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/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58 static void autostart_arrays(int part);
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
64 static void md_print_devices(void);
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
91 static inline int speed_max(mddev_t *mddev)
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
97 static struct ctl_table_header *raid_table_header;
99 static ctl_table raid_table[] = {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
119 static ctl_table raid_dir_table[] = {
121 .ctl_name = DEV_RAID,
124 .mode = S_IRUGO|S_IXUGO,
130 static ctl_table raid_root_table[] = {
136 .child = raid_dir_table,
141 static struct block_device_operations md_fops;
143 static int start_readonly;
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
162 EXPORT_SYMBOL_GPL(md_new_event);
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
167 static void md_new_event_inintr(mddev_t *mddev)
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
188 #define for_each_mddev(mddev,tmp) \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
204 static int md_fail_request(struct request_queue *q, struct bio *bio)
210 static inline mddev_t *mddev_get(mddev_t *mddev)
212 atomic_inc(&mddev->active);
216 static void mddev_delayed_delete(struct work_struct *ws)
218 mddev_t *mddev = container_of(ws, mddev_t, del_work);
219 kobject_del(&mddev->kobj);
220 kobject_put(&mddev->kobj);
223 static void mddev_put(mddev_t *mddev)
225 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
227 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
228 !mddev->hold_active) {
229 list_del(&mddev->all_mddevs);
230 if (mddev->gendisk) {
231 /* we did a probe so need to clean up.
232 * Call schedule_work inside the spinlock
233 * so that flush_scheduled_work() after
234 * mddev_find will succeed in waiting for the
237 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
238 schedule_work(&mddev->del_work);
242 spin_unlock(&all_mddevs_lock);
245 static mddev_t * mddev_find(dev_t unit)
247 mddev_t *mddev, *new = NULL;
250 spin_lock(&all_mddevs_lock);
253 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
254 if (mddev->unit == unit) {
256 spin_unlock(&all_mddevs_lock);
262 list_add(&new->all_mddevs, &all_mddevs);
263 spin_unlock(&all_mddevs_lock);
264 new->hold_active = UNTIL_IOCTL;
268 /* find an unused unit number */
269 static int next_minor = 512;
270 int start = next_minor;
274 dev = MKDEV(MD_MAJOR, next_minor);
276 if (next_minor > MINORMASK)
278 if (next_minor == start) {
279 /* Oh dear, all in use. */
280 spin_unlock(&all_mddevs_lock);
286 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
287 if (mddev->unit == dev) {
293 new->md_minor = MINOR(dev);
294 new->hold_active = UNTIL_STOP;
295 list_add(&new->all_mddevs, &all_mddevs);
296 spin_unlock(&all_mddevs_lock);
299 spin_unlock(&all_mddevs_lock);
301 new = kzalloc(sizeof(*new), GFP_KERNEL);
306 if (MAJOR(unit) == MD_MAJOR)
307 new->md_minor = MINOR(unit);
309 new->md_minor = MINOR(unit) >> MdpMinorShift;
311 mutex_init(&new->reconfig_mutex);
312 INIT_LIST_HEAD(&new->disks);
313 INIT_LIST_HEAD(&new->all_mddevs);
314 init_timer(&new->safemode_timer);
315 atomic_set(&new->active, 1);
316 atomic_set(&new->openers, 0);
317 spin_lock_init(&new->write_lock);
318 init_waitqueue_head(&new->sb_wait);
319 init_waitqueue_head(&new->recovery_wait);
320 new->reshape_position = MaxSector;
322 new->resync_max = MaxSector;
323 new->level = LEVEL_NONE;
328 static inline int mddev_lock(mddev_t * mddev)
330 return mutex_lock_interruptible(&mddev->reconfig_mutex);
333 static inline int mddev_trylock(mddev_t * mddev)
335 return mutex_trylock(&mddev->reconfig_mutex);
338 static inline void mddev_unlock(mddev_t * mddev)
340 mutex_unlock(&mddev->reconfig_mutex);
342 md_wakeup_thread(mddev->thread);
345 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
349 list_for_each_entry(rdev, &mddev->disks, same_set)
350 if (rdev->desc_nr == nr)
356 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
360 list_for_each_entry(rdev, &mddev->disks, same_set)
361 if (rdev->bdev->bd_dev == dev)
367 static struct mdk_personality *find_pers(int level, char *clevel)
369 struct mdk_personality *pers;
370 list_for_each_entry(pers, &pers_list, list) {
371 if (level != LEVEL_NONE && pers->level == level)
373 if (strcmp(pers->name, clevel)==0)
379 /* return the offset of the super block in 512byte sectors */
380 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
382 sector_t num_sectors = bdev->bd_inode->i_size / 512;
383 return MD_NEW_SIZE_SECTORS(num_sectors);
386 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
388 sector_t num_sectors = rdev->sb_start;
391 num_sectors &= ~((sector_t)chunk_size/512 - 1);
395 static int alloc_disk_sb(mdk_rdev_t * rdev)
400 rdev->sb_page = alloc_page(GFP_KERNEL);
401 if (!rdev->sb_page) {
402 printk(KERN_ALERT "md: out of memory.\n");
409 static void free_disk_sb(mdk_rdev_t * rdev)
412 put_page(rdev->sb_page);
414 rdev->sb_page = NULL;
421 static void super_written(struct bio *bio, int error)
423 mdk_rdev_t *rdev = bio->bi_private;
424 mddev_t *mddev = rdev->mddev;
426 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
427 printk("md: super_written gets error=%d, uptodate=%d\n",
428 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
429 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
430 md_error(mddev, rdev);
433 if (atomic_dec_and_test(&mddev->pending_writes))
434 wake_up(&mddev->sb_wait);
438 static void super_written_barrier(struct bio *bio, int error)
440 struct bio *bio2 = bio->bi_private;
441 mdk_rdev_t *rdev = bio2->bi_private;
442 mddev_t *mddev = rdev->mddev;
444 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
445 error == -EOPNOTSUPP) {
447 /* barriers don't appear to be supported :-( */
448 set_bit(BarriersNotsupp, &rdev->flags);
449 mddev->barriers_work = 0;
450 spin_lock_irqsave(&mddev->write_lock, flags);
451 bio2->bi_next = mddev->biolist;
452 mddev->biolist = bio2;
453 spin_unlock_irqrestore(&mddev->write_lock, flags);
454 wake_up(&mddev->sb_wait);
458 bio->bi_private = rdev;
459 super_written(bio, error);
463 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
464 sector_t sector, int size, struct page *page)
466 /* write first size bytes of page to sector of rdev
467 * Increment mddev->pending_writes before returning
468 * and decrement it on completion, waking up sb_wait
469 * if zero is reached.
470 * If an error occurred, call md_error
472 * As we might need to resubmit the request if BIO_RW_BARRIER
473 * causes ENOTSUPP, we allocate a spare bio...
475 struct bio *bio = bio_alloc(GFP_NOIO, 1);
476 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
478 bio->bi_bdev = rdev->bdev;
479 bio->bi_sector = sector;
480 bio_add_page(bio, page, size, 0);
481 bio->bi_private = rdev;
482 bio->bi_end_io = super_written;
485 atomic_inc(&mddev->pending_writes);
486 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
488 rw |= (1<<BIO_RW_BARRIER);
489 rbio = bio_clone(bio, GFP_NOIO);
490 rbio->bi_private = bio;
491 rbio->bi_end_io = super_written_barrier;
492 submit_bio(rw, rbio);
497 void md_super_wait(mddev_t *mddev)
499 /* wait for all superblock writes that were scheduled to complete.
500 * if any had to be retried (due to BARRIER problems), retry them
504 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
505 if (atomic_read(&mddev->pending_writes)==0)
507 while (mddev->biolist) {
509 spin_lock_irq(&mddev->write_lock);
510 bio = mddev->biolist;
511 mddev->biolist = bio->bi_next ;
513 spin_unlock_irq(&mddev->write_lock);
514 submit_bio(bio->bi_rw, bio);
518 finish_wait(&mddev->sb_wait, &wq);
521 static void bi_complete(struct bio *bio, int error)
523 complete((struct completion*)bio->bi_private);
526 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
527 struct page *page, int rw)
529 struct bio *bio = bio_alloc(GFP_NOIO, 1);
530 struct completion event;
533 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
536 bio->bi_sector = sector;
537 bio_add_page(bio, page, size, 0);
538 init_completion(&event);
539 bio->bi_private = &event;
540 bio->bi_end_io = bi_complete;
542 wait_for_completion(&event);
544 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
548 EXPORT_SYMBOL_GPL(sync_page_io);
550 static int read_disk_sb(mdk_rdev_t * rdev, int size)
552 char b[BDEVNAME_SIZE];
553 if (!rdev->sb_page) {
561 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
567 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
568 bdevname(rdev->bdev,b));
572 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
574 return sb1->set_uuid0 == sb2->set_uuid0 &&
575 sb1->set_uuid1 == sb2->set_uuid1 &&
576 sb1->set_uuid2 == sb2->set_uuid2 &&
577 sb1->set_uuid3 == sb2->set_uuid3;
580 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
583 mdp_super_t *tmp1, *tmp2;
585 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
586 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
588 if (!tmp1 || !tmp2) {
590 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
598 * nr_disks is not constant
603 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
611 static u32 md_csum_fold(u32 csum)
613 csum = (csum & 0xffff) + (csum >> 16);
614 return (csum & 0xffff) + (csum >> 16);
617 static unsigned int calc_sb_csum(mdp_super_t * sb)
620 u32 *sb32 = (u32*)sb;
622 unsigned int disk_csum, csum;
624 disk_csum = sb->sb_csum;
627 for (i = 0; i < MD_SB_BYTES/4 ; i++)
629 csum = (newcsum & 0xffffffff) + (newcsum>>32);
633 /* This used to use csum_partial, which was wrong for several
634 * reasons including that different results are returned on
635 * different architectures. It isn't critical that we get exactly
636 * the same return value as before (we always csum_fold before
637 * testing, and that removes any differences). However as we
638 * know that csum_partial always returned a 16bit value on
639 * alphas, do a fold to maximise conformity to previous behaviour.
641 sb->sb_csum = md_csum_fold(disk_csum);
643 sb->sb_csum = disk_csum;
650 * Handle superblock details.
651 * We want to be able to handle multiple superblock formats
652 * so we have a common interface to them all, and an array of
653 * different handlers.
654 * We rely on user-space to write the initial superblock, and support
655 * reading and updating of superblocks.
656 * Interface methods are:
657 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
658 * loads and validates a superblock on dev.
659 * if refdev != NULL, compare superblocks on both devices
661 * 0 - dev has a superblock that is compatible with refdev
662 * 1 - dev has a superblock that is compatible and newer than refdev
663 * so dev should be used as the refdev in future
664 * -EINVAL superblock incompatible or invalid
665 * -othererror e.g. -EIO
667 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
668 * Verify that dev is acceptable into mddev.
669 * The first time, mddev->raid_disks will be 0, and data from
670 * dev should be merged in. Subsequent calls check that dev
671 * is new enough. Return 0 or -EINVAL
673 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
674 * Update the superblock for rdev with data in mddev
675 * This does not write to disc.
681 struct module *owner;
682 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
684 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
685 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
686 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
687 sector_t num_sectors);
691 * load_super for 0.90.0
693 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
695 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
700 * Calculate the position of the superblock (512byte sectors),
701 * it's at the end of the disk.
703 * It also happens to be a multiple of 4Kb.
705 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
707 ret = read_disk_sb(rdev, MD_SB_BYTES);
712 bdevname(rdev->bdev, b);
713 sb = (mdp_super_t*)page_address(rdev->sb_page);
715 if (sb->md_magic != MD_SB_MAGIC) {
716 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
721 if (sb->major_version != 0 ||
722 sb->minor_version < 90 ||
723 sb->minor_version > 91) {
724 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
725 sb->major_version, sb->minor_version,
730 if (sb->raid_disks <= 0)
733 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
734 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
739 rdev->preferred_minor = sb->md_minor;
740 rdev->data_offset = 0;
741 rdev->sb_size = MD_SB_BYTES;
743 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
744 if (sb->level != 1 && sb->level != 4
745 && sb->level != 5 && sb->level != 6
746 && sb->level != 10) {
747 /* FIXME use a better test */
749 "md: bitmaps not supported for this level.\n");
754 if (sb->level == LEVEL_MULTIPATH)
757 rdev->desc_nr = sb->this_disk.number;
763 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
764 if (!uuid_equal(refsb, sb)) {
765 printk(KERN_WARNING "md: %s has different UUID to %s\n",
766 b, bdevname(refdev->bdev,b2));
769 if (!sb_equal(refsb, sb)) {
770 printk(KERN_WARNING "md: %s has same UUID"
771 " but different superblock to %s\n",
772 b, bdevname(refdev->bdev, b2));
776 ev2 = md_event(refsb);
782 rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
784 if (rdev->size < sb->size && sb->level > 1)
785 /* "this cannot possibly happen" ... */
793 * validate_super for 0.90.0
795 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
798 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
799 __u64 ev1 = md_event(sb);
801 rdev->raid_disk = -1;
802 clear_bit(Faulty, &rdev->flags);
803 clear_bit(In_sync, &rdev->flags);
804 clear_bit(WriteMostly, &rdev->flags);
805 clear_bit(BarriersNotsupp, &rdev->flags);
807 if (mddev->raid_disks == 0) {
808 mddev->major_version = 0;
809 mddev->minor_version = sb->minor_version;
810 mddev->patch_version = sb->patch_version;
812 mddev->chunk_size = sb->chunk_size;
813 mddev->ctime = sb->ctime;
814 mddev->utime = sb->utime;
815 mddev->level = sb->level;
816 mddev->clevel[0] = 0;
817 mddev->layout = sb->layout;
818 mddev->raid_disks = sb->raid_disks;
819 mddev->size = sb->size;
821 mddev->bitmap_offset = 0;
822 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
824 if (mddev->minor_version >= 91) {
825 mddev->reshape_position = sb->reshape_position;
826 mddev->delta_disks = sb->delta_disks;
827 mddev->new_level = sb->new_level;
828 mddev->new_layout = sb->new_layout;
829 mddev->new_chunk = sb->new_chunk;
831 mddev->reshape_position = MaxSector;
832 mddev->delta_disks = 0;
833 mddev->new_level = mddev->level;
834 mddev->new_layout = mddev->layout;
835 mddev->new_chunk = mddev->chunk_size;
838 if (sb->state & (1<<MD_SB_CLEAN))
839 mddev->recovery_cp = MaxSector;
841 if (sb->events_hi == sb->cp_events_hi &&
842 sb->events_lo == sb->cp_events_lo) {
843 mddev->recovery_cp = sb->recovery_cp;
845 mddev->recovery_cp = 0;
848 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
849 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
850 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
851 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
853 mddev->max_disks = MD_SB_DISKS;
855 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
856 mddev->bitmap_file == NULL)
857 mddev->bitmap_offset = mddev->default_bitmap_offset;
859 } else if (mddev->pers == NULL) {
860 /* Insist on good event counter while assembling */
862 if (ev1 < mddev->events)
864 } else if (mddev->bitmap) {
865 /* if adding to array with a bitmap, then we can accept an
866 * older device ... but not too old.
868 if (ev1 < mddev->bitmap->events_cleared)
871 if (ev1 < mddev->events)
872 /* just a hot-add of a new device, leave raid_disk at -1 */
876 if (mddev->level != LEVEL_MULTIPATH) {
877 desc = sb->disks + rdev->desc_nr;
879 if (desc->state & (1<<MD_DISK_FAULTY))
880 set_bit(Faulty, &rdev->flags);
881 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
882 desc->raid_disk < mddev->raid_disks */) {
883 set_bit(In_sync, &rdev->flags);
884 rdev->raid_disk = desc->raid_disk;
886 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
887 set_bit(WriteMostly, &rdev->flags);
888 } else /* MULTIPATH are always insync */
889 set_bit(In_sync, &rdev->flags);
894 * sync_super for 0.90.0
896 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
900 int next_spare = mddev->raid_disks;
903 /* make rdev->sb match mddev data..
906 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
907 * 3/ any empty disks < next_spare become removed
909 * disks[0] gets initialised to REMOVED because
910 * we cannot be sure from other fields if it has
911 * been initialised or not.
914 int active=0, working=0,failed=0,spare=0,nr_disks=0;
916 rdev->sb_size = MD_SB_BYTES;
918 sb = (mdp_super_t*)page_address(rdev->sb_page);
920 memset(sb, 0, sizeof(*sb));
922 sb->md_magic = MD_SB_MAGIC;
923 sb->major_version = mddev->major_version;
924 sb->patch_version = mddev->patch_version;
925 sb->gvalid_words = 0; /* ignored */
926 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
927 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
928 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
929 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
931 sb->ctime = mddev->ctime;
932 sb->level = mddev->level;
933 sb->size = mddev->size;
934 sb->raid_disks = mddev->raid_disks;
935 sb->md_minor = mddev->md_minor;
936 sb->not_persistent = 0;
937 sb->utime = mddev->utime;
939 sb->events_hi = (mddev->events>>32);
940 sb->events_lo = (u32)mddev->events;
942 if (mddev->reshape_position == MaxSector)
943 sb->minor_version = 90;
945 sb->minor_version = 91;
946 sb->reshape_position = mddev->reshape_position;
947 sb->new_level = mddev->new_level;
948 sb->delta_disks = mddev->delta_disks;
949 sb->new_layout = mddev->new_layout;
950 sb->new_chunk = mddev->new_chunk;
952 mddev->minor_version = sb->minor_version;
955 sb->recovery_cp = mddev->recovery_cp;
956 sb->cp_events_hi = (mddev->events>>32);
957 sb->cp_events_lo = (u32)mddev->events;
958 if (mddev->recovery_cp == MaxSector)
959 sb->state = (1<< MD_SB_CLEAN);
963 sb->layout = mddev->layout;
964 sb->chunk_size = mddev->chunk_size;
966 if (mddev->bitmap && mddev->bitmap_file == NULL)
967 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
969 sb->disks[0].state = (1<<MD_DISK_REMOVED);
970 list_for_each_entry(rdev2, &mddev->disks, same_set) {
973 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
974 && !test_bit(Faulty, &rdev2->flags))
975 desc_nr = rdev2->raid_disk;
977 desc_nr = next_spare++;
978 rdev2->desc_nr = desc_nr;
979 d = &sb->disks[rdev2->desc_nr];
981 d->number = rdev2->desc_nr;
982 d->major = MAJOR(rdev2->bdev->bd_dev);
983 d->minor = MINOR(rdev2->bdev->bd_dev);
984 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
985 && !test_bit(Faulty, &rdev2->flags))
986 d->raid_disk = rdev2->raid_disk;
988 d->raid_disk = rdev2->desc_nr; /* compatibility */
989 if (test_bit(Faulty, &rdev2->flags))
990 d->state = (1<<MD_DISK_FAULTY);
991 else if (test_bit(In_sync, &rdev2->flags)) {
992 d->state = (1<<MD_DISK_ACTIVE);
993 d->state |= (1<<MD_DISK_SYNC);
1001 if (test_bit(WriteMostly, &rdev2->flags))
1002 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1004 /* now set the "removed" and "faulty" bits on any missing devices */
1005 for (i=0 ; i < mddev->raid_disks ; i++) {
1006 mdp_disk_t *d = &sb->disks[i];
1007 if (d->state == 0 && d->number == 0) {
1010 d->state = (1<<MD_DISK_REMOVED);
1011 d->state |= (1<<MD_DISK_FAULTY);
1015 sb->nr_disks = nr_disks;
1016 sb->active_disks = active;
1017 sb->working_disks = working;
1018 sb->failed_disks = failed;
1019 sb->spare_disks = spare;
1021 sb->this_disk = sb->disks[rdev->desc_nr];
1022 sb->sb_csum = calc_sb_csum(sb);
1026 * rdev_size_change for 0.90.0
1028 static unsigned long long
1029 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1031 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1032 return 0; /* component must fit device */
1033 if (rdev->mddev->bitmap_offset)
1034 return 0; /* can't move bitmap */
1035 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1036 if (!num_sectors || num_sectors > rdev->sb_start)
1037 num_sectors = rdev->sb_start;
1038 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1040 md_super_wait(rdev->mddev);
1041 return num_sectors / 2; /* kB for sysfs */
1046 * version 1 superblock
1049 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1053 unsigned long long newcsum;
1054 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1055 __le32 *isuper = (__le32*)sb;
1058 disk_csum = sb->sb_csum;
1061 for (i=0; size>=4; size -= 4 )
1062 newcsum += le32_to_cpu(*isuper++);
1065 newcsum += le16_to_cpu(*(__le16*) isuper);
1067 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1068 sb->sb_csum = disk_csum;
1069 return cpu_to_le32(csum);
1072 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1074 struct mdp_superblock_1 *sb;
1077 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1081 * Calculate the position of the superblock in 512byte sectors.
1082 * It is always aligned to a 4K boundary and
1083 * depeding on minor_version, it can be:
1084 * 0: At least 8K, but less than 12K, from end of device
1085 * 1: At start of device
1086 * 2: 4K from start of device.
1088 switch(minor_version) {
1090 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1092 sb_start &= ~(sector_t)(4*2-1);
1103 rdev->sb_start = sb_start;
1105 /* superblock is rarely larger than 1K, but it can be larger,
1106 * and it is safe to read 4k, so we do that
1108 ret = read_disk_sb(rdev, 4096);
1109 if (ret) return ret;
1112 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1114 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1115 sb->major_version != cpu_to_le32(1) ||
1116 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1117 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1118 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1121 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1122 printk("md: invalid superblock checksum on %s\n",
1123 bdevname(rdev->bdev,b));
1126 if (le64_to_cpu(sb->data_size) < 10) {
1127 printk("md: data_size too small on %s\n",
1128 bdevname(rdev->bdev,b));
1131 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1132 if (sb->level != cpu_to_le32(1) &&
1133 sb->level != cpu_to_le32(4) &&
1134 sb->level != cpu_to_le32(5) &&
1135 sb->level != cpu_to_le32(6) &&
1136 sb->level != cpu_to_le32(10)) {
1138 "md: bitmaps not supported for this level.\n");
1143 rdev->preferred_minor = 0xffff;
1144 rdev->data_offset = le64_to_cpu(sb->data_offset);
1145 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1147 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1148 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1149 if (rdev->sb_size & bmask)
1150 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1153 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1156 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1159 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1165 struct mdp_superblock_1 *refsb =
1166 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1168 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1169 sb->level != refsb->level ||
1170 sb->layout != refsb->layout ||
1171 sb->chunksize != refsb->chunksize) {
1172 printk(KERN_WARNING "md: %s has strangely different"
1173 " superblock to %s\n",
1174 bdevname(rdev->bdev,b),
1175 bdevname(refdev->bdev,b2));
1178 ev1 = le64_to_cpu(sb->events);
1179 ev2 = le64_to_cpu(refsb->events);
1187 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1189 rdev->size = rdev->sb_start / 2;
1190 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1192 rdev->size = le64_to_cpu(sb->data_size)/2;
1193 if (le32_to_cpu(sb->chunksize))
1194 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1196 if (le64_to_cpu(sb->size) > rdev->size*2)
1201 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1203 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1204 __u64 ev1 = le64_to_cpu(sb->events);
1206 rdev->raid_disk = -1;
1207 clear_bit(Faulty, &rdev->flags);
1208 clear_bit(In_sync, &rdev->flags);
1209 clear_bit(WriteMostly, &rdev->flags);
1210 clear_bit(BarriersNotsupp, &rdev->flags);
1212 if (mddev->raid_disks == 0) {
1213 mddev->major_version = 1;
1214 mddev->patch_version = 0;
1215 mddev->external = 0;
1216 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1217 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1218 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1219 mddev->level = le32_to_cpu(sb->level);
1220 mddev->clevel[0] = 0;
1221 mddev->layout = le32_to_cpu(sb->layout);
1222 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1223 mddev->size = le64_to_cpu(sb->size)/2;
1224 mddev->events = ev1;
1225 mddev->bitmap_offset = 0;
1226 mddev->default_bitmap_offset = 1024 >> 9;
1228 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1229 memcpy(mddev->uuid, sb->set_uuid, 16);
1231 mddev->max_disks = (4096-256)/2;
1233 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1234 mddev->bitmap_file == NULL )
1235 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1237 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1238 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1239 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1240 mddev->new_level = le32_to_cpu(sb->new_level);
1241 mddev->new_layout = le32_to_cpu(sb->new_layout);
1242 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1244 mddev->reshape_position = MaxSector;
1245 mddev->delta_disks = 0;
1246 mddev->new_level = mddev->level;
1247 mddev->new_layout = mddev->layout;
1248 mddev->new_chunk = mddev->chunk_size;
1251 } else if (mddev->pers == NULL) {
1252 /* Insist of good event counter while assembling */
1254 if (ev1 < mddev->events)
1256 } else if (mddev->bitmap) {
1257 /* If adding to array with a bitmap, then we can accept an
1258 * older device, but not too old.
1260 if (ev1 < mddev->bitmap->events_cleared)
1263 if (ev1 < mddev->events)
1264 /* just a hot-add of a new device, leave raid_disk at -1 */
1267 if (mddev->level != LEVEL_MULTIPATH) {
1269 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1271 case 0xffff: /* spare */
1273 case 0xfffe: /* faulty */
1274 set_bit(Faulty, &rdev->flags);
1277 if ((le32_to_cpu(sb->feature_map) &
1278 MD_FEATURE_RECOVERY_OFFSET))
1279 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1281 set_bit(In_sync, &rdev->flags);
1282 rdev->raid_disk = role;
1285 if (sb->devflags & WriteMostly1)
1286 set_bit(WriteMostly, &rdev->flags);
1287 } else /* MULTIPATH are always insync */
1288 set_bit(In_sync, &rdev->flags);
1293 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1295 struct mdp_superblock_1 *sb;
1298 /* make rdev->sb match mddev and rdev data. */
1300 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1302 sb->feature_map = 0;
1304 sb->recovery_offset = cpu_to_le64(0);
1305 memset(sb->pad1, 0, sizeof(sb->pad1));
1306 memset(sb->pad2, 0, sizeof(sb->pad2));
1307 memset(sb->pad3, 0, sizeof(sb->pad3));
1309 sb->utime = cpu_to_le64((__u64)mddev->utime);
1310 sb->events = cpu_to_le64(mddev->events);
1312 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1314 sb->resync_offset = cpu_to_le64(0);
1316 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1318 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1319 sb->size = cpu_to_le64(mddev->size<<1);
1321 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1322 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1323 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1326 if (rdev->raid_disk >= 0 &&
1327 !test_bit(In_sync, &rdev->flags)) {
1328 if (mddev->curr_resync_completed > rdev->recovery_offset)
1329 rdev->recovery_offset = mddev->curr_resync_completed;
1330 if (rdev->recovery_offset > 0) {
1332 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1333 sb->recovery_offset =
1334 cpu_to_le64(rdev->recovery_offset);
1338 if (mddev->reshape_position != MaxSector) {
1339 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1340 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1341 sb->new_layout = cpu_to_le32(mddev->new_layout);
1342 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1343 sb->new_level = cpu_to_le32(mddev->new_level);
1344 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1348 list_for_each_entry(rdev2, &mddev->disks, same_set)
1349 if (rdev2->desc_nr+1 > max_dev)
1350 max_dev = rdev2->desc_nr+1;
1352 if (max_dev > le32_to_cpu(sb->max_dev))
1353 sb->max_dev = cpu_to_le32(max_dev);
1354 for (i=0; i<max_dev;i++)
1355 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1357 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1359 if (test_bit(Faulty, &rdev2->flags))
1360 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1361 else if (test_bit(In_sync, &rdev2->flags))
1362 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1363 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1364 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1366 sb->dev_roles[i] = cpu_to_le16(0xffff);
1369 sb->sb_csum = calc_sb_1_csum(sb);
1372 static unsigned long long
1373 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1375 struct mdp_superblock_1 *sb;
1376 sector_t max_sectors;
1377 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1378 return 0; /* component must fit device */
1379 if (rdev->sb_start < rdev->data_offset) {
1380 /* minor versions 1 and 2; superblock before data */
1381 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1382 max_sectors -= rdev->data_offset;
1383 if (!num_sectors || num_sectors > max_sectors)
1384 num_sectors = max_sectors;
1385 } else if (rdev->mddev->bitmap_offset) {
1386 /* minor version 0 with bitmap we can't move */
1389 /* minor version 0; superblock after data */
1391 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1392 sb_start &= ~(sector_t)(4*2 - 1);
1393 max_sectors = rdev->size * 2 + sb_start - rdev->sb_start;
1394 if (!num_sectors || num_sectors > max_sectors)
1395 num_sectors = max_sectors;
1396 rdev->sb_start = sb_start;
1398 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1399 sb->data_size = cpu_to_le64(num_sectors);
1400 sb->super_offset = rdev->sb_start;
1401 sb->sb_csum = calc_sb_1_csum(sb);
1402 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1404 md_super_wait(rdev->mddev);
1405 return num_sectors / 2; /* kB for sysfs */
1408 static struct super_type super_types[] = {
1411 .owner = THIS_MODULE,
1412 .load_super = super_90_load,
1413 .validate_super = super_90_validate,
1414 .sync_super = super_90_sync,
1415 .rdev_size_change = super_90_rdev_size_change,
1419 .owner = THIS_MODULE,
1420 .load_super = super_1_load,
1421 .validate_super = super_1_validate,
1422 .sync_super = super_1_sync,
1423 .rdev_size_change = super_1_rdev_size_change,
1427 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1429 mdk_rdev_t *rdev, *rdev2;
1432 rdev_for_each_rcu(rdev, mddev1)
1433 rdev_for_each_rcu(rdev2, mddev2)
1434 if (rdev->bdev->bd_contains ==
1435 rdev2->bdev->bd_contains) {
1443 static LIST_HEAD(pending_raid_disks);
1445 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1447 struct mdk_personality *pers = mddev->pers;
1448 struct gendisk *disk = mddev->gendisk;
1449 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1450 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1452 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1453 if (pers && pers->level >= 4 && pers->level <= 6)
1456 /* If rdev is integrity capable, register profile for mddev */
1457 if (!bi_mddev && bi_rdev) {
1458 if (blk_integrity_register(disk, bi_rdev))
1459 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1460 __func__, disk->disk_name);
1462 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1467 /* Check that mddev and rdev have matching profiles */
1468 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1469 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1470 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1471 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1473 blk_integrity_unregister(disk);
1477 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1479 char b[BDEVNAME_SIZE];
1489 /* prevent duplicates */
1490 if (find_rdev(mddev, rdev->bdev->bd_dev))
1493 /* make sure rdev->size exceeds mddev->size */
1494 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1496 /* Cannot change size, so fail
1497 * If mddev->level <= 0, then we don't care
1498 * about aligning sizes (e.g. linear)
1500 if (mddev->level > 0)
1503 mddev->size = rdev->size;
1506 /* Verify rdev->desc_nr is unique.
1507 * If it is -1, assign a free number, else
1508 * check number is not in use
1510 if (rdev->desc_nr < 0) {
1512 if (mddev->pers) choice = mddev->raid_disks;
1513 while (find_rdev_nr(mddev, choice))
1515 rdev->desc_nr = choice;
1517 if (find_rdev_nr(mddev, rdev->desc_nr))
1520 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1521 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1522 mdname(mddev), mddev->max_disks);
1525 bdevname(rdev->bdev,b);
1526 while ( (s=strchr(b, '/')) != NULL)
1529 rdev->mddev = mddev;
1530 printk(KERN_INFO "md: bind<%s>\n", b);
1532 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1535 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1536 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1537 kobject_del(&rdev->kobj);
1540 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1542 list_add_rcu(&rdev->same_set, &mddev->disks);
1543 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1545 /* May as well allow recovery to be retried once */
1546 mddev->recovery_disabled = 0;
1548 md_integrity_check(rdev, mddev);
1552 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1557 static void md_delayed_delete(struct work_struct *ws)
1559 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1560 kobject_del(&rdev->kobj);
1561 kobject_put(&rdev->kobj);
1564 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1566 char b[BDEVNAME_SIZE];
1571 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1572 list_del_rcu(&rdev->same_set);
1573 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1575 sysfs_remove_link(&rdev->kobj, "block");
1576 sysfs_put(rdev->sysfs_state);
1577 rdev->sysfs_state = NULL;
1578 /* We need to delay this, otherwise we can deadlock when
1579 * writing to 'remove' to "dev/state". We also need
1580 * to delay it due to rcu usage.
1583 INIT_WORK(&rdev->del_work, md_delayed_delete);
1584 kobject_get(&rdev->kobj);
1585 schedule_work(&rdev->del_work);
1589 * prevent the device from being mounted, repartitioned or
1590 * otherwise reused by a RAID array (or any other kernel
1591 * subsystem), by bd_claiming the device.
1593 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1596 struct block_device *bdev;
1597 char b[BDEVNAME_SIZE];
1599 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1601 printk(KERN_ERR "md: could not open %s.\n",
1602 __bdevname(dev, b));
1603 return PTR_ERR(bdev);
1605 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1607 printk(KERN_ERR "md: could not bd_claim %s.\n",
1609 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1613 set_bit(AllReserved, &rdev->flags);
1618 static void unlock_rdev(mdk_rdev_t *rdev)
1620 struct block_device *bdev = rdev->bdev;
1625 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1628 void md_autodetect_dev(dev_t dev);
1630 static void export_rdev(mdk_rdev_t * rdev)
1632 char b[BDEVNAME_SIZE];
1633 printk(KERN_INFO "md: export_rdev(%s)\n",
1634 bdevname(rdev->bdev,b));
1639 if (test_bit(AutoDetected, &rdev->flags))
1640 md_autodetect_dev(rdev->bdev->bd_dev);
1643 kobject_put(&rdev->kobj);
1646 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1648 unbind_rdev_from_array(rdev);
1652 static void export_array(mddev_t *mddev)
1654 mdk_rdev_t *rdev, *tmp;
1656 rdev_for_each(rdev, tmp, mddev) {
1661 kick_rdev_from_array(rdev);
1663 if (!list_empty(&mddev->disks))
1665 mddev->raid_disks = 0;
1666 mddev->major_version = 0;
1669 static void print_desc(mdp_disk_t *desc)
1671 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1672 desc->major,desc->minor,desc->raid_disk,desc->state);
1675 static void print_sb_90(mdp_super_t *sb)
1680 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1681 sb->major_version, sb->minor_version, sb->patch_version,
1682 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1684 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1685 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1686 sb->md_minor, sb->layout, sb->chunk_size);
1687 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1688 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1689 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1690 sb->failed_disks, sb->spare_disks,
1691 sb->sb_csum, (unsigned long)sb->events_lo);
1694 for (i = 0; i < MD_SB_DISKS; i++) {
1697 desc = sb->disks + i;
1698 if (desc->number || desc->major || desc->minor ||
1699 desc->raid_disk || (desc->state && (desc->state != 4))) {
1700 printk(" D %2d: ", i);
1704 printk(KERN_INFO "md: THIS: ");
1705 print_desc(&sb->this_disk);
1708 static void print_sb_1(struct mdp_superblock_1 *sb)
1712 uuid = sb->set_uuid;
1713 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1714 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1715 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1716 le32_to_cpu(sb->major_version),
1717 le32_to_cpu(sb->feature_map),
1718 uuid[0], uuid[1], uuid[2], uuid[3],
1719 uuid[4], uuid[5], uuid[6], uuid[7],
1720 uuid[8], uuid[9], uuid[10], uuid[11],
1721 uuid[12], uuid[13], uuid[14], uuid[15],
1723 (unsigned long long)le64_to_cpu(sb->ctime)
1724 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1726 uuid = sb->device_uuid;
1727 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1729 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1730 ":%02x%02x%02x%02x%02x%02x\n"
1731 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1732 KERN_INFO "md: (MaxDev:%u) \n",
1733 le32_to_cpu(sb->level),
1734 (unsigned long long)le64_to_cpu(sb->size),
1735 le32_to_cpu(sb->raid_disks),
1736 le32_to_cpu(sb->layout),
1737 le32_to_cpu(sb->chunksize),
1738 (unsigned long long)le64_to_cpu(sb->data_offset),
1739 (unsigned long long)le64_to_cpu(sb->data_size),
1740 (unsigned long long)le64_to_cpu(sb->super_offset),
1741 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1742 le32_to_cpu(sb->dev_number),
1743 uuid[0], uuid[1], uuid[2], uuid[3],
1744 uuid[4], uuid[5], uuid[6], uuid[7],
1745 uuid[8], uuid[9], uuid[10], uuid[11],
1746 uuid[12], uuid[13], uuid[14], uuid[15],
1748 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1749 (unsigned long long)le64_to_cpu(sb->events),
1750 (unsigned long long)le64_to_cpu(sb->resync_offset),
1751 le32_to_cpu(sb->sb_csum),
1752 le32_to_cpu(sb->max_dev)
1756 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1758 char b[BDEVNAME_SIZE];
1759 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1760 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1761 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1763 if (rdev->sb_loaded) {
1764 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1765 switch (major_version) {
1767 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1770 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1774 printk(KERN_INFO "md: no rdev superblock!\n");
1777 static void md_print_devices(void)
1779 struct list_head *tmp;
1782 char b[BDEVNAME_SIZE];
1785 printk("md: **********************************\n");
1786 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1787 printk("md: **********************************\n");
1788 for_each_mddev(mddev, tmp) {
1791 bitmap_print_sb(mddev->bitmap);
1793 printk("%s: ", mdname(mddev));
1794 list_for_each_entry(rdev, &mddev->disks, same_set)
1795 printk("<%s>", bdevname(rdev->bdev,b));
1798 list_for_each_entry(rdev, &mddev->disks, same_set)
1799 print_rdev(rdev, mddev->major_version);
1801 printk("md: **********************************\n");
1806 static void sync_sbs(mddev_t * mddev, int nospares)
1808 /* Update each superblock (in-memory image), but
1809 * if we are allowed to, skip spares which already
1810 * have the right event counter, or have one earlier
1811 * (which would mean they aren't being marked as dirty
1812 * with the rest of the array)
1816 list_for_each_entry(rdev, &mddev->disks, same_set) {
1817 if (rdev->sb_events == mddev->events ||
1819 rdev->raid_disk < 0 &&
1820 (rdev->sb_events&1)==0 &&
1821 rdev->sb_events+1 == mddev->events)) {
1822 /* Don't update this superblock */
1823 rdev->sb_loaded = 2;
1825 super_types[mddev->major_version].
1826 sync_super(mddev, rdev);
1827 rdev->sb_loaded = 1;
1832 static void md_update_sb(mddev_t * mddev, int force_change)
1838 if (mddev->external)
1841 spin_lock_irq(&mddev->write_lock);
1843 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1844 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1846 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1847 /* just a clean<-> dirty transition, possibly leave spares alone,
1848 * though if events isn't the right even/odd, we will have to do
1854 if (mddev->degraded)
1855 /* If the array is degraded, then skipping spares is both
1856 * dangerous and fairly pointless.
1857 * Dangerous because a device that was removed from the array
1858 * might have a event_count that still looks up-to-date,
1859 * so it can be re-added without a resync.
1860 * Pointless because if there are any spares to skip,
1861 * then a recovery will happen and soon that array won't
1862 * be degraded any more and the spare can go back to sleep then.
1866 sync_req = mddev->in_sync;
1867 mddev->utime = get_seconds();
1869 /* If this is just a dirty<->clean transition, and the array is clean
1870 * and 'events' is odd, we can roll back to the previous clean state */
1872 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1873 && (mddev->events & 1)
1874 && mddev->events != 1)
1877 /* otherwise we have to go forward and ... */
1879 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1880 /* .. if the array isn't clean, insist on an odd 'events' */
1881 if ((mddev->events&1)==0) {
1886 /* otherwise insist on an even 'events' (for clean states) */
1887 if ((mddev->events&1)) {
1894 if (!mddev->events) {
1896 * oops, this 64-bit counter should never wrap.
1897 * Either we are in around ~1 trillion A.C., assuming
1898 * 1 reboot per second, or we have a bug:
1905 * do not write anything to disk if using
1906 * nonpersistent superblocks
1908 if (!mddev->persistent) {
1909 if (!mddev->external)
1910 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1912 spin_unlock_irq(&mddev->write_lock);
1913 wake_up(&mddev->sb_wait);
1916 sync_sbs(mddev, nospares);
1917 spin_unlock_irq(&mddev->write_lock);
1920 "md: updating %s RAID superblock on device (in sync %d)\n",
1921 mdname(mddev),mddev->in_sync);
1923 bitmap_update_sb(mddev->bitmap);
1924 list_for_each_entry(rdev, &mddev->disks, same_set) {
1925 char b[BDEVNAME_SIZE];
1926 dprintk(KERN_INFO "md: ");
1927 if (rdev->sb_loaded != 1)
1928 continue; /* no noise on spare devices */
1929 if (test_bit(Faulty, &rdev->flags))
1930 dprintk("(skipping faulty ");
1932 dprintk("%s ", bdevname(rdev->bdev,b));
1933 if (!test_bit(Faulty, &rdev->flags)) {
1934 md_super_write(mddev,rdev,
1935 rdev->sb_start, rdev->sb_size,
1937 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1938 bdevname(rdev->bdev,b),
1939 (unsigned long long)rdev->sb_start);
1940 rdev->sb_events = mddev->events;
1944 if (mddev->level == LEVEL_MULTIPATH)
1945 /* only need to write one superblock... */
1948 md_super_wait(mddev);
1949 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1951 spin_lock_irq(&mddev->write_lock);
1952 if (mddev->in_sync != sync_req ||
1953 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1954 /* have to write it out again */
1955 spin_unlock_irq(&mddev->write_lock);
1958 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1959 spin_unlock_irq(&mddev->write_lock);
1960 wake_up(&mddev->sb_wait);
1964 /* words written to sysfs files may, or may not, be \n terminated.
1965 * We want to accept with case. For this we use cmd_match.
1967 static int cmd_match(const char *cmd, const char *str)
1969 /* See if cmd, written into a sysfs file, matches
1970 * str. They must either be the same, or cmd can
1971 * have a trailing newline
1973 while (*cmd && *str && *cmd == *str) {
1984 struct rdev_sysfs_entry {
1985 struct attribute attr;
1986 ssize_t (*show)(mdk_rdev_t *, char *);
1987 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1991 state_show(mdk_rdev_t *rdev, char *page)
1996 if (test_bit(Faulty, &rdev->flags)) {
1997 len+= sprintf(page+len, "%sfaulty",sep);
2000 if (test_bit(In_sync, &rdev->flags)) {
2001 len += sprintf(page+len, "%sin_sync",sep);
2004 if (test_bit(WriteMostly, &rdev->flags)) {
2005 len += sprintf(page+len, "%swrite_mostly",sep);
2008 if (test_bit(Blocked, &rdev->flags)) {
2009 len += sprintf(page+len, "%sblocked", sep);
2012 if (!test_bit(Faulty, &rdev->flags) &&
2013 !test_bit(In_sync, &rdev->flags)) {
2014 len += sprintf(page+len, "%sspare", sep);
2017 return len+sprintf(page+len, "\n");
2021 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2024 * faulty - simulates and error
2025 * remove - disconnects the device
2026 * writemostly - sets write_mostly
2027 * -writemostly - clears write_mostly
2028 * blocked - sets the Blocked flag
2029 * -blocked - clears the Blocked flag
2032 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2033 md_error(rdev->mddev, rdev);
2035 } else if (cmd_match(buf, "remove")) {
2036 if (rdev->raid_disk >= 0)
2039 mddev_t *mddev = rdev->mddev;
2040 kick_rdev_from_array(rdev);
2042 md_update_sb(mddev, 1);
2043 md_new_event(mddev);
2046 } else if (cmd_match(buf, "writemostly")) {
2047 set_bit(WriteMostly, &rdev->flags);
2049 } else if (cmd_match(buf, "-writemostly")) {
2050 clear_bit(WriteMostly, &rdev->flags);
2052 } else if (cmd_match(buf, "blocked")) {
2053 set_bit(Blocked, &rdev->flags);
2055 } else if (cmd_match(buf, "-blocked")) {
2056 clear_bit(Blocked, &rdev->flags);
2057 wake_up(&rdev->blocked_wait);
2058 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2059 md_wakeup_thread(rdev->mddev->thread);
2063 if (!err && rdev->sysfs_state)
2064 sysfs_notify_dirent(rdev->sysfs_state);
2065 return err ? err : len;
2067 static struct rdev_sysfs_entry rdev_state =
2068 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2071 errors_show(mdk_rdev_t *rdev, char *page)
2073 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2077 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2080 unsigned long n = simple_strtoul(buf, &e, 10);
2081 if (*buf && (*e == 0 || *e == '\n')) {
2082 atomic_set(&rdev->corrected_errors, n);
2087 static struct rdev_sysfs_entry rdev_errors =
2088 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2091 slot_show(mdk_rdev_t *rdev, char *page)
2093 if (rdev->raid_disk < 0)
2094 return sprintf(page, "none\n");
2096 return sprintf(page, "%d\n", rdev->raid_disk);
2100 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2105 int slot = simple_strtoul(buf, &e, 10);
2106 if (strncmp(buf, "none", 4)==0)
2108 else if (e==buf || (*e && *e!= '\n'))
2110 if (rdev->mddev->pers && slot == -1) {
2111 /* Setting 'slot' on an active array requires also
2112 * updating the 'rd%d' link, and communicating
2113 * with the personality with ->hot_*_disk.
2114 * For now we only support removing
2115 * failed/spare devices. This normally happens automatically,
2116 * but not when the metadata is externally managed.
2118 if (rdev->raid_disk == -1)
2120 /* personality does all needed checks */
2121 if (rdev->mddev->pers->hot_add_disk == NULL)
2123 err = rdev->mddev->pers->
2124 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2127 sprintf(nm, "rd%d", rdev->raid_disk);
2128 sysfs_remove_link(&rdev->mddev->kobj, nm);
2129 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2130 md_wakeup_thread(rdev->mddev->thread);
2131 } else if (rdev->mddev->pers) {
2133 /* Activating a spare .. or possibly reactivating
2134 * if we every get bitmaps working here.
2137 if (rdev->raid_disk != -1)
2140 if (rdev->mddev->pers->hot_add_disk == NULL)
2143 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2144 if (rdev2->raid_disk == slot)
2147 rdev->raid_disk = slot;
2148 if (test_bit(In_sync, &rdev->flags))
2149 rdev->saved_raid_disk = slot;
2151 rdev->saved_raid_disk = -1;
2152 err = rdev->mddev->pers->
2153 hot_add_disk(rdev->mddev, rdev);
2155 rdev->raid_disk = -1;
2158 sysfs_notify_dirent(rdev->sysfs_state);
2159 sprintf(nm, "rd%d", rdev->raid_disk);
2160 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2162 "md: cannot register "
2164 nm, mdname(rdev->mddev));
2166 /* don't wakeup anyone, leave that to userspace. */
2168 if (slot >= rdev->mddev->raid_disks)
2170 rdev->raid_disk = slot;
2171 /* assume it is working */
2172 clear_bit(Faulty, &rdev->flags);
2173 clear_bit(WriteMostly, &rdev->flags);
2174 set_bit(In_sync, &rdev->flags);
2175 sysfs_notify_dirent(rdev->sysfs_state);
2181 static struct rdev_sysfs_entry rdev_slot =
2182 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2185 offset_show(mdk_rdev_t *rdev, char *page)
2187 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2191 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2194 unsigned long long offset = simple_strtoull(buf, &e, 10);
2195 if (e==buf || (*e && *e != '\n'))
2197 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2199 if (rdev->size && rdev->mddev->external)
2200 /* Must set offset before size, so overlap checks
2203 rdev->data_offset = offset;
2207 static struct rdev_sysfs_entry rdev_offset =
2208 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2211 rdev_size_show(mdk_rdev_t *rdev, char *page)
2213 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2216 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2218 /* check if two start/length pairs overlap */
2227 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2229 unsigned long long size;
2230 unsigned long long oldsize = rdev->size;
2231 mddev_t *my_mddev = rdev->mddev;
2233 if (strict_strtoull(buf, 10, &size) < 0)
2235 if (my_mddev->pers && rdev->raid_disk >= 0) {
2236 if (my_mddev->persistent) {
2237 size = super_types[my_mddev->major_version].
2238 rdev_size_change(rdev, size * 2);
2242 size = (rdev->bdev->bd_inode->i_size >> 10);
2243 size -= rdev->data_offset/2;
2246 if (size < my_mddev->size)
2247 return -EINVAL; /* component must fit device */
2250 if (size > oldsize && my_mddev->external) {
2251 /* need to check that all other rdevs with the same ->bdev
2252 * do not overlap. We need to unlock the mddev to avoid
2253 * a deadlock. We have already changed rdev->size, and if
2254 * we have to change it back, we will have the lock again.
2258 struct list_head *tmp;
2260 mddev_unlock(my_mddev);
2261 for_each_mddev(mddev, tmp) {
2265 list_for_each_entry(rdev2, &mddev->disks, same_set)
2266 if (test_bit(AllReserved, &rdev2->flags) ||
2267 (rdev->bdev == rdev2->bdev &&
2269 overlaps(rdev->data_offset, rdev->size * 2,
2271 rdev2->size * 2))) {
2275 mddev_unlock(mddev);
2281 mddev_lock(my_mddev);
2283 /* Someone else could have slipped in a size
2284 * change here, but doing so is just silly.
2285 * We put oldsize back because we *know* it is
2286 * safe, and trust userspace not to race with
2289 rdev->size = oldsize;
2296 static struct rdev_sysfs_entry rdev_size =
2297 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2299 static struct attribute *rdev_default_attrs[] = {
2308 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2310 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2311 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2312 mddev_t *mddev = rdev->mddev;
2318 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2320 if (rdev->mddev == NULL)
2323 rv = entry->show(rdev, page);
2324 mddev_unlock(mddev);
2330 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2331 const char *page, size_t length)
2333 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2334 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2336 mddev_t *mddev = rdev->mddev;
2340 if (!capable(CAP_SYS_ADMIN))
2342 rv = mddev ? mddev_lock(mddev): -EBUSY;
2344 if (rdev->mddev == NULL)
2347 rv = entry->store(rdev, page, length);
2348 mddev_unlock(mddev);
2353 static void rdev_free(struct kobject *ko)
2355 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2358 static struct sysfs_ops rdev_sysfs_ops = {
2359 .show = rdev_attr_show,
2360 .store = rdev_attr_store,
2362 static struct kobj_type rdev_ktype = {
2363 .release = rdev_free,
2364 .sysfs_ops = &rdev_sysfs_ops,
2365 .default_attrs = rdev_default_attrs,
2369 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2371 * mark the device faulty if:
2373 * - the device is nonexistent (zero size)
2374 * - the device has no valid superblock
2376 * a faulty rdev _never_ has rdev->sb set.
2378 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2380 char b[BDEVNAME_SIZE];
2385 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2387 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2388 return ERR_PTR(-ENOMEM);
2391 if ((err = alloc_disk_sb(rdev)))
2394 err = lock_rdev(rdev, newdev, super_format == -2);
2398 kobject_init(&rdev->kobj, &rdev_ktype);
2401 rdev->saved_raid_disk = -1;
2402 rdev->raid_disk = -1;
2404 rdev->data_offset = 0;
2405 rdev->sb_events = 0;
2406 atomic_set(&rdev->nr_pending, 0);
2407 atomic_set(&rdev->read_errors, 0);
2408 atomic_set(&rdev->corrected_errors, 0);
2410 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2413 "md: %s has zero or unknown size, marking faulty!\n",
2414 bdevname(rdev->bdev,b));
2419 if (super_format >= 0) {
2420 err = super_types[super_format].
2421 load_super(rdev, NULL, super_minor);
2422 if (err == -EINVAL) {
2424 "md: %s does not have a valid v%d.%d "
2425 "superblock, not importing!\n",
2426 bdevname(rdev->bdev,b),
2427 super_format, super_minor);
2432 "md: could not read %s's sb, not importing!\n",
2433 bdevname(rdev->bdev,b));
2438 INIT_LIST_HEAD(&rdev->same_set);
2439 init_waitqueue_head(&rdev->blocked_wait);
2444 if (rdev->sb_page) {
2450 return ERR_PTR(err);
2454 * Check a full RAID array for plausibility
2458 static void analyze_sbs(mddev_t * mddev)
2461 mdk_rdev_t *rdev, *freshest, *tmp;
2462 char b[BDEVNAME_SIZE];
2465 rdev_for_each(rdev, tmp, mddev)
2466 switch (super_types[mddev->major_version].
2467 load_super(rdev, freshest, mddev->minor_version)) {
2475 "md: fatal superblock inconsistency in %s"
2476 " -- removing from array\n",
2477 bdevname(rdev->bdev,b));
2478 kick_rdev_from_array(rdev);
2482 super_types[mddev->major_version].
2483 validate_super(mddev, freshest);
2486 rdev_for_each(rdev, tmp, mddev) {
2487 if (rdev->desc_nr >= mddev->max_disks ||
2488 i > mddev->max_disks) {
2490 "md: %s: %s: only %d devices permitted\n",
2491 mdname(mddev), bdevname(rdev->bdev, b),
2493 kick_rdev_from_array(rdev);
2496 if (rdev != freshest)
2497 if (super_types[mddev->major_version].
2498 validate_super(mddev, rdev)) {
2499 printk(KERN_WARNING "md: kicking non-fresh %s"
2501 bdevname(rdev->bdev,b));
2502 kick_rdev_from_array(rdev);
2505 if (mddev->level == LEVEL_MULTIPATH) {
2506 rdev->desc_nr = i++;
2507 rdev->raid_disk = rdev->desc_nr;
2508 set_bit(In_sync, &rdev->flags);
2509 } else if (rdev->raid_disk >= mddev->raid_disks) {
2510 rdev->raid_disk = -1;
2511 clear_bit(In_sync, &rdev->flags);
2517 if (mddev->recovery_cp != MaxSector &&
2519 printk(KERN_ERR "md: %s: raid array is not clean"
2520 " -- starting background reconstruction\n",
2525 static void md_safemode_timeout(unsigned long data);
2528 safe_delay_show(mddev_t *mddev, char *page)
2530 int msec = (mddev->safemode_delay*1000)/HZ;
2531 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2534 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2542 /* remove a period, and count digits after it */
2543 if (len >= sizeof(buf))
2545 strlcpy(buf, cbuf, sizeof(buf));
2546 for (i=0; i<len; i++) {
2548 if (isdigit(buf[i])) {
2553 } else if (buf[i] == '.') {
2558 if (strict_strtoul(buf, 10, &msec) < 0)
2560 msec = (msec * 1000) / scale;
2562 mddev->safemode_delay = 0;
2564 unsigned long old_delay = mddev->safemode_delay;
2565 mddev->safemode_delay = (msec*HZ)/1000;
2566 if (mddev->safemode_delay == 0)
2567 mddev->safemode_delay = 1;
2568 if (mddev->safemode_delay < old_delay)
2569 md_safemode_timeout((unsigned long)mddev);
2573 static struct md_sysfs_entry md_safe_delay =
2574 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2577 level_show(mddev_t *mddev, char *page)
2579 struct mdk_personality *p = mddev->pers;
2581 return sprintf(page, "%s\n", p->name);
2582 else if (mddev->clevel[0])
2583 return sprintf(page, "%s\n", mddev->clevel);
2584 else if (mddev->level != LEVEL_NONE)
2585 return sprintf(page, "%d\n", mddev->level);
2591 level_store(mddev_t *mddev, const char *buf, size_t len)
2598 if (len >= sizeof(mddev->clevel))
2600 strncpy(mddev->clevel, buf, len);
2601 if (mddev->clevel[len-1] == '\n')
2603 mddev->clevel[len] = 0;
2604 mddev->level = LEVEL_NONE;
2608 static struct md_sysfs_entry md_level =
2609 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2613 layout_show(mddev_t *mddev, char *page)
2615 /* just a number, not meaningful for all levels */
2616 if (mddev->reshape_position != MaxSector &&
2617 mddev->layout != mddev->new_layout)
2618 return sprintf(page, "%d (%d)\n",
2619 mddev->new_layout, mddev->layout);
2620 return sprintf(page, "%d\n", mddev->layout);
2624 layout_store(mddev_t *mddev, const char *buf, size_t len)
2627 unsigned long n = simple_strtoul(buf, &e, 10);
2629 if (!*buf || (*e && *e != '\n'))
2634 if (mddev->reshape_position != MaxSector)
2635 mddev->new_layout = n;
2640 static struct md_sysfs_entry md_layout =
2641 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2645 raid_disks_show(mddev_t *mddev, char *page)
2647 if (mddev->raid_disks == 0)
2649 if (mddev->reshape_position != MaxSector &&
2650 mddev->delta_disks != 0)
2651 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2652 mddev->raid_disks - mddev->delta_disks);
2653 return sprintf(page, "%d\n", mddev->raid_disks);
2656 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2659 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2663 unsigned long n = simple_strtoul(buf, &e, 10);
2665 if (!*buf || (*e && *e != '\n'))
2669 rv = update_raid_disks(mddev, n);
2670 else if (mddev->reshape_position != MaxSector) {
2671 int olddisks = mddev->raid_disks - mddev->delta_disks;
2672 mddev->delta_disks = n - olddisks;
2673 mddev->raid_disks = n;
2675 mddev->raid_disks = n;
2676 return rv ? rv : len;
2678 static struct md_sysfs_entry md_raid_disks =
2679 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2682 chunk_size_show(mddev_t *mddev, char *page)
2684 if (mddev->reshape_position != MaxSector &&
2685 mddev->chunk_size != mddev->new_chunk)
2686 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2688 return sprintf(page, "%d\n", mddev->chunk_size);
2692 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2694 /* can only set chunk_size if array is not yet active */
2696 unsigned long n = simple_strtoul(buf, &e, 10);
2698 if (!*buf || (*e && *e != '\n'))
2703 else if (mddev->reshape_position != MaxSector)
2704 mddev->new_chunk = n;
2706 mddev->chunk_size = n;
2709 static struct md_sysfs_entry md_chunk_size =
2710 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2713 resync_start_show(mddev_t *mddev, char *page)
2715 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2719 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2722 unsigned long long n = simple_strtoull(buf, &e, 10);
2726 if (!*buf || (*e && *e != '\n'))
2729 mddev->recovery_cp = n;
2732 static struct md_sysfs_entry md_resync_start =
2733 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2736 * The array state can be:
2739 * No devices, no size, no level
2740 * Equivalent to STOP_ARRAY ioctl
2742 * May have some settings, but array is not active
2743 * all IO results in error
2744 * When written, doesn't tear down array, but just stops it
2745 * suspended (not supported yet)
2746 * All IO requests will block. The array can be reconfigured.
2747 * Writing this, if accepted, will block until array is quiescent
2749 * no resync can happen. no superblocks get written.
2750 * write requests fail
2752 * like readonly, but behaves like 'clean' on a write request.
2754 * clean - no pending writes, but otherwise active.
2755 * When written to inactive array, starts without resync
2756 * If a write request arrives then
2757 * if metadata is known, mark 'dirty' and switch to 'active'.
2758 * if not known, block and switch to write-pending
2759 * If written to an active array that has pending writes, then fails.
2761 * fully active: IO and resync can be happening.
2762 * When written to inactive array, starts with resync
2765 * clean, but writes are blocked waiting for 'active' to be written.
2768 * like active, but no writes have been seen for a while (100msec).
2771 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2772 write_pending, active_idle, bad_word};
2773 static char *array_states[] = {
2774 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2775 "write-pending", "active-idle", NULL };
2777 static int match_word(const char *word, char **list)
2780 for (n=0; list[n]; n++)
2781 if (cmd_match(word, list[n]))
2787 array_state_show(mddev_t *mddev, char *page)
2789 enum array_state st = inactive;
2802 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2804 else if (mddev->safemode)
2810 if (list_empty(&mddev->disks) &&
2811 mddev->raid_disks == 0 &&
2817 return sprintf(page, "%s\n", array_states[st]);
2820 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2821 static int do_md_run(mddev_t * mddev);
2822 static int restart_array(mddev_t *mddev);
2825 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2828 enum array_state st = match_word(buf, array_states);
2833 /* stopping an active array */
2834 if (atomic_read(&mddev->openers) > 0)
2836 err = do_md_stop(mddev, 0, 0);
2839 /* stopping an active array */
2841 if (atomic_read(&mddev->openers) > 0)
2843 err = do_md_stop(mddev, 2, 0);
2845 err = 0; /* already inactive */
2848 break; /* not supported yet */
2851 err = do_md_stop(mddev, 1, 0);
2854 set_disk_ro(mddev->gendisk, 1);
2855 err = do_md_run(mddev);
2861 err = do_md_stop(mddev, 1, 0);
2862 else if (mddev->ro == 1)
2863 err = restart_array(mddev);
2866 set_disk_ro(mddev->gendisk, 0);
2870 err = do_md_run(mddev);
2875 restart_array(mddev);
2876 spin_lock_irq(&mddev->write_lock);
2877 if (atomic_read(&mddev->writes_pending) == 0) {
2878 if (mddev->in_sync == 0) {
2880 if (mddev->safemode == 1)
2881 mddev->safemode = 0;
2882 if (mddev->persistent)
2883 set_bit(MD_CHANGE_CLEAN,
2889 spin_unlock_irq(&mddev->write_lock);
2892 mddev->recovery_cp = MaxSector;
2893 err = do_md_run(mddev);
2898 restart_array(mddev);
2899 if (mddev->external)
2900 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2901 wake_up(&mddev->sb_wait);
2905 set_disk_ro(mddev->gendisk, 0);
2906 err = do_md_run(mddev);
2911 /* these cannot be set */
2917 sysfs_notify_dirent(mddev->sysfs_state);
2921 static struct md_sysfs_entry md_array_state =
2922 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2925 null_show(mddev_t *mddev, char *page)
2931 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2933 /* buf must be %d:%d\n? giving major and minor numbers */
2934 /* The new device is added to the array.
2935 * If the array has a persistent superblock, we read the
2936 * superblock to initialise info and check validity.
2937 * Otherwise, only checking done is that in bind_rdev_to_array,
2938 * which mainly checks size.
2941 int major = simple_strtoul(buf, &e, 10);
2947 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2949 minor = simple_strtoul(e+1, &e, 10);
2950 if (*e && *e != '\n')
2952 dev = MKDEV(major, minor);
2953 if (major != MAJOR(dev) ||
2954 minor != MINOR(dev))
2958 if (mddev->persistent) {
2959 rdev = md_import_device(dev, mddev->major_version,
2960 mddev->minor_version);
2961 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2962 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2963 mdk_rdev_t, same_set);
2964 err = super_types[mddev->major_version]
2965 .load_super(rdev, rdev0, mddev->minor_version);
2969 } else if (mddev->external)
2970 rdev = md_import_device(dev, -2, -1);
2972 rdev = md_import_device(dev, -1, -1);
2975 return PTR_ERR(rdev);
2976 err = bind_rdev_to_array(rdev, mddev);
2980 return err ? err : len;
2983 static struct md_sysfs_entry md_new_device =
2984 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2987 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2990 unsigned long chunk, end_chunk;
2994 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2996 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2997 if (buf == end) break;
2998 if (*end == '-') { /* range */
3000 end_chunk = simple_strtoul(buf, &end, 0);
3001 if (buf == end) break;
3003 if (*end && !isspace(*end)) break;
3004 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3006 while (isspace(*buf)) buf++;
3008 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3013 static struct md_sysfs_entry md_bitmap =
3014 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3017 size_show(mddev_t *mddev, char *page)
3019 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
3022 static int update_size(mddev_t *mddev, sector_t num_sectors);
3025 size_store(mddev_t *mddev, const char *buf, size_t len)
3027 /* If array is inactive, we can reduce the component size, but
3028 * not increase it (except from 0).
3029 * If array is active, we can try an on-line resize
3033 unsigned long long size = simple_strtoull(buf, &e, 10);
3034 if (!*buf || *buf == '\n' ||
3039 err = update_size(mddev, size * 2);
3040 md_update_sb(mddev, 1);
3042 if (mddev->size == 0 ||
3048 return err ? err : len;
3051 static struct md_sysfs_entry md_size =
3052 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3057 * 'none' for arrays with no metadata (good luck...)
3058 * 'external' for arrays with externally managed metadata,
3059 * or N.M for internally known formats
3062 metadata_show(mddev_t *mddev, char *page)
3064 if (mddev->persistent)
3065 return sprintf(page, "%d.%d\n",
3066 mddev->major_version, mddev->minor_version);
3067 else if (mddev->external)
3068 return sprintf(page, "external:%s\n", mddev->metadata_type);
3070 return sprintf(page, "none\n");
3074 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3078 /* Changing the details of 'external' metadata is
3079 * always permitted. Otherwise there must be
3080 * no devices attached to the array.
3082 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3084 else if (!list_empty(&mddev->disks))
3087 if (cmd_match(buf, "none")) {
3088 mddev->persistent = 0;
3089 mddev->external = 0;
3090 mddev->major_version = 0;
3091 mddev->minor_version = 90;
3094 if (strncmp(buf, "external:", 9) == 0) {
3095 size_t namelen = len-9;
3096 if (namelen >= sizeof(mddev->metadata_type))
3097 namelen = sizeof(mddev->metadata_type)-1;
3098 strncpy(mddev->metadata_type, buf+9, namelen);
3099 mddev->metadata_type[namelen] = 0;
3100 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3101 mddev->metadata_type[--namelen] = 0;
3102 mddev->persistent = 0;
3103 mddev->external = 1;
3104 mddev->major_version = 0;
3105 mddev->minor_version = 90;
3108 major = simple_strtoul(buf, &e, 10);
3109 if (e==buf || *e != '.')
3112 minor = simple_strtoul(buf, &e, 10);
3113 if (e==buf || (*e && *e != '\n') )
3115 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3117 mddev->major_version = major;
3118 mddev->minor_version = minor;
3119 mddev->persistent = 1;
3120 mddev->external = 0;
3124 static struct md_sysfs_entry md_metadata =
3125 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3128 action_show(mddev_t *mddev, char *page)
3130 char *type = "idle";
3131 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3132 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3133 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3135 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3136 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3138 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3142 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3145 return sprintf(page, "%s\n", type);
3149 action_store(mddev_t *mddev, const char *page, size_t len)
3151 if (!mddev->pers || !mddev->pers->sync_request)
3154 if (cmd_match(page, "idle")) {
3155 if (mddev->sync_thread) {
3156 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3157 md_unregister_thread(mddev->sync_thread);
3158 mddev->sync_thread = NULL;
3159 mddev->recovery = 0;
3161 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3162 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3164 else if (cmd_match(page, "resync"))
3165 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3166 else if (cmd_match(page, "recover")) {
3167 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3168 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3169 } else if (cmd_match(page, "reshape")) {
3171 if (mddev->pers->start_reshape == NULL)
3173 err = mddev->pers->start_reshape(mddev);
3176 sysfs_notify(&mddev->kobj, NULL, "degraded");
3178 if (cmd_match(page, "check"))
3179 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3180 else if (!cmd_match(page, "repair"))
3182 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3183 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3185 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3186 md_wakeup_thread(mddev->thread);
3187 sysfs_notify_dirent(mddev->sysfs_action);
3192 mismatch_cnt_show(mddev_t *mddev, char *page)
3194 return sprintf(page, "%llu\n",
3195 (unsigned long long) mddev->resync_mismatches);
3198 static struct md_sysfs_entry md_scan_mode =
3199 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3202 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3205 sync_min_show(mddev_t *mddev, char *page)
3207 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3208 mddev->sync_speed_min ? "local": "system");
3212 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3216 if (strncmp(buf, "system", 6)==0) {
3217 mddev->sync_speed_min = 0;
3220 min = simple_strtoul(buf, &e, 10);
3221 if (buf == e || (*e && *e != '\n') || min <= 0)
3223 mddev->sync_speed_min = min;
3227 static struct md_sysfs_entry md_sync_min =
3228 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3231 sync_max_show(mddev_t *mddev, char *page)
3233 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3234 mddev->sync_speed_max ? "local": "system");
3238 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3242 if (strncmp(buf, "system", 6)==0) {
3243 mddev->sync_speed_max = 0;
3246 max = simple_strtoul(buf, &e, 10);
3247 if (buf == e || (*e && *e != '\n') || max <= 0)
3249 mddev->sync_speed_max = max;
3253 static struct md_sysfs_entry md_sync_max =
3254 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3257 degraded_show(mddev_t *mddev, char *page)
3259 return sprintf(page, "%d\n", mddev->degraded);
3261 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3264 sync_force_parallel_show(mddev_t *mddev, char *page)
3266 return sprintf(page, "%d\n", mddev->parallel_resync);
3270 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3274 if (strict_strtol(buf, 10, &n))
3277 if (n != 0 && n != 1)
3280 mddev->parallel_resync = n;
3282 if (mddev->sync_thread)
3283 wake_up(&resync_wait);
3288 /* force parallel resync, even with shared block devices */
3289 static struct md_sysfs_entry md_sync_force_parallel =
3290 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3291 sync_force_parallel_show, sync_force_parallel_store);
3294 sync_speed_show(mddev_t *mddev, char *page)
3296 unsigned long resync, dt, db;
3297 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3298 dt = (jiffies - mddev->resync_mark) / HZ;
3300 db = resync - mddev->resync_mark_cnt;
3301 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3304 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3307 sync_completed_show(mddev_t *mddev, char *page)
3309 unsigned long max_blocks, resync;
3311 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3312 max_blocks = mddev->resync_max_sectors;
3314 max_blocks = mddev->size << 1;
3316 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3317 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3320 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3323 min_sync_show(mddev_t *mddev, char *page)
3325 return sprintf(page, "%llu\n",
3326 (unsigned long long)mddev->resync_min);
3329 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3331 unsigned long long min;
3332 if (strict_strtoull(buf, 10, &min))
3334 if (min > mddev->resync_max)
3336 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3339 /* Must be a multiple of chunk_size */
3340 if (mddev->chunk_size) {
3341 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3344 mddev->resync_min = min;
3349 static struct md_sysfs_entry md_min_sync =
3350 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3353 max_sync_show(mddev_t *mddev, char *page)
3355 if (mddev->resync_max == MaxSector)
3356 return sprintf(page, "max\n");
3358 return sprintf(page, "%llu\n",
3359 (unsigned long long)mddev->resync_max);
3362 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3364 if (strncmp(buf, "max", 3) == 0)
3365 mddev->resync_max = MaxSector;
3367 unsigned long long max;
3368 if (strict_strtoull(buf, 10, &max))
3370 if (max < mddev->resync_min)
3372 if (max < mddev->resync_max &&
3373 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3376 /* Must be a multiple of chunk_size */
3377 if (mddev->chunk_size) {
3378 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3381 mddev->resync_max = max;
3383 wake_up(&mddev->recovery_wait);
3387 static struct md_sysfs_entry md_max_sync =
3388 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3391 suspend_lo_show(mddev_t *mddev, char *page)
3393 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3397 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3400 unsigned long long new = simple_strtoull(buf, &e, 10);
3402 if (mddev->pers->quiesce == NULL)
3404 if (buf == e || (*e && *e != '\n'))
3406 if (new >= mddev->suspend_hi ||
3407 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3408 mddev->suspend_lo = new;
3409 mddev->pers->quiesce(mddev, 2);
3414 static struct md_sysfs_entry md_suspend_lo =
3415 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3419 suspend_hi_show(mddev_t *mddev, char *page)
3421 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3425 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3428 unsigned long long new = simple_strtoull(buf, &e, 10);
3430 if (mddev->pers->quiesce == NULL)
3432 if (buf == e || (*e && *e != '\n'))
3434 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3435 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3436 mddev->suspend_hi = new;
3437 mddev->pers->quiesce(mddev, 1);
3438 mddev->pers->quiesce(mddev, 0);
3443 static struct md_sysfs_entry md_suspend_hi =
3444 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3447 reshape_position_show(mddev_t *mddev, char *page)
3449 if (mddev->reshape_position != MaxSector)
3450 return sprintf(page, "%llu\n",
3451 (unsigned long long)mddev->reshape_position);
3452 strcpy(page, "none\n");
3457 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3460 unsigned long long new = simple_strtoull(buf, &e, 10);
3463 if (buf == e || (*e && *e != '\n'))
3465 mddev->reshape_position = new;
3466 mddev->delta_disks = 0;
3467 mddev->new_level = mddev->level;
3468 mddev->new_layout = mddev->layout;
3469 mddev->new_chunk = mddev->chunk_size;
3473 static struct md_sysfs_entry md_reshape_position =
3474 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3475 reshape_position_store);
3478 static struct attribute *md_default_attrs[] = {
3481 &md_raid_disks.attr,
3482 &md_chunk_size.attr,
3484 &md_resync_start.attr,
3486 &md_new_device.attr,
3487 &md_safe_delay.attr,
3488 &md_array_state.attr,
3489 &md_reshape_position.attr,
3493 static struct attribute *md_redundancy_attrs[] = {
3495 &md_mismatches.attr,
3498 &md_sync_speed.attr,
3499 &md_sync_force_parallel.attr,
3500 &md_sync_completed.attr,
3503 &md_suspend_lo.attr,
3504 &md_suspend_hi.attr,
3509 static struct attribute_group md_redundancy_group = {
3511 .attrs = md_redundancy_attrs,
3516 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3518 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3519 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3524 rv = mddev_lock(mddev);
3526 rv = entry->show(mddev, page);
3527 mddev_unlock(mddev);
3533 md_attr_store(struct kobject *kobj, struct attribute *attr,
3534 const char *page, size_t length)
3536 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3537 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3542 if (!capable(CAP_SYS_ADMIN))
3544 rv = mddev_lock(mddev);
3545 if (mddev->hold_active == UNTIL_IOCTL)
3546 mddev->hold_active = 0;
3548 rv = entry->store(mddev, page, length);
3549 mddev_unlock(mddev);
3554 static void md_free(struct kobject *ko)
3556 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3558 if (mddev->sysfs_state)
3559 sysfs_put(mddev->sysfs_state);
3561 if (mddev->gendisk) {
3562 del_gendisk(mddev->gendisk);
3563 put_disk(mddev->gendisk);
3566 blk_cleanup_queue(mddev->queue);
3571 static struct sysfs_ops md_sysfs_ops = {
3572 .show = md_attr_show,
3573 .store = md_attr_store,
3575 static struct kobj_type md_ktype = {
3577 .sysfs_ops = &md_sysfs_ops,
3578 .default_attrs = md_default_attrs,
3583 static int md_alloc(dev_t dev, char *name)
3585 static DEFINE_MUTEX(disks_mutex);
3586 mddev_t *mddev = mddev_find(dev);
3587 struct gendisk *disk;
3596 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3597 shift = partitioned ? MdpMinorShift : 0;
3598 unit = MINOR(mddev->unit) >> shift;
3600 /* wait for any previous instance if this device
3601 * to be completed removed (mddev_delayed_delete).
3603 flush_scheduled_work();
3605 mutex_lock(&disks_mutex);
3606 if (mddev->gendisk) {
3607 mutex_unlock(&disks_mutex);
3613 /* Need to ensure that 'name' is not a duplicate.
3616 spin_lock(&all_mddevs_lock);
3618 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3619 if (mddev2->gendisk &&
3620 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3621 spin_unlock(&all_mddevs_lock);
3624 spin_unlock(&all_mddevs_lock);
3627 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3628 if (!mddev->queue) {
3629 mutex_unlock(&disks_mutex);
3633 /* Can be unlocked because the queue is new: no concurrency */
3634 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3636 blk_queue_make_request(mddev->queue, md_fail_request);
3638 disk = alloc_disk(1 << shift);
3640 mutex_unlock(&disks_mutex);
3641 blk_cleanup_queue(mddev->queue);
3642 mddev->queue = NULL;
3646 disk->major = MAJOR(mddev->unit);
3647 disk->first_minor = unit << shift;
3649 strcpy(disk->disk_name, name);
3650 else if (partitioned)
3651 sprintf(disk->disk_name, "md_d%d", unit);
3653 sprintf(disk->disk_name, "md%d", unit);
3654 disk->fops = &md_fops;
3655 disk->private_data = mddev;
3656 disk->queue = mddev->queue;
3657 /* Allow extended partitions. This makes the
3658 * 'mdp' device redundant, but we can't really
3661 disk->flags |= GENHD_FL_EXT_DEVT;
3663 mddev->gendisk = disk;
3664 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3665 &disk_to_dev(disk)->kobj, "%s", "md");
3666 mutex_unlock(&disks_mutex);
3668 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3671 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3672 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3678 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3680 md_alloc(dev, NULL);
3684 static int add_named_array(const char *val, struct kernel_param *kp)
3686 /* val must be "md_*" where * is not all digits.
3687 * We allocate an array with a large free minor number, and
3688 * set the name to val. val must not already be an active name.
3690 int len = strlen(val);
3691 char buf[DISK_NAME_LEN];
3693 while (len && val[len-1] == '\n')
3695 if (len >= DISK_NAME_LEN)
3697 strlcpy(buf, val, len+1);
3698 if (strncmp(buf, "md_", 3) != 0)
3700 return md_alloc(0, buf);
3703 static void md_safemode_timeout(unsigned long data)
3705 mddev_t *mddev = (mddev_t *) data;
3707 if (!atomic_read(&mddev->writes_pending)) {
3708 mddev->safemode = 1;
3709 if (mddev->external)
3710 sysfs_notify_dirent(mddev->sysfs_state);
3712 md_wakeup_thread(mddev->thread);
3715 static int start_dirty_degraded;
3717 static int do_md_run(mddev_t * mddev)
3722 struct gendisk *disk;
3723 struct mdk_personality *pers;
3724 char b[BDEVNAME_SIZE];
3726 if (list_empty(&mddev->disks))
3727 /* cannot run an array with no devices.. */
3734 * Analyze all RAID superblock(s)
3736 if (!mddev->raid_disks) {
3737 if (!mddev->persistent)
3742 chunk_size = mddev->chunk_size;
3745 if (chunk_size > MAX_CHUNK_SIZE) {
3746 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3747 chunk_size, MAX_CHUNK_SIZE);
3751 * chunk-size has to be a power of 2
3753 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3754 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3758 /* devices must have minimum size of one chunk */
3759 list_for_each_entry(rdev, &mddev->disks, same_set) {
3760 if (test_bit(Faulty, &rdev->flags))
3762 if (rdev->size < chunk_size / 1024) {
3764 "md: Dev %s smaller than chunk_size:"
3766 bdevname(rdev->bdev,b),
3767 (unsigned long long)rdev->size,
3774 if (mddev->level != LEVEL_NONE)
3775 request_module("md-level-%d", mddev->level);
3776 else if (mddev->clevel[0])
3777 request_module("md-%s", mddev->clevel);
3780 * Drop all container device buffers, from now on
3781 * the only valid external interface is through the md
3784 list_for_each_entry(rdev, &mddev->disks, same_set) {
3785 if (test_bit(Faulty, &rdev->flags))
3787 sync_blockdev(rdev->bdev);
3788 invalidate_bdev(rdev->bdev);
3790 /* perform some consistency tests on the device.
3791 * We don't want the data to overlap the metadata,
3792 * Internal Bitmap issues has handled elsewhere.
3794 if (rdev->data_offset < rdev->sb_start) {
3796 rdev->data_offset + mddev->size*2
3798 printk("md: %s: data overlaps metadata\n",
3803 if (rdev->sb_start + rdev->sb_size/512
3804 > rdev->data_offset) {
3805 printk("md: %s: metadata overlaps data\n",
3810 sysfs_notify_dirent(rdev->sysfs_state);
3813 md_probe(mddev->unit, NULL, NULL);
3814 disk = mddev->gendisk;
3818 spin_lock(&pers_lock);
3819 pers = find_pers(mddev->level, mddev->clevel);
3820 if (!pers || !try_module_get(pers->owner)) {
3821 spin_unlock(&pers_lock);
3822 if (mddev->level != LEVEL_NONE)
3823 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3826 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3831 spin_unlock(&pers_lock);
3832 mddev->level = pers->level;
3833 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3835 if (pers->level >= 4 && pers->level <= 6)
3836 /* Cannot support integrity (yet) */
3837 blk_integrity_unregister(mddev->gendisk);
3839 if (mddev->reshape_position != MaxSector &&
3840 pers->start_reshape == NULL) {
3841 /* This personality cannot handle reshaping... */
3843 module_put(pers->owner);
3847 if (pers->sync_request) {
3848 /* Warn if this is a potentially silly
3851 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3855 list_for_each_entry(rdev, &mddev->disks, same_set)
3856 list_for_each_entry(rdev2, &mddev->disks, same_set) {
3858 rdev->bdev->bd_contains ==
3859 rdev2->bdev->bd_contains) {
3861 "%s: WARNING: %s appears to be"
3862 " on the same physical disk as"
3865 bdevname(rdev->bdev,b),
3866 bdevname(rdev2->bdev,b2));
3873 "True protection against single-disk"
3874 " failure might be compromised.\n");
3877 mddev->recovery = 0;
3878 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3879 mddev->barriers_work = 1;
3880 mddev->ok_start_degraded = start_dirty_degraded;
3883 mddev->ro = 2; /* read-only, but switch on first write */
3885 err = mddev->pers->run(mddev);
3887 printk(KERN_ERR "md: pers->run() failed ...\n");
3888 else if (mddev->pers->sync_request) {
3889 err = bitmap_create(mddev);
3891 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3892 mdname(mddev), err);
3893 mddev->pers->stop(mddev);
3897 module_put(mddev->pers->owner);
3899 bitmap_destroy(mddev);
3902 if (mddev->pers->sync_request) {
3903 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3905 "md: cannot register extra attributes for %s\n",
3907 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3908 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3911 atomic_set(&mddev->writes_pending,0);
3912 mddev->safemode = 0;
3913 mddev->safemode_timer.function = md_safemode_timeout;
3914 mddev->safemode_timer.data = (unsigned long) mddev;
3915 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3918 list_for_each_entry(rdev, &mddev->disks, same_set)
3919 if (rdev->raid_disk >= 0) {
3921 sprintf(nm, "rd%d", rdev->raid_disk);
3922 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3923 printk("md: cannot register %s for %s\n",
3927 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3930 md_update_sb(mddev, 0);
3932 set_capacity(disk, mddev->array_sectors);
3934 /* If we call blk_queue_make_request here, it will
3935 * re-initialise max_sectors etc which may have been
3936 * refined inside -> run. So just set the bits we need to set.
3937 * Most initialisation happended when we called
3938 * blk_queue_make_request(..., md_fail_request)
3941 mddev->queue->queuedata = mddev;
3942 mddev->queue->make_request_fn = mddev->pers->make_request;
3944 /* If there is a partially-recovered drive we need to
3945 * start recovery here. If we leave it to md_check_recovery,
3946 * it will remove the drives and not do the right thing
3948 if (mddev->degraded && !mddev->sync_thread) {
3950 list_for_each_entry(rdev, &mddev->disks, same_set)
3951 if (rdev->raid_disk >= 0 &&
3952 !test_bit(In_sync, &rdev->flags) &&
3953 !test_bit(Faulty, &rdev->flags))
3954 /* complete an interrupted recovery */
3956 if (spares && mddev->pers->sync_request) {
3957 mddev->recovery = 0;
3958 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3959 mddev->sync_thread = md_register_thread(md_do_sync,
3962 if (!mddev->sync_thread) {
3963 printk(KERN_ERR "%s: could not start resync"
3966 /* leave the spares where they are, it shouldn't hurt */
3967 mddev->recovery = 0;
3971 md_wakeup_thread(mddev->thread);
3972 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3975 md_new_event(mddev);
3976 sysfs_notify_dirent(mddev->sysfs_state);
3977 if (mddev->sysfs_action)
3978 sysfs_notify_dirent(mddev->sysfs_action);
3979 sysfs_notify(&mddev->kobj, NULL, "degraded");
3980 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3984 static int restart_array(mddev_t *mddev)
3986 struct gendisk *disk = mddev->gendisk;
3988 /* Complain if it has no devices */
3989 if (list_empty(&mddev->disks))
3995 mddev->safemode = 0;
3997 set_disk_ro(disk, 0);
3998 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4000 /* Kick recovery or resync if necessary */
4001 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4002 md_wakeup_thread(mddev->thread);
4003 md_wakeup_thread(mddev->sync_thread);
4004 sysfs_notify_dirent(mddev->sysfs_state);
4008 /* similar to deny_write_access, but accounts for our holding a reference
4009 * to the file ourselves */
4010 static int deny_bitmap_write_access(struct file * file)
4012 struct inode *inode = file->f_mapping->host;
4014 spin_lock(&inode->i_lock);
4015 if (atomic_read(&inode->i_writecount) > 1) {
4016 spin_unlock(&inode->i_lock);
4019 atomic_set(&inode->i_writecount, -1);
4020 spin_unlock(&inode->i_lock);
4025 static void restore_bitmap_write_access(struct file *file)
4027 struct inode *inode = file->f_mapping->host;
4029 spin_lock(&inode->i_lock);
4030 atomic_set(&inode->i_writecount, 1);
4031 spin_unlock(&inode->i_lock);
4035 * 0 - completely stop and dis-assemble array
4036 * 1 - switch to readonly
4037 * 2 - stop but do not disassemble array
4039 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4042 struct gendisk *disk = mddev->gendisk;
4044 if (atomic_read(&mddev->openers) > is_open) {
4045 printk("md: %s still in use.\n",mdname(mddev));
4051 if (mddev->sync_thread) {
4052 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4053 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4054 md_unregister_thread(mddev->sync_thread);
4055 mddev->sync_thread = NULL;
4058 del_timer_sync(&mddev->safemode_timer);
4061 case 1: /* readonly */
4067 case 0: /* disassemble */
4069 bitmap_flush(mddev);
4070 md_super_wait(mddev);
4072 set_disk_ro(disk, 0);
4073 blk_queue_make_request(mddev->queue, md_fail_request);
4074 mddev->pers->stop(mddev);
4075 mddev->queue->merge_bvec_fn = NULL;
4076 mddev->queue->unplug_fn = NULL;
4077 mddev->queue->backing_dev_info.congested_fn = NULL;
4078 if (mddev->pers->sync_request) {
4079 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4080 if (mddev->sysfs_action)
4081 sysfs_put(mddev->sysfs_action);
4082 mddev->sysfs_action = NULL;
4084 module_put(mddev->pers->owner);
4086 /* tell userspace to handle 'inactive' */
4087 sysfs_notify_dirent(mddev->sysfs_state);
4089 set_capacity(disk, 0);
4095 if (!mddev->in_sync || mddev->flags) {
4096 /* mark array as shutdown cleanly */
4098 md_update_sb(mddev, 1);
4101 set_disk_ro(disk, 1);
4102 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4106 * Free resources if final stop
4111 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4113 bitmap_destroy(mddev);
4114 if (mddev->bitmap_file) {
4115 restore_bitmap_write_access(mddev->bitmap_file);
4116 fput(mddev->bitmap_file);
4117 mddev->bitmap_file = NULL;
4119 mddev->bitmap_offset = 0;
4121 list_for_each_entry(rdev, &mddev->disks, same_set)
4122 if (rdev->raid_disk >= 0) {
4124 sprintf(nm, "rd%d", rdev->raid_disk);
4125 sysfs_remove_link(&mddev->kobj, nm);
4128 /* make sure all md_delayed_delete calls have finished */
4129 flush_scheduled_work();
4131 export_array(mddev);
4133 mddev->array_sectors = 0;
4135 mddev->raid_disks = 0;
4136 mddev->recovery_cp = 0;
4137 mddev->resync_min = 0;
4138 mddev->resync_max = MaxSector;
4139 mddev->reshape_position = MaxSector;
4140 mddev->external = 0;
4141 mddev->persistent = 0;
4142 mddev->level = LEVEL_NONE;
4143 mddev->clevel[0] = 0;
4146 mddev->metadata_type[0] = 0;
4147 mddev->chunk_size = 0;
4148 mddev->ctime = mddev->utime = 0;
4150 mddev->max_disks = 0;
4152 mddev->delta_disks = 0;
4153 mddev->new_level = LEVEL_NONE;
4154 mddev->new_layout = 0;
4155 mddev->new_chunk = 0;
4156 mddev->curr_resync = 0;
4157 mddev->resync_mismatches = 0;
4158 mddev->suspend_lo = mddev->suspend_hi = 0;
4159 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4160 mddev->recovery = 0;
4163 mddev->degraded = 0;
4164 mddev->barriers_work = 0;
4165 mddev->safemode = 0;
4166 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4167 if (mddev->hold_active == UNTIL_STOP)
4168 mddev->hold_active = 0;
4170 } else if (mddev->pers)
4171 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4174 blk_integrity_unregister(disk);
4175 md_new_event(mddev);
4176 sysfs_notify_dirent(mddev->sysfs_state);
4182 static void autorun_array(mddev_t *mddev)
4187 if (list_empty(&mddev->disks))
4190 printk(KERN_INFO "md: running: ");
4192 list_for_each_entry(rdev, &mddev->disks, same_set) {
4193 char b[BDEVNAME_SIZE];
4194 printk("<%s>", bdevname(rdev->bdev,b));
4198 err = do_md_run(mddev);
4200 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4201 do_md_stop(mddev, 0, 0);
4206 * lets try to run arrays based on all disks that have arrived
4207 * until now. (those are in pending_raid_disks)
4209 * the method: pick the first pending disk, collect all disks with
4210 * the same UUID, remove all from the pending list and put them into
4211 * the 'same_array' list. Then order this list based on superblock
4212 * update time (freshest comes first), kick out 'old' disks and
4213 * compare superblocks. If everything's fine then run it.
4215 * If "unit" is allocated, then bump its reference count
4217 static void autorun_devices(int part)
4219 mdk_rdev_t *rdev0, *rdev, *tmp;
4221 char b[BDEVNAME_SIZE];
4223 printk(KERN_INFO "md: autorun ...\n");
4224 while (!list_empty(&pending_raid_disks)) {
4227 LIST_HEAD(candidates);
4228 rdev0 = list_entry(pending_raid_disks.next,
4229 mdk_rdev_t, same_set);
4231 printk(KERN_INFO "md: considering %s ...\n",
4232 bdevname(rdev0->bdev,b));
4233 INIT_LIST_HEAD(&candidates);
4234 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4235 if (super_90_load(rdev, rdev0, 0) >= 0) {
4236 printk(KERN_INFO "md: adding %s ...\n",
4237 bdevname(rdev->bdev,b));
4238 list_move(&rdev->same_set, &candidates);
4241 * now we have a set of devices, with all of them having
4242 * mostly sane superblocks. It's time to allocate the
4246 dev = MKDEV(mdp_major,
4247 rdev0->preferred_minor << MdpMinorShift);
4248 unit = MINOR(dev) >> MdpMinorShift;
4250 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4253 if (rdev0->preferred_minor != unit) {
4254 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4255 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4259 md_probe(dev, NULL, NULL);
4260 mddev = mddev_find(dev);
4261 if (!mddev || !mddev->gendisk) {
4265 "md: cannot allocate memory for md drive.\n");
4268 if (mddev_lock(mddev))
4269 printk(KERN_WARNING "md: %s locked, cannot run\n",
4271 else if (mddev->raid_disks || mddev->major_version
4272 || !list_empty(&mddev->disks)) {
4274 "md: %s already running, cannot run %s\n",
4275 mdname(mddev), bdevname(rdev0->bdev,b));
4276 mddev_unlock(mddev);
4278 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4279 mddev->persistent = 1;
4280 rdev_for_each_list(rdev, tmp, &candidates) {
4281 list_del_init(&rdev->same_set);
4282 if (bind_rdev_to_array(rdev, mddev))
4285 autorun_array(mddev);
4286 mddev_unlock(mddev);
4288 /* on success, candidates will be empty, on error
4291 rdev_for_each_list(rdev, tmp, &candidates) {
4292 list_del_init(&rdev->same_set);
4297 printk(KERN_INFO "md: ... autorun DONE.\n");
4299 #endif /* !MODULE */
4301 static int get_version(void __user * arg)
4305 ver.major = MD_MAJOR_VERSION;
4306 ver.minor = MD_MINOR_VERSION;
4307 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4309 if (copy_to_user(arg, &ver, sizeof(ver)))
4315 static int get_array_info(mddev_t * mddev, void __user * arg)
4317 mdu_array_info_t info;
4318 int nr,working,active,failed,spare;
4321 nr=working=active=failed=spare=0;
4322 list_for_each_entry(rdev, &mddev->disks, same_set) {
4324 if (test_bit(Faulty, &rdev->flags))
4328 if (test_bit(In_sync, &rdev->flags))
4335 info.major_version = mddev->major_version;
4336 info.minor_version = mddev->minor_version;
4337 info.patch_version = MD_PATCHLEVEL_VERSION;
4338 info.ctime = mddev->ctime;
4339 info.level = mddev->level;
4340 info.size = mddev->size;
4341 if (info.size != mddev->size) /* overflow */
4344 info.raid_disks = mddev->raid_disks;
4345 info.md_minor = mddev->md_minor;
4346 info.not_persistent= !mddev->persistent;
4348 info.utime = mddev->utime;
4351 info.state = (1<<MD_SB_CLEAN);
4352 if (mddev->bitmap && mddev->bitmap_offset)
4353 info.state = (1<<MD_SB_BITMAP_PRESENT);
4354 info.active_disks = active;
4355 info.working_disks = working;
4356 info.failed_disks = failed;
4357 info.spare_disks = spare;
4359 info.layout = mddev->layout;
4360 info.chunk_size = mddev->chunk_size;
4362 if (copy_to_user(arg, &info, sizeof(info)))
4368 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4370 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4371 char *ptr, *buf = NULL;
4374 if (md_allow_write(mddev))
4375 file = kmalloc(sizeof(*file), GFP_NOIO);
4377 file = kmalloc(sizeof(*file), GFP_KERNEL);
4382 /* bitmap disabled, zero the first byte and copy out */
4383 if (!mddev->bitmap || !mddev->bitmap->file) {
4384 file->pathname[0] = '\0';
4388 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4392 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4396 strcpy(file->pathname, ptr);
4400 if (copy_to_user(arg, file, sizeof(*file)))
4408 static int get_disk_info(mddev_t * mddev, void __user * arg)
4410 mdu_disk_info_t info;
4413 if (copy_from_user(&info, arg, sizeof(info)))
4416 rdev = find_rdev_nr(mddev, info.number);
4418 info.major = MAJOR(rdev->bdev->bd_dev);
4419 info.minor = MINOR(rdev->bdev->bd_dev);
4420 info.raid_disk = rdev->raid_disk;
4422 if (test_bit(Faulty, &rdev->flags))
4423 info.state |= (1<<MD_DISK_FAULTY);
4424 else if (test_bit(In_sync, &rdev->flags)) {
4425 info.state |= (1<<MD_DISK_ACTIVE);
4426 info.state |= (1<<MD_DISK_SYNC);
4428 if (test_bit(WriteMostly, &rdev->flags))
4429 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4431 info.major = info.minor = 0;
4432 info.raid_disk = -1;
4433 info.state = (1<<MD_DISK_REMOVED);
4436 if (copy_to_user(arg, &info, sizeof(info)))
4442 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4444 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4446 dev_t dev = MKDEV(info->major,info->minor);
4448 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4451 if (!mddev->raid_disks) {
4453 /* expecting a device which has a superblock */
4454 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4457 "md: md_import_device returned %ld\n",
4459 return PTR_ERR(rdev);
4461 if (!list_empty(&mddev->disks)) {
4462 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4463 mdk_rdev_t, same_set);
4464 int err = super_types[mddev->major_version]
4465 .load_super(rdev, rdev0, mddev->minor_version);
4468 "md: %s has different UUID to %s\n",
4469 bdevname(rdev->bdev,b),
4470 bdevname(rdev0->bdev,b2));
4475 err = bind_rdev_to_array(rdev, mddev);
4482 * add_new_disk can be used once the array is assembled
4483 * to add "hot spares". They must already have a superblock
4488 if (!mddev->pers->hot_add_disk) {
4490 "%s: personality does not support diskops!\n",
4494 if (mddev->persistent)
4495 rdev = md_import_device(dev, mddev->major_version,
4496 mddev->minor_version);
4498 rdev = md_import_device(dev, -1, -1);
4501 "md: md_import_device returned %ld\n",
4503 return PTR_ERR(rdev);
4505 /* set save_raid_disk if appropriate */
4506 if (!mddev->persistent) {
4507 if (info->state & (1<<MD_DISK_SYNC) &&
4508 info->raid_disk < mddev->raid_disks)
4509 rdev->raid_disk = info->raid_disk;
4511 rdev->raid_disk = -1;
4513 super_types[mddev->major_version].
4514 validate_super(mddev, rdev);
4515 rdev->saved_raid_disk = rdev->raid_disk;
4517 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4518 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4519 set_bit(WriteMostly, &rdev->flags);
4521 rdev->raid_disk = -1;
4522 err = bind_rdev_to_array(rdev, mddev);
4523 if (!err && !mddev->pers->hot_remove_disk) {
4524 /* If there is hot_add_disk but no hot_remove_disk
4525 * then added disks for geometry changes,
4526 * and should be added immediately.
4528 super_types[mddev->major_version].
4529 validate_super(mddev, rdev);
4530 err = mddev->pers->hot_add_disk(mddev, rdev);
4532 unbind_rdev_from_array(rdev);
4537 sysfs_notify_dirent(rdev->sysfs_state);
4539 md_update_sb(mddev, 1);
4540 if (mddev->degraded)
4541 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4542 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4543 md_wakeup_thread(mddev->thread);
4547 /* otherwise, add_new_disk is only allowed
4548 * for major_version==0 superblocks
4550 if (mddev->major_version != 0) {
4551 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4556 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4558 rdev = md_import_device(dev, -1, 0);
4561 "md: error, md_import_device() returned %ld\n",
4563 return PTR_ERR(rdev);
4565 rdev->desc_nr = info->number;
4566 if (info->raid_disk < mddev->raid_disks)
4567 rdev->raid_disk = info->raid_disk;
4569 rdev->raid_disk = -1;
4571 if (rdev->raid_disk < mddev->raid_disks)
4572 if (info->state & (1<<MD_DISK_SYNC))
4573 set_bit(In_sync, &rdev->flags);
4575 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4576 set_bit(WriteMostly, &rdev->flags);
4578 if (!mddev->persistent) {
4579 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4580 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4582 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4583 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4585 err = bind_rdev_to_array(rdev, mddev);
4595 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4597 char b[BDEVNAME_SIZE];
4600 rdev = find_rdev(mddev, dev);
4604 if (rdev->raid_disk >= 0)
4607 kick_rdev_from_array(rdev);
4608 md_update_sb(mddev, 1);
4609 md_new_event(mddev);
4613 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4614 bdevname(rdev->bdev,b), mdname(mddev));
4618 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4620 char b[BDEVNAME_SIZE];
4627 if (mddev->major_version != 0) {
4628 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4629 " version-0 superblocks.\n",
4633 if (!mddev->pers->hot_add_disk) {
4635 "%s: personality does not support diskops!\n",
4640 rdev = md_import_device(dev, -1, 0);
4643 "md: error, md_import_device() returned %ld\n",
4648 if (mddev->persistent)
4649 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4651 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4653 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4655 if (test_bit(Faulty, &rdev->flags)) {
4657 "md: can not hot-add faulty %s disk to %s!\n",
4658 bdevname(rdev->bdev,b), mdname(mddev));
4662 clear_bit(In_sync, &rdev->flags);
4664 rdev->saved_raid_disk = -1;
4665 err = bind_rdev_to_array(rdev, mddev);
4670 * The rest should better be atomic, we can have disk failures
4671 * noticed in interrupt contexts ...
4674 rdev->raid_disk = -1;
4676 md_update_sb(mddev, 1);
4679 * Kick recovery, maybe this spare has to be added to the
4680 * array immediately.
4682 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4683 md_wakeup_thread(mddev->thread);
4684 md_new_event(mddev);
4692 static int set_bitmap_file(mddev_t *mddev, int fd)
4697 if (!mddev->pers->quiesce)
4699 if (mddev->recovery || mddev->sync_thread)
4701 /* we should be able to change the bitmap.. */
4707 return -EEXIST; /* cannot add when bitmap is present */
4708 mddev->bitmap_file = fget(fd);
4710 if (mddev->bitmap_file == NULL) {
4711 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4716 err = deny_bitmap_write_access(mddev->bitmap_file);
4718 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4720 fput(mddev->bitmap_file);
4721 mddev->bitmap_file = NULL;
4724 mddev->bitmap_offset = 0; /* file overrides offset */
4725 } else if (mddev->bitmap == NULL)
4726 return -ENOENT; /* cannot remove what isn't there */
4729 mddev->pers->quiesce(mddev, 1);
4731 err = bitmap_create(mddev);
4732 if (fd < 0 || err) {
4733 bitmap_destroy(mddev);
4734 fd = -1; /* make sure to put the file */
4736 mddev->pers->quiesce(mddev, 0);
4739 if (mddev->bitmap_file) {
4740 restore_bitmap_write_access(mddev->bitmap_file);
4741 fput(mddev->bitmap_file);
4743 mddev->bitmap_file = NULL;
4750 * set_array_info is used two different ways
4751 * The original usage is when creating a new array.
4752 * In this usage, raid_disks is > 0 and it together with
4753 * level, size, not_persistent,layout,chunksize determine the
4754 * shape of the array.
4755 * This will always create an array with a type-0.90.0 superblock.
4756 * The newer usage is when assembling an array.
4757 * In this case raid_disks will be 0, and the major_version field is
4758 * use to determine which style super-blocks are to be found on the devices.
4759 * The minor and patch _version numbers are also kept incase the
4760 * super_block handler wishes to interpret them.
4762 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4765 if (info->raid_disks == 0) {
4766 /* just setting version number for superblock loading */
4767 if (info->major_version < 0 ||
4768 info->major_version >= ARRAY_SIZE(super_types) ||
4769 super_types[info->major_version].name == NULL) {
4770 /* maybe try to auto-load a module? */
4772 "md: superblock version %d not known\n",
4773 info->major_version);
4776 mddev->major_version = info->major_version;
4777 mddev->minor_version = info->minor_version;
4778 mddev->patch_version = info->patch_version;
4779 mddev->persistent = !info->not_persistent;
4782 mddev->major_version = MD_MAJOR_VERSION;
4783 mddev->minor_version = MD_MINOR_VERSION;
4784 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4785 mddev->ctime = get_seconds();
4787 mddev->level = info->level;
4788 mddev->clevel[0] = 0;
4789 mddev->size = info->size;
4790 mddev->raid_disks = info->raid_disks;
4791 /* don't set md_minor, it is determined by which /dev/md* was
4794 if (info->state & (1<<MD_SB_CLEAN))
4795 mddev->recovery_cp = MaxSector;
4797 mddev->recovery_cp = 0;
4798 mddev->persistent = ! info->not_persistent;
4799 mddev->external = 0;
4801 mddev->layout = info->layout;
4802 mddev->chunk_size = info->chunk_size;
4804 mddev->max_disks = MD_SB_DISKS;
4806 if (mddev->persistent)
4808 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4810 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4811 mddev->bitmap_offset = 0;
4813 mddev->reshape_position = MaxSector;
4816 * Generate a 128 bit UUID
4818 get_random_bytes(mddev->uuid, 16);
4820 mddev->new_level = mddev->level;
4821 mddev->new_chunk = mddev->chunk_size;
4822 mddev->new_layout = mddev->layout;
4823 mddev->delta_disks = 0;
4828 static int update_size(mddev_t *mddev, sector_t num_sectors)
4832 int fit = (num_sectors == 0);
4834 if (mddev->pers->resize == NULL)
4836 /* The "num_sectors" is the number of sectors of each device that
4837 * is used. This can only make sense for arrays with redundancy.
4838 * linear and raid0 always use whatever space is available. We can only
4839 * consider changing this number if no resync or reconstruction is
4840 * happening, and if the new size is acceptable. It must fit before the
4841 * sb_start or, if that is <data_offset, it must fit before the size
4842 * of each device. If num_sectors is zero, we find the largest size
4846 if (mddev->sync_thread)
4849 /* Sorry, cannot grow a bitmap yet, just remove it,
4853 list_for_each_entry(rdev, &mddev->disks, same_set) {
4855 avail = rdev->size * 2;
4857 if (fit && (num_sectors == 0 || num_sectors > avail))
4858 num_sectors = avail;
4859 if (avail < num_sectors)
4862 rv = mddev->pers->resize(mddev, num_sectors);
4864 struct block_device *bdev;
4866 bdev = bdget_disk(mddev->gendisk, 0);
4868 mutex_lock(&bdev->bd_inode->i_mutex);
4869 i_size_write(bdev->bd_inode,
4870 (loff_t)mddev->array_sectors << 9);
4871 mutex_unlock(&bdev->bd_inode->i_mutex);
4878 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4881 /* change the number of raid disks */
4882 if (mddev->pers->check_reshape == NULL)
4884 if (raid_disks <= 0 ||
4885 raid_disks >= mddev->max_disks)
4887 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4889 mddev->delta_disks = raid_disks - mddev->raid_disks;
4891 rv = mddev->pers->check_reshape(mddev);
4897 * update_array_info is used to change the configuration of an
4899 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4900 * fields in the info are checked against the array.
4901 * Any differences that cannot be handled will cause an error.
4902 * Normally, only one change can be managed at a time.
4904 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4910 /* calculate expected state,ignoring low bits */
4911 if (mddev->bitmap && mddev->bitmap_offset)
4912 state |= (1 << MD_SB_BITMAP_PRESENT);
4914 if (mddev->major_version != info->major_version ||
4915 mddev->minor_version != info->minor_version ||
4916 /* mddev->patch_version != info->patch_version || */
4917 mddev->ctime != info->ctime ||
4918 mddev->level != info->level ||
4919 /* mddev->layout != info->layout || */
4920 !mddev->persistent != info->not_persistent||
4921 mddev->chunk_size != info->chunk_size ||
4922 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4923 ((state^info->state) & 0xfffffe00)
4926 /* Check there is only one change */
4927 if (info->size >= 0 && mddev->size != info->size) cnt++;
4928 if (mddev->raid_disks != info->raid_disks) cnt++;
4929 if (mddev->layout != info->layout) cnt++;
4930 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4931 if (cnt == 0) return 0;
4932 if (cnt > 1) return -EINVAL;
4934 if (mddev->layout != info->layout) {
4936 * we don't need to do anything at the md level, the
4937 * personality will take care of it all.
4939 if (mddev->pers->reconfig == NULL)
4942 return mddev->pers->reconfig(mddev, info->layout, -1);
4944 if (info->size >= 0 && mddev->size != info->size)
4945 rv = update_size(mddev, (sector_t)info->size * 2);
4947 if (mddev->raid_disks != info->raid_disks)
4948 rv = update_raid_disks(mddev, info->raid_disks);
4950 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4951 if (mddev->pers->quiesce == NULL)
4953 if (mddev->recovery || mddev->sync_thread)
4955 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4956 /* add the bitmap */
4959 if (mddev->default_bitmap_offset == 0)
4961 mddev->bitmap_offset = mddev->default_bitmap_offset;
4962 mddev->pers->quiesce(mddev, 1);
4963 rv = bitmap_create(mddev);
4965 bitmap_destroy(mddev);
4966 mddev->pers->quiesce(mddev, 0);
4968 /* remove the bitmap */
4971 if (mddev->bitmap->file)
4973 mddev->pers->quiesce(mddev, 1);
4974 bitmap_destroy(mddev);
4975 mddev->pers->quiesce(mddev, 0);
4976 mddev->bitmap_offset = 0;
4979 md_update_sb(mddev, 1);
4983 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4987 if (mddev->pers == NULL)
4990 rdev = find_rdev(mddev, dev);
4994 md_error(mddev, rdev);
4999 * We have a problem here : there is no easy way to give a CHS
5000 * virtual geometry. We currently pretend that we have a 2 heads
5001 * 4 sectors (with a BIG number of cylinders...). This drives
5002 * dosfs just mad... ;-)
5004 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5006 mddev_t *mddev = bdev->bd_disk->private_data;
5010 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5014 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5015 unsigned int cmd, unsigned long arg)
5018 void __user *argp = (void __user *)arg;
5019 mddev_t *mddev = NULL;
5021 if (!capable(CAP_SYS_ADMIN))
5025 * Commands dealing with the RAID driver but not any
5031 err = get_version(argp);
5034 case PRINT_RAID_DEBUG:
5042 autostart_arrays(arg);
5049 * Commands creating/starting a new array:
5052 mddev = bdev->bd_disk->private_data;
5059 err = mddev_lock(mddev);
5062 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5069 case SET_ARRAY_INFO:
5071 mdu_array_info_t info;
5073 memset(&info, 0, sizeof(info));
5074 else if (copy_from_user(&info, argp, sizeof(info))) {
5079 err = update_array_info(mddev, &info);
5081 printk(KERN_WARNING "md: couldn't update"
5082 " array info. %d\n", err);
5087 if (!list_empty(&mddev->disks)) {
5089 "md: array %s already has disks!\n",
5094 if (mddev->raid_disks) {
5096 "md: array %s already initialised!\n",
5101 err = set_array_info(mddev, &info);
5103 printk(KERN_WARNING "md: couldn't set"
5104 " array info. %d\n", err);
5114 * Commands querying/configuring an existing array:
5116 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5117 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5118 if ((!mddev->raid_disks && !mddev->external)
5119 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5120 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5121 && cmd != GET_BITMAP_FILE) {
5127 * Commands even a read-only array can execute:
5131 case GET_ARRAY_INFO:
5132 err = get_array_info(mddev, argp);
5135 case GET_BITMAP_FILE:
5136 err = get_bitmap_file(mddev, argp);
5140 err = get_disk_info(mddev, argp);
5143 case RESTART_ARRAY_RW:
5144 err = restart_array(mddev);
5148 err = do_md_stop(mddev, 0, 1);
5152 err = do_md_stop(mddev, 1, 1);
5158 * The remaining ioctls are changing the state of the
5159 * superblock, so we do not allow them on read-only arrays.
5160 * However non-MD ioctls (e.g. get-size) will still come through
5161 * here and hit the 'default' below, so only disallow
5162 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5164 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5165 if (mddev->ro == 2) {
5167 sysfs_notify_dirent(mddev->sysfs_state);
5168 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5169 md_wakeup_thread(mddev->thread);
5180 mdu_disk_info_t info;
5181 if (copy_from_user(&info, argp, sizeof(info)))
5184 err = add_new_disk(mddev, &info);
5188 case HOT_REMOVE_DISK:
5189 err = hot_remove_disk(mddev, new_decode_dev(arg));
5193 err = hot_add_disk(mddev, new_decode_dev(arg));
5196 case SET_DISK_FAULTY:
5197 err = set_disk_faulty(mddev, new_decode_dev(arg));
5201 err = do_md_run(mddev);
5204 case SET_BITMAP_FILE:
5205 err = set_bitmap_file(mddev, (int)arg);
5215 if (mddev->hold_active == UNTIL_IOCTL &&
5217 mddev->hold_active = 0;
5218 mddev_unlock(mddev);
5228 static int md_open(struct block_device *bdev, fmode_t mode)
5231 * Succeed if we can lock the mddev, which confirms that
5232 * it isn't being stopped right now.
5234 mddev_t *mddev = mddev_find(bdev->bd_dev);
5237 if (mddev->gendisk != bdev->bd_disk) {
5238 /* we are racing with mddev_put which is discarding this
5242 /* Wait until bdev->bd_disk is definitely gone */
5243 flush_scheduled_work();
5244 /* Then retry the open from the top */
5245 return -ERESTARTSYS;
5247 BUG_ON(mddev != bdev->bd_disk->private_data);
5249 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5253 atomic_inc(&mddev->openers);
5254 mddev_unlock(mddev);
5256 check_disk_change(bdev);
5261 static int md_release(struct gendisk *disk, fmode_t mode)
5263 mddev_t *mddev = disk->private_data;
5266 atomic_dec(&mddev->openers);
5272 static int md_media_changed(struct gendisk *disk)
5274 mddev_t *mddev = disk->private_data;
5276 return mddev->changed;
5279 static int md_revalidate(struct gendisk *disk)
5281 mddev_t *mddev = disk->private_data;
5286 static struct block_device_operations md_fops =
5288 .owner = THIS_MODULE,
5290 .release = md_release,
5291 .locked_ioctl = md_ioctl,
5292 .getgeo = md_getgeo,
5293 .media_changed = md_media_changed,
5294 .revalidate_disk= md_revalidate,
5297 static int md_thread(void * arg)
5299 mdk_thread_t *thread = arg;
5302 * md_thread is a 'system-thread', it's priority should be very
5303 * high. We avoid resource deadlocks individually in each
5304 * raid personality. (RAID5 does preallocation) We also use RR and
5305 * the very same RT priority as kswapd, thus we will never get
5306 * into a priority inversion deadlock.
5308 * we definitely have to have equal or higher priority than
5309 * bdflush, otherwise bdflush will deadlock if there are too
5310 * many dirty RAID5 blocks.
5313 allow_signal(SIGKILL);
5314 while (!kthread_should_stop()) {
5316 /* We need to wait INTERRUPTIBLE so that
5317 * we don't add to the load-average.
5318 * That means we need to be sure no signals are
5321 if (signal_pending(current))
5322 flush_signals(current);
5324 wait_event_interruptible_timeout
5326 test_bit(THREAD_WAKEUP, &thread->flags)
5327 || kthread_should_stop(),
5330 clear_bit(THREAD_WAKEUP, &thread->flags);
5332 thread->run(thread->mddev);
5338 void md_wakeup_thread(mdk_thread_t *thread)
5341 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5342 set_bit(THREAD_WAKEUP, &thread->flags);
5343 wake_up(&thread->wqueue);
5347 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5350 mdk_thread_t *thread;
5352 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5356 init_waitqueue_head(&thread->wqueue);
5359 thread->mddev = mddev;
5360 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5361 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5362 if (IS_ERR(thread->tsk)) {
5369 void md_unregister_thread(mdk_thread_t *thread)
5371 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5373 kthread_stop(thread->tsk);
5377 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5384 if (!rdev || test_bit(Faulty, &rdev->flags))
5387 if (mddev->external)
5388 set_bit(Blocked, &rdev->flags);
5390 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5392 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5393 __builtin_return_address(0),__builtin_return_address(1),
5394 __builtin_return_address(2),__builtin_return_address(3));
5398 if (!mddev->pers->error_handler)
5400 mddev->pers->error_handler(mddev,rdev);
5401 if (mddev->degraded)
5402 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5403 set_bit(StateChanged, &rdev->flags);
5404 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5406 md_wakeup_thread(mddev->thread);
5407 md_new_event_inintr(mddev);
5410 /* seq_file implementation /proc/mdstat */
5412 static void status_unused(struct seq_file *seq)
5417 seq_printf(seq, "unused devices: ");
5419 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5420 char b[BDEVNAME_SIZE];
5422 seq_printf(seq, "%s ",
5423 bdevname(rdev->bdev,b));
5426 seq_printf(seq, "<none>");
5428 seq_printf(seq, "\n");
5432 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5434 sector_t max_blocks, resync, res;
5435 unsigned long dt, db, rt;
5437 unsigned int per_milli;
5439 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5441 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5442 max_blocks = mddev->resync_max_sectors >> 1;
5444 max_blocks = mddev->size;
5447 * Should not happen.
5453 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5454 * in a sector_t, and (max_blocks>>scale) will fit in a
5455 * u32, as those are the requirements for sector_div.
5456 * Thus 'scale' must be at least 10
5459 if (sizeof(sector_t) > sizeof(unsigned long)) {
5460 while ( max_blocks/2 > (1ULL<<(scale+32)))
5463 res = (resync>>scale)*1000;
5464 sector_div(res, (u32)((max_blocks>>scale)+1));
5468 int i, x = per_milli/50, y = 20-x;
5469 seq_printf(seq, "[");
5470 for (i = 0; i < x; i++)
5471 seq_printf(seq, "=");
5472 seq_printf(seq, ">");
5473 for (i = 0; i < y; i++)
5474 seq_printf(seq, ".");
5475 seq_printf(seq, "] ");
5477 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5478 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5480 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5482 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5483 "resync" : "recovery"))),
5484 per_milli/10, per_milli % 10,
5485 (unsigned long long) resync,
5486 (unsigned long long) max_blocks);
5489 * We do not want to overflow, so the order of operands and
5490 * the * 100 / 100 trick are important. We do a +1 to be
5491 * safe against division by zero. We only estimate anyway.
5493 * dt: time from mark until now
5494 * db: blocks written from mark until now
5495 * rt: remaining time
5497 dt = ((jiffies - mddev->resync_mark) / HZ);
5499 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5500 - mddev->resync_mark_cnt;
5501 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5503 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5505 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5508 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5510 struct list_head *tmp;
5520 spin_lock(&all_mddevs_lock);
5521 list_for_each(tmp,&all_mddevs)
5523 mddev = list_entry(tmp, mddev_t, all_mddevs);
5525 spin_unlock(&all_mddevs_lock);
5528 spin_unlock(&all_mddevs_lock);
5530 return (void*)2;/* tail */
5534 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5536 struct list_head *tmp;
5537 mddev_t *next_mddev, *mddev = v;
5543 spin_lock(&all_mddevs_lock);
5545 tmp = all_mddevs.next;
5547 tmp = mddev->all_mddevs.next;
5548 if (tmp != &all_mddevs)
5549 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5551 next_mddev = (void*)2;
5554 spin_unlock(&all_mddevs_lock);
5562 static void md_seq_stop(struct seq_file *seq, void *v)
5566 if (mddev && v != (void*)1 && v != (void*)2)
5570 struct mdstat_info {
5574 static int md_seq_show(struct seq_file *seq, void *v)
5579 struct mdstat_info *mi = seq->private;
5580 struct bitmap *bitmap;
5582 if (v == (void*)1) {
5583 struct mdk_personality *pers;
5584 seq_printf(seq, "Personalities : ");
5585 spin_lock(&pers_lock);
5586 list_for_each_entry(pers, &pers_list, list)
5587 seq_printf(seq, "[%s] ", pers->name);
5589 spin_unlock(&pers_lock);
5590 seq_printf(seq, "\n");
5591 mi->event = atomic_read(&md_event_count);
5594 if (v == (void*)2) {
5599 if (mddev_lock(mddev) < 0)
5602 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5603 seq_printf(seq, "%s : %sactive", mdname(mddev),
5604 mddev->pers ? "" : "in");
5607 seq_printf(seq, " (read-only)");
5609 seq_printf(seq, " (auto-read-only)");
5610 seq_printf(seq, " %s", mddev->pers->name);
5614 list_for_each_entry(rdev, &mddev->disks, same_set) {
5615 char b[BDEVNAME_SIZE];
5616 seq_printf(seq, " %s[%d]",
5617 bdevname(rdev->bdev,b), rdev->desc_nr);
5618 if (test_bit(WriteMostly, &rdev->flags))
5619 seq_printf(seq, "(W)");
5620 if (test_bit(Faulty, &rdev->flags)) {
5621 seq_printf(seq, "(F)");
5623 } else if (rdev->raid_disk < 0)
5624 seq_printf(seq, "(S)"); /* spare */
5628 if (!list_empty(&mddev->disks)) {
5630 seq_printf(seq, "\n %llu blocks",
5631 (unsigned long long)
5632 mddev->array_sectors / 2);
5634 seq_printf(seq, "\n %llu blocks",
5635 (unsigned long long)size);
5637 if (mddev->persistent) {
5638 if (mddev->major_version != 0 ||
5639 mddev->minor_version != 90) {
5640 seq_printf(seq," super %d.%d",
5641 mddev->major_version,
5642 mddev->minor_version);
5644 } else if (mddev->external)
5645 seq_printf(seq, " super external:%s",
5646 mddev->metadata_type);
5648 seq_printf(seq, " super non-persistent");
5651 mddev->pers->status(seq, mddev);
5652 seq_printf(seq, "\n ");
5653 if (mddev->pers->sync_request) {
5654 if (mddev->curr_resync > 2) {
5655 status_resync(seq, mddev);
5656 seq_printf(seq, "\n ");
5657 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5658 seq_printf(seq, "\tresync=DELAYED\n ");
5659 else if (mddev->recovery_cp < MaxSector)
5660 seq_printf(seq, "\tresync=PENDING\n ");
5663 seq_printf(seq, "\n ");
5665 if ((bitmap = mddev->bitmap)) {
5666 unsigned long chunk_kb;
5667 unsigned long flags;
5668 spin_lock_irqsave(&bitmap->lock, flags);
5669 chunk_kb = bitmap->chunksize >> 10;
5670 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5672 bitmap->pages - bitmap->missing_pages,
5674 (bitmap->pages - bitmap->missing_pages)
5675 << (PAGE_SHIFT - 10),
5676 chunk_kb ? chunk_kb : bitmap->chunksize,
5677 chunk_kb ? "KB" : "B");
5679 seq_printf(seq, ", file: ");
5680 seq_path(seq, &bitmap->file->f_path, " \t\n");
5683 seq_printf(seq, "\n");
5684 spin_unlock_irqrestore(&bitmap->lock, flags);
5687 seq_printf(seq, "\n");
5689 mddev_unlock(mddev);
5694 static struct seq_operations md_seq_ops = {
5695 .start = md_seq_start,
5696 .next = md_seq_next,
5697 .stop = md_seq_stop,
5698 .show = md_seq_show,
5701 static int md_seq_open(struct inode *inode, struct file *file)
5704 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5708 error = seq_open(file, &md_seq_ops);
5712 struct seq_file *p = file->private_data;
5714 mi->event = atomic_read(&md_event_count);
5719 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5721 struct seq_file *m = filp->private_data;
5722 struct mdstat_info *mi = m->private;
5725 poll_wait(filp, &md_event_waiters, wait);
5727 /* always allow read */
5728 mask = POLLIN | POLLRDNORM;
5730 if (mi->event != atomic_read(&md_event_count))
5731 mask |= POLLERR | POLLPRI;
5735 static const struct file_operations md_seq_fops = {
5736 .owner = THIS_MODULE,
5737 .open = md_seq_open,
5739 .llseek = seq_lseek,
5740 .release = seq_release_private,
5741 .poll = mdstat_poll,
5744 int register_md_personality(struct mdk_personality *p)
5746 spin_lock(&pers_lock);
5747 list_add_tail(&p->list, &pers_list);
5748 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5749 spin_unlock(&pers_lock);
5753 int unregister_md_personality(struct mdk_personality *p)
5755 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5756 spin_lock(&pers_lock);
5757 list_del_init(&p->list);
5758 spin_unlock(&pers_lock);
5762 static int is_mddev_idle(mddev_t *mddev, int init)
5770 rdev_for_each_rcu(rdev, mddev) {
5771 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5772 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
5773 (int)part_stat_read(&disk->part0, sectors[1]) -
5774 atomic_read(&disk->sync_io);
5775 /* sync IO will cause sync_io to increase before the disk_stats
5776 * as sync_io is counted when a request starts, and
5777 * disk_stats is counted when it completes.
5778 * So resync activity will cause curr_events to be smaller than
5779 * when there was no such activity.
5780 * non-sync IO will cause disk_stat to increase without
5781 * increasing sync_io so curr_events will (eventually)
5782 * be larger than it was before. Once it becomes
5783 * substantially larger, the test below will cause
5784 * the array to appear non-idle, and resync will slow
5786 * If there is a lot of outstanding resync activity when
5787 * we set last_event to curr_events, then all that activity
5788 * completing might cause the array to appear non-idle
5789 * and resync will be slowed down even though there might
5790 * not have been non-resync activity. This will only
5791 * happen once though. 'last_events' will soon reflect
5792 * the state where there is little or no outstanding
5793 * resync requests, and further resync activity will
5794 * always make curr_events less than last_events.
5797 if (init || curr_events - rdev->last_events > 64) {
5798 rdev->last_events = curr_events;
5806 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5808 /* another "blocks" (512byte) blocks have been synced */
5809 atomic_sub(blocks, &mddev->recovery_active);
5810 wake_up(&mddev->recovery_wait);
5812 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5813 md_wakeup_thread(mddev->thread);
5814 // stop recovery, signal do_sync ....
5819 /* md_write_start(mddev, bi)
5820 * If we need to update some array metadata (e.g. 'active' flag
5821 * in superblock) before writing, schedule a superblock update
5822 * and wait for it to complete.
5824 void md_write_start(mddev_t *mddev, struct bio *bi)
5827 if (bio_data_dir(bi) != WRITE)
5830 BUG_ON(mddev->ro == 1);
5831 if (mddev->ro == 2) {
5832 /* need to switch to read/write */
5834 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5835 md_wakeup_thread(mddev->thread);
5836 md_wakeup_thread(mddev->sync_thread);
5839 atomic_inc(&mddev->writes_pending);
5840 if (mddev->safemode == 1)
5841 mddev->safemode = 0;
5842 if (mddev->in_sync) {
5843 spin_lock_irq(&mddev->write_lock);
5844 if (mddev->in_sync) {
5846 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5847 md_wakeup_thread(mddev->thread);
5850 spin_unlock_irq(&mddev->write_lock);
5853 sysfs_notify_dirent(mddev->sysfs_state);
5854 wait_event(mddev->sb_wait,
5855 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5856 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5859 void md_write_end(mddev_t *mddev)
5861 if (atomic_dec_and_test(&mddev->writes_pending)) {
5862 if (mddev->safemode == 2)
5863 md_wakeup_thread(mddev->thread);
5864 else if (mddev->safemode_delay)
5865 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5869 /* md_allow_write(mddev)
5870 * Calling this ensures that the array is marked 'active' so that writes
5871 * may proceed without blocking. It is important to call this before
5872 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5873 * Must be called with mddev_lock held.
5875 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5876 * is dropped, so return -EAGAIN after notifying userspace.
5878 int md_allow_write(mddev_t *mddev)
5884 if (!mddev->pers->sync_request)
5887 spin_lock_irq(&mddev->write_lock);
5888 if (mddev->in_sync) {
5890 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5891 if (mddev->safemode_delay &&
5892 mddev->safemode == 0)
5893 mddev->safemode = 1;
5894 spin_unlock_irq(&mddev->write_lock);
5895 md_update_sb(mddev, 0);
5896 sysfs_notify_dirent(mddev->sysfs_state);
5898 spin_unlock_irq(&mddev->write_lock);
5900 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5905 EXPORT_SYMBOL_GPL(md_allow_write);
5907 #define SYNC_MARKS 10
5908 #define SYNC_MARK_STEP (3*HZ)
5909 void md_do_sync(mddev_t *mddev)
5912 unsigned int currspeed = 0,
5914 sector_t max_sectors,j, io_sectors;
5915 unsigned long mark[SYNC_MARKS];
5916 sector_t mark_cnt[SYNC_MARKS];
5918 struct list_head *tmp;
5919 sector_t last_check;
5924 /* just incase thread restarts... */
5925 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5927 if (mddev->ro) /* never try to sync a read-only array */
5930 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5931 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5932 desc = "data-check";
5933 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5934 desc = "requested-resync";
5937 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5942 /* we overload curr_resync somewhat here.
5943 * 0 == not engaged in resync at all
5944 * 2 == checking that there is no conflict with another sync
5945 * 1 == like 2, but have yielded to allow conflicting resync to
5947 * other == active in resync - this many blocks
5949 * Before starting a resync we must have set curr_resync to
5950 * 2, and then checked that every "conflicting" array has curr_resync
5951 * less than ours. When we find one that is the same or higher
5952 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5953 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5954 * This will mean we have to start checking from the beginning again.
5959 mddev->curr_resync = 2;
5962 if (kthread_should_stop()) {
5963 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5966 for_each_mddev(mddev2, tmp) {
5967 if (mddev2 == mddev)
5969 if (!mddev->parallel_resync
5970 && mddev2->curr_resync
5971 && match_mddev_units(mddev, mddev2)) {
5973 if (mddev < mddev2 && mddev->curr_resync == 2) {
5974 /* arbitrarily yield */
5975 mddev->curr_resync = 1;
5976 wake_up(&resync_wait);
5978 if (mddev > mddev2 && mddev->curr_resync == 1)
5979 /* no need to wait here, we can wait the next
5980 * time 'round when curr_resync == 2
5983 /* We need to wait 'interruptible' so as not to
5984 * contribute to the load average, and not to
5985 * be caught by 'softlockup'
5987 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5988 if (!kthread_should_stop() &&
5989 mddev2->curr_resync >= mddev->curr_resync) {
5990 printk(KERN_INFO "md: delaying %s of %s"
5991 " until %s has finished (they"
5992 " share one or more physical units)\n",
5993 desc, mdname(mddev), mdname(mddev2));
5995 if (signal_pending(current))
5996 flush_signals(current);
5998 finish_wait(&resync_wait, &wq);
6001 finish_wait(&resync_wait, &wq);
6004 } while (mddev->curr_resync < 2);
6007 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6008 /* resync follows the size requested by the personality,
6009 * which defaults to physical size, but can be virtual size
6011 max_sectors = mddev->resync_max_sectors;
6012 mddev->resync_mismatches = 0;
6013 /* we don't use the checkpoint if there's a bitmap */
6014 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6015 j = mddev->resync_min;
6016 else if (!mddev->bitmap)
6017 j = mddev->recovery_cp;
6019 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6020 max_sectors = mddev->size << 1;
6022 /* recovery follows the physical size of devices */
6023 max_sectors = mddev->size << 1;
6025 list_for_each_entry(rdev, &mddev->disks, same_set)
6026 if (rdev->raid_disk >= 0 &&
6027 !test_bit(Faulty, &rdev->flags) &&
6028 !test_bit(In_sync, &rdev->flags) &&
6029 rdev->recovery_offset < j)
6030 j = rdev->recovery_offset;
6033 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6034 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6035 " %d KB/sec/disk.\n", speed_min(mddev));
6036 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6037 "(but not more than %d KB/sec) for %s.\n",
6038 speed_max(mddev), desc);
6040 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6043 for (m = 0; m < SYNC_MARKS; m++) {
6045 mark_cnt[m] = io_sectors;
6048 mddev->resync_mark = mark[last_mark];
6049 mddev->resync_mark_cnt = mark_cnt[last_mark];
6052 * Tune reconstruction:
6054 window = 32*(PAGE_SIZE/512);
6055 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6056 window/2,(unsigned long long) max_sectors/2);
6058 atomic_set(&mddev->recovery_active, 0);
6063 "md: resuming %s of %s from checkpoint.\n",
6064 desc, mdname(mddev));
6065 mddev->curr_resync = j;
6068 while (j < max_sectors) {
6072 if (j >= mddev->resync_max) {
6073 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6074 wait_event(mddev->recovery_wait,
6075 mddev->resync_max > j
6076 || kthread_should_stop());
6078 if (kthread_should_stop())
6081 if (mddev->curr_resync > mddev->curr_resync_completed &&
6082 (mddev->curr_resync - mddev->curr_resync_completed)
6083 > (max_sectors >> 4)) {
6084 /* time to update curr_resync_completed */
6085 blk_unplug(mddev->queue);
6086 wait_event(mddev->recovery_wait,
6087 atomic_read(&mddev->recovery_active) == 0);
6088 mddev->curr_resync_completed =
6090 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6092 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6093 currspeed < speed_min(mddev));
6095 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6099 if (!skipped) { /* actual IO requested */
6100 io_sectors += sectors;
6101 atomic_add(sectors, &mddev->recovery_active);
6105 if (j>1) mddev->curr_resync = j;
6106 mddev->curr_mark_cnt = io_sectors;
6107 if (last_check == 0)
6108 /* this is the earliers that rebuilt will be
6109 * visible in /proc/mdstat
6111 md_new_event(mddev);
6113 if (last_check + window > io_sectors || j == max_sectors)
6116 last_check = io_sectors;
6118 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6122 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6124 int next = (last_mark+1) % SYNC_MARKS;
6126 mddev->resync_mark = mark[next];
6127 mddev->resync_mark_cnt = mark_cnt[next];
6128 mark[next] = jiffies;
6129 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6134 if (kthread_should_stop())
6139 * this loop exits only if either when we are slower than
6140 * the 'hard' speed limit, or the system was IO-idle for
6142 * the system might be non-idle CPU-wise, but we only care
6143 * about not overloading the IO subsystem. (things like an
6144 * e2fsck being done on the RAID array should execute fast)
6146 blk_unplug(mddev->queue);
6149 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6150 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6152 if (currspeed > speed_min(mddev)) {
6153 if ((currspeed > speed_max(mddev)) ||
6154 !is_mddev_idle(mddev, 0)) {
6160 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6162 * this also signals 'finished resyncing' to md_stop
6165 blk_unplug(mddev->queue);
6167 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6169 /* tell personality that we are finished */
6170 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6172 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6173 mddev->curr_resync > 2) {
6174 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6175 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6176 if (mddev->curr_resync >= mddev->recovery_cp) {
6178 "md: checkpointing %s of %s.\n",
6179 desc, mdname(mddev));
6180 mddev->recovery_cp = mddev->curr_resync;
6183 mddev->recovery_cp = MaxSector;
6185 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6186 mddev->curr_resync = MaxSector;
6187 list_for_each_entry(rdev, &mddev->disks, same_set)
6188 if (rdev->raid_disk >= 0 &&
6189 !test_bit(Faulty, &rdev->flags) &&
6190 !test_bit(In_sync, &rdev->flags) &&
6191 rdev->recovery_offset < mddev->curr_resync)
6192 rdev->recovery_offset = mddev->curr_resync;
6195 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6198 mddev->curr_resync = 0;
6199 mddev->resync_min = 0;
6200 mddev->resync_max = MaxSector;
6201 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6202 wake_up(&resync_wait);
6203 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6204 md_wakeup_thread(mddev->thread);
6209 * got a signal, exit.
6212 "md: md_do_sync() got signal ... exiting\n");
6213 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6217 EXPORT_SYMBOL_GPL(md_do_sync);
6220 static int remove_and_add_spares(mddev_t *mddev)
6225 mddev->curr_resync_completed = 0;
6227 list_for_each_entry(rdev, &mddev->disks, same_set)
6228 if (rdev->raid_disk >= 0 &&
6229 !test_bit(Blocked, &rdev->flags) &&
6230 (test_bit(Faulty, &rdev->flags) ||
6231 ! test_bit(In_sync, &rdev->flags)) &&
6232 atomic_read(&rdev->nr_pending)==0) {
6233 if (mddev->pers->hot_remove_disk(
6234 mddev, rdev->raid_disk)==0) {
6236 sprintf(nm,"rd%d", rdev->raid_disk);
6237 sysfs_remove_link(&mddev->kobj, nm);
6238 rdev->raid_disk = -1;
6242 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6243 list_for_each_entry(rdev, &mddev->disks, same_set) {
6244 if (rdev->raid_disk >= 0 &&
6245 !test_bit(In_sync, &rdev->flags) &&
6246 !test_bit(Blocked, &rdev->flags))
6248 if (rdev->raid_disk < 0
6249 && !test_bit(Faulty, &rdev->flags)) {
6250 rdev->recovery_offset = 0;
6252 hot_add_disk(mddev, rdev) == 0) {
6254 sprintf(nm, "rd%d", rdev->raid_disk);
6255 if (sysfs_create_link(&mddev->kobj,
6258 "md: cannot register "
6262 md_new_event(mddev);
6271 * This routine is regularly called by all per-raid-array threads to
6272 * deal with generic issues like resync and super-block update.
6273 * Raid personalities that don't have a thread (linear/raid0) do not
6274 * need this as they never do any recovery or update the superblock.
6276 * It does not do any resync itself, but rather "forks" off other threads
6277 * to do that as needed.
6278 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6279 * "->recovery" and create a thread at ->sync_thread.
6280 * When the thread finishes it sets MD_RECOVERY_DONE
6281 * and wakeups up this thread which will reap the thread and finish up.
6282 * This thread also removes any faulty devices (with nr_pending == 0).
6284 * The overall approach is:
6285 * 1/ if the superblock needs updating, update it.
6286 * 2/ If a recovery thread is running, don't do anything else.
6287 * 3/ If recovery has finished, clean up, possibly marking spares active.
6288 * 4/ If there are any faulty devices, remove them.
6289 * 5/ If array is degraded, try to add spares devices
6290 * 6/ If array has spares or is not in-sync, start a resync thread.
6292 void md_check_recovery(mddev_t *mddev)
6298 bitmap_daemon_work(mddev->bitmap);
6303 if (signal_pending(current)) {
6304 if (mddev->pers->sync_request && !mddev->external) {
6305 printk(KERN_INFO "md: %s in immediate safe mode\n",
6307 mddev->safemode = 2;
6309 flush_signals(current);
6312 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6315 (mddev->flags && !mddev->external) ||
6316 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6317 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6318 (mddev->external == 0 && mddev->safemode == 1) ||
6319 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6320 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6324 if (mddev_trylock(mddev)) {
6328 /* Only thing we do on a ro array is remove
6331 remove_and_add_spares(mddev);
6332 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6336 if (!mddev->external) {
6338 spin_lock_irq(&mddev->write_lock);
6339 if (mddev->safemode &&
6340 !atomic_read(&mddev->writes_pending) &&
6342 mddev->recovery_cp == MaxSector) {
6345 if (mddev->persistent)
6346 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6348 if (mddev->safemode == 1)
6349 mddev->safemode = 0;
6350 spin_unlock_irq(&mddev->write_lock);
6352 sysfs_notify_dirent(mddev->sysfs_state);
6356 md_update_sb(mddev, 0);
6358 list_for_each_entry(rdev, &mddev->disks, same_set)
6359 if (test_and_clear_bit(StateChanged, &rdev->flags))
6360 sysfs_notify_dirent(rdev->sysfs_state);
6363 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6364 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6365 /* resync/recovery still happening */
6366 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6369 if (mddev->sync_thread) {
6370 /* resync has finished, collect result */
6371 md_unregister_thread(mddev->sync_thread);
6372 mddev->sync_thread = NULL;
6373 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6374 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6376 /* activate any spares */
6377 if (mddev->pers->spare_active(mddev))
6378 sysfs_notify(&mddev->kobj, NULL,
6381 md_update_sb(mddev, 1);
6383 /* if array is no-longer degraded, then any saved_raid_disk
6384 * information must be scrapped
6386 if (!mddev->degraded)
6387 list_for_each_entry(rdev, &mddev->disks, same_set)
6388 rdev->saved_raid_disk = -1;
6390 mddev->recovery = 0;
6391 /* flag recovery needed just to double check */
6392 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6393 sysfs_notify_dirent(mddev->sysfs_action);
6394 md_new_event(mddev);
6397 /* Set RUNNING before clearing NEEDED to avoid
6398 * any transients in the value of "sync_action".
6400 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6401 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6402 /* Clear some bits that don't mean anything, but
6405 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6406 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6408 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6410 /* no recovery is running.
6411 * remove any failed drives, then
6412 * add spares if possible.
6413 * Spare are also removed and re-added, to allow
6414 * the personality to fail the re-add.
6417 if (mddev->reshape_position != MaxSector) {
6418 if (mddev->pers->check_reshape(mddev) != 0)
6419 /* Cannot proceed */
6421 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6422 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6423 } else if ((spares = remove_and_add_spares(mddev))) {
6424 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6425 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6426 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6427 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6428 } else if (mddev->recovery_cp < MaxSector) {
6429 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6430 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6431 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6432 /* nothing to be done ... */
6435 if (mddev->pers->sync_request) {
6436 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6437 /* We are adding a device or devices to an array
6438 * which has the bitmap stored on all devices.
6439 * So make sure all bitmap pages get written
6441 bitmap_write_all(mddev->bitmap);
6443 mddev->sync_thread = md_register_thread(md_do_sync,
6446 if (!mddev->sync_thread) {
6447 printk(KERN_ERR "%s: could not start resync"
6450 /* leave the spares where they are, it shouldn't hurt */
6451 mddev->recovery = 0;
6453 md_wakeup_thread(mddev->sync_thread);
6454 sysfs_notify_dirent(mddev->sysfs_action);
6455 md_new_event(mddev);
6458 if (!mddev->sync_thread) {
6459 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6460 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6462 if (mddev->sysfs_action)
6463 sysfs_notify_dirent(mddev->sysfs_action);
6465 mddev_unlock(mddev);
6469 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6471 sysfs_notify_dirent(rdev->sysfs_state);
6472 wait_event_timeout(rdev->blocked_wait,
6473 !test_bit(Blocked, &rdev->flags),
6474 msecs_to_jiffies(5000));
6475 rdev_dec_pending(rdev, mddev);
6477 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6479 static int md_notify_reboot(struct notifier_block *this,
6480 unsigned long code, void *x)
6482 struct list_head *tmp;
6485 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6487 printk(KERN_INFO "md: stopping all md devices.\n");
6489 for_each_mddev(mddev, tmp)
6490 if (mddev_trylock(mddev)) {
6491 /* Force a switch to readonly even array
6492 * appears to still be in use. Hence
6495 do_md_stop(mddev, 1, 100);
6496 mddev_unlock(mddev);
6499 * certain more exotic SCSI devices are known to be
6500 * volatile wrt too early system reboots. While the
6501 * right place to handle this issue is the given
6502 * driver, we do want to have a safe RAID driver ...
6509 static struct notifier_block md_notifier = {
6510 .notifier_call = md_notify_reboot,
6512 .priority = INT_MAX, /* before any real devices */
6515 static void md_geninit(void)
6517 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6519 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6522 static int __init md_init(void)
6524 if (register_blkdev(MD_MAJOR, "md"))
6526 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6527 unregister_blkdev(MD_MAJOR, "md");
6530 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6531 md_probe, NULL, NULL);
6532 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6533 md_probe, NULL, NULL);
6535 register_reboot_notifier(&md_notifier);
6536 raid_table_header = register_sysctl_table(raid_root_table);
6546 * Searches all registered partitions for autorun RAID arrays
6550 static LIST_HEAD(all_detected_devices);
6551 struct detected_devices_node {
6552 struct list_head list;
6556 void md_autodetect_dev(dev_t dev)
6558 struct detected_devices_node *node_detected_dev;
6560 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6561 if (node_detected_dev) {
6562 node_detected_dev->dev = dev;
6563 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6565 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6566 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6571 static void autostart_arrays(int part)
6574 struct detected_devices_node *node_detected_dev;
6576 int i_scanned, i_passed;
6581 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6583 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6585 node_detected_dev = list_entry(all_detected_devices.next,
6586 struct detected_devices_node, list);
6587 list_del(&node_detected_dev->list);
6588 dev = node_detected_dev->dev;
6589 kfree(node_detected_dev);
6590 rdev = md_import_device(dev,0, 90);
6594 if (test_bit(Faulty, &rdev->flags)) {
6598 set_bit(AutoDetected, &rdev->flags);
6599 list_add(&rdev->same_set, &pending_raid_disks);
6603 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6604 i_scanned, i_passed);
6606 autorun_devices(part);
6609 #endif /* !MODULE */
6611 static __exit void md_exit(void)
6614 struct list_head *tmp;
6616 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6617 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6619 unregister_blkdev(MD_MAJOR,"md");
6620 unregister_blkdev(mdp_major, "mdp");
6621 unregister_reboot_notifier(&md_notifier);
6622 unregister_sysctl_table(raid_table_header);
6623 remove_proc_entry("mdstat", NULL);
6624 for_each_mddev(mddev, tmp) {
6625 export_array(mddev);
6626 mddev->hold_active = 0;
6630 subsys_initcall(md_init);
6631 module_exit(md_exit)
6633 static int get_ro(char *buffer, struct kernel_param *kp)
6635 return sprintf(buffer, "%d", start_readonly);
6637 static int set_ro(const char *val, struct kernel_param *kp)
6640 int num = simple_strtoul(val, &e, 10);
6641 if (*val && (*e == '\0' || *e == '\n')) {
6642 start_readonly = num;
6648 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6649 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6651 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6653 EXPORT_SYMBOL(register_md_personality);
6654 EXPORT_SYMBOL(unregister_md_personality);
6655 EXPORT_SYMBOL(md_error);
6656 EXPORT_SYMBOL(md_done_sync);
6657 EXPORT_SYMBOL(md_write_start);
6658 EXPORT_SYMBOL(md_write_end);
6659 EXPORT_SYMBOL(md_register_thread);
6660 EXPORT_SYMBOL(md_unregister_thread);
6661 EXPORT_SYMBOL(md_wakeup_thread);
6662 EXPORT_SYMBOL(md_check_recovery);
6663 MODULE_LICENSE("GPL");
6665 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);