2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45 #include <linux/poll.h>
46 #include <linux/mutex.h>
47 #include <linux/ctype.h>
49 #include <linux/init.h>
51 #include <linux/file.h>
54 #include <linux/kmod.h>
57 #include <asm/unaligned.h>
59 #define MAJOR_NR MD_MAJOR
62 /* 63 partitions with the alternate major number (mdp) */
63 #define MdpMinorShift 6
66 #define dprintk(x...) ((void)(DEBUG && printk(x)))
70 static void autostart_arrays (int part);
73 static LIST_HEAD(pers_list);
74 static DEFINE_SPINLOCK(pers_lock);
76 static void md_print_devices(void);
78 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
81 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
82 * is 1000 KB/sec, so the extra system load does not show up that much.
83 * Increase it if you want to have more _guaranteed_ speed. Note that
84 * the RAID driver will use the maximum available bandwidth if the IO
85 * subsystem is idle. There is also an 'absolute maximum' reconstruction
86 * speed limit - in case reconstruction slows down your system despite
89 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
90 * or /sys/block/mdX/md/sync_speed_{min,max}
93 static int sysctl_speed_limit_min = 1000;
94 static int sysctl_speed_limit_max = 200000;
95 static inline int speed_min(mddev_t *mddev)
97 return mddev->sync_speed_min ?
98 mddev->sync_speed_min : sysctl_speed_limit_min;
101 static inline int speed_max(mddev_t *mddev)
103 return mddev->sync_speed_max ?
104 mddev->sync_speed_max : sysctl_speed_limit_max;
107 static struct ctl_table_header *raid_table_header;
109 static ctl_table raid_table[] = {
111 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
112 .procname = "speed_limit_min",
113 .data = &sysctl_speed_limit_min,
114 .maxlen = sizeof(int),
116 .proc_handler = &proc_dointvec,
119 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
120 .procname = "speed_limit_max",
121 .data = &sysctl_speed_limit_max,
122 .maxlen = sizeof(int),
124 .proc_handler = &proc_dointvec,
129 static ctl_table raid_dir_table[] = {
131 .ctl_name = DEV_RAID,
140 static ctl_table raid_root_table[] = {
146 .child = raid_dir_table,
151 static struct block_device_operations md_fops;
153 static int start_readonly;
156 * We have a system wide 'event count' that is incremented
157 * on any 'interesting' event, and readers of /proc/mdstat
158 * can use 'poll' or 'select' to find out when the event
162 * start array, stop array, error, add device, remove device,
163 * start build, activate spare
165 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
166 static atomic_t md_event_count;
167 void md_new_event(mddev_t *mddev)
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
171 sysfs_notify(&mddev->kobj, NULL, "sync_action");
173 EXPORT_SYMBOL_GPL(md_new_event);
175 /* Alternate version that can be called from interrupts
176 * when calling sysfs_notify isn't needed.
178 void md_new_event_inintr(mddev_t *mddev)
180 atomic_inc(&md_event_count);
181 wake_up(&md_event_waiters);
185 * Enables to iterate over all existing md arrays
186 * all_mddevs_lock protects this list.
188 static LIST_HEAD(all_mddevs);
189 static DEFINE_SPINLOCK(all_mddevs_lock);
193 * iterates through all used mddevs in the system.
194 * We take care to grab the all_mddevs_lock whenever navigating
195 * the list, and to always hold a refcount when unlocked.
196 * Any code which breaks out of this loop while own
197 * a reference to the current mddev and must mddev_put it.
199 #define ITERATE_MDDEV(mddev,tmp) \
201 for (({ spin_lock(&all_mddevs_lock); \
202 tmp = all_mddevs.next; \
204 ({ if (tmp != &all_mddevs) \
205 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
206 spin_unlock(&all_mddevs_lock); \
207 if (mddev) mddev_put(mddev); \
208 mddev = list_entry(tmp, mddev_t, all_mddevs); \
209 tmp != &all_mddevs;}); \
210 ({ spin_lock(&all_mddevs_lock); \
215 static int md_fail_request (request_queue_t *q, struct bio *bio)
217 bio_io_error(bio, bio->bi_size);
221 static inline mddev_t *mddev_get(mddev_t *mddev)
223 atomic_inc(&mddev->active);
227 static void mddev_put(mddev_t *mddev)
229 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
231 if (!mddev->raid_disks && list_empty(&mddev->disks)) {
232 list_del(&mddev->all_mddevs);
233 spin_unlock(&all_mddevs_lock);
234 blk_cleanup_queue(mddev->queue);
235 kobject_unregister(&mddev->kobj);
237 spin_unlock(&all_mddevs_lock);
240 static mddev_t * mddev_find(dev_t unit)
242 mddev_t *mddev, *new = NULL;
245 spin_lock(&all_mddevs_lock);
246 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
247 if (mddev->unit == unit) {
249 spin_unlock(&all_mddevs_lock);
255 list_add(&new->all_mddevs, &all_mddevs);
256 spin_unlock(&all_mddevs_lock);
259 spin_unlock(&all_mddevs_lock);
261 new = kzalloc(sizeof(*new), GFP_KERNEL);
266 if (MAJOR(unit) == MD_MAJOR)
267 new->md_minor = MINOR(unit);
269 new->md_minor = MINOR(unit) >> MdpMinorShift;
271 mutex_init(&new->reconfig_mutex);
272 INIT_LIST_HEAD(&new->disks);
273 INIT_LIST_HEAD(&new->all_mddevs);
274 init_timer(&new->safemode_timer);
275 atomic_set(&new->active, 1);
276 spin_lock_init(&new->write_lock);
277 init_waitqueue_head(&new->sb_wait);
279 new->queue = blk_alloc_queue(GFP_KERNEL);
284 set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
286 blk_queue_make_request(new->queue, md_fail_request);
291 static inline int mddev_lock(mddev_t * mddev)
293 return mutex_lock_interruptible(&mddev->reconfig_mutex);
296 static inline int mddev_trylock(mddev_t * mddev)
298 return mutex_trylock(&mddev->reconfig_mutex);
301 static inline void mddev_unlock(mddev_t * mddev)
303 mutex_unlock(&mddev->reconfig_mutex);
305 md_wakeup_thread(mddev->thread);
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
311 struct list_head *tmp;
313 ITERATE_RDEV(mddev,rdev,tmp) {
314 if (rdev->desc_nr == nr)
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
322 struct list_head *tmp;
325 ITERATE_RDEV(mddev,rdev,tmp) {
326 if (rdev->bdev->bd_dev == dev)
332 static struct mdk_personality *find_pers(int level, char *clevel)
334 struct mdk_personality *pers;
335 list_for_each_entry(pers, &pers_list, list) {
336 if (level != LEVEL_NONE && pers->level == level)
338 if (strcmp(pers->name, clevel)==0)
344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
346 sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
347 return MD_NEW_SIZE_BLOCKS(size);
350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
354 size = rdev->sb_offset;
357 size &= ~((sector_t)chunk_size/1024 - 1);
361 static int alloc_disk_sb(mdk_rdev_t * rdev)
366 rdev->sb_page = alloc_page(GFP_KERNEL);
367 if (!rdev->sb_page) {
368 printk(KERN_ALERT "md: out of memory.\n");
375 static void free_disk_sb(mdk_rdev_t * rdev)
378 put_page(rdev->sb_page);
380 rdev->sb_page = NULL;
387 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
389 mdk_rdev_t *rdev = bio->bi_private;
390 mddev_t *mddev = rdev->mddev;
394 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
395 md_error(mddev, rdev);
397 if (atomic_dec_and_test(&mddev->pending_writes))
398 wake_up(&mddev->sb_wait);
403 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
405 struct bio *bio2 = bio->bi_private;
406 mdk_rdev_t *rdev = bio2->bi_private;
407 mddev_t *mddev = rdev->mddev;
411 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
412 error == -EOPNOTSUPP) {
414 /* barriers don't appear to be supported :-( */
415 set_bit(BarriersNotsupp, &rdev->flags);
416 mddev->barriers_work = 0;
417 spin_lock_irqsave(&mddev->write_lock, flags);
418 bio2->bi_next = mddev->biolist;
419 mddev->biolist = bio2;
420 spin_unlock_irqrestore(&mddev->write_lock, flags);
421 wake_up(&mddev->sb_wait);
426 bio->bi_private = rdev;
427 return super_written(bio, bytes_done, error);
430 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
431 sector_t sector, int size, struct page *page)
433 /* write first size bytes of page to sector of rdev
434 * Increment mddev->pending_writes before returning
435 * and decrement it on completion, waking up sb_wait
436 * if zero is reached.
437 * If an error occurred, call md_error
439 * As we might need to resubmit the request if BIO_RW_BARRIER
440 * causes ENOTSUPP, we allocate a spare bio...
442 struct bio *bio = bio_alloc(GFP_NOIO, 1);
443 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
445 bio->bi_bdev = rdev->bdev;
446 bio->bi_sector = sector;
447 bio_add_page(bio, page, size, 0);
448 bio->bi_private = rdev;
449 bio->bi_end_io = super_written;
452 atomic_inc(&mddev->pending_writes);
453 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
455 rw |= (1<<BIO_RW_BARRIER);
456 rbio = bio_clone(bio, GFP_NOIO);
457 rbio->bi_private = bio;
458 rbio->bi_end_io = super_written_barrier;
459 submit_bio(rw, rbio);
464 void md_super_wait(mddev_t *mddev)
466 /* wait for all superblock writes that were scheduled to complete.
467 * if any had to be retried (due to BARRIER problems), retry them
471 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
472 if (atomic_read(&mddev->pending_writes)==0)
474 while (mddev->biolist) {
476 spin_lock_irq(&mddev->write_lock);
477 bio = mddev->biolist;
478 mddev->biolist = bio->bi_next ;
480 spin_unlock_irq(&mddev->write_lock);
481 submit_bio(bio->bi_rw, bio);
485 finish_wait(&mddev->sb_wait, &wq);
488 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
493 complete((struct completion*)bio->bi_private);
497 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
498 struct page *page, int rw)
500 struct bio *bio = bio_alloc(GFP_NOIO, 1);
501 struct completion event;
504 rw |= (1 << BIO_RW_SYNC);
507 bio->bi_sector = sector;
508 bio_add_page(bio, page, size, 0);
509 init_completion(&event);
510 bio->bi_private = &event;
511 bio->bi_end_io = bi_complete;
513 wait_for_completion(&event);
515 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
519 EXPORT_SYMBOL_GPL(sync_page_io);
521 static int read_disk_sb(mdk_rdev_t * rdev, int size)
523 char b[BDEVNAME_SIZE];
524 if (!rdev->sb_page) {
532 if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
538 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
539 bdevname(rdev->bdev,b));
543 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
545 if ( (sb1->set_uuid0 == sb2->set_uuid0) &&
546 (sb1->set_uuid1 == sb2->set_uuid1) &&
547 (sb1->set_uuid2 == sb2->set_uuid2) &&
548 (sb1->set_uuid3 == sb2->set_uuid3))
556 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
559 mdp_super_t *tmp1, *tmp2;
561 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
562 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
564 if (!tmp1 || !tmp2) {
566 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
574 * nr_disks is not constant
579 if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
590 static unsigned int calc_sb_csum(mdp_super_t * sb)
592 unsigned int disk_csum, csum;
594 disk_csum = sb->sb_csum;
596 csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
597 sb->sb_csum = disk_csum;
603 * Handle superblock details.
604 * We want to be able to handle multiple superblock formats
605 * so we have a common interface to them all, and an array of
606 * different handlers.
607 * We rely on user-space to write the initial superblock, and support
608 * reading and updating of superblocks.
609 * Interface methods are:
610 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
611 * loads and validates a superblock on dev.
612 * if refdev != NULL, compare superblocks on both devices
614 * 0 - dev has a superblock that is compatible with refdev
615 * 1 - dev has a superblock that is compatible and newer than refdev
616 * so dev should be used as the refdev in future
617 * -EINVAL superblock incompatible or invalid
618 * -othererror e.g. -EIO
620 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
621 * Verify that dev is acceptable into mddev.
622 * The first time, mddev->raid_disks will be 0, and data from
623 * dev should be merged in. Subsequent calls check that dev
624 * is new enough. Return 0 or -EINVAL
626 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
627 * Update the superblock for rdev with data in mddev
628 * This does not write to disc.
634 struct module *owner;
635 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
636 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
637 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
641 * load_super for 0.90.0
643 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
645 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
651 * Calculate the position of the superblock,
652 * it's at the end of the disk.
654 * It also happens to be a multiple of 4Kb.
656 sb_offset = calc_dev_sboffset(rdev->bdev);
657 rdev->sb_offset = sb_offset;
659 ret = read_disk_sb(rdev, MD_SB_BYTES);
664 bdevname(rdev->bdev, b);
665 sb = (mdp_super_t*)page_address(rdev->sb_page);
667 if (sb->md_magic != MD_SB_MAGIC) {
668 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
673 if (sb->major_version != 0 ||
674 sb->minor_version < 90 ||
675 sb->minor_version > 91) {
676 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
677 sb->major_version, sb->minor_version,
682 if (sb->raid_disks <= 0)
685 if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
686 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
691 rdev->preferred_minor = sb->md_minor;
692 rdev->data_offset = 0;
693 rdev->sb_size = MD_SB_BYTES;
695 if (sb->level == LEVEL_MULTIPATH)
698 rdev->desc_nr = sb->this_disk.number;
704 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
705 if (!uuid_equal(refsb, sb)) {
706 printk(KERN_WARNING "md: %s has different UUID to %s\n",
707 b, bdevname(refdev->bdev,b2));
710 if (!sb_equal(refsb, sb)) {
711 printk(KERN_WARNING "md: %s has same UUID"
712 " but different superblock to %s\n",
713 b, bdevname(refdev->bdev, b2));
717 ev2 = md_event(refsb);
723 rdev->size = calc_dev_size(rdev, sb->chunk_size);
725 if (rdev->size < sb->size && sb->level > 1)
726 /* "this cannot possibly happen" ... */
734 * validate_super for 0.90.0
736 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
739 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
740 __u64 ev1 = md_event(sb);
742 rdev->raid_disk = -1;
744 if (mddev->raid_disks == 0) {
745 mddev->major_version = 0;
746 mddev->minor_version = sb->minor_version;
747 mddev->patch_version = sb->patch_version;
748 mddev->persistent = ! sb->not_persistent;
749 mddev->chunk_size = sb->chunk_size;
750 mddev->ctime = sb->ctime;
751 mddev->utime = sb->utime;
752 mddev->level = sb->level;
753 mddev->clevel[0] = 0;
754 mddev->layout = sb->layout;
755 mddev->raid_disks = sb->raid_disks;
756 mddev->size = sb->size;
758 mddev->bitmap_offset = 0;
759 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
761 if (mddev->minor_version >= 91) {
762 mddev->reshape_position = sb->reshape_position;
763 mddev->delta_disks = sb->delta_disks;
764 mddev->new_level = sb->new_level;
765 mddev->new_layout = sb->new_layout;
766 mddev->new_chunk = sb->new_chunk;
768 mddev->reshape_position = MaxSector;
769 mddev->delta_disks = 0;
770 mddev->new_level = mddev->level;
771 mddev->new_layout = mddev->layout;
772 mddev->new_chunk = mddev->chunk_size;
775 if (sb->state & (1<<MD_SB_CLEAN))
776 mddev->recovery_cp = MaxSector;
778 if (sb->events_hi == sb->cp_events_hi &&
779 sb->events_lo == sb->cp_events_lo) {
780 mddev->recovery_cp = sb->recovery_cp;
782 mddev->recovery_cp = 0;
785 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
786 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
787 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
788 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
790 mddev->max_disks = MD_SB_DISKS;
792 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
793 mddev->bitmap_file == NULL) {
794 if (mddev->level != 1 && mddev->level != 4
795 && mddev->level != 5 && mddev->level != 6
796 && mddev->level != 10) {
797 /* FIXME use a better test */
798 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
801 mddev->bitmap_offset = mddev->default_bitmap_offset;
804 } else if (mddev->pers == NULL) {
805 /* Insist on good event counter while assembling */
807 if (ev1 < mddev->events)
809 } else if (mddev->bitmap) {
810 /* if adding to array with a bitmap, then we can accept an
811 * older device ... but not too old.
813 if (ev1 < mddev->bitmap->events_cleared)
816 if (ev1 < mddev->events)
817 /* just a hot-add of a new device, leave raid_disk at -1 */
821 if (mddev->level != LEVEL_MULTIPATH) {
822 desc = sb->disks + rdev->desc_nr;
824 if (desc->state & (1<<MD_DISK_FAULTY))
825 set_bit(Faulty, &rdev->flags);
826 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
827 desc->raid_disk < mddev->raid_disks */) {
828 set_bit(In_sync, &rdev->flags);
829 rdev->raid_disk = desc->raid_disk;
831 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
832 set_bit(WriteMostly, &rdev->flags);
833 } else /* MULTIPATH are always insync */
834 set_bit(In_sync, &rdev->flags);
839 * sync_super for 0.90.0
841 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
844 struct list_head *tmp;
846 int next_spare = mddev->raid_disks;
849 /* make rdev->sb match mddev data..
852 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
853 * 3/ any empty disks < next_spare become removed
855 * disks[0] gets initialised to REMOVED because
856 * we cannot be sure from other fields if it has
857 * been initialised or not.
860 int active=0, working=0,failed=0,spare=0,nr_disks=0;
862 rdev->sb_size = MD_SB_BYTES;
864 sb = (mdp_super_t*)page_address(rdev->sb_page);
866 memset(sb, 0, sizeof(*sb));
868 sb->md_magic = MD_SB_MAGIC;
869 sb->major_version = mddev->major_version;
870 sb->patch_version = mddev->patch_version;
871 sb->gvalid_words = 0; /* ignored */
872 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
873 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
874 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
875 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
877 sb->ctime = mddev->ctime;
878 sb->level = mddev->level;
879 sb->size = mddev->size;
880 sb->raid_disks = mddev->raid_disks;
881 sb->md_minor = mddev->md_minor;
882 sb->not_persistent = !mddev->persistent;
883 sb->utime = mddev->utime;
885 sb->events_hi = (mddev->events>>32);
886 sb->events_lo = (u32)mddev->events;
888 if (mddev->reshape_position == MaxSector)
889 sb->minor_version = 90;
891 sb->minor_version = 91;
892 sb->reshape_position = mddev->reshape_position;
893 sb->new_level = mddev->new_level;
894 sb->delta_disks = mddev->delta_disks;
895 sb->new_layout = mddev->new_layout;
896 sb->new_chunk = mddev->new_chunk;
898 mddev->minor_version = sb->minor_version;
901 sb->recovery_cp = mddev->recovery_cp;
902 sb->cp_events_hi = (mddev->events>>32);
903 sb->cp_events_lo = (u32)mddev->events;
904 if (mddev->recovery_cp == MaxSector)
905 sb->state = (1<< MD_SB_CLEAN);
909 sb->layout = mddev->layout;
910 sb->chunk_size = mddev->chunk_size;
912 if (mddev->bitmap && mddev->bitmap_file == NULL)
913 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
915 sb->disks[0].state = (1<<MD_DISK_REMOVED);
916 ITERATE_RDEV(mddev,rdev2,tmp) {
919 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
920 && !test_bit(Faulty, &rdev2->flags))
921 desc_nr = rdev2->raid_disk;
923 desc_nr = next_spare++;
924 rdev2->desc_nr = desc_nr;
925 d = &sb->disks[rdev2->desc_nr];
927 d->number = rdev2->desc_nr;
928 d->major = MAJOR(rdev2->bdev->bd_dev);
929 d->minor = MINOR(rdev2->bdev->bd_dev);
930 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
931 && !test_bit(Faulty, &rdev2->flags))
932 d->raid_disk = rdev2->raid_disk;
934 d->raid_disk = rdev2->desc_nr; /* compatibility */
935 if (test_bit(Faulty, &rdev2->flags))
936 d->state = (1<<MD_DISK_FAULTY);
937 else if (test_bit(In_sync, &rdev2->flags)) {
938 d->state = (1<<MD_DISK_ACTIVE);
939 d->state |= (1<<MD_DISK_SYNC);
947 if (test_bit(WriteMostly, &rdev2->flags))
948 d->state |= (1<<MD_DISK_WRITEMOSTLY);
950 /* now set the "removed" and "faulty" bits on any missing devices */
951 for (i=0 ; i < mddev->raid_disks ; i++) {
952 mdp_disk_t *d = &sb->disks[i];
953 if (d->state == 0 && d->number == 0) {
956 d->state = (1<<MD_DISK_REMOVED);
957 d->state |= (1<<MD_DISK_FAULTY);
961 sb->nr_disks = nr_disks;
962 sb->active_disks = active;
963 sb->working_disks = working;
964 sb->failed_disks = failed;
965 sb->spare_disks = spare;
967 sb->this_disk = sb->disks[rdev->desc_nr];
968 sb->sb_csum = calc_sb_csum(sb);
972 * version 1 superblock
975 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
977 unsigned int disk_csum, csum;
978 unsigned long long newcsum;
979 int size = 256 + le32_to_cpu(sb->max_dev)*2;
980 unsigned int *isuper = (unsigned int*)sb;
983 disk_csum = sb->sb_csum;
986 for (i=0; size>=4; size -= 4 )
987 newcsum += le32_to_cpu(*isuper++);
990 newcsum += le16_to_cpu(*(unsigned short*) isuper);
992 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
993 sb->sb_csum = disk_csum;
994 return cpu_to_le32(csum);
997 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
999 struct mdp_superblock_1 *sb;
1002 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1006 * Calculate the position of the superblock.
1007 * It is always aligned to a 4K boundary and
1008 * depeding on minor_version, it can be:
1009 * 0: At least 8K, but less than 12K, from end of device
1010 * 1: At start of device
1011 * 2: 4K from start of device.
1013 switch(minor_version) {
1015 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1017 sb_offset &= ~(sector_t)(4*2-1);
1018 /* convert from sectors to K */
1030 rdev->sb_offset = sb_offset;
1032 /* superblock is rarely larger than 1K, but it can be larger,
1033 * and it is safe to read 4k, so we do that
1035 ret = read_disk_sb(rdev, 4096);
1036 if (ret) return ret;
1039 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1041 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1042 sb->major_version != cpu_to_le32(1) ||
1043 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1044 le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1045 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1048 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1049 printk("md: invalid superblock checksum on %s\n",
1050 bdevname(rdev->bdev,b));
1053 if (le64_to_cpu(sb->data_size) < 10) {
1054 printk("md: data_size too small on %s\n",
1055 bdevname(rdev->bdev,b));
1058 rdev->preferred_minor = 0xffff;
1059 rdev->data_offset = le64_to_cpu(sb->data_offset);
1060 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1062 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1063 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1064 if (rdev->sb_size & bmask)
1065 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1071 struct mdp_superblock_1 *refsb =
1072 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1074 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1075 sb->level != refsb->level ||
1076 sb->layout != refsb->layout ||
1077 sb->chunksize != refsb->chunksize) {
1078 printk(KERN_WARNING "md: %s has strangely different"
1079 " superblock to %s\n",
1080 bdevname(rdev->bdev,b),
1081 bdevname(refdev->bdev,b2));
1084 ev1 = le64_to_cpu(sb->events);
1085 ev2 = le64_to_cpu(refsb->events);
1093 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1095 rdev->size = rdev->sb_offset;
1096 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1098 rdev->size = le64_to_cpu(sb->data_size)/2;
1099 if (le32_to_cpu(sb->chunksize))
1100 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1102 if (le32_to_cpu(sb->size) > rdev->size*2)
1107 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1109 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1110 __u64 ev1 = le64_to_cpu(sb->events);
1112 rdev->raid_disk = -1;
1114 if (mddev->raid_disks == 0) {
1115 mddev->major_version = 1;
1116 mddev->patch_version = 0;
1117 mddev->persistent = 1;
1118 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1119 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1120 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1121 mddev->level = le32_to_cpu(sb->level);
1122 mddev->clevel[0] = 0;
1123 mddev->layout = le32_to_cpu(sb->layout);
1124 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1125 mddev->size = le64_to_cpu(sb->size)/2;
1126 mddev->events = ev1;
1127 mddev->bitmap_offset = 0;
1128 mddev->default_bitmap_offset = 1024 >> 9;
1130 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1131 memcpy(mddev->uuid, sb->set_uuid, 16);
1133 mddev->max_disks = (4096-256)/2;
1135 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1136 mddev->bitmap_file == NULL ) {
1137 if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1138 && mddev->level != 10) {
1139 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1142 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1144 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1145 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1146 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1147 mddev->new_level = le32_to_cpu(sb->new_level);
1148 mddev->new_layout = le32_to_cpu(sb->new_layout);
1149 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1151 mddev->reshape_position = MaxSector;
1152 mddev->delta_disks = 0;
1153 mddev->new_level = mddev->level;
1154 mddev->new_layout = mddev->layout;
1155 mddev->new_chunk = mddev->chunk_size;
1158 } else if (mddev->pers == NULL) {
1159 /* Insist of good event counter while assembling */
1161 if (ev1 < mddev->events)
1163 } else if (mddev->bitmap) {
1164 /* If adding to array with a bitmap, then we can accept an
1165 * older device, but not too old.
1167 if (ev1 < mddev->bitmap->events_cleared)
1170 if (ev1 < mddev->events)
1171 /* just a hot-add of a new device, leave raid_disk at -1 */
1174 if (mddev->level != LEVEL_MULTIPATH) {
1176 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1177 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1179 case 0xffff: /* spare */
1181 case 0xfffe: /* faulty */
1182 set_bit(Faulty, &rdev->flags);
1185 if ((le32_to_cpu(sb->feature_map) &
1186 MD_FEATURE_RECOVERY_OFFSET))
1187 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1189 set_bit(In_sync, &rdev->flags);
1190 rdev->raid_disk = role;
1193 if (sb->devflags & WriteMostly1)
1194 set_bit(WriteMostly, &rdev->flags);
1195 } else /* MULTIPATH are always insync */
1196 set_bit(In_sync, &rdev->flags);
1201 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1203 struct mdp_superblock_1 *sb;
1204 struct list_head *tmp;
1207 /* make rdev->sb match mddev and rdev data. */
1209 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1211 sb->feature_map = 0;
1213 sb->recovery_offset = cpu_to_le64(0);
1214 memset(sb->pad1, 0, sizeof(sb->pad1));
1215 memset(sb->pad2, 0, sizeof(sb->pad2));
1216 memset(sb->pad3, 0, sizeof(sb->pad3));
1218 sb->utime = cpu_to_le64((__u64)mddev->utime);
1219 sb->events = cpu_to_le64(mddev->events);
1221 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1223 sb->resync_offset = cpu_to_le64(0);
1225 sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1227 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1228 sb->size = cpu_to_le64(mddev->size<<1);
1230 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1231 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1232 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1235 if (rdev->raid_disk >= 0 &&
1236 !test_bit(In_sync, &rdev->flags) &&
1237 rdev->recovery_offset > 0) {
1238 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1239 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1242 if (mddev->reshape_position != MaxSector) {
1243 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1244 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1245 sb->new_layout = cpu_to_le32(mddev->new_layout);
1246 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1247 sb->new_level = cpu_to_le32(mddev->new_level);
1248 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1252 ITERATE_RDEV(mddev,rdev2,tmp)
1253 if (rdev2->desc_nr+1 > max_dev)
1254 max_dev = rdev2->desc_nr+1;
1256 sb->max_dev = cpu_to_le32(max_dev);
1257 for (i=0; i<max_dev;i++)
1258 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1260 ITERATE_RDEV(mddev,rdev2,tmp) {
1262 if (test_bit(Faulty, &rdev2->flags))
1263 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1264 else if (test_bit(In_sync, &rdev2->flags))
1265 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1266 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1267 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1269 sb->dev_roles[i] = cpu_to_le16(0xffff);
1272 sb->sb_csum = calc_sb_1_csum(sb);
1276 static struct super_type super_types[] = {
1279 .owner = THIS_MODULE,
1280 .load_super = super_90_load,
1281 .validate_super = super_90_validate,
1282 .sync_super = super_90_sync,
1286 .owner = THIS_MODULE,
1287 .load_super = super_1_load,
1288 .validate_super = super_1_validate,
1289 .sync_super = super_1_sync,
1293 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1295 struct list_head *tmp;
1298 ITERATE_RDEV(mddev,rdev,tmp)
1299 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1305 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1307 struct list_head *tmp;
1310 ITERATE_RDEV(mddev1,rdev,tmp)
1311 if (match_dev_unit(mddev2, rdev))
1317 static LIST_HEAD(pending_raid_disks);
1319 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1321 mdk_rdev_t *same_pdev;
1322 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1330 /* make sure rdev->size exceeds mddev->size */
1331 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1333 /* Cannot change size, so fail */
1336 mddev->size = rdev->size;
1338 same_pdev = match_dev_unit(mddev, rdev);
1341 "%s: WARNING: %s appears to be on the same physical"
1342 " disk as %s. True\n protection against single-disk"
1343 " failure might be compromised.\n",
1344 mdname(mddev), bdevname(rdev->bdev,b),
1345 bdevname(same_pdev->bdev,b2));
1347 /* Verify rdev->desc_nr is unique.
1348 * If it is -1, assign a free number, else
1349 * check number is not in use
1351 if (rdev->desc_nr < 0) {
1353 if (mddev->pers) choice = mddev->raid_disks;
1354 while (find_rdev_nr(mddev, choice))
1356 rdev->desc_nr = choice;
1358 if (find_rdev_nr(mddev, rdev->desc_nr))
1361 bdevname(rdev->bdev,b);
1362 if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1364 while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1367 list_add(&rdev->same_set, &mddev->disks);
1368 rdev->mddev = mddev;
1369 printk(KERN_INFO "md: bind<%s>\n", b);
1371 rdev->kobj.parent = &mddev->kobj;
1372 kobject_add(&rdev->kobj);
1374 if (rdev->bdev->bd_part)
1375 ko = &rdev->bdev->bd_part->kobj;
1377 ko = &rdev->bdev->bd_disk->kobj;
1378 sysfs_create_link(&rdev->kobj, ko, "block");
1379 bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1383 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1385 char b[BDEVNAME_SIZE];
1390 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1391 list_del_init(&rdev->same_set);
1392 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1394 sysfs_remove_link(&rdev->kobj, "block");
1395 kobject_del(&rdev->kobj);
1399 * prevent the device from being mounted, repartitioned or
1400 * otherwise reused by a RAID array (or any other kernel
1401 * subsystem), by bd_claiming the device.
1403 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1406 struct block_device *bdev;
1407 char b[BDEVNAME_SIZE];
1409 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1411 printk(KERN_ERR "md: could not open %s.\n",
1412 __bdevname(dev, b));
1413 return PTR_ERR(bdev);
1415 err = bd_claim(bdev, rdev);
1417 printk(KERN_ERR "md: could not bd_claim %s.\n",
1426 static void unlock_rdev(mdk_rdev_t *rdev)
1428 struct block_device *bdev = rdev->bdev;
1436 void md_autodetect_dev(dev_t dev);
1438 static void export_rdev(mdk_rdev_t * rdev)
1440 char b[BDEVNAME_SIZE];
1441 printk(KERN_INFO "md: export_rdev(%s)\n",
1442 bdevname(rdev->bdev,b));
1446 list_del_init(&rdev->same_set);
1448 md_autodetect_dev(rdev->bdev->bd_dev);
1451 kobject_put(&rdev->kobj);
1454 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1456 unbind_rdev_from_array(rdev);
1460 static void export_array(mddev_t *mddev)
1462 struct list_head *tmp;
1465 ITERATE_RDEV(mddev,rdev,tmp) {
1470 kick_rdev_from_array(rdev);
1472 if (!list_empty(&mddev->disks))
1474 mddev->raid_disks = 0;
1475 mddev->major_version = 0;
1478 static void print_desc(mdp_disk_t *desc)
1480 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1481 desc->major,desc->minor,desc->raid_disk,desc->state);
1484 static void print_sb(mdp_super_t *sb)
1489 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1490 sb->major_version, sb->minor_version, sb->patch_version,
1491 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1493 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1494 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1495 sb->md_minor, sb->layout, sb->chunk_size);
1496 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1497 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1498 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1499 sb->failed_disks, sb->spare_disks,
1500 sb->sb_csum, (unsigned long)sb->events_lo);
1503 for (i = 0; i < MD_SB_DISKS; i++) {
1506 desc = sb->disks + i;
1507 if (desc->number || desc->major || desc->minor ||
1508 desc->raid_disk || (desc->state && (desc->state != 4))) {
1509 printk(" D %2d: ", i);
1513 printk(KERN_INFO "md: THIS: ");
1514 print_desc(&sb->this_disk);
1518 static void print_rdev(mdk_rdev_t *rdev)
1520 char b[BDEVNAME_SIZE];
1521 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1522 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1523 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1525 if (rdev->sb_loaded) {
1526 printk(KERN_INFO "md: rdev superblock:\n");
1527 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1529 printk(KERN_INFO "md: no rdev superblock!\n");
1532 static void md_print_devices(void)
1534 struct list_head *tmp, *tmp2;
1537 char b[BDEVNAME_SIZE];
1540 printk("md: **********************************\n");
1541 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1542 printk("md: **********************************\n");
1543 ITERATE_MDDEV(mddev,tmp) {
1546 bitmap_print_sb(mddev->bitmap);
1548 printk("%s: ", mdname(mddev));
1549 ITERATE_RDEV(mddev,rdev,tmp2)
1550 printk("<%s>", bdevname(rdev->bdev,b));
1553 ITERATE_RDEV(mddev,rdev,tmp2)
1556 printk("md: **********************************\n");
1561 static void sync_sbs(mddev_t * mddev, int nospares)
1563 /* Update each superblock (in-memory image), but
1564 * if we are allowed to, skip spares which already
1565 * have the right event counter, or have one earlier
1566 * (which would mean they aren't being marked as dirty
1567 * with the rest of the array)
1570 struct list_head *tmp;
1572 ITERATE_RDEV(mddev,rdev,tmp) {
1573 if (rdev->sb_events == mddev->events ||
1575 rdev->raid_disk < 0 &&
1576 (rdev->sb_events&1)==0 &&
1577 rdev->sb_events+1 == mddev->events)) {
1578 /* Don't update this superblock */
1579 rdev->sb_loaded = 2;
1581 super_types[mddev->major_version].
1582 sync_super(mddev, rdev);
1583 rdev->sb_loaded = 1;
1588 void md_update_sb(mddev_t * mddev)
1591 struct list_head *tmp;
1597 spin_lock_irq(&mddev->write_lock);
1598 sync_req = mddev->in_sync;
1599 mddev->utime = get_seconds();
1600 if (mddev->sb_dirty == 3)
1601 /* just a clean<-> dirty transition, possibly leave spares alone,
1602 * though if events isn't the right even/odd, we will have to do
1607 /* If this is just a dirty<->clean transition, and the array is clean
1608 * and 'events' is odd, we can roll back to the previous clean state */
1609 if (mddev->sb_dirty == 3
1610 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1611 && (mddev->events & 1))
1614 /* otherwise we have to go forward and ... */
1616 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1617 /* .. if the array isn't clean, insist on an odd 'events' */
1618 if ((mddev->events&1)==0) {
1623 /* otherwise insist on an even 'events' (for clean states) */
1624 if ((mddev->events&1)) {
1631 if (!mddev->events) {
1633 * oops, this 64-bit counter should never wrap.
1634 * Either we are in around ~1 trillion A.C., assuming
1635 * 1 reboot per second, or we have a bug:
1640 mddev->sb_dirty = 2;
1641 sync_sbs(mddev, nospares);
1644 * do not write anything to disk if using
1645 * nonpersistent superblocks
1647 if (!mddev->persistent) {
1648 mddev->sb_dirty = 0;
1649 spin_unlock_irq(&mddev->write_lock);
1650 wake_up(&mddev->sb_wait);
1653 spin_unlock_irq(&mddev->write_lock);
1656 "md: updating %s RAID superblock on device (in sync %d)\n",
1657 mdname(mddev),mddev->in_sync);
1659 err = bitmap_update_sb(mddev->bitmap);
1660 ITERATE_RDEV(mddev,rdev,tmp) {
1661 char b[BDEVNAME_SIZE];
1662 dprintk(KERN_INFO "md: ");
1663 if (rdev->sb_loaded != 1)
1664 continue; /* no noise on spare devices */
1665 if (test_bit(Faulty, &rdev->flags))
1666 dprintk("(skipping faulty ");
1668 dprintk("%s ", bdevname(rdev->bdev,b));
1669 if (!test_bit(Faulty, &rdev->flags)) {
1670 md_super_write(mddev,rdev,
1671 rdev->sb_offset<<1, rdev->sb_size,
1673 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1674 bdevname(rdev->bdev,b),
1675 (unsigned long long)rdev->sb_offset);
1676 rdev->sb_events = mddev->events;
1680 if (mddev->level == LEVEL_MULTIPATH)
1681 /* only need to write one superblock... */
1684 md_super_wait(mddev);
1685 /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1687 spin_lock_irq(&mddev->write_lock);
1688 if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1689 /* have to write it out again */
1690 spin_unlock_irq(&mddev->write_lock);
1693 mddev->sb_dirty = 0;
1694 spin_unlock_irq(&mddev->write_lock);
1695 wake_up(&mddev->sb_wait);
1698 EXPORT_SYMBOL_GPL(md_update_sb);
1700 /* words written to sysfs files may, or my not, be \n terminated.
1701 * We want to accept with case. For this we use cmd_match.
1703 static int cmd_match(const char *cmd, const char *str)
1705 /* See if cmd, written into a sysfs file, matches
1706 * str. They must either be the same, or cmd can
1707 * have a trailing newline
1709 while (*cmd && *str && *cmd == *str) {
1720 struct rdev_sysfs_entry {
1721 struct attribute attr;
1722 ssize_t (*show)(mdk_rdev_t *, char *);
1723 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1727 state_show(mdk_rdev_t *rdev, char *page)
1732 if (test_bit(Faulty, &rdev->flags)) {
1733 len+= sprintf(page+len, "%sfaulty",sep);
1736 if (test_bit(In_sync, &rdev->flags)) {
1737 len += sprintf(page+len, "%sin_sync",sep);
1740 if (!test_bit(Faulty, &rdev->flags) &&
1741 !test_bit(In_sync, &rdev->flags)) {
1742 len += sprintf(page+len, "%sspare", sep);
1745 return len+sprintf(page+len, "\n");
1749 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1752 * faulty - simulates and error
1753 * remove - disconnects the device
1756 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1757 md_error(rdev->mddev, rdev);
1759 } else if (cmd_match(buf, "remove")) {
1760 if (rdev->raid_disk >= 0)
1763 mddev_t *mddev = rdev->mddev;
1764 kick_rdev_from_array(rdev);
1765 md_update_sb(mddev);
1766 md_new_event(mddev);
1770 return err ? err : len;
1772 static struct rdev_sysfs_entry
1773 rdev_state = __ATTR(state, 0644, state_show, state_store);
1776 super_show(mdk_rdev_t *rdev, char *page)
1778 if (rdev->sb_loaded && rdev->sb_size) {
1779 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1780 return rdev->sb_size;
1784 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1787 errors_show(mdk_rdev_t *rdev, char *page)
1789 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1793 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1796 unsigned long n = simple_strtoul(buf, &e, 10);
1797 if (*buf && (*e == 0 || *e == '\n')) {
1798 atomic_set(&rdev->corrected_errors, n);
1803 static struct rdev_sysfs_entry rdev_errors =
1804 __ATTR(errors, 0644, errors_show, errors_store);
1807 slot_show(mdk_rdev_t *rdev, char *page)
1809 if (rdev->raid_disk < 0)
1810 return sprintf(page, "none\n");
1812 return sprintf(page, "%d\n", rdev->raid_disk);
1816 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1819 int slot = simple_strtoul(buf, &e, 10);
1820 if (strncmp(buf, "none", 4)==0)
1822 else if (e==buf || (*e && *e!= '\n'))
1824 if (rdev->mddev->pers)
1825 /* Cannot set slot in active array (yet) */
1827 if (slot >= rdev->mddev->raid_disks)
1829 rdev->raid_disk = slot;
1830 /* assume it is working */
1832 set_bit(In_sync, &rdev->flags);
1837 static struct rdev_sysfs_entry rdev_slot =
1838 __ATTR(slot, 0644, slot_show, slot_store);
1841 offset_show(mdk_rdev_t *rdev, char *page)
1843 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1847 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1850 unsigned long long offset = simple_strtoull(buf, &e, 10);
1851 if (e==buf || (*e && *e != '\n'))
1853 if (rdev->mddev->pers)
1855 rdev->data_offset = offset;
1859 static struct rdev_sysfs_entry rdev_offset =
1860 __ATTR(offset, 0644, offset_show, offset_store);
1863 rdev_size_show(mdk_rdev_t *rdev, char *page)
1865 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1869 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1872 unsigned long long size = simple_strtoull(buf, &e, 10);
1873 if (e==buf || (*e && *e != '\n'))
1875 if (rdev->mddev->pers)
1878 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1879 rdev->mddev->size = size;
1883 static struct rdev_sysfs_entry rdev_size =
1884 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1886 static struct attribute *rdev_default_attrs[] = {
1896 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1898 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1899 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1903 return entry->show(rdev, page);
1907 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1908 const char *page, size_t length)
1910 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1911 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1915 return entry->store(rdev, page, length);
1918 static void rdev_free(struct kobject *ko)
1920 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1923 static struct sysfs_ops rdev_sysfs_ops = {
1924 .show = rdev_attr_show,
1925 .store = rdev_attr_store,
1927 static struct kobj_type rdev_ktype = {
1928 .release = rdev_free,
1929 .sysfs_ops = &rdev_sysfs_ops,
1930 .default_attrs = rdev_default_attrs,
1934 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1936 * mark the device faulty if:
1938 * - the device is nonexistent (zero size)
1939 * - the device has no valid superblock
1941 * a faulty rdev _never_ has rdev->sb set.
1943 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1945 char b[BDEVNAME_SIZE];
1950 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1952 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1953 return ERR_PTR(-ENOMEM);
1956 if ((err = alloc_disk_sb(rdev)))
1959 err = lock_rdev(rdev, newdev);
1963 rdev->kobj.parent = NULL;
1964 rdev->kobj.ktype = &rdev_ktype;
1965 kobject_init(&rdev->kobj);
1969 rdev->data_offset = 0;
1970 rdev->sb_events = 0;
1971 atomic_set(&rdev->nr_pending, 0);
1972 atomic_set(&rdev->read_errors, 0);
1973 atomic_set(&rdev->corrected_errors, 0);
1975 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1978 "md: %s has zero or unknown size, marking faulty!\n",
1979 bdevname(rdev->bdev,b));
1984 if (super_format >= 0) {
1985 err = super_types[super_format].
1986 load_super(rdev, NULL, super_minor);
1987 if (err == -EINVAL) {
1989 "md: %s has invalid sb, not importing!\n",
1990 bdevname(rdev->bdev,b));
1995 "md: could not read %s's sb, not importing!\n",
1996 bdevname(rdev->bdev,b));
2000 INIT_LIST_HEAD(&rdev->same_set);
2005 if (rdev->sb_page) {
2011 return ERR_PTR(err);
2015 * Check a full RAID array for plausibility
2019 static void analyze_sbs(mddev_t * mddev)
2022 struct list_head *tmp;
2023 mdk_rdev_t *rdev, *freshest;
2024 char b[BDEVNAME_SIZE];
2027 ITERATE_RDEV(mddev,rdev,tmp)
2028 switch (super_types[mddev->major_version].
2029 load_super(rdev, freshest, mddev->minor_version)) {
2037 "md: fatal superblock inconsistency in %s"
2038 " -- removing from array\n",
2039 bdevname(rdev->bdev,b));
2040 kick_rdev_from_array(rdev);
2044 super_types[mddev->major_version].
2045 validate_super(mddev, freshest);
2048 ITERATE_RDEV(mddev,rdev,tmp) {
2049 if (rdev != freshest)
2050 if (super_types[mddev->major_version].
2051 validate_super(mddev, rdev)) {
2052 printk(KERN_WARNING "md: kicking non-fresh %s"
2054 bdevname(rdev->bdev,b));
2055 kick_rdev_from_array(rdev);
2058 if (mddev->level == LEVEL_MULTIPATH) {
2059 rdev->desc_nr = i++;
2060 rdev->raid_disk = rdev->desc_nr;
2061 set_bit(In_sync, &rdev->flags);
2067 if (mddev->recovery_cp != MaxSector &&
2069 printk(KERN_ERR "md: %s: raid array is not clean"
2070 " -- starting background reconstruction\n",
2076 safe_delay_show(mddev_t *mddev, char *page)
2078 int msec = (mddev->safemode_delay*1000)/HZ;
2079 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2082 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2090 /* remove a period, and count digits after it */
2091 if (len >= sizeof(buf))
2093 strlcpy(buf, cbuf, len);
2095 for (i=0; i<len; i++) {
2097 if (isdigit(buf[i])) {
2102 } else if (buf[i] == '.') {
2107 msec = simple_strtoul(buf, &e, 10);
2108 if (e == buf || (*e && *e != '\n'))
2110 msec = (msec * 1000) / scale;
2112 mddev->safemode_delay = 0;
2114 mddev->safemode_delay = (msec*HZ)/1000;
2115 if (mddev->safemode_delay == 0)
2116 mddev->safemode_delay = 1;
2120 static struct md_sysfs_entry md_safe_delay =
2121 __ATTR(safe_mode_delay, 0644,safe_delay_show, safe_delay_store);
2124 level_show(mddev_t *mddev, char *page)
2126 struct mdk_personality *p = mddev->pers;
2128 return sprintf(page, "%s\n", p->name);
2129 else if (mddev->clevel[0])
2130 return sprintf(page, "%s\n", mddev->clevel);
2131 else if (mddev->level != LEVEL_NONE)
2132 return sprintf(page, "%d\n", mddev->level);
2138 level_store(mddev_t *mddev, const char *buf, size_t len)
2145 if (len >= sizeof(mddev->clevel))
2147 strncpy(mddev->clevel, buf, len);
2148 if (mddev->clevel[len-1] == '\n')
2150 mddev->clevel[len] = 0;
2151 mddev->level = LEVEL_NONE;
2155 static struct md_sysfs_entry md_level =
2156 __ATTR(level, 0644, level_show, level_store);
2160 layout_show(mddev_t *mddev, char *page)
2162 /* just a number, not meaningful for all levels */
2163 return sprintf(page, "%d\n", mddev->layout);
2167 layout_store(mddev_t *mddev, const char *buf, size_t len)
2170 unsigned long n = simple_strtoul(buf, &e, 10);
2174 if (!*buf || (*e && *e != '\n'))
2180 static struct md_sysfs_entry md_layout =
2181 __ATTR(layout, 0655, layout_show, layout_store);
2185 raid_disks_show(mddev_t *mddev, char *page)
2187 if (mddev->raid_disks == 0)
2189 return sprintf(page, "%d\n", mddev->raid_disks);
2192 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2195 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2197 /* can only set raid_disks if array is not yet active */
2200 unsigned long n = simple_strtoul(buf, &e, 10);
2202 if (!*buf || (*e && *e != '\n'))
2206 rv = update_raid_disks(mddev, n);
2208 mddev->raid_disks = n;
2209 return rv ? rv : len;
2211 static struct md_sysfs_entry md_raid_disks =
2212 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
2215 chunk_size_show(mddev_t *mddev, char *page)
2217 return sprintf(page, "%d\n", mddev->chunk_size);
2221 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2223 /* can only set chunk_size if array is not yet active */
2225 unsigned long n = simple_strtoul(buf, &e, 10);
2229 if (!*buf || (*e && *e != '\n'))
2232 mddev->chunk_size = n;
2235 static struct md_sysfs_entry md_chunk_size =
2236 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2239 resync_start_show(mddev_t *mddev, char *page)
2241 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2245 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2247 /* can only set chunk_size if array is not yet active */
2249 unsigned long long n = simple_strtoull(buf, &e, 10);
2253 if (!*buf || (*e && *e != '\n'))
2256 mddev->recovery_cp = n;
2259 static struct md_sysfs_entry md_resync_start =
2260 __ATTR(resync_start, 0644, resync_start_show, resync_start_store);
2263 * The array state can be:
2266 * No devices, no size, no level
2267 * Equivalent to STOP_ARRAY ioctl
2269 * May have some settings, but array is not active
2270 * all IO results in error
2271 * When written, doesn't tear down array, but just stops it
2272 * suspended (not supported yet)
2273 * All IO requests will block. The array can be reconfigured.
2274 * Writing this, if accepted, will block until array is quiessent
2276 * no resync can happen. no superblocks get written.
2277 * write requests fail
2279 * like readonly, but behaves like 'clean' on a write request.
2281 * clean - no pending writes, but otherwise active.
2282 * When written to inactive array, starts without resync
2283 * If a write request arrives then
2284 * if metadata is known, mark 'dirty' and switch to 'active'.
2285 * if not known, block and switch to write-pending
2286 * If written to an active array that has pending writes, then fails.
2288 * fully active: IO and resync can be happening.
2289 * When written to inactive array, starts with resync
2292 * clean, but writes are blocked waiting for 'active' to be written.
2295 * like active, but no writes have been seen for a while (100msec).
2298 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2299 write_pending, active_idle, bad_word};
2300 char *array_states[] = {
2301 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2302 "write-pending", "active-idle", NULL };
2304 static int match_word(const char *word, char **list)
2307 for (n=0; list[n]; n++)
2308 if (cmd_match(word, list[n]))
2314 array_state_show(mddev_t *mddev, char *page)
2316 enum array_state st = inactive;
2329 else if (mddev->safemode)
2335 if (list_empty(&mddev->disks) &&
2336 mddev->raid_disks == 0 &&
2342 return sprintf(page, "%s\n", array_states[st]);
2345 static int do_md_stop(mddev_t * mddev, int ro);
2346 static int do_md_run(mddev_t * mddev);
2347 static int restart_array(mddev_t *mddev);
2350 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2353 enum array_state st = match_word(buf, array_states);
2358 /* stopping an active array */
2360 if (atomic_read(&mddev->active) > 1)
2362 err = do_md_stop(mddev, 0);
2366 /* stopping an active array */
2368 if (atomic_read(&mddev->active) > 1)
2370 err = do_md_stop(mddev, 2);
2374 break; /* not supported yet */
2377 err = do_md_stop(mddev, 1);
2380 err = do_md_run(mddev);
2384 /* stopping an active array */
2386 err = do_md_stop(mddev, 1);
2388 mddev->ro = 2; /* FIXME mark devices writable */
2391 err = do_md_run(mddev);
2396 restart_array(mddev);
2397 spin_lock_irq(&mddev->write_lock);
2398 if (atomic_read(&mddev->writes_pending) == 0) {
2400 mddev->sb_dirty = 1;
2402 spin_unlock_irq(&mddev->write_lock);
2405 mddev->recovery_cp = MaxSector;
2406 err = do_md_run(mddev);
2411 restart_array(mddev);
2412 mddev->sb_dirty = 0;
2413 wake_up(&mddev->sb_wait);
2417 err = do_md_run(mddev);
2422 /* these cannot be set */
2430 static struct md_sysfs_entry md_array_state = __ATTR(array_state, 0644, array_state_show, array_state_store);
2433 null_show(mddev_t *mddev, char *page)
2439 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2441 /* buf must be %d:%d\n? giving major and minor numbers */
2442 /* The new device is added to the array.
2443 * If the array has a persistent superblock, we read the
2444 * superblock to initialise info and check validity.
2445 * Otherwise, only checking done is that in bind_rdev_to_array,
2446 * which mainly checks size.
2449 int major = simple_strtoul(buf, &e, 10);
2455 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2457 minor = simple_strtoul(e+1, &e, 10);
2458 if (*e && *e != '\n')
2460 dev = MKDEV(major, minor);
2461 if (major != MAJOR(dev) ||
2462 minor != MINOR(dev))
2466 if (mddev->persistent) {
2467 rdev = md_import_device(dev, mddev->major_version,
2468 mddev->minor_version);
2469 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2470 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2471 mdk_rdev_t, same_set);
2472 err = super_types[mddev->major_version]
2473 .load_super(rdev, rdev0, mddev->minor_version);
2478 rdev = md_import_device(dev, -1, -1);
2481 return PTR_ERR(rdev);
2482 err = bind_rdev_to_array(rdev, mddev);
2486 return err ? err : len;
2489 static struct md_sysfs_entry md_new_device =
2490 __ATTR(new_dev, 0200, null_show, new_dev_store);
2493 size_show(mddev_t *mddev, char *page)
2495 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2498 static int update_size(mddev_t *mddev, unsigned long size);
2501 size_store(mddev_t *mddev, const char *buf, size_t len)
2503 /* If array is inactive, we can reduce the component size, but
2504 * not increase it (except from 0).
2505 * If array is active, we can try an on-line resize
2509 unsigned long long size = simple_strtoull(buf, &e, 10);
2510 if (!*buf || *buf == '\n' ||
2515 err = update_size(mddev, size);
2516 md_update_sb(mddev);
2518 if (mddev->size == 0 ||
2524 return err ? err : len;
2527 static struct md_sysfs_entry md_size =
2528 __ATTR(component_size, 0644, size_show, size_store);
2532 * This is either 'none' for arrays with externally managed metadata,
2533 * or N.M for internally known formats
2536 metadata_show(mddev_t *mddev, char *page)
2538 if (mddev->persistent)
2539 return sprintf(page, "%d.%d\n",
2540 mddev->major_version, mddev->minor_version);
2542 return sprintf(page, "none\n");
2546 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2550 if (!list_empty(&mddev->disks))
2553 if (cmd_match(buf, "none")) {
2554 mddev->persistent = 0;
2555 mddev->major_version = 0;
2556 mddev->minor_version = 90;
2559 major = simple_strtoul(buf, &e, 10);
2560 if (e==buf || *e != '.')
2563 minor = simple_strtoul(buf, &e, 10);
2564 if (e==buf || *e != '\n')
2566 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2567 super_types[major].name == NULL)
2569 mddev->major_version = major;
2570 mddev->minor_version = minor;
2571 mddev->persistent = 1;
2575 static struct md_sysfs_entry md_metadata =
2576 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2579 action_show(mddev_t *mddev, char *page)
2581 char *type = "idle";
2582 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2583 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2584 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2586 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2587 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2589 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2596 return sprintf(page, "%s\n", type);
2600 action_store(mddev_t *mddev, const char *page, size_t len)
2602 if (!mddev->pers || !mddev->pers->sync_request)
2605 if (cmd_match(page, "idle")) {
2606 if (mddev->sync_thread) {
2607 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2608 md_unregister_thread(mddev->sync_thread);
2609 mddev->sync_thread = NULL;
2610 mddev->recovery = 0;
2612 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2613 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2615 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2616 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2617 else if (cmd_match(page, "reshape")) {
2619 if (mddev->pers->start_reshape == NULL)
2621 err = mddev->pers->start_reshape(mddev);
2625 if (cmd_match(page, "check"))
2626 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2627 else if (!cmd_match(page, "repair"))
2629 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2630 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2632 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2633 md_wakeup_thread(mddev->thread);
2638 mismatch_cnt_show(mddev_t *mddev, char *page)
2640 return sprintf(page, "%llu\n",
2641 (unsigned long long) mddev->resync_mismatches);
2644 static struct md_sysfs_entry
2645 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2648 static struct md_sysfs_entry
2649 md_mismatches = __ATTR_RO(mismatch_cnt);
2652 sync_min_show(mddev_t *mddev, char *page)
2654 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2655 mddev->sync_speed_min ? "local": "system");
2659 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2663 if (strncmp(buf, "system", 6)==0) {
2664 mddev->sync_speed_min = 0;
2667 min = simple_strtoul(buf, &e, 10);
2668 if (buf == e || (*e && *e != '\n') || min <= 0)
2670 mddev->sync_speed_min = min;
2674 static struct md_sysfs_entry md_sync_min =
2675 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2678 sync_max_show(mddev_t *mddev, char *page)
2680 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2681 mddev->sync_speed_max ? "local": "system");
2685 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2689 if (strncmp(buf, "system", 6)==0) {
2690 mddev->sync_speed_max = 0;
2693 max = simple_strtoul(buf, &e, 10);
2694 if (buf == e || (*e && *e != '\n') || max <= 0)
2696 mddev->sync_speed_max = max;
2700 static struct md_sysfs_entry md_sync_max =
2701 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2705 sync_speed_show(mddev_t *mddev, char *page)
2707 unsigned long resync, dt, db;
2708 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2709 dt = ((jiffies - mddev->resync_mark) / HZ);
2711 db = resync - (mddev->resync_mark_cnt);
2712 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2715 static struct md_sysfs_entry
2716 md_sync_speed = __ATTR_RO(sync_speed);
2719 sync_completed_show(mddev_t *mddev, char *page)
2721 unsigned long max_blocks, resync;
2723 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2724 max_blocks = mddev->resync_max_sectors;
2726 max_blocks = mddev->size << 1;
2728 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2729 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2732 static struct md_sysfs_entry
2733 md_sync_completed = __ATTR_RO(sync_completed);
2736 suspend_lo_show(mddev_t *mddev, char *page)
2738 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2742 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2745 unsigned long long new = simple_strtoull(buf, &e, 10);
2747 if (mddev->pers->quiesce == NULL)
2749 if (buf == e || (*e && *e != '\n'))
2751 if (new >= mddev->suspend_hi ||
2752 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2753 mddev->suspend_lo = new;
2754 mddev->pers->quiesce(mddev, 2);
2759 static struct md_sysfs_entry md_suspend_lo =
2760 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2764 suspend_hi_show(mddev_t *mddev, char *page)
2766 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2770 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2773 unsigned long long new = simple_strtoull(buf, &e, 10);
2775 if (mddev->pers->quiesce == NULL)
2777 if (buf == e || (*e && *e != '\n'))
2779 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2780 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2781 mddev->suspend_hi = new;
2782 mddev->pers->quiesce(mddev, 1);
2783 mddev->pers->quiesce(mddev, 0);
2788 static struct md_sysfs_entry md_suspend_hi =
2789 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2792 static struct attribute *md_default_attrs[] = {
2795 &md_raid_disks.attr,
2796 &md_chunk_size.attr,
2798 &md_resync_start.attr,
2800 &md_new_device.attr,
2801 &md_safe_delay.attr,
2802 &md_array_state.attr,
2806 static struct attribute *md_redundancy_attrs[] = {
2808 &md_mismatches.attr,
2811 &md_sync_speed.attr,
2812 &md_sync_completed.attr,
2813 &md_suspend_lo.attr,
2814 &md_suspend_hi.attr,
2817 static struct attribute_group md_redundancy_group = {
2819 .attrs = md_redundancy_attrs,
2824 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2826 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2827 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2832 rv = mddev_lock(mddev);
2834 rv = entry->show(mddev, page);
2835 mddev_unlock(mddev);
2841 md_attr_store(struct kobject *kobj, struct attribute *attr,
2842 const char *page, size_t length)
2844 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2845 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2850 rv = mddev_lock(mddev);
2852 rv = entry->store(mddev, page, length);
2853 mddev_unlock(mddev);
2858 static void md_free(struct kobject *ko)
2860 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2864 static struct sysfs_ops md_sysfs_ops = {
2865 .show = md_attr_show,
2866 .store = md_attr_store,
2868 static struct kobj_type md_ktype = {
2870 .sysfs_ops = &md_sysfs_ops,
2871 .default_attrs = md_default_attrs,
2876 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2878 static DEFINE_MUTEX(disks_mutex);
2879 mddev_t *mddev = mddev_find(dev);
2880 struct gendisk *disk;
2881 int partitioned = (MAJOR(dev) != MD_MAJOR);
2882 int shift = partitioned ? MdpMinorShift : 0;
2883 int unit = MINOR(dev) >> shift;
2888 mutex_lock(&disks_mutex);
2889 if (mddev->gendisk) {
2890 mutex_unlock(&disks_mutex);
2894 disk = alloc_disk(1 << shift);
2896 mutex_unlock(&disks_mutex);
2900 disk->major = MAJOR(dev);
2901 disk->first_minor = unit << shift;
2903 sprintf(disk->disk_name, "md_d%d", unit);
2904 sprintf(disk->devfs_name, "md/d%d", unit);
2906 sprintf(disk->disk_name, "md%d", unit);
2907 sprintf(disk->devfs_name, "md/%d", unit);
2909 disk->fops = &md_fops;
2910 disk->private_data = mddev;
2911 disk->queue = mddev->queue;
2913 mddev->gendisk = disk;
2914 mutex_unlock(&disks_mutex);
2915 mddev->kobj.parent = &disk->kobj;
2916 mddev->kobj.k_name = NULL;
2917 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2918 mddev->kobj.ktype = &md_ktype;
2919 kobject_register(&mddev->kobj);
2923 static void md_safemode_timeout(unsigned long data)
2925 mddev_t *mddev = (mddev_t *) data;
2927 mddev->safemode = 1;
2928 md_wakeup_thread(mddev->thread);
2931 static int start_dirty_degraded;
2933 static int do_md_run(mddev_t * mddev)
2937 struct list_head *tmp;
2939 struct gendisk *disk;
2940 struct mdk_personality *pers;
2941 char b[BDEVNAME_SIZE];
2943 if (list_empty(&mddev->disks))
2944 /* cannot run an array with no devices.. */
2951 * Analyze all RAID superblock(s)
2953 if (!mddev->raid_disks)
2956 chunk_size = mddev->chunk_size;
2959 if (chunk_size > MAX_CHUNK_SIZE) {
2960 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2961 chunk_size, MAX_CHUNK_SIZE);
2965 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2967 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2968 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2971 if (chunk_size < PAGE_SIZE) {
2972 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2973 chunk_size, PAGE_SIZE);
2977 /* devices must have minimum size of one chunk */
2978 ITERATE_RDEV(mddev,rdev,tmp) {
2979 if (test_bit(Faulty, &rdev->flags))
2981 if (rdev->size < chunk_size / 1024) {
2983 "md: Dev %s smaller than chunk_size:"
2985 bdevname(rdev->bdev,b),
2986 (unsigned long long)rdev->size,
2994 if (mddev->level != LEVEL_NONE)
2995 request_module("md-level-%d", mddev->level);
2996 else if (mddev->clevel[0])
2997 request_module("md-%s", mddev->clevel);
3001 * Drop all container device buffers, from now on
3002 * the only valid external interface is through the md
3004 * Also find largest hardsector size
3006 ITERATE_RDEV(mddev,rdev,tmp) {
3007 if (test_bit(Faulty, &rdev->flags))
3009 sync_blockdev(rdev->bdev);
3010 invalidate_bdev(rdev->bdev, 0);
3013 md_probe(mddev->unit, NULL, NULL);
3014 disk = mddev->gendisk;
3018 spin_lock(&pers_lock);
3019 pers = find_pers(mddev->level, mddev->clevel);
3020 if (!pers || !try_module_get(pers->owner)) {
3021 spin_unlock(&pers_lock);
3022 if (mddev->level != LEVEL_NONE)
3023 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3026 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3031 spin_unlock(&pers_lock);
3032 mddev->level = pers->level;
3033 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3035 if (mddev->reshape_position != MaxSector &&
3036 pers->start_reshape == NULL) {
3037 /* This personality cannot handle reshaping... */
3039 module_put(pers->owner);
3043 mddev->recovery = 0;
3044 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3045 mddev->barriers_work = 1;
3046 mddev->ok_start_degraded = start_dirty_degraded;
3049 mddev->ro = 2; /* read-only, but switch on first write */
3051 err = mddev->pers->run(mddev);
3052 if (!err && mddev->pers->sync_request) {
3053 err = bitmap_create(mddev);
3055 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3056 mdname(mddev), err);
3057 mddev->pers->stop(mddev);
3061 printk(KERN_ERR "md: pers->run() failed ...\n");
3062 module_put(mddev->pers->owner);
3064 bitmap_destroy(mddev);
3067 if (mddev->pers->sync_request)
3068 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
3069 else if (mddev->ro == 2) /* auto-readonly not meaningful */
3072 atomic_set(&mddev->writes_pending,0);
3073 mddev->safemode = 0;
3074 mddev->safemode_timer.function = md_safemode_timeout;
3075 mddev->safemode_timer.data = (unsigned long) mddev;
3076 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3079 ITERATE_RDEV(mddev,rdev,tmp)
3080 if (rdev->raid_disk >= 0) {
3082 sprintf(nm, "rd%d", rdev->raid_disk);
3083 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3086 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3087 md_wakeup_thread(mddev->thread);
3089 if (mddev->sb_dirty)
3090 md_update_sb(mddev);
3092 set_capacity(disk, mddev->array_size<<1);
3094 /* If we call blk_queue_make_request here, it will
3095 * re-initialise max_sectors etc which may have been
3096 * refined inside -> run. So just set the bits we need to set.
3097 * Most initialisation happended when we called
3098 * blk_queue_make_request(..., md_fail_request)
3101 mddev->queue->queuedata = mddev;
3102 mddev->queue->make_request_fn = mddev->pers->make_request;
3104 /* If there is a partially-recovered drive we need to
3105 * start recovery here. If we leave it to md_check_recovery,
3106 * it will remove the drives and not do the right thing
3108 if (mddev->degraded) {
3109 struct list_head *rtmp;
3111 ITERATE_RDEV(mddev,rdev,rtmp)
3112 if (rdev->raid_disk >= 0 &&
3113 !test_bit(In_sync, &rdev->flags) &&
3114 !test_bit(Faulty, &rdev->flags))
3115 /* complete an interrupted recovery */
3117 if (spares && mddev->pers->sync_request) {
3118 mddev->recovery = 0;
3119 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3120 mddev->sync_thread = md_register_thread(md_do_sync,
3123 if (!mddev->sync_thread) {
3124 printk(KERN_ERR "%s: could not start resync"
3127 /* leave the spares where they are, it shouldn't hurt */
3128 mddev->recovery = 0;
3130 md_wakeup_thread(mddev->sync_thread);
3135 md_new_event(mddev);
3139 static int restart_array(mddev_t *mddev)
3141 struct gendisk *disk = mddev->gendisk;
3145 * Complain if it has no devices
3148 if (list_empty(&mddev->disks))
3156 mddev->safemode = 0;
3158 set_disk_ro(disk, 0);
3160 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3163 * Kick recovery or resync if necessary
3165 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3166 md_wakeup_thread(mddev->thread);
3167 md_wakeup_thread(mddev->sync_thread);
3176 /* similar to deny_write_access, but accounts for our holding a reference
3177 * to the file ourselves */
3178 static int deny_bitmap_write_access(struct file * file)
3180 struct inode *inode = file->f_mapping->host;
3182 spin_lock(&inode->i_lock);
3183 if (atomic_read(&inode->i_writecount) > 1) {
3184 spin_unlock(&inode->i_lock);
3187 atomic_set(&inode->i_writecount, -1);
3188 spin_unlock(&inode->i_lock);
3193 static void restore_bitmap_write_access(struct file *file)
3195 struct inode *inode = file->f_mapping->host;
3197 spin_lock(&inode->i_lock);
3198 atomic_set(&inode->i_writecount, 1);
3199 spin_unlock(&inode->i_lock);
3203 * 0 - completely stop and dis-assemble array
3204 * 1 - switch to readonly
3205 * 2 - stop but do not disassemble array
3207 static int do_md_stop(mddev_t * mddev, int mode)
3210 struct gendisk *disk = mddev->gendisk;
3213 if (atomic_read(&mddev->active)>2) {
3214 printk("md: %s still in use.\n",mdname(mddev));
3218 if (mddev->sync_thread) {
3219 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3220 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3221 md_unregister_thread(mddev->sync_thread);
3222 mddev->sync_thread = NULL;
3225 del_timer_sync(&mddev->safemode_timer);
3227 invalidate_partition(disk, 0);
3230 case 1: /* readonly */
3236 case 0: /* disassemble */
3238 bitmap_flush(mddev);
3239 md_super_wait(mddev);
3241 set_disk_ro(disk, 0);
3242 blk_queue_make_request(mddev->queue, md_fail_request);
3243 mddev->pers->stop(mddev);
3244 if (mddev->pers->sync_request)
3245 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3247 module_put(mddev->pers->owner);
3252 if (!mddev->in_sync || mddev->sb_dirty) {
3253 /* mark array as shutdown cleanly */
3255 md_update_sb(mddev);
3258 set_disk_ro(disk, 1);
3259 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3263 * Free resources if final stop
3267 struct list_head *tmp;
3268 struct gendisk *disk;
3269 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3271 bitmap_destroy(mddev);
3272 if (mddev->bitmap_file) {
3273 restore_bitmap_write_access(mddev->bitmap_file);
3274 fput(mddev->bitmap_file);
3275 mddev->bitmap_file = NULL;
3277 mddev->bitmap_offset = 0;
3279 ITERATE_RDEV(mddev,rdev,tmp)
3280 if (rdev->raid_disk >= 0) {
3282 sprintf(nm, "rd%d", rdev->raid_disk);
3283 sysfs_remove_link(&mddev->kobj, nm);
3286 export_array(mddev);
3288 mddev->array_size = 0;
3290 mddev->raid_disks = 0;
3291 mddev->recovery_cp = 0;
3293 disk = mddev->gendisk;
3295 set_capacity(disk, 0);
3297 } else if (mddev->pers)
3298 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3301 md_new_event(mddev);
3306 static void autorun_array(mddev_t *mddev)
3309 struct list_head *tmp;
3312 if (list_empty(&mddev->disks))
3315 printk(KERN_INFO "md: running: ");
3317 ITERATE_RDEV(mddev,rdev,tmp) {
3318 char b[BDEVNAME_SIZE];
3319 printk("<%s>", bdevname(rdev->bdev,b));
3323 err = do_md_run (mddev);
3325 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3326 do_md_stop (mddev, 0);
3331 * lets try to run arrays based on all disks that have arrived
3332 * until now. (those are in pending_raid_disks)
3334 * the method: pick the first pending disk, collect all disks with
3335 * the same UUID, remove all from the pending list and put them into
3336 * the 'same_array' list. Then order this list based on superblock
3337 * update time (freshest comes first), kick out 'old' disks and
3338 * compare superblocks. If everything's fine then run it.
3340 * If "unit" is allocated, then bump its reference count
3342 static void autorun_devices(int part)
3344 struct list_head *tmp;
3345 mdk_rdev_t *rdev0, *rdev;
3347 char b[BDEVNAME_SIZE];
3349 printk(KERN_INFO "md: autorun ...\n");
3350 while (!list_empty(&pending_raid_disks)) {
3352 LIST_HEAD(candidates);
3353 rdev0 = list_entry(pending_raid_disks.next,
3354 mdk_rdev_t, same_set);
3356 printk(KERN_INFO "md: considering %s ...\n",
3357 bdevname(rdev0->bdev,b));
3358 INIT_LIST_HEAD(&candidates);
3359 ITERATE_RDEV_PENDING(rdev,tmp)
3360 if (super_90_load(rdev, rdev0, 0) >= 0) {
3361 printk(KERN_INFO "md: adding %s ...\n",
3362 bdevname(rdev->bdev,b));
3363 list_move(&rdev->same_set, &candidates);
3366 * now we have a set of devices, with all of them having
3367 * mostly sane superblocks. It's time to allocate the
3370 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
3371 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3372 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3376 dev = MKDEV(mdp_major,
3377 rdev0->preferred_minor << MdpMinorShift);
3379 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3381 md_probe(dev, NULL, NULL);
3382 mddev = mddev_find(dev);
3385 "md: cannot allocate memory for md drive.\n");
3388 if (mddev_lock(mddev))
3389 printk(KERN_WARNING "md: %s locked, cannot run\n",
3391 else if (mddev->raid_disks || mddev->major_version
3392 || !list_empty(&mddev->disks)) {
3394 "md: %s already running, cannot run %s\n",
3395 mdname(mddev), bdevname(rdev0->bdev,b));
3396 mddev_unlock(mddev);
3398 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3399 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3400 list_del_init(&rdev->same_set);
3401 if (bind_rdev_to_array(rdev, mddev))
3404 autorun_array(mddev);
3405 mddev_unlock(mddev);
3407 /* on success, candidates will be empty, on error
3410 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3414 printk(KERN_INFO "md: ... autorun DONE.\n");
3418 * import RAID devices based on one partition
3419 * if possible, the array gets run as well.
3422 static int autostart_array(dev_t startdev)
3424 char b[BDEVNAME_SIZE];
3425 int err = -EINVAL, i;
3426 mdp_super_t *sb = NULL;
3427 mdk_rdev_t *start_rdev = NULL, *rdev;
3429 start_rdev = md_import_device(startdev, 0, 0);
3430 if (IS_ERR(start_rdev))
3434 /* NOTE: this can only work for 0.90.0 superblocks */
3435 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
3436 if (sb->major_version != 0 ||
3437 sb->minor_version != 90 ) {
3438 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
3439 export_rdev(start_rdev);
3443 if (test_bit(Faulty, &start_rdev->flags)) {
3445 "md: can not autostart based on faulty %s!\n",
3446 bdevname(start_rdev->bdev,b));
3447 export_rdev(start_rdev);
3450 list_add(&start_rdev->same_set, &pending_raid_disks);
3452 for (i = 0; i < MD_SB_DISKS; i++) {
3453 mdp_disk_t *desc = sb->disks + i;
3454 dev_t dev = MKDEV(desc->major, desc->minor);
3458 if (dev == startdev)
3460 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3462 rdev = md_import_device(dev, 0, 0);
3466 list_add(&rdev->same_set, &pending_raid_disks);
3470 * possibly return codes
3478 static int get_version(void __user * arg)
3482 ver.major = MD_MAJOR_VERSION;
3483 ver.minor = MD_MINOR_VERSION;
3484 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3486 if (copy_to_user(arg, &ver, sizeof(ver)))
3492 static int get_array_info(mddev_t * mddev, void __user * arg)
3494 mdu_array_info_t info;
3495 int nr,working,active,failed,spare;
3497 struct list_head *tmp;
3499 nr=working=active=failed=spare=0;
3500 ITERATE_RDEV(mddev,rdev,tmp) {
3502 if (test_bit(Faulty, &rdev->flags))
3506 if (test_bit(In_sync, &rdev->flags))
3513 info.major_version = mddev->major_version;
3514 info.minor_version = mddev->minor_version;
3515 info.patch_version = MD_PATCHLEVEL_VERSION;
3516 info.ctime = mddev->ctime;
3517 info.level = mddev->level;
3518 info.size = mddev->size;
3519 if (info.size != mddev->size) /* overflow */
3522 info.raid_disks = mddev->raid_disks;
3523 info.md_minor = mddev->md_minor;
3524 info.not_persistent= !mddev->persistent;
3526 info.utime = mddev->utime;
3529 info.state = (1<<MD_SB_CLEAN);
3530 if (mddev->bitmap && mddev->bitmap_offset)
3531 info.state = (1<<MD_SB_BITMAP_PRESENT);
3532 info.active_disks = active;
3533 info.working_disks = working;
3534 info.failed_disks = failed;
3535 info.spare_disks = spare;
3537 info.layout = mddev->layout;
3538 info.chunk_size = mddev->chunk_size;
3540 if (copy_to_user(arg, &info, sizeof(info)))
3546 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3548 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3549 char *ptr, *buf = NULL;
3552 file = kmalloc(sizeof(*file), GFP_KERNEL);
3556 /* bitmap disabled, zero the first byte and copy out */
3557 if (!mddev->bitmap || !mddev->bitmap->file) {
3558 file->pathname[0] = '\0';
3562 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3566 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3570 strcpy(file->pathname, ptr);
3574 if (copy_to_user(arg, file, sizeof(*file)))
3582 static int get_disk_info(mddev_t * mddev, void __user * arg)
3584 mdu_disk_info_t info;
3588 if (copy_from_user(&info, arg, sizeof(info)))
3593 rdev = find_rdev_nr(mddev, nr);
3595 info.major = MAJOR(rdev->bdev->bd_dev);
3596 info.minor = MINOR(rdev->bdev->bd_dev);
3597 info.raid_disk = rdev->raid_disk;
3599 if (test_bit(Faulty, &rdev->flags))
3600 info.state |= (1<<MD_DISK_FAULTY);
3601 else if (test_bit(In_sync, &rdev->flags)) {
3602 info.state |= (1<<MD_DISK_ACTIVE);
3603 info.state |= (1<<MD_DISK_SYNC);
3605 if (test_bit(WriteMostly, &rdev->flags))
3606 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3608 info.major = info.minor = 0;
3609 info.raid_disk = -1;
3610 info.state = (1<<MD_DISK_REMOVED);
3613 if (copy_to_user(arg, &info, sizeof(info)))
3619 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3621 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3623 dev_t dev = MKDEV(info->major,info->minor);
3625 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3628 if (!mddev->raid_disks) {
3630 /* expecting a device which has a superblock */
3631 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3634 "md: md_import_device returned %ld\n",
3636 return PTR_ERR(rdev);
3638 if (!list_empty(&mddev->disks)) {
3639 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3640 mdk_rdev_t, same_set);
3641 int err = super_types[mddev->major_version]
3642 .load_super(rdev, rdev0, mddev->minor_version);
3645 "md: %s has different UUID to %s\n",
3646 bdevname(rdev->bdev,b),
3647 bdevname(rdev0->bdev,b2));
3652 err = bind_rdev_to_array(rdev, mddev);
3659 * add_new_disk can be used once the array is assembled
3660 * to add "hot spares". They must already have a superblock
3665 if (!mddev->pers->hot_add_disk) {
3667 "%s: personality does not support diskops!\n",
3671 if (mddev->persistent)
3672 rdev = md_import_device(dev, mddev->major_version,
3673 mddev->minor_version);
3675 rdev = md_import_device(dev, -1, -1);
3678 "md: md_import_device returned %ld\n",
3680 return PTR_ERR(rdev);
3682 /* set save_raid_disk if appropriate */
3683 if (!mddev->persistent) {
3684 if (info->state & (1<<MD_DISK_SYNC) &&
3685 info->raid_disk < mddev->raid_disks)
3686 rdev->raid_disk = info->raid_disk;
3688 rdev->raid_disk = -1;
3690 super_types[mddev->major_version].
3691 validate_super(mddev, rdev);
3692 rdev->saved_raid_disk = rdev->raid_disk;
3694 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3695 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3696 set_bit(WriteMostly, &rdev->flags);
3698 rdev->raid_disk = -1;
3699 err = bind_rdev_to_array(rdev, mddev);
3700 if (!err && !mddev->pers->hot_remove_disk) {
3701 /* If there is hot_add_disk but no hot_remove_disk
3702 * then added disks for geometry changes,
3703 * and should be added immediately.
3705 super_types[mddev->major_version].
3706 validate_super(mddev, rdev);
3707 err = mddev->pers->hot_add_disk(mddev, rdev);
3709 unbind_rdev_from_array(rdev);
3714 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3715 md_wakeup_thread(mddev->thread);
3719 /* otherwise, add_new_disk is only allowed
3720 * for major_version==0 superblocks
3722 if (mddev->major_version != 0) {
3723 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3728 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3730 rdev = md_import_device (dev, -1, 0);
3733 "md: error, md_import_device() returned %ld\n",
3735 return PTR_ERR(rdev);
3737 rdev->desc_nr = info->number;
3738 if (info->raid_disk < mddev->raid_disks)
3739 rdev->raid_disk = info->raid_disk;
3741 rdev->raid_disk = -1;
3745 if (rdev->raid_disk < mddev->raid_disks)
3746 if (info->state & (1<<MD_DISK_SYNC))
3747 set_bit(In_sync, &rdev->flags);
3749 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3750 set_bit(WriteMostly, &rdev->flags);
3752 if (!mddev->persistent) {
3753 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3754 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3756 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3757 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3759 err = bind_rdev_to_array(rdev, mddev);
3769 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3771 char b[BDEVNAME_SIZE];
3777 rdev = find_rdev(mddev, dev);
3781 if (rdev->raid_disk >= 0)
3784 kick_rdev_from_array(rdev);
3785 md_update_sb(mddev);
3786 md_new_event(mddev);
3790 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3791 bdevname(rdev->bdev,b), mdname(mddev));
3795 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3797 char b[BDEVNAME_SIZE];
3805 if (mddev->major_version != 0) {
3806 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3807 " version-0 superblocks.\n",
3811 if (!mddev->pers->hot_add_disk) {
3813 "%s: personality does not support diskops!\n",
3818 rdev = md_import_device (dev, -1, 0);
3821 "md: error, md_import_device() returned %ld\n",
3826 if (mddev->persistent)
3827 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3830 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3832 size = calc_dev_size(rdev, mddev->chunk_size);
3835 if (test_bit(Faulty, &rdev->flags)) {
3837 "md: can not hot-add faulty %s disk to %s!\n",
3838 bdevname(rdev->bdev,b), mdname(mddev));
3842 clear_bit(In_sync, &rdev->flags);
3844 err = bind_rdev_to_array(rdev, mddev);
3849 * The rest should better be atomic, we can have disk failures
3850 * noticed in interrupt contexts ...
3853 if (rdev->desc_nr == mddev->max_disks) {
3854 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3857 goto abort_unbind_export;
3860 rdev->raid_disk = -1;
3862 md_update_sb(mddev);
3865 * Kick recovery, maybe this spare has to be added to the
3866 * array immediately.
3868 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3869 md_wakeup_thread(mddev->thread);
3870 md_new_event(mddev);
3873 abort_unbind_export:
3874 unbind_rdev_from_array(rdev);
3881 static int set_bitmap_file(mddev_t *mddev, int fd)
3886 if (!mddev->pers->quiesce)
3888 if (mddev->recovery || mddev->sync_thread)
3890 /* we should be able to change the bitmap.. */
3896 return -EEXIST; /* cannot add when bitmap is present */
3897 mddev->bitmap_file = fget(fd);
3899 if (mddev->bitmap_file == NULL) {
3900 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3905 err = deny_bitmap_write_access(mddev->bitmap_file);
3907 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3909 fput(mddev->bitmap_file);
3910 mddev->bitmap_file = NULL;
3913 mddev->bitmap_offset = 0; /* file overrides offset */
3914 } else if (mddev->bitmap == NULL)
3915 return -ENOENT; /* cannot remove what isn't there */
3918 mddev->pers->quiesce(mddev, 1);
3920 err = bitmap_create(mddev);
3921 if (fd < 0 || err) {
3922 bitmap_destroy(mddev);
3923 fd = -1; /* make sure to put the file */
3925 mddev->pers->quiesce(mddev, 0);
3928 if (mddev->bitmap_file) {
3929 restore_bitmap_write_access(mddev->bitmap_file);
3930 fput(mddev->bitmap_file);
3932 mddev->bitmap_file = NULL;
3939 * set_array_info is used two different ways
3940 * The original usage is when creating a new array.
3941 * In this usage, raid_disks is > 0 and it together with
3942 * level, size, not_persistent,layout,chunksize determine the
3943 * shape of the array.
3944 * This will always create an array with a type-0.90.0 superblock.
3945 * The newer usage is when assembling an array.
3946 * In this case raid_disks will be 0, and the major_version field is
3947 * use to determine which style super-blocks are to be found on the devices.
3948 * The minor and patch _version numbers are also kept incase the
3949 * super_block handler wishes to interpret them.
3951 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3954 if (info->raid_disks == 0) {
3955 /* just setting version number for superblock loading */
3956 if (info->major_version < 0 ||
3957 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3958 super_types[info->major_version].name == NULL) {
3959 /* maybe try to auto-load a module? */
3961 "md: superblock version %d not known\n",
3962 info->major_version);
3965 mddev->major_version = info->major_version;
3966 mddev->minor_version = info->minor_version;
3967 mddev->patch_version = info->patch_version;
3970 mddev->major_version = MD_MAJOR_VERSION;
3971 mddev->minor_version = MD_MINOR_VERSION;
3972 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3973 mddev->ctime = get_seconds();
3975 mddev->level = info->level;
3976 mddev->clevel[0] = 0;
3977 mddev->size = info->size;
3978 mddev->raid_disks = info->raid_disks;
3979 /* don't set md_minor, it is determined by which /dev/md* was
3982 if (info->state & (1<<MD_SB_CLEAN))
3983 mddev->recovery_cp = MaxSector;
3985 mddev->recovery_cp = 0;
3986 mddev->persistent = ! info->not_persistent;
3988 mddev->layout = info->layout;
3989 mddev->chunk_size = info->chunk_size;
3991 mddev->max_disks = MD_SB_DISKS;
3993 mddev->sb_dirty = 1;
3995 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3996 mddev->bitmap_offset = 0;
3998 mddev->reshape_position = MaxSector;
4001 * Generate a 128 bit UUID
4003 get_random_bytes(mddev->uuid, 16);
4005 mddev->new_level = mddev->level;
4006 mddev->new_chunk = mddev->chunk_size;
4007 mddev->new_layout = mddev->layout;
4008 mddev->delta_disks = 0;
4013 static int update_size(mddev_t *mddev, unsigned long size)
4017 struct list_head *tmp;
4018 int fit = (size == 0);
4020 if (mddev->pers->resize == NULL)
4022 /* The "size" is the amount of each device that is used.
4023 * This can only make sense for arrays with redundancy.
4024 * linear and raid0 always use whatever space is available
4025 * We can only consider changing the size if no resync
4026 * or reconstruction is happening, and if the new size
4027 * is acceptable. It must fit before the sb_offset or,
4028 * if that is <data_offset, it must fit before the
4029 * size of each device.
4030 * If size is zero, we find the largest size that fits.
4032 if (mddev->sync_thread)
4034 ITERATE_RDEV(mddev,rdev,tmp) {
4036 if (rdev->sb_offset > rdev->data_offset)
4037 avail = (rdev->sb_offset*2) - rdev->data_offset;
4039 avail = get_capacity(rdev->bdev->bd_disk)
4040 - rdev->data_offset;
4041 if (fit && (size == 0 || size > avail/2))
4043 if (avail < ((sector_t)size << 1))
4046 rv = mddev->pers->resize(mddev, (sector_t)size *2);
4048 struct block_device *bdev;
4050 bdev = bdget_disk(mddev->gendisk, 0);
4052 mutex_lock(&bdev->bd_inode->i_mutex);
4053 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4054 mutex_unlock(&bdev->bd_inode->i_mutex);
4061 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4064 /* change the number of raid disks */
4065 if (mddev->pers->check_reshape == NULL)
4067 if (raid_disks <= 0 ||
4068 raid_disks >= mddev->max_disks)
4070 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4072 mddev->delta_disks = raid_disks - mddev->raid_disks;
4074 rv = mddev->pers->check_reshape(mddev);
4080 * update_array_info is used to change the configuration of an
4082 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4083 * fields in the info are checked against the array.
4084 * Any differences that cannot be handled will cause an error.
4085 * Normally, only one change can be managed at a time.
4087 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4093 /* calculate expected state,ignoring low bits */
4094 if (mddev->bitmap && mddev->bitmap_offset)
4095 state |= (1 << MD_SB_BITMAP_PRESENT);
4097 if (mddev->major_version != info->major_version ||
4098 mddev->minor_version != info->minor_version ||
4099 /* mddev->patch_version != info->patch_version || */
4100 mddev->ctime != info->ctime ||
4101 mddev->level != info->level ||
4102 /* mddev->layout != info->layout || */
4103 !mddev->persistent != info->not_persistent||
4104 mddev->chunk_size != info->chunk_size ||
4105 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4106 ((state^info->state) & 0xfffffe00)
4109 /* Check there is only one change */
4110 if (info->size >= 0 && mddev->size != info->size) cnt++;
4111 if (mddev->raid_disks != info->raid_disks) cnt++;
4112 if (mddev->layout != info->layout) cnt++;
4113 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4114 if (cnt == 0) return 0;
4115 if (cnt > 1) return -EINVAL;
4117 if (mddev->layout != info->layout) {
4119 * we don't need to do anything at the md level, the
4120 * personality will take care of it all.
4122 if (mddev->pers->reconfig == NULL)
4125 return mddev->pers->reconfig(mddev, info->layout, -1);
4127 if (info->size >= 0 && mddev->size != info->size)
4128 rv = update_size(mddev, info->size);
4130 if (mddev->raid_disks != info->raid_disks)
4131 rv = update_raid_disks(mddev, info->raid_disks);
4133 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4134 if (mddev->pers->quiesce == NULL)
4136 if (mddev->recovery || mddev->sync_thread)
4138 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4139 /* add the bitmap */
4142 if (mddev->default_bitmap_offset == 0)
4144 mddev->bitmap_offset = mddev->default_bitmap_offset;
4145 mddev->pers->quiesce(mddev, 1);
4146 rv = bitmap_create(mddev);
4148 bitmap_destroy(mddev);
4149 mddev->pers->quiesce(mddev, 0);
4151 /* remove the bitmap */
4154 if (mddev->bitmap->file)
4156 mddev->pers->quiesce(mddev, 1);
4157 bitmap_destroy(mddev);
4158 mddev->pers->quiesce(mddev, 0);
4159 mddev->bitmap_offset = 0;
4162 md_update_sb(mddev);
4166 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4170 if (mddev->pers == NULL)
4173 rdev = find_rdev(mddev, dev);
4177 md_error(mddev, rdev);
4181 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4183 mddev_t *mddev = bdev->bd_disk->private_data;
4187 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4191 static int md_ioctl(struct inode *inode, struct file *file,
4192 unsigned int cmd, unsigned long arg)
4195 void __user *argp = (void __user *)arg;
4196 mddev_t *mddev = NULL;
4198 if (!capable(CAP_SYS_ADMIN))
4202 * Commands dealing with the RAID driver but not any
4208 err = get_version(argp);
4211 case PRINT_RAID_DEBUG:
4219 autostart_arrays(arg);
4226 * Commands creating/starting a new array:
4229 mddev = inode->i_bdev->bd_disk->private_data;
4237 if (cmd == START_ARRAY) {
4238 /* START_ARRAY doesn't need to lock the array as autostart_array
4239 * does the locking, and it could even be a different array
4244 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
4245 "This will not be supported beyond July 2006\n",
4246 current->comm, current->pid);
4249 err = autostart_array(new_decode_dev(arg));
4251 printk(KERN_WARNING "md: autostart failed!\n");
4257 err = mddev_lock(mddev);
4260 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4267 case SET_ARRAY_INFO:
4269 mdu_array_info_t info;
4271 memset(&info, 0, sizeof(info));
4272 else if (copy_from_user(&info, argp, sizeof(info))) {
4277 err = update_array_info(mddev, &info);
4279 printk(KERN_WARNING "md: couldn't update"
4280 " array info. %d\n", err);
4285 if (!list_empty(&mddev->disks)) {
4287 "md: array %s already has disks!\n",
4292 if (mddev->raid_disks) {
4294 "md: array %s already initialised!\n",
4299 err = set_array_info(mddev, &info);
4301 printk(KERN_WARNING "md: couldn't set"
4302 " array info. %d\n", err);
4312 * Commands querying/configuring an existing array:
4314 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4315 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4316 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4317 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
4323 * Commands even a read-only array can execute:
4327 case GET_ARRAY_INFO:
4328 err = get_array_info(mddev, argp);
4331 case GET_BITMAP_FILE:
4332 err = get_bitmap_file(mddev, argp);
4336 err = get_disk_info(mddev, argp);
4339 case RESTART_ARRAY_RW:
4340 err = restart_array(mddev);
4344 err = do_md_stop (mddev, 0);
4348 err = do_md_stop (mddev, 1);
4352 * We have a problem here : there is no easy way to give a CHS
4353 * virtual geometry. We currently pretend that we have a 2 heads
4354 * 4 sectors (with a BIG number of cylinders...). This drives
4355 * dosfs just mad... ;-)
4360 * The remaining ioctls are changing the state of the
4361 * superblock, so we do not allow them on read-only arrays.
4362 * However non-MD ioctls (e.g. get-size) will still come through
4363 * here and hit the 'default' below, so only disallow
4364 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4366 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4367 mddev->ro && mddev->pers) {
4368 if (mddev->ro == 2) {
4370 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4371 md_wakeup_thread(mddev->thread);
4383 mdu_disk_info_t info;
4384 if (copy_from_user(&info, argp, sizeof(info)))
4387 err = add_new_disk(mddev, &info);
4391 case HOT_REMOVE_DISK:
4392 err = hot_remove_disk(mddev, new_decode_dev(arg));
4396 err = hot_add_disk(mddev, new_decode_dev(arg));
4399 case SET_DISK_FAULTY:
4400 err = set_disk_faulty(mddev, new_decode_dev(arg));
4404 err = do_md_run (mddev);
4407 case SET_BITMAP_FILE:
4408 err = set_bitmap_file(mddev, (int)arg);
4418 mddev_unlock(mddev);
4428 static int md_open(struct inode *inode, struct file *file)
4431 * Succeed if we can lock the mddev, which confirms that
4432 * it isn't being stopped right now.
4434 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4437 if ((err = mddev_lock(mddev)))
4442 mddev_unlock(mddev);
4444 check_disk_change(inode->i_bdev);
4449 static int md_release(struct inode *inode, struct file * file)
4451 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4460 static int md_media_changed(struct gendisk *disk)
4462 mddev_t *mddev = disk->private_data;
4464 return mddev->changed;
4467 static int md_revalidate(struct gendisk *disk)
4469 mddev_t *mddev = disk->private_data;
4474 static struct block_device_operations md_fops =
4476 .owner = THIS_MODULE,
4478 .release = md_release,
4480 .getgeo = md_getgeo,
4481 .media_changed = md_media_changed,
4482 .revalidate_disk= md_revalidate,
4485 static int md_thread(void * arg)
4487 mdk_thread_t *thread = arg;
4490 * md_thread is a 'system-thread', it's priority should be very
4491 * high. We avoid resource deadlocks individually in each
4492 * raid personality. (RAID5 does preallocation) We also use RR and
4493 * the very same RT priority as kswapd, thus we will never get
4494 * into a priority inversion deadlock.
4496 * we definitely have to have equal or higher priority than
4497 * bdflush, otherwise bdflush will deadlock if there are too
4498 * many dirty RAID5 blocks.
4501 allow_signal(SIGKILL);
4502 while (!kthread_should_stop()) {
4504 /* We need to wait INTERRUPTIBLE so that
4505 * we don't add to the load-average.
4506 * That means we need to be sure no signals are
4509 if (signal_pending(current))
4510 flush_signals(current);
4512 wait_event_interruptible_timeout
4514 test_bit(THREAD_WAKEUP, &thread->flags)
4515 || kthread_should_stop(),
4519 clear_bit(THREAD_WAKEUP, &thread->flags);
4521 thread->run(thread->mddev);
4527 void md_wakeup_thread(mdk_thread_t *thread)
4530 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4531 set_bit(THREAD_WAKEUP, &thread->flags);
4532 wake_up(&thread->wqueue);
4536 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4539 mdk_thread_t *thread;
4541 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4545 init_waitqueue_head(&thread->wqueue);
4548 thread->mddev = mddev;
4549 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4550 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4551 if (IS_ERR(thread->tsk)) {
4558 void md_unregister_thread(mdk_thread_t *thread)
4560 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4562 kthread_stop(thread->tsk);
4566 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4573 if (!rdev || test_bit(Faulty, &rdev->flags))
4576 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4578 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4579 __builtin_return_address(0),__builtin_return_address(1),
4580 __builtin_return_address(2),__builtin_return_address(3));
4582 if (!mddev->pers->error_handler)
4584 mddev->pers->error_handler(mddev,rdev);
4585 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4586 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4587 md_wakeup_thread(mddev->thread);
4588 md_new_event_inintr(mddev);
4591 /* seq_file implementation /proc/mdstat */
4593 static void status_unused(struct seq_file *seq)
4597 struct list_head *tmp;
4599 seq_printf(seq, "unused devices: ");
4601 ITERATE_RDEV_PENDING(rdev,tmp) {
4602 char b[BDEVNAME_SIZE];
4604 seq_printf(seq, "%s ",
4605 bdevname(rdev->bdev,b));
4608 seq_printf(seq, "<none>");
4610 seq_printf(seq, "\n");
4614 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4616 sector_t max_blocks, resync, res;
4617 unsigned long dt, db, rt;
4619 unsigned int per_milli;
4621 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4623 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4624 max_blocks = mddev->resync_max_sectors >> 1;
4626 max_blocks = mddev->size;
4629 * Should not happen.
4635 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4636 * in a sector_t, and (max_blocks>>scale) will fit in a
4637 * u32, as those are the requirements for sector_div.
4638 * Thus 'scale' must be at least 10
4641 if (sizeof(sector_t) > sizeof(unsigned long)) {
4642 while ( max_blocks/2 > (1ULL<<(scale+32)))
4645 res = (resync>>scale)*1000;
4646 sector_div(res, (u32)((max_blocks>>scale)+1));
4650 int i, x = per_milli/50, y = 20-x;
4651 seq_printf(seq, "[");
4652 for (i = 0; i < x; i++)
4653 seq_printf(seq, "=");
4654 seq_printf(seq, ">");
4655 for (i = 0; i < y; i++)
4656 seq_printf(seq, ".");
4657 seq_printf(seq, "] ");
4659 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4660 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4662 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4663 "resync" : "recovery")),
4664 per_milli/10, per_milli % 10,
4665 (unsigned long long) resync,
4666 (unsigned long long) max_blocks);
4669 * We do not want to overflow, so the order of operands and
4670 * the * 100 / 100 trick are important. We do a +1 to be
4671 * safe against division by zero. We only estimate anyway.
4673 * dt: time from mark until now
4674 * db: blocks written from mark until now
4675 * rt: remaining time
4677 dt = ((jiffies - mddev->resync_mark) / HZ);
4679 db = resync - (mddev->resync_mark_cnt/2);
4680 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100;
4682 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4684 seq_printf(seq, " speed=%ldK/sec", db/dt);
4687 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4689 struct list_head *tmp;
4699 spin_lock(&all_mddevs_lock);
4700 list_for_each(tmp,&all_mddevs)
4702 mddev = list_entry(tmp, mddev_t, all_mddevs);
4704 spin_unlock(&all_mddevs_lock);
4707 spin_unlock(&all_mddevs_lock);
4709 return (void*)2;/* tail */
4713 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4715 struct list_head *tmp;
4716 mddev_t *next_mddev, *mddev = v;
4722 spin_lock(&all_mddevs_lock);
4724 tmp = all_mddevs.next;
4726 tmp = mddev->all_mddevs.next;
4727 if (tmp != &all_mddevs)
4728 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4730 next_mddev = (void*)2;
4733 spin_unlock(&all_mddevs_lock);
4741 static void md_seq_stop(struct seq_file *seq, void *v)
4745 if (mddev && v != (void*)1 && v != (void*)2)
4749 struct mdstat_info {
4753 static int md_seq_show(struct seq_file *seq, void *v)
4757 struct list_head *tmp2;
4759 struct mdstat_info *mi = seq->private;
4760 struct bitmap *bitmap;
4762 if (v == (void*)1) {
4763 struct mdk_personality *pers;
4764 seq_printf(seq, "Personalities : ");
4765 spin_lock(&pers_lock);
4766 list_for_each_entry(pers, &pers_list, list)
4767 seq_printf(seq, "[%s] ", pers->name);
4769 spin_unlock(&pers_lock);
4770 seq_printf(seq, "\n");
4771 mi->event = atomic_read(&md_event_count);
4774 if (v == (void*)2) {
4779 if (mddev_lock(mddev) < 0)
4782 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4783 seq_printf(seq, "%s : %sactive", mdname(mddev),
4784 mddev->pers ? "" : "in");
4787 seq_printf(seq, " (read-only)");
4789 seq_printf(seq, "(auto-read-only)");
4790 seq_printf(seq, " %s", mddev->pers->name);
4794 ITERATE_RDEV(mddev,rdev,tmp2) {
4795 char b[BDEVNAME_SIZE];
4796 seq_printf(seq, " %s[%d]",
4797 bdevname(rdev->bdev,b), rdev->desc_nr);
4798 if (test_bit(WriteMostly, &rdev->flags))
4799 seq_printf(seq, "(W)");
4800 if (test_bit(Faulty, &rdev->flags)) {
4801 seq_printf(seq, "(F)");
4803 } else if (rdev->raid_disk < 0)
4804 seq_printf(seq, "(S)"); /* spare */
4808 if (!list_empty(&mddev->disks)) {
4810 seq_printf(seq, "\n %llu blocks",
4811 (unsigned long long)mddev->array_size);
4813 seq_printf(seq, "\n %llu blocks",
4814 (unsigned long long)size);
4816 if (mddev->persistent) {
4817 if (mddev->major_version != 0 ||
4818 mddev->minor_version != 90) {
4819 seq_printf(seq," super %d.%d",
4820 mddev->major_version,
4821 mddev->minor_version);
4824 seq_printf(seq, " super non-persistent");
4827 mddev->pers->status (seq, mddev);
4828 seq_printf(seq, "\n ");
4829 if (mddev->pers->sync_request) {
4830 if (mddev->curr_resync > 2) {
4831 status_resync (seq, mddev);
4832 seq_printf(seq, "\n ");
4833 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4834 seq_printf(seq, "\tresync=DELAYED\n ");
4835 else if (mddev->recovery_cp < MaxSector)
4836 seq_printf(seq, "\tresync=PENDING\n ");
4839 seq_printf(seq, "\n ");
4841 if ((bitmap = mddev->bitmap)) {
4842 unsigned long chunk_kb;
4843 unsigned long flags;
4844 spin_lock_irqsave(&bitmap->lock, flags);
4845 chunk_kb = bitmap->chunksize >> 10;
4846 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4848 bitmap->pages - bitmap->missing_pages,
4850 (bitmap->pages - bitmap->missing_pages)
4851 << (PAGE_SHIFT - 10),
4852 chunk_kb ? chunk_kb : bitmap->chunksize,
4853 chunk_kb ? "KB" : "B");
4855 seq_printf(seq, ", file: ");
4856 seq_path(seq, bitmap->file->f_vfsmnt,
4857 bitmap->file->f_dentry," \t\n");
4860 seq_printf(seq, "\n");
4861 spin_unlock_irqrestore(&bitmap->lock, flags);
4864 seq_printf(seq, "\n");
4866 mddev_unlock(mddev);
4871 static struct seq_operations md_seq_ops = {
4872 .start = md_seq_start,
4873 .next = md_seq_next,
4874 .stop = md_seq_stop,
4875 .show = md_seq_show,
4878 static int md_seq_open(struct inode *inode, struct file *file)
4881 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4885 error = seq_open(file, &md_seq_ops);
4889 struct seq_file *p = file->private_data;
4891 mi->event = atomic_read(&md_event_count);
4896 static int md_seq_release(struct inode *inode, struct file *file)
4898 struct seq_file *m = file->private_data;
4899 struct mdstat_info *mi = m->private;
4902 return seq_release(inode, file);
4905 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4907 struct seq_file *m = filp->private_data;
4908 struct mdstat_info *mi = m->private;
4911 poll_wait(filp, &md_event_waiters, wait);
4913 /* always allow read */
4914 mask = POLLIN | POLLRDNORM;
4916 if (mi->event != atomic_read(&md_event_count))
4917 mask |= POLLERR | POLLPRI;
4921 static struct file_operations md_seq_fops = {
4922 .open = md_seq_open,
4924 .llseek = seq_lseek,
4925 .release = md_seq_release,
4926 .poll = mdstat_poll,
4929 int register_md_personality(struct mdk_personality *p)
4931 spin_lock(&pers_lock);
4932 list_add_tail(&p->list, &pers_list);
4933 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4934 spin_unlock(&pers_lock);
4938 int unregister_md_personality(struct mdk_personality *p)
4940 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4941 spin_lock(&pers_lock);
4942 list_del_init(&p->list);
4943 spin_unlock(&pers_lock);
4947 static int is_mddev_idle(mddev_t *mddev)
4950 struct list_head *tmp;
4952 unsigned long curr_events;
4955 ITERATE_RDEV(mddev,rdev,tmp) {
4956 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4957 curr_events = disk_stat_read(disk, sectors[0]) +
4958 disk_stat_read(disk, sectors[1]) -
4959 atomic_read(&disk->sync_io);
4960 /* The difference between curr_events and last_events
4961 * will be affected by any new non-sync IO (making
4962 * curr_events bigger) and any difference in the amount of
4963 * in-flight syncio (making current_events bigger or smaller)
4964 * The amount in-flight is currently limited to
4965 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4966 * which is at most 4096 sectors.
4967 * These numbers are fairly fragile and should be made
4968 * more robust, probably by enforcing the
4969 * 'window size' that md_do_sync sort-of uses.
4971 * Note: the following is an unsigned comparison.
4973 if ((curr_events - rdev->last_events + 4096) > 8192) {
4974 rdev->last_events = curr_events;
4981 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4983 /* another "blocks" (512byte) blocks have been synced */
4984 atomic_sub(blocks, &mddev->recovery_active);
4985 wake_up(&mddev->recovery_wait);
4987 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4988 md_wakeup_thread(mddev->thread);
4989 // stop recovery, signal do_sync ....
4994 /* md_write_start(mddev, bi)
4995 * If we need to update some array metadata (e.g. 'active' flag
4996 * in superblock) before writing, schedule a superblock update
4997 * and wait for it to complete.
4999 void md_write_start(mddev_t *mddev, struct bio *bi)
5001 if (bio_data_dir(bi) != WRITE)
5004 BUG_ON(mddev->ro == 1);
5005 if (mddev->ro == 2) {
5006 /* need to switch to read/write */
5008 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5009 md_wakeup_thread(mddev->thread);
5011 atomic_inc(&mddev->writes_pending);
5012 if (mddev->in_sync) {
5013 spin_lock_irq(&mddev->write_lock);
5014 if (mddev->in_sync) {
5016 mddev->sb_dirty = 3;
5017 md_wakeup_thread(mddev->thread);
5019 spin_unlock_irq(&mddev->write_lock);
5021 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
5024 void md_write_end(mddev_t *mddev)
5026 if (atomic_dec_and_test(&mddev->writes_pending)) {
5027 if (mddev->safemode == 2)
5028 md_wakeup_thread(mddev->thread);
5029 else if (mddev->safemode_delay)
5030 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5034 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5036 #define SYNC_MARKS 10
5037 #define SYNC_MARK_STEP (3*HZ)
5038 void md_do_sync(mddev_t *mddev)
5041 unsigned int currspeed = 0,
5043 sector_t max_sectors,j, io_sectors;
5044 unsigned long mark[SYNC_MARKS];
5045 sector_t mark_cnt[SYNC_MARKS];
5047 struct list_head *tmp;
5048 sector_t last_check;
5050 struct list_head *rtmp;
5053 /* just incase thread restarts... */
5054 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5056 if (mddev->ro) /* never try to sync a read-only array */
5059 /* we overload curr_resync somewhat here.
5060 * 0 == not engaged in resync at all
5061 * 2 == checking that there is no conflict with another sync
5062 * 1 == like 2, but have yielded to allow conflicting resync to
5064 * other == active in resync - this many blocks
5066 * Before starting a resync we must have set curr_resync to
5067 * 2, and then checked that every "conflicting" array has curr_resync
5068 * less than ours. When we find one that is the same or higher
5069 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5070 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5071 * This will mean we have to start checking from the beginning again.
5076 mddev->curr_resync = 2;
5079 if (kthread_should_stop()) {
5080 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5083 ITERATE_MDDEV(mddev2,tmp) {
5084 if (mddev2 == mddev)
5086 if (mddev2->curr_resync &&
5087 match_mddev_units(mddev,mddev2)) {
5089 if (mddev < mddev2 && mddev->curr_resync == 2) {
5090 /* arbitrarily yield */
5091 mddev->curr_resync = 1;
5092 wake_up(&resync_wait);
5094 if (mddev > mddev2 && mddev->curr_resync == 1)
5095 /* no need to wait here, we can wait the next
5096 * time 'round when curr_resync == 2
5099 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5100 if (!kthread_should_stop() &&
5101 mddev2->curr_resync >= mddev->curr_resync) {
5102 printk(KERN_INFO "md: delaying resync of %s"
5103 " until %s has finished resync (they"
5104 " share one or more physical units)\n",
5105 mdname(mddev), mdname(mddev2));
5108 finish_wait(&resync_wait, &wq);
5111 finish_wait(&resync_wait, &wq);
5114 } while (mddev->curr_resync < 2);
5117 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5118 /* resync follows the size requested by the personality,
5119 * which defaults to physical size, but can be virtual size
5121 max_sectors = mddev->resync_max_sectors;
5122 mddev->resync_mismatches = 0;
5123 /* we don't use the checkpoint if there's a bitmap */
5124 if (!mddev->bitmap &&
5125 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5126 j = mddev->recovery_cp;
5127 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5128 max_sectors = mddev->size << 1;
5130 /* recovery follows the physical size of devices */
5131 max_sectors = mddev->size << 1;
5133 ITERATE_RDEV(mddev,rdev,rtmp)
5134 if (rdev->raid_disk >= 0 &&
5135 !test_bit(Faulty, &rdev->flags) &&
5136 !test_bit(In_sync, &rdev->flags) &&
5137 rdev->recovery_offset < j)
5138 j = rdev->recovery_offset;
5141 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
5142 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
5143 " %d KB/sec/disc.\n", speed_min(mddev));
5144 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5145 "(but not more than %d KB/sec) for reconstruction.\n",
5148 is_mddev_idle(mddev); /* this also initializes IO event counters */
5151 for (m = 0; m < SYNC_MARKS; m++) {
5153 mark_cnt[m] = io_sectors;
5156 mddev->resync_mark = mark[last_mark];
5157 mddev->resync_mark_cnt = mark_cnt[last_mark];
5160 * Tune reconstruction:
5162 window = 32*(PAGE_SIZE/512);
5163 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5164 window/2,(unsigned long long) max_sectors/2);
5166 atomic_set(&mddev->recovery_active, 0);
5167 init_waitqueue_head(&mddev->recovery_wait);
5172 "md: resuming recovery of %s from checkpoint.\n",
5174 mddev->curr_resync = j;
5177 while (j < max_sectors) {
5181 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5182 currspeed < speed_min(mddev));
5184 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5188 if (!skipped) { /* actual IO requested */
5189 io_sectors += sectors;
5190 atomic_add(sectors, &mddev->recovery_active);
5194 if (j>1) mddev->curr_resync = j;
5195 if (last_check == 0)
5196 /* this is the earliers that rebuilt will be
5197 * visible in /proc/mdstat
5199 md_new_event(mddev);
5201 if (last_check + window > io_sectors || j == max_sectors)
5204 last_check = io_sectors;
5206 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5207 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5211 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5213 int next = (last_mark+1) % SYNC_MARKS;
5215 mddev->resync_mark = mark[next];
5216 mddev->resync_mark_cnt = mark_cnt[next];
5217 mark[next] = jiffies;
5218 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5223 if (kthread_should_stop()) {
5225 * got a signal, exit.
5228 "md: md_do_sync() got signal ... exiting\n");
5229 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5234 * this loop exits only if either when we are slower than
5235 * the 'hard' speed limit, or the system was IO-idle for
5237 * the system might be non-idle CPU-wise, but we only care
5238 * about not overloading the IO subsystem. (things like an
5239 * e2fsck being done on the RAID array should execute fast)
5241 mddev->queue->unplug_fn(mddev->queue);
5244 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5245 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5247 if (currspeed > speed_min(mddev)) {
5248 if ((currspeed > speed_max(mddev)) ||
5249 !is_mddev_idle(mddev)) {
5255 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
5257 * this also signals 'finished resyncing' to md_stop
5260 mddev->queue->unplug_fn(mddev->queue);
5262 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5264 /* tell personality that we are finished */
5265 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5267 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5268 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
5269 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5270 mddev->curr_resync > 2) {
5271 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5272 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5273 if (mddev->curr_resync >= mddev->recovery_cp) {
5275 "md: checkpointing recovery of %s.\n",
5277 mddev->recovery_cp = mddev->curr_resync;
5280 mddev->recovery_cp = MaxSector;
5282 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5283 mddev->curr_resync = MaxSector;
5284 ITERATE_RDEV(mddev,rdev,rtmp)
5285 if (rdev->raid_disk >= 0 &&
5286 !test_bit(Faulty, &rdev->flags) &&
5287 !test_bit(In_sync, &rdev->flags) &&
5288 rdev->recovery_offset < mddev->curr_resync)
5289 rdev->recovery_offset = mddev->curr_resync;
5290 mddev->sb_dirty = 1;
5295 mddev->curr_resync = 0;
5296 wake_up(&resync_wait);
5297 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5298 md_wakeup_thread(mddev->thread);
5300 EXPORT_SYMBOL_GPL(md_do_sync);
5304 * This routine is regularly called by all per-raid-array threads to
5305 * deal with generic issues like resync and super-block update.
5306 * Raid personalities that don't have a thread (linear/raid0) do not
5307 * need this as they never do any recovery or update the superblock.
5309 * It does not do any resync itself, but rather "forks" off other threads
5310 * to do that as needed.
5311 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5312 * "->recovery" and create a thread at ->sync_thread.
5313 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5314 * and wakeups up this thread which will reap the thread and finish up.
5315 * This thread also removes any faulty devices (with nr_pending == 0).
5317 * The overall approach is:
5318 * 1/ if the superblock needs updating, update it.
5319 * 2/ If a recovery thread is running, don't do anything else.
5320 * 3/ If recovery has finished, clean up, possibly marking spares active.
5321 * 4/ If there are any faulty devices, remove them.
5322 * 5/ If array is degraded, try to add spares devices
5323 * 6/ If array has spares or is not in-sync, start a resync thread.
5325 void md_check_recovery(mddev_t *mddev)
5328 struct list_head *rtmp;
5332 bitmap_daemon_work(mddev->bitmap);
5337 if (signal_pending(current)) {
5338 if (mddev->pers->sync_request) {
5339 printk(KERN_INFO "md: %s in immediate safe mode\n",
5341 mddev->safemode = 2;
5343 flush_signals(current);
5348 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5349 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5350 (mddev->safemode == 1) ||
5351 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5352 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5356 if (mddev_trylock(mddev)) {
5359 spin_lock_irq(&mddev->write_lock);
5360 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5361 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5363 mddev->sb_dirty = 3;
5365 if (mddev->safemode == 1)
5366 mddev->safemode = 0;
5367 spin_unlock_irq(&mddev->write_lock);
5369 if (mddev->sb_dirty)
5370 md_update_sb(mddev);
5373 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5374 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5375 /* resync/recovery still happening */
5376 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5379 if (mddev->sync_thread) {
5380 /* resync has finished, collect result */
5381 md_unregister_thread(mddev->sync_thread);
5382 mddev->sync_thread = NULL;
5383 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5384 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5386 /* activate any spares */
5387 mddev->pers->spare_active(mddev);
5389 md_update_sb(mddev);
5391 /* if array is no-longer degraded, then any saved_raid_disk
5392 * information must be scrapped
5394 if (!mddev->degraded)
5395 ITERATE_RDEV(mddev,rdev,rtmp)
5396 rdev->saved_raid_disk = -1;
5398 mddev->recovery = 0;
5399 /* flag recovery needed just to double check */
5400 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5401 md_new_event(mddev);
5404 /* Clear some bits that don't mean anything, but
5407 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5408 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5409 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5410 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5412 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5414 /* no recovery is running.
5415 * remove any failed drives, then
5416 * add spares if possible.
5417 * Spare are also removed and re-added, to allow
5418 * the personality to fail the re-add.
5420 ITERATE_RDEV(mddev,rdev,rtmp)
5421 if (rdev->raid_disk >= 0 &&
5422 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
5423 atomic_read(&rdev->nr_pending)==0) {
5424 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5426 sprintf(nm,"rd%d", rdev->raid_disk);
5427 sysfs_remove_link(&mddev->kobj, nm);
5428 rdev->raid_disk = -1;
5432 if (mddev->degraded) {
5433 ITERATE_RDEV(mddev,rdev,rtmp)
5434 if (rdev->raid_disk < 0
5435 && !test_bit(Faulty, &rdev->flags)) {
5436 rdev->recovery_offset = 0;
5437 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5439 sprintf(nm, "rd%d", rdev->raid_disk);
5440 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
5442 md_new_event(mddev);
5449 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5450 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5451 } else if (mddev->recovery_cp < MaxSector) {
5452 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5453 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5454 /* nothing to be done ... */
5457 if (mddev->pers->sync_request) {
5458 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5459 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5460 /* We are adding a device or devices to an array
5461 * which has the bitmap stored on all devices.
5462 * So make sure all bitmap pages get written
5464 bitmap_write_all(mddev->bitmap);
5466 mddev->sync_thread = md_register_thread(md_do_sync,
5469 if (!mddev->sync_thread) {
5470 printk(KERN_ERR "%s: could not start resync"
5473 /* leave the spares where they are, it shouldn't hurt */
5474 mddev->recovery = 0;
5476 md_wakeup_thread(mddev->sync_thread);
5477 md_new_event(mddev);
5480 mddev_unlock(mddev);
5484 static int md_notify_reboot(struct notifier_block *this,
5485 unsigned long code, void *x)
5487 struct list_head *tmp;
5490 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5492 printk(KERN_INFO "md: stopping all md devices.\n");
5494 ITERATE_MDDEV(mddev,tmp)
5495 if (mddev_trylock(mddev)) {
5496 do_md_stop (mddev, 1);
5497 mddev_unlock(mddev);
5500 * certain more exotic SCSI devices are known to be
5501 * volatile wrt too early system reboots. While the
5502 * right place to handle this issue is the given
5503 * driver, we do want to have a safe RAID driver ...
5510 static struct notifier_block md_notifier = {
5511 .notifier_call = md_notify_reboot,
5513 .priority = INT_MAX, /* before any real devices */
5516 static void md_geninit(void)
5518 struct proc_dir_entry *p;
5520 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5522 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5524 p->proc_fops = &md_seq_fops;
5527 static int __init md_init(void)
5531 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5532 " MD_SB_DISKS=%d\n",
5533 MD_MAJOR_VERSION, MD_MINOR_VERSION,
5534 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
5535 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
5538 if (register_blkdev(MAJOR_NR, "md"))
5540 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5541 unregister_blkdev(MAJOR_NR, "md");
5545 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5546 md_probe, NULL, NULL);
5547 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5548 md_probe, NULL, NULL);
5550 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5551 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
5552 S_IFBLK|S_IRUSR|S_IWUSR,
5555 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5556 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
5557 S_IFBLK|S_IRUSR|S_IWUSR,
5561 register_reboot_notifier(&md_notifier);
5562 raid_table_header = register_sysctl_table(raid_root_table, 1);
5572 * Searches all registered partitions for autorun RAID arrays
5575 static dev_t detected_devices[128];
5578 void md_autodetect_dev(dev_t dev)
5580 if (dev_cnt >= 0 && dev_cnt < 127)
5581 detected_devices[dev_cnt++] = dev;
5585 static void autostart_arrays(int part)
5590 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5592 for (i = 0; i < dev_cnt; i++) {
5593 dev_t dev = detected_devices[i];
5595 rdev = md_import_device(dev,0, 0);
5599 if (test_bit(Faulty, &rdev->flags)) {
5603 list_add(&rdev->same_set, &pending_raid_disks);
5607 autorun_devices(part);
5612 static __exit void md_exit(void)
5615 struct list_head *tmp;
5617 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5618 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
5619 for (i=0; i < MAX_MD_DEVS; i++)
5620 devfs_remove("md/%d", i);
5621 for (i=0; i < MAX_MD_DEVS; i++)
5622 devfs_remove("md/d%d", i);
5626 unregister_blkdev(MAJOR_NR,"md");
5627 unregister_blkdev(mdp_major, "mdp");
5628 unregister_reboot_notifier(&md_notifier);
5629 unregister_sysctl_table(raid_table_header);
5630 remove_proc_entry("mdstat", NULL);
5631 ITERATE_MDDEV(mddev,tmp) {
5632 struct gendisk *disk = mddev->gendisk;
5635 export_array(mddev);
5638 mddev->gendisk = NULL;
5643 module_init(md_init)
5644 module_exit(md_exit)
5646 static int get_ro(char *buffer, struct kernel_param *kp)
5648 return sprintf(buffer, "%d", start_readonly);
5650 static int set_ro(const char *val, struct kernel_param *kp)
5653 int num = simple_strtoul(val, &e, 10);
5654 if (*val && (*e == '\0' || *e == '\n')) {
5655 start_readonly = num;
5661 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5662 module_param(start_dirty_degraded, int, 0644);
5665 EXPORT_SYMBOL(register_md_personality);
5666 EXPORT_SYMBOL(unregister_md_personality);
5667 EXPORT_SYMBOL(md_error);
5668 EXPORT_SYMBOL(md_done_sync);
5669 EXPORT_SYMBOL(md_write_start);
5670 EXPORT_SYMBOL(md_write_end);
5671 EXPORT_SYMBOL(md_register_thread);
5672 EXPORT_SYMBOL(md_unregister_thread);
5673 EXPORT_SYMBOL(md_wakeup_thread);
5674 EXPORT_SYMBOL(md_check_recovery);
5675 MODULE_LICENSE("GPL");
5677 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);