->array_size<<1);
3348 /* If we call blk_queue_make_request here, it will
3349 * re-initialise max_sectors etc which may have been
3350 * refined inside -> run. So just set the bits we need to set.
3351 * Most initialisation happended when we called
3352 * blk_queue_make_request(..., md_fail_request)
3355 mddev->queue->queuedata = mddev;
3356 mddev->queue->make_request_fn = mddev->pers->make_request;
3358 /* If there is a partially-recovered drive we need to
3359 * start recovery here. If we leave it to md_check_recovery,
3360 * it will remove the drives and not do the right thing
3362 if (mddev->degraded && !mddev->sync_thread) {
3363 struct list_head *rtmp;
3365 ITERATE_RDEV(mddev,rdev,rtmp)
3366 if (rdev->raid_disk >= 0 &&
3367 !test_bit(In_sync, &rdev->flags) &&
3368 !test_bit(Faulty, &rdev->flags))
3369 /* complete an interrupted recovery */
3371 if (spares && mddev->pers->sync_request) {
3372 mddev->recovery = 0;
3373 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3374 mddev->sync_thread = md_register_thread(md_do_sync,
3377 if (!mddev->sync_thread) {
3378 printk(KERN_ERR "%s: could not start resync"
3381 /* leave the spares where they are, it shouldn't hurt */
3382 mddev->recovery = 0;
3386 md_wakeup_thread(mddev->thread);
3387 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3390 md_new_event(mddev);
3391 kobject_uevent(&mddev->gendisk->dev.kobj, KOBJ_CHANGE);
3395 static int restart_array(mddev_t *mddev)
3397 struct gendisk *disk = mddev->gendisk;
3401 * Complain if it has no devices
3404 if (list_empty(&mddev->disks))
3412 mddev->safemode = 0;
3414 set_disk_ro(disk, 0);
3416 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3419 * Kick recovery or resync if necessary
3421 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3422 md_wakeup_thread(mddev->thread);
3423 md_wakeup_thread(mddev->sync_thread);
3432 /* similar to deny_write_access, but accounts for our holding a reference
3433 * to the file ourselves */
3434 static int deny_bitmap_write_access(struct file * file)
3436 struct inode *inode = file->f_mapping->host;
3438 spin_lock(&inode->i_lock);
3439 if (atomic_read(&inode->i_writecount) > 1) {
3440 spin_unlock(&inode->i_lock);
3443 atomic_set(&inode->i_writecount, -1);
3444 spin_unlock(&inode->i_lock);
3449 static void restore_bitmap_write_access(struct file *file)
3451 struct inode *inode = file->f_mapping->host;
3453 spin_lock(&inode->i_lock);
3454 atomic_set(&inode->i_writecount, 1);
3455 spin_unlock(&inode->i_lock);
3459 * 0 - completely stop and dis-assemble array
3460 * 1 - switch to readonly
3461 * 2 - stop but do not disassemble array
3463 static int do_md_stop(mddev_t * mddev, int mode)
3466 struct gendisk *disk = mddev->gendisk;
3469 if (atomic_read(&mddev->active)>2) {
3470 printk("md: %s still in use.\n",mdname(mddev));
3474 if (mddev->sync_thread) {
3475 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3476 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3477 md_unregister_thread(mddev->sync_thread);
3478 mddev->sync_thread = NULL;
3481 del_timer_sync(&mddev->safemode_timer);
3483 invalidate_partition(disk, 0);
3486 case 1: /* readonly */
3492 case 0: /* disassemble */
3494 bitmap_flush(mddev);
3495 md_super_wait(mddev);
3497 set_disk_ro(disk, 0);
3498 blk_queue_make_request(mddev->queue, md_fail_request);
3499 mddev->pers->stop(mddev);
3500 mddev->queue->merge_bvec_fn = NULL;
3501 mddev->queue->unplug_fn = NULL;
3502 mddev->queue->backing_dev_info.congested_fn = NULL;
3503 if (mddev->pers->sync_request)
3504 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3506 module_put(mddev->pers->owner);
3509 set_capacity(disk, 0);
3515 if (!mddev->in_sync || mddev->flags) {
3516 /* mark array as shutdown cleanly */
3518 md_update_sb(mddev, 1);
3521 set_disk_ro(disk, 1);
3522 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3526 * Free resources if final stop
3530 struct list_head *tmp;
3532 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3534 bitmap_destroy(mddev);
3535 if (mddev->bitmap_file) {
3536 restore_bitmap_write_access(mddev->bitmap_file);
3537 fput(mddev->bitmap_file);
3538 mddev->bitmap_file = NULL;
3540 mddev->bitmap_offset = 0;
3542 ITERATE_RDEV(mddev,rdev,tmp)
3543 if (rdev->raid_disk >= 0) {
3545 sprintf(nm, "rd%d", rdev->raid_disk);
3546 sysfs_remove_link(&mddev->kobj, nm);
3549 /* make sure all delayed_delete calls have finished */
3550 flush_scheduled_work();
3552 export_array(mddev);
3554 mddev->array_size = 0;
3556 mddev->raid_disks = 0;
3557 mddev->recovery_cp = 0;
3558 mddev->reshape_position = MaxSector;
3559 mddev->external = 0;
3561 } else if (mddev->pers)
3562 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3565 md_new_event(mddev);
3571 static void autorun_array(mddev_t *mddev)
3574 struct list_head *tmp;
3577 if (list_empty(&mddev->disks))
3580 printk(KERN_INFO "md: running: ");
3582 ITERATE_RDEV(mddev,rdev,tmp) {
3583 char b[BDEVNAME_SIZE];
3584 printk("<%s>", bdevname(rdev->bdev,b));
3588 err = do_md_run (mddev);
3590 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3591 do_md_stop (mddev, 0);
3596 * lets try to run arrays based on all disks that have arrived
3597 * until now. (those are in pending_raid_disks)
3599 * the method: pick the first pending disk, collect all disks with
3600 * the same UUID, remove all from the pending list and put them into
3601 * the 'same_array' list. Then order this list based on superblock
3602 * update time (freshest comes first), kick out 'old' disks and
3603 * compare superblocks. If everything's fine then run it.
3605 * If "unit" is allocated, then bump its reference count
3607 static void autorun_devices(int part)
3609 struct list_head *tmp;
3610 mdk_rdev_t *rdev0, *rdev;
3612 char b[BDEVNAME_SIZE];
3614 printk(KERN_INFO "md: autorun ...\n");
3615 while (!list_empty(&pending_raid_disks)) {
3618 LIST_HEAD(candidates);
3619 rdev0 = list_entry(pending_raid_disks.next,
3620 mdk_rdev_t, same_set);
3622 printk(KERN_INFO "md: considering %s ...\n",
3623 bdevname(rdev0->bdev,b));
3624 INIT_LIST_HEAD(&candidates);
3625 ITERATE_RDEV_PENDING(rdev,tmp)
3626 if (super_90_load(rdev, rdev0, 0) >= 0) {
3627 printk(KERN_INFO "md: adding %s ...\n",
3628 bdevname(rdev->bdev,b));
3629 list_move(&rdev->same_set, &candidates);
3632 * now we have a set of devices, with all of them having
3633 * mostly sane superblocks. It's time to allocate the
3637 dev = MKDEV(mdp_major,
3638 rdev0->preferred_minor << MdpMinorShift);
3639 unit = MINOR(dev) >> MdpMinorShift;
3641 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3644 if (rdev0->preferred_minor != unit) {
3645 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3646 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3650 md_probe(dev, NULL, NULL);
3651 mddev = mddev_find(dev);
3654 "md: cannot allocate memory for md drive.\n");
3657 if (mddev_lock(mddev))
3658 printk(KERN_WARNING "md: %s locked, cannot run\n",
3660 else if (mddev->raid_disks || mddev->major_version
3661 || !list_empty(&mddev->disks)) {
3663 "md: %s already running, cannot run %s\n",
3664 mdname(mddev), bdevname(rdev0->bdev,b));
3665 mddev_unlock(mddev);
3667 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3668 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3669 list_del_init(&rdev->same_set);
3670 if (bind_rdev_to_array(rdev, mddev))
3673 autorun_array(mddev);
3674 mddev_unlock(mddev);
3676 /* on success, candidates will be empty, on error
3679 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3683 printk(KERN_INFO "md: ... autorun DONE.\n");
3685 #endif /* !MODULE */
3687 static int get_version(void __user * arg)
3691 ver.major = MD_MAJOR_VERSION;
3692 ver.minor = MD_MINOR_VERSION;
3693 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3695 if (copy_to_user(arg, &ver, sizeof(ver)))
3701 static int get_array_info(mddev_t * mddev, void __user * arg)
3703 mdu_array_info_t info;
3704 int nr,working,active,failed,spare;
3706 struct list_head *tmp;
3708 nr=working=active=failed=spare=0;
3709 ITERATE_RDEV(mddev,rdev,tmp) {
3711 if (test_bit(Faulty, &rdev->flags))
3715 if (test_bit(In_sync, &rdev->flags))
3722 info.major_version = mddev->major_version;
3723 info.minor_version = mddev->minor_version;
3724 info.patch_version = MD_PATCHLEVEL_VERSION;
3725 info.ctime = mddev->ctime;
3726 info.level = mddev->level;
3727 info.size = mddev->size;
3728 if (info.size != mddev->size) /* overflow */
3731 info.raid_disks = mddev->raid_disks;
3732 info.md_minor = mddev->md_minor;
3733 info.not_persistent= !mddev->persistent;
3735 info.utime = mddev->utime;
3738 info.state = (1<<MD_SB_CLEAN);
3739 if (mddev->bitmap && mddev->bitmap_offset)
3740 info.state = (1<<MD_SB_BITMAP_PRESENT);
3741 info.active_disks = active;
3742 info.working_disks = working;
3743 info.failed_disks = failed;
3744 info.spare_disks = spare;
3746 info.layout = mddev->layout;
3747 info.chunk_size = mddev->chunk_size;
3749 if (copy_to_user(arg, &info, sizeof(info)))
3755 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3757 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3758 char *ptr, *buf = NULL;
3761 md_allow_write(mddev);
3763 file = kmalloc(sizeof(*file), GFP_KERNEL);
3767 /* bitmap disabled, zero the first byte and copy out */
3768 if (!mddev->bitmap || !mddev->bitmap->file) {
3769 file->pathname[0] = '\0';
3773 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3777 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3781 strcpy(file->pathname, ptr);
3785 if (copy_to_user(arg, file, sizeof(*file)))
3793 static int get_disk_info(mddev_t * mddev, void __user * arg)
3795 mdu_disk_info_t info;
3799 if (copy_from_user(&info, arg, sizeof(info)))
3804 rdev = find_rdev_nr(mddev, nr);
3806 info.major = MAJOR(rdev->bdev->bd_dev);
3807 info.minor = MINOR(rdev->bdev->bd_dev);
3808 info.raid_disk = rdev->raid_disk;
3810 if (test_bit(Faulty, &rdev->flags))
3811 info.state |= (1<<MD_DISK_FAULTY);
3812 else if (test_bit(In_sync, &rdev->flags)) {
3813 info.state |= (1<<MD_DISK_ACTIVE);
3814 info.state |= (1<<MD_DISK_SYNC);
3816 if (test_bit(WriteMostly, &rdev->flags))
3817 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3819 info.major = info.minor = 0;
3820 info.raid_disk = -1;
3821 info.state = (1<<MD_DISK_REMOVED);
3824 if (copy_to_user(arg, &info, sizeof(info)))
3830 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3832 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3834 dev_t dev = MKDEV(info->major,info->minor);
3836 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3839 if (!mddev->raid_disks) {
3841 /* expecting a device which has a superblock */
3842 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3845 "md: md_import_device returned %ld\n",
3847 return PTR_ERR(rdev);
3849 if (!list_empty(&mddev->disks)) {
3850 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3851 mdk_rdev_t, same_set);
3852 int err = super_types[mddev->major_version]
3853 .load_super(rdev, rdev0, mddev->minor_version);
3856 "md: %s has different UUID to %s\n",
3857 bdevname(rdev->bdev,b),
3858 bdevname(rdev0->bdev,b2));
3863 err = bind_rdev_to_array(rdev, mddev);
3870 * add_new_disk can be used once the array is assembled
3871 * to add "hot spares". They must already have a superblock
3876 if (!mddev->pers->hot_add_disk) {
3878 "%s: personality does not support diskops!\n",
3882 if (mddev->persistent)
3883 rdev = md_import_device(dev, mddev->major_version,
3884 mddev->minor_version);
3886 rdev = md_import_device(dev, -1, -1);
3889 "md: md_import_device returned %ld\n",
3891 return PTR_ERR(rdev);
3893 /* set save_raid_disk if appropriate */
3894 if (!mddev->persistent) {
3895 if (info->state & (1<<MD_DISK_SYNC) &&
3896 info->raid_disk < mddev->raid_disks)
3897 rdev->raid_disk = info->raid_disk;
3899 rdev->raid_disk = -1;
3901 super_types[mddev->major_version].
3902 validate_super(mddev, rdev);
3903 rdev->saved_raid_disk = rdev->raid_disk;
3905 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3906 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3907 set_bit(WriteMostly, &rdev->flags);
3909 rdev->raid_disk = -1;
3910 err = bind_rdev_to_array(rdev, mddev);
3911 if (!err && !mddev->pers->hot_remove_disk) {
3912 /* If there is hot_add_disk but no hot_remove_disk
3913 * then added disks for geometry changes,
3914 * and should be added immediately.
3916 super_types[mddev->major_version].
3917 validate_super(mddev, rdev);
3918 err = mddev->pers->hot_add_disk(mddev, rdev);
3920 unbind_rdev_from_array(rdev);
3925 md_update_sb(mddev, 1);
3926 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3927 md_wakeup_thread(mddev->thread);
3931 /* otherwise, add_new_disk is only allowed
3932 * for major_version==0 superblocks
3934 if (mddev->major_version != 0) {
3935 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3940 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3942 rdev = md_import_device (dev, -1, 0);
3945 "md: error, md_import_device() returned %ld\n",
3947 return PTR_ERR(rdev);
3949 rdev->desc_nr = info->number;
3950 if (info->raid_disk < mddev->raid_disks)
3951 rdev->raid_disk = info->raid_disk;
3953 rdev->raid_disk = -1;
3957 if (rdev->raid_disk < mddev->raid_disks)
3958 if (info->state & (1<<MD_DISK_SYNC))
3959 set_bit(In_sync, &rdev->flags);
3961 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3962 set_bit(WriteMostly, &rdev->flags);
3964 if (!mddev->persistent) {
3965 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3966 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3968 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3969 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3971 err = bind_rdev_to_array(rdev, mddev);
3981 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3983 char b[BDEVNAME_SIZE];
3989 rdev = find_rdev(mddev, dev);
3993 if (rdev->raid_disk >= 0)
3996 kick_rdev_from_array(rdev);
3997 md_update_sb(mddev, 1);
3998 md_new_event(mddev);
4002 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
4003 bdevname(rdev->bdev,b), mdname(mddev));
4007 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4009 char b[BDEVNAME_SIZE];
4017 if (mddev->major_version != 0) {
4018 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4019 " version-0 superblocks.\n",
4023 if (!mddev->pers->hot_add_disk) {
4025 "%s: personality does not support diskops!\n",
4030 rdev = md_import_device (dev, -1, 0);
4033 "md: error, md_import_device() returned %ld\n",
4038 if (mddev->persistent)
4039 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
4042 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
4044 size = calc_dev_size(rdev, mddev->chunk_size);
4047 if (test_bit(Faulty, &rdev->flags)) {
4049 "md: can not hot-add faulty %s disk to %s!\n",
4050 bdevname(rdev->bdev,b), mdname(mddev));
4054 clear_bit(In_sync, &rdev->flags);
4056 rdev->saved_raid_disk = -1;
4057 err = bind_rdev_to_array(rdev, mddev);
4062 * The rest should better be atomic, we can have disk failures
4063 * noticed in interrupt contexts ...
4066 if (rdev->desc_nr == mddev->max_disks) {
4067 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4070 goto abort_unbind_export;
4073 rdev->raid_disk = -1;
4075 md_update_sb(mddev, 1);
4078 * Kick recovery, maybe this spare has to be added to the
4079 * array immediately.
4081 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4082 md_wakeup_thread(mddev->thread);
4083 md_new_event(mddev);
4086 abort_unbind_export:
4087 unbind_rdev_from_array(rdev);
4094 static int set_bitmap_file(mddev_t *mddev, int fd)
4099 if (!mddev->pers->quiesce)
4101 if (mddev->recovery || mddev->sync_thread)
4103 /* we should be able to change the bitmap.. */
4109 return -EEXIST; /* cannot add when bitmap is present */
4110 mddev->bitmap_file = fget(fd);
4112 if (mddev->bitmap_file == NULL) {
4113 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4118 err = deny_bitmap_write_access(mddev->bitmap_file);
4120 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4122 fput(mddev->bitmap_file);
4123 mddev->bitmap_file = NULL;
4126 mddev->bitmap_offset = 0; /* file overrides offset */
4127 } else if (mddev->bitmap == NULL)
4128 return -ENOENT; /* cannot remove what isn't there */
4131 mddev->pers->quiesce(mddev, 1);
4133 err = bitmap_create(mddev);
4134 if (fd < 0 || err) {
4135 bitmap_destroy(mddev);
4136 fd = -1; /* make sure to put the file */
4138 mddev->pers->quiesce(mddev, 0);
4141 if (mddev->bitmap_file) {
4142 restore_bitmap_write_access(mddev->bitmap_file);
4143 fput(mddev->bitmap_file);
4145 mddev->bitmap_file = NULL;
4152 * set_array_info is used two different ways
4153 * The original usage is when creating a new array.
4154 * In this usage, raid_disks is > 0 and it together with
4155 * level, size, not_persistent,layout,chunksize determine the
4156 * shape of the array.
4157 * This will always create an array with a type-0.90.0 superblock.
4158 * The newer usage is when assembling an array.
4159 * In this case raid_disks will be 0, and the major_version field is
4160 * use to determine which style super-blocks are to be found on the devices.
4161 * The minor and patch _version numbers are also kept incase the
4162 * super_block handler wishes to interpret them.
4164 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4167 if (info->raid_disks == 0) {
4168 /* just setting version number for superblock loading */
4169 if (info->major_version < 0 ||
4170 info->major_version >= ARRAY_SIZE(super_types) ||
4171 super_types[info->major_version].name == NULL) {
4172 /* maybe try to auto-load a module? */
4174 "md: superblock version %d not known\n",
4175 info->major_version);
4178 mddev->major_version = info->major_version;
4179 mddev->minor_version = info->minor_version;
4180 mddev->patch_version = info->patch_version;
4181 mddev->persistent = !info->not_persistent;
4184 mddev->major_version = MD_MAJOR_VERSION;
4185 mddev->minor_version = MD_MINOR_VERSION;
4186 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4187 mddev->ctime = get_seconds();
4189 mddev->level = info->level;
4190 mddev->clevel[0] = 0;
4191 mddev->size = info->size;
4192 mddev->raid_disks = info->raid_disks;
4193 /* don't set md_minor, it is determined by which /dev/md* was
4196 if (info->state & (1<<MD_SB_CLEAN))
4197 mddev->recovery_cp = MaxSector;
4199 mddev->recovery_cp = 0;
4200 mddev->persistent = ! info->not_persistent;
4201 mddev->external = 0;
4203 mddev->layout = info->layout;
4204 mddev->chunk_size = info->chunk_size;
4206 mddev->max_disks = MD_SB_DISKS;
4208 if (mddev->persistent)
4210 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4212 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4213 mddev->bitmap_offset = 0;
4215 mddev->reshape_position = MaxSector;
4218 * Generate a 128 bit UUID
4220 get_random_bytes(mddev->uuid, 16);
4222 mddev->new_level = mddev->level;
4223 mddev->new_chunk = mddev->chunk_size;
4224 mddev->new_layout = mddev->layout;
4225 mddev->delta_disks = 0;
4230 static int update_size(mddev_t *mddev, unsigned long size)
4234 struct list_head *tmp;
4235 int fit = (size == 0);
4237 if (mddev->pers->resize == NULL)
4239 /* The "size" is the amount of each device that is used.
4240 * This can only make sense for arrays with redundancy.
4241 * linear and raid0 always use whatever space is available
4242 * We can only consider changing the size if no resync
4243 * or reconstruction is happening, and if the new size
4244 * is acceptable. It must fit before the sb_offset or,
4245 * if that is <data_offset, it must fit before the
4246 * size of each device.
4247 * If size is zero, we find the largest size that fits.
4249 if (mddev->sync_thread)
4251 ITERATE_RDEV(mddev,rdev,tmp) {
4253 avail = rdev->size * 2;
4255 if (fit && (size == 0 || size > avail/2))
4257 if (avail < ((sector_t)size << 1))
4260 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4262 struct block_device *bdev;
4264 bdev = bdget_disk(mddev->gendisk, 0);
4266 mutex_lock(&bdev->bd_inode->i_mutex);
4267 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4268 mutex_unlock(&bdev->bd_inode->i_mutex);
4275 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4278 /* change the number of raid disks */
4279 if (mddev->pers->check_reshape == NULL)
4281 if (raid_disks <= 0 ||
4282 raid_disks >= mddev->max_disks)
4284 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4286 mddev->delta_disks = raid_disks - mddev->raid_disks;
4288 rv = mddev->pers->check_reshape(mddev);
4294 * update_array_info is used to change the configuration of an
4296 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4297 * fields in the info are checked against the array.
4298 * Any differences that cannot be handled will cause an error.
4299 * Normally, only one change can be managed at a time.
4301 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4307 /* calculate expected state,ignoring low bits */
4308 if (mddev->bitmap && mddev->bitmap_offset)
4309 state |= (1 << MD_SB_BITMAP_PRESENT);
4311 if (mddev->major_version != info->major_version ||
4312 mddev->minor_version != info->minor_version ||
4313 /* mddev->patch_version != info->patch_version || */
4314 mddev->ctime != info->ctime ||
4315 mddev->level != info->level ||
4316 /* mddev->layout != info->layout || */
4317 !mddev->persistent != info->not_persistent||
4318 mddev->chunk_size != info->chunk_size ||
4319 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4320 ((state^info->state) & 0xfffffe00)
4323 /* Check there is only one change */
4324 if (info->size >= 0 && mddev->size != info->size) cnt++;
4325 if (mddev->raid_disks != info->raid_disks) cnt++;
4326 if (mddev->layout != info->layout) cnt++;
4327 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4328 if (cnt == 0) return 0;
4329 if (cnt > 1) return -EINVAL;
4331 if (mddev->layout != info->layout) {
4333 * we don't need to do anything at the md level, the
4334 * personality will take care of it all.
4336 if (mddev->pers->reconfig == NULL)
4339 return mddev->pers->reconfig(mddev, info->layout, -1);
4341 if (info->size >= 0 && mddev->size != info->size)
4342 rv = update_size(mddev, info->size);
4344 if (mddev->raid_disks != info->raid_disks)
4345 rv = update_raid_disks(mddev, info->raid_disks);
4347 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4348 if (mddev->pers->quiesce == NULL)
4350 if (mddev->recovery || mddev->sync_thread)
4352 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4353 /* add the bitmap */
4356 if (mddev->default_bitmap_offset == 0)
4358 mddev->bitmap_offset = mddev->default_bitmap_offset;
4359 mddev->pers->quiesce(mddev, 1);
4360 rv = bitmap_create(mddev);
4362 bitmap_destroy(mddev);
4363 mddev->pers->quiesce(mddev, 0);
4365 /* remove the bitmap */
4368 if (mddev->bitmap->file)
4370 mddev->pers->quiesce(mddev, 1);
4371 bitmap_destroy(mddev);
4372 mddev->pers->quiesce(mddev, 0);
4373 mddev->bitmap_offset = 0;
4376 md_update_sb(mddev, 1);
4380 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4384 if (mddev->pers == NULL)
4387 rdev = find_rdev(mddev, dev);
4391 md_error(mddev, rdev);
4395 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4397 mddev_t *mddev = bdev->bd_disk->private_data;
4401 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4405 static int md_ioctl(struct inode *inode, struct file *file,
4406 unsigned int cmd, unsigned long arg)
4409 void __user *argp = (void __user *)arg;
4410 mddev_t *mddev = NULL;
4412 if (!capable(CAP_SYS_ADMIN))
4416 * Commands dealing with the RAID driver but not any
4422 err = get_version(argp);
4425 case PRINT_RAID_DEBUG:
4433 autostart_arrays(arg);
4440 * Commands creating/starting a new array:
4443 mddev = inode->i_bdev->bd_disk->private_data;
4450 err = mddev_lock(mddev);
4453 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4460 case SET_ARRAY_INFO:
4462 mdu_array_info_t info;
4464 memset(&info, 0, sizeof(info));
4465 else if (copy_from_user(&info, argp, sizeof(info))) {
4470 err = update_array_info(mddev, &info);
4472 printk(KERN_WARNING "md: couldn't update"
4473 " array info. %d\n", err);
4478 if (!list_empty(&mddev->disks)) {
4480 "md: array %s already has disks!\n",
4485 if (mddev->raid_disks) {
4487 "md: array %s already initialised!\n",
4492 err = set_array_info(mddev, &info);
4494 printk(KERN_WARNING "md: couldn't set"
4495 " array info. %d\n", err);
4505 * Commands querying/configuring an existing array:
4507 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4508 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4509 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4510 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4511 && cmd != GET_BITMAP_FILE) {
4517 * Commands even a read-only array can execute:
4521 case GET_ARRAY_INFO:
4522 err = get_array_info(mddev, argp);
4525 case GET_BITMAP_FILE:
4526 err = get_bitmap_file(mddev, argp);
4530 err = get_disk_info(mddev, argp);
4533 case RESTART_ARRAY_RW:
4534 err = restart_array(mddev);
4538 err = do_md_stop (mddev, 0);
4542 err = do_md_stop (mddev, 1);
4546 * We have a problem here : there is no easy way to give a CHS
4547 * virtual geometry. We currently pretend that we have a 2 heads
4548 * 4 sectors (with a BIG number of cylinders...). This drives
4549 * dosfs just mad... ;-)
4554 * The remaining ioctls are changing the state of the
4555 * superblock, so we do not allow them on read-only arrays.
4556 * However non-MD ioctls (e.g. get-size) will still come through
4557 * here and hit the 'default' below, so only disallow
4558 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4560 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4561 mddev->ro && mddev->pers) {
4562 if (mddev->ro == 2) {
4564 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4565 md_wakeup_thread(mddev->thread);
4577 mdu_disk_info_t info;
4578 if (copy_from_user(&info, argp, sizeof(info)))
4581 err = add_new_disk(mddev, &info);
4585 case HOT_REMOVE_DISK:
4586 err = hot_remove_disk(mddev, new_decode_dev(arg));
4590 err = hot_add_disk(mddev, new_decode_dev(arg));
4593 case SET_DISK_FAULTY:
4594 err = set_disk_faulty(mddev, new_decode_dev(arg));
4598 err = do_md_run (mddev);
4601 case SET_BITMAP_FILE:
4602 err = set_bitmap_file(mddev, (int)arg);
4612 mddev_unlock(mddev);
4622 static int md_open(struct inode *inode, struct file *file)
4625 * Succeed if we can lock the mddev, which confirms that
4626 * it isn't being stopped right now.
4628 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4631 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4636 mddev_unlock(mddev);
4638 check_disk_change(inode->i_bdev);
4643 static int md_release(struct inode *inode, struct file * file)
4645 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4653 static int md_media_changed(struct gendisk *disk)
4655 mddev_t *mddev = disk->private_data;
4657 return mddev->changed;
4660 static int md_revalidate(struct gendisk *disk)
4662 mddev_t *mddev = disk->private_data;
4667 static struct block_device_operations md_fops =
4669 .owner = THIS_MODULE,
4671 .release = md_release,
4673 .getgeo = md_getgeo,
4674 .media_changed = md_media_changed,
4675 .revalidate_disk= md_revalidate,
4678 static int md_thread(void * arg)
4680 mdk_thread_t *thread = arg;
4683 * md_thread is a 'system-thread', it's priority should be very
4684 * high. We avoid resource deadlocks individually in each
4685 * raid personality. (RAID5 does preallocation) We also use RR and
4686 * the very same RT priority as kswapd, thus we will never get
4687 * into a priority inversion deadlock.
4689 * we definitely have to have equal or higher priority than
4690 * bdflush, otherwise bdflush will deadlock if there are too
4691 * many dirty RAID5 blocks.
4694 allow_signal(SIGKILL);
4695 while (!kthread_should_stop()) {
4697 /* We need to wait INTERRUPTIBLE so that
4698 * we don't add to the load-average.
4699 * That means we need to be sure no signals are
4702 if (signal_pending(current))
4703 flush_signals(current);
4705 wait_event_interruptible_timeout
4707 test_bit(THREAD_WAKEUP, &thread->flags)
4708 || kthread_should_stop(),
4711 clear_bit(THREAD_WAKEUP, &thread->flags);
4713 thread->run(thread->mddev);
4719 void md_wakeup_thread(mdk_thread_t *thread)
4722 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4723 set_bit(THREAD_WAKEUP, &thread->flags);
4724 wake_up(&thread->wqueue);
4728 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4731 mdk_thread_t *thread;
4733 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4737 init_waitqueue_head(&thread->wqueue);
4740 thread->mddev = mddev;
4741 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4742 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4743 if (IS_ERR(thread->tsk)) {
4750 void md_unregister_thread(mdk_thread_t *thread)
4752 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
4754 kthread_stop(thread->tsk);
4758 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4765 if (!rdev || test_bit(Faulty, &rdev->flags))
4768 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4770 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4771 __builtin_return_address(0),__builtin_return_address(1),
4772 __builtin_return_address(2),__builtin_return_address(3));
4776 if (!mddev->pers->error_handler)
4778 mddev->pers->error_handler(mddev,rdev);
4779 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4780 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4781 md_wakeup_thread(mddev->thread);
4782 md_new_event_inintr(mddev);
4785 /* seq_file implementation /proc/mdstat */
4787 static void status_unused(struct seq_file *seq)
4791 struct list_head *tmp;
4793 seq_printf(seq, "unused devices: ");
4795 ITERATE_RDEV_PENDING(rdev,tmp) {
4796 char b[BDEVNAME_SIZE];
4798 seq_printf(seq, "%s ",
4799 bdevname(rdev->bdev,b));
4802 seq_printf(seq, "<none>");
4804 seq_printf(seq, "\n");
4808 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4810 sector_t max_blocks, resync, res;
4811 unsigned long dt, db, rt;
4813 unsigned int per_milli;
4815 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4817 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4818 max_blocks = mddev->resync_max_sectors >> 1;
4820 max_blocks = mddev->size;
4823 * Should not happen.
4829 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4830 * in a sector_t, and (max_blocks>>scale) will fit in a
4831 * u32, as those are the requirements for sector_div.
4832 * Thus 'scale' must be at least 10
4835 if (sizeof(sector_t) > sizeof(unsigned long)) {
4836 while ( max_blocks/2 > (1ULL<<(scale+32)))
4839 res = (resync>>scale)*1000;
4840 sector_div(res, (u32)((max_blocks>>scale)+1));
4844 int i, x = per_milli/50, y = 20-x;
4845 seq_printf(seq, "[");
4846 for (i = 0; i < x; i++)
4847 seq_printf(seq, "=");
4848 seq_printf(seq, ">");
4849 for (i = 0; i < y; i++)
4850 seq_printf(seq, ".");
4851 seq_printf(seq, "] ");
4853 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4854 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4856 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4858 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4859 "resync" : "recovery"))),
4860 per_milli/10, per_milli % 10,
4861 (unsigned long long) resync,
4862 (unsigned long long) max_blocks);
4865 * We do not want to overflow, so the order of operands and
4866 * the * 100 / 100 trick are important. We do a +1 to be
4867 * safe against division by zero. We only estimate anyway.
4869 * dt: time from mark until now
4870 * db: blocks written from mark until now
4871 * rt: remaining time
4873 dt = ((jiffies - mddev->resync_mark) / HZ);
4875 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4876 - mddev->resync_mark_cnt;
4877 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4879 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4881 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4884 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4886 struct list_head *tmp;
4896 spin_lock(&all_mddevs_lock);
4897 list_for_each(tmp,&all_mddevs)
4899 mddev = list_entry(tmp, mddev_t, all_mddevs);
4901 spin_unlock(&all_mddevs_lock);
4904 spin_unlock(&all_mddevs_lock);
4906 return (void*)2;/* tail */
4910 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4912 struct list_head *tmp;
4913 mddev_t *next_mddev, *mddev = v;
4919 spin_lock(&all_mddevs_lock);
4921 tmp = all_mddevs.next;
4923 tmp = mddev->all_mddevs.next;
4924 if (tmp != &all_mddevs)
4925 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4927 next_mddev = (void*)2;
4930 spin_unlock(&all_mddevs_lock);
4938 static void md_seq_stop(struct seq_file *seq, void *v)
4942 if (mddev && v != (void*)1 && v != (void*)2)
4946 struct mdstat_info {
4950 static int md_seq_show(struct seq_file *seq, void *v)
4954 struct list_head *tmp2;
4956 struct mdstat_info *mi = seq->private;
4957 struct bitmap *bitmap;
4959 if (v == (void*)1) {
4960 struct mdk_personality *pers;
4961 seq_printf(seq, "Personalities : ");
4962 spin_lock(&pers_lock);
4963 list_for_each_entry(pers, &pers_list, list)
4964 seq_printf(seq, "[%s] ", pers->name);
4966 spin_unlock(&pers_lock);
4967 seq_printf(seq, "\n");
4968 mi->event = atomic_read(&md_event_count);
4971 if (v == (void*)2) {
4976 if (mddev_lock(mddev) < 0)
4979 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4980 seq_printf(seq, "%s : %sactive", mdname(mddev),
4981 mddev->pers ? "" : "in");
4984 seq_printf(seq, " (read-only)");
4986 seq_printf(seq, "(auto-read-only)");
4987 seq_printf(seq, " %s", mddev->pers->name);
4991 ITERATE_RDEV(mddev,rdev,tmp2) {
4992 char b[BDEVNAME_SIZE];
4993 seq_printf(seq, " %s[%d]",
4994 bdevname(rdev->bdev,b), rdev->desc_nr);
4995 if (test_bit(WriteMostly, &rdev->flags))
4996 seq_printf(seq, "(W)");
4997 if (test_bit(Faulty, &rdev->flags)) {
4998 seq_printf(seq, "(F)");
5000 } else if (rdev->raid_disk < 0)
5001 seq_printf(seq, "(S)"); /* spare */
5005 if (!list_empty(&mddev->disks)) {
5007 seq_printf(seq, "\n %llu blocks",
5008 (unsigned long long)mddev->array_size);
5010 seq_printf(seq, "\n %llu blocks",
5011 (unsigned long long)size);
5013 if (mddev->persistent) {
5014 if (mddev->major_version != 0 ||
5015 mddev->minor_version != 90) {
5016 seq_printf(seq," super %d.%d",
5017 mddev->major_version,
5018 mddev->minor_version);
5020 } else if (mddev->external)
5021 seq_printf(seq, " super external:%s",
5022 mddev->metadata_type);
5024 seq_printf(seq, " super non-persistent");
5027 mddev->pers->status (seq, mddev);
5028 seq_printf(seq, "\n ");
5029 if (mddev->pers->sync_request) {
5030 if (mddev->curr_resync > 2) {
5031 status_resync (seq, mddev);
5032 seq_printf(seq, "\n ");
5033 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5034 seq_printf(seq, "\tresync=DELAYED\n ");
5035 else if (mddev->recovery_cp < MaxSector)
5036 seq_printf(seq, "\tresync=PENDING\n ");
5039 seq_printf(seq, "\n ");
5041 if ((bitmap = mddev->bitmap)) {
5042 unsigned long chunk_kb;
5043 unsigned long flags;
5044 spin_lock_irqsave(&bitmap->lock, flags);
5045 chunk_kb = bitmap->chunksize >> 10;
5046 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5048 bitmap->pages - bitmap->missing_pages,
5050 (bitmap->pages - bitmap->missing_pages)
5051 << (PAGE_SHIFT - 10),
5052 chunk_kb ? chunk_kb : bitmap->chunksize,
5053 chunk_kb ? "KB" : "B");
5055 seq_printf(seq, ", file: ");
5056 seq_path(seq, bitmap->file->f_path.mnt,
5057 bitmap->file->f_path.dentry," \t\n");
5060 seq_printf(seq, "\n");
5061 spin_unlock_irqrestore(&bitmap->lock, flags);
5064 seq_printf(seq, "\n");
5066 mddev_unlock(mddev);
5071 static struct seq_operations md_seq_ops = {
5072 .start = md_seq_start,
5073 .next = md_seq_next,
5074 .stop = md_seq_stop,
5075 .show = md_seq_show,
5078 static int md_seq_open(struct inode *inode, struct file *file)
5081 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5085 error = seq_open(file, &md_seq_ops);
5089 struct seq_file *p = file->private_data;
5091 mi->event = atomic_read(&md_event_count);
5096 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5098 struct seq_file *m = filp->private_data;
5099 struct mdstat_info *mi = m->private;
5102 poll_wait(filp, &md_event_waiters, wait);
5104 /* always allow read */
5105 mask = POLLIN | POLLRDNORM;
5107 if (mi->event != atomic_read(&md_event_count))
5108 mask |= POLLERR | POLLPRI;
5112 static const struct file_operations md_seq_fops = {
5113 .owner = THIS_MODULE,
5114 .open = md_seq_open,
5116 .llseek = seq_lseek,
5117 .release = seq_release_private,
5118 .poll = mdstat_poll,
5121 int register_md_personality(struct mdk_personality *p)
5123 spin_lock(&pers_lock);
5124 list_add_tail(&p->list, &pers_list);
5125 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5126 spin_unlock(&pers_lock);
5130 int unregister_md_personality(struct mdk_personality *p)
5132 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5133 spin_lock(&pers_lock);
5134 list_del_init(&p->list);
5135 spin_unlock(&pers_lock);
5139 static int is_mddev_idle(mddev_t *mddev)
5142 struct list_head *tmp;
5147 ITERATE_RDEV(mddev,rdev,tmp) {
5148 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5149 curr_events = disk_stat_read(disk, sectors[0]) +
5150 disk_stat_read(disk, sectors[1]) -
5151 atomic_read(&disk->sync_io);
5152 /* sync IO will cause sync_io to increase before the disk_stats
5153 * as sync_io is counted when a request starts, and
5154 * disk_stats is counted when it completes.
5155 * So resync activity will cause curr_events to be smaller than
5156 * when there was no such activity.
5157 * non-sync IO will cause disk_stat to increase without
5158 * increasing sync_io so curr_events will (eventually)
5159 * be larger than it was before. Once it becomes
5160 * substantially larger, the test below will cause
5161 * the array to appear non-idle, and resync will slow
5163 * If there is a lot of outstanding resync activity when
5164 * we set last_event to curr_events, then all that activity
5165 * completing might cause the array to appear non-idle
5166 * and resync will be slowed down even though there might
5167 * not have been non-resync activity. This will only
5168 * happen once though. 'last_events' will soon reflect
5169 * the state where there is little or no outstanding
5170 * resync requests, and further resync activity will
5171 * always make curr_events less than last_events.
5174 if (curr_events - rdev->last_events > 4096) {
5175 rdev->last_events = curr_events;
5182 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5184 /* another "blocks" (512byte) blocks have been synced */
5185 atomic_sub(blocks, &mddev->recovery_active);
5186 wake_up(&mddev->recovery_wait);
5188 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5189 md_wakeup_thread(mddev->thread);
5190 // stop recovery, signal do_sync ....
5195 /* md_write_start(mddev, bi)
5196 * If we need to update some array metadata (e.g. 'active' flag
5197 * in superblock) before writing, schedule a superblock update
5198 * and wait for it to complete.
5200 void md_write_start(mddev_t *mddev, struct bio *bi)
5202 if (bio_data_dir(bi) != WRITE)
5205 BUG_ON(mddev->ro == 1);
5206 if (mddev->ro == 2) {
5207 /* need to switch to read/write */
5209 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5210 md_wakeup_thread(mddev->thread);
5212 atomic_inc(&mddev->writes_pending);
5213 if (mddev->in_sync) {
5214 spin_lock_irq(&mddev->write_lock);
5215 if (mddev->in_sync) {
5217 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5218 md_wakeup_thread(mddev->thread);
5220 spin_unlock_irq(&mddev->write_lock);
5222 wait_event(mddev->sb_wait, mddev->flags==0);
5225 void md_write_end(mddev_t *mddev)
5227 if (atomic_dec_and_test(&mddev->writes_pending)) {
5228 if (mddev->safemode == 2)
5229 md_wakeup_thread(mddev->thread);
5230 else if (mddev->safemode_delay)
5231 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5235 /* md_allow_write(mddev)
5236 * Calling this ensures that the array is marked 'active' so that writes
5237 * may proceed without blocking. It is important to call this before
5238 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5239 * Must be called with mddev_lock held.
5241 void md_allow_write(mddev_t *mddev)
5248 spin_lock_irq(&mddev->write_lock);
5249 if (mddev->in_sync) {
5251 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5252 if (mddev->safemode_delay &&
5253 mddev->safemode == 0)
5254 mddev->safemode = 1;
5255 spin_unlock_irq(&mddev->write_lock);
5256 md_update_sb(mddev, 0);
5258 spin_unlock_irq(&mddev->write_lock);
5260 EXPORT_SYMBOL_GPL(md_allow_write);
5262 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5264 #define SYNC_MARKS 10
5265 #define SYNC_MARK_STEP (3*HZ)
5266 void md_do_sync(mddev_t *mddev)
5269 unsigned int currspeed = 0,
5271 sector_t max_sectors,j, io_sectors;
5272 unsigned long mark[SYNC_MARKS];
5273 sector_t mark_cnt[SYNC_MARKS];
5275 struct list_head *tmp;
5276 sector_t last_check;
5278 struct list_head *rtmp;
5282 /* just incase thread restarts... */
5283 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5285 if (mddev->ro) /* never try to sync a read-only array */
5288 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5289 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5290 desc = "data-check";
5291 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5292 desc = "requested-resync";
5295 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5300 /* we overload curr_resync somewhat here.
5301 * 0 == not engaged in resync at all
5302 * 2 == checking that there is no conflict with another sync
5303 * 1 == like 2, but have yielded to allow conflicting resync to
5305 * other == active in resync - this many blocks
5307 * Before starting a resync we must have set curr_resync to
5308 * 2, and then checked that every "conflicting" array has curr_resync
5309 * less than ours. When we find one that is the same or higher
5310 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5311 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5312 * This will mean we have to start checking from the beginning again.
5317 mddev->curr_resync = 2;
5320 if (kthread_should_stop()) {
5321 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5324 ITERATE_MDDEV(mddev2,tmp) {
5325 if (mddev2 == mddev)
5327 if (mddev2->curr_resync &&
5328 match_mddev_units(mddev,mddev2)) {
5330 if (mddev < mddev2 && mddev->curr_resync == 2) {
5331 /* arbitrarily yield */
5332 mddev->curr_resync = 1;
5333 wake_up(&resync_wait);
5335 if (mddev > mddev2 && mddev->curr_resync == 1)
5336 /* no need to wait here, we can wait the next
5337 * time 'round when curr_resync == 2
5340 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5341 if (!kthread_should_stop() &&
5342 mddev2->curr_resync >= mddev->curr_resync) {
5343 printk(KERN_INFO "md: delaying %s of %s"
5344 " until %s has finished (they"
5345 " share one or more physical units)\n",
5346 desc, mdname(mddev), mdname(mddev2));
5349 finish_wait(&resync_wait, &wq);
5352 finish_wait(&resync_wait, &wq);
5355 } while (mddev->curr_resync < 2);
5358 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5359 /* resync follows the size requested by the personality,
5360 * which defaults to physical size, but can be virtual size
5362 max_sectors = mddev->resync_max_sectors;
5363 mddev->resync_mismatches = 0;
5364 /* we don't use the checkpoint if there's a bitmap */
5365 if (!mddev->bitmap &&
5366 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5367 j = mddev->recovery_cp;
5368 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5369 max_sectors = mddev->size << 1;
5371 /* recovery follows the physical size of devices */
5372 max_sectors = mddev->size << 1;
5374 ITERATE_RDEV(mddev,rdev,rtmp)
5375 if (rdev->raid_disk >= 0 &&
5376 !test_bit(Faulty, &rdev->flags) &&
5377 !test_bit(In_sync, &rdev->flags) &&
5378 rdev->recovery_offset < j)
5379 j = rdev->recovery_offset;
5382 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5383 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
5384 " %d KB/sec/disk.\n", speed_min(mddev));
5385 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5386 "(but not more than %d KB/sec) for %s.\n",
5387 speed_max(mddev), desc);
5389 is_mddev_idle(mddev); /* this also initializes IO event counters */
5392 for (m = 0; m < SYNC_MARKS; m++) {
5394 mark_cnt[m] = io_sectors;
5397 mddev->resync_mark = mark[last_mark];
5398 mddev->resync_mark_cnt = mark_cnt[last_mark];
5401 * Tune reconstruction:
5403 window = 32*(PAGE_SIZE/512);
5404 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5405 window/2,(unsigned long long) max_sectors/2);
5407 atomic_set(&mddev->recovery_active, 0);
5408 init_waitqueue_head(&mddev->recovery_wait);
5413 "md: resuming %s of %s from checkpoint.\n",
5414 desc, mdname(mddev));
5415 mddev->curr_resync = j;
5418 while (j < max_sectors) {
5422 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5423 currspeed < speed_min(mddev));
5425 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5429 if (!skipped) { /* actual IO requested */
5430 io_sectors += sectors;
5431 atomic_add(sectors, &mddev->recovery_active);
5435 if (j>1) mddev->curr_resync = j;
5436 mddev->curr_mark_cnt = io_sectors;
5437 if (last_check == 0)
5438 /* this is the earliers that rebuilt will be
5439 * visible in /proc/mdstat
5441 md_new_event(mddev);
5443 if (last_check + window > io_sectors || j == max_sectors)
5446 last_check = io_sectors;
5448 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5449 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5453 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5455 int next = (last_mark+1) % SYNC_MARKS;
5457 mddev->resync_mark = mark[next];
5458 mddev->resync_mark_cnt = mark_cnt[next];
5459 mark[next] = jiffies;
5460 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5465 if (kthread_should_stop()) {
5467 * got a signal, exit.
5470 "md: md_do_sync() got signal ... exiting\n");
5471 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5476 * this loop exits only if either when we are slower than
5477 * the 'hard' speed limit, or the system was IO-idle for
5479 * the system might be non-idle CPU-wise, but we only care
5480 * about not overloading the IO subsystem. (things like an
5481 * e2fsck being done on the RAID array should execute fast)
5483 blk_unplug(mddev->queue);
5486 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5487 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5489 if (currspeed > speed_min(mddev)) {
5490 if ((currspeed > speed_max(mddev)) ||
5491 !is_mddev_idle(mddev)) {
5497 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5499 * this also signals 'finished resyncing' to md_stop
5502 blk_unplug(mddev->queue);
5504 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5506 /* tell personality that we are finished */
5507 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5509 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5510 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5511 mddev->curr_resync > 2) {
5512 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5513 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5514 if (mddev->curr_resync >= mddev->recovery_cp) {
5516 "md: checkpointing %s of %s.\n",
5517 desc, mdname(mddev));
5518 mddev->recovery_cp = mddev->curr_resync;
5521 mddev->recovery_cp = MaxSector;
5523 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5524 mddev->curr_resync = MaxSector;
5525 ITERATE_RDEV(mddev,rdev,rtmp)
5526 if (rdev->raid_disk >= 0 &&
5527 !test_bit(Faulty, &rdev->flags) &&
5528 !test_bit(In_sync, &rdev->flags) &&
5529 rdev->recovery_offset < mddev->curr_resync)
5530 rdev->recovery_offset = mddev->curr_resync;
5533 set_bit(MD_CHANGE_DEVS, &mddev->flags);
5536 mddev->curr_resync = 0;
5537 wake_up(&resync_wait);
5538 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5539 md_wakeup_thread(mddev->thread);
5541 EXPORT_SYMBOL_GPL(md_do_sync);
5544 static int remove_and_add_spares(mddev_t *mddev)
5547 struct list_head *rtmp;
5550 ITERATE_RDEV(mddev,rdev,rtmp)
5551 if (rdev->raid_disk >= 0 &&
5552 (test_bit(Faulty, &rdev->flags) ||
5553 ! test_bit(In_sync, &rdev->flags)) &&
5554 atomic_read(&rdev->nr_pending)==0) {
5555 if (mddev->pers->hot_remove_disk(
5556 mddev, rdev->raid_disk)==0) {
5558 sprintf(nm,"rd%d", rdev->raid_disk);
5559 sysfs_remove_link(&mddev->kobj, nm);
5560 rdev->raid_disk = -1;
5564 if (mddev->degraded) {
5565 ITERATE_RDEV(mddev,rdev,rtmp)
5566 if (rdev->raid_disk < 0
5567 && !test_bit(Faulty, &rdev->flags)) {
5568 rdev->recovery_offset = 0;
5569 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5571 sprintf(nm, "rd%d", rdev->raid_disk);
5572 if (sysfs_create_link(&mddev->kobj,
5575 "md: cannot register "
5579 md_new_event(mddev);
5587 * This routine is regularly called by all per-raid-array threads to
5588 * deal with generic issues like resync and super-block update.
5589 * Raid personalities that don't have a thread (linear/raid0) do not
5590 * need this as they never do any recovery or update the superblock.
5592 * It does not do any resync itself, but rather "forks" off other threads
5593 * to do that as needed.
5594 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5595 * "->recovery" and create a thread at ->sync_thread.
5596 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5597 * and wakeups up this thread which will reap the thread and finish up.
5598 * This thread also removes any faulty devices (with nr_pending == 0).
5600 * The overall approach is:
5601 * 1/ if the superblock needs updating, update it.
5602 * 2/ If a recovery thread is running, don't do anything else.
5603 * 3/ If recovery has finished, clean up, possibly marking spares active.
5604 * 4/ If there are any faulty devices, remove them.
5605 * 5/ If array is degraded, try to add spares devices
5606 * 6/ If array has spares or is not in-sync, start a resync thread.
5608 void md_check_recovery(mddev_t *mddev)
5611 struct list_head *rtmp;
5615 bitmap_daemon_work(mddev->bitmap);
5620 if (signal_pending(current)) {
5621 if (mddev->pers->sync_request) {
5622 printk(KERN_INFO "md: %s in immediate safe mode\n",
5624 mddev->safemode = 2;
5626 flush_signals(current);
5630 (mddev->flags && !mddev->external) ||
5631 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5632 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5633 (mddev->safemode == 1) ||
5634 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5635 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5639 if (mddev_trylock(mddev)) {
5642 spin_lock_irq(&mddev->write_lock);
5643 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5644 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5646 if (mddev->persistent)
5647 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5649 if (mddev->safemode == 1)
5650 mddev->safemode = 0;
5651 spin_unlock_irq(&mddev->write_lock);
5654 md_update_sb(mddev, 0);
5657 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5658 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5659 /* resync/recovery still happening */
5660 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5663 if (mddev->sync_thread) {
5664 /* resync has finished, collect result */
5665 md_unregister_thread(mddev->sync_thread);
5666 mddev->sync_thread = NULL;
5667 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5668 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5670 /* activate any spares */
5671 mddev->pers->spare_active(mddev);
5673 md_update_sb(mddev, 1);
5675 /* if array is no-longer degraded, then any saved_raid_disk
5676 * information must be scrapped
5678 if (!mddev->degraded)
5679 ITERATE_RDEV(mddev,rdev,rtmp)
5680 rdev->saved_raid_disk = -1;
5682 mddev->recovery = 0;
5683 /* flag recovery needed just to double check */
5684 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5685 md_new_event(mddev);
5688 /* Clear some bits that don't mean anything, but
5691 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5692 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5693 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5694 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5696 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5698 /* no recovery is running.
5699 * remove any failed drives, then
5700 * add spares if possible.
5701 * Spare are also removed and re-added, to allow
5702 * the personality to fail the re-add.
5705 if (mddev->reshape_position != MaxSector) {
5706 if (mddev->pers->check_reshape(mddev) != 0)
5707 /* Cannot proceed */
5709 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5710 } else if ((spares = remove_and_add_spares(mddev))) {
5711 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5712 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5713 } else if (mddev->recovery_cp < MaxSector) {
5714 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5715 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5716 /* nothing to be done ... */
5719 if (mddev->pers->sync_request) {
5720 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5721 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5722 /* We are adding a device or devices to an array
5723 * which has the bitmap stored on all devices.
5724 * So make sure all bitmap pages get written
5726 bitmap_write_all(mddev->bitmap);
5728 mddev->sync_thread = md_register_thread(md_do_sync,
5731 if (!mddev->sync_thread) {
5732 printk(KERN_ERR "%s: could not start resync"
5735 /* leave the spares where they are, it shouldn't hurt */
5736 mddev->recovery = 0;
5738 md_wakeup_thread(mddev->sync_thread);
5739 md_new_event(mddev);
5742 mddev_unlock(mddev);
5746 static int md_notify_reboot(struct notifier_block *this,
5747 unsigned long code, void *x)
5749 struct list_head *tmp;
5752 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5754 printk(KERN_INFO "md: stopping all md devices.\n");
5756 ITERATE_MDDEV(mddev,tmp)
5757 if (mddev_trylock(mddev)) {
5758 do_md_stop (mddev, 1);
5759 mddev_unlock(mddev);
5762 * certain more exotic SCSI devices are known to be
5763 * volatile wrt too early system reboots. While the
5764 * right place to handle this issue is the given
5765 * driver, we do want to have a safe RAID driver ...
5772 static struct notifier_block md_notifier = {
5773 .notifier_call = md_notify_reboot,
5775 .priority = INT_MAX, /* before any real devices */
5778 static void md_geninit(void)
5780 struct proc_dir_entry *p;
5782 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5784 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5786 p->proc_fops = &md_seq_fops;
5789 static int __init md_init(void)
5791 if (register_blkdev(MAJOR_NR, "md"))
5793 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5794 unregister_blkdev(MAJOR_NR, "md");
5797 blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5798 md_probe, NULL, NULL);
5799 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5800 md_probe, NULL, NULL);
5802 register_reboot_notifier(&md_notifier);
5803 raid_table_header = register_sysctl_table(raid_root_table);
5813 * Searches all registered partitions for autorun RAID arrays
5817 static LIST_HEAD(all_detected_devices);
5818 struct detected_devices_node {
5819 struct list_head list;
5823 void md_autodetect_dev(dev_t dev)
5825 struct detected_devices_node *node_detected_dev;
5827 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
5828 if (node_detected_dev) {
5829 node_detected_dev->dev = dev;
5830 list_add_tail(&node_detected_dev->list, &all_detected_devices);
5832 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
5833 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
5838 static void autostart_arrays(int part)
5841 struct detected_devices_node *node_detected_dev;
5843 int i_scanned, i_passed;
5848 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5850 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
5852 node_detected_dev = list_entry(all_detected_devices.next,
5853 struct detected_devices_node, list);
5854 list_del(&node_detected_dev->list);
5855 dev = node_detected_dev->dev;
5856 kfree(node_detected_dev);
5857 rdev = md_import_device(dev,0, 90);
5861 if (test_bit(Faulty, &rdev->flags)) {
5865 list_add(&rdev->same_set, &pending_raid_disks);
5869 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
5870 i_scanned, i_passed);
5872 autorun_devices(part);
5875 #endif /* !MODULE */
5877 static __exit void md_exit(void)
5880 struct list_head *tmp;
5882 blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5883 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5885 unregister_blkdev(MAJOR_NR,"md");
5886 unregister_blkdev(mdp_major, "mdp");
5887 unregister_reboot_notifier(&md_notifier);
5888 unregister_sysctl_table(raid_table_header);
5889 remove_proc_entry("mdstat", NULL);
5890 ITERATE_MDDEV(mddev,tmp) {
5891 struct gendisk *disk = mddev->gendisk;
5894 export_array(mddev);
5897 mddev->gendisk = NULL;
5902 subsys_initcall(md_init);
5903 module_exit(md_exit)
5905 static int get_ro(char *buffer, struct kernel_param *kp)
5907 return sprintf(buffer, "%d", start_readonly);
5909 static int set_ro(const char *val, struct kernel_param *kp)
5912 int num = simple_strtoul(val, &e, 10);
5913 if (*val && (*e == '\0' || *e == '\n')) {
5914 start_readonly = num;
5920 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5921 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5924 EXPORT_SYMBOL(register_md_personality);
5925 EXPORT_SYMBOL(unregister_md_personality);
5926 EXPORT_SYMBOL(md_error);
5927 EXPORT_SYMBOL(md_done_sync);
5928 EXPORT_SYMBOL(md_write_start);
5929 EXPORT_SYMBOL(md_write_end);
5930 EXPORT_SYMBOL(md_register_thread);
5931 EXPORT_SYMBOL(md_unregister_thread);
5932 EXPORT_SYMBOL(md_wakeup_thread);
5933 EXPORT_SYMBOL(md_check_recovery);
5934 MODULE_LICENSE("GPL");
5936 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);