2 md.c : Multiple Devices driver for Linux
3 Copyright (C) 1998, 1999, 2000 Ingo Molnar
5 completely rewritten, based on the MD driver code from Marc Zyngier
9 - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10 - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11 - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12 - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13 - kmod support by: Cyrus Durgin
14 - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15 - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
17 - lots of fixes and improvements to the RAID1/RAID5 and generic
18 RAID code (such as request based resynchronization):
20 Neil Brown <neilb@cse.unsw.edu.au>.
22 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
25 This program is free software; you can redistribute it and/or modify
26 it under the terms of the GNU General Public License as published by
27 the Free Software Foundation; either version 2, or (at your option)
30 You should have received a copy of the GNU General Public License
31 (for example /usr/src/linux/COPYING); if not, write to the Free
32 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
58 static void autostart_arrays(int part);
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
64 static void md_print_devices(void);
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
71 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72 * is 1000 KB/sec, so the extra system load does not show up that much.
73 * Increase it if you want to have more _guaranteed_ speed. Note that
74 * the RAID driver will use the maximum available bandwidth if the IO
75 * subsystem is idle. There is also an 'absolute maximum' reconstruction
76 * speed limit - in case reconstruction slows down your system despite
79 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80 * or /sys/block/mdX/md/sync_speed_{min,max}
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
87 return mddev->sync_speed_min ?
88 mddev->sync_speed_min : sysctl_speed_limit_min;
91 static inline int speed_max(mddev_t *mddev)
93 return mddev->sync_speed_max ?
94 mddev->sync_speed_max : sysctl_speed_limit_max;
97 static struct ctl_table_header *raid_table_header;
99 static ctl_table raid_table[] = {
101 .ctl_name = DEV_RAID_SPEED_LIMIT_MIN,
102 .procname = "speed_limit_min",
103 .data = &sysctl_speed_limit_min,
104 .maxlen = sizeof(int),
105 .mode = S_IRUGO|S_IWUSR,
106 .proc_handler = &proc_dointvec,
109 .ctl_name = DEV_RAID_SPEED_LIMIT_MAX,
110 .procname = "speed_limit_max",
111 .data = &sysctl_speed_limit_max,
112 .maxlen = sizeof(int),
113 .mode = S_IRUGO|S_IWUSR,
114 .proc_handler = &proc_dointvec,
119 static ctl_table raid_dir_table[] = {
121 .ctl_name = DEV_RAID,
124 .mode = S_IRUGO|S_IXUGO,
130 static ctl_table raid_root_table[] = {
136 .child = raid_dir_table,
141 static struct block_device_operations md_fops;
143 static int start_readonly;
146 * We have a system wide 'event count' that is incremented
147 * on any 'interesting' event, and readers of /proc/mdstat
148 * can use 'poll' or 'select' to find out when the event
152 * start array, stop array, error, add device, remove device,
153 * start build, activate spare
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
159 atomic_inc(&md_event_count);
160 wake_up(&md_event_waiters);
162 EXPORT_SYMBOL_GPL(md_new_event);
164 /* Alternate version that can be called from interrupts
165 * when calling sysfs_notify isn't needed.
167 static void md_new_event_inintr(mddev_t *mddev)
169 atomic_inc(&md_event_count);
170 wake_up(&md_event_waiters);
174 * Enables to iterate over all existing md arrays
175 * all_mddevs_lock protects this list.
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
182 * iterates through all used mddevs in the system.
183 * We take care to grab the all_mddevs_lock whenever navigating
184 * the list, and to always hold a refcount when unlocked.
185 * Any code which breaks out of this loop while own
186 * a reference to the current mddev and must mddev_put it.
188 #define for_each_mddev(mddev,tmp) \
190 for (({ spin_lock(&all_mddevs_lock); \
191 tmp = all_mddevs.next; \
193 ({ if (tmp != &all_mddevs) \
194 mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195 spin_unlock(&all_mddevs_lock); \
196 if (mddev) mddev_put(mddev); \
197 mddev = list_entry(tmp, mddev_t, all_mddevs); \
198 tmp != &all_mddevs;}); \
199 ({ spin_lock(&all_mddevs_lock); \
204 static int md_fail_request(struct request_queue *q, struct bio *bio)
210 static inline mddev_t *mddev_get(mddev_t *mddev)
212 atomic_inc(&mddev->active);
216 static void mddev_delayed_delete(struct work_struct *ws)
218 mddev_t *mddev = container_of(ws, mddev_t, del_work);
219 kobject_del(&mddev->kobj);
220 kobject_put(&mddev->kobj);
223 static void mddev_put(mddev_t *mddev)
225 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
227 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
228 !mddev->hold_active) {
229 list_del(&mddev->all_mddevs);
230 if (mddev->gendisk) {
231 /* we did a probe so need to clean up.
232 * Call schedule_work inside the spinlock
233 * so that flush_scheduled_work() after
234 * mddev_find will succeed in waiting for the
237 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
238 schedule_work(&mddev->del_work);
242 spin_unlock(&all_mddevs_lock);
245 static mddev_t * mddev_find(dev_t unit)
247 mddev_t *mddev, *new = NULL;
250 spin_lock(&all_mddevs_lock);
253 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
254 if (mddev->unit == unit) {
256 spin_unlock(&all_mddevs_lock);
262 list_add(&new->all_mddevs, &all_mddevs);
263 spin_unlock(&all_mddevs_lock);
264 new->hold_active = UNTIL_IOCTL;
268 /* find an unused unit number */
269 static int next_minor = 512;
270 int start = next_minor;
274 dev = MKDEV(MD_MAJOR, next_minor);
276 if (next_minor > MINORMASK)
278 if (next_minor == start) {
279 /* Oh dear, all in use. */
280 spin_unlock(&all_mddevs_lock);
286 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
287 if (mddev->unit == dev) {
293 new->md_minor = MINOR(dev);
294 new->hold_active = UNTIL_STOP;
295 list_add(&new->all_mddevs, &all_mddevs);
296 spin_unlock(&all_mddevs_lock);
299 spin_unlock(&all_mddevs_lock);
301 new = kzalloc(sizeof(*new), GFP_KERNEL);
306 if (MAJOR(unit) == MD_MAJOR)
307 new->md_minor = MINOR(unit);
309 new->md_minor = MINOR(unit) >> MdpMinorShift;
311 mutex_init(&new->reconfig_mutex);
312 INIT_LIST_HEAD(&new->disks);
313 INIT_LIST_HEAD(&new->all_mddevs);
314 init_timer(&new->safemode_timer);
315 atomic_set(&new->active, 1);
316 atomic_set(&new->openers, 0);
317 spin_lock_init(&new->write_lock);
318 init_waitqueue_head(&new->sb_wait);
319 init_waitqueue_head(&new->recovery_wait);
320 new->reshape_position = MaxSector;
322 new->resync_max = MaxSector;
323 new->level = LEVEL_NONE;
328 static inline int mddev_lock(mddev_t * mddev)
330 return mutex_lock_interruptible(&mddev->reconfig_mutex);
333 static inline int mddev_trylock(mddev_t * mddev)
335 return mutex_trylock(&mddev->reconfig_mutex);
338 static inline void mddev_unlock(mddev_t * mddev)
340 mutex_unlock(&mddev->reconfig_mutex);
342 md_wakeup_thread(mddev->thread);
345 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
349 list_for_each_entry(rdev, &mddev->disks, same_set)
350 if (rdev->desc_nr == nr)
356 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
360 list_for_each_entry(rdev, &mddev->disks, same_set)
361 if (rdev->bdev->bd_dev == dev)
367 static struct mdk_personality *find_pers(int level, char *clevel)
369 struct mdk_personality *pers;
370 list_for_each_entry(pers, &pers_list, list) {
371 if (level != LEVEL_NONE && pers->level == level)
373 if (strcmp(pers->name, clevel)==0)
379 /* return the offset of the super block in 512byte sectors */
380 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
382 sector_t num_sectors = bdev->bd_inode->i_size / 512;
383 return MD_NEW_SIZE_SECTORS(num_sectors);
386 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
388 sector_t num_sectors = rdev->sb_start;
391 num_sectors &= ~((sector_t)chunk_size/512 - 1);
395 static int alloc_disk_sb(mdk_rdev_t * rdev)
400 rdev->sb_page = alloc_page(GFP_KERNEL);
401 if (!rdev->sb_page) {
402 printk(KERN_ALERT "md: out of memory.\n");
409 static void free_disk_sb(mdk_rdev_t * rdev)
412 put_page(rdev->sb_page);
414 rdev->sb_page = NULL;
421 static void super_written(struct bio *bio, int error)
423 mdk_rdev_t *rdev = bio->bi_private;
424 mddev_t *mddev = rdev->mddev;
426 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
427 printk("md: super_written gets error=%d, uptodate=%d\n",
428 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
429 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
430 md_error(mddev, rdev);
433 if (atomic_dec_and_test(&mddev->pending_writes))
434 wake_up(&mddev->sb_wait);
438 static void super_written_barrier(struct bio *bio, int error)
440 struct bio *bio2 = bio->bi_private;
441 mdk_rdev_t *rdev = bio2->bi_private;
442 mddev_t *mddev = rdev->mddev;
444 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
445 error == -EOPNOTSUPP) {
447 /* barriers don't appear to be supported :-( */
448 set_bit(BarriersNotsupp, &rdev->flags);
449 mddev->barriers_work = 0;
450 spin_lock_irqsave(&mddev->write_lock, flags);
451 bio2->bi_next = mddev->biolist;
452 mddev->biolist = bio2;
453 spin_unlock_irqrestore(&mddev->write_lock, flags);
454 wake_up(&mddev->sb_wait);
458 bio->bi_private = rdev;
459 super_written(bio, error);
463 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
464 sector_t sector, int size, struct page *page)
466 /* write first size bytes of page to sector of rdev
467 * Increment mddev->pending_writes before returning
468 * and decrement it on completion, waking up sb_wait
469 * if zero is reached.
470 * If an error occurred, call md_error
472 * As we might need to resubmit the request if BIO_RW_BARRIER
473 * causes ENOTSUPP, we allocate a spare bio...
475 struct bio *bio = bio_alloc(GFP_NOIO, 1);
476 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
478 bio->bi_bdev = rdev->bdev;
479 bio->bi_sector = sector;
480 bio_add_page(bio, page, size, 0);
481 bio->bi_private = rdev;
482 bio->bi_end_io = super_written;
485 atomic_inc(&mddev->pending_writes);
486 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
488 rw |= (1<<BIO_RW_BARRIER);
489 rbio = bio_clone(bio, GFP_NOIO);
490 rbio->bi_private = bio;
491 rbio->bi_end_io = super_written_barrier;
492 submit_bio(rw, rbio);
497 void md_super_wait(mddev_t *mddev)
499 /* wait for all superblock writes that were scheduled to complete.
500 * if any had to be retried (due to BARRIER problems), retry them
504 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
505 if (atomic_read(&mddev->pending_writes)==0)
507 while (mddev->biolist) {
509 spin_lock_irq(&mddev->write_lock);
510 bio = mddev->biolist;
511 mddev->biolist = bio->bi_next ;
513 spin_unlock_irq(&mddev->write_lock);
514 submit_bio(bio->bi_rw, bio);
518 finish_wait(&mddev->sb_wait, &wq);
521 static void bi_complete(struct bio *bio, int error)
523 complete((struct completion*)bio->bi_private);
526 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
527 struct page *page, int rw)
529 struct bio *bio = bio_alloc(GFP_NOIO, 1);
530 struct completion event;
533 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
536 bio->bi_sector = sector;
537 bio_add_page(bio, page, size, 0);
538 init_completion(&event);
539 bio->bi_private = &event;
540 bio->bi_end_io = bi_complete;
542 wait_for_completion(&event);
544 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
548 EXPORT_SYMBOL_GPL(sync_page_io);
550 static int read_disk_sb(mdk_rdev_t * rdev, int size)
552 char b[BDEVNAME_SIZE];
553 if (!rdev->sb_page) {
561 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
567 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
568 bdevname(rdev->bdev,b));
572 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
574 return sb1->set_uuid0 == sb2->set_uuid0 &&
575 sb1->set_uuid1 == sb2->set_uuid1 &&
576 sb1->set_uuid2 == sb2->set_uuid2 &&
577 sb1->set_uuid3 == sb2->set_uuid3;
580 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
583 mdp_super_t *tmp1, *tmp2;
585 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
586 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
588 if (!tmp1 || !tmp2) {
590 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
598 * nr_disks is not constant
603 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
611 static u32 md_csum_fold(u32 csum)
613 csum = (csum & 0xffff) + (csum >> 16);
614 return (csum & 0xffff) + (csum >> 16);
617 static unsigned int calc_sb_csum(mdp_super_t * sb)
620 u32 *sb32 = (u32*)sb;
622 unsigned int disk_csum, csum;
624 disk_csum = sb->sb_csum;
627 for (i = 0; i < MD_SB_BYTES/4 ; i++)
629 csum = (newcsum & 0xffffffff) + (newcsum>>32);
633 /* This used to use csum_partial, which was wrong for several
634 * reasons including that different results are returned on
635 * different architectures. It isn't critical that we get exactly
636 * the same return value as before (we always csum_fold before
637 * testing, and that removes any differences). However as we
638 * know that csum_partial always returned a 16bit value on
639 * alphas, do a fold to maximise conformity to previous behaviour.
641 sb->sb_csum = md_csum_fold(disk_csum);
643 sb->sb_csum = disk_csum;
650 * Handle superblock details.
651 * We want to be able to handle multiple superblock formats
652 * so we have a common interface to them all, and an array of
653 * different handlers.
654 * We rely on user-space to write the initial superblock, and support
655 * reading and updating of superblocks.
656 * Interface methods are:
657 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
658 * loads and validates a superblock on dev.
659 * if refdev != NULL, compare superblocks on both devices
661 * 0 - dev has a superblock that is compatible with refdev
662 * 1 - dev has a superblock that is compatible and newer than refdev
663 * so dev should be used as the refdev in future
664 * -EINVAL superblock incompatible or invalid
665 * -othererror e.g. -EIO
667 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
668 * Verify that dev is acceptable into mddev.
669 * The first time, mddev->raid_disks will be 0, and data from
670 * dev should be merged in. Subsequent calls check that dev
671 * is new enough. Return 0 or -EINVAL
673 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
674 * Update the superblock for rdev with data in mddev
675 * This does not write to disc.
681 struct module *owner;
682 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
684 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
685 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
686 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
687 sector_t num_sectors);
691 * load_super for 0.90.0
693 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
695 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
700 * Calculate the position of the superblock (512byte sectors),
701 * it's at the end of the disk.
703 * It also happens to be a multiple of 4Kb.
705 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
707 ret = read_disk_sb(rdev, MD_SB_BYTES);
712 bdevname(rdev->bdev, b);
713 sb = (mdp_super_t*)page_address(rdev->sb_page);
715 if (sb->md_magic != MD_SB_MAGIC) {
716 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
721 if (sb->major_version != 0 ||
722 sb->minor_version < 90 ||
723 sb->minor_version > 91) {
724 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
725 sb->major_version, sb->minor_version,
730 if (sb->raid_disks <= 0)
733 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
734 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
739 rdev->preferred_minor = sb->md_minor;
740 rdev->data_offset = 0;
741 rdev->sb_size = MD_SB_BYTES;
743 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
744 if (sb->level != 1 && sb->level != 4
745 && sb->level != 5 && sb->level != 6
746 && sb->level != 10) {
747 /* FIXME use a better test */
749 "md: bitmaps not supported for this level.\n");
754 if (sb->level == LEVEL_MULTIPATH)
757 rdev->desc_nr = sb->this_disk.number;
763 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
764 if (!uuid_equal(refsb, sb)) {
765 printk(KERN_WARNING "md: %s has different UUID to %s\n",
766 b, bdevname(refdev->bdev,b2));
769 if (!sb_equal(refsb, sb)) {
770 printk(KERN_WARNING "md: %s has same UUID"
771 " but different superblock to %s\n",
772 b, bdevname(refdev->bdev, b2));
776 ev2 = md_event(refsb);
782 rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
784 if (rdev->size < sb->size && sb->level > 1)
785 /* "this cannot possibly happen" ... */
793 * validate_super for 0.90.0
795 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
798 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
799 __u64 ev1 = md_event(sb);
801 rdev->raid_disk = -1;
802 clear_bit(Faulty, &rdev->flags);
803 clear_bit(In_sync, &rdev->flags);
804 clear_bit(WriteMostly, &rdev->flags);
805 clear_bit(BarriersNotsupp, &rdev->flags);
807 if (mddev->raid_disks == 0) {
808 mddev->major_version = 0;
809 mddev->minor_version = sb->minor_version;
810 mddev->patch_version = sb->patch_version;
812 mddev->chunk_size = sb->chunk_size;
813 mddev->ctime = sb->ctime;
814 mddev->utime = sb->utime;
815 mddev->level = sb->level;
816 mddev->clevel[0] = 0;
817 mddev->layout = sb->layout;
818 mddev->raid_disks = sb->raid_disks;
819 mddev->size = sb->size;
821 mddev->bitmap_offset = 0;
822 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
824 if (mddev->minor_version >= 91) {
825 mddev->reshape_position = sb->reshape_position;
826 mddev->delta_disks = sb->delta_disks;
827 mddev->new_level = sb->new_level;
828 mddev->new_layout = sb->new_layout;
829 mddev->new_chunk = sb->new_chunk;
831 mddev->reshape_position = MaxSector;
832 mddev->delta_disks = 0;
833 mddev->new_level = mddev->level;
834 mddev->new_layout = mddev->layout;
835 mddev->new_chunk = mddev->chunk_size;
838 if (sb->state & (1<<MD_SB_CLEAN))
839 mddev->recovery_cp = MaxSector;
841 if (sb->events_hi == sb->cp_events_hi &&
842 sb->events_lo == sb->cp_events_lo) {
843 mddev->recovery_cp = sb->recovery_cp;
845 mddev->recovery_cp = 0;
848 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
849 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
850 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
851 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
853 mddev->max_disks = MD_SB_DISKS;
855 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
856 mddev->bitmap_file == NULL)
857 mddev->bitmap_offset = mddev->default_bitmap_offset;
859 } else if (mddev->pers == NULL) {
860 /* Insist on good event counter while assembling */
862 if (ev1 < mddev->events)
864 } else if (mddev->bitmap) {
865 /* if adding to array with a bitmap, then we can accept an
866 * older device ... but not too old.
868 if (ev1 < mddev->bitmap->events_cleared)
871 if (ev1 < mddev->events)
872 /* just a hot-add of a new device, leave raid_disk at -1 */
876 if (mddev->level != LEVEL_MULTIPATH) {
877 desc = sb->disks + rdev->desc_nr;
879 if (desc->state & (1<<MD_DISK_FAULTY))
880 set_bit(Faulty, &rdev->flags);
881 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
882 desc->raid_disk < mddev->raid_disks */) {
883 set_bit(In_sync, &rdev->flags);
884 rdev->raid_disk = desc->raid_disk;
886 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
887 set_bit(WriteMostly, &rdev->flags);
888 } else /* MULTIPATH are always insync */
889 set_bit(In_sync, &rdev->flags);
894 * sync_super for 0.90.0
896 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
900 int next_spare = mddev->raid_disks;
903 /* make rdev->sb match mddev data..
906 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
907 * 3/ any empty disks < next_spare become removed
909 * disks[0] gets initialised to REMOVED because
910 * we cannot be sure from other fields if it has
911 * been initialised or not.
914 int active=0, working=0,failed=0,spare=0,nr_disks=0;
916 rdev->sb_size = MD_SB_BYTES;
918 sb = (mdp_super_t*)page_address(rdev->sb_page);
920 memset(sb, 0, sizeof(*sb));
922 sb->md_magic = MD_SB_MAGIC;
923 sb->major_version = mddev->major_version;
924 sb->patch_version = mddev->patch_version;
925 sb->gvalid_words = 0; /* ignored */
926 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
927 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
928 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
929 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
931 sb->ctime = mddev->ctime;
932 sb->level = mddev->level;
933 sb->size = mddev->size;
934 sb->raid_disks = mddev->raid_disks;
935 sb->md_minor = mddev->md_minor;
936 sb->not_persistent = 0;
937 sb->utime = mddev->utime;
939 sb->events_hi = (mddev->events>>32);
940 sb->events_lo = (u32)mddev->events;
942 if (mddev->reshape_position == MaxSector)
943 sb->minor_version = 90;
945 sb->minor_version = 91;
946 sb->reshape_position = mddev->reshape_position;
947 sb->new_level = mddev->new_level;
948 sb->delta_disks = mddev->delta_disks;
949 sb->new_layout = mddev->new_layout;
950 sb->new_chunk = mddev->new_chunk;
952 mddev->minor_version = sb->minor_version;
955 sb->recovery_cp = mddev->recovery_cp;
956 sb->cp_events_hi = (mddev->events>>32);
957 sb->cp_events_lo = (u32)mddev->events;
958 if (mddev->recovery_cp == MaxSector)
959 sb->state = (1<< MD_SB_CLEAN);
963 sb->layout = mddev->layout;
964 sb->chunk_size = mddev->chunk_size;
966 if (mddev->bitmap && mddev->bitmap_file == NULL)
967 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
969 sb->disks[0].state = (1<<MD_DISK_REMOVED);
970 list_for_each_entry(rdev2, &mddev->disks, same_set) {
973 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
974 && !test_bit(Faulty, &rdev2->flags))
975 desc_nr = rdev2->raid_disk;
977 desc_nr = next_spare++;
978 rdev2->desc_nr = desc_nr;
979 d = &sb->disks[rdev2->desc_nr];
981 d->number = rdev2->desc_nr;
982 d->major = MAJOR(rdev2->bdev->bd_dev);
983 d->minor = MINOR(rdev2->bdev->bd_dev);
984 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
985 && !test_bit(Faulty, &rdev2->flags))
986 d->raid_disk = rdev2->raid_disk;
988 d->raid_disk = rdev2->desc_nr; /* compatibility */
989 if (test_bit(Faulty, &rdev2->flags))
990 d->state = (1<<MD_DISK_FAULTY);
991 else if (test_bit(In_sync, &rdev2->flags)) {
992 d->state = (1<<MD_DISK_ACTIVE);
993 d->state |= (1<<MD_DISK_SYNC);
1001 if (test_bit(WriteMostly, &rdev2->flags))
1002 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1004 /* now set the "removed" and "faulty" bits on any missing devices */
1005 for (i=0 ; i < mddev->raid_disks ; i++) {
1006 mdp_disk_t *d = &sb->disks[i];
1007 if (d->state == 0 && d->number == 0) {
1010 d->state = (1<<MD_DISK_REMOVED);
1011 d->state |= (1<<MD_DISK_FAULTY);
1015 sb->nr_disks = nr_disks;
1016 sb->active_disks = active;
1017 sb->working_disks = working;
1018 sb->failed_disks = failed;
1019 sb->spare_disks = spare;
1021 sb->this_disk = sb->disks[rdev->desc_nr];
1022 sb->sb_csum = calc_sb_csum(sb);
1026 * rdev_size_change for 0.90.0
1028 static unsigned long long
1029 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1031 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1032 return 0; /* component must fit device */
1033 if (rdev->mddev->bitmap_offset)
1034 return 0; /* can't move bitmap */
1035 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1036 if (!num_sectors || num_sectors > rdev->sb_start)
1037 num_sectors = rdev->sb_start;
1038 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1040 md_super_wait(rdev->mddev);
1041 return num_sectors / 2; /* kB for sysfs */
1046 * version 1 superblock
1049 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1053 unsigned long long newcsum;
1054 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1055 __le32 *isuper = (__le32*)sb;
1058 disk_csum = sb->sb_csum;
1061 for (i=0; size>=4; size -= 4 )
1062 newcsum += le32_to_cpu(*isuper++);
1065 newcsum += le16_to_cpu(*(__le16*) isuper);
1067 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1068 sb->sb_csum = disk_csum;
1069 return cpu_to_le32(csum);
1072 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1074 struct mdp_superblock_1 *sb;
1077 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1081 * Calculate the position of the superblock in 512byte sectors.
1082 * It is always aligned to a 4K boundary and
1083 * depeding on minor_version, it can be:
1084 * 0: At least 8K, but less than 12K, from end of device
1085 * 1: At start of device
1086 * 2: 4K from start of device.
1088 switch(minor_version) {
1090 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1092 sb_start &= ~(sector_t)(4*2-1);
1103 rdev->sb_start = sb_start;
1105 /* superblock is rarely larger than 1K, but it can be larger,
1106 * and it is safe to read 4k, so we do that
1108 ret = read_disk_sb(rdev, 4096);
1109 if (ret) return ret;
1112 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1114 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1115 sb->major_version != cpu_to_le32(1) ||
1116 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1117 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1118 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1121 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1122 printk("md: invalid superblock checksum on %s\n",
1123 bdevname(rdev->bdev,b));
1126 if (le64_to_cpu(sb->data_size) < 10) {
1127 printk("md: data_size too small on %s\n",
1128 bdevname(rdev->bdev,b));
1131 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1132 if (sb->level != cpu_to_le32(1) &&
1133 sb->level != cpu_to_le32(4) &&
1134 sb->level != cpu_to_le32(5) &&
1135 sb->level != cpu_to_le32(6) &&
1136 sb->level != cpu_to_le32(10)) {
1138 "md: bitmaps not supported for this level.\n");
1143 rdev->preferred_minor = 0xffff;
1144 rdev->data_offset = le64_to_cpu(sb->data_offset);
1145 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1147 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1148 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1149 if (rdev->sb_size & bmask)
1150 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1153 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1156 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1159 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1165 struct mdp_superblock_1 *refsb =
1166 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1168 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1169 sb->level != refsb->level ||
1170 sb->layout != refsb->layout ||
1171 sb->chunksize != refsb->chunksize) {
1172 printk(KERN_WARNING "md: %s has strangely different"
1173 " superblock to %s\n",
1174 bdevname(rdev->bdev,b),
1175 bdevname(refdev->bdev,b2));
1178 ev1 = le64_to_cpu(sb->events);
1179 ev2 = le64_to_cpu(refsb->events);
1187 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1189 rdev->size = rdev->sb_start / 2;
1190 if (rdev->size < le64_to_cpu(sb->data_size)/2)
1192 rdev->size = le64_to_cpu(sb->data_size)/2;
1193 if (le32_to_cpu(sb->chunksize))
1194 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1196 if (le64_to_cpu(sb->size) > rdev->size*2)
1201 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1203 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1204 __u64 ev1 = le64_to_cpu(sb->events);
1206 rdev->raid_disk = -1;
1207 clear_bit(Faulty, &rdev->flags);
1208 clear_bit(In_sync, &rdev->flags);
1209 clear_bit(WriteMostly, &rdev->flags);
1210 clear_bit(BarriersNotsupp, &rdev->flags);
1212 if (mddev->raid_disks == 0) {
1213 mddev->major_version = 1;
1214 mddev->patch_version = 0;
1215 mddev->external = 0;
1216 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1217 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1218 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1219 mddev->level = le32_to_cpu(sb->level);
1220 mddev->clevel[0] = 0;
1221 mddev->layout = le32_to_cpu(sb->layout);
1222 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1223 mddev->size = le64_to_cpu(sb->size)/2;
1224 mddev->events = ev1;
1225 mddev->bitmap_offset = 0;
1226 mddev->default_bitmap_offset = 1024 >> 9;
1228 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1229 memcpy(mddev->uuid, sb->set_uuid, 16);
1231 mddev->max_disks = (4096-256)/2;
1233 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1234 mddev->bitmap_file == NULL )
1235 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1237 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1238 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1239 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1240 mddev->new_level = le32_to_cpu(sb->new_level);
1241 mddev->new_layout = le32_to_cpu(sb->new_layout);
1242 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1244 mddev->reshape_position = MaxSector;
1245 mddev->delta_disks = 0;
1246 mddev->new_level = mddev->level;
1247 mddev->new_layout = mddev->layout;
1248 mddev->new_chunk = mddev->chunk_size;
1251 } else if (mddev->pers == NULL) {
1252 /* Insist of good event counter while assembling */
1254 if (ev1 < mddev->events)
1256 } else if (mddev->bitmap) {
1257 /* If adding to array with a bitmap, then we can accept an
1258 * older device, but not too old.
1260 if (ev1 < mddev->bitmap->events_cleared)
1263 if (ev1 < mddev->events)
1264 /* just a hot-add of a new device, leave raid_disk at -1 */
1267 if (mddev->level != LEVEL_MULTIPATH) {
1269 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1271 case 0xffff: /* spare */
1273 case 0xfffe: /* faulty */
1274 set_bit(Faulty, &rdev->flags);
1277 if ((le32_to_cpu(sb->feature_map) &
1278 MD_FEATURE_RECOVERY_OFFSET))
1279 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1281 set_bit(In_sync, &rdev->flags);
1282 rdev->raid_disk = role;
1285 if (sb->devflags & WriteMostly1)
1286 set_bit(WriteMostly, &rdev->flags);
1287 } else /* MULTIPATH are always insync */
1288 set_bit(In_sync, &rdev->flags);
1293 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1295 struct mdp_superblock_1 *sb;
1298 /* make rdev->sb match mddev and rdev data. */
1300 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1302 sb->feature_map = 0;
1304 sb->recovery_offset = cpu_to_le64(0);
1305 memset(sb->pad1, 0, sizeof(sb->pad1));
1306 memset(sb->pad2, 0, sizeof(sb->pad2));
1307 memset(sb->pad3, 0, sizeof(sb->pad3));
1309 sb->utime = cpu_to_le64((__u64)mddev->utime);
1310 sb->events = cpu_to_le64(mddev->events);
1312 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1314 sb->resync_offset = cpu_to_le64(0);
1316 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1318 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1319 sb->size = cpu_to_le64(mddev->size<<1);
1321 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1322 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1323 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1326 if (rdev->raid_disk >= 0 &&
1327 !test_bit(In_sync, &rdev->flags) &&
1328 rdev->recovery_offset > 0) {
1329 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1330 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1333 if (mddev->reshape_position != MaxSector) {
1334 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1335 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1336 sb->new_layout = cpu_to_le32(mddev->new_layout);
1337 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1338 sb->new_level = cpu_to_le32(mddev->new_level);
1339 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1343 list_for_each_entry(rdev2, &mddev->disks, same_set)
1344 if (rdev2->desc_nr+1 > max_dev)
1345 max_dev = rdev2->desc_nr+1;
1347 if (max_dev > le32_to_cpu(sb->max_dev))
1348 sb->max_dev = cpu_to_le32(max_dev);
1349 for (i=0; i<max_dev;i++)
1350 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1352 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1354 if (test_bit(Faulty, &rdev2->flags))
1355 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1356 else if (test_bit(In_sync, &rdev2->flags))
1357 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1358 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1359 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1361 sb->dev_roles[i] = cpu_to_le16(0xffff);
1364 sb->sb_csum = calc_sb_1_csum(sb);
1367 static unsigned long long
1368 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1370 struct mdp_superblock_1 *sb;
1371 sector_t max_sectors;
1372 if (num_sectors && num_sectors < rdev->mddev->size * 2)
1373 return 0; /* component must fit device */
1374 if (rdev->sb_start < rdev->data_offset) {
1375 /* minor versions 1 and 2; superblock before data */
1376 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1377 max_sectors -= rdev->data_offset;
1378 if (!num_sectors || num_sectors > max_sectors)
1379 num_sectors = max_sectors;
1380 } else if (rdev->mddev->bitmap_offset) {
1381 /* minor version 0 with bitmap we can't move */
1384 /* minor version 0; superblock after data */
1386 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1387 sb_start &= ~(sector_t)(4*2 - 1);
1388 max_sectors = rdev->size * 2 + sb_start - rdev->sb_start;
1389 if (!num_sectors || num_sectors > max_sectors)
1390 num_sectors = max_sectors;
1391 rdev->sb_start = sb_start;
1393 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1394 sb->data_size = cpu_to_le64(num_sectors);
1395 sb->super_offset = rdev->sb_start;
1396 sb->sb_csum = calc_sb_1_csum(sb);
1397 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1399 md_super_wait(rdev->mddev);
1400 return num_sectors / 2; /* kB for sysfs */
1403 static struct super_type super_types[] = {
1406 .owner = THIS_MODULE,
1407 .load_super = super_90_load,
1408 .validate_super = super_90_validate,
1409 .sync_super = super_90_sync,
1410 .rdev_size_change = super_90_rdev_size_change,
1414 .owner = THIS_MODULE,
1415 .load_super = super_1_load,
1416 .validate_super = super_1_validate,
1417 .sync_super = super_1_sync,
1418 .rdev_size_change = super_1_rdev_size_change,
1422 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1424 mdk_rdev_t *rdev, *rdev2;
1427 rdev_for_each_rcu(rdev, mddev1)
1428 rdev_for_each_rcu(rdev2, mddev2)
1429 if (rdev->bdev->bd_contains ==
1430 rdev2->bdev->bd_contains) {
1438 static LIST_HEAD(pending_raid_disks);
1440 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1442 struct mdk_personality *pers = mddev->pers;
1443 struct gendisk *disk = mddev->gendisk;
1444 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1445 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1447 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1448 if (pers && pers->level >= 4 && pers->level <= 6)
1451 /* If rdev is integrity capable, register profile for mddev */
1452 if (!bi_mddev && bi_rdev) {
1453 if (blk_integrity_register(disk, bi_rdev))
1454 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1455 __func__, disk->disk_name);
1457 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1462 /* Check that mddev and rdev have matching profiles */
1463 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1464 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1465 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1466 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1468 blk_integrity_unregister(disk);
1472 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1474 char b[BDEVNAME_SIZE];
1484 /* prevent duplicates */
1485 if (find_rdev(mddev, rdev->bdev->bd_dev))
1488 /* make sure rdev->size exceeds mddev->size */
1489 if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1491 /* Cannot change size, so fail
1492 * If mddev->level <= 0, then we don't care
1493 * about aligning sizes (e.g. linear)
1495 if (mddev->level > 0)
1498 mddev->size = rdev->size;
1501 /* Verify rdev->desc_nr is unique.
1502 * If it is -1, assign a free number, else
1503 * check number is not in use
1505 if (rdev->desc_nr < 0) {
1507 if (mddev->pers) choice = mddev->raid_disks;
1508 while (find_rdev_nr(mddev, choice))
1510 rdev->desc_nr = choice;
1512 if (find_rdev_nr(mddev, rdev->desc_nr))
1515 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1516 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1517 mdname(mddev), mddev->max_disks);
1520 bdevname(rdev->bdev,b);
1521 while ( (s=strchr(b, '/')) != NULL)
1524 rdev->mddev = mddev;
1525 printk(KERN_INFO "md: bind<%s>\n", b);
1527 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1530 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1531 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1532 kobject_del(&rdev->kobj);
1535 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1537 list_add_rcu(&rdev->same_set, &mddev->disks);
1538 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1540 /* May as well allow recovery to be retried once */
1541 mddev->recovery_disabled = 0;
1543 md_integrity_check(rdev, mddev);
1547 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1552 static void md_delayed_delete(struct work_struct *ws)
1554 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1555 kobject_del(&rdev->kobj);
1556 kobject_put(&rdev->kobj);
1559 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1561 char b[BDEVNAME_SIZE];
1566 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1567 list_del_rcu(&rdev->same_set);
1568 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1570 sysfs_remove_link(&rdev->kobj, "block");
1571 sysfs_put(rdev->sysfs_state);
1572 rdev->sysfs_state = NULL;
1573 /* We need to delay this, otherwise we can deadlock when
1574 * writing to 'remove' to "dev/state". We also need
1575 * to delay it due to rcu usage.
1578 INIT_WORK(&rdev->del_work, md_delayed_delete);
1579 kobject_get(&rdev->kobj);
1580 schedule_work(&rdev->del_work);
1584 * prevent the device from being mounted, repartitioned or
1585 * otherwise reused by a RAID array (or any other kernel
1586 * subsystem), by bd_claiming the device.
1588 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1591 struct block_device *bdev;
1592 char b[BDEVNAME_SIZE];
1594 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1596 printk(KERN_ERR "md: could not open %s.\n",
1597 __bdevname(dev, b));
1598 return PTR_ERR(bdev);
1600 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1602 printk(KERN_ERR "md: could not bd_claim %s.\n",
1604 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1608 set_bit(AllReserved, &rdev->flags);
1613 static void unlock_rdev(mdk_rdev_t *rdev)
1615 struct block_device *bdev = rdev->bdev;
1620 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1623 void md_autodetect_dev(dev_t dev);
1625 static void export_rdev(mdk_rdev_t * rdev)
1627 char b[BDEVNAME_SIZE];
1628 printk(KERN_INFO "md: export_rdev(%s)\n",
1629 bdevname(rdev->bdev,b));
1634 if (test_bit(AutoDetected, &rdev->flags))
1635 md_autodetect_dev(rdev->bdev->bd_dev);
1638 kobject_put(&rdev->kobj);
1641 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1643 unbind_rdev_from_array(rdev);
1647 static void export_array(mddev_t *mddev)
1649 mdk_rdev_t *rdev, *tmp;
1651 rdev_for_each(rdev, tmp, mddev) {
1656 kick_rdev_from_array(rdev);
1658 if (!list_empty(&mddev->disks))
1660 mddev->raid_disks = 0;
1661 mddev->major_version = 0;
1664 static void print_desc(mdp_disk_t *desc)
1666 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1667 desc->major,desc->minor,desc->raid_disk,desc->state);
1670 static void print_sb_90(mdp_super_t *sb)
1675 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1676 sb->major_version, sb->minor_version, sb->patch_version,
1677 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1679 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1680 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1681 sb->md_minor, sb->layout, sb->chunk_size);
1682 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1683 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1684 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1685 sb->failed_disks, sb->spare_disks,
1686 sb->sb_csum, (unsigned long)sb->events_lo);
1689 for (i = 0; i < MD_SB_DISKS; i++) {
1692 desc = sb->disks + i;
1693 if (desc->number || desc->major || desc->minor ||
1694 desc->raid_disk || (desc->state && (desc->state != 4))) {
1695 printk(" D %2d: ", i);
1699 printk(KERN_INFO "md: THIS: ");
1700 print_desc(&sb->this_disk);
1703 static void print_sb_1(struct mdp_superblock_1 *sb)
1707 uuid = sb->set_uuid;
1708 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1709 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1710 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1711 le32_to_cpu(sb->major_version),
1712 le32_to_cpu(sb->feature_map),
1713 uuid[0], uuid[1], uuid[2], uuid[3],
1714 uuid[4], uuid[5], uuid[6], uuid[7],
1715 uuid[8], uuid[9], uuid[10], uuid[11],
1716 uuid[12], uuid[13], uuid[14], uuid[15],
1718 (unsigned long long)le64_to_cpu(sb->ctime)
1719 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1721 uuid = sb->device_uuid;
1722 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1724 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1725 ":%02x%02x%02x%02x%02x%02x\n"
1726 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1727 KERN_INFO "md: (MaxDev:%u) \n",
1728 le32_to_cpu(sb->level),
1729 (unsigned long long)le64_to_cpu(sb->size),
1730 le32_to_cpu(sb->raid_disks),
1731 le32_to_cpu(sb->layout),
1732 le32_to_cpu(sb->chunksize),
1733 (unsigned long long)le64_to_cpu(sb->data_offset),
1734 (unsigned long long)le64_to_cpu(sb->data_size),
1735 (unsigned long long)le64_to_cpu(sb->super_offset),
1736 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1737 le32_to_cpu(sb->dev_number),
1738 uuid[0], uuid[1], uuid[2], uuid[3],
1739 uuid[4], uuid[5], uuid[6], uuid[7],
1740 uuid[8], uuid[9], uuid[10], uuid[11],
1741 uuid[12], uuid[13], uuid[14], uuid[15],
1743 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1744 (unsigned long long)le64_to_cpu(sb->events),
1745 (unsigned long long)le64_to_cpu(sb->resync_offset),
1746 le32_to_cpu(sb->sb_csum),
1747 le32_to_cpu(sb->max_dev)
1751 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1753 char b[BDEVNAME_SIZE];
1754 printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1755 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1756 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1758 if (rdev->sb_loaded) {
1759 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1760 switch (major_version) {
1762 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1765 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1769 printk(KERN_INFO "md: no rdev superblock!\n");
1772 static void md_print_devices(void)
1774 struct list_head *tmp;
1777 char b[BDEVNAME_SIZE];
1780 printk("md: **********************************\n");
1781 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1782 printk("md: **********************************\n");
1783 for_each_mddev(mddev, tmp) {
1786 bitmap_print_sb(mddev->bitmap);
1788 printk("%s: ", mdname(mddev));
1789 list_for_each_entry(rdev, &mddev->disks, same_set)
1790 printk("<%s>", bdevname(rdev->bdev,b));
1793 list_for_each_entry(rdev, &mddev->disks, same_set)
1794 print_rdev(rdev, mddev->major_version);
1796 printk("md: **********************************\n");
1801 static void sync_sbs(mddev_t * mddev, int nospares)
1803 /* Update each superblock (in-memory image), but
1804 * if we are allowed to, skip spares which already
1805 * have the right event counter, or have one earlier
1806 * (which would mean they aren't being marked as dirty
1807 * with the rest of the array)
1811 list_for_each_entry(rdev, &mddev->disks, same_set) {
1812 if (rdev->sb_events == mddev->events ||
1814 rdev->raid_disk < 0 &&
1815 (rdev->sb_events&1)==0 &&
1816 rdev->sb_events+1 == mddev->events)) {
1817 /* Don't update this superblock */
1818 rdev->sb_loaded = 2;
1820 super_types[mddev->major_version].
1821 sync_super(mddev, rdev);
1822 rdev->sb_loaded = 1;
1827 static void md_update_sb(mddev_t * mddev, int force_change)
1833 if (mddev->external)
1836 spin_lock_irq(&mddev->write_lock);
1838 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1839 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1841 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1842 /* just a clean<-> dirty transition, possibly leave spares alone,
1843 * though if events isn't the right even/odd, we will have to do
1849 if (mddev->degraded)
1850 /* If the array is degraded, then skipping spares is both
1851 * dangerous and fairly pointless.
1852 * Dangerous because a device that was removed from the array
1853 * might have a event_count that still looks up-to-date,
1854 * so it can be re-added without a resync.
1855 * Pointless because if there are any spares to skip,
1856 * then a recovery will happen and soon that array won't
1857 * be degraded any more and the spare can go back to sleep then.
1861 sync_req = mddev->in_sync;
1862 mddev->utime = get_seconds();
1864 /* If this is just a dirty<->clean transition, and the array is clean
1865 * and 'events' is odd, we can roll back to the previous clean state */
1867 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1868 && (mddev->events & 1)
1869 && mddev->events != 1)
1872 /* otherwise we have to go forward and ... */
1874 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1875 /* .. if the array isn't clean, insist on an odd 'events' */
1876 if ((mddev->events&1)==0) {
1881 /* otherwise insist on an even 'events' (for clean states) */
1882 if ((mddev->events&1)) {
1889 if (!mddev->events) {
1891 * oops, this 64-bit counter should never wrap.
1892 * Either we are in around ~1 trillion A.C., assuming
1893 * 1 reboot per second, or we have a bug:
1900 * do not write anything to disk if using
1901 * nonpersistent superblocks
1903 if (!mddev->persistent) {
1904 if (!mddev->external)
1905 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1907 spin_unlock_irq(&mddev->write_lock);
1908 wake_up(&mddev->sb_wait);
1911 sync_sbs(mddev, nospares);
1912 spin_unlock_irq(&mddev->write_lock);
1915 "md: updating %s RAID superblock on device (in sync %d)\n",
1916 mdname(mddev),mddev->in_sync);
1918 bitmap_update_sb(mddev->bitmap);
1919 list_for_each_entry(rdev, &mddev->disks, same_set) {
1920 char b[BDEVNAME_SIZE];
1921 dprintk(KERN_INFO "md: ");
1922 if (rdev->sb_loaded != 1)
1923 continue; /* no noise on spare devices */
1924 if (test_bit(Faulty, &rdev->flags))
1925 dprintk("(skipping faulty ");
1927 dprintk("%s ", bdevname(rdev->bdev,b));
1928 if (!test_bit(Faulty, &rdev->flags)) {
1929 md_super_write(mddev,rdev,
1930 rdev->sb_start, rdev->sb_size,
1932 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1933 bdevname(rdev->bdev,b),
1934 (unsigned long long)rdev->sb_start);
1935 rdev->sb_events = mddev->events;
1939 if (mddev->level == LEVEL_MULTIPATH)
1940 /* only need to write one superblock... */
1943 md_super_wait(mddev);
1944 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1946 spin_lock_irq(&mddev->write_lock);
1947 if (mddev->in_sync != sync_req ||
1948 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1949 /* have to write it out again */
1950 spin_unlock_irq(&mddev->write_lock);
1953 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1954 spin_unlock_irq(&mddev->write_lock);
1955 wake_up(&mddev->sb_wait);
1959 /* words written to sysfs files may, or may not, be \n terminated.
1960 * We want to accept with case. For this we use cmd_match.
1962 static int cmd_match(const char *cmd, const char *str)
1964 /* See if cmd, written into a sysfs file, matches
1965 * str. They must either be the same, or cmd can
1966 * have a trailing newline
1968 while (*cmd && *str && *cmd == *str) {
1979 struct rdev_sysfs_entry {
1980 struct attribute attr;
1981 ssize_t (*show)(mdk_rdev_t *, char *);
1982 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1986 state_show(mdk_rdev_t *rdev, char *page)
1991 if (test_bit(Faulty, &rdev->flags)) {
1992 len+= sprintf(page+len, "%sfaulty",sep);
1995 if (test_bit(In_sync, &rdev->flags)) {
1996 len += sprintf(page+len, "%sin_sync",sep);
1999 if (test_bit(WriteMostly, &rdev->flags)) {
2000 len += sprintf(page+len, "%swrite_mostly",sep);
2003 if (test_bit(Blocked, &rdev->flags)) {
2004 len += sprintf(page+len, "%sblocked", sep);
2007 if (!test_bit(Faulty, &rdev->flags) &&
2008 !test_bit(In_sync, &rdev->flags)) {
2009 len += sprintf(page+len, "%sspare", sep);
2012 return len+sprintf(page+len, "\n");
2016 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2019 * faulty - simulates and error
2020 * remove - disconnects the device
2021 * writemostly - sets write_mostly
2022 * -writemostly - clears write_mostly
2023 * blocked - sets the Blocked flag
2024 * -blocked - clears the Blocked flag
2027 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2028 md_error(rdev->mddev, rdev);
2030 } else if (cmd_match(buf, "remove")) {
2031 if (rdev->raid_disk >= 0)
2034 mddev_t *mddev = rdev->mddev;
2035 kick_rdev_from_array(rdev);
2037 md_update_sb(mddev, 1);
2038 md_new_event(mddev);
2041 } else if (cmd_match(buf, "writemostly")) {
2042 set_bit(WriteMostly, &rdev->flags);
2044 } else if (cmd_match(buf, "-writemostly")) {
2045 clear_bit(WriteMostly, &rdev->flags);
2047 } else if (cmd_match(buf, "blocked")) {
2048 set_bit(Blocked, &rdev->flags);
2050 } else if (cmd_match(buf, "-blocked")) {
2051 clear_bit(Blocked, &rdev->flags);
2052 wake_up(&rdev->blocked_wait);
2053 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2054 md_wakeup_thread(rdev->mddev->thread);
2058 if (!err && rdev->sysfs_state)
2059 sysfs_notify_dirent(rdev->sysfs_state);
2060 return err ? err : len;
2062 static struct rdev_sysfs_entry rdev_state =
2063 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2066 errors_show(mdk_rdev_t *rdev, char *page)
2068 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2072 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2075 unsigned long n = simple_strtoul(buf, &e, 10);
2076 if (*buf && (*e == 0 || *e == '\n')) {
2077 atomic_set(&rdev->corrected_errors, n);
2082 static struct rdev_sysfs_entry rdev_errors =
2083 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2086 slot_show(mdk_rdev_t *rdev, char *page)
2088 if (rdev->raid_disk < 0)
2089 return sprintf(page, "none\n");
2091 return sprintf(page, "%d\n", rdev->raid_disk);
2095 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2100 int slot = simple_strtoul(buf, &e, 10);
2101 if (strncmp(buf, "none", 4)==0)
2103 else if (e==buf || (*e && *e!= '\n'))
2105 if (rdev->mddev->pers && slot == -1) {
2106 /* Setting 'slot' on an active array requires also
2107 * updating the 'rd%d' link, and communicating
2108 * with the personality with ->hot_*_disk.
2109 * For now we only support removing
2110 * failed/spare devices. This normally happens automatically,
2111 * but not when the metadata is externally managed.
2113 if (rdev->raid_disk == -1)
2115 /* personality does all needed checks */
2116 if (rdev->mddev->pers->hot_add_disk == NULL)
2118 err = rdev->mddev->pers->
2119 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2122 sprintf(nm, "rd%d", rdev->raid_disk);
2123 sysfs_remove_link(&rdev->mddev->kobj, nm);
2124 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2125 md_wakeup_thread(rdev->mddev->thread);
2126 } else if (rdev->mddev->pers) {
2128 /* Activating a spare .. or possibly reactivating
2129 * if we every get bitmaps working here.
2132 if (rdev->raid_disk != -1)
2135 if (rdev->mddev->pers->hot_add_disk == NULL)
2138 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2139 if (rdev2->raid_disk == slot)
2142 rdev->raid_disk = slot;
2143 if (test_bit(In_sync, &rdev->flags))
2144 rdev->saved_raid_disk = slot;
2146 rdev->saved_raid_disk = -1;
2147 err = rdev->mddev->pers->
2148 hot_add_disk(rdev->mddev, rdev);
2150 rdev->raid_disk = -1;
2153 sysfs_notify_dirent(rdev->sysfs_state);
2154 sprintf(nm, "rd%d", rdev->raid_disk);
2155 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2157 "md: cannot register "
2159 nm, mdname(rdev->mddev));
2161 /* don't wakeup anyone, leave that to userspace. */
2163 if (slot >= rdev->mddev->raid_disks)
2165 rdev->raid_disk = slot;
2166 /* assume it is working */
2167 clear_bit(Faulty, &rdev->flags);
2168 clear_bit(WriteMostly, &rdev->flags);
2169 set_bit(In_sync, &rdev->flags);
2170 sysfs_notify_dirent(rdev->sysfs_state);
2176 static struct rdev_sysfs_entry rdev_slot =
2177 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2180 offset_show(mdk_rdev_t *rdev, char *page)
2182 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2186 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2189 unsigned long long offset = simple_strtoull(buf, &e, 10);
2190 if (e==buf || (*e && *e != '\n'))
2192 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2194 if (rdev->size && rdev->mddev->external)
2195 /* Must set offset before size, so overlap checks
2198 rdev->data_offset = offset;
2202 static struct rdev_sysfs_entry rdev_offset =
2203 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2206 rdev_size_show(mdk_rdev_t *rdev, char *page)
2208 return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2211 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2213 /* check if two start/length pairs overlap */
2222 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2224 unsigned long long size;
2225 unsigned long long oldsize = rdev->size;
2226 mddev_t *my_mddev = rdev->mddev;
2228 if (strict_strtoull(buf, 10, &size) < 0)
2230 if (my_mddev->pers && rdev->raid_disk >= 0) {
2231 if (my_mddev->persistent) {
2232 size = super_types[my_mddev->major_version].
2233 rdev_size_change(rdev, size * 2);
2237 size = (rdev->bdev->bd_inode->i_size >> 10);
2238 size -= rdev->data_offset/2;
2241 if (size < my_mddev->size)
2242 return -EINVAL; /* component must fit device */
2245 if (size > oldsize && my_mddev->external) {
2246 /* need to check that all other rdevs with the same ->bdev
2247 * do not overlap. We need to unlock the mddev to avoid
2248 * a deadlock. We have already changed rdev->size, and if
2249 * we have to change it back, we will have the lock again.
2253 struct list_head *tmp;
2255 mddev_unlock(my_mddev);
2256 for_each_mddev(mddev, tmp) {
2260 list_for_each_entry(rdev2, &mddev->disks, same_set)
2261 if (test_bit(AllReserved, &rdev2->flags) ||
2262 (rdev->bdev == rdev2->bdev &&
2264 overlaps(rdev->data_offset, rdev->size * 2,
2266 rdev2->size * 2))) {
2270 mddev_unlock(mddev);
2276 mddev_lock(my_mddev);
2278 /* Someone else could have slipped in a size
2279 * change here, but doing so is just silly.
2280 * We put oldsize back because we *know* it is
2281 * safe, and trust userspace not to race with
2284 rdev->size = oldsize;
2291 static struct rdev_sysfs_entry rdev_size =
2292 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2294 static struct attribute *rdev_default_attrs[] = {
2303 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2305 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2306 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2307 mddev_t *mddev = rdev->mddev;
2313 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2315 if (rdev->mddev == NULL)
2318 rv = entry->show(rdev, page);
2319 mddev_unlock(mddev);
2325 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2326 const char *page, size_t length)
2328 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2329 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2331 mddev_t *mddev = rdev->mddev;
2335 if (!capable(CAP_SYS_ADMIN))
2337 rv = mddev ? mddev_lock(mddev): -EBUSY;
2339 if (rdev->mddev == NULL)
2342 rv = entry->store(rdev, page, length);
2343 mddev_unlock(mddev);
2348 static void rdev_free(struct kobject *ko)
2350 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2353 static struct sysfs_ops rdev_sysfs_ops = {
2354 .show = rdev_attr_show,
2355 .store = rdev_attr_store,
2357 static struct kobj_type rdev_ktype = {
2358 .release = rdev_free,
2359 .sysfs_ops = &rdev_sysfs_ops,
2360 .default_attrs = rdev_default_attrs,
2364 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2366 * mark the device faulty if:
2368 * - the device is nonexistent (zero size)
2369 * - the device has no valid superblock
2371 * a faulty rdev _never_ has rdev->sb set.
2373 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2375 char b[BDEVNAME_SIZE];
2380 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2382 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2383 return ERR_PTR(-ENOMEM);
2386 if ((err = alloc_disk_sb(rdev)))
2389 err = lock_rdev(rdev, newdev, super_format == -2);
2393 kobject_init(&rdev->kobj, &rdev_ktype);
2396 rdev->saved_raid_disk = -1;
2397 rdev->raid_disk = -1;
2399 rdev->data_offset = 0;
2400 rdev->sb_events = 0;
2401 atomic_set(&rdev->nr_pending, 0);
2402 atomic_set(&rdev->read_errors, 0);
2403 atomic_set(&rdev->corrected_errors, 0);
2405 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2408 "md: %s has zero or unknown size, marking faulty!\n",
2409 bdevname(rdev->bdev,b));
2414 if (super_format >= 0) {
2415 err = super_types[super_format].
2416 load_super(rdev, NULL, super_minor);
2417 if (err == -EINVAL) {
2419 "md: %s does not have a valid v%d.%d "
2420 "superblock, not importing!\n",
2421 bdevname(rdev->bdev,b),
2422 super_format, super_minor);
2427 "md: could not read %s's sb, not importing!\n",
2428 bdevname(rdev->bdev,b));
2433 INIT_LIST_HEAD(&rdev->same_set);
2434 init_waitqueue_head(&rdev->blocked_wait);
2439 if (rdev->sb_page) {
2445 return ERR_PTR(err);
2449 * Check a full RAID array for plausibility
2453 static void analyze_sbs(mddev_t * mddev)
2456 mdk_rdev_t *rdev, *freshest, *tmp;
2457 char b[BDEVNAME_SIZE];
2460 rdev_for_each(rdev, tmp, mddev)
2461 switch (super_types[mddev->major_version].
2462 load_super(rdev, freshest, mddev->minor_version)) {
2470 "md: fatal superblock inconsistency in %s"
2471 " -- removing from array\n",
2472 bdevname(rdev->bdev,b));
2473 kick_rdev_from_array(rdev);
2477 super_types[mddev->major_version].
2478 validate_super(mddev, freshest);
2481 rdev_for_each(rdev, tmp, mddev) {
2482 if (rdev->desc_nr >= mddev->max_disks ||
2483 i > mddev->max_disks) {
2485 "md: %s: %s: only %d devices permitted\n",
2486 mdname(mddev), bdevname(rdev->bdev, b),
2488 kick_rdev_from_array(rdev);
2491 if (rdev != freshest)
2492 if (super_types[mddev->major_version].
2493 validate_super(mddev, rdev)) {
2494 printk(KERN_WARNING "md: kicking non-fresh %s"
2496 bdevname(rdev->bdev,b));
2497 kick_rdev_from_array(rdev);
2500 if (mddev->level == LEVEL_MULTIPATH) {
2501 rdev->desc_nr = i++;
2502 rdev->raid_disk = rdev->desc_nr;
2503 set_bit(In_sync, &rdev->flags);
2504 } else if (rdev->raid_disk >= mddev->raid_disks) {
2505 rdev->raid_disk = -1;
2506 clear_bit(In_sync, &rdev->flags);
2512 if (mddev->recovery_cp != MaxSector &&
2514 printk(KERN_ERR "md: %s: raid array is not clean"
2515 " -- starting background reconstruction\n",
2520 static void md_safemode_timeout(unsigned long data);
2523 safe_delay_show(mddev_t *mddev, char *page)
2525 int msec = (mddev->safemode_delay*1000)/HZ;
2526 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2529 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2537 /* remove a period, and count digits after it */
2538 if (len >= sizeof(buf))
2540 strlcpy(buf, cbuf, sizeof(buf));
2541 for (i=0; i<len; i++) {
2543 if (isdigit(buf[i])) {
2548 } else if (buf[i] == '.') {
2553 if (strict_strtoul(buf, 10, &msec) < 0)
2555 msec = (msec * 1000) / scale;
2557 mddev->safemode_delay = 0;
2559 unsigned long old_delay = mddev->safemode_delay;
2560 mddev->safemode_delay = (msec*HZ)/1000;
2561 if (mddev->safemode_delay == 0)
2562 mddev->safemode_delay = 1;
2563 if (mddev->safemode_delay < old_delay)
2564 md_safemode_timeout((unsigned long)mddev);
2568 static struct md_sysfs_entry md_safe_delay =
2569 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2572 level_show(mddev_t *mddev, char *page)
2574 struct mdk_personality *p = mddev->pers;
2576 return sprintf(page, "%s\n", p->name);
2577 else if (mddev->clevel[0])
2578 return sprintf(page, "%s\n", mddev->clevel);
2579 else if (mddev->level != LEVEL_NONE)
2580 return sprintf(page, "%d\n", mddev->level);
2586 level_store(mddev_t *mddev, const char *buf, size_t len)
2593 if (len >= sizeof(mddev->clevel))
2595 strncpy(mddev->clevel, buf, len);
2596 if (mddev->clevel[len-1] == '\n')
2598 mddev->clevel[len] = 0;
2599 mddev->level = LEVEL_NONE;
2603 static struct md_sysfs_entry md_level =
2604 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2608 layout_show(mddev_t *mddev, char *page)
2610 /* just a number, not meaningful for all levels */
2611 if (mddev->reshape_position != MaxSector &&
2612 mddev->layout != mddev->new_layout)
2613 return sprintf(page, "%d (%d)\n",
2614 mddev->new_layout, mddev->layout);
2615 return sprintf(page, "%d\n", mddev->layout);
2619 layout_store(mddev_t *mddev, const char *buf, size_t len)
2622 unsigned long n = simple_strtoul(buf, &e, 10);
2624 if (!*buf || (*e && *e != '\n'))
2629 if (mddev->reshape_position != MaxSector)
2630 mddev->new_layout = n;
2635 static struct md_sysfs_entry md_layout =
2636 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2640 raid_disks_show(mddev_t *mddev, char *page)
2642 if (mddev->raid_disks == 0)
2644 if (mddev->reshape_position != MaxSector &&
2645 mddev->delta_disks != 0)
2646 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2647 mddev->raid_disks - mddev->delta_disks);
2648 return sprintf(page, "%d\n", mddev->raid_disks);
2651 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2654 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2658 unsigned long n = simple_strtoul(buf, &e, 10);
2660 if (!*buf || (*e && *e != '\n'))
2664 rv = update_raid_disks(mddev, n);
2665 else if (mddev->reshape_position != MaxSector) {
2666 int olddisks = mddev->raid_disks - mddev->delta_disks;
2667 mddev->delta_disks = n - olddisks;
2668 mddev->raid_disks = n;
2670 mddev->raid_disks = n;
2671 return rv ? rv : len;
2673 static struct md_sysfs_entry md_raid_disks =
2674 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2677 chunk_size_show(mddev_t *mddev, char *page)
2679 if (mddev->reshape_position != MaxSector &&
2680 mddev->chunk_size != mddev->new_chunk)
2681 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2683 return sprintf(page, "%d\n", mddev->chunk_size);
2687 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2689 /* can only set chunk_size if array is not yet active */
2691 unsigned long n = simple_strtoul(buf, &e, 10);
2693 if (!*buf || (*e && *e != '\n'))
2698 else if (mddev->reshape_position != MaxSector)
2699 mddev->new_chunk = n;
2701 mddev->chunk_size = n;
2704 static struct md_sysfs_entry md_chunk_size =
2705 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2708 resync_start_show(mddev_t *mddev, char *page)
2710 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2714 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2717 unsigned long long n = simple_strtoull(buf, &e, 10);
2721 if (!*buf || (*e && *e != '\n'))
2724 mddev->recovery_cp = n;
2727 static struct md_sysfs_entry md_resync_start =
2728 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2731 * The array state can be:
2734 * No devices, no size, no level
2735 * Equivalent to STOP_ARRAY ioctl
2737 * May have some settings, but array is not active
2738 * all IO results in error
2739 * When written, doesn't tear down array, but just stops it
2740 * suspended (not supported yet)
2741 * All IO requests will block. The array can be reconfigured.
2742 * Writing this, if accepted, will block until array is quiescent
2744 * no resync can happen. no superblocks get written.
2745 * write requests fail
2747 * like readonly, but behaves like 'clean' on a write request.
2749 * clean - no pending writes, but otherwise active.
2750 * When written to inactive array, starts without resync
2751 * If a write request arrives then
2752 * if metadata is known, mark 'dirty' and switch to 'active'.
2753 * if not known, block and switch to write-pending
2754 * If written to an active array that has pending writes, then fails.
2756 * fully active: IO and resync can be happening.
2757 * When written to inactive array, starts with resync
2760 * clean, but writes are blocked waiting for 'active' to be written.
2763 * like active, but no writes have been seen for a while (100msec).
2766 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2767 write_pending, active_idle, bad_word};
2768 static char *array_states[] = {
2769 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2770 "write-pending", "active-idle", NULL };
2772 static int match_word(const char *word, char **list)
2775 for (n=0; list[n]; n++)
2776 if (cmd_match(word, list[n]))
2782 array_state_show(mddev_t *mddev, char *page)
2784 enum array_state st = inactive;
2797 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2799 else if (mddev->safemode)
2805 if (list_empty(&mddev->disks) &&
2806 mddev->raid_disks == 0 &&
2812 return sprintf(page, "%s\n", array_states[st]);
2815 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2816 static int do_md_run(mddev_t * mddev);
2817 static int restart_array(mddev_t *mddev);
2820 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2823 enum array_state st = match_word(buf, array_states);
2828 /* stopping an active array */
2829 if (atomic_read(&mddev->openers) > 0)
2831 err = do_md_stop(mddev, 0, 0);
2834 /* stopping an active array */
2836 if (atomic_read(&mddev->openers) > 0)
2838 err = do_md_stop(mddev, 2, 0);
2840 err = 0; /* already inactive */
2843 break; /* not supported yet */
2846 err = do_md_stop(mddev, 1, 0);
2849 set_disk_ro(mddev->gendisk, 1);
2850 err = do_md_run(mddev);
2856 err = do_md_stop(mddev, 1, 0);
2857 else if (mddev->ro == 1)
2858 err = restart_array(mddev);
2861 set_disk_ro(mddev->gendisk, 0);
2865 err = do_md_run(mddev);
2870 restart_array(mddev);
2871 spin_lock_irq(&mddev->write_lock);
2872 if (atomic_read(&mddev->writes_pending) == 0) {
2873 if (mddev->in_sync == 0) {
2875 if (mddev->safemode == 1)
2876 mddev->safemode = 0;
2877 if (mddev->persistent)
2878 set_bit(MD_CHANGE_CLEAN,
2884 spin_unlock_irq(&mddev->write_lock);
2887 mddev->recovery_cp = MaxSector;
2888 err = do_md_run(mddev);
2893 restart_array(mddev);
2894 if (mddev->external)
2895 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2896 wake_up(&mddev->sb_wait);
2900 set_disk_ro(mddev->gendisk, 0);
2901 err = do_md_run(mddev);
2906 /* these cannot be set */
2912 sysfs_notify_dirent(mddev->sysfs_state);
2916 static struct md_sysfs_entry md_array_state =
2917 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2920 null_show(mddev_t *mddev, char *page)
2926 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2928 /* buf must be %d:%d\n? giving major and minor numbers */
2929 /* The new device is added to the array.
2930 * If the array has a persistent superblock, we read the
2931 * superblock to initialise info and check validity.
2932 * Otherwise, only checking done is that in bind_rdev_to_array,
2933 * which mainly checks size.
2936 int major = simple_strtoul(buf, &e, 10);
2942 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2944 minor = simple_strtoul(e+1, &e, 10);
2945 if (*e && *e != '\n')
2947 dev = MKDEV(major, minor);
2948 if (major != MAJOR(dev) ||
2949 minor != MINOR(dev))
2953 if (mddev->persistent) {
2954 rdev = md_import_device(dev, mddev->major_version,
2955 mddev->minor_version);
2956 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2957 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2958 mdk_rdev_t, same_set);
2959 err = super_types[mddev->major_version]
2960 .load_super(rdev, rdev0, mddev->minor_version);
2964 } else if (mddev->external)
2965 rdev = md_import_device(dev, -2, -1);
2967 rdev = md_import_device(dev, -1, -1);
2970 return PTR_ERR(rdev);
2971 err = bind_rdev_to_array(rdev, mddev);
2975 return err ? err : len;
2978 static struct md_sysfs_entry md_new_device =
2979 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2982 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2985 unsigned long chunk, end_chunk;
2989 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2991 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2992 if (buf == end) break;
2993 if (*end == '-') { /* range */
2995 end_chunk = simple_strtoul(buf, &end, 0);
2996 if (buf == end) break;
2998 if (*end && !isspace(*end)) break;
2999 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3001 while (isspace(*buf)) buf++;
3003 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3008 static struct md_sysfs_entry md_bitmap =
3009 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3012 size_show(mddev_t *mddev, char *page)
3014 return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
3017 static int update_size(mddev_t *mddev, sector_t num_sectors);
3020 size_store(mddev_t *mddev, const char *buf, size_t len)
3022 /* If array is inactive, we can reduce the component size, but
3023 * not increase it (except from 0).
3024 * If array is active, we can try an on-line resize
3028 unsigned long long size = simple_strtoull(buf, &e, 10);
3029 if (!*buf || *buf == '\n' ||
3034 err = update_size(mddev, size * 2);
3035 md_update_sb(mddev, 1);
3037 if (mddev->size == 0 ||
3043 return err ? err : len;
3046 static struct md_sysfs_entry md_size =
3047 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3052 * 'none' for arrays with no metadata (good luck...)
3053 * 'external' for arrays with externally managed metadata,
3054 * or N.M for internally known formats
3057 metadata_show(mddev_t *mddev, char *page)
3059 if (mddev->persistent)
3060 return sprintf(page, "%d.%d\n",
3061 mddev->major_version, mddev->minor_version);
3062 else if (mddev->external)
3063 return sprintf(page, "external:%s\n", mddev->metadata_type);
3065 return sprintf(page, "none\n");
3069 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3073 /* Changing the details of 'external' metadata is
3074 * always permitted. Otherwise there must be
3075 * no devices attached to the array.
3077 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3079 else if (!list_empty(&mddev->disks))
3082 if (cmd_match(buf, "none")) {
3083 mddev->persistent = 0;
3084 mddev->external = 0;
3085 mddev->major_version = 0;
3086 mddev->minor_version = 90;
3089 if (strncmp(buf, "external:", 9) == 0) {
3090 size_t namelen = len-9;
3091 if (namelen >= sizeof(mddev->metadata_type))
3092 namelen = sizeof(mddev->metadata_type)-1;
3093 strncpy(mddev->metadata_type, buf+9, namelen);
3094 mddev->metadata_type[namelen] = 0;
3095 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3096 mddev->metadata_type[--namelen] = 0;
3097 mddev->persistent = 0;
3098 mddev->external = 1;
3099 mddev->major_version = 0;
3100 mddev->minor_version = 90;
3103 major = simple_strtoul(buf, &e, 10);
3104 if (e==buf || *e != '.')
3107 minor = simple_strtoul(buf, &e, 10);
3108 if (e==buf || (*e && *e != '\n') )
3110 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3112 mddev->major_version = major;
3113 mddev->minor_version = minor;
3114 mddev->persistent = 1;
3115 mddev->external = 0;
3119 static struct md_sysfs_entry md_metadata =
3120 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3123 action_show(mddev_t *mddev, char *page)
3125 char *type = "idle";
3126 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3127 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3128 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3130 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3131 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3133 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3137 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3140 return sprintf(page, "%s\n", type);
3144 action_store(mddev_t *mddev, const char *page, size_t len)
3146 if (!mddev->pers || !mddev->pers->sync_request)
3149 if (cmd_match(page, "idle")) {
3150 if (mddev->sync_thread) {
3151 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3152 md_unregister_thread(mddev->sync_thread);
3153 mddev->sync_thread = NULL;
3154 mddev->recovery = 0;
3156 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3157 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3159 else if (cmd_match(page, "resync"))
3160 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3161 else if (cmd_match(page, "recover")) {
3162 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3163 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3164 } else if (cmd_match(page, "reshape")) {
3166 if (mddev->pers->start_reshape == NULL)
3168 err = mddev->pers->start_reshape(mddev);
3171 sysfs_notify(&mddev->kobj, NULL, "degraded");
3173 if (cmd_match(page, "check"))
3174 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3175 else if (!cmd_match(page, "repair"))
3177 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3178 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3180 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3181 md_wakeup_thread(mddev->thread);
3182 sysfs_notify_dirent(mddev->sysfs_action);
3187 mismatch_cnt_show(mddev_t *mddev, char *page)
3189 return sprintf(page, "%llu\n",
3190 (unsigned long long) mddev->resync_mismatches);
3193 static struct md_sysfs_entry md_scan_mode =
3194 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3197 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3200 sync_min_show(mddev_t *mddev, char *page)
3202 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3203 mddev->sync_speed_min ? "local": "system");
3207 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3211 if (strncmp(buf, "system", 6)==0) {
3212 mddev->sync_speed_min = 0;
3215 min = simple_strtoul(buf, &e, 10);
3216 if (buf == e || (*e && *e != '\n') || min <= 0)
3218 mddev->sync_speed_min = min;
3222 static struct md_sysfs_entry md_sync_min =
3223 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3226 sync_max_show(mddev_t *mddev, char *page)
3228 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3229 mddev->sync_speed_max ? "local": "system");
3233 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3237 if (strncmp(buf, "system", 6)==0) {
3238 mddev->sync_speed_max = 0;
3241 max = simple_strtoul(buf, &e, 10);
3242 if (buf == e || (*e && *e != '\n') || max <= 0)
3244 mddev->sync_speed_max = max;
3248 static struct md_sysfs_entry md_sync_max =
3249 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3252 degraded_show(mddev_t *mddev, char *page)
3254 return sprintf(page, "%d\n", mddev->degraded);
3256 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3259 sync_force_parallel_show(mddev_t *mddev, char *page)
3261 return sprintf(page, "%d\n", mddev->parallel_resync);
3265 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3269 if (strict_strtol(buf, 10, &n))
3272 if (n != 0 && n != 1)
3275 mddev->parallel_resync = n;
3277 if (mddev->sync_thread)
3278 wake_up(&resync_wait);
3283 /* force parallel resync, even with shared block devices */
3284 static struct md_sysfs_entry md_sync_force_parallel =
3285 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3286 sync_force_parallel_show, sync_force_parallel_store);
3289 sync_speed_show(mddev_t *mddev, char *page)
3291 unsigned long resync, dt, db;
3292 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3293 dt = (jiffies - mddev->resync_mark) / HZ;
3295 db = resync - mddev->resync_mark_cnt;
3296 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3299 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3302 sync_completed_show(mddev_t *mddev, char *page)
3304 unsigned long max_blocks, resync;
3306 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3307 max_blocks = mddev->resync_max_sectors;
3309 max_blocks = mddev->size << 1;
3311 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3312 return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3315 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3318 min_sync_show(mddev_t *mddev, char *page)
3320 return sprintf(page, "%llu\n",
3321 (unsigned long long)mddev->resync_min);
3324 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3326 unsigned long long min;
3327 if (strict_strtoull(buf, 10, &min))
3329 if (min > mddev->resync_max)
3331 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3334 /* Must be a multiple of chunk_size */
3335 if (mddev->chunk_size) {
3336 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3339 mddev->resync_min = min;
3344 static struct md_sysfs_entry md_min_sync =
3345 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3348 max_sync_show(mddev_t *mddev, char *page)
3350 if (mddev->resync_max == MaxSector)
3351 return sprintf(page, "max\n");
3353 return sprintf(page, "%llu\n",
3354 (unsigned long long)mddev->resync_max);
3357 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3359 if (strncmp(buf, "max", 3) == 0)
3360 mddev->resync_max = MaxSector;
3362 unsigned long long max;
3363 if (strict_strtoull(buf, 10, &max))
3365 if (max < mddev->resync_min)
3367 if (max < mddev->resync_max &&
3368 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3371 /* Must be a multiple of chunk_size */
3372 if (mddev->chunk_size) {
3373 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3376 mddev->resync_max = max;
3378 wake_up(&mddev->recovery_wait);
3382 static struct md_sysfs_entry md_max_sync =
3383 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3386 suspend_lo_show(mddev_t *mddev, char *page)
3388 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3392 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3395 unsigned long long new = simple_strtoull(buf, &e, 10);
3397 if (mddev->pers->quiesce == NULL)
3399 if (buf == e || (*e && *e != '\n'))
3401 if (new >= mddev->suspend_hi ||
3402 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3403 mddev->suspend_lo = new;
3404 mddev->pers->quiesce(mddev, 2);
3409 static struct md_sysfs_entry md_suspend_lo =
3410 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3414 suspend_hi_show(mddev_t *mddev, char *page)
3416 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3420 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3423 unsigned long long new = simple_strtoull(buf, &e, 10);
3425 if (mddev->pers->quiesce == NULL)
3427 if (buf == e || (*e && *e != '\n'))
3429 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3430 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3431 mddev->suspend_hi = new;
3432 mddev->pers->quiesce(mddev, 1);
3433 mddev->pers->quiesce(mddev, 0);
3438 static struct md_sysfs_entry md_suspend_hi =
3439 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3442 reshape_position_show(mddev_t *mddev, char *page)
3444 if (mddev->reshape_position != MaxSector)
3445 return sprintf(page, "%llu\n",
3446 (unsigned long long)mddev->reshape_position);
3447 strcpy(page, "none\n");
3452 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3455 unsigned long long new = simple_strtoull(buf, &e, 10);
3458 if (buf == e || (*e && *e != '\n'))
3460 mddev->reshape_position = new;
3461 mddev->delta_disks = 0;
3462 mddev->new_level = mddev->level;
3463 mddev->new_layout = mddev->layout;
3464 mddev->new_chunk = mddev->chunk_size;
3468 static struct md_sysfs_entry md_reshape_position =
3469 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3470 reshape_position_store);
3473 static struct attribute *md_default_attrs[] = {
3476 &md_raid_disks.attr,
3477 &md_chunk_size.attr,
3479 &md_resync_start.attr,
3481 &md_new_device.attr,
3482 &md_safe_delay.attr,
3483 &md_array_state.attr,
3484 &md_reshape_position.attr,
3488 static struct attribute *md_redundancy_attrs[] = {
3490 &md_mismatches.attr,
3493 &md_sync_speed.attr,
3494 &md_sync_force_parallel.attr,
3495 &md_sync_completed.attr,
3498 &md_suspend_lo.attr,
3499 &md_suspend_hi.attr,
3504 static struct attribute_group md_redundancy_group = {
3506 .attrs = md_redundancy_attrs,
3511 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3513 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3514 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3519 rv = mddev_lock(mddev);
3521 rv = entry->show(mddev, page);
3522 mddev_unlock(mddev);
3528 md_attr_store(struct kobject *kobj, struct attribute *attr,
3529 const char *page, size_t length)
3531 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3532 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3537 if (!capable(CAP_SYS_ADMIN))
3539 rv = mddev_lock(mddev);
3540 if (mddev->hold_active == UNTIL_IOCTL)
3541 mddev->hold_active = 0;
3543 rv = entry->store(mddev, page, length);
3544 mddev_unlock(mddev);
3549 static void md_free(struct kobject *ko)
3551 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3553 if (mddev->sysfs_state)
3554 sysfs_put(mddev->sysfs_state);
3556 if (mddev->gendisk) {
3557 del_gendisk(mddev->gendisk);
3558 put_disk(mddev->gendisk);
3561 blk_cleanup_queue(mddev->queue);
3566 static struct sysfs_ops md_sysfs_ops = {
3567 .show = md_attr_show,
3568 .store = md_attr_store,
3570 static struct kobj_type md_ktype = {
3572 .sysfs_ops = &md_sysfs_ops,
3573 .default_attrs = md_default_attrs,
3578 static int md_alloc(dev_t dev, char *name)
3580 static DEFINE_MUTEX(disks_mutex);
3581 mddev_t *mddev = mddev_find(dev);
3582 struct gendisk *disk;
3591 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3592 shift = partitioned ? MdpMinorShift : 0;
3593 unit = MINOR(mddev->unit) >> shift;
3595 /* wait for any previous instance if this device
3596 * to be completed removed (mddev_delayed_delete).
3598 flush_scheduled_work();
3600 mutex_lock(&disks_mutex);
3601 if (mddev->gendisk) {
3602 mutex_unlock(&disks_mutex);
3608 /* Need to ensure that 'name' is not a duplicate.
3611 spin_lock(&all_mddevs_lock);
3613 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3614 if (mddev2->gendisk &&
3615 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3616 spin_unlock(&all_mddevs_lock);
3619 spin_unlock(&all_mddevs_lock);
3622 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3623 if (!mddev->queue) {
3624 mutex_unlock(&disks_mutex);
3628 /* Can be unlocked because the queue is new: no concurrency */
3629 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3631 blk_queue_make_request(mddev->queue, md_fail_request);
3633 disk = alloc_disk(1 << shift);
3635 mutex_unlock(&disks_mutex);
3636 blk_cleanup_queue(mddev->queue);
3637 mddev->queue = NULL;
3641 disk->major = MAJOR(mddev->unit);
3642 disk->first_minor = unit << shift;
3644 strcpy(disk->disk_name, name);
3645 else if (partitioned)
3646 sprintf(disk->disk_name, "md_d%d", unit);
3648 sprintf(disk->disk_name, "md%d", unit);
3649 disk->fops = &md_fops;
3650 disk->private_data = mddev;
3651 disk->queue = mddev->queue;
3652 /* Allow extended partitions. This makes the
3653 * 'mdp' device redundant, but we can't really
3656 disk->flags |= GENHD_FL_EXT_DEVT;
3658 mddev->gendisk = disk;
3659 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3660 &disk_to_dev(disk)->kobj, "%s", "md");
3661 mutex_unlock(&disks_mutex);
3663 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3666 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3667 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3673 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3675 md_alloc(dev, NULL);
3679 static int add_named_array(const char *val, struct kernel_param *kp)
3681 /* val must be "md_*" where * is not all digits.
3682 * We allocate an array with a large free minor number, and
3683 * set the name to val. val must not already be an active name.
3685 int len = strlen(val);
3686 char buf[DISK_NAME_LEN];
3688 while (len && val[len-1] == '\n')
3690 if (len >= DISK_NAME_LEN)
3692 strlcpy(buf, val, len+1);
3693 if (strncmp(buf, "md_", 3) != 0)
3695 return md_alloc(0, buf);
3698 static void md_safemode_timeout(unsigned long data)
3700 mddev_t *mddev = (mddev_t *) data;
3702 if (!atomic_read(&mddev->writes_pending)) {
3703 mddev->safemode = 1;
3704 if (mddev->external)
3705 sysfs_notify_dirent(mddev->sysfs_state);
3707 md_wakeup_thread(mddev->thread);
3710 static int start_dirty_degraded;
3712 static int do_md_run(mddev_t * mddev)
3717 struct gendisk *disk;
3718 struct mdk_personality *pers;
3719 char b[BDEVNAME_SIZE];
3721 if (list_empty(&mddev->disks))
3722 /* cannot run an array with no devices.. */
3729 * Analyze all RAID superblock(s)
3731 if (!mddev->raid_disks) {
3732 if (!mddev->persistent)
3737 chunk_size = mddev->chunk_size;
3740 if (chunk_size > MAX_CHUNK_SIZE) {
3741 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3742 chunk_size, MAX_CHUNK_SIZE);
3746 * chunk-size has to be a power of 2
3748 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3749 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3753 /* devices must have minimum size of one chunk */
3754 list_for_each_entry(rdev, &mddev->disks, same_set) {
3755 if (test_bit(Faulty, &rdev->flags))
3757 if (rdev->size < chunk_size / 1024) {
3759 "md: Dev %s smaller than chunk_size:"
3761 bdevname(rdev->bdev,b),
3762 (unsigned long long)rdev->size,
3769 if (mddev->level != LEVEL_NONE)
3770 request_module("md-level-%d", mddev->level);
3771 else if (mddev->clevel[0])
3772 request_module("md-%s", mddev->clevel);
3775 * Drop all container device buffers, from now on
3776 * the only valid external interface is through the md
3779 list_for_each_entry(rdev, &mddev->disks, same_set) {
3780 if (test_bit(Faulty, &rdev->flags))
3782 sync_blockdev(rdev->bdev);
3783 invalidate_bdev(rdev->bdev);
3785 /* perform some consistency tests on the device.
3786 * We don't want the data to overlap the metadata,
3787 * Internal Bitmap issues has handled elsewhere.
3789 if (rdev->data_offset < rdev->sb_start) {
3791 rdev->data_offset + mddev->size*2
3793 printk("md: %s: data overlaps metadata\n",
3798 if (rdev->sb_start + rdev->sb_size/512
3799 > rdev->data_offset) {
3800 printk("md: %s: metadata overlaps data\n",
3805 sysfs_notify_dirent(rdev->sysfs_state);
3808 md_probe(mddev->unit, NULL, NULL);
3809 disk = mddev->gendisk;
3813 spin_lock(&pers_lock);
3814 pers = find_pers(mddev->level, mddev->clevel);
3815 if (!pers || !try_module_get(pers->owner)) {
3816 spin_unlock(&pers_lock);
3817 if (mddev->level != LEVEL_NONE)
3818 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3821 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3826 spin_unlock(&pers_lock);
3827 mddev->level = pers->level;
3828 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3830 if (pers->level >= 4 && pers->level <= 6)
3831 /* Cannot support integrity (yet) */
3832 blk_integrity_unregister(mddev->gendisk);
3834 if (mddev->reshape_position != MaxSector &&
3835 pers->start_reshape == NULL) {
3836 /* This personality cannot handle reshaping... */
3838 module_put(pers->owner);
3842 if (pers->sync_request) {
3843 /* Warn if this is a potentially silly
3846 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3850 list_for_each_entry(rdev, &mddev->disks, same_set)
3851 list_for_each_entry(rdev2, &mddev->disks, same_set) {
3853 rdev->bdev->bd_contains ==
3854 rdev2->bdev->bd_contains) {
3856 "%s: WARNING: %s appears to be"
3857 " on the same physical disk as"
3860 bdevname(rdev->bdev,b),
3861 bdevname(rdev2->bdev,b2));
3868 "True protection against single-disk"
3869 " failure might be compromised.\n");
3872 mddev->recovery = 0;
3873 mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3874 mddev->barriers_work = 1;
3875 mddev->ok_start_degraded = start_dirty_degraded;
3878 mddev->ro = 2; /* read-only, but switch on first write */
3880 err = mddev->pers->run(mddev);
3882 printk(KERN_ERR "md: pers->run() failed ...\n");
3883 else if (mddev->pers->sync_request) {
3884 err = bitmap_create(mddev);
3886 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3887 mdname(mddev), err);
3888 mddev->pers->stop(mddev);
3892 module_put(mddev->pers->owner);
3894 bitmap_destroy(mddev);
3897 if (mddev->pers->sync_request) {
3898 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3900 "md: cannot register extra attributes for %s\n",
3902 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3903 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3906 atomic_set(&mddev->writes_pending,0);
3907 mddev->safemode = 0;
3908 mddev->safemode_timer.function = md_safemode_timeout;
3909 mddev->safemode_timer.data = (unsigned long) mddev;
3910 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3913 list_for_each_entry(rdev, &mddev->disks, same_set)
3914 if (rdev->raid_disk >= 0) {
3916 sprintf(nm, "rd%d", rdev->raid_disk);
3917 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3918 printk("md: cannot register %s for %s\n",
3922 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3925 md_update_sb(mddev, 0);
3927 set_capacity(disk, mddev->array_sectors);
3929 /* If we call blk_queue_make_request here, it will
3930 * re-initialise max_sectors etc which may have been
3931 * refined inside -> run. So just set the bits we need to set.
3932 * Most initialisation happended when we called
3933 * blk_queue_make_request(..., md_fail_request)
3936 mddev->queue->queuedata = mddev;
3937 mddev->queue->make_request_fn = mddev->pers->make_request;
3939 /* If there is a partially-recovered drive we need to
3940 * start recovery here. If we leave it to md_check_recovery,
3941 * it will remove the drives and not do the right thing
3943 if (mddev->degraded && !mddev->sync_thread) {
3945 list_for_each_entry(rdev, &mddev->disks, same_set)
3946 if (rdev->raid_disk >= 0 &&
3947 !test_bit(In_sync, &rdev->flags) &&
3948 !test_bit(Faulty, &rdev->flags))
3949 /* complete an interrupted recovery */
3951 if (spares && mddev->pers->sync_request) {
3952 mddev->recovery = 0;
3953 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3954 mddev->sync_thread = md_register_thread(md_do_sync,
3957 if (!mddev->sync_thread) {
3958 printk(KERN_ERR "%s: could not start resync"
3961 /* leave the spares where they are, it shouldn't hurt */
3962 mddev->recovery = 0;
3966 md_wakeup_thread(mddev->thread);
3967 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3970 md_new_event(mddev);
3971 sysfs_notify_dirent(mddev->sysfs_state);
3972 if (mddev->sysfs_action)
3973 sysfs_notify_dirent(mddev->sysfs_action);
3974 sysfs_notify(&mddev->kobj, NULL, "degraded");
3975 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3979 static int restart_array(mddev_t *mddev)
3981 struct gendisk *disk = mddev->gendisk;
3983 /* Complain if it has no devices */
3984 if (list_empty(&mddev->disks))
3990 mddev->safemode = 0;
3992 set_disk_ro(disk, 0);
3993 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3995 /* Kick recovery or resync if necessary */
3996 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3997 md_wakeup_thread(mddev->thread);
3998 md_wakeup_thread(mddev->sync_thread);
3999 sysfs_notify_dirent(mddev->sysfs_state);
4003 /* similar to deny_write_access, but accounts for our holding a reference
4004 * to the file ourselves */
4005 static int deny_bitmap_write_access(struct file * file)
4007 struct inode *inode = file->f_mapping->host;
4009 spin_lock(&inode->i_lock);
4010 if (atomic_read(&inode->i_writecount) > 1) {
4011 spin_unlock(&inode->i_lock);
4014 atomic_set(&inode->i_writecount, -1);
4015 spin_unlock(&inode->i_lock);
4020 static void restore_bitmap_write_access(struct file *file)
4022 struct inode *inode = file->f_mapping->host;
4024 spin_lock(&inode->i_lock);
4025 atomic_set(&inode->i_writecount, 1);
4026 spin_unlock(&inode->i_lock);
4030 * 0 - completely stop and dis-assemble array
4031 * 1 - switch to readonly
4032 * 2 - stop but do not disassemble array
4034 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4037 struct gendisk *disk = mddev->gendisk;
4039 if (atomic_read(&mddev->openers) > is_open) {
4040 printk("md: %s still in use.\n",mdname(mddev));
4046 if (mddev->sync_thread) {
4047 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4048 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4049 md_unregister_thread(mddev->sync_thread);
4050 mddev->sync_thread = NULL;
4053 del_timer_sync(&mddev->safemode_timer);
4056 case 1: /* readonly */
4062 case 0: /* disassemble */
4064 bitmap_flush(mddev);
4065 md_super_wait(mddev);
4067 set_disk_ro(disk, 0);
4068 blk_queue_make_request(mddev->queue, md_fail_request);
4069 mddev->pers->stop(mddev);
4070 mddev->queue->merge_bvec_fn = NULL;
4071 mddev->queue->unplug_fn = NULL;
4072 mddev->queue->backing_dev_info.congested_fn = NULL;
4073 if (mddev->pers->sync_request) {
4074 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4075 if (mddev->sysfs_action)
4076 sysfs_put(mddev->sysfs_action);
4077 mddev->sysfs_action = NULL;
4079 module_put(mddev->pers->owner);
4081 /* tell userspace to handle 'inactive' */
4082 sysfs_notify_dirent(mddev->sysfs_state);
4084 set_capacity(disk, 0);
4090 if (!mddev->in_sync || mddev->flags) {
4091 /* mark array as shutdown cleanly */
4093 md_update_sb(mddev, 1);
4096 set_disk_ro(disk, 1);
4097 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4101 * Free resources if final stop
4106 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4108 bitmap_destroy(mddev);
4109 if (mddev->bitmap_file) {
4110 restore_bitmap_write_access(mddev->bitmap_file);
4111 fput(mddev->bitmap_file);
4112 mddev->bitmap_file = NULL;
4114 mddev->bitmap_offset = 0;
4116 list_for_each_entry(rdev, &mddev->disks, same_set)
4117 if (rdev->raid_disk >= 0) {
4119 sprintf(nm, "rd%d", rdev->raid_disk);
4120 sysfs_remove_link(&mddev->kobj, nm);
4123 /* make sure all md_delayed_delete calls have finished */
4124 flush_scheduled_work();
4126 export_array(mddev);
4128 mddev->array_sectors = 0;
4130 mddev->raid_disks = 0;
4131 mddev->recovery_cp = 0;
4132 mddev->resync_min = 0;
4133 mddev->resync_max = MaxSector;
4134 mddev->reshape_position = MaxSector;
4135 mddev->external = 0;
4136 mddev->persistent = 0;
4137 mddev->level = LEVEL_NONE;
4138 mddev->clevel[0] = 0;
4141 mddev->metadata_type[0] = 0;
4142 mddev->chunk_size = 0;
4143 mddev->ctime = mddev->utime = 0;
4145 mddev->max_disks = 0;
4147 mddev->delta_disks = 0;
4148 mddev->new_level = LEVEL_NONE;
4149 mddev->new_layout = 0;
4150 mddev->new_chunk = 0;
4151 mddev->curr_resync = 0;
4152 mddev->resync_mismatches = 0;
4153 mddev->suspend_lo = mddev->suspend_hi = 0;
4154 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4155 mddev->recovery = 0;
4158 mddev->degraded = 0;
4159 mddev->barriers_work = 0;
4160 mddev->safemode = 0;
4161 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4162 if (mddev->hold_active == UNTIL_STOP)
4163 mddev->hold_active = 0;
4165 } else if (mddev->pers)
4166 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4169 blk_integrity_unregister(disk);
4170 md_new_event(mddev);
4171 sysfs_notify_dirent(mddev->sysfs_state);
4177 static void autorun_array(mddev_t *mddev)
4182 if (list_empty(&mddev->disks))
4185 printk(KERN_INFO "md: running: ");
4187 list_for_each_entry(rdev, &mddev->disks, same_set) {
4188 char b[BDEVNAME_SIZE];
4189 printk("<%s>", bdevname(rdev->bdev,b));
4193 err = do_md_run(mddev);
4195 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4196 do_md_stop(mddev, 0, 0);
4201 * lets try to run arrays based on all disks that have arrived
4202 * until now. (those are in pending_raid_disks)
4204 * the method: pick the first pending disk, collect all disks with
4205 * the same UUID, remove all from the pending list and put them into
4206 * the 'same_array' list. Then order this list based on superblock
4207 * update time (freshest comes first), kick out 'old' disks and
4208 * compare superblocks. If everything's fine then run it.
4210 * If "unit" is allocated, then bump its reference count
4212 static void autorun_devices(int part)
4214 mdk_rdev_t *rdev0, *rdev, *tmp;
4216 char b[BDEVNAME_SIZE];
4218 printk(KERN_INFO "md: autorun ...\n");
4219 while (!list_empty(&pending_raid_disks)) {
4222 LIST_HEAD(candidates);
4223 rdev0 = list_entry(pending_raid_disks.next,
4224 mdk_rdev_t, same_set);
4226 printk(KERN_INFO "md: considering %s ...\n",
4227 bdevname(rdev0->bdev,b));
4228 INIT_LIST_HEAD(&candidates);
4229 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4230 if (super_90_load(rdev, rdev0, 0) >= 0) {
4231 printk(KERN_INFO "md: adding %s ...\n",
4232 bdevname(rdev->bdev,b));
4233 list_move(&rdev->same_set, &candidates);
4236 * now we have a set of devices, with all of them having
4237 * mostly sane superblocks. It's time to allocate the
4241 dev = MKDEV(mdp_major,
4242 rdev0->preferred_minor << MdpMinorShift);
4243 unit = MINOR(dev) >> MdpMinorShift;
4245 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4248 if (rdev0->preferred_minor != unit) {
4249 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4250 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4254 md_probe(dev, NULL, NULL);
4255 mddev = mddev_find(dev);
4256 if (!mddev || !mddev->gendisk) {
4260 "md: cannot allocate memory for md drive.\n");
4263 if (mddev_lock(mddev))
4264 printk(KERN_WARNING "md: %s locked, cannot run\n",
4266 else if (mddev->raid_disks || mddev->major_version
4267 || !list_empty(&mddev->disks)) {
4269 "md: %s already running, cannot run %s\n",
4270 mdname(mddev), bdevname(rdev0->bdev,b));
4271 mddev_unlock(mddev);
4273 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4274 mddev->persistent = 1;
4275 rdev_for_each_list(rdev, tmp, &candidates) {
4276 list_del_init(&rdev->same_set);
4277 if (bind_rdev_to_array(rdev, mddev))
4280 autorun_array(mddev);
4281 mddev_unlock(mddev);
4283 /* on success, candidates will be empty, on error
4286 rdev_for_each_list(rdev, tmp, &candidates) {
4287 list_del_init(&rdev->same_set);
4292 printk(KERN_INFO "md: ... autorun DONE.\n");
4294 #endif /* !MODULE */
4296 static int get_version(void __user * arg)
4300 ver.major = MD_MAJOR_VERSION;
4301 ver.minor = MD_MINOR_VERSION;
4302 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4304 if (copy_to_user(arg, &ver, sizeof(ver)))
4310 static int get_array_info(mddev_t * mddev, void __user * arg)
4312 mdu_array_info_t info;
4313 int nr,working,active,failed,spare;
4316 nr=working=active=failed=spare=0;
4317 list_for_each_entry(rdev, &mddev->disks, same_set) {
4319 if (test_bit(Faulty, &rdev->flags))
4323 if (test_bit(In_sync, &rdev->flags))
4330 info.major_version = mddev->major_version;
4331 info.minor_version = mddev->minor_version;
4332 info.patch_version = MD_PATCHLEVEL_VERSION;
4333 info.ctime = mddev->ctime;
4334 info.level = mddev->level;
4335 info.size = mddev->size;
4336 if (info.size != mddev->size) /* overflow */
4339 info.raid_disks = mddev->raid_disks;
4340 info.md_minor = mddev->md_minor;
4341 info.not_persistent= !mddev->persistent;
4343 info.utime = mddev->utime;
4346 info.state = (1<<MD_SB_CLEAN);
4347 if (mddev->bitmap && mddev->bitmap_offset)
4348 info.state = (1<<MD_SB_BITMAP_PRESENT);
4349 info.active_disks = active;
4350 info.working_disks = working;
4351 info.failed_disks = failed;
4352 info.spare_disks = spare;
4354 info.layout = mddev->layout;
4355 info.chunk_size = mddev->chunk_size;
4357 if (copy_to_user(arg, &info, sizeof(info)))
4363 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4365 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4366 char *ptr, *buf = NULL;
4369 if (md_allow_write(mddev))
4370 file = kmalloc(sizeof(*file), GFP_NOIO);
4372 file = kmalloc(sizeof(*file), GFP_KERNEL);
4377 /* bitmap disabled, zero the first byte and copy out */
4378 if (!mddev->bitmap || !mddev->bitmap->file) {
4379 file->pathname[0] = '\0';
4383 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4387 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4391 strcpy(file->pathname, ptr);
4395 if (copy_to_user(arg, file, sizeof(*file)))
4403 static int get_disk_info(mddev_t * mddev, void __user * arg)
4405 mdu_disk_info_t info;
4408 if (copy_from_user(&info, arg, sizeof(info)))
4411 rdev = find_rdev_nr(mddev, info.number);
4413 info.major = MAJOR(rdev->bdev->bd_dev);
4414 info.minor = MINOR(rdev->bdev->bd_dev);
4415 info.raid_disk = rdev->raid_disk;
4417 if (test_bit(Faulty, &rdev->flags))
4418 info.state |= (1<<MD_DISK_FAULTY);
4419 else if (test_bit(In_sync, &rdev->flags)) {
4420 info.state |= (1<<MD_DISK_ACTIVE);
4421 info.state |= (1<<MD_DISK_SYNC);
4423 if (test_bit(WriteMostly, &rdev->flags))
4424 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4426 info.major = info.minor = 0;
4427 info.raid_disk = -1;
4428 info.state = (1<<MD_DISK_REMOVED);
4431 if (copy_to_user(arg, &info, sizeof(info)))
4437 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4439 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4441 dev_t dev = MKDEV(info->major,info->minor);
4443 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4446 if (!mddev->raid_disks) {
4448 /* expecting a device which has a superblock */
4449 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4452 "md: md_import_device returned %ld\n",
4454 return PTR_ERR(rdev);
4456 if (!list_empty(&mddev->disks)) {
4457 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4458 mdk_rdev_t, same_set);
4459 int err = super_types[mddev->major_version]
4460 .load_super(rdev, rdev0, mddev->minor_version);
4463 "md: %s has different UUID to %s\n",
4464 bdevname(rdev->bdev,b),
4465 bdevname(rdev0->bdev,b2));
4470 err = bind_rdev_to_array(rdev, mddev);
4477 * add_new_disk can be used once the array is assembled
4478 * to add "hot spares". They must already have a superblock
4483 if (!mddev->pers->hot_add_disk) {
4485 "%s: personality does not support diskops!\n",
4489 if (mddev->persistent)
4490 rdev = md_import_device(dev, mddev->major_version,
4491 mddev->minor_version);
4493 rdev = md_import_device(dev, -1, -1);
4496 "md: md_import_device returned %ld\n",
4498 return PTR_ERR(rdev);
4500 /* set save_raid_disk if appropriate */
4501 if (!mddev->persistent) {
4502 if (info->state & (1<<MD_DISK_SYNC) &&
4503 info->raid_disk < mddev->raid_disks)
4504 rdev->raid_disk = info->raid_disk;
4506 rdev->raid_disk = -1;
4508 super_types[mddev->major_version].
4509 validate_super(mddev, rdev);
4510 rdev->saved_raid_disk = rdev->raid_disk;
4512 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4513 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4514 set_bit(WriteMostly, &rdev->flags);
4516 rdev->raid_disk = -1;
4517 err = bind_rdev_to_array(rdev, mddev);
4518 if (!err && !mddev->pers->hot_remove_disk) {
4519 /* If there is hot_add_disk but no hot_remove_disk
4520 * then added disks for geometry changes,
4521 * and should be added immediately.
4523 super_types[mddev->major_version].
4524 validate_super(mddev, rdev);
4525 err = mddev->pers->hot_add_disk(mddev, rdev);
4527 unbind_rdev_from_array(rdev);
4532 sysfs_notify_dirent(rdev->sysfs_state);
4534 md_update_sb(mddev, 1);
4535 if (mddev->degraded)
4536 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4537 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4538 md_wakeup_thread(mddev->thread);
4542 /* otherwise, add_new_disk is only allowed
4543 * for major_version==0 superblocks
4545 if (mddev->major_version != 0) {
4546 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4551 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4553 rdev = md_import_device(dev, -1, 0);
4556 "md: error, md_import_device() returned %ld\n",
4558 return PTR_ERR(rdev);
4560 rdev->desc_nr = info->number;
4561 if (info->raid_disk < mddev->raid_disks)
4562 rdev->raid_disk = info->raid_disk;
4564 rdev->raid_disk = -1;
4566 if (rdev->raid_disk < mddev->raid_disks)
4567 if (info->state & (1<<MD_DISK_SYNC))
4568 set_bit(In_sync, &rdev->flags);
4570 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4571 set_bit(WriteMostly, &rdev->flags);
4573 if (!mddev->persistent) {
4574 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4575 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4577 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4578 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4580 err = bind_rdev_to_array(rdev, mddev);
4590 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4592 char b[BDEVNAME_SIZE];
4595 rdev = find_rdev(mddev, dev);
4599 if (rdev->raid_disk >= 0)
4602 kick_rdev_from_array(rdev);
4603 md_update_sb(mddev, 1);
4604 md_new_event(mddev);
4608 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4609 bdevname(rdev->bdev,b), mdname(mddev));
4613 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4615 char b[BDEVNAME_SIZE];
4622 if (mddev->major_version != 0) {
4623 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4624 " version-0 superblocks.\n",
4628 if (!mddev->pers->hot_add_disk) {
4630 "%s: personality does not support diskops!\n",
4635 rdev = md_import_device(dev, -1, 0);
4638 "md: error, md_import_device() returned %ld\n",
4643 if (mddev->persistent)
4644 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4646 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4648 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4650 if (test_bit(Faulty, &rdev->flags)) {
4652 "md: can not hot-add faulty %s disk to %s!\n",
4653 bdevname(rdev->bdev,b), mdname(mddev));
4657 clear_bit(In_sync, &rdev->flags);
4659 rdev->saved_raid_disk = -1;
4660 err = bind_rdev_to_array(rdev, mddev);
4665 * The rest should better be atomic, we can have disk failures
4666 * noticed in interrupt contexts ...
4669 rdev->raid_disk = -1;
4671 md_update_sb(mddev, 1);
4674 * Kick recovery, maybe this spare has to be added to the
4675 * array immediately.
4677 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4678 md_wakeup_thread(mddev->thread);
4679 md_new_event(mddev);
4687 static int set_bitmap_file(mddev_t *mddev, int fd)
4692 if (!mddev->pers->quiesce)
4694 if (mddev->recovery || mddev->sync_thread)
4696 /* we should be able to change the bitmap.. */
4702 return -EEXIST; /* cannot add when bitmap is present */
4703 mddev->bitmap_file = fget(fd);
4705 if (mddev->bitmap_file == NULL) {
4706 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4711 err = deny_bitmap_write_access(mddev->bitmap_file);
4713 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4715 fput(mddev->bitmap_file);
4716 mddev->bitmap_file = NULL;
4719 mddev->bitmap_offset = 0; /* file overrides offset */
4720 } else if (mddev->bitmap == NULL)
4721 return -ENOENT; /* cannot remove what isn't there */
4724 mddev->pers->quiesce(mddev, 1);
4726 err = bitmap_create(mddev);
4727 if (fd < 0 || err) {
4728 bitmap_destroy(mddev);
4729 fd = -1; /* make sure to put the file */
4731 mddev->pers->quiesce(mddev, 0);
4734 if (mddev->bitmap_file) {
4735 restore_bitmap_write_access(mddev->bitmap_file);
4736 fput(mddev->bitmap_file);
4738 mddev->bitmap_file = NULL;
4745 * set_array_info is used two different ways
4746 * The original usage is when creating a new array.
4747 * In this usage, raid_disks is > 0 and it together with
4748 * level, size, not_persistent,layout,chunksize determine the
4749 * shape of the array.
4750 * This will always create an array with a type-0.90.0 superblock.
4751 * The newer usage is when assembling an array.
4752 * In this case raid_disks will be 0, and the major_version field is
4753 * use to determine which style super-blocks are to be found on the devices.
4754 * The minor and patch _version numbers are also kept incase the
4755 * super_block handler wishes to interpret them.
4757 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4760 if (info->raid_disks == 0) {
4761 /* just setting version number for superblock loading */
4762 if (info->major_version < 0 ||
4763 info->major_version >= ARRAY_SIZE(super_types) ||
4764 super_types[info->major_version].name == NULL) {
4765 /* maybe try to auto-load a module? */
4767 "md: superblock version %d not known\n",
4768 info->major_version);
4771 mddev->major_version = info->major_version;
4772 mddev->minor_version = info->minor_version;
4773 mddev->patch_version = info->patch_version;
4774 mddev->persistent = !info->not_persistent;
4777 mddev->major_version = MD_MAJOR_VERSION;
4778 mddev->minor_version = MD_MINOR_VERSION;
4779 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4780 mddev->ctime = get_seconds();
4782 mddev->level = info->level;
4783 mddev->clevel[0] = 0;
4784 mddev->size = info->size;
4785 mddev->raid_disks = info->raid_disks;
4786 /* don't set md_minor, it is determined by which /dev/md* was
4789 if (info->state & (1<<MD_SB_CLEAN))
4790 mddev->recovery_cp = MaxSector;
4792 mddev->recovery_cp = 0;
4793 mddev->persistent = ! info->not_persistent;
4794 mddev->external = 0;
4796 mddev->layout = info->layout;
4797 mddev->chunk_size = info->chunk_size;
4799 mddev->max_disks = MD_SB_DISKS;
4801 if (mddev->persistent)
4803 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4805 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4806 mddev->bitmap_offset = 0;
4808 mddev->reshape_position = MaxSector;
4811 * Generate a 128 bit UUID
4813 get_random_bytes(mddev->uuid, 16);
4815 mddev->new_level = mddev->level;
4816 mddev->new_chunk = mddev->chunk_size;
4817 mddev->new_layout = mddev->layout;
4818 mddev->delta_disks = 0;
4823 static int update_size(mddev_t *mddev, sector_t num_sectors)
4827 int fit = (num_sectors == 0);
4829 if (mddev->pers->resize == NULL)
4831 /* The "num_sectors" is the number of sectors of each device that
4832 * is used. This can only make sense for arrays with redundancy.
4833 * linear and raid0 always use whatever space is available. We can only
4834 * consider changing this number if no resync or reconstruction is
4835 * happening, and if the new size is acceptable. It must fit before the
4836 * sb_start or, if that is <data_offset, it must fit before the size
4837 * of each device. If num_sectors is zero, we find the largest size
4841 if (mddev->sync_thread)
4844 /* Sorry, cannot grow a bitmap yet, just remove it,
4848 list_for_each_entry(rdev, &mddev->disks, same_set) {
4850 avail = rdev->size * 2;
4852 if (fit && (num_sectors == 0 || num_sectors > avail))
4853 num_sectors = avail;
4854 if (avail < num_sectors)
4857 rv = mddev->pers->resize(mddev, num_sectors);
4859 struct block_device *bdev;
4861 bdev = bdget_disk(mddev->gendisk, 0);
4863 mutex_lock(&bdev->bd_inode->i_mutex);
4864 i_size_write(bdev->bd_inode,
4865 (loff_t)mddev->array_sectors << 9);
4866 mutex_unlock(&bdev->bd_inode->i_mutex);
4873 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4876 /* change the number of raid disks */
4877 if (mddev->pers->check_reshape == NULL)
4879 if (raid_disks <= 0 ||
4880 raid_disks >= mddev->max_disks)
4882 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4884 mddev->delta_disks = raid_disks - mddev->raid_disks;
4886 rv = mddev->pers->check_reshape(mddev);
4892 * update_array_info is used to change the configuration of an
4894 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4895 * fields in the info are checked against the array.
4896 * Any differences that cannot be handled will cause an error.
4897 * Normally, only one change can be managed at a time.
4899 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4905 /* calculate expected state,ignoring low bits */
4906 if (mddev->bitmap && mddev->bitmap_offset)
4907 state |= (1 << MD_SB_BITMAP_PRESENT);
4909 if (mddev->major_version != info->major_version ||
4910 mddev->minor_version != info->minor_version ||
4911 /* mddev->patch_version != info->patch_version || */
4912 mddev->ctime != info->ctime ||
4913 mddev->level != info->level ||
4914 /* mddev->layout != info->layout || */
4915 !mddev->persistent != info->not_persistent||
4916 mddev->chunk_size != info->chunk_size ||
4917 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4918 ((state^info->state) & 0xfffffe00)
4921 /* Check there is only one change */
4922 if (info->size >= 0 && mddev->size != info->size) cnt++;
4923 if (mddev->raid_disks != info->raid_disks) cnt++;
4924 if (mddev->layout != info->layout) cnt++;
4925 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4926 if (cnt == 0) return 0;
4927 if (cnt > 1) return -EINVAL;
4929 if (mddev->layout != info->layout) {
4931 * we don't need to do anything at the md level, the
4932 * personality will take care of it all.
4934 if (mddev->pers->reconfig == NULL)
4937 return mddev->pers->reconfig(mddev, info->layout, -1);
4939 if (info->size >= 0 && mddev->size != info->size)
4940 rv = update_size(mddev, (sector_t)info->size * 2);
4942 if (mddev->raid_disks != info->raid_disks)
4943 rv = update_raid_disks(mddev, info->raid_disks);
4945 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4946 if (mddev->pers->quiesce == NULL)
4948 if (mddev->recovery || mddev->sync_thread)
4950 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4951 /* add the bitmap */
4954 if (mddev->default_bitmap_offset == 0)
4956 mddev->bitmap_offset = mddev->default_bitmap_offset;
4957 mddev->pers->quiesce(mddev, 1);
4958 rv = bitmap_create(mddev);
4960 bitmap_destroy(mddev);
4961 mddev->pers->quiesce(mddev, 0);
4963 /* remove the bitmap */
4966 if (mddev->bitmap->file)
4968 mddev->pers->quiesce(mddev, 1);
4969 bitmap_destroy(mddev);
4970 mddev->pers->quiesce(mddev, 0);
4971 mddev->bitmap_offset = 0;
4974 md_update_sb(mddev, 1);
4978 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4982 if (mddev->pers == NULL)
4985 rdev = find_rdev(mddev, dev);
4989 md_error(mddev, rdev);
4994 * We have a problem here : there is no easy way to give a CHS
4995 * virtual geometry. We currently pretend that we have a 2 heads
4996 * 4 sectors (with a BIG number of cylinders...). This drives
4997 * dosfs just mad... ;-)
4999 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5001 mddev_t *mddev = bdev->bd_disk->private_data;
5005 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5009 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5010 unsigned int cmd, unsigned long arg)
5013 void __user *argp = (void __user *)arg;
5014 mddev_t *mddev = NULL;
5016 if (!capable(CAP_SYS_ADMIN))
5020 * Commands dealing with the RAID driver but not any
5026 err = get_version(argp);
5029 case PRINT_RAID_DEBUG:
5037 autostart_arrays(arg);
5044 * Commands creating/starting a new array:
5047 mddev = bdev->bd_disk->private_data;
5054 err = mddev_lock(mddev);
5057 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5064 case SET_ARRAY_INFO:
5066 mdu_array_info_t info;
5068 memset(&info, 0, sizeof(info));
5069 else if (copy_from_user(&info, argp, sizeof(info))) {
5074 err = update_array_info(mddev, &info);
5076 printk(KERN_WARNING "md: couldn't update"
5077 " array info. %d\n", err);
5082 if (!list_empty(&mddev->disks)) {
5084 "md: array %s already has disks!\n",
5089 if (mddev->raid_disks) {
5091 "md: array %s already initialised!\n",
5096 err = set_array_info(mddev, &info);
5098 printk(KERN_WARNING "md: couldn't set"
5099 " array info. %d\n", err);
5109 * Commands querying/configuring an existing array:
5111 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5112 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5113 if ((!mddev->raid_disks && !mddev->external)
5114 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5115 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5116 && cmd != GET_BITMAP_FILE) {
5122 * Commands even a read-only array can execute:
5126 case GET_ARRAY_INFO:
5127 err = get_array_info(mddev, argp);
5130 case GET_BITMAP_FILE:
5131 err = get_bitmap_file(mddev, argp);
5135 err = get_disk_info(mddev, argp);
5138 case RESTART_ARRAY_RW:
5139 err = restart_array(mddev);
5143 err = do_md_stop(mddev, 0, 1);
5147 err = do_md_stop(mddev, 1, 1);
5153 * The remaining ioctls are changing the state of the
5154 * superblock, so we do not allow them on read-only arrays.
5155 * However non-MD ioctls (e.g. get-size) will still come through
5156 * here and hit the 'default' below, so only disallow
5157 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5159 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5160 if (mddev->ro == 2) {
5162 sysfs_notify_dirent(mddev->sysfs_state);
5163 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5164 md_wakeup_thread(mddev->thread);
5175 mdu_disk_info_t info;
5176 if (copy_from_user(&info, argp, sizeof(info)))
5179 err = add_new_disk(mddev, &info);
5183 case HOT_REMOVE_DISK:
5184 err = hot_remove_disk(mddev, new_decode_dev(arg));
5188 err = hot_add_disk(mddev, new_decode_dev(arg));
5191 case SET_DISK_FAULTY:
5192 err = set_disk_faulty(mddev, new_decode_dev(arg));
5196 err = do_md_run(mddev);
5199 case SET_BITMAP_FILE:
5200 err = set_bitmap_file(mddev, (int)arg);
5210 if (mddev->hold_active == UNTIL_IOCTL &&
5212 mddev->hold_active = 0;
5213 mddev_unlock(mddev);
5223 static int md_open(struct block_device *bdev, fmode_t mode)
5226 * Succeed if we can lock the mddev, which confirms that
5227 * it isn't being stopped right now.
5229 mddev_t *mddev = mddev_find(bdev->bd_dev);
5232 if (mddev->gendisk != bdev->bd_disk) {
5233 /* we are racing with mddev_put which is discarding this
5237 /* Wait until bdev->bd_disk is definitely gone */
5238 flush_scheduled_work();
5239 /* Then retry the open from the top */
5240 return -ERESTARTSYS;
5242 BUG_ON(mddev != bdev->bd_disk->private_data);
5244 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5248 atomic_inc(&mddev->openers);
5249 mddev_unlock(mddev);
5251 check_disk_change(bdev);
5256 static int md_release(struct gendisk *disk, fmode_t mode)
5258 mddev_t *mddev = disk->private_data;
5261 atomic_dec(&mddev->openers);
5267 static int md_media_changed(struct gendisk *disk)
5269 mddev_t *mddev = disk->private_data;
5271 return mddev->changed;
5274 static int md_revalidate(struct gendisk *disk)
5276 mddev_t *mddev = disk->private_data;
5281 static struct block_device_operations md_fops =
5283 .owner = THIS_MODULE,
5285 .release = md_release,
5286 .locked_ioctl = md_ioctl,
5287 .getgeo = md_getgeo,
5288 .media_changed = md_media_changed,
5289 .revalidate_disk= md_revalidate,
5292 static int md_thread(void * arg)
5294 mdk_thread_t *thread = arg;
5297 * md_thread is a 'system-thread', it's priority should be very
5298 * high. We avoid resource deadlocks individually in each
5299 * raid personality. (RAID5 does preallocation) We also use RR and
5300 * the very same RT priority as kswapd, thus we will never get
5301 * into a priority inversion deadlock.
5303 * we definitely have to have equal or higher priority than
5304 * bdflush, otherwise bdflush will deadlock if there are too
5305 * many dirty RAID5 blocks.
5308 allow_signal(SIGKILL);
5309 while (!kthread_should_stop()) {
5311 /* We need to wait INTERRUPTIBLE so that
5312 * we don't add to the load-average.
5313 * That means we need to be sure no signals are
5316 if (signal_pending(current))
5317 flush_signals(current);
5319 wait_event_interruptible_timeout
5321 test_bit(THREAD_WAKEUP, &thread->flags)
5322 || kthread_should_stop(),
5325 clear_bit(THREAD_WAKEUP, &thread->flags);
5327 thread->run(thread->mddev);
5333 void md_wakeup_thread(mdk_thread_t *thread)
5336 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5337 set_bit(THREAD_WAKEUP, &thread->flags);
5338 wake_up(&thread->wqueue);
5342 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5345 mdk_thread_t *thread;
5347 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5351 init_waitqueue_head(&thread->wqueue);
5354 thread->mddev = mddev;
5355 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5356 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5357 if (IS_ERR(thread->tsk)) {
5364 void md_unregister_thread(mdk_thread_t *thread)
5366 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5368 kthread_stop(thread->tsk);
5372 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5379 if (!rdev || test_bit(Faulty, &rdev->flags))
5382 if (mddev->external)
5383 set_bit(Blocked, &rdev->flags);
5385 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5387 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5388 __builtin_return_address(0),__builtin_return_address(1),
5389 __builtin_return_address(2),__builtin_return_address(3));
5393 if (!mddev->pers->error_handler)
5395 mddev->pers->error_handler(mddev,rdev);
5396 if (mddev->degraded)
5397 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5398 set_bit(StateChanged, &rdev->flags);
5399 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5400 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5401 md_wakeup_thread(mddev->thread);
5402 md_new_event_inintr(mddev);
5405 /* seq_file implementation /proc/mdstat */
5407 static void status_unused(struct seq_file *seq)
5412 seq_printf(seq, "unused devices: ");
5414 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5415 char b[BDEVNAME_SIZE];
5417 seq_printf(seq, "%s ",
5418 bdevname(rdev->bdev,b));
5421 seq_printf(seq, "<none>");
5423 seq_printf(seq, "\n");
5427 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5429 sector_t max_blocks, resync, res;
5430 unsigned long dt, db, rt;
5432 unsigned int per_milli;
5434 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5436 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5437 max_blocks = mddev->resync_max_sectors >> 1;
5439 max_blocks = mddev->size;
5442 * Should not happen.
5448 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5449 * in a sector_t, and (max_blocks>>scale) will fit in a
5450 * u32, as those are the requirements for sector_div.
5451 * Thus 'scale' must be at least 10
5454 if (sizeof(sector_t) > sizeof(unsigned long)) {
5455 while ( max_blocks/2 > (1ULL<<(scale+32)))
5458 res = (resync>>scale)*1000;
5459 sector_div(res, (u32)((max_blocks>>scale)+1));
5463 int i, x = per_milli/50, y = 20-x;
5464 seq_printf(seq, "[");
5465 for (i = 0; i < x; i++)
5466 seq_printf(seq, "=");
5467 seq_printf(seq, ">");
5468 for (i = 0; i < y; i++)
5469 seq_printf(seq, ".");
5470 seq_printf(seq, "] ");
5472 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5473 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5475 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5477 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5478 "resync" : "recovery"))),
5479 per_milli/10, per_milli % 10,
5480 (unsigned long long) resync,
5481 (unsigned long long) max_blocks);
5484 * We do not want to overflow, so the order of operands and
5485 * the * 100 / 100 trick are important. We do a +1 to be
5486 * safe against division by zero. We only estimate anyway.
5488 * dt: time from mark until now
5489 * db: blocks written from mark until now
5490 * rt: remaining time
5492 dt = ((jiffies - mddev->resync_mark) / HZ);
5494 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5495 - mddev->resync_mark_cnt;
5496 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5498 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5500 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5503 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5505 struct list_head *tmp;
5515 spin_lock(&all_mddevs_lock);
5516 list_for_each(tmp,&all_mddevs)
5518 mddev = list_entry(tmp, mddev_t, all_mddevs);
5520 spin_unlock(&all_mddevs_lock);
5523 spin_unlock(&all_mddevs_lock);
5525 return (void*)2;/* tail */
5529 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5531 struct list_head *tmp;
5532 mddev_t *next_mddev, *mddev = v;
5538 spin_lock(&all_mddevs_lock);
5540 tmp = all_mddevs.next;
5542 tmp = mddev->all_mddevs.next;
5543 if (tmp != &all_mddevs)
5544 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5546 next_mddev = (void*)2;
5549 spin_unlock(&all_mddevs_lock);
5557 static void md_seq_stop(struct seq_file *seq, void *v)
5561 if (mddev && v != (void*)1 && v != (void*)2)
5565 struct mdstat_info {
5569 static int md_seq_show(struct seq_file *seq, void *v)
5574 struct mdstat_info *mi = seq->private;
5575 struct bitmap *bitmap;
5577 if (v == (void*)1) {
5578 struct mdk_personality *pers;
5579 seq_printf(seq, "Personalities : ");
5580 spin_lock(&pers_lock);
5581 list_for_each_entry(pers, &pers_list, list)
5582 seq_printf(seq, "[%s] ", pers->name);
5584 spin_unlock(&pers_lock);
5585 seq_printf(seq, "\n");
5586 mi->event = atomic_read(&md_event_count);
5589 if (v == (void*)2) {
5594 if (mddev_lock(mddev) < 0)
5597 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5598 seq_printf(seq, "%s : %sactive", mdname(mddev),
5599 mddev->pers ? "" : "in");
5602 seq_printf(seq, " (read-only)");
5604 seq_printf(seq, " (auto-read-only)");
5605 seq_printf(seq, " %s", mddev->pers->name);
5609 list_for_each_entry(rdev, &mddev->disks, same_set) {
5610 char b[BDEVNAME_SIZE];
5611 seq_printf(seq, " %s[%d]",
5612 bdevname(rdev->bdev,b), rdev->desc_nr);
5613 if (test_bit(WriteMostly, &rdev->flags))
5614 seq_printf(seq, "(W)");
5615 if (test_bit(Faulty, &rdev->flags)) {
5616 seq_printf(seq, "(F)");
5618 } else if (rdev->raid_disk < 0)
5619 seq_printf(seq, "(S)"); /* spare */
5623 if (!list_empty(&mddev->disks)) {
5625 seq_printf(seq, "\n %llu blocks",
5626 (unsigned long long)
5627 mddev->array_sectors / 2);
5629 seq_printf(seq, "\n %llu blocks",
5630 (unsigned long long)size);
5632 if (mddev->persistent) {
5633 if (mddev->major_version != 0 ||
5634 mddev->minor_version != 90) {
5635 seq_printf(seq," super %d.%d",
5636 mddev->major_version,
5637 mddev->minor_version);
5639 } else if (mddev->external)
5640 seq_printf(seq, " super external:%s",
5641 mddev->metadata_type);
5643 seq_printf(seq, " super non-persistent");
5646 mddev->pers->status(seq, mddev);
5647 seq_printf(seq, "\n ");
5648 if (mddev->pers->sync_request) {
5649 if (mddev->curr_resync > 2) {
5650 status_resync(seq, mddev);
5651 seq_printf(seq, "\n ");
5652 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5653 seq_printf(seq, "\tresync=DELAYED\n ");
5654 else if (mddev->recovery_cp < MaxSector)
5655 seq_printf(seq, "\tresync=PENDING\n ");
5658 seq_printf(seq, "\n ");
5660 if ((bitmap = mddev->bitmap)) {
5661 unsigned long chunk_kb;
5662 unsigned long flags;
5663 spin_lock_irqsave(&bitmap->lock, flags);
5664 chunk_kb = bitmap->chunksize >> 10;
5665 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5667 bitmap->pages - bitmap->missing_pages,
5669 (bitmap->pages - bitmap->missing_pages)
5670 << (PAGE_SHIFT - 10),
5671 chunk_kb ? chunk_kb : bitmap->chunksize,
5672 chunk_kb ? "KB" : "B");
5674 seq_printf(seq, ", file: ");
5675 seq_path(seq, &bitmap->file->f_path, " \t\n");
5678 seq_printf(seq, "\n");
5679 spin_unlock_irqrestore(&bitmap->lock, flags);
5682 seq_printf(seq, "\n");
5684 mddev_unlock(mddev);
5689 static struct seq_operations md_seq_ops = {
5690 .start = md_seq_start,
5691 .next = md_seq_next,
5692 .stop = md_seq_stop,
5693 .show = md_seq_show,
5696 static int md_seq_open(struct inode *inode, struct file *file)
5699 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5703 error = seq_open(file, &md_seq_ops);
5707 struct seq_file *p = file->private_data;
5709 mi->event = atomic_read(&md_event_count);
5714 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5716 struct seq_file *m = filp->private_data;
5717 struct mdstat_info *mi = m->private;
5720 poll_wait(filp, &md_event_waiters, wait);
5722 /* always allow read */
5723 mask = POLLIN | POLLRDNORM;
5725 if (mi->event != atomic_read(&md_event_count))
5726 mask |= POLLERR | POLLPRI;
5730 static const struct file_operations md_seq_fops = {
5731 .owner = THIS_MODULE,
5732 .open = md_seq_open,
5734 .llseek = seq_lseek,
5735 .release = seq_release_private,
5736 .poll = mdstat_poll,
5739 int register_md_personality(struct mdk_personality *p)
5741 spin_lock(&pers_lock);
5742 list_add_tail(&p->list, &pers_list);
5743 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5744 spin_unlock(&pers_lock);
5748 int unregister_md_personality(struct mdk_personality *p)
5750 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5751 spin_lock(&pers_lock);
5752 list_del_init(&p->list);
5753 spin_unlock(&pers_lock);
5757 static int is_mddev_idle(mddev_t *mddev, int init)
5765 rdev_for_each_rcu(rdev, mddev) {
5766 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5767 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
5768 (int)part_stat_read(&disk->part0, sectors[1]) -
5769 atomic_read(&disk->sync_io);
5770 /* sync IO will cause sync_io to increase before the disk_stats
5771 * as sync_io is counted when a request starts, and
5772 * disk_stats is counted when it completes.
5773 * So resync activity will cause curr_events to be smaller than
5774 * when there was no such activity.
5775 * non-sync IO will cause disk_stat to increase without
5776 * increasing sync_io so curr_events will (eventually)
5777 * be larger than it was before. Once it becomes
5778 * substantially larger, the test below will cause
5779 * the array to appear non-idle, and resync will slow
5781 * If there is a lot of outstanding resync activity when
5782 * we set last_event to curr_events, then all that activity
5783 * completing might cause the array to appear non-idle
5784 * and resync will be slowed down even though there might
5785 * not have been non-resync activity. This will only
5786 * happen once though. 'last_events' will soon reflect
5787 * the state where there is little or no outstanding
5788 * resync requests, and further resync activity will
5789 * always make curr_events less than last_events.
5792 if (init || curr_events - rdev->last_events > 64) {
5793 rdev->last_events = curr_events;
5801 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5803 /* another "blocks" (512byte) blocks have been synced */
5804 atomic_sub(blocks, &mddev->recovery_active);
5805 wake_up(&mddev->recovery_wait);
5807 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5808 md_wakeup_thread(mddev->thread);
5809 // stop recovery, signal do_sync ....
5814 /* md_write_start(mddev, bi)
5815 * If we need to update some array metadata (e.g. 'active' flag
5816 * in superblock) before writing, schedule a superblock update
5817 * and wait for it to complete.
5819 void md_write_start(mddev_t *mddev, struct bio *bi)
5822 if (bio_data_dir(bi) != WRITE)
5825 BUG_ON(mddev->ro == 1);
5826 if (mddev->ro == 2) {
5827 /* need to switch to read/write */
5829 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5830 md_wakeup_thread(mddev->thread);
5831 md_wakeup_thread(mddev->sync_thread);
5834 atomic_inc(&mddev->writes_pending);
5835 if (mddev->safemode == 1)
5836 mddev->safemode = 0;
5837 if (mddev->in_sync) {
5838 spin_lock_irq(&mddev->write_lock);
5839 if (mddev->in_sync) {
5841 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5842 md_wakeup_thread(mddev->thread);
5845 spin_unlock_irq(&mddev->write_lock);
5848 sysfs_notify_dirent(mddev->sysfs_state);
5849 wait_event(mddev->sb_wait,
5850 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5851 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5854 void md_write_end(mddev_t *mddev)
5856 if (atomic_dec_and_test(&mddev->writes_pending)) {
5857 if (mddev->safemode == 2)
5858 md_wakeup_thread(mddev->thread);
5859 else if (mddev->safemode_delay)
5860 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5864 /* md_allow_write(mddev)
5865 * Calling this ensures that the array is marked 'active' so that writes
5866 * may proceed without blocking. It is important to call this before
5867 * attempting a GFP_KERNEL allocation while holding the mddev lock.
5868 * Must be called with mddev_lock held.
5870 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5871 * is dropped, so return -EAGAIN after notifying userspace.
5873 int md_allow_write(mddev_t *mddev)
5879 if (!mddev->pers->sync_request)
5882 spin_lock_irq(&mddev->write_lock);
5883 if (mddev->in_sync) {
5885 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5886 if (mddev->safemode_delay &&
5887 mddev->safemode == 0)
5888 mddev->safemode = 1;
5889 spin_unlock_irq(&mddev->write_lock);
5890 md_update_sb(mddev, 0);
5891 sysfs_notify_dirent(mddev->sysfs_state);
5893 spin_unlock_irq(&mddev->write_lock);
5895 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5900 EXPORT_SYMBOL_GPL(md_allow_write);
5902 #define SYNC_MARKS 10
5903 #define SYNC_MARK_STEP (3*HZ)
5904 void md_do_sync(mddev_t *mddev)
5907 unsigned int currspeed = 0,
5909 sector_t max_sectors,j, io_sectors;
5910 unsigned long mark[SYNC_MARKS];
5911 sector_t mark_cnt[SYNC_MARKS];
5913 struct list_head *tmp;
5914 sector_t last_check;
5919 /* just incase thread restarts... */
5920 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5922 if (mddev->ro) /* never try to sync a read-only array */
5925 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5926 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5927 desc = "data-check";
5928 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5929 desc = "requested-resync";
5932 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5937 /* we overload curr_resync somewhat here.
5938 * 0 == not engaged in resync at all
5939 * 2 == checking that there is no conflict with another sync
5940 * 1 == like 2, but have yielded to allow conflicting resync to
5942 * other == active in resync - this many blocks
5944 * Before starting a resync we must have set curr_resync to
5945 * 2, and then checked that every "conflicting" array has curr_resync
5946 * less than ours. When we find one that is the same or higher
5947 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
5948 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5949 * This will mean we have to start checking from the beginning again.
5954 mddev->curr_resync = 2;
5957 if (kthread_should_stop()) {
5958 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5961 for_each_mddev(mddev2, tmp) {
5962 if (mddev2 == mddev)
5964 if (!mddev->parallel_resync
5965 && mddev2->curr_resync
5966 && match_mddev_units(mddev, mddev2)) {
5968 if (mddev < mddev2 && mddev->curr_resync == 2) {
5969 /* arbitrarily yield */
5970 mddev->curr_resync = 1;
5971 wake_up(&resync_wait);
5973 if (mddev > mddev2 && mddev->curr_resync == 1)
5974 /* no need to wait here, we can wait the next
5975 * time 'round when curr_resync == 2
5978 /* We need to wait 'interruptible' so as not to
5979 * contribute to the load average, and not to
5980 * be caught by 'softlockup'
5982 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5983 if (!kthread_should_stop() &&
5984 mddev2->curr_resync >= mddev->curr_resync) {
5985 printk(KERN_INFO "md: delaying %s of %s"
5986 " until %s has finished (they"
5987 " share one or more physical units)\n",
5988 desc, mdname(mddev), mdname(mddev2));
5990 if (signal_pending(current))
5991 flush_signals(current);
5993 finish_wait(&resync_wait, &wq);
5996 finish_wait(&resync_wait, &wq);
5999 } while (mddev->curr_resync < 2);
6002 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6003 /* resync follows the size requested by the personality,
6004 * which defaults to physical size, but can be virtual size
6006 max_sectors = mddev->resync_max_sectors;
6007 mddev->resync_mismatches = 0;
6008 /* we don't use the checkpoint if there's a bitmap */
6009 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6010 j = mddev->resync_min;
6011 else if (!mddev->bitmap)
6012 j = mddev->recovery_cp;
6014 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6015 max_sectors = mddev->size << 1;
6017 /* recovery follows the physical size of devices */
6018 max_sectors = mddev->size << 1;
6020 list_for_each_entry(rdev, &mddev->disks, same_set)
6021 if (rdev->raid_disk >= 0 &&
6022 !test_bit(Faulty, &rdev->flags) &&
6023 !test_bit(In_sync, &rdev->flags) &&
6024 rdev->recovery_offset < j)
6025 j = rdev->recovery_offset;
6028 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6029 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6030 " %d KB/sec/disk.\n", speed_min(mddev));
6031 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6032 "(but not more than %d KB/sec) for %s.\n",
6033 speed_max(mddev), desc);
6035 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6038 for (m = 0; m < SYNC_MARKS; m++) {
6040 mark_cnt[m] = io_sectors;
6043 mddev->resync_mark = mark[last_mark];
6044 mddev->resync_mark_cnt = mark_cnt[last_mark];
6047 * Tune reconstruction:
6049 window = 32*(PAGE_SIZE/512);
6050 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6051 window/2,(unsigned long long) max_sectors/2);
6053 atomic_set(&mddev->recovery_active, 0);
6058 "md: resuming %s of %s from checkpoint.\n",
6059 desc, mdname(mddev));
6060 mddev->curr_resync = j;
6063 while (j < max_sectors) {
6067 if (j >= mddev->resync_max) {
6068 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6069 wait_event(mddev->recovery_wait,
6070 mddev->resync_max > j
6071 || kthread_should_stop());
6073 if (kthread_should_stop())
6075 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6076 currspeed < speed_min(mddev));
6078 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6082 if (!skipped) { /* actual IO requested */
6083 io_sectors += sectors;
6084 atomic_add(sectors, &mddev->recovery_active);
6088 if (j>1) mddev->curr_resync = j;
6089 mddev->curr_mark_cnt = io_sectors;
6090 if (last_check == 0)
6091 /* this is the earliers that rebuilt will be
6092 * visible in /proc/mdstat
6094 md_new_event(mddev);
6096 if (last_check + window > io_sectors || j == max_sectors)
6099 last_check = io_sectors;
6101 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6105 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6107 int next = (last_mark+1) % SYNC_MARKS;
6109 mddev->resync_mark = mark[next];
6110 mddev->resync_mark_cnt = mark_cnt[next];
6111 mark[next] = jiffies;
6112 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6117 if (kthread_should_stop())
6122 * this loop exits only if either when we are slower than
6123 * the 'hard' speed limit, or the system was IO-idle for
6125 * the system might be non-idle CPU-wise, but we only care
6126 * about not overloading the IO subsystem. (things like an
6127 * e2fsck being done on the RAID array should execute fast)
6129 blk_unplug(mddev->queue);
6132 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6133 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6135 if (currspeed > speed_min(mddev)) {
6136 if ((currspeed > speed_max(mddev)) ||
6137 !is_mddev_idle(mddev, 0)) {
6143 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6145 * this also signals 'finished resyncing' to md_stop
6148 blk_unplug(mddev->queue);
6150 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6152 /* tell personality that we are finished */
6153 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6155 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6156 mddev->curr_resync > 2) {
6157 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6158 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6159 if (mddev->curr_resync >= mddev->recovery_cp) {
6161 "md: checkpointing %s of %s.\n",
6162 desc, mdname(mddev));
6163 mddev->recovery_cp = mddev->curr_resync;
6166 mddev->recovery_cp = MaxSector;
6168 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6169 mddev->curr_resync = MaxSector;
6170 list_for_each_entry(rdev, &mddev->disks, same_set)
6171 if (rdev->raid_disk >= 0 &&
6172 !test_bit(Faulty, &rdev->flags) &&
6173 !test_bit(In_sync, &rdev->flags) &&
6174 rdev->recovery_offset < mddev->curr_resync)
6175 rdev->recovery_offset = mddev->curr_resync;
6178 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6181 mddev->curr_resync = 0;
6182 mddev->resync_min = 0;
6183 mddev->resync_max = MaxSector;
6184 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6185 wake_up(&resync_wait);
6186 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6187 md_wakeup_thread(mddev->thread);
6192 * got a signal, exit.
6195 "md: md_do_sync() got signal ... exiting\n");
6196 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6200 EXPORT_SYMBOL_GPL(md_do_sync);
6203 static int remove_and_add_spares(mddev_t *mddev)
6208 list_for_each_entry(rdev, &mddev->disks, same_set)
6209 if (rdev->raid_disk >= 0 &&
6210 !test_bit(Blocked, &rdev->flags) &&
6211 (test_bit(Faulty, &rdev->flags) ||
6212 ! test_bit(In_sync, &rdev->flags)) &&
6213 atomic_read(&rdev->nr_pending)==0) {
6214 if (mddev->pers->hot_remove_disk(
6215 mddev, rdev->raid_disk)==0) {
6217 sprintf(nm,"rd%d", rdev->raid_disk);
6218 sysfs_remove_link(&mddev->kobj, nm);
6219 rdev->raid_disk = -1;
6223 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6224 list_for_each_entry(rdev, &mddev->disks, same_set) {
6225 if (rdev->raid_disk >= 0 &&
6226 !test_bit(In_sync, &rdev->flags) &&
6227 !test_bit(Blocked, &rdev->flags))
6229 if (rdev->raid_disk < 0
6230 && !test_bit(Faulty, &rdev->flags)) {
6231 rdev->recovery_offset = 0;
6233 hot_add_disk(mddev, rdev) == 0) {
6235 sprintf(nm, "rd%d", rdev->raid_disk);
6236 if (sysfs_create_link(&mddev->kobj,
6239 "md: cannot register "
6243 md_new_event(mddev);
6252 * This routine is regularly called by all per-raid-array threads to
6253 * deal with generic issues like resync and super-block update.
6254 * Raid personalities that don't have a thread (linear/raid0) do not
6255 * need this as they never do any recovery or update the superblock.
6257 * It does not do any resync itself, but rather "forks" off other threads
6258 * to do that as needed.
6259 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6260 * "->recovery" and create a thread at ->sync_thread.
6261 * When the thread finishes it sets MD_RECOVERY_DONE
6262 * and wakeups up this thread which will reap the thread and finish up.
6263 * This thread also removes any faulty devices (with nr_pending == 0).
6265 * The overall approach is:
6266 * 1/ if the superblock needs updating, update it.
6267 * 2/ If a recovery thread is running, don't do anything else.
6268 * 3/ If recovery has finished, clean up, possibly marking spares active.
6269 * 4/ If there are any faulty devices, remove them.
6270 * 5/ If array is degraded, try to add spares devices
6271 * 6/ If array has spares or is not in-sync, start a resync thread.
6273 void md_check_recovery(mddev_t *mddev)
6279 bitmap_daemon_work(mddev->bitmap);
6284 if (signal_pending(current)) {
6285 if (mddev->pers->sync_request && !mddev->external) {
6286 printk(KERN_INFO "md: %s in immediate safe mode\n",
6288 mddev->safemode = 2;
6290 flush_signals(current);
6293 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6296 (mddev->flags && !mddev->external) ||
6297 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6298 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6299 (mddev->external == 0 && mddev->safemode == 1) ||
6300 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6301 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6305 if (mddev_trylock(mddev)) {
6309 /* Only thing we do on a ro array is remove
6312 remove_and_add_spares(mddev);
6313 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6317 if (!mddev->external) {
6319 spin_lock_irq(&mddev->write_lock);
6320 if (mddev->safemode &&
6321 !atomic_read(&mddev->writes_pending) &&
6323 mddev->recovery_cp == MaxSector) {
6326 if (mddev->persistent)
6327 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6329 if (mddev->safemode == 1)
6330 mddev->safemode = 0;
6331 spin_unlock_irq(&mddev->write_lock);
6333 sysfs_notify_dirent(mddev->sysfs_state);
6337 md_update_sb(mddev, 0);
6339 list_for_each_entry(rdev, &mddev->disks, same_set)
6340 if (test_and_clear_bit(StateChanged, &rdev->flags))
6341 sysfs_notify_dirent(rdev->sysfs_state);
6344 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6345 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6346 /* resync/recovery still happening */
6347 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6350 if (mddev->sync_thread) {
6351 /* resync has finished, collect result */
6352 md_unregister_thread(mddev->sync_thread);
6353 mddev->sync_thread = NULL;
6354 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6355 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6357 /* activate any spares */
6358 if (mddev->pers->spare_active(mddev))
6359 sysfs_notify(&mddev->kobj, NULL,
6362 md_update_sb(mddev, 1);
6364 /* if array is no-longer degraded, then any saved_raid_disk
6365 * information must be scrapped
6367 if (!mddev->degraded)
6368 list_for_each_entry(rdev, &mddev->disks, same_set)
6369 rdev->saved_raid_disk = -1;
6371 mddev->recovery = 0;
6372 /* flag recovery needed just to double check */
6373 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6374 sysfs_notify_dirent(mddev->sysfs_action);
6375 md_new_event(mddev);
6378 /* Set RUNNING before clearing NEEDED to avoid
6379 * any transients in the value of "sync_action".
6381 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6382 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6383 /* Clear some bits that don't mean anything, but
6386 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6387 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6389 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6391 /* no recovery is running.
6392 * remove any failed drives, then
6393 * add spares if possible.
6394 * Spare are also removed and re-added, to allow
6395 * the personality to fail the re-add.
6398 if (mddev->reshape_position != MaxSector) {
6399 if (mddev->pers->check_reshape(mddev) != 0)
6400 /* Cannot proceed */
6402 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6403 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6404 } else if ((spares = remove_and_add_spares(mddev))) {
6405 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6406 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6407 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6408 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6409 } else if (mddev->recovery_cp < MaxSector) {
6410 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6411 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6412 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6413 /* nothing to be done ... */
6416 if (mddev->pers->sync_request) {
6417 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6418 /* We are adding a device or devices to an array
6419 * which has the bitmap stored on all devices.
6420 * So make sure all bitmap pages get written
6422 bitmap_write_all(mddev->bitmap);
6424 mddev->sync_thread = md_register_thread(md_do_sync,
6427 if (!mddev->sync_thread) {
6428 printk(KERN_ERR "%s: could not start resync"
6431 /* leave the spares where they are, it shouldn't hurt */
6432 mddev->recovery = 0;
6434 md_wakeup_thread(mddev->sync_thread);
6435 sysfs_notify_dirent(mddev->sysfs_action);
6436 md_new_event(mddev);
6439 if (!mddev->sync_thread) {
6440 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6441 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6443 if (mddev->sysfs_action)
6444 sysfs_notify_dirent(mddev->sysfs_action);
6446 mddev_unlock(mddev);
6450 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6452 sysfs_notify_dirent(rdev->sysfs_state);
6453 wait_event_timeout(rdev->blocked_wait,
6454 !test_bit(Blocked, &rdev->flags),
6455 msecs_to_jiffies(5000));
6456 rdev_dec_pending(rdev, mddev);
6458 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6460 static int md_notify_reboot(struct notifier_block *this,
6461 unsigned long code, void *x)
6463 struct list_head *tmp;
6466 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6468 printk(KERN_INFO "md: stopping all md devices.\n");
6470 for_each_mddev(mddev, tmp)
6471 if (mddev_trylock(mddev)) {
6472 /* Force a switch to readonly even array
6473 * appears to still be in use. Hence
6476 do_md_stop(mddev, 1, 100);
6477 mddev_unlock(mddev);
6480 * certain more exotic SCSI devices are known to be
6481 * volatile wrt too early system reboots. While the
6482 * right place to handle this issue is the given
6483 * driver, we do want to have a safe RAID driver ...
6490 static struct notifier_block md_notifier = {
6491 .notifier_call = md_notify_reboot,
6493 .priority = INT_MAX, /* before any real devices */
6496 static void md_geninit(void)
6498 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6500 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6503 static int __init md_init(void)
6505 if (register_blkdev(MD_MAJOR, "md"))
6507 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6508 unregister_blkdev(MD_MAJOR, "md");
6511 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6512 md_probe, NULL, NULL);
6513 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6514 md_probe, NULL, NULL);
6516 register_reboot_notifier(&md_notifier);
6517 raid_table_header = register_sysctl_table(raid_root_table);
6527 * Searches all registered partitions for autorun RAID arrays
6531 static LIST_HEAD(all_detected_devices);
6532 struct detected_devices_node {
6533 struct list_head list;
6537 void md_autodetect_dev(dev_t dev)
6539 struct detected_devices_node *node_detected_dev;
6541 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6542 if (node_detected_dev) {
6543 node_detected_dev->dev = dev;
6544 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6546 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6547 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6552 static void autostart_arrays(int part)
6555 struct detected_devices_node *node_detected_dev;
6557 int i_scanned, i_passed;
6562 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6564 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6566 node_detected_dev = list_entry(all_detected_devices.next,
6567 struct detected_devices_node, list);
6568 list_del(&node_detected_dev->list);
6569 dev = node_detected_dev->dev;
6570 kfree(node_detected_dev);
6571 rdev = md_import_device(dev,0, 90);
6575 if (test_bit(Faulty, &rdev->flags)) {
6579 set_bit(AutoDetected, &rdev->flags);
6580 list_add(&rdev->same_set, &pending_raid_disks);
6584 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6585 i_scanned, i_passed);
6587 autorun_devices(part);
6590 #endif /* !MODULE */
6592 static __exit void md_exit(void)
6595 struct list_head *tmp;
6597 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6598 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6600 unregister_blkdev(MD_MAJOR,"md");
6601 unregister_blkdev(mdp_major, "mdp");
6602 unregister_reboot_notifier(&md_notifier);
6603 unregister_sysctl_table(raid_table_header);
6604 remove_proc_entry("mdstat", NULL);
6605 for_each_mddev(mddev, tmp) {
6606 export_array(mddev);
6607 mddev->hold_active = 0;
6611 subsys_initcall(md_init);
6612 module_exit(md_exit)
6614 static int get_ro(char *buffer, struct kernel_param *kp)
6616 return sprintf(buffer, "%d", start_readonly);
6618 static int set_ro(const char *val, struct kernel_param *kp)
6621 int num = simple_strtoul(val, &e, 10);
6622 if (*val && (*e == '\0' || *e == '\n')) {
6623 start_readonly = num;
6629 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6630 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6632 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6634 EXPORT_SYMBOL(register_md_personality);
6635 EXPORT_SYMBOL(unregister_md_personality);
6636 EXPORT_SYMBOL(md_error);
6637 EXPORT_SYMBOL(md_done_sync);
6638 EXPORT_SYMBOL(md_write_start);
6639 EXPORT_SYMBOL(md_write_end);
6640 EXPORT_SYMBOL(md_register_thread);
6641 EXPORT_SYMBOL(md_unregister_thread);
6642 EXPORT_SYMBOL(md_wakeup_thread);
6643 EXPORT_SYMBOL(md_check_recovery);
6644 MODULE_LICENSE("GPL");
6646 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);