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 /* Rather than calling directly into the personality make_request function,
205 * IO requests come here first so that we can check if the device is
206 * being suspended pending a reconfiguration.
207 * We hold a refcount over the call to ->make_request. By the time that
208 * call has finished, the bio has been linked into some internal structure
209 * and so is visible to ->quiesce(), so we don't need the refcount any more.
211 static int md_make_request(struct request_queue *q, struct bio *bio)
213 mddev_t *mddev = q->queuedata;
215 if (mddev == NULL || mddev->pers == NULL) {
220 if (mddev->suspended) {
223 prepare_to_wait(&mddev->sb_wait, &__wait,
224 TASK_UNINTERRUPTIBLE);
225 if (!mddev->suspended)
231 finish_wait(&mddev->sb_wait, &__wait);
233 atomic_inc(&mddev->active_io);
235 rv = mddev->pers->make_request(q, bio);
236 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237 wake_up(&mddev->sb_wait);
242 static void mddev_suspend(mddev_t *mddev)
244 BUG_ON(mddev->suspended);
245 mddev->suspended = 1;
247 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248 mddev->pers->quiesce(mddev, 1);
249 md_unregister_thread(mddev->thread);
250 mddev->thread = NULL;
251 /* we now know that no code is executing in the personality module,
252 * except possibly the tail end of a ->bi_end_io function, but that
253 * is certain to complete before the module has a chance to get
258 static void mddev_resume(mddev_t *mddev)
260 mddev->suspended = 0;
261 wake_up(&mddev->sb_wait);
262 mddev->pers->quiesce(mddev, 0);
266 static inline mddev_t *mddev_get(mddev_t *mddev)
268 atomic_inc(&mddev->active);
272 static void mddev_delayed_delete(struct work_struct *ws)
274 mddev_t *mddev = container_of(ws, mddev_t, del_work);
275 kobject_del(&mddev->kobj);
276 kobject_put(&mddev->kobj);
279 static void mddev_put(mddev_t *mddev)
281 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
283 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
284 !mddev->hold_active) {
285 list_del(&mddev->all_mddevs);
286 if (mddev->gendisk) {
287 /* we did a probe so need to clean up.
288 * Call schedule_work inside the spinlock
289 * so that flush_scheduled_work() after
290 * mddev_find will succeed in waiting for the
293 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
294 schedule_work(&mddev->del_work);
298 spin_unlock(&all_mddevs_lock);
301 static mddev_t * mddev_find(dev_t unit)
303 mddev_t *mddev, *new = NULL;
306 spin_lock(&all_mddevs_lock);
309 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
310 if (mddev->unit == unit) {
312 spin_unlock(&all_mddevs_lock);
318 list_add(&new->all_mddevs, &all_mddevs);
319 spin_unlock(&all_mddevs_lock);
320 new->hold_active = UNTIL_IOCTL;
324 /* find an unused unit number */
325 static int next_minor = 512;
326 int start = next_minor;
330 dev = MKDEV(MD_MAJOR, next_minor);
332 if (next_minor > MINORMASK)
334 if (next_minor == start) {
335 /* Oh dear, all in use. */
336 spin_unlock(&all_mddevs_lock);
342 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
343 if (mddev->unit == dev) {
349 new->md_minor = MINOR(dev);
350 new->hold_active = UNTIL_STOP;
351 list_add(&new->all_mddevs, &all_mddevs);
352 spin_unlock(&all_mddevs_lock);
355 spin_unlock(&all_mddevs_lock);
357 new = kzalloc(sizeof(*new), GFP_KERNEL);
362 if (MAJOR(unit) == MD_MAJOR)
363 new->md_minor = MINOR(unit);
365 new->md_minor = MINOR(unit) >> MdpMinorShift;
367 mutex_init(&new->reconfig_mutex);
368 INIT_LIST_HEAD(&new->disks);
369 INIT_LIST_HEAD(&new->all_mddevs);
370 init_timer(&new->safemode_timer);
371 atomic_set(&new->active, 1);
372 atomic_set(&new->openers, 0);
373 atomic_set(&new->active_io, 0);
374 spin_lock_init(&new->write_lock);
375 init_waitqueue_head(&new->sb_wait);
376 init_waitqueue_head(&new->recovery_wait);
377 new->reshape_position = MaxSector;
379 new->resync_max = MaxSector;
380 new->level = LEVEL_NONE;
385 static inline int mddev_lock(mddev_t * mddev)
387 return mutex_lock_interruptible(&mddev->reconfig_mutex);
390 static inline int mddev_trylock(mddev_t * mddev)
392 return mutex_trylock(&mddev->reconfig_mutex);
395 static inline void mddev_unlock(mddev_t * mddev)
397 mutex_unlock(&mddev->reconfig_mutex);
399 md_wakeup_thread(mddev->thread);
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
406 list_for_each_entry(rdev, &mddev->disks, same_set)
407 if (rdev->desc_nr == nr)
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
417 list_for_each_entry(rdev, &mddev->disks, same_set)
418 if (rdev->bdev->bd_dev == dev)
424 static struct mdk_personality *find_pers(int level, char *clevel)
426 struct mdk_personality *pers;
427 list_for_each_entry(pers, &pers_list, list) {
428 if (level != LEVEL_NONE && pers->level == level)
430 if (strcmp(pers->name, clevel)==0)
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
439 sector_t num_sectors = bdev->bd_inode->i_size / 512;
440 return MD_NEW_SIZE_SECTORS(num_sectors);
443 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
445 sector_t num_sectors = rdev->sb_start;
448 num_sectors &= ~((sector_t)chunk_size/512 - 1);
452 static int alloc_disk_sb(mdk_rdev_t * rdev)
457 rdev->sb_page = alloc_page(GFP_KERNEL);
458 if (!rdev->sb_page) {
459 printk(KERN_ALERT "md: out of memory.\n");
466 static void free_disk_sb(mdk_rdev_t * rdev)
469 put_page(rdev->sb_page);
471 rdev->sb_page = NULL;
478 static void super_written(struct bio *bio, int error)
480 mdk_rdev_t *rdev = bio->bi_private;
481 mddev_t *mddev = rdev->mddev;
483 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
484 printk("md: super_written gets error=%d, uptodate=%d\n",
485 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
486 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
487 md_error(mddev, rdev);
490 if (atomic_dec_and_test(&mddev->pending_writes))
491 wake_up(&mddev->sb_wait);
495 static void super_written_barrier(struct bio *bio, int error)
497 struct bio *bio2 = bio->bi_private;
498 mdk_rdev_t *rdev = bio2->bi_private;
499 mddev_t *mddev = rdev->mddev;
501 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
502 error == -EOPNOTSUPP) {
504 /* barriers don't appear to be supported :-( */
505 set_bit(BarriersNotsupp, &rdev->flags);
506 mddev->barriers_work = 0;
507 spin_lock_irqsave(&mddev->write_lock, flags);
508 bio2->bi_next = mddev->biolist;
509 mddev->biolist = bio2;
510 spin_unlock_irqrestore(&mddev->write_lock, flags);
511 wake_up(&mddev->sb_wait);
515 bio->bi_private = rdev;
516 super_written(bio, error);
520 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
521 sector_t sector, int size, struct page *page)
523 /* write first size bytes of page to sector of rdev
524 * Increment mddev->pending_writes before returning
525 * and decrement it on completion, waking up sb_wait
526 * if zero is reached.
527 * If an error occurred, call md_error
529 * As we might need to resubmit the request if BIO_RW_BARRIER
530 * causes ENOTSUPP, we allocate a spare bio...
532 struct bio *bio = bio_alloc(GFP_NOIO, 1);
533 int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
535 bio->bi_bdev = rdev->bdev;
536 bio->bi_sector = sector;
537 bio_add_page(bio, page, size, 0);
538 bio->bi_private = rdev;
539 bio->bi_end_io = super_written;
542 atomic_inc(&mddev->pending_writes);
543 if (!test_bit(BarriersNotsupp, &rdev->flags)) {
545 rw |= (1<<BIO_RW_BARRIER);
546 rbio = bio_clone(bio, GFP_NOIO);
547 rbio->bi_private = bio;
548 rbio->bi_end_io = super_written_barrier;
549 submit_bio(rw, rbio);
554 void md_super_wait(mddev_t *mddev)
556 /* wait for all superblock writes that were scheduled to complete.
557 * if any had to be retried (due to BARRIER problems), retry them
561 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
562 if (atomic_read(&mddev->pending_writes)==0)
564 while (mddev->biolist) {
566 spin_lock_irq(&mddev->write_lock);
567 bio = mddev->biolist;
568 mddev->biolist = bio->bi_next ;
570 spin_unlock_irq(&mddev->write_lock);
571 submit_bio(bio->bi_rw, bio);
575 finish_wait(&mddev->sb_wait, &wq);
578 static void bi_complete(struct bio *bio, int error)
580 complete((struct completion*)bio->bi_private);
583 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
584 struct page *page, int rw)
586 struct bio *bio = bio_alloc(GFP_NOIO, 1);
587 struct completion event;
590 rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
593 bio->bi_sector = sector;
594 bio_add_page(bio, page, size, 0);
595 init_completion(&event);
596 bio->bi_private = &event;
597 bio->bi_end_io = bi_complete;
599 wait_for_completion(&event);
601 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
605 EXPORT_SYMBOL_GPL(sync_page_io);
607 static int read_disk_sb(mdk_rdev_t * rdev, int size)
609 char b[BDEVNAME_SIZE];
610 if (!rdev->sb_page) {
618 if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
624 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
625 bdevname(rdev->bdev,b));
629 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
631 return sb1->set_uuid0 == sb2->set_uuid0 &&
632 sb1->set_uuid1 == sb2->set_uuid1 &&
633 sb1->set_uuid2 == sb2->set_uuid2 &&
634 sb1->set_uuid3 == sb2->set_uuid3;
637 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
640 mdp_super_t *tmp1, *tmp2;
642 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
643 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
645 if (!tmp1 || !tmp2) {
647 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
655 * nr_disks is not constant
660 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
668 static u32 md_csum_fold(u32 csum)
670 csum = (csum & 0xffff) + (csum >> 16);
671 return (csum & 0xffff) + (csum >> 16);
674 static unsigned int calc_sb_csum(mdp_super_t * sb)
677 u32 *sb32 = (u32*)sb;
679 unsigned int disk_csum, csum;
681 disk_csum = sb->sb_csum;
684 for (i = 0; i < MD_SB_BYTES/4 ; i++)
686 csum = (newcsum & 0xffffffff) + (newcsum>>32);
690 /* This used to use csum_partial, which was wrong for several
691 * reasons including that different results are returned on
692 * different architectures. It isn't critical that we get exactly
693 * the same return value as before (we always csum_fold before
694 * testing, and that removes any differences). However as we
695 * know that csum_partial always returned a 16bit value on
696 * alphas, do a fold to maximise conformity to previous behaviour.
698 sb->sb_csum = md_csum_fold(disk_csum);
700 sb->sb_csum = disk_csum;
707 * Handle superblock details.
708 * We want to be able to handle multiple superblock formats
709 * so we have a common interface to them all, and an array of
710 * different handlers.
711 * We rely on user-space to write the initial superblock, and support
712 * reading and updating of superblocks.
713 * Interface methods are:
714 * int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
715 * loads and validates a superblock on dev.
716 * if refdev != NULL, compare superblocks on both devices
718 * 0 - dev has a superblock that is compatible with refdev
719 * 1 - dev has a superblock that is compatible and newer than refdev
720 * so dev should be used as the refdev in future
721 * -EINVAL superblock incompatible or invalid
722 * -othererror e.g. -EIO
724 * int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
725 * Verify that dev is acceptable into mddev.
726 * The first time, mddev->raid_disks will be 0, and data from
727 * dev should be merged in. Subsequent calls check that dev
728 * is new enough. Return 0 or -EINVAL
730 * void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
731 * Update the superblock for rdev with data in mddev
732 * This does not write to disc.
738 struct module *owner;
739 int (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
741 int (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
742 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
743 unsigned long long (*rdev_size_change)(mdk_rdev_t *rdev,
744 sector_t num_sectors);
748 * load_super for 0.90.0
750 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
752 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
757 * Calculate the position of the superblock (512byte sectors),
758 * it's at the end of the disk.
760 * It also happens to be a multiple of 4Kb.
762 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
764 ret = read_disk_sb(rdev, MD_SB_BYTES);
769 bdevname(rdev->bdev, b);
770 sb = (mdp_super_t*)page_address(rdev->sb_page);
772 if (sb->md_magic != MD_SB_MAGIC) {
773 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
778 if (sb->major_version != 0 ||
779 sb->minor_version < 90 ||
780 sb->minor_version > 91) {
781 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
782 sb->major_version, sb->minor_version,
787 if (sb->raid_disks <= 0)
790 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
791 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
796 rdev->preferred_minor = sb->md_minor;
797 rdev->data_offset = 0;
798 rdev->sb_size = MD_SB_BYTES;
800 if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
801 if (sb->level != 1 && sb->level != 4
802 && sb->level != 5 && sb->level != 6
803 && sb->level != 10) {
804 /* FIXME use a better test */
806 "md: bitmaps not supported for this level.\n");
811 if (sb->level == LEVEL_MULTIPATH)
814 rdev->desc_nr = sb->this_disk.number;
820 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
821 if (!uuid_equal(refsb, sb)) {
822 printk(KERN_WARNING "md: %s has different UUID to %s\n",
823 b, bdevname(refdev->bdev,b2));
826 if (!sb_equal(refsb, sb)) {
827 printk(KERN_WARNING "md: %s has same UUID"
828 " but different superblock to %s\n",
829 b, bdevname(refdev->bdev, b2));
833 ev2 = md_event(refsb);
839 rdev->sectors = calc_num_sectors(rdev, sb->chunk_size);
841 if (rdev->sectors < sb->size * 2 && sb->level > 1)
842 /* "this cannot possibly happen" ... */
850 * validate_super for 0.90.0
852 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
855 mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
856 __u64 ev1 = md_event(sb);
858 rdev->raid_disk = -1;
859 clear_bit(Faulty, &rdev->flags);
860 clear_bit(In_sync, &rdev->flags);
861 clear_bit(WriteMostly, &rdev->flags);
862 clear_bit(BarriersNotsupp, &rdev->flags);
864 if (mddev->raid_disks == 0) {
865 mddev->major_version = 0;
866 mddev->minor_version = sb->minor_version;
867 mddev->patch_version = sb->patch_version;
869 mddev->chunk_size = sb->chunk_size;
870 mddev->ctime = sb->ctime;
871 mddev->utime = sb->utime;
872 mddev->level = sb->level;
873 mddev->clevel[0] = 0;
874 mddev->layout = sb->layout;
875 mddev->raid_disks = sb->raid_disks;
876 mddev->dev_sectors = sb->size * 2;
878 mddev->bitmap_offset = 0;
879 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
881 if (mddev->minor_version >= 91) {
882 mddev->reshape_position = sb->reshape_position;
883 mddev->delta_disks = sb->delta_disks;
884 mddev->new_level = sb->new_level;
885 mddev->new_layout = sb->new_layout;
886 mddev->new_chunk = sb->new_chunk;
888 mddev->reshape_position = MaxSector;
889 mddev->delta_disks = 0;
890 mddev->new_level = mddev->level;
891 mddev->new_layout = mddev->layout;
892 mddev->new_chunk = mddev->chunk_size;
895 if (sb->state & (1<<MD_SB_CLEAN))
896 mddev->recovery_cp = MaxSector;
898 if (sb->events_hi == sb->cp_events_hi &&
899 sb->events_lo == sb->cp_events_lo) {
900 mddev->recovery_cp = sb->recovery_cp;
902 mddev->recovery_cp = 0;
905 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
906 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
907 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
908 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
910 mddev->max_disks = MD_SB_DISKS;
912 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
913 mddev->bitmap_file == NULL)
914 mddev->bitmap_offset = mddev->default_bitmap_offset;
916 } else if (mddev->pers == NULL) {
917 /* Insist on good event counter while assembling */
919 if (ev1 < mddev->events)
921 } else if (mddev->bitmap) {
922 /* if adding to array with a bitmap, then we can accept an
923 * older device ... but not too old.
925 if (ev1 < mddev->bitmap->events_cleared)
928 if (ev1 < mddev->events)
929 /* just a hot-add of a new device, leave raid_disk at -1 */
933 if (mddev->level != LEVEL_MULTIPATH) {
934 desc = sb->disks + rdev->desc_nr;
936 if (desc->state & (1<<MD_DISK_FAULTY))
937 set_bit(Faulty, &rdev->flags);
938 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
939 desc->raid_disk < mddev->raid_disks */) {
940 set_bit(In_sync, &rdev->flags);
941 rdev->raid_disk = desc->raid_disk;
943 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
944 set_bit(WriteMostly, &rdev->flags);
945 } else /* MULTIPATH are always insync */
946 set_bit(In_sync, &rdev->flags);
951 * sync_super for 0.90.0
953 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
957 int next_spare = mddev->raid_disks;
960 /* make rdev->sb match mddev data..
963 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
964 * 3/ any empty disks < next_spare become removed
966 * disks[0] gets initialised to REMOVED because
967 * we cannot be sure from other fields if it has
968 * been initialised or not.
971 int active=0, working=0,failed=0,spare=0,nr_disks=0;
973 rdev->sb_size = MD_SB_BYTES;
975 sb = (mdp_super_t*)page_address(rdev->sb_page);
977 memset(sb, 0, sizeof(*sb));
979 sb->md_magic = MD_SB_MAGIC;
980 sb->major_version = mddev->major_version;
981 sb->patch_version = mddev->patch_version;
982 sb->gvalid_words = 0; /* ignored */
983 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
984 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
985 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
986 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
988 sb->ctime = mddev->ctime;
989 sb->level = mddev->level;
990 sb->size = mddev->dev_sectors / 2;
991 sb->raid_disks = mddev->raid_disks;
992 sb->md_minor = mddev->md_minor;
993 sb->not_persistent = 0;
994 sb->utime = mddev->utime;
996 sb->events_hi = (mddev->events>>32);
997 sb->events_lo = (u32)mddev->events;
999 if (mddev->reshape_position == MaxSector)
1000 sb->minor_version = 90;
1002 sb->minor_version = 91;
1003 sb->reshape_position = mddev->reshape_position;
1004 sb->new_level = mddev->new_level;
1005 sb->delta_disks = mddev->delta_disks;
1006 sb->new_layout = mddev->new_layout;
1007 sb->new_chunk = mddev->new_chunk;
1009 mddev->minor_version = sb->minor_version;
1012 sb->recovery_cp = mddev->recovery_cp;
1013 sb->cp_events_hi = (mddev->events>>32);
1014 sb->cp_events_lo = (u32)mddev->events;
1015 if (mddev->recovery_cp == MaxSector)
1016 sb->state = (1<< MD_SB_CLEAN);
1018 sb->recovery_cp = 0;
1020 sb->layout = mddev->layout;
1021 sb->chunk_size = mddev->chunk_size;
1023 if (mddev->bitmap && mddev->bitmap_file == NULL)
1024 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1026 sb->disks[0].state = (1<<MD_DISK_REMOVED);
1027 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1030 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1031 && !test_bit(Faulty, &rdev2->flags))
1032 desc_nr = rdev2->raid_disk;
1034 desc_nr = next_spare++;
1035 rdev2->desc_nr = desc_nr;
1036 d = &sb->disks[rdev2->desc_nr];
1038 d->number = rdev2->desc_nr;
1039 d->major = MAJOR(rdev2->bdev->bd_dev);
1040 d->minor = MINOR(rdev2->bdev->bd_dev);
1041 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1042 && !test_bit(Faulty, &rdev2->flags))
1043 d->raid_disk = rdev2->raid_disk;
1045 d->raid_disk = rdev2->desc_nr; /* compatibility */
1046 if (test_bit(Faulty, &rdev2->flags))
1047 d->state = (1<<MD_DISK_FAULTY);
1048 else if (test_bit(In_sync, &rdev2->flags)) {
1049 d->state = (1<<MD_DISK_ACTIVE);
1050 d->state |= (1<<MD_DISK_SYNC);
1058 if (test_bit(WriteMostly, &rdev2->flags))
1059 d->state |= (1<<MD_DISK_WRITEMOSTLY);
1061 /* now set the "removed" and "faulty" bits on any missing devices */
1062 for (i=0 ; i < mddev->raid_disks ; i++) {
1063 mdp_disk_t *d = &sb->disks[i];
1064 if (d->state == 0 && d->number == 0) {
1067 d->state = (1<<MD_DISK_REMOVED);
1068 d->state |= (1<<MD_DISK_FAULTY);
1072 sb->nr_disks = nr_disks;
1073 sb->active_disks = active;
1074 sb->working_disks = working;
1075 sb->failed_disks = failed;
1076 sb->spare_disks = spare;
1078 sb->this_disk = sb->disks[rdev->desc_nr];
1079 sb->sb_csum = calc_sb_csum(sb);
1083 * rdev_size_change for 0.90.0
1085 static unsigned long long
1086 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1088 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1089 return 0; /* component must fit device */
1090 if (rdev->mddev->bitmap_offset)
1091 return 0; /* can't move bitmap */
1092 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1093 if (!num_sectors || num_sectors > rdev->sb_start)
1094 num_sectors = rdev->sb_start;
1095 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1097 md_super_wait(rdev->mddev);
1098 return num_sectors / 2; /* kB for sysfs */
1103 * version 1 superblock
1106 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1110 unsigned long long newcsum;
1111 int size = 256 + le32_to_cpu(sb->max_dev)*2;
1112 __le32 *isuper = (__le32*)sb;
1115 disk_csum = sb->sb_csum;
1118 for (i=0; size>=4; size -= 4 )
1119 newcsum += le32_to_cpu(*isuper++);
1122 newcsum += le16_to_cpu(*(__le16*) isuper);
1124 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1125 sb->sb_csum = disk_csum;
1126 return cpu_to_le32(csum);
1129 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1131 struct mdp_superblock_1 *sb;
1134 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1138 * Calculate the position of the superblock in 512byte sectors.
1139 * It is always aligned to a 4K boundary and
1140 * depeding on minor_version, it can be:
1141 * 0: At least 8K, but less than 12K, from end of device
1142 * 1: At start of device
1143 * 2: 4K from start of device.
1145 switch(minor_version) {
1147 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1149 sb_start &= ~(sector_t)(4*2-1);
1160 rdev->sb_start = sb_start;
1162 /* superblock is rarely larger than 1K, but it can be larger,
1163 * and it is safe to read 4k, so we do that
1165 ret = read_disk_sb(rdev, 4096);
1166 if (ret) return ret;
1169 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1171 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1172 sb->major_version != cpu_to_le32(1) ||
1173 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1174 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1175 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1178 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1179 printk("md: invalid superblock checksum on %s\n",
1180 bdevname(rdev->bdev,b));
1183 if (le64_to_cpu(sb->data_size) < 10) {
1184 printk("md: data_size too small on %s\n",
1185 bdevname(rdev->bdev,b));
1188 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1189 if (sb->level != cpu_to_le32(1) &&
1190 sb->level != cpu_to_le32(4) &&
1191 sb->level != cpu_to_le32(5) &&
1192 sb->level != cpu_to_le32(6) &&
1193 sb->level != cpu_to_le32(10)) {
1195 "md: bitmaps not supported for this level.\n");
1200 rdev->preferred_minor = 0xffff;
1201 rdev->data_offset = le64_to_cpu(sb->data_offset);
1202 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1204 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1205 bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1206 if (rdev->sb_size & bmask)
1207 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1210 && rdev->data_offset < sb_start + (rdev->sb_size/512))
1213 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1216 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1222 struct mdp_superblock_1 *refsb =
1223 (struct mdp_superblock_1*)page_address(refdev->sb_page);
1225 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1226 sb->level != refsb->level ||
1227 sb->layout != refsb->layout ||
1228 sb->chunksize != refsb->chunksize) {
1229 printk(KERN_WARNING "md: %s has strangely different"
1230 " superblock to %s\n",
1231 bdevname(rdev->bdev,b),
1232 bdevname(refdev->bdev,b2));
1235 ev1 = le64_to_cpu(sb->events);
1236 ev2 = le64_to_cpu(refsb->events);
1244 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1245 le64_to_cpu(sb->data_offset);
1247 rdev->sectors = rdev->sb_start;
1248 if (rdev->sectors < le64_to_cpu(sb->data_size))
1250 rdev->sectors = le64_to_cpu(sb->data_size);
1251 if (le32_to_cpu(sb->chunksize))
1252 rdev->sectors &= ~((sector_t)le32_to_cpu(sb->chunksize) - 1);
1254 if (le64_to_cpu(sb->size) > rdev->sectors)
1259 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1261 struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1262 __u64 ev1 = le64_to_cpu(sb->events);
1264 rdev->raid_disk = -1;
1265 clear_bit(Faulty, &rdev->flags);
1266 clear_bit(In_sync, &rdev->flags);
1267 clear_bit(WriteMostly, &rdev->flags);
1268 clear_bit(BarriersNotsupp, &rdev->flags);
1270 if (mddev->raid_disks == 0) {
1271 mddev->major_version = 1;
1272 mddev->patch_version = 0;
1273 mddev->external = 0;
1274 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1275 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1276 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1277 mddev->level = le32_to_cpu(sb->level);
1278 mddev->clevel[0] = 0;
1279 mddev->layout = le32_to_cpu(sb->layout);
1280 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1281 mddev->dev_sectors = le64_to_cpu(sb->size);
1282 mddev->events = ev1;
1283 mddev->bitmap_offset = 0;
1284 mddev->default_bitmap_offset = 1024 >> 9;
1286 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1287 memcpy(mddev->uuid, sb->set_uuid, 16);
1289 mddev->max_disks = (4096-256)/2;
1291 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1292 mddev->bitmap_file == NULL )
1293 mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1295 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1296 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1297 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1298 mddev->new_level = le32_to_cpu(sb->new_level);
1299 mddev->new_layout = le32_to_cpu(sb->new_layout);
1300 mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1302 mddev->reshape_position = MaxSector;
1303 mddev->delta_disks = 0;
1304 mddev->new_level = mddev->level;
1305 mddev->new_layout = mddev->layout;
1306 mddev->new_chunk = mddev->chunk_size;
1309 } else if (mddev->pers == NULL) {
1310 /* Insist of good event counter while assembling */
1312 if (ev1 < mddev->events)
1314 } else if (mddev->bitmap) {
1315 /* If adding to array with a bitmap, then we can accept an
1316 * older device, but not too old.
1318 if (ev1 < mddev->bitmap->events_cleared)
1321 if (ev1 < mddev->events)
1322 /* just a hot-add of a new device, leave raid_disk at -1 */
1325 if (mddev->level != LEVEL_MULTIPATH) {
1327 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1329 case 0xffff: /* spare */
1331 case 0xfffe: /* faulty */
1332 set_bit(Faulty, &rdev->flags);
1335 if ((le32_to_cpu(sb->feature_map) &
1336 MD_FEATURE_RECOVERY_OFFSET))
1337 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1339 set_bit(In_sync, &rdev->flags);
1340 rdev->raid_disk = role;
1343 if (sb->devflags & WriteMostly1)
1344 set_bit(WriteMostly, &rdev->flags);
1345 } else /* MULTIPATH are always insync */
1346 set_bit(In_sync, &rdev->flags);
1351 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1353 struct mdp_superblock_1 *sb;
1356 /* make rdev->sb match mddev and rdev data. */
1358 sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1360 sb->feature_map = 0;
1362 sb->recovery_offset = cpu_to_le64(0);
1363 memset(sb->pad1, 0, sizeof(sb->pad1));
1364 memset(sb->pad2, 0, sizeof(sb->pad2));
1365 memset(sb->pad3, 0, sizeof(sb->pad3));
1367 sb->utime = cpu_to_le64((__u64)mddev->utime);
1368 sb->events = cpu_to_le64(mddev->events);
1370 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1372 sb->resync_offset = cpu_to_le64(0);
1374 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1376 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1377 sb->size = cpu_to_le64(mddev->dev_sectors);
1379 if (mddev->bitmap && mddev->bitmap_file == NULL) {
1380 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1381 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1384 if (rdev->raid_disk >= 0 &&
1385 !test_bit(In_sync, &rdev->flags)) {
1386 if (mddev->curr_resync_completed > rdev->recovery_offset)
1387 rdev->recovery_offset = mddev->curr_resync_completed;
1388 if (rdev->recovery_offset > 0) {
1390 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1391 sb->recovery_offset =
1392 cpu_to_le64(rdev->recovery_offset);
1396 if (mddev->reshape_position != MaxSector) {
1397 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1398 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1399 sb->new_layout = cpu_to_le32(mddev->new_layout);
1400 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1401 sb->new_level = cpu_to_le32(mddev->new_level);
1402 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1406 list_for_each_entry(rdev2, &mddev->disks, same_set)
1407 if (rdev2->desc_nr+1 > max_dev)
1408 max_dev = rdev2->desc_nr+1;
1410 if (max_dev > le32_to_cpu(sb->max_dev))
1411 sb->max_dev = cpu_to_le32(max_dev);
1412 for (i=0; i<max_dev;i++)
1413 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1415 list_for_each_entry(rdev2, &mddev->disks, same_set) {
1417 if (test_bit(Faulty, &rdev2->flags))
1418 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1419 else if (test_bit(In_sync, &rdev2->flags))
1420 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1421 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1422 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1424 sb->dev_roles[i] = cpu_to_le16(0xffff);
1427 sb->sb_csum = calc_sb_1_csum(sb);
1430 static unsigned long long
1431 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1433 struct mdp_superblock_1 *sb;
1434 sector_t max_sectors;
1435 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1436 return 0; /* component must fit device */
1437 if (rdev->sb_start < rdev->data_offset) {
1438 /* minor versions 1 and 2; superblock before data */
1439 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1440 max_sectors -= rdev->data_offset;
1441 if (!num_sectors || num_sectors > max_sectors)
1442 num_sectors = max_sectors;
1443 } else if (rdev->mddev->bitmap_offset) {
1444 /* minor version 0 with bitmap we can't move */
1447 /* minor version 0; superblock after data */
1449 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1450 sb_start &= ~(sector_t)(4*2 - 1);
1451 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1452 if (!num_sectors || num_sectors > max_sectors)
1453 num_sectors = max_sectors;
1454 rdev->sb_start = sb_start;
1456 sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1457 sb->data_size = cpu_to_le64(num_sectors);
1458 sb->super_offset = rdev->sb_start;
1459 sb->sb_csum = calc_sb_1_csum(sb);
1460 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1462 md_super_wait(rdev->mddev);
1463 return num_sectors / 2; /* kB for sysfs */
1466 static struct super_type super_types[] = {
1469 .owner = THIS_MODULE,
1470 .load_super = super_90_load,
1471 .validate_super = super_90_validate,
1472 .sync_super = super_90_sync,
1473 .rdev_size_change = super_90_rdev_size_change,
1477 .owner = THIS_MODULE,
1478 .load_super = super_1_load,
1479 .validate_super = super_1_validate,
1480 .sync_super = super_1_sync,
1481 .rdev_size_change = super_1_rdev_size_change,
1485 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1487 mdk_rdev_t *rdev, *rdev2;
1490 rdev_for_each_rcu(rdev, mddev1)
1491 rdev_for_each_rcu(rdev2, mddev2)
1492 if (rdev->bdev->bd_contains ==
1493 rdev2->bdev->bd_contains) {
1501 static LIST_HEAD(pending_raid_disks);
1503 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1505 struct mdk_personality *pers = mddev->pers;
1506 struct gendisk *disk = mddev->gendisk;
1507 struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1508 struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1510 /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1511 if (pers && pers->level >= 4 && pers->level <= 6)
1514 /* If rdev is integrity capable, register profile for mddev */
1515 if (!bi_mddev && bi_rdev) {
1516 if (blk_integrity_register(disk, bi_rdev))
1517 printk(KERN_ERR "%s: %s Could not register integrity!\n",
1518 __func__, disk->disk_name);
1520 printk(KERN_NOTICE "Enabling data integrity on %s\n",
1525 /* Check that mddev and rdev have matching profiles */
1526 if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1527 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1528 disk->disk_name, rdev->bdev->bd_disk->disk_name);
1529 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1531 blk_integrity_unregister(disk);
1535 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1537 char b[BDEVNAME_SIZE];
1547 /* prevent duplicates */
1548 if (find_rdev(mddev, rdev->bdev->bd_dev))
1551 /* make sure rdev->sectors exceeds mddev->dev_sectors */
1552 if (rdev->sectors && (mddev->dev_sectors == 0 ||
1553 rdev->sectors < mddev->dev_sectors)) {
1555 /* Cannot change size, so fail
1556 * If mddev->level <= 0, then we don't care
1557 * about aligning sizes (e.g. linear)
1559 if (mddev->level > 0)
1562 mddev->dev_sectors = rdev->sectors;
1565 /* Verify rdev->desc_nr is unique.
1566 * If it is -1, assign a free number, else
1567 * check number is not in use
1569 if (rdev->desc_nr < 0) {
1571 if (mddev->pers) choice = mddev->raid_disks;
1572 while (find_rdev_nr(mddev, choice))
1574 rdev->desc_nr = choice;
1576 if (find_rdev_nr(mddev, rdev->desc_nr))
1579 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1580 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1581 mdname(mddev), mddev->max_disks);
1584 bdevname(rdev->bdev,b);
1585 while ( (s=strchr(b, '/')) != NULL)
1588 rdev->mddev = mddev;
1589 printk(KERN_INFO "md: bind<%s>\n", b);
1591 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1594 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1595 if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1596 kobject_del(&rdev->kobj);
1599 rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1601 list_add_rcu(&rdev->same_set, &mddev->disks);
1602 bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1604 /* May as well allow recovery to be retried once */
1605 mddev->recovery_disabled = 0;
1607 md_integrity_check(rdev, mddev);
1611 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1616 static void md_delayed_delete(struct work_struct *ws)
1618 mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1619 kobject_del(&rdev->kobj);
1620 kobject_put(&rdev->kobj);
1623 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1625 char b[BDEVNAME_SIZE];
1630 bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1631 list_del_rcu(&rdev->same_set);
1632 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1634 sysfs_remove_link(&rdev->kobj, "block");
1635 sysfs_put(rdev->sysfs_state);
1636 rdev->sysfs_state = NULL;
1637 /* We need to delay this, otherwise we can deadlock when
1638 * writing to 'remove' to "dev/state". We also need
1639 * to delay it due to rcu usage.
1642 INIT_WORK(&rdev->del_work, md_delayed_delete);
1643 kobject_get(&rdev->kobj);
1644 schedule_work(&rdev->del_work);
1648 * prevent the device from being mounted, repartitioned or
1649 * otherwise reused by a RAID array (or any other kernel
1650 * subsystem), by bd_claiming the device.
1652 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1655 struct block_device *bdev;
1656 char b[BDEVNAME_SIZE];
1658 bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1660 printk(KERN_ERR "md: could not open %s.\n",
1661 __bdevname(dev, b));
1662 return PTR_ERR(bdev);
1664 err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1666 printk(KERN_ERR "md: could not bd_claim %s.\n",
1668 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1672 set_bit(AllReserved, &rdev->flags);
1677 static void unlock_rdev(mdk_rdev_t *rdev)
1679 struct block_device *bdev = rdev->bdev;
1684 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1687 void md_autodetect_dev(dev_t dev);
1689 static void export_rdev(mdk_rdev_t * rdev)
1691 char b[BDEVNAME_SIZE];
1692 printk(KERN_INFO "md: export_rdev(%s)\n",
1693 bdevname(rdev->bdev,b));
1698 if (test_bit(AutoDetected, &rdev->flags))
1699 md_autodetect_dev(rdev->bdev->bd_dev);
1702 kobject_put(&rdev->kobj);
1705 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1707 unbind_rdev_from_array(rdev);
1711 static void export_array(mddev_t *mddev)
1713 mdk_rdev_t *rdev, *tmp;
1715 rdev_for_each(rdev, tmp, mddev) {
1720 kick_rdev_from_array(rdev);
1722 if (!list_empty(&mddev->disks))
1724 mddev->raid_disks = 0;
1725 mddev->major_version = 0;
1728 static void print_desc(mdp_disk_t *desc)
1730 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1731 desc->major,desc->minor,desc->raid_disk,desc->state);
1734 static void print_sb_90(mdp_super_t *sb)
1739 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1740 sb->major_version, sb->minor_version, sb->patch_version,
1741 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1743 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1744 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1745 sb->md_minor, sb->layout, sb->chunk_size);
1746 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
1747 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1748 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1749 sb->failed_disks, sb->spare_disks,
1750 sb->sb_csum, (unsigned long)sb->events_lo);
1753 for (i = 0; i < MD_SB_DISKS; i++) {
1756 desc = sb->disks + i;
1757 if (desc->number || desc->major || desc->minor ||
1758 desc->raid_disk || (desc->state && (desc->state != 4))) {
1759 printk(" D %2d: ", i);
1763 printk(KERN_INFO "md: THIS: ");
1764 print_desc(&sb->this_disk);
1767 static void print_sb_1(struct mdp_superblock_1 *sb)
1771 uuid = sb->set_uuid;
1772 printk(KERN_INFO "md: SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1773 ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1774 KERN_INFO "md: Name: \"%s\" CT:%llu\n",
1775 le32_to_cpu(sb->major_version),
1776 le32_to_cpu(sb->feature_map),
1777 uuid[0], uuid[1], uuid[2], uuid[3],
1778 uuid[4], uuid[5], uuid[6], uuid[7],
1779 uuid[8], uuid[9], uuid[10], uuid[11],
1780 uuid[12], uuid[13], uuid[14], uuid[15],
1782 (unsigned long long)le64_to_cpu(sb->ctime)
1783 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1785 uuid = sb->device_uuid;
1786 printk(KERN_INFO "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1788 KERN_INFO "md: Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1789 ":%02x%02x%02x%02x%02x%02x\n"
1790 KERN_INFO "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1791 KERN_INFO "md: (MaxDev:%u) \n",
1792 le32_to_cpu(sb->level),
1793 (unsigned long long)le64_to_cpu(sb->size),
1794 le32_to_cpu(sb->raid_disks),
1795 le32_to_cpu(sb->layout),
1796 le32_to_cpu(sb->chunksize),
1797 (unsigned long long)le64_to_cpu(sb->data_offset),
1798 (unsigned long long)le64_to_cpu(sb->data_size),
1799 (unsigned long long)le64_to_cpu(sb->super_offset),
1800 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1801 le32_to_cpu(sb->dev_number),
1802 uuid[0], uuid[1], uuid[2], uuid[3],
1803 uuid[4], uuid[5], uuid[6], uuid[7],
1804 uuid[8], uuid[9], uuid[10], uuid[11],
1805 uuid[12], uuid[13], uuid[14], uuid[15],
1807 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1808 (unsigned long long)le64_to_cpu(sb->events),
1809 (unsigned long long)le64_to_cpu(sb->resync_offset),
1810 le32_to_cpu(sb->sb_csum),
1811 le32_to_cpu(sb->max_dev)
1815 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1817 char b[BDEVNAME_SIZE];
1818 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1819 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1820 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1822 if (rdev->sb_loaded) {
1823 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1824 switch (major_version) {
1826 print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1829 print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1833 printk(KERN_INFO "md: no rdev superblock!\n");
1836 static void md_print_devices(void)
1838 struct list_head *tmp;
1841 char b[BDEVNAME_SIZE];
1844 printk("md: **********************************\n");
1845 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
1846 printk("md: **********************************\n");
1847 for_each_mddev(mddev, tmp) {
1850 bitmap_print_sb(mddev->bitmap);
1852 printk("%s: ", mdname(mddev));
1853 list_for_each_entry(rdev, &mddev->disks, same_set)
1854 printk("<%s>", bdevname(rdev->bdev,b));
1857 list_for_each_entry(rdev, &mddev->disks, same_set)
1858 print_rdev(rdev, mddev->major_version);
1860 printk("md: **********************************\n");
1865 static void sync_sbs(mddev_t * mddev, int nospares)
1867 /* Update each superblock (in-memory image), but
1868 * if we are allowed to, skip spares which already
1869 * have the right event counter, or have one earlier
1870 * (which would mean they aren't being marked as dirty
1871 * with the rest of the array)
1875 list_for_each_entry(rdev, &mddev->disks, same_set) {
1876 if (rdev->sb_events == mddev->events ||
1878 rdev->raid_disk < 0 &&
1879 (rdev->sb_events&1)==0 &&
1880 rdev->sb_events+1 == mddev->events)) {
1881 /* Don't update this superblock */
1882 rdev->sb_loaded = 2;
1884 super_types[mddev->major_version].
1885 sync_super(mddev, rdev);
1886 rdev->sb_loaded = 1;
1891 static void md_update_sb(mddev_t * mddev, int force_change)
1897 if (mddev->external)
1900 spin_lock_irq(&mddev->write_lock);
1902 set_bit(MD_CHANGE_PENDING, &mddev->flags);
1903 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1905 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1906 /* just a clean<-> dirty transition, possibly leave spares alone,
1907 * though if events isn't the right even/odd, we will have to do
1913 if (mddev->degraded)
1914 /* If the array is degraded, then skipping spares is both
1915 * dangerous and fairly pointless.
1916 * Dangerous because a device that was removed from the array
1917 * might have a event_count that still looks up-to-date,
1918 * so it can be re-added without a resync.
1919 * Pointless because if there are any spares to skip,
1920 * then a recovery will happen and soon that array won't
1921 * be degraded any more and the spare can go back to sleep then.
1925 sync_req = mddev->in_sync;
1926 mddev->utime = get_seconds();
1928 /* If this is just a dirty<->clean transition, and the array is clean
1929 * and 'events' is odd, we can roll back to the previous clean state */
1931 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1932 && (mddev->events & 1)
1933 && mddev->events != 1)
1936 /* otherwise we have to go forward and ... */
1938 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1939 /* .. if the array isn't clean, insist on an odd 'events' */
1940 if ((mddev->events&1)==0) {
1945 /* otherwise insist on an even 'events' (for clean states) */
1946 if ((mddev->events&1)) {
1953 if (!mddev->events) {
1955 * oops, this 64-bit counter should never wrap.
1956 * Either we are in around ~1 trillion A.C., assuming
1957 * 1 reboot per second, or we have a bug:
1964 * do not write anything to disk if using
1965 * nonpersistent superblocks
1967 if (!mddev->persistent) {
1968 if (!mddev->external)
1969 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1971 spin_unlock_irq(&mddev->write_lock);
1972 wake_up(&mddev->sb_wait);
1975 sync_sbs(mddev, nospares);
1976 spin_unlock_irq(&mddev->write_lock);
1979 "md: updating %s RAID superblock on device (in sync %d)\n",
1980 mdname(mddev),mddev->in_sync);
1982 bitmap_update_sb(mddev->bitmap);
1983 list_for_each_entry(rdev, &mddev->disks, same_set) {
1984 char b[BDEVNAME_SIZE];
1985 dprintk(KERN_INFO "md: ");
1986 if (rdev->sb_loaded != 1)
1987 continue; /* no noise on spare devices */
1988 if (test_bit(Faulty, &rdev->flags))
1989 dprintk("(skipping faulty ");
1991 dprintk("%s ", bdevname(rdev->bdev,b));
1992 if (!test_bit(Faulty, &rdev->flags)) {
1993 md_super_write(mddev,rdev,
1994 rdev->sb_start, rdev->sb_size,
1996 dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1997 bdevname(rdev->bdev,b),
1998 (unsigned long long)rdev->sb_start);
1999 rdev->sb_events = mddev->events;
2003 if (mddev->level == LEVEL_MULTIPATH)
2004 /* only need to write one superblock... */
2007 md_super_wait(mddev);
2008 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2010 spin_lock_irq(&mddev->write_lock);
2011 if (mddev->in_sync != sync_req ||
2012 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2013 /* have to write it out again */
2014 spin_unlock_irq(&mddev->write_lock);
2017 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2018 spin_unlock_irq(&mddev->write_lock);
2019 wake_up(&mddev->sb_wait);
2023 /* words written to sysfs files may, or may not, be \n terminated.
2024 * We want to accept with case. For this we use cmd_match.
2026 static int cmd_match(const char *cmd, const char *str)
2028 /* See if cmd, written into a sysfs file, matches
2029 * str. They must either be the same, or cmd can
2030 * have a trailing newline
2032 while (*cmd && *str && *cmd == *str) {
2043 struct rdev_sysfs_entry {
2044 struct attribute attr;
2045 ssize_t (*show)(mdk_rdev_t *, char *);
2046 ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2050 state_show(mdk_rdev_t *rdev, char *page)
2055 if (test_bit(Faulty, &rdev->flags)) {
2056 len+= sprintf(page+len, "%sfaulty",sep);
2059 if (test_bit(In_sync, &rdev->flags)) {
2060 len += sprintf(page+len, "%sin_sync",sep);
2063 if (test_bit(WriteMostly, &rdev->flags)) {
2064 len += sprintf(page+len, "%swrite_mostly",sep);
2067 if (test_bit(Blocked, &rdev->flags)) {
2068 len += sprintf(page+len, "%sblocked", sep);
2071 if (!test_bit(Faulty, &rdev->flags) &&
2072 !test_bit(In_sync, &rdev->flags)) {
2073 len += sprintf(page+len, "%sspare", sep);
2076 return len+sprintf(page+len, "\n");
2080 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2083 * faulty - simulates and error
2084 * remove - disconnects the device
2085 * writemostly - sets write_mostly
2086 * -writemostly - clears write_mostly
2087 * blocked - sets the Blocked flag
2088 * -blocked - clears the Blocked flag
2091 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2092 md_error(rdev->mddev, rdev);
2094 } else if (cmd_match(buf, "remove")) {
2095 if (rdev->raid_disk >= 0)
2098 mddev_t *mddev = rdev->mddev;
2099 kick_rdev_from_array(rdev);
2101 md_update_sb(mddev, 1);
2102 md_new_event(mddev);
2105 } else if (cmd_match(buf, "writemostly")) {
2106 set_bit(WriteMostly, &rdev->flags);
2108 } else if (cmd_match(buf, "-writemostly")) {
2109 clear_bit(WriteMostly, &rdev->flags);
2111 } else if (cmd_match(buf, "blocked")) {
2112 set_bit(Blocked, &rdev->flags);
2114 } else if (cmd_match(buf, "-blocked")) {
2115 clear_bit(Blocked, &rdev->flags);
2116 wake_up(&rdev->blocked_wait);
2117 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2118 md_wakeup_thread(rdev->mddev->thread);
2122 if (!err && rdev->sysfs_state)
2123 sysfs_notify_dirent(rdev->sysfs_state);
2124 return err ? err : len;
2126 static struct rdev_sysfs_entry rdev_state =
2127 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2130 errors_show(mdk_rdev_t *rdev, char *page)
2132 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2136 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2139 unsigned long n = simple_strtoul(buf, &e, 10);
2140 if (*buf && (*e == 0 || *e == '\n')) {
2141 atomic_set(&rdev->corrected_errors, n);
2146 static struct rdev_sysfs_entry rdev_errors =
2147 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2150 slot_show(mdk_rdev_t *rdev, char *page)
2152 if (rdev->raid_disk < 0)
2153 return sprintf(page, "none\n");
2155 return sprintf(page, "%d\n", rdev->raid_disk);
2159 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2164 int slot = simple_strtoul(buf, &e, 10);
2165 if (strncmp(buf, "none", 4)==0)
2167 else if (e==buf || (*e && *e!= '\n'))
2169 if (rdev->mddev->pers && slot == -1) {
2170 /* Setting 'slot' on an active array requires also
2171 * updating the 'rd%d' link, and communicating
2172 * with the personality with ->hot_*_disk.
2173 * For now we only support removing
2174 * failed/spare devices. This normally happens automatically,
2175 * but not when the metadata is externally managed.
2177 if (rdev->raid_disk == -1)
2179 /* personality does all needed checks */
2180 if (rdev->mddev->pers->hot_add_disk == NULL)
2182 err = rdev->mddev->pers->
2183 hot_remove_disk(rdev->mddev, rdev->raid_disk);
2186 sprintf(nm, "rd%d", rdev->raid_disk);
2187 sysfs_remove_link(&rdev->mddev->kobj, nm);
2188 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2189 md_wakeup_thread(rdev->mddev->thread);
2190 } else if (rdev->mddev->pers) {
2192 /* Activating a spare .. or possibly reactivating
2193 * if we every get bitmaps working here.
2196 if (rdev->raid_disk != -1)
2199 if (rdev->mddev->pers->hot_add_disk == NULL)
2202 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2203 if (rdev2->raid_disk == slot)
2206 rdev->raid_disk = slot;
2207 if (test_bit(In_sync, &rdev->flags))
2208 rdev->saved_raid_disk = slot;
2210 rdev->saved_raid_disk = -1;
2211 err = rdev->mddev->pers->
2212 hot_add_disk(rdev->mddev, rdev);
2214 rdev->raid_disk = -1;
2217 sysfs_notify_dirent(rdev->sysfs_state);
2218 sprintf(nm, "rd%d", rdev->raid_disk);
2219 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2221 "md: cannot register "
2223 nm, mdname(rdev->mddev));
2225 /* don't wakeup anyone, leave that to userspace. */
2227 if (slot >= rdev->mddev->raid_disks)
2229 rdev->raid_disk = slot;
2230 /* assume it is working */
2231 clear_bit(Faulty, &rdev->flags);
2232 clear_bit(WriteMostly, &rdev->flags);
2233 set_bit(In_sync, &rdev->flags);
2234 sysfs_notify_dirent(rdev->sysfs_state);
2240 static struct rdev_sysfs_entry rdev_slot =
2241 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2244 offset_show(mdk_rdev_t *rdev, char *page)
2246 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2250 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2253 unsigned long long offset = simple_strtoull(buf, &e, 10);
2254 if (e==buf || (*e && *e != '\n'))
2256 if (rdev->mddev->pers && rdev->raid_disk >= 0)
2258 if (rdev->sectors && rdev->mddev->external)
2259 /* Must set offset before size, so overlap checks
2262 rdev->data_offset = offset;
2266 static struct rdev_sysfs_entry rdev_offset =
2267 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2270 rdev_size_show(mdk_rdev_t *rdev, char *page)
2272 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2275 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2277 /* check if two start/length pairs overlap */
2286 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2288 mddev_t *my_mddev = rdev->mddev;
2289 sector_t oldsectors = rdev->sectors;
2290 unsigned long long sectors;
2292 if (strict_strtoull(buf, 10, §ors) < 0)
2295 if (my_mddev->pers && rdev->raid_disk >= 0) {
2296 if (my_mddev->persistent) {
2297 sectors = super_types[my_mddev->major_version].
2298 rdev_size_change(rdev, sectors);
2301 } else if (!sectors)
2302 sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2305 if (sectors < my_mddev->dev_sectors)
2306 return -EINVAL; /* component must fit device */
2308 rdev->sectors = sectors;
2309 if (sectors > oldsectors && my_mddev->external) {
2310 /* need to check that all other rdevs with the same ->bdev
2311 * do not overlap. We need to unlock the mddev to avoid
2312 * a deadlock. We have already changed rdev->sectors, and if
2313 * we have to change it back, we will have the lock again.
2317 struct list_head *tmp;
2319 mddev_unlock(my_mddev);
2320 for_each_mddev(mddev, tmp) {
2324 list_for_each_entry(rdev2, &mddev->disks, same_set)
2325 if (test_bit(AllReserved, &rdev2->flags) ||
2326 (rdev->bdev == rdev2->bdev &&
2328 overlaps(rdev->data_offset, rdev->sectors,
2334 mddev_unlock(mddev);
2340 mddev_lock(my_mddev);
2342 /* Someone else could have slipped in a size
2343 * change here, but doing so is just silly.
2344 * We put oldsectors back because we *know* it is
2345 * safe, and trust userspace not to race with
2348 rdev->sectors = oldsectors;
2355 static struct rdev_sysfs_entry rdev_size =
2356 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2358 static struct attribute *rdev_default_attrs[] = {
2367 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2369 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2370 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2371 mddev_t *mddev = rdev->mddev;
2377 rv = mddev ? mddev_lock(mddev) : -EBUSY;
2379 if (rdev->mddev == NULL)
2382 rv = entry->show(rdev, page);
2383 mddev_unlock(mddev);
2389 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2390 const char *page, size_t length)
2392 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2393 mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2395 mddev_t *mddev = rdev->mddev;
2399 if (!capable(CAP_SYS_ADMIN))
2401 rv = mddev ? mddev_lock(mddev): -EBUSY;
2403 if (rdev->mddev == NULL)
2406 rv = entry->store(rdev, page, length);
2407 mddev_unlock(mddev);
2412 static void rdev_free(struct kobject *ko)
2414 mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2417 static struct sysfs_ops rdev_sysfs_ops = {
2418 .show = rdev_attr_show,
2419 .store = rdev_attr_store,
2421 static struct kobj_type rdev_ktype = {
2422 .release = rdev_free,
2423 .sysfs_ops = &rdev_sysfs_ops,
2424 .default_attrs = rdev_default_attrs,
2428 * Import a device. If 'super_format' >= 0, then sanity check the superblock
2430 * mark the device faulty if:
2432 * - the device is nonexistent (zero size)
2433 * - the device has no valid superblock
2435 * a faulty rdev _never_ has rdev->sb set.
2437 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2439 char b[BDEVNAME_SIZE];
2444 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2446 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2447 return ERR_PTR(-ENOMEM);
2450 if ((err = alloc_disk_sb(rdev)))
2453 err = lock_rdev(rdev, newdev, super_format == -2);
2457 kobject_init(&rdev->kobj, &rdev_ktype);
2460 rdev->saved_raid_disk = -1;
2461 rdev->raid_disk = -1;
2463 rdev->data_offset = 0;
2464 rdev->sb_events = 0;
2465 atomic_set(&rdev->nr_pending, 0);
2466 atomic_set(&rdev->read_errors, 0);
2467 atomic_set(&rdev->corrected_errors, 0);
2469 size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2472 "md: %s has zero or unknown size, marking faulty!\n",
2473 bdevname(rdev->bdev,b));
2478 if (super_format >= 0) {
2479 err = super_types[super_format].
2480 load_super(rdev, NULL, super_minor);
2481 if (err == -EINVAL) {
2483 "md: %s does not have a valid v%d.%d "
2484 "superblock, not importing!\n",
2485 bdevname(rdev->bdev,b),
2486 super_format, super_minor);
2491 "md: could not read %s's sb, not importing!\n",
2492 bdevname(rdev->bdev,b));
2497 INIT_LIST_HEAD(&rdev->same_set);
2498 init_waitqueue_head(&rdev->blocked_wait);
2503 if (rdev->sb_page) {
2509 return ERR_PTR(err);
2513 * Check a full RAID array for plausibility
2517 static void analyze_sbs(mddev_t * mddev)
2520 mdk_rdev_t *rdev, *freshest, *tmp;
2521 char b[BDEVNAME_SIZE];
2524 rdev_for_each(rdev, tmp, mddev)
2525 switch (super_types[mddev->major_version].
2526 load_super(rdev, freshest, mddev->minor_version)) {
2534 "md: fatal superblock inconsistency in %s"
2535 " -- removing from array\n",
2536 bdevname(rdev->bdev,b));
2537 kick_rdev_from_array(rdev);
2541 super_types[mddev->major_version].
2542 validate_super(mddev, freshest);
2545 rdev_for_each(rdev, tmp, mddev) {
2546 if (rdev->desc_nr >= mddev->max_disks ||
2547 i > mddev->max_disks) {
2549 "md: %s: %s: only %d devices permitted\n",
2550 mdname(mddev), bdevname(rdev->bdev, b),
2552 kick_rdev_from_array(rdev);
2555 if (rdev != freshest)
2556 if (super_types[mddev->major_version].
2557 validate_super(mddev, rdev)) {
2558 printk(KERN_WARNING "md: kicking non-fresh %s"
2560 bdevname(rdev->bdev,b));
2561 kick_rdev_from_array(rdev);
2564 if (mddev->level == LEVEL_MULTIPATH) {
2565 rdev->desc_nr = i++;
2566 rdev->raid_disk = rdev->desc_nr;
2567 set_bit(In_sync, &rdev->flags);
2568 } else if (rdev->raid_disk >= mddev->raid_disks) {
2569 rdev->raid_disk = -1;
2570 clear_bit(In_sync, &rdev->flags);
2576 if (mddev->recovery_cp != MaxSector &&
2578 printk(KERN_ERR "md: %s: raid array is not clean"
2579 " -- starting background reconstruction\n",
2584 static void md_safemode_timeout(unsigned long data);
2587 safe_delay_show(mddev_t *mddev, char *page)
2589 int msec = (mddev->safemode_delay*1000)/HZ;
2590 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2593 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2601 /* remove a period, and count digits after it */
2602 if (len >= sizeof(buf))
2604 strlcpy(buf, cbuf, sizeof(buf));
2605 for (i=0; i<len; i++) {
2607 if (isdigit(buf[i])) {
2612 } else if (buf[i] == '.') {
2617 if (strict_strtoul(buf, 10, &msec) < 0)
2619 msec = (msec * 1000) / scale;
2621 mddev->safemode_delay = 0;
2623 unsigned long old_delay = mddev->safemode_delay;
2624 mddev->safemode_delay = (msec*HZ)/1000;
2625 if (mddev->safemode_delay == 0)
2626 mddev->safemode_delay = 1;
2627 if (mddev->safemode_delay < old_delay)
2628 md_safemode_timeout((unsigned long)mddev);
2632 static struct md_sysfs_entry md_safe_delay =
2633 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2636 level_show(mddev_t *mddev, char *page)
2638 struct mdk_personality *p = mddev->pers;
2640 return sprintf(page, "%s\n", p->name);
2641 else if (mddev->clevel[0])
2642 return sprintf(page, "%s\n", mddev->clevel);
2643 else if (mddev->level != LEVEL_NONE)
2644 return sprintf(page, "%d\n", mddev->level);
2650 level_store(mddev_t *mddev, const char *buf, size_t len)
2654 struct mdk_personality *pers;
2657 if (mddev->pers == NULL) {
2660 if (len >= sizeof(mddev->clevel))
2662 strncpy(mddev->clevel, buf, len);
2663 if (mddev->clevel[len-1] == '\n')
2665 mddev->clevel[len] = 0;
2666 mddev->level = LEVEL_NONE;
2670 /* request to change the personality. Need to ensure:
2671 * - array is not engaged in resync/recovery/reshape
2672 * - old personality can be suspended
2673 * - new personality will access other array.
2676 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2679 if (!mddev->pers->quiesce) {
2680 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2681 mdname(mddev), mddev->pers->name);
2685 /* Now find the new personality */
2686 if (len == 0 || len >= sizeof(level))
2688 strncpy(level, buf, len);
2689 if (level[len-1] == '\n')
2693 request_module("md-%s", level);
2694 spin_lock(&pers_lock);
2695 pers = find_pers(LEVEL_NONE, level);
2696 if (!pers || !try_module_get(pers->owner)) {
2697 spin_unlock(&pers_lock);
2698 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2701 spin_unlock(&pers_lock);
2703 if (pers == mddev->pers) {
2704 /* Nothing to do! */
2705 module_put(pers->owner);
2708 if (!pers->takeover) {
2709 module_put(pers->owner);
2710 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2711 mdname(mddev), level);
2715 /* ->takeover must set new_* and/or delta_disks
2716 * if it succeeds, and may set them when it fails.
2718 priv = pers->takeover(mddev);
2720 mddev->new_level = mddev->level;
2721 mddev->new_layout = mddev->layout;
2722 mddev->new_chunk = mddev->chunk_size;
2723 mddev->raid_disks -= mddev->delta_disks;
2724 mddev->delta_disks = 0;
2725 module_put(pers->owner);
2726 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2727 mdname(mddev), level);
2728 return PTR_ERR(priv);
2731 /* Looks like we have a winner */
2732 mddev_suspend(mddev);
2733 mddev->pers->stop(mddev);
2734 module_put(mddev->pers->owner);
2736 mddev->private = priv;
2737 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2738 mddev->level = mddev->new_level;
2739 mddev->layout = mddev->new_layout;
2740 mddev->chunk_size = mddev->new_chunk;
2741 mddev->delta_disks = 0;
2743 mddev_resume(mddev);
2744 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2745 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2746 md_wakeup_thread(mddev->thread);
2750 static struct md_sysfs_entry md_level =
2751 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2755 layout_show(mddev_t *mddev, char *page)
2757 /* just a number, not meaningful for all levels */
2758 if (mddev->reshape_position != MaxSector &&
2759 mddev->layout != mddev->new_layout)
2760 return sprintf(page, "%d (%d)\n",
2761 mddev->new_layout, mddev->layout);
2762 return sprintf(page, "%d\n", mddev->layout);
2766 layout_store(mddev_t *mddev, const char *buf, size_t len)
2769 unsigned long n = simple_strtoul(buf, &e, 10);
2771 if (!*buf || (*e && *e != '\n'))
2776 if (mddev->pers->reconfig == NULL)
2778 err = mddev->pers->reconfig(mddev, n, -1);
2782 mddev->new_layout = n;
2783 if (mddev->reshape_position == MaxSector)
2788 static struct md_sysfs_entry md_layout =
2789 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2793 raid_disks_show(mddev_t *mddev, char *page)
2795 if (mddev->raid_disks == 0)
2797 if (mddev->reshape_position != MaxSector &&
2798 mddev->delta_disks != 0)
2799 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2800 mddev->raid_disks - mddev->delta_disks);
2801 return sprintf(page, "%d\n", mddev->raid_disks);
2804 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2807 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2811 unsigned long n = simple_strtoul(buf, &e, 10);
2813 if (!*buf || (*e && *e != '\n'))
2817 rv = update_raid_disks(mddev, n);
2818 else if (mddev->reshape_position != MaxSector) {
2819 int olddisks = mddev->raid_disks - mddev->delta_disks;
2820 mddev->delta_disks = n - olddisks;
2821 mddev->raid_disks = n;
2823 mddev->raid_disks = n;
2824 return rv ? rv : len;
2826 static struct md_sysfs_entry md_raid_disks =
2827 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2830 chunk_size_show(mddev_t *mddev, char *page)
2832 if (mddev->reshape_position != MaxSector &&
2833 mddev->chunk_size != mddev->new_chunk)
2834 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2836 return sprintf(page, "%d\n", mddev->chunk_size);
2840 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2843 unsigned long n = simple_strtoul(buf, &e, 10);
2845 if (!*buf || (*e && *e != '\n'))
2850 if (mddev->pers->reconfig == NULL)
2852 err = mddev->pers->reconfig(mddev, -1, n);
2856 mddev->new_chunk = n;
2857 if (mddev->reshape_position == MaxSector)
2858 mddev->chunk_size = n;
2862 static struct md_sysfs_entry md_chunk_size =
2863 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2866 resync_start_show(mddev_t *mddev, char *page)
2868 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2872 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2875 unsigned long long n = simple_strtoull(buf, &e, 10);
2879 if (!*buf || (*e && *e != '\n'))
2882 mddev->recovery_cp = n;
2885 static struct md_sysfs_entry md_resync_start =
2886 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2889 * The array state can be:
2892 * No devices, no size, no level
2893 * Equivalent to STOP_ARRAY ioctl
2895 * May have some settings, but array is not active
2896 * all IO results in error
2897 * When written, doesn't tear down array, but just stops it
2898 * suspended (not supported yet)
2899 * All IO requests will block. The array can be reconfigured.
2900 * Writing this, if accepted, will block until array is quiescent
2902 * no resync can happen. no superblocks get written.
2903 * write requests fail
2905 * like readonly, but behaves like 'clean' on a write request.
2907 * clean - no pending writes, but otherwise active.
2908 * When written to inactive array, starts without resync
2909 * If a write request arrives then
2910 * if metadata is known, mark 'dirty' and switch to 'active'.
2911 * if not known, block and switch to write-pending
2912 * If written to an active array that has pending writes, then fails.
2914 * fully active: IO and resync can be happening.
2915 * When written to inactive array, starts with resync
2918 * clean, but writes are blocked waiting for 'active' to be written.
2921 * like active, but no writes have been seen for a while (100msec).
2924 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2925 write_pending, active_idle, bad_word};
2926 static char *array_states[] = {
2927 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2928 "write-pending", "active-idle", NULL };
2930 static int match_word(const char *word, char **list)
2933 for (n=0; list[n]; n++)
2934 if (cmd_match(word, list[n]))
2940 array_state_show(mddev_t *mddev, char *page)
2942 enum array_state st = inactive;
2955 else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2957 else if (mddev->safemode)
2963 if (list_empty(&mddev->disks) &&
2964 mddev->raid_disks == 0 &&
2965 mddev->dev_sectors == 0)
2970 return sprintf(page, "%s\n", array_states[st]);
2973 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2974 static int do_md_run(mddev_t * mddev);
2975 static int restart_array(mddev_t *mddev);
2978 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2981 enum array_state st = match_word(buf, array_states);
2986 /* stopping an active array */
2987 if (atomic_read(&mddev->openers) > 0)
2989 err = do_md_stop(mddev, 0, 0);
2992 /* stopping an active array */
2994 if (atomic_read(&mddev->openers) > 0)
2996 err = do_md_stop(mddev, 2, 0);
2998 err = 0; /* already inactive */
3001 break; /* not supported yet */
3004 err = do_md_stop(mddev, 1, 0);
3007 set_disk_ro(mddev->gendisk, 1);
3008 err = do_md_run(mddev);
3014 err = do_md_stop(mddev, 1, 0);
3015 else if (mddev->ro == 1)
3016 err = restart_array(mddev);
3019 set_disk_ro(mddev->gendisk, 0);
3023 err = do_md_run(mddev);
3028 restart_array(mddev);
3029 spin_lock_irq(&mddev->write_lock);
3030 if (atomic_read(&mddev->writes_pending) == 0) {
3031 if (mddev->in_sync == 0) {
3033 if (mddev->safemode == 1)
3034 mddev->safemode = 0;
3035 if (mddev->persistent)
3036 set_bit(MD_CHANGE_CLEAN,
3042 spin_unlock_irq(&mddev->write_lock);
3045 mddev->recovery_cp = MaxSector;
3046 err = do_md_run(mddev);
3051 restart_array(mddev);
3052 if (mddev->external)
3053 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3054 wake_up(&mddev->sb_wait);
3058 set_disk_ro(mddev->gendisk, 0);
3059 err = do_md_run(mddev);
3064 /* these cannot be set */
3070 sysfs_notify_dirent(mddev->sysfs_state);
3074 static struct md_sysfs_entry md_array_state =
3075 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3078 null_show(mddev_t *mddev, char *page)
3084 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3086 /* buf must be %d:%d\n? giving major and minor numbers */
3087 /* The new device is added to the array.
3088 * If the array has a persistent superblock, we read the
3089 * superblock to initialise info and check validity.
3090 * Otherwise, only checking done is that in bind_rdev_to_array,
3091 * which mainly checks size.
3094 int major = simple_strtoul(buf, &e, 10);
3100 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3102 minor = simple_strtoul(e+1, &e, 10);
3103 if (*e && *e != '\n')
3105 dev = MKDEV(major, minor);
3106 if (major != MAJOR(dev) ||
3107 minor != MINOR(dev))
3111 if (mddev->persistent) {
3112 rdev = md_import_device(dev, mddev->major_version,
3113 mddev->minor_version);
3114 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3115 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3116 mdk_rdev_t, same_set);
3117 err = super_types[mddev->major_version]
3118 .load_super(rdev, rdev0, mddev->minor_version);
3122 } else if (mddev->external)
3123 rdev = md_import_device(dev, -2, -1);
3125 rdev = md_import_device(dev, -1, -1);
3128 return PTR_ERR(rdev);
3129 err = bind_rdev_to_array(rdev, mddev);
3133 return err ? err : len;
3136 static struct md_sysfs_entry md_new_device =
3137 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3140 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3143 unsigned long chunk, end_chunk;
3147 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3149 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3150 if (buf == end) break;
3151 if (*end == '-') { /* range */
3153 end_chunk = simple_strtoul(buf, &end, 0);
3154 if (buf == end) break;
3156 if (*end && !isspace(*end)) break;
3157 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3159 while (isspace(*buf)) buf++;
3161 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3166 static struct md_sysfs_entry md_bitmap =
3167 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3170 size_show(mddev_t *mddev, char *page)
3172 return sprintf(page, "%llu\n",
3173 (unsigned long long)mddev->dev_sectors / 2);
3176 static int update_size(mddev_t *mddev, sector_t num_sectors);
3179 size_store(mddev_t *mddev, const char *buf, size_t len)
3181 /* If array is inactive, we can reduce the component size, but
3182 * not increase it (except from 0).
3183 * If array is active, we can try an on-line resize
3185 unsigned long long sectors;
3186 int err = strict_strtoull(buf, 10, §ors);
3192 err = update_size(mddev, sectors);
3193 md_update_sb(mddev, 1);
3195 if (mddev->dev_sectors == 0 ||
3196 mddev->dev_sectors > sectors)
3197 mddev->dev_sectors = sectors;
3201 return err ? err : len;
3204 static struct md_sysfs_entry md_size =
3205 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3210 * 'none' for arrays with no metadata (good luck...)
3211 * 'external' for arrays with externally managed metadata,
3212 * or N.M for internally known formats
3215 metadata_show(mddev_t *mddev, char *page)
3217 if (mddev->persistent)
3218 return sprintf(page, "%d.%d\n",
3219 mddev->major_version, mddev->minor_version);
3220 else if (mddev->external)
3221 return sprintf(page, "external:%s\n", mddev->metadata_type);
3223 return sprintf(page, "none\n");
3227 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3231 /* Changing the details of 'external' metadata is
3232 * always permitted. Otherwise there must be
3233 * no devices attached to the array.
3235 if (mddev->external && strncmp(buf, "external:", 9) == 0)
3237 else if (!list_empty(&mddev->disks))
3240 if (cmd_match(buf, "none")) {
3241 mddev->persistent = 0;
3242 mddev->external = 0;
3243 mddev->major_version = 0;
3244 mddev->minor_version = 90;
3247 if (strncmp(buf, "external:", 9) == 0) {
3248 size_t namelen = len-9;
3249 if (namelen >= sizeof(mddev->metadata_type))
3250 namelen = sizeof(mddev->metadata_type)-1;
3251 strncpy(mddev->metadata_type, buf+9, namelen);
3252 mddev->metadata_type[namelen] = 0;
3253 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3254 mddev->metadata_type[--namelen] = 0;
3255 mddev->persistent = 0;
3256 mddev->external = 1;
3257 mddev->major_version = 0;
3258 mddev->minor_version = 90;
3261 major = simple_strtoul(buf, &e, 10);
3262 if (e==buf || *e != '.')
3265 minor = simple_strtoul(buf, &e, 10);
3266 if (e==buf || (*e && *e != '\n') )
3268 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3270 mddev->major_version = major;
3271 mddev->minor_version = minor;
3272 mddev->persistent = 1;
3273 mddev->external = 0;
3277 static struct md_sysfs_entry md_metadata =
3278 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3281 action_show(mddev_t *mddev, char *page)
3283 char *type = "idle";
3284 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3285 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3286 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3288 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3289 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3291 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3295 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3298 return sprintf(page, "%s\n", type);
3302 action_store(mddev_t *mddev, const char *page, size_t len)
3304 if (!mddev->pers || !mddev->pers->sync_request)
3307 if (cmd_match(page, "idle")) {
3308 if (mddev->sync_thread) {
3309 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3310 md_unregister_thread(mddev->sync_thread);
3311 mddev->sync_thread = NULL;
3312 mddev->recovery = 0;
3314 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3315 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3317 else if (cmd_match(page, "resync"))
3318 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3319 else if (cmd_match(page, "recover")) {
3320 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3321 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3322 } else if (cmd_match(page, "reshape")) {
3324 if (mddev->pers->start_reshape == NULL)
3326 err = mddev->pers->start_reshape(mddev);
3329 sysfs_notify(&mddev->kobj, NULL, "degraded");
3331 if (cmd_match(page, "check"))
3332 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3333 else if (!cmd_match(page, "repair"))
3335 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3336 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3338 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3339 md_wakeup_thread(mddev->thread);
3340 sysfs_notify_dirent(mddev->sysfs_action);
3345 mismatch_cnt_show(mddev_t *mddev, char *page)
3347 return sprintf(page, "%llu\n",
3348 (unsigned long long) mddev->resync_mismatches);
3351 static struct md_sysfs_entry md_scan_mode =
3352 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3355 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3358 sync_min_show(mddev_t *mddev, char *page)
3360 return sprintf(page, "%d (%s)\n", speed_min(mddev),
3361 mddev->sync_speed_min ? "local": "system");
3365 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3369 if (strncmp(buf, "system", 6)==0) {
3370 mddev->sync_speed_min = 0;
3373 min = simple_strtoul(buf, &e, 10);
3374 if (buf == e || (*e && *e != '\n') || min <= 0)
3376 mddev->sync_speed_min = min;
3380 static struct md_sysfs_entry md_sync_min =
3381 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3384 sync_max_show(mddev_t *mddev, char *page)
3386 return sprintf(page, "%d (%s)\n", speed_max(mddev),
3387 mddev->sync_speed_max ? "local": "system");
3391 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3395 if (strncmp(buf, "system", 6)==0) {
3396 mddev->sync_speed_max = 0;
3399 max = simple_strtoul(buf, &e, 10);
3400 if (buf == e || (*e && *e != '\n') || max <= 0)
3402 mddev->sync_speed_max = max;
3406 static struct md_sysfs_entry md_sync_max =
3407 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3410 degraded_show(mddev_t *mddev, char *page)
3412 return sprintf(page, "%d\n", mddev->degraded);
3414 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3417 sync_force_parallel_show(mddev_t *mddev, char *page)
3419 return sprintf(page, "%d\n", mddev->parallel_resync);
3423 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3427 if (strict_strtol(buf, 10, &n))
3430 if (n != 0 && n != 1)
3433 mddev->parallel_resync = n;
3435 if (mddev->sync_thread)
3436 wake_up(&resync_wait);
3441 /* force parallel resync, even with shared block devices */
3442 static struct md_sysfs_entry md_sync_force_parallel =
3443 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3444 sync_force_parallel_show, sync_force_parallel_store);
3447 sync_speed_show(mddev_t *mddev, char *page)
3449 unsigned long resync, dt, db;
3450 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3451 dt = (jiffies - mddev->resync_mark) / HZ;
3453 db = resync - mddev->resync_mark_cnt;
3454 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3457 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3460 sync_completed_show(mddev_t *mddev, char *page)
3462 unsigned long max_sectors, resync;
3464 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3465 max_sectors = mddev->resync_max_sectors;
3467 max_sectors = mddev->dev_sectors;
3469 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3470 return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3473 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3476 min_sync_show(mddev_t *mddev, char *page)
3478 return sprintf(page, "%llu\n",
3479 (unsigned long long)mddev->resync_min);
3482 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3484 unsigned long long min;
3485 if (strict_strtoull(buf, 10, &min))
3487 if (min > mddev->resync_max)
3489 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3492 /* Must be a multiple of chunk_size */
3493 if (mddev->chunk_size) {
3494 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3497 mddev->resync_min = min;
3502 static struct md_sysfs_entry md_min_sync =
3503 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3506 max_sync_show(mddev_t *mddev, char *page)
3508 if (mddev->resync_max == MaxSector)
3509 return sprintf(page, "max\n");
3511 return sprintf(page, "%llu\n",
3512 (unsigned long long)mddev->resync_max);
3515 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3517 if (strncmp(buf, "max", 3) == 0)
3518 mddev->resync_max = MaxSector;
3520 unsigned long long max;
3521 if (strict_strtoull(buf, 10, &max))
3523 if (max < mddev->resync_min)
3525 if (max < mddev->resync_max &&
3526 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3529 /* Must be a multiple of chunk_size */
3530 if (mddev->chunk_size) {
3531 if (max & (sector_t)((mddev->chunk_size>>9)-1))
3534 mddev->resync_max = max;
3536 wake_up(&mddev->recovery_wait);
3540 static struct md_sysfs_entry md_max_sync =
3541 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3544 suspend_lo_show(mddev_t *mddev, char *page)
3546 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3550 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3553 unsigned long long new = simple_strtoull(buf, &e, 10);
3555 if (mddev->pers->quiesce == NULL)
3557 if (buf == e || (*e && *e != '\n'))
3559 if (new >= mddev->suspend_hi ||
3560 (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3561 mddev->suspend_lo = new;
3562 mddev->pers->quiesce(mddev, 2);
3567 static struct md_sysfs_entry md_suspend_lo =
3568 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3572 suspend_hi_show(mddev_t *mddev, char *page)
3574 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3578 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3581 unsigned long long new = simple_strtoull(buf, &e, 10);
3583 if (mddev->pers->quiesce == NULL)
3585 if (buf == e || (*e && *e != '\n'))
3587 if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3588 (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3589 mddev->suspend_hi = new;
3590 mddev->pers->quiesce(mddev, 1);
3591 mddev->pers->quiesce(mddev, 0);
3596 static struct md_sysfs_entry md_suspend_hi =
3597 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3600 reshape_position_show(mddev_t *mddev, char *page)
3602 if (mddev->reshape_position != MaxSector)
3603 return sprintf(page, "%llu\n",
3604 (unsigned long long)mddev->reshape_position);
3605 strcpy(page, "none\n");
3610 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3613 unsigned long long new = simple_strtoull(buf, &e, 10);
3616 if (buf == e || (*e && *e != '\n'))
3618 mddev->reshape_position = new;
3619 mddev->delta_disks = 0;
3620 mddev->new_level = mddev->level;
3621 mddev->new_layout = mddev->layout;
3622 mddev->new_chunk = mddev->chunk_size;
3626 static struct md_sysfs_entry md_reshape_position =
3627 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3628 reshape_position_store);
3631 static struct attribute *md_default_attrs[] = {
3634 &md_raid_disks.attr,
3635 &md_chunk_size.attr,
3637 &md_resync_start.attr,
3639 &md_new_device.attr,
3640 &md_safe_delay.attr,
3641 &md_array_state.attr,
3642 &md_reshape_position.attr,
3646 static struct attribute *md_redundancy_attrs[] = {
3648 &md_mismatches.attr,
3651 &md_sync_speed.attr,
3652 &md_sync_force_parallel.attr,
3653 &md_sync_completed.attr,
3656 &md_suspend_lo.attr,
3657 &md_suspend_hi.attr,
3662 static struct attribute_group md_redundancy_group = {
3664 .attrs = md_redundancy_attrs,
3669 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3671 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3672 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3677 rv = mddev_lock(mddev);
3679 rv = entry->show(mddev, page);
3680 mddev_unlock(mddev);
3686 md_attr_store(struct kobject *kobj, struct attribute *attr,
3687 const char *page, size_t length)
3689 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3690 mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3695 if (!capable(CAP_SYS_ADMIN))
3697 rv = mddev_lock(mddev);
3698 if (mddev->hold_active == UNTIL_IOCTL)
3699 mddev->hold_active = 0;
3701 rv = entry->store(mddev, page, length);
3702 mddev_unlock(mddev);
3707 static void md_free(struct kobject *ko)
3709 mddev_t *mddev = container_of(ko, mddev_t, kobj);
3711 if (mddev->sysfs_state)
3712 sysfs_put(mddev->sysfs_state);
3714 if (mddev->gendisk) {
3715 del_gendisk(mddev->gendisk);
3716 put_disk(mddev->gendisk);
3719 blk_cleanup_queue(mddev->queue);
3724 static struct sysfs_ops md_sysfs_ops = {
3725 .show = md_attr_show,
3726 .store = md_attr_store,
3728 static struct kobj_type md_ktype = {
3730 .sysfs_ops = &md_sysfs_ops,
3731 .default_attrs = md_default_attrs,
3736 static int md_alloc(dev_t dev, char *name)
3738 static DEFINE_MUTEX(disks_mutex);
3739 mddev_t *mddev = mddev_find(dev);
3740 struct gendisk *disk;
3749 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3750 shift = partitioned ? MdpMinorShift : 0;
3751 unit = MINOR(mddev->unit) >> shift;
3753 /* wait for any previous instance if this device
3754 * to be completed removed (mddev_delayed_delete).
3756 flush_scheduled_work();
3758 mutex_lock(&disks_mutex);
3759 if (mddev->gendisk) {
3760 mutex_unlock(&disks_mutex);
3766 /* Need to ensure that 'name' is not a duplicate.
3769 spin_lock(&all_mddevs_lock);
3771 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3772 if (mddev2->gendisk &&
3773 strcmp(mddev2->gendisk->disk_name, name) == 0) {
3774 spin_unlock(&all_mddevs_lock);
3777 spin_unlock(&all_mddevs_lock);
3780 mddev->queue = blk_alloc_queue(GFP_KERNEL);
3781 if (!mddev->queue) {
3782 mutex_unlock(&disks_mutex);
3786 mddev->queue->queuedata = mddev;
3788 /* Can be unlocked because the queue is new: no concurrency */
3789 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3791 blk_queue_make_request(mddev->queue, md_make_request);
3793 disk = alloc_disk(1 << shift);
3795 mutex_unlock(&disks_mutex);
3796 blk_cleanup_queue(mddev->queue);
3797 mddev->queue = NULL;
3801 disk->major = MAJOR(mddev->unit);
3802 disk->first_minor = unit << shift;
3804 strcpy(disk->disk_name, name);
3805 else if (partitioned)
3806 sprintf(disk->disk_name, "md_d%d", unit);
3808 sprintf(disk->disk_name, "md%d", unit);
3809 disk->fops = &md_fops;
3810 disk->private_data = mddev;
3811 disk->queue = mddev->queue;
3812 /* Allow extended partitions. This makes the
3813 * 'mdp' device redundant, but we can't really
3816 disk->flags |= GENHD_FL_EXT_DEVT;
3818 mddev->gendisk = disk;
3819 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3820 &disk_to_dev(disk)->kobj, "%s", "md");
3821 mutex_unlock(&disks_mutex);
3823 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3826 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3827 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3833 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3835 md_alloc(dev, NULL);
3839 static int add_named_array(const char *val, struct kernel_param *kp)
3841 /* val must be "md_*" where * is not all digits.
3842 * We allocate an array with a large free minor number, and
3843 * set the name to val. val must not already be an active name.
3845 int len = strlen(val);
3846 char buf[DISK_NAME_LEN];
3848 while (len && val[len-1] == '\n')
3850 if (len >= DISK_NAME_LEN)
3852 strlcpy(buf, val, len+1);
3853 if (strncmp(buf, "md_", 3) != 0)
3855 return md_alloc(0, buf);
3858 static void md_safemode_timeout(unsigned long data)
3860 mddev_t *mddev = (mddev_t *) data;
3862 if (!atomic_read(&mddev->writes_pending)) {
3863 mddev->safemode = 1;
3864 if (mddev->external)
3865 sysfs_notify_dirent(mddev->sysfs_state);
3867 md_wakeup_thread(mddev->thread);
3870 static int start_dirty_degraded;
3872 static int do_md_run(mddev_t * mddev)
3877 struct gendisk *disk;
3878 struct mdk_personality *pers;
3879 char b[BDEVNAME_SIZE];
3881 if (list_empty(&mddev->disks))
3882 /* cannot run an array with no devices.. */
3889 * Analyze all RAID superblock(s)
3891 if (!mddev->raid_disks) {
3892 if (!mddev->persistent)
3897 chunk_size = mddev->chunk_size;
3900 if (chunk_size > MAX_CHUNK_SIZE) {
3901 printk(KERN_ERR "too big chunk_size: %d > %d\n",
3902 chunk_size, MAX_CHUNK_SIZE);
3906 * chunk-size has to be a power of 2
3908 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3909 printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3913 /* devices must have minimum size of one chunk */
3914 list_for_each_entry(rdev, &mddev->disks, same_set) {
3915 if (test_bit(Faulty, &rdev->flags))
3917 if (rdev->sectors < chunk_size / 512) {
3919 "md: Dev %s smaller than chunk_size:"
3921 bdevname(rdev->bdev,b),
3922 (unsigned long long)rdev->sectors,
3929 if (mddev->level != LEVEL_NONE)
3930 request_module("md-level-%d", mddev->level);
3931 else if (mddev->clevel[0])
3932 request_module("md-%s", mddev->clevel);
3935 * Drop all container device buffers, from now on
3936 * the only valid external interface is through the md
3939 list_for_each_entry(rdev, &mddev->disks, same_set) {
3940 if (test_bit(Faulty, &rdev->flags))
3942 sync_blockdev(rdev->bdev);
3943 invalidate_bdev(rdev->bdev);
3945 /* perform some consistency tests on the device.
3946 * We don't want the data to overlap the metadata,
3947 * Internal Bitmap issues have been handled elsewhere.
3949 if (rdev->data_offset < rdev->sb_start) {
3950 if (mddev->dev_sectors &&
3951 rdev->data_offset + mddev->dev_sectors
3953 printk("md: %s: data overlaps metadata\n",
3958 if (rdev->sb_start + rdev->sb_size/512
3959 > rdev->data_offset) {
3960 printk("md: %s: metadata overlaps data\n",
3965 sysfs_notify_dirent(rdev->sysfs_state);
3968 md_probe(mddev->unit, NULL, NULL);
3969 disk = mddev->gendisk;
3973 spin_lock(&pers_lock);
3974 pers = find_pers(mddev->level, mddev->clevel);
3975 if (!pers || !try_module_get(pers->owner)) {
3976 spin_unlock(&pers_lock);
3977 if (mddev->level != LEVEL_NONE)
3978 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3981 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3986 spin_unlock(&pers_lock);
3987 if (mddev->level != pers->level) {
3988 mddev->level = pers->level;
3989 mddev->new_level = pers->level;
3991 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3993 if (pers->level >= 4 && pers->level <= 6)
3994 /* Cannot support integrity (yet) */
3995 blk_integrity_unregister(mddev->gendisk);
3997 if (mddev->reshape_position != MaxSector &&
3998 pers->start_reshape == NULL) {
3999 /* This personality cannot handle reshaping... */
4001 module_put(pers->owner);
4005 if (pers->sync_request) {
4006 /* Warn if this is a potentially silly
4009 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4013 list_for_each_entry(rdev, &mddev->disks, same_set)
4014 list_for_each_entry(rdev2, &mddev->disks, same_set) {
4016 rdev->bdev->bd_contains ==
4017 rdev2->bdev->bd_contains) {
4019 "%s: WARNING: %s appears to be"
4020 " on the same physical disk as"
4023 bdevname(rdev->bdev,b),
4024 bdevname(rdev2->bdev,b2));
4031 "True protection against single-disk"
4032 " failure might be compromised.\n");
4035 mddev->recovery = 0;
4036 /* may be over-ridden by personality */
4037 mddev->resync_max_sectors = mddev->dev_sectors;
4039 mddev->barriers_work = 1;
4040 mddev->ok_start_degraded = start_dirty_degraded;
4043 mddev->ro = 2; /* read-only, but switch on first write */
4045 err = mddev->pers->run(mddev);
4047 printk(KERN_ERR "md: pers->run() failed ...\n");
4048 else if (mddev->pers->sync_request) {
4049 err = bitmap_create(mddev);
4051 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4052 mdname(mddev), err);
4053 mddev->pers->stop(mddev);
4057 module_put(mddev->pers->owner);
4059 bitmap_destroy(mddev);
4062 if (mddev->pers->sync_request) {
4063 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4065 "md: cannot register extra attributes for %s\n",
4067 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4068 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4071 atomic_set(&mddev->writes_pending,0);
4072 mddev->safemode = 0;
4073 mddev->safemode_timer.function = md_safemode_timeout;
4074 mddev->safemode_timer.data = (unsigned long) mddev;
4075 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4078 list_for_each_entry(rdev, &mddev->disks, same_set)
4079 if (rdev->raid_disk >= 0) {
4081 sprintf(nm, "rd%d", rdev->raid_disk);
4082 if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4083 printk("md: cannot register %s for %s\n",
4087 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4090 md_update_sb(mddev, 0);
4092 set_capacity(disk, mddev->array_sectors);
4094 /* If there is a partially-recovered drive we need to
4095 * start recovery here. If we leave it to md_check_recovery,
4096 * it will remove the drives and not do the right thing
4098 if (mddev->degraded && !mddev->sync_thread) {
4100 list_for_each_entry(rdev, &mddev->disks, same_set)
4101 if (rdev->raid_disk >= 0 &&
4102 !test_bit(In_sync, &rdev->flags) &&
4103 !test_bit(Faulty, &rdev->flags))
4104 /* complete an interrupted recovery */
4106 if (spares && mddev->pers->sync_request) {
4107 mddev->recovery = 0;
4108 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4109 mddev->sync_thread = md_register_thread(md_do_sync,
4112 if (!mddev->sync_thread) {
4113 printk(KERN_ERR "%s: could not start resync"
4116 /* leave the spares where they are, it shouldn't hurt */
4117 mddev->recovery = 0;
4121 md_wakeup_thread(mddev->thread);
4122 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4125 md_new_event(mddev);
4126 sysfs_notify_dirent(mddev->sysfs_state);
4127 if (mddev->sysfs_action)
4128 sysfs_notify_dirent(mddev->sysfs_action);
4129 sysfs_notify(&mddev->kobj, NULL, "degraded");
4130 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4134 static int restart_array(mddev_t *mddev)
4136 struct gendisk *disk = mddev->gendisk;
4138 /* Complain if it has no devices */
4139 if (list_empty(&mddev->disks))
4145 mddev->safemode = 0;
4147 set_disk_ro(disk, 0);
4148 printk(KERN_INFO "md: %s switched to read-write mode.\n",
4150 /* Kick recovery or resync if necessary */
4151 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4152 md_wakeup_thread(mddev->thread);
4153 md_wakeup_thread(mddev->sync_thread);
4154 sysfs_notify_dirent(mddev->sysfs_state);
4158 /* similar to deny_write_access, but accounts for our holding a reference
4159 * to the file ourselves */
4160 static int deny_bitmap_write_access(struct file * file)
4162 struct inode *inode = file->f_mapping->host;
4164 spin_lock(&inode->i_lock);
4165 if (atomic_read(&inode->i_writecount) > 1) {
4166 spin_unlock(&inode->i_lock);
4169 atomic_set(&inode->i_writecount, -1);
4170 spin_unlock(&inode->i_lock);
4175 static void restore_bitmap_write_access(struct file *file)
4177 struct inode *inode = file->f_mapping->host;
4179 spin_lock(&inode->i_lock);
4180 atomic_set(&inode->i_writecount, 1);
4181 spin_unlock(&inode->i_lock);
4185 * 0 - completely stop and dis-assemble array
4186 * 1 - switch to readonly
4187 * 2 - stop but do not disassemble array
4189 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4192 struct gendisk *disk = mddev->gendisk;
4194 if (atomic_read(&mddev->openers) > is_open) {
4195 printk("md: %s still in use.\n",mdname(mddev));
4201 if (mddev->sync_thread) {
4202 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4203 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4204 md_unregister_thread(mddev->sync_thread);
4205 mddev->sync_thread = NULL;
4208 del_timer_sync(&mddev->safemode_timer);
4211 case 1: /* readonly */
4217 case 0: /* disassemble */
4219 bitmap_flush(mddev);
4220 md_super_wait(mddev);
4222 set_disk_ro(disk, 0);
4224 mddev->pers->stop(mddev);
4225 mddev->queue->merge_bvec_fn = NULL;
4226 mddev->queue->unplug_fn = NULL;
4227 mddev->queue->backing_dev_info.congested_fn = NULL;
4228 if (mddev->pers->sync_request) {
4229 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
4230 if (mddev->sysfs_action)
4231 sysfs_put(mddev->sysfs_action);
4232 mddev->sysfs_action = NULL;
4234 module_put(mddev->pers->owner);
4236 /* tell userspace to handle 'inactive' */
4237 sysfs_notify_dirent(mddev->sysfs_state);
4239 set_capacity(disk, 0);
4245 if (!mddev->in_sync || mddev->flags) {
4246 /* mark array as shutdown cleanly */
4248 md_update_sb(mddev, 1);
4251 set_disk_ro(disk, 1);
4252 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4256 * Free resources if final stop
4261 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4263 bitmap_destroy(mddev);
4264 if (mddev->bitmap_file) {
4265 restore_bitmap_write_access(mddev->bitmap_file);
4266 fput(mddev->bitmap_file);
4267 mddev->bitmap_file = NULL;
4269 mddev->bitmap_offset = 0;
4271 list_for_each_entry(rdev, &mddev->disks, same_set)
4272 if (rdev->raid_disk >= 0) {
4274 sprintf(nm, "rd%d", rdev->raid_disk);
4275 sysfs_remove_link(&mddev->kobj, nm);
4278 /* make sure all md_delayed_delete calls have finished */
4279 flush_scheduled_work();
4281 export_array(mddev);
4283 mddev->array_sectors = 0;
4284 mddev->dev_sectors = 0;
4285 mddev->raid_disks = 0;
4286 mddev->recovery_cp = 0;
4287 mddev->resync_min = 0;
4288 mddev->resync_max = MaxSector;
4289 mddev->reshape_position = MaxSector;
4290 mddev->external = 0;
4291 mddev->persistent = 0;
4292 mddev->level = LEVEL_NONE;
4293 mddev->clevel[0] = 0;
4296 mddev->metadata_type[0] = 0;
4297 mddev->chunk_size = 0;
4298 mddev->ctime = mddev->utime = 0;
4300 mddev->max_disks = 0;
4302 mddev->delta_disks = 0;
4303 mddev->new_level = LEVEL_NONE;
4304 mddev->new_layout = 0;
4305 mddev->new_chunk = 0;
4306 mddev->curr_resync = 0;
4307 mddev->resync_mismatches = 0;
4308 mddev->suspend_lo = mddev->suspend_hi = 0;
4309 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4310 mddev->recovery = 0;
4313 mddev->degraded = 0;
4314 mddev->barriers_work = 0;
4315 mddev->safemode = 0;
4316 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4317 if (mddev->hold_active == UNTIL_STOP)
4318 mddev->hold_active = 0;
4320 } else if (mddev->pers)
4321 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4324 blk_integrity_unregister(disk);
4325 md_new_event(mddev);
4326 sysfs_notify_dirent(mddev->sysfs_state);
4332 static void autorun_array(mddev_t *mddev)
4337 if (list_empty(&mddev->disks))
4340 printk(KERN_INFO "md: running: ");
4342 list_for_each_entry(rdev, &mddev->disks, same_set) {
4343 char b[BDEVNAME_SIZE];
4344 printk("<%s>", bdevname(rdev->bdev,b));
4348 err = do_md_run(mddev);
4350 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4351 do_md_stop(mddev, 0, 0);
4356 * lets try to run arrays based on all disks that have arrived
4357 * until now. (those are in pending_raid_disks)
4359 * the method: pick the first pending disk, collect all disks with
4360 * the same UUID, remove all from the pending list and put them into
4361 * the 'same_array' list. Then order this list based on superblock
4362 * update time (freshest comes first), kick out 'old' disks and
4363 * compare superblocks. If everything's fine then run it.
4365 * If "unit" is allocated, then bump its reference count
4367 static void autorun_devices(int part)
4369 mdk_rdev_t *rdev0, *rdev, *tmp;
4371 char b[BDEVNAME_SIZE];
4373 printk(KERN_INFO "md: autorun ...\n");
4374 while (!list_empty(&pending_raid_disks)) {
4377 LIST_HEAD(candidates);
4378 rdev0 = list_entry(pending_raid_disks.next,
4379 mdk_rdev_t, same_set);
4381 printk(KERN_INFO "md: considering %s ...\n",
4382 bdevname(rdev0->bdev,b));
4383 INIT_LIST_HEAD(&candidates);
4384 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4385 if (super_90_load(rdev, rdev0, 0) >= 0) {
4386 printk(KERN_INFO "md: adding %s ...\n",
4387 bdevname(rdev->bdev,b));
4388 list_move(&rdev->same_set, &candidates);
4391 * now we have a set of devices, with all of them having
4392 * mostly sane superblocks. It's time to allocate the
4396 dev = MKDEV(mdp_major,
4397 rdev0->preferred_minor << MdpMinorShift);
4398 unit = MINOR(dev) >> MdpMinorShift;
4400 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4403 if (rdev0->preferred_minor != unit) {
4404 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4405 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4409 md_probe(dev, NULL, NULL);
4410 mddev = mddev_find(dev);
4411 if (!mddev || !mddev->gendisk) {
4415 "md: cannot allocate memory for md drive.\n");
4418 if (mddev_lock(mddev))
4419 printk(KERN_WARNING "md: %s locked, cannot run\n",
4421 else if (mddev->raid_disks || mddev->major_version
4422 || !list_empty(&mddev->disks)) {
4424 "md: %s already running, cannot run %s\n",
4425 mdname(mddev), bdevname(rdev0->bdev,b));
4426 mddev_unlock(mddev);
4428 printk(KERN_INFO "md: created %s\n", mdname(mddev));
4429 mddev->persistent = 1;
4430 rdev_for_each_list(rdev, tmp, &candidates) {
4431 list_del_init(&rdev->same_set);
4432 if (bind_rdev_to_array(rdev, mddev))
4435 autorun_array(mddev);
4436 mddev_unlock(mddev);
4438 /* on success, candidates will be empty, on error
4441 rdev_for_each_list(rdev, tmp, &candidates) {
4442 list_del_init(&rdev->same_set);
4447 printk(KERN_INFO "md: ... autorun DONE.\n");
4449 #endif /* !MODULE */
4451 static int get_version(void __user * arg)
4455 ver.major = MD_MAJOR_VERSION;
4456 ver.minor = MD_MINOR_VERSION;
4457 ver.patchlevel = MD_PATCHLEVEL_VERSION;
4459 if (copy_to_user(arg, &ver, sizeof(ver)))
4465 static int get_array_info(mddev_t * mddev, void __user * arg)
4467 mdu_array_info_t info;
4468 int nr,working,active,failed,spare;
4471 nr=working=active=failed=spare=0;
4472 list_for_each_entry(rdev, &mddev->disks, same_set) {
4474 if (test_bit(Faulty, &rdev->flags))
4478 if (test_bit(In_sync, &rdev->flags))
4485 info.major_version = mddev->major_version;
4486 info.minor_version = mddev->minor_version;
4487 info.patch_version = MD_PATCHLEVEL_VERSION;
4488 info.ctime = mddev->ctime;
4489 info.level = mddev->level;
4490 info.size = mddev->dev_sectors / 2;
4491 if (info.size != mddev->dev_sectors / 2) /* overflow */
4494 info.raid_disks = mddev->raid_disks;
4495 info.md_minor = mddev->md_minor;
4496 info.not_persistent= !mddev->persistent;
4498 info.utime = mddev->utime;
4501 info.state = (1<<MD_SB_CLEAN);
4502 if (mddev->bitmap && mddev->bitmap_offset)
4503 info.state = (1<<MD_SB_BITMAP_PRESENT);
4504 info.active_disks = active;
4505 info.working_disks = working;
4506 info.failed_disks = failed;
4507 info.spare_disks = spare;
4509 info.layout = mddev->layout;
4510 info.chunk_size = mddev->chunk_size;
4512 if (copy_to_user(arg, &info, sizeof(info)))
4518 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4520 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4521 char *ptr, *buf = NULL;
4524 if (md_allow_write(mddev))
4525 file = kmalloc(sizeof(*file), GFP_NOIO);
4527 file = kmalloc(sizeof(*file), GFP_KERNEL);
4532 /* bitmap disabled, zero the first byte and copy out */
4533 if (!mddev->bitmap || !mddev->bitmap->file) {
4534 file->pathname[0] = '\0';
4538 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4542 ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4546 strcpy(file->pathname, ptr);
4550 if (copy_to_user(arg, file, sizeof(*file)))
4558 static int get_disk_info(mddev_t * mddev, void __user * arg)
4560 mdu_disk_info_t info;
4563 if (copy_from_user(&info, arg, sizeof(info)))
4566 rdev = find_rdev_nr(mddev, info.number);
4568 info.major = MAJOR(rdev->bdev->bd_dev);
4569 info.minor = MINOR(rdev->bdev->bd_dev);
4570 info.raid_disk = rdev->raid_disk;
4572 if (test_bit(Faulty, &rdev->flags))
4573 info.state |= (1<<MD_DISK_FAULTY);
4574 else if (test_bit(In_sync, &rdev->flags)) {
4575 info.state |= (1<<MD_DISK_ACTIVE);
4576 info.state |= (1<<MD_DISK_SYNC);
4578 if (test_bit(WriteMostly, &rdev->flags))
4579 info.state |= (1<<MD_DISK_WRITEMOSTLY);
4581 info.major = info.minor = 0;
4582 info.raid_disk = -1;
4583 info.state = (1<<MD_DISK_REMOVED);
4586 if (copy_to_user(arg, &info, sizeof(info)))
4592 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4594 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4596 dev_t dev = MKDEV(info->major,info->minor);
4598 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4601 if (!mddev->raid_disks) {
4603 /* expecting a device which has a superblock */
4604 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4607 "md: md_import_device returned %ld\n",
4609 return PTR_ERR(rdev);
4611 if (!list_empty(&mddev->disks)) {
4612 mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4613 mdk_rdev_t, same_set);
4614 int err = super_types[mddev->major_version]
4615 .load_super(rdev, rdev0, mddev->minor_version);
4618 "md: %s has different UUID to %s\n",
4619 bdevname(rdev->bdev,b),
4620 bdevname(rdev0->bdev,b2));
4625 err = bind_rdev_to_array(rdev, mddev);
4632 * add_new_disk can be used once the array is assembled
4633 * to add "hot spares". They must already have a superblock
4638 if (!mddev->pers->hot_add_disk) {
4640 "%s: personality does not support diskops!\n",
4644 if (mddev->persistent)
4645 rdev = md_import_device(dev, mddev->major_version,
4646 mddev->minor_version);
4648 rdev = md_import_device(dev, -1, -1);
4651 "md: md_import_device returned %ld\n",
4653 return PTR_ERR(rdev);
4655 /* set save_raid_disk if appropriate */
4656 if (!mddev->persistent) {
4657 if (info->state & (1<<MD_DISK_SYNC) &&
4658 info->raid_disk < mddev->raid_disks)
4659 rdev->raid_disk = info->raid_disk;
4661 rdev->raid_disk = -1;
4663 super_types[mddev->major_version].
4664 validate_super(mddev, rdev);
4665 rdev->saved_raid_disk = rdev->raid_disk;
4667 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4668 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4669 set_bit(WriteMostly, &rdev->flags);
4671 clear_bit(WriteMostly, &rdev->flags);
4673 rdev->raid_disk = -1;
4674 err = bind_rdev_to_array(rdev, mddev);
4675 if (!err && !mddev->pers->hot_remove_disk) {
4676 /* If there is hot_add_disk but no hot_remove_disk
4677 * then added disks for geometry changes,
4678 * and should be added immediately.
4680 super_types[mddev->major_version].
4681 validate_super(mddev, rdev);
4682 err = mddev->pers->hot_add_disk(mddev, rdev);
4684 unbind_rdev_from_array(rdev);
4689 sysfs_notify_dirent(rdev->sysfs_state);
4691 md_update_sb(mddev, 1);
4692 if (mddev->degraded)
4693 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4694 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4695 md_wakeup_thread(mddev->thread);
4699 /* otherwise, add_new_disk is only allowed
4700 * for major_version==0 superblocks
4702 if (mddev->major_version != 0) {
4703 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4708 if (!(info->state & (1<<MD_DISK_FAULTY))) {
4710 rdev = md_import_device(dev, -1, 0);
4713 "md: error, md_import_device() returned %ld\n",
4715 return PTR_ERR(rdev);
4717 rdev->desc_nr = info->number;
4718 if (info->raid_disk < mddev->raid_disks)
4719 rdev->raid_disk = info->raid_disk;
4721 rdev->raid_disk = -1;
4723 if (rdev->raid_disk < mddev->raid_disks)
4724 if (info->state & (1<<MD_DISK_SYNC))
4725 set_bit(In_sync, &rdev->flags);
4727 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4728 set_bit(WriteMostly, &rdev->flags);
4730 if (!mddev->persistent) {
4731 printk(KERN_INFO "md: nonpersistent superblock ...\n");
4732 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4734 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4735 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4737 err = bind_rdev_to_array(rdev, mddev);
4747 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4749 char b[BDEVNAME_SIZE];
4752 rdev = find_rdev(mddev, dev);
4756 if (rdev->raid_disk >= 0)
4759 kick_rdev_from_array(rdev);
4760 md_update_sb(mddev, 1);
4761 md_new_event(mddev);
4765 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4766 bdevname(rdev->bdev,b), mdname(mddev));
4770 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4772 char b[BDEVNAME_SIZE];
4779 if (mddev->major_version != 0) {
4780 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4781 " version-0 superblocks.\n",
4785 if (!mddev->pers->hot_add_disk) {
4787 "%s: personality does not support diskops!\n",
4792 rdev = md_import_device(dev, -1, 0);
4795 "md: error, md_import_device() returned %ld\n",
4800 if (mddev->persistent)
4801 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4803 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4805 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4807 if (test_bit(Faulty, &rdev->flags)) {
4809 "md: can not hot-add faulty %s disk to %s!\n",
4810 bdevname(rdev->bdev,b), mdname(mddev));
4814 clear_bit(In_sync, &rdev->flags);
4816 rdev->saved_raid_disk = -1;
4817 err = bind_rdev_to_array(rdev, mddev);
4822 * The rest should better be atomic, we can have disk failures
4823 * noticed in interrupt contexts ...
4826 rdev->raid_disk = -1;
4828 md_update_sb(mddev, 1);
4831 * Kick recovery, maybe this spare has to be added to the
4832 * array immediately.
4834 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4835 md_wakeup_thread(mddev->thread);
4836 md_new_event(mddev);
4844 static int set_bitmap_file(mddev_t *mddev, int fd)
4849 if (!mddev->pers->quiesce)
4851 if (mddev->recovery || mddev->sync_thread)
4853 /* we should be able to change the bitmap.. */
4859 return -EEXIST; /* cannot add when bitmap is present */
4860 mddev->bitmap_file = fget(fd);
4862 if (mddev->bitmap_file == NULL) {
4863 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4868 err = deny_bitmap_write_access(mddev->bitmap_file);
4870 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4872 fput(mddev->bitmap_file);
4873 mddev->bitmap_file = NULL;
4876 mddev->bitmap_offset = 0; /* file overrides offset */
4877 } else if (mddev->bitmap == NULL)
4878 return -ENOENT; /* cannot remove what isn't there */
4881 mddev->pers->quiesce(mddev, 1);
4883 err = bitmap_create(mddev);
4884 if (fd < 0 || err) {
4885 bitmap_destroy(mddev);
4886 fd = -1; /* make sure to put the file */
4888 mddev->pers->quiesce(mddev, 0);
4891 if (mddev->bitmap_file) {
4892 restore_bitmap_write_access(mddev->bitmap_file);
4893 fput(mddev->bitmap_file);
4895 mddev->bitmap_file = NULL;
4902 * set_array_info is used two different ways
4903 * The original usage is when creating a new array.
4904 * In this usage, raid_disks is > 0 and it together with
4905 * level, size, not_persistent,layout,chunksize determine the
4906 * shape of the array.
4907 * This will always create an array with a type-0.90.0 superblock.
4908 * The newer usage is when assembling an array.
4909 * In this case raid_disks will be 0, and the major_version field is
4910 * use to determine which style super-blocks are to be found on the devices.
4911 * The minor and patch _version numbers are also kept incase the
4912 * super_block handler wishes to interpret them.
4914 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4917 if (info->raid_disks == 0) {
4918 /* just setting version number for superblock loading */
4919 if (info->major_version < 0 ||
4920 info->major_version >= ARRAY_SIZE(super_types) ||
4921 super_types[info->major_version].name == NULL) {
4922 /* maybe try to auto-load a module? */
4924 "md: superblock version %d not known\n",
4925 info->major_version);
4928 mddev->major_version = info->major_version;
4929 mddev->minor_version = info->minor_version;
4930 mddev->patch_version = info->patch_version;
4931 mddev->persistent = !info->not_persistent;
4934 mddev->major_version = MD_MAJOR_VERSION;
4935 mddev->minor_version = MD_MINOR_VERSION;
4936 mddev->patch_version = MD_PATCHLEVEL_VERSION;
4937 mddev->ctime = get_seconds();
4939 mddev->level = info->level;
4940 mddev->clevel[0] = 0;
4941 mddev->dev_sectors = 2 * (sector_t)info->size;
4942 mddev->raid_disks = info->raid_disks;
4943 /* don't set md_minor, it is determined by which /dev/md* was
4946 if (info->state & (1<<MD_SB_CLEAN))
4947 mddev->recovery_cp = MaxSector;
4949 mddev->recovery_cp = 0;
4950 mddev->persistent = ! info->not_persistent;
4951 mddev->external = 0;
4953 mddev->layout = info->layout;
4954 mddev->chunk_size = info->chunk_size;
4956 mddev->max_disks = MD_SB_DISKS;
4958 if (mddev->persistent)
4960 set_bit(MD_CHANGE_DEVS, &mddev->flags);
4962 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4963 mddev->bitmap_offset = 0;
4965 mddev->reshape_position = MaxSector;
4968 * Generate a 128 bit UUID
4970 get_random_bytes(mddev->uuid, 16);
4972 mddev->new_level = mddev->level;
4973 mddev->new_chunk = mddev->chunk_size;
4974 mddev->new_layout = mddev->layout;
4975 mddev->delta_disks = 0;
4980 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
4982 mddev->array_sectors = array_sectors;
4984 EXPORT_SYMBOL(md_set_array_sectors);
4986 static int update_size(mddev_t *mddev, sector_t num_sectors)
4990 int fit = (num_sectors == 0);
4992 if (mddev->pers->resize == NULL)
4994 /* The "num_sectors" is the number of sectors of each device that
4995 * is used. This can only make sense for arrays with redundancy.
4996 * linear and raid0 always use whatever space is available. We can only
4997 * consider changing this number if no resync or reconstruction is
4998 * happening, and if the new size is acceptable. It must fit before the
4999 * sb_start or, if that is <data_offset, it must fit before the size
5000 * of each device. If num_sectors is zero, we find the largest size
5004 if (mddev->sync_thread)
5007 /* Sorry, cannot grow a bitmap yet, just remove it,
5011 list_for_each_entry(rdev, &mddev->disks, same_set) {
5012 sector_t avail = rdev->sectors;
5014 if (fit && (num_sectors == 0 || num_sectors > avail))
5015 num_sectors = avail;
5016 if (avail < num_sectors)
5019 rv = mddev->pers->resize(mddev, num_sectors);
5021 struct block_device *bdev;
5023 bdev = bdget_disk(mddev->gendisk, 0);
5025 mutex_lock(&bdev->bd_inode->i_mutex);
5026 i_size_write(bdev->bd_inode,
5027 (loff_t)mddev->array_sectors << 9);
5028 mutex_unlock(&bdev->bd_inode->i_mutex);
5035 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5038 /* change the number of raid disks */
5039 if (mddev->pers->check_reshape == NULL)
5041 if (raid_disks <= 0 ||
5042 raid_disks >= mddev->max_disks)
5044 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5046 mddev->delta_disks = raid_disks - mddev->raid_disks;
5048 rv = mddev->pers->check_reshape(mddev);
5054 * update_array_info is used to change the configuration of an
5056 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5057 * fields in the info are checked against the array.
5058 * Any differences that cannot be handled will cause an error.
5059 * Normally, only one change can be managed at a time.
5061 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5067 /* calculate expected state,ignoring low bits */
5068 if (mddev->bitmap && mddev->bitmap_offset)
5069 state |= (1 << MD_SB_BITMAP_PRESENT);
5071 if (mddev->major_version != info->major_version ||
5072 mddev->minor_version != info->minor_version ||
5073 /* mddev->patch_version != info->patch_version || */
5074 mddev->ctime != info->ctime ||
5075 mddev->level != info->level ||
5076 /* mddev->layout != info->layout || */
5077 !mddev->persistent != info->not_persistent||
5078 mddev->chunk_size != info->chunk_size ||
5079 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5080 ((state^info->state) & 0xfffffe00)
5083 /* Check there is only one change */
5084 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5086 if (mddev->raid_disks != info->raid_disks)
5088 if (mddev->layout != info->layout)
5090 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5097 if (mddev->layout != info->layout) {
5099 * we don't need to do anything at the md level, the
5100 * personality will take care of it all.
5102 if (mddev->pers->reconfig == NULL)
5105 return mddev->pers->reconfig(mddev, info->layout, -1);
5107 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5108 rv = update_size(mddev, (sector_t)info->size * 2);
5110 if (mddev->raid_disks != info->raid_disks)
5111 rv = update_raid_disks(mddev, info->raid_disks);
5113 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5114 if (mddev->pers->quiesce == NULL)
5116 if (mddev->recovery || mddev->sync_thread)
5118 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5119 /* add the bitmap */
5122 if (mddev->default_bitmap_offset == 0)
5124 mddev->bitmap_offset = mddev->default_bitmap_offset;
5125 mddev->pers->quiesce(mddev, 1);
5126 rv = bitmap_create(mddev);
5128 bitmap_destroy(mddev);
5129 mddev->pers->quiesce(mddev, 0);
5131 /* remove the bitmap */
5134 if (mddev->bitmap->file)
5136 mddev->pers->quiesce(mddev, 1);
5137 bitmap_destroy(mddev);
5138 mddev->pers->quiesce(mddev, 0);
5139 mddev->bitmap_offset = 0;
5142 md_update_sb(mddev, 1);
5146 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5150 if (mddev->pers == NULL)
5153 rdev = find_rdev(mddev, dev);
5157 md_error(mddev, rdev);
5162 * We have a problem here : there is no easy way to give a CHS
5163 * virtual geometry. We currently pretend that we have a 2 heads
5164 * 4 sectors (with a BIG number of cylinders...). This drives
5165 * dosfs just mad... ;-)
5167 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5169 mddev_t *mddev = bdev->bd_disk->private_data;
5173 geo->cylinders = get_capacity(mddev->gendisk) / 8;
5177 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5178 unsigned int cmd, unsigned long arg)
5181 void __user *argp = (void __user *)arg;
5182 mddev_t *mddev = NULL;
5184 if (!capable(CAP_SYS_ADMIN))
5188 * Commands dealing with the RAID driver but not any
5194 err = get_version(argp);
5197 case PRINT_RAID_DEBUG:
5205 autostart_arrays(arg);
5212 * Commands creating/starting a new array:
5215 mddev = bdev->bd_disk->private_data;
5222 err = mddev_lock(mddev);
5225 "md: ioctl lock interrupted, reason %d, cmd %d\n",
5232 case SET_ARRAY_INFO:
5234 mdu_array_info_t info;
5236 memset(&info, 0, sizeof(info));
5237 else if (copy_from_user(&info, argp, sizeof(info))) {
5242 err = update_array_info(mddev, &info);
5244 printk(KERN_WARNING "md: couldn't update"
5245 " array info. %d\n", err);
5250 if (!list_empty(&mddev->disks)) {
5252 "md: array %s already has disks!\n",
5257 if (mddev->raid_disks) {
5259 "md: array %s already initialised!\n",
5264 err = set_array_info(mddev, &info);
5266 printk(KERN_WARNING "md: couldn't set"
5267 " array info. %d\n", err);
5277 * Commands querying/configuring an existing array:
5279 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5280 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5281 if ((!mddev->raid_disks && !mddev->external)
5282 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5283 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5284 && cmd != GET_BITMAP_FILE) {
5290 * Commands even a read-only array can execute:
5294 case GET_ARRAY_INFO:
5295 err = get_array_info(mddev, argp);
5298 case GET_BITMAP_FILE:
5299 err = get_bitmap_file(mddev, argp);
5303 err = get_disk_info(mddev, argp);
5306 case RESTART_ARRAY_RW:
5307 err = restart_array(mddev);
5311 err = do_md_stop(mddev, 0, 1);
5315 err = do_md_stop(mddev, 1, 1);
5321 * The remaining ioctls are changing the state of the
5322 * superblock, so we do not allow them on read-only arrays.
5323 * However non-MD ioctls (e.g. get-size) will still come through
5324 * here and hit the 'default' below, so only disallow
5325 * 'md' ioctls, and switch to rw mode if started auto-readonly.
5327 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5328 if (mddev->ro == 2) {
5330 sysfs_notify_dirent(mddev->sysfs_state);
5331 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5332 md_wakeup_thread(mddev->thread);
5343 mdu_disk_info_t info;
5344 if (copy_from_user(&info, argp, sizeof(info)))
5347 err = add_new_disk(mddev, &info);
5351 case HOT_REMOVE_DISK:
5352 err = hot_remove_disk(mddev, new_decode_dev(arg));
5356 err = hot_add_disk(mddev, new_decode_dev(arg));
5359 case SET_DISK_FAULTY:
5360 err = set_disk_faulty(mddev, new_decode_dev(arg));
5364 err = do_md_run(mddev);
5367 case SET_BITMAP_FILE:
5368 err = set_bitmap_file(mddev, (int)arg);
5378 if (mddev->hold_active == UNTIL_IOCTL &&
5380 mddev->hold_active = 0;
5381 mddev_unlock(mddev);
5391 static int md_open(struct block_device *bdev, fmode_t mode)
5394 * Succeed if we can lock the mddev, which confirms that
5395 * it isn't being stopped right now.
5397 mddev_t *mddev = mddev_find(bdev->bd_dev);
5400 if (mddev->gendisk != bdev->bd_disk) {
5401 /* we are racing with mddev_put which is discarding this
5405 /* Wait until bdev->bd_disk is definitely gone */
5406 flush_scheduled_work();
5407 /* Then retry the open from the top */
5408 return -ERESTARTSYS;
5410 BUG_ON(mddev != bdev->bd_disk->private_data);
5412 if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5416 atomic_inc(&mddev->openers);
5417 mddev_unlock(mddev);
5419 check_disk_change(bdev);
5424 static int md_release(struct gendisk *disk, fmode_t mode)
5426 mddev_t *mddev = disk->private_data;
5429 atomic_dec(&mddev->openers);
5435 static int md_media_changed(struct gendisk *disk)
5437 mddev_t *mddev = disk->private_data;
5439 return mddev->changed;
5442 static int md_revalidate(struct gendisk *disk)
5444 mddev_t *mddev = disk->private_data;
5449 static struct block_device_operations md_fops =
5451 .owner = THIS_MODULE,
5453 .release = md_release,
5454 .locked_ioctl = md_ioctl,
5455 .getgeo = md_getgeo,
5456 .media_changed = md_media_changed,
5457 .revalidate_disk= md_revalidate,
5460 static int md_thread(void * arg)
5462 mdk_thread_t *thread = arg;
5465 * md_thread is a 'system-thread', it's priority should be very
5466 * high. We avoid resource deadlocks individually in each
5467 * raid personality. (RAID5 does preallocation) We also use RR and
5468 * the very same RT priority as kswapd, thus we will never get
5469 * into a priority inversion deadlock.
5471 * we definitely have to have equal or higher priority than
5472 * bdflush, otherwise bdflush will deadlock if there are too
5473 * many dirty RAID5 blocks.
5476 allow_signal(SIGKILL);
5477 while (!kthread_should_stop()) {
5479 /* We need to wait INTERRUPTIBLE so that
5480 * we don't add to the load-average.
5481 * That means we need to be sure no signals are
5484 if (signal_pending(current))
5485 flush_signals(current);
5487 wait_event_interruptible_timeout
5489 test_bit(THREAD_WAKEUP, &thread->flags)
5490 || kthread_should_stop(),
5493 clear_bit(THREAD_WAKEUP, &thread->flags);
5495 thread->run(thread->mddev);
5501 void md_wakeup_thread(mdk_thread_t *thread)
5504 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5505 set_bit(THREAD_WAKEUP, &thread->flags);
5506 wake_up(&thread->wqueue);
5510 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5513 mdk_thread_t *thread;
5515 thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5519 init_waitqueue_head(&thread->wqueue);
5522 thread->mddev = mddev;
5523 thread->timeout = MAX_SCHEDULE_TIMEOUT;
5524 thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5525 if (IS_ERR(thread->tsk)) {
5532 void md_unregister_thread(mdk_thread_t *thread)
5536 dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5538 kthread_stop(thread->tsk);
5542 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5549 if (!rdev || test_bit(Faulty, &rdev->flags))
5552 if (mddev->external)
5553 set_bit(Blocked, &rdev->flags);
5555 dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5557 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5558 __builtin_return_address(0),__builtin_return_address(1),
5559 __builtin_return_address(2),__builtin_return_address(3));
5563 if (!mddev->pers->error_handler)
5565 mddev->pers->error_handler(mddev,rdev);
5566 if (mddev->degraded)
5567 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5568 set_bit(StateChanged, &rdev->flags);
5569 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5570 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5571 md_wakeup_thread(mddev->thread);
5572 md_new_event_inintr(mddev);
5575 /* seq_file implementation /proc/mdstat */
5577 static void status_unused(struct seq_file *seq)
5582 seq_printf(seq, "unused devices: ");
5584 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5585 char b[BDEVNAME_SIZE];
5587 seq_printf(seq, "%s ",
5588 bdevname(rdev->bdev,b));
5591 seq_printf(seq, "<none>");
5593 seq_printf(seq, "\n");
5597 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5599 sector_t max_blocks, resync, res;
5600 unsigned long dt, db, rt;
5602 unsigned int per_milli;
5604 resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5606 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5607 max_blocks = mddev->resync_max_sectors >> 1;
5609 max_blocks = mddev->dev_sectors / 2;
5612 * Should not happen.
5618 /* Pick 'scale' such that (resync>>scale)*1000 will fit
5619 * in a sector_t, and (max_blocks>>scale) will fit in a
5620 * u32, as those are the requirements for sector_div.
5621 * Thus 'scale' must be at least 10
5624 if (sizeof(sector_t) > sizeof(unsigned long)) {
5625 while ( max_blocks/2 > (1ULL<<(scale+32)))
5628 res = (resync>>scale)*1000;
5629 sector_div(res, (u32)((max_blocks>>scale)+1));
5633 int i, x = per_milli/50, y = 20-x;
5634 seq_printf(seq, "[");
5635 for (i = 0; i < x; i++)
5636 seq_printf(seq, "=");
5637 seq_printf(seq, ">");
5638 for (i = 0; i < y; i++)
5639 seq_printf(seq, ".");
5640 seq_printf(seq, "] ");
5642 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5643 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5645 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5647 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5648 "resync" : "recovery"))),
5649 per_milli/10, per_milli % 10,
5650 (unsigned long long) resync,
5651 (unsigned long long) max_blocks);
5654 * We do not want to overflow, so the order of operands and
5655 * the * 100 / 100 trick are important. We do a +1 to be
5656 * safe against division by zero. We only estimate anyway.
5658 * dt: time from mark until now
5659 * db: blocks written from mark until now
5660 * rt: remaining time
5662 dt = ((jiffies - mddev->resync_mark) / HZ);
5664 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5665 - mddev->resync_mark_cnt;
5666 rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5668 seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5670 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5673 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5675 struct list_head *tmp;
5685 spin_lock(&all_mddevs_lock);
5686 list_for_each(tmp,&all_mddevs)
5688 mddev = list_entry(tmp, mddev_t, all_mddevs);
5690 spin_unlock(&all_mddevs_lock);
5693 spin_unlock(&all_mddevs_lock);
5695 return (void*)2;/* tail */
5699 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5701 struct list_head *tmp;
5702 mddev_t *next_mddev, *mddev = v;
5708 spin_lock(&all_mddevs_lock);
5710 tmp = all_mddevs.next;
5712 tmp = mddev->all_mddevs.next;
5713 if (tmp != &all_mddevs)
5714 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5716 next_mddev = (void*)2;
5719 spin_unlock(&all_mddevs_lock);
5727 static void md_seq_stop(struct seq_file *seq, void *v)
5731 if (mddev && v != (void*)1 && v != (void*)2)
5735 struct mdstat_info {
5739 static int md_seq_show(struct seq_file *seq, void *v)
5744 struct mdstat_info *mi = seq->private;
5745 struct bitmap *bitmap;
5747 if (v == (void*)1) {
5748 struct mdk_personality *pers;
5749 seq_printf(seq, "Personalities : ");
5750 spin_lock(&pers_lock);
5751 list_for_each_entry(pers, &pers_list, list)
5752 seq_printf(seq, "[%s] ", pers->name);
5754 spin_unlock(&pers_lock);
5755 seq_printf(seq, "\n");
5756 mi->event = atomic_read(&md_event_count);
5759 if (v == (void*)2) {
5764 if (mddev_lock(mddev) < 0)
5767 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5768 seq_printf(seq, "%s : %sactive", mdname(mddev),
5769 mddev->pers ? "" : "in");
5772 seq_printf(seq, " (read-only)");
5774 seq_printf(seq, " (auto-read-only)");
5775 seq_printf(seq, " %s", mddev->pers->name);
5779 list_for_each_entry(rdev, &mddev->disks, same_set) {
5780 char b[BDEVNAME_SIZE];
5781 seq_printf(seq, " %s[%d]",
5782 bdevname(rdev->bdev,b), rdev->desc_nr);
5783 if (test_bit(WriteMostly, &rdev->flags))
5784 seq_printf(seq, "(W)");
5785 if (test_bit(Faulty, &rdev->flags)) {
5786 seq_printf(seq, "(F)");
5788 } else if (rdev->raid_disk < 0)
5789 seq_printf(seq, "(S)"); /* spare */
5790 sectors += rdev->sectors;
5793 if (!list_empty(&mddev->disks)) {
5795 seq_printf(seq, "\n %llu blocks",
5796 (unsigned long long)
5797 mddev->array_sectors / 2);
5799 seq_printf(seq, "\n %llu blocks",
5800 (unsigned long long)sectors / 2);
5802 if (mddev->persistent) {
5803 if (mddev->major_version != 0 ||
5804 mddev->minor_version != 90) {
5805 seq_printf(seq," super %d.%d",
5806 mddev->major_version,
5807 mddev->minor_version);
5809 } else if (mddev->external)
5810 seq_printf(seq, " super external:%s",
5811 mddev->metadata_type);
5813 seq_printf(seq, " super non-persistent");
5816 mddev->pers->status(seq, mddev);
5817 seq_printf(seq, "\n ");
5818 if (mddev->pers->sync_request) {
5819 if (mddev->curr_resync > 2) {
5820 status_resync(seq, mddev);
5821 seq_printf(seq, "\n ");
5822 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5823 seq_printf(seq, "\tresync=DELAYED\n ");
5824 else if (mddev->recovery_cp < MaxSector)
5825 seq_printf(seq, "\tresync=PENDING\n ");
5828 seq_printf(seq, "\n ");
5830 if ((bitmap = mddev->bitmap)) {
5831 unsigned long chunk_kb;
5832 unsigned long flags;
5833 spin_lock_irqsave(&bitmap->lock, flags);
5834 chunk_kb = bitmap->chunksize >> 10;
5835 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5837 bitmap->pages - bitmap->missing_pages,
5839 (bitmap->pages - bitmap->missing_pages)
5840 << (PAGE_SHIFT - 10),
5841 chunk_kb ? chunk_kb : bitmap->chunksize,
5842 chunk_kb ? "KB" : "B");
5844 seq_printf(seq, ", file: ");
5845 seq_path(seq, &bitmap->file->f_path, " \t\n");
5848 seq_printf(seq, "\n");
5849 spin_unlock_irqrestore(&bitmap->lock, flags);
5852 seq_printf(seq, "\n");
5854 mddev_unlock(mddev);
5859 static struct seq_operations md_seq_ops = {
5860 .start = md_seq_start,
5861 .next = md_seq_next,
5862 .stop = md_seq_stop,
5863 .show = md_seq_show,
5866 static int md_seq_open(struct inode *inode, struct file *file)
5869 struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5873 error = seq_open(file, &md_seq_ops);
5877 struct seq_file *p = file->private_data;
5879 mi->event = atomic_read(&md_event_count);
5884 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5886 struct seq_file *m = filp->private_data;
5887 struct mdstat_info *mi = m->private;
5890 poll_wait(filp, &md_event_waiters, wait);
5892 /* always allow read */
5893 mask = POLLIN | POLLRDNORM;
5895 if (mi->event != atomic_read(&md_event_count))
5896 mask |= POLLERR | POLLPRI;
5900 static const struct file_operations md_seq_fops = {
5901 .owner = THIS_MODULE,
5902 .open = md_seq_open,
5904 .llseek = seq_lseek,
5905 .release = seq_release_private,
5906 .poll = mdstat_poll,
5909 int register_md_personality(struct mdk_personality *p)
5911 spin_lock(&pers_lock);
5912 list_add_tail(&p->list, &pers_list);
5913 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5914 spin_unlock(&pers_lock);
5918 int unregister_md_personality(struct mdk_personality *p)
5920 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5921 spin_lock(&pers_lock);
5922 list_del_init(&p->list);
5923 spin_unlock(&pers_lock);
5927 static int is_mddev_idle(mddev_t *mddev, int init)
5935 rdev_for_each_rcu(rdev, mddev) {
5936 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5937 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
5938 (int)part_stat_read(&disk->part0, sectors[1]) -
5939 atomic_read(&disk->sync_io);
5940 /* sync IO will cause sync_io to increase before the disk_stats
5941 * as sync_io is counted when a request starts, and
5942 * disk_stats is counted when it completes.
5943 * So resync activity will cause curr_events to be smaller than
5944 * when there was no such activity.
5945 * non-sync IO will cause disk_stat to increase without
5946 * increasing sync_io so curr_events will (eventually)
5947 * be larger than it was before. Once it becomes
5948 * substantially larger, the test below will cause
5949 * the array to appear non-idle, and resync will slow
5951 * If there is a lot of outstanding resync activity when
5952 * we set last_event to curr_events, then all that activity
5953 * completing might cause the array to appear non-idle
5954 * and resync will be slowed down even though there might
5955 * not have been non-resync activity. This will only
5956 * happen once though. 'last_events' will soon reflect
5957 * the state where there is little or no outstanding
5958 * resync requests, and further resync activity will
5959 * always make curr_events less than last_events.
5962 if (init || curr_events - rdev->last_events > 64) {
5963 rdev->last_events = curr_events;
5971 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5973 /* another "blocks" (512byte) blocks have been synced */
5974 atomic_sub(blocks, &mddev->recovery_active);
5975 wake_up(&mddev->recovery_wait);
5977 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5978 md_wakeup_thread(mddev->thread);
5979 // stop recovery, signal do_sync ....
5984 /* md_write_start(mddev, bi)
5985 * If we need to update some array metadata (e.g. 'active' flag
5986 * in superblock) before writing, schedule a superblock update
5987 * and wait for it to complete.
5989 void md_write_start(mddev_t *mddev, struct bio *bi)
5992 if (bio_data_dir(bi) != WRITE)
5995 BUG_ON(mddev->ro == 1);
5996 if (mddev->ro == 2) {
5997 /* need to switch to read/write */
5999 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6000 md_wakeup_thread(mddev->thread);
6001 md_wakeup_thread(mddev->sync_thread);
6004 atomic_inc(&mddev->writes_pending);
6005 if (mddev->safemode == 1)
6006 mddev->safemode = 0;
6007 if (mddev->in_sync) {
6008 spin_lock_irq(&mddev->write_lock);
6009 if (mddev->in_sync) {
6011 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6012 md_wakeup_thread(mddev->thread);
6015 spin_unlock_irq(&mddev->write_lock);
6018 sysfs_notify_dirent(mddev->sysfs_state);
6019 wait_event(mddev->sb_wait,
6020 !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6021 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6024 void md_write_end(mddev_t *mddev)
6026 if (atomic_dec_and_test(&mddev->writes_pending)) {
6027 if (mddev->safemode == 2)
6028 md_wakeup_thread(mddev->thread);
6029 else if (mddev->safemode_delay)
6030 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6034 /* md_allow_write(mddev)
6035 * Calling this ensures that the array is marked 'active' so that writes
6036 * may proceed without blocking. It is important to call this before
6037 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6038 * Must be called with mddev_lock held.
6040 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6041 * is dropped, so return -EAGAIN after notifying userspace.
6043 int md_allow_write(mddev_t *mddev)
6049 if (!mddev->pers->sync_request)
6052 spin_lock_irq(&mddev->write_lock);
6053 if (mddev->in_sync) {
6055 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6056 if (mddev->safemode_delay &&
6057 mddev->safemode == 0)
6058 mddev->safemode = 1;
6059 spin_unlock_irq(&mddev->write_lock);
6060 md_update_sb(mddev, 0);
6061 sysfs_notify_dirent(mddev->sysfs_state);
6063 spin_unlock_irq(&mddev->write_lock);
6065 if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6070 EXPORT_SYMBOL_GPL(md_allow_write);
6072 #define SYNC_MARKS 10
6073 #define SYNC_MARK_STEP (3*HZ)
6074 void md_do_sync(mddev_t *mddev)
6077 unsigned int currspeed = 0,
6079 sector_t max_sectors,j, io_sectors;
6080 unsigned long mark[SYNC_MARKS];
6081 sector_t mark_cnt[SYNC_MARKS];
6083 struct list_head *tmp;
6084 sector_t last_check;
6089 /* just incase thread restarts... */
6090 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6092 if (mddev->ro) /* never try to sync a read-only array */
6095 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6096 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6097 desc = "data-check";
6098 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6099 desc = "requested-resync";
6102 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6107 /* we overload curr_resync somewhat here.
6108 * 0 == not engaged in resync at all
6109 * 2 == checking that there is no conflict with another sync
6110 * 1 == like 2, but have yielded to allow conflicting resync to
6112 * other == active in resync - this many blocks
6114 * Before starting a resync we must have set curr_resync to
6115 * 2, and then checked that every "conflicting" array has curr_resync
6116 * less than ours. When we find one that is the same or higher
6117 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
6118 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6119 * This will mean we have to start checking from the beginning again.
6124 mddev->curr_resync = 2;
6127 if (kthread_should_stop()) {
6128 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6131 for_each_mddev(mddev2, tmp) {
6132 if (mddev2 == mddev)
6134 if (!mddev->parallel_resync
6135 && mddev2->curr_resync
6136 && match_mddev_units(mddev, mddev2)) {
6138 if (mddev < mddev2 && mddev->curr_resync == 2) {
6139 /* arbitrarily yield */
6140 mddev->curr_resync = 1;
6141 wake_up(&resync_wait);
6143 if (mddev > mddev2 && mddev->curr_resync == 1)
6144 /* no need to wait here, we can wait the next
6145 * time 'round when curr_resync == 2
6148 /* We need to wait 'interruptible' so as not to
6149 * contribute to the load average, and not to
6150 * be caught by 'softlockup'
6152 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6153 if (!kthread_should_stop() &&
6154 mddev2->curr_resync >= mddev->curr_resync) {
6155 printk(KERN_INFO "md: delaying %s of %s"
6156 " until %s has finished (they"
6157 " share one or more physical units)\n",
6158 desc, mdname(mddev), mdname(mddev2));
6160 if (signal_pending(current))
6161 flush_signals(current);
6163 finish_wait(&resync_wait, &wq);
6166 finish_wait(&resync_wait, &wq);
6169 } while (mddev->curr_resync < 2);
6172 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6173 /* resync follows the size requested by the personality,
6174 * which defaults to physical size, but can be virtual size
6176 max_sectors = mddev->resync_max_sectors;
6177 mddev->resync_mismatches = 0;
6178 /* we don't use the checkpoint if there's a bitmap */
6179 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6180 j = mddev->resync_min;
6181 else if (!mddev->bitmap)
6182 j = mddev->recovery_cp;
6184 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6185 max_sectors = mddev->dev_sectors;
6187 /* recovery follows the physical size of devices */
6188 max_sectors = mddev->dev_sectors;
6190 list_for_each_entry(rdev, &mddev->disks, same_set)
6191 if (rdev->raid_disk >= 0 &&
6192 !test_bit(Faulty, &rdev->flags) &&
6193 !test_bit(In_sync, &rdev->flags) &&
6194 rdev->recovery_offset < j)
6195 j = rdev->recovery_offset;
6198 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6199 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
6200 " %d KB/sec/disk.\n", speed_min(mddev));
6201 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6202 "(but not more than %d KB/sec) for %s.\n",
6203 speed_max(mddev), desc);
6205 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6208 for (m = 0; m < SYNC_MARKS; m++) {
6210 mark_cnt[m] = io_sectors;
6213 mddev->resync_mark = mark[last_mark];
6214 mddev->resync_mark_cnt = mark_cnt[last_mark];
6217 * Tune reconstruction:
6219 window = 32*(PAGE_SIZE/512);
6220 printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6221 window/2,(unsigned long long) max_sectors/2);
6223 atomic_set(&mddev->recovery_active, 0);
6228 "md: resuming %s of %s from checkpoint.\n",
6229 desc, mdname(mddev));
6230 mddev->curr_resync = j;
6233 while (j < max_sectors) {
6237 if (j >= mddev->resync_max) {
6238 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6239 wait_event(mddev->recovery_wait,
6240 mddev->resync_max > j
6241 || kthread_should_stop());
6243 if (kthread_should_stop())
6246 if (mddev->curr_resync > mddev->curr_resync_completed &&
6247 (mddev->curr_resync - mddev->curr_resync_completed)
6248 > (max_sectors >> 4)) {
6249 /* time to update curr_resync_completed */
6250 blk_unplug(mddev->queue);
6251 wait_event(mddev->recovery_wait,
6252 atomic_read(&mddev->recovery_active) == 0);
6253 mddev->curr_resync_completed =
6255 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6257 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6258 currspeed < speed_min(mddev));
6260 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6264 if (!skipped) { /* actual IO requested */
6265 io_sectors += sectors;
6266 atomic_add(sectors, &mddev->recovery_active);
6270 if (j>1) mddev->curr_resync = j;
6271 mddev->curr_mark_cnt = io_sectors;
6272 if (last_check == 0)
6273 /* this is the earliers that rebuilt will be
6274 * visible in /proc/mdstat
6276 md_new_event(mddev);
6278 if (last_check + window > io_sectors || j == max_sectors)
6281 last_check = io_sectors;
6283 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6287 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6289 int next = (last_mark+1) % SYNC_MARKS;
6291 mddev->resync_mark = mark[next];
6292 mddev->resync_mark_cnt = mark_cnt[next];
6293 mark[next] = jiffies;
6294 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6299 if (kthread_should_stop())
6304 * this loop exits only if either when we are slower than
6305 * the 'hard' speed limit, or the system was IO-idle for
6307 * the system might be non-idle CPU-wise, but we only care
6308 * about not overloading the IO subsystem. (things like an
6309 * e2fsck being done on the RAID array should execute fast)
6311 blk_unplug(mddev->queue);
6314 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6315 /((jiffies-mddev->resync_mark)/HZ +1) +1;
6317 if (currspeed > speed_min(mddev)) {
6318 if ((currspeed > speed_max(mddev)) ||
6319 !is_mddev_idle(mddev, 0)) {
6325 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6327 * this also signals 'finished resyncing' to md_stop
6330 blk_unplug(mddev->queue);
6332 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6334 /* tell personality that we are finished */
6335 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6337 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6338 mddev->curr_resync > 2) {
6339 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6340 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6341 if (mddev->curr_resync >= mddev->recovery_cp) {
6343 "md: checkpointing %s of %s.\n",
6344 desc, mdname(mddev));
6345 mddev->recovery_cp = mddev->curr_resync;
6348 mddev->recovery_cp = MaxSector;
6350 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6351 mddev->curr_resync = MaxSector;
6352 list_for_each_entry(rdev, &mddev->disks, same_set)
6353 if (rdev->raid_disk >= 0 &&
6354 !test_bit(Faulty, &rdev->flags) &&
6355 !test_bit(In_sync, &rdev->flags) &&
6356 rdev->recovery_offset < mddev->curr_resync)
6357 rdev->recovery_offset = mddev->curr_resync;
6360 set_bit(MD_CHANGE_DEVS, &mddev->flags);
6363 mddev->curr_resync = 0;
6364 mddev->resync_min = 0;
6365 mddev->resync_max = MaxSector;
6366 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6367 wake_up(&resync_wait);
6368 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6369 md_wakeup_thread(mddev->thread);
6374 * got a signal, exit.
6377 "md: md_do_sync() got signal ... exiting\n");
6378 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6382 EXPORT_SYMBOL_GPL(md_do_sync);
6385 static int remove_and_add_spares(mddev_t *mddev)
6390 mddev->curr_resync_completed = 0;
6392 list_for_each_entry(rdev, &mddev->disks, same_set)
6393 if (rdev->raid_disk >= 0 &&
6394 !test_bit(Blocked, &rdev->flags) &&
6395 (test_bit(Faulty, &rdev->flags) ||
6396 ! test_bit(In_sync, &rdev->flags)) &&
6397 atomic_read(&rdev->nr_pending)==0) {
6398 if (mddev->pers->hot_remove_disk(
6399 mddev, rdev->raid_disk)==0) {
6401 sprintf(nm,"rd%d", rdev->raid_disk);
6402 sysfs_remove_link(&mddev->kobj, nm);
6403 rdev->raid_disk = -1;
6407 if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6408 list_for_each_entry(rdev, &mddev->disks, same_set) {
6409 if (rdev->raid_disk >= 0 &&
6410 !test_bit(In_sync, &rdev->flags) &&
6411 !test_bit(Blocked, &rdev->flags))
6413 if (rdev->raid_disk < 0
6414 && !test_bit(Faulty, &rdev->flags)) {
6415 rdev->recovery_offset = 0;
6417 hot_add_disk(mddev, rdev) == 0) {
6419 sprintf(nm, "rd%d", rdev->raid_disk);
6420 if (sysfs_create_link(&mddev->kobj,
6423 "md: cannot register "
6427 md_new_event(mddev);
6436 * This routine is regularly called by all per-raid-array threads to
6437 * deal with generic issues like resync and super-block update.
6438 * Raid personalities that don't have a thread (linear/raid0) do not
6439 * need this as they never do any recovery or update the superblock.
6441 * It does not do any resync itself, but rather "forks" off other threads
6442 * to do that as needed.
6443 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6444 * "->recovery" and create a thread at ->sync_thread.
6445 * When the thread finishes it sets MD_RECOVERY_DONE
6446 * and wakeups up this thread which will reap the thread and finish up.
6447 * This thread also removes any faulty devices (with nr_pending == 0).
6449 * The overall approach is:
6450 * 1/ if the superblock needs updating, update it.
6451 * 2/ If a recovery thread is running, don't do anything else.
6452 * 3/ If recovery has finished, clean up, possibly marking spares active.
6453 * 4/ If there are any faulty devices, remove them.
6454 * 5/ If array is degraded, try to add spares devices
6455 * 6/ If array has spares or is not in-sync, start a resync thread.
6457 void md_check_recovery(mddev_t *mddev)
6463 bitmap_daemon_work(mddev->bitmap);
6468 if (signal_pending(current)) {
6469 if (mddev->pers->sync_request && !mddev->external) {
6470 printk(KERN_INFO "md: %s in immediate safe mode\n",
6472 mddev->safemode = 2;
6474 flush_signals(current);
6477 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6480 (mddev->flags && !mddev->external) ||
6481 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6482 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6483 (mddev->external == 0 && mddev->safemode == 1) ||
6484 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6485 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6489 if (mddev_trylock(mddev)) {
6493 /* Only thing we do on a ro array is remove
6496 remove_and_add_spares(mddev);
6497 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6501 if (!mddev->external) {
6503 spin_lock_irq(&mddev->write_lock);
6504 if (mddev->safemode &&
6505 !atomic_read(&mddev->writes_pending) &&
6507 mddev->recovery_cp == MaxSector) {
6510 if (mddev->persistent)
6511 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6513 if (mddev->safemode == 1)
6514 mddev->safemode = 0;
6515 spin_unlock_irq(&mddev->write_lock);
6517 sysfs_notify_dirent(mddev->sysfs_state);
6521 md_update_sb(mddev, 0);
6523 list_for_each_entry(rdev, &mddev->disks, same_set)
6524 if (test_and_clear_bit(StateChanged, &rdev->flags))
6525 sysfs_notify_dirent(rdev->sysfs_state);
6528 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6529 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6530 /* resync/recovery still happening */
6531 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6534 if (mddev->sync_thread) {
6535 /* resync has finished, collect result */
6536 md_unregister_thread(mddev->sync_thread);
6537 mddev->sync_thread = NULL;
6538 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6539 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6541 /* activate any spares */
6542 if (mddev->pers->spare_active(mddev))
6543 sysfs_notify(&mddev->kobj, NULL,
6546 md_update_sb(mddev, 1);
6548 /* if array is no-longer degraded, then any saved_raid_disk
6549 * information must be scrapped
6551 if (!mddev->degraded)
6552 list_for_each_entry(rdev, &mddev->disks, same_set)
6553 rdev->saved_raid_disk = -1;
6555 mddev->recovery = 0;
6556 /* flag recovery needed just to double check */
6557 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6558 sysfs_notify_dirent(mddev->sysfs_action);
6559 md_new_event(mddev);
6562 /* Set RUNNING before clearing NEEDED to avoid
6563 * any transients in the value of "sync_action".
6565 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6566 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6567 /* Clear some bits that don't mean anything, but
6570 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6571 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6573 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6575 /* no recovery is running.
6576 * remove any failed drives, then
6577 * add spares if possible.
6578 * Spare are also removed and re-added, to allow
6579 * the personality to fail the re-add.
6582 if (mddev->reshape_position != MaxSector) {
6583 if (mddev->pers->check_reshape(mddev) != 0)
6584 /* Cannot proceed */
6586 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6587 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6588 } else if ((spares = remove_and_add_spares(mddev))) {
6589 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6590 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6591 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6592 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6593 } else if (mddev->recovery_cp < MaxSector) {
6594 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6595 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6596 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6597 /* nothing to be done ... */
6600 if (mddev->pers->sync_request) {
6601 if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6602 /* We are adding a device or devices to an array
6603 * which has the bitmap stored on all devices.
6604 * So make sure all bitmap pages get written
6606 bitmap_write_all(mddev->bitmap);
6608 mddev->sync_thread = md_register_thread(md_do_sync,
6611 if (!mddev->sync_thread) {
6612 printk(KERN_ERR "%s: could not start resync"
6615 /* leave the spares where they are, it shouldn't hurt */
6616 mddev->recovery = 0;
6618 md_wakeup_thread(mddev->sync_thread);
6619 sysfs_notify_dirent(mddev->sysfs_action);
6620 md_new_event(mddev);
6623 if (!mddev->sync_thread) {
6624 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6625 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6627 if (mddev->sysfs_action)
6628 sysfs_notify_dirent(mddev->sysfs_action);
6630 mddev_unlock(mddev);
6634 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6636 sysfs_notify_dirent(rdev->sysfs_state);
6637 wait_event_timeout(rdev->blocked_wait,
6638 !test_bit(Blocked, &rdev->flags),
6639 msecs_to_jiffies(5000));
6640 rdev_dec_pending(rdev, mddev);
6642 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6644 static int md_notify_reboot(struct notifier_block *this,
6645 unsigned long code, void *x)
6647 struct list_head *tmp;
6650 if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6652 printk(KERN_INFO "md: stopping all md devices.\n");
6654 for_each_mddev(mddev, tmp)
6655 if (mddev_trylock(mddev)) {
6656 /* Force a switch to readonly even array
6657 * appears to still be in use. Hence
6660 do_md_stop(mddev, 1, 100);
6661 mddev_unlock(mddev);
6664 * certain more exotic SCSI devices are known to be
6665 * volatile wrt too early system reboots. While the
6666 * right place to handle this issue is the given
6667 * driver, we do want to have a safe RAID driver ...
6674 static struct notifier_block md_notifier = {
6675 .notifier_call = md_notify_reboot,
6677 .priority = INT_MAX, /* before any real devices */
6680 static void md_geninit(void)
6682 dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6684 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6687 static int __init md_init(void)
6689 if (register_blkdev(MD_MAJOR, "md"))
6691 if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6692 unregister_blkdev(MD_MAJOR, "md");
6695 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6696 md_probe, NULL, NULL);
6697 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6698 md_probe, NULL, NULL);
6700 register_reboot_notifier(&md_notifier);
6701 raid_table_header = register_sysctl_table(raid_root_table);
6711 * Searches all registered partitions for autorun RAID arrays
6715 static LIST_HEAD(all_detected_devices);
6716 struct detected_devices_node {
6717 struct list_head list;
6721 void md_autodetect_dev(dev_t dev)
6723 struct detected_devices_node *node_detected_dev;
6725 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6726 if (node_detected_dev) {
6727 node_detected_dev->dev = dev;
6728 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6730 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6731 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6736 static void autostart_arrays(int part)
6739 struct detected_devices_node *node_detected_dev;
6741 int i_scanned, i_passed;
6746 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6748 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6750 node_detected_dev = list_entry(all_detected_devices.next,
6751 struct detected_devices_node, list);
6752 list_del(&node_detected_dev->list);
6753 dev = node_detected_dev->dev;
6754 kfree(node_detected_dev);
6755 rdev = md_import_device(dev,0, 90);
6759 if (test_bit(Faulty, &rdev->flags)) {
6763 set_bit(AutoDetected, &rdev->flags);
6764 list_add(&rdev->same_set, &pending_raid_disks);
6768 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6769 i_scanned, i_passed);
6771 autorun_devices(part);
6774 #endif /* !MODULE */
6776 static __exit void md_exit(void)
6779 struct list_head *tmp;
6781 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6782 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6784 unregister_blkdev(MD_MAJOR,"md");
6785 unregister_blkdev(mdp_major, "mdp");
6786 unregister_reboot_notifier(&md_notifier);
6787 unregister_sysctl_table(raid_table_header);
6788 remove_proc_entry("mdstat", NULL);
6789 for_each_mddev(mddev, tmp) {
6790 export_array(mddev);
6791 mddev->hold_active = 0;
6795 subsys_initcall(md_init);
6796 module_exit(md_exit)
6798 static int get_ro(char *buffer, struct kernel_param *kp)
6800 return sprintf(buffer, "%d", start_readonly);
6802 static int set_ro(const char *val, struct kernel_param *kp)
6805 int num = simple_strtoul(val, &e, 10);
6806 if (*val && (*e == '\0' || *e == '\n')) {
6807 start_readonly = num;
6813 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6814 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6816 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6818 EXPORT_SYMBOL(register_md_personality);
6819 EXPORT_SYMBOL(unregister_md_personality);
6820 EXPORT_SYMBOL(md_error);
6821 EXPORT_SYMBOL(md_done_sync);
6822 EXPORT_SYMBOL(md_write_start);
6823 EXPORT_SYMBOL(md_write_end);
6824 EXPORT_SYMBOL(md_register_thread);
6825 EXPORT_SYMBOL(md_unregister_thread);
6826 EXPORT_SYMBOL(md_wakeup_thread);
6827 EXPORT_SYMBOL(md_check_recovery);
6828 MODULE_LICENSE("GPL");
6830 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);