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->sectors = calc_num_sectors(rdev, sb->chunk_size);
784 if (rdev->sectors < sb->size * 2 && 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->dev_sectors = sb->size * 2;
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->dev_sectors / 2;
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->dev_sectors)
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->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1188 le64_to_cpu(sb->data_offset);
1190 rdev->sectors = rdev->sb_start;
1191 if (rdev->sectors < le64_to_cpu(sb->data_size))
1193 rdev->sectors = le64_to_cpu(sb->data_size);
1194 if (le32_to_cpu(sb->chunksize))
1195 rdev->sectors &= ~((sector_t)le32_to_cpu(sb->chunksize) - 1);
1197 if (le64_to_cpu(sb->size) > rdev->sectors)
1202 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1204 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1205 __u64 ev1 = le64_to_cpu(sb->events);
1207 rdev->raid_disk = -1;
1208 clear_bit(Faulty, &rdev->flags);
1209 clear_bit(In_sync, &rdev->flags);
1210 clear_bit(WriteMostly, &rdev->flags);
1211 clear_bit(BarriersNotsupp, &rdev->flags);
1213 if (mddev->raid_disks == 0) {
1214 mddev->major_version = 1;
1215 mddev->patch_version = 0;
1216 mddev->external = 0;
1217 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1218 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1219 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1220 mddev->level = le32_to_cpu(sb->level);
1221 mddev->clevel[0] = 0;
1222 mddev->layout = le32_to_cpu(sb->layout);
1223 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1224 mddev->dev_sectors = le64_to_cpu(sb->size);
1225 mddev->events = ev1;
1226 mddev->bitmap_offset = 0;
1227 mddev->default_bitmap_offset = 1024 >> 9;
1229 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1230 memcpy(mddev->uuid, sb->set_uuid, 16);
1232 mddev->max_disks = (4096-256)/2;
1234 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1235 mddev->bitmap_file == NULL )
1236 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1238 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1239 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1240 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1241 mddev->new_level = le32_to_cpu(sb->new_level);
1242 mddev->new_layout = le32_to_cpu(sb->new_layout);
1243 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1245 mddev->reshape_position = MaxSector;
1246 mddev->delta_disks = 0;
1247 mddev->new_level = mddev->level;
1248 mddev->new_layout = mddev->layout;
1249 mddev->new_chunk = mddev->chunk_size;
1252 } else if (mddev->pers == NULL) {
1253 /* Insist of good event counter while assembling */
1255 if (ev1 < mddev->events)
1257 } else if (mddev->bitmap) {
1258 /* If adding to array with a bitmap, then we can accept an
1259 * older device, but not too old.
1261 if (ev1 < mddev->bitmap->events_cleared)
1264 if (ev1 < mddev->events)
1265 /* just a hot-add of a new device, leave raid_disk at -1 */
1268 if (mddev->level != LEVEL_MULTIPATH) {
1270 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1272 case 0xffff: /* spare */
1274 case 0xfffe: /* faulty */
1275 set_bit(Faulty, &rdev->flags);
1278 if ((le32_to_cpu(sb->feature_map) &
1279 MD_FEATURE_RECOVERY_OFFSET))
1280 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1282 set_bit(In_sync, &rdev->flags);
1283 rdev->raid_disk = role;
1286 if (sb->devflags & WriteMostly1)
1287 set_bit(WriteMostly, &rdev->flags);
1288 } else /* MULTIPATH are always insync */
1289 set_bit(In_sync, &rdev->flags);
1294 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1296 struct mdp_superblock_1 *sb;
1299 /* make rdev->sb match mddev and rdev data. */
1301 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1303 sb->feature_map = 0;
1305 sb->recovery_offset = cpu_to_le64(0);
1306 memset(sb->pad1, 0, sizeof(sb->pad1));
1307 memset(sb->pad2, 0, sizeof(sb->pad2));
1308 memset(sb->pad3, 0, sizeof(sb->pad3));
1310 sb->utime = cpu_to_le64((__u64)mddev->utime);
1311 sb->events = cpu_to_le64(mddev->events);
1313 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1315 sb->resync_offset = cpu_to_le64(0);
1317 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1319 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1320 sb->size = cpu_to_le64(mddev->dev_sectors);
1322 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1323 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1324 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1327 if (rdev->raid_disk >= 0 &&
1328 !test_bit(In_sync, &rdev->flags)) {
1329 if (mddev->curr_resync_completed > rdev->recovery_offset)
1330 rdev->recovery_offset = mddev->curr_resync_completed;
1331 if (rdev->recovery_offset > 0) {
1333 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1334 sb->recovery_offset =
1335 cpu_to_le64(rdev->recovery_offset);
1339 if (mddev->reshape_position != MaxSector) {
1340 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1341 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1342 sb->new_layout = cpu_to_le32(mddev->new_layout);
1343 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1344 sb->new_level = cpu_to_le32(mddev->new_level);
1345 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1349 list_for_each_entry(rdev2, &mddev->disks, same_set)
1350 if (rdev2->desc_nr+1 > max_dev)
1351 max_dev = rdev2->desc_nr+1;
1353 if (max_dev > le32_to_cpu(sb->max_dev))
1354 sb->max_dev = cpu_to_le32(max_dev);
1355 for (i=0; i<max_dev;i++)
1356 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1358 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1360 if (test_bit(Faulty, &rdev2->flags))
1361 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1362 else if (test_bit(In_sync, &rdev2->flags))
1363 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1364 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1365 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1367 sb->dev_roles[i] = cpu_to_le16(0xffff);
1370 sb->sb_csum = calc_sb_1_csum(sb);
1373 static unsigned long long
1374 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1376 struct mdp_superblock_1 *sb;
1377 sector_t max_sectors;
1378 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1379 return 0; /* component must fit device */
1380 if (rdev->sb_start < rdev->data_offset) {
1381 /* minor versions 1 and 2; superblock before data */
1382 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1383 max_sectors -= rdev->data_offset;
1384 if (!num_sectors || num_sectors > max_sectors)
1385 num_sectors = max_sectors;
1386 } else if (rdev->mddev->bitmap_offset) {
1387 /* minor version 0 with bitmap we can't move */
1390 /* minor version 0; superblock after data */
1392 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1393 sb_start &= ~(sector_t)(4*2 - 1);
1394 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1395 if (!num_sectors || num_sectors > max_sectors)
1396 num_sectors = max_sectors;
1397 rdev->sb_start = sb_start;
1399 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1400 sb->data_size = cpu_to_le64(num_sectors);
1401 sb->super_offset = rdev->sb_start;
1402 sb->sb_csum = calc_sb_1_csum(sb);
1403 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1405 md_super_wait(rdev->mddev);
1406 return num_sectors / 2; /* kB for sysfs */
1409 static struct super_type super_types[] = {
1412 .owner = THIS_MODULE,
1413 .load_super = super_90_load,
1414 .validate_super = super_90_validate,
1415 .sync_super = super_90_sync,
1416 .rdev_size_change = super_90_rdev_size_change,
1420 .owner = THIS_MODULE,
1421 .load_super = super_1_load,
1422 .validate_super = super_1_validate,
1423 .sync_super = super_1_sync,
1424 .rdev_size_change = super_1_rdev_size_change,
1428 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1430 mdk_rdev_t *rdev, *rdev2;
1433 rdev_for_each_rcu(rdev, mddev1)
1434 rdev_for_each_rcu(rdev2, mddev2)
1435 if (rdev->bdev->bd_contains ==
1436 rdev2->bdev->bd_contains) {
1444 static LIST_HEAD(pending_raid_disks);
1446 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1448 struct mdk_personality *pers = mddev->pers;
1449 struct gendisk *disk = mddev->gendisk;
1450 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1451 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1453 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1454 if (pers && pers->level >= 4 && pers->level <= 6)
1457 /* If rdev is integrity capable, register profile for mddev */
1458 if (!bi_mddev && bi_rdev) {
1459 if (blk_integrity_register(disk, bi_rdev))
1460 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1461 __func__, disk->disk_name);
1463 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1468 /* Check that mddev and rdev have matching profiles */
1469 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1470 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1471 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1472 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1474 blk_integrity_unregister(disk);
1478 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1480 char b[BDEVNAME_SIZE];
1490 /* prevent duplicates */
1491 if (find_rdev(mddev, rdev->bdev->bd_dev))
1494 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1495 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1496 rdev->sectors < mddev->dev_sectors)) {
1498 /* Cannot change size, so fail
1499 * If mddev->level <= 0, then we don't care
1500 * about aligning sizes (e.g. linear)
1502 if (mddev->level > 0)
1505 mddev->dev_sectors = rdev->sectors;
1508 /* Verify rdev->desc_nr is unique.
1509 * If it is -1, assign a free number, else
1510 * check number is not in use
1512 if (rdev->desc_nr < 0) {
1514 if (mddev->pers) choice = mddev->raid_disks;
1515 while (find_rdev_nr(mddev, choice))
1517 rdev->desc_nr = choice;
1519 if (find_rdev_nr(mddev, rdev->desc_nr))
1522 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1523 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1524 mdname(mddev), mddev->max_disks);
1527 bdevname(rdev->bdev,b);
1528 while ( (s=strchr(b, '/')) != NULL)
1531 rdev->mddev = mddev;
1532 printk(KERN_INFO "md: bind<%s>\n", b);
1534 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1537 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1538 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1539 kobject_del(&rdev->kobj);
1542 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1544 list_add_rcu(&rdev->same_set, &mddev->disks);
1545 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1547 /* May as well allow recovery to be retried once */
1548 mddev->recovery_disabled = 0;
1550 md_integrity_check(rdev, mddev);
1554 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1559 static void md_delayed_delete(struct work_struct *ws)
1561 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1562 kobject_del(&rdev->kobj);
1563 kobject_put(&rdev->kobj);
1566 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1568 char b[BDEVNAME_SIZE];
1573 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1574 list_del_rcu(&rdev->same_set);
1575 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1577 sysfs_remove_link(&rdev->kobj, "block");
1578 sysfs_put(rdev->sysfs_state);
1579 rdev->sysfs_state = NULL;
1580 /* We need to delay this, otherwise we can deadlock when
1581 * writing to 'remove' to "dev/state". We also need
1582 * to delay it due to rcu usage.
1585 INIT_WORK(&rdev->del_work, md_delayed_delete);
1586 kobject_get(&rdev->kobj);
1587 schedule_work(&rdev->del_work);
1591 * prevent the device from being mounted, repartitioned or
1592 * otherwise reused by a RAID array (or any other kernel
1593 * subsystem), by bd_claiming the device.
1595 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1598 struct block_device *bdev;
1599 char b[BDEVNAME_SIZE];
1601 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1603 printk(KERN_ERR "md: could not open %s.\n",
1604 __bdevname(dev, b));
1605 return PTR_ERR(bdev);
1607 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1609 printk(KERN_ERR "md: could not bd_claim %s.\n",
1611 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1615 set_bit(AllReserved, &rdev->flags);
1620 static void unlock_rdev(mdk_rdev_t *rdev)
1622 struct block_device *bdev = rdev->bdev;
1627 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1630 void md_autodetect_dev(dev_t dev);
1632 static void export_rdev(mdk_rdev_t * rdev)
1634 char b[BDEVNAME_SIZE];
1635 printk(KERN_INFO "md: export_rdev(%s)\n",
1636 bdevname(rdev->bdev,b));
1641 if (test_bit(AutoDetected, &rdev->flags))
1642 md_autodetect_dev(rdev->bdev->bd_dev);
1645 kobject_put(&rdev->kobj);
1648 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1650 unbind_rdev_from_array(rdev);
1654 static void export_array(mddev_t *mddev)
1656 mdk_rdev_t *rdev, *tmp;
1658 rdev_for_each(rdev, tmp, mddev) {
1663 kick_rdev_from_array(rdev);
1665 if (!list_empty(&mddev->disks))
1667 mddev->raid_disks = 0;
1668 mddev->major_version = 0;
1671 static void print_desc(mdp_disk_t *desc)
1673 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1674 desc->major,desc->minor,desc->raid_disk,desc->state);
1677 static void print_sb_90(mdp_super_t *sb)
1682 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1683 sb->major_version, sb->minor_version, sb->patch_version,
1684 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1686 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1687 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1688 sb->md_minor, sb->layout, sb->chunk_size);
1689 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1690 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1691 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1692 sb->failed_disks, sb->spare_disks,
1693 sb->sb_csum, (unsigned long)sb->events_lo);
1696 for (i = 0; i < MD_SB_DISKS; i++) {
1699 desc = sb->disks + i;
1700 if (desc->number || desc->major || desc->minor ||
1701 desc->raid_disk || (desc->state && (desc->state != 4))) {
1702 printk(" D %2d: ", i);
1706 printk(KERN_INFO "md: THIS: ");
1707 print_desc(&sb->this_disk);
1710 static void print_sb_1(struct mdp_superblock_1 *sb)
1714 uuid = sb->set_uuid;
1715 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1716 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1717 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1718 le32_to_cpu(sb->major_version),
1719 le32_to_cpu(sb->feature_map),
1720 uuid[0], uuid[1], uuid[2], uuid[3],
1721 uuid[4], uuid[5], uuid[6], uuid[7],
1722 uuid[8], uuid[9], uuid[10], uuid[11],
1723 uuid[12], uuid[13], uuid[14], uuid[15],
1725 (unsigned long long)le64_to_cpu(sb->ctime)
1726 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1728 uuid = sb->device_uuid;
1729 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1731 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1732 ":%02x%02x%02x%02x%02x%02x\n"
1733 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1734 KERN_INFO "md: (MaxDev:%u) \n",
1735 le32_to_cpu(sb->level),
1736 (unsigned long long)le64_to_cpu(sb->size),
1737 le32_to_cpu(sb->raid_disks),
1738 le32_to_cpu(sb->layout),
1739 le32_to_cpu(sb->chunksize),
1740 (unsigned long long)le64_to_cpu(sb->data_offset),
1741 (unsigned long long)le64_to_cpu(sb->data_size),
1742 (unsigned long long)le64_to_cpu(sb->super_offset),
1743 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1744 le32_to_cpu(sb->dev_number),
1745 uuid[0], uuid[1], uuid[2], uuid[3],
1746 uuid[4], uuid[5], uuid[6], uuid[7],
1747 uuid[8], uuid[9], uuid[10], uuid[11],
1748 uuid[12], uuid[13], uuid[14], uuid[15],
1750 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1751 (unsigned long long)le64_to_cpu(sb->events),
1752 (unsigned long long)le64_to_cpu(sb->resync_offset),
1753 le32_to_cpu(sb->sb_csum),
1754 le32_to_cpu(sb->max_dev)
1758 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1760 char b[BDEVNAME_SIZE];
1761 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1762 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1763 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1765 if (rdev->sb_loaded) {
1766 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1767 switch (major_version) {
1769 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1772 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1776 printk(KERN_INFO "md: no rdev superblock!\n");
1779 static void md_print_devices(void)
1781 struct list_head *tmp;
1784 char b[BDEVNAME_SIZE];
1787 printk("md: **********************************\n");
1788 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1789 printk("md: **********************************\n");
1790 for_each_mddev(mddev, tmp) {
1793 bitmap_print_sb(mddev->bitmap);
1795 printk("%s: ", mdname(mddev));
1796 list_for_each_entry(rdev, &mddev->disks, same_set)
1797 printk("<%s>", bdevname(rdev->bdev,b));
1800 list_for_each_entry(rdev, &mddev->disks, same_set)
1801 print_rdev(rdev, mddev->major_version);
1803 printk("md: **********************************\n");
1808 static void sync_sbs(mddev_t * mddev, int nospares)
1810 /* Update each superblock (in-memory image), but
1811 * if we are allowed to, skip spares which already
1812 * have the right event counter, or have one earlier
1813 * (which would mean they aren't being marked as dirty
1814 * with the rest of the array)
1818 list_for_each_entry(rdev, &mddev->disks, same_set) {
1819 if (rdev->sb_events == mddev->events ||
1821 rdev->raid_disk < 0 &&
1822 (rdev->sb_events&1)==0 &&
1823 rdev->sb_events+1 == mddev->events)) {
1824 /* Don't update this superblock */
1825 rdev->sb_loaded = 2;
1827 super_types[mddev->major_version].
1828 sync_super(mddev, rdev);
1829 rdev->sb_loaded = 1;
1834 static void md_update_sb(mddev_t * mddev, int force_change)
1840 if (mddev->external)
1843 spin_lock_irq(&mddev->write_lock);
1845 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1846 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1848 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1849 /* just a clean<-> dirty transition, possibly leave spares alone,
1850 * though if events isn't the right even/odd, we will have to do
1856 if (mddev->degraded)
1857 /* If the array is degraded, then skipping spares is both
1858 * dangerous and fairly pointless.
1859 * Dangerous because a device that was removed from the array
1860 * might have a event_count that still looks up-to-date,
1861 * so it can be re-added without a resync.
1862 * Pointless because if there are any spares to skip,
1863 * then a recovery will happen and soon that array won't
1864 * be degraded any more and the spare can go back to sleep then.
1868 sync_req = mddev->in_sync;
1869 mddev->utime = get_seconds();
1871 /* If this is just a dirty<->clean transition, and the array is clean
1872 * and 'events' is odd, we can roll back to the previous clean state */
1874 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1875 && (mddev->events & 1)
1876 && mddev->events != 1)
1879 /* otherwise we have to go forward and ... */
1881 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1882 /* .. if the array isn't clean, insist on an odd 'events' */
1883 if ((mddev->events&1)==0) {
1888 /* otherwise insist on an even 'events' (for clean states) */
1889 if ((mddev->events&1)) {
1896 if (!mddev->events) {
1898 * oops, this 64-bit counter should never wrap.
1899 * Either we are in around ~1 trillion A.C., assuming
1900 * 1 reboot per second, or we have a bug:
1907 * do not write anything to disk if using
1908 * nonpersistent superblocks
1910 if (!mddev->persistent) {
1911 if (!mddev->external)
1912 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1914 spin_unlock_irq(&mddev->write_lock);
1915 wake_up(&mddev->sb_wait);
1918 sync_sbs(mddev, nospares);
1919 spin_unlock_irq(&mddev->write_lock);
1922 "md: updating %s RAID superblock on device (in sync %d)\n",
1923 mdname(mddev),mddev->in_sync);
1925 bitmap_update_sb(mddev->bitmap);
1926 list_for_each_entry(rdev, &mddev->disks, same_set) {
1927 char b[BDEVNAME_SIZE];
1928 dprintk(KERN_INFO "md: ");
1929 if (rdev->sb_loaded != 1)
1930 continue; /* no noise on spare devices */
1931 if (test_bit(Faulty, &rdev->flags))
1932 dprintk("(skipping faulty ");
1934 dprintk("%s ", bdevname(rdev->bdev,b));
1935 if (!test_bit(Faulty, &rdev->flags)) {
1936 md_super_write(mddev,rdev,
1937 rdev->sb_start, rdev->sb_size,
1939 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1940 bdevname(rdev->bdev,b),
1941 (unsigned long long)rdev->sb_start);
1942 rdev->sb_events = mddev->events;
1946 if (mddev->level == LEVEL_MULTIPATH)
1947 /* only need to write one superblock... */
1950 md_super_wait(mddev);
1951 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1953 spin_lock_irq(&mddev->write_lock);
1954 if (mddev->in_sync != sync_req ||
1955 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1956 /* have to write it out again */
1957 spin_unlock_irq(&mddev->write_lock);
1960 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1961 spin_unlock_irq(&mddev->write_lock);
1962 wake_up(&mddev->sb_wait);
1966 /* words written to sysfs files may, or may not, be \n terminated.
1967 * We want to accept with case. For this we use cmd_match.
1969 static int cmd_match(const char *cmd, const char *str)
1971 /* See if cmd, written into a sysfs file, matches
1972 * str. They must either be the same, or cmd can
1973 * have a trailing newline
1975 while (*cmd && *str && *cmd == *str) {
1986 struct rdev_sysfs_entry {
1987 struct attribute attr;
1988 ssize_t (*show)(mdk_rdev_t *, char *);
1989 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1993 state_show(mdk_rdev_t *rdev, char *page)
1998 if (test_bit(Faulty, &rdev->flags)) {
1999 len+= sprintf(page+len, "%sfaulty",sep);
2002 if (test_bit(In_sync, &rdev->flags)) {
2003 len += sprintf(page+len, "%sin_sync",sep);
2006 if (test_bit(WriteMostly, &rdev->flags)) {
2007 len += sprintf(page+len, "%swrite_mostly",sep);
2010 if (test_bit(Blocked, &rdev->flags)) {
2011 len += sprintf(page+len, "%sblocked", sep);
2014 if (!test_bit(Faulty, &rdev->flags) &&
2015 !test_bit(In_sync, &rdev->flags)) {
2016 len += sprintf(page+len, "%sspare", sep);
2019 return len+sprintf(page+len, "\n");
2023 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2026 * faulty - simulates and error
2027 * remove - disconnects the device
2028 * writemostly - sets write_mostly
2029 * -writemostly - clears write_mostly
2030 * blocked - sets the Blocked flag
2031 * -blocked - clears the Blocked flag
2034 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2035 md_error(rdev->mddev, rdev);
2037 } else if (cmd_match(buf, "remove")) {
2038 if (rdev->raid_disk >= 0)
2041 mddev_t *mddev = rdev->mddev;
2042 kick_rdev_from_array(rdev);
2044 md_update_sb(mddev, 1);
2045 md_new_event(mddev);
2048 } else if (cmd_match(buf, "writemostly")) {
2049 set_bit(WriteMostly, &rdev->flags);
2051 } else if (cmd_match(buf, "-writemostly")) {
2052 clear_bit(WriteMostly, &rdev->flags);
2054 } else if (cmd_match(buf, "blocked")) {
2055 set_bit(Blocked, &rdev->flags);
2057 } else if (cmd_match(buf, "-blocked")) {
2058 clear_bit(Blocked, &rdev->flags);
2059 wake_up(&rdev->blocked_wait);
2060 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2061 md_wakeup_thread(rdev->mddev->thread);
2065 if (!err && rdev->sysfs_state)
2066 sysfs_notify_dirent(rdev->sysfs_state);
2067 return err ? err : len;
2069 static struct rdev_sysfs_entry rdev_state =
2070 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2073 errors_show(mdk_rdev_t *rdev, char *page)
2075 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2079 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2082 unsigned long n = simple_strtoul(buf, &e, 10);
2083 if (*buf && (*e == 0 || *e == '\n')) {
2084 atomic_set(&rdev->corrected_errors, n);
2089 static struct rdev_sysfs_entry rdev_errors =
2090 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2093 slot_show(mdk_rdev_t *rdev, char *page)
2095 if (rdev->raid_disk < 0)
2096 return sprintf(page, "none\n");
2098 return sprintf(page, "%d\n", rdev->raid_disk);
2102 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2107 int slot = simple_strtoul(buf, &e, 10);
2108 if (strncmp(buf, "none", 4)==0)
2110 else if (e==buf || (*e && *e!= '\n'))
2112 if (rdev->mddev->pers && slot == -1) {
2113 /* Setting 'slot' on an active array requires also
2114 * updating the 'rd%d' link, and communicating
2115 * with the personality with ->hot_*_disk.
2116 * For now we only support removing
2117 * failed/spare devices. This normally happens automatically,
2118 * but not when the metadata is externally managed.
2120 if (rdev->raid_disk == -1)
2122 /* personality does all needed checks */
2123 if (rdev->mddev->pers->hot_add_disk == NULL)
2125 err = rdev->mddev->pers->
2126 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2129 sprintf(nm, "rd%d", rdev->raid_disk);
2130 sysfs_remove_link(&rdev->mddev->kobj, nm);
2131 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2132 md_wakeup_thread(rdev->mddev->thread);
2133 } else if (rdev->mddev->pers) {
2135 /* Activating a spare .. or possibly reactivating
2136 * if we every get bitmaps working here.
2139 if (rdev->raid_disk != -1)
2142 if (rdev->mddev->pers->hot_add_disk == NULL)
2145 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2146 if (rdev2->raid_disk == slot)
2149 rdev->raid_disk = slot;
2150 if (test_bit(In_sync, &rdev->flags))
2151 rdev->saved_raid_disk = slot;
2153 rdev->saved_raid_disk = -1;
2154 err = rdev->mddev->pers->
2155 hot_add_disk(rdev->mddev, rdev);
2157 rdev->raid_disk = -1;
2160 sysfs_notify_dirent(rdev->sysfs_state);
2161 sprintf(nm, "rd%d", rdev->raid_disk);
2162 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2164 "md: cannot register "
2166 nm, mdname(rdev->mddev));
2168 /* don't wakeup anyone, leave that to userspace. */
2170 if (slot >= rdev->mddev->raid_disks)
2172 rdev->raid_disk = slot;
2173 /* assume it is working */
2174 clear_bit(Faulty, &rdev->flags);
2175 clear_bit(WriteMostly, &rdev->flags);
2176 set_bit(In_sync, &rdev->flags);
2177 sysfs_notify_dirent(rdev->sysfs_state);
2183 static struct rdev_sysfs_entry rdev_slot =
2184 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2187 offset_show(mdk_rdev_t *rdev, char *page)
2189 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2193 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2196 unsigned long long offset = simple_strtoull(buf, &e, 10);
2197 if (e==buf || (*e && *e != '\n'))
2199 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2201 if (rdev->sectors && rdev->mddev->external)
2202 /* Must set offset before size, so overlap checks
2205 rdev->data_offset = offset;
2209 static struct rdev_sysfs_entry rdev_offset =
2210 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2213 rdev_size_show(mdk_rdev_t *rdev, char *page)
2215 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2218 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2220 /* check if two start/length pairs overlap */
2229 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2231 mddev_t *my_mddev = rdev->mddev;
2232 sector_t oldsectors = rdev->sectors;
2233 unsigned long long sectors;
2235 if (strict_strtoull(buf, 10, §ors) < 0)
2238 if (my_mddev->pers && rdev->raid_disk >= 0) {
2239 if (my_mddev->persistent) {
2240 sectors = super_types[my_mddev->major_version].
2241 rdev_size_change(rdev, sectors);
2244 } else if (!sectors)
2245 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2248 if (sectors < my_mddev->dev_sectors)
2249 return -EINVAL; /* component must fit device */
2251 rdev->sectors = sectors;
2252 if (sectors > oldsectors && my_mddev->external) {
2253 /* need to check that all other rdevs with the same ->bdev
2254 * do not overlap. We need to unlock the mddev to avoid
2255 * a deadlock. We have already changed rdev->sectors, and if
2256 * we have to change it back, we will have the lock again.
2260 struct list_head *tmp;
2262 mddev_unlock(my_mddev);
2263 for_each_mddev(mddev, tmp) {
2267 list_for_each_entry(rdev2, &mddev->disks, same_set)
2268 if (test_bit(AllReserved, &rdev2->flags) ||
2269 (rdev->bdev == rdev2->bdev &&
2271 overlaps(rdev->data_offset, rdev->sectors,
2277 mddev_unlock(mddev);
2283 mddev_lock(my_mddev);
2285 /* Someone else could have slipped in a size
2286 * change here, but doing so is just silly.
2287 * We put oldsectors back because we *know* it is
2288 * safe, and trust userspace not to race with
2291 rdev->sectors = oldsectors;
2298 static struct rdev_sysfs_entry rdev_size =
2299 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2301 static struct attribute *rdev_default_attrs[] = {
2310 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2312 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2313 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2314 mddev_t *mddev = rdev->mddev;
2320 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2322 if (rdev->mddev == NULL)
2325 rv = entry->show(rdev, page);
2326 mddev_unlock(mddev);
2332 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2333 const char *page, size_t length)
2335 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2336 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2338 mddev_t *mddev = rdev->mddev;
2342 if (!capable(CAP_SYS_ADMIN))
2344 rv = mddev ? mddev_lock(mddev): -EBUSY;
2346 if (rdev->mddev == NULL)
2349 rv = entry->store(rdev, page, length);
2350 mddev_unlock(mddev);
2355 static void rdev_free(struct kobject *ko)
2357 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2360 static struct sysfs_ops rdev_sysfs_ops = {
2361 .show = rdev_attr_show,
2362 .store = rdev_attr_store,
2364 static struct kobj_type rdev_ktype = {
2365 .release = rdev_free,
2366 .sysfs_ops = &rdev_sysfs_ops,
2367 .default_attrs = rdev_default_attrs,
2371 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2373 * mark the device faulty if:
2375 * - the device is nonexistent (zero size)
2376 * - the device has no valid superblock
2378 * a faulty rdev _never_ has rdev->sb set.
2380 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2382 char b[BDEVNAME_SIZE];
2387 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2389 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2390 return ERR_PTR(-ENOMEM);
2393 if ((err = alloc_disk_sb(rdev)))
2396 err = lock_rdev(rdev, newdev, super_format == -2);
2400 kobject_init(&rdev->kobj, &rdev_ktype);
2403 rdev->saved_raid_disk = -1;
2404 rdev->raid_disk = -1;
2406 rdev->data_offset = 0;
2407 rdev->sb_events = 0;
2408 atomic_set(&rdev->nr_pending, 0);
2409 atomic_set(&rdev->read_errors, 0);
2410 atomic_set(&rdev->corrected_errors, 0);
2412 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2415 "md: %s has zero or unknown size, marking faulty!\n",
2416 bdevname(rdev->bdev,b));
2421 if (super_format >= 0) {
2422 err = super_types[super_format].
2423 load_super(rdev, NULL, super_minor);
2424 if (err == -EINVAL) {
2426 "md: %s does not have a valid v%d.%d "
2427 "superblock, not importing!\n",
2428 bdevname(rdev->bdev,b),
2429 super_format, super_minor);
2434 "md: could not read %s's sb, not importing!\n",
2435 bdevname(rdev->bdev,b));
2440 INIT_LIST_HEAD(&rdev->same_set);
2441 init_waitqueue_head(&rdev->blocked_wait);
2446 if (rdev->sb_page) {
2452 return ERR_PTR(err);
2456 * Check a full RAID array for plausibility
2460 static void analyze_sbs(mddev_t * mddev)
2463 mdk_rdev_t *rdev, *freshest, *tmp;
2464 char b[BDEVNAME_SIZE];
2467 rdev_for_each(rdev, tmp, mddev)
2468 switch (super_types[mddev->major_version].
2469 load_super(rdev, freshest, mddev->minor_version)) {
2477 "md: fatal superblock inconsistency in %s"
2478 " -- removing from array\n",
2479 bdevname(rdev->bdev,b));
2480 kick_rdev_from_array(rdev);
2484 super_types[mddev->major_version].
2485 validate_super(mddev, freshest);
2488 rdev_for_each(rdev, tmp, mddev) {
2489 if (rdev->desc_nr >= mddev->max_disks ||
2490 i > mddev->max_disks) {
2492 "md: %s: %s: only %d devices permitted\n",
2493 mdname(mddev), bdevname(rdev->bdev, b),
2495 kick_rdev_from_array(rdev);
2498 if (rdev != freshest)
2499 if (super_types[mddev->major_version].
2500 validate_super(mddev, rdev)) {
2501 printk(KERN_WARNING "md: kicking non-fresh %s"
2503 bdevname(rdev->bdev,b));
2504 kick_rdev_from_array(rdev);
2507 if (mddev->level == LEVEL_MULTIPATH) {
2508 rdev->desc_nr = i++;
2509 rdev->raid_disk = rdev->desc_nr;
2510 set_bit(In_sync, &rdev->flags);
2511 } else if (rdev->raid_disk >= mddev->raid_disks) {
2512 rdev->raid_disk = -1;
2513 clear_bit(In_sync, &rdev->flags);
2519 if (mddev->recovery_cp != MaxSector &&
2521 printk(KERN_ERR "md: %s: raid array is not clean"
2522 " -- starting background reconstruction\n",
2527 static void md_safemode_timeout(unsigned long data);
2530 safe_delay_show(mddev_t *mddev, char *page)
2532 int msec = (mddev->safemode_delay*1000)/HZ;
2533 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2536 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2544 /* remove a period, and count digits after it */
2545 if (len >= sizeof(buf))
2547 strlcpy(buf, cbuf, sizeof(buf));
2548 for (i=0; i<len; i++) {
2550 if (isdigit(buf[i])) {
2555 } else if (buf[i] == '.') {
2560 if (strict_strtoul(buf, 10, &msec) < 0)
2562 msec = (msec * 1000) / scale;
2564 mddev->safemode_delay = 0;
2566 unsigned long old_delay = mddev->safemode_delay;
2567 mddev->safemode_delay = (msec*HZ)/1000;
2568 if (mddev->safemode_delay == 0)
2569 mddev->safemode_delay = 1;
2570 if (mddev->safemode_delay < old_delay)
2571 md_safemode_timeout((unsigned long)mddev);
2575 static struct md_sysfs_entry md_safe_delay =
2576 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2579 level_show(mddev_t *mddev, char *page)
2581 struct mdk_personality *p = mddev->pers;
2583 return sprintf(page, "%s\n", p->name);
2584 else if (mddev->clevel[0])
2585 return sprintf(page, "%s\n", mddev->clevel);
2586 else if (mddev->level != LEVEL_NONE)
2587 return sprintf(page, "%d\n", mddev->level);
2593 level_store(mddev_t *mddev, const char *buf, size_t len)
2600 if (len >= sizeof(mddev->clevel))
2602 strncpy(mddev->clevel, buf, len);
2603 if (mddev->clevel[len-1] == '\n')
2605 mddev->clevel[len] = 0;
2606 mddev->level = LEVEL_NONE;
2610 static struct md_sysfs_entry md_level =
2611 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2615 layout_show(mddev_t *mddev, char *page)
2617 /* just a number, not meaningful for all levels */
2618 if (mddev->reshape_position != MaxSector &&
2619 mddev->layout != mddev->new_layout)
2620 return sprintf(page, "%d (%d)\n",
2621 mddev->new_layout, mddev->layout);
2622 return sprintf(page, "%d\n", mddev->layout);
2626 layout_store(mddev_t *mddev, const char *buf, size_t len)
2629 unsigned long n = simple_strtoul(buf, &e, 10);
2631 if (!*buf || (*e && *e != '\n'))
2636 if (mddev->reshape_position != MaxSector)
2637 mddev->new_layout = n;
2642 static struct md_sysfs_entry md_layout =
2643 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2647 raid_disks_show(mddev_t *mddev, char *page)
2649 if (mddev->raid_disks == 0)
2651 if (mddev->reshape_position != MaxSector &&
2652 mddev->delta_disks != 0)
2653 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2654 mddev->raid_disks - mddev->delta_disks);
2655 return sprintf(page, "%d\n", mddev->raid_disks);
2658 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2661 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2665 unsigned long n = simple_strtoul(buf, &e, 10);
2667 if (!*buf || (*e && *e != '\n'))
2671 rv = update_raid_disks(mddev, n);
2672 else if (mddev->reshape_position != MaxSector) {
2673 int olddisks = mddev->raid_disks - mddev->delta_disks;
2674 mddev->delta_disks = n - olddisks;
2675 mddev->raid_disks = n;
2677 mddev->raid_disks = n;
2678 return rv ? rv : len;
2680 static struct md_sysfs_entry md_raid_disks =
2681 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2684 chunk_size_show(mddev_t *mddev, char *page)
2686 if (mddev->reshape_position != MaxSector &&
2687 mddev->chunk_size != mddev->new_chunk)
2688 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2690 return sprintf(page, "%d\n", mddev->chunk_size);
2694 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2696 /* can only set chunk_size if array is not yet active */
2698 unsigned long n = simple_strtoul(buf, &e, 10);
2700 if (!*buf || (*e && *e != '\n'))
2705 else if (mddev->reshape_position != MaxSector)
2706 mddev->new_chunk = n;
2708 mddev->chunk_size = n;
2711 static struct md_sysfs_entry md_chunk_size =
2712 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2715 resync_start_show(mddev_t *mddev, char *page)
2717 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2721 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2724 unsigned long long n = simple_strtoull(buf, &e, 10);
2728 if (!*buf || (*e && *e != '\n'))
2731 mddev->recovery_cp = n;
2734 static struct md_sysfs_entry md_resync_start =
2735 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2738 * The array state can be:
2741 * No devices, no size, no level
2742 * Equivalent to STOP_ARRAY ioctl
2744 * May have some settings, but array is not active
2745 * all IO results in error
2746 * When written, doesn't tear down array, but just stops it
2747 * suspended (not supported yet)
2748 * All IO requests will block. The array can be reconfigured.
2749 * Writing this, if accepted, will block until array is quiescent
2751 * no resync can happen. no superblocks get written.
2752 * write requests fail
2754 * like readonly, but behaves like 'clean' on a write request.
2756 * clean - no pending writes, but otherwise active.
2757 * When written to inactive array, starts without resync
2758 * If a write request arrives then
2759 * if metadata is known, mark 'dirty' and switch to 'active'.
2760 * if not known, block and switch to write-pending
2761 * If written to an active array that has pending writes, then fails.
2763 * fully active: IO and resync can be happening.
2764 * When written to inactive array, starts with resync
2767 * clean, but writes are blocked waiting for 'active' to be written.
2770 * like active, but no writes have been seen for a while (100msec).
2773 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2774 write_pending, active_idle, bad_word};
2775 static char *array_states[] = {
2776 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2777 "write-pending", "active-idle", NULL };
2779 static int match_word(const char *word, char **list)
2782 for (n=0; list[n]; n++)
2783 if (cmd_match(word, list[n]))
2789 array_state_show(mddev_t *mddev, char *page)
2791 enum array_state st = inactive;
2804 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2806 else if (mddev->safemode)
2812 if (list_empty(&mddev->disks) &&
2813 mddev->raid_disks == 0 &&
2814 mddev->dev_sectors == 0)
2819 return sprintf(page, "%s\n", array_states[st]);
2822 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2823 static int do_md_run(mddev_t * mddev);
2824 static int restart_array(mddev_t *mddev);
2827 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2830 enum array_state st = match_word(buf, array_states);
2835 /* stopping an active array */
2836 if (atomic_read(&mddev->openers) > 0)
2838 err = do_md_stop(mddev, 0, 0);
2841 /* stopping an active array */
2843 if (atomic_read(&mddev->openers) > 0)
2845 err = do_md_stop(mddev, 2, 0);
2847 err = 0; /* already inactive */
2850 break; /* not supported yet */
2853 err = do_md_stop(mddev, 1, 0);
2856 set_disk_ro(mddev->gendisk, 1);
2857 err = do_md_run(mddev);
2863 err = do_md_stop(mddev, 1, 0);
2864 else if (mddev->ro == 1)
2865 err = restart_array(mddev);
2868 set_disk_ro(mddev->gendisk, 0);
2872 err = do_md_run(mddev);
2877 restart_array(mddev);
2878 spin_lock_irq(&mddev->write_lock);
2879 if (atomic_read(&mddev->writes_pending) == 0) {
2880 if (mddev->in_sync == 0) {
2882 if (mddev->safemode == 1)
2883 mddev->safemode = 0;
2884 if (mddev->persistent)
2885 set_bit(MD_CHANGE_CLEAN,
2891 spin_unlock_irq(&mddev->write_lock);
2894 mddev->recovery_cp = MaxSector;
2895 err = do_md_run(mddev);
2900 restart_array(mddev);
2901 if (mddev->external)
2902 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2903 wake_up(&mddev->sb_wait);
2907 set_disk_ro(mddev->gendisk, 0);
2908 err = do_md_run(mddev);
2913 /* these cannot be set */
2919 sysfs_notify_dirent(mddev->sysfs_state);
2923 static struct md_sysfs_entry md_array_state =
2924 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2927 null_show(mddev_t *mddev, char *page)
2933 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2935 /* buf must be %d:%d\n? giving major and minor numbers */
2936 /* The new device is added to the array.
2937 * If the array has a persistent superblock, we read the
2938 * superblock to initialise info and check validity.
2939 * Otherwise, only checking done is that in bind_rdev_to_array,
2940 * which mainly checks size.
2943 int major = simple_strtoul(buf, &e, 10);
2949 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2951 minor = simple_strtoul(e+1, &e, 10);
2952 if (*e && *e != '\n')
2954 dev = MKDEV(major, minor);
2955 if (major != MAJOR(dev) ||
2956 minor != MINOR(dev))
2960 if (mddev->persistent) {
2961 rdev = md_import_device(dev, mddev->major_version,
2962 mddev->minor_version);
2963 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2964 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2965 mdk_rdev_t, same_set);
2966 err = super_types[mddev->major_version]
2967 .load_super(rdev, rdev0, mddev->minor_version);
2971 } else if (mddev->external)
2972 rdev = md_import_device(dev, -2, -1);
2974 rdev = md_import_device(dev, -1, -1);
2977 return PTR_ERR(rdev);
2978 err = bind_rdev_to_array(rdev, mddev);
2982 return err ? err : len;
2985 static struct md_sysfs_entry md_new_device =
2986 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2989 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2992 unsigned long chunk, end_chunk;
2996 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2998 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2999 if (buf == end) break;
3000 if (*end == '-') { /* range */
3002 end_chunk = simple_strtoul(buf, &end, 0);
3003 if (buf == end) break;
3005 if (*end && !isspace(*end)) break;
3006 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3008 while (isspace(*buf)) buf++;
3010 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3015 static struct md_sysfs_entry md_bitmap =
3016 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3019 size_show(mddev_t *mddev, char *page)
3021 return sprintf(page, "%llu\n",
3022 (unsigned long long)mddev->dev_sectors / 2);
3025 static int update_size(mddev_t *mddev, sector_t num_sectors);
3028 size_store(mddev_t *mddev, const char *buf, size_t len)
3030 /* If array is inactive, we can reduce the component size, but
3031 * not increase it (except from 0).
3032 * If array is active, we can try an on-line resize
3034 unsigned long long sectors;
3035 int err = strict_strtoull(buf, 10, §ors);
3041 err = update_size(mddev, sectors);
3042 md_update_sb(mddev, 1);
3044 if (mddev->dev_sectors == 0 ||
3045 mddev->dev_sectors > sectors)
3046 mddev->dev_sectors = sectors;
3050 return err ? err : len;
3053 static struct md_sysfs_entry md_size =
3054 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3059 * 'none' for arrays with no metadata (good luck...)
3060 * 'external' for arrays with externally managed metadata,
3061 * or N.M for internally known formats
3064 metadata_show(mddev_t *mddev, char *page)
3066 if (mddev->persistent)
3067 return sprintf(page, "%d.%d\n",
3068 mddev->major_version, mddev->minor_version);
3069 else if (mddev->external)
3070 return sprintf(page, "external:%s\n", mddev->metadata_type);
3072 return sprintf(page, "none\n");
3076 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3080 /* Changing the details of 'external' metadata is
3081 * always permitted. Otherwise there must be
3082 * no devices attached to the array.
3084 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3086 else if (!list_empty(&mddev->disks))
3089 if (cmd_match(buf, "none")) {
3090 mddev->persistent = 0;
3091 mddev->external = 0;
3092 mddev->major_version = 0;
3093 mddev->minor_version = 90;
3096 if (strncmp(buf, "external:", 9) == 0) {
3097 size_t namelen = len-9;
3098 if (namelen >= sizeof(mddev->metadata_type))
3099 namelen = sizeof(mddev->metadata_type)-1;
3100 strncpy(mddev->metadata_type, buf+9, namelen);
3101 mddev->metadata_type[namelen] = 0;
3102 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3103 mddev->metadata_type[--namelen] = 0;
3104 mddev->persistent = 0;
3105 mddev->external = 1;
3106 mddev->major_version = 0;
3107 mddev->minor_version = 90;
3110 major = simple_strtoul(buf, &e, 10);
3111 if (e==buf || *e != '.')
3114 minor = simple_strtoul(buf, &e, 10);
3115 if (e==buf || (*e && *e != '\n') )
3117 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3119 mddev->major_version = major;
3120 mddev->minor_version = minor;
3121 mddev->persistent = 1;
3122 mddev->external = 0;
3126 static struct md_sysfs_entry md_metadata =
3127 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3130 action_show(mddev_t *mddev, char *page)
3132 char *type = "idle";
3133 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3134 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3135 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3137 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3138 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3140 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3144 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3147 return sprintf(page, "%s\n", type);
3151 action_store(mddev_t *mddev, const char *page, size_t len)
3153 if (!mddev->pers || !mddev->pers->sync_request)
3156 if (cmd_match(page, "idle")) {
3157 if (mddev->sync_thread) {
3158 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3159 md_unregister_thread(mddev->sync_thread);
3160 mddev->sync_thread = NULL;
3161 mddev->recovery = 0;
3163 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3164 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3166 else if (cmd_match(page, "resync"))
3167 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3168 else if (cmd_match(page, "recover")) {
3169 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3170 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3171 } else if (cmd_match(page, "reshape")) {
3173 if (mddev->pers->start_reshape == NULL)
3175 err = mddev->pers->start_reshape(mddev);
3178 sysfs_notify(&mddev->kobj, NULL, "degraded");
3180 if (cmd_match(page, "check"))
3181 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3182 else if (!cmd_match(page, "repair"))
3184 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3185 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3187 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3188 md_wakeup_thread(mddev->thread);
3189 sysfs_notify_dirent(mddev->sysfs_action);
3194 mismatch_cnt_show(mddev_t *mddev, char *page)
3196 return sprintf(page, "%llu\n",
3197 (unsigned long long) mddev->resync_mismatches);
3200 static struct md_sysfs_entry md_scan_mode =
3201 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3204 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3207 sync_min_show(mddev_t *mddev, char *page)
3209 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3210 mddev->sync_speed_min ? "local": "system");
3214 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3218 if (strncmp(buf, "system", 6)==0) {
3219 mddev->sync_speed_min = 0;
3222 min = simple_strtoul(buf, &e, 10);
3223 if (buf == e || (*e && *e != '\n') || min <= 0)
3225 mddev->sync_speed_min = min;
3229 static struct md_sysfs_entry md_sync_min =
3230 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3233 sync_max_show(mddev_t *mddev, char *page)
3235 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3236 mddev->sync_speed_max ? "local": "system");
3240 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3244 if (strncmp(buf, "system", 6)==0) {
3245 mddev->sync_speed_max = 0;
3248 max = simple_strtoul(buf, &e, 10);
3249 if (buf == e || (*e && *e != '\n') || max <= 0)
3251 mddev->sync_speed_max = max;
3255 static struct md_sysfs_entry md_sync_max =
3256 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3259 degraded_show(mddev_t *mddev, char *page)
3261 return sprintf(page, "%d\n", mddev->degraded);
3263 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3266 sync_force_parallel_show(mddev_t *mddev, char *page)
3268 return sprintf(page, "%d\n", mddev->parallel_resync);
3272 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3276 if (strict_strtol(buf, 10, &n))
3279 if (n != 0 && n != 1)
3282 mddev->parallel_resync = n;
3284 if (mddev->sync_thread)
3285 wake_up(&resync_wait);
3290 /* force parallel resync, even with shared block devices */
3291 static struct md_sysfs_entry md_sync_force_parallel =
3292 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3293 sync_force_parallel_show, sync_force_parallel_store);
3296 sync_speed_show(mddev_t *mddev, char *page)
3298 unsigned long resync, dt, db;
3299 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3300 dt = (jiffies - mddev->resync_mark) / HZ;
3302 db = resync - mddev->resync_mark_cnt;
3303 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3306 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3309 sync_completed_show(mddev_t *mddev, char *page)
3311 unsigned long max_sectors, resync;
3313 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3314 max_sectors = mddev->resync_max_sectors;
3316 max_sectors = mddev->dev_sectors;
3318 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3319 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3322 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3325 min_sync_show(mddev_t *mddev, char *page)
3327 return sprintf(page, "%llu\n",
3328 (unsigned long long)mddev->resync_min);
3331 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3333 unsigned long long min;
3334 if (strict_strtoull(buf, 10, &min))
3336 if (min > mddev->resync_max)
3338 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3341 /* Must be a multiple of chunk_size */
3342 if (mddev->chunk_size) {
3343 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3346 mddev->resync_min = min;
3351 static struct md_sysfs_entry md_min_sync =
3352 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3355 max_sync_show(mddev_t *mddev, char *page)
3357 if (mddev->resync_max == MaxSector)
3358 return sprintf(page, "max\n");
3360 return sprintf(page, "%llu\n",
3361 (unsigned long long)mddev->resync_max);
3364 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3366 if (strncmp(buf, "max", 3) == 0)
3367 mddev->resync_max = MaxSector;
3369 unsigned long long max;
3370 if (strict_strtoull(buf, 10, &max))
3372 if (max < mddev->resync_min)
3374 if (max < mddev->resync_max &&
3375 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3378 /* Must be a multiple of chunk_size */
3379 if (mddev->chunk_size) {
3380 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3383 mddev->resync_max = max;
3385 wake_up(&mddev->recovery_wait);
3389 static struct md_sysfs_entry md_max_sync =
3390 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3393 suspend_lo_show(mddev_t *mddev, char *page)
3395 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3399 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3402 unsigned long long new = simple_strtoull(buf, &e, 10);
3404 if (mddev->pers->quiesce == NULL)
3406 if (buf == e || (*e && *e != '\n'))
3408 if (new >= mddev->suspend_hi ||
3409 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3410 mddev->suspend_lo = new;
3411 mddev->pers->quiesce(mddev, 2);
3416 static struct md_sysfs_entry md_suspend_lo =
3417 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3421 suspend_hi_show(mddev_t *mddev, char *page)
3423 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3427 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3430 unsigned long long new = simple_strtoull(buf, &e, 10);
3432 if (mddev->pers->quiesce == NULL)
3434 if (buf == e || (*e && *e != '\n'))
3436 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3437 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3438 mddev->suspend_hi = new;
3439 mddev->pers->quiesce(mddev, 1);
3440 mddev->pers->quiesce(mddev, 0);
3445 static struct md_sysfs_entry md_suspend_hi =
3446 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3449 reshape_position_show(mddev_t *mddev, char *page)
3451 if (mddev->reshape_position != MaxSector)
3452 return sprintf(page, "%llu\n",
3453 (unsigned long long)mddev->reshape_position);
3454 strcpy(page, "none\n");
3459 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3462 unsigned long long new = simple_strtoull(buf, &e, 10);
3465 if (buf == e || (*e && *e != '\n'))
3467 mddev->reshape_position = new;
3468 mddev->delta_disks = 0;
3469 mddev->new_level = mddev->level;
3470 mddev->new_layout = mddev->layout;
3471 mddev->new_chunk = mddev->chunk_size;
3475 static struct md_sysfs_entry md_reshape_position =
3476 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3477 reshape_position_store);
3480 static struct attribute *md_default_attrs[] = {
3483 &md_raid_disks.attr,
3484 &md_chunk_size.attr,
3486 &md_resync_start.attr,
3488 &md_new_device.attr,
3489 &md_safe_delay.attr,
3490 &md_array_state.attr,
3491 &md_reshape_position.attr,
3495 static struct attribute *md_redundancy_attrs[] = {
3497 &md_mismatches.attr,
3500 &md_sync_speed.attr,
3501 &md_sync_force_parallel.attr,
3502 &md_sync_completed.attr,
3505 &md_suspend_lo.attr,
3506 &md_suspend_hi.attr,
3511 static struct attribute_group md_redundancy_group = {
3513 .attrs = md_redundancy_attrs,
3518 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3520 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3521 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3526 rv = mddev_lock(mddev);
3528 rv = entry->show(mddev, page);
3529 mddev_unlock(mddev);
3535 md_attr_store(struct kobject *kobj, struct attribute *attr,
3536 const char *page, size_t length)
3538 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3539 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3544 if (!capable(CAP_SYS_ADMIN))
3546 rv = mddev_lock(mddev);
3547 if (mddev->hold_active == UNTIL_IOCTL)
3548 mddev->hold_active = 0;
3550 rv = entry->store(mddev, page, length);
3551 mddev_unlock(mddev);
3556 static void md_free(struct kobject *ko)
3558 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3560 if (mddev->sysfs_state)
3561 sysfs_put(mddev->sysfs_state);
3563 if (mddev->gendisk) {
3564 del_gendisk(mddev->gendisk);
3565 put_disk(mddev->gendisk);
3568 blk_cleanup_queue(mddev->queue);
3573 static struct sysfs_ops md_sysfs_ops = {
3574 .show = md_attr_show,
3575 .store = md_attr_store,
3577 static struct kobj_type md_ktype = {
3579 .sysfs_ops = &md_sysfs_ops,
3580 .default_attrs = md_default_attrs,
3585 static int md_alloc(dev_t dev, char *name)
3587 static DEFINE_MUTEX(disks_mutex);
3588 mddev_t *mddev = mddev_find(dev);
3589 struct gendisk *disk;
3598 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3599 shift = partitioned ? MdpMinorShift : 0;
3600 unit = MINOR(mddev->unit) >> shift;
3602 /* wait for any previous instance if this device
3603 * to be completed removed (mddev_delayed_delete).
3605 flush_scheduled_work();
3607 mutex_lock(&disks_mutex);
3608 if (mddev->gendisk) {
3609 mutex_unlock(&disks_mutex);
3615 /* Need to ensure that 'name' is not a duplicate.
3618 spin_lock(&all_mddevs_lock);
3620 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3621 if (mddev2->gendisk &&
3622 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3623 spin_unlock(&all_mddevs_lock);
3626 spin_unlock(&all_mddevs_lock);
3629 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3630 if (!mddev->queue) {
3631 mutex_unlock(&disks_mutex);
3635 /* Can be unlocked because the queue is new: no concurrency */
3636 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3638 blk_queue_make_request(mddev->queue, md_fail_request);
3640 disk = alloc_disk(1 << shift);
3642 mutex_unlock(&disks_mutex);
3643 blk_cleanup_queue(mddev->queue);
3644 mddev->queue = NULL;
3648 disk->major = MAJOR(mddev->unit);
3649 disk->first_minor = unit << shift;
3651 strcpy(disk->disk_name, name);
3652 else if (partitioned)
3653 sprintf(disk->disk_name, "md_d%d", unit);
3655 sprintf(disk->disk_name, "md%d", unit);
3656 disk->fops = &md_fops;
3657 disk->private_data = mddev;
3658 disk->queue = mddev->queue;
3659 /* Allow extended partitions. This makes the
3660 * 'mdp' device redundant, but we can't really
3663 disk->flags |= GENHD_FL_EXT_DEVT;
3665 mddev->gendisk = disk;
3666 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3667 &disk_to_dev(disk)->kobj, "%s", "md");
3668 mutex_unlock(&disks_mutex);
3670 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3673 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3674 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3680 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3682 md_alloc(dev, NULL);
3686 static int add_named_array(const char *val, struct kernel_param *kp)
3688 /* val must be "md_*" where * is not all digits.
3689 * We allocate an array with a large free minor number, and
3690 * set the name to val. val must not already be an active name.
3692 int len = strlen(val);
3693 char buf[DISK_NAME_LEN];
3695 while (len && val[len-1] == '\n')
3697 if (len >= DISK_NAME_LEN)
3699 strlcpy(buf, val, len+1);
3700 if (strncmp(buf, "md_", 3) != 0)
3702 return md_alloc(0, buf);
3705 static void md_safemode_timeout(unsigned long data)
3707 mddev_t *mddev = (mddev_t *) data;
3709 if (!atomic_read(&mddev->writes_pending)) {
3710 mddev->safemode = 1;
3711 if (mddev->external)
3712 sysfs_notify_dirent(mddev->sysfs_state);
3714 md_wakeup_thread(mddev->thread);
3717 static int start_dirty_degraded;
3719 static int do_md_run(mddev_t * mddev)
3724 struct gendisk *disk;
3725 struct mdk_personality *pers;
3726 char b[BDEVNAME_SIZE];
3728 if (list_empty(&mddev->disks))
3729 /* cannot run an array with no devices.. */
3736 * Analyze all RAID superblock(s)
3738 if (!mddev->raid_disks) {
3739 if (!mddev->persistent)
3744 chunk_size = mddev->chunk_size;
3747 if (chunk_size > MAX_CHUNK_SIZE) {
3748 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3749 chunk_size, MAX_CHUNK_SIZE);
3753 * chunk-size has to be a power of 2
3755 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3756 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3760 /* devices must have minimum size of one chunk */
3761 list_for_each_entry(rdev, &mddev->disks, same_set) {
3762 if (test_bit(Faulty, &rdev->flags))
3764 if (rdev->sectors < chunk_size / 512) {
3766 "md: Dev %s smaller than chunk_size:"
3768 bdevname(rdev->bdev,b),
3769 (unsigned long long)rdev->sectors,
3776 if (mddev->level != LEVEL_NONE)
3777 request_module("md-level-%d", mddev->level);
3778 else if (mddev->clevel[0])
3779 request_module("md-%s", mddev->clevel);
3782 * Drop all container device buffers, from now on
3783 * the only valid external interface is through the md
3786 list_for_each_entry(rdev, &mddev->disks, same_set) {
3787 if (test_bit(Faulty, &rdev->flags))
3789 sync_blockdev(rdev->bdev);
3790 invalidate_bdev(rdev->bdev);
3792 /* perform some consistency tests on the device.
3793 * We don't want the data to overlap the metadata,
3794 * Internal Bitmap issues have been handled elsewhere.
3796 if (rdev->data_offset < rdev->sb_start) {
3797 if (mddev->dev_sectors &&
3798 rdev->data_offset + mddev->dev_sectors
3800 printk("md: %s: data overlaps metadata\n",
3805 if (rdev->sb_start + rdev->sb_size/512
3806 > rdev->data_offset) {
3807 printk("md: %s: metadata overlaps data\n",
3812 sysfs_notify_dirent(rdev->sysfs_state);
3815 md_probe(mddev->unit, NULL, NULL);
3816 disk = mddev->gendisk;
3820 spin_lock(&pers_lock);
3821 pers = find_pers(mddev->level, mddev->clevel);
3822 if (!pers || !try_module_get(pers->owner)) {
3823 spin_unlock(&pers_lock);
3824 if (mddev->level != LEVEL_NONE)
3825 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3828 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3833 spin_unlock(&pers_lock);
3834 mddev->level = pers->level;
3835 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3837 if (pers->level >= 4 && pers->level <= 6)
3838 /* Cannot support integrity (yet) */
3839 blk_integrity_unregister(mddev->gendisk);
3841 if (mddev->reshape_position != MaxSector &&
3842 pers->start_reshape == NULL) {
3843 /* This personality cannot handle reshaping... */
3845 module_put(pers->owner);
3849 if (pers->sync_request) {
3850 /* Warn if this is a potentially silly
3853 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3857 list_for_each_entry(rdev, &mddev->disks, same_set)
3858 list_for_each_entry(rdev2, &mddev->disks, same_set) {
3860 rdev->bdev->bd_contains ==
3861 rdev2->bdev->bd_contains) {
3863 "%s: WARNING: %s appears to be"
3864 " on the same physical disk as"
3867 bdevname(rdev->bdev,b),
3868 bdevname(rdev2->bdev,b2));
3875 "True protection against single-disk"
3876 " failure might be compromised.\n");
3879 mddev->recovery = 0;
3880 /* may be over-ridden by personality */
3881 mddev->resync_max_sectors = mddev->dev_sectors;
3883 mddev->barriers_work = 1;
3884 mddev->ok_start_degraded = start_dirty_degraded;
3887 mddev->ro = 2; /* read-only, but switch on first write */
3889 err = mddev->pers->run(mddev);
3891 printk(KERN_ERR "md: pers->run() failed ...\n");
3892 else if (mddev->pers->sync_request) {
3893 err = bitmap_create(mddev);
3895 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3896 mdname(mddev), err);
3897 mddev->pers->stop(mddev);
3901 module_put(mddev->pers->owner);
3903 bitmap_destroy(mddev);
3906 if (mddev->pers->sync_request) {
3907 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3909 "md: cannot register extra attributes for %s\n",
3911 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3912 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3915 atomic_set(&mddev->writes_pending,0);
3916 mddev->safemode = 0;
3917 mddev->safemode_timer.function = md_safemode_timeout;
3918 mddev->safemode_timer.data = (unsigned long) mddev;
3919 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3922 list_for_each_entry(rdev, &mddev->disks, same_set)
3923 if (rdev->raid_disk >= 0) {
3925 sprintf(nm, "rd%d", rdev->raid_disk);
3926 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3927 printk("md: cannot register %s for %s\n",
3931 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3934 md_update_sb(mddev, 0);
3936 set_capacity(disk, mddev->array_sectors);
3938 /* If we call blk_queue_make_request here, it will
3939 * re-initialise max_sectors etc which may have been
3940 * refined inside -> run. So just set the bits we need to set.
3941 * Most initialisation happended when we called
3942 * blk_queue_make_request(..., md_fail_request)
3945 mddev->queue->queuedata = mddev;
3946 mddev->queue->make_request_fn = mddev->pers->make_request;
3948 /* If there is a partially-recovered drive we need to
3949 * start recovery here. If we leave it to md_check_recovery,
3950 * it will remove the drives and not do the right thing
3952 if (mddev->degraded && !mddev->sync_thread) {
3954 list_for_each_entry(rdev, &mddev->disks, same_set)
3955 if (rdev->raid_disk >= 0 &&
3956 !test_bit(In_sync, &rdev->flags) &&
3957 !test_bit(Faulty, &rdev->flags))
3958 /* complete an interrupted recovery */
3960 if (spares && mddev->pers->sync_request) {
3961 mddev->recovery = 0;
3962 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3963 mddev->sync_thread = md_register_thread(md_do_sync,
3966 if (!mddev->sync_thread) {
3967 printk(KERN_ERR "%s: could not start resync"
3970 /* leave the spares where they are, it shouldn't hurt */
3971 mddev->recovery = 0;
3975 md_wakeup_thread(mddev->thread);
3976 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3979 md_new_event(mddev);
3980 sysfs_notify_dirent(mddev->sysfs_state);
3981 if (mddev->sysfs_action)
3982 sysfs_notify_dirent(mddev->sysfs_action);
3983 sysfs_notify(&mddev->kobj, NULL, "degraded");
3984 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3988 static int restart_array(mddev_t *mddev)
3990 struct gendisk *disk = mddev->gendisk;
3992 /* Complain if it has no devices */
3993 if (list_empty(&mddev->disks))
3999 mddev->safemode = 0;
4001 set_disk_ro(disk, 0);
4002 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4004 /* Kick recovery or resync if necessary */
4005 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4006 md_wakeup_thread(mddev->thread);
4007 md_wakeup_thread(mddev->sync_thread);
4008 sysfs_notify_dirent(mddev->sysfs_state);
4012 /* similar to deny_write_access, but accounts for our holding a reference
4013 * to the file ourselves */
4014 static int deny_bitmap_write_access(struct file * file)
4016 struct inode *inode = file->f_mapping->host;
4018 spin_lock(&inode->i_lock);
4019 if (atomic_read(&inode->i_writecount) > 1) {
4020 spin_unlock(&inode->i_lock);
4023 atomic_set(&inode->i_writecount, -1);
4024 spin_unlock(&inode->i_lock);
4029 static void restore_bitmap_write_access(struct file *file)
4031 struct inode *inode = file->f_mapping->host;
4033 spin_lock(&inode->i_lock);
4034 atomic_set(&inode->i_writecount, 1);
4035 spin_unlock(&inode->i_lock);
4039 * 0 - completely stop and dis-assemble array
4040 * 1 - switch to readonly
4041 * 2 - stop but do not disassemble array
4043 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4046 struct gendisk *disk = mddev->gendisk;
4048 if (atomic_read(&mddev->openers) > is_open) {
4049 printk("md: %s still in use.\n",mdname(mddev));
4055 if (mddev->sync_thread) {
4056 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4057 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4058 md_unregister_thread(mddev->sync_thread);
4059 mddev->sync_thread = NULL;
4062 del_timer_sync(&mddev->safemode_timer);
4065 case 1: /* readonly */
4071 case 0: /* disassemble */
4073 bitmap_flush(mddev);
4074 md_super_wait(mddev);
4076 set_disk_ro(disk, 0);
4077 blk_queue_make_request(mddev->queue, md_fail_request);
4078 mddev->pers->stop(mddev);
4079 mddev->queue->merge_bvec_fn = NULL;
4080 mddev->queue->unplug_fn = NULL;
4081 mddev->queue->backing_dev_info.congested_fn = NULL;
4082 if (mddev->pers->sync_request) {
4083 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4084 if (mddev->sysfs_action)
4085 sysfs_put(mddev->sysfs_action);
4086 mddev->sysfs_action = NULL;
4088 module_put(mddev->pers->owner);
4090 /* tell userspace to handle 'inactive' */
4091 sysfs_notify_dirent(mddev->sysfs_state);
4093 set_capacity(disk, 0);
4099 if (!mddev->in_sync || mddev->flags) {
4100 /* mark array as shutdown cleanly */
4102 md_update_sb(mddev, 1);
4105 set_disk_ro(disk, 1);
4106 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4110 * Free resources if final stop
4115 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4117 bitmap_destroy(mddev);
4118 if (mddev->bitmap_file) {
4119 restore_bitmap_write_access(mddev->bitmap_file);
4120 fput(mddev->bitmap_file);
4121 mddev->bitmap_file = NULL;
4123 mddev->bitmap_offset = 0;
4125 list_for_each_entry(rdev, &mddev->disks, same_set)
4126 if (rdev->raid_disk >= 0) {
4128 sprintf(nm, "rd%d", rdev->raid_disk);
4129 sysfs_remove_link(&mddev->kobj, nm);
4132 /* make sure all md_delayed_delete calls have finished */
4133 flush_scheduled_work();
4135 export_array(mddev);
4137 mddev->array_sectors = 0;
4138 mddev->dev_sectors = 0;
4139 mddev->raid_disks = 0;
4140 mddev->recovery_cp = 0;
4141 mddev->resync_min = 0;
4142 mddev->resync_max = MaxSector;
4143 mddev->reshape_position = MaxSector;
4144 mddev->external = 0;
4145 mddev->persistent = 0;
4146 mddev->level = LEVEL_NONE;
4147 mddev->clevel[0] = 0;
4150 mddev->metadata_type[0] = 0;
4151 mddev->chunk_size = 0;
4152 mddev->ctime = mddev->utime = 0;
4154 mddev->max_disks = 0;
4156 mddev->delta_disks = 0;
4157 mddev->new_level = LEVEL_NONE;
4158 mddev->new_layout = 0;
4159 mddev->new_chunk = 0;
4160 mddev->curr_resync = 0;
4161 mddev->resync_mismatches = 0;
4162 mddev->suspend_lo = mddev->suspend_hi = 0;
4163 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4164 mddev->recovery = 0;
4167 mddev->degraded = 0;
4168 mddev->barriers_work = 0;
4169 mddev->safemode = 0;
4170 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4171 if (mddev->hold_active == UNTIL_STOP)
4172 mddev->hold_active = 0;
4174 } else if (mddev->pers)
4175 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4178 blk_integrity_unregister(disk);
4179 md_new_event(mddev);
4180 sysfs_notify_dirent(mddev->sysfs_state);
4186 static void autorun_array(mddev_t *mddev)
4191 if (list_empty(&mddev->disks))
4194 printk(KERN_INFO "md: running: ");
4196 list_for_each_entry(rdev, &mddev->disks, same_set) {
4197 char b[BDEVNAME_SIZE];
4198 printk("<%s>", bdevname(rdev->bdev,b));
4202 err = do_md_run(mddev);
4204 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4205 do_md_stop(mddev, 0, 0);
4210 * lets try to run arrays based on all disks that have arrived
4211 * until now. (those are in pending_raid_disks)
4213 * the method: pick the first pending disk, collect all disks with
4214 * the same UUID, remove all from the pending list and put them into
4215 * the 'same_array' list. Then order this list based on superblock
4216 * update time (freshest comes first), kick out 'old' disks and
4217 * compare superblocks. If everything's fine then run it.
4219 * If "unit" is allocated, then bump its reference count
4221 static void autorun_devices(int part)
4223 mdk_rdev_t *rdev0, *rdev, *tmp;
4225 char b[BDEVNAME_SIZE];
4227 printk(KERN_INFO "md: autorun ...\n");
4228 while (!list_empty(&pending_raid_disks)) {
4231 LIST_HEAD(candidates);
4232 rdev0 = list_entry(pending_raid_disks.next,
4233 mdk_rdev_t, same_set);
4235 printk(KERN_INFO "md: considering %s ...\n",
4236 bdevname(rdev0->bdev,b));
4237 INIT_LIST_HEAD(&candidates);
4238 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4239 if (super_90_load(rdev, rdev0, 0) >= 0) {
4240 printk(KERN_INFO "md: adding %s ...\n",
4241 bdevname(rdev->bdev,b));
4242 list_move(&rdev->same_set, &candidates);
4245 * now we have a set of devices, with all of them having
4246 * mostly sane superblocks. It's time to allocate the
4250 dev = MKDEV(mdp_major,
4251 rdev0->preferred_minor << MdpMinorShift);
4252 unit = MINOR(dev) >> MdpMinorShift;
4254 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4257 if (rdev0->preferred_minor != unit) {
4258 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4259 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4263 md_probe(dev, NULL, NULL);
4264 mddev = mddev_find(dev);
4265 if (!mddev || !mddev->gendisk) {
4269 "md: cannot allocate memory for md drive.\n");
4272 if (mddev_lock(mddev))
4273 printk(KERN_WARNING "md: %s locked, cannot run\n",
4275 else if (mddev->raid_disks || mddev->major_version
4276 || !list_empty(&mddev->disks)) {
4278 "md: %s already running, cannot run %s\n",
4279 mdname(mddev), bdevname(rdev0->bdev,b));
4280 mddev_unlock(mddev);
4282 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4283 mddev->persistent = 1;
4284 rdev_for_each_list(rdev, tmp, &candidates) {
4285 list_del_init(&rdev->same_set);
4286 if (bind_rdev_to_array(rdev, mddev))
4289 autorun_array(mddev);
4290 mddev_unlock(mddev);
4292 /* on success, candidates will be empty, on error
4295 rdev_for_each_list(rdev, tmp, &candidates) {
4296 list_del_init(&rdev->same_set);
4301 printk(KERN_INFO "md: ... autorun DONE.\n");
4303 #endif /* !MODULE */
4305 static int get_version(void __user * arg)
4309 ver.major = MD_MAJOR_VERSION;
4310 ver.minor = MD_MINOR_VERSION;
4311 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4313 if (copy_to_user(arg, &ver, sizeof(ver)))
4319 static int get_array_info(mddev_t * mddev, void __user * arg)
4321 mdu_array_info_t info;
4322 int nr,working,active,failed,spare;
4325 nr=working=active=failed=spare=0;
4326 list_for_each_entry(rdev, &mddev->disks, same_set) {
4328 if (test_bit(Faulty, &rdev->flags))
4332 if (test_bit(In_sync, &rdev->flags))
4339 info.major_version = mddev->major_version;
4340 info.minor_version = mddev->minor_version;
4341 info.patch_version = MD_PATCHLEVEL_VERSION;
4342 info.ctime = mddev->ctime;
4343 info.level = mddev->level;
4344 info.size = mddev->dev_sectors / 2;
4345 if (info.size != mddev->dev_sectors / 2) /* overflow */
4348 info.raid_disks = mddev->raid_disks;
4349 info.md_minor = mddev->md_minor;
4350 info.not_persistent= !mddev->persistent;
4352 info.utime = mddev->utime;
4355 info.state = (1<<MD_SB_CLEAN);
4356 if (mddev->bitmap && mddev->bitmap_offset)
4357 info.state = (1<<MD_SB_BITMAP_PRESENT);
4358 info.active_disks = active;
4359 info.working_disks = working;
4360 info.failed_disks = failed;
4361 info.spare_disks = spare;
4363 info.layout = mddev->layout;
4364 info.chunk_size = mddev->chunk_size;
4366 if (copy_to_user(arg, &info, sizeof(info)))
4372 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4374 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4375 char *ptr, *buf = NULL;
4378 if (md_allow_write(mddev))
4379 file = kmalloc(sizeof(*file), GFP_NOIO);
4381 file = kmalloc(sizeof(*file), GFP_KERNEL);
4386 /* bitmap disabled, zero the first byte and copy out */
4387 if (!mddev->bitmap || !mddev->bitmap->file) {
4388 file->pathname[0] = '\0';
4392 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4396 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4400 strcpy(file->pathname, ptr);
4404 if (copy_to_user(arg, file, sizeof(*file)))
4412 static int get_disk_info(mddev_t * mddev, void __user * arg)
4414 mdu_disk_info_t info;
4417 if (copy_from_user(&info, arg, sizeof(info)))
4420 rdev = find_rdev_nr(mddev, info.number);
4422 info.major = MAJOR(rdev->bdev->bd_dev);
4423 info.minor = MINOR(rdev->bdev->bd_dev);
4424 info.raid_disk = rdev->raid_disk;
4426 if (test_bit(Faulty, &rdev->flags))
4427 info.state |= (1<<MD_DISK_FAULTY);
4428 else if (test_bit(In_sync, &rdev->flags)) {
4429 info.state |= (1<<MD_DISK_ACTIVE);
4430 info.state |= (1<<MD_DISK_SYNC);
4432 if (test_bit(WriteMostly, &rdev->flags))
4433 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4435 info.major = info.minor = 0;
4436 info.raid_disk = -1;
4437 info.state = (1<<MD_DISK_REMOVED);
4440 if (copy_to_user(arg, &info, sizeof(info)))
4446 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4448 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4450 dev_t dev = MKDEV(info->major,info->minor);
4452 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4455 if (!mddev->raid_disks) {
4457 /* expecting a device which has a superblock */
4458 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4461 "md: md_import_device returned %ld\n",
4463 return PTR_ERR(rdev);
4465 if (!list_empty(&mddev->disks)) {
4466 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4467 mdk_rdev_t, same_set);
4468 int err = super_types[mddev->major_version]
4469 .load_super(rdev, rdev0, mddev->minor_version);
4472 "md: %s has different UUID to %s\n",
4473 bdevname(rdev->bdev,b),
4474 bdevname(rdev0->bdev,b2));
4479 err = bind_rdev_to_array(rdev, mddev);
4486 * add_new_disk can be used once the array is assembled
4487 * to add "hot spares". They must already have a superblock
4492 if (!mddev->pers->hot_add_disk) {
4494 "%s: personality does not support diskops!\n",
4498 if (mddev->persistent)
4499 rdev = md_import_device(dev, mddev->major_version,
4500 mddev->minor_version);
4502 rdev = md_import_device(dev, -1, -1);
4505 "md: md_import_device returned %ld\n",
4507 return PTR_ERR(rdev);
4509 /* set save_raid_disk if appropriate */
4510 if (!mddev->persistent) {
4511 if (info->state & (1<<MD_DISK_SYNC) &&
4512 info->raid_disk < mddev->raid_disks)
4513 rdev->raid_disk = info->raid_disk;
4515 rdev->raid_disk = -1;
4517 super_types[mddev->major_version].
4518 validate_super(mddev, rdev);
4519 rdev->saved_raid_disk = rdev->raid_disk;
4521 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4522 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4523 set_bit(WriteMostly, &rdev->flags);
4525 clear_bit(WriteMostly, &rdev->flags);
4527 rdev->raid_disk = -1;
4528 err = bind_rdev_to_array(rdev, mddev);
4529 if (!err && !mddev->pers->hot_remove_disk) {
4530 /* If there is hot_add_disk but no hot_remove_disk
4531 * then added disks for geometry changes,
4532 * and should be added immediately.
4534 super_types[mddev->major_version].
4535 validate_super(mddev, rdev);
4536 err = mddev->pers->hot_add_disk(mddev, rdev);
4538 unbind_rdev_from_array(rdev);
4543 sysfs_notify_dirent(rdev->sysfs_state);
4545 md_update_sb(mddev, 1);
4546 if (mddev->degraded)
4547 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4548 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4549 md_wakeup_thread(mddev->thread);
4553 /* otherwise, add_new_disk is only allowed
4554 * for major_version==0 superblocks
4556 if (mddev->major_version != 0) {
4557 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4562 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4564 rdev = md_import_device(dev, -1, 0);
4567 "md: error, md_import_device() returned %ld\n",
4569 return PTR_ERR(rdev);
4571 rdev->desc_nr = info->number;
4572 if (info->raid_disk < mddev->raid_disks)
4573 rdev->raid_disk = info->raid_disk;
4575 rdev->raid_disk = -1;
4577 if (rdev->raid_disk < mddev->raid_disks)
4578 if (info->state & (1<<MD_DISK_SYNC))
4579 set_bit(In_sync, &rdev->flags);
4581 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4582 set_bit(WriteMostly, &rdev->flags);
4584 if (!mddev->persistent) {
4585 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4586 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4588 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4589 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4591 err = bind_rdev_to_array(rdev, mddev);
4601 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4603 char b[BDEVNAME_SIZE];
4606 rdev = find_rdev(mddev, dev);
4610 if (rdev->raid_disk >= 0)
4613 kick_rdev_from_array(rdev);
4614 md_update_sb(mddev, 1);
4615 md_new_event(mddev);
4619 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4620 bdevname(rdev->bdev,b), mdname(mddev));
4624 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4626 char b[BDEVNAME_SIZE];
4633 if (mddev->major_version != 0) {
4634 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4635 " version-0 superblocks.\n",
4639 if (!mddev->pers->hot_add_disk) {
4641 "%s: personality does not support diskops!\n",
4646 rdev = md_import_device(dev, -1, 0);
4649 "md: error, md_import_device() returned %ld\n",
4654 if (mddev->persistent)
4655 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4657 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4659 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4661 if (test_bit(Faulty, &rdev->flags)) {
4663 "md: can not hot-add faulty %s disk to %s!\n",
4664 bdevname(rdev->bdev,b), mdname(mddev));
4668 clear_bit(In_sync, &rdev->flags);
4670 rdev->saved_raid_disk = -1;
4671 err = bind_rdev_to_array(rdev, mddev);
4676 * The rest should better be atomic, we can have disk failures
4677 * noticed in interrupt contexts ...
4680 rdev->raid_disk = -1;
4682 md_update_sb(mddev, 1);
4685 * Kick recovery, maybe this spare has to be added to the
4686 * array immediately.
4688 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4689 md_wakeup_thread(mddev->thread);
4690 md_new_event(mddev);
4698 static int set_bitmap_file(mddev_t *mddev, int fd)
4703 if (!mddev->pers->quiesce)
4705 if (mddev->recovery || mddev->sync_thread)
4707 /* we should be able to change the bitmap.. */
4713 return -EEXIST; /* cannot add when bitmap is present */
4714 mddev->bitmap_file = fget(fd);
4716 if (mddev->bitmap_file == NULL) {
4717 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4722 err = deny_bitmap_write_access(mddev->bitmap_file);
4724 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4726 fput(mddev->bitmap_file);
4727 mddev->bitmap_file = NULL;
4730 mddev->bitmap_offset = 0; /* file overrides offset */
4731 } else if (mddev->bitmap == NULL)
4732 return -ENOENT; /* cannot remove what isn't there */
4735 mddev->pers->quiesce(mddev, 1);
4737 err = bitmap_create(mddev);
4738 if (fd < 0 || err) {
4739 bitmap_destroy(mddev);
4740 fd = -1; /* make sure to put the file */
4742 mddev->pers->quiesce(mddev, 0);
4745 if (mddev->bitmap_file) {
4746 restore_bitmap_write_access(mddev->bitmap_file);
4747 fput(mddev->bitmap_file);
4749 mddev->bitmap_file = NULL;
4756 * set_array_info is used two different ways
4757 * The original usage is when creating a new array.
4758 * In this usage, raid_disks is > 0 and it together with
4759 * level, size, not_persistent,layout,chunksize determine the
4760 * shape of the array.
4761 * This will always create an array with a type-0.90.0 superblock.
4762 * The newer usage is when assembling an array.
4763 * In this case raid_disks will be 0, and the major_version field is
4764 * use to determine which style super-blocks are to be found on the devices.
4765 * The minor and patch _version numbers are also kept incase the
4766 * super_block handler wishes to interpret them.
4768 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4771 if (info->raid_disks == 0) {
4772 /* just setting version number for superblock loading */
4773 if (info->major_version < 0 ||
4774 info->major_version >= ARRAY_SIZE(super_types) ||
4775 super_types[info->major_version].name == NULL) {
4776 /* maybe try to auto-load a module? */
4778 "md: superblock version %d not known\n",
4779 info->major_version);
4782 mddev->major_version = info->major_version;
4783 mddev->minor_version = info->minor_version;
4784 mddev->patch_version = info->patch_version;
4785 mddev->persistent = !info->not_persistent;
4788 mddev->major_version = MD_MAJOR_VERSION;
4789 mddev->minor_version = MD_MINOR_VERSION;
4790 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4791 mddev->ctime = get_seconds();
4793 mddev->level = info->level;
4794 mddev->clevel[0] = 0;
4795 mddev->dev_sectors = 2 * (sector_t)info->size;
4796 mddev->raid_disks = info->raid_disks;
4797 /* don't set md_minor, it is determined by which /dev/md* was
4800 if (info->state & (1<<MD_SB_CLEAN))
4801 mddev->recovery_cp = MaxSector;
4803 mddev->recovery_cp = 0;
4804 mddev->persistent = ! info->not_persistent;
4805 mddev->external = 0;
4807 mddev->layout = info->layout;
4808 mddev->chunk_size = info->chunk_size;
4810 mddev->max_disks = MD_SB_DISKS;
4812 if (mddev->persistent)
4814 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4816 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4817 mddev->bitmap_offset = 0;
4819 mddev->reshape_position = MaxSector;
4822 * Generate a 128 bit UUID
4824 get_random_bytes(mddev->uuid, 16);
4826 mddev->new_level = mddev->level;
4827 mddev->new_chunk = mddev->chunk_size;
4828 mddev->new_layout = mddev->layout;
4829 mddev->delta_disks = 0;
4834 static int update_size(mddev_t *mddev, sector_t num_sectors)
4838 int fit = (num_sectors == 0);
4840 if (mddev->pers->resize == NULL)
4842 /* The "num_sectors" is the number of sectors of each device that
4843 * is used. This can only make sense for arrays with redundancy.
4844 * linear and raid0 always use whatever space is available. We can only
4845 * consider changing this number if no resync or reconstruction is
4846 * happening, and if the new size is acceptable. It must fit before the
4847 * sb_start or, if that is <data_offset, it must fit before the size
4848 * of each device. If num_sectors is zero, we find the largest size
4852 if (mddev->sync_thread)
4855 /* Sorry, cannot grow a bitmap yet, just remove it,
4859 list_for_each_entry(rdev, &mddev->disks, same_set) {
4860 sector_t avail = rdev->sectors;
4862 if (fit && (num_sectors == 0 || num_sectors > avail))
4863 num_sectors = avail;
4864 if (avail < num_sectors)
4867 rv = mddev->pers->resize(mddev, num_sectors);
4869 struct block_device *bdev;
4871 bdev = bdget_disk(mddev->gendisk, 0);
4873 mutex_lock(&bdev->bd_inode->i_mutex);
4874 i_size_write(bdev->bd_inode,
4875 (loff_t)mddev->array_sectors << 9);
4876 mutex_unlock(&bdev->bd_inode->i_mutex);
4883 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4886 /* change the number of raid disks */
4887 if (mddev->pers->check_reshape == NULL)
4889 if (raid_disks <= 0 ||
4890 raid_disks >= mddev->max_disks)
4892 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4894 mddev->delta_disks = raid_disks - mddev->raid_disks;
4896 rv = mddev->pers->check_reshape(mddev);
4902 * update_array_info is used to change the configuration of an
4904 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4905 * fields in the info are checked against the array.
4906 * Any differences that cannot be handled will cause an error.
4907 * Normally, only one change can be managed at a time.
4909 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4915 /* calculate expected state,ignoring low bits */
4916 if (mddev->bitmap && mddev->bitmap_offset)
4917 state |= (1 << MD_SB_BITMAP_PRESENT);
4919 if (mddev->major_version != info->major_version ||
4920 mddev->minor_version != info->minor_version ||
4921 /* mddev->patch_version != info->patch_version || */
4922 mddev->ctime != info->ctime ||
4923 mddev->level != info->level ||
4924 /* mddev->layout != info->layout || */
4925 !mddev->persistent != info->not_persistent||
4926 mddev->chunk_size != info->chunk_size ||
4927 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4928 ((state^info->state) & 0xfffffe00)
4931 /* Check there is only one change */
4932 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
4934 if (mddev->raid_disks != info->raid_disks)
4936 if (mddev->layout != info->layout)
4938 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
4945 if (mddev->layout != info->layout) {
4947 * we don't need to do anything at the md level, the
4948 * personality will take care of it all.
4950 if (mddev->pers->reconfig == NULL)
4953 return mddev->pers->reconfig(mddev, info->layout, -1);
4955 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
4956 rv = update_size(mddev, (sector_t)info->size * 2);
4958 if (mddev->raid_disks != info->raid_disks)
4959 rv = update_raid_disks(mddev, info->raid_disks);
4961 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4962 if (mddev->pers->quiesce == NULL)
4964 if (mddev->recovery || mddev->sync_thread)
4966 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4967 /* add the bitmap */
4970 if (mddev->default_bitmap_offset == 0)
4972 mddev->bitmap_offset = mddev->default_bitmap_offset;
4973 mddev->pers->quiesce(mddev, 1);
4974 rv = bitmap_create(mddev);
4976 bitmap_destroy(mddev);
4977 mddev->pers->quiesce(mddev, 0);
4979 /* remove the bitmap */
4982 if (mddev->bitmap->file)
4984 mddev->pers->quiesce(mddev, 1);
4985 bitmap_destroy(mddev);
4986 mddev->pers->quiesce(mddev, 0);
4987 mddev->bitmap_offset = 0;
4990 md_update_sb(mddev, 1);
4994 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4998 if (mddev->pers == NULL)
5001 rdev = find_rdev(mddev, dev);
5005 md_error(mddev, rdev);
5010 * We have a problem here : there is no easy way to give a CHS
5011 * virtual geometry. We currently pretend that we have a 2 heads
5012 * 4 sectors (with a BIG number of cylinders...). This drives
5013 * dosfs just mad... ;-)
5015 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5017 mddev_t *mddev = bdev->bd_disk->private_data;
5021 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5025 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5026 unsigned int cmd, unsigned long arg)
5029 void __user *argp = (void __user *)arg;
5030 mddev_t *mddev = NULL;
5032 if (!capable(CAP_SYS_ADMIN))
5036 * Commands dealing with the RAID driver but not any
5042 err = get_version(argp);
5045 case PRINT_RAID_DEBUG:
5053 autostart_arrays(arg);
5060 * Commands creating/starting a new array:
5063 mddev = bdev->bd_disk->private_data;
5070 err = mddev_lock(mddev);
5073 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5080 case SET_ARRAY_INFO:
5082 mdu_array_info_t info;
5084 memset(&info, 0, sizeof(info));
5085 else if (copy_from_user(&info, argp, sizeof(info))) {
5090 err = update_array_info(mddev, &info);
5092 printk(KERN_WARNING "md: couldn't update"
5093 " array info. %d\n", err);
5098 if (!list_empty(&mddev->disks)) {
5100 "md: array %s already has disks!\n",
5105 if (mddev->raid_disks) {
5107 "md: array %s already initialised!\n",
5112 err = set_array_info(mddev, &info);
5114 printk(KERN_WARNING "md: couldn't set"
5115 " array info. %d\n", err);
5125 * Commands querying/configuring an existing array:
5127 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5128 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5129 if ((!mddev->raid_disks && !mddev->external)
5130 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5131 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5132 && cmd != GET_BITMAP_FILE) {
5138 * Commands even a read-only array can execute:
5142 case GET_ARRAY_INFO:
5143 err = get_array_info(mddev, argp);
5146 case GET_BITMAP_FILE:
5147 err = get_bitmap_file(mddev, argp);
5151 err = get_disk_info(mddev, argp);
5154 case RESTART_ARRAY_RW:
5155 err = restart_array(mddev);
5159 err = do_md_stop(mddev, 0, 1);
5163 err = do_md_stop(mddev, 1, 1);
5169 * The remaining ioctls are changing the state of the
5170 * superblock, so we do not allow them on read-only arrays.
5171 * However non-MD ioctls (e.g. get-size) will still come through
5172 * here and hit the 'default' below, so only disallow
5173 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5175 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5176 if (mddev->ro == 2) {
5178 sysfs_notify_dirent(mddev->sysfs_state);
5179 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5180 md_wakeup_thread(mddev->thread);
5191 mdu_disk_info_t info;
5192 if (copy_from_user(&info, argp, sizeof(info)))
5195 err = add_new_disk(mddev, &info);
5199 case HOT_REMOVE_DISK:
5200 err = hot_remove_disk(mddev, new_decode_dev(arg));
5204 err = hot_add_disk(mddev, new_decode_dev(arg));
5207 case SET_DISK_FAULTY:
5208 err = set_disk_faulty(mddev, new_decode_dev(arg));
5212 err = do_md_run(mddev);
5215 case SET_BITMAP_FILE:
5216 err = set_bitmap_file(mddev, (int)arg);
5226 if (mddev->hold_active == UNTIL_IOCTL &&
5228 mddev->hold_active = 0;
5229 mddev_unlock(mddev);
5239 static int md_open(struct block_device *bdev, fmode_t mode)
5242 * Succeed if we can lock the mddev, which confirms that
5243 * it isn't being stopped right now.
5245 mddev_t *mddev = mddev_find(bdev->bd_dev);
5248 if (mddev->gendisk != bdev->bd_disk) {
5249 /* we are racing with mddev_put which is discarding this
5253 /* Wait until bdev->bd_disk is definitely gone */
5254 flush_scheduled_work();
5255 /* Then retry the open from the top */
5256 return -ERESTARTSYS;
5258 BUG_ON(mddev != bdev->bd_disk->private_data);
5260 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5264 atomic_inc(&mddev->openers);
5265 mddev_unlock(mddev);
5267 check_disk_change(bdev);
5272 static int md_release(struct gendisk *disk, fmode_t mode)
5274 mddev_t *mddev = disk->private_data;
5277 atomic_dec(&mddev->openers);
5283 static int md_media_changed(struct gendisk *disk)
5285 mddev_t *mddev = disk->private_data;
5287 return mddev->changed;
5290 static int md_revalidate(struct gendisk *disk)
5292 mddev_t *mddev = disk->private_data;
5297 static struct block_device_operations md_fops =
5299 .owner = THIS_MODULE,
5301 .release = md_release,
5302 .locked_ioctl = md_ioctl,
5303 .getgeo = md_getgeo,
5304 .media_changed = md_media_changed,
5305 .revalidate_disk= md_revalidate,
5308 static int md_thread(void * arg)
5310 mdk_thread_t *thread = arg;
5313 * md_thread is a 'system-thread', it's priority should be very
5314 * high. We avoid resource deadlocks individually in each
5315 * raid personality. (RAID5 does preallocation) We also use RR and
5316 * the very same RT priority as kswapd, thus we will never get
5317 * into a priority inversion deadlock.
5319 * we definitely have to have equal or higher priority than
5320 * bdflush, otherwise bdflush will deadlock if there are too
5321 * many dirty RAID5 blocks.
5324 allow_signal(SIGKILL);
5325 while (!kthread_should_stop()) {
5327 /* We need to wait INTERRUPTIBLE so that
5328 * we don't add to the load-average.
5329 * That means we need to be sure no signals are
5332 if (signal_pending(current))
5333 flush_signals(current);
5335 wait_event_interruptible_timeout
5337 test_bit(THREAD_WAKEUP, &thread->flags)
5338 || kthread_should_stop(),
5341 clear_bit(THREAD_WAKEUP, &thread->flags);
5343 thread->run(thread->mddev);
5349 void md_wakeup_thread(mdk_thread_t *thread)
5352 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5353 set_bit(THREAD_WAKEUP, &thread->flags);
5354 wake_up(&thread->wqueue);
5358 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5361 mdk_thread_t *thread;
5363 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5367 init_waitqueue_head(&thread->wqueue);
5370 thread->mddev = mddev;
5371 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5372 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5373 if (IS_ERR(thread->tsk)) {
5380 void md_unregister_thread(mdk_thread_t *thread)
5382 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5384 kthread_stop(thread->tsk);
5388 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5395 if (!rdev || test_bit(Faulty, &rdev->flags))
5398 if (mddev->external)
5399 set_bit(Blocked, &rdev->flags);
5401 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5403 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5404 __builtin_return_address(0),__builtin_return_address(1),
5405 __builtin_return_address(2),__builtin_return_address(3));
5409 if (!mddev->pers->error_handler)
5411 mddev->pers->error_handler(mddev,rdev);
5412 if (mddev->degraded)
5413 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5414 set_bit(StateChanged, &rdev->flags);
5415 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5416 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5417 md_wakeup_thread(mddev->thread);
5418 md_new_event_inintr(mddev);
5421 /* seq_file implementation /proc/mdstat */
5423 static void status_unused(struct seq_file *seq)
5428 seq_printf(seq, "unused devices: ");
5430 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5431 char b[BDEVNAME_SIZE];
5433 seq_printf(seq, "%s ",
5434 bdevname(rdev->bdev,b));
5437 seq_printf(seq, "<none>");
5439 seq_printf(seq, "\n");
5443 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5445 sector_t max_blocks, resync, res;
5446 unsigned long dt, db, rt;
5448 unsigned int per_milli;
5450 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5452 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5453 max_blocks = mddev->resync_max_sectors >> 1;
5455 max_blocks = mddev->dev_sectors / 2;
5458 * Should not happen.
5464 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5465 * in a sector_t, and (max_blocks>>scale) will fit in a
5466 * u32, as those are the requirements for sector_div.
5467 * Thus 'scale' must be at least 10
5470 if (sizeof(sector_t) > sizeof(unsigned long)) {
5471 while ( max_blocks/2 > (1ULL<<(scale+32)))
5474 res = (resync>>scale)*1000;
5475 sector_div(res, (u32)((max_blocks>>scale)+1));
5479 int i, x = per_milli/50, y = 20-x;
5480 seq_printf(seq, "[");
5481 for (i = 0; i < x; i++)
5482 seq_printf(seq, "=");
5483 seq_printf(seq, ">");
5484 for (i = 0; i < y; i++)
5485 seq_printf(seq, ".");
5486 seq_printf(seq, "] ");
5488 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5489 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5491 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5493 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5494 "resync" : "recovery"))),
5495 per_milli/10, per_milli % 10,
5496 (unsigned long long) resync,
5497 (unsigned long long) max_blocks);
5500 * We do not want to overflow, so the order of operands and
5501 * the * 100 / 100 trick are important. We do a +1 to be
5502 * safe against division by zero. We only estimate anyway.
5504 * dt: time from mark until now
5505 * db: blocks written from mark until now
5506 * rt: remaining time
5508 dt = ((jiffies - mddev->resync_mark) / HZ);
5510 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5511 - mddev->resync_mark_cnt;
5512 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5514 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5516 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5519 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5521 struct list_head *tmp;
5531 spin_lock(&all_mddevs_lock);
5532 list_for_each(tmp,&all_mddevs)
5534 mddev = list_entry(tmp, mddev_t, all_mddevs);
5536 spin_unlock(&all_mddevs_lock);
5539 spin_unlock(&all_mddevs_lock);
5541 return (void*)2;/* tail */
5545 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5547 struct list_head *tmp;
5548 mddev_t *next_mddev, *mddev = v;
5554 spin_lock(&all_mddevs_lock);
5556 tmp = all_mddevs.next;
5558 tmp = mddev->all_mddevs.next;
5559 if (tmp != &all_mddevs)
5560 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5562 next_mddev = (void*)2;
5565 spin_unlock(&all_mddevs_lock);
5573 static void md_seq_stop(struct seq_file *seq, void *v)
5577 if (mddev && v != (void*)1 && v != (void*)2)
5581 struct mdstat_info {
5585 static int md_seq_show(struct seq_file *seq, void *v)
5590 struct mdstat_info *mi = seq->private;
5591 struct bitmap *bitmap;
5593 if (v == (void*)1) {
5594 struct mdk_personality *pers;
5595 seq_printf(seq, "Personalities : ");
5596 spin_lock(&pers_lock);
5597 list_for_each_entry(pers, &pers_list, list)
5598 seq_printf(seq, "[%s] ", pers->name);
5600 spin_unlock(&pers_lock);
5601 seq_printf(seq, "\n");
5602 mi->event = atomic_read(&md_event_count);
5605 if (v == (void*)2) {
5610 if (mddev_lock(mddev) < 0)
5613 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5614 seq_printf(seq, "%s : %sactive", mdname(mddev),
5615 mddev->pers ? "" : "in");
5618 seq_printf(seq, " (read-only)");
5620 seq_printf(seq, " (auto-read-only)");
5621 seq_printf(seq, " %s", mddev->pers->name);
5625 list_for_each_entry(rdev, &mddev->disks, same_set) {
5626 char b[BDEVNAME_SIZE];
5627 seq_printf(seq, " %s[%d]",
5628 bdevname(rdev->bdev,b), rdev->desc_nr);
5629 if (test_bit(WriteMostly, &rdev->flags))
5630 seq_printf(seq, "(W)");
5631 if (test_bit(Faulty, &rdev->flags)) {
5632 seq_printf(seq, "(F)");
5634 } else if (rdev->raid_disk < 0)
5635 seq_printf(seq, "(S)"); /* spare */
5636 sectors += rdev->sectors;
5639 if (!list_empty(&mddev->disks)) {
5641 seq_printf(seq, "\n %llu blocks",
5642 (unsigned long long)
5643 mddev->array_sectors / 2);
5645 seq_printf(seq, "\n %llu blocks",
5646 (unsigned long long)sectors / 2);
5648 if (mddev->persistent) {
5649 if (mddev->major_version != 0 ||
5650 mddev->minor_version != 90) {
5651 seq_printf(seq," super %d.%d",
5652 mddev->major_version,
5653 mddev->minor_version);
5655 } else if (mddev->external)
5656 seq_printf(seq, " super external:%s",
5657 mddev->metadata_type);
5659 seq_printf(seq, " super non-persistent");
5662 mddev->pers->status(seq, mddev);
5663 seq_printf(seq, "\n ");
5664 if (mddev->pers->sync_request) {
5665 if (mddev->curr_resync > 2) {
5666 status_resync(seq, mddev);
5667 seq_printf(seq, "\n ");
5668 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5669 seq_printf(seq, "\tresync=DELAYED\n ");
5670 else if (mddev->recovery_cp < MaxSector)
5671 seq_printf(seq, "\tresync=PENDING\n ");
5674 seq_printf(seq, "\n ");
5676 if ((bitmap = mddev->bitmap)) {
5677 unsigned long chunk_kb;
5678 unsigned long flags;
5679 spin_lock_irqsave(&bitmap->lock, flags);
5680 chunk_kb = bitmap->chunksize >> 10;
5681 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5683 bitmap->pages - bitmap->missing_pages,
5685 (bitmap->pages - bitmap->missing_pages)
5686 << (PAGE_SHIFT - 10),
5687 chunk_kb ? chunk_kb : bitmap->chunksize,
5688 chunk_kb ? "KB" : "B");
5690 seq_printf(seq, ", file: ");
5691 seq_path(seq, &bitmap->file->f_path, " \t\n");
5694 seq_printf(seq, "\n");
5695 spin_unlock_irqrestore(&bitmap->lock, flags);
5698 seq_printf(seq, "\n");
5700 mddev_unlock(mddev);
5705 static struct seq_operations md_seq_ops = {
5706 .start = md_seq_start,
5707 .next = md_seq_next,
5708 .stop = md_seq_stop,
5709 .show = md_seq_show,
5712 static int md_seq_open(struct inode *inode, struct file *file)
5715 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5719 error = seq_open(file, &md_seq_ops);
5723 struct seq_file *p = file->private_data;
5725 mi->event = atomic_read(&md_event_count);
5730 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5732 struct seq_file *m = filp->private_data;
5733 struct mdstat_info *mi = m->private;
5736 poll_wait(filp, &md_event_waiters, wait);
5738 /* always allow read */
5739 mask = POLLIN | POLLRDNORM;
5741 if (mi->event != atomic_read(&md_event_count))
5742 mask |= POLLERR | POLLPRI;
5746 static const struct file_operations md_seq_fops = {
5747 .owner = THIS_MODULE,
5748 .open = md_seq_open,
5750 .llseek = seq_lseek,
5751 .release = seq_release_private,
5752 .poll = mdstat_poll,
5755 int register_md_personality(struct mdk_personality *p)
5757 spin_lock(&pers_lock);
5758 list_add_tail(&p->list, &pers_list);
5759 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5760 spin_unlock(&pers_lock);
5764 int unregister_md_personality(struct mdk_personality *p)
5766 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5767 spin_lock(&pers_lock);
5768 list_del_init(&p->list);
5769 spin_unlock(&pers_lock);
5773 static int is_mddev_idle(mddev_t *mddev, int init)
5781 rdev_for_each_rcu(rdev, mddev) {
5782 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5783 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
5784 (int)part_stat_read(&disk->part0, sectors[1]) -
5785 atomic_read(&disk->sync_io);
5786 /* sync IO will cause sync_io to increase before the disk_stats
5787 * as sync_io is counted when a request starts, and
5788 * disk_stats is counted when it completes.
5789 * So resync activity will cause curr_events to be smaller than
5790 * when there was no such activity.
5791 * non-sync IO will cause disk_stat to increase without
5792 * increasing sync_io so curr_events will (eventually)
5793 * be larger than it was before. Once it becomes
5794 * substantially larger, the test below will cause
5795 * the array to appear non-idle, and resync will slow
5797 * If there is a lot of outstanding resync activity when
5798 * we set last_event to curr_events, then all that activity
5799 * completing might cause the array to appear non-idle
5800 * and resync will be slowed down even though there might
5801 * not have been non-resync activity. This will only
5802 * happen once though. 'last_events' will soon reflect
5803 * the state where there is little or no outstanding
5804 * resync requests, and further resync activity will
5805 * always make curr_events less than last_events.
5808 if (init || curr_events - rdev->last_events > 64) {
5809 rdev->last_events = curr_events;
5817 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5819 /* another "blocks" (512byte) blocks have been synced */
5820 atomic_sub(blocks, &mddev->recovery_active);
5821 wake_up(&mddev->recovery_wait);
5823 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5824 md_wakeup_thread(mddev->thread);
5825 // stop recovery, signal do_sync ....
5830 /* md_write_start(mddev, bi)
5831 * If we need to update some array metadata (e.g. 'active' flag
5832 * in superblock) before writing, schedule a superblock update
5833 * and wait for it to complete.
5835 void md_write_start(mddev_t *mddev, struct bio *bi)
5838 if (bio_data_dir(bi) != WRITE)
5841 BUG_ON(mddev->ro == 1);
5842 if (mddev->ro == 2) {
5843 /* need to switch to read/write */
5845 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5846 md_wakeup_thread(mddev->thread);
5847 md_wakeup_thread(mddev->sync_thread);
5850 atomic_inc(&mddev->writes_pending);
5851 if (mddev->safemode == 1)
5852 mddev->safemode = 0;
5853 if (mddev->in_sync) {
5854 spin_lock_irq(&mddev->write_lock);
5855 if (mddev->in_sync) {
5857 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5858 md_wakeup_thread(mddev->thread);
5861 spin_unlock_irq(&mddev->write_lock);
5864 sysfs_notify_dirent(mddev->sysfs_state);
5865 wait_event(mddev->sb_wait,
5866 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5867 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5870 void md_write_end(mddev_t *mddev)
5872 if (atomic_dec_and_test(&mddev->writes_pending)) {
5873 if (mddev->safemode == 2)
5874 md_wakeup_thread(mddev->thread);
5875 else if (mddev->safemode_delay)
5876 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5880 /* md_allow_write(mddev)
5881 * Calling this ensures that the array is marked 'active' so that writes
5882 * may proceed without blocking. It is important to call this before
5883 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5884 * Must be called with mddev_lock held.
5886 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5887 * is dropped, so return -EAGAIN after notifying userspace.
5889 int md_allow_write(mddev_t *mddev)
5895 if (!mddev->pers->sync_request)
5898 spin_lock_irq(&mddev->write_lock);
5899 if (mddev->in_sync) {
5901 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5902 if (mddev->safemode_delay &&
5903 mddev->safemode == 0)
5904 mddev->safemode = 1;
5905 spin_unlock_irq(&mddev->write_lock);
5906 md_update_sb(mddev, 0);
5907 sysfs_notify_dirent(mddev->sysfs_state);
5909 spin_unlock_irq(&mddev->write_lock);
5911 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5916 EXPORT_SYMBOL_GPL(md_allow_write);
5918 #define SYNC_MARKS 10
5919 #define SYNC_MARK_STEP (3*HZ)
5920 void md_do_sync(mddev_t *mddev)
5923 unsigned int currspeed = 0,
5925 sector_t max_sectors,j, io_sectors;
5926 unsigned long mark[SYNC_MARKS];
5927 sector_t mark_cnt[SYNC_MARKS];
5929 struct list_head *tmp;
5930 sector_t last_check;
5935 /* just incase thread restarts... */
5936 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5938 if (mddev->ro) /* never try to sync a read-only array */
5941 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5942 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5943 desc = "data-check";
5944 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5945 desc = "requested-resync";
5948 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5953 /* we overload curr_resync somewhat here.
5954 * 0 == not engaged in resync at all
5955 * 2 == checking that there is no conflict with another sync
5956 * 1 == like 2, but have yielded to allow conflicting resync to
5958 * other == active in resync - this many blocks
5960 * Before starting a resync we must have set curr_resync to
5961 * 2, and then checked that every "conflicting" array has curr_resync
5962 * less than ours. When we find one that is the same or higher
5963 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5964 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5965 * This will mean we have to start checking from the beginning again.
5970 mddev->curr_resync = 2;
5973 if (kthread_should_stop()) {
5974 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5977 for_each_mddev(mddev2, tmp) {
5978 if (mddev2 == mddev)
5980 if (!mddev->parallel_resync
5981 && mddev2->curr_resync
5982 && match_mddev_units(mddev, mddev2)) {
5984 if (mddev < mddev2 && mddev->curr_resync == 2) {
5985 /* arbitrarily yield */
5986 mddev->curr_resync = 1;
5987 wake_up(&resync_wait);
5989 if (mddev > mddev2 && mddev->curr_resync == 1)
5990 /* no need to wait here, we can wait the next
5991 * time 'round when curr_resync == 2
5994 /* We need to wait 'interruptible' so as not to
5995 * contribute to the load average, and not to
5996 * be caught by 'softlockup'
5998 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5999 if (!kthread_should_stop() &&
6000 mddev2->curr_resync >= mddev->curr_resync) {
6001 printk(KERN_INFO "md: delaying %s of %s"
6002 " until %s has finished (they"
6003 " share one or more physical units)\n",
6004 desc, mdname(mddev), mdname(mddev2));
6006 if (signal_pending(current))
6007 flush_signals(current);
6009 finish_wait(&resync_wait, &wq);
6012 finish_wait(&resync_wait, &wq);
6015 } while (mddev->curr_resync < 2);
6018 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6019 /* resync follows the size requested by the personality,
6020 * which defaults to physical size, but can be virtual size
6022 max_sectors = mddev->resync_max_sectors;
6023 mddev->resync_mismatches = 0;
6024 /* we don't use the checkpoint if there's a bitmap */
6025 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6026 j = mddev->resync_min;
6027 else if (!mddev->bitmap)
6028 j = mddev->recovery_cp;
6030 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6031 max_sectors = mddev->dev_sectors;
6033 /* recovery follows the physical size of devices */
6034 max_sectors = mddev->dev_sectors;
6036 list_for_each_entry(rdev, &mddev->disks, same_set)
6037 if (rdev->raid_disk >= 0 &&
6038 !test_bit(Faulty, &rdev->flags) &&
6039 !test_bit(In_sync, &rdev->flags) &&
6040 rdev->recovery_offset < j)
6041 j = rdev->recovery_offset;
6044 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6045 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6046 " %d KB/sec/disk.\n", speed_min(mddev));
6047 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6048 "(but not more than %d KB/sec) for %s.\n",
6049 speed_max(mddev), desc);
6051 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6054 for (m = 0; m < SYNC_MARKS; m++) {
6056 mark_cnt[m] = io_sectors;
6059 mddev->resync_mark = mark[last_mark];
6060 mddev->resync_mark_cnt = mark_cnt[last_mark];
6063 * Tune reconstruction:
6065 window = 32*(PAGE_SIZE/512);
6066 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6067 window/2,(unsigned long long) max_sectors/2);
6069 atomic_set(&mddev->recovery_active, 0);
6074 "md: resuming %s of %s from checkpoint.\n",
6075 desc, mdname(mddev));
6076 mddev->curr_resync = j;
6079 while (j < max_sectors) {
6083 if (j >= mddev->resync_max) {
6084 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6085 wait_event(mddev->recovery_wait,
6086 mddev->resync_max > j
6087 || kthread_should_stop());
6089 if (kthread_should_stop())
6092 if (mddev->curr_resync > mddev->curr_resync_completed &&
6093 (mddev->curr_resync - mddev->curr_resync_completed)
6094 > (max_sectors >> 4)) {
6095 /* time to update curr_resync_completed */
6096 blk_unplug(mddev->queue);
6097 wait_event(mddev->recovery_wait,
6098 atomic_read(&mddev->recovery_active) == 0);
6099 mddev->curr_resync_completed =
6101 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6103 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6104 currspeed < speed_min(mddev));
6106 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6110 if (!skipped) { /* actual IO requested */
6111 io_sectors += sectors;
6112 atomic_add(sectors, &mddev->recovery_active);
6116 if (j>1) mddev->curr_resync = j;
6117 mddev->curr_mark_cnt = io_sectors;
6118 if (last_check == 0)
6119 /* this is the earliers that rebuilt will be
6120 * visible in /proc/mdstat
6122 md_new_event(mddev);
6124 if (last_check + window > io_sectors || j == max_sectors)
6127 last_check = io_sectors;
6129 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6133 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6135 int next = (last_mark+1) % SYNC_MARKS;
6137 mddev->resync_mark = mark[next];
6138 mddev->resync_mark_cnt = mark_cnt[next];
6139 mark[next] = jiffies;
6140 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6145 if (kthread_should_stop())
6150 * this loop exits only if either when we are slower than
6151 * the 'hard' speed limit, or the system was IO-idle for
6153 * the system might be non-idle CPU-wise, but we only care
6154 * about not overloading the IO subsystem. (things like an
6155 * e2fsck being done on the RAID array should execute fast)
6157 blk_unplug(mddev->queue);
6160 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6161 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6163 if (currspeed > speed_min(mddev)) {
6164 if ((currspeed > speed_max(mddev)) ||
6165 !is_mddev_idle(mddev, 0)) {
6171 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6173 * this also signals 'finished resyncing' to md_stop
6176 blk_unplug(mddev->queue);
6178 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6180 /* tell personality that we are finished */
6181 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6183 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6184 mddev->curr_resync > 2) {
6185 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6186 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6187 if (mddev->curr_resync >= mddev->recovery_cp) {
6189 "md: checkpointing %s of %s.\n",
6190 desc, mdname(mddev));
6191 mddev->recovery_cp = mddev->curr_resync;
6194 mddev->recovery_cp = MaxSector;
6196 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6197 mddev->curr_resync = MaxSector;
6198 list_for_each_entry(rdev, &mddev->disks, same_set)
6199 if (rdev->raid_disk >= 0 &&
6200 !test_bit(Faulty, &rdev->flags) &&
6201 !test_bit(In_sync, &rdev->flags) &&
6202 rdev->recovery_offset < mddev->curr_resync)
6203 rdev->recovery_offset = mddev->curr_resync;
6206 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6209 mddev->curr_resync = 0;
6210 mddev->resync_min = 0;
6211 mddev->resync_max = MaxSector;
6212 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6213 wake_up(&resync_wait);
6214 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6215 md_wakeup_thread(mddev->thread);
6220 * got a signal, exit.
6223 "md: md_do_sync() got signal ... exiting\n");
6224 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6228 EXPORT_SYMBOL_GPL(md_do_sync);
6231 static int remove_and_add_spares(mddev_t *mddev)
6236 mddev->curr_resync_completed = 0;
6238 list_for_each_entry(rdev, &mddev->disks, same_set)
6239 if (rdev->raid_disk >= 0 &&
6240 !test_bit(Blocked, &rdev->flags) &&
6241 (test_bit(Faulty, &rdev->flags) ||
6242 ! test_bit(In_sync, &rdev->flags)) &&
6243 atomic_read(&rdev->nr_pending)==0) {
6244 if (mddev->pers->hot_remove_disk(
6245 mddev, rdev->raid_disk)==0) {
6247 sprintf(nm,"rd%d", rdev->raid_disk);
6248 sysfs_remove_link(&mddev->kobj, nm);
6249 rdev->raid_disk = -1;
6253 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6254 list_for_each_entry(rdev, &mddev->disks, same_set) {
6255 if (rdev->raid_disk >= 0 &&
6256 !test_bit(In_sync, &rdev->flags) &&
6257 !test_bit(Blocked, &rdev->flags))
6259 if (rdev->raid_disk < 0
6260 && !test_bit(Faulty, &rdev->flags)) {
6261 rdev->recovery_offset = 0;
6263 hot_add_disk(mddev, rdev) == 0) {
6265 sprintf(nm, "rd%d", rdev->raid_disk);
6266 if (sysfs_create_link(&mddev->kobj,
6269 "md: cannot register "
6273 md_new_event(mddev);
6282 * This routine is regularly called by all per-raid-array threads to
6283 * deal with generic issues like resync and super-block update.
6284 * Raid personalities that don't have a thread (linear/raid0) do not
6285 * need this as they never do any recovery or update the superblock.
6287 * It does not do any resync itself, but rather "forks" off other threads
6288 * to do that as needed.
6289 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6290 * "->recovery" and create a thread at ->sync_thread.
6291 * When the thread finishes it sets MD_RECOVERY_DONE
6292 * and wakeups up this thread which will reap the thread and finish up.
6293 * This thread also removes any faulty devices (with nr_pending == 0).
6295 * The overall approach is:
6296 * 1/ if the superblock needs updating, update it.
6297 * 2/ If a recovery thread is running, don't do anything else.
6298 * 3/ If recovery has finished, clean up, possibly marking spares active.
6299 * 4/ If there are any faulty devices, remove them.
6300 * 5/ If array is degraded, try to add spares devices
6301 * 6/ If array has spares or is not in-sync, start a resync thread.
6303 void md_check_recovery(mddev_t *mddev)
6309 bitmap_daemon_work(mddev->bitmap);
6314 if (signal_pending(current)) {
6315 if (mddev->pers->sync_request && !mddev->external) {
6316 printk(KERN_INFO "md: %s in immediate safe mode\n",
6318 mddev->safemode = 2;
6320 flush_signals(current);
6323 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6326 (mddev->flags && !mddev->external) ||
6327 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6328 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6329 (mddev->external == 0 && mddev->safemode == 1) ||
6330 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6331 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6335 if (mddev_trylock(mddev)) {
6339 /* Only thing we do on a ro array is remove
6342 remove_and_add_spares(mddev);
6343 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6347 if (!mddev->external) {
6349 spin_lock_irq(&mddev->write_lock);
6350 if (mddev->safemode &&
6351 !atomic_read(&mddev->writes_pending) &&
6353 mddev->recovery_cp == MaxSector) {
6356 if (mddev->persistent)
6357 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6359 if (mddev->safemode == 1)
6360 mddev->safemode = 0;
6361 spin_unlock_irq(&mddev->write_lock);
6363 sysfs_notify_dirent(mddev->sysfs_state);
6367 md_update_sb(mddev, 0);
6369 list_for_each_entry(rdev, &mddev->disks, same_set)
6370 if (test_and_clear_bit(StateChanged, &rdev->flags))
6371 sysfs_notify_dirent(rdev->sysfs_state);
6374 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6375 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6376 /* resync/recovery still happening */
6377 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6380 if (mddev->sync_thread) {
6381 /* resync has finished, collect result */
6382 md_unregister_thread(mddev->sync_thread);
6383 mddev->sync_thread = NULL;
6384 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6385 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6387 /* activate any spares */
6388 if (mddev->pers->spare_active(mddev))
6389 sysfs_notify(&mddev->kobj, NULL,
6392 md_update_sb(mddev, 1);
6394 /* if array is no-longer degraded, then any saved_raid_disk
6395 * information must be scrapped
6397 if (!mddev->degraded)
6398 list_for_each_entry(rdev, &mddev->disks, same_set)
6399 rdev->saved_raid_disk = -1;
6401 mddev->recovery = 0;
6402 /* flag recovery needed just to double check */
6403 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6404 sysfs_notify_dirent(mddev->sysfs_action);
6405 md_new_event(mddev);
6408 /* Set RUNNING before clearing NEEDED to avoid
6409 * any transients in the value of "sync_action".
6411 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6412 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6413 /* Clear some bits that don't mean anything, but
6416 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6417 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6419 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6421 /* no recovery is running.
6422 * remove any failed drives, then
6423 * add spares if possible.
6424 * Spare are also removed and re-added, to allow
6425 * the personality to fail the re-add.
6428 if (mddev->reshape_position != MaxSector) {
6429 if (mddev->pers->check_reshape(mddev) != 0)
6430 /* Cannot proceed */
6432 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6433 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6434 } else if ((spares = remove_and_add_spares(mddev))) {
6435 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6436 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6437 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6438 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6439 } else if (mddev->recovery_cp < MaxSector) {
6440 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6441 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6442 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6443 /* nothing to be done ... */
6446 if (mddev->pers->sync_request) {
6447 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6448 /* We are adding a device or devices to an array
6449 * which has the bitmap stored on all devices.
6450 * So make sure all bitmap pages get written
6452 bitmap_write_all(mddev->bitmap);
6454 mddev->sync_thread = md_register_thread(md_do_sync,
6457 if (!mddev->sync_thread) {
6458 printk(KERN_ERR "%s: could not start resync"
6461 /* leave the spares where they are, it shouldn't hurt */
6462 mddev->recovery = 0;
6464 md_wakeup_thread(mddev->sync_thread);
6465 sysfs_notify_dirent(mddev->sysfs_action);
6466 md_new_event(mddev);
6469 if (!mddev->sync_thread) {
6470 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6471 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6473 if (mddev->sysfs_action)
6474 sysfs_notify_dirent(mddev->sysfs_action);
6476 mddev_unlock(mddev);
6480 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6482 sysfs_notify_dirent(rdev->sysfs_state);
6483 wait_event_timeout(rdev->blocked_wait,
6484 !test_bit(Blocked, &rdev->flags),
6485 msecs_to_jiffies(5000));
6486 rdev_dec_pending(rdev, mddev);
6488 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6490 static int md_notify_reboot(struct notifier_block *this,
6491 unsigned long code, void *x)
6493 struct list_head *tmp;
6496 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6498 printk(KERN_INFO "md: stopping all md devices.\n");
6500 for_each_mddev(mddev, tmp)
6501 if (mddev_trylock(mddev)) {
6502 /* Force a switch to readonly even array
6503 * appears to still be in use. Hence
6506 do_md_stop(mddev, 1, 100);
6507 mddev_unlock(mddev);
6510 * certain more exotic SCSI devices are known to be
6511 * volatile wrt too early system reboots. While the
6512 * right place to handle this issue is the given
6513 * driver, we do want to have a safe RAID driver ...
6520 static struct notifier_block md_notifier = {
6521 .notifier_call = md_notify_reboot,
6523 .priority = INT_MAX, /* before any real devices */
6526 static void md_geninit(void)
6528 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6530 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6533 static int __init md_init(void)
6535 if (register_blkdev(MD_MAJOR, "md"))
6537 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6538 unregister_blkdev(MD_MAJOR, "md");
6541 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6542 md_probe, NULL, NULL);
6543 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6544 md_probe, NULL, NULL);
6546 register_reboot_notifier(&md_notifier);
6547 raid_table_header = register_sysctl_table(raid_root_table);
6557 * Searches all registered partitions for autorun RAID arrays
6561 static LIST_HEAD(all_detected_devices);
6562 struct detected_devices_node {
6563 struct list_head list;
6567 void md_autodetect_dev(dev_t dev)
6569 struct detected_devices_node *node_detected_dev;
6571 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6572 if (node_detected_dev) {
6573 node_detected_dev->dev = dev;
6574 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6576 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6577 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6582 static void autostart_arrays(int part)
6585 struct detected_devices_node *node_detected_dev;
6587 int i_scanned, i_passed;
6592 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6594 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6596 node_detected_dev = list_entry(all_detected_devices.next,
6597 struct detected_devices_node, list);
6598 list_del(&node_detected_dev->list);
6599 dev = node_detected_dev->dev;
6600 kfree(node_detected_dev);
6601 rdev = md_import_device(dev,0, 90);
6605 if (test_bit(Faulty, &rdev->flags)) {
6609 set_bit(AutoDetected, &rdev->flags);
6610 list_add(&rdev->same_set, &pending_raid_disks);
6614 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6615 i_scanned, i_passed);
6617 autorun_devices(part);
6620 #endif /* !MODULE */
6622 static __exit void md_exit(void)
6625 struct list_head *tmp;
6627 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6628 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6630 unregister_blkdev(MD_MAJOR,"md");
6631 unregister_blkdev(mdp_major, "mdp");
6632 unregister_reboot_notifier(&md_notifier);
6633 unregister_sysctl_table(raid_table_header);
6634 remove_proc_entry("mdstat", NULL);
6635 for_each_mddev(mddev, tmp) {
6636 export_array(mddev);
6637 mddev->hold_active = 0;
6641 subsys_initcall(md_init);
6642 module_exit(md_exit)
6644 static int get_ro(char *buffer, struct kernel_param *kp)
6646 return sprintf(buffer, "%d", start_readonly);
6648 static int set_ro(const char *val, struct kernel_param *kp)
6651 int num = simple_strtoul(val, &e, 10);
6652 if (*val && (*e == '\0' || *e == '\n')) {
6653 start_readonly = num;
6659 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6660 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6662 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6664 EXPORT_SYMBOL(register_md_personality);
6665 EXPORT_SYMBOL(unregister_md_personality);
6666 EXPORT_SYMBOL(md_error);
6667 EXPORT_SYMBOL(md_done_sync);
6668 EXPORT_SYMBOL(md_write_start);
6669 EXPORT_SYMBOL(md_write_end);
6670 EXPORT_SYMBOL(md_register_thread);
6671 EXPORT_SYMBOL(md_unregister_thread);
6672 EXPORT_SYMBOL(md_wakeup_thread);
6673 EXPORT_SYMBOL(md_check_recovery);
6674 MODULE_LICENSE("GPL");
6676 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);