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");
1748 static struct rdev_sysfs_entry
1749 rdev_state = __ATTR_RO(state);
1752 super_show(mdk_rdev_t *rdev, char *page)
1754 if (rdev->sb_loaded && rdev->sb_size) {
1755 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1756 return rdev->sb_size;
1760 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1763 errors_show(mdk_rdev_t *rdev, char *page)
1765 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1769 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1772 unsigned long n = simple_strtoul(buf, &e, 10);
1773 if (*buf && (*e == 0 || *e == '\n')) {
1774 atomic_set(&rdev->corrected_errors, n);
1779 static struct rdev_sysfs_entry rdev_errors =
1780 __ATTR(errors, 0644, errors_show, errors_store);
1783 slot_show(mdk_rdev_t *rdev, char *page)
1785 if (rdev->raid_disk < 0)
1786 return sprintf(page, "none\n");
1788 return sprintf(page, "%d\n", rdev->raid_disk);
1792 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1795 int slot = simple_strtoul(buf, &e, 10);
1796 if (strncmp(buf, "none", 4)==0)
1798 else if (e==buf || (*e && *e!= '\n'))
1800 if (rdev->mddev->pers)
1801 /* Cannot set slot in active array (yet) */
1803 if (slot >= rdev->mddev->raid_disks)
1805 rdev->raid_disk = slot;
1806 /* assume it is working */
1808 set_bit(In_sync, &rdev->flags);
1813 static struct rdev_sysfs_entry rdev_slot =
1814 __ATTR(slot, 0644, slot_show, slot_store);
1817 offset_show(mdk_rdev_t *rdev, char *page)
1819 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1823 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1826 unsigned long long offset = simple_strtoull(buf, &e, 10);
1827 if (e==buf || (*e && *e != '\n'))
1829 if (rdev->mddev->pers)
1831 rdev->data_offset = offset;
1835 static struct rdev_sysfs_entry rdev_offset =
1836 __ATTR(offset, 0644, offset_show, offset_store);
1839 rdev_size_show(mdk_rdev_t *rdev, char *page)
1841 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1845 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1848 unsigned long long size = simple_strtoull(buf, &e, 10);
1849 if (e==buf || (*e && *e != '\n'))
1851 if (rdev->mddev->pers)
1854 if (size < rdev->mddev->size || rdev->mddev->size == 0)
1855 rdev->mddev->size = size;
1859 static struct rdev_sysfs_entry rdev_size =
1860 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1862 static struct attribute *rdev_default_attrs[] = {
1872 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1874 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1875 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1879 return entry->show(rdev, page);
1883 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1884 const char *page, size_t length)
1886 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1887 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1891 return entry->store(rdev, page, length);
1894 static void rdev_free(struct kobject *ko)
1896 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1899 static struct sysfs_ops rdev_sysfs_ops = {
1900 .show = rdev_attr_show,
1901 .store = rdev_attr_store,
1903 static struct kobj_type rdev_ktype = {
1904 .release = rdev_free,
1905 .sysfs_ops = &rdev_sysfs_ops,
1906 .default_attrs = rdev_default_attrs,
1910 * Import a device. If 'super_format' >= 0, then sanity check the superblock
1912 * mark the device faulty if:
1914 * - the device is nonexistent (zero size)
1915 * - the device has no valid superblock
1917 * a faulty rdev _never_ has rdev->sb set.
1919 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1921 char b[BDEVNAME_SIZE];
1926 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1928 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1929 return ERR_PTR(-ENOMEM);
1932 if ((err = alloc_disk_sb(rdev)))
1935 err = lock_rdev(rdev, newdev);
1939 rdev->kobj.parent = NULL;
1940 rdev->kobj.ktype = &rdev_ktype;
1941 kobject_init(&rdev->kobj);
1945 rdev->data_offset = 0;
1946 rdev->sb_events = 0;
1947 atomic_set(&rdev->nr_pending, 0);
1948 atomic_set(&rdev->read_errors, 0);
1949 atomic_set(&rdev->corrected_errors, 0);
1951 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1954 "md: %s has zero or unknown size, marking faulty!\n",
1955 bdevname(rdev->bdev,b));
1960 if (super_format >= 0) {
1961 err = super_types[super_format].
1962 load_super(rdev, NULL, super_minor);
1963 if (err == -EINVAL) {
1965 "md: %s has invalid sb, not importing!\n",
1966 bdevname(rdev->bdev,b));
1971 "md: could not read %s's sb, not importing!\n",
1972 bdevname(rdev->bdev,b));
1976 INIT_LIST_HEAD(&rdev->same_set);
1981 if (rdev->sb_page) {
1987 return ERR_PTR(err);
1991 * Check a full RAID array for plausibility
1995 static void analyze_sbs(mddev_t * mddev)
1998 struct list_head *tmp;
1999 mdk_rdev_t *rdev, *freshest;
2000 char b[BDEVNAME_SIZE];
2003 ITERATE_RDEV(mddev,rdev,tmp)
2004 switch (super_types[mddev->major_version].
2005 load_super(rdev, freshest, mddev->minor_version)) {
2013 "md: fatal superblock inconsistency in %s"
2014 " -- removing from array\n",
2015 bdevname(rdev->bdev,b));
2016 kick_rdev_from_array(rdev);
2020 super_types[mddev->major_version].
2021 validate_super(mddev, freshest);
2024 ITERATE_RDEV(mddev,rdev,tmp) {
2025 if (rdev != freshest)
2026 if (super_types[mddev->major_version].
2027 validate_super(mddev, rdev)) {
2028 printk(KERN_WARNING "md: kicking non-fresh %s"
2030 bdevname(rdev->bdev,b));
2031 kick_rdev_from_array(rdev);
2034 if (mddev->level == LEVEL_MULTIPATH) {
2035 rdev->desc_nr = i++;
2036 rdev->raid_disk = rdev->desc_nr;
2037 set_bit(In_sync, &rdev->flags);
2043 if (mddev->recovery_cp != MaxSector &&
2045 printk(KERN_ERR "md: %s: raid array is not clean"
2046 " -- starting background reconstruction\n",
2052 safe_delay_show(mddev_t *mddev, char *page)
2054 int msec = (mddev->safemode_delay*1000)/HZ;
2055 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2058 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2066 /* remove a period, and count digits after it */
2067 if (len >= sizeof(buf))
2069 strlcpy(buf, cbuf, len);
2071 for (i=0; i<len; i++) {
2073 if (isdigit(buf[i])) {
2078 } else if (buf[i] == '.') {
2083 msec = simple_strtoul(buf, &e, 10);
2084 if (e == buf || (*e && *e != '\n'))
2086 msec = (msec * 1000) / scale;
2088 mddev->safemode_delay = 0;
2090 mddev->safemode_delay = (msec*HZ)/1000;
2091 if (mddev->safemode_delay == 0)
2092 mddev->safemode_delay = 1;
2096 static struct md_sysfs_entry md_safe_delay =
2097 __ATTR(safe_mode_delay, 0644,safe_delay_show, safe_delay_store);
2100 level_show(mddev_t *mddev, char *page)
2102 struct mdk_personality *p = mddev->pers;
2104 return sprintf(page, "%s\n", p->name);
2105 else if (mddev->clevel[0])
2106 return sprintf(page, "%s\n", mddev->clevel);
2107 else if (mddev->level != LEVEL_NONE)
2108 return sprintf(page, "%d\n", mddev->level);
2114 level_store(mddev_t *mddev, const char *buf, size_t len)
2121 if (len >= sizeof(mddev->clevel))
2123 strncpy(mddev->clevel, buf, len);
2124 if (mddev->clevel[len-1] == '\n')
2126 mddev->clevel[len] = 0;
2127 mddev->level = LEVEL_NONE;
2131 static struct md_sysfs_entry md_level =
2132 __ATTR(level, 0644, level_show, level_store);
2135 raid_disks_show(mddev_t *mddev, char *page)
2137 if (mddev->raid_disks == 0)
2139 return sprintf(page, "%d\n", mddev->raid_disks);
2142 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2145 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2147 /* can only set raid_disks if array is not yet active */
2150 unsigned long n = simple_strtoul(buf, &e, 10);
2152 if (!*buf || (*e && *e != '\n'))
2156 rv = update_raid_disks(mddev, n);
2158 mddev->raid_disks = n;
2159 return rv ? rv : len;
2161 static struct md_sysfs_entry md_raid_disks =
2162 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
2165 chunk_size_show(mddev_t *mddev, char *page)
2167 return sprintf(page, "%d\n", mddev->chunk_size);
2171 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2173 /* can only set chunk_size if array is not yet active */
2175 unsigned long n = simple_strtoul(buf, &e, 10);
2179 if (!*buf || (*e && *e != '\n'))
2182 mddev->chunk_size = n;
2185 static struct md_sysfs_entry md_chunk_size =
2186 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2189 * The array state can be:
2192 * No devices, no size, no level
2193 * Equivalent to STOP_ARRAY ioctl
2195 * May have some settings, but array is not active
2196 * all IO results in error
2197 * When written, doesn't tear down array, but just stops it
2198 * suspended (not supported yet)
2199 * All IO requests will block. The array can be reconfigured.
2200 * Writing this, if accepted, will block until array is quiessent
2202 * no resync can happen. no superblocks get written.
2203 * write requests fail
2205 * like readonly, but behaves like 'clean' on a write request.
2207 * clean - no pending writes, but otherwise active.
2208 * When written to inactive array, starts without resync
2209 * If a write request arrives then
2210 * if metadata is known, mark 'dirty' and switch to 'active'.
2211 * if not known, block and switch to write-pending
2212 * If written to an active array that has pending writes, then fails.
2214 * fully active: IO and resync can be happening.
2215 * When written to inactive array, starts with resync
2218 * clean, but writes are blocked waiting for 'active' to be written.
2221 * like active, but no writes have been seen for a while (100msec).
2224 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2225 write_pending, active_idle, bad_word};
2226 char *array_states[] = {
2227 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2228 "write-pending", "active-idle", NULL };
2230 static int match_word(const char *word, char **list)
2233 for (n=0; list[n]; n++)
2234 if (cmd_match(word, list[n]))
2240 array_state_show(mddev_t *mddev, char *page)
2242 enum array_state st = inactive;
2255 else if (mddev->safemode)
2261 if (list_empty(&mddev->disks) &&
2262 mddev->raid_disks == 0 &&
2268 return sprintf(page, "%s\n", array_states[st]);
2271 static int do_md_stop(mddev_t * mddev, int ro);
2272 static int do_md_run(mddev_t * mddev);
2273 static int restart_array(mddev_t *mddev);
2276 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2279 enum array_state st = match_word(buf, array_states);
2284 /* stopping an active array */
2286 if (atomic_read(&mddev->active) > 1)
2288 err = do_md_stop(mddev, 0);
2292 /* stopping an active array */
2294 if (atomic_read(&mddev->active) > 1)
2296 err = do_md_stop(mddev, 2);
2300 break; /* not supported yet */
2303 err = do_md_stop(mddev, 1);
2306 err = do_md_run(mddev);
2310 /* stopping an active array */
2312 err = do_md_stop(mddev, 1);
2314 mddev->ro = 2; /* FIXME mark devices writable */
2317 err = do_md_run(mddev);
2322 restart_array(mddev);
2323 spin_lock_irq(&mddev->write_lock);
2324 if (atomic_read(&mddev->writes_pending) == 0) {
2326 mddev->sb_dirty = 1;
2328 spin_unlock_irq(&mddev->write_lock);
2331 mddev->recovery_cp = MaxSector;
2332 err = do_md_run(mddev);
2337 restart_array(mddev);
2338 mddev->sb_dirty = 0;
2339 wake_up(&mddev->sb_wait);
2343 err = do_md_run(mddev);
2348 /* these cannot be set */
2356 static struct md_sysfs_entry md_array_state = __ATTR(array_state, 0644, array_state_show, array_state_store);
2359 null_show(mddev_t *mddev, char *page)
2365 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2367 /* buf must be %d:%d\n? giving major and minor numbers */
2368 /* The new device is added to the array.
2369 * If the array has a persistent superblock, we read the
2370 * superblock to initialise info and check validity.
2371 * Otherwise, only checking done is that in bind_rdev_to_array,
2372 * which mainly checks size.
2375 int major = simple_strtoul(buf, &e, 10);
2381 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2383 minor = simple_strtoul(e+1, &e, 10);
2384 if (*e && *e != '\n')
2386 dev = MKDEV(major, minor);
2387 if (major != MAJOR(dev) ||
2388 minor != MINOR(dev))
2392 if (mddev->persistent) {
2393 rdev = md_import_device(dev, mddev->major_version,
2394 mddev->minor_version);
2395 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2396 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2397 mdk_rdev_t, same_set);
2398 err = super_types[mddev->major_version]
2399 .load_super(rdev, rdev0, mddev->minor_version);
2404 rdev = md_import_device(dev, -1, -1);
2407 return PTR_ERR(rdev);
2408 err = bind_rdev_to_array(rdev, mddev);
2412 return err ? err : len;
2415 static struct md_sysfs_entry md_new_device =
2416 __ATTR(new_dev, 0200, null_show, new_dev_store);
2419 size_show(mddev_t *mddev, char *page)
2421 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2424 static int update_size(mddev_t *mddev, unsigned long size);
2427 size_store(mddev_t *mddev, const char *buf, size_t len)
2429 /* If array is inactive, we can reduce the component size, but
2430 * not increase it (except from 0).
2431 * If array is active, we can try an on-line resize
2435 unsigned long long size = simple_strtoull(buf, &e, 10);
2436 if (!*buf || *buf == '\n' ||
2441 err = update_size(mddev, size);
2442 md_update_sb(mddev);
2444 if (mddev->size == 0 ||
2450 return err ? err : len;
2453 static struct md_sysfs_entry md_size =
2454 __ATTR(component_size, 0644, size_show, size_store);
2458 * This is either 'none' for arrays with externally managed metadata,
2459 * or N.M for internally known formats
2462 metadata_show(mddev_t *mddev, char *page)
2464 if (mddev->persistent)
2465 return sprintf(page, "%d.%d\n",
2466 mddev->major_version, mddev->minor_version);
2468 return sprintf(page, "none\n");
2472 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2476 if (!list_empty(&mddev->disks))
2479 if (cmd_match(buf, "none")) {
2480 mddev->persistent = 0;
2481 mddev->major_version = 0;
2482 mddev->minor_version = 90;
2485 major = simple_strtoul(buf, &e, 10);
2486 if (e==buf || *e != '.')
2489 minor = simple_strtoul(buf, &e, 10);
2490 if (e==buf || *e != '\n')
2492 if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2493 super_types[major].name == NULL)
2495 mddev->major_version = major;
2496 mddev->minor_version = minor;
2497 mddev->persistent = 1;
2501 static struct md_sysfs_entry md_metadata =
2502 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2505 action_show(mddev_t *mddev, char *page)
2507 char *type = "idle";
2508 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2509 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2510 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2512 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2513 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2515 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2522 return sprintf(page, "%s\n", type);
2526 action_store(mddev_t *mddev, const char *page, size_t len)
2528 if (!mddev->pers || !mddev->pers->sync_request)
2531 if (cmd_match(page, "idle")) {
2532 if (mddev->sync_thread) {
2533 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2534 md_unregister_thread(mddev->sync_thread);
2535 mddev->sync_thread = NULL;
2536 mddev->recovery = 0;
2538 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2539 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2541 else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2542 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2543 else if (cmd_match(page, "reshape")) {
2545 if (mddev->pers->start_reshape == NULL)
2547 err = mddev->pers->start_reshape(mddev);
2551 if (cmd_match(page, "check"))
2552 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2553 else if (!cmd_match(page, "repair"))
2555 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2556 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2558 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2559 md_wakeup_thread(mddev->thread);
2564 mismatch_cnt_show(mddev_t *mddev, char *page)
2566 return sprintf(page, "%llu\n",
2567 (unsigned long long) mddev->resync_mismatches);
2570 static struct md_sysfs_entry
2571 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2574 static struct md_sysfs_entry
2575 md_mismatches = __ATTR_RO(mismatch_cnt);
2578 sync_min_show(mddev_t *mddev, char *page)
2580 return sprintf(page, "%d (%s)\n", speed_min(mddev),
2581 mddev->sync_speed_min ? "local": "system");
2585 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2589 if (strncmp(buf, "system", 6)==0) {
2590 mddev->sync_speed_min = 0;
2593 min = simple_strtoul(buf, &e, 10);
2594 if (buf == e || (*e && *e != '\n') || min <= 0)
2596 mddev->sync_speed_min = min;
2600 static struct md_sysfs_entry md_sync_min =
2601 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2604 sync_max_show(mddev_t *mddev, char *page)
2606 return sprintf(page, "%d (%s)\n", speed_max(mddev),
2607 mddev->sync_speed_max ? "local": "system");
2611 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2615 if (strncmp(buf, "system", 6)==0) {
2616 mddev->sync_speed_max = 0;
2619 max = simple_strtoul(buf, &e, 10);
2620 if (buf == e || (*e && *e != '\n') || max <= 0)
2622 mddev->sync_speed_max = max;
2626 static struct md_sysfs_entry md_sync_max =
2627 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2631 sync_speed_show(mddev_t *mddev, char *page)
2633 unsigned long resync, dt, db;
2634 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2635 dt = ((jiffies - mddev->resync_mark) / HZ);
2637 db = resync - (mddev->resync_mark_cnt);
2638 return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2641 static struct md_sysfs_entry
2642 md_sync_speed = __ATTR_RO(sync_speed);
2645 sync_completed_show(mddev_t *mddev, char *page)
2647 unsigned long max_blocks, resync;
2649 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2650 max_blocks = mddev->resync_max_sectors;
2652 max_blocks = mddev->size << 1;
2654 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2655 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2658 static struct md_sysfs_entry
2659 md_sync_completed = __ATTR_RO(sync_completed);
2662 suspend_lo_show(mddev_t *mddev, char *page)
2664 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2668 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2671 unsigned long long new = simple_strtoull(buf, &e, 10);
2673 if (mddev->pers->quiesce == NULL)
2675 if (buf == e || (*e && *e != '\n'))
2677 if (new >= mddev->suspend_hi ||
2678 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2679 mddev->suspend_lo = new;
2680 mddev->pers->quiesce(mddev, 2);
2685 static struct md_sysfs_entry md_suspend_lo =
2686 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2690 suspend_hi_show(mddev_t *mddev, char *page)
2692 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2696 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2699 unsigned long long new = simple_strtoull(buf, &e, 10);
2701 if (mddev->pers->quiesce == NULL)
2703 if (buf == e || (*e && *e != '\n'))
2705 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2706 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2707 mddev->suspend_hi = new;
2708 mddev->pers->quiesce(mddev, 1);
2709 mddev->pers->quiesce(mddev, 0);
2714 static struct md_sysfs_entry md_suspend_hi =
2715 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2718 static struct attribute *md_default_attrs[] = {
2720 &md_raid_disks.attr,
2721 &md_chunk_size.attr,
2724 &md_new_device.attr,
2725 &md_safe_delay.attr,
2726 &md_array_state.attr,
2730 static struct attribute *md_redundancy_attrs[] = {
2732 &md_mismatches.attr,
2735 &md_sync_speed.attr,
2736 &md_sync_completed.attr,
2737 &md_suspend_lo.attr,
2738 &md_suspend_hi.attr,
2741 static struct attribute_group md_redundancy_group = {
2743 .attrs = md_redundancy_attrs,
2748 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2750 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2751 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2756 rv = mddev_lock(mddev);
2758 rv = entry->show(mddev, page);
2759 mddev_unlock(mddev);
2765 md_attr_store(struct kobject *kobj, struct attribute *attr,
2766 const char *page, size_t length)
2768 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2769 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2774 rv = mddev_lock(mddev);
2776 rv = entry->store(mddev, page, length);
2777 mddev_unlock(mddev);
2782 static void md_free(struct kobject *ko)
2784 mddev_t *mddev = container_of(ko, mddev_t, kobj);
2788 static struct sysfs_ops md_sysfs_ops = {
2789 .show = md_attr_show,
2790 .store = md_attr_store,
2792 static struct kobj_type md_ktype = {
2794 .sysfs_ops = &md_sysfs_ops,
2795 .default_attrs = md_default_attrs,
2800 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2802 static DEFINE_MUTEX(disks_mutex);
2803 mddev_t *mddev = mddev_find(dev);
2804 struct gendisk *disk;
2805 int partitioned = (MAJOR(dev) != MD_MAJOR);
2806 int shift = partitioned ? MdpMinorShift : 0;
2807 int unit = MINOR(dev) >> shift;
2812 mutex_lock(&disks_mutex);
2813 if (mddev->gendisk) {
2814 mutex_unlock(&disks_mutex);
2818 disk = alloc_disk(1 << shift);
2820 mutex_unlock(&disks_mutex);
2824 disk->major = MAJOR(dev);
2825 disk->first_minor = unit << shift;
2827 sprintf(disk->disk_name, "md_d%d", unit);
2828 sprintf(disk->devfs_name, "md/d%d", unit);
2830 sprintf(disk->disk_name, "md%d", unit);
2831 sprintf(disk->devfs_name, "md/%d", unit);
2833 disk->fops = &md_fops;
2834 disk->private_data = mddev;
2835 disk->queue = mddev->queue;
2837 mddev->gendisk = disk;
2838 mutex_unlock(&disks_mutex);
2839 mddev->kobj.parent = &disk->kobj;
2840 mddev->kobj.k_name = NULL;
2841 snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2842 mddev->kobj.ktype = &md_ktype;
2843 kobject_register(&mddev->kobj);
2847 static void md_safemode_timeout(unsigned long data)
2849 mddev_t *mddev = (mddev_t *) data;
2851 mddev->safemode = 1;
2852 md_wakeup_thread(mddev->thread);
2855 static int start_dirty_degraded;
2857 static int do_md_run(mddev_t * mddev)
2861 struct list_head *tmp;
2863 struct gendisk *disk;
2864 struct mdk_personality *pers;
2865 char b[BDEVNAME_SIZE];
2867 if (list_empty(&mddev->disks))
2868 /* cannot run an array with no devices.. */
2875 * Analyze all RAID superblock(s)
2877 if (!mddev->raid_disks)
2880 chunk_size = mddev->chunk_size;
2883 if (chunk_size > MAX_CHUNK_SIZE) {
2884 printk(KERN_ERR "too big chunk_size: %d > %d\n",
2885 chunk_size, MAX_CHUNK_SIZE);
2889 * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2891 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2892 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2895 if (chunk_size < PAGE_SIZE) {
2896 printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2897 chunk_size, PAGE_SIZE);
2901 /* devices must have minimum size of one chunk */
2902 ITERATE_RDEV(mddev,rdev,tmp) {
2903 if (test_bit(Faulty, &rdev->flags))
2905 if (rdev->size < chunk_size / 1024) {
2907 "md: Dev %s smaller than chunk_size:"
2909 bdevname(rdev->bdev,b),
2910 (unsigned long long)rdev->size,
2918 if (mddev->level != LEVEL_NONE)
2919 request_module("md-level-%d", mddev->level);
2920 else if (mddev->clevel[0])
2921 request_module("md-%s", mddev->clevel);
2925 * Drop all container device buffers, from now on
2926 * the only valid external interface is through the md
2928 * Also find largest hardsector size
2930 ITERATE_RDEV(mddev,rdev,tmp) {
2931 if (test_bit(Faulty, &rdev->flags))
2933 sync_blockdev(rdev->bdev);
2934 invalidate_bdev(rdev->bdev, 0);
2937 md_probe(mddev->unit, NULL, NULL);
2938 disk = mddev->gendisk;
2942 spin_lock(&pers_lock);
2943 pers = find_pers(mddev->level, mddev->clevel);
2944 if (!pers || !try_module_get(pers->owner)) {
2945 spin_unlock(&pers_lock);
2946 if (mddev->level != LEVEL_NONE)
2947 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2950 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2955 spin_unlock(&pers_lock);
2956 mddev->level = pers->level;
2957 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2959 if (mddev->reshape_position != MaxSector &&
2960 pers->start_reshape == NULL) {
2961 /* This personality cannot handle reshaping... */
2963 module_put(pers->owner);
2967 mddev->recovery = 0;
2968 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2969 mddev->barriers_work = 1;
2970 mddev->ok_start_degraded = start_dirty_degraded;
2973 mddev->ro = 2; /* read-only, but switch on first write */
2975 err = mddev->pers->run(mddev);
2976 if (!err && mddev->pers->sync_request) {
2977 err = bitmap_create(mddev);
2979 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2980 mdname(mddev), err);
2981 mddev->pers->stop(mddev);
2985 printk(KERN_ERR "md: pers->run() failed ...\n");
2986 module_put(mddev->pers->owner);
2988 bitmap_destroy(mddev);
2991 if (mddev->pers->sync_request)
2992 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2993 else if (mddev->ro == 2) /* auto-readonly not meaningful */
2996 atomic_set(&mddev->writes_pending,0);
2997 mddev->safemode = 0;
2998 mddev->safemode_timer.function = md_safemode_timeout;
2999 mddev->safemode_timer.data = (unsigned long) mddev;
3000 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3003 ITERATE_RDEV(mddev,rdev,tmp)
3004 if (rdev->raid_disk >= 0) {
3006 sprintf(nm, "rd%d", rdev->raid_disk);
3007 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3010 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3011 md_wakeup_thread(mddev->thread);
3013 if (mddev->sb_dirty)
3014 md_update_sb(mddev);
3016 set_capacity(disk, mddev->array_size<<1);
3018 /* If we call blk_queue_make_request here, it will
3019 * re-initialise max_sectors etc which may have been
3020 * refined inside -> run. So just set the bits we need to set.
3021 * Most initialisation happended when we called
3022 * blk_queue_make_request(..., md_fail_request)
3025 mddev->queue->queuedata = mddev;
3026 mddev->queue->make_request_fn = mddev->pers->make_request;
3028 /* If there is a partially-recovered drive we need to
3029 * start recovery here. If we leave it to md_check_recovery,
3030 * it will remove the drives and not do the right thing
3032 if (mddev->degraded) {
3033 struct list_head *rtmp;
3035 ITERATE_RDEV(mddev,rdev,rtmp)
3036 if (rdev->raid_disk >= 0 &&
3037 !test_bit(In_sync, &rdev->flags) &&
3038 !test_bit(Faulty, &rdev->flags))
3039 /* complete an interrupted recovery */
3041 if (spares && mddev->pers->sync_request) {
3042 mddev->recovery = 0;
3043 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3044 mddev->sync_thread = md_register_thread(md_do_sync,
3047 if (!mddev->sync_thread) {
3048 printk(KERN_ERR "%s: could not start resync"
3051 /* leave the spares where they are, it shouldn't hurt */
3052 mddev->recovery = 0;
3054 md_wakeup_thread(mddev->sync_thread);
3059 md_new_event(mddev);
3063 static int restart_array(mddev_t *mddev)
3065 struct gendisk *disk = mddev->gendisk;
3069 * Complain if it has no devices
3072 if (list_empty(&mddev->disks))
3080 mddev->safemode = 0;
3082 set_disk_ro(disk, 0);
3084 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3087 * Kick recovery or resync if necessary
3089 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3090 md_wakeup_thread(mddev->thread);
3091 md_wakeup_thread(mddev->sync_thread);
3100 /* similar to deny_write_access, but accounts for our holding a reference
3101 * to the file ourselves */
3102 static int deny_bitmap_write_access(struct file * file)
3104 struct inode *inode = file->f_mapping->host;
3106 spin_lock(&inode->i_lock);
3107 if (atomic_read(&inode->i_writecount) > 1) {
3108 spin_unlock(&inode->i_lock);
3111 atomic_set(&inode->i_writecount, -1);
3112 spin_unlock(&inode->i_lock);
3117 static void restore_bitmap_write_access(struct file *file)
3119 struct inode *inode = file->f_mapping->host;
3121 spin_lock(&inode->i_lock);
3122 atomic_set(&inode->i_writecount, 1);
3123 spin_unlock(&inode->i_lock);
3127 * 0 - completely stop and dis-assemble array
3128 * 1 - switch to readonly
3129 * 2 - stop but do not disassemble array
3131 static int do_md_stop(mddev_t * mddev, int mode)
3134 struct gendisk *disk = mddev->gendisk;
3137 if (atomic_read(&mddev->active)>2) {
3138 printk("md: %s still in use.\n",mdname(mddev));
3142 if (mddev->sync_thread) {
3143 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3144 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3145 md_unregister_thread(mddev->sync_thread);
3146 mddev->sync_thread = NULL;
3149 del_timer_sync(&mddev->safemode_timer);
3151 invalidate_partition(disk, 0);
3154 case 1: /* readonly */
3160 case 0: /* disassemble */
3162 bitmap_flush(mddev);
3163 md_super_wait(mddev);
3165 set_disk_ro(disk, 0);
3166 blk_queue_make_request(mddev->queue, md_fail_request);
3167 mddev->pers->stop(mddev);
3168 if (mddev->pers->sync_request)
3169 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3171 module_put(mddev->pers->owner);
3176 if (!mddev->in_sync || mddev->sb_dirty) {
3177 /* mark array as shutdown cleanly */
3179 md_update_sb(mddev);
3182 set_disk_ro(disk, 1);
3183 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3187 * Free resources if final stop
3191 struct list_head *tmp;
3192 struct gendisk *disk;
3193 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3195 bitmap_destroy(mddev);
3196 if (mddev->bitmap_file) {
3197 restore_bitmap_write_access(mddev->bitmap_file);
3198 fput(mddev->bitmap_file);
3199 mddev->bitmap_file = NULL;
3201 mddev->bitmap_offset = 0;
3203 ITERATE_RDEV(mddev,rdev,tmp)
3204 if (rdev->raid_disk >= 0) {
3206 sprintf(nm, "rd%d", rdev->raid_disk);
3207 sysfs_remove_link(&mddev->kobj, nm);
3210 export_array(mddev);
3212 mddev->array_size = 0;
3214 mddev->raid_disks = 0;
3216 disk = mddev->gendisk;
3218 set_capacity(disk, 0);
3220 } else if (mddev->pers)
3221 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3224 md_new_event(mddev);
3229 static void autorun_array(mddev_t *mddev)
3232 struct list_head *tmp;
3235 if (list_empty(&mddev->disks))
3238 printk(KERN_INFO "md: running: ");
3240 ITERATE_RDEV(mddev,rdev,tmp) {
3241 char b[BDEVNAME_SIZE];
3242 printk("<%s>", bdevname(rdev->bdev,b));
3246 err = do_md_run (mddev);
3248 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3249 do_md_stop (mddev, 0);
3254 * lets try to run arrays based on all disks that have arrived
3255 * until now. (those are in pending_raid_disks)
3257 * the method: pick the first pending disk, collect all disks with
3258 * the same UUID, remove all from the pending list and put them into
3259 * the 'same_array' list. Then order this list based on superblock
3260 * update time (freshest comes first), kick out 'old' disks and
3261 * compare superblocks. If everything's fine then run it.
3263 * If "unit" is allocated, then bump its reference count
3265 static void autorun_devices(int part)
3267 struct list_head *tmp;
3268 mdk_rdev_t *rdev0, *rdev;
3270 char b[BDEVNAME_SIZE];
3272 printk(KERN_INFO "md: autorun ...\n");
3273 while (!list_empty(&pending_raid_disks)) {
3275 LIST_HEAD(candidates);
3276 rdev0 = list_entry(pending_raid_disks.next,
3277 mdk_rdev_t, same_set);
3279 printk(KERN_INFO "md: considering %s ...\n",
3280 bdevname(rdev0->bdev,b));
3281 INIT_LIST_HEAD(&candidates);
3282 ITERATE_RDEV_PENDING(rdev,tmp)
3283 if (super_90_load(rdev, rdev0, 0) >= 0) {
3284 printk(KERN_INFO "md: adding %s ...\n",
3285 bdevname(rdev->bdev,b));
3286 list_move(&rdev->same_set, &candidates);
3289 * now we have a set of devices, with all of them having
3290 * mostly sane superblocks. It's time to allocate the
3293 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
3294 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3295 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3299 dev = MKDEV(mdp_major,
3300 rdev0->preferred_minor << MdpMinorShift);
3302 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3304 md_probe(dev, NULL, NULL);
3305 mddev = mddev_find(dev);
3308 "md: cannot allocate memory for md drive.\n");
3311 if (mddev_lock(mddev))
3312 printk(KERN_WARNING "md: %s locked, cannot run\n",
3314 else if (mddev->raid_disks || mddev->major_version
3315 || !list_empty(&mddev->disks)) {
3317 "md: %s already running, cannot run %s\n",
3318 mdname(mddev), bdevname(rdev0->bdev,b));
3319 mddev_unlock(mddev);
3321 printk(KERN_INFO "md: created %s\n", mdname(mddev));
3322 ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3323 list_del_init(&rdev->same_set);
3324 if (bind_rdev_to_array(rdev, mddev))
3327 autorun_array(mddev);
3328 mddev_unlock(mddev);
3330 /* on success, candidates will be empty, on error
3333 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3337 printk(KERN_INFO "md: ... autorun DONE.\n");
3341 * import RAID devices based on one partition
3342 * if possible, the array gets run as well.
3345 static int autostart_array(dev_t startdev)
3347 char b[BDEVNAME_SIZE];
3348 int err = -EINVAL, i;
3349 mdp_super_t *sb = NULL;
3350 mdk_rdev_t *start_rdev = NULL, *rdev;
3352 start_rdev = md_import_device(startdev, 0, 0);
3353 if (IS_ERR(start_rdev))
3357 /* NOTE: this can only work for 0.90.0 superblocks */
3358 sb = (mdp_super_t*)page_address(start_rdev->sb_page);
3359 if (sb->major_version != 0 ||
3360 sb->minor_version != 90 ) {
3361 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
3362 export_rdev(start_rdev);
3366 if (test_bit(Faulty, &start_rdev->flags)) {
3368 "md: can not autostart based on faulty %s!\n",
3369 bdevname(start_rdev->bdev,b));
3370 export_rdev(start_rdev);
3373 list_add(&start_rdev->same_set, &pending_raid_disks);
3375 for (i = 0; i < MD_SB_DISKS; i++) {
3376 mdp_disk_t *desc = sb->disks + i;
3377 dev_t dev = MKDEV(desc->major, desc->minor);
3381 if (dev == startdev)
3383 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
3385 rdev = md_import_device(dev, 0, 0);
3389 list_add(&rdev->same_set, &pending_raid_disks);
3393 * possibly return codes
3401 static int get_version(void __user * arg)
3405 ver.major = MD_MAJOR_VERSION;
3406 ver.minor = MD_MINOR_VERSION;
3407 ver.patchlevel = MD_PATCHLEVEL_VERSION;
3409 if (copy_to_user(arg, &ver, sizeof(ver)))
3415 static int get_array_info(mddev_t * mddev, void __user * arg)
3417 mdu_array_info_t info;
3418 int nr,working,active,failed,spare;
3420 struct list_head *tmp;
3422 nr=working=active=failed=spare=0;
3423 ITERATE_RDEV(mddev,rdev,tmp) {
3425 if (test_bit(Faulty, &rdev->flags))
3429 if (test_bit(In_sync, &rdev->flags))
3436 info.major_version = mddev->major_version;
3437 info.minor_version = mddev->minor_version;
3438 info.patch_version = MD_PATCHLEVEL_VERSION;
3439 info.ctime = mddev->ctime;
3440 info.level = mddev->level;
3441 info.size = mddev->size;
3442 if (info.size != mddev->size) /* overflow */
3445 info.raid_disks = mddev->raid_disks;
3446 info.md_minor = mddev->md_minor;
3447 info.not_persistent= !mddev->persistent;
3449 info.utime = mddev->utime;
3452 info.state = (1<<MD_SB_CLEAN);
3453 if (mddev->bitmap && mddev->bitmap_offset)
3454 info.state = (1<<MD_SB_BITMAP_PRESENT);
3455 info.active_disks = active;
3456 info.working_disks = working;
3457 info.failed_disks = failed;
3458 info.spare_disks = spare;
3460 info.layout = mddev->layout;
3461 info.chunk_size = mddev->chunk_size;
3463 if (copy_to_user(arg, &info, sizeof(info)))
3469 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3471 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3472 char *ptr, *buf = NULL;
3475 file = kmalloc(sizeof(*file), GFP_KERNEL);
3479 /* bitmap disabled, zero the first byte and copy out */
3480 if (!mddev->bitmap || !mddev->bitmap->file) {
3481 file->pathname[0] = '\0';
3485 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3489 ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3493 strcpy(file->pathname, ptr);
3497 if (copy_to_user(arg, file, sizeof(*file)))
3505 static int get_disk_info(mddev_t * mddev, void __user * arg)
3507 mdu_disk_info_t info;
3511 if (copy_from_user(&info, arg, sizeof(info)))
3516 rdev = find_rdev_nr(mddev, nr);
3518 info.major = MAJOR(rdev->bdev->bd_dev);
3519 info.minor = MINOR(rdev->bdev->bd_dev);
3520 info.raid_disk = rdev->raid_disk;
3522 if (test_bit(Faulty, &rdev->flags))
3523 info.state |= (1<<MD_DISK_FAULTY);
3524 else if (test_bit(In_sync, &rdev->flags)) {
3525 info.state |= (1<<MD_DISK_ACTIVE);
3526 info.state |= (1<<MD_DISK_SYNC);
3528 if (test_bit(WriteMostly, &rdev->flags))
3529 info.state |= (1<<MD_DISK_WRITEMOSTLY);
3531 info.major = info.minor = 0;
3532 info.raid_disk = -1;
3533 info.state = (1<<MD_DISK_REMOVED);
3536 if (copy_to_user(arg, &info, sizeof(info)))
3542 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3544 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3546 dev_t dev = MKDEV(info->major,info->minor);
3548 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3551 if (!mddev->raid_disks) {
3553 /* expecting a device which has a superblock */
3554 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3557 "md: md_import_device returned %ld\n",
3559 return PTR_ERR(rdev);
3561 if (!list_empty(&mddev->disks)) {
3562 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3563 mdk_rdev_t, same_set);
3564 int err = super_types[mddev->major_version]
3565 .load_super(rdev, rdev0, mddev->minor_version);
3568 "md: %s has different UUID to %s\n",
3569 bdevname(rdev->bdev,b),
3570 bdevname(rdev0->bdev,b2));
3575 err = bind_rdev_to_array(rdev, mddev);
3582 * add_new_disk can be used once the array is assembled
3583 * to add "hot spares". They must already have a superblock
3588 if (!mddev->pers->hot_add_disk) {
3590 "%s: personality does not support diskops!\n",
3594 if (mddev->persistent)
3595 rdev = md_import_device(dev, mddev->major_version,
3596 mddev->minor_version);
3598 rdev = md_import_device(dev, -1, -1);
3601 "md: md_import_device returned %ld\n",
3603 return PTR_ERR(rdev);
3605 /* set save_raid_disk if appropriate */
3606 if (!mddev->persistent) {
3607 if (info->state & (1<<MD_DISK_SYNC) &&
3608 info->raid_disk < mddev->raid_disks)
3609 rdev->raid_disk = info->raid_disk;
3611 rdev->raid_disk = -1;
3613 super_types[mddev->major_version].
3614 validate_super(mddev, rdev);
3615 rdev->saved_raid_disk = rdev->raid_disk;
3617 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3618 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3619 set_bit(WriteMostly, &rdev->flags);
3621 rdev->raid_disk = -1;
3622 err = bind_rdev_to_array(rdev, mddev);
3623 if (!err && !mddev->pers->hot_remove_disk) {
3624 /* If there is hot_add_disk but no hot_remove_disk
3625 * then added disks for geometry changes,
3626 * and should be added immediately.
3628 super_types[mddev->major_version].
3629 validate_super(mddev, rdev);
3630 err = mddev->pers->hot_add_disk(mddev, rdev);
3632 unbind_rdev_from_array(rdev);
3637 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3638 md_wakeup_thread(mddev->thread);
3642 /* otherwise, add_new_disk is only allowed
3643 * for major_version==0 superblocks
3645 if (mddev->major_version != 0) {
3646 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3651 if (!(info->state & (1<<MD_DISK_FAULTY))) {
3653 rdev = md_import_device (dev, -1, 0);
3656 "md: error, md_import_device() returned %ld\n",
3658 return PTR_ERR(rdev);
3660 rdev->desc_nr = info->number;
3661 if (info->raid_disk < mddev->raid_disks)
3662 rdev->raid_disk = info->raid_disk;
3664 rdev->raid_disk = -1;
3668 if (rdev->raid_disk < mddev->raid_disks)
3669 if (info->state & (1<<MD_DISK_SYNC))
3670 set_bit(In_sync, &rdev->flags);
3672 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3673 set_bit(WriteMostly, &rdev->flags);
3675 if (!mddev->persistent) {
3676 printk(KERN_INFO "md: nonpersistent superblock ...\n");
3677 rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3679 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3680 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3682 err = bind_rdev_to_array(rdev, mddev);
3692 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3694 char b[BDEVNAME_SIZE];
3700 rdev = find_rdev(mddev, dev);
3704 if (rdev->raid_disk >= 0)
3707 kick_rdev_from_array(rdev);
3708 md_update_sb(mddev);
3709 md_new_event(mddev);
3713 printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3714 bdevname(rdev->bdev,b), mdname(mddev));
3718 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3720 char b[BDEVNAME_SIZE];
3728 if (mddev->major_version != 0) {
3729 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3730 " version-0 superblocks.\n",
3734 if (!mddev->pers->hot_add_disk) {
3736 "%s: personality does not support diskops!\n",
3741 rdev = md_import_device (dev, -1, 0);
3744 "md: error, md_import_device() returned %ld\n",
3749 if (mddev->persistent)
3750 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3753 rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3755 size = calc_dev_size(rdev, mddev->chunk_size);
3758 if (test_bit(Faulty, &rdev->flags)) {
3760 "md: can not hot-add faulty %s disk to %s!\n",
3761 bdevname(rdev->bdev,b), mdname(mddev));
3765 clear_bit(In_sync, &rdev->flags);
3767 err = bind_rdev_to_array(rdev, mddev);
3772 * The rest should better be atomic, we can have disk failures
3773 * noticed in interrupt contexts ...
3776 if (rdev->desc_nr == mddev->max_disks) {
3777 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3780 goto abort_unbind_export;
3783 rdev->raid_disk = -1;
3785 md_update_sb(mddev);
3788 * Kick recovery, maybe this spare has to be added to the
3789 * array immediately.
3791 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3792 md_wakeup_thread(mddev->thread);
3793 md_new_event(mddev);
3796 abort_unbind_export:
3797 unbind_rdev_from_array(rdev);
3804 static int set_bitmap_file(mddev_t *mddev, int fd)
3809 if (!mddev->pers->quiesce)
3811 if (mddev->recovery || mddev->sync_thread)
3813 /* we should be able to change the bitmap.. */
3819 return -EEXIST; /* cannot add when bitmap is present */
3820 mddev->bitmap_file = fget(fd);
3822 if (mddev->bitmap_file == NULL) {
3823 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3828 err = deny_bitmap_write_access(mddev->bitmap_file);
3830 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3832 fput(mddev->bitmap_file);
3833 mddev->bitmap_file = NULL;
3836 mddev->bitmap_offset = 0; /* file overrides offset */
3837 } else if (mddev->bitmap == NULL)
3838 return -ENOENT; /* cannot remove what isn't there */
3841 mddev->pers->quiesce(mddev, 1);
3843 err = bitmap_create(mddev);
3844 if (fd < 0 || err) {
3845 bitmap_destroy(mddev);
3846 fd = -1; /* make sure to put the file */
3848 mddev->pers->quiesce(mddev, 0);
3851 if (mddev->bitmap_file) {
3852 restore_bitmap_write_access(mddev->bitmap_file);
3853 fput(mddev->bitmap_file);
3855 mddev->bitmap_file = NULL;
3862 * set_array_info is used two different ways
3863 * The original usage is when creating a new array.
3864 * In this usage, raid_disks is > 0 and it together with
3865 * level, size, not_persistent,layout,chunksize determine the
3866 * shape of the array.
3867 * This will always create an array with a type-0.90.0 superblock.
3868 * The newer usage is when assembling an array.
3869 * In this case raid_disks will be 0, and the major_version field is
3870 * use to determine which style super-blocks are to be found on the devices.
3871 * The minor and patch _version numbers are also kept incase the
3872 * super_block handler wishes to interpret them.
3874 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3877 if (info->raid_disks == 0) {
3878 /* just setting version number for superblock loading */
3879 if (info->major_version < 0 ||
3880 info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3881 super_types[info->major_version].name == NULL) {
3882 /* maybe try to auto-load a module? */
3884 "md: superblock version %d not known\n",
3885 info->major_version);
3888 mddev->major_version = info->major_version;
3889 mddev->minor_version = info->minor_version;
3890 mddev->patch_version = info->patch_version;
3893 mddev->major_version = MD_MAJOR_VERSION;
3894 mddev->minor_version = MD_MINOR_VERSION;
3895 mddev->patch_version = MD_PATCHLEVEL_VERSION;
3896 mddev->ctime = get_seconds();
3898 mddev->level = info->level;
3899 mddev->clevel[0] = 0;
3900 mddev->size = info->size;
3901 mddev->raid_disks = info->raid_disks;
3902 /* don't set md_minor, it is determined by which /dev/md* was
3905 if (info->state & (1<<MD_SB_CLEAN))
3906 mddev->recovery_cp = MaxSector;
3908 mddev->recovery_cp = 0;
3909 mddev->persistent = ! info->not_persistent;
3911 mddev->layout = info->layout;
3912 mddev->chunk_size = info->chunk_size;
3914 mddev->max_disks = MD_SB_DISKS;
3916 mddev->sb_dirty = 1;
3918 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3919 mddev->bitmap_offset = 0;
3921 mddev->reshape_position = MaxSector;
3924 * Generate a 128 bit UUID
3926 get_random_bytes(mddev->uuid, 16);
3928 mddev->new_level = mddev->level;
3929 mddev->new_chunk = mddev->chunk_size;
3930 mddev->new_layout = mddev->layout;
3931 mddev->delta_disks = 0;
3936 static int update_size(mddev_t *mddev, unsigned long size)
3940 struct list_head *tmp;
3941 int fit = (size == 0);
3943 if (mddev->pers->resize == NULL)
3945 /* The "size" is the amount of each device that is used.
3946 * This can only make sense for arrays with redundancy.
3947 * linear and raid0 always use whatever space is available
3948 * We can only consider changing the size if no resync
3949 * or reconstruction is happening, and if the new size
3950 * is acceptable. It must fit before the sb_offset or,
3951 * if that is <data_offset, it must fit before the
3952 * size of each device.
3953 * If size is zero, we find the largest size that fits.
3955 if (mddev->sync_thread)
3957 ITERATE_RDEV(mddev,rdev,tmp) {
3959 if (rdev->sb_offset > rdev->data_offset)
3960 avail = (rdev->sb_offset*2) - rdev->data_offset;
3962 avail = get_capacity(rdev->bdev->bd_disk)
3963 - rdev->data_offset;
3964 if (fit && (size == 0 || size > avail/2))
3966 if (avail < ((sector_t)size << 1))
3969 rv = mddev->pers->resize(mddev, (sector_t)size *2);
3971 struct block_device *bdev;
3973 bdev = bdget_disk(mddev->gendisk, 0);
3975 mutex_lock(&bdev->bd_inode->i_mutex);
3976 i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
3977 mutex_unlock(&bdev->bd_inode->i_mutex);
3984 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3987 /* change the number of raid disks */
3988 if (mddev->pers->check_reshape == NULL)
3990 if (raid_disks <= 0 ||
3991 raid_disks >= mddev->max_disks)
3993 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
3995 mddev->delta_disks = raid_disks - mddev->raid_disks;
3997 rv = mddev->pers->check_reshape(mddev);
4003 * update_array_info is used to change the configuration of an
4005 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4006 * fields in the info are checked against the array.
4007 * Any differences that cannot be handled will cause an error.
4008 * Normally, only one change can be managed at a time.
4010 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4016 /* calculate expected state,ignoring low bits */
4017 if (mddev->bitmap && mddev->bitmap_offset)
4018 state |= (1 << MD_SB_BITMAP_PRESENT);
4020 if (mddev->major_version != info->major_version ||
4021 mddev->minor_version != info->minor_version ||
4022 /* mddev->patch_version != info->patch_version || */
4023 mddev->ctime != info->ctime ||
4024 mddev->level != info->level ||
4025 /* mddev->layout != info->layout || */
4026 !mddev->persistent != info->not_persistent||
4027 mddev->chunk_size != info->chunk_size ||
4028 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4029 ((state^info->state) & 0xfffffe00)
4032 /* Check there is only one change */
4033 if (info->size >= 0 && mddev->size != info->size) cnt++;
4034 if (mddev->raid_disks != info->raid_disks) cnt++;
4035 if (mddev->layout != info->layout) cnt++;
4036 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4037 if (cnt == 0) return 0;
4038 if (cnt > 1) return -EINVAL;
4040 if (mddev->layout != info->layout) {
4042 * we don't need to do anything at the md level, the
4043 * personality will take care of it all.
4045 if (mddev->pers->reconfig == NULL)
4048 return mddev->pers->reconfig(mddev, info->layout, -1);
4050 if (info->size >= 0 && mddev->size != info->size)
4051 rv = update_size(mddev, info->size);
4053 if (mddev->raid_disks != info->raid_disks)
4054 rv = update_raid_disks(mddev, info->raid_disks);
4056 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4057 if (mddev->pers->quiesce == NULL)
4059 if (mddev->recovery || mddev->sync_thread)
4061 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4062 /* add the bitmap */
4065 if (mddev->default_bitmap_offset == 0)
4067 mddev->bitmap_offset = mddev->default_bitmap_offset;
4068 mddev->pers->quiesce(mddev, 1);
4069 rv = bitmap_create(mddev);
4071 bitmap_destroy(mddev);
4072 mddev->pers->quiesce(mddev, 0);
4074 /* remove the bitmap */
4077 if (mddev->bitmap->file)
4079 mddev->pers->quiesce(mddev, 1);
4080 bitmap_destroy(mddev);
4081 mddev->pers->quiesce(mddev, 0);
4082 mddev->bitmap_offset = 0;
4085 md_update_sb(mddev);
4089 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4093 if (mddev->pers == NULL)
4096 rdev = find_rdev(mddev, dev);
4100 md_error(mddev, rdev);
4104 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4106 mddev_t *mddev = bdev->bd_disk->private_data;
4110 geo->cylinders = get_capacity(mddev->gendisk) / 8;
4114 static int md_ioctl(struct inode *inode, struct file *file,
4115 unsigned int cmd, unsigned long arg)
4118 void __user *argp = (void __user *)arg;
4119 mddev_t *mddev = NULL;
4121 if (!capable(CAP_SYS_ADMIN))
4125 * Commands dealing with the RAID driver but not any
4131 err = get_version(argp);
4134 case PRINT_RAID_DEBUG:
4142 autostart_arrays(arg);
4149 * Commands creating/starting a new array:
4152 mddev = inode->i_bdev->bd_disk->private_data;
4160 if (cmd == START_ARRAY) {
4161 /* START_ARRAY doesn't need to lock the array as autostart_array
4162 * does the locking, and it could even be a different array
4167 "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
4168 "This will not be supported beyond July 2006\n",
4169 current->comm, current->pid);
4172 err = autostart_array(new_decode_dev(arg));
4174 printk(KERN_WARNING "md: autostart failed!\n");
4180 err = mddev_lock(mddev);
4183 "md: ioctl lock interrupted, reason %d, cmd %d\n",
4190 case SET_ARRAY_INFO:
4192 mdu_array_info_t info;
4194 memset(&info, 0, sizeof(info));
4195 else if (copy_from_user(&info, argp, sizeof(info))) {
4200 err = update_array_info(mddev, &info);
4202 printk(KERN_WARNING "md: couldn't update"
4203 " array info. %d\n", err);
4208 if (!list_empty(&mddev->disks)) {
4210 "md: array %s already has disks!\n",
4215 if (mddev->raid_disks) {
4217 "md: array %s already initialised!\n",
4222 err = set_array_info(mddev, &info);
4224 printk(KERN_WARNING "md: couldn't set"
4225 " array info. %d\n", err);
4235 * Commands querying/configuring an existing array:
4237 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4238 * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
4239 if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4240 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
4246 * Commands even a read-only array can execute:
4250 case GET_ARRAY_INFO:
4251 err = get_array_info(mddev, argp);
4254 case GET_BITMAP_FILE:
4255 err = get_bitmap_file(mddev, argp);
4259 err = get_disk_info(mddev, argp);
4262 case RESTART_ARRAY_RW:
4263 err = restart_array(mddev);
4267 err = do_md_stop (mddev, 0);
4271 err = do_md_stop (mddev, 1);
4275 * We have a problem here : there is no easy way to give a CHS
4276 * virtual geometry. We currently pretend that we have a 2 heads
4277 * 4 sectors (with a BIG number of cylinders...). This drives
4278 * dosfs just mad... ;-)
4283 * The remaining ioctls are changing the state of the
4284 * superblock, so we do not allow them on read-only arrays.
4285 * However non-MD ioctls (e.g. get-size) will still come through
4286 * here and hit the 'default' below, so only disallow
4287 * 'md' ioctls, and switch to rw mode if started auto-readonly.
4289 if (_IOC_TYPE(cmd) == MD_MAJOR &&
4290 mddev->ro && mddev->pers) {
4291 if (mddev->ro == 2) {
4293 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4294 md_wakeup_thread(mddev->thread);
4306 mdu_disk_info_t info;
4307 if (copy_from_user(&info, argp, sizeof(info)))
4310 err = add_new_disk(mddev, &info);
4314 case HOT_REMOVE_DISK:
4315 err = hot_remove_disk(mddev, new_decode_dev(arg));
4319 err = hot_add_disk(mddev, new_decode_dev(arg));
4322 case SET_DISK_FAULTY:
4323 err = set_disk_faulty(mddev, new_decode_dev(arg));
4327 err = do_md_run (mddev);
4330 case SET_BITMAP_FILE:
4331 err = set_bitmap_file(mddev, (int)arg);
4341 mddev_unlock(mddev);
4351 static int md_open(struct inode *inode, struct file *file)
4354 * Succeed if we can lock the mddev, which confirms that
4355 * it isn't being stopped right now.
4357 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4360 if ((err = mddev_lock(mddev)))
4365 mddev_unlock(mddev);
4367 check_disk_change(inode->i_bdev);
4372 static int md_release(struct inode *inode, struct file * file)
4374 mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4383 static int md_media_changed(struct gendisk *disk)
4385 mddev_t *mddev = disk->private_data;
4387 return mddev->changed;
4390 static int md_revalidate(struct gendisk *disk)
4392 mddev_t *mddev = disk->private_data;
4397 static struct block_device_operations md_fops =
4399 .owner = THIS_MODULE,
4401 .release = md_release,
4403 .getgeo = md_getgeo,
4404 .media_changed = md_media_changed,
4405 .revalidate_disk= md_revalidate,
4408 static int md_thread(void * arg)
4410 mdk_thread_t *thread = arg;
4413 * md_thread is a 'system-thread', it's priority should be very
4414 * high. We avoid resource deadlocks individually in each
4415 * raid personality. (RAID5 does preallocation) We also use RR and
4416 * the very same RT priority as kswapd, thus we will never get
4417 * into a priority inversion deadlock.
4419 * we definitely have to have equal or higher priority than
4420 * bdflush, otherwise bdflush will deadlock if there are too
4421 * many dirty RAID5 blocks.
4424 allow_signal(SIGKILL);
4425 while (!kthread_should_stop()) {
4427 /* We need to wait INTERRUPTIBLE so that
4428 * we don't add to the load-average.
4429 * That means we need to be sure no signals are
4432 if (signal_pending(current))
4433 flush_signals(current);
4435 wait_event_interruptible_timeout
4437 test_bit(THREAD_WAKEUP, &thread->flags)
4438 || kthread_should_stop(),
4442 clear_bit(THREAD_WAKEUP, &thread->flags);
4444 thread->run(thread->mddev);
4450 void md_wakeup_thread(mdk_thread_t *thread)
4453 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4454 set_bit(THREAD_WAKEUP, &thread->flags);
4455 wake_up(&thread->wqueue);
4459 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4462 mdk_thread_t *thread;
4464 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4468 init_waitqueue_head(&thread->wqueue);
4471 thread->mddev = mddev;
4472 thread->timeout = MAX_SCHEDULE_TIMEOUT;
4473 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4474 if (IS_ERR(thread->tsk)) {
4481 void md_unregister_thread(mdk_thread_t *thread)
4483 dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4485 kthread_stop(thread->tsk);
4489 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4496 if (!rdev || test_bit(Faulty, &rdev->flags))
4499 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4501 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4502 __builtin_return_address(0),__builtin_return_address(1),
4503 __builtin_return_address(2),__builtin_return_address(3));
4505 if (!mddev->pers->error_handler)
4507 mddev->pers->error_handler(mddev,rdev);
4508 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4509 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4510 md_wakeup_thread(mddev->thread);
4511 md_new_event_inintr(mddev);
4514 /* seq_file implementation /proc/mdstat */
4516 static void status_unused(struct seq_file *seq)
4520 struct list_head *tmp;
4522 seq_printf(seq, "unused devices: ");
4524 ITERATE_RDEV_PENDING(rdev,tmp) {
4525 char b[BDEVNAME_SIZE];
4527 seq_printf(seq, "%s ",
4528 bdevname(rdev->bdev,b));
4531 seq_printf(seq, "<none>");
4533 seq_printf(seq, "\n");
4537 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4539 sector_t max_blocks, resync, res;
4540 unsigned long dt, db, rt;
4542 unsigned int per_milli;
4544 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4546 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4547 max_blocks = mddev->resync_max_sectors >> 1;
4549 max_blocks = mddev->size;
4552 * Should not happen.
4558 /* Pick 'scale' such that (resync>>scale)*1000 will fit
4559 * in a sector_t, and (max_blocks>>scale) will fit in a
4560 * u32, as those are the requirements for sector_div.
4561 * Thus 'scale' must be at least 10
4564 if (sizeof(sector_t) > sizeof(unsigned long)) {
4565 while ( max_blocks/2 > (1ULL<<(scale+32)))
4568 res = (resync>>scale)*1000;
4569 sector_div(res, (u32)((max_blocks>>scale)+1));
4573 int i, x = per_milli/50, y = 20-x;
4574 seq_printf(seq, "[");
4575 for (i = 0; i < x; i++)
4576 seq_printf(seq, "=");
4577 seq_printf(seq, ">");
4578 for (i = 0; i < y; i++)
4579 seq_printf(seq, ".");
4580 seq_printf(seq, "] ");
4582 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4583 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4585 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4586 "resync" : "recovery")),
4587 per_milli/10, per_milli % 10,
4588 (unsigned long long) resync,
4589 (unsigned long long) max_blocks);
4592 * We do not want to overflow, so the order of operands and
4593 * the * 100 / 100 trick are important. We do a +1 to be
4594 * safe against division by zero. We only estimate anyway.
4596 * dt: time from mark until now
4597 * db: blocks written from mark until now
4598 * rt: remaining time
4600 dt = ((jiffies - mddev->resync_mark) / HZ);
4602 db = resync - (mddev->resync_mark_cnt/2);
4603 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100;
4605 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4607 seq_printf(seq, " speed=%ldK/sec", db/dt);
4610 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4612 struct list_head *tmp;
4622 spin_lock(&all_mddevs_lock);
4623 list_for_each(tmp,&all_mddevs)
4625 mddev = list_entry(tmp, mddev_t, all_mddevs);
4627 spin_unlock(&all_mddevs_lock);
4630 spin_unlock(&all_mddevs_lock);
4632 return (void*)2;/* tail */
4636 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4638 struct list_head *tmp;
4639 mddev_t *next_mddev, *mddev = v;
4645 spin_lock(&all_mddevs_lock);
4647 tmp = all_mddevs.next;
4649 tmp = mddev->all_mddevs.next;
4650 if (tmp != &all_mddevs)
4651 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4653 next_mddev = (void*)2;
4656 spin_unlock(&all_mddevs_lock);
4664 static void md_seq_stop(struct seq_file *seq, void *v)
4668 if (mddev && v != (void*)1 && v != (void*)2)
4672 struct mdstat_info {
4676 static int md_seq_show(struct seq_file *seq, void *v)
4680 struct list_head *tmp2;
4682 struct mdstat_info *mi = seq->private;
4683 struct bitmap *bitmap;
4685 if (v == (void*)1) {
4686 struct mdk_personality *pers;
4687 seq_printf(seq, "Personalities : ");
4688 spin_lock(&pers_lock);
4689 list_for_each_entry(pers, &pers_list, list)
4690 seq_printf(seq, "[%s] ", pers->name);
4692 spin_unlock(&pers_lock);
4693 seq_printf(seq, "\n");
4694 mi->event = atomic_read(&md_event_count);
4697 if (v == (void*)2) {
4702 if (mddev_lock(mddev) < 0)
4705 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4706 seq_printf(seq, "%s : %sactive", mdname(mddev),
4707 mddev->pers ? "" : "in");
4710 seq_printf(seq, " (read-only)");
4712 seq_printf(seq, "(auto-read-only)");
4713 seq_printf(seq, " %s", mddev->pers->name);
4717 ITERATE_RDEV(mddev,rdev,tmp2) {
4718 char b[BDEVNAME_SIZE];
4719 seq_printf(seq, " %s[%d]",
4720 bdevname(rdev->bdev,b), rdev->desc_nr);
4721 if (test_bit(WriteMostly, &rdev->flags))
4722 seq_printf(seq, "(W)");
4723 if (test_bit(Faulty, &rdev->flags)) {
4724 seq_printf(seq, "(F)");
4726 } else if (rdev->raid_disk < 0)
4727 seq_printf(seq, "(S)"); /* spare */
4731 if (!list_empty(&mddev->disks)) {
4733 seq_printf(seq, "\n %llu blocks",
4734 (unsigned long long)mddev->array_size);
4736 seq_printf(seq, "\n %llu blocks",
4737 (unsigned long long)size);
4739 if (mddev->persistent) {
4740 if (mddev->major_version != 0 ||
4741 mddev->minor_version != 90) {
4742 seq_printf(seq," super %d.%d",
4743 mddev->major_version,
4744 mddev->minor_version);
4747 seq_printf(seq, " super non-persistent");
4750 mddev->pers->status (seq, mddev);
4751 seq_printf(seq, "\n ");
4752 if (mddev->pers->sync_request) {
4753 if (mddev->curr_resync > 2) {
4754 status_resync (seq, mddev);
4755 seq_printf(seq, "\n ");
4756 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4757 seq_printf(seq, "\tresync=DELAYED\n ");
4758 else if (mddev->recovery_cp < MaxSector)
4759 seq_printf(seq, "\tresync=PENDING\n ");
4762 seq_printf(seq, "\n ");
4764 if ((bitmap = mddev->bitmap)) {
4765 unsigned long chunk_kb;
4766 unsigned long flags;
4767 spin_lock_irqsave(&bitmap->lock, flags);
4768 chunk_kb = bitmap->chunksize >> 10;
4769 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4771 bitmap->pages - bitmap->missing_pages,
4773 (bitmap->pages - bitmap->missing_pages)
4774 << (PAGE_SHIFT - 10),
4775 chunk_kb ? chunk_kb : bitmap->chunksize,
4776 chunk_kb ? "KB" : "B");
4778 seq_printf(seq, ", file: ");
4779 seq_path(seq, bitmap->file->f_vfsmnt,
4780 bitmap->file->f_dentry," \t\n");
4783 seq_printf(seq, "\n");
4784 spin_unlock_irqrestore(&bitmap->lock, flags);
4787 seq_printf(seq, "\n");
4789 mddev_unlock(mddev);
4794 static struct seq_operations md_seq_ops = {
4795 .start = md_seq_start,
4796 .next = md_seq_next,
4797 .stop = md_seq_stop,
4798 .show = md_seq_show,
4801 static int md_seq_open(struct inode *inode, struct file *file)
4804 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4808 error = seq_open(file, &md_seq_ops);
4812 struct seq_file *p = file->private_data;
4814 mi->event = atomic_read(&md_event_count);
4819 static int md_seq_release(struct inode *inode, struct file *file)
4821 struct seq_file *m = file->private_data;
4822 struct mdstat_info *mi = m->private;
4825 return seq_release(inode, file);
4828 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4830 struct seq_file *m = filp->private_data;
4831 struct mdstat_info *mi = m->private;
4834 poll_wait(filp, &md_event_waiters, wait);
4836 /* always allow read */
4837 mask = POLLIN | POLLRDNORM;
4839 if (mi->event != atomic_read(&md_event_count))
4840 mask |= POLLERR | POLLPRI;
4844 static struct file_operations md_seq_fops = {
4845 .open = md_seq_open,
4847 .llseek = seq_lseek,
4848 .release = md_seq_release,
4849 .poll = mdstat_poll,
4852 int register_md_personality(struct mdk_personality *p)
4854 spin_lock(&pers_lock);
4855 list_add_tail(&p->list, &pers_list);
4856 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4857 spin_unlock(&pers_lock);
4861 int unregister_md_personality(struct mdk_personality *p)
4863 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4864 spin_lock(&pers_lock);
4865 list_del_init(&p->list);
4866 spin_unlock(&pers_lock);
4870 static int is_mddev_idle(mddev_t *mddev)
4873 struct list_head *tmp;
4875 unsigned long curr_events;
4878 ITERATE_RDEV(mddev,rdev,tmp) {
4879 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4880 curr_events = disk_stat_read(disk, sectors[0]) +
4881 disk_stat_read(disk, sectors[1]) -
4882 atomic_read(&disk->sync_io);
4883 /* The difference between curr_events and last_events
4884 * will be affected by any new non-sync IO (making
4885 * curr_events bigger) and any difference in the amount of
4886 * in-flight syncio (making current_events bigger or smaller)
4887 * The amount in-flight is currently limited to
4888 * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4889 * which is at most 4096 sectors.
4890 * These numbers are fairly fragile and should be made
4891 * more robust, probably by enforcing the
4892 * 'window size' that md_do_sync sort-of uses.
4894 * Note: the following is an unsigned comparison.
4896 if ((curr_events - rdev->last_events + 4096) > 8192) {
4897 rdev->last_events = curr_events;
4904 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4906 /* another "blocks" (512byte) blocks have been synced */
4907 atomic_sub(blocks, &mddev->recovery_active);
4908 wake_up(&mddev->recovery_wait);
4910 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4911 md_wakeup_thread(mddev->thread);
4912 // stop recovery, signal do_sync ....
4917 /* md_write_start(mddev, bi)
4918 * If we need to update some array metadata (e.g. 'active' flag
4919 * in superblock) before writing, schedule a superblock update
4920 * and wait for it to complete.
4922 void md_write_start(mddev_t *mddev, struct bio *bi)
4924 if (bio_data_dir(bi) != WRITE)
4927 BUG_ON(mddev->ro == 1);
4928 if (mddev->ro == 2) {
4929 /* need to switch to read/write */
4931 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4932 md_wakeup_thread(mddev->thread);
4934 atomic_inc(&mddev->writes_pending);
4935 if (mddev->in_sync) {
4936 spin_lock_irq(&mddev->write_lock);
4937 if (mddev->in_sync) {
4939 mddev->sb_dirty = 3;
4940 md_wakeup_thread(mddev->thread);
4942 spin_unlock_irq(&mddev->write_lock);
4944 wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4947 void md_write_end(mddev_t *mddev)
4949 if (atomic_dec_and_test(&mddev->writes_pending)) {
4950 if (mddev->safemode == 2)
4951 md_wakeup_thread(mddev->thread);
4952 else if (mddev->safemode_delay)
4953 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4957 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4959 #define SYNC_MARKS 10
4960 #define SYNC_MARK_STEP (3*HZ)
4961 void md_do_sync(mddev_t *mddev)
4964 unsigned int currspeed = 0,
4966 sector_t max_sectors,j, io_sectors;
4967 unsigned long mark[SYNC_MARKS];
4968 sector_t mark_cnt[SYNC_MARKS];
4970 struct list_head *tmp;
4971 sector_t last_check;
4973 struct list_head *rtmp;
4976 /* just incase thread restarts... */
4977 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4979 if (mddev->ro) /* never try to sync a read-only array */
4982 /* we overload curr_resync somewhat here.
4983 * 0 == not engaged in resync at all
4984 * 2 == checking that there is no conflict with another sync
4985 * 1 == like 2, but have yielded to allow conflicting resync to
4987 * other == active in resync - this many blocks
4989 * Before starting a resync we must have set curr_resync to
4990 * 2, and then checked that every "conflicting" array has curr_resync
4991 * less than ours. When we find one that is the same or higher
4992 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
4993 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4994 * This will mean we have to start checking from the beginning again.
4999 mddev->curr_resync = 2;
5002 if (kthread_should_stop()) {
5003 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5006 ITERATE_MDDEV(mddev2,tmp) {
5007 if (mddev2 == mddev)
5009 if (mddev2->curr_resync &&
5010 match_mddev_units(mddev,mddev2)) {
5012 if (mddev < mddev2 && mddev->curr_resync == 2) {
5013 /* arbitrarily yield */
5014 mddev->curr_resync = 1;
5015 wake_up(&resync_wait);
5017 if (mddev > mddev2 && mddev->curr_resync == 1)
5018 /* no need to wait here, we can wait the next
5019 * time 'round when curr_resync == 2
5022 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5023 if (!kthread_should_stop() &&
5024 mddev2->curr_resync >= mddev->curr_resync) {
5025 printk(KERN_INFO "md: delaying resync of %s"
5026 " until %s has finished resync (they"
5027 " share one or more physical units)\n",
5028 mdname(mddev), mdname(mddev2));
5031 finish_wait(&resync_wait, &wq);
5034 finish_wait(&resync_wait, &wq);
5037 } while (mddev->curr_resync < 2);
5040 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5041 /* resync follows the size requested by the personality,
5042 * which defaults to physical size, but can be virtual size
5044 max_sectors = mddev->resync_max_sectors;
5045 mddev->resync_mismatches = 0;
5046 /* we don't use the checkpoint if there's a bitmap */
5047 if (!mddev->bitmap &&
5048 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5049 j = mddev->recovery_cp;
5050 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5051 max_sectors = mddev->size << 1;
5053 /* recovery follows the physical size of devices */
5054 max_sectors = mddev->size << 1;
5056 ITERATE_RDEV(mddev,rdev,rtmp)
5057 if (rdev->raid_disk >= 0 &&
5058 !test_bit(Faulty, &rdev->flags) &&
5059 !test_bit(In_sync, &rdev->flags) &&
5060 rdev->recovery_offset < j)
5061 j = rdev->recovery_offset;
5064 printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
5065 printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
5066 " %d KB/sec/disc.\n", speed_min(mddev));
5067 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5068 "(but not more than %d KB/sec) for reconstruction.\n",
5071 is_mddev_idle(mddev); /* this also initializes IO event counters */
5074 for (m = 0; m < SYNC_MARKS; m++) {
5076 mark_cnt[m] = io_sectors;
5079 mddev->resync_mark = mark[last_mark];
5080 mddev->resync_mark_cnt = mark_cnt[last_mark];
5083 * Tune reconstruction:
5085 window = 32*(PAGE_SIZE/512);
5086 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5087 window/2,(unsigned long long) max_sectors/2);
5089 atomic_set(&mddev->recovery_active, 0);
5090 init_waitqueue_head(&mddev->recovery_wait);
5095 "md: resuming recovery of %s from checkpoint.\n",
5097 mddev->curr_resync = j;
5100 while (j < max_sectors) {
5104 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5105 currspeed < speed_min(mddev));
5107 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5111 if (!skipped) { /* actual IO requested */
5112 io_sectors += sectors;
5113 atomic_add(sectors, &mddev->recovery_active);
5117 if (j>1) mddev->curr_resync = j;
5118 if (last_check == 0)
5119 /* this is the earliers that rebuilt will be
5120 * visible in /proc/mdstat
5122 md_new_event(mddev);
5124 if (last_check + window > io_sectors || j == max_sectors)
5127 last_check = io_sectors;
5129 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5130 test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5134 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5136 int next = (last_mark+1) % SYNC_MARKS;
5138 mddev->resync_mark = mark[next];
5139 mddev->resync_mark_cnt = mark_cnt[next];
5140 mark[next] = jiffies;
5141 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5146 if (kthread_should_stop()) {
5148 * got a signal, exit.
5151 "md: md_do_sync() got signal ... exiting\n");
5152 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5157 * this loop exits only if either when we are slower than
5158 * the 'hard' speed limit, or the system was IO-idle for
5160 * the system might be non-idle CPU-wise, but we only care
5161 * about not overloading the IO subsystem. (things like an
5162 * e2fsck being done on the RAID array should execute fast)
5164 mddev->queue->unplug_fn(mddev->queue);
5167 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5168 /((jiffies-mddev->resync_mark)/HZ +1) +1;
5170 if (currspeed > speed_min(mddev)) {
5171 if ((currspeed > speed_max(mddev)) ||
5172 !is_mddev_idle(mddev)) {
5178 printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
5180 * this also signals 'finished resyncing' to md_stop
5183 mddev->queue->unplug_fn(mddev->queue);
5185 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5187 /* tell personality that we are finished */
5188 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5190 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5191 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
5192 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5193 mddev->curr_resync > 2) {
5194 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5195 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5196 if (mddev->curr_resync >= mddev->recovery_cp) {
5198 "md: checkpointing recovery of %s.\n",
5200 mddev->recovery_cp = mddev->curr_resync;
5203 mddev->recovery_cp = MaxSector;
5205 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5206 mddev->curr_resync = MaxSector;
5207 ITERATE_RDEV(mddev,rdev,rtmp)
5208 if (rdev->raid_disk >= 0 &&
5209 !test_bit(Faulty, &rdev->flags) &&
5210 !test_bit(In_sync, &rdev->flags) &&
5211 rdev->recovery_offset < mddev->curr_resync)
5212 rdev->recovery_offset = mddev->curr_resync;
5213 mddev->sb_dirty = 1;
5218 mddev->curr_resync = 0;
5219 wake_up(&resync_wait);
5220 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5221 md_wakeup_thread(mddev->thread);
5223 EXPORT_SYMBOL_GPL(md_do_sync);
5227 * This routine is regularly called by all per-raid-array threads to
5228 * deal with generic issues like resync and super-block update.
5229 * Raid personalities that don't have a thread (linear/raid0) do not
5230 * need this as they never do any recovery or update the superblock.
5232 * It does not do any resync itself, but rather "forks" off other threads
5233 * to do that as needed.
5234 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5235 * "->recovery" and create a thread at ->sync_thread.
5236 * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5237 * and wakeups up this thread which will reap the thread and finish up.
5238 * This thread also removes any faulty devices (with nr_pending == 0).
5240 * The overall approach is:
5241 * 1/ if the superblock needs updating, update it.
5242 * 2/ If a recovery thread is running, don't do anything else.
5243 * 3/ If recovery has finished, clean up, possibly marking spares active.
5244 * 4/ If there are any faulty devices, remove them.
5245 * 5/ If array is degraded, try to add spares devices
5246 * 6/ If array has spares or is not in-sync, start a resync thread.
5248 void md_check_recovery(mddev_t *mddev)
5251 struct list_head *rtmp;
5255 bitmap_daemon_work(mddev->bitmap);
5260 if (signal_pending(current)) {
5261 if (mddev->pers->sync_request) {
5262 printk(KERN_INFO "md: %s in immediate safe mode\n",
5264 mddev->safemode = 2;
5266 flush_signals(current);
5271 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5272 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5273 (mddev->safemode == 1) ||
5274 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5275 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5279 if (mddev_trylock(mddev)) {
5282 spin_lock_irq(&mddev->write_lock);
5283 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5284 !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5286 mddev->sb_dirty = 3;
5288 if (mddev->safemode == 1)
5289 mddev->safemode = 0;
5290 spin_unlock_irq(&mddev->write_lock);
5292 if (mddev->sb_dirty)
5293 md_update_sb(mddev);
5296 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5297 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5298 /* resync/recovery still happening */
5299 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5302 if (mddev->sync_thread) {
5303 /* resync has finished, collect result */
5304 md_unregister_thread(mddev->sync_thread);
5305 mddev->sync_thread = NULL;
5306 if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5307 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5309 /* activate any spares */
5310 mddev->pers->spare_active(mddev);
5312 md_update_sb(mddev);
5314 /* if array is no-longer degraded, then any saved_raid_disk
5315 * information must be scrapped
5317 if (!mddev->degraded)
5318 ITERATE_RDEV(mddev,rdev,rtmp)
5319 rdev->saved_raid_disk = -1;
5321 mddev->recovery = 0;
5322 /* flag recovery needed just to double check */
5323 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5324 md_new_event(mddev);
5327 /* Clear some bits that don't mean anything, but
5330 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5331 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5332 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5333 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5335 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5337 /* no recovery is running.
5338 * remove any failed drives, then
5339 * add spares if possible.
5340 * Spare are also removed and re-added, to allow
5341 * the personality to fail the re-add.
5343 ITERATE_RDEV(mddev,rdev,rtmp)
5344 if (rdev->raid_disk >= 0 &&
5345 (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
5346 atomic_read(&rdev->nr_pending)==0) {
5347 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5349 sprintf(nm,"rd%d", rdev->raid_disk);
5350 sysfs_remove_link(&mddev->kobj, nm);
5351 rdev->raid_disk = -1;
5355 if (mddev->degraded) {
5356 ITERATE_RDEV(mddev,rdev,rtmp)
5357 if (rdev->raid_disk < 0
5358 && !test_bit(Faulty, &rdev->flags)) {
5359 rdev->recovery_offset = 0;
5360 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5362 sprintf(nm, "rd%d", rdev->raid_disk);
5363 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
5365 md_new_event(mddev);
5372 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5373 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5374 } else if (mddev->recovery_cp < MaxSector) {
5375 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5376 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5377 /* nothing to be done ... */
5380 if (mddev->pers->sync_request) {
5381 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5382 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5383 /* We are adding a device or devices to an array
5384 * which has the bitmap stored on all devices.
5385 * So make sure all bitmap pages get written
5387 bitmap_write_all(mddev->bitmap);
5389 mddev->sync_thread = md_register_thread(md_do_sync,
5392 if (!mddev->sync_thread) {
5393 printk(KERN_ERR "%s: could not start resync"
5396 /* leave the spares where they are, it shouldn't hurt */
5397 mddev->recovery = 0;
5399 md_wakeup_thread(mddev->sync_thread);
5400 md_new_event(mddev);
5403 mddev_unlock(mddev);
5407 static int md_notify_reboot(struct notifier_block *this,
5408 unsigned long code, void *x)
5410 struct list_head *tmp;
5413 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5415 printk(KERN_INFO "md: stopping all md devices.\n");
5417 ITERATE_MDDEV(mddev,tmp)
5418 if (mddev_trylock(mddev)) {
5419 do_md_stop (mddev, 1);
5420 mddev_unlock(mddev);
5423 * certain more exotic SCSI devices are known to be
5424 * volatile wrt too early system reboots. While the
5425 * right place to handle this issue is the given
5426 * driver, we do want to have a safe RAID driver ...
5433 static struct notifier_block md_notifier = {
5434 .notifier_call = md_notify_reboot,
5436 .priority = INT_MAX, /* before any real devices */
5439 static void md_geninit(void)
5441 struct proc_dir_entry *p;
5443 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5445 p = create_proc_entry("mdstat", S_IRUGO, NULL);
5447 p->proc_fops = &md_seq_fops;
5450 static int __init md_init(void)
5454 printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5455 " MD_SB_DISKS=%d\n",
5456 MD_MAJOR_VERSION, MD_MINOR_VERSION,
5457 MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
5458 printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
5461 if (register_blkdev(MAJOR_NR, "md"))
5463 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5464 unregister_blkdev(MAJOR_NR, "md");
5468 blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5469 md_probe, NULL, NULL);
5470 blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5471 md_probe, NULL, NULL);
5473 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5474 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
5475 S_IFBLK|S_IRUSR|S_IWUSR,
5478 for (minor=0; minor < MAX_MD_DEVS; ++minor)
5479 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
5480 S_IFBLK|S_IRUSR|S_IWUSR,
5484 register_reboot_notifier(&md_notifier);
5485 raid_table_header = register_sysctl_table(raid_root_table, 1);
5495 * Searches all registered partitions for autorun RAID arrays
5498 static dev_t detected_devices[128];
5501 void md_autodetect_dev(dev_t dev)
5503 if (dev_cnt >= 0 && dev_cnt < 127)
5504 detected_devices[dev_cnt++] = dev;
5508 static void autostart_arrays(int part)
5513 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5515 for (i = 0; i < dev_cnt; i++) {
5516 dev_t dev = detected_devices[i];
5518 rdev = md_import_device(dev,0, 0);
5522 if (test_bit(Faulty, &rdev->flags)) {
5526 list_add(&rdev->same_set, &pending_raid_disks);
5530 autorun_devices(part);
5535 static __exit void md_exit(void)
5538 struct list_head *tmp;
5540 blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5541 blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
5542 for (i=0; i < MAX_MD_DEVS; i++)
5543 devfs_remove("md/%d", i);
5544 for (i=0; i < MAX_MD_DEVS; i++)
5545 devfs_remove("md/d%d", i);
5549 unregister_blkdev(MAJOR_NR,"md");
5550 unregister_blkdev(mdp_major, "mdp");
5551 unregister_reboot_notifier(&md_notifier);
5552 unregister_sysctl_table(raid_table_header);
5553 remove_proc_entry("mdstat", NULL);
5554 ITERATE_MDDEV(mddev,tmp) {
5555 struct gendisk *disk = mddev->gendisk;
5558 export_array(mddev);
5561 mddev->gendisk = NULL;
5566 module_init(md_init)
5567 module_exit(md_exit)
5569 static int get_ro(char *buffer, struct kernel_param *kp)
5571 return sprintf(buffer, "%d", start_readonly);
5573 static int set_ro(const char *val, struct kernel_param *kp)
5576 int num = simple_strtoul(val, &e, 10);
5577 if (*val && (*e == '\0' || *e == '\n')) {
5578 start_readonly = num;
5584 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5585 module_param(start_dirty_degraded, int, 0644);
5588 EXPORT_SYMBOL(register_md_personality);
5589 EXPORT_SYMBOL(unregister_md_personality);
5590 EXPORT_SYMBOL(md_error);
5591 EXPORT_SYMBOL(md_done_sync);
5592 EXPORT_SYMBOL(md_write_start);
5593 EXPORT_SYMBOL(md_write_end);
5594 EXPORT_SYMBOL(md_register_thread);
5595 EXPORT_SYMBOL(md_unregister_thread);
5596 EXPORT_SYMBOL(md_wakeup_thread);
5597 EXPORT_SYMBOL(md_check_recovery);
5598 MODULE_LICENSE("GPL");
5600 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);